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->params.mode == BOND_MODE_TLB) ||
298 (bond->params.mode == BOND_MODE_ALB)) {
299 bond_alb_clear_vlan(bond, vlan_id);
302 dprintk("removed VLAN ID %d from bond %s\n", vlan_id,
307 if (list_empty(&bond->vlan_list) &&
308 (bond->slave_cnt == 0)) {
309 /* Last VLAN removed and no slaves, so
310 * restore block on adding VLANs. This will
311 * be removed once new slaves that are not
312 * VLAN challenged will be added.
314 bond->dev->features |= NETIF_F_VLAN_CHALLENGED;
322 dprintk("couldn't find VLAN ID %d in bond %s\n", vlan_id,
326 write_unlock_bh(&bond->lock);
331 * bond_has_challenged_slaves
332 * @bond: the bond we're working on
334 * Searches the slave list. Returns 1 if a vlan challenged slave
335 * was found, 0 otherwise.
337 * Assumes bond->lock is held.
339 static int bond_has_challenged_slaves(struct bonding *bond)
344 bond_for_each_slave(bond, slave, i) {
345 if (slave->dev->features & NETIF_F_VLAN_CHALLENGED) {
346 dprintk("found VLAN challenged slave - %s\n",
352 dprintk("no VLAN challenged slaves found\n");
357 * bond_next_vlan - safely skip to the next item in the vlans list.
358 * @bond: the bond we're working on
359 * @curr: item we're advancing from
361 * Returns %NULL if list is empty, bond->next_vlan if @curr is %NULL,
362 * or @curr->next otherwise (even if it is @curr itself again).
364 * Caller must hold bond->lock
366 struct vlan_entry *bond_next_vlan(struct bonding *bond, struct vlan_entry *curr)
368 struct vlan_entry *next, *last;
370 if (list_empty(&bond->vlan_list)) {
375 next = list_entry(bond->vlan_list.next,
376 struct vlan_entry, vlan_list);
378 last = list_entry(bond->vlan_list.prev,
379 struct vlan_entry, vlan_list);
381 next = list_entry(bond->vlan_list.next,
382 struct vlan_entry, vlan_list);
384 next = list_entry(curr->vlan_list.next,
385 struct vlan_entry, vlan_list);
393 * bond_dev_queue_xmit - Prepare skb for xmit.
395 * @bond: bond device that got this skb for tx.
396 * @skb: hw accel VLAN tagged skb to transmit
397 * @slave_dev: slave that is supposed to xmit this skbuff
399 * When the bond gets an skb to transmit that is
400 * already hardware accelerated VLAN tagged, and it
401 * needs to relay this skb to a slave that is not
402 * hw accel capable, the skb needs to be "unaccelerated",
403 * i.e. strip the hwaccel tag and re-insert it as part
406 int bond_dev_queue_xmit(struct bonding *bond, struct sk_buff *skb, struct net_device *slave_dev)
408 unsigned short uninitialized_var(vlan_id);
410 if (!list_empty(&bond->vlan_list) &&
411 !(slave_dev->features & NETIF_F_HW_VLAN_TX) &&
412 vlan_get_tag(skb, &vlan_id) == 0) {
413 skb->dev = slave_dev;
414 skb = vlan_put_tag(skb, vlan_id);
416 /* vlan_put_tag() frees the skb in case of error,
417 * so return success here so the calling functions
418 * won't attempt to free is again.
423 skb->dev = slave_dev;
433 * In the following 3 functions, bond_vlan_rx_register(), bond_vlan_rx_add_vid
434 * and bond_vlan_rx_kill_vid, We don't protect the slave list iteration with a
436 * a. This operation is performed in IOCTL context,
437 * b. The operation is protected by the RTNL semaphore in the 8021q code,
438 * c. Holding a lock with BH disabled while directly calling a base driver
439 * entry point is generally a BAD idea.
441 * The design of synchronization/protection for this operation in the 8021q
442 * module is good for one or more VLAN devices over a single physical device
443 * and cannot be extended for a teaming solution like bonding, so there is a
444 * potential race condition here where a net device from the vlan group might
445 * be referenced (either by a base driver or the 8021q code) while it is being
446 * removed from the system. However, it turns out we're not making matters
447 * worse, and if it works for regular VLAN usage it will work here too.
451 * bond_vlan_rx_register - Propagates registration to slaves
452 * @bond_dev: bonding net device that got called
453 * @grp: vlan group being registered
455 static void bond_vlan_rx_register(struct net_device *bond_dev, struct vlan_group *grp)
457 struct bonding *bond = netdev_priv(bond_dev);
463 bond_for_each_slave(bond, slave, i) {
464 struct net_device *slave_dev = slave->dev;
465 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
467 if ((slave_dev->features & NETIF_F_HW_VLAN_RX) &&
468 slave_ops->ndo_vlan_rx_register) {
469 slave_ops->ndo_vlan_rx_register(slave_dev, grp);
475 * bond_vlan_rx_add_vid - Propagates adding an id to slaves
476 * @bond_dev: bonding net device that got called
477 * @vid: vlan id being added
479 static void bond_vlan_rx_add_vid(struct net_device *bond_dev, uint16_t vid)
481 struct bonding *bond = netdev_priv(bond_dev);
485 bond_for_each_slave(bond, slave, i) {
486 struct net_device *slave_dev = slave->dev;
487 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
489 if ((slave_dev->features & NETIF_F_HW_VLAN_FILTER) &&
490 slave_ops->ndo_vlan_rx_add_vid) {
491 slave_ops->ndo_vlan_rx_add_vid(slave_dev, vid);
495 res = bond_add_vlan(bond, vid);
497 printk(KERN_ERR DRV_NAME
498 ": %s: Error: Failed to add vlan id %d\n",
499 bond_dev->name, vid);
504 * bond_vlan_rx_kill_vid - Propagates deleting an id to slaves
505 * @bond_dev: bonding net device that got called
506 * @vid: vlan id being removed
508 static void bond_vlan_rx_kill_vid(struct net_device *bond_dev, uint16_t vid)
510 struct bonding *bond = netdev_priv(bond_dev);
512 struct net_device *vlan_dev;
515 bond_for_each_slave(bond, slave, i) {
516 struct net_device *slave_dev = slave->dev;
517 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
519 if ((slave_dev->features & NETIF_F_HW_VLAN_FILTER) &&
520 slave_ops->ndo_vlan_rx_kill_vid) {
521 /* Save and then restore vlan_dev in the grp array,
522 * since the slave's driver might clear it.
524 vlan_dev = vlan_group_get_device(bond->vlgrp, vid);
525 slave_ops->ndo_vlan_rx_kill_vid(slave_dev, vid);
526 vlan_group_set_device(bond->vlgrp, vid, vlan_dev);
530 res = bond_del_vlan(bond, vid);
532 printk(KERN_ERR DRV_NAME
533 ": %s: Error: Failed to remove vlan id %d\n",
534 bond_dev->name, vid);
538 static void bond_add_vlans_on_slave(struct bonding *bond, struct net_device *slave_dev)
540 struct vlan_entry *vlan;
541 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
543 write_lock_bh(&bond->lock);
545 if (list_empty(&bond->vlan_list))
548 if ((slave_dev->features & NETIF_F_HW_VLAN_RX) &&
549 slave_ops->ndo_vlan_rx_register)
550 slave_ops->ndo_vlan_rx_register(slave_dev, bond->vlgrp);
552 if (!(slave_dev->features & NETIF_F_HW_VLAN_FILTER) ||
553 !(slave_ops->ndo_vlan_rx_add_vid))
556 list_for_each_entry(vlan, &bond->vlan_list, vlan_list)
557 slave_ops->ndo_vlan_rx_add_vid(slave_dev, vlan->vlan_id);
560 write_unlock_bh(&bond->lock);
563 static void bond_del_vlans_from_slave(struct bonding *bond, struct net_device *slave_dev)
565 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
566 struct vlan_entry *vlan;
567 struct net_device *vlan_dev;
569 write_lock_bh(&bond->lock);
571 if (list_empty(&bond->vlan_list))
574 if (!(slave_dev->features & NETIF_F_HW_VLAN_FILTER) ||
575 !(slave_ops->ndo_vlan_rx_kill_vid))
578 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
579 /* Save and then restore vlan_dev in the grp array,
580 * since the slave's driver might clear it.
582 vlan_dev = vlan_group_get_device(bond->vlgrp, vlan->vlan_id);
583 slave_ops->ndo_vlan_rx_kill_vid(slave_dev, vlan->vlan_id);
584 vlan_group_set_device(bond->vlgrp, vlan->vlan_id, vlan_dev);
588 if ((slave_dev->features & NETIF_F_HW_VLAN_RX) &&
589 slave_ops->ndo_vlan_rx_register)
590 slave_ops->ndo_vlan_rx_register(slave_dev, NULL);
593 write_unlock_bh(&bond->lock);
596 /*------------------------------- Link status -------------------------------*/
599 * Set the carrier state for the master according to the state of its
600 * slaves. If any slaves are up, the master is up. In 802.3ad mode,
601 * do special 802.3ad magic.
603 * Returns zero if carrier state does not change, nonzero if it does.
605 static int bond_set_carrier(struct bonding *bond)
610 if (bond->slave_cnt == 0)
613 if (bond->params.mode == BOND_MODE_8023AD)
614 return bond_3ad_set_carrier(bond);
616 bond_for_each_slave(bond, slave, i) {
617 if (slave->link == BOND_LINK_UP) {
618 if (!netif_carrier_ok(bond->dev)) {
619 netif_carrier_on(bond->dev);
627 if (netif_carrier_ok(bond->dev)) {
628 netif_carrier_off(bond->dev);
635 * Get link speed and duplex from the slave's base driver
636 * using ethtool. If for some reason the call fails or the
637 * values are invalid, fake speed and duplex to 100/Full
640 static int bond_update_speed_duplex(struct slave *slave)
642 struct net_device *slave_dev = slave->dev;
643 struct ethtool_cmd etool;
646 /* Fake speed and duplex */
647 slave->speed = SPEED_100;
648 slave->duplex = DUPLEX_FULL;
650 if (!slave_dev->ethtool_ops || !slave_dev->ethtool_ops->get_settings)
653 res = slave_dev->ethtool_ops->get_settings(slave_dev, &etool);
657 switch (etool.speed) {
667 switch (etool.duplex) {
675 slave->speed = etool.speed;
676 slave->duplex = etool.duplex;
682 * if <dev> supports MII link status reporting, check its link status.
684 * We either do MII/ETHTOOL ioctls, or check netif_carrier_ok(),
685 * depening upon the setting of the use_carrier parameter.
687 * Return either BMSR_LSTATUS, meaning that the link is up (or we
688 * can't tell and just pretend it is), or 0, meaning that the link is
691 * If reporting is non-zero, instead of faking link up, return -1 if
692 * both ETHTOOL and MII ioctls fail (meaning the device does not
693 * support them). If use_carrier is set, return whatever it says.
694 * It'd be nice if there was a good way to tell if a driver supports
695 * netif_carrier, but there really isn't.
697 static int bond_check_dev_link(struct bonding *bond, struct net_device *slave_dev, int reporting)
699 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
700 static int (* ioctl)(struct net_device *, struct ifreq *, int);
702 struct mii_ioctl_data *mii;
704 if (bond->params.use_carrier)
705 return netif_carrier_ok(slave_dev) ? BMSR_LSTATUS : 0;
707 ioctl = slave_ops->ndo_do_ioctl;
709 /* TODO: set pointer to correct ioctl on a per team member */
710 /* bases to make this more efficient. that is, once */
711 /* we determine the correct ioctl, we will always */
712 /* call it and not the others for that team */
716 * We cannot assume that SIOCGMIIPHY will also read a
717 * register; not all network drivers (e.g., e100)
721 /* Yes, the mii is overlaid on the ifreq.ifr_ifru */
722 strncpy(ifr.ifr_name, slave_dev->name, IFNAMSIZ);
724 if (IOCTL(slave_dev, &ifr, SIOCGMIIPHY) == 0) {
725 mii->reg_num = MII_BMSR;
726 if (IOCTL(slave_dev, &ifr, SIOCGMIIREG) == 0) {
727 return (mii->val_out & BMSR_LSTATUS);
733 * Some drivers cache ETHTOOL_GLINK for a period of time so we only
734 * attempt to get link status from it if the above MII ioctls fail.
736 if (slave_dev->ethtool_ops) {
737 if (slave_dev->ethtool_ops->get_link) {
740 link = slave_dev->ethtool_ops->get_link(slave_dev);
742 return link ? BMSR_LSTATUS : 0;
747 * If reporting, report that either there's no dev->do_ioctl,
748 * or both SIOCGMIIREG and get_link failed (meaning that we
749 * cannot report link status). If not reporting, pretend
752 return (reporting ? -1 : BMSR_LSTATUS);
755 /*----------------------------- Multicast list ------------------------------*/
758 * Returns 0 if dmi1 and dmi2 are the same, non-0 otherwise
760 static inline int bond_is_dmi_same(struct dev_mc_list *dmi1, struct dev_mc_list *dmi2)
762 return memcmp(dmi1->dmi_addr, dmi2->dmi_addr, dmi1->dmi_addrlen) == 0 &&
763 dmi1->dmi_addrlen == dmi2->dmi_addrlen;
767 * returns dmi entry if found, NULL otherwise
769 static struct dev_mc_list *bond_mc_list_find_dmi(struct dev_mc_list *dmi, struct dev_mc_list *mc_list)
771 struct dev_mc_list *idmi;
773 for (idmi = mc_list; idmi; idmi = idmi->next) {
774 if (bond_is_dmi_same(dmi, idmi)) {
783 * Push the promiscuity flag down to appropriate slaves
785 static int bond_set_promiscuity(struct bonding *bond, int inc)
788 if (USES_PRIMARY(bond->params.mode)) {
789 /* write lock already acquired */
790 if (bond->curr_active_slave) {
791 err = dev_set_promiscuity(bond->curr_active_slave->dev,
797 bond_for_each_slave(bond, slave, i) {
798 err = dev_set_promiscuity(slave->dev, inc);
807 * Push the allmulti flag down to all slaves
809 static int bond_set_allmulti(struct bonding *bond, int inc)
812 if (USES_PRIMARY(bond->params.mode)) {
813 /* write lock already acquired */
814 if (bond->curr_active_slave) {
815 err = dev_set_allmulti(bond->curr_active_slave->dev,
821 bond_for_each_slave(bond, slave, i) {
822 err = dev_set_allmulti(slave->dev, inc);
831 * Add a Multicast address to slaves
834 static void bond_mc_add(struct bonding *bond, void *addr, int alen)
836 if (USES_PRIMARY(bond->params.mode)) {
837 /* write lock already acquired */
838 if (bond->curr_active_slave) {
839 dev_mc_add(bond->curr_active_slave->dev, addr, alen, 0);
844 bond_for_each_slave(bond, slave, i) {
845 dev_mc_add(slave->dev, addr, alen, 0);
851 * Remove a multicast address from slave
854 static void bond_mc_delete(struct bonding *bond, void *addr, int alen)
856 if (USES_PRIMARY(bond->params.mode)) {
857 /* write lock already acquired */
858 if (bond->curr_active_slave) {
859 dev_mc_delete(bond->curr_active_slave->dev, addr, alen, 0);
864 bond_for_each_slave(bond, slave, i) {
865 dev_mc_delete(slave->dev, addr, alen, 0);
872 * Retrieve the list of registered multicast addresses for the bonding
873 * device and retransmit an IGMP JOIN request to the current active
876 static void bond_resend_igmp_join_requests(struct bonding *bond)
878 struct in_device *in_dev;
879 struct ip_mc_list *im;
882 in_dev = __in_dev_get_rcu(bond->dev);
884 for (im = in_dev->mc_list; im; im = im->next) {
885 ip_mc_rejoin_group(im);
893 * Totally destroys the mc_list in bond
895 static void bond_mc_list_destroy(struct bonding *bond)
897 struct dev_mc_list *dmi;
901 bond->mc_list = dmi->next;
905 bond->mc_list = NULL;
909 * Copy all the Multicast addresses from src to the bonding device dst
911 static int bond_mc_list_copy(struct dev_mc_list *mc_list, struct bonding *bond,
914 struct dev_mc_list *dmi, *new_dmi;
916 for (dmi = mc_list; dmi; dmi = dmi->next) {
917 new_dmi = kmalloc(sizeof(struct dev_mc_list), gfp_flag);
920 /* FIXME: Potential memory leak !!! */
924 new_dmi->next = bond->mc_list;
925 bond->mc_list = new_dmi;
926 new_dmi->dmi_addrlen = dmi->dmi_addrlen;
927 memcpy(new_dmi->dmi_addr, dmi->dmi_addr, dmi->dmi_addrlen);
928 new_dmi->dmi_users = dmi->dmi_users;
929 new_dmi->dmi_gusers = dmi->dmi_gusers;
936 * flush all members of flush->mc_list from device dev->mc_list
938 static void bond_mc_list_flush(struct net_device *bond_dev, struct net_device *slave_dev)
940 struct bonding *bond = netdev_priv(bond_dev);
941 struct dev_mc_list *dmi;
943 for (dmi = bond_dev->mc_list; dmi; dmi = dmi->next) {
944 dev_mc_delete(slave_dev, dmi->dmi_addr, dmi->dmi_addrlen, 0);
947 if (bond->params.mode == BOND_MODE_8023AD) {
948 /* del lacpdu mc addr from mc list */
949 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
951 dev_mc_delete(slave_dev, lacpdu_multicast, ETH_ALEN, 0);
955 /*--------------------------- Active slave change ---------------------------*/
958 * Update the mc list and multicast-related flags for the new and
959 * old active slaves (if any) according to the multicast mode, and
960 * promiscuous flags unconditionally.
962 static void bond_mc_swap(struct bonding *bond, struct slave *new_active, struct slave *old_active)
964 struct dev_mc_list *dmi;
966 if (!USES_PRIMARY(bond->params.mode)) {
967 /* nothing to do - mc list is already up-to-date on
974 if (bond->dev->flags & IFF_PROMISC) {
975 dev_set_promiscuity(old_active->dev, -1);
978 if (bond->dev->flags & IFF_ALLMULTI) {
979 dev_set_allmulti(old_active->dev, -1);
982 for (dmi = bond->dev->mc_list; dmi; dmi = dmi->next) {
983 dev_mc_delete(old_active->dev, dmi->dmi_addr, dmi->dmi_addrlen, 0);
988 /* FIXME: Signal errors upstream. */
989 if (bond->dev->flags & IFF_PROMISC) {
990 dev_set_promiscuity(new_active->dev, 1);
993 if (bond->dev->flags & IFF_ALLMULTI) {
994 dev_set_allmulti(new_active->dev, 1);
997 for (dmi = bond->dev->mc_list; dmi; dmi = dmi->next) {
998 dev_mc_add(new_active->dev, dmi->dmi_addr, dmi->dmi_addrlen, 0);
1000 bond_resend_igmp_join_requests(bond);
1005 * bond_do_fail_over_mac
1007 * Perform special MAC address swapping for fail_over_mac settings
1009 * Called with RTNL, bond->lock for read, curr_slave_lock for write_bh.
1011 static void bond_do_fail_over_mac(struct bonding *bond,
1012 struct slave *new_active,
1013 struct slave *old_active)
1015 u8 tmp_mac[ETH_ALEN];
1016 struct sockaddr saddr;
1019 switch (bond->params.fail_over_mac) {
1020 case BOND_FOM_ACTIVE:
1022 memcpy(bond->dev->dev_addr, new_active->dev->dev_addr,
1023 new_active->dev->addr_len);
1025 case BOND_FOM_FOLLOW:
1027 * if new_active && old_active, swap them
1028 * if just old_active, do nothing (going to no active slave)
1029 * if just new_active, set new_active to bond's MAC
1034 write_unlock_bh(&bond->curr_slave_lock);
1035 read_unlock(&bond->lock);
1038 memcpy(tmp_mac, new_active->dev->dev_addr, ETH_ALEN);
1039 memcpy(saddr.sa_data, old_active->dev->dev_addr,
1041 saddr.sa_family = new_active->dev->type;
1043 memcpy(saddr.sa_data, bond->dev->dev_addr, ETH_ALEN);
1044 saddr.sa_family = bond->dev->type;
1047 rv = dev_set_mac_address(new_active->dev, &saddr);
1049 printk(KERN_ERR DRV_NAME
1050 ": %s: Error %d setting MAC of slave %s\n",
1051 bond->dev->name, -rv, new_active->dev->name);
1058 memcpy(saddr.sa_data, tmp_mac, ETH_ALEN);
1059 saddr.sa_family = old_active->dev->type;
1061 rv = dev_set_mac_address(old_active->dev, &saddr);
1063 printk(KERN_ERR DRV_NAME
1064 ": %s: Error %d setting MAC of slave %s\n",
1065 bond->dev->name, -rv, new_active->dev->name);
1067 read_lock(&bond->lock);
1068 write_lock_bh(&bond->curr_slave_lock);
1071 printk(KERN_ERR DRV_NAME
1072 ": %s: bond_do_fail_over_mac impossible: bad policy %d\n",
1073 bond->dev->name, bond->params.fail_over_mac);
1081 * find_best_interface - select the best available slave to be the active one
1082 * @bond: our bonding struct
1084 * Warning: Caller must hold curr_slave_lock for writing.
1086 static struct slave *bond_find_best_slave(struct bonding *bond)
1088 struct slave *new_active, *old_active;
1089 struct slave *bestslave = NULL;
1090 int mintime = bond->params.updelay;
1093 new_active = old_active = bond->curr_active_slave;
1095 if (!new_active) { /* there were no active slaves left */
1096 if (bond->slave_cnt > 0) { /* found one slave */
1097 new_active = bond->first_slave;
1099 return NULL; /* still no slave, return NULL */
1103 /* first try the primary link; if arping, a link must tx/rx traffic
1104 * before it can be considered the curr_active_slave - also, we would skip
1105 * slaves between the curr_active_slave and primary_slave that may be up
1108 if ((bond->primary_slave) &&
1109 (!bond->params.arp_interval) &&
1110 (IS_UP(bond->primary_slave->dev))) {
1111 new_active = bond->primary_slave;
1114 /* remember where to stop iterating over the slaves */
1115 old_active = new_active;
1117 bond_for_each_slave_from(bond, new_active, i, old_active) {
1118 if (IS_UP(new_active->dev)) {
1119 if (new_active->link == BOND_LINK_UP) {
1121 } else if (new_active->link == BOND_LINK_BACK) {
1122 /* link up, but waiting for stabilization */
1123 if (new_active->delay < mintime) {
1124 mintime = new_active->delay;
1125 bestslave = new_active;
1135 * change_active_interface - change the active slave into the specified one
1136 * @bond: our bonding struct
1137 * @new: the new slave to make the active one
1139 * Set the new slave to the bond's settings and unset them on the old
1140 * curr_active_slave.
1141 * Setting include flags, mc-list, promiscuity, allmulti, etc.
1143 * If @new's link state is %BOND_LINK_BACK we'll set it to %BOND_LINK_UP,
1144 * because it is apparently the best available slave we have, even though its
1145 * updelay hasn't timed out yet.
1147 * If new_active is not NULL, caller must hold bond->lock for read and
1148 * curr_slave_lock for write_bh.
1150 void bond_change_active_slave(struct bonding *bond, struct slave *new_active)
1152 struct slave *old_active = bond->curr_active_slave;
1154 if (old_active == new_active) {
1159 new_active->jiffies = jiffies;
1161 if (new_active->link == BOND_LINK_BACK) {
1162 if (USES_PRIMARY(bond->params.mode)) {
1163 printk(KERN_INFO DRV_NAME
1164 ": %s: making interface %s the new "
1165 "active one %d ms earlier.\n",
1166 bond->dev->name, new_active->dev->name,
1167 (bond->params.updelay - new_active->delay) * bond->params.miimon);
1170 new_active->delay = 0;
1171 new_active->link = BOND_LINK_UP;
1173 if (bond->params.mode == BOND_MODE_8023AD) {
1174 bond_3ad_handle_link_change(new_active, BOND_LINK_UP);
1177 if ((bond->params.mode == BOND_MODE_TLB) ||
1178 (bond->params.mode == BOND_MODE_ALB)) {
1179 bond_alb_handle_link_change(bond, new_active, BOND_LINK_UP);
1182 if (USES_PRIMARY(bond->params.mode)) {
1183 printk(KERN_INFO DRV_NAME
1184 ": %s: making interface %s the new "
1186 bond->dev->name, new_active->dev->name);
1191 if (USES_PRIMARY(bond->params.mode)) {
1192 bond_mc_swap(bond, new_active, old_active);
1195 if ((bond->params.mode == BOND_MODE_TLB) ||
1196 (bond->params.mode == BOND_MODE_ALB)) {
1197 bond_alb_handle_active_change(bond, new_active);
1199 bond_set_slave_inactive_flags(old_active);
1201 bond_set_slave_active_flags(new_active);
1203 bond->curr_active_slave = new_active;
1206 if (bond->params.mode == BOND_MODE_ACTIVEBACKUP) {
1208 bond_set_slave_inactive_flags(old_active);
1212 bond_set_slave_active_flags(new_active);
1214 if (bond->params.fail_over_mac)
1215 bond_do_fail_over_mac(bond, new_active,
1218 bond->send_grat_arp = bond->params.num_grat_arp;
1219 bond_send_gratuitous_arp(bond);
1221 bond->send_unsol_na = bond->params.num_unsol_na;
1222 bond_send_unsolicited_na(bond);
1224 write_unlock_bh(&bond->curr_slave_lock);
1225 read_unlock(&bond->lock);
1227 netdev_bonding_change(bond->dev);
1229 read_lock(&bond->lock);
1230 write_lock_bh(&bond->curr_slave_lock);
1236 * bond_select_active_slave - select a new active slave, if needed
1237 * @bond: our bonding struct
1239 * This functions shoud be called when one of the following occurs:
1240 * - The old curr_active_slave has been released or lost its link.
1241 * - The primary_slave has got its link back.
1242 * - A slave has got its link back and there's no old curr_active_slave.
1244 * Caller must hold bond->lock for read and curr_slave_lock for write_bh.
1246 void bond_select_active_slave(struct bonding *bond)
1248 struct slave *best_slave;
1251 best_slave = bond_find_best_slave(bond);
1252 if (best_slave != bond->curr_active_slave) {
1253 bond_change_active_slave(bond, best_slave);
1254 rv = bond_set_carrier(bond);
1258 if (netif_carrier_ok(bond->dev)) {
1259 printk(KERN_INFO DRV_NAME
1260 ": %s: first active interface up!\n",
1263 printk(KERN_INFO DRV_NAME ": %s: "
1264 "now running without any active interface !\n",
1270 /*--------------------------- slave list handling ---------------------------*/
1273 * This function attaches the slave to the end of list.
1275 * bond->lock held for writing by caller.
1277 static void bond_attach_slave(struct bonding *bond, struct slave *new_slave)
1279 if (bond->first_slave == NULL) { /* attaching the first slave */
1280 new_slave->next = new_slave;
1281 new_slave->prev = new_slave;
1282 bond->first_slave = new_slave;
1284 new_slave->next = bond->first_slave;
1285 new_slave->prev = bond->first_slave->prev;
1286 new_slave->next->prev = new_slave;
1287 new_slave->prev->next = new_slave;
1294 * This function detaches the slave from the list.
1295 * WARNING: no check is made to verify if the slave effectively
1296 * belongs to <bond>.
1297 * Nothing is freed on return, structures are just unchained.
1298 * If any slave pointer in bond was pointing to <slave>,
1299 * it should be changed by the calling function.
1301 * bond->lock held for writing by caller.
1303 static void bond_detach_slave(struct bonding *bond, struct slave *slave)
1306 slave->next->prev = slave->prev;
1310 slave->prev->next = slave->next;
1313 if (bond->first_slave == slave) { /* slave is the first slave */
1314 if (bond->slave_cnt > 1) { /* there are more slave */
1315 bond->first_slave = slave->next;
1317 bond->first_slave = NULL; /* slave was the last one */
1326 /*---------------------------------- IOCTL ----------------------------------*/
1328 static int bond_sethwaddr(struct net_device *bond_dev,
1329 struct net_device *slave_dev)
1331 dprintk("bond_dev=%p\n", bond_dev);
1332 dprintk("slave_dev=%p\n", slave_dev);
1333 dprintk("slave_dev->addr_len=%d\n", slave_dev->addr_len);
1334 memcpy(bond_dev->dev_addr, slave_dev->dev_addr, slave_dev->addr_len);
1338 #define BOND_VLAN_FEATURES \
1339 (NETIF_F_VLAN_CHALLENGED | NETIF_F_HW_VLAN_RX | NETIF_F_HW_VLAN_TX | \
1340 NETIF_F_HW_VLAN_FILTER)
1343 * Compute the common dev->feature set available to all slaves. Some
1344 * feature bits are managed elsewhere, so preserve those feature bits
1345 * on the master device.
1347 static int bond_compute_features(struct bonding *bond)
1349 struct slave *slave;
1350 struct net_device *bond_dev = bond->dev;
1351 unsigned long features = bond_dev->features;
1352 unsigned short max_hard_header_len = max((u16)ETH_HLEN,
1353 bond_dev->hard_header_len);
1356 features &= ~(NETIF_F_ALL_CSUM | BOND_VLAN_FEATURES);
1357 features |= NETIF_F_GSO_MASK | NETIF_F_NO_CSUM;
1359 if (!bond->first_slave)
1362 features &= ~NETIF_F_ONE_FOR_ALL;
1364 bond_for_each_slave(bond, slave, i) {
1365 features = netdev_increment_features(features,
1366 slave->dev->features,
1367 NETIF_F_ONE_FOR_ALL);
1368 if (slave->dev->hard_header_len > max_hard_header_len)
1369 max_hard_header_len = slave->dev->hard_header_len;
1373 features |= (bond_dev->features & BOND_VLAN_FEATURES);
1374 bond_dev->features = netdev_fix_features(features, NULL);
1375 bond_dev->hard_header_len = max_hard_header_len;
1380 static void bond_setup_by_slave(struct net_device *bond_dev,
1381 struct net_device *slave_dev)
1383 struct bonding *bond = netdev_priv(bond_dev);
1385 bond_dev->header_ops = slave_dev->header_ops;
1387 bond_dev->type = slave_dev->type;
1388 bond_dev->hard_header_len = slave_dev->hard_header_len;
1389 bond_dev->addr_len = slave_dev->addr_len;
1391 memcpy(bond_dev->broadcast, slave_dev->broadcast,
1392 slave_dev->addr_len);
1393 bond->setup_by_slave = 1;
1396 /* enslave device <slave> to bond device <master> */
1397 int bond_enslave(struct net_device *bond_dev, struct net_device *slave_dev)
1399 struct bonding *bond = netdev_priv(bond_dev);
1400 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
1401 struct slave *new_slave = NULL;
1402 struct dev_mc_list *dmi;
1403 struct sockaddr addr;
1405 int old_features = bond_dev->features;
1408 if (!bond->params.use_carrier && slave_dev->ethtool_ops == NULL &&
1409 slave_ops->ndo_do_ioctl == NULL) {
1410 printk(KERN_WARNING DRV_NAME
1411 ": %s: Warning: no link monitoring support for %s\n",
1412 bond_dev->name, slave_dev->name);
1415 /* bond must be initialized by bond_open() before enslaving */
1416 if (!(bond_dev->flags & IFF_UP)) {
1417 printk(KERN_WARNING DRV_NAME
1418 " %s: master_dev is not up in bond_enslave\n",
1422 /* already enslaved */
1423 if (slave_dev->flags & IFF_SLAVE) {
1424 dprintk("Error, Device was already enslaved\n");
1428 /* vlan challenged mutual exclusion */
1429 /* no need to lock since we're protected by rtnl_lock */
1430 if (slave_dev->features & NETIF_F_VLAN_CHALLENGED) {
1431 dprintk("%s: NETIF_F_VLAN_CHALLENGED\n", slave_dev->name);
1432 if (!list_empty(&bond->vlan_list)) {
1433 printk(KERN_ERR DRV_NAME
1434 ": %s: Error: cannot enslave VLAN "
1435 "challenged slave %s on VLAN enabled "
1436 "bond %s\n", bond_dev->name, slave_dev->name,
1440 printk(KERN_WARNING DRV_NAME
1441 ": %s: Warning: enslaved VLAN challenged "
1442 "slave %s. Adding VLANs will be blocked as "
1443 "long as %s is part of bond %s\n",
1444 bond_dev->name, slave_dev->name, slave_dev->name,
1446 bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
1449 dprintk("%s: ! NETIF_F_VLAN_CHALLENGED\n", slave_dev->name);
1450 if (bond->slave_cnt == 0) {
1451 /* First slave, and it is not VLAN challenged,
1452 * so remove the block of adding VLANs over the bond.
1454 bond_dev->features &= ~NETIF_F_VLAN_CHALLENGED;
1459 * Old ifenslave binaries are no longer supported. These can
1460 * be identified with moderate accurary by the state of the slave:
1461 * the current ifenslave will set the interface down prior to
1462 * enslaving it; the old ifenslave will not.
1464 if ((slave_dev->flags & IFF_UP)) {
1465 printk(KERN_ERR DRV_NAME ": %s is up. "
1466 "This may be due to an out of date ifenslave.\n",
1469 goto err_undo_flags;
1472 /* set bonding device ether type by slave - bonding netdevices are
1473 * created with ether_setup, so when the slave type is not ARPHRD_ETHER
1474 * there is a need to override some of the type dependent attribs/funcs.
1476 * bond ether type mutual exclusion - don't allow slaves of dissimilar
1477 * ether type (eg ARPHRD_ETHER and ARPHRD_INFINIBAND) share the same bond
1479 if (bond->slave_cnt == 0) {
1480 if (slave_dev->type != ARPHRD_ETHER)
1481 bond_setup_by_slave(bond_dev, slave_dev);
1482 } else if (bond_dev->type != slave_dev->type) {
1483 printk(KERN_ERR DRV_NAME ": %s ether type (%d) is different "
1484 "from other slaves (%d), can not enslave it.\n",
1486 slave_dev->type, bond_dev->type);
1488 goto err_undo_flags;
1491 if (slave_ops->ndo_set_mac_address == NULL) {
1492 if (bond->slave_cnt == 0) {
1493 printk(KERN_WARNING DRV_NAME
1494 ": %s: Warning: The first slave device "
1495 "specified does not support setting the MAC "
1496 "address. Setting fail_over_mac to active.",
1498 bond->params.fail_over_mac = BOND_FOM_ACTIVE;
1499 } else if (bond->params.fail_over_mac != BOND_FOM_ACTIVE) {
1500 printk(KERN_ERR DRV_NAME
1501 ": %s: Error: The slave device specified "
1502 "does not support setting the MAC address, "
1503 "but fail_over_mac is not set to active.\n"
1506 goto err_undo_flags;
1510 new_slave = kzalloc(sizeof(struct slave), GFP_KERNEL);
1513 goto err_undo_flags;
1516 /* save slave's original flags before calling
1517 * netdev_set_master and dev_open
1519 new_slave->original_flags = slave_dev->flags;
1522 * Save slave's original ("permanent") mac address for modes
1523 * that need it, and for restoring it upon release, and then
1524 * set it to the master's address
1526 memcpy(new_slave->perm_hwaddr, slave_dev->dev_addr, ETH_ALEN);
1528 if (!bond->params.fail_over_mac) {
1530 * Set slave to master's mac address. The application already
1531 * set the master's mac address to that of the first slave
1533 memcpy(addr.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
1534 addr.sa_family = slave_dev->type;
1535 res = dev_set_mac_address(slave_dev, &addr);
1537 dprintk("Error %d calling set_mac_address\n", res);
1542 res = netdev_set_master(slave_dev, bond_dev);
1544 dprintk("Error %d calling netdev_set_master\n", res);
1545 goto err_restore_mac;
1547 /* open the slave since the application closed it */
1548 res = dev_open(slave_dev);
1550 dprintk("Openning slave %s failed\n", slave_dev->name);
1551 goto err_unset_master;
1554 new_slave->dev = slave_dev;
1555 slave_dev->priv_flags |= IFF_BONDING;
1557 if ((bond->params.mode == BOND_MODE_TLB) ||
1558 (bond->params.mode == BOND_MODE_ALB)) {
1559 /* bond_alb_init_slave() must be called before all other stages since
1560 * it might fail and we do not want to have to undo everything
1562 res = bond_alb_init_slave(bond, new_slave);
1568 /* If the mode USES_PRIMARY, then the new slave gets the
1569 * master's promisc (and mc) settings only if it becomes the
1570 * curr_active_slave, and that is taken care of later when calling
1571 * bond_change_active()
1573 if (!USES_PRIMARY(bond->params.mode)) {
1574 /* set promiscuity level to new slave */
1575 if (bond_dev->flags & IFF_PROMISC) {
1576 res = dev_set_promiscuity(slave_dev, 1);
1581 /* set allmulti level to new slave */
1582 if (bond_dev->flags & IFF_ALLMULTI) {
1583 res = dev_set_allmulti(slave_dev, 1);
1588 netif_addr_lock_bh(bond_dev);
1589 /* upload master's mc_list to new slave */
1590 for (dmi = bond_dev->mc_list; dmi; dmi = dmi->next) {
1591 dev_mc_add (slave_dev, dmi->dmi_addr, dmi->dmi_addrlen, 0);
1593 netif_addr_unlock_bh(bond_dev);
1596 if (bond->params.mode == BOND_MODE_8023AD) {
1597 /* add lacpdu mc addr to mc list */
1598 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
1600 dev_mc_add(slave_dev, lacpdu_multicast, ETH_ALEN, 0);
1603 bond_add_vlans_on_slave(bond, slave_dev);
1605 write_lock_bh(&bond->lock);
1607 bond_attach_slave(bond, new_slave);
1609 new_slave->delay = 0;
1610 new_slave->link_failure_count = 0;
1612 bond_compute_features(bond);
1614 write_unlock_bh(&bond->lock);
1616 read_lock(&bond->lock);
1618 new_slave->last_arp_rx = jiffies;
1620 if (bond->params.miimon && !bond->params.use_carrier) {
1621 link_reporting = bond_check_dev_link(bond, slave_dev, 1);
1623 if ((link_reporting == -1) && !bond->params.arp_interval) {
1625 * miimon is set but a bonded network driver
1626 * does not support ETHTOOL/MII and
1627 * arp_interval is not set. Note: if
1628 * use_carrier is enabled, we will never go
1629 * here (because netif_carrier is always
1630 * supported); thus, we don't need to change
1631 * the messages for netif_carrier.
1633 printk(KERN_WARNING DRV_NAME
1634 ": %s: Warning: MII and ETHTOOL support not "
1635 "available for interface %s, and "
1636 "arp_interval/arp_ip_target module parameters "
1637 "not specified, thus bonding will not detect "
1638 "link failures! see bonding.txt for details.\n",
1639 bond_dev->name, slave_dev->name);
1640 } else if (link_reporting == -1) {
1641 /* unable get link status using mii/ethtool */
1642 printk(KERN_WARNING DRV_NAME
1643 ": %s: Warning: can't get link status from "
1644 "interface %s; the network driver associated "
1645 "with this interface does not support MII or "
1646 "ETHTOOL link status reporting, thus miimon "
1647 "has no effect on this interface.\n",
1648 bond_dev->name, slave_dev->name);
1652 /* check for initial state */
1653 if (!bond->params.miimon ||
1654 (bond_check_dev_link(bond, slave_dev, 0) == BMSR_LSTATUS)) {
1655 if (bond->params.updelay) {
1656 dprintk("Initial state of slave_dev is "
1657 "BOND_LINK_BACK\n");
1658 new_slave->link = BOND_LINK_BACK;
1659 new_slave->delay = bond->params.updelay;
1661 dprintk("Initial state of slave_dev is "
1663 new_slave->link = BOND_LINK_UP;
1665 new_slave->jiffies = jiffies;
1667 dprintk("Initial state of slave_dev is "
1668 "BOND_LINK_DOWN\n");
1669 new_slave->link = BOND_LINK_DOWN;
1672 if (bond_update_speed_duplex(new_slave) &&
1673 (new_slave->link != BOND_LINK_DOWN)) {
1674 printk(KERN_WARNING DRV_NAME
1675 ": %s: Warning: failed to get speed and duplex from %s, "
1676 "assumed to be 100Mb/sec and Full.\n",
1677 bond_dev->name, new_slave->dev->name);
1679 if (bond->params.mode == BOND_MODE_8023AD) {
1680 printk(KERN_WARNING DRV_NAME
1681 ": %s: Warning: Operation of 802.3ad mode requires ETHTOOL "
1682 "support in base driver for proper aggregator "
1683 "selection.\n", bond_dev->name);
1687 if (USES_PRIMARY(bond->params.mode) && bond->params.primary[0]) {
1688 /* if there is a primary slave, remember it */
1689 if (strcmp(bond->params.primary, new_slave->dev->name) == 0) {
1690 bond->primary_slave = new_slave;
1694 write_lock_bh(&bond->curr_slave_lock);
1696 switch (bond->params.mode) {
1697 case BOND_MODE_ACTIVEBACKUP:
1698 bond_set_slave_inactive_flags(new_slave);
1699 bond_select_active_slave(bond);
1701 case BOND_MODE_8023AD:
1702 /* in 802.3ad mode, the internal mechanism
1703 * will activate the slaves in the selected
1706 bond_set_slave_inactive_flags(new_slave);
1707 /* if this is the first slave */
1708 if (bond->slave_cnt == 1) {
1709 SLAVE_AD_INFO(new_slave).id = 1;
1710 /* Initialize AD with the number of times that the AD timer is called in 1 second
1711 * can be called only after the mac address of the bond is set
1713 bond_3ad_initialize(bond, 1000/AD_TIMER_INTERVAL,
1714 bond->params.lacp_fast);
1716 SLAVE_AD_INFO(new_slave).id =
1717 SLAVE_AD_INFO(new_slave->prev).id + 1;
1720 bond_3ad_bind_slave(new_slave);
1724 new_slave->state = BOND_STATE_ACTIVE;
1725 bond_set_slave_inactive_flags(new_slave);
1728 dprintk("This slave is always active in trunk mode\n");
1730 /* always active in trunk mode */
1731 new_slave->state = BOND_STATE_ACTIVE;
1733 /* In trunking mode there is little meaning to curr_active_slave
1734 * anyway (it holds no special properties of the bond device),
1735 * so we can change it without calling change_active_interface()
1737 if (!bond->curr_active_slave) {
1738 bond->curr_active_slave = new_slave;
1741 } /* switch(bond_mode) */
1743 write_unlock_bh(&bond->curr_slave_lock);
1745 bond_set_carrier(bond);
1747 read_unlock(&bond->lock);
1749 res = bond_create_slave_symlinks(bond_dev, slave_dev);
1753 printk(KERN_INFO DRV_NAME
1754 ": %s: enslaving %s as a%s interface with a%s link.\n",
1755 bond_dev->name, slave_dev->name,
1756 new_slave->state == BOND_STATE_ACTIVE ? "n active" : " backup",
1757 new_slave->link != BOND_LINK_DOWN ? "n up" : " down");
1759 /* enslave is successful */
1762 /* Undo stages on error */
1764 dev_close(slave_dev);
1767 netdev_set_master(slave_dev, NULL);
1770 if (!bond->params.fail_over_mac) {
1771 /* XXX TODO - fom follow mode needs to change master's
1772 * MAC if this slave's MAC is in use by the bond, or at
1773 * least print a warning.
1775 memcpy(addr.sa_data, new_slave->perm_hwaddr, ETH_ALEN);
1776 addr.sa_family = slave_dev->type;
1777 dev_set_mac_address(slave_dev, &addr);
1784 bond_dev->features = old_features;
1790 * Try to release the slave device <slave> from the bond device <master>
1791 * It is legal to access curr_active_slave without a lock because all the function
1794 * The rules for slave state should be:
1795 * for Active/Backup:
1796 * Active stays on all backups go down
1797 * for Bonded connections:
1798 * The first up interface should be left on and all others downed.
1800 int bond_release(struct net_device *bond_dev, struct net_device *slave_dev)
1802 struct bonding *bond = netdev_priv(bond_dev);
1803 struct slave *slave, *oldcurrent;
1804 struct sockaddr addr;
1805 int mac_addr_differ;
1807 /* slave is not a slave or master is not master of this slave */
1808 if (!(slave_dev->flags & IFF_SLAVE) ||
1809 (slave_dev->master != bond_dev)) {
1810 printk(KERN_ERR DRV_NAME
1811 ": %s: Error: cannot release %s.\n",
1812 bond_dev->name, slave_dev->name);
1816 write_lock_bh(&bond->lock);
1818 slave = bond_get_slave_by_dev(bond, slave_dev);
1820 /* not a slave of this bond */
1821 printk(KERN_INFO DRV_NAME
1822 ": %s: %s not enslaved\n",
1823 bond_dev->name, slave_dev->name);
1824 write_unlock_bh(&bond->lock);
1828 if (!bond->params.fail_over_mac) {
1829 mac_addr_differ = memcmp(bond_dev->dev_addr, slave->perm_hwaddr,
1831 if (!mac_addr_differ && (bond->slave_cnt > 1))
1832 printk(KERN_WARNING DRV_NAME
1833 ": %s: Warning: the permanent HWaddr of %s - "
1834 "%pM - is still in use by %s. "
1835 "Set the HWaddr of %s to a different address "
1836 "to avoid conflicts.\n",
1837 bond_dev->name, slave_dev->name,
1839 bond_dev->name, slave_dev->name);
1842 /* Inform AD package of unbinding of slave. */
1843 if (bond->params.mode == BOND_MODE_8023AD) {
1844 /* must be called before the slave is
1845 * detached from the list
1847 bond_3ad_unbind_slave(slave);
1850 printk(KERN_INFO DRV_NAME
1851 ": %s: releasing %s interface %s\n",
1853 (slave->state == BOND_STATE_ACTIVE)
1854 ? "active" : "backup",
1857 oldcurrent = bond->curr_active_slave;
1859 bond->current_arp_slave = NULL;
1861 /* release the slave from its bond */
1862 bond_detach_slave(bond, slave);
1864 bond_compute_features(bond);
1866 if (bond->primary_slave == slave) {
1867 bond->primary_slave = NULL;
1870 if (oldcurrent == slave) {
1871 bond_change_active_slave(bond, NULL);
1874 if ((bond->params.mode == BOND_MODE_TLB) ||
1875 (bond->params.mode == BOND_MODE_ALB)) {
1876 /* Must be called only after the slave has been
1877 * detached from the list and the curr_active_slave
1878 * has been cleared (if our_slave == old_current),
1879 * but before a new active slave is selected.
1881 write_unlock_bh(&bond->lock);
1882 bond_alb_deinit_slave(bond, slave);
1883 write_lock_bh(&bond->lock);
1886 if (oldcurrent == slave) {
1888 * Note that we hold RTNL over this sequence, so there
1889 * is no concern that another slave add/remove event
1892 write_unlock_bh(&bond->lock);
1893 read_lock(&bond->lock);
1894 write_lock_bh(&bond->curr_slave_lock);
1896 bond_select_active_slave(bond);
1898 write_unlock_bh(&bond->curr_slave_lock);
1899 read_unlock(&bond->lock);
1900 write_lock_bh(&bond->lock);
1903 if (bond->slave_cnt == 0) {
1904 bond_set_carrier(bond);
1906 /* if the last slave was removed, zero the mac address
1907 * of the master so it will be set by the application
1908 * to the mac address of the first slave
1910 memset(bond_dev->dev_addr, 0, bond_dev->addr_len);
1912 if (list_empty(&bond->vlan_list)) {
1913 bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
1915 printk(KERN_WARNING DRV_NAME
1916 ": %s: Warning: clearing HW address of %s while it "
1917 "still has VLANs.\n",
1918 bond_dev->name, bond_dev->name);
1919 printk(KERN_WARNING DRV_NAME
1920 ": %s: When re-adding slaves, make sure the bond's "
1921 "HW address matches its VLANs'.\n",
1924 } else if ((bond_dev->features & NETIF_F_VLAN_CHALLENGED) &&
1925 !bond_has_challenged_slaves(bond)) {
1926 printk(KERN_INFO DRV_NAME
1927 ": %s: last VLAN challenged slave %s "
1928 "left bond %s. VLAN blocking is removed\n",
1929 bond_dev->name, slave_dev->name, bond_dev->name);
1930 bond_dev->features &= ~NETIF_F_VLAN_CHALLENGED;
1933 write_unlock_bh(&bond->lock);
1935 /* must do this from outside any spinlocks */
1936 bond_destroy_slave_symlinks(bond_dev, slave_dev);
1938 bond_del_vlans_from_slave(bond, slave_dev);
1940 /* If the mode USES_PRIMARY, then we should only remove its
1941 * promisc and mc settings if it was the curr_active_slave, but that was
1942 * already taken care of above when we detached the slave
1944 if (!USES_PRIMARY(bond->params.mode)) {
1945 /* unset promiscuity level from slave */
1946 if (bond_dev->flags & IFF_PROMISC) {
1947 dev_set_promiscuity(slave_dev, -1);
1950 /* unset allmulti level from slave */
1951 if (bond_dev->flags & IFF_ALLMULTI) {
1952 dev_set_allmulti(slave_dev, -1);
1955 /* flush master's mc_list from slave */
1956 netif_addr_lock_bh(bond_dev);
1957 bond_mc_list_flush(bond_dev, slave_dev);
1958 netif_addr_unlock_bh(bond_dev);
1961 netdev_set_master(slave_dev, NULL);
1963 /* close slave before restoring its mac address */
1964 dev_close(slave_dev);
1966 if (bond->params.fail_over_mac != BOND_FOM_ACTIVE) {
1967 /* restore original ("permanent") mac address */
1968 memcpy(addr.sa_data, slave->perm_hwaddr, ETH_ALEN);
1969 addr.sa_family = slave_dev->type;
1970 dev_set_mac_address(slave_dev, &addr);
1973 slave_dev->priv_flags &= ~(IFF_MASTER_8023AD | IFF_MASTER_ALB |
1974 IFF_SLAVE_INACTIVE | IFF_BONDING |
1979 return 0; /* deletion OK */
1983 * Destroy a bonding device.
1984 * Must be under rtnl_lock when this function is called.
1986 void bond_destroy(struct bonding *bond)
1988 bond_deinit(bond->dev);
1989 bond_destroy_sysfs_entry(bond);
1990 unregister_netdevice(bond->dev);
1993 static void bond_destructor(struct net_device *bond_dev)
1995 struct bonding *bond = bond_dev->priv;
1998 destroy_workqueue(bond->wq);
2000 netif_addr_lock_bh(bond_dev);
2001 bond_mc_list_destroy(bond);
2002 netif_addr_unlock_bh(bond_dev);
2004 free_netdev(bond_dev);
2008 * First release a slave and than destroy the bond if no more slaves iare left.
2009 * Must be under rtnl_lock when this function is called.
2011 int bond_release_and_destroy(struct net_device *bond_dev, struct net_device *slave_dev)
2013 struct bonding *bond = netdev_priv(bond_dev);
2016 ret = bond_release(bond_dev, slave_dev);
2017 if ((ret == 0) && (bond->slave_cnt == 0)) {
2018 printk(KERN_INFO DRV_NAME ": %s: destroying bond %s.\n",
2019 bond_dev->name, bond_dev->name);
2026 * This function releases all slaves.
2028 static int bond_release_all(struct net_device *bond_dev)
2030 struct bonding *bond = netdev_priv(bond_dev);
2031 struct slave *slave;
2032 struct net_device *slave_dev;
2033 struct sockaddr addr;
2035 write_lock_bh(&bond->lock);
2037 netif_carrier_off(bond_dev);
2039 if (bond->slave_cnt == 0) {
2043 bond->current_arp_slave = NULL;
2044 bond->primary_slave = NULL;
2045 bond_change_active_slave(bond, NULL);
2047 while ((slave = bond->first_slave) != NULL) {
2048 /* Inform AD package of unbinding of slave
2049 * before slave is detached from the list.
2051 if (bond->params.mode == BOND_MODE_8023AD) {
2052 bond_3ad_unbind_slave(slave);
2055 slave_dev = slave->dev;
2056 bond_detach_slave(bond, slave);
2058 /* now that the slave is detached, unlock and perform
2059 * all the undo steps that should not be called from
2062 write_unlock_bh(&bond->lock);
2064 if ((bond->params.mode == BOND_MODE_TLB) ||
2065 (bond->params.mode == BOND_MODE_ALB)) {
2066 /* must be called only after the slave
2067 * has been detached from the list
2069 bond_alb_deinit_slave(bond, slave);
2072 bond_compute_features(bond);
2074 bond_destroy_slave_symlinks(bond_dev, slave_dev);
2075 bond_del_vlans_from_slave(bond, slave_dev);
2077 /* If the mode USES_PRIMARY, then we should only remove its
2078 * promisc and mc settings if it was the curr_active_slave, but that was
2079 * already taken care of above when we detached the slave
2081 if (!USES_PRIMARY(bond->params.mode)) {
2082 /* unset promiscuity level from slave */
2083 if (bond_dev->flags & IFF_PROMISC) {
2084 dev_set_promiscuity(slave_dev, -1);
2087 /* unset allmulti level from slave */
2088 if (bond_dev->flags & IFF_ALLMULTI) {
2089 dev_set_allmulti(slave_dev, -1);
2092 /* flush master's mc_list from slave */
2093 netif_addr_lock_bh(bond_dev);
2094 bond_mc_list_flush(bond_dev, slave_dev);
2095 netif_addr_unlock_bh(bond_dev);
2098 netdev_set_master(slave_dev, NULL);
2100 /* close slave before restoring its mac address */
2101 dev_close(slave_dev);
2103 if (!bond->params.fail_over_mac) {
2104 /* restore original ("permanent") mac address*/
2105 memcpy(addr.sa_data, slave->perm_hwaddr, ETH_ALEN);
2106 addr.sa_family = slave_dev->type;
2107 dev_set_mac_address(slave_dev, &addr);
2110 slave_dev->priv_flags &= ~(IFF_MASTER_8023AD | IFF_MASTER_ALB |
2111 IFF_SLAVE_INACTIVE);
2115 /* re-acquire the lock before getting the next slave */
2116 write_lock_bh(&bond->lock);
2119 /* zero the mac address of the master so it will be
2120 * set by the application to the mac address of the
2123 memset(bond_dev->dev_addr, 0, bond_dev->addr_len);
2125 if (list_empty(&bond->vlan_list)) {
2126 bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
2128 printk(KERN_WARNING DRV_NAME
2129 ": %s: Warning: clearing HW address of %s while it "
2130 "still has VLANs.\n",
2131 bond_dev->name, bond_dev->name);
2132 printk(KERN_WARNING DRV_NAME
2133 ": %s: When re-adding slaves, make sure the bond's "
2134 "HW address matches its VLANs'.\n",
2138 printk(KERN_INFO DRV_NAME
2139 ": %s: released all slaves\n",
2143 write_unlock_bh(&bond->lock);
2149 * This function changes the active slave to slave <slave_dev>.
2150 * It returns -EINVAL in the following cases.
2151 * - <slave_dev> is not found in the list.
2152 * - There is not active slave now.
2153 * - <slave_dev> is already active.
2154 * - The link state of <slave_dev> is not BOND_LINK_UP.
2155 * - <slave_dev> is not running.
2156 * In these cases, this fuction does nothing.
2157 * In the other cases, currnt_slave pointer is changed and 0 is returned.
2159 static int bond_ioctl_change_active(struct net_device *bond_dev, struct net_device *slave_dev)
2161 struct bonding *bond = netdev_priv(bond_dev);
2162 struct slave *old_active = NULL;
2163 struct slave *new_active = NULL;
2166 if (!USES_PRIMARY(bond->params.mode)) {
2170 /* Verify that master_dev is indeed the master of slave_dev */
2171 if (!(slave_dev->flags & IFF_SLAVE) ||
2172 (slave_dev->master != bond_dev)) {
2176 read_lock(&bond->lock);
2178 read_lock(&bond->curr_slave_lock);
2179 old_active = bond->curr_active_slave;
2180 read_unlock(&bond->curr_slave_lock);
2182 new_active = bond_get_slave_by_dev(bond, slave_dev);
2185 * Changing to the current active: do nothing; return success.
2187 if (new_active && (new_active == old_active)) {
2188 read_unlock(&bond->lock);
2194 (new_active->link == BOND_LINK_UP) &&
2195 IS_UP(new_active->dev)) {
2196 write_lock_bh(&bond->curr_slave_lock);
2197 bond_change_active_slave(bond, new_active);
2198 write_unlock_bh(&bond->curr_slave_lock);
2203 read_unlock(&bond->lock);
2208 static int bond_info_query(struct net_device *bond_dev, struct ifbond *info)
2210 struct bonding *bond = netdev_priv(bond_dev);
2212 info->bond_mode = bond->params.mode;
2213 info->miimon = bond->params.miimon;
2215 read_lock(&bond->lock);
2216 info->num_slaves = bond->slave_cnt;
2217 read_unlock(&bond->lock);
2222 static int bond_slave_info_query(struct net_device *bond_dev, struct ifslave *info)
2224 struct bonding *bond = netdev_priv(bond_dev);
2225 struct slave *slave;
2228 if (info->slave_id < 0) {
2232 read_lock(&bond->lock);
2234 bond_for_each_slave(bond, slave, i) {
2235 if (i == (int)info->slave_id) {
2241 read_unlock(&bond->lock);
2244 strcpy(info->slave_name, slave->dev->name);
2245 info->link = slave->link;
2246 info->state = slave->state;
2247 info->link_failure_count = slave->link_failure_count;
2255 /*-------------------------------- Monitoring -------------------------------*/
2258 static int bond_miimon_inspect(struct bonding *bond)
2260 struct slave *slave;
2261 int i, link_state, commit = 0;
2263 bond_for_each_slave(bond, slave, i) {
2264 slave->new_link = BOND_LINK_NOCHANGE;
2266 link_state = bond_check_dev_link(bond, slave->dev, 0);
2268 switch (slave->link) {
2273 slave->link = BOND_LINK_FAIL;
2274 slave->delay = bond->params.downdelay;
2276 printk(KERN_INFO DRV_NAME
2277 ": %s: link status down for %s"
2278 "interface %s, disabling it in %d ms.\n",
2280 (bond->params.mode ==
2281 BOND_MODE_ACTIVEBACKUP) ?
2282 ((slave->state == BOND_STATE_ACTIVE) ?
2283 "active " : "backup ") : "",
2285 bond->params.downdelay * bond->params.miimon);
2288 case BOND_LINK_FAIL:
2291 * recovered before downdelay expired
2293 slave->link = BOND_LINK_UP;
2294 slave->jiffies = jiffies;
2295 printk(KERN_INFO DRV_NAME
2296 ": %s: link status up again after %d "
2297 "ms for interface %s.\n",
2299 (bond->params.downdelay - slave->delay) *
2300 bond->params.miimon,
2305 if (slave->delay <= 0) {
2306 slave->new_link = BOND_LINK_DOWN;
2314 case BOND_LINK_DOWN:
2318 slave->link = BOND_LINK_BACK;
2319 slave->delay = bond->params.updelay;
2322 printk(KERN_INFO DRV_NAME
2323 ": %s: link status up for "
2324 "interface %s, enabling it in %d ms.\n",
2325 bond->dev->name, slave->dev->name,
2326 bond->params.updelay *
2327 bond->params.miimon);
2330 case BOND_LINK_BACK:
2332 slave->link = BOND_LINK_DOWN;
2333 printk(KERN_INFO DRV_NAME
2334 ": %s: link status down again after %d "
2335 "ms for interface %s.\n",
2337 (bond->params.updelay - slave->delay) *
2338 bond->params.miimon,
2344 if (slave->delay <= 0) {
2345 slave->new_link = BOND_LINK_UP;
2358 static void bond_miimon_commit(struct bonding *bond)
2360 struct slave *slave;
2363 bond_for_each_slave(bond, slave, i) {
2364 switch (slave->new_link) {
2365 case BOND_LINK_NOCHANGE:
2369 slave->link = BOND_LINK_UP;
2370 slave->jiffies = jiffies;
2372 if (bond->params.mode == BOND_MODE_8023AD) {
2373 /* prevent it from being the active one */
2374 slave->state = BOND_STATE_BACKUP;
2375 } else if (bond->params.mode != BOND_MODE_ACTIVEBACKUP) {
2376 /* make it immediately active */
2377 slave->state = BOND_STATE_ACTIVE;
2378 } else if (slave != bond->primary_slave) {
2379 /* prevent it from being the active one */
2380 slave->state = BOND_STATE_BACKUP;
2383 printk(KERN_INFO DRV_NAME
2384 ": %s: link status definitely "
2385 "up for interface %s.\n",
2386 bond->dev->name, slave->dev->name);
2388 /* notify ad that the link status has changed */
2389 if (bond->params.mode == BOND_MODE_8023AD)
2390 bond_3ad_handle_link_change(slave, BOND_LINK_UP);
2392 if ((bond->params.mode == BOND_MODE_TLB) ||
2393 (bond->params.mode == BOND_MODE_ALB))
2394 bond_alb_handle_link_change(bond, slave,
2397 if (!bond->curr_active_slave ||
2398 (slave == bond->primary_slave))
2403 case BOND_LINK_DOWN:
2404 if (slave->link_failure_count < UINT_MAX)
2405 slave->link_failure_count++;
2407 slave->link = BOND_LINK_DOWN;
2409 if (bond->params.mode == BOND_MODE_ACTIVEBACKUP ||
2410 bond->params.mode == BOND_MODE_8023AD)
2411 bond_set_slave_inactive_flags(slave);
2413 printk(KERN_INFO DRV_NAME
2414 ": %s: link status definitely down for "
2415 "interface %s, disabling it\n",
2416 bond->dev->name, slave->dev->name);
2418 if (bond->params.mode == BOND_MODE_8023AD)
2419 bond_3ad_handle_link_change(slave,
2422 if (bond->params.mode == BOND_MODE_TLB ||
2423 bond->params.mode == BOND_MODE_ALB)
2424 bond_alb_handle_link_change(bond, slave,
2427 if (slave == bond->curr_active_slave)
2433 printk(KERN_ERR DRV_NAME
2434 ": %s: invalid new link %d on slave %s\n",
2435 bond->dev->name, slave->new_link,
2437 slave->new_link = BOND_LINK_NOCHANGE;
2444 write_lock_bh(&bond->curr_slave_lock);
2445 bond_select_active_slave(bond);
2446 write_unlock_bh(&bond->curr_slave_lock);
2449 bond_set_carrier(bond);
2455 * Really a wrapper that splits the mii monitor into two phases: an
2456 * inspection, then (if inspection indicates something needs to be done)
2457 * an acquisition of appropriate locks followed by a commit phase to
2458 * implement whatever link state changes are indicated.
2460 void bond_mii_monitor(struct work_struct *work)
2462 struct bonding *bond = container_of(work, struct bonding,
2465 read_lock(&bond->lock);
2466 if (bond->kill_timers)
2469 if (bond->slave_cnt == 0)
2472 if (bond->send_grat_arp) {
2473 read_lock(&bond->curr_slave_lock);
2474 bond_send_gratuitous_arp(bond);
2475 read_unlock(&bond->curr_slave_lock);
2478 if (bond->send_unsol_na) {
2479 read_lock(&bond->curr_slave_lock);
2480 bond_send_unsolicited_na(bond);
2481 read_unlock(&bond->curr_slave_lock);
2484 if (bond_miimon_inspect(bond)) {
2485 read_unlock(&bond->lock);
2487 read_lock(&bond->lock);
2489 bond_miimon_commit(bond);
2491 read_unlock(&bond->lock);
2492 rtnl_unlock(); /* might sleep, hold no other locks */
2493 read_lock(&bond->lock);
2497 if (bond->params.miimon)
2498 queue_delayed_work(bond->wq, &bond->mii_work,
2499 msecs_to_jiffies(bond->params.miimon));
2501 read_unlock(&bond->lock);
2504 static __be32 bond_glean_dev_ip(struct net_device *dev)
2506 struct in_device *idev;
2507 struct in_ifaddr *ifa;
2514 idev = __in_dev_get_rcu(dev);
2518 ifa = idev->ifa_list;
2522 addr = ifa->ifa_local;
2528 static int bond_has_this_ip(struct bonding *bond, __be32 ip)
2530 struct vlan_entry *vlan;
2532 if (ip == bond->master_ip)
2535 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
2536 if (ip == vlan->vlan_ip)
2544 * We go to the (large) trouble of VLAN tagging ARP frames because
2545 * switches in VLAN mode (especially if ports are configured as
2546 * "native" to a VLAN) might not pass non-tagged frames.
2548 static void bond_arp_send(struct net_device *slave_dev, int arp_op, __be32 dest_ip, __be32 src_ip, unsigned short vlan_id)
2550 struct sk_buff *skb;
2552 dprintk("arp %d on slave %s: dst %x src %x vid %d\n", arp_op,
2553 slave_dev->name, dest_ip, src_ip, vlan_id);
2555 skb = arp_create(arp_op, ETH_P_ARP, dest_ip, slave_dev, src_ip,
2556 NULL, slave_dev->dev_addr, NULL);
2559 printk(KERN_ERR DRV_NAME ": ARP packet allocation failed\n");
2563 skb = vlan_put_tag(skb, vlan_id);
2565 printk(KERN_ERR DRV_NAME ": failed to insert VLAN tag\n");
2573 static void bond_arp_send_all(struct bonding *bond, struct slave *slave)
2576 __be32 *targets = bond->params.arp_targets;
2577 struct vlan_entry *vlan;
2578 struct net_device *vlan_dev;
2582 for (i = 0; (i < BOND_MAX_ARP_TARGETS); i++) {
2585 dprintk("basa: target %x\n", targets[i]);
2586 if (list_empty(&bond->vlan_list)) {
2587 dprintk("basa: empty vlan: arp_send\n");
2588 bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2589 bond->master_ip, 0);
2594 * If VLANs are configured, we do a route lookup to
2595 * determine which VLAN interface would be used, so we
2596 * can tag the ARP with the proper VLAN tag.
2598 memset(&fl, 0, sizeof(fl));
2599 fl.fl4_dst = targets[i];
2600 fl.fl4_tos = RTO_ONLINK;
2602 rv = ip_route_output_key(&init_net, &rt, &fl);
2604 if (net_ratelimit()) {
2605 printk(KERN_WARNING DRV_NAME
2606 ": %s: no route to arp_ip_target %pI4\n",
2607 bond->dev->name, &fl.fl4_dst);
2613 * This target is not on a VLAN
2615 if (rt->u.dst.dev == bond->dev) {
2617 dprintk("basa: rtdev == bond->dev: arp_send\n");
2618 bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2619 bond->master_ip, 0);
2624 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
2625 vlan_dev = vlan_group_get_device(bond->vlgrp, vlan->vlan_id);
2626 if (vlan_dev == rt->u.dst.dev) {
2627 vlan_id = vlan->vlan_id;
2628 dprintk("basa: vlan match on %s %d\n",
2629 vlan_dev->name, vlan_id);
2636 bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2637 vlan->vlan_ip, vlan_id);
2641 if (net_ratelimit()) {
2642 printk(KERN_WARNING DRV_NAME
2643 ": %s: no path to arp_ip_target %pI4 via rt.dev %s\n",
2644 bond->dev->name, &fl.fl4_dst,
2645 rt->u.dst.dev ? rt->u.dst.dev->name : "NULL");
2652 * Kick out a gratuitous ARP for an IP on the bonding master plus one
2653 * for each VLAN above us.
2655 * Caller must hold curr_slave_lock for read or better
2657 static void bond_send_gratuitous_arp(struct bonding *bond)
2659 struct slave *slave = bond->curr_active_slave;
2660 struct vlan_entry *vlan;
2661 struct net_device *vlan_dev;
2663 dprintk("bond_send_grat_arp: bond %s slave %s\n", bond->dev->name,
2664 slave ? slave->dev->name : "NULL");
2666 if (!slave || !bond->send_grat_arp ||
2667 test_bit(__LINK_STATE_LINKWATCH_PENDING, &slave->dev->state))
2670 bond->send_grat_arp--;
2672 if (bond->master_ip) {
2673 bond_arp_send(slave->dev, ARPOP_REPLY, bond->master_ip,
2674 bond->master_ip, 0);
2677 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
2678 vlan_dev = vlan_group_get_device(bond->vlgrp, vlan->vlan_id);
2679 if (vlan->vlan_ip) {
2680 bond_arp_send(slave->dev, ARPOP_REPLY, vlan->vlan_ip,
2681 vlan->vlan_ip, vlan->vlan_id);
2686 static void bond_validate_arp(struct bonding *bond, struct slave *slave, __be32 sip, __be32 tip)
2689 __be32 *targets = bond->params.arp_targets;
2691 targets = bond->params.arp_targets;
2692 for (i = 0; (i < BOND_MAX_ARP_TARGETS) && targets[i]; i++) {
2693 dprintk("bva: sip %pI4 tip %pI4 t[%d] %pI4 bhti(tip) %d\n",
2694 &sip, &tip, i, &targets[i], bond_has_this_ip(bond, tip));
2695 if (sip == targets[i]) {
2696 if (bond_has_this_ip(bond, tip))
2697 slave->last_arp_rx = jiffies;
2703 static int bond_arp_rcv(struct sk_buff *skb, struct net_device *dev, struct packet_type *pt, struct net_device *orig_dev)
2706 struct slave *slave;
2707 struct bonding *bond;
2708 unsigned char *arp_ptr;
2711 if (dev_net(dev) != &init_net)
2714 if (!(dev->priv_flags & IFF_BONDING) || !(dev->flags & IFF_MASTER))
2717 bond = netdev_priv(dev);
2718 read_lock(&bond->lock);
2720 dprintk("bond_arp_rcv: bond %s skb->dev %s orig_dev %s\n",
2721 bond->dev->name, skb->dev ? skb->dev->name : "NULL",
2722 orig_dev ? orig_dev->name : "NULL");
2724 slave = bond_get_slave_by_dev(bond, orig_dev);
2725 if (!slave || !slave_do_arp_validate(bond, slave))
2728 if (!pskb_may_pull(skb, arp_hdr_len(dev)))
2732 if (arp->ar_hln != dev->addr_len ||
2733 skb->pkt_type == PACKET_OTHERHOST ||
2734 skb->pkt_type == PACKET_LOOPBACK ||
2735 arp->ar_hrd != htons(ARPHRD_ETHER) ||
2736 arp->ar_pro != htons(ETH_P_IP) ||
2740 arp_ptr = (unsigned char *)(arp + 1);
2741 arp_ptr += dev->addr_len;
2742 memcpy(&sip, arp_ptr, 4);
2743 arp_ptr += 4 + dev->addr_len;
2744 memcpy(&tip, arp_ptr, 4);
2746 dprintk("bond_arp_rcv: %s %s/%d av %d sv %d sip %pI4 tip %pI4\n",
2747 bond->dev->name, slave->dev->name, slave->state,
2748 bond->params.arp_validate, slave_do_arp_validate(bond, slave),
2752 * Backup slaves won't see the ARP reply, but do come through
2753 * here for each ARP probe (so we swap the sip/tip to validate
2754 * the probe). In a "redundant switch, common router" type of
2755 * configuration, the ARP probe will (hopefully) travel from
2756 * the active, through one switch, the router, then the other
2757 * switch before reaching the backup.
2759 if (slave->state == BOND_STATE_ACTIVE)
2760 bond_validate_arp(bond, slave, sip, tip);
2762 bond_validate_arp(bond, slave, tip, sip);
2765 read_unlock(&bond->lock);
2768 return NET_RX_SUCCESS;
2772 * this function is called regularly to monitor each slave's link
2773 * ensuring that traffic is being sent and received when arp monitoring
2774 * is used in load-balancing mode. if the adapter has been dormant, then an
2775 * arp is transmitted to generate traffic. see activebackup_arp_monitor for
2776 * arp monitoring in active backup mode.
2778 void bond_loadbalance_arp_mon(struct work_struct *work)
2780 struct bonding *bond = container_of(work, struct bonding,
2782 struct slave *slave, *oldcurrent;
2783 int do_failover = 0;
2787 read_lock(&bond->lock);
2789 delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
2791 if (bond->kill_timers) {
2795 if (bond->slave_cnt == 0) {
2799 read_lock(&bond->curr_slave_lock);
2800 oldcurrent = bond->curr_active_slave;
2801 read_unlock(&bond->curr_slave_lock);
2803 /* see if any of the previous devices are up now (i.e. they have
2804 * xmt and rcv traffic). the curr_active_slave does not come into
2805 * the picture unless it is null. also, slave->jiffies is not needed
2806 * here because we send an arp on each slave and give a slave as
2807 * long as it needs to get the tx/rx within the delta.
2808 * TODO: what about up/down delay in arp mode? it wasn't here before
2811 bond_for_each_slave(bond, slave, i) {
2812 if (slave->link != BOND_LINK_UP) {
2813 if (time_before_eq(jiffies, slave->dev->trans_start + delta_in_ticks) &&
2814 time_before_eq(jiffies, slave->dev->last_rx + delta_in_ticks)) {
2816 slave->link = BOND_LINK_UP;
2817 slave->state = BOND_STATE_ACTIVE;
2819 /* primary_slave has no meaning in round-robin
2820 * mode. the window of a slave being up and
2821 * curr_active_slave being null after enslaving
2825 printk(KERN_INFO DRV_NAME
2826 ": %s: link status definitely "
2827 "up for interface %s, ",
2832 printk(KERN_INFO DRV_NAME
2833 ": %s: interface %s is now up\n",
2839 /* slave->link == BOND_LINK_UP */
2841 /* not all switches will respond to an arp request
2842 * when the source ip is 0, so don't take the link down
2843 * if we don't know our ip yet
2845 if (time_after_eq(jiffies, slave->dev->trans_start + 2*delta_in_ticks) ||
2846 (time_after_eq(jiffies, slave->dev->last_rx + 2*delta_in_ticks))) {
2848 slave->link = BOND_LINK_DOWN;
2849 slave->state = BOND_STATE_BACKUP;
2851 if (slave->link_failure_count < UINT_MAX) {
2852 slave->link_failure_count++;
2855 printk(KERN_INFO DRV_NAME
2856 ": %s: interface %s is now down.\n",
2860 if (slave == oldcurrent) {
2866 /* note: if switch is in round-robin mode, all links
2867 * must tx arp to ensure all links rx an arp - otherwise
2868 * links may oscillate or not come up at all; if switch is
2869 * in something like xor mode, there is nothing we can
2870 * do - all replies will be rx'ed on same link causing slaves
2871 * to be unstable during low/no traffic periods
2873 if (IS_UP(slave->dev)) {
2874 bond_arp_send_all(bond, slave);
2879 write_lock_bh(&bond->curr_slave_lock);
2881 bond_select_active_slave(bond);
2883 write_unlock_bh(&bond->curr_slave_lock);
2887 if (bond->params.arp_interval)
2888 queue_delayed_work(bond->wq, &bond->arp_work, delta_in_ticks);
2890 read_unlock(&bond->lock);
2894 * Called to inspect slaves for active-backup mode ARP monitor link state
2895 * changes. Sets new_link in slaves to specify what action should take
2896 * place for the slave. Returns 0 if no changes are found, >0 if changes
2897 * to link states must be committed.
2899 * Called with bond->lock held for read.
2901 static int bond_ab_arp_inspect(struct bonding *bond, int delta_in_ticks)
2903 struct slave *slave;
2906 bond_for_each_slave(bond, slave, i) {
2907 slave->new_link = BOND_LINK_NOCHANGE;
2909 if (slave->link != BOND_LINK_UP) {
2910 if (time_before_eq(jiffies, slave_last_rx(bond, slave) +
2912 slave->new_link = BOND_LINK_UP;
2920 * Give slaves 2*delta after being enslaved or made
2921 * active. This avoids bouncing, as the last receive
2922 * times need a full ARP monitor cycle to be updated.
2924 if (!time_after_eq(jiffies, slave->jiffies +
2925 2 * delta_in_ticks))
2929 * Backup slave is down if:
2930 * - No current_arp_slave AND
2931 * - more than 3*delta since last receive AND
2932 * - the bond has an IP address
2934 * Note: a non-null current_arp_slave indicates
2935 * the curr_active_slave went down and we are
2936 * searching for a new one; under this condition
2937 * we only take the curr_active_slave down - this
2938 * gives each slave a chance to tx/rx traffic
2939 * before being taken out
2941 if (slave->state == BOND_STATE_BACKUP &&
2942 !bond->current_arp_slave &&
2943 time_after(jiffies, slave_last_rx(bond, slave) +
2944 3 * delta_in_ticks)) {
2945 slave->new_link = BOND_LINK_DOWN;
2950 * Active slave is down if:
2951 * - more than 2*delta since transmitting OR
2952 * - (more than 2*delta since receive AND
2953 * the bond has an IP address)
2955 if ((slave->state == BOND_STATE_ACTIVE) &&
2956 (time_after_eq(jiffies, slave->dev->trans_start +
2957 2 * delta_in_ticks) ||
2958 (time_after_eq(jiffies, slave_last_rx(bond, slave)
2959 + 2 * delta_in_ticks)))) {
2960 slave->new_link = BOND_LINK_DOWN;
2965 read_lock(&bond->curr_slave_lock);
2968 * Trigger a commit if the primary option setting has changed.
2970 if (bond->primary_slave &&
2971 (bond->primary_slave != bond->curr_active_slave) &&
2972 (bond->primary_slave->link == BOND_LINK_UP))
2975 read_unlock(&bond->curr_slave_lock);
2981 * Called to commit link state changes noted by inspection step of
2982 * active-backup mode ARP monitor.
2984 * Called with RTNL and bond->lock for read.
2986 static void bond_ab_arp_commit(struct bonding *bond, int delta_in_ticks)
2988 struct slave *slave;
2991 bond_for_each_slave(bond, slave, i) {
2992 switch (slave->new_link) {
2993 case BOND_LINK_NOCHANGE:
2997 write_lock_bh(&bond->curr_slave_lock);
2999 if (!bond->curr_active_slave &&
3000 time_before_eq(jiffies, slave->dev->trans_start +
3002 slave->link = BOND_LINK_UP;
3003 bond_change_active_slave(bond, slave);
3004 bond->current_arp_slave = NULL;
3006 printk(KERN_INFO DRV_NAME
3007 ": %s: %s is up and now the "
3008 "active interface\n",
3009 bond->dev->name, slave->dev->name);
3011 } else if (bond->curr_active_slave != slave) {
3012 /* this slave has just come up but we
3013 * already have a current slave; this can
3014 * also happen if bond_enslave adds a new
3015 * slave that is up while we are searching
3018 slave->link = BOND_LINK_UP;
3019 bond_set_slave_inactive_flags(slave);
3020 bond->current_arp_slave = NULL;
3022 printk(KERN_INFO DRV_NAME
3023 ": %s: backup interface %s is now up\n",
3024 bond->dev->name, slave->dev->name);
3027 write_unlock_bh(&bond->curr_slave_lock);
3031 case BOND_LINK_DOWN:
3032 if (slave->link_failure_count < UINT_MAX)
3033 slave->link_failure_count++;
3035 slave->link = BOND_LINK_DOWN;
3037 if (slave == bond->curr_active_slave) {
3038 printk(KERN_INFO DRV_NAME
3039 ": %s: link status down for active "
3040 "interface %s, disabling it\n",
3041 bond->dev->name, slave->dev->name);
3043 bond_set_slave_inactive_flags(slave);
3045 write_lock_bh(&bond->curr_slave_lock);
3047 bond_select_active_slave(bond);
3048 if (bond->curr_active_slave)
3049 bond->curr_active_slave->jiffies =
3052 write_unlock_bh(&bond->curr_slave_lock);
3054 bond->current_arp_slave = NULL;
3056 } else if (slave->state == BOND_STATE_BACKUP) {
3057 printk(KERN_INFO DRV_NAME
3058 ": %s: backup interface %s is now down\n",
3059 bond->dev->name, slave->dev->name);
3061 bond_set_slave_inactive_flags(slave);
3066 printk(KERN_ERR DRV_NAME
3067 ": %s: impossible: new_link %d on slave %s\n",
3068 bond->dev->name, slave->new_link,
3074 * No race with changes to primary via sysfs, as we hold rtnl.
3076 if (bond->primary_slave &&
3077 (bond->primary_slave != bond->curr_active_slave) &&
3078 (bond->primary_slave->link == BOND_LINK_UP)) {
3079 write_lock_bh(&bond->curr_slave_lock);
3080 bond_change_active_slave(bond, bond->primary_slave);
3081 write_unlock_bh(&bond->curr_slave_lock);
3084 bond_set_carrier(bond);
3088 * Send ARP probes for active-backup mode ARP monitor.
3090 * Called with bond->lock held for read.
3092 static void bond_ab_arp_probe(struct bonding *bond)
3094 struct slave *slave;
3097 read_lock(&bond->curr_slave_lock);
3099 if (bond->current_arp_slave && bond->curr_active_slave)
3100 printk("PROBE: c_arp %s && cas %s BAD\n",
3101 bond->current_arp_slave->dev->name,
3102 bond->curr_active_slave->dev->name);
3104 if (bond->curr_active_slave) {
3105 bond_arp_send_all(bond, bond->curr_active_slave);
3106 read_unlock(&bond->curr_slave_lock);
3110 read_unlock(&bond->curr_slave_lock);
3112 /* if we don't have a curr_active_slave, search for the next available
3113 * backup slave from the current_arp_slave and make it the candidate
3114 * for becoming the curr_active_slave
3117 if (!bond->current_arp_slave) {
3118 bond->current_arp_slave = bond->first_slave;
3119 if (!bond->current_arp_slave)
3123 bond_set_slave_inactive_flags(bond->current_arp_slave);
3125 /* search for next candidate */
3126 bond_for_each_slave_from(bond, slave, i, bond->current_arp_slave->next) {
3127 if (IS_UP(slave->dev)) {
3128 slave->link = BOND_LINK_BACK;
3129 bond_set_slave_active_flags(slave);
3130 bond_arp_send_all(bond, slave);
3131 slave->jiffies = jiffies;
3132 bond->current_arp_slave = slave;
3136 /* if the link state is up at this point, we
3137 * mark it down - this can happen if we have
3138 * simultaneous link failures and
3139 * reselect_active_interface doesn't make this
3140 * one the current slave so it is still marked
3141 * up when it is actually down
3143 if (slave->link == BOND_LINK_UP) {
3144 slave->link = BOND_LINK_DOWN;
3145 if (slave->link_failure_count < UINT_MAX)
3146 slave->link_failure_count++;
3148 bond_set_slave_inactive_flags(slave);
3150 printk(KERN_INFO DRV_NAME
3151 ": %s: backup interface %s is now down.\n",
3152 bond->dev->name, slave->dev->name);
3157 void bond_activebackup_arp_mon(struct work_struct *work)
3159 struct bonding *bond = container_of(work, struct bonding,
3163 read_lock(&bond->lock);
3165 if (bond->kill_timers)
3168 delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
3170 if (bond->slave_cnt == 0)
3173 if (bond->send_grat_arp) {
3174 read_lock(&bond->curr_slave_lock);
3175 bond_send_gratuitous_arp(bond);
3176 read_unlock(&bond->curr_slave_lock);
3179 if (bond->send_unsol_na) {
3180 read_lock(&bond->curr_slave_lock);
3181 bond_send_unsolicited_na(bond);
3182 read_unlock(&bond->curr_slave_lock);
3185 if (bond_ab_arp_inspect(bond, delta_in_ticks)) {
3186 read_unlock(&bond->lock);
3188 read_lock(&bond->lock);
3190 bond_ab_arp_commit(bond, delta_in_ticks);
3192 read_unlock(&bond->lock);
3194 read_lock(&bond->lock);
3197 bond_ab_arp_probe(bond);
3200 if (bond->params.arp_interval) {
3201 queue_delayed_work(bond->wq, &bond->arp_work, delta_in_ticks);
3204 read_unlock(&bond->lock);
3207 /*------------------------------ proc/seq_file-------------------------------*/
3209 #ifdef CONFIG_PROC_FS
3211 static void *bond_info_seq_start(struct seq_file *seq, loff_t *pos)
3213 struct bonding *bond = seq->private;
3215 struct slave *slave;
3218 /* make sure the bond won't be taken away */
3219 read_lock(&dev_base_lock);
3220 read_lock(&bond->lock);
3223 return SEQ_START_TOKEN;
3226 bond_for_each_slave(bond, slave, i) {
3227 if (++off == *pos) {
3235 static void *bond_info_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3237 struct bonding *bond = seq->private;
3238 struct slave *slave = v;
3241 if (v == SEQ_START_TOKEN) {
3242 return bond->first_slave;
3245 slave = slave->next;
3247 return (slave == bond->first_slave) ? NULL : slave;
3250 static void bond_info_seq_stop(struct seq_file *seq, void *v)
3252 struct bonding *bond = seq->private;
3254 read_unlock(&bond->lock);
3255 read_unlock(&dev_base_lock);
3258 static void bond_info_show_master(struct seq_file *seq)
3260 struct bonding *bond = seq->private;
3264 read_lock(&bond->curr_slave_lock);
3265 curr = bond->curr_active_slave;
3266 read_unlock(&bond->curr_slave_lock);
3268 seq_printf(seq, "Bonding Mode: %s",
3269 bond_mode_name(bond->params.mode));
3271 if (bond->params.mode == BOND_MODE_ACTIVEBACKUP &&
3272 bond->params.fail_over_mac)
3273 seq_printf(seq, " (fail_over_mac %s)",
3274 fail_over_mac_tbl[bond->params.fail_over_mac].modename);
3276 seq_printf(seq, "\n");
3278 if (bond->params.mode == BOND_MODE_XOR ||
3279 bond->params.mode == BOND_MODE_8023AD) {
3280 seq_printf(seq, "Transmit Hash Policy: %s (%d)\n",
3281 xmit_hashtype_tbl[bond->params.xmit_policy].modename,
3282 bond->params.xmit_policy);
3285 if (USES_PRIMARY(bond->params.mode)) {
3286 seq_printf(seq, "Primary Slave: %s\n",
3287 (bond->primary_slave) ?
3288 bond->primary_slave->dev->name : "None");
3290 seq_printf(seq, "Currently Active Slave: %s\n",
3291 (curr) ? curr->dev->name : "None");
3294 seq_printf(seq, "MII Status: %s\n", netif_carrier_ok(bond->dev) ?
3296 seq_printf(seq, "MII Polling Interval (ms): %d\n", bond->params.miimon);
3297 seq_printf(seq, "Up Delay (ms): %d\n",
3298 bond->params.updelay * bond->params.miimon);
3299 seq_printf(seq, "Down Delay (ms): %d\n",
3300 bond->params.downdelay * bond->params.miimon);
3303 /* ARP information */
3304 if(bond->params.arp_interval > 0) {
3306 seq_printf(seq, "ARP Polling Interval (ms): %d\n",
3307 bond->params.arp_interval);
3309 seq_printf(seq, "ARP IP target/s (n.n.n.n form):");
3311 for(i = 0; (i < BOND_MAX_ARP_TARGETS) ;i++) {
3312 if (!bond->params.arp_targets[i])
3315 seq_printf(seq, ",");
3316 seq_printf(seq, " %pI4", &bond->params.arp_targets[i]);
3319 seq_printf(seq, "\n");
3322 if (bond->params.mode == BOND_MODE_8023AD) {
3323 struct ad_info ad_info;
3325 seq_puts(seq, "\n802.3ad info\n");
3326 seq_printf(seq, "LACP rate: %s\n",
3327 (bond->params.lacp_fast) ? "fast" : "slow");
3328 seq_printf(seq, "Aggregator selection policy (ad_select): %s\n",
3329 ad_select_tbl[bond->params.ad_select].modename);
3331 if (bond_3ad_get_active_agg_info(bond, &ad_info)) {
3332 seq_printf(seq, "bond %s has no active aggregator\n",
3335 seq_printf(seq, "Active Aggregator Info:\n");
3337 seq_printf(seq, "\tAggregator ID: %d\n",
3338 ad_info.aggregator_id);
3339 seq_printf(seq, "\tNumber of ports: %d\n",
3341 seq_printf(seq, "\tActor Key: %d\n",
3343 seq_printf(seq, "\tPartner Key: %d\n",
3344 ad_info.partner_key);
3345 seq_printf(seq, "\tPartner Mac Address: %pM\n",
3346 ad_info.partner_system);
3351 static void bond_info_show_slave(struct seq_file *seq, const struct slave *slave)
3353 struct bonding *bond = seq->private;
3355 seq_printf(seq, "\nSlave Interface: %s\n", slave->dev->name);
3356 seq_printf(seq, "MII Status: %s\n",
3357 (slave->link == BOND_LINK_UP) ? "up" : "down");
3358 seq_printf(seq, "Link Failure Count: %u\n",
3359 slave->link_failure_count);
3361 seq_printf(seq, "Permanent HW addr: %pM\n", slave->perm_hwaddr);
3363 if (bond->params.mode == BOND_MODE_8023AD) {
3364 const struct aggregator *agg
3365 = SLAVE_AD_INFO(slave).port.aggregator;
3368 seq_printf(seq, "Aggregator ID: %d\n",
3369 agg->aggregator_identifier);
3371 seq_puts(seq, "Aggregator ID: N/A\n");
3376 static int bond_info_seq_show(struct seq_file *seq, void *v)
3378 if (v == SEQ_START_TOKEN) {
3379 seq_printf(seq, "%s\n", version);
3380 bond_info_show_master(seq);
3382 bond_info_show_slave(seq, v);
3388 static struct seq_operations bond_info_seq_ops = {
3389 .start = bond_info_seq_start,
3390 .next = bond_info_seq_next,
3391 .stop = bond_info_seq_stop,
3392 .show = bond_info_seq_show,
3395 static int bond_info_open(struct inode *inode, struct file *file)
3397 struct seq_file *seq;
3398 struct proc_dir_entry *proc;
3401 res = seq_open(file, &bond_info_seq_ops);
3403 /* recover the pointer buried in proc_dir_entry data */
3404 seq = file->private_data;
3406 seq->private = proc->data;
3412 static const struct file_operations bond_info_fops = {
3413 .owner = THIS_MODULE,
3414 .open = bond_info_open,
3416 .llseek = seq_lseek,
3417 .release = seq_release,
3420 static int bond_create_proc_entry(struct bonding *bond)
3422 struct net_device *bond_dev = bond->dev;
3424 if (bond_proc_dir) {
3425 bond->proc_entry = proc_create_data(bond_dev->name,
3426 S_IRUGO, bond_proc_dir,
3427 &bond_info_fops, bond);
3428 if (bond->proc_entry == NULL) {
3429 printk(KERN_WARNING DRV_NAME
3430 ": Warning: Cannot create /proc/net/%s/%s\n",
3431 DRV_NAME, bond_dev->name);
3433 memcpy(bond->proc_file_name, bond_dev->name, IFNAMSIZ);
3440 static void bond_remove_proc_entry(struct bonding *bond)
3442 if (bond_proc_dir && bond->proc_entry) {
3443 remove_proc_entry(bond->proc_file_name, bond_proc_dir);
3444 memset(bond->proc_file_name, 0, IFNAMSIZ);
3445 bond->proc_entry = NULL;
3449 /* Create the bonding directory under /proc/net, if doesn't exist yet.
3450 * Caller must hold rtnl_lock.
3452 static void bond_create_proc_dir(void)
3454 int len = strlen(DRV_NAME);
3456 for (bond_proc_dir = init_net.proc_net->subdir; bond_proc_dir;
3457 bond_proc_dir = bond_proc_dir->next) {
3458 if ((bond_proc_dir->namelen == len) &&
3459 !memcmp(bond_proc_dir->name, DRV_NAME, len)) {
3464 if (!bond_proc_dir) {
3465 bond_proc_dir = proc_mkdir(DRV_NAME, init_net.proc_net);
3466 if (bond_proc_dir) {
3467 bond_proc_dir->owner = THIS_MODULE;
3469 printk(KERN_WARNING DRV_NAME
3470 ": Warning: cannot create /proc/net/%s\n",
3476 /* Destroy the bonding directory under /proc/net, if empty.
3477 * Caller must hold rtnl_lock.
3479 static void bond_destroy_proc_dir(void)
3481 struct proc_dir_entry *de;
3483 if (!bond_proc_dir) {
3487 /* verify that the /proc dir is empty */
3488 for (de = bond_proc_dir->subdir; de; de = de->next) {
3489 /* ignore . and .. */
3490 if (*(de->name) != '.') {
3496 if (bond_proc_dir->owner == THIS_MODULE) {
3497 bond_proc_dir->owner = NULL;
3500 remove_proc_entry(DRV_NAME, init_net.proc_net);
3501 bond_proc_dir = NULL;
3504 #endif /* CONFIG_PROC_FS */
3506 /*-------------------------- netdev event handling --------------------------*/
3509 * Change device name
3511 static int bond_event_changename(struct bonding *bond)
3513 #ifdef CONFIG_PROC_FS
3514 bond_remove_proc_entry(bond);
3515 bond_create_proc_entry(bond);
3517 down_write(&(bonding_rwsem));
3518 bond_destroy_sysfs_entry(bond);
3519 bond_create_sysfs_entry(bond);
3520 up_write(&(bonding_rwsem));
3524 static int bond_master_netdev_event(unsigned long event, struct net_device *bond_dev)
3526 struct bonding *event_bond = netdev_priv(bond_dev);
3529 case NETDEV_CHANGENAME:
3530 return bond_event_changename(event_bond);
3531 case NETDEV_UNREGISTER:
3532 bond_release_all(event_bond->dev);
3541 static int bond_slave_netdev_event(unsigned long event, struct net_device *slave_dev)
3543 struct net_device *bond_dev = slave_dev->master;
3544 struct bonding *bond = netdev_priv(bond_dev);
3547 case NETDEV_UNREGISTER:
3549 if (bond->setup_by_slave)
3550 bond_release_and_destroy(bond_dev, slave_dev);
3552 bond_release(bond_dev, slave_dev);
3557 * TODO: is this what we get if somebody
3558 * sets up a hierarchical bond, then rmmod's
3559 * one of the slave bonding devices?
3564 * ... Or is it this?
3567 case NETDEV_CHANGEMTU:
3569 * TODO: Should slaves be allowed to
3570 * independently alter their MTU? For
3571 * an active-backup bond, slaves need
3572 * not be the same type of device, so
3573 * MTUs may vary. For other modes,
3574 * slaves arguably should have the
3575 * same MTUs. To do this, we'd need to
3576 * take over the slave's change_mtu
3577 * function for the duration of their
3581 case NETDEV_CHANGENAME:
3583 * TODO: handle changing the primary's name
3586 case NETDEV_FEAT_CHANGE:
3587 bond_compute_features(bond);
3597 * bond_netdev_event: handle netdev notifier chain events.
3599 * This function receives events for the netdev chain. The caller (an
3600 * ioctl handler calling blocking_notifier_call_chain) holds the necessary
3601 * locks for us to safely manipulate the slave devices (RTNL lock,
3604 static int bond_netdev_event(struct notifier_block *this, unsigned long event, void *ptr)
3606 struct net_device *event_dev = (struct net_device *)ptr;
3608 if (dev_net(event_dev) != &init_net)
3611 dprintk("event_dev: %s, event: %lx\n",
3612 (event_dev ? event_dev->name : "None"),
3615 if (!(event_dev->priv_flags & IFF_BONDING))
3618 if (event_dev->flags & IFF_MASTER) {
3619 dprintk("IFF_MASTER\n");
3620 return bond_master_netdev_event(event, event_dev);
3623 if (event_dev->flags & IFF_SLAVE) {
3624 dprintk("IFF_SLAVE\n");
3625 return bond_slave_netdev_event(event, event_dev);
3632 * bond_inetaddr_event: handle inetaddr notifier chain events.
3634 * We keep track of device IPs primarily to use as source addresses in
3635 * ARP monitor probes (rather than spewing out broadcasts all the time).
3637 * We track one IP for the main device (if it has one), plus one per VLAN.
3639 static int bond_inetaddr_event(struct notifier_block *this, unsigned long event, void *ptr)
3641 struct in_ifaddr *ifa = ptr;
3642 struct net_device *vlan_dev, *event_dev = ifa->ifa_dev->dev;
3643 struct bonding *bond;
3644 struct vlan_entry *vlan;
3646 if (dev_net(ifa->ifa_dev->dev) != &init_net)
3649 list_for_each_entry(bond, &bond_dev_list, bond_list) {
3650 if (bond->dev == event_dev) {
3653 bond->master_ip = ifa->ifa_local;
3656 bond->master_ip = bond_glean_dev_ip(bond->dev);
3663 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
3664 vlan_dev = vlan_group_get_device(bond->vlgrp, vlan->vlan_id);
3665 if (vlan_dev == event_dev) {
3668 vlan->vlan_ip = ifa->ifa_local;
3672 bond_glean_dev_ip(vlan_dev);
3683 static struct notifier_block bond_netdev_notifier = {
3684 .notifier_call = bond_netdev_event,
3687 static struct notifier_block bond_inetaddr_notifier = {
3688 .notifier_call = bond_inetaddr_event,
3691 /*-------------------------- Packet type handling ---------------------------*/
3693 /* register to receive lacpdus on a bond */
3694 static void bond_register_lacpdu(struct bonding *bond)
3696 struct packet_type *pk_type = &(BOND_AD_INFO(bond).ad_pkt_type);
3698 /* initialize packet type */
3699 pk_type->type = PKT_TYPE_LACPDU;
3700 pk_type->dev = bond->dev;
3701 pk_type->func = bond_3ad_lacpdu_recv;
3703 dev_add_pack(pk_type);
3706 /* unregister to receive lacpdus on a bond */
3707 static void bond_unregister_lacpdu(struct bonding *bond)
3709 dev_remove_pack(&(BOND_AD_INFO(bond).ad_pkt_type));
3712 void bond_register_arp(struct bonding *bond)
3714 struct packet_type *pt = &bond->arp_mon_pt;
3719 pt->type = htons(ETH_P_ARP);
3720 pt->dev = bond->dev;
3721 pt->func = bond_arp_rcv;
3725 void bond_unregister_arp(struct bonding *bond)
3727 struct packet_type *pt = &bond->arp_mon_pt;
3729 dev_remove_pack(pt);
3733 /*---------------------------- Hashing Policies -----------------------------*/
3736 * Hash for the output device based upon layer 2 and layer 3 data. If
3737 * the packet is not IP mimic bond_xmit_hash_policy_l2()
3739 static int bond_xmit_hash_policy_l23(struct sk_buff *skb,
3740 struct net_device *bond_dev, int count)
3742 struct ethhdr *data = (struct ethhdr *)skb->data;
3743 struct iphdr *iph = ip_hdr(skb);
3745 if (skb->protocol == htons(ETH_P_IP)) {
3746 return ((ntohl(iph->saddr ^ iph->daddr) & 0xffff) ^
3747 (data->h_dest[5] ^ bond_dev->dev_addr[5])) % count;
3750 return (data->h_dest[5] ^ bond_dev->dev_addr[5]) % count;
3754 * Hash for the output device based upon layer 3 and layer 4 data. If
3755 * the packet is a frag or not TCP or UDP, just use layer 3 data. If it is
3756 * altogether not IP, mimic bond_xmit_hash_policy_l2()
3758 static int bond_xmit_hash_policy_l34(struct sk_buff *skb,
3759 struct net_device *bond_dev, int count)
3761 struct ethhdr *data = (struct ethhdr *)skb->data;
3762 struct iphdr *iph = ip_hdr(skb);
3763 __be16 *layer4hdr = (__be16 *)((u32 *)iph + iph->ihl);
3766 if (skb->protocol == htons(ETH_P_IP)) {
3767 if (!(iph->frag_off & htons(IP_MF|IP_OFFSET)) &&
3768 (iph->protocol == IPPROTO_TCP ||
3769 iph->protocol == IPPROTO_UDP)) {
3770 layer4_xor = ntohs((*layer4hdr ^ *(layer4hdr + 1)));
3772 return (layer4_xor ^
3773 ((ntohl(iph->saddr ^ iph->daddr)) & 0xffff)) % count;
3777 return (data->h_dest[5] ^ bond_dev->dev_addr[5]) % count;
3781 * Hash for the output device based upon layer 2 data
3783 static int bond_xmit_hash_policy_l2(struct sk_buff *skb,
3784 struct net_device *bond_dev, int count)
3786 struct ethhdr *data = (struct ethhdr *)skb->data;
3788 return (data->h_dest[5] ^ bond_dev->dev_addr[5]) % count;
3791 /*-------------------------- Device entry points ----------------------------*/
3793 static int bond_open(struct net_device *bond_dev)
3795 struct bonding *bond = netdev_priv(bond_dev);
3797 bond->kill_timers = 0;
3799 if ((bond->params.mode == BOND_MODE_TLB) ||
3800 (bond->params.mode == BOND_MODE_ALB)) {
3801 /* bond_alb_initialize must be called before the timer
3804 if (bond_alb_initialize(bond, (bond->params.mode == BOND_MODE_ALB))) {
3805 /* something went wrong - fail the open operation */
3809 INIT_DELAYED_WORK(&bond->alb_work, bond_alb_monitor);
3810 queue_delayed_work(bond->wq, &bond->alb_work, 0);
3813 if (bond->params.miimon) { /* link check interval, in milliseconds. */
3814 INIT_DELAYED_WORK(&bond->mii_work, bond_mii_monitor);
3815 queue_delayed_work(bond->wq, &bond->mii_work, 0);
3818 if (bond->params.arp_interval) { /* arp interval, in milliseconds. */
3819 if (bond->params.mode == BOND_MODE_ACTIVEBACKUP)
3820 INIT_DELAYED_WORK(&bond->arp_work,
3821 bond_activebackup_arp_mon);
3823 INIT_DELAYED_WORK(&bond->arp_work,
3824 bond_loadbalance_arp_mon);
3826 queue_delayed_work(bond->wq, &bond->arp_work, 0);
3827 if (bond->params.arp_validate)
3828 bond_register_arp(bond);
3831 if (bond->params.mode == BOND_MODE_8023AD) {
3832 INIT_DELAYED_WORK(&bond->ad_work, bond_3ad_state_machine_handler);
3833 queue_delayed_work(bond->wq, &bond->ad_work, 0);
3834 /* register to receive LACPDUs */
3835 bond_register_lacpdu(bond);
3836 bond_3ad_initiate_agg_selection(bond, 1);
3842 static int bond_close(struct net_device *bond_dev)
3844 struct bonding *bond = netdev_priv(bond_dev);
3846 if (bond->params.mode == BOND_MODE_8023AD) {
3847 /* Unregister the receive of LACPDUs */
3848 bond_unregister_lacpdu(bond);
3851 if (bond->params.arp_validate)
3852 bond_unregister_arp(bond);
3854 write_lock_bh(&bond->lock);
3856 bond->send_grat_arp = 0;
3857 bond->send_unsol_na = 0;
3859 /* signal timers not to re-arm */
3860 bond->kill_timers = 1;
3862 write_unlock_bh(&bond->lock);
3864 if (bond->params.miimon) { /* link check interval, in milliseconds. */
3865 cancel_delayed_work(&bond->mii_work);
3868 if (bond->params.arp_interval) { /* arp interval, in milliseconds. */
3869 cancel_delayed_work(&bond->arp_work);
3872 switch (bond->params.mode) {
3873 case BOND_MODE_8023AD:
3874 cancel_delayed_work(&bond->ad_work);
3878 cancel_delayed_work(&bond->alb_work);
3885 if ((bond->params.mode == BOND_MODE_TLB) ||
3886 (bond->params.mode == BOND_MODE_ALB)) {
3887 /* Must be called only after all
3888 * slaves have been released
3890 bond_alb_deinitialize(bond);
3896 static struct net_device_stats *bond_get_stats(struct net_device *bond_dev)
3898 struct bonding *bond = netdev_priv(bond_dev);
3899 struct net_device_stats *stats = &bond->stats;
3900 struct net_device_stats local_stats;
3901 struct slave *slave;
3904 memset(&local_stats, 0, sizeof(struct net_device_stats));
3906 read_lock_bh(&bond->lock);
3908 bond_for_each_slave(bond, slave, i) {
3909 const struct net_device_stats *sstats = dev_get_stats(slave->dev);
3911 local_stats.rx_packets += sstats->rx_packets;
3912 local_stats.rx_bytes += sstats->rx_bytes;
3913 local_stats.rx_errors += sstats->rx_errors;
3914 local_stats.rx_dropped += sstats->rx_dropped;
3916 local_stats.tx_packets += sstats->tx_packets;
3917 local_stats.tx_bytes += sstats->tx_bytes;
3918 local_stats.tx_errors += sstats->tx_errors;
3919 local_stats.tx_dropped += sstats->tx_dropped;
3921 local_stats.multicast += sstats->multicast;
3922 local_stats.collisions += sstats->collisions;
3924 local_stats.rx_length_errors += sstats->rx_length_errors;
3925 local_stats.rx_over_errors += sstats->rx_over_errors;
3926 local_stats.rx_crc_errors += sstats->rx_crc_errors;
3927 local_stats.rx_frame_errors += sstats->rx_frame_errors;
3928 local_stats.rx_fifo_errors += sstats->rx_fifo_errors;
3929 local_stats.rx_missed_errors += sstats->rx_missed_errors;
3931 local_stats.tx_aborted_errors += sstats->tx_aborted_errors;
3932 local_stats.tx_carrier_errors += sstats->tx_carrier_errors;
3933 local_stats.tx_fifo_errors += sstats->tx_fifo_errors;
3934 local_stats.tx_heartbeat_errors += sstats->tx_heartbeat_errors;
3935 local_stats.tx_window_errors += sstats->tx_window_errors;
3938 memcpy(stats, &local_stats, sizeof(struct net_device_stats));
3940 read_unlock_bh(&bond->lock);
3945 static int bond_do_ioctl(struct net_device *bond_dev, struct ifreq *ifr, int cmd)
3947 struct net_device *slave_dev = NULL;
3948 struct ifbond k_binfo;
3949 struct ifbond __user *u_binfo = NULL;
3950 struct ifslave k_sinfo;
3951 struct ifslave __user *u_sinfo = NULL;
3952 struct mii_ioctl_data *mii = NULL;
3955 dprintk("bond_ioctl: master=%s, cmd=%d\n",
3956 bond_dev->name, cmd);
3968 * We do this again just in case we were called by SIOCGMIIREG
3969 * instead of SIOCGMIIPHY.
3976 if (mii->reg_num == 1) {
3977 struct bonding *bond = netdev_priv(bond_dev);
3979 read_lock(&bond->lock);
3980 read_lock(&bond->curr_slave_lock);
3981 if (netif_carrier_ok(bond->dev)) {
3982 mii->val_out = BMSR_LSTATUS;
3984 read_unlock(&bond->curr_slave_lock);
3985 read_unlock(&bond->lock);
3989 case BOND_INFO_QUERY_OLD:
3990 case SIOCBONDINFOQUERY:
3991 u_binfo = (struct ifbond __user *)ifr->ifr_data;
3993 if (copy_from_user(&k_binfo, u_binfo, sizeof(ifbond))) {
3997 res = bond_info_query(bond_dev, &k_binfo);
3999 if (copy_to_user(u_binfo, &k_binfo, sizeof(ifbond))) {
4005 case BOND_SLAVE_INFO_QUERY_OLD:
4006 case SIOCBONDSLAVEINFOQUERY:
4007 u_sinfo = (struct ifslave __user *)ifr->ifr_data;
4009 if (copy_from_user(&k_sinfo, u_sinfo, sizeof(ifslave))) {
4013 res = bond_slave_info_query(bond_dev, &k_sinfo);
4015 if (copy_to_user(u_sinfo, &k_sinfo, sizeof(ifslave))) {
4026 if (!capable(CAP_NET_ADMIN)) {
4030 down_write(&(bonding_rwsem));
4031 slave_dev = dev_get_by_name(&init_net, ifr->ifr_slave);
4033 dprintk("slave_dev=%p: \n", slave_dev);
4038 dprintk("slave_dev->name=%s: \n", slave_dev->name);
4040 case BOND_ENSLAVE_OLD:
4041 case SIOCBONDENSLAVE:
4042 res = bond_enslave(bond_dev, slave_dev);
4044 case BOND_RELEASE_OLD:
4045 case SIOCBONDRELEASE:
4046 res = bond_release(bond_dev, slave_dev);
4048 case BOND_SETHWADDR_OLD:
4049 case SIOCBONDSETHWADDR:
4050 res = bond_sethwaddr(bond_dev, slave_dev);
4052 case BOND_CHANGE_ACTIVE_OLD:
4053 case SIOCBONDCHANGEACTIVE:
4054 res = bond_ioctl_change_active(bond_dev, slave_dev);
4063 up_write(&(bonding_rwsem));
4067 static void bond_set_multicast_list(struct net_device *bond_dev)
4069 struct bonding *bond = netdev_priv(bond_dev);
4070 struct dev_mc_list *dmi;
4073 * Do promisc before checking multicast_mode
4075 if ((bond_dev->flags & IFF_PROMISC) && !(bond->flags & IFF_PROMISC)) {
4077 * FIXME: Need to handle the error when one of the multi-slaves
4080 bond_set_promiscuity(bond, 1);
4083 if (!(bond_dev->flags & IFF_PROMISC) && (bond->flags & IFF_PROMISC)) {
4084 bond_set_promiscuity(bond, -1);
4087 /* set allmulti flag to slaves */
4088 if ((bond_dev->flags & IFF_ALLMULTI) && !(bond->flags & IFF_ALLMULTI)) {
4090 * FIXME: Need to handle the error when one of the multi-slaves
4093 bond_set_allmulti(bond, 1);
4096 if (!(bond_dev->flags & IFF_ALLMULTI) && (bond->flags & IFF_ALLMULTI)) {
4097 bond_set_allmulti(bond, -1);
4100 read_lock(&bond->lock);
4102 bond->flags = bond_dev->flags;
4104 /* looking for addresses to add to slaves' mc list */
4105 for (dmi = bond_dev->mc_list; dmi; dmi = dmi->next) {
4106 if (!bond_mc_list_find_dmi(dmi, bond->mc_list)) {
4107 bond_mc_add(bond, dmi->dmi_addr, dmi->dmi_addrlen);
4111 /* looking for addresses to delete from slaves' list */
4112 for (dmi = bond->mc_list; dmi; dmi = dmi->next) {
4113 if (!bond_mc_list_find_dmi(dmi, bond_dev->mc_list)) {
4114 bond_mc_delete(bond, dmi->dmi_addr, dmi->dmi_addrlen);
4118 /* save master's multicast list */
4119 bond_mc_list_destroy(bond);
4120 bond_mc_list_copy(bond_dev->mc_list, bond, GFP_ATOMIC);
4122 read_unlock(&bond->lock);
4125 static int bond_neigh_setup(struct net_device *dev, struct neigh_parms *parms)
4127 struct bonding *bond = netdev_priv(dev);
4128 struct slave *slave = bond->first_slave;
4131 const struct net_device_ops *slave_ops
4132 = slave->dev->netdev_ops;
4133 if (slave_ops->ndo_neigh_setup)
4134 return slave_ops->ndo_neigh_setup(dev, parms);
4140 * Change the MTU of all of a master's slaves to match the master
4142 static int bond_change_mtu(struct net_device *bond_dev, int new_mtu)
4144 struct bonding *bond = netdev_priv(bond_dev);
4145 struct slave *slave, *stop_at;
4149 dprintk("bond=%p, name=%s, new_mtu=%d\n", bond,
4150 (bond_dev ? bond_dev->name : "None"), new_mtu);
4152 /* Can't hold bond->lock with bh disabled here since
4153 * some base drivers panic. On the other hand we can't
4154 * hold bond->lock without bh disabled because we'll
4155 * deadlock. The only solution is to rely on the fact
4156 * that we're under rtnl_lock here, and the slaves
4157 * list won't change. This doesn't solve the problem
4158 * of setting the slave's MTU while it is
4159 * transmitting, but the assumption is that the base
4160 * driver can handle that.
4162 * TODO: figure out a way to safely iterate the slaves
4163 * list, but without holding a lock around the actual
4164 * call to the base driver.
4167 bond_for_each_slave(bond, slave, i) {
4168 dprintk("s %p s->p %p c_m %p\n", slave,
4169 slave->prev, slave->dev->change_mtu);
4171 res = dev_set_mtu(slave->dev, new_mtu);
4174 /* If we failed to set the slave's mtu to the new value
4175 * we must abort the operation even in ACTIVE_BACKUP
4176 * mode, because if we allow the backup slaves to have
4177 * different mtu values than the active slave we'll
4178 * need to change their mtu when doing a failover. That
4179 * means changing their mtu from timer context, which
4180 * is probably not a good idea.
4182 dprintk("err %d %s\n", res, slave->dev->name);
4187 bond_dev->mtu = new_mtu;
4192 /* unwind from head to the slave that failed */
4194 bond_for_each_slave_from_to(bond, slave, i, bond->first_slave, stop_at) {
4197 tmp_res = dev_set_mtu(slave->dev, bond_dev->mtu);
4199 dprintk("unwind err %d dev %s\n", tmp_res,
4210 * Note that many devices must be down to change the HW address, and
4211 * downing the master releases all slaves. We can make bonds full of
4212 * bonding devices to test this, however.
4214 static int bond_set_mac_address(struct net_device *bond_dev, void *addr)
4216 struct bonding *bond = netdev_priv(bond_dev);
4217 struct sockaddr *sa = addr, tmp_sa;
4218 struct slave *slave, *stop_at;
4222 if (bond->params.mode == BOND_MODE_ALB)
4223 return bond_alb_set_mac_address(bond_dev, addr);
4226 dprintk("bond=%p, name=%s\n", bond, (bond_dev ? bond_dev->name : "None"));
4229 * If fail_over_mac is set to active, do nothing and return
4230 * success. Returning an error causes ifenslave to fail.
4232 if (bond->params.fail_over_mac == BOND_FOM_ACTIVE)
4235 if (!is_valid_ether_addr(sa->sa_data)) {
4236 return -EADDRNOTAVAIL;
4239 /* Can't hold bond->lock with bh disabled here since
4240 * some base drivers panic. On the other hand we can't
4241 * hold bond->lock without bh disabled because we'll
4242 * deadlock. The only solution is to rely on the fact
4243 * that we're under rtnl_lock here, and the slaves
4244 * list won't change. This doesn't solve the problem
4245 * of setting the slave's hw address while it is
4246 * transmitting, but the assumption is that the base
4247 * driver can handle that.
4249 * TODO: figure out a way to safely iterate the slaves
4250 * list, but without holding a lock around the actual
4251 * call to the base driver.
4254 bond_for_each_slave(bond, slave, i) {
4255 const struct net_device_ops *slave_ops = slave->dev->netdev_ops;
4256 dprintk("slave %p %s\n", slave, slave->dev->name);
4258 if (slave_ops->ndo_set_mac_address == NULL) {
4260 dprintk("EOPNOTSUPP %s\n", slave->dev->name);
4264 res = dev_set_mac_address(slave->dev, addr);
4266 /* TODO: consider downing the slave
4268 * User should expect communications
4269 * breakage anyway until ARP finish
4272 dprintk("err %d %s\n", res, slave->dev->name);
4278 memcpy(bond_dev->dev_addr, sa->sa_data, bond_dev->addr_len);
4282 memcpy(tmp_sa.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
4283 tmp_sa.sa_family = bond_dev->type;
4285 /* unwind from head to the slave that failed */
4287 bond_for_each_slave_from_to(bond, slave, i, bond->first_slave, stop_at) {
4290 tmp_res = dev_set_mac_address(slave->dev, &tmp_sa);
4292 dprintk("unwind err %d dev %s\n", tmp_res,
4300 static int bond_xmit_roundrobin(struct sk_buff *skb, struct net_device *bond_dev)
4302 struct bonding *bond = netdev_priv(bond_dev);
4303 struct slave *slave, *start_at;
4304 int i, slave_no, res = 1;
4306 read_lock(&bond->lock);
4308 if (!BOND_IS_OK(bond)) {
4313 * Concurrent TX may collide on rr_tx_counter; we accept that
4314 * as being rare enough not to justify using an atomic op here
4316 slave_no = bond->rr_tx_counter++ % bond->slave_cnt;
4318 bond_for_each_slave(bond, slave, i) {
4326 bond_for_each_slave_from(bond, slave, i, start_at) {
4327 if (IS_UP(slave->dev) &&
4328 (slave->link == BOND_LINK_UP) &&
4329 (slave->state == BOND_STATE_ACTIVE)) {
4330 res = bond_dev_queue_xmit(bond, skb, slave->dev);
4337 /* no suitable interface, frame not sent */
4340 read_unlock(&bond->lock);
4346 * in active-backup mode, we know that bond->curr_active_slave is always valid if
4347 * the bond has a usable interface.
4349 static int bond_xmit_activebackup(struct sk_buff *skb, struct net_device *bond_dev)
4351 struct bonding *bond = netdev_priv(bond_dev);
4354 read_lock(&bond->lock);
4355 read_lock(&bond->curr_slave_lock);
4357 if (!BOND_IS_OK(bond)) {
4361 if (!bond->curr_active_slave)
4364 res = bond_dev_queue_xmit(bond, skb, bond->curr_active_slave->dev);
4368 /* no suitable interface, frame not sent */
4371 read_unlock(&bond->curr_slave_lock);
4372 read_unlock(&bond->lock);
4377 * In bond_xmit_xor() , we determine the output device by using a pre-
4378 * determined xmit_hash_policy(), If the selected device is not enabled,
4379 * find the next active slave.
4381 static int bond_xmit_xor(struct sk_buff *skb, struct net_device *bond_dev)
4383 struct bonding *bond = netdev_priv(bond_dev);
4384 struct slave *slave, *start_at;
4389 read_lock(&bond->lock);
4391 if (!BOND_IS_OK(bond)) {
4395 slave_no = bond->xmit_hash_policy(skb, bond_dev, bond->slave_cnt);
4397 bond_for_each_slave(bond, slave, i) {
4406 bond_for_each_slave_from(bond, slave, i, start_at) {
4407 if (IS_UP(slave->dev) &&
4408 (slave->link == BOND_LINK_UP) &&
4409 (slave->state == BOND_STATE_ACTIVE)) {
4410 res = bond_dev_queue_xmit(bond, skb, slave->dev);
4417 /* no suitable interface, frame not sent */
4420 read_unlock(&bond->lock);
4425 * in broadcast mode, we send everything to all usable interfaces.
4427 static int bond_xmit_broadcast(struct sk_buff *skb, struct net_device *bond_dev)
4429 struct bonding *bond = netdev_priv(bond_dev);
4430 struct slave *slave, *start_at;
4431 struct net_device *tx_dev = NULL;
4435 read_lock(&bond->lock);
4437 if (!BOND_IS_OK(bond)) {
4441 read_lock(&bond->curr_slave_lock);
4442 start_at = bond->curr_active_slave;
4443 read_unlock(&bond->curr_slave_lock);
4449 bond_for_each_slave_from(bond, slave, i, start_at) {
4450 if (IS_UP(slave->dev) &&
4451 (slave->link == BOND_LINK_UP) &&
4452 (slave->state == BOND_STATE_ACTIVE)) {
4454 struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
4456 printk(KERN_ERR DRV_NAME
4457 ": %s: Error: bond_xmit_broadcast(): "
4458 "skb_clone() failed\n",
4463 res = bond_dev_queue_xmit(bond, skb2, tx_dev);
4465 dev_kfree_skb(skb2);
4469 tx_dev = slave->dev;
4474 res = bond_dev_queue_xmit(bond, skb, tx_dev);
4479 /* no suitable interface, frame not sent */
4482 /* frame sent to all suitable interfaces */
4483 read_unlock(&bond->lock);
4487 /*------------------------- Device initialization ---------------------------*/
4489 static void bond_set_xmit_hash_policy(struct bonding *bond)
4491 switch (bond->params.xmit_policy) {
4492 case BOND_XMIT_POLICY_LAYER23:
4493 bond->xmit_hash_policy = bond_xmit_hash_policy_l23;
4495 case BOND_XMIT_POLICY_LAYER34:
4496 bond->xmit_hash_policy = bond_xmit_hash_policy_l34;
4498 case BOND_XMIT_POLICY_LAYER2:
4500 bond->xmit_hash_policy = bond_xmit_hash_policy_l2;
4505 static int bond_start_xmit(struct sk_buff *skb, struct net_device *dev)
4507 const struct bonding *bond = netdev_priv(dev);
4509 switch (bond->params.mode) {
4510 case BOND_MODE_ROUNDROBIN:
4511 return bond_xmit_roundrobin(skb, dev);
4512 case BOND_MODE_ACTIVEBACKUP:
4513 return bond_xmit_activebackup(skb, dev);
4515 return bond_xmit_xor(skb, dev);
4516 case BOND_MODE_BROADCAST:
4517 return bond_xmit_broadcast(skb, dev);
4518 case BOND_MODE_8023AD:
4519 return bond_3ad_xmit_xor(skb, dev);
4522 return bond_alb_xmit(skb, dev);
4524 /* Should never happen, mode already checked */
4525 printk(KERN_ERR DRV_NAME ": %s: Error: Unknown bonding mode %d\n",
4526 dev->name, bond->params.mode);
4529 return NETDEV_TX_OK;
4535 * set bond mode specific net device operations
4537 void bond_set_mode_ops(struct bonding *bond, int mode)
4539 struct net_device *bond_dev = bond->dev;
4542 case BOND_MODE_ROUNDROBIN:
4544 case BOND_MODE_ACTIVEBACKUP:
4547 bond_set_xmit_hash_policy(bond);
4549 case BOND_MODE_BROADCAST:
4551 case BOND_MODE_8023AD:
4552 bond_set_master_3ad_flags(bond);
4553 bond_set_xmit_hash_policy(bond);
4556 bond_set_master_alb_flags(bond);
4561 /* Should never happen, mode already checked */
4562 printk(KERN_ERR DRV_NAME
4563 ": %s: Error: Unknown bonding mode %d\n",
4570 static void bond_ethtool_get_drvinfo(struct net_device *bond_dev,
4571 struct ethtool_drvinfo *drvinfo)
4573 strncpy(drvinfo->driver, DRV_NAME, 32);
4574 strncpy(drvinfo->version, DRV_VERSION, 32);
4575 snprintf(drvinfo->fw_version, 32, "%d", BOND_ABI_VERSION);
4578 static const struct ethtool_ops bond_ethtool_ops = {
4579 .get_drvinfo = bond_ethtool_get_drvinfo,
4580 .get_link = ethtool_op_get_link,
4581 .get_tx_csum = ethtool_op_get_tx_csum,
4582 .get_sg = ethtool_op_get_sg,
4583 .get_tso = ethtool_op_get_tso,
4584 .get_ufo = ethtool_op_get_ufo,
4585 .get_flags = ethtool_op_get_flags,
4588 static const struct net_device_ops bond_netdev_ops = {
4589 .ndo_open = bond_open,
4590 .ndo_stop = bond_close,
4591 .ndo_start_xmit = bond_start_xmit,
4592 .ndo_get_stats = bond_get_stats,
4593 .ndo_do_ioctl = bond_do_ioctl,
4594 .ndo_set_multicast_list = bond_set_multicast_list,
4595 .ndo_change_mtu = bond_change_mtu,
4596 .ndo_set_mac_address = bond_set_mac_address,
4597 .ndo_neigh_setup = bond_neigh_setup,
4598 .ndo_vlan_rx_register = bond_vlan_rx_register,
4599 .ndo_vlan_rx_add_vid = bond_vlan_rx_add_vid,
4600 .ndo_vlan_rx_kill_vid = bond_vlan_rx_kill_vid,
4604 * Does not allocate but creates a /proc entry.
4607 static int bond_init(struct net_device *bond_dev, struct bond_params *params)
4609 struct bonding *bond = netdev_priv(bond_dev);
4611 dprintk("Begin bond_init for %s\n", bond_dev->name);
4613 /* initialize rwlocks */
4614 rwlock_init(&bond->lock);
4615 rwlock_init(&bond->curr_slave_lock);
4617 bond->params = *params; /* copy params struct */
4619 bond->wq = create_singlethread_workqueue(bond_dev->name);
4623 /* Initialize pointers */
4624 bond->first_slave = NULL;
4625 bond->curr_active_slave = NULL;
4626 bond->current_arp_slave = NULL;
4627 bond->primary_slave = NULL;
4628 bond->dev = bond_dev;
4629 bond->send_grat_arp = 0;
4630 bond->send_unsol_na = 0;
4631 bond->setup_by_slave = 0;
4632 INIT_LIST_HEAD(&bond->vlan_list);
4634 /* Initialize the device entry points */
4635 bond_dev->netdev_ops = &bond_netdev_ops;
4636 bond_dev->ethtool_ops = &bond_ethtool_ops;
4637 bond_set_mode_ops(bond, bond->params.mode);
4639 bond_dev->destructor = bond_destructor;
4641 /* Initialize the device options */
4642 bond_dev->tx_queue_len = 0;
4643 bond_dev->flags |= IFF_MASTER|IFF_MULTICAST;
4644 bond_dev->priv_flags |= IFF_BONDING;
4645 if (bond->params.arp_interval)
4646 bond_dev->priv_flags |= IFF_MASTER_ARPMON;
4648 /* At first, we block adding VLANs. That's the only way to
4649 * prevent problems that occur when adding VLANs over an
4650 * empty bond. The block will be removed once non-challenged
4651 * slaves are enslaved.
4653 bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
4655 /* don't acquire bond device's netif_tx_lock when
4657 bond_dev->features |= NETIF_F_LLTX;
4659 /* By default, we declare the bond to be fully
4660 * VLAN hardware accelerated capable. Special
4661 * care is taken in the various xmit functions
4662 * when there are slaves that are not hw accel
4665 bond_dev->features |= (NETIF_F_HW_VLAN_TX |
4666 NETIF_F_HW_VLAN_RX |
4667 NETIF_F_HW_VLAN_FILTER);
4669 #ifdef CONFIG_PROC_FS
4670 bond_create_proc_entry(bond);
4672 list_add_tail(&bond->bond_list, &bond_dev_list);
4677 static void bond_work_cancel_all(struct bonding *bond)
4679 write_lock_bh(&bond->lock);
4680 bond->kill_timers = 1;
4681 write_unlock_bh(&bond->lock);
4683 if (bond->params.miimon && delayed_work_pending(&bond->mii_work))
4684 cancel_delayed_work(&bond->mii_work);
4686 if (bond->params.arp_interval && delayed_work_pending(&bond->arp_work))
4687 cancel_delayed_work(&bond->arp_work);
4689 if (bond->params.mode == BOND_MODE_ALB &&
4690 delayed_work_pending(&bond->alb_work))
4691 cancel_delayed_work(&bond->alb_work);
4693 if (bond->params.mode == BOND_MODE_8023AD &&
4694 delayed_work_pending(&bond->ad_work))
4695 cancel_delayed_work(&bond->ad_work);
4698 /* De-initialize device specific data.
4699 * Caller must hold rtnl_lock.
4701 static void bond_deinit(struct net_device *bond_dev)
4703 struct bonding *bond = netdev_priv(bond_dev);
4705 list_del(&bond->bond_list);
4707 bond_work_cancel_all(bond);
4709 #ifdef CONFIG_PROC_FS
4710 bond_remove_proc_entry(bond);
4714 /* Unregister and free all bond devices.
4715 * Caller must hold rtnl_lock.
4717 static void bond_free_all(void)
4719 struct bonding *bond, *nxt;
4721 list_for_each_entry_safe(bond, nxt, &bond_dev_list, bond_list) {
4722 struct net_device *bond_dev = bond->dev;
4724 bond_work_cancel_all(bond);
4725 /* Release the bonded slaves */
4726 bond_release_all(bond_dev);
4730 #ifdef CONFIG_PROC_FS
4731 bond_destroy_proc_dir();
4735 /*------------------------- Module initialization ---------------------------*/
4738 * Convert string input module parms. Accept either the
4739 * number of the mode or its string name. A bit complicated because
4740 * some mode names are substrings of other names, and calls from sysfs
4741 * may have whitespace in the name (trailing newlines, for example).
4743 int bond_parse_parm(const char *buf, struct bond_parm_tbl *tbl)
4745 int mode = -1, i, rv;
4746 char *p, modestr[BOND_MAX_MODENAME_LEN + 1] = { 0, };
4748 for (p = (char *)buf; *p; p++)
4749 if (!(isdigit(*p) || isspace(*p)))
4753 rv = sscanf(buf, "%20s", modestr);
4755 rv = sscanf(buf, "%d", &mode);
4760 for (i = 0; tbl[i].modename; i++) {
4761 if (mode == tbl[i].mode)
4763 if (strcmp(modestr, tbl[i].modename) == 0)
4770 static int bond_check_params(struct bond_params *params)
4772 int arp_validate_value, fail_over_mac_value;
4775 * Convert string parameters.
4778 bond_mode = bond_parse_parm(mode, bond_mode_tbl);
4779 if (bond_mode == -1) {
4780 printk(KERN_ERR DRV_NAME
4781 ": Error: Invalid bonding mode \"%s\"\n",
4782 mode == NULL ? "NULL" : mode);
4787 if (xmit_hash_policy) {
4788 if ((bond_mode != BOND_MODE_XOR) &&
4789 (bond_mode != BOND_MODE_8023AD)) {
4790 printk(KERN_INFO DRV_NAME
4791 ": xor_mode param is irrelevant in mode %s\n",
4792 bond_mode_name(bond_mode));
4794 xmit_hashtype = bond_parse_parm(xmit_hash_policy,
4796 if (xmit_hashtype == -1) {
4797 printk(KERN_ERR DRV_NAME
4798 ": Error: Invalid xmit_hash_policy \"%s\"\n",
4799 xmit_hash_policy == NULL ? "NULL" :
4807 if (bond_mode != BOND_MODE_8023AD) {
4808 printk(KERN_INFO DRV_NAME
4809 ": lacp_rate param is irrelevant in mode %s\n",
4810 bond_mode_name(bond_mode));
4812 lacp_fast = bond_parse_parm(lacp_rate, bond_lacp_tbl);
4813 if (lacp_fast == -1) {
4814 printk(KERN_ERR DRV_NAME
4815 ": Error: Invalid lacp rate \"%s\"\n",
4816 lacp_rate == NULL ? "NULL" : lacp_rate);
4823 params->ad_select = bond_parse_parm(ad_select, ad_select_tbl);
4824 if (params->ad_select == -1) {
4825 printk(KERN_ERR DRV_NAME
4826 ": Error: Invalid ad_select \"%s\"\n",
4827 ad_select == NULL ? "NULL" : ad_select);
4831 if (bond_mode != BOND_MODE_8023AD) {
4832 printk(KERN_WARNING DRV_NAME
4833 ": ad_select param only affects 802.3ad mode\n");
4836 params->ad_select = BOND_AD_STABLE;
4839 if (max_bonds < 0 || max_bonds > INT_MAX) {
4840 printk(KERN_WARNING DRV_NAME
4841 ": Warning: max_bonds (%d) not in range %d-%d, so it "
4842 "was reset to BOND_DEFAULT_MAX_BONDS (%d)\n",
4843 max_bonds, 0, INT_MAX, BOND_DEFAULT_MAX_BONDS);
4844 max_bonds = BOND_DEFAULT_MAX_BONDS;
4848 printk(KERN_WARNING DRV_NAME
4849 ": Warning: miimon module parameter (%d), "
4850 "not in range 0-%d, so it was reset to %d\n",
4851 miimon, INT_MAX, BOND_LINK_MON_INTERV);
4852 miimon = BOND_LINK_MON_INTERV;
4856 printk(KERN_WARNING DRV_NAME
4857 ": Warning: updelay module parameter (%d), "
4858 "not in range 0-%d, so it was reset to 0\n",
4863 if (downdelay < 0) {
4864 printk(KERN_WARNING DRV_NAME
4865 ": Warning: downdelay module parameter (%d), "
4866 "not in range 0-%d, so it was reset to 0\n",
4867 downdelay, INT_MAX);
4871 if ((use_carrier != 0) && (use_carrier != 1)) {
4872 printk(KERN_WARNING DRV_NAME
4873 ": Warning: use_carrier module parameter (%d), "
4874 "not of valid value (0/1), so it was set to 1\n",
4879 if (num_grat_arp < 0 || num_grat_arp > 255) {
4880 printk(KERN_WARNING DRV_NAME
4881 ": Warning: num_grat_arp (%d) not in range 0-255 so it "
4882 "was reset to 1 \n", num_grat_arp);
4886 if (num_unsol_na < 0 || num_unsol_na > 255) {
4887 printk(KERN_WARNING DRV_NAME
4888 ": Warning: num_unsol_na (%d) not in range 0-255 so it "
4889 "was reset to 1 \n", num_unsol_na);
4893 /* reset values for 802.3ad */
4894 if (bond_mode == BOND_MODE_8023AD) {
4896 printk(KERN_WARNING DRV_NAME
4897 ": Warning: miimon must be specified, "
4898 "otherwise bonding will not detect link "
4899 "failure, speed and duplex which are "
4900 "essential for 802.3ad operation\n");
4901 printk(KERN_WARNING "Forcing miimon to 100msec\n");
4906 /* reset values for TLB/ALB */
4907 if ((bond_mode == BOND_MODE_TLB) ||
4908 (bond_mode == BOND_MODE_ALB)) {
4910 printk(KERN_WARNING DRV_NAME
4911 ": Warning: miimon must be specified, "
4912 "otherwise bonding will not detect link "
4913 "failure and link speed which are essential "
4914 "for TLB/ALB load balancing\n");
4915 printk(KERN_WARNING "Forcing miimon to 100msec\n");
4920 if (bond_mode == BOND_MODE_ALB) {
4921 printk(KERN_NOTICE DRV_NAME
4922 ": In ALB mode you might experience client "
4923 "disconnections upon reconnection of a link if the "
4924 "bonding module updelay parameter (%d msec) is "
4925 "incompatible with the forwarding delay time of the "
4931 if (updelay || downdelay) {
4932 /* just warn the user the up/down delay will have
4933 * no effect since miimon is zero...
4935 printk(KERN_WARNING DRV_NAME
4936 ": Warning: miimon module parameter not set "
4937 "and updelay (%d) or downdelay (%d) module "
4938 "parameter is set; updelay and downdelay have "
4939 "no effect unless miimon is set\n",
4940 updelay, downdelay);
4943 /* don't allow arp monitoring */
4945 printk(KERN_WARNING DRV_NAME
4946 ": Warning: miimon (%d) and arp_interval (%d) "
4947 "can't be used simultaneously, disabling ARP "
4949 miimon, arp_interval);
4953 if ((updelay % miimon) != 0) {
4954 printk(KERN_WARNING DRV_NAME
4955 ": Warning: updelay (%d) is not a multiple "
4956 "of miimon (%d), updelay rounded to %d ms\n",
4957 updelay, miimon, (updelay / miimon) * miimon);
4962 if ((downdelay % miimon) != 0) {
4963 printk(KERN_WARNING DRV_NAME
4964 ": Warning: downdelay (%d) is not a multiple "
4965 "of miimon (%d), downdelay rounded to %d ms\n",
4967 (downdelay / miimon) * miimon);
4970 downdelay /= miimon;
4973 if (arp_interval < 0) {
4974 printk(KERN_WARNING DRV_NAME
4975 ": Warning: arp_interval module parameter (%d) "
4976 ", not in range 0-%d, so it was reset to %d\n",
4977 arp_interval, INT_MAX, BOND_LINK_ARP_INTERV);
4978 arp_interval = BOND_LINK_ARP_INTERV;
4981 for (arp_ip_count = 0;
4982 (arp_ip_count < BOND_MAX_ARP_TARGETS) && arp_ip_target[arp_ip_count];
4984 /* not complete check, but should be good enough to
4986 if (!isdigit(arp_ip_target[arp_ip_count][0])) {
4987 printk(KERN_WARNING DRV_NAME
4988 ": Warning: bad arp_ip_target module parameter "
4989 "(%s), ARP monitoring will not be performed\n",
4990 arp_ip_target[arp_ip_count]);
4993 __be32 ip = in_aton(arp_ip_target[arp_ip_count]);
4994 arp_target[arp_ip_count] = ip;
4998 if (arp_interval && !arp_ip_count) {
4999 /* don't allow arping if no arp_ip_target given... */
5000 printk(KERN_WARNING DRV_NAME
5001 ": Warning: arp_interval module parameter (%d) "
5002 "specified without providing an arp_ip_target "
5003 "parameter, arp_interval was reset to 0\n",
5009 if (bond_mode != BOND_MODE_ACTIVEBACKUP) {
5010 printk(KERN_ERR DRV_NAME
5011 ": arp_validate only supported in active-backup mode\n");
5014 if (!arp_interval) {
5015 printk(KERN_ERR DRV_NAME
5016 ": arp_validate requires arp_interval\n");
5020 arp_validate_value = bond_parse_parm(arp_validate,
5022 if (arp_validate_value == -1) {
5023 printk(KERN_ERR DRV_NAME
5024 ": Error: invalid arp_validate \"%s\"\n",
5025 arp_validate == NULL ? "NULL" : arp_validate);
5029 arp_validate_value = 0;
5032 printk(KERN_INFO DRV_NAME
5033 ": MII link monitoring set to %d ms\n",
5035 } else if (arp_interval) {
5038 printk(KERN_INFO DRV_NAME
5039 ": ARP monitoring set to %d ms, validate %s, with %d target(s):",
5041 arp_validate_tbl[arp_validate_value].modename,
5044 for (i = 0; i < arp_ip_count; i++)
5045 printk (" %s", arp_ip_target[i]);
5049 } else if (max_bonds) {
5050 /* miimon and arp_interval not set, we need one so things
5051 * work as expected, see bonding.txt for details
5053 printk(KERN_WARNING DRV_NAME
5054 ": Warning: either miimon or arp_interval and "
5055 "arp_ip_target module parameters must be specified, "
5056 "otherwise bonding will not detect link failures! see "
5057 "bonding.txt for details.\n");
5060 if (primary && !USES_PRIMARY(bond_mode)) {
5061 /* currently, using a primary only makes sense
5062 * in active backup, TLB or ALB modes
5064 printk(KERN_WARNING DRV_NAME
5065 ": Warning: %s primary device specified but has no "
5066 "effect in %s mode\n",
5067 primary, bond_mode_name(bond_mode));
5071 if (fail_over_mac) {
5072 fail_over_mac_value = bond_parse_parm(fail_over_mac,
5074 if (fail_over_mac_value == -1) {
5075 printk(KERN_ERR DRV_NAME
5076 ": Error: invalid fail_over_mac \"%s\"\n",
5077 arp_validate == NULL ? "NULL" : arp_validate);
5081 if (bond_mode != BOND_MODE_ACTIVEBACKUP)
5082 printk(KERN_WARNING DRV_NAME
5083 ": Warning: fail_over_mac only affects "
5084 "active-backup mode.\n");
5086 fail_over_mac_value = BOND_FOM_NONE;
5089 /* fill params struct with the proper values */
5090 params->mode = bond_mode;
5091 params->xmit_policy = xmit_hashtype;
5092 params->miimon = miimon;
5093 params->num_grat_arp = num_grat_arp;
5094 params->num_unsol_na = num_unsol_na;
5095 params->arp_interval = arp_interval;
5096 params->arp_validate = arp_validate_value;
5097 params->updelay = updelay;
5098 params->downdelay = downdelay;
5099 params->use_carrier = use_carrier;
5100 params->lacp_fast = lacp_fast;
5101 params->primary[0] = 0;
5102 params->fail_over_mac = fail_over_mac_value;
5105 strncpy(params->primary, primary, IFNAMSIZ);
5106 params->primary[IFNAMSIZ - 1] = 0;
5109 memcpy(params->arp_targets, arp_target, sizeof(arp_target));
5114 static struct lock_class_key bonding_netdev_xmit_lock_key;
5115 static struct lock_class_key bonding_netdev_addr_lock_key;
5117 static void bond_set_lockdep_class_one(struct net_device *dev,
5118 struct netdev_queue *txq,
5121 lockdep_set_class(&txq->_xmit_lock,
5122 &bonding_netdev_xmit_lock_key);
5125 static void bond_set_lockdep_class(struct net_device *dev)
5127 lockdep_set_class(&dev->addr_list_lock,
5128 &bonding_netdev_addr_lock_key);
5129 netdev_for_each_tx_queue(dev, bond_set_lockdep_class_one, NULL);
5132 /* Create a new bond based on the specified name and bonding parameters.
5133 * If name is NULL, obtain a suitable "bond%d" name for us.
5134 * Caller must NOT hold rtnl_lock; we need to release it here before we
5135 * set up our sysfs entries.
5137 int bond_create(char *name, struct bond_params *params)
5139 struct net_device *bond_dev;
5140 struct bonding *bond;
5144 down_write(&bonding_rwsem);
5146 /* Check to see if the bond already exists. */
5148 list_for_each_entry(bond, &bond_dev_list, bond_list)
5149 if (strnicmp(bond->dev->name, name, IFNAMSIZ) == 0) {
5150 printk(KERN_ERR DRV_NAME
5151 ": cannot add bond %s; it already exists\n",
5158 bond_dev = alloc_netdev(sizeof(struct bonding), name ? name : "",
5161 printk(KERN_ERR DRV_NAME
5162 ": %s: eek! can't alloc netdev!\n",
5169 res = dev_alloc_name(bond_dev, "bond%d");
5174 /* bond_init() must be called after dev_alloc_name() (for the
5175 * /proc files), but before register_netdevice(), because we
5176 * need to set function pointers.
5179 res = bond_init(bond_dev, params);
5184 res = register_netdevice(bond_dev);
5189 bond_set_lockdep_class(bond_dev);
5191 netif_carrier_off(bond_dev);
5193 up_write(&bonding_rwsem);
5194 rtnl_unlock(); /* allows sysfs registration of net device */
5195 res = bond_create_sysfs_entry(netdev_priv(bond_dev));
5198 down_write(&bonding_rwsem);
5199 bond_deinit(bond_dev);
5200 unregister_netdevice(bond_dev);
5207 bond_deinit(bond_dev);
5209 free_netdev(bond_dev);
5211 up_write(&bonding_rwsem);
5216 static int __init bonding_init(void)
5220 struct bonding *bond;
5222 printk(KERN_INFO "%s", version);
5224 res = bond_check_params(&bonding_defaults);
5229 #ifdef CONFIG_PROC_FS
5230 bond_create_proc_dir();
5233 init_rwsem(&bonding_rwsem);
5235 for (i = 0; i < max_bonds; i++) {
5236 res = bond_create(NULL, &bonding_defaults);
5241 res = bond_create_sysfs();
5245 register_netdevice_notifier(&bond_netdev_notifier);
5246 register_inetaddr_notifier(&bond_inetaddr_notifier);
5247 bond_register_ipv6_notifier();
5251 list_for_each_entry(bond, &bond_dev_list, bond_list) {
5252 bond_work_cancel_all(bond);
5253 destroy_workqueue(bond->wq);
5256 bond_destroy_sysfs();
5266 static void __exit bonding_exit(void)
5268 unregister_netdevice_notifier(&bond_netdev_notifier);
5269 unregister_inetaddr_notifier(&bond_inetaddr_notifier);
5270 bond_unregister_ipv6_notifier();
5272 bond_destroy_sysfs();
5279 module_init(bonding_init);
5280 module_exit(bonding_exit);
5281 MODULE_LICENSE("GPL");
5282 MODULE_VERSION(DRV_VERSION);
5283 MODULE_DESCRIPTION(DRV_DESCRIPTION ", v" DRV_VERSION);
5284 MODULE_AUTHOR("Thomas Davis, tadavis@lbl.gov and many others");
5285 MODULE_SUPPORTED_DEVICE("most ethernet devices");