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 BONDING_DEBUG 1
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>
56 #include <asm/system.h>
59 #include <asm/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/proc_fs.h>
69 #include <linux/seq_file.h>
70 #include <linux/smp.h>
71 #include <linux/if_ether.h>
73 #include <linux/mii.h>
74 #include <linux/ethtool.h>
75 #include <linux/if_vlan.h>
76 #include <linux/if_bonding.h>
77 #include <linux/jiffies.h>
78 #include <net/route.h>
79 #include <net/net_namespace.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 num_grat_arp = 1;
92 static int num_unsol_na = 1;
93 static int miimon = BOND_LINK_MON_INTERV;
94 static int updelay = 0;
95 static int downdelay = 0;
96 static int use_carrier = 1;
97 static char *mode = NULL;
98 static char *primary = NULL;
99 static char *lacp_rate = NULL;
100 static char *ad_select = NULL;
101 static char *xmit_hash_policy = NULL;
102 static int arp_interval = BOND_LINK_ARP_INTERV;
103 static char *arp_ip_target[BOND_MAX_ARP_TARGETS] = { NULL, };
104 static char *arp_validate = NULL;
105 static char *fail_over_mac = NULL;
106 struct bond_params bonding_defaults;
108 module_param(max_bonds, int, 0);
109 MODULE_PARM_DESC(max_bonds, "Max number of bonded devices");
110 module_param(num_grat_arp, int, 0644);
111 MODULE_PARM_DESC(num_grat_arp, "Number of gratuitous ARP packets to send on failover event");
112 module_param(num_unsol_na, int, 0644);
113 MODULE_PARM_DESC(num_unsol_na, "Number of unsolicited IPv6 Neighbor Advertisements packets to send on failover event");
114 module_param(miimon, int, 0);
115 MODULE_PARM_DESC(miimon, "Link check interval in milliseconds");
116 module_param(updelay, int, 0);
117 MODULE_PARM_DESC(updelay, "Delay before considering link up, in milliseconds");
118 module_param(downdelay, int, 0);
119 MODULE_PARM_DESC(downdelay, "Delay before considering link down, "
121 module_param(use_carrier, int, 0);
122 MODULE_PARM_DESC(use_carrier, "Use netif_carrier_ok (vs MII ioctls) in miimon; "
123 "0 for off, 1 for on (default)");
124 module_param(mode, charp, 0);
125 MODULE_PARM_DESC(mode, "Mode of operation : 0 for balance-rr, "
126 "1 for active-backup, 2 for balance-xor, "
127 "3 for broadcast, 4 for 802.3ad, 5 for balance-tlb, "
128 "6 for balance-alb");
129 module_param(primary, charp, 0);
130 MODULE_PARM_DESC(primary, "Primary network device to use");
131 module_param(lacp_rate, charp, 0);
132 MODULE_PARM_DESC(lacp_rate, "LACPDU tx rate to request from 802.3ad partner "
134 module_param(ad_select, charp, 0);
135 MODULE_PARM_DESC(ad_select, "803.ad aggregation selection logic: stable (0, default), bandwidth (1), count (2)");
136 module_param(xmit_hash_policy, charp, 0);
137 MODULE_PARM_DESC(xmit_hash_policy, "XOR hashing method: 0 for layer 2 (default)"
138 ", 1 for layer 3+4");
139 module_param(arp_interval, int, 0);
140 MODULE_PARM_DESC(arp_interval, "arp interval in milliseconds");
141 module_param_array(arp_ip_target, charp, NULL, 0);
142 MODULE_PARM_DESC(arp_ip_target, "arp targets in n.n.n.n form");
143 module_param(arp_validate, charp, 0);
144 MODULE_PARM_DESC(arp_validate, "validate src/dst of ARP probes: none (default), active, backup or all");
145 module_param(fail_over_mac, charp, 0);
146 MODULE_PARM_DESC(fail_over_mac, "For active-backup, do not set all slaves to the same MAC. none (default), active or follow");
148 /*----------------------------- Global variables ----------------------------*/
150 static const char * const version =
151 DRV_DESCRIPTION ": v" DRV_VERSION " (" DRV_RELDATE ")\n";
153 LIST_HEAD(bond_dev_list);
155 #ifdef CONFIG_PROC_FS
156 static struct proc_dir_entry *bond_proc_dir = NULL;
159 extern struct rw_semaphore bonding_rwsem;
160 static __be32 arp_target[BOND_MAX_ARP_TARGETS] = { 0, } ;
161 static int arp_ip_count = 0;
162 static int bond_mode = BOND_MODE_ROUNDROBIN;
163 static int xmit_hashtype= BOND_XMIT_POLICY_LAYER2;
164 static int lacp_fast = 0;
167 struct bond_parm_tbl bond_lacp_tbl[] = {
168 { "slow", AD_LACP_SLOW},
169 { "fast", AD_LACP_FAST},
173 struct bond_parm_tbl bond_mode_tbl[] = {
174 { "balance-rr", BOND_MODE_ROUNDROBIN},
175 { "active-backup", BOND_MODE_ACTIVEBACKUP},
176 { "balance-xor", BOND_MODE_XOR},
177 { "broadcast", BOND_MODE_BROADCAST},
178 { "802.3ad", BOND_MODE_8023AD},
179 { "balance-tlb", BOND_MODE_TLB},
180 { "balance-alb", BOND_MODE_ALB},
184 struct bond_parm_tbl xmit_hashtype_tbl[] = {
185 { "layer2", BOND_XMIT_POLICY_LAYER2},
186 { "layer3+4", BOND_XMIT_POLICY_LAYER34},
187 { "layer2+3", BOND_XMIT_POLICY_LAYER23},
191 struct bond_parm_tbl arp_validate_tbl[] = {
192 { "none", BOND_ARP_VALIDATE_NONE},
193 { "active", BOND_ARP_VALIDATE_ACTIVE},
194 { "backup", BOND_ARP_VALIDATE_BACKUP},
195 { "all", BOND_ARP_VALIDATE_ALL},
199 struct bond_parm_tbl fail_over_mac_tbl[] = {
200 { "none", BOND_FOM_NONE},
201 { "active", BOND_FOM_ACTIVE},
202 { "follow", BOND_FOM_FOLLOW},
206 struct bond_parm_tbl ad_select_tbl[] = {
207 { "stable", BOND_AD_STABLE},
208 { "bandwidth", BOND_AD_BANDWIDTH},
209 { "count", BOND_AD_COUNT},
213 /*-------------------------- Forward declarations ---------------------------*/
215 static void bond_send_gratuitous_arp(struct bonding *bond);
216 static void bond_deinit(struct net_device *bond_dev);
218 /*---------------------------- General routines -----------------------------*/
220 static const char *bond_mode_name(int mode)
223 case BOND_MODE_ROUNDROBIN :
224 return "load balancing (round-robin)";
225 case BOND_MODE_ACTIVEBACKUP :
226 return "fault-tolerance (active-backup)";
228 return "load balancing (xor)";
229 case BOND_MODE_BROADCAST :
230 return "fault-tolerance (broadcast)";
231 case BOND_MODE_8023AD:
232 return "IEEE 802.3ad Dynamic link aggregation";
234 return "transmit load balancing";
236 return "adaptive load balancing";
242 /*---------------------------------- VLAN -----------------------------------*/
245 * bond_add_vlan - add a new vlan id on bond
246 * @bond: bond that got the notification
247 * @vlan_id: the vlan id to add
249 * Returns -ENOMEM if allocation failed.
251 static int bond_add_vlan(struct bonding *bond, unsigned short vlan_id)
253 struct vlan_entry *vlan;
255 dprintk("bond: %s, vlan id %d\n",
256 (bond ? bond->dev->name: "None"), vlan_id);
258 vlan = kzalloc(sizeof(struct vlan_entry), GFP_KERNEL);
263 INIT_LIST_HEAD(&vlan->vlan_list);
264 vlan->vlan_id = vlan_id;
266 write_lock_bh(&bond->lock);
268 list_add_tail(&vlan->vlan_list, &bond->vlan_list);
270 write_unlock_bh(&bond->lock);
272 dprintk("added VLAN ID %d on bond %s\n", vlan_id, bond->dev->name);
278 * bond_del_vlan - delete a vlan id from bond
279 * @bond: bond that got the notification
280 * @vlan_id: the vlan id to delete
282 * returns -ENODEV if @vlan_id was not found in @bond.
284 static int bond_del_vlan(struct bonding *bond, unsigned short vlan_id)
286 struct vlan_entry *vlan;
289 dprintk("bond: %s, vlan id %d\n", bond->dev->name, vlan_id);
291 write_lock_bh(&bond->lock);
293 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
294 if (vlan->vlan_id == vlan_id) {
295 list_del(&vlan->vlan_list);
297 if (bond_is_lb(bond))
298 bond_alb_clear_vlan(bond, vlan_id);
300 dprintk("removed VLAN ID %d from bond %s\n", vlan_id,
305 if (list_empty(&bond->vlan_list) &&
306 (bond->slave_cnt == 0)) {
307 /* Last VLAN removed and no slaves, so
308 * restore block on adding VLANs. This will
309 * be removed once new slaves that are not
310 * VLAN challenged will be added.
312 bond->dev->features |= NETIF_F_VLAN_CHALLENGED;
320 dprintk("couldn't find VLAN ID %d in bond %s\n", vlan_id,
324 write_unlock_bh(&bond->lock);
329 * bond_has_challenged_slaves
330 * @bond: the bond we're working on
332 * Searches the slave list. Returns 1 if a vlan challenged slave
333 * was found, 0 otherwise.
335 * Assumes bond->lock is held.
337 static int bond_has_challenged_slaves(struct bonding *bond)
342 bond_for_each_slave(bond, slave, i) {
343 if (slave->dev->features & NETIF_F_VLAN_CHALLENGED) {
344 dprintk("found VLAN challenged slave - %s\n",
350 dprintk("no VLAN challenged slaves found\n");
355 * bond_next_vlan - safely skip to the next item in the vlans list.
356 * @bond: the bond we're working on
357 * @curr: item we're advancing from
359 * Returns %NULL if list is empty, bond->next_vlan if @curr is %NULL,
360 * or @curr->next otherwise (even if it is @curr itself again).
362 * Caller must hold bond->lock
364 struct vlan_entry *bond_next_vlan(struct bonding *bond, struct vlan_entry *curr)
366 struct vlan_entry *next, *last;
368 if (list_empty(&bond->vlan_list)) {
373 next = list_entry(bond->vlan_list.next,
374 struct vlan_entry, vlan_list);
376 last = list_entry(bond->vlan_list.prev,
377 struct vlan_entry, vlan_list);
379 next = list_entry(bond->vlan_list.next,
380 struct vlan_entry, vlan_list);
382 next = list_entry(curr->vlan_list.next,
383 struct vlan_entry, vlan_list);
391 * bond_dev_queue_xmit - Prepare skb for xmit.
393 * @bond: bond device that got this skb for tx.
394 * @skb: hw accel VLAN tagged skb to transmit
395 * @slave_dev: slave that is supposed to xmit this skbuff
397 * When the bond gets an skb to transmit that is
398 * already hardware accelerated VLAN tagged, and it
399 * needs to relay this skb to a slave that is not
400 * hw accel capable, the skb needs to be "unaccelerated",
401 * i.e. strip the hwaccel tag and re-insert it as part
404 int bond_dev_queue_xmit(struct bonding *bond, struct sk_buff *skb, struct net_device *slave_dev)
406 unsigned short uninitialized_var(vlan_id);
408 if (!list_empty(&bond->vlan_list) &&
409 !(slave_dev->features & NETIF_F_HW_VLAN_TX) &&
410 vlan_get_tag(skb, &vlan_id) == 0) {
411 skb->dev = slave_dev;
412 skb = vlan_put_tag(skb, vlan_id);
414 /* vlan_put_tag() frees the skb in case of error,
415 * so return success here so the calling functions
416 * won't attempt to free is again.
421 skb->dev = slave_dev;
431 * In the following 3 functions, bond_vlan_rx_register(), bond_vlan_rx_add_vid
432 * and bond_vlan_rx_kill_vid, We don't protect the slave list iteration with a
434 * a. This operation is performed in IOCTL context,
435 * b. The operation is protected by the RTNL semaphore in the 8021q code,
436 * c. Holding a lock with BH disabled while directly calling a base driver
437 * entry point is generally a BAD idea.
439 * The design of synchronization/protection for this operation in the 8021q
440 * module is good for one or more VLAN devices over a single physical device
441 * and cannot be extended for a teaming solution like bonding, so there is a
442 * potential race condition here where a net device from the vlan group might
443 * be referenced (either by a base driver or the 8021q code) while it is being
444 * removed from the system. However, it turns out we're not making matters
445 * worse, and if it works for regular VLAN usage it will work here too.
449 * bond_vlan_rx_register - Propagates registration to slaves
450 * @bond_dev: bonding net device that got called
451 * @grp: vlan group being registered
453 static void bond_vlan_rx_register(struct net_device *bond_dev, struct vlan_group *grp)
455 struct bonding *bond = netdev_priv(bond_dev);
461 bond_for_each_slave(bond, slave, i) {
462 struct net_device *slave_dev = slave->dev;
463 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
465 if ((slave_dev->features & NETIF_F_HW_VLAN_RX) &&
466 slave_ops->ndo_vlan_rx_register) {
467 slave_ops->ndo_vlan_rx_register(slave_dev, grp);
473 * bond_vlan_rx_add_vid - Propagates adding an id to slaves
474 * @bond_dev: bonding net device that got called
475 * @vid: vlan id being added
477 static void bond_vlan_rx_add_vid(struct net_device *bond_dev, uint16_t vid)
479 struct bonding *bond = netdev_priv(bond_dev);
483 bond_for_each_slave(bond, slave, i) {
484 struct net_device *slave_dev = slave->dev;
485 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
487 if ((slave_dev->features & NETIF_F_HW_VLAN_FILTER) &&
488 slave_ops->ndo_vlan_rx_add_vid) {
489 slave_ops->ndo_vlan_rx_add_vid(slave_dev, vid);
493 res = bond_add_vlan(bond, vid);
495 printk(KERN_ERR DRV_NAME
496 ": %s: Error: Failed to add vlan id %d\n",
497 bond_dev->name, vid);
502 * bond_vlan_rx_kill_vid - Propagates deleting an id to slaves
503 * @bond_dev: bonding net device that got called
504 * @vid: vlan id being removed
506 static void bond_vlan_rx_kill_vid(struct net_device *bond_dev, uint16_t vid)
508 struct bonding *bond = netdev_priv(bond_dev);
510 struct net_device *vlan_dev;
513 bond_for_each_slave(bond, slave, i) {
514 struct net_device *slave_dev = slave->dev;
515 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
517 if ((slave_dev->features & NETIF_F_HW_VLAN_FILTER) &&
518 slave_ops->ndo_vlan_rx_kill_vid) {
519 /* Save and then restore vlan_dev in the grp array,
520 * since the slave's driver might clear it.
522 vlan_dev = vlan_group_get_device(bond->vlgrp, vid);
523 slave_ops->ndo_vlan_rx_kill_vid(slave_dev, vid);
524 vlan_group_set_device(bond->vlgrp, vid, vlan_dev);
528 res = bond_del_vlan(bond, vid);
530 printk(KERN_ERR DRV_NAME
531 ": %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;
541 write_lock_bh(&bond->lock);
543 if (list_empty(&bond->vlan_list))
546 if ((slave_dev->features & NETIF_F_HW_VLAN_RX) &&
547 slave_ops->ndo_vlan_rx_register)
548 slave_ops->ndo_vlan_rx_register(slave_dev, bond->vlgrp);
550 if (!(slave_dev->features & NETIF_F_HW_VLAN_FILTER) ||
551 !(slave_ops->ndo_vlan_rx_add_vid))
554 list_for_each_entry(vlan, &bond->vlan_list, vlan_list)
555 slave_ops->ndo_vlan_rx_add_vid(slave_dev, vlan->vlan_id);
558 write_unlock_bh(&bond->lock);
561 static void bond_del_vlans_from_slave(struct bonding *bond, struct net_device *slave_dev)
563 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
564 struct vlan_entry *vlan;
565 struct net_device *vlan_dev;
567 write_lock_bh(&bond->lock);
569 if (list_empty(&bond->vlan_list))
572 if (!(slave_dev->features & NETIF_F_HW_VLAN_FILTER) ||
573 !(slave_ops->ndo_vlan_rx_kill_vid))
576 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
577 /* Save and then restore vlan_dev in the grp array,
578 * since the slave's driver might clear it.
580 vlan_dev = vlan_group_get_device(bond->vlgrp, vlan->vlan_id);
581 slave_ops->ndo_vlan_rx_kill_vid(slave_dev, vlan->vlan_id);
582 vlan_group_set_device(bond->vlgrp, vlan->vlan_id, vlan_dev);
586 if ((slave_dev->features & NETIF_F_HW_VLAN_RX) &&
587 slave_ops->ndo_vlan_rx_register)
588 slave_ops->ndo_vlan_rx_register(slave_dev, NULL);
591 write_unlock_bh(&bond->lock);
594 /*------------------------------- Link status -------------------------------*/
597 * Set the carrier state for the master according to the state of its
598 * slaves. If any slaves are up, the master is up. In 802.3ad mode,
599 * do special 802.3ad magic.
601 * Returns zero if carrier state does not change, nonzero if it does.
603 static int bond_set_carrier(struct bonding *bond)
608 if (bond->slave_cnt == 0)
611 if (bond->params.mode == BOND_MODE_8023AD)
612 return bond_3ad_set_carrier(bond);
614 bond_for_each_slave(bond, slave, i) {
615 if (slave->link == BOND_LINK_UP) {
616 if (!netif_carrier_ok(bond->dev)) {
617 netif_carrier_on(bond->dev);
625 if (netif_carrier_ok(bond->dev)) {
626 netif_carrier_off(bond->dev);
633 * Get link speed and duplex from the slave's base driver
634 * using ethtool. If for some reason the call fails or the
635 * values are invalid, fake speed and duplex to 100/Full
638 static int bond_update_speed_duplex(struct slave *slave)
640 struct net_device *slave_dev = slave->dev;
641 struct ethtool_cmd etool;
644 /* Fake speed and duplex */
645 slave->speed = SPEED_100;
646 slave->duplex = DUPLEX_FULL;
648 if (!slave_dev->ethtool_ops || !slave_dev->ethtool_ops->get_settings)
651 res = slave_dev->ethtool_ops->get_settings(slave_dev, &etool);
655 switch (etool.speed) {
665 switch (etool.duplex) {
673 slave->speed = etool.speed;
674 slave->duplex = etool.duplex;
680 * if <dev> supports MII link status reporting, check its link status.
682 * We either do MII/ETHTOOL ioctls, or check netif_carrier_ok(),
683 * depening upon the setting of the use_carrier parameter.
685 * Return either BMSR_LSTATUS, meaning that the link is up (or we
686 * can't tell and just pretend it is), or 0, meaning that the link is
689 * If reporting is non-zero, instead of faking link up, return -1 if
690 * both ETHTOOL and MII ioctls fail (meaning the device does not
691 * support them). If use_carrier is set, return whatever it says.
692 * It'd be nice if there was a good way to tell if a driver supports
693 * netif_carrier, but there really isn't.
695 static int bond_check_dev_link(struct bonding *bond, struct net_device *slave_dev, int reporting)
697 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
698 static int (* ioctl)(struct net_device *, struct ifreq *, int);
700 struct mii_ioctl_data *mii;
702 if (bond->params.use_carrier)
703 return netif_carrier_ok(slave_dev) ? BMSR_LSTATUS : 0;
705 ioctl = slave_ops->ndo_do_ioctl;
707 /* TODO: set pointer to correct ioctl on a per team member */
708 /* bases to make this more efficient. that is, once */
709 /* we determine the correct ioctl, we will always */
710 /* call it and not the others for that team */
714 * We cannot assume that SIOCGMIIPHY will also read a
715 * register; not all network drivers (e.g., e100)
719 /* Yes, the mii is overlaid on the ifreq.ifr_ifru */
720 strncpy(ifr.ifr_name, slave_dev->name, IFNAMSIZ);
722 if (IOCTL(slave_dev, &ifr, SIOCGMIIPHY) == 0) {
723 mii->reg_num = MII_BMSR;
724 if (IOCTL(slave_dev, &ifr, SIOCGMIIREG) == 0) {
725 return (mii->val_out & BMSR_LSTATUS);
731 * Some drivers cache ETHTOOL_GLINK for a period of time so we only
732 * attempt to get link status from it if the above MII ioctls fail.
734 if (slave_dev->ethtool_ops) {
735 if (slave_dev->ethtool_ops->get_link) {
738 link = slave_dev->ethtool_ops->get_link(slave_dev);
740 return link ? BMSR_LSTATUS : 0;
745 * If reporting, report that either there's no dev->do_ioctl,
746 * or both SIOCGMIIREG and get_link failed (meaning that we
747 * cannot report link status). If not reporting, pretend
750 return (reporting ? -1 : BMSR_LSTATUS);
753 /*----------------------------- Multicast list ------------------------------*/
756 * Returns 0 if dmi1 and dmi2 are the same, non-0 otherwise
758 static inline int bond_is_dmi_same(struct dev_mc_list *dmi1, struct dev_mc_list *dmi2)
760 return memcmp(dmi1->dmi_addr, dmi2->dmi_addr, dmi1->dmi_addrlen) == 0 &&
761 dmi1->dmi_addrlen == dmi2->dmi_addrlen;
765 * returns dmi entry if found, NULL otherwise
767 static struct dev_mc_list *bond_mc_list_find_dmi(struct dev_mc_list *dmi, struct dev_mc_list *mc_list)
769 struct dev_mc_list *idmi;
771 for (idmi = mc_list; idmi; idmi = idmi->next) {
772 if (bond_is_dmi_same(dmi, idmi)) {
781 * Push the promiscuity flag down to appropriate slaves
783 static int bond_set_promiscuity(struct bonding *bond, int inc)
786 if (USES_PRIMARY(bond->params.mode)) {
787 /* write lock already acquired */
788 if (bond->curr_active_slave) {
789 err = dev_set_promiscuity(bond->curr_active_slave->dev,
795 bond_for_each_slave(bond, slave, i) {
796 err = dev_set_promiscuity(slave->dev, inc);
805 * Push the allmulti flag down to all slaves
807 static int bond_set_allmulti(struct bonding *bond, int inc)
810 if (USES_PRIMARY(bond->params.mode)) {
811 /* write lock already acquired */
812 if (bond->curr_active_slave) {
813 err = dev_set_allmulti(bond->curr_active_slave->dev,
819 bond_for_each_slave(bond, slave, i) {
820 err = dev_set_allmulti(slave->dev, inc);
829 * Add a Multicast address to slaves
832 static void bond_mc_add(struct bonding *bond, void *addr, int alen)
834 if (USES_PRIMARY(bond->params.mode)) {
835 /* write lock already acquired */
836 if (bond->curr_active_slave) {
837 dev_mc_add(bond->curr_active_slave->dev, addr, alen, 0);
842 bond_for_each_slave(bond, slave, i) {
843 dev_mc_add(slave->dev, addr, alen, 0);
849 * Remove a multicast address from slave
852 static void bond_mc_delete(struct bonding *bond, void *addr, int alen)
854 if (USES_PRIMARY(bond->params.mode)) {
855 /* write lock already acquired */
856 if (bond->curr_active_slave) {
857 dev_mc_delete(bond->curr_active_slave->dev, addr, alen, 0);
862 bond_for_each_slave(bond, slave, i) {
863 dev_mc_delete(slave->dev, addr, alen, 0);
870 * Retrieve the list of registered multicast addresses for the bonding
871 * device and retransmit an IGMP JOIN request to the current active
874 static void bond_resend_igmp_join_requests(struct bonding *bond)
876 struct in_device *in_dev;
877 struct ip_mc_list *im;
880 in_dev = __in_dev_get_rcu(bond->dev);
882 for (im = in_dev->mc_list; im; im = im->next) {
883 ip_mc_rejoin_group(im);
891 * Totally destroys the mc_list in bond
893 static void bond_mc_list_destroy(struct bonding *bond)
895 struct dev_mc_list *dmi;
899 bond->mc_list = dmi->next;
903 bond->mc_list = NULL;
907 * Copy all the Multicast addresses from src to the bonding device dst
909 static int bond_mc_list_copy(struct dev_mc_list *mc_list, struct bonding *bond,
912 struct dev_mc_list *dmi, *new_dmi;
914 for (dmi = mc_list; dmi; dmi = dmi->next) {
915 new_dmi = kmalloc(sizeof(struct dev_mc_list), gfp_flag);
918 /* FIXME: Potential memory leak !!! */
922 new_dmi->next = bond->mc_list;
923 bond->mc_list = new_dmi;
924 new_dmi->dmi_addrlen = dmi->dmi_addrlen;
925 memcpy(new_dmi->dmi_addr, dmi->dmi_addr, dmi->dmi_addrlen);
926 new_dmi->dmi_users = dmi->dmi_users;
927 new_dmi->dmi_gusers = dmi->dmi_gusers;
934 * flush all members of flush->mc_list from device dev->mc_list
936 static void bond_mc_list_flush(struct net_device *bond_dev, struct net_device *slave_dev)
938 struct bonding *bond = netdev_priv(bond_dev);
939 struct dev_mc_list *dmi;
941 for (dmi = bond_dev->mc_list; dmi; dmi = dmi->next) {
942 dev_mc_delete(slave_dev, dmi->dmi_addr, dmi->dmi_addrlen, 0);
945 if (bond->params.mode == BOND_MODE_8023AD) {
946 /* del lacpdu mc addr from mc list */
947 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
949 dev_mc_delete(slave_dev, lacpdu_multicast, ETH_ALEN, 0);
953 /*--------------------------- Active slave change ---------------------------*/
956 * Update the mc list and multicast-related flags for the new and
957 * old active slaves (if any) according to the multicast mode, and
958 * promiscuous flags unconditionally.
960 static void bond_mc_swap(struct bonding *bond, struct slave *new_active, struct slave *old_active)
962 struct dev_mc_list *dmi;
964 if (!USES_PRIMARY(bond->params.mode)) {
965 /* nothing to do - mc list is already up-to-date on
972 if (bond->dev->flags & IFF_PROMISC) {
973 dev_set_promiscuity(old_active->dev, -1);
976 if (bond->dev->flags & IFF_ALLMULTI) {
977 dev_set_allmulti(old_active->dev, -1);
980 for (dmi = bond->dev->mc_list; dmi; dmi = dmi->next) {
981 dev_mc_delete(old_active->dev, dmi->dmi_addr, dmi->dmi_addrlen, 0);
986 /* FIXME: Signal errors upstream. */
987 if (bond->dev->flags & IFF_PROMISC) {
988 dev_set_promiscuity(new_active->dev, 1);
991 if (bond->dev->flags & IFF_ALLMULTI) {
992 dev_set_allmulti(new_active->dev, 1);
995 for (dmi = bond->dev->mc_list; dmi; dmi = dmi->next) {
996 dev_mc_add(new_active->dev, dmi->dmi_addr, dmi->dmi_addrlen, 0);
998 bond_resend_igmp_join_requests(bond);
1003 * bond_do_fail_over_mac
1005 * Perform special MAC address swapping for fail_over_mac settings
1007 * Called with RTNL, bond->lock for read, curr_slave_lock for write_bh.
1009 static void bond_do_fail_over_mac(struct bonding *bond,
1010 struct slave *new_active,
1011 struct slave *old_active)
1013 u8 tmp_mac[ETH_ALEN];
1014 struct sockaddr saddr;
1017 switch (bond->params.fail_over_mac) {
1018 case BOND_FOM_ACTIVE:
1020 memcpy(bond->dev->dev_addr, new_active->dev->dev_addr,
1021 new_active->dev->addr_len);
1023 case BOND_FOM_FOLLOW:
1025 * if new_active && old_active, swap them
1026 * if just old_active, do nothing (going to no active slave)
1027 * if just new_active, set new_active to bond's MAC
1032 write_unlock_bh(&bond->curr_slave_lock);
1033 read_unlock(&bond->lock);
1036 memcpy(tmp_mac, new_active->dev->dev_addr, ETH_ALEN);
1037 memcpy(saddr.sa_data, old_active->dev->dev_addr,
1039 saddr.sa_family = new_active->dev->type;
1041 memcpy(saddr.sa_data, bond->dev->dev_addr, ETH_ALEN);
1042 saddr.sa_family = bond->dev->type;
1045 rv = dev_set_mac_address(new_active->dev, &saddr);
1047 printk(KERN_ERR DRV_NAME
1048 ": %s: Error %d setting MAC of slave %s\n",
1049 bond->dev->name, -rv, new_active->dev->name);
1056 memcpy(saddr.sa_data, tmp_mac, ETH_ALEN);
1057 saddr.sa_family = old_active->dev->type;
1059 rv = dev_set_mac_address(old_active->dev, &saddr);
1061 printk(KERN_ERR DRV_NAME
1062 ": %s: Error %d setting MAC of slave %s\n",
1063 bond->dev->name, -rv, new_active->dev->name);
1065 read_lock(&bond->lock);
1066 write_lock_bh(&bond->curr_slave_lock);
1069 printk(KERN_ERR DRV_NAME
1070 ": %s: bond_do_fail_over_mac impossible: bad policy %d\n",
1071 bond->dev->name, bond->params.fail_over_mac);
1079 * find_best_interface - select the best available slave to be the active one
1080 * @bond: our bonding struct
1082 * Warning: Caller must hold curr_slave_lock for writing.
1084 static struct slave *bond_find_best_slave(struct bonding *bond)
1086 struct slave *new_active, *old_active;
1087 struct slave *bestslave = NULL;
1088 int mintime = bond->params.updelay;
1091 new_active = old_active = bond->curr_active_slave;
1093 if (!new_active) { /* there were no active slaves left */
1094 if (bond->slave_cnt > 0) { /* found one slave */
1095 new_active = bond->first_slave;
1097 return NULL; /* still no slave, return NULL */
1101 /* first try the primary link; if arping, a link must tx/rx traffic
1102 * before it can be considered the curr_active_slave - also, we would skip
1103 * slaves between the curr_active_slave and primary_slave that may be up
1106 if ((bond->primary_slave) &&
1107 (!bond->params.arp_interval) &&
1108 (IS_UP(bond->primary_slave->dev))) {
1109 new_active = bond->primary_slave;
1112 /* remember where to stop iterating over the slaves */
1113 old_active = new_active;
1115 bond_for_each_slave_from(bond, new_active, i, old_active) {
1116 if (IS_UP(new_active->dev)) {
1117 if (new_active->link == BOND_LINK_UP) {
1119 } else if (new_active->link == BOND_LINK_BACK) {
1120 /* link up, but waiting for stabilization */
1121 if (new_active->delay < mintime) {
1122 mintime = new_active->delay;
1123 bestslave = new_active;
1133 * change_active_interface - change the active slave into the specified one
1134 * @bond: our bonding struct
1135 * @new: the new slave to make the active one
1137 * Set the new slave to the bond's settings and unset them on the old
1138 * curr_active_slave.
1139 * Setting include flags, mc-list, promiscuity, allmulti, etc.
1141 * If @new's link state is %BOND_LINK_BACK we'll set it to %BOND_LINK_UP,
1142 * because it is apparently the best available slave we have, even though its
1143 * updelay hasn't timed out yet.
1145 * If new_active is not NULL, caller must hold bond->lock for read and
1146 * curr_slave_lock for write_bh.
1148 void bond_change_active_slave(struct bonding *bond, struct slave *new_active)
1150 struct slave *old_active = bond->curr_active_slave;
1152 if (old_active == new_active) {
1157 new_active->jiffies = jiffies;
1159 if (new_active->link == BOND_LINK_BACK) {
1160 if (USES_PRIMARY(bond->params.mode)) {
1161 printk(KERN_INFO DRV_NAME
1162 ": %s: making interface %s the new "
1163 "active one %d ms earlier.\n",
1164 bond->dev->name, new_active->dev->name,
1165 (bond->params.updelay - new_active->delay) * bond->params.miimon);
1168 new_active->delay = 0;
1169 new_active->link = BOND_LINK_UP;
1171 if (bond->params.mode == BOND_MODE_8023AD) {
1172 bond_3ad_handle_link_change(new_active, BOND_LINK_UP);
1175 if (bond_is_lb(bond))
1176 bond_alb_handle_link_change(bond, new_active, BOND_LINK_UP);
1178 if (USES_PRIMARY(bond->params.mode)) {
1179 printk(KERN_INFO DRV_NAME
1180 ": %s: making interface %s the new "
1182 bond->dev->name, new_active->dev->name);
1187 if (USES_PRIMARY(bond->params.mode)) {
1188 bond_mc_swap(bond, new_active, old_active);
1191 if (bond_is_lb(bond)) {
1192 bond_alb_handle_active_change(bond, new_active);
1194 bond_set_slave_inactive_flags(old_active);
1196 bond_set_slave_active_flags(new_active);
1198 bond->curr_active_slave = new_active;
1201 if (bond->params.mode == BOND_MODE_ACTIVEBACKUP) {
1203 bond_set_slave_inactive_flags(old_active);
1207 bond_set_slave_active_flags(new_active);
1209 if (bond->params.fail_over_mac)
1210 bond_do_fail_over_mac(bond, new_active,
1213 bond->send_grat_arp = bond->params.num_grat_arp;
1214 bond_send_gratuitous_arp(bond);
1216 bond->send_unsol_na = bond->params.num_unsol_na;
1217 bond_send_unsolicited_na(bond);
1219 write_unlock_bh(&bond->curr_slave_lock);
1220 read_unlock(&bond->lock);
1222 netdev_bonding_change(bond->dev);
1224 read_lock(&bond->lock);
1225 write_lock_bh(&bond->curr_slave_lock);
1231 * bond_select_active_slave - select a new active slave, if needed
1232 * @bond: our bonding struct
1234 * This functions shoud be called when one of the following occurs:
1235 * - The old curr_active_slave has been released or lost its link.
1236 * - The primary_slave has got its link back.
1237 * - A slave has got its link back and there's no old curr_active_slave.
1239 * Caller must hold bond->lock for read and curr_slave_lock for write_bh.
1241 void bond_select_active_slave(struct bonding *bond)
1243 struct slave *best_slave;
1246 best_slave = bond_find_best_slave(bond);
1247 if (best_slave != bond->curr_active_slave) {
1248 bond_change_active_slave(bond, best_slave);
1249 rv = bond_set_carrier(bond);
1253 if (netif_carrier_ok(bond->dev)) {
1254 printk(KERN_INFO DRV_NAME
1255 ": %s: first active interface up!\n",
1258 printk(KERN_INFO DRV_NAME ": %s: "
1259 "now running without any active interface !\n",
1265 /*--------------------------- slave list handling ---------------------------*/
1268 * This function attaches the slave to the end of list.
1270 * bond->lock held for writing by caller.
1272 static void bond_attach_slave(struct bonding *bond, struct slave *new_slave)
1274 if (bond->first_slave == NULL) { /* attaching the first slave */
1275 new_slave->next = new_slave;
1276 new_slave->prev = new_slave;
1277 bond->first_slave = new_slave;
1279 new_slave->next = bond->first_slave;
1280 new_slave->prev = bond->first_slave->prev;
1281 new_slave->next->prev = new_slave;
1282 new_slave->prev->next = new_slave;
1289 * This function detaches the slave from the list.
1290 * WARNING: no check is made to verify if the slave effectively
1291 * belongs to <bond>.
1292 * Nothing is freed on return, structures are just unchained.
1293 * If any slave pointer in bond was pointing to <slave>,
1294 * it should be changed by the calling function.
1296 * bond->lock held for writing by caller.
1298 static void bond_detach_slave(struct bonding *bond, struct slave *slave)
1301 slave->next->prev = slave->prev;
1305 slave->prev->next = slave->next;
1308 if (bond->first_slave == slave) { /* slave is the first slave */
1309 if (bond->slave_cnt > 1) { /* there are more slave */
1310 bond->first_slave = slave->next;
1312 bond->first_slave = NULL; /* slave was the last one */
1321 /*---------------------------------- IOCTL ----------------------------------*/
1323 static int bond_sethwaddr(struct net_device *bond_dev,
1324 struct net_device *slave_dev)
1326 dprintk("bond_dev=%p\n", bond_dev);
1327 dprintk("slave_dev=%p\n", slave_dev);
1328 dprintk("slave_dev->addr_len=%d\n", slave_dev->addr_len);
1329 memcpy(bond_dev->dev_addr, slave_dev->dev_addr, slave_dev->addr_len);
1333 #define BOND_VLAN_FEATURES \
1334 (NETIF_F_VLAN_CHALLENGED | NETIF_F_HW_VLAN_RX | NETIF_F_HW_VLAN_TX | \
1335 NETIF_F_HW_VLAN_FILTER)
1338 * Compute the common dev->feature set available to all slaves. Some
1339 * feature bits are managed elsewhere, so preserve those feature bits
1340 * on the master device.
1342 static int bond_compute_features(struct bonding *bond)
1344 struct slave *slave;
1345 struct net_device *bond_dev = bond->dev;
1346 unsigned long features = bond_dev->features;
1347 unsigned short max_hard_header_len = max((u16)ETH_HLEN,
1348 bond_dev->hard_header_len);
1351 features &= ~(NETIF_F_ALL_CSUM | BOND_VLAN_FEATURES);
1352 features |= NETIF_F_GSO_MASK | NETIF_F_NO_CSUM;
1354 if (!bond->first_slave)
1357 features &= ~NETIF_F_ONE_FOR_ALL;
1359 bond_for_each_slave(bond, slave, i) {
1360 features = netdev_increment_features(features,
1361 slave->dev->features,
1362 NETIF_F_ONE_FOR_ALL);
1363 if (slave->dev->hard_header_len > max_hard_header_len)
1364 max_hard_header_len = slave->dev->hard_header_len;
1368 features |= (bond_dev->features & BOND_VLAN_FEATURES);
1369 bond_dev->features = netdev_fix_features(features, NULL);
1370 bond_dev->hard_header_len = max_hard_header_len;
1375 static void bond_setup_by_slave(struct net_device *bond_dev,
1376 struct net_device *slave_dev)
1378 struct bonding *bond = netdev_priv(bond_dev);
1380 bond_dev->header_ops = slave_dev->header_ops;
1382 bond_dev->type = slave_dev->type;
1383 bond_dev->hard_header_len = slave_dev->hard_header_len;
1384 bond_dev->addr_len = slave_dev->addr_len;
1386 memcpy(bond_dev->broadcast, slave_dev->broadcast,
1387 slave_dev->addr_len);
1388 bond->setup_by_slave = 1;
1391 /* enslave device <slave> to bond device <master> */
1392 int bond_enslave(struct net_device *bond_dev, struct net_device *slave_dev)
1394 struct bonding *bond = netdev_priv(bond_dev);
1395 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
1396 struct slave *new_slave = NULL;
1397 struct dev_mc_list *dmi;
1398 struct sockaddr addr;
1400 int old_features = bond_dev->features;
1403 if (!bond->params.use_carrier && slave_dev->ethtool_ops == NULL &&
1404 slave_ops->ndo_do_ioctl == NULL) {
1405 printk(KERN_WARNING DRV_NAME
1406 ": %s: Warning: no link monitoring support for %s\n",
1407 bond_dev->name, slave_dev->name);
1410 /* bond must be initialized by bond_open() before enslaving */
1411 if (!(bond_dev->flags & IFF_UP)) {
1412 printk(KERN_WARNING DRV_NAME
1413 " %s: master_dev is not up in bond_enslave\n",
1417 /* already enslaved */
1418 if (slave_dev->flags & IFF_SLAVE) {
1419 dprintk("Error, Device was already enslaved\n");
1423 /* vlan challenged mutual exclusion */
1424 /* no need to lock since we're protected by rtnl_lock */
1425 if (slave_dev->features & NETIF_F_VLAN_CHALLENGED) {
1426 dprintk("%s: NETIF_F_VLAN_CHALLENGED\n", slave_dev->name);
1427 if (!list_empty(&bond->vlan_list)) {
1428 printk(KERN_ERR DRV_NAME
1429 ": %s: Error: cannot enslave VLAN "
1430 "challenged slave %s on VLAN enabled "
1431 "bond %s\n", bond_dev->name, slave_dev->name,
1435 printk(KERN_WARNING DRV_NAME
1436 ": %s: Warning: enslaved VLAN challenged "
1437 "slave %s. Adding VLANs will be blocked as "
1438 "long as %s is part of bond %s\n",
1439 bond_dev->name, slave_dev->name, slave_dev->name,
1441 bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
1444 dprintk("%s: ! NETIF_F_VLAN_CHALLENGED\n", slave_dev->name);
1445 if (bond->slave_cnt == 0) {
1446 /* First slave, and it is not VLAN challenged,
1447 * so remove the block of adding VLANs over the bond.
1449 bond_dev->features &= ~NETIF_F_VLAN_CHALLENGED;
1454 * Old ifenslave binaries are no longer supported. These can
1455 * be identified with moderate accurary by the state of the slave:
1456 * the current ifenslave will set the interface down prior to
1457 * enslaving it; the old ifenslave will not.
1459 if ((slave_dev->flags & IFF_UP)) {
1460 printk(KERN_ERR DRV_NAME ": %s is up. "
1461 "This may be due to an out of date ifenslave.\n",
1464 goto err_undo_flags;
1467 /* set bonding device ether type by slave - bonding netdevices are
1468 * created with ether_setup, so when the slave type is not ARPHRD_ETHER
1469 * there is a need to override some of the type dependent attribs/funcs.
1471 * bond ether type mutual exclusion - don't allow slaves of dissimilar
1472 * ether type (eg ARPHRD_ETHER and ARPHRD_INFINIBAND) share the same bond
1474 if (bond->slave_cnt == 0) {
1475 if (slave_dev->type != ARPHRD_ETHER)
1476 bond_setup_by_slave(bond_dev, slave_dev);
1477 } else if (bond_dev->type != slave_dev->type) {
1478 printk(KERN_ERR DRV_NAME ": %s ether type (%d) is different "
1479 "from other slaves (%d), can not enslave it.\n",
1481 slave_dev->type, bond_dev->type);
1483 goto err_undo_flags;
1486 if (slave_ops->ndo_set_mac_address == NULL) {
1487 if (bond->slave_cnt == 0) {
1488 printk(KERN_WARNING DRV_NAME
1489 ": %s: Warning: The first slave device "
1490 "specified does not support setting the MAC "
1491 "address. Setting fail_over_mac to active.",
1493 bond->params.fail_over_mac = BOND_FOM_ACTIVE;
1494 } else if (bond->params.fail_over_mac != BOND_FOM_ACTIVE) {
1495 printk(KERN_ERR DRV_NAME
1496 ": %s: Error: The slave device specified "
1497 "does not support setting the MAC address, "
1498 "but fail_over_mac is not set to active.\n"
1501 goto err_undo_flags;
1505 new_slave = kzalloc(sizeof(struct slave), GFP_KERNEL);
1508 goto err_undo_flags;
1511 /* save slave's original flags before calling
1512 * netdev_set_master and dev_open
1514 new_slave->original_flags = slave_dev->flags;
1517 * Save slave's original ("permanent") mac address for modes
1518 * that need it, and for restoring it upon release, and then
1519 * set it to the master's address
1521 memcpy(new_slave->perm_hwaddr, slave_dev->dev_addr, ETH_ALEN);
1523 if (!bond->params.fail_over_mac) {
1525 * Set slave to master's mac address. The application already
1526 * set the master's mac address to that of the first slave
1528 memcpy(addr.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
1529 addr.sa_family = slave_dev->type;
1530 res = dev_set_mac_address(slave_dev, &addr);
1532 dprintk("Error %d calling set_mac_address\n", res);
1537 res = netdev_set_master(slave_dev, bond_dev);
1539 dprintk("Error %d calling netdev_set_master\n", res);
1540 goto err_restore_mac;
1542 /* open the slave since the application closed it */
1543 res = dev_open(slave_dev);
1545 dprintk("Openning slave %s failed\n", slave_dev->name);
1546 goto err_unset_master;
1549 new_slave->dev = slave_dev;
1550 slave_dev->priv_flags |= IFF_BONDING;
1552 if (bond_is_lb(bond)) {
1553 /* bond_alb_init_slave() must be called before all other stages since
1554 * it might fail and we do not want to have to undo everything
1556 res = bond_alb_init_slave(bond, new_slave);
1562 /* If the mode USES_PRIMARY, then the new slave gets the
1563 * master's promisc (and mc) settings only if it becomes the
1564 * curr_active_slave, and that is taken care of later when calling
1565 * bond_change_active()
1567 if (!USES_PRIMARY(bond->params.mode)) {
1568 /* set promiscuity level to new slave */
1569 if (bond_dev->flags & IFF_PROMISC) {
1570 res = dev_set_promiscuity(slave_dev, 1);
1575 /* set allmulti level to new slave */
1576 if (bond_dev->flags & IFF_ALLMULTI) {
1577 res = dev_set_allmulti(slave_dev, 1);
1582 netif_addr_lock_bh(bond_dev);
1583 /* upload master's mc_list to new slave */
1584 for (dmi = bond_dev->mc_list; dmi; dmi = dmi->next) {
1585 dev_mc_add (slave_dev, dmi->dmi_addr, dmi->dmi_addrlen, 0);
1587 netif_addr_unlock_bh(bond_dev);
1590 if (bond->params.mode == BOND_MODE_8023AD) {
1591 /* add lacpdu mc addr to mc list */
1592 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
1594 dev_mc_add(slave_dev, lacpdu_multicast, ETH_ALEN, 0);
1597 bond_add_vlans_on_slave(bond, slave_dev);
1599 write_lock_bh(&bond->lock);
1601 bond_attach_slave(bond, new_slave);
1603 new_slave->delay = 0;
1604 new_slave->link_failure_count = 0;
1606 bond_compute_features(bond);
1608 write_unlock_bh(&bond->lock);
1610 read_lock(&bond->lock);
1612 new_slave->last_arp_rx = jiffies;
1614 if (bond->params.miimon && !bond->params.use_carrier) {
1615 link_reporting = bond_check_dev_link(bond, slave_dev, 1);
1617 if ((link_reporting == -1) && !bond->params.arp_interval) {
1619 * miimon is set but a bonded network driver
1620 * does not support ETHTOOL/MII and
1621 * arp_interval is not set. Note: if
1622 * use_carrier is enabled, we will never go
1623 * here (because netif_carrier is always
1624 * supported); thus, we don't need to change
1625 * the messages for netif_carrier.
1627 printk(KERN_WARNING DRV_NAME
1628 ": %s: Warning: MII and ETHTOOL support not "
1629 "available for interface %s, and "
1630 "arp_interval/arp_ip_target module parameters "
1631 "not specified, thus bonding will not detect "
1632 "link failures! see bonding.txt for details.\n",
1633 bond_dev->name, slave_dev->name);
1634 } else if (link_reporting == -1) {
1635 /* unable get link status using mii/ethtool */
1636 printk(KERN_WARNING DRV_NAME
1637 ": %s: Warning: can't get link status from "
1638 "interface %s; the network driver associated "
1639 "with this interface does not support MII or "
1640 "ETHTOOL link status reporting, thus miimon "
1641 "has no effect on this interface.\n",
1642 bond_dev->name, slave_dev->name);
1646 /* check for initial state */
1647 if (!bond->params.miimon ||
1648 (bond_check_dev_link(bond, slave_dev, 0) == BMSR_LSTATUS)) {
1649 if (bond->params.updelay) {
1650 dprintk("Initial state of slave_dev is "
1651 "BOND_LINK_BACK\n");
1652 new_slave->link = BOND_LINK_BACK;
1653 new_slave->delay = bond->params.updelay;
1655 dprintk("Initial state of slave_dev is "
1657 new_slave->link = BOND_LINK_UP;
1659 new_slave->jiffies = jiffies;
1661 dprintk("Initial state of slave_dev is "
1662 "BOND_LINK_DOWN\n");
1663 new_slave->link = BOND_LINK_DOWN;
1666 if (bond_update_speed_duplex(new_slave) &&
1667 (new_slave->link != BOND_LINK_DOWN)) {
1668 printk(KERN_WARNING DRV_NAME
1669 ": %s: Warning: failed to get speed and duplex from %s, "
1670 "assumed to be 100Mb/sec and Full.\n",
1671 bond_dev->name, new_slave->dev->name);
1673 if (bond->params.mode == BOND_MODE_8023AD) {
1674 printk(KERN_WARNING DRV_NAME
1675 ": %s: Warning: Operation of 802.3ad mode requires ETHTOOL "
1676 "support in base driver for proper aggregator "
1677 "selection.\n", bond_dev->name);
1681 if (USES_PRIMARY(bond->params.mode) && bond->params.primary[0]) {
1682 /* if there is a primary slave, remember it */
1683 if (strcmp(bond->params.primary, new_slave->dev->name) == 0) {
1684 bond->primary_slave = new_slave;
1688 write_lock_bh(&bond->curr_slave_lock);
1690 switch (bond->params.mode) {
1691 case BOND_MODE_ACTIVEBACKUP:
1692 bond_set_slave_inactive_flags(new_slave);
1693 bond_select_active_slave(bond);
1695 case BOND_MODE_8023AD:
1696 /* in 802.3ad mode, the internal mechanism
1697 * will activate the slaves in the selected
1700 bond_set_slave_inactive_flags(new_slave);
1701 /* if this is the first slave */
1702 if (bond->slave_cnt == 1) {
1703 SLAVE_AD_INFO(new_slave).id = 1;
1704 /* Initialize AD with the number of times that the AD timer is called in 1 second
1705 * can be called only after the mac address of the bond is set
1707 bond_3ad_initialize(bond, 1000/AD_TIMER_INTERVAL,
1708 bond->params.lacp_fast);
1710 SLAVE_AD_INFO(new_slave).id =
1711 SLAVE_AD_INFO(new_slave->prev).id + 1;
1714 bond_3ad_bind_slave(new_slave);
1718 new_slave->state = BOND_STATE_ACTIVE;
1719 bond_set_slave_inactive_flags(new_slave);
1722 dprintk("This slave is always active in trunk mode\n");
1724 /* always active in trunk mode */
1725 new_slave->state = BOND_STATE_ACTIVE;
1727 /* In trunking mode there is little meaning to curr_active_slave
1728 * anyway (it holds no special properties of the bond device),
1729 * so we can change it without calling change_active_interface()
1731 if (!bond->curr_active_slave) {
1732 bond->curr_active_slave = new_slave;
1735 } /* switch(bond_mode) */
1737 write_unlock_bh(&bond->curr_slave_lock);
1739 bond_set_carrier(bond);
1741 read_unlock(&bond->lock);
1743 res = bond_create_slave_symlinks(bond_dev, slave_dev);
1747 printk(KERN_INFO DRV_NAME
1748 ": %s: enslaving %s as a%s interface with a%s link.\n",
1749 bond_dev->name, slave_dev->name,
1750 new_slave->state == BOND_STATE_ACTIVE ? "n active" : " backup",
1751 new_slave->link != BOND_LINK_DOWN ? "n up" : " down");
1753 /* enslave is successful */
1756 /* Undo stages on error */
1758 dev_close(slave_dev);
1761 netdev_set_master(slave_dev, NULL);
1764 if (!bond->params.fail_over_mac) {
1765 /* XXX TODO - fom follow mode needs to change master's
1766 * MAC if this slave's MAC is in use by the bond, or at
1767 * least print a warning.
1769 memcpy(addr.sa_data, new_slave->perm_hwaddr, ETH_ALEN);
1770 addr.sa_family = slave_dev->type;
1771 dev_set_mac_address(slave_dev, &addr);
1778 bond_dev->features = old_features;
1784 * Try to release the slave device <slave> from the bond device <master>
1785 * It is legal to access curr_active_slave without a lock because all the function
1788 * The rules for slave state should be:
1789 * for Active/Backup:
1790 * Active stays on all backups go down
1791 * for Bonded connections:
1792 * The first up interface should be left on and all others downed.
1794 int bond_release(struct net_device *bond_dev, struct net_device *slave_dev)
1796 struct bonding *bond = netdev_priv(bond_dev);
1797 struct slave *slave, *oldcurrent;
1798 struct sockaddr addr;
1799 int mac_addr_differ;
1801 /* slave is not a slave or master is not master of this slave */
1802 if (!(slave_dev->flags & IFF_SLAVE) ||
1803 (slave_dev->master != bond_dev)) {
1804 printk(KERN_ERR DRV_NAME
1805 ": %s: Error: cannot release %s.\n",
1806 bond_dev->name, slave_dev->name);
1810 write_lock_bh(&bond->lock);
1812 slave = bond_get_slave_by_dev(bond, slave_dev);
1814 /* not a slave of this bond */
1815 printk(KERN_INFO DRV_NAME
1816 ": %s: %s not enslaved\n",
1817 bond_dev->name, slave_dev->name);
1818 write_unlock_bh(&bond->lock);
1822 if (!bond->params.fail_over_mac) {
1823 mac_addr_differ = memcmp(bond_dev->dev_addr, slave->perm_hwaddr,
1825 if (!mac_addr_differ && (bond->slave_cnt > 1))
1826 printk(KERN_WARNING DRV_NAME
1827 ": %s: Warning: the permanent HWaddr of %s - "
1828 "%pM - is still in use by %s. "
1829 "Set the HWaddr of %s to a different address "
1830 "to avoid conflicts.\n",
1831 bond_dev->name, slave_dev->name,
1833 bond_dev->name, slave_dev->name);
1836 /* Inform AD package of unbinding of slave. */
1837 if (bond->params.mode == BOND_MODE_8023AD) {
1838 /* must be called before the slave is
1839 * detached from the list
1841 bond_3ad_unbind_slave(slave);
1844 printk(KERN_INFO DRV_NAME
1845 ": %s: releasing %s interface %s\n",
1847 (slave->state == BOND_STATE_ACTIVE)
1848 ? "active" : "backup",
1851 oldcurrent = bond->curr_active_slave;
1853 bond->current_arp_slave = NULL;
1855 /* release the slave from its bond */
1856 bond_detach_slave(bond, slave);
1858 bond_compute_features(bond);
1860 if (bond->primary_slave == slave) {
1861 bond->primary_slave = NULL;
1864 if (oldcurrent == slave) {
1865 bond_change_active_slave(bond, NULL);
1868 if (bond_is_lb(bond)) {
1869 /* Must be called only after the slave has been
1870 * detached from the list and the curr_active_slave
1871 * has been cleared (if our_slave == old_current),
1872 * but before a new active slave is selected.
1874 write_unlock_bh(&bond->lock);
1875 bond_alb_deinit_slave(bond, slave);
1876 write_lock_bh(&bond->lock);
1879 if (oldcurrent == slave) {
1881 * Note that we hold RTNL over this sequence, so there
1882 * is no concern that another slave add/remove event
1885 write_unlock_bh(&bond->lock);
1886 read_lock(&bond->lock);
1887 write_lock_bh(&bond->curr_slave_lock);
1889 bond_select_active_slave(bond);
1891 write_unlock_bh(&bond->curr_slave_lock);
1892 read_unlock(&bond->lock);
1893 write_lock_bh(&bond->lock);
1896 if (bond->slave_cnt == 0) {
1897 bond_set_carrier(bond);
1899 /* if the last slave was removed, zero the mac address
1900 * of the master so it will be set by the application
1901 * to the mac address of the first slave
1903 memset(bond_dev->dev_addr, 0, bond_dev->addr_len);
1905 if (list_empty(&bond->vlan_list)) {
1906 bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
1908 printk(KERN_WARNING DRV_NAME
1909 ": %s: Warning: clearing HW address of %s while it "
1910 "still has VLANs.\n",
1911 bond_dev->name, bond_dev->name);
1912 printk(KERN_WARNING DRV_NAME
1913 ": %s: When re-adding slaves, make sure the bond's "
1914 "HW address matches its VLANs'.\n",
1917 } else if ((bond_dev->features & NETIF_F_VLAN_CHALLENGED) &&
1918 !bond_has_challenged_slaves(bond)) {
1919 printk(KERN_INFO DRV_NAME
1920 ": %s: last VLAN challenged slave %s "
1921 "left bond %s. VLAN blocking is removed\n",
1922 bond_dev->name, slave_dev->name, bond_dev->name);
1923 bond_dev->features &= ~NETIF_F_VLAN_CHALLENGED;
1926 write_unlock_bh(&bond->lock);
1928 /* must do this from outside any spinlocks */
1929 bond_destroy_slave_symlinks(bond_dev, slave_dev);
1931 bond_del_vlans_from_slave(bond, slave_dev);
1933 /* If the mode USES_PRIMARY, then we should only remove its
1934 * promisc and mc settings if it was the curr_active_slave, but that was
1935 * already taken care of above when we detached the slave
1937 if (!USES_PRIMARY(bond->params.mode)) {
1938 /* unset promiscuity level from slave */
1939 if (bond_dev->flags & IFF_PROMISC) {
1940 dev_set_promiscuity(slave_dev, -1);
1943 /* unset allmulti level from slave */
1944 if (bond_dev->flags & IFF_ALLMULTI) {
1945 dev_set_allmulti(slave_dev, -1);
1948 /* flush master's mc_list from slave */
1949 netif_addr_lock_bh(bond_dev);
1950 bond_mc_list_flush(bond_dev, slave_dev);
1951 netif_addr_unlock_bh(bond_dev);
1954 netdev_set_master(slave_dev, NULL);
1956 /* close slave before restoring its mac address */
1957 dev_close(slave_dev);
1959 if (bond->params.fail_over_mac != BOND_FOM_ACTIVE) {
1960 /* restore original ("permanent") mac address */
1961 memcpy(addr.sa_data, slave->perm_hwaddr, ETH_ALEN);
1962 addr.sa_family = slave_dev->type;
1963 dev_set_mac_address(slave_dev, &addr);
1966 slave_dev->priv_flags &= ~(IFF_MASTER_8023AD | IFF_MASTER_ALB |
1967 IFF_SLAVE_INACTIVE | IFF_BONDING |
1972 return 0; /* deletion OK */
1976 * Destroy a bonding device.
1977 * Must be under rtnl_lock when this function is called.
1979 void bond_destroy(struct bonding *bond)
1981 bond_deinit(bond->dev);
1982 bond_destroy_sysfs_entry(bond);
1983 unregister_netdevice(bond->dev);
1986 static void bond_destructor(struct net_device *bond_dev)
1988 struct bonding *bond = netdev_priv(bond_dev);
1991 destroy_workqueue(bond->wq);
1993 netif_addr_lock_bh(bond_dev);
1994 bond_mc_list_destroy(bond);
1995 netif_addr_unlock_bh(bond_dev);
1997 free_netdev(bond_dev);
2001 * First release a slave and than destroy the bond if no more slaves iare left.
2002 * Must be under rtnl_lock when this function is called.
2004 int bond_release_and_destroy(struct net_device *bond_dev, struct net_device *slave_dev)
2006 struct bonding *bond = netdev_priv(bond_dev);
2009 ret = bond_release(bond_dev, slave_dev);
2010 if ((ret == 0) && (bond->slave_cnt == 0)) {
2011 printk(KERN_INFO DRV_NAME ": %s: destroying bond %s.\n",
2012 bond_dev->name, bond_dev->name);
2019 * This function releases all slaves.
2021 static int bond_release_all(struct net_device *bond_dev)
2023 struct bonding *bond = netdev_priv(bond_dev);
2024 struct slave *slave;
2025 struct net_device *slave_dev;
2026 struct sockaddr addr;
2028 write_lock_bh(&bond->lock);
2030 netif_carrier_off(bond_dev);
2032 if (bond->slave_cnt == 0) {
2036 bond->current_arp_slave = NULL;
2037 bond->primary_slave = NULL;
2038 bond_change_active_slave(bond, NULL);
2040 while ((slave = bond->first_slave) != NULL) {
2041 /* Inform AD package of unbinding of slave
2042 * before slave is detached from the list.
2044 if (bond->params.mode == BOND_MODE_8023AD) {
2045 bond_3ad_unbind_slave(slave);
2048 slave_dev = slave->dev;
2049 bond_detach_slave(bond, slave);
2051 /* now that the slave is detached, unlock and perform
2052 * all the undo steps that should not be called from
2055 write_unlock_bh(&bond->lock);
2057 if (bond_is_lb(bond)) {
2058 /* must be called only after the slave
2059 * has been detached from the list
2061 bond_alb_deinit_slave(bond, slave);
2064 bond_compute_features(bond);
2066 bond_destroy_slave_symlinks(bond_dev, slave_dev);
2067 bond_del_vlans_from_slave(bond, slave_dev);
2069 /* If the mode USES_PRIMARY, then we should only remove its
2070 * promisc and mc settings if it was the curr_active_slave, but that was
2071 * already taken care of above when we detached the slave
2073 if (!USES_PRIMARY(bond->params.mode)) {
2074 /* unset promiscuity level from slave */
2075 if (bond_dev->flags & IFF_PROMISC) {
2076 dev_set_promiscuity(slave_dev, -1);
2079 /* unset allmulti level from slave */
2080 if (bond_dev->flags & IFF_ALLMULTI) {
2081 dev_set_allmulti(slave_dev, -1);
2084 /* flush master's mc_list from slave */
2085 netif_addr_lock_bh(bond_dev);
2086 bond_mc_list_flush(bond_dev, slave_dev);
2087 netif_addr_unlock_bh(bond_dev);
2090 netdev_set_master(slave_dev, NULL);
2092 /* close slave before restoring its mac address */
2093 dev_close(slave_dev);
2095 if (!bond->params.fail_over_mac) {
2096 /* restore original ("permanent") mac address*/
2097 memcpy(addr.sa_data, slave->perm_hwaddr, ETH_ALEN);
2098 addr.sa_family = slave_dev->type;
2099 dev_set_mac_address(slave_dev, &addr);
2102 slave_dev->priv_flags &= ~(IFF_MASTER_8023AD | IFF_MASTER_ALB |
2103 IFF_SLAVE_INACTIVE);
2107 /* re-acquire the lock before getting the next slave */
2108 write_lock_bh(&bond->lock);
2111 /* zero the mac address of the master so it will be
2112 * set by the application to the mac address of the
2115 memset(bond_dev->dev_addr, 0, bond_dev->addr_len);
2117 if (list_empty(&bond->vlan_list)) {
2118 bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
2120 printk(KERN_WARNING DRV_NAME
2121 ": %s: Warning: clearing HW address of %s while it "
2122 "still has VLANs.\n",
2123 bond_dev->name, bond_dev->name);
2124 printk(KERN_WARNING DRV_NAME
2125 ": %s: When re-adding slaves, make sure the bond's "
2126 "HW address matches its VLANs'.\n",
2130 printk(KERN_INFO DRV_NAME
2131 ": %s: released all slaves\n",
2135 write_unlock_bh(&bond->lock);
2141 * This function changes the active slave to slave <slave_dev>.
2142 * It returns -EINVAL in the following cases.
2143 * - <slave_dev> is not found in the list.
2144 * - There is not active slave now.
2145 * - <slave_dev> is already active.
2146 * - The link state of <slave_dev> is not BOND_LINK_UP.
2147 * - <slave_dev> is not running.
2148 * In these cases, this fuction does nothing.
2149 * In the other cases, currnt_slave pointer is changed and 0 is returned.
2151 static int bond_ioctl_change_active(struct net_device *bond_dev, struct net_device *slave_dev)
2153 struct bonding *bond = netdev_priv(bond_dev);
2154 struct slave *old_active = NULL;
2155 struct slave *new_active = NULL;
2158 if (!USES_PRIMARY(bond->params.mode)) {
2162 /* Verify that master_dev is indeed the master of slave_dev */
2163 if (!(slave_dev->flags & IFF_SLAVE) ||
2164 (slave_dev->master != bond_dev)) {
2168 read_lock(&bond->lock);
2170 read_lock(&bond->curr_slave_lock);
2171 old_active = bond->curr_active_slave;
2172 read_unlock(&bond->curr_slave_lock);
2174 new_active = bond_get_slave_by_dev(bond, slave_dev);
2177 * Changing to the current active: do nothing; return success.
2179 if (new_active && (new_active == old_active)) {
2180 read_unlock(&bond->lock);
2186 (new_active->link == BOND_LINK_UP) &&
2187 IS_UP(new_active->dev)) {
2188 write_lock_bh(&bond->curr_slave_lock);
2189 bond_change_active_slave(bond, new_active);
2190 write_unlock_bh(&bond->curr_slave_lock);
2195 read_unlock(&bond->lock);
2200 static int bond_info_query(struct net_device *bond_dev, struct ifbond *info)
2202 struct bonding *bond = netdev_priv(bond_dev);
2204 info->bond_mode = bond->params.mode;
2205 info->miimon = bond->params.miimon;
2207 read_lock(&bond->lock);
2208 info->num_slaves = bond->slave_cnt;
2209 read_unlock(&bond->lock);
2214 static int bond_slave_info_query(struct net_device *bond_dev, struct ifslave *info)
2216 struct bonding *bond = netdev_priv(bond_dev);
2217 struct slave *slave;
2220 if (info->slave_id < 0) {
2224 read_lock(&bond->lock);
2226 bond_for_each_slave(bond, slave, i) {
2227 if (i == (int)info->slave_id) {
2233 read_unlock(&bond->lock);
2236 strcpy(info->slave_name, slave->dev->name);
2237 info->link = slave->link;
2238 info->state = slave->state;
2239 info->link_failure_count = slave->link_failure_count;
2247 /*-------------------------------- Monitoring -------------------------------*/
2250 static int bond_miimon_inspect(struct bonding *bond)
2252 struct slave *slave;
2253 int i, link_state, commit = 0;
2255 bond_for_each_slave(bond, slave, i) {
2256 slave->new_link = BOND_LINK_NOCHANGE;
2258 link_state = bond_check_dev_link(bond, slave->dev, 0);
2260 switch (slave->link) {
2265 slave->link = BOND_LINK_FAIL;
2266 slave->delay = bond->params.downdelay;
2268 printk(KERN_INFO DRV_NAME
2269 ": %s: link status down for %s"
2270 "interface %s, disabling it in %d ms.\n",
2272 (bond->params.mode ==
2273 BOND_MODE_ACTIVEBACKUP) ?
2274 ((slave->state == BOND_STATE_ACTIVE) ?
2275 "active " : "backup ") : "",
2277 bond->params.downdelay * bond->params.miimon);
2280 case BOND_LINK_FAIL:
2283 * recovered before downdelay expired
2285 slave->link = BOND_LINK_UP;
2286 slave->jiffies = jiffies;
2287 printk(KERN_INFO DRV_NAME
2288 ": %s: link status up again after %d "
2289 "ms for interface %s.\n",
2291 (bond->params.downdelay - slave->delay) *
2292 bond->params.miimon,
2297 if (slave->delay <= 0) {
2298 slave->new_link = BOND_LINK_DOWN;
2306 case BOND_LINK_DOWN:
2310 slave->link = BOND_LINK_BACK;
2311 slave->delay = bond->params.updelay;
2314 printk(KERN_INFO DRV_NAME
2315 ": %s: link status up for "
2316 "interface %s, enabling it in %d ms.\n",
2317 bond->dev->name, slave->dev->name,
2318 bond->params.updelay *
2319 bond->params.miimon);
2322 case BOND_LINK_BACK:
2324 slave->link = BOND_LINK_DOWN;
2325 printk(KERN_INFO DRV_NAME
2326 ": %s: link status down again after %d "
2327 "ms for interface %s.\n",
2329 (bond->params.updelay - slave->delay) *
2330 bond->params.miimon,
2336 if (slave->delay <= 0) {
2337 slave->new_link = BOND_LINK_UP;
2350 static void bond_miimon_commit(struct bonding *bond)
2352 struct slave *slave;
2355 bond_for_each_slave(bond, slave, i) {
2356 switch (slave->new_link) {
2357 case BOND_LINK_NOCHANGE:
2361 slave->link = BOND_LINK_UP;
2362 slave->jiffies = jiffies;
2364 if (bond->params.mode == BOND_MODE_8023AD) {
2365 /* prevent it from being the active one */
2366 slave->state = BOND_STATE_BACKUP;
2367 } else if (bond->params.mode != BOND_MODE_ACTIVEBACKUP) {
2368 /* make it immediately active */
2369 slave->state = BOND_STATE_ACTIVE;
2370 } else if (slave != bond->primary_slave) {
2371 /* prevent it from being the active one */
2372 slave->state = BOND_STATE_BACKUP;
2375 printk(KERN_INFO DRV_NAME
2376 ": %s: link status definitely "
2377 "up for interface %s.\n",
2378 bond->dev->name, slave->dev->name);
2380 /* notify ad that the link status has changed */
2381 if (bond->params.mode == BOND_MODE_8023AD)
2382 bond_3ad_handle_link_change(slave, BOND_LINK_UP);
2384 if (bond_is_lb(bond))
2385 bond_alb_handle_link_change(bond, slave,
2388 if (!bond->curr_active_slave ||
2389 (slave == bond->primary_slave))
2394 case BOND_LINK_DOWN:
2395 if (slave->link_failure_count < UINT_MAX)
2396 slave->link_failure_count++;
2398 slave->link = BOND_LINK_DOWN;
2400 if (bond->params.mode == BOND_MODE_ACTIVEBACKUP ||
2401 bond->params.mode == BOND_MODE_8023AD)
2402 bond_set_slave_inactive_flags(slave);
2404 printk(KERN_INFO DRV_NAME
2405 ": %s: link status definitely down for "
2406 "interface %s, disabling it\n",
2407 bond->dev->name, slave->dev->name);
2409 if (bond->params.mode == BOND_MODE_8023AD)
2410 bond_3ad_handle_link_change(slave,
2413 if (bond->params.mode == BOND_MODE_TLB ||
2414 bond->params.mode == BOND_MODE_ALB)
2415 bond_alb_handle_link_change(bond, slave,
2418 if (slave == bond->curr_active_slave)
2424 printk(KERN_ERR DRV_NAME
2425 ": %s: invalid new link %d on slave %s\n",
2426 bond->dev->name, slave->new_link,
2428 slave->new_link = BOND_LINK_NOCHANGE;
2435 write_lock_bh(&bond->curr_slave_lock);
2436 bond_select_active_slave(bond);
2437 write_unlock_bh(&bond->curr_slave_lock);
2440 bond_set_carrier(bond);
2446 * Really a wrapper that splits the mii monitor into two phases: an
2447 * inspection, then (if inspection indicates something needs to be done)
2448 * an acquisition of appropriate locks followed by a commit phase to
2449 * implement whatever link state changes are indicated.
2451 void bond_mii_monitor(struct work_struct *work)
2453 struct bonding *bond = container_of(work, struct bonding,
2456 read_lock(&bond->lock);
2457 if (bond->kill_timers)
2460 if (bond->slave_cnt == 0)
2463 if (bond->send_grat_arp) {
2464 read_lock(&bond->curr_slave_lock);
2465 bond_send_gratuitous_arp(bond);
2466 read_unlock(&bond->curr_slave_lock);
2469 if (bond->send_unsol_na) {
2470 read_lock(&bond->curr_slave_lock);
2471 bond_send_unsolicited_na(bond);
2472 read_unlock(&bond->curr_slave_lock);
2475 if (bond_miimon_inspect(bond)) {
2476 read_unlock(&bond->lock);
2478 read_lock(&bond->lock);
2480 bond_miimon_commit(bond);
2482 read_unlock(&bond->lock);
2483 rtnl_unlock(); /* might sleep, hold no other locks */
2484 read_lock(&bond->lock);
2488 if (bond->params.miimon)
2489 queue_delayed_work(bond->wq, &bond->mii_work,
2490 msecs_to_jiffies(bond->params.miimon));
2492 read_unlock(&bond->lock);
2495 static __be32 bond_glean_dev_ip(struct net_device *dev)
2497 struct in_device *idev;
2498 struct in_ifaddr *ifa;
2505 idev = __in_dev_get_rcu(dev);
2509 ifa = idev->ifa_list;
2513 addr = ifa->ifa_local;
2519 static int bond_has_this_ip(struct bonding *bond, __be32 ip)
2521 struct vlan_entry *vlan;
2523 if (ip == bond->master_ip)
2526 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
2527 if (ip == vlan->vlan_ip)
2535 * We go to the (large) trouble of VLAN tagging ARP frames because
2536 * switches in VLAN mode (especially if ports are configured as
2537 * "native" to a VLAN) might not pass non-tagged frames.
2539 static void bond_arp_send(struct net_device *slave_dev, int arp_op, __be32 dest_ip, __be32 src_ip, unsigned short vlan_id)
2541 struct sk_buff *skb;
2543 dprintk("arp %d on slave %s: dst %x src %x vid %d\n", arp_op,
2544 slave_dev->name, dest_ip, src_ip, vlan_id);
2546 skb = arp_create(arp_op, ETH_P_ARP, dest_ip, slave_dev, src_ip,
2547 NULL, slave_dev->dev_addr, NULL);
2550 printk(KERN_ERR DRV_NAME ": ARP packet allocation failed\n");
2554 skb = vlan_put_tag(skb, vlan_id);
2556 printk(KERN_ERR DRV_NAME ": failed to insert VLAN tag\n");
2564 static void bond_arp_send_all(struct bonding *bond, struct slave *slave)
2567 __be32 *targets = bond->params.arp_targets;
2568 struct vlan_entry *vlan;
2569 struct net_device *vlan_dev;
2573 for (i = 0; (i < BOND_MAX_ARP_TARGETS); i++) {
2576 dprintk("basa: target %x\n", targets[i]);
2577 if (list_empty(&bond->vlan_list)) {
2578 dprintk("basa: empty vlan: arp_send\n");
2579 bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2580 bond->master_ip, 0);
2585 * If VLANs are configured, we do a route lookup to
2586 * determine which VLAN interface would be used, so we
2587 * can tag the ARP with the proper VLAN tag.
2589 memset(&fl, 0, sizeof(fl));
2590 fl.fl4_dst = targets[i];
2591 fl.fl4_tos = RTO_ONLINK;
2593 rv = ip_route_output_key(&init_net, &rt, &fl);
2595 if (net_ratelimit()) {
2596 printk(KERN_WARNING DRV_NAME
2597 ": %s: no route to arp_ip_target %pI4\n",
2598 bond->dev->name, &fl.fl4_dst);
2604 * This target is not on a VLAN
2606 if (rt->u.dst.dev == bond->dev) {
2608 dprintk("basa: rtdev == bond->dev: arp_send\n");
2609 bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2610 bond->master_ip, 0);
2615 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
2616 vlan_dev = vlan_group_get_device(bond->vlgrp, vlan->vlan_id);
2617 if (vlan_dev == rt->u.dst.dev) {
2618 vlan_id = vlan->vlan_id;
2619 dprintk("basa: vlan match on %s %d\n",
2620 vlan_dev->name, vlan_id);
2627 bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2628 vlan->vlan_ip, vlan_id);
2632 if (net_ratelimit()) {
2633 printk(KERN_WARNING DRV_NAME
2634 ": %s: no path to arp_ip_target %pI4 via rt.dev %s\n",
2635 bond->dev->name, &fl.fl4_dst,
2636 rt->u.dst.dev ? rt->u.dst.dev->name : "NULL");
2643 * Kick out a gratuitous ARP for an IP on the bonding master plus one
2644 * for each VLAN above us.
2646 * Caller must hold curr_slave_lock for read or better
2648 static void bond_send_gratuitous_arp(struct bonding *bond)
2650 struct slave *slave = bond->curr_active_slave;
2651 struct vlan_entry *vlan;
2652 struct net_device *vlan_dev;
2654 dprintk("bond_send_grat_arp: bond %s slave %s\n", bond->dev->name,
2655 slave ? slave->dev->name : "NULL");
2657 if (!slave || !bond->send_grat_arp ||
2658 test_bit(__LINK_STATE_LINKWATCH_PENDING, &slave->dev->state))
2661 bond->send_grat_arp--;
2663 if (bond->master_ip) {
2664 bond_arp_send(slave->dev, ARPOP_REPLY, bond->master_ip,
2665 bond->master_ip, 0);
2668 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
2669 vlan_dev = vlan_group_get_device(bond->vlgrp, vlan->vlan_id);
2670 if (vlan->vlan_ip) {
2671 bond_arp_send(slave->dev, ARPOP_REPLY, vlan->vlan_ip,
2672 vlan->vlan_ip, vlan->vlan_id);
2677 static void bond_validate_arp(struct bonding *bond, struct slave *slave, __be32 sip, __be32 tip)
2680 __be32 *targets = bond->params.arp_targets;
2682 targets = bond->params.arp_targets;
2683 for (i = 0; (i < BOND_MAX_ARP_TARGETS) && targets[i]; i++) {
2684 dprintk("bva: sip %pI4 tip %pI4 t[%d] %pI4 bhti(tip) %d\n",
2685 &sip, &tip, i, &targets[i], bond_has_this_ip(bond, tip));
2686 if (sip == targets[i]) {
2687 if (bond_has_this_ip(bond, tip))
2688 slave->last_arp_rx = jiffies;
2694 static int bond_arp_rcv(struct sk_buff *skb, struct net_device *dev, struct packet_type *pt, struct net_device *orig_dev)
2697 struct slave *slave;
2698 struct bonding *bond;
2699 unsigned char *arp_ptr;
2702 if (dev_net(dev) != &init_net)
2705 if (!(dev->priv_flags & IFF_BONDING) || !(dev->flags & IFF_MASTER))
2708 bond = netdev_priv(dev);
2709 read_lock(&bond->lock);
2711 dprintk("bond_arp_rcv: bond %s skb->dev %s orig_dev %s\n",
2712 bond->dev->name, skb->dev ? skb->dev->name : "NULL",
2713 orig_dev ? orig_dev->name : "NULL");
2715 slave = bond_get_slave_by_dev(bond, orig_dev);
2716 if (!slave || !slave_do_arp_validate(bond, slave))
2719 if (!pskb_may_pull(skb, arp_hdr_len(dev)))
2723 if (arp->ar_hln != dev->addr_len ||
2724 skb->pkt_type == PACKET_OTHERHOST ||
2725 skb->pkt_type == PACKET_LOOPBACK ||
2726 arp->ar_hrd != htons(ARPHRD_ETHER) ||
2727 arp->ar_pro != htons(ETH_P_IP) ||
2731 arp_ptr = (unsigned char *)(arp + 1);
2732 arp_ptr += dev->addr_len;
2733 memcpy(&sip, arp_ptr, 4);
2734 arp_ptr += 4 + dev->addr_len;
2735 memcpy(&tip, arp_ptr, 4);
2737 dprintk("bond_arp_rcv: %s %s/%d av %d sv %d sip %pI4 tip %pI4\n",
2738 bond->dev->name, slave->dev->name, slave->state,
2739 bond->params.arp_validate, slave_do_arp_validate(bond, slave),
2743 * Backup slaves won't see the ARP reply, but do come through
2744 * here for each ARP probe (so we swap the sip/tip to validate
2745 * the probe). In a "redundant switch, common router" type of
2746 * configuration, the ARP probe will (hopefully) travel from
2747 * the active, through one switch, the router, then the other
2748 * switch before reaching the backup.
2750 if (slave->state == BOND_STATE_ACTIVE)
2751 bond_validate_arp(bond, slave, sip, tip);
2753 bond_validate_arp(bond, slave, tip, sip);
2756 read_unlock(&bond->lock);
2759 return NET_RX_SUCCESS;
2763 * this function is called regularly to monitor each slave's link
2764 * ensuring that traffic is being sent and received when arp monitoring
2765 * is used in load-balancing mode. if the adapter has been dormant, then an
2766 * arp is transmitted to generate traffic. see activebackup_arp_monitor for
2767 * arp monitoring in active backup mode.
2769 void bond_loadbalance_arp_mon(struct work_struct *work)
2771 struct bonding *bond = container_of(work, struct bonding,
2773 struct slave *slave, *oldcurrent;
2774 int do_failover = 0;
2778 read_lock(&bond->lock);
2780 delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
2782 if (bond->kill_timers) {
2786 if (bond->slave_cnt == 0) {
2790 read_lock(&bond->curr_slave_lock);
2791 oldcurrent = bond->curr_active_slave;
2792 read_unlock(&bond->curr_slave_lock);
2794 /* see if any of the previous devices are up now (i.e. they have
2795 * xmt and rcv traffic). the curr_active_slave does not come into
2796 * the picture unless it is null. also, slave->jiffies is not needed
2797 * here because we send an arp on each slave and give a slave as
2798 * long as it needs to get the tx/rx within the delta.
2799 * TODO: what about up/down delay in arp mode? it wasn't here before
2802 bond_for_each_slave(bond, slave, i) {
2803 if (slave->link != BOND_LINK_UP) {
2804 if (time_before_eq(jiffies, slave->dev->trans_start + delta_in_ticks) &&
2805 time_before_eq(jiffies, slave->dev->last_rx + delta_in_ticks)) {
2807 slave->link = BOND_LINK_UP;
2808 slave->state = BOND_STATE_ACTIVE;
2810 /* primary_slave has no meaning in round-robin
2811 * mode. the window of a slave being up and
2812 * curr_active_slave being null after enslaving
2816 printk(KERN_INFO DRV_NAME
2817 ": %s: link status definitely "
2818 "up for interface %s, ",
2823 printk(KERN_INFO DRV_NAME
2824 ": %s: interface %s is now up\n",
2830 /* slave->link == BOND_LINK_UP */
2832 /* not all switches will respond to an arp request
2833 * when the source ip is 0, so don't take the link down
2834 * if we don't know our ip yet
2836 if (time_after_eq(jiffies, slave->dev->trans_start + 2*delta_in_ticks) ||
2837 (time_after_eq(jiffies, slave->dev->last_rx + 2*delta_in_ticks))) {
2839 slave->link = BOND_LINK_DOWN;
2840 slave->state = BOND_STATE_BACKUP;
2842 if (slave->link_failure_count < UINT_MAX) {
2843 slave->link_failure_count++;
2846 printk(KERN_INFO DRV_NAME
2847 ": %s: interface %s is now down.\n",
2851 if (slave == oldcurrent) {
2857 /* note: if switch is in round-robin mode, all links
2858 * must tx arp to ensure all links rx an arp - otherwise
2859 * links may oscillate or not come up at all; if switch is
2860 * in something like xor mode, there is nothing we can
2861 * do - all replies will be rx'ed on same link causing slaves
2862 * to be unstable during low/no traffic periods
2864 if (IS_UP(slave->dev)) {
2865 bond_arp_send_all(bond, slave);
2870 write_lock_bh(&bond->curr_slave_lock);
2872 bond_select_active_slave(bond);
2874 write_unlock_bh(&bond->curr_slave_lock);
2878 if (bond->params.arp_interval)
2879 queue_delayed_work(bond->wq, &bond->arp_work, delta_in_ticks);
2881 read_unlock(&bond->lock);
2885 * Called to inspect slaves for active-backup mode ARP monitor link state
2886 * changes. Sets new_link in slaves to specify what action should take
2887 * place for the slave. Returns 0 if no changes are found, >0 if changes
2888 * to link states must be committed.
2890 * Called with bond->lock held for read.
2892 static int bond_ab_arp_inspect(struct bonding *bond, int delta_in_ticks)
2894 struct slave *slave;
2897 bond_for_each_slave(bond, slave, i) {
2898 slave->new_link = BOND_LINK_NOCHANGE;
2900 if (slave->link != BOND_LINK_UP) {
2901 if (time_before_eq(jiffies, slave_last_rx(bond, slave) +
2903 slave->new_link = BOND_LINK_UP;
2911 * Give slaves 2*delta after being enslaved or made
2912 * active. This avoids bouncing, as the last receive
2913 * times need a full ARP monitor cycle to be updated.
2915 if (!time_after_eq(jiffies, slave->jiffies +
2916 2 * delta_in_ticks))
2920 * Backup slave is down if:
2921 * - No current_arp_slave AND
2922 * - more than 3*delta since last receive AND
2923 * - the bond has an IP address
2925 * Note: a non-null current_arp_slave indicates
2926 * the curr_active_slave went down and we are
2927 * searching for a new one; under this condition
2928 * we only take the curr_active_slave down - this
2929 * gives each slave a chance to tx/rx traffic
2930 * before being taken out
2932 if (slave->state == BOND_STATE_BACKUP &&
2933 !bond->current_arp_slave &&
2934 time_after(jiffies, slave_last_rx(bond, slave) +
2935 3 * delta_in_ticks)) {
2936 slave->new_link = BOND_LINK_DOWN;
2941 * Active slave is down if:
2942 * - more than 2*delta since transmitting OR
2943 * - (more than 2*delta since receive AND
2944 * the bond has an IP address)
2946 if ((slave->state == BOND_STATE_ACTIVE) &&
2947 (time_after_eq(jiffies, slave->dev->trans_start +
2948 2 * delta_in_ticks) ||
2949 (time_after_eq(jiffies, slave_last_rx(bond, slave)
2950 + 2 * delta_in_ticks)))) {
2951 slave->new_link = BOND_LINK_DOWN;
2956 read_lock(&bond->curr_slave_lock);
2959 * Trigger a commit if the primary option setting has changed.
2961 if (bond->primary_slave &&
2962 (bond->primary_slave != bond->curr_active_slave) &&
2963 (bond->primary_slave->link == BOND_LINK_UP))
2966 read_unlock(&bond->curr_slave_lock);
2972 * Called to commit link state changes noted by inspection step of
2973 * active-backup mode ARP monitor.
2975 * Called with RTNL and bond->lock for read.
2977 static void bond_ab_arp_commit(struct bonding *bond, int delta_in_ticks)
2979 struct slave *slave;
2982 bond_for_each_slave(bond, slave, i) {
2983 switch (slave->new_link) {
2984 case BOND_LINK_NOCHANGE:
2988 write_lock_bh(&bond->curr_slave_lock);
2990 if (!bond->curr_active_slave &&
2991 time_before_eq(jiffies, slave->dev->trans_start +
2993 slave->link = BOND_LINK_UP;
2994 bond_change_active_slave(bond, slave);
2995 bond->current_arp_slave = NULL;
2997 printk(KERN_INFO DRV_NAME
2998 ": %s: %s is up and now the "
2999 "active interface\n",
3000 bond->dev->name, slave->dev->name);
3002 } else if (bond->curr_active_slave != slave) {
3003 /* this slave has just come up but we
3004 * already have a current slave; this can
3005 * also happen if bond_enslave adds a new
3006 * slave that is up while we are searching
3009 slave->link = BOND_LINK_UP;
3010 bond_set_slave_inactive_flags(slave);
3011 bond->current_arp_slave = NULL;
3013 printk(KERN_INFO DRV_NAME
3014 ": %s: backup interface %s is now up\n",
3015 bond->dev->name, slave->dev->name);
3018 write_unlock_bh(&bond->curr_slave_lock);
3022 case BOND_LINK_DOWN:
3023 if (slave->link_failure_count < UINT_MAX)
3024 slave->link_failure_count++;
3026 slave->link = BOND_LINK_DOWN;
3028 if (slave == bond->curr_active_slave) {
3029 printk(KERN_INFO DRV_NAME
3030 ": %s: link status down for active "
3031 "interface %s, disabling it\n",
3032 bond->dev->name, slave->dev->name);
3034 bond_set_slave_inactive_flags(slave);
3036 write_lock_bh(&bond->curr_slave_lock);
3038 bond_select_active_slave(bond);
3039 if (bond->curr_active_slave)
3040 bond->curr_active_slave->jiffies =
3043 write_unlock_bh(&bond->curr_slave_lock);
3045 bond->current_arp_slave = NULL;
3047 } else if (slave->state == BOND_STATE_BACKUP) {
3048 printk(KERN_INFO DRV_NAME
3049 ": %s: backup interface %s is now down\n",
3050 bond->dev->name, slave->dev->name);
3052 bond_set_slave_inactive_flags(slave);
3057 printk(KERN_ERR DRV_NAME
3058 ": %s: impossible: new_link %d on slave %s\n",
3059 bond->dev->name, slave->new_link,
3065 * No race with changes to primary via sysfs, as we hold rtnl.
3067 if (bond->primary_slave &&
3068 (bond->primary_slave != bond->curr_active_slave) &&
3069 (bond->primary_slave->link == BOND_LINK_UP)) {
3070 write_lock_bh(&bond->curr_slave_lock);
3071 bond_change_active_slave(bond, bond->primary_slave);
3072 write_unlock_bh(&bond->curr_slave_lock);
3075 bond_set_carrier(bond);
3079 * Send ARP probes for active-backup mode ARP monitor.
3081 * Called with bond->lock held for read.
3083 static void bond_ab_arp_probe(struct bonding *bond)
3085 struct slave *slave;
3088 read_lock(&bond->curr_slave_lock);
3090 if (bond->current_arp_slave && bond->curr_active_slave)
3091 printk("PROBE: c_arp %s && cas %s BAD\n",
3092 bond->current_arp_slave->dev->name,
3093 bond->curr_active_slave->dev->name);
3095 if (bond->curr_active_slave) {
3096 bond_arp_send_all(bond, bond->curr_active_slave);
3097 read_unlock(&bond->curr_slave_lock);
3101 read_unlock(&bond->curr_slave_lock);
3103 /* if we don't have a curr_active_slave, search for the next available
3104 * backup slave from the current_arp_slave and make it the candidate
3105 * for becoming the curr_active_slave
3108 if (!bond->current_arp_slave) {
3109 bond->current_arp_slave = bond->first_slave;
3110 if (!bond->current_arp_slave)
3114 bond_set_slave_inactive_flags(bond->current_arp_slave);
3116 /* search for next candidate */
3117 bond_for_each_slave_from(bond, slave, i, bond->current_arp_slave->next) {
3118 if (IS_UP(slave->dev)) {
3119 slave->link = BOND_LINK_BACK;
3120 bond_set_slave_active_flags(slave);
3121 bond_arp_send_all(bond, slave);
3122 slave->jiffies = jiffies;
3123 bond->current_arp_slave = slave;
3127 /* if the link state is up at this point, we
3128 * mark it down - this can happen if we have
3129 * simultaneous link failures and
3130 * reselect_active_interface doesn't make this
3131 * one the current slave so it is still marked
3132 * up when it is actually down
3134 if (slave->link == BOND_LINK_UP) {
3135 slave->link = BOND_LINK_DOWN;
3136 if (slave->link_failure_count < UINT_MAX)
3137 slave->link_failure_count++;
3139 bond_set_slave_inactive_flags(slave);
3141 printk(KERN_INFO DRV_NAME
3142 ": %s: backup interface %s is now down.\n",
3143 bond->dev->name, slave->dev->name);
3148 void bond_activebackup_arp_mon(struct work_struct *work)
3150 struct bonding *bond = container_of(work, struct bonding,
3154 read_lock(&bond->lock);
3156 if (bond->kill_timers)
3159 delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
3161 if (bond->slave_cnt == 0)
3164 if (bond->send_grat_arp) {
3165 read_lock(&bond->curr_slave_lock);
3166 bond_send_gratuitous_arp(bond);
3167 read_unlock(&bond->curr_slave_lock);
3170 if (bond->send_unsol_na) {
3171 read_lock(&bond->curr_slave_lock);
3172 bond_send_unsolicited_na(bond);
3173 read_unlock(&bond->curr_slave_lock);
3176 if (bond_ab_arp_inspect(bond, delta_in_ticks)) {
3177 read_unlock(&bond->lock);
3179 read_lock(&bond->lock);
3181 bond_ab_arp_commit(bond, delta_in_ticks);
3183 read_unlock(&bond->lock);
3185 read_lock(&bond->lock);
3188 bond_ab_arp_probe(bond);
3191 if (bond->params.arp_interval) {
3192 queue_delayed_work(bond->wq, &bond->arp_work, delta_in_ticks);
3195 read_unlock(&bond->lock);
3198 /*------------------------------ proc/seq_file-------------------------------*/
3200 #ifdef CONFIG_PROC_FS
3202 static void *bond_info_seq_start(struct seq_file *seq, loff_t *pos)
3204 struct bonding *bond = seq->private;
3206 struct slave *slave;
3209 /* make sure the bond won't be taken away */
3210 read_lock(&dev_base_lock);
3211 read_lock(&bond->lock);
3214 return SEQ_START_TOKEN;
3217 bond_for_each_slave(bond, slave, i) {
3218 if (++off == *pos) {
3226 static void *bond_info_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3228 struct bonding *bond = seq->private;
3229 struct slave *slave = v;
3232 if (v == SEQ_START_TOKEN) {
3233 return bond->first_slave;
3236 slave = slave->next;
3238 return (slave == bond->first_slave) ? NULL : slave;
3241 static void bond_info_seq_stop(struct seq_file *seq, void *v)
3243 struct bonding *bond = seq->private;
3245 read_unlock(&bond->lock);
3246 read_unlock(&dev_base_lock);
3249 static void bond_info_show_master(struct seq_file *seq)
3251 struct bonding *bond = seq->private;
3255 read_lock(&bond->curr_slave_lock);
3256 curr = bond->curr_active_slave;
3257 read_unlock(&bond->curr_slave_lock);
3259 seq_printf(seq, "Bonding Mode: %s",
3260 bond_mode_name(bond->params.mode));
3262 if (bond->params.mode == BOND_MODE_ACTIVEBACKUP &&
3263 bond->params.fail_over_mac)
3264 seq_printf(seq, " (fail_over_mac %s)",
3265 fail_over_mac_tbl[bond->params.fail_over_mac].modename);
3267 seq_printf(seq, "\n");
3269 if (bond->params.mode == BOND_MODE_XOR ||
3270 bond->params.mode == BOND_MODE_8023AD) {
3271 seq_printf(seq, "Transmit Hash Policy: %s (%d)\n",
3272 xmit_hashtype_tbl[bond->params.xmit_policy].modename,
3273 bond->params.xmit_policy);
3276 if (USES_PRIMARY(bond->params.mode)) {
3277 seq_printf(seq, "Primary Slave: %s\n",
3278 (bond->primary_slave) ?
3279 bond->primary_slave->dev->name : "None");
3281 seq_printf(seq, "Currently Active Slave: %s\n",
3282 (curr) ? curr->dev->name : "None");
3285 seq_printf(seq, "MII Status: %s\n", netif_carrier_ok(bond->dev) ?
3287 seq_printf(seq, "MII Polling Interval (ms): %d\n", bond->params.miimon);
3288 seq_printf(seq, "Up Delay (ms): %d\n",
3289 bond->params.updelay * bond->params.miimon);
3290 seq_printf(seq, "Down Delay (ms): %d\n",
3291 bond->params.downdelay * bond->params.miimon);
3294 /* ARP information */
3295 if(bond->params.arp_interval > 0) {
3297 seq_printf(seq, "ARP Polling Interval (ms): %d\n",
3298 bond->params.arp_interval);
3300 seq_printf(seq, "ARP IP target/s (n.n.n.n form):");
3302 for(i = 0; (i < BOND_MAX_ARP_TARGETS) ;i++) {
3303 if (!bond->params.arp_targets[i])
3306 seq_printf(seq, ",");
3307 seq_printf(seq, " %pI4", &bond->params.arp_targets[i]);
3310 seq_printf(seq, "\n");
3313 if (bond->params.mode == BOND_MODE_8023AD) {
3314 struct ad_info ad_info;
3316 seq_puts(seq, "\n802.3ad info\n");
3317 seq_printf(seq, "LACP rate: %s\n",
3318 (bond->params.lacp_fast) ? "fast" : "slow");
3319 seq_printf(seq, "Aggregator selection policy (ad_select): %s\n",
3320 ad_select_tbl[bond->params.ad_select].modename);
3322 if (bond_3ad_get_active_agg_info(bond, &ad_info)) {
3323 seq_printf(seq, "bond %s has no active aggregator\n",
3326 seq_printf(seq, "Active Aggregator Info:\n");
3328 seq_printf(seq, "\tAggregator ID: %d\n",
3329 ad_info.aggregator_id);
3330 seq_printf(seq, "\tNumber of ports: %d\n",
3332 seq_printf(seq, "\tActor Key: %d\n",
3334 seq_printf(seq, "\tPartner Key: %d\n",
3335 ad_info.partner_key);
3336 seq_printf(seq, "\tPartner Mac Address: %pM\n",
3337 ad_info.partner_system);
3342 static void bond_info_show_slave(struct seq_file *seq, const struct slave *slave)
3344 struct bonding *bond = seq->private;
3346 seq_printf(seq, "\nSlave Interface: %s\n", slave->dev->name);
3347 seq_printf(seq, "MII Status: %s\n",
3348 (slave->link == BOND_LINK_UP) ? "up" : "down");
3349 seq_printf(seq, "Link Failure Count: %u\n",
3350 slave->link_failure_count);
3352 seq_printf(seq, "Permanent HW addr: %pM\n", slave->perm_hwaddr);
3354 if (bond->params.mode == BOND_MODE_8023AD) {
3355 const struct aggregator *agg
3356 = SLAVE_AD_INFO(slave).port.aggregator;
3359 seq_printf(seq, "Aggregator ID: %d\n",
3360 agg->aggregator_identifier);
3362 seq_puts(seq, "Aggregator ID: N/A\n");
3367 static int bond_info_seq_show(struct seq_file *seq, void *v)
3369 if (v == SEQ_START_TOKEN) {
3370 seq_printf(seq, "%s\n", version);
3371 bond_info_show_master(seq);
3373 bond_info_show_slave(seq, v);
3379 static struct seq_operations bond_info_seq_ops = {
3380 .start = bond_info_seq_start,
3381 .next = bond_info_seq_next,
3382 .stop = bond_info_seq_stop,
3383 .show = bond_info_seq_show,
3386 static int bond_info_open(struct inode *inode, struct file *file)
3388 struct seq_file *seq;
3389 struct proc_dir_entry *proc;
3392 res = seq_open(file, &bond_info_seq_ops);
3394 /* recover the pointer buried in proc_dir_entry data */
3395 seq = file->private_data;
3397 seq->private = proc->data;
3403 static const struct file_operations bond_info_fops = {
3404 .owner = THIS_MODULE,
3405 .open = bond_info_open,
3407 .llseek = seq_lseek,
3408 .release = seq_release,
3411 static int bond_create_proc_entry(struct bonding *bond)
3413 struct net_device *bond_dev = bond->dev;
3415 if (bond_proc_dir) {
3416 bond->proc_entry = proc_create_data(bond_dev->name,
3417 S_IRUGO, bond_proc_dir,
3418 &bond_info_fops, bond);
3419 if (bond->proc_entry == NULL) {
3420 printk(KERN_WARNING DRV_NAME
3421 ": Warning: Cannot create /proc/net/%s/%s\n",
3422 DRV_NAME, bond_dev->name);
3424 memcpy(bond->proc_file_name, bond_dev->name, IFNAMSIZ);
3431 static void bond_remove_proc_entry(struct bonding *bond)
3433 if (bond_proc_dir && bond->proc_entry) {
3434 remove_proc_entry(bond->proc_file_name, bond_proc_dir);
3435 memset(bond->proc_file_name, 0, IFNAMSIZ);
3436 bond->proc_entry = NULL;
3440 /* Create the bonding directory under /proc/net, if doesn't exist yet.
3441 * Caller must hold rtnl_lock.
3443 static void bond_create_proc_dir(void)
3445 int len = strlen(DRV_NAME);
3447 for (bond_proc_dir = init_net.proc_net->subdir; bond_proc_dir;
3448 bond_proc_dir = bond_proc_dir->next) {
3449 if ((bond_proc_dir->namelen == len) &&
3450 !memcmp(bond_proc_dir->name, DRV_NAME, len)) {
3455 if (!bond_proc_dir) {
3456 bond_proc_dir = proc_mkdir(DRV_NAME, init_net.proc_net);
3457 if (bond_proc_dir) {
3458 bond_proc_dir->owner = THIS_MODULE;
3460 printk(KERN_WARNING DRV_NAME
3461 ": Warning: cannot create /proc/net/%s\n",
3467 /* Destroy the bonding directory under /proc/net, if empty.
3468 * Caller must hold rtnl_lock.
3470 static void bond_destroy_proc_dir(void)
3472 struct proc_dir_entry *de;
3474 if (!bond_proc_dir) {
3478 /* verify that the /proc dir is empty */
3479 for (de = bond_proc_dir->subdir; de; de = de->next) {
3480 /* ignore . and .. */
3481 if (*(de->name) != '.') {
3487 if (bond_proc_dir->owner == THIS_MODULE) {
3488 bond_proc_dir->owner = NULL;
3491 remove_proc_entry(DRV_NAME, init_net.proc_net);
3492 bond_proc_dir = NULL;
3495 #endif /* CONFIG_PROC_FS */
3497 /*-------------------------- netdev event handling --------------------------*/
3500 * Change device name
3502 static int bond_event_changename(struct bonding *bond)
3504 #ifdef CONFIG_PROC_FS
3505 bond_remove_proc_entry(bond);
3506 bond_create_proc_entry(bond);
3508 down_write(&(bonding_rwsem));
3509 bond_destroy_sysfs_entry(bond);
3510 bond_create_sysfs_entry(bond);
3511 up_write(&(bonding_rwsem));
3515 static int bond_master_netdev_event(unsigned long event, struct net_device *bond_dev)
3517 struct bonding *event_bond = netdev_priv(bond_dev);
3520 case NETDEV_CHANGENAME:
3521 return bond_event_changename(event_bond);
3522 case NETDEV_UNREGISTER:
3523 bond_release_all(event_bond->dev);
3532 static int bond_slave_netdev_event(unsigned long event, struct net_device *slave_dev)
3534 struct net_device *bond_dev = slave_dev->master;
3535 struct bonding *bond = netdev_priv(bond_dev);
3538 case NETDEV_UNREGISTER:
3540 if (bond->setup_by_slave)
3541 bond_release_and_destroy(bond_dev, slave_dev);
3543 bond_release(bond_dev, slave_dev);
3548 * TODO: is this what we get if somebody
3549 * sets up a hierarchical bond, then rmmod's
3550 * one of the slave bonding devices?
3555 * ... Or is it this?
3558 case NETDEV_CHANGEMTU:
3560 * TODO: Should slaves be allowed to
3561 * independently alter their MTU? For
3562 * an active-backup bond, slaves need
3563 * not be the same type of device, so
3564 * MTUs may vary. For other modes,
3565 * slaves arguably should have the
3566 * same MTUs. To do this, we'd need to
3567 * take over the slave's change_mtu
3568 * function for the duration of their
3572 case NETDEV_CHANGENAME:
3574 * TODO: handle changing the primary's name
3577 case NETDEV_FEAT_CHANGE:
3578 bond_compute_features(bond);
3588 * bond_netdev_event: handle netdev notifier chain events.
3590 * This function receives events for the netdev chain. The caller (an
3591 * ioctl handler calling blocking_notifier_call_chain) holds the necessary
3592 * locks for us to safely manipulate the slave devices (RTNL lock,
3595 static int bond_netdev_event(struct notifier_block *this, unsigned long event, void *ptr)
3597 struct net_device *event_dev = (struct net_device *)ptr;
3599 if (dev_net(event_dev) != &init_net)
3602 dprintk("event_dev: %s, event: %lx\n",
3603 (event_dev ? event_dev->name : "None"),
3606 if (!(event_dev->priv_flags & IFF_BONDING))
3609 if (event_dev->flags & IFF_MASTER) {
3610 dprintk("IFF_MASTER\n");
3611 return bond_master_netdev_event(event, event_dev);
3614 if (event_dev->flags & IFF_SLAVE) {
3615 dprintk("IFF_SLAVE\n");
3616 return bond_slave_netdev_event(event, event_dev);
3623 * bond_inetaddr_event: handle inetaddr notifier chain events.
3625 * We keep track of device IPs primarily to use as source addresses in
3626 * ARP monitor probes (rather than spewing out broadcasts all the time).
3628 * We track one IP for the main device (if it has one), plus one per VLAN.
3630 static int bond_inetaddr_event(struct notifier_block *this, unsigned long event, void *ptr)
3632 struct in_ifaddr *ifa = ptr;
3633 struct net_device *vlan_dev, *event_dev = ifa->ifa_dev->dev;
3634 struct bonding *bond;
3635 struct vlan_entry *vlan;
3637 if (dev_net(ifa->ifa_dev->dev) != &init_net)
3640 list_for_each_entry(bond, &bond_dev_list, bond_list) {
3641 if (bond->dev == event_dev) {
3644 bond->master_ip = ifa->ifa_local;
3647 bond->master_ip = bond_glean_dev_ip(bond->dev);
3654 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
3655 vlan_dev = vlan_group_get_device(bond->vlgrp, vlan->vlan_id);
3656 if (vlan_dev == event_dev) {
3659 vlan->vlan_ip = ifa->ifa_local;
3663 bond_glean_dev_ip(vlan_dev);
3674 static struct notifier_block bond_netdev_notifier = {
3675 .notifier_call = bond_netdev_event,
3678 static struct notifier_block bond_inetaddr_notifier = {
3679 .notifier_call = bond_inetaddr_event,
3682 /*-------------------------- Packet type handling ---------------------------*/
3684 /* register to receive lacpdus on a bond */
3685 static void bond_register_lacpdu(struct bonding *bond)
3687 struct packet_type *pk_type = &(BOND_AD_INFO(bond).ad_pkt_type);
3689 /* initialize packet type */
3690 pk_type->type = PKT_TYPE_LACPDU;
3691 pk_type->dev = bond->dev;
3692 pk_type->func = bond_3ad_lacpdu_recv;
3694 dev_add_pack(pk_type);
3697 /* unregister to receive lacpdus on a bond */
3698 static void bond_unregister_lacpdu(struct bonding *bond)
3700 dev_remove_pack(&(BOND_AD_INFO(bond).ad_pkt_type));
3703 void bond_register_arp(struct bonding *bond)
3705 struct packet_type *pt = &bond->arp_mon_pt;
3710 pt->type = htons(ETH_P_ARP);
3711 pt->dev = bond->dev;
3712 pt->func = bond_arp_rcv;
3716 void bond_unregister_arp(struct bonding *bond)
3718 struct packet_type *pt = &bond->arp_mon_pt;
3720 dev_remove_pack(pt);
3724 /*---------------------------- Hashing Policies -----------------------------*/
3727 * Hash for the output device based upon layer 2 and layer 3 data. If
3728 * the packet is not IP mimic bond_xmit_hash_policy_l2()
3730 static int bond_xmit_hash_policy_l23(struct sk_buff *skb,
3731 struct net_device *bond_dev, int count)
3733 struct ethhdr *data = (struct ethhdr *)skb->data;
3734 struct iphdr *iph = ip_hdr(skb);
3736 if (skb->protocol == htons(ETH_P_IP)) {
3737 return ((ntohl(iph->saddr ^ iph->daddr) & 0xffff) ^
3738 (data->h_dest[5] ^ bond_dev->dev_addr[5])) % count;
3741 return (data->h_dest[5] ^ bond_dev->dev_addr[5]) % count;
3745 * Hash for the output device based upon layer 3 and layer 4 data. If
3746 * the packet is a frag or not TCP or UDP, just use layer 3 data. If it is
3747 * altogether not IP, mimic bond_xmit_hash_policy_l2()
3749 static int bond_xmit_hash_policy_l34(struct sk_buff *skb,
3750 struct net_device *bond_dev, int count)
3752 struct ethhdr *data = (struct ethhdr *)skb->data;
3753 struct iphdr *iph = ip_hdr(skb);
3754 __be16 *layer4hdr = (__be16 *)((u32 *)iph + iph->ihl);
3757 if (skb->protocol == htons(ETH_P_IP)) {
3758 if (!(iph->frag_off & htons(IP_MF|IP_OFFSET)) &&
3759 (iph->protocol == IPPROTO_TCP ||
3760 iph->protocol == IPPROTO_UDP)) {
3761 layer4_xor = ntohs((*layer4hdr ^ *(layer4hdr + 1)));
3763 return (layer4_xor ^
3764 ((ntohl(iph->saddr ^ iph->daddr)) & 0xffff)) % count;
3768 return (data->h_dest[5] ^ bond_dev->dev_addr[5]) % count;
3772 * Hash for the output device based upon layer 2 data
3774 static int bond_xmit_hash_policy_l2(struct sk_buff *skb,
3775 struct net_device *bond_dev, int count)
3777 struct ethhdr *data = (struct ethhdr *)skb->data;
3779 return (data->h_dest[5] ^ bond_dev->dev_addr[5]) % count;
3782 /*-------------------------- Device entry points ----------------------------*/
3784 static int bond_open(struct net_device *bond_dev)
3786 struct bonding *bond = netdev_priv(bond_dev);
3788 bond->kill_timers = 0;
3790 if (bond_is_lb(bond)) {
3791 /* bond_alb_initialize must be called before the timer
3794 if (bond_alb_initialize(bond, (bond->params.mode == BOND_MODE_ALB))) {
3795 /* something went wrong - fail the open operation */
3799 INIT_DELAYED_WORK(&bond->alb_work, bond_alb_monitor);
3800 queue_delayed_work(bond->wq, &bond->alb_work, 0);
3803 if (bond->params.miimon) { /* link check interval, in milliseconds. */
3804 INIT_DELAYED_WORK(&bond->mii_work, bond_mii_monitor);
3805 queue_delayed_work(bond->wq, &bond->mii_work, 0);
3808 if (bond->params.arp_interval) { /* arp interval, in milliseconds. */
3809 if (bond->params.mode == BOND_MODE_ACTIVEBACKUP)
3810 INIT_DELAYED_WORK(&bond->arp_work,
3811 bond_activebackup_arp_mon);
3813 INIT_DELAYED_WORK(&bond->arp_work,
3814 bond_loadbalance_arp_mon);
3816 queue_delayed_work(bond->wq, &bond->arp_work, 0);
3817 if (bond->params.arp_validate)
3818 bond_register_arp(bond);
3821 if (bond->params.mode == BOND_MODE_8023AD) {
3822 INIT_DELAYED_WORK(&bond->ad_work, bond_3ad_state_machine_handler);
3823 queue_delayed_work(bond->wq, &bond->ad_work, 0);
3824 /* register to receive LACPDUs */
3825 bond_register_lacpdu(bond);
3826 bond_3ad_initiate_agg_selection(bond, 1);
3832 static int bond_close(struct net_device *bond_dev)
3834 struct bonding *bond = netdev_priv(bond_dev);
3836 if (bond->params.mode == BOND_MODE_8023AD) {
3837 /* Unregister the receive of LACPDUs */
3838 bond_unregister_lacpdu(bond);
3841 if (bond->params.arp_validate)
3842 bond_unregister_arp(bond);
3844 write_lock_bh(&bond->lock);
3846 bond->send_grat_arp = 0;
3847 bond->send_unsol_na = 0;
3849 /* signal timers not to re-arm */
3850 bond->kill_timers = 1;
3852 write_unlock_bh(&bond->lock);
3854 if (bond->params.miimon) { /* link check interval, in milliseconds. */
3855 cancel_delayed_work(&bond->mii_work);
3858 if (bond->params.arp_interval) { /* arp interval, in milliseconds. */
3859 cancel_delayed_work(&bond->arp_work);
3862 switch (bond->params.mode) {
3863 case BOND_MODE_8023AD:
3864 cancel_delayed_work(&bond->ad_work);
3868 cancel_delayed_work(&bond->alb_work);
3875 if (bond_is_lb(bond)) {
3876 /* Must be called only after all
3877 * slaves have been released
3879 bond_alb_deinitialize(bond);
3885 static struct net_device_stats *bond_get_stats(struct net_device *bond_dev)
3887 struct bonding *bond = netdev_priv(bond_dev);
3888 struct net_device_stats *stats = &bond->stats;
3889 struct net_device_stats local_stats;
3890 struct slave *slave;
3893 memset(&local_stats, 0, sizeof(struct net_device_stats));
3895 read_lock_bh(&bond->lock);
3897 bond_for_each_slave(bond, slave, i) {
3898 const struct net_device_stats *sstats = dev_get_stats(slave->dev);
3900 local_stats.rx_packets += sstats->rx_packets;
3901 local_stats.rx_bytes += sstats->rx_bytes;
3902 local_stats.rx_errors += sstats->rx_errors;
3903 local_stats.rx_dropped += sstats->rx_dropped;
3905 local_stats.tx_packets += sstats->tx_packets;
3906 local_stats.tx_bytes += sstats->tx_bytes;
3907 local_stats.tx_errors += sstats->tx_errors;
3908 local_stats.tx_dropped += sstats->tx_dropped;
3910 local_stats.multicast += sstats->multicast;
3911 local_stats.collisions += sstats->collisions;
3913 local_stats.rx_length_errors += sstats->rx_length_errors;
3914 local_stats.rx_over_errors += sstats->rx_over_errors;
3915 local_stats.rx_crc_errors += sstats->rx_crc_errors;
3916 local_stats.rx_frame_errors += sstats->rx_frame_errors;
3917 local_stats.rx_fifo_errors += sstats->rx_fifo_errors;
3918 local_stats.rx_missed_errors += sstats->rx_missed_errors;
3920 local_stats.tx_aborted_errors += sstats->tx_aborted_errors;
3921 local_stats.tx_carrier_errors += sstats->tx_carrier_errors;
3922 local_stats.tx_fifo_errors += sstats->tx_fifo_errors;
3923 local_stats.tx_heartbeat_errors += sstats->tx_heartbeat_errors;
3924 local_stats.tx_window_errors += sstats->tx_window_errors;
3927 memcpy(stats, &local_stats, sizeof(struct net_device_stats));
3929 read_unlock_bh(&bond->lock);
3934 static int bond_do_ioctl(struct net_device *bond_dev, struct ifreq *ifr, int cmd)
3936 struct net_device *slave_dev = NULL;
3937 struct ifbond k_binfo;
3938 struct ifbond __user *u_binfo = NULL;
3939 struct ifslave k_sinfo;
3940 struct ifslave __user *u_sinfo = NULL;
3941 struct mii_ioctl_data *mii = NULL;
3944 dprintk("bond_ioctl: master=%s, cmd=%d\n",
3945 bond_dev->name, cmd);
3957 * We do this again just in case we were called by SIOCGMIIREG
3958 * instead of SIOCGMIIPHY.
3965 if (mii->reg_num == 1) {
3966 struct bonding *bond = netdev_priv(bond_dev);
3968 read_lock(&bond->lock);
3969 read_lock(&bond->curr_slave_lock);
3970 if (netif_carrier_ok(bond->dev)) {
3971 mii->val_out = BMSR_LSTATUS;
3973 read_unlock(&bond->curr_slave_lock);
3974 read_unlock(&bond->lock);
3978 case BOND_INFO_QUERY_OLD:
3979 case SIOCBONDINFOQUERY:
3980 u_binfo = (struct ifbond __user *)ifr->ifr_data;
3982 if (copy_from_user(&k_binfo, u_binfo, sizeof(ifbond))) {
3986 res = bond_info_query(bond_dev, &k_binfo);
3988 if (copy_to_user(u_binfo, &k_binfo, sizeof(ifbond))) {
3994 case BOND_SLAVE_INFO_QUERY_OLD:
3995 case SIOCBONDSLAVEINFOQUERY:
3996 u_sinfo = (struct ifslave __user *)ifr->ifr_data;
3998 if (copy_from_user(&k_sinfo, u_sinfo, sizeof(ifslave))) {
4002 res = bond_slave_info_query(bond_dev, &k_sinfo);
4004 if (copy_to_user(u_sinfo, &k_sinfo, sizeof(ifslave))) {
4015 if (!capable(CAP_NET_ADMIN)) {
4019 down_write(&(bonding_rwsem));
4020 slave_dev = dev_get_by_name(&init_net, ifr->ifr_slave);
4022 dprintk("slave_dev=%p: \n", slave_dev);
4027 dprintk("slave_dev->name=%s: \n", slave_dev->name);
4029 case BOND_ENSLAVE_OLD:
4030 case SIOCBONDENSLAVE:
4031 res = bond_enslave(bond_dev, slave_dev);
4033 case BOND_RELEASE_OLD:
4034 case SIOCBONDRELEASE:
4035 res = bond_release(bond_dev, slave_dev);
4037 case BOND_SETHWADDR_OLD:
4038 case SIOCBONDSETHWADDR:
4039 res = bond_sethwaddr(bond_dev, slave_dev);
4041 case BOND_CHANGE_ACTIVE_OLD:
4042 case SIOCBONDCHANGEACTIVE:
4043 res = bond_ioctl_change_active(bond_dev, slave_dev);
4052 up_write(&(bonding_rwsem));
4056 static void bond_set_multicast_list(struct net_device *bond_dev)
4058 struct bonding *bond = netdev_priv(bond_dev);
4059 struct dev_mc_list *dmi;
4062 * Do promisc before checking multicast_mode
4064 if ((bond_dev->flags & IFF_PROMISC) && !(bond->flags & IFF_PROMISC)) {
4066 * FIXME: Need to handle the error when one of the multi-slaves
4069 bond_set_promiscuity(bond, 1);
4072 if (!(bond_dev->flags & IFF_PROMISC) && (bond->flags & IFF_PROMISC)) {
4073 bond_set_promiscuity(bond, -1);
4076 /* set allmulti flag to slaves */
4077 if ((bond_dev->flags & IFF_ALLMULTI) && !(bond->flags & IFF_ALLMULTI)) {
4079 * FIXME: Need to handle the error when one of the multi-slaves
4082 bond_set_allmulti(bond, 1);
4085 if (!(bond_dev->flags & IFF_ALLMULTI) && (bond->flags & IFF_ALLMULTI)) {
4086 bond_set_allmulti(bond, -1);
4089 read_lock(&bond->lock);
4091 bond->flags = bond_dev->flags;
4093 /* looking for addresses to add to slaves' mc list */
4094 for (dmi = bond_dev->mc_list; dmi; dmi = dmi->next) {
4095 if (!bond_mc_list_find_dmi(dmi, bond->mc_list)) {
4096 bond_mc_add(bond, dmi->dmi_addr, dmi->dmi_addrlen);
4100 /* looking for addresses to delete from slaves' list */
4101 for (dmi = bond->mc_list; dmi; dmi = dmi->next) {
4102 if (!bond_mc_list_find_dmi(dmi, bond_dev->mc_list)) {
4103 bond_mc_delete(bond, dmi->dmi_addr, dmi->dmi_addrlen);
4107 /* save master's multicast list */
4108 bond_mc_list_destroy(bond);
4109 bond_mc_list_copy(bond_dev->mc_list, bond, GFP_ATOMIC);
4111 read_unlock(&bond->lock);
4114 static int bond_neigh_setup(struct net_device *dev, struct neigh_parms *parms)
4116 struct bonding *bond = netdev_priv(dev);
4117 struct slave *slave = bond->first_slave;
4120 const struct net_device_ops *slave_ops
4121 = slave->dev->netdev_ops;
4122 if (slave_ops->ndo_neigh_setup)
4123 return slave_ops->ndo_neigh_setup(dev, parms);
4129 * Change the MTU of all of a master's slaves to match the master
4131 static int bond_change_mtu(struct net_device *bond_dev, int new_mtu)
4133 struct bonding *bond = netdev_priv(bond_dev);
4134 struct slave *slave, *stop_at;
4138 dprintk("bond=%p, name=%s, new_mtu=%d\n", bond,
4139 (bond_dev ? bond_dev->name : "None"), new_mtu);
4141 /* Can't hold bond->lock with bh disabled here since
4142 * some base drivers panic. On the other hand we can't
4143 * hold bond->lock without bh disabled because we'll
4144 * deadlock. The only solution is to rely on the fact
4145 * that we're under rtnl_lock here, and the slaves
4146 * list won't change. This doesn't solve the problem
4147 * of setting the slave's MTU while it is
4148 * transmitting, but the assumption is that the base
4149 * driver can handle that.
4151 * TODO: figure out a way to safely iterate the slaves
4152 * list, but without holding a lock around the actual
4153 * call to the base driver.
4156 bond_for_each_slave(bond, slave, i) {
4157 dprintk("s %p s->p %p c_m %p\n", slave,
4158 slave->prev, slave->dev->change_mtu);
4160 res = dev_set_mtu(slave->dev, new_mtu);
4163 /* If we failed to set the slave's mtu to the new value
4164 * we must abort the operation even in ACTIVE_BACKUP
4165 * mode, because if we allow the backup slaves to have
4166 * different mtu values than the active slave we'll
4167 * need to change their mtu when doing a failover. That
4168 * means changing their mtu from timer context, which
4169 * is probably not a good idea.
4171 dprintk("err %d %s\n", res, slave->dev->name);
4176 bond_dev->mtu = new_mtu;
4181 /* unwind from head to the slave that failed */
4183 bond_for_each_slave_from_to(bond, slave, i, bond->first_slave, stop_at) {
4186 tmp_res = dev_set_mtu(slave->dev, bond_dev->mtu);
4188 dprintk("unwind err %d dev %s\n", tmp_res,
4199 * Note that many devices must be down to change the HW address, and
4200 * downing the master releases all slaves. We can make bonds full of
4201 * bonding devices to test this, however.
4203 static int bond_set_mac_address(struct net_device *bond_dev, void *addr)
4205 struct bonding *bond = netdev_priv(bond_dev);
4206 struct sockaddr *sa = addr, tmp_sa;
4207 struct slave *slave, *stop_at;
4211 if (bond->params.mode == BOND_MODE_ALB)
4212 return bond_alb_set_mac_address(bond_dev, addr);
4215 dprintk("bond=%p, name=%s\n", bond, (bond_dev ? bond_dev->name : "None"));
4218 * If fail_over_mac is set to active, do nothing and return
4219 * success. Returning an error causes ifenslave to fail.
4221 if (bond->params.fail_over_mac == BOND_FOM_ACTIVE)
4224 if (!is_valid_ether_addr(sa->sa_data)) {
4225 return -EADDRNOTAVAIL;
4228 /* Can't hold bond->lock with bh disabled here since
4229 * some base drivers panic. On the other hand we can't
4230 * hold bond->lock without bh disabled because we'll
4231 * deadlock. The only solution is to rely on the fact
4232 * that we're under rtnl_lock here, and the slaves
4233 * list won't change. This doesn't solve the problem
4234 * of setting the slave's hw address while it is
4235 * transmitting, but the assumption is that the base
4236 * driver can handle that.
4238 * TODO: figure out a way to safely iterate the slaves
4239 * list, but without holding a lock around the actual
4240 * call to the base driver.
4243 bond_for_each_slave(bond, slave, i) {
4244 const struct net_device_ops *slave_ops = slave->dev->netdev_ops;
4245 dprintk("slave %p %s\n", slave, slave->dev->name);
4247 if (slave_ops->ndo_set_mac_address == NULL) {
4249 dprintk("EOPNOTSUPP %s\n", slave->dev->name);
4253 res = dev_set_mac_address(slave->dev, addr);
4255 /* TODO: consider downing the slave
4257 * User should expect communications
4258 * breakage anyway until ARP finish
4261 dprintk("err %d %s\n", res, slave->dev->name);
4267 memcpy(bond_dev->dev_addr, sa->sa_data, bond_dev->addr_len);
4271 memcpy(tmp_sa.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
4272 tmp_sa.sa_family = bond_dev->type;
4274 /* unwind from head to the slave that failed */
4276 bond_for_each_slave_from_to(bond, slave, i, bond->first_slave, stop_at) {
4279 tmp_res = dev_set_mac_address(slave->dev, &tmp_sa);
4281 dprintk("unwind err %d dev %s\n", tmp_res,
4289 static int bond_xmit_roundrobin(struct sk_buff *skb, struct net_device *bond_dev)
4291 struct bonding *bond = netdev_priv(bond_dev);
4292 struct slave *slave, *start_at;
4293 int i, slave_no, res = 1;
4295 read_lock(&bond->lock);
4297 if (!BOND_IS_OK(bond)) {
4302 * Concurrent TX may collide on rr_tx_counter; we accept that
4303 * as being rare enough not to justify using an atomic op here
4305 slave_no = bond->rr_tx_counter++ % bond->slave_cnt;
4307 bond_for_each_slave(bond, slave, i) {
4315 bond_for_each_slave_from(bond, slave, i, start_at) {
4316 if (IS_UP(slave->dev) &&
4317 (slave->link == BOND_LINK_UP) &&
4318 (slave->state == BOND_STATE_ACTIVE)) {
4319 res = bond_dev_queue_xmit(bond, skb, slave->dev);
4326 /* no suitable interface, frame not sent */
4329 read_unlock(&bond->lock);
4335 * in active-backup mode, we know that bond->curr_active_slave is always valid if
4336 * the bond has a usable interface.
4338 static int bond_xmit_activebackup(struct sk_buff *skb, struct net_device *bond_dev)
4340 struct bonding *bond = netdev_priv(bond_dev);
4343 read_lock(&bond->lock);
4344 read_lock(&bond->curr_slave_lock);
4346 if (!BOND_IS_OK(bond)) {
4350 if (!bond->curr_active_slave)
4353 res = bond_dev_queue_xmit(bond, skb, bond->curr_active_slave->dev);
4357 /* no suitable interface, frame not sent */
4360 read_unlock(&bond->curr_slave_lock);
4361 read_unlock(&bond->lock);
4366 * In bond_xmit_xor() , we determine the output device by using a pre-
4367 * determined xmit_hash_policy(), If the selected device is not enabled,
4368 * find the next active slave.
4370 static int bond_xmit_xor(struct sk_buff *skb, struct net_device *bond_dev)
4372 struct bonding *bond = netdev_priv(bond_dev);
4373 struct slave *slave, *start_at;
4378 read_lock(&bond->lock);
4380 if (!BOND_IS_OK(bond)) {
4384 slave_no = bond->xmit_hash_policy(skb, bond_dev, bond->slave_cnt);
4386 bond_for_each_slave(bond, slave, i) {
4395 bond_for_each_slave_from(bond, slave, i, start_at) {
4396 if (IS_UP(slave->dev) &&
4397 (slave->link == BOND_LINK_UP) &&
4398 (slave->state == BOND_STATE_ACTIVE)) {
4399 res = bond_dev_queue_xmit(bond, skb, slave->dev);
4406 /* no suitable interface, frame not sent */
4409 read_unlock(&bond->lock);
4414 * in broadcast mode, we send everything to all usable interfaces.
4416 static int bond_xmit_broadcast(struct sk_buff *skb, struct net_device *bond_dev)
4418 struct bonding *bond = netdev_priv(bond_dev);
4419 struct slave *slave, *start_at;
4420 struct net_device *tx_dev = NULL;
4424 read_lock(&bond->lock);
4426 if (!BOND_IS_OK(bond)) {
4430 read_lock(&bond->curr_slave_lock);
4431 start_at = bond->curr_active_slave;
4432 read_unlock(&bond->curr_slave_lock);
4438 bond_for_each_slave_from(bond, slave, i, start_at) {
4439 if (IS_UP(slave->dev) &&
4440 (slave->link == BOND_LINK_UP) &&
4441 (slave->state == BOND_STATE_ACTIVE)) {
4443 struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
4445 printk(KERN_ERR DRV_NAME
4446 ": %s: Error: bond_xmit_broadcast(): "
4447 "skb_clone() failed\n",
4452 res = bond_dev_queue_xmit(bond, skb2, tx_dev);
4454 dev_kfree_skb(skb2);
4458 tx_dev = slave->dev;
4463 res = bond_dev_queue_xmit(bond, skb, tx_dev);
4468 /* no suitable interface, frame not sent */
4471 /* frame sent to all suitable interfaces */
4472 read_unlock(&bond->lock);
4476 /*------------------------- Device initialization ---------------------------*/
4478 static void bond_set_xmit_hash_policy(struct bonding *bond)
4480 switch (bond->params.xmit_policy) {
4481 case BOND_XMIT_POLICY_LAYER23:
4482 bond->xmit_hash_policy = bond_xmit_hash_policy_l23;
4484 case BOND_XMIT_POLICY_LAYER34:
4485 bond->xmit_hash_policy = bond_xmit_hash_policy_l34;
4487 case BOND_XMIT_POLICY_LAYER2:
4489 bond->xmit_hash_policy = bond_xmit_hash_policy_l2;
4494 static int bond_start_xmit(struct sk_buff *skb, struct net_device *dev)
4496 const struct bonding *bond = netdev_priv(dev);
4498 switch (bond->params.mode) {
4499 case BOND_MODE_ROUNDROBIN:
4500 return bond_xmit_roundrobin(skb, dev);
4501 case BOND_MODE_ACTIVEBACKUP:
4502 return bond_xmit_activebackup(skb, dev);
4504 return bond_xmit_xor(skb, dev);
4505 case BOND_MODE_BROADCAST:
4506 return bond_xmit_broadcast(skb, dev);
4507 case BOND_MODE_8023AD:
4508 return bond_3ad_xmit_xor(skb, dev);
4511 return bond_alb_xmit(skb, dev);
4513 /* Should never happen, mode already checked */
4514 printk(KERN_ERR DRV_NAME ": %s: Error: Unknown bonding mode %d\n",
4515 dev->name, bond->params.mode);
4518 return NETDEV_TX_OK;
4524 * set bond mode specific net device operations
4526 void bond_set_mode_ops(struct bonding *bond, int mode)
4528 struct net_device *bond_dev = bond->dev;
4531 case BOND_MODE_ROUNDROBIN:
4533 case BOND_MODE_ACTIVEBACKUP:
4536 bond_set_xmit_hash_policy(bond);
4538 case BOND_MODE_BROADCAST:
4540 case BOND_MODE_8023AD:
4541 bond_set_master_3ad_flags(bond);
4542 bond_set_xmit_hash_policy(bond);
4545 bond_set_master_alb_flags(bond);
4550 /* Should never happen, mode already checked */
4551 printk(KERN_ERR DRV_NAME
4552 ": %s: Error: Unknown bonding mode %d\n",
4559 static void bond_ethtool_get_drvinfo(struct net_device *bond_dev,
4560 struct ethtool_drvinfo *drvinfo)
4562 strncpy(drvinfo->driver, DRV_NAME, 32);
4563 strncpy(drvinfo->version, DRV_VERSION, 32);
4564 snprintf(drvinfo->fw_version, 32, "%d", BOND_ABI_VERSION);
4567 static const struct ethtool_ops bond_ethtool_ops = {
4568 .get_drvinfo = bond_ethtool_get_drvinfo,
4569 .get_link = ethtool_op_get_link,
4570 .get_tx_csum = ethtool_op_get_tx_csum,
4571 .get_sg = ethtool_op_get_sg,
4572 .get_tso = ethtool_op_get_tso,
4573 .get_ufo = ethtool_op_get_ufo,
4574 .get_flags = ethtool_op_get_flags,
4577 static const struct net_device_ops bond_netdev_ops = {
4578 .ndo_open = bond_open,
4579 .ndo_stop = bond_close,
4580 .ndo_start_xmit = bond_start_xmit,
4581 .ndo_get_stats = bond_get_stats,
4582 .ndo_do_ioctl = bond_do_ioctl,
4583 .ndo_set_multicast_list = bond_set_multicast_list,
4584 .ndo_change_mtu = bond_change_mtu,
4585 .ndo_set_mac_address = bond_set_mac_address,
4586 .ndo_neigh_setup = bond_neigh_setup,
4587 .ndo_vlan_rx_register = bond_vlan_rx_register,
4588 .ndo_vlan_rx_add_vid = bond_vlan_rx_add_vid,
4589 .ndo_vlan_rx_kill_vid = bond_vlan_rx_kill_vid,
4593 * Does not allocate but creates a /proc entry.
4596 static int bond_init(struct net_device *bond_dev, struct bond_params *params)
4598 struct bonding *bond = netdev_priv(bond_dev);
4600 dprintk("Begin bond_init for %s\n", bond_dev->name);
4602 /* initialize rwlocks */
4603 rwlock_init(&bond->lock);
4604 rwlock_init(&bond->curr_slave_lock);
4606 bond->params = *params; /* copy params struct */
4608 bond->wq = create_singlethread_workqueue(bond_dev->name);
4612 /* Initialize pointers */
4613 bond->first_slave = NULL;
4614 bond->curr_active_slave = NULL;
4615 bond->current_arp_slave = NULL;
4616 bond->primary_slave = NULL;
4617 bond->dev = bond_dev;
4618 bond->send_grat_arp = 0;
4619 bond->send_unsol_na = 0;
4620 bond->setup_by_slave = 0;
4621 INIT_LIST_HEAD(&bond->vlan_list);
4623 /* Initialize the device entry points */
4624 bond_dev->netdev_ops = &bond_netdev_ops;
4625 bond_dev->ethtool_ops = &bond_ethtool_ops;
4626 bond_set_mode_ops(bond, bond->params.mode);
4628 bond_dev->destructor = bond_destructor;
4630 /* Initialize the device options */
4631 bond_dev->tx_queue_len = 0;
4632 bond_dev->flags |= IFF_MASTER|IFF_MULTICAST;
4633 bond_dev->priv_flags |= IFF_BONDING;
4634 if (bond->params.arp_interval)
4635 bond_dev->priv_flags |= IFF_MASTER_ARPMON;
4637 /* At first, we block adding VLANs. That's the only way to
4638 * prevent problems that occur when adding VLANs over an
4639 * empty bond. The block will be removed once non-challenged
4640 * slaves are enslaved.
4642 bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
4644 /* don't acquire bond device's netif_tx_lock when
4646 bond_dev->features |= NETIF_F_LLTX;
4648 /* By default, we declare the bond to be fully
4649 * VLAN hardware accelerated capable. Special
4650 * care is taken in the various xmit functions
4651 * when there are slaves that are not hw accel
4654 bond_dev->features |= (NETIF_F_HW_VLAN_TX |
4655 NETIF_F_HW_VLAN_RX |
4656 NETIF_F_HW_VLAN_FILTER);
4658 #ifdef CONFIG_PROC_FS
4659 bond_create_proc_entry(bond);
4661 list_add_tail(&bond->bond_list, &bond_dev_list);
4666 static void bond_work_cancel_all(struct bonding *bond)
4668 write_lock_bh(&bond->lock);
4669 bond->kill_timers = 1;
4670 write_unlock_bh(&bond->lock);
4672 if (bond->params.miimon && delayed_work_pending(&bond->mii_work))
4673 cancel_delayed_work(&bond->mii_work);
4675 if (bond->params.arp_interval && delayed_work_pending(&bond->arp_work))
4676 cancel_delayed_work(&bond->arp_work);
4678 if (bond->params.mode == BOND_MODE_ALB &&
4679 delayed_work_pending(&bond->alb_work))
4680 cancel_delayed_work(&bond->alb_work);
4682 if (bond->params.mode == BOND_MODE_8023AD &&
4683 delayed_work_pending(&bond->ad_work))
4684 cancel_delayed_work(&bond->ad_work);
4687 /* De-initialize device specific data.
4688 * Caller must hold rtnl_lock.
4690 static void bond_deinit(struct net_device *bond_dev)
4692 struct bonding *bond = netdev_priv(bond_dev);
4694 list_del(&bond->bond_list);
4696 bond_work_cancel_all(bond);
4698 #ifdef CONFIG_PROC_FS
4699 bond_remove_proc_entry(bond);
4703 /* Unregister and free all bond devices.
4704 * Caller must hold rtnl_lock.
4706 static void bond_free_all(void)
4708 struct bonding *bond, *nxt;
4710 list_for_each_entry_safe(bond, nxt, &bond_dev_list, bond_list) {
4711 struct net_device *bond_dev = bond->dev;
4713 bond_work_cancel_all(bond);
4714 /* Release the bonded slaves */
4715 bond_release_all(bond_dev);
4719 #ifdef CONFIG_PROC_FS
4720 bond_destroy_proc_dir();
4724 /*------------------------- Module initialization ---------------------------*/
4727 * Convert string input module parms. Accept either the
4728 * number of the mode or its string name. A bit complicated because
4729 * some mode names are substrings of other names, and calls from sysfs
4730 * may have whitespace in the name (trailing newlines, for example).
4732 int bond_parse_parm(const char *buf, struct bond_parm_tbl *tbl)
4734 int mode = -1, i, rv;
4735 char *p, modestr[BOND_MAX_MODENAME_LEN + 1] = { 0, };
4737 for (p = (char *)buf; *p; p++)
4738 if (!(isdigit(*p) || isspace(*p)))
4742 rv = sscanf(buf, "%20s", modestr);
4744 rv = sscanf(buf, "%d", &mode);
4749 for (i = 0; tbl[i].modename; i++) {
4750 if (mode == tbl[i].mode)
4752 if (strcmp(modestr, tbl[i].modename) == 0)
4759 static int bond_check_params(struct bond_params *params)
4761 int arp_validate_value, fail_over_mac_value;
4764 * Convert string parameters.
4767 bond_mode = bond_parse_parm(mode, bond_mode_tbl);
4768 if (bond_mode == -1) {
4769 printk(KERN_ERR DRV_NAME
4770 ": Error: Invalid bonding mode \"%s\"\n",
4771 mode == NULL ? "NULL" : mode);
4776 if (xmit_hash_policy) {
4777 if ((bond_mode != BOND_MODE_XOR) &&
4778 (bond_mode != BOND_MODE_8023AD)) {
4779 printk(KERN_INFO DRV_NAME
4780 ": xor_mode param is irrelevant in mode %s\n",
4781 bond_mode_name(bond_mode));
4783 xmit_hashtype = bond_parse_parm(xmit_hash_policy,
4785 if (xmit_hashtype == -1) {
4786 printk(KERN_ERR DRV_NAME
4787 ": Error: Invalid xmit_hash_policy \"%s\"\n",
4788 xmit_hash_policy == NULL ? "NULL" :
4796 if (bond_mode != BOND_MODE_8023AD) {
4797 printk(KERN_INFO DRV_NAME
4798 ": lacp_rate param is irrelevant in mode %s\n",
4799 bond_mode_name(bond_mode));
4801 lacp_fast = bond_parse_parm(lacp_rate, bond_lacp_tbl);
4802 if (lacp_fast == -1) {
4803 printk(KERN_ERR DRV_NAME
4804 ": Error: Invalid lacp rate \"%s\"\n",
4805 lacp_rate == NULL ? "NULL" : lacp_rate);
4812 params->ad_select = bond_parse_parm(ad_select, ad_select_tbl);
4813 if (params->ad_select == -1) {
4814 printk(KERN_ERR DRV_NAME
4815 ": Error: Invalid ad_select \"%s\"\n",
4816 ad_select == NULL ? "NULL" : ad_select);
4820 if (bond_mode != BOND_MODE_8023AD) {
4821 printk(KERN_WARNING DRV_NAME
4822 ": ad_select param only affects 802.3ad mode\n");
4825 params->ad_select = BOND_AD_STABLE;
4828 if (max_bonds < 0 || max_bonds > INT_MAX) {
4829 printk(KERN_WARNING DRV_NAME
4830 ": Warning: max_bonds (%d) not in range %d-%d, so it "
4831 "was reset to BOND_DEFAULT_MAX_BONDS (%d)\n",
4832 max_bonds, 0, INT_MAX, BOND_DEFAULT_MAX_BONDS);
4833 max_bonds = BOND_DEFAULT_MAX_BONDS;
4837 printk(KERN_WARNING DRV_NAME
4838 ": Warning: miimon module parameter (%d), "
4839 "not in range 0-%d, so it was reset to %d\n",
4840 miimon, INT_MAX, BOND_LINK_MON_INTERV);
4841 miimon = BOND_LINK_MON_INTERV;
4845 printk(KERN_WARNING DRV_NAME
4846 ": Warning: updelay module parameter (%d), "
4847 "not in range 0-%d, so it was reset to 0\n",
4852 if (downdelay < 0) {
4853 printk(KERN_WARNING DRV_NAME
4854 ": Warning: downdelay module parameter (%d), "
4855 "not in range 0-%d, so it was reset to 0\n",
4856 downdelay, INT_MAX);
4860 if ((use_carrier != 0) && (use_carrier != 1)) {
4861 printk(KERN_WARNING DRV_NAME
4862 ": Warning: use_carrier module parameter (%d), "
4863 "not of valid value (0/1), so it was set to 1\n",
4868 if (num_grat_arp < 0 || num_grat_arp > 255) {
4869 printk(KERN_WARNING DRV_NAME
4870 ": Warning: num_grat_arp (%d) not in range 0-255 so it "
4871 "was reset to 1 \n", num_grat_arp);
4875 if (num_unsol_na < 0 || num_unsol_na > 255) {
4876 printk(KERN_WARNING DRV_NAME
4877 ": Warning: num_unsol_na (%d) not in range 0-255 so it "
4878 "was reset to 1 \n", num_unsol_na);
4882 /* reset values for 802.3ad */
4883 if (bond_mode == BOND_MODE_8023AD) {
4885 printk(KERN_WARNING DRV_NAME
4886 ": Warning: miimon must be specified, "
4887 "otherwise bonding will not detect link "
4888 "failure, speed and duplex which are "
4889 "essential for 802.3ad operation\n");
4890 printk(KERN_WARNING "Forcing miimon to 100msec\n");
4895 /* reset values for TLB/ALB */
4896 if ((bond_mode == BOND_MODE_TLB) ||
4897 (bond_mode == BOND_MODE_ALB)) {
4899 printk(KERN_WARNING DRV_NAME
4900 ": Warning: miimon must be specified, "
4901 "otherwise bonding will not detect link "
4902 "failure and link speed which are essential "
4903 "for TLB/ALB load balancing\n");
4904 printk(KERN_WARNING "Forcing miimon to 100msec\n");
4909 if (bond_mode == BOND_MODE_ALB) {
4910 printk(KERN_NOTICE DRV_NAME
4911 ": In ALB mode you might experience client "
4912 "disconnections upon reconnection of a link if the "
4913 "bonding module updelay parameter (%d msec) is "
4914 "incompatible with the forwarding delay time of the "
4920 if (updelay || downdelay) {
4921 /* just warn the user the up/down delay will have
4922 * no effect since miimon is zero...
4924 printk(KERN_WARNING DRV_NAME
4925 ": Warning: miimon module parameter not set "
4926 "and updelay (%d) or downdelay (%d) module "
4927 "parameter is set; updelay and downdelay have "
4928 "no effect unless miimon is set\n",
4929 updelay, downdelay);
4932 /* don't allow arp monitoring */
4934 printk(KERN_WARNING DRV_NAME
4935 ": Warning: miimon (%d) and arp_interval (%d) "
4936 "can't be used simultaneously, disabling ARP "
4938 miimon, arp_interval);
4942 if ((updelay % miimon) != 0) {
4943 printk(KERN_WARNING DRV_NAME
4944 ": Warning: updelay (%d) is not a multiple "
4945 "of miimon (%d), updelay rounded to %d ms\n",
4946 updelay, miimon, (updelay / miimon) * miimon);
4951 if ((downdelay % miimon) != 0) {
4952 printk(KERN_WARNING DRV_NAME
4953 ": Warning: downdelay (%d) is not a multiple "
4954 "of miimon (%d), downdelay rounded to %d ms\n",
4956 (downdelay / miimon) * miimon);
4959 downdelay /= miimon;
4962 if (arp_interval < 0) {
4963 printk(KERN_WARNING DRV_NAME
4964 ": Warning: arp_interval module parameter (%d) "
4965 ", not in range 0-%d, so it was reset to %d\n",
4966 arp_interval, INT_MAX, BOND_LINK_ARP_INTERV);
4967 arp_interval = BOND_LINK_ARP_INTERV;
4970 for (arp_ip_count = 0;
4971 (arp_ip_count < BOND_MAX_ARP_TARGETS) && arp_ip_target[arp_ip_count];
4973 /* not complete check, but should be good enough to
4975 if (!isdigit(arp_ip_target[arp_ip_count][0])) {
4976 printk(KERN_WARNING DRV_NAME
4977 ": Warning: bad arp_ip_target module parameter "
4978 "(%s), ARP monitoring will not be performed\n",
4979 arp_ip_target[arp_ip_count]);
4982 __be32 ip = in_aton(arp_ip_target[arp_ip_count]);
4983 arp_target[arp_ip_count] = ip;
4987 if (arp_interval && !arp_ip_count) {
4988 /* don't allow arping if no arp_ip_target given... */
4989 printk(KERN_WARNING DRV_NAME
4990 ": Warning: arp_interval module parameter (%d) "
4991 "specified without providing an arp_ip_target "
4992 "parameter, arp_interval was reset to 0\n",
4998 if (bond_mode != BOND_MODE_ACTIVEBACKUP) {
4999 printk(KERN_ERR DRV_NAME
5000 ": arp_validate only supported in active-backup mode\n");
5003 if (!arp_interval) {
5004 printk(KERN_ERR DRV_NAME
5005 ": arp_validate requires arp_interval\n");
5009 arp_validate_value = bond_parse_parm(arp_validate,
5011 if (arp_validate_value == -1) {
5012 printk(KERN_ERR DRV_NAME
5013 ": Error: invalid arp_validate \"%s\"\n",
5014 arp_validate == NULL ? "NULL" : arp_validate);
5018 arp_validate_value = 0;
5021 printk(KERN_INFO DRV_NAME
5022 ": MII link monitoring set to %d ms\n",
5024 } else if (arp_interval) {
5027 printk(KERN_INFO DRV_NAME
5028 ": ARP monitoring set to %d ms, validate %s, with %d target(s):",
5030 arp_validate_tbl[arp_validate_value].modename,
5033 for (i = 0; i < arp_ip_count; i++)
5034 printk (" %s", arp_ip_target[i]);
5038 } else if (max_bonds) {
5039 /* miimon and arp_interval not set, we need one so things
5040 * work as expected, see bonding.txt for details
5042 printk(KERN_WARNING DRV_NAME
5043 ": Warning: either miimon or arp_interval and "
5044 "arp_ip_target module parameters must be specified, "
5045 "otherwise bonding will not detect link failures! see "
5046 "bonding.txt for details.\n");
5049 if (primary && !USES_PRIMARY(bond_mode)) {
5050 /* currently, using a primary only makes sense
5051 * in active backup, TLB or ALB modes
5053 printk(KERN_WARNING DRV_NAME
5054 ": Warning: %s primary device specified but has no "
5055 "effect in %s mode\n",
5056 primary, bond_mode_name(bond_mode));
5060 if (fail_over_mac) {
5061 fail_over_mac_value = bond_parse_parm(fail_over_mac,
5063 if (fail_over_mac_value == -1) {
5064 printk(KERN_ERR DRV_NAME
5065 ": Error: invalid fail_over_mac \"%s\"\n",
5066 arp_validate == NULL ? "NULL" : arp_validate);
5070 if (bond_mode != BOND_MODE_ACTIVEBACKUP)
5071 printk(KERN_WARNING DRV_NAME
5072 ": Warning: fail_over_mac only affects "
5073 "active-backup mode.\n");
5075 fail_over_mac_value = BOND_FOM_NONE;
5078 /* fill params struct with the proper values */
5079 params->mode = bond_mode;
5080 params->xmit_policy = xmit_hashtype;
5081 params->miimon = miimon;
5082 params->num_grat_arp = num_grat_arp;
5083 params->num_unsol_na = num_unsol_na;
5084 params->arp_interval = arp_interval;
5085 params->arp_validate = arp_validate_value;
5086 params->updelay = updelay;
5087 params->downdelay = downdelay;
5088 params->use_carrier = use_carrier;
5089 params->lacp_fast = lacp_fast;
5090 params->primary[0] = 0;
5091 params->fail_over_mac = fail_over_mac_value;
5094 strncpy(params->primary, primary, IFNAMSIZ);
5095 params->primary[IFNAMSIZ - 1] = 0;
5098 memcpy(params->arp_targets, arp_target, sizeof(arp_target));
5103 static struct lock_class_key bonding_netdev_xmit_lock_key;
5104 static struct lock_class_key bonding_netdev_addr_lock_key;
5106 static void bond_set_lockdep_class_one(struct net_device *dev,
5107 struct netdev_queue *txq,
5110 lockdep_set_class(&txq->_xmit_lock,
5111 &bonding_netdev_xmit_lock_key);
5114 static void bond_set_lockdep_class(struct net_device *dev)
5116 lockdep_set_class(&dev->addr_list_lock,
5117 &bonding_netdev_addr_lock_key);
5118 netdev_for_each_tx_queue(dev, bond_set_lockdep_class_one, NULL);
5121 /* Create a new bond based on the specified name and bonding parameters.
5122 * If name is NULL, obtain a suitable "bond%d" name for us.
5123 * Caller must NOT hold rtnl_lock; we need to release it here before we
5124 * set up our sysfs entries.
5126 int bond_create(char *name, struct bond_params *params)
5128 struct net_device *bond_dev;
5129 struct bonding *bond;
5133 down_write(&bonding_rwsem);
5135 /* Check to see if the bond already exists. */
5137 list_for_each_entry(bond, &bond_dev_list, bond_list)
5138 if (strnicmp(bond->dev->name, name, IFNAMSIZ) == 0) {
5139 printk(KERN_ERR DRV_NAME
5140 ": cannot add bond %s; it already exists\n",
5147 bond_dev = alloc_netdev(sizeof(struct bonding), name ? name : "",
5150 printk(KERN_ERR DRV_NAME
5151 ": %s: eek! can't alloc netdev!\n",
5158 res = dev_alloc_name(bond_dev, "bond%d");
5163 /* bond_init() must be called after dev_alloc_name() (for the
5164 * /proc files), but before register_netdevice(), because we
5165 * need to set function pointers.
5168 res = bond_init(bond_dev, params);
5173 res = register_netdevice(bond_dev);
5178 bond_set_lockdep_class(bond_dev);
5180 netif_carrier_off(bond_dev);
5182 up_write(&bonding_rwsem);
5183 rtnl_unlock(); /* allows sysfs registration of net device */
5184 res = bond_create_sysfs_entry(netdev_priv(bond_dev));
5187 down_write(&bonding_rwsem);
5188 bond_deinit(bond_dev);
5189 unregister_netdevice(bond_dev);
5196 bond_deinit(bond_dev);
5198 free_netdev(bond_dev);
5200 up_write(&bonding_rwsem);
5205 static int __init bonding_init(void)
5209 struct bonding *bond;
5211 printk(KERN_INFO "%s", version);
5213 res = bond_check_params(&bonding_defaults);
5218 #ifdef CONFIG_PROC_FS
5219 bond_create_proc_dir();
5222 init_rwsem(&bonding_rwsem);
5224 for (i = 0; i < max_bonds; i++) {
5225 res = bond_create(NULL, &bonding_defaults);
5230 res = bond_create_sysfs();
5234 register_netdevice_notifier(&bond_netdev_notifier);
5235 register_inetaddr_notifier(&bond_inetaddr_notifier);
5236 bond_register_ipv6_notifier();
5240 list_for_each_entry(bond, &bond_dev_list, bond_list) {
5241 bond_work_cancel_all(bond);
5242 destroy_workqueue(bond->wq);
5245 bond_destroy_sysfs();
5255 static void __exit bonding_exit(void)
5257 unregister_netdevice_notifier(&bond_netdev_notifier);
5258 unregister_inetaddr_notifier(&bond_inetaddr_notifier);
5259 bond_unregister_ipv6_notifier();
5261 bond_destroy_sysfs();
5268 module_init(bonding_init);
5269 module_exit(bonding_exit);
5270 MODULE_LICENSE("GPL");
5271 MODULE_VERSION(DRV_VERSION);
5272 MODULE_DESCRIPTION(DRV_DESCRIPTION ", v" DRV_VERSION);
5273 MODULE_AUTHOR("Thomas Davis, tadavis@lbl.gov and many others");
5274 MODULE_SUPPORTED_DEVICE("most ethernet devices");