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 <net/route.h>
82 /*---------------------------- Module parameters ----------------------------*/
84 /* monitor all links that often (in milliseconds). <=0 disables monitoring */
85 #define BOND_LINK_MON_INTERV 0
86 #define BOND_LINK_ARP_INTERV 0
88 static int max_bonds = BOND_DEFAULT_MAX_BONDS;
89 static int miimon = BOND_LINK_MON_INTERV;
90 static int updelay = 0;
91 static int downdelay = 0;
92 static int use_carrier = 1;
93 static char *mode = NULL;
94 static char *primary = NULL;
95 static char *lacp_rate = NULL;
96 static char *xmit_hash_policy = NULL;
97 static int arp_interval = BOND_LINK_ARP_INTERV;
98 static char *arp_ip_target[BOND_MAX_ARP_TARGETS] = { NULL, };
99 static char *arp_validate = NULL;
100 struct bond_params bonding_defaults;
102 module_param(max_bonds, int, 0);
103 MODULE_PARM_DESC(max_bonds, "Max number of bonded devices");
104 module_param(miimon, int, 0);
105 MODULE_PARM_DESC(miimon, "Link check interval in milliseconds");
106 module_param(updelay, int, 0);
107 MODULE_PARM_DESC(updelay, "Delay before considering link up, in milliseconds");
108 module_param(downdelay, int, 0);
109 MODULE_PARM_DESC(downdelay, "Delay before considering link down, "
111 module_param(use_carrier, int, 0);
112 MODULE_PARM_DESC(use_carrier, "Use netif_carrier_ok (vs MII ioctls) in miimon; "
113 "0 for off, 1 for on (default)");
114 module_param(mode, charp, 0);
115 MODULE_PARM_DESC(mode, "Mode of operation : 0 for balance-rr, "
116 "1 for active-backup, 2 for balance-xor, "
117 "3 for broadcast, 4 for 802.3ad, 5 for balance-tlb, "
118 "6 for balance-alb");
119 module_param(primary, charp, 0);
120 MODULE_PARM_DESC(primary, "Primary network device to use");
121 module_param(lacp_rate, charp, 0);
122 MODULE_PARM_DESC(lacp_rate, "LACPDU tx rate to request from 802.3ad partner "
124 module_param(xmit_hash_policy, charp, 0);
125 MODULE_PARM_DESC(xmit_hash_policy, "XOR hashing method: 0 for layer 2 (default)"
126 ", 1 for layer 3+4");
127 module_param(arp_interval, int, 0);
128 MODULE_PARM_DESC(arp_interval, "arp interval in milliseconds");
129 module_param_array(arp_ip_target, charp, NULL, 0);
130 MODULE_PARM_DESC(arp_ip_target, "arp targets in n.n.n.n form");
131 module_param(arp_validate, charp, 0);
132 MODULE_PARM_DESC(arp_validate, "validate src/dst of ARP probes: none (default), active, backup or all");
134 /*----------------------------- Global variables ----------------------------*/
136 static const char * const version =
137 DRV_DESCRIPTION ": v" DRV_VERSION " (" DRV_RELDATE ")\n";
139 LIST_HEAD(bond_dev_list);
141 #ifdef CONFIG_PROC_FS
142 static struct proc_dir_entry *bond_proc_dir = NULL;
145 extern struct rw_semaphore bonding_rwsem;
146 static u32 arp_target[BOND_MAX_ARP_TARGETS] = { 0, } ;
147 static int arp_ip_count = 0;
148 static int bond_mode = BOND_MODE_ROUNDROBIN;
149 static int xmit_hashtype= BOND_XMIT_POLICY_LAYER2;
150 static int lacp_fast = 0;
153 struct bond_parm_tbl bond_lacp_tbl[] = {
154 { "slow", AD_LACP_SLOW},
155 { "fast", AD_LACP_FAST},
159 struct bond_parm_tbl bond_mode_tbl[] = {
160 { "balance-rr", BOND_MODE_ROUNDROBIN},
161 { "active-backup", BOND_MODE_ACTIVEBACKUP},
162 { "balance-xor", BOND_MODE_XOR},
163 { "broadcast", BOND_MODE_BROADCAST},
164 { "802.3ad", BOND_MODE_8023AD},
165 { "balance-tlb", BOND_MODE_TLB},
166 { "balance-alb", BOND_MODE_ALB},
170 struct bond_parm_tbl xmit_hashtype_tbl[] = {
171 { "layer2", BOND_XMIT_POLICY_LAYER2},
172 { "layer3+4", BOND_XMIT_POLICY_LAYER34},
176 struct bond_parm_tbl arp_validate_tbl[] = {
177 { "none", BOND_ARP_VALIDATE_NONE},
178 { "active", BOND_ARP_VALIDATE_ACTIVE},
179 { "backup", BOND_ARP_VALIDATE_BACKUP},
180 { "all", BOND_ARP_VALIDATE_ALL},
184 /*-------------------------- Forward declarations ---------------------------*/
186 static void bond_send_gratuitous_arp(struct bonding *bond);
188 /*---------------------------- General routines -----------------------------*/
190 static const char *bond_mode_name(int mode)
193 case BOND_MODE_ROUNDROBIN :
194 return "load balancing (round-robin)";
195 case BOND_MODE_ACTIVEBACKUP :
196 return "fault-tolerance (active-backup)";
198 return "load balancing (xor)";
199 case BOND_MODE_BROADCAST :
200 return "fault-tolerance (broadcast)";
201 case BOND_MODE_8023AD:
202 return "IEEE 802.3ad Dynamic link aggregation";
204 return "transmit load balancing";
206 return "adaptive load balancing";
212 /*---------------------------------- VLAN -----------------------------------*/
215 * bond_add_vlan - add a new vlan id on bond
216 * @bond: bond that got the notification
217 * @vlan_id: the vlan id to add
219 * Returns -ENOMEM if allocation failed.
221 static int bond_add_vlan(struct bonding *bond, unsigned short vlan_id)
223 struct vlan_entry *vlan;
225 dprintk("bond: %s, vlan id %d\n",
226 (bond ? bond->dev->name: "None"), vlan_id);
228 vlan = kmalloc(sizeof(struct vlan_entry), GFP_KERNEL);
233 INIT_LIST_HEAD(&vlan->vlan_list);
234 vlan->vlan_id = vlan_id;
237 write_lock_bh(&bond->lock);
239 list_add_tail(&vlan->vlan_list, &bond->vlan_list);
241 write_unlock_bh(&bond->lock);
243 dprintk("added VLAN ID %d on bond %s\n", vlan_id, bond->dev->name);
249 * bond_del_vlan - delete a vlan id from bond
250 * @bond: bond that got the notification
251 * @vlan_id: the vlan id to delete
253 * returns -ENODEV if @vlan_id was not found in @bond.
255 static int bond_del_vlan(struct bonding *bond, unsigned short vlan_id)
257 struct vlan_entry *vlan, *next;
260 dprintk("bond: %s, vlan id %d\n", bond->dev->name, vlan_id);
262 write_lock_bh(&bond->lock);
264 list_for_each_entry_safe(vlan, next, &bond->vlan_list, vlan_list) {
265 if (vlan->vlan_id == vlan_id) {
266 list_del(&vlan->vlan_list);
268 if ((bond->params.mode == BOND_MODE_TLB) ||
269 (bond->params.mode == BOND_MODE_ALB)) {
270 bond_alb_clear_vlan(bond, vlan_id);
273 dprintk("removed VLAN ID %d from bond %s\n", vlan_id,
278 if (list_empty(&bond->vlan_list) &&
279 (bond->slave_cnt == 0)) {
280 /* Last VLAN removed and no slaves, so
281 * restore block on adding VLANs. This will
282 * be removed once new slaves that are not
283 * VLAN challenged will be added.
285 bond->dev->features |= NETIF_F_VLAN_CHALLENGED;
293 dprintk("couldn't find VLAN ID %d in bond %s\n", vlan_id,
297 write_unlock_bh(&bond->lock);
302 * bond_has_challenged_slaves
303 * @bond: the bond we're working on
305 * Searches the slave list. Returns 1 if a vlan challenged slave
306 * was found, 0 otherwise.
308 * Assumes bond->lock is held.
310 static int bond_has_challenged_slaves(struct bonding *bond)
315 bond_for_each_slave(bond, slave, i) {
316 if (slave->dev->features & NETIF_F_VLAN_CHALLENGED) {
317 dprintk("found VLAN challenged slave - %s\n",
323 dprintk("no VLAN challenged slaves found\n");
328 * bond_next_vlan - safely skip to the next item in the vlans list.
329 * @bond: the bond we're working on
330 * @curr: item we're advancing from
332 * Returns %NULL if list is empty, bond->next_vlan if @curr is %NULL,
333 * or @curr->next otherwise (even if it is @curr itself again).
335 * Caller must hold bond->lock
337 struct vlan_entry *bond_next_vlan(struct bonding *bond, struct vlan_entry *curr)
339 struct vlan_entry *next, *last;
341 if (list_empty(&bond->vlan_list)) {
346 next = list_entry(bond->vlan_list.next,
347 struct vlan_entry, vlan_list);
349 last = list_entry(bond->vlan_list.prev,
350 struct vlan_entry, vlan_list);
352 next = list_entry(bond->vlan_list.next,
353 struct vlan_entry, vlan_list);
355 next = list_entry(curr->vlan_list.next,
356 struct vlan_entry, vlan_list);
364 * bond_dev_queue_xmit - Prepare skb for xmit.
366 * @bond: bond device that got this skb for tx.
367 * @skb: hw accel VLAN tagged skb to transmit
368 * @slave_dev: slave that is supposed to xmit this skbuff
370 * When the bond gets an skb to transmit that is
371 * already hardware accelerated VLAN tagged, and it
372 * needs to relay this skb to a slave that is not
373 * hw accel capable, the skb needs to be "unaccelerated",
374 * i.e. strip the hwaccel tag and re-insert it as part
377 int bond_dev_queue_xmit(struct bonding *bond, struct sk_buff *skb, struct net_device *slave_dev)
379 unsigned short vlan_id;
381 if (!list_empty(&bond->vlan_list) &&
382 !(slave_dev->features & NETIF_F_HW_VLAN_TX) &&
383 vlan_get_tag(skb, &vlan_id) == 0) {
384 skb->dev = slave_dev;
385 skb = vlan_put_tag(skb, vlan_id);
387 /* vlan_put_tag() frees the skb in case of error,
388 * so return success here so the calling functions
389 * won't attempt to free is again.
394 skb->dev = slave_dev;
404 * In the following 3 functions, bond_vlan_rx_register(), bond_vlan_rx_add_vid
405 * and bond_vlan_rx_kill_vid, We don't protect the slave list iteration with a
407 * a. This operation is performed in IOCTL context,
408 * b. The operation is protected by the RTNL semaphore in the 8021q code,
409 * c. Holding a lock with BH disabled while directly calling a base driver
410 * entry point is generally a BAD idea.
412 * The design of synchronization/protection for this operation in the 8021q
413 * module is good for one or more VLAN devices over a single physical device
414 * and cannot be extended for a teaming solution like bonding, so there is a
415 * potential race condition here where a net device from the vlan group might
416 * be referenced (either by a base driver or the 8021q code) while it is being
417 * removed from the system. However, it turns out we're not making matters
418 * worse, and if it works for regular VLAN usage it will work here too.
422 * bond_vlan_rx_register - Propagates registration to slaves
423 * @bond_dev: bonding net device that got called
424 * @grp: vlan group being registered
426 static void bond_vlan_rx_register(struct net_device *bond_dev, struct vlan_group *grp)
428 struct bonding *bond = bond_dev->priv;
434 bond_for_each_slave(bond, slave, i) {
435 struct net_device *slave_dev = slave->dev;
437 if ((slave_dev->features & NETIF_F_HW_VLAN_RX) &&
438 slave_dev->vlan_rx_register) {
439 slave_dev->vlan_rx_register(slave_dev, grp);
445 * bond_vlan_rx_add_vid - Propagates adding an id to slaves
446 * @bond_dev: bonding net device that got called
447 * @vid: vlan id being added
449 static void bond_vlan_rx_add_vid(struct net_device *bond_dev, uint16_t vid)
451 struct bonding *bond = bond_dev->priv;
455 bond_for_each_slave(bond, slave, i) {
456 struct net_device *slave_dev = slave->dev;
458 if ((slave_dev->features & NETIF_F_HW_VLAN_FILTER) &&
459 slave_dev->vlan_rx_add_vid) {
460 slave_dev->vlan_rx_add_vid(slave_dev, vid);
464 res = bond_add_vlan(bond, vid);
466 printk(KERN_ERR DRV_NAME
467 ": %s: Error: Failed to add vlan id %d\n",
468 bond_dev->name, vid);
473 * bond_vlan_rx_kill_vid - Propagates deleting an id to slaves
474 * @bond_dev: bonding net device that got called
475 * @vid: vlan id being removed
477 static void bond_vlan_rx_kill_vid(struct net_device *bond_dev, uint16_t vid)
479 struct bonding *bond = bond_dev->priv;
481 struct net_device *vlan_dev;
484 bond_for_each_slave(bond, slave, i) {
485 struct net_device *slave_dev = slave->dev;
487 if ((slave_dev->features & NETIF_F_HW_VLAN_FILTER) &&
488 slave_dev->vlan_rx_kill_vid) {
489 /* Save and then restore vlan_dev in the grp array,
490 * since the slave's driver might clear it.
492 vlan_dev = vlan_group_get_device(bond->vlgrp, vid);
493 slave_dev->vlan_rx_kill_vid(slave_dev, vid);
494 vlan_group_set_device(bond->vlgrp, vid, vlan_dev);
498 res = bond_del_vlan(bond, vid);
500 printk(KERN_ERR DRV_NAME
501 ": %s: Error: Failed to remove vlan id %d\n",
502 bond_dev->name, vid);
506 static void bond_add_vlans_on_slave(struct bonding *bond, struct net_device *slave_dev)
508 struct vlan_entry *vlan;
510 write_lock_bh(&bond->lock);
512 if (list_empty(&bond->vlan_list)) {
516 if ((slave_dev->features & NETIF_F_HW_VLAN_RX) &&
517 slave_dev->vlan_rx_register) {
518 slave_dev->vlan_rx_register(slave_dev, bond->vlgrp);
521 if (!(slave_dev->features & NETIF_F_HW_VLAN_FILTER) ||
522 !(slave_dev->vlan_rx_add_vid)) {
526 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
527 slave_dev->vlan_rx_add_vid(slave_dev, vlan->vlan_id);
531 write_unlock_bh(&bond->lock);
534 static void bond_del_vlans_from_slave(struct bonding *bond, struct net_device *slave_dev)
536 struct vlan_entry *vlan;
537 struct net_device *vlan_dev;
539 write_lock_bh(&bond->lock);
541 if (list_empty(&bond->vlan_list)) {
545 if (!(slave_dev->features & NETIF_F_HW_VLAN_FILTER) ||
546 !(slave_dev->vlan_rx_kill_vid)) {
550 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
551 /* Save and then restore vlan_dev in the grp array,
552 * since the slave's driver might clear it.
554 vlan_dev = vlan_group_get_device(bond->vlgrp, vlan->vlan_id);
555 slave_dev->vlan_rx_kill_vid(slave_dev, vlan->vlan_id);
556 vlan_group_set_device(bond->vlgrp, vlan->vlan_id, vlan_dev);
560 if ((slave_dev->features & NETIF_F_HW_VLAN_RX) &&
561 slave_dev->vlan_rx_register) {
562 slave_dev->vlan_rx_register(slave_dev, NULL);
566 write_unlock_bh(&bond->lock);
569 /*------------------------------- Link status -------------------------------*/
572 * Set the carrier state for the master according to the state of its
573 * slaves. If any slaves are up, the master is up. In 802.3ad mode,
574 * do special 802.3ad magic.
576 * Returns zero if carrier state does not change, nonzero if it does.
578 static int bond_set_carrier(struct bonding *bond)
583 if (bond->slave_cnt == 0)
586 if (bond->params.mode == BOND_MODE_8023AD)
587 return bond_3ad_set_carrier(bond);
589 bond_for_each_slave(bond, slave, i) {
590 if (slave->link == BOND_LINK_UP) {
591 if (!netif_carrier_ok(bond->dev)) {
592 netif_carrier_on(bond->dev);
600 if (netif_carrier_ok(bond->dev)) {
601 netif_carrier_off(bond->dev);
608 * Get link speed and duplex from the slave's base driver
609 * using ethtool. If for some reason the call fails or the
610 * values are invalid, fake speed and duplex to 100/Full
613 static int bond_update_speed_duplex(struct slave *slave)
615 struct net_device *slave_dev = slave->dev;
616 static int (* ioctl)(struct net_device *, struct ifreq *, int);
618 struct ethtool_cmd etool;
620 /* Fake speed and duplex */
621 slave->speed = SPEED_100;
622 slave->duplex = DUPLEX_FULL;
624 if (slave_dev->ethtool_ops) {
627 if (!slave_dev->ethtool_ops->get_settings) {
631 res = slave_dev->ethtool_ops->get_settings(slave_dev, &etool);
639 ioctl = slave_dev->do_ioctl;
640 strncpy(ifr.ifr_name, slave_dev->name, IFNAMSIZ);
641 etool.cmd = ETHTOOL_GSET;
642 ifr.ifr_data = (char*)&etool;
643 if (!ioctl || (IOCTL(slave_dev, &ifr, SIOCETHTOOL) < 0)) {
648 switch (etool.speed) {
658 switch (etool.duplex) {
666 slave->speed = etool.speed;
667 slave->duplex = etool.duplex;
673 * if <dev> supports MII link status reporting, check its link status.
675 * We either do MII/ETHTOOL ioctls, or check netif_carrier_ok(),
676 * depening upon the setting of the use_carrier parameter.
678 * Return either BMSR_LSTATUS, meaning that the link is up (or we
679 * can't tell and just pretend it is), or 0, meaning that the link is
682 * If reporting is non-zero, instead of faking link up, return -1 if
683 * both ETHTOOL and MII ioctls fail (meaning the device does not
684 * support them). If use_carrier is set, return whatever it says.
685 * It'd be nice if there was a good way to tell if a driver supports
686 * netif_carrier, but there really isn't.
688 static int bond_check_dev_link(struct bonding *bond, struct net_device *slave_dev, int reporting)
690 static int (* ioctl)(struct net_device *, struct ifreq *, int);
692 struct mii_ioctl_data *mii;
693 struct ethtool_value etool;
695 if (bond->params.use_carrier) {
696 return netif_carrier_ok(slave_dev) ? BMSR_LSTATUS : 0;
699 ioctl = slave_dev->do_ioctl;
701 /* TODO: set pointer to correct ioctl on a per team member */
702 /* bases to make this more efficient. that is, once */
703 /* we determine the correct ioctl, we will always */
704 /* call it and not the others for that team */
708 * We cannot assume that SIOCGMIIPHY will also read a
709 * register; not all network drivers (e.g., e100)
713 /* Yes, the mii is overlaid on the ifreq.ifr_ifru */
714 strncpy(ifr.ifr_name, slave_dev->name, IFNAMSIZ);
716 if (IOCTL(slave_dev, &ifr, SIOCGMIIPHY) == 0) {
717 mii->reg_num = MII_BMSR;
718 if (IOCTL(slave_dev, &ifr, SIOCGMIIREG) == 0) {
719 return (mii->val_out & BMSR_LSTATUS);
724 /* try SIOCETHTOOL ioctl, some drivers cache ETHTOOL_GLINK */
725 /* for a period of time so we attempt to get link status */
726 /* from it last if the above MII ioctls fail... */
727 if (slave_dev->ethtool_ops) {
728 if (slave_dev->ethtool_ops->get_link) {
731 link = slave_dev->ethtool_ops->get_link(slave_dev);
733 return link ? BMSR_LSTATUS : 0;
738 strncpy(ifr.ifr_name, slave_dev->name, IFNAMSIZ);
739 etool.cmd = ETHTOOL_GLINK;
740 ifr.ifr_data = (char*)&etool;
741 if (IOCTL(slave_dev, &ifr, SIOCETHTOOL) == 0) {
742 if (etool.data == 1) {
745 dprintk("SIOCETHTOOL shows link down\n");
752 * If reporting, report that either there's no dev->do_ioctl,
753 * or both SIOCGMIIREG and SIOCETHTOOL failed (meaning that we
754 * cannot report link status). If not reporting, pretend
757 return (reporting ? -1 : BMSR_LSTATUS);
760 /*----------------------------- Multicast list ------------------------------*/
763 * Returns 0 if dmi1 and dmi2 are the same, non-0 otherwise
765 static inline int bond_is_dmi_same(struct dev_mc_list *dmi1, struct dev_mc_list *dmi2)
767 return memcmp(dmi1->dmi_addr, dmi2->dmi_addr, dmi1->dmi_addrlen) == 0 &&
768 dmi1->dmi_addrlen == dmi2->dmi_addrlen;
772 * returns dmi entry if found, NULL otherwise
774 static struct dev_mc_list *bond_mc_list_find_dmi(struct dev_mc_list *dmi, struct dev_mc_list *mc_list)
776 struct dev_mc_list *idmi;
778 for (idmi = mc_list; idmi; idmi = idmi->next) {
779 if (bond_is_dmi_same(dmi, idmi)) {
788 * Push the promiscuity flag down to appropriate slaves
790 static void bond_set_promiscuity(struct bonding *bond, int inc)
792 if (USES_PRIMARY(bond->params.mode)) {
793 /* write lock already acquired */
794 if (bond->curr_active_slave) {
795 dev_set_promiscuity(bond->curr_active_slave->dev, inc);
800 bond_for_each_slave(bond, slave, i) {
801 dev_set_promiscuity(slave->dev, inc);
807 * Push the allmulti flag down to all slaves
809 static void bond_set_allmulti(struct bonding *bond, int inc)
811 if (USES_PRIMARY(bond->params.mode)) {
812 /* write lock already acquired */
813 if (bond->curr_active_slave) {
814 dev_set_allmulti(bond->curr_active_slave->dev, inc);
819 bond_for_each_slave(bond, slave, i) {
820 dev_set_allmulti(slave->dev, inc);
826 * Add a Multicast address to slaves
829 static void bond_mc_add(struct bonding *bond, void *addr, int alen)
831 if (USES_PRIMARY(bond->params.mode)) {
832 /* write lock already acquired */
833 if (bond->curr_active_slave) {
834 dev_mc_add(bond->curr_active_slave->dev, addr, alen, 0);
839 bond_for_each_slave(bond, slave, i) {
840 dev_mc_add(slave->dev, addr, alen, 0);
846 * Remove a multicast address from slave
849 static void bond_mc_delete(struct bonding *bond, void *addr, int alen)
851 if (USES_PRIMARY(bond->params.mode)) {
852 /* write lock already acquired */
853 if (bond->curr_active_slave) {
854 dev_mc_delete(bond->curr_active_slave->dev, addr, alen, 0);
859 bond_for_each_slave(bond, slave, i) {
860 dev_mc_delete(slave->dev, addr, alen, 0);
867 * Retrieve the list of registered multicast addresses for the bonding
868 * device and retransmit an IGMP JOIN request to the current active
871 static void bond_resend_igmp_join_requests(struct bonding *bond)
873 struct in_device *in_dev;
874 struct ip_mc_list *im;
877 in_dev = __in_dev_get_rcu(bond->dev);
879 for (im = in_dev->mc_list; im; im = im->next) {
880 ip_mc_rejoin_group(im);
888 * Totally destroys the mc_list in bond
890 static void bond_mc_list_destroy(struct bonding *bond)
892 struct dev_mc_list *dmi;
896 bond->mc_list = dmi->next;
900 bond->mc_list = NULL;
904 * Copy all the Multicast addresses from src to the bonding device dst
906 static int bond_mc_list_copy(struct dev_mc_list *mc_list, struct bonding *bond,
909 struct dev_mc_list *dmi, *new_dmi;
911 for (dmi = mc_list; dmi; dmi = dmi->next) {
912 new_dmi = kmalloc(sizeof(struct dev_mc_list), gfp_flag);
915 /* FIXME: Potential memory leak !!! */
919 new_dmi->next = bond->mc_list;
920 bond->mc_list = new_dmi;
921 new_dmi->dmi_addrlen = dmi->dmi_addrlen;
922 memcpy(new_dmi->dmi_addr, dmi->dmi_addr, dmi->dmi_addrlen);
923 new_dmi->dmi_users = dmi->dmi_users;
924 new_dmi->dmi_gusers = dmi->dmi_gusers;
931 * flush all members of flush->mc_list from device dev->mc_list
933 static void bond_mc_list_flush(struct net_device *bond_dev, struct net_device *slave_dev)
935 struct bonding *bond = bond_dev->priv;
936 struct dev_mc_list *dmi;
938 for (dmi = bond_dev->mc_list; dmi; dmi = dmi->next) {
939 dev_mc_delete(slave_dev, dmi->dmi_addr, dmi->dmi_addrlen, 0);
942 if (bond->params.mode == BOND_MODE_8023AD) {
943 /* del lacpdu mc addr from mc list */
944 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
946 dev_mc_delete(slave_dev, lacpdu_multicast, ETH_ALEN, 0);
950 /*--------------------------- Active slave change ---------------------------*/
953 * Update the mc list and multicast-related flags for the new and
954 * old active slaves (if any) according to the multicast mode, and
955 * promiscuous flags unconditionally.
957 static void bond_mc_swap(struct bonding *bond, struct slave *new_active, struct slave *old_active)
959 struct dev_mc_list *dmi;
961 if (!USES_PRIMARY(bond->params.mode)) {
962 /* nothing to do - mc list is already up-to-date on
969 if (bond->dev->flags & IFF_PROMISC) {
970 dev_set_promiscuity(old_active->dev, -1);
973 if (bond->dev->flags & IFF_ALLMULTI) {
974 dev_set_allmulti(old_active->dev, -1);
977 for (dmi = bond->dev->mc_list; dmi; dmi = dmi->next) {
978 dev_mc_delete(old_active->dev, dmi->dmi_addr, dmi->dmi_addrlen, 0);
983 if (bond->dev->flags & IFF_PROMISC) {
984 dev_set_promiscuity(new_active->dev, 1);
987 if (bond->dev->flags & IFF_ALLMULTI) {
988 dev_set_allmulti(new_active->dev, 1);
991 for (dmi = bond->dev->mc_list; dmi; dmi = dmi->next) {
992 dev_mc_add(new_active->dev, dmi->dmi_addr, dmi->dmi_addrlen, 0);
994 bond_resend_igmp_join_requests(bond);
999 * find_best_interface - select the best available slave to be the active one
1000 * @bond: our bonding struct
1002 * Warning: Caller must hold curr_slave_lock for writing.
1004 static struct slave *bond_find_best_slave(struct bonding *bond)
1006 struct slave *new_active, *old_active;
1007 struct slave *bestslave = NULL;
1008 int mintime = bond->params.updelay;
1011 new_active = old_active = bond->curr_active_slave;
1013 if (!new_active) { /* there were no active slaves left */
1014 if (bond->slave_cnt > 0) { /* found one slave */
1015 new_active = bond->first_slave;
1017 return NULL; /* still no slave, return NULL */
1021 /* first try the primary link; if arping, a link must tx/rx traffic
1022 * before it can be considered the curr_active_slave - also, we would skip
1023 * slaves between the curr_active_slave and primary_slave that may be up
1026 if ((bond->primary_slave) &&
1027 (!bond->params.arp_interval) &&
1028 (IS_UP(bond->primary_slave->dev))) {
1029 new_active = bond->primary_slave;
1032 /* remember where to stop iterating over the slaves */
1033 old_active = new_active;
1035 bond_for_each_slave_from(bond, new_active, i, old_active) {
1036 if (IS_UP(new_active->dev)) {
1037 if (new_active->link == BOND_LINK_UP) {
1039 } else if (new_active->link == BOND_LINK_BACK) {
1040 /* link up, but waiting for stabilization */
1041 if (new_active->delay < mintime) {
1042 mintime = new_active->delay;
1043 bestslave = new_active;
1053 * change_active_interface - change the active slave into the specified one
1054 * @bond: our bonding struct
1055 * @new: the new slave to make the active one
1057 * Set the new slave to the bond's settings and unset them on the old
1058 * curr_active_slave.
1059 * Setting include flags, mc-list, promiscuity, allmulti, etc.
1061 * If @new's link state is %BOND_LINK_BACK we'll set it to %BOND_LINK_UP,
1062 * because it is apparently the best available slave we have, even though its
1063 * updelay hasn't timed out yet.
1065 * Warning: Caller must hold curr_slave_lock for writing.
1067 void bond_change_active_slave(struct bonding *bond, struct slave *new_active)
1069 struct slave *old_active = bond->curr_active_slave;
1071 if (old_active == new_active) {
1076 if (new_active->link == BOND_LINK_BACK) {
1077 if (USES_PRIMARY(bond->params.mode)) {
1078 printk(KERN_INFO DRV_NAME
1079 ": %s: making interface %s the new "
1080 "active one %d ms earlier.\n",
1081 bond->dev->name, new_active->dev->name,
1082 (bond->params.updelay - new_active->delay) * bond->params.miimon);
1085 new_active->delay = 0;
1086 new_active->link = BOND_LINK_UP;
1087 new_active->jiffies = jiffies;
1089 if (bond->params.mode == BOND_MODE_8023AD) {
1090 bond_3ad_handle_link_change(new_active, BOND_LINK_UP);
1093 if ((bond->params.mode == BOND_MODE_TLB) ||
1094 (bond->params.mode == BOND_MODE_ALB)) {
1095 bond_alb_handle_link_change(bond, new_active, BOND_LINK_UP);
1098 if (USES_PRIMARY(bond->params.mode)) {
1099 printk(KERN_INFO DRV_NAME
1100 ": %s: making interface %s the new "
1102 bond->dev->name, new_active->dev->name);
1107 if (USES_PRIMARY(bond->params.mode)) {
1108 bond_mc_swap(bond, new_active, old_active);
1111 if ((bond->params.mode == BOND_MODE_TLB) ||
1112 (bond->params.mode == BOND_MODE_ALB)) {
1113 bond_alb_handle_active_change(bond, new_active);
1115 bond_set_slave_inactive_flags(old_active);
1117 bond_set_slave_active_flags(new_active);
1119 bond->curr_active_slave = new_active;
1122 if (bond->params.mode == BOND_MODE_ACTIVEBACKUP) {
1124 bond_set_slave_inactive_flags(old_active);
1128 bond_set_slave_active_flags(new_active);
1130 bond_send_gratuitous_arp(bond);
1135 * bond_select_active_slave - select a new active slave, if needed
1136 * @bond: our bonding struct
1138 * This functions shoud be called when one of the following occurs:
1139 * - The old curr_active_slave has been released or lost its link.
1140 * - The primary_slave has got its link back.
1141 * - A slave has got its link back and there's no old curr_active_slave.
1143 * Warning: Caller must hold curr_slave_lock for writing.
1145 void bond_select_active_slave(struct bonding *bond)
1147 struct slave *best_slave;
1150 best_slave = bond_find_best_slave(bond);
1151 if (best_slave != bond->curr_active_slave) {
1152 bond_change_active_slave(bond, best_slave);
1153 rv = bond_set_carrier(bond);
1157 if (netif_carrier_ok(bond->dev)) {
1158 printk(KERN_INFO DRV_NAME
1159 ": %s: first active interface up!\n",
1162 printk(KERN_INFO DRV_NAME ": %s: "
1163 "now running without any active interface !\n",
1169 /*--------------------------- slave list handling ---------------------------*/
1172 * This function attaches the slave to the end of list.
1174 * bond->lock held for writing by caller.
1176 static void bond_attach_slave(struct bonding *bond, struct slave *new_slave)
1178 if (bond->first_slave == NULL) { /* attaching the first slave */
1179 new_slave->next = new_slave;
1180 new_slave->prev = new_slave;
1181 bond->first_slave = new_slave;
1183 new_slave->next = bond->first_slave;
1184 new_slave->prev = bond->first_slave->prev;
1185 new_slave->next->prev = new_slave;
1186 new_slave->prev->next = new_slave;
1193 * This function detaches the slave from the list.
1194 * WARNING: no check is made to verify if the slave effectively
1195 * belongs to <bond>.
1196 * Nothing is freed on return, structures are just unchained.
1197 * If any slave pointer in bond was pointing to <slave>,
1198 * it should be changed by the calling function.
1200 * bond->lock held for writing by caller.
1202 static void bond_detach_slave(struct bonding *bond, struct slave *slave)
1205 slave->next->prev = slave->prev;
1209 slave->prev->next = slave->next;
1212 if (bond->first_slave == slave) { /* slave is the first slave */
1213 if (bond->slave_cnt > 1) { /* there are more slave */
1214 bond->first_slave = slave->next;
1216 bond->first_slave = NULL; /* slave was the last one */
1225 /*---------------------------------- IOCTL ----------------------------------*/
1227 static int bond_sethwaddr(struct net_device *bond_dev,
1228 struct net_device *slave_dev)
1230 dprintk("bond_dev=%p\n", bond_dev);
1231 dprintk("slave_dev=%p\n", slave_dev);
1232 dprintk("slave_dev->addr_len=%d\n", slave_dev->addr_len);
1233 memcpy(bond_dev->dev_addr, slave_dev->dev_addr, slave_dev->addr_len);
1237 #define BOND_INTERSECT_FEATURES \
1238 (NETIF_F_SG | NETIF_F_ALL_CSUM | NETIF_F_TSO | NETIF_F_UFO)
1241 * Compute the common dev->feature set available to all slaves. Some
1242 * feature bits are managed elsewhere, so preserve feature bits set on
1243 * master device that are not part of the examined set.
1245 static int bond_compute_features(struct bonding *bond)
1247 unsigned long features = BOND_INTERSECT_FEATURES;
1248 struct slave *slave;
1249 struct net_device *bond_dev = bond->dev;
1250 unsigned short max_hard_header_len = ETH_HLEN;
1253 bond_for_each_slave(bond, slave, i) {
1254 features &= (slave->dev->features & BOND_INTERSECT_FEATURES);
1255 if (slave->dev->hard_header_len > max_hard_header_len)
1256 max_hard_header_len = slave->dev->hard_header_len;
1259 if ((features & NETIF_F_SG) &&
1260 !(features & NETIF_F_ALL_CSUM))
1261 features &= ~NETIF_F_SG;
1264 * features will include NETIF_F_TSO (NETIF_F_UFO) iff all
1265 * slave devices support NETIF_F_TSO (NETIF_F_UFO), which
1266 * implies that all slaves also support scatter-gather
1267 * (NETIF_F_SG), which implies that features also includes
1268 * NETIF_F_SG. So no need to check whether we have an
1269 * illegal combination of NETIF_F_{TSO,UFO} and
1273 features |= (bond_dev->features & ~BOND_INTERSECT_FEATURES);
1274 bond_dev->features = features;
1275 bond_dev->hard_header_len = max_hard_header_len;
1280 /* enslave device <slave> to bond device <master> */
1281 int bond_enslave(struct net_device *bond_dev, struct net_device *slave_dev)
1283 struct bonding *bond = bond_dev->priv;
1284 struct slave *new_slave = NULL;
1285 struct dev_mc_list *dmi;
1286 struct sockaddr addr;
1288 int old_features = bond_dev->features;
1291 if (!bond->params.use_carrier && slave_dev->ethtool_ops == NULL &&
1292 slave_dev->do_ioctl == NULL) {
1293 printk(KERN_WARNING DRV_NAME
1294 ": %s: Warning: no link monitoring support for %s\n",
1295 bond_dev->name, slave_dev->name);
1298 /* bond must be initialized by bond_open() before enslaving */
1299 if (!(bond_dev->flags & IFF_UP)) {
1300 dprintk("Error, master_dev is not up\n");
1304 /* already enslaved */
1305 if (slave_dev->flags & IFF_SLAVE) {
1306 dprintk("Error, Device was already enslaved\n");
1310 /* vlan challenged mutual exclusion */
1311 /* no need to lock since we're protected by rtnl_lock */
1312 if (slave_dev->features & NETIF_F_VLAN_CHALLENGED) {
1313 dprintk("%s: NETIF_F_VLAN_CHALLENGED\n", slave_dev->name);
1314 if (!list_empty(&bond->vlan_list)) {
1315 printk(KERN_ERR DRV_NAME
1316 ": %s: Error: cannot enslave VLAN "
1317 "challenged slave %s on VLAN enabled "
1318 "bond %s\n", bond_dev->name, slave_dev->name,
1322 printk(KERN_WARNING DRV_NAME
1323 ": %s: Warning: enslaved VLAN challenged "
1324 "slave %s. Adding VLANs will be blocked as "
1325 "long as %s is part of bond %s\n",
1326 bond_dev->name, slave_dev->name, slave_dev->name,
1328 bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
1331 dprintk("%s: ! NETIF_F_VLAN_CHALLENGED\n", slave_dev->name);
1332 if (bond->slave_cnt == 0) {
1333 /* First slave, and it is not VLAN challenged,
1334 * so remove the block of adding VLANs over the bond.
1336 bond_dev->features &= ~NETIF_F_VLAN_CHALLENGED;
1341 * Old ifenslave binaries are no longer supported. These can
1342 * be identified with moderate accurary by the state of the slave:
1343 * the current ifenslave will set the interface down prior to
1344 * enslaving it; the old ifenslave will not.
1346 if ((slave_dev->flags & IFF_UP)) {
1347 printk(KERN_ERR DRV_NAME ": %s is up. "
1348 "This may be due to an out of date ifenslave.\n",
1351 goto err_undo_flags;
1354 if (slave_dev->set_mac_address == NULL) {
1355 printk(KERN_ERR DRV_NAME
1356 ": %s: Error: The slave device you specified does "
1357 "not support setting the MAC address. "
1358 "Your kernel likely does not support slave "
1359 "devices.\n", bond_dev->name);
1361 goto err_undo_flags;
1364 new_slave = kzalloc(sizeof(struct slave), GFP_KERNEL);
1367 goto err_undo_flags;
1370 /* save slave's original flags before calling
1371 * netdev_set_master and dev_open
1373 new_slave->original_flags = slave_dev->flags;
1376 * Save slave's original ("permanent") mac address for modes
1377 * that need it, and for restoring it upon release, and then
1378 * set it to the master's address
1380 memcpy(new_slave->perm_hwaddr, slave_dev->dev_addr, ETH_ALEN);
1383 * Set slave to master's mac address. The application already
1384 * set the master's mac address to that of the first slave
1386 memcpy(addr.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
1387 addr.sa_family = slave_dev->type;
1388 res = dev_set_mac_address(slave_dev, &addr);
1390 dprintk("Error %d calling set_mac_address\n", res);
1394 res = netdev_set_master(slave_dev, bond_dev);
1396 dprintk("Error %d calling netdev_set_master\n", res);
1399 /* open the slave since the application closed it */
1400 res = dev_open(slave_dev);
1402 dprintk("Openning slave %s failed\n", slave_dev->name);
1403 goto err_restore_mac;
1406 new_slave->dev = slave_dev;
1407 slave_dev->priv_flags |= IFF_BONDING;
1409 if ((bond->params.mode == BOND_MODE_TLB) ||
1410 (bond->params.mode == BOND_MODE_ALB)) {
1411 /* bond_alb_init_slave() must be called before all other stages since
1412 * it might fail and we do not want to have to undo everything
1414 res = bond_alb_init_slave(bond, new_slave);
1416 goto err_unset_master;
1420 /* If the mode USES_PRIMARY, then the new slave gets the
1421 * master's promisc (and mc) settings only if it becomes the
1422 * curr_active_slave, and that is taken care of later when calling
1423 * bond_change_active()
1425 if (!USES_PRIMARY(bond->params.mode)) {
1426 /* set promiscuity level to new slave */
1427 if (bond_dev->flags & IFF_PROMISC) {
1428 dev_set_promiscuity(slave_dev, 1);
1431 /* set allmulti level to new slave */
1432 if (bond_dev->flags & IFF_ALLMULTI) {
1433 dev_set_allmulti(slave_dev, 1);
1436 /* upload master's mc_list to new slave */
1437 for (dmi = bond_dev->mc_list; dmi; dmi = dmi->next) {
1438 dev_mc_add (slave_dev, dmi->dmi_addr, dmi->dmi_addrlen, 0);
1442 if (bond->params.mode == BOND_MODE_8023AD) {
1443 /* add lacpdu mc addr to mc list */
1444 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
1446 dev_mc_add(slave_dev, lacpdu_multicast, ETH_ALEN, 0);
1449 bond_add_vlans_on_slave(bond, slave_dev);
1451 write_lock_bh(&bond->lock);
1453 bond_attach_slave(bond, new_slave);
1455 new_slave->delay = 0;
1456 new_slave->link_failure_count = 0;
1458 bond_compute_features(bond);
1460 new_slave->last_arp_rx = jiffies;
1462 if (bond->params.miimon && !bond->params.use_carrier) {
1463 link_reporting = bond_check_dev_link(bond, slave_dev, 1);
1465 if ((link_reporting == -1) && !bond->params.arp_interval) {
1467 * miimon is set but a bonded network driver
1468 * does not support ETHTOOL/MII and
1469 * arp_interval is not set. Note: if
1470 * use_carrier is enabled, we will never go
1471 * here (because netif_carrier is always
1472 * supported); thus, we don't need to change
1473 * the messages for netif_carrier.
1475 printk(KERN_WARNING DRV_NAME
1476 ": %s: Warning: MII and ETHTOOL support not "
1477 "available for interface %s, and "
1478 "arp_interval/arp_ip_target module parameters "
1479 "not specified, thus bonding will not detect "
1480 "link failures! see bonding.txt for details.\n",
1481 bond_dev->name, slave_dev->name);
1482 } else if (link_reporting == -1) {
1483 /* unable get link status using mii/ethtool */
1484 printk(KERN_WARNING DRV_NAME
1485 ": %s: Warning: can't get link status from "
1486 "interface %s; the network driver associated "
1487 "with this interface does not support MII or "
1488 "ETHTOOL link status reporting, thus miimon "
1489 "has no effect on this interface.\n",
1490 bond_dev->name, slave_dev->name);
1494 /* check for initial state */
1495 if (!bond->params.miimon ||
1496 (bond_check_dev_link(bond, slave_dev, 0) == BMSR_LSTATUS)) {
1497 if (bond->params.updelay) {
1498 dprintk("Initial state of slave_dev is "
1499 "BOND_LINK_BACK\n");
1500 new_slave->link = BOND_LINK_BACK;
1501 new_slave->delay = bond->params.updelay;
1503 dprintk("Initial state of slave_dev is "
1505 new_slave->link = BOND_LINK_UP;
1507 new_slave->jiffies = jiffies;
1509 dprintk("Initial state of slave_dev is "
1510 "BOND_LINK_DOWN\n");
1511 new_slave->link = BOND_LINK_DOWN;
1514 if (bond_update_speed_duplex(new_slave) &&
1515 (new_slave->link != BOND_LINK_DOWN)) {
1516 printk(KERN_WARNING DRV_NAME
1517 ": %s: Warning: failed to get speed and duplex from %s, "
1518 "assumed to be 100Mb/sec and Full.\n",
1519 bond_dev->name, new_slave->dev->name);
1521 if (bond->params.mode == BOND_MODE_8023AD) {
1522 printk(KERN_WARNING DRV_NAME
1523 ": %s: Warning: Operation of 802.3ad mode requires ETHTOOL "
1524 "support in base driver for proper aggregator "
1525 "selection.\n", bond_dev->name);
1529 if (USES_PRIMARY(bond->params.mode) && bond->params.primary[0]) {
1530 /* if there is a primary slave, remember it */
1531 if (strcmp(bond->params.primary, new_slave->dev->name) == 0) {
1532 bond->primary_slave = new_slave;
1536 switch (bond->params.mode) {
1537 case BOND_MODE_ACTIVEBACKUP:
1538 bond_set_slave_inactive_flags(new_slave);
1539 bond_select_active_slave(bond);
1541 case BOND_MODE_8023AD:
1542 /* in 802.3ad mode, the internal mechanism
1543 * will activate the slaves in the selected
1546 bond_set_slave_inactive_flags(new_slave);
1547 /* if this is the first slave */
1548 if (bond->slave_cnt == 1) {
1549 SLAVE_AD_INFO(new_slave).id = 1;
1550 /* Initialize AD with the number of times that the AD timer is called in 1 second
1551 * can be called only after the mac address of the bond is set
1553 bond_3ad_initialize(bond, 1000/AD_TIMER_INTERVAL,
1554 bond->params.lacp_fast);
1556 SLAVE_AD_INFO(new_slave).id =
1557 SLAVE_AD_INFO(new_slave->prev).id + 1;
1560 bond_3ad_bind_slave(new_slave);
1564 new_slave->state = BOND_STATE_ACTIVE;
1565 if ((!bond->curr_active_slave) &&
1566 (new_slave->link != BOND_LINK_DOWN)) {
1567 /* first slave or no active slave yet, and this link
1568 * is OK, so make this interface the active one
1570 bond_change_active_slave(bond, new_slave);
1572 bond_set_slave_inactive_flags(new_slave);
1576 dprintk("This slave is always active in trunk mode\n");
1578 /* always active in trunk mode */
1579 new_slave->state = BOND_STATE_ACTIVE;
1581 /* In trunking mode there is little meaning to curr_active_slave
1582 * anyway (it holds no special properties of the bond device),
1583 * so we can change it without calling change_active_interface()
1585 if (!bond->curr_active_slave) {
1586 bond->curr_active_slave = new_slave;
1589 } /* switch(bond_mode) */
1591 bond_set_carrier(bond);
1593 write_unlock_bh(&bond->lock);
1595 res = bond_create_slave_symlinks(bond_dev, slave_dev);
1597 goto err_unset_master;
1599 printk(KERN_INFO DRV_NAME
1600 ": %s: enslaving %s as a%s interface with a%s link.\n",
1601 bond_dev->name, slave_dev->name,
1602 new_slave->state == BOND_STATE_ACTIVE ? "n active" : " backup",
1603 new_slave->link != BOND_LINK_DOWN ? "n up" : " down");
1605 /* enslave is successful */
1608 /* Undo stages on error */
1610 netdev_set_master(slave_dev, NULL);
1613 dev_close(slave_dev);
1616 memcpy(addr.sa_data, new_slave->perm_hwaddr, ETH_ALEN);
1617 addr.sa_family = slave_dev->type;
1618 dev_set_mac_address(slave_dev, &addr);
1624 bond_dev->features = old_features;
1630 * Try to release the slave device <slave> from the bond device <master>
1631 * It is legal to access curr_active_slave without a lock because all the function
1634 * The rules for slave state should be:
1635 * for Active/Backup:
1636 * Active stays on all backups go down
1637 * for Bonded connections:
1638 * The first up interface should be left on and all others downed.
1640 int bond_release(struct net_device *bond_dev, struct net_device *slave_dev)
1642 struct bonding *bond = bond_dev->priv;
1643 struct slave *slave, *oldcurrent;
1644 struct sockaddr addr;
1645 int mac_addr_differ;
1647 /* slave is not a slave or master is not master of this slave */
1648 if (!(slave_dev->flags & IFF_SLAVE) ||
1649 (slave_dev->master != bond_dev)) {
1650 printk(KERN_ERR DRV_NAME
1651 ": %s: Error: cannot release %s.\n",
1652 bond_dev->name, slave_dev->name);
1656 write_lock_bh(&bond->lock);
1658 slave = bond_get_slave_by_dev(bond, slave_dev);
1660 /* not a slave of this bond */
1661 printk(KERN_INFO DRV_NAME
1662 ": %s: %s not enslaved\n",
1663 bond_dev->name, slave_dev->name);
1664 write_unlock_bh(&bond->lock);
1668 mac_addr_differ = memcmp(bond_dev->dev_addr,
1671 if (!mac_addr_differ && (bond->slave_cnt > 1)) {
1672 printk(KERN_WARNING DRV_NAME
1673 ": %s: Warning: the permanent HWaddr of %s "
1674 "- %02X:%02X:%02X:%02X:%02X:%02X - is "
1675 "still in use by %s. Set the HWaddr of "
1676 "%s to a different address to avoid "
1680 slave->perm_hwaddr[0],
1681 slave->perm_hwaddr[1],
1682 slave->perm_hwaddr[2],
1683 slave->perm_hwaddr[3],
1684 slave->perm_hwaddr[4],
1685 slave->perm_hwaddr[5],
1690 /* Inform AD package of unbinding of slave. */
1691 if (bond->params.mode == BOND_MODE_8023AD) {
1692 /* must be called before the slave is
1693 * detached from the list
1695 bond_3ad_unbind_slave(slave);
1698 printk(KERN_INFO DRV_NAME
1699 ": %s: releasing %s interface %s\n",
1701 (slave->state == BOND_STATE_ACTIVE)
1702 ? "active" : "backup",
1705 oldcurrent = bond->curr_active_slave;
1707 bond->current_arp_slave = NULL;
1709 /* release the slave from its bond */
1710 bond_detach_slave(bond, slave);
1712 bond_compute_features(bond);
1714 if (bond->primary_slave == slave) {
1715 bond->primary_slave = NULL;
1718 if (oldcurrent == slave) {
1719 bond_change_active_slave(bond, NULL);
1722 if ((bond->params.mode == BOND_MODE_TLB) ||
1723 (bond->params.mode == BOND_MODE_ALB)) {
1724 /* Must be called only after the slave has been
1725 * detached from the list and the curr_active_slave
1726 * has been cleared (if our_slave == old_current),
1727 * but before a new active slave is selected.
1729 bond_alb_deinit_slave(bond, slave);
1732 if (oldcurrent == slave)
1733 bond_select_active_slave(bond);
1735 if (bond->slave_cnt == 0) {
1736 bond_set_carrier(bond);
1738 /* if the last slave was removed, zero the mac address
1739 * of the master so it will be set by the application
1740 * to the mac address of the first slave
1742 memset(bond_dev->dev_addr, 0, bond_dev->addr_len);
1744 if (list_empty(&bond->vlan_list)) {
1745 bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
1747 printk(KERN_WARNING DRV_NAME
1748 ": %s: Warning: clearing HW address of %s while it "
1749 "still has VLANs.\n",
1750 bond_dev->name, bond_dev->name);
1751 printk(KERN_WARNING DRV_NAME
1752 ": %s: When re-adding slaves, make sure the bond's "
1753 "HW address matches its VLANs'.\n",
1756 } else if ((bond_dev->features & NETIF_F_VLAN_CHALLENGED) &&
1757 !bond_has_challenged_slaves(bond)) {
1758 printk(KERN_INFO DRV_NAME
1759 ": %s: last VLAN challenged slave %s "
1760 "left bond %s. VLAN blocking is removed\n",
1761 bond_dev->name, slave_dev->name, bond_dev->name);
1762 bond_dev->features &= ~NETIF_F_VLAN_CHALLENGED;
1765 write_unlock_bh(&bond->lock);
1767 /* must do this from outside any spinlocks */
1768 bond_destroy_slave_symlinks(bond_dev, slave_dev);
1770 bond_del_vlans_from_slave(bond, slave_dev);
1772 /* If the mode USES_PRIMARY, then we should only remove its
1773 * promisc and mc settings if it was the curr_active_slave, but that was
1774 * already taken care of above when we detached the slave
1776 if (!USES_PRIMARY(bond->params.mode)) {
1777 /* unset promiscuity level from slave */
1778 if (bond_dev->flags & IFF_PROMISC) {
1779 dev_set_promiscuity(slave_dev, -1);
1782 /* unset allmulti level from slave */
1783 if (bond_dev->flags & IFF_ALLMULTI) {
1784 dev_set_allmulti(slave_dev, -1);
1787 /* flush master's mc_list from slave */
1788 bond_mc_list_flush(bond_dev, slave_dev);
1791 netdev_set_master(slave_dev, NULL);
1793 /* close slave before restoring its mac address */
1794 dev_close(slave_dev);
1796 /* restore original ("permanent") mac address */
1797 memcpy(addr.sa_data, slave->perm_hwaddr, ETH_ALEN);
1798 addr.sa_family = slave_dev->type;
1799 dev_set_mac_address(slave_dev, &addr);
1801 slave_dev->priv_flags &= ~(IFF_MASTER_8023AD | IFF_MASTER_ALB |
1802 IFF_SLAVE_INACTIVE | IFF_BONDING |
1807 return 0; /* deletion OK */
1811 * This function releases all slaves.
1813 static int bond_release_all(struct net_device *bond_dev)
1815 struct bonding *bond = bond_dev->priv;
1816 struct slave *slave;
1817 struct net_device *slave_dev;
1818 struct sockaddr addr;
1820 write_lock_bh(&bond->lock);
1822 netif_carrier_off(bond_dev);
1824 if (bond->slave_cnt == 0) {
1828 bond->current_arp_slave = NULL;
1829 bond->primary_slave = NULL;
1830 bond_change_active_slave(bond, NULL);
1832 while ((slave = bond->first_slave) != NULL) {
1833 /* Inform AD package of unbinding of slave
1834 * before slave is detached from the list.
1836 if (bond->params.mode == BOND_MODE_8023AD) {
1837 bond_3ad_unbind_slave(slave);
1840 slave_dev = slave->dev;
1841 bond_detach_slave(bond, slave);
1843 if ((bond->params.mode == BOND_MODE_TLB) ||
1844 (bond->params.mode == BOND_MODE_ALB)) {
1845 /* must be called only after the slave
1846 * has been detached from the list
1848 bond_alb_deinit_slave(bond, slave);
1851 bond_compute_features(bond);
1853 /* now that the slave is detached, unlock and perform
1854 * all the undo steps that should not be called from
1857 write_unlock_bh(&bond->lock);
1859 bond_destroy_slave_symlinks(bond_dev, slave_dev);
1860 bond_del_vlans_from_slave(bond, slave_dev);
1862 /* If the mode USES_PRIMARY, then we should only remove its
1863 * promisc and mc settings if it was the curr_active_slave, but that was
1864 * already taken care of above when we detached the slave
1866 if (!USES_PRIMARY(bond->params.mode)) {
1867 /* unset promiscuity level from slave */
1868 if (bond_dev->flags & IFF_PROMISC) {
1869 dev_set_promiscuity(slave_dev, -1);
1872 /* unset allmulti level from slave */
1873 if (bond_dev->flags & IFF_ALLMULTI) {
1874 dev_set_allmulti(slave_dev, -1);
1877 /* flush master's mc_list from slave */
1878 bond_mc_list_flush(bond_dev, slave_dev);
1881 netdev_set_master(slave_dev, NULL);
1883 /* close slave before restoring its mac address */
1884 dev_close(slave_dev);
1886 /* restore original ("permanent") mac address*/
1887 memcpy(addr.sa_data, slave->perm_hwaddr, ETH_ALEN);
1888 addr.sa_family = slave_dev->type;
1889 dev_set_mac_address(slave_dev, &addr);
1891 slave_dev->priv_flags &= ~(IFF_MASTER_8023AD | IFF_MASTER_ALB |
1892 IFF_SLAVE_INACTIVE);
1896 /* re-acquire the lock before getting the next slave */
1897 write_lock_bh(&bond->lock);
1900 /* zero the mac address of the master so it will be
1901 * set by the application to the mac address of the
1904 memset(bond_dev->dev_addr, 0, bond_dev->addr_len);
1906 if (list_empty(&bond->vlan_list)) {
1907 bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
1909 printk(KERN_WARNING DRV_NAME
1910 ": %s: Warning: clearing HW address of %s while it "
1911 "still has VLANs.\n",
1912 bond_dev->name, bond_dev->name);
1913 printk(KERN_WARNING DRV_NAME
1914 ": %s: When re-adding slaves, make sure the bond's "
1915 "HW address matches its VLANs'.\n",
1919 printk(KERN_INFO DRV_NAME
1920 ": %s: released all slaves\n",
1924 write_unlock_bh(&bond->lock);
1930 * This function changes the active slave to slave <slave_dev>.
1931 * It returns -EINVAL in the following cases.
1932 * - <slave_dev> is not found in the list.
1933 * - There is not active slave now.
1934 * - <slave_dev> is already active.
1935 * - The link state of <slave_dev> is not BOND_LINK_UP.
1936 * - <slave_dev> is not running.
1937 * In these cases, this fuction does nothing.
1938 * In the other cases, currnt_slave pointer is changed and 0 is returned.
1940 static int bond_ioctl_change_active(struct net_device *bond_dev, struct net_device *slave_dev)
1942 struct bonding *bond = bond_dev->priv;
1943 struct slave *old_active = NULL;
1944 struct slave *new_active = NULL;
1947 if (!USES_PRIMARY(bond->params.mode)) {
1951 /* Verify that master_dev is indeed the master of slave_dev */
1952 if (!(slave_dev->flags & IFF_SLAVE) ||
1953 (slave_dev->master != bond_dev)) {
1957 write_lock_bh(&bond->lock);
1959 old_active = bond->curr_active_slave;
1960 new_active = bond_get_slave_by_dev(bond, slave_dev);
1963 * Changing to the current active: do nothing; return success.
1965 if (new_active && (new_active == old_active)) {
1966 write_unlock_bh(&bond->lock);
1972 (new_active->link == BOND_LINK_UP) &&
1973 IS_UP(new_active->dev)) {
1974 bond_change_active_slave(bond, new_active);
1979 write_unlock_bh(&bond->lock);
1984 static int bond_info_query(struct net_device *bond_dev, struct ifbond *info)
1986 struct bonding *bond = bond_dev->priv;
1988 info->bond_mode = bond->params.mode;
1989 info->miimon = bond->params.miimon;
1991 read_lock_bh(&bond->lock);
1992 info->num_slaves = bond->slave_cnt;
1993 read_unlock_bh(&bond->lock);
1998 static int bond_slave_info_query(struct net_device *bond_dev, struct ifslave *info)
2000 struct bonding *bond = bond_dev->priv;
2001 struct slave *slave;
2004 if (info->slave_id < 0) {
2008 read_lock_bh(&bond->lock);
2010 bond_for_each_slave(bond, slave, i) {
2011 if (i == (int)info->slave_id) {
2017 read_unlock_bh(&bond->lock);
2020 strcpy(info->slave_name, slave->dev->name);
2021 info->link = slave->link;
2022 info->state = slave->state;
2023 info->link_failure_count = slave->link_failure_count;
2031 /*-------------------------------- Monitoring -------------------------------*/
2033 /* this function is called regularly to monitor each slave's link. */
2034 void bond_mii_monitor(struct net_device *bond_dev)
2036 struct bonding *bond = bond_dev->priv;
2037 struct slave *slave, *oldcurrent;
2038 int do_failover = 0;
2042 read_lock(&bond->lock);
2044 delta_in_ticks = (bond->params.miimon * HZ) / 1000;
2046 if (bond->kill_timers) {
2050 if (bond->slave_cnt == 0) {
2054 /* we will try to read the link status of each of our slaves, and
2055 * set their IFF_RUNNING flag appropriately. For each slave not
2056 * supporting MII status, we won't do anything so that a user-space
2057 * program could monitor the link itself if needed.
2060 read_lock(&bond->curr_slave_lock);
2061 oldcurrent = bond->curr_active_slave;
2062 read_unlock(&bond->curr_slave_lock);
2064 bond_for_each_slave(bond, slave, i) {
2065 struct net_device *slave_dev = slave->dev;
2067 u16 old_speed = slave->speed;
2068 u8 old_duplex = slave->duplex;
2070 link_state = bond_check_dev_link(bond, slave_dev, 0);
2072 switch (slave->link) {
2073 case BOND_LINK_UP: /* the link was up */
2074 if (link_state == BMSR_LSTATUS) {
2075 /* link stays up, nothing more to do */
2077 } else { /* link going down */
2078 slave->link = BOND_LINK_FAIL;
2079 slave->delay = bond->params.downdelay;
2081 if (slave->link_failure_count < UINT_MAX) {
2082 slave->link_failure_count++;
2085 if (bond->params.downdelay) {
2086 printk(KERN_INFO DRV_NAME
2087 ": %s: link status down for %s "
2088 "interface %s, disabling it in "
2092 ? ((bond->params.mode == BOND_MODE_ACTIVEBACKUP)
2093 ? ((slave == oldcurrent)
2094 ? "active " : "backup ")
2098 bond->params.downdelay * bond->params.miimon);
2101 /* no break ! fall through the BOND_LINK_FAIL test to
2102 ensure proper action to be taken
2104 case BOND_LINK_FAIL: /* the link has just gone down */
2105 if (link_state != BMSR_LSTATUS) {
2106 /* link stays down */
2107 if (slave->delay <= 0) {
2108 /* link down for too long time */
2109 slave->link = BOND_LINK_DOWN;
2111 /* in active/backup mode, we must
2112 * completely disable this interface
2114 if ((bond->params.mode == BOND_MODE_ACTIVEBACKUP) ||
2115 (bond->params.mode == BOND_MODE_8023AD)) {
2116 bond_set_slave_inactive_flags(slave);
2119 printk(KERN_INFO DRV_NAME
2120 ": %s: link status definitely "
2121 "down for interface %s, "
2126 /* notify ad that the link status has changed */
2127 if (bond->params.mode == BOND_MODE_8023AD) {
2128 bond_3ad_handle_link_change(slave, BOND_LINK_DOWN);
2131 if ((bond->params.mode == BOND_MODE_TLB) ||
2132 (bond->params.mode == BOND_MODE_ALB)) {
2133 bond_alb_handle_link_change(bond, slave, BOND_LINK_DOWN);
2136 if (slave == oldcurrent) {
2144 slave->link = BOND_LINK_UP;
2145 slave->jiffies = jiffies;
2146 printk(KERN_INFO DRV_NAME
2147 ": %s: link status up again after %d "
2148 "ms for interface %s.\n",
2150 (bond->params.downdelay - slave->delay) * bond->params.miimon,
2154 case BOND_LINK_DOWN: /* the link was down */
2155 if (link_state != BMSR_LSTATUS) {
2156 /* the link stays down, nothing more to do */
2158 } else { /* link going up */
2159 slave->link = BOND_LINK_BACK;
2160 slave->delay = bond->params.updelay;
2162 if (bond->params.updelay) {
2163 /* if updelay == 0, no need to
2164 advertise about a 0 ms delay */
2165 printk(KERN_INFO DRV_NAME
2166 ": %s: link status up for "
2167 "interface %s, enabling it "
2171 bond->params.updelay * bond->params.miimon);
2174 /* no break ! fall through the BOND_LINK_BACK state in
2175 case there's something to do.
2177 case BOND_LINK_BACK: /* the link has just come back */
2178 if (link_state != BMSR_LSTATUS) {
2179 /* link down again */
2180 slave->link = BOND_LINK_DOWN;
2182 printk(KERN_INFO DRV_NAME
2183 ": %s: link status down again after %d "
2184 "ms for interface %s.\n",
2186 (bond->params.updelay - slave->delay) * bond->params.miimon,
2190 if (slave->delay == 0) {
2191 /* now the link has been up for long time enough */
2192 slave->link = BOND_LINK_UP;
2193 slave->jiffies = jiffies;
2195 if (bond->params.mode == BOND_MODE_8023AD) {
2196 /* prevent it from being the active one */
2197 slave->state = BOND_STATE_BACKUP;
2198 } else if (bond->params.mode != BOND_MODE_ACTIVEBACKUP) {
2199 /* make it immediately active */
2200 slave->state = BOND_STATE_ACTIVE;
2201 } else if (slave != bond->primary_slave) {
2202 /* prevent it from being the active one */
2203 slave->state = BOND_STATE_BACKUP;
2206 printk(KERN_INFO DRV_NAME
2207 ": %s: link status definitely "
2208 "up for interface %s.\n",
2212 /* notify ad that the link status has changed */
2213 if (bond->params.mode == BOND_MODE_8023AD) {
2214 bond_3ad_handle_link_change(slave, BOND_LINK_UP);
2217 if ((bond->params.mode == BOND_MODE_TLB) ||
2218 (bond->params.mode == BOND_MODE_ALB)) {
2219 bond_alb_handle_link_change(bond, slave, BOND_LINK_UP);
2222 if ((!oldcurrent) ||
2223 (slave == bond->primary_slave)) {
2232 /* Should not happen */
2233 printk(KERN_ERR DRV_NAME
2234 ": %s: Error: %s Illegal value (link=%d)\n",
2239 } /* end of switch (slave->link) */
2241 bond_update_speed_duplex(slave);
2243 if (bond->params.mode == BOND_MODE_8023AD) {
2244 if (old_speed != slave->speed) {
2245 bond_3ad_adapter_speed_changed(slave);
2248 if (old_duplex != slave->duplex) {
2249 bond_3ad_adapter_duplex_changed(slave);
2256 write_lock(&bond->curr_slave_lock);
2258 bond_select_active_slave(bond);
2260 write_unlock(&bond->curr_slave_lock);
2262 bond_set_carrier(bond);
2265 if (bond->params.miimon) {
2266 mod_timer(&bond->mii_timer, jiffies + delta_in_ticks);
2269 read_unlock(&bond->lock);
2273 static u32 bond_glean_dev_ip(struct net_device *dev)
2275 struct in_device *idev;
2276 struct in_ifaddr *ifa;
2283 idev = __in_dev_get_rcu(dev);
2287 ifa = idev->ifa_list;
2291 addr = ifa->ifa_local;
2297 static int bond_has_ip(struct bonding *bond)
2299 struct vlan_entry *vlan, *vlan_next;
2301 if (bond->master_ip)
2304 if (list_empty(&bond->vlan_list))
2307 list_for_each_entry_safe(vlan, vlan_next, &bond->vlan_list,
2316 static int bond_has_this_ip(struct bonding *bond, u32 ip)
2318 struct vlan_entry *vlan, *vlan_next;
2320 if (ip == bond->master_ip)
2323 if (list_empty(&bond->vlan_list))
2326 list_for_each_entry_safe(vlan, vlan_next, &bond->vlan_list,
2328 if (ip == vlan->vlan_ip)
2336 * We go to the (large) trouble of VLAN tagging ARP frames because
2337 * switches in VLAN mode (especially if ports are configured as
2338 * "native" to a VLAN) might not pass non-tagged frames.
2340 static void bond_arp_send(struct net_device *slave_dev, int arp_op, u32 dest_ip, u32 src_ip, unsigned short vlan_id)
2342 struct sk_buff *skb;
2344 dprintk("arp %d on slave %s: dst %x src %x vid %d\n", arp_op,
2345 slave_dev->name, dest_ip, src_ip, vlan_id);
2347 skb = arp_create(arp_op, ETH_P_ARP, dest_ip, slave_dev, src_ip,
2348 NULL, slave_dev->dev_addr, NULL);
2351 printk(KERN_ERR DRV_NAME ": ARP packet allocation failed\n");
2355 skb = vlan_put_tag(skb, vlan_id);
2357 printk(KERN_ERR DRV_NAME ": failed to insert VLAN tag\n");
2365 static void bond_arp_send_all(struct bonding *bond, struct slave *slave)
2368 u32 *targets = bond->params.arp_targets;
2369 struct vlan_entry *vlan, *vlan_next;
2370 struct net_device *vlan_dev;
2374 for (i = 0; (i < BOND_MAX_ARP_TARGETS); i++) {
2377 dprintk("basa: target %x\n", targets[i]);
2378 if (list_empty(&bond->vlan_list)) {
2379 dprintk("basa: empty vlan: arp_send\n");
2380 bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2381 bond->master_ip, 0);
2386 * If VLANs are configured, we do a route lookup to
2387 * determine which VLAN interface would be used, so we
2388 * can tag the ARP with the proper VLAN tag.
2390 memset(&fl, 0, sizeof(fl));
2391 fl.fl4_dst = targets[i];
2392 fl.fl4_tos = RTO_ONLINK;
2394 rv = ip_route_output_key(&rt, &fl);
2396 if (net_ratelimit()) {
2397 printk(KERN_WARNING DRV_NAME
2398 ": %s: no route to arp_ip_target %u.%u.%u.%u\n",
2399 bond->dev->name, NIPQUAD(fl.fl4_dst));
2405 * This target is not on a VLAN
2407 if (rt->u.dst.dev == bond->dev) {
2409 dprintk("basa: rtdev == bond->dev: arp_send\n");
2410 bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2411 bond->master_ip, 0);
2416 list_for_each_entry_safe(vlan, vlan_next, &bond->vlan_list,
2418 vlan_dev = vlan_group_get_device(bond->vlgrp, vlan->vlan_id);
2419 if (vlan_dev == rt->u.dst.dev) {
2420 vlan_id = vlan->vlan_id;
2421 dprintk("basa: vlan match on %s %d\n",
2422 vlan_dev->name, vlan_id);
2429 bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2430 vlan->vlan_ip, vlan_id);
2434 if (net_ratelimit()) {
2435 printk(KERN_WARNING DRV_NAME
2436 ": %s: no path to arp_ip_target %u.%u.%u.%u via rt.dev %s\n",
2437 bond->dev->name, NIPQUAD(fl.fl4_dst),
2438 rt->u.dst.dev ? rt->u.dst.dev->name : "NULL");
2445 * Kick out a gratuitous ARP for an IP on the bonding master plus one
2446 * for each VLAN above us.
2448 static void bond_send_gratuitous_arp(struct bonding *bond)
2450 struct slave *slave = bond->curr_active_slave;
2451 struct vlan_entry *vlan;
2452 struct net_device *vlan_dev;
2454 dprintk("bond_send_grat_arp: bond %s slave %s\n", bond->dev->name,
2455 slave ? slave->dev->name : "NULL");
2459 if (bond->master_ip) {
2460 bond_arp_send(slave->dev, ARPOP_REPLY, bond->master_ip,
2461 bond->master_ip, 0);
2464 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
2465 vlan_dev = vlan_group_get_device(bond->vlgrp, vlan->vlan_id);
2466 if (vlan->vlan_ip) {
2467 bond_arp_send(slave->dev, ARPOP_REPLY, vlan->vlan_ip,
2468 vlan->vlan_ip, vlan->vlan_id);
2473 static void bond_validate_arp(struct bonding *bond, struct slave *slave, u32 sip, u32 tip)
2476 u32 *targets = bond->params.arp_targets;
2478 targets = bond->params.arp_targets;
2479 for (i = 0; (i < BOND_MAX_ARP_TARGETS) && targets[i]; i++) {
2480 dprintk("bva: sip %u.%u.%u.%u tip %u.%u.%u.%u t[%d] "
2481 "%u.%u.%u.%u bhti(tip) %d\n",
2482 NIPQUAD(sip), NIPQUAD(tip), i, NIPQUAD(targets[i]),
2483 bond_has_this_ip(bond, tip));
2484 if (sip == targets[i]) {
2485 if (bond_has_this_ip(bond, tip))
2486 slave->last_arp_rx = jiffies;
2492 static int bond_arp_rcv(struct sk_buff *skb, struct net_device *dev, struct packet_type *pt, struct net_device *orig_dev)
2495 struct slave *slave;
2496 struct bonding *bond;
2497 unsigned char *arp_ptr;
2500 if (!(dev->priv_flags & IFF_BONDING) || !(dev->flags & IFF_MASTER))
2504 read_lock(&bond->lock);
2506 dprintk("bond_arp_rcv: bond %s skb->dev %s orig_dev %s\n",
2507 bond->dev->name, skb->dev ? skb->dev->name : "NULL",
2508 orig_dev ? orig_dev->name : "NULL");
2510 slave = bond_get_slave_by_dev(bond, orig_dev);
2511 if (!slave || !slave_do_arp_validate(bond, slave))
2514 /* ARP header, plus 2 device addresses, plus 2 IP addresses. */
2515 if (!pskb_may_pull(skb, (sizeof(struct arphdr) +
2516 (2 * dev->addr_len) +
2517 (2 * sizeof(u32)))))
2521 if (arp->ar_hln != dev->addr_len ||
2522 skb->pkt_type == PACKET_OTHERHOST ||
2523 skb->pkt_type == PACKET_LOOPBACK ||
2524 arp->ar_hrd != htons(ARPHRD_ETHER) ||
2525 arp->ar_pro != htons(ETH_P_IP) ||
2529 arp_ptr = (unsigned char *)(arp + 1);
2530 arp_ptr += dev->addr_len;
2531 memcpy(&sip, arp_ptr, 4);
2532 arp_ptr += 4 + dev->addr_len;
2533 memcpy(&tip, arp_ptr, 4);
2535 dprintk("bond_arp_rcv: %s %s/%d av %d sv %d sip %u.%u.%u.%u"
2536 " tip %u.%u.%u.%u\n", bond->dev->name, slave->dev->name,
2537 slave->state, bond->params.arp_validate,
2538 slave_do_arp_validate(bond, slave), NIPQUAD(sip), NIPQUAD(tip));
2541 * Backup slaves won't see the ARP reply, but do come through
2542 * here for each ARP probe (so we swap the sip/tip to validate
2543 * the probe). In a "redundant switch, common router" type of
2544 * configuration, the ARP probe will (hopefully) travel from
2545 * the active, through one switch, the router, then the other
2546 * switch before reaching the backup.
2548 if (slave->state == BOND_STATE_ACTIVE)
2549 bond_validate_arp(bond, slave, sip, tip);
2551 bond_validate_arp(bond, slave, tip, sip);
2554 read_unlock(&bond->lock);
2557 return NET_RX_SUCCESS;
2561 * this function is called regularly to monitor each slave's link
2562 * ensuring that traffic is being sent and received when arp monitoring
2563 * is used in load-balancing mode. if the adapter has been dormant, then an
2564 * arp is transmitted to generate traffic. see activebackup_arp_monitor for
2565 * arp monitoring in active backup mode.
2567 void bond_loadbalance_arp_mon(struct net_device *bond_dev)
2569 struct bonding *bond = bond_dev->priv;
2570 struct slave *slave, *oldcurrent;
2571 int do_failover = 0;
2575 read_lock(&bond->lock);
2577 delta_in_ticks = (bond->params.arp_interval * HZ) / 1000;
2579 if (bond->kill_timers) {
2583 if (bond->slave_cnt == 0) {
2587 read_lock(&bond->curr_slave_lock);
2588 oldcurrent = bond->curr_active_slave;
2589 read_unlock(&bond->curr_slave_lock);
2591 /* see if any of the previous devices are up now (i.e. they have
2592 * xmt and rcv traffic). the curr_active_slave does not come into
2593 * the picture unless it is null. also, slave->jiffies is not needed
2594 * here because we send an arp on each slave and give a slave as
2595 * long as it needs to get the tx/rx within the delta.
2596 * TODO: what about up/down delay in arp mode? it wasn't here before
2599 bond_for_each_slave(bond, slave, i) {
2600 if (slave->link != BOND_LINK_UP) {
2601 if (((jiffies - slave->dev->trans_start) <= delta_in_ticks) &&
2602 ((jiffies - slave->dev->last_rx) <= delta_in_ticks)) {
2604 slave->link = BOND_LINK_UP;
2605 slave->state = BOND_STATE_ACTIVE;
2607 /* primary_slave has no meaning in round-robin
2608 * mode. the window of a slave being up and
2609 * curr_active_slave being null after enslaving
2613 printk(KERN_INFO DRV_NAME
2614 ": %s: link status definitely "
2615 "up for interface %s, ",
2620 printk(KERN_INFO DRV_NAME
2621 ": %s: interface %s is now up\n",
2627 /* slave->link == BOND_LINK_UP */
2629 /* not all switches will respond to an arp request
2630 * when the source ip is 0, so don't take the link down
2631 * if we don't know our ip yet
2633 if (((jiffies - slave->dev->trans_start) >= (2*delta_in_ticks)) ||
2634 (((jiffies - slave->dev->last_rx) >= (2*delta_in_ticks)) &&
2635 bond_has_ip(bond))) {
2637 slave->link = BOND_LINK_DOWN;
2638 slave->state = BOND_STATE_BACKUP;
2640 if (slave->link_failure_count < UINT_MAX) {
2641 slave->link_failure_count++;
2644 printk(KERN_INFO DRV_NAME
2645 ": %s: interface %s is now down.\n",
2649 if (slave == oldcurrent) {
2655 /* note: if switch is in round-robin mode, all links
2656 * must tx arp to ensure all links rx an arp - otherwise
2657 * links may oscillate or not come up at all; if switch is
2658 * in something like xor mode, there is nothing we can
2659 * do - all replies will be rx'ed on same link causing slaves
2660 * to be unstable during low/no traffic periods
2662 if (IS_UP(slave->dev)) {
2663 bond_arp_send_all(bond, slave);
2668 write_lock(&bond->curr_slave_lock);
2670 bond_select_active_slave(bond);
2672 write_unlock(&bond->curr_slave_lock);
2676 if (bond->params.arp_interval) {
2677 mod_timer(&bond->arp_timer, jiffies + delta_in_ticks);
2680 read_unlock(&bond->lock);
2684 * When using arp monitoring in active-backup mode, this function is
2685 * called to determine if any backup slaves have went down or a new
2686 * current slave needs to be found.
2687 * The backup slaves never generate traffic, they are considered up by merely
2688 * receiving traffic. If the current slave goes down, each backup slave will
2689 * be given the opportunity to tx/rx an arp before being taken down - this
2690 * prevents all slaves from being taken down due to the current slave not
2691 * sending any traffic for the backups to receive. The arps are not necessarily
2692 * necessary, any tx and rx traffic will keep the current slave up. While any
2693 * rx traffic will keep the backup slaves up, the current slave is responsible
2694 * for generating traffic to keep them up regardless of any other traffic they
2695 * may have received.
2696 * see loadbalance_arp_monitor for arp monitoring in load balancing mode
2698 void bond_activebackup_arp_mon(struct net_device *bond_dev)
2700 struct bonding *bond = bond_dev->priv;
2701 struct slave *slave;
2705 read_lock(&bond->lock);
2707 delta_in_ticks = (bond->params.arp_interval * HZ) / 1000;
2709 if (bond->kill_timers) {
2713 if (bond->slave_cnt == 0) {
2717 /* determine if any slave has come up or any backup slave has
2719 * TODO: what about up/down delay in arp mode? it wasn't here before
2722 bond_for_each_slave(bond, slave, i) {
2723 if (slave->link != BOND_LINK_UP) {
2724 if ((jiffies - slave_last_rx(bond, slave)) <=
2727 slave->link = BOND_LINK_UP;
2729 write_lock(&bond->curr_slave_lock);
2731 if ((!bond->curr_active_slave) &&
2732 ((jiffies - slave->dev->trans_start) <= delta_in_ticks)) {
2733 bond_change_active_slave(bond, slave);
2734 bond->current_arp_slave = NULL;
2735 } else if (bond->curr_active_slave != slave) {
2736 /* this slave has just come up but we
2737 * already have a current slave; this
2738 * can also happen if bond_enslave adds
2739 * a new slave that is up while we are
2740 * searching for a new slave
2742 bond_set_slave_inactive_flags(slave);
2743 bond->current_arp_slave = NULL;
2746 bond_set_carrier(bond);
2748 if (slave == bond->curr_active_slave) {
2749 printk(KERN_INFO DRV_NAME
2750 ": %s: %s is up and now the "
2751 "active interface\n",
2754 netif_carrier_on(bond->dev);
2756 printk(KERN_INFO DRV_NAME
2757 ": %s: backup interface %s is "
2763 write_unlock(&bond->curr_slave_lock);
2766 read_lock(&bond->curr_slave_lock);
2768 if ((slave != bond->curr_active_slave) &&
2769 (!bond->current_arp_slave) &&
2770 (((jiffies - slave_last_rx(bond, slave)) >= 3*delta_in_ticks) &&
2771 bond_has_ip(bond))) {
2772 /* a backup slave has gone down; three times
2773 * the delta allows the current slave to be
2774 * taken out before the backup slave.
2775 * note: a non-null current_arp_slave indicates
2776 * the curr_active_slave went down and we are
2777 * searching for a new one; under this
2778 * condition we only take the curr_active_slave
2779 * down - this gives each slave a chance to
2780 * tx/rx traffic before being taken out
2783 read_unlock(&bond->curr_slave_lock);
2785 slave->link = BOND_LINK_DOWN;
2787 if (slave->link_failure_count < UINT_MAX) {
2788 slave->link_failure_count++;
2791 bond_set_slave_inactive_flags(slave);
2793 printk(KERN_INFO DRV_NAME
2794 ": %s: backup interface %s is now down\n",
2798 read_unlock(&bond->curr_slave_lock);
2803 read_lock(&bond->curr_slave_lock);
2804 slave = bond->curr_active_slave;
2805 read_unlock(&bond->curr_slave_lock);
2808 /* if we have sent traffic in the past 2*arp_intervals but
2809 * haven't xmit and rx traffic in that time interval, select
2810 * a different slave. slave->jiffies is only updated when
2811 * a slave first becomes the curr_active_slave - not necessarily
2812 * after every arp; this ensures the slave has a full 2*delta
2813 * before being taken out. if a primary is being used, check
2814 * if it is up and needs to take over as the curr_active_slave
2816 if ((((jiffies - slave->dev->trans_start) >= (2*delta_in_ticks)) ||
2817 (((jiffies - slave_last_rx(bond, slave)) >= (2*delta_in_ticks)) &&
2818 bond_has_ip(bond))) &&
2819 ((jiffies - slave->jiffies) >= 2*delta_in_ticks)) {
2821 slave->link = BOND_LINK_DOWN;
2823 if (slave->link_failure_count < UINT_MAX) {
2824 slave->link_failure_count++;
2827 printk(KERN_INFO DRV_NAME
2828 ": %s: link status down for active interface "
2829 "%s, disabling it\n",
2833 write_lock(&bond->curr_slave_lock);
2835 bond_select_active_slave(bond);
2836 slave = bond->curr_active_slave;
2838 write_unlock(&bond->curr_slave_lock);
2840 bond->current_arp_slave = slave;
2843 slave->jiffies = jiffies;
2845 } else if ((bond->primary_slave) &&
2846 (bond->primary_slave != slave) &&
2847 (bond->primary_slave->link == BOND_LINK_UP)) {
2848 /* at this point, slave is the curr_active_slave */
2849 printk(KERN_INFO DRV_NAME
2850 ": %s: changing from interface %s to primary "
2854 bond->primary_slave->dev->name);
2856 /* primary is up so switch to it */
2857 write_lock(&bond->curr_slave_lock);
2858 bond_change_active_slave(bond, bond->primary_slave);
2859 write_unlock(&bond->curr_slave_lock);
2861 slave = bond->primary_slave;
2862 slave->jiffies = jiffies;
2864 bond->current_arp_slave = NULL;
2867 /* the current slave must tx an arp to ensure backup slaves
2870 if (slave && bond_has_ip(bond)) {
2871 bond_arp_send_all(bond, slave);
2875 /* if we don't have a curr_active_slave, search for the next available
2876 * backup slave from the current_arp_slave and make it the candidate
2877 * for becoming the curr_active_slave
2880 if (!bond->current_arp_slave) {
2881 bond->current_arp_slave = bond->first_slave;
2884 if (bond->current_arp_slave) {
2885 bond_set_slave_inactive_flags(bond->current_arp_slave);
2887 /* search for next candidate */
2888 bond_for_each_slave_from(bond, slave, i, bond->current_arp_slave->next) {
2889 if (IS_UP(slave->dev)) {
2890 slave->link = BOND_LINK_BACK;
2891 bond_set_slave_active_flags(slave);
2892 bond_arp_send_all(bond, slave);
2893 slave->jiffies = jiffies;
2894 bond->current_arp_slave = slave;
2898 /* if the link state is up at this point, we
2899 * mark it down - this can happen if we have
2900 * simultaneous link failures and
2901 * reselect_active_interface doesn't make this
2902 * one the current slave so it is still marked
2903 * up when it is actually down
2905 if (slave->link == BOND_LINK_UP) {
2906 slave->link = BOND_LINK_DOWN;
2907 if (slave->link_failure_count < UINT_MAX) {
2908 slave->link_failure_count++;
2911 bond_set_slave_inactive_flags(slave);
2913 printk(KERN_INFO DRV_NAME
2914 ": %s: backup interface %s is "
2924 if (bond->params.arp_interval) {
2925 mod_timer(&bond->arp_timer, jiffies + delta_in_ticks);
2928 read_unlock(&bond->lock);
2931 /*------------------------------ proc/seq_file-------------------------------*/
2933 #ifdef CONFIG_PROC_FS
2935 #define SEQ_START_TOKEN ((void *)1)
2937 static void *bond_info_seq_start(struct seq_file *seq, loff_t *pos)
2939 struct bonding *bond = seq->private;
2941 struct slave *slave;
2944 /* make sure the bond won't be taken away */
2945 read_lock(&dev_base_lock);
2946 read_lock_bh(&bond->lock);
2949 return SEQ_START_TOKEN;
2952 bond_for_each_slave(bond, slave, i) {
2953 if (++off == *pos) {
2961 static void *bond_info_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2963 struct bonding *bond = seq->private;
2964 struct slave *slave = v;
2967 if (v == SEQ_START_TOKEN) {
2968 return bond->first_slave;
2971 slave = slave->next;
2973 return (slave == bond->first_slave) ? NULL : slave;
2976 static void bond_info_seq_stop(struct seq_file *seq, void *v)
2978 struct bonding *bond = seq->private;
2980 read_unlock_bh(&bond->lock);
2981 read_unlock(&dev_base_lock);
2984 static void bond_info_show_master(struct seq_file *seq)
2986 struct bonding *bond = seq->private;
2991 read_lock(&bond->curr_slave_lock);
2992 curr = bond->curr_active_slave;
2993 read_unlock(&bond->curr_slave_lock);
2995 seq_printf(seq, "Bonding Mode: %s\n",
2996 bond_mode_name(bond->params.mode));
2998 if (bond->params.mode == BOND_MODE_XOR ||
2999 bond->params.mode == BOND_MODE_8023AD) {
3000 seq_printf(seq, "Transmit Hash Policy: %s (%d)\n",
3001 xmit_hashtype_tbl[bond->params.xmit_policy].modename,
3002 bond->params.xmit_policy);
3005 if (USES_PRIMARY(bond->params.mode)) {
3006 seq_printf(seq, "Primary Slave: %s\n",
3007 (bond->primary_slave) ?
3008 bond->primary_slave->dev->name : "None");
3010 seq_printf(seq, "Currently Active Slave: %s\n",
3011 (curr) ? curr->dev->name : "None");
3014 seq_printf(seq, "MII Status: %s\n", netif_carrier_ok(bond->dev) ?
3016 seq_printf(seq, "MII Polling Interval (ms): %d\n", bond->params.miimon);
3017 seq_printf(seq, "Up Delay (ms): %d\n",
3018 bond->params.updelay * bond->params.miimon);
3019 seq_printf(seq, "Down Delay (ms): %d\n",
3020 bond->params.downdelay * bond->params.miimon);
3023 /* ARP information */
3024 if(bond->params.arp_interval > 0) {
3026 seq_printf(seq, "ARP Polling Interval (ms): %d\n",
3027 bond->params.arp_interval);
3029 seq_printf(seq, "ARP IP target/s (n.n.n.n form):");
3031 for(i = 0; (i < BOND_MAX_ARP_TARGETS) ;i++) {
3032 if (!bond->params.arp_targets[i])
3035 seq_printf(seq, ",");
3036 target = ntohl(bond->params.arp_targets[i]);
3037 seq_printf(seq, " %d.%d.%d.%d", HIPQUAD(target));
3040 seq_printf(seq, "\n");
3043 if (bond->params.mode == BOND_MODE_8023AD) {
3044 struct ad_info ad_info;
3046 seq_puts(seq, "\n802.3ad info\n");
3047 seq_printf(seq, "LACP rate: %s\n",
3048 (bond->params.lacp_fast) ? "fast" : "slow");
3050 if (bond_3ad_get_active_agg_info(bond, &ad_info)) {
3051 seq_printf(seq, "bond %s has no active aggregator\n",
3054 seq_printf(seq, "Active Aggregator Info:\n");
3056 seq_printf(seq, "\tAggregator ID: %d\n",
3057 ad_info.aggregator_id);
3058 seq_printf(seq, "\tNumber of ports: %d\n",
3060 seq_printf(seq, "\tActor Key: %d\n",
3062 seq_printf(seq, "\tPartner Key: %d\n",
3063 ad_info.partner_key);
3064 seq_printf(seq, "\tPartner Mac Address: %02x:%02x:%02x:%02x:%02x:%02x\n",
3065 ad_info.partner_system[0],
3066 ad_info.partner_system[1],
3067 ad_info.partner_system[2],
3068 ad_info.partner_system[3],
3069 ad_info.partner_system[4],
3070 ad_info.partner_system[5]);
3075 static void bond_info_show_slave(struct seq_file *seq, const struct slave *slave)
3077 struct bonding *bond = seq->private;
3079 seq_printf(seq, "\nSlave Interface: %s\n", slave->dev->name);
3080 seq_printf(seq, "MII Status: %s\n",
3081 (slave->link == BOND_LINK_UP) ? "up" : "down");
3082 seq_printf(seq, "Link Failure Count: %u\n",
3083 slave->link_failure_count);
3086 "Permanent HW addr: %02x:%02x:%02x:%02x:%02x:%02x\n",
3087 slave->perm_hwaddr[0], slave->perm_hwaddr[1],
3088 slave->perm_hwaddr[2], slave->perm_hwaddr[3],
3089 slave->perm_hwaddr[4], slave->perm_hwaddr[5]);
3091 if (bond->params.mode == BOND_MODE_8023AD) {
3092 const struct aggregator *agg
3093 = SLAVE_AD_INFO(slave).port.aggregator;
3096 seq_printf(seq, "Aggregator ID: %d\n",
3097 agg->aggregator_identifier);
3099 seq_puts(seq, "Aggregator ID: N/A\n");
3104 static int bond_info_seq_show(struct seq_file *seq, void *v)
3106 if (v == SEQ_START_TOKEN) {
3107 seq_printf(seq, "%s\n", version);
3108 bond_info_show_master(seq);
3110 bond_info_show_slave(seq, v);
3116 static struct seq_operations bond_info_seq_ops = {
3117 .start = bond_info_seq_start,
3118 .next = bond_info_seq_next,
3119 .stop = bond_info_seq_stop,
3120 .show = bond_info_seq_show,
3123 static int bond_info_open(struct inode *inode, struct file *file)
3125 struct seq_file *seq;
3126 struct proc_dir_entry *proc;
3129 res = seq_open(file, &bond_info_seq_ops);
3131 /* recover the pointer buried in proc_dir_entry data */
3132 seq = file->private_data;
3134 seq->private = proc->data;
3140 static const struct file_operations bond_info_fops = {
3141 .owner = THIS_MODULE,
3142 .open = bond_info_open,
3144 .llseek = seq_lseek,
3145 .release = seq_release,
3148 static int bond_create_proc_entry(struct bonding *bond)
3150 struct net_device *bond_dev = bond->dev;
3152 if (bond_proc_dir) {
3153 bond->proc_entry = create_proc_entry(bond_dev->name,
3156 if (bond->proc_entry == NULL) {
3157 printk(KERN_WARNING DRV_NAME
3158 ": Warning: Cannot create /proc/net/%s/%s\n",
3159 DRV_NAME, bond_dev->name);
3161 bond->proc_entry->data = bond;
3162 bond->proc_entry->proc_fops = &bond_info_fops;
3163 bond->proc_entry->owner = THIS_MODULE;
3164 memcpy(bond->proc_file_name, bond_dev->name, IFNAMSIZ);
3171 static void bond_remove_proc_entry(struct bonding *bond)
3173 if (bond_proc_dir && bond->proc_entry) {
3174 remove_proc_entry(bond->proc_file_name, bond_proc_dir);
3175 memset(bond->proc_file_name, 0, IFNAMSIZ);
3176 bond->proc_entry = NULL;
3180 /* Create the bonding directory under /proc/net, if doesn't exist yet.
3181 * Caller must hold rtnl_lock.
3183 static void bond_create_proc_dir(void)
3185 int len = strlen(DRV_NAME);
3187 for (bond_proc_dir = proc_net->subdir; bond_proc_dir;
3188 bond_proc_dir = bond_proc_dir->next) {
3189 if ((bond_proc_dir->namelen == len) &&
3190 !memcmp(bond_proc_dir->name, DRV_NAME, len)) {
3195 if (!bond_proc_dir) {
3196 bond_proc_dir = proc_mkdir(DRV_NAME, proc_net);
3197 if (bond_proc_dir) {
3198 bond_proc_dir->owner = THIS_MODULE;
3200 printk(KERN_WARNING DRV_NAME
3201 ": Warning: cannot create /proc/net/%s\n",
3207 /* Destroy the bonding directory under /proc/net, if empty.
3208 * Caller must hold rtnl_lock.
3210 static void bond_destroy_proc_dir(void)
3212 struct proc_dir_entry *de;
3214 if (!bond_proc_dir) {
3218 /* verify that the /proc dir is empty */
3219 for (de = bond_proc_dir->subdir; de; de = de->next) {
3220 /* ignore . and .. */
3221 if (*(de->name) != '.') {
3227 if (bond_proc_dir->owner == THIS_MODULE) {
3228 bond_proc_dir->owner = NULL;
3231 remove_proc_entry(DRV_NAME, proc_net);
3232 bond_proc_dir = NULL;
3235 #endif /* CONFIG_PROC_FS */
3237 /*-------------------------- netdev event handling --------------------------*/
3240 * Change device name
3242 static int bond_event_changename(struct bonding *bond)
3244 #ifdef CONFIG_PROC_FS
3245 bond_remove_proc_entry(bond);
3246 bond_create_proc_entry(bond);
3248 down_write(&(bonding_rwsem));
3249 bond_destroy_sysfs_entry(bond);
3250 bond_create_sysfs_entry(bond);
3251 up_write(&(bonding_rwsem));
3255 static int bond_master_netdev_event(unsigned long event, struct net_device *bond_dev)
3257 struct bonding *event_bond = bond_dev->priv;
3260 case NETDEV_CHANGENAME:
3261 return bond_event_changename(event_bond);
3262 case NETDEV_UNREGISTER:
3264 * TODO: remove a bond from the list?
3274 static int bond_slave_netdev_event(unsigned long event, struct net_device *slave_dev)
3276 struct net_device *bond_dev = slave_dev->master;
3277 struct bonding *bond = bond_dev->priv;
3280 case NETDEV_UNREGISTER:
3282 bond_release(bond_dev, slave_dev);
3287 * TODO: is this what we get if somebody
3288 * sets up a hierarchical bond, then rmmod's
3289 * one of the slave bonding devices?
3294 * ... Or is it this?
3297 case NETDEV_CHANGEMTU:
3299 * TODO: Should slaves be allowed to
3300 * independently alter their MTU? For
3301 * an active-backup bond, slaves need
3302 * not be the same type of device, so
3303 * MTUs may vary. For other modes,
3304 * slaves arguably should have the
3305 * same MTUs. To do this, we'd need to
3306 * take over the slave's change_mtu
3307 * function for the duration of their
3311 case NETDEV_CHANGENAME:
3313 * TODO: handle changing the primary's name
3316 case NETDEV_FEAT_CHANGE:
3317 bond_compute_features(bond);
3327 * bond_netdev_event: handle netdev notifier chain events.
3329 * This function receives events for the netdev chain. The caller (an
3330 * ioctl handler calling blocking_notifier_call_chain) holds the necessary
3331 * locks for us to safely manipulate the slave devices (RTNL lock,
3334 static int bond_netdev_event(struct notifier_block *this, unsigned long event, void *ptr)
3336 struct net_device *event_dev = (struct net_device *)ptr;
3338 dprintk("event_dev: %s, event: %lx\n",
3339 (event_dev ? event_dev->name : "None"),
3342 if (!(event_dev->priv_flags & IFF_BONDING))
3345 if (event_dev->flags & IFF_MASTER) {
3346 dprintk("IFF_MASTER\n");
3347 return bond_master_netdev_event(event, event_dev);
3350 if (event_dev->flags & IFF_SLAVE) {
3351 dprintk("IFF_SLAVE\n");
3352 return bond_slave_netdev_event(event, event_dev);
3359 * bond_inetaddr_event: handle inetaddr notifier chain events.
3361 * We keep track of device IPs primarily to use as source addresses in
3362 * ARP monitor probes (rather than spewing out broadcasts all the time).
3364 * We track one IP for the main device (if it has one), plus one per VLAN.
3366 static int bond_inetaddr_event(struct notifier_block *this, unsigned long event, void *ptr)
3368 struct in_ifaddr *ifa = ptr;
3369 struct net_device *vlan_dev, *event_dev = ifa->ifa_dev->dev;
3370 struct bonding *bond, *bond_next;
3371 struct vlan_entry *vlan, *vlan_next;
3373 list_for_each_entry_safe(bond, bond_next, &bond_dev_list, bond_list) {
3374 if (bond->dev == event_dev) {
3377 bond->master_ip = ifa->ifa_local;
3380 bond->master_ip = bond_glean_dev_ip(bond->dev);
3387 if (list_empty(&bond->vlan_list))
3390 list_for_each_entry_safe(vlan, vlan_next, &bond->vlan_list,
3392 vlan_dev = vlan_group_get_device(bond->vlgrp, vlan->vlan_id);
3393 if (vlan_dev == event_dev) {
3396 vlan->vlan_ip = ifa->ifa_local;
3400 bond_glean_dev_ip(vlan_dev);
3411 static struct notifier_block bond_netdev_notifier = {
3412 .notifier_call = bond_netdev_event,
3415 static struct notifier_block bond_inetaddr_notifier = {
3416 .notifier_call = bond_inetaddr_event,
3419 /*-------------------------- Packet type handling ---------------------------*/
3421 /* register to receive lacpdus on a bond */
3422 static void bond_register_lacpdu(struct bonding *bond)
3424 struct packet_type *pk_type = &(BOND_AD_INFO(bond).ad_pkt_type);
3426 /* initialize packet type */
3427 pk_type->type = PKT_TYPE_LACPDU;
3428 pk_type->dev = bond->dev;
3429 pk_type->func = bond_3ad_lacpdu_recv;
3431 dev_add_pack(pk_type);
3434 /* unregister to receive lacpdus on a bond */
3435 static void bond_unregister_lacpdu(struct bonding *bond)
3437 dev_remove_pack(&(BOND_AD_INFO(bond).ad_pkt_type));
3440 void bond_register_arp(struct bonding *bond)
3442 struct packet_type *pt = &bond->arp_mon_pt;
3447 pt->type = htons(ETH_P_ARP);
3448 pt->dev = bond->dev;
3449 pt->func = bond_arp_rcv;
3453 void bond_unregister_arp(struct bonding *bond)
3455 struct packet_type *pt = &bond->arp_mon_pt;
3457 dev_remove_pack(pt);
3461 /*---------------------------- Hashing Policies -----------------------------*/
3464 * Hash for the output device based upon layer 3 and layer 4 data. If
3465 * the packet is a frag or not TCP or UDP, just use layer 3 data. If it is
3466 * altogether not IP, mimic bond_xmit_hash_policy_l2()
3468 static int bond_xmit_hash_policy_l34(struct sk_buff *skb,
3469 struct net_device *bond_dev, int count)
3471 struct ethhdr *data = (struct ethhdr *)skb->data;
3472 struct iphdr *iph = ip_hdr(skb);
3473 u16 *layer4hdr = (u16 *)((u32 *)iph + iph->ihl);
3476 if (skb->protocol == __constant_htons(ETH_P_IP)) {
3477 if (!(iph->frag_off & __constant_htons(IP_MF|IP_OFFSET)) &&
3478 (iph->protocol == IPPROTO_TCP ||
3479 iph->protocol == IPPROTO_UDP)) {
3480 layer4_xor = htons((*layer4hdr ^ *(layer4hdr + 1)));
3482 return (layer4_xor ^
3483 ((ntohl(iph->saddr ^ iph->daddr)) & 0xffff)) % count;
3487 return (data->h_dest[5] ^ bond_dev->dev_addr[5]) % count;
3491 * Hash for the output device based upon layer 2 data
3493 static int bond_xmit_hash_policy_l2(struct sk_buff *skb,
3494 struct net_device *bond_dev, int count)
3496 struct ethhdr *data = (struct ethhdr *)skb->data;
3498 return (data->h_dest[5] ^ bond_dev->dev_addr[5]) % count;
3501 /*-------------------------- Device entry points ----------------------------*/
3503 static int bond_open(struct net_device *bond_dev)
3505 struct bonding *bond = bond_dev->priv;
3506 struct timer_list *mii_timer = &bond->mii_timer;
3507 struct timer_list *arp_timer = &bond->arp_timer;
3509 bond->kill_timers = 0;
3511 if ((bond->params.mode == BOND_MODE_TLB) ||
3512 (bond->params.mode == BOND_MODE_ALB)) {
3513 struct timer_list *alb_timer = &(BOND_ALB_INFO(bond).alb_timer);
3515 /* bond_alb_initialize must be called before the timer
3518 if (bond_alb_initialize(bond, (bond->params.mode == BOND_MODE_ALB))) {
3519 /* something went wrong - fail the open operation */
3523 init_timer(alb_timer);
3524 alb_timer->expires = jiffies + 1;
3525 alb_timer->data = (unsigned long)bond;
3526 alb_timer->function = (void *)&bond_alb_monitor;
3527 add_timer(alb_timer);
3530 if (bond->params.miimon) { /* link check interval, in milliseconds. */
3531 init_timer(mii_timer);
3532 mii_timer->expires = jiffies + 1;
3533 mii_timer->data = (unsigned long)bond_dev;
3534 mii_timer->function = (void *)&bond_mii_monitor;
3535 add_timer(mii_timer);
3538 if (bond->params.arp_interval) { /* arp interval, in milliseconds. */
3539 init_timer(arp_timer);
3540 arp_timer->expires = jiffies + 1;
3541 arp_timer->data = (unsigned long)bond_dev;
3542 if (bond->params.mode == BOND_MODE_ACTIVEBACKUP) {
3543 arp_timer->function = (void *)&bond_activebackup_arp_mon;
3545 arp_timer->function = (void *)&bond_loadbalance_arp_mon;
3547 if (bond->params.arp_validate)
3548 bond_register_arp(bond);
3550 add_timer(arp_timer);
3553 if (bond->params.mode == BOND_MODE_8023AD) {
3554 struct timer_list *ad_timer = &(BOND_AD_INFO(bond).ad_timer);
3555 init_timer(ad_timer);
3556 ad_timer->expires = jiffies + 1;
3557 ad_timer->data = (unsigned long)bond;
3558 ad_timer->function = (void *)&bond_3ad_state_machine_handler;
3559 add_timer(ad_timer);
3561 /* register to receive LACPDUs */
3562 bond_register_lacpdu(bond);
3568 static int bond_close(struct net_device *bond_dev)
3570 struct bonding *bond = bond_dev->priv;
3572 if (bond->params.mode == BOND_MODE_8023AD) {
3573 /* Unregister the receive of LACPDUs */
3574 bond_unregister_lacpdu(bond);
3577 if (bond->params.arp_validate)
3578 bond_unregister_arp(bond);
3580 write_lock_bh(&bond->lock);
3583 /* signal timers not to re-arm */
3584 bond->kill_timers = 1;
3586 write_unlock_bh(&bond->lock);
3588 /* del_timer_sync must run without holding the bond->lock
3589 * because a running timer might be trying to hold it too
3592 if (bond->params.miimon) { /* link check interval, in milliseconds. */
3593 del_timer_sync(&bond->mii_timer);
3596 if (bond->params.arp_interval) { /* arp interval, in milliseconds. */
3597 del_timer_sync(&bond->arp_timer);
3600 switch (bond->params.mode) {
3601 case BOND_MODE_8023AD:
3602 del_timer_sync(&(BOND_AD_INFO(bond).ad_timer));
3606 del_timer_sync(&(BOND_ALB_INFO(bond).alb_timer));
3613 if ((bond->params.mode == BOND_MODE_TLB) ||
3614 (bond->params.mode == BOND_MODE_ALB)) {
3615 /* Must be called only after all
3616 * slaves have been released
3618 bond_alb_deinitialize(bond);
3624 static struct net_device_stats *bond_get_stats(struct net_device *bond_dev)
3626 struct bonding *bond = bond_dev->priv;
3627 struct net_device_stats *stats = &(bond->stats), *sstats;
3628 struct slave *slave;
3631 memset(stats, 0, sizeof(struct net_device_stats));
3633 read_lock_bh(&bond->lock);
3635 bond_for_each_slave(bond, slave, i) {
3636 sstats = slave->dev->get_stats(slave->dev);
3637 stats->rx_packets += sstats->rx_packets;
3638 stats->rx_bytes += sstats->rx_bytes;
3639 stats->rx_errors += sstats->rx_errors;
3640 stats->rx_dropped += sstats->rx_dropped;
3642 stats->tx_packets += sstats->tx_packets;
3643 stats->tx_bytes += sstats->tx_bytes;
3644 stats->tx_errors += sstats->tx_errors;
3645 stats->tx_dropped += sstats->tx_dropped;
3647 stats->multicast += sstats->multicast;
3648 stats->collisions += sstats->collisions;
3650 stats->rx_length_errors += sstats->rx_length_errors;
3651 stats->rx_over_errors += sstats->rx_over_errors;
3652 stats->rx_crc_errors += sstats->rx_crc_errors;
3653 stats->rx_frame_errors += sstats->rx_frame_errors;
3654 stats->rx_fifo_errors += sstats->rx_fifo_errors;
3655 stats->rx_missed_errors += sstats->rx_missed_errors;
3657 stats->tx_aborted_errors += sstats->tx_aborted_errors;
3658 stats->tx_carrier_errors += sstats->tx_carrier_errors;
3659 stats->tx_fifo_errors += sstats->tx_fifo_errors;
3660 stats->tx_heartbeat_errors += sstats->tx_heartbeat_errors;
3661 stats->tx_window_errors += sstats->tx_window_errors;
3664 read_unlock_bh(&bond->lock);
3669 static int bond_do_ioctl(struct net_device *bond_dev, struct ifreq *ifr, int cmd)
3671 struct net_device *slave_dev = NULL;
3672 struct ifbond k_binfo;
3673 struct ifbond __user *u_binfo = NULL;
3674 struct ifslave k_sinfo;
3675 struct ifslave __user *u_sinfo = NULL;
3676 struct mii_ioctl_data *mii = NULL;
3679 dprintk("bond_ioctl: master=%s, cmd=%d\n",
3680 bond_dev->name, cmd);
3692 * We do this again just in case we were called by SIOCGMIIREG
3693 * instead of SIOCGMIIPHY.
3700 if (mii->reg_num == 1) {
3701 struct bonding *bond = bond_dev->priv;
3703 read_lock_bh(&bond->lock);
3704 read_lock(&bond->curr_slave_lock);
3705 if (netif_carrier_ok(bond->dev)) {
3706 mii->val_out = BMSR_LSTATUS;
3708 read_unlock(&bond->curr_slave_lock);
3709 read_unlock_bh(&bond->lock);
3713 case BOND_INFO_QUERY_OLD:
3714 case SIOCBONDINFOQUERY:
3715 u_binfo = (struct ifbond __user *)ifr->ifr_data;
3717 if (copy_from_user(&k_binfo, u_binfo, sizeof(ifbond))) {
3721 res = bond_info_query(bond_dev, &k_binfo);
3723 if (copy_to_user(u_binfo, &k_binfo, sizeof(ifbond))) {
3729 case BOND_SLAVE_INFO_QUERY_OLD:
3730 case SIOCBONDSLAVEINFOQUERY:
3731 u_sinfo = (struct ifslave __user *)ifr->ifr_data;
3733 if (copy_from_user(&k_sinfo, u_sinfo, sizeof(ifslave))) {
3737 res = bond_slave_info_query(bond_dev, &k_sinfo);
3739 if (copy_to_user(u_sinfo, &k_sinfo, sizeof(ifslave))) {
3750 if (!capable(CAP_NET_ADMIN)) {
3754 down_write(&(bonding_rwsem));
3755 slave_dev = dev_get_by_name(ifr->ifr_slave);
3757 dprintk("slave_dev=%p: \n", slave_dev);
3762 dprintk("slave_dev->name=%s: \n", slave_dev->name);
3764 case BOND_ENSLAVE_OLD:
3765 case SIOCBONDENSLAVE:
3766 res = bond_enslave(bond_dev, slave_dev);
3768 case BOND_RELEASE_OLD:
3769 case SIOCBONDRELEASE:
3770 res = bond_release(bond_dev, slave_dev);
3772 case BOND_SETHWADDR_OLD:
3773 case SIOCBONDSETHWADDR:
3774 res = bond_sethwaddr(bond_dev, slave_dev);
3776 case BOND_CHANGE_ACTIVE_OLD:
3777 case SIOCBONDCHANGEACTIVE:
3778 res = bond_ioctl_change_active(bond_dev, slave_dev);
3787 up_write(&(bonding_rwsem));
3791 static void bond_set_multicast_list(struct net_device *bond_dev)
3793 struct bonding *bond = bond_dev->priv;
3794 struct dev_mc_list *dmi;
3796 write_lock_bh(&bond->lock);
3799 * Do promisc before checking multicast_mode
3801 if ((bond_dev->flags & IFF_PROMISC) && !(bond->flags & IFF_PROMISC)) {
3802 bond_set_promiscuity(bond, 1);
3805 if (!(bond_dev->flags & IFF_PROMISC) && (bond->flags & IFF_PROMISC)) {
3806 bond_set_promiscuity(bond, -1);
3809 /* set allmulti flag to slaves */
3810 if ((bond_dev->flags & IFF_ALLMULTI) && !(bond->flags & IFF_ALLMULTI)) {
3811 bond_set_allmulti(bond, 1);
3814 if (!(bond_dev->flags & IFF_ALLMULTI) && (bond->flags & IFF_ALLMULTI)) {
3815 bond_set_allmulti(bond, -1);
3818 bond->flags = bond_dev->flags;
3820 /* looking for addresses to add to slaves' mc list */
3821 for (dmi = bond_dev->mc_list; dmi; dmi = dmi->next) {
3822 if (!bond_mc_list_find_dmi(dmi, bond->mc_list)) {
3823 bond_mc_add(bond, dmi->dmi_addr, dmi->dmi_addrlen);
3827 /* looking for addresses to delete from slaves' list */
3828 for (dmi = bond->mc_list; dmi; dmi = dmi->next) {
3829 if (!bond_mc_list_find_dmi(dmi, bond_dev->mc_list)) {
3830 bond_mc_delete(bond, dmi->dmi_addr, dmi->dmi_addrlen);
3834 /* save master's multicast list */
3835 bond_mc_list_destroy(bond);
3836 bond_mc_list_copy(bond_dev->mc_list, bond, GFP_ATOMIC);
3838 write_unlock_bh(&bond->lock);
3842 * Change the MTU of all of a master's slaves to match the master
3844 static int bond_change_mtu(struct net_device *bond_dev, int new_mtu)
3846 struct bonding *bond = bond_dev->priv;
3847 struct slave *slave, *stop_at;
3851 dprintk("bond=%p, name=%s, new_mtu=%d\n", bond,
3852 (bond_dev ? bond_dev->name : "None"), new_mtu);
3854 /* Can't hold bond->lock with bh disabled here since
3855 * some base drivers panic. On the other hand we can't
3856 * hold bond->lock without bh disabled because we'll
3857 * deadlock. The only solution is to rely on the fact
3858 * that we're under rtnl_lock here, and the slaves
3859 * list won't change. This doesn't solve the problem
3860 * of setting the slave's MTU while it is
3861 * transmitting, but the assumption is that the base
3862 * driver can handle that.
3864 * TODO: figure out a way to safely iterate the slaves
3865 * list, but without holding a lock around the actual
3866 * call to the base driver.
3869 bond_for_each_slave(bond, slave, i) {
3870 dprintk("s %p s->p %p c_m %p\n", slave,
3871 slave->prev, slave->dev->change_mtu);
3873 res = dev_set_mtu(slave->dev, new_mtu);
3876 /* If we failed to set the slave's mtu to the new value
3877 * we must abort the operation even in ACTIVE_BACKUP
3878 * mode, because if we allow the backup slaves to have
3879 * different mtu values than the active slave we'll
3880 * need to change their mtu when doing a failover. That
3881 * means changing their mtu from timer context, which
3882 * is probably not a good idea.
3884 dprintk("err %d %s\n", res, slave->dev->name);
3889 bond_dev->mtu = new_mtu;
3894 /* unwind from head to the slave that failed */
3896 bond_for_each_slave_from_to(bond, slave, i, bond->first_slave, stop_at) {
3899 tmp_res = dev_set_mtu(slave->dev, bond_dev->mtu);
3901 dprintk("unwind err %d dev %s\n", tmp_res,
3912 * Note that many devices must be down to change the HW address, and
3913 * downing the master releases all slaves. We can make bonds full of
3914 * bonding devices to test this, however.
3916 static int bond_set_mac_address(struct net_device *bond_dev, void *addr)
3918 struct bonding *bond = bond_dev->priv;
3919 struct sockaddr *sa = addr, tmp_sa;
3920 struct slave *slave, *stop_at;
3924 dprintk("bond=%p, name=%s\n", bond, (bond_dev ? bond_dev->name : "None"));
3926 if (!is_valid_ether_addr(sa->sa_data)) {
3927 return -EADDRNOTAVAIL;
3930 /* Can't hold bond->lock with bh disabled here since
3931 * some base drivers panic. On the other hand we can't
3932 * hold bond->lock without bh disabled because we'll
3933 * deadlock. The only solution is to rely on the fact
3934 * that we're under rtnl_lock here, and the slaves
3935 * list won't change. This doesn't solve the problem
3936 * of setting the slave's hw address while it is
3937 * transmitting, but the assumption is that the base
3938 * driver can handle that.
3940 * TODO: figure out a way to safely iterate the slaves
3941 * list, but without holding a lock around the actual
3942 * call to the base driver.
3945 bond_for_each_slave(bond, slave, i) {
3946 dprintk("slave %p %s\n", slave, slave->dev->name);
3948 if (slave->dev->set_mac_address == NULL) {
3950 dprintk("EOPNOTSUPP %s\n", slave->dev->name);
3954 res = dev_set_mac_address(slave->dev, addr);
3956 /* TODO: consider downing the slave
3958 * User should expect communications
3959 * breakage anyway until ARP finish
3962 dprintk("err %d %s\n", res, slave->dev->name);
3968 memcpy(bond_dev->dev_addr, sa->sa_data, bond_dev->addr_len);
3972 memcpy(tmp_sa.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
3973 tmp_sa.sa_family = bond_dev->type;
3975 /* unwind from head to the slave that failed */
3977 bond_for_each_slave_from_to(bond, slave, i, bond->first_slave, stop_at) {
3980 tmp_res = dev_set_mac_address(slave->dev, &tmp_sa);
3982 dprintk("unwind err %d dev %s\n", tmp_res,
3990 static int bond_xmit_roundrobin(struct sk_buff *skb, struct net_device *bond_dev)
3992 struct bonding *bond = bond_dev->priv;
3993 struct slave *slave, *start_at;
3997 read_lock(&bond->lock);
3999 if (!BOND_IS_OK(bond)) {
4003 read_lock(&bond->curr_slave_lock);
4004 slave = start_at = bond->curr_active_slave;
4005 read_unlock(&bond->curr_slave_lock);
4011 bond_for_each_slave_from(bond, slave, i, start_at) {
4012 if (IS_UP(slave->dev) &&
4013 (slave->link == BOND_LINK_UP) &&
4014 (slave->state == BOND_STATE_ACTIVE)) {
4015 res = bond_dev_queue_xmit(bond, skb, slave->dev);
4017 write_lock(&bond->curr_slave_lock);
4018 bond->curr_active_slave = slave->next;
4019 write_unlock(&bond->curr_slave_lock);
4028 /* no suitable interface, frame not sent */
4031 read_unlock(&bond->lock);
4037 * in active-backup mode, we know that bond->curr_active_slave is always valid if
4038 * the bond has a usable interface.
4040 static int bond_xmit_activebackup(struct sk_buff *skb, struct net_device *bond_dev)
4042 struct bonding *bond = bond_dev->priv;
4045 read_lock(&bond->lock);
4046 read_lock(&bond->curr_slave_lock);
4048 if (!BOND_IS_OK(bond)) {
4052 if (!bond->curr_active_slave)
4055 res = bond_dev_queue_xmit(bond, skb, bond->curr_active_slave->dev);
4059 /* no suitable interface, frame not sent */
4062 read_unlock(&bond->curr_slave_lock);
4063 read_unlock(&bond->lock);
4068 * In bond_xmit_xor() , we determine the output device by using a pre-
4069 * determined xmit_hash_policy(), If the selected device is not enabled,
4070 * find the next active slave.
4072 static int bond_xmit_xor(struct sk_buff *skb, struct net_device *bond_dev)
4074 struct bonding *bond = bond_dev->priv;
4075 struct slave *slave, *start_at;
4080 read_lock(&bond->lock);
4082 if (!BOND_IS_OK(bond)) {
4086 slave_no = bond->xmit_hash_policy(skb, bond_dev, bond->slave_cnt);
4088 bond_for_each_slave(bond, slave, i) {
4097 bond_for_each_slave_from(bond, slave, i, start_at) {
4098 if (IS_UP(slave->dev) &&
4099 (slave->link == BOND_LINK_UP) &&
4100 (slave->state == BOND_STATE_ACTIVE)) {
4101 res = bond_dev_queue_xmit(bond, skb, slave->dev);
4108 /* no suitable interface, frame not sent */
4111 read_unlock(&bond->lock);
4116 * in broadcast mode, we send everything to all usable interfaces.
4118 static int bond_xmit_broadcast(struct sk_buff *skb, struct net_device *bond_dev)
4120 struct bonding *bond = bond_dev->priv;
4121 struct slave *slave, *start_at;
4122 struct net_device *tx_dev = NULL;
4126 read_lock(&bond->lock);
4128 if (!BOND_IS_OK(bond)) {
4132 read_lock(&bond->curr_slave_lock);
4133 start_at = bond->curr_active_slave;
4134 read_unlock(&bond->curr_slave_lock);
4140 bond_for_each_slave_from(bond, slave, i, start_at) {
4141 if (IS_UP(slave->dev) &&
4142 (slave->link == BOND_LINK_UP) &&
4143 (slave->state == BOND_STATE_ACTIVE)) {
4145 struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
4147 printk(KERN_ERR DRV_NAME
4148 ": %s: Error: bond_xmit_broadcast(): "
4149 "skb_clone() failed\n",
4154 res = bond_dev_queue_xmit(bond, skb2, tx_dev);
4156 dev_kfree_skb(skb2);
4160 tx_dev = slave->dev;
4165 res = bond_dev_queue_xmit(bond, skb, tx_dev);
4170 /* no suitable interface, frame not sent */
4173 /* frame sent to all suitable interfaces */
4174 read_unlock(&bond->lock);
4178 /*------------------------- Device initialization ---------------------------*/
4181 * set bond mode specific net device operations
4183 void bond_set_mode_ops(struct bonding *bond, int mode)
4185 struct net_device *bond_dev = bond->dev;
4188 case BOND_MODE_ROUNDROBIN:
4189 bond_dev->hard_start_xmit = bond_xmit_roundrobin;
4191 case BOND_MODE_ACTIVEBACKUP:
4192 bond_dev->hard_start_xmit = bond_xmit_activebackup;
4195 bond_dev->hard_start_xmit = bond_xmit_xor;
4196 if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER34)
4197 bond->xmit_hash_policy = bond_xmit_hash_policy_l34;
4199 bond->xmit_hash_policy = bond_xmit_hash_policy_l2;
4201 case BOND_MODE_BROADCAST:
4202 bond_dev->hard_start_xmit = bond_xmit_broadcast;
4204 case BOND_MODE_8023AD:
4205 bond_set_master_3ad_flags(bond);
4206 bond_dev->hard_start_xmit = bond_3ad_xmit_xor;
4207 if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER34)
4208 bond->xmit_hash_policy = bond_xmit_hash_policy_l34;
4210 bond->xmit_hash_policy = bond_xmit_hash_policy_l2;
4213 bond_set_master_alb_flags(bond);
4216 bond_dev->hard_start_xmit = bond_alb_xmit;
4217 bond_dev->set_mac_address = bond_alb_set_mac_address;
4220 /* Should never happen, mode already checked */
4221 printk(KERN_ERR DRV_NAME
4222 ": %s: Error: Unknown bonding mode %d\n",
4229 static void bond_ethtool_get_drvinfo(struct net_device *bond_dev,
4230 struct ethtool_drvinfo *drvinfo)
4232 strncpy(drvinfo->driver, DRV_NAME, 32);
4233 strncpy(drvinfo->version, DRV_VERSION, 32);
4234 snprintf(drvinfo->fw_version, 32, "%d", BOND_ABI_VERSION);
4237 static const struct ethtool_ops bond_ethtool_ops = {
4238 .get_tx_csum = ethtool_op_get_tx_csum,
4239 .get_tso = ethtool_op_get_tso,
4240 .get_ufo = ethtool_op_get_ufo,
4241 .get_sg = ethtool_op_get_sg,
4242 .get_drvinfo = bond_ethtool_get_drvinfo,
4246 * Does not allocate but creates a /proc entry.
4249 static int bond_init(struct net_device *bond_dev, struct bond_params *params)
4251 struct bonding *bond = bond_dev->priv;
4253 dprintk("Begin bond_init for %s\n", bond_dev->name);
4255 /* initialize rwlocks */
4256 rwlock_init(&bond->lock);
4257 rwlock_init(&bond->curr_slave_lock);
4259 bond->params = *params; /* copy params struct */
4261 /* Initialize pointers */
4262 bond->first_slave = NULL;
4263 bond->curr_active_slave = NULL;
4264 bond->current_arp_slave = NULL;
4265 bond->primary_slave = NULL;
4266 bond->dev = bond_dev;
4267 INIT_LIST_HEAD(&bond->vlan_list);
4269 /* Initialize the device entry points */
4270 bond_dev->open = bond_open;
4271 bond_dev->stop = bond_close;
4272 bond_dev->get_stats = bond_get_stats;
4273 bond_dev->do_ioctl = bond_do_ioctl;
4274 bond_dev->ethtool_ops = &bond_ethtool_ops;
4275 bond_dev->set_multicast_list = bond_set_multicast_list;
4276 bond_dev->change_mtu = bond_change_mtu;
4277 bond_dev->set_mac_address = bond_set_mac_address;
4279 bond_set_mode_ops(bond, bond->params.mode);
4281 bond_dev->destructor = free_netdev;
4283 /* Initialize the device options */
4284 bond_dev->tx_queue_len = 0;
4285 bond_dev->flags |= IFF_MASTER|IFF_MULTICAST;
4286 bond_dev->priv_flags |= IFF_BONDING;
4288 /* At first, we block adding VLANs. That's the only way to
4289 * prevent problems that occur when adding VLANs over an
4290 * empty bond. The block will be removed once non-challenged
4291 * slaves are enslaved.
4293 bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
4295 /* don't acquire bond device's netif_tx_lock when
4297 bond_dev->features |= NETIF_F_LLTX;
4299 /* By default, we declare the bond to be fully
4300 * VLAN hardware accelerated capable. Special
4301 * care is taken in the various xmit functions
4302 * when there are slaves that are not hw accel
4305 bond_dev->vlan_rx_register = bond_vlan_rx_register;
4306 bond_dev->vlan_rx_add_vid = bond_vlan_rx_add_vid;
4307 bond_dev->vlan_rx_kill_vid = bond_vlan_rx_kill_vid;
4308 bond_dev->features |= (NETIF_F_HW_VLAN_TX |
4309 NETIF_F_HW_VLAN_RX |
4310 NETIF_F_HW_VLAN_FILTER);
4312 #ifdef CONFIG_PROC_FS
4313 bond_create_proc_entry(bond);
4316 list_add_tail(&bond->bond_list, &bond_dev_list);
4321 /* De-initialize device specific data.
4322 * Caller must hold rtnl_lock.
4324 void bond_deinit(struct net_device *bond_dev)
4326 struct bonding *bond = bond_dev->priv;
4328 list_del(&bond->bond_list);
4330 #ifdef CONFIG_PROC_FS
4331 bond_remove_proc_entry(bond);
4335 /* Unregister and free all bond devices.
4336 * Caller must hold rtnl_lock.
4338 static void bond_free_all(void)
4340 struct bonding *bond, *nxt;
4342 list_for_each_entry_safe(bond, nxt, &bond_dev_list, bond_list) {
4343 struct net_device *bond_dev = bond->dev;
4345 bond_mc_list_destroy(bond);
4346 /* Release the bonded slaves */
4347 bond_release_all(bond_dev);
4348 bond_deinit(bond_dev);
4349 unregister_netdevice(bond_dev);
4352 #ifdef CONFIG_PROC_FS
4353 bond_destroy_proc_dir();
4357 /*------------------------- Module initialization ---------------------------*/
4360 * Convert string input module parms. Accept either the
4361 * number of the mode or its string name.
4363 int bond_parse_parm(char *mode_arg, struct bond_parm_tbl *tbl)
4367 for (i = 0; tbl[i].modename; i++) {
4368 if ((isdigit(*mode_arg) &&
4369 tbl[i].mode == simple_strtol(mode_arg, NULL, 0)) ||
4370 (strncmp(mode_arg, tbl[i].modename,
4371 strlen(tbl[i].modename)) == 0)) {
4379 static int bond_check_params(struct bond_params *params)
4381 int arp_validate_value;
4384 * Convert string parameters.
4387 bond_mode = bond_parse_parm(mode, bond_mode_tbl);
4388 if (bond_mode == -1) {
4389 printk(KERN_ERR DRV_NAME
4390 ": Error: Invalid bonding mode \"%s\"\n",
4391 mode == NULL ? "NULL" : mode);
4396 if (xmit_hash_policy) {
4397 if ((bond_mode != BOND_MODE_XOR) &&
4398 (bond_mode != BOND_MODE_8023AD)) {
4399 printk(KERN_INFO DRV_NAME
4400 ": xor_mode param is irrelevant in mode %s\n",
4401 bond_mode_name(bond_mode));
4403 xmit_hashtype = bond_parse_parm(xmit_hash_policy,
4405 if (xmit_hashtype == -1) {
4406 printk(KERN_ERR DRV_NAME
4407 ": Error: Invalid xmit_hash_policy \"%s\"\n",
4408 xmit_hash_policy == NULL ? "NULL" :
4416 if (bond_mode != BOND_MODE_8023AD) {
4417 printk(KERN_INFO DRV_NAME
4418 ": lacp_rate param is irrelevant in mode %s\n",
4419 bond_mode_name(bond_mode));
4421 lacp_fast = bond_parse_parm(lacp_rate, bond_lacp_tbl);
4422 if (lacp_fast == -1) {
4423 printk(KERN_ERR DRV_NAME
4424 ": Error: Invalid lacp rate \"%s\"\n",
4425 lacp_rate == NULL ? "NULL" : lacp_rate);
4431 if (max_bonds < 1 || max_bonds > INT_MAX) {
4432 printk(KERN_WARNING DRV_NAME
4433 ": Warning: max_bonds (%d) not in range %d-%d, so it "
4434 "was reset to BOND_DEFAULT_MAX_BONDS (%d)\n",
4435 max_bonds, 1, INT_MAX, BOND_DEFAULT_MAX_BONDS);
4436 max_bonds = BOND_DEFAULT_MAX_BONDS;
4440 printk(KERN_WARNING DRV_NAME
4441 ": Warning: miimon module parameter (%d), "
4442 "not in range 0-%d, so it was reset to %d\n",
4443 miimon, INT_MAX, BOND_LINK_MON_INTERV);
4444 miimon = BOND_LINK_MON_INTERV;
4448 printk(KERN_WARNING DRV_NAME
4449 ": Warning: updelay module parameter (%d), "
4450 "not in range 0-%d, so it was reset to 0\n",
4455 if (downdelay < 0) {
4456 printk(KERN_WARNING DRV_NAME
4457 ": Warning: downdelay module parameter (%d), "
4458 "not in range 0-%d, so it was reset to 0\n",
4459 downdelay, INT_MAX);
4463 if ((use_carrier != 0) && (use_carrier != 1)) {
4464 printk(KERN_WARNING DRV_NAME
4465 ": Warning: use_carrier module parameter (%d), "
4466 "not of valid value (0/1), so it was set to 1\n",
4471 /* reset values for 802.3ad */
4472 if (bond_mode == BOND_MODE_8023AD) {
4474 printk(KERN_WARNING DRV_NAME
4475 ": Warning: miimon must be specified, "
4476 "otherwise bonding will not detect link "
4477 "failure, speed and duplex which are "
4478 "essential for 802.3ad operation\n");
4479 printk(KERN_WARNING "Forcing miimon to 100msec\n");
4484 /* reset values for TLB/ALB */
4485 if ((bond_mode == BOND_MODE_TLB) ||
4486 (bond_mode == BOND_MODE_ALB)) {
4488 printk(KERN_WARNING DRV_NAME
4489 ": Warning: miimon must be specified, "
4490 "otherwise bonding will not detect link "
4491 "failure and link speed which are essential "
4492 "for TLB/ALB load balancing\n");
4493 printk(KERN_WARNING "Forcing miimon to 100msec\n");
4498 if (bond_mode == BOND_MODE_ALB) {
4499 printk(KERN_NOTICE DRV_NAME
4500 ": In ALB mode you might experience client "
4501 "disconnections upon reconnection of a link if the "
4502 "bonding module updelay parameter (%d msec) is "
4503 "incompatible with the forwarding delay time of the "
4509 if (updelay || downdelay) {
4510 /* just warn the user the up/down delay will have
4511 * no effect since miimon is zero...
4513 printk(KERN_WARNING DRV_NAME
4514 ": Warning: miimon module parameter not set "
4515 "and updelay (%d) or downdelay (%d) module "
4516 "parameter is set; updelay and downdelay have "
4517 "no effect unless miimon is set\n",
4518 updelay, downdelay);
4521 /* don't allow arp monitoring */
4523 printk(KERN_WARNING DRV_NAME
4524 ": Warning: miimon (%d) and arp_interval (%d) "
4525 "can't be used simultaneously, disabling ARP "
4527 miimon, arp_interval);
4531 if ((updelay % miimon) != 0) {
4532 printk(KERN_WARNING DRV_NAME
4533 ": Warning: updelay (%d) is not a multiple "
4534 "of miimon (%d), updelay rounded to %d ms\n",
4535 updelay, miimon, (updelay / miimon) * miimon);
4540 if ((downdelay % miimon) != 0) {
4541 printk(KERN_WARNING DRV_NAME
4542 ": Warning: downdelay (%d) is not a multiple "
4543 "of miimon (%d), downdelay rounded to %d ms\n",
4545 (downdelay / miimon) * miimon);
4548 downdelay /= miimon;
4551 if (arp_interval < 0) {
4552 printk(KERN_WARNING DRV_NAME
4553 ": Warning: arp_interval module parameter (%d) "
4554 ", not in range 0-%d, so it was reset to %d\n",
4555 arp_interval, INT_MAX, BOND_LINK_ARP_INTERV);
4556 arp_interval = BOND_LINK_ARP_INTERV;
4559 for (arp_ip_count = 0;
4560 (arp_ip_count < BOND_MAX_ARP_TARGETS) && arp_ip_target[arp_ip_count];
4562 /* not complete check, but should be good enough to
4564 if (!isdigit(arp_ip_target[arp_ip_count][0])) {
4565 printk(KERN_WARNING DRV_NAME
4566 ": Warning: bad arp_ip_target module parameter "
4567 "(%s), ARP monitoring will not be performed\n",
4568 arp_ip_target[arp_ip_count]);
4571 u32 ip = in_aton(arp_ip_target[arp_ip_count]);
4572 arp_target[arp_ip_count] = ip;
4576 if (arp_interval && !arp_ip_count) {
4577 /* don't allow arping if no arp_ip_target given... */
4578 printk(KERN_WARNING DRV_NAME
4579 ": Warning: arp_interval module parameter (%d) "
4580 "specified without providing an arp_ip_target "
4581 "parameter, arp_interval was reset to 0\n",
4587 if (bond_mode != BOND_MODE_ACTIVEBACKUP) {
4588 printk(KERN_ERR DRV_NAME
4589 ": arp_validate only supported in active-backup mode\n");
4592 if (!arp_interval) {
4593 printk(KERN_ERR DRV_NAME
4594 ": arp_validate requires arp_interval\n");
4598 arp_validate_value = bond_parse_parm(arp_validate,
4600 if (arp_validate_value == -1) {
4601 printk(KERN_ERR DRV_NAME
4602 ": Error: invalid arp_validate \"%s\"\n",
4603 arp_validate == NULL ? "NULL" : arp_validate);
4607 arp_validate_value = 0;
4610 printk(KERN_INFO DRV_NAME
4611 ": MII link monitoring set to %d ms\n",
4613 } else if (arp_interval) {
4616 printk(KERN_INFO DRV_NAME
4617 ": ARP monitoring set to %d ms, validate %s, with %d target(s):",
4619 arp_validate_tbl[arp_validate_value].modename,
4622 for (i = 0; i < arp_ip_count; i++)
4623 printk (" %s", arp_ip_target[i]);
4628 /* miimon and arp_interval not set, we need one so things
4629 * work as expected, see bonding.txt for details
4631 printk(KERN_WARNING DRV_NAME
4632 ": Warning: either miimon or arp_interval and "
4633 "arp_ip_target module parameters must be specified, "
4634 "otherwise bonding will not detect link failures! see "
4635 "bonding.txt for details.\n");
4638 if (primary && !USES_PRIMARY(bond_mode)) {
4639 /* currently, using a primary only makes sense
4640 * in active backup, TLB or ALB modes
4642 printk(KERN_WARNING DRV_NAME
4643 ": Warning: %s primary device specified but has no "
4644 "effect in %s mode\n",
4645 primary, bond_mode_name(bond_mode));
4649 /* fill params struct with the proper values */
4650 params->mode = bond_mode;
4651 params->xmit_policy = xmit_hashtype;
4652 params->miimon = miimon;
4653 params->arp_interval = arp_interval;
4654 params->arp_validate = arp_validate_value;
4655 params->updelay = updelay;
4656 params->downdelay = downdelay;
4657 params->use_carrier = use_carrier;
4658 params->lacp_fast = lacp_fast;
4659 params->primary[0] = 0;
4662 strncpy(params->primary, primary, IFNAMSIZ);
4663 params->primary[IFNAMSIZ - 1] = 0;
4666 memcpy(params->arp_targets, arp_target, sizeof(arp_target));
4671 static struct lock_class_key bonding_netdev_xmit_lock_key;
4673 /* Create a new bond based on the specified name and bonding parameters.
4674 * If name is NULL, obtain a suitable "bond%d" name for us.
4675 * Caller must NOT hold rtnl_lock; we need to release it here before we
4676 * set up our sysfs entries.
4678 int bond_create(char *name, struct bond_params *params, struct bonding **newbond)
4680 struct net_device *bond_dev;
4684 bond_dev = alloc_netdev(sizeof(struct bonding), name ? name : "",
4687 printk(KERN_ERR DRV_NAME
4688 ": %s: eek! can't alloc netdev!\n",
4695 res = dev_alloc_name(bond_dev, "bond%d");
4700 /* bond_init() must be called after dev_alloc_name() (for the
4701 * /proc files), but before register_netdevice(), because we
4702 * need to set function pointers.
4705 res = bond_init(bond_dev, params);
4710 SET_MODULE_OWNER(bond_dev);
4712 res = register_netdevice(bond_dev);
4717 lockdep_set_class(&bond_dev->_xmit_lock, &bonding_netdev_xmit_lock_key);
4720 *newbond = bond_dev->priv;
4722 netif_carrier_off(bond_dev);
4724 rtnl_unlock(); /* allows sysfs registration of net device */
4725 res = bond_create_sysfs_entry(bond_dev->priv);
4734 bond_deinit(bond_dev);
4736 free_netdev(bond_dev);
4742 static int __init bonding_init(void)
4747 printk(KERN_INFO "%s", version);
4749 res = bond_check_params(&bonding_defaults);
4754 #ifdef CONFIG_PROC_FS
4755 bond_create_proc_dir();
4757 for (i = 0; i < max_bonds; i++) {
4758 res = bond_create(NULL, &bonding_defaults, NULL);
4763 res = bond_create_sysfs();
4767 register_netdevice_notifier(&bond_netdev_notifier);
4768 register_inetaddr_notifier(&bond_inetaddr_notifier);
4774 bond_destroy_sysfs();
4781 static void __exit bonding_exit(void)
4783 unregister_netdevice_notifier(&bond_netdev_notifier);
4784 unregister_inetaddr_notifier(&bond_inetaddr_notifier);
4788 bond_destroy_sysfs();
4792 module_init(bonding_init);
4793 module_exit(bonding_exit);
4794 MODULE_LICENSE("GPL");
4795 MODULE_VERSION(DRV_VERSION);
4796 MODULE_DESCRIPTION(DRV_DESCRIPTION ", v" DRV_VERSION);
4797 MODULE_AUTHOR("Thomas Davis, tadavis@lbl.gov and many others");
4798 MODULE_SUPPORTED_DEVICE("most ethernet devices");