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/config.h>
37 #include <linux/kernel.h>
38 #include <linux/module.h>
39 #include <linux/sched.h>
40 #include <linux/types.h>
41 #include <linux/fcntl.h>
42 #include <linux/interrupt.h>
43 #include <linux/ptrace.h>
44 #include <linux/ioport.h>
48 #include <linux/tcp.h>
49 #include <linux/udp.h>
50 #include <linux/slab.h>
51 #include <linux/string.h>
52 #include <linux/init.h>
53 #include <linux/timer.h>
54 #include <linux/socket.h>
55 #include <linux/ctype.h>
56 #include <linux/inet.h>
57 #include <linux/bitops.h>
58 #include <asm/system.h>
61 #include <asm/uaccess.h>
62 #include <linux/errno.h>
63 #include <linux/netdevice.h>
64 #include <linux/inetdevice.h>
65 #include <linux/etherdevice.h>
66 #include <linux/skbuff.h>
68 #include <linux/rtnetlink.h>
69 #include <linux/proc_fs.h>
70 #include <linux/seq_file.h>
71 #include <linux/smp.h>
72 #include <linux/if_ether.h>
74 #include <linux/mii.h>
75 #include <linux/ethtool.h>
76 #include <linux/if_vlan.h>
77 #include <linux/if_bonding.h>
78 #include <net/route.h>
83 /*---------------------------- Module parameters ----------------------------*/
85 /* monitor all links that often (in milliseconds). <=0 disables monitoring */
86 #define BOND_LINK_MON_INTERV 0
87 #define BOND_LINK_ARP_INTERV 0
89 static int max_bonds = BOND_DEFAULT_MAX_BONDS;
90 static int miimon = BOND_LINK_MON_INTERV;
91 static int updelay = 0;
92 static int downdelay = 0;
93 static int use_carrier = 1;
94 static char *mode = NULL;
95 static char *primary = NULL;
96 static char *lacp_rate = NULL;
97 static char *xmit_hash_policy = NULL;
98 static int arp_interval = BOND_LINK_ARP_INTERV;
99 static char *arp_ip_target[BOND_MAX_ARP_TARGETS] = { 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");
132 /*----------------------------- Global variables ----------------------------*/
134 static const char * const version =
135 DRV_DESCRIPTION ": v" DRV_VERSION " (" DRV_RELDATE ")\n";
137 LIST_HEAD(bond_dev_list);
139 #ifdef CONFIG_PROC_FS
140 static struct proc_dir_entry *bond_proc_dir = NULL;
143 extern struct rw_semaphore bonding_rwsem;
144 static u32 arp_target[BOND_MAX_ARP_TARGETS] = { 0, } ;
145 static int arp_ip_count = 0;
146 static int bond_mode = BOND_MODE_ROUNDROBIN;
147 static int xmit_hashtype= BOND_XMIT_POLICY_LAYER2;
148 static int lacp_fast = 0;
151 struct bond_parm_tbl bond_lacp_tbl[] = {
152 { "slow", AD_LACP_SLOW},
153 { "fast", AD_LACP_FAST},
157 struct bond_parm_tbl bond_mode_tbl[] = {
158 { "balance-rr", BOND_MODE_ROUNDROBIN},
159 { "active-backup", BOND_MODE_ACTIVEBACKUP},
160 { "balance-xor", BOND_MODE_XOR},
161 { "broadcast", BOND_MODE_BROADCAST},
162 { "802.3ad", BOND_MODE_8023AD},
163 { "balance-tlb", BOND_MODE_TLB},
164 { "balance-alb", BOND_MODE_ALB},
168 struct bond_parm_tbl xmit_hashtype_tbl[] = {
169 { "layer2", BOND_XMIT_POLICY_LAYER2},
170 { "layer3+4", BOND_XMIT_POLICY_LAYER34},
174 /*-------------------------- Forward declarations ---------------------------*/
176 static void bond_send_gratuitous_arp(struct bonding *bond);
178 /*---------------------------- General routines -----------------------------*/
180 const char *bond_mode_name(int mode)
183 case BOND_MODE_ROUNDROBIN :
184 return "load balancing (round-robin)";
185 case BOND_MODE_ACTIVEBACKUP :
186 return "fault-tolerance (active-backup)";
188 return "load balancing (xor)";
189 case BOND_MODE_BROADCAST :
190 return "fault-tolerance (broadcast)";
191 case BOND_MODE_8023AD:
192 return "IEEE 802.3ad Dynamic link aggregation";
194 return "transmit load balancing";
196 return "adaptive load balancing";
202 /*---------------------------------- VLAN -----------------------------------*/
205 * bond_add_vlan - add a new vlan id on bond
206 * @bond: bond that got the notification
207 * @vlan_id: the vlan id to add
209 * Returns -ENOMEM if allocation failed.
211 static int bond_add_vlan(struct bonding *bond, unsigned short vlan_id)
213 struct vlan_entry *vlan;
215 dprintk("bond: %s, vlan id %d\n",
216 (bond ? bond->dev->name: "None"), vlan_id);
218 vlan = kmalloc(sizeof(struct vlan_entry), GFP_KERNEL);
223 INIT_LIST_HEAD(&vlan->vlan_list);
224 vlan->vlan_id = vlan_id;
227 write_lock_bh(&bond->lock);
229 list_add_tail(&vlan->vlan_list, &bond->vlan_list);
231 write_unlock_bh(&bond->lock);
233 dprintk("added VLAN ID %d on bond %s\n", vlan_id, bond->dev->name);
239 * bond_del_vlan - delete a vlan id from bond
240 * @bond: bond that got the notification
241 * @vlan_id: the vlan id to delete
243 * returns -ENODEV if @vlan_id was not found in @bond.
245 static int bond_del_vlan(struct bonding *bond, unsigned short vlan_id)
247 struct vlan_entry *vlan, *next;
250 dprintk("bond: %s, vlan id %d\n", bond->dev->name, vlan_id);
252 write_lock_bh(&bond->lock);
254 list_for_each_entry_safe(vlan, next, &bond->vlan_list, vlan_list) {
255 if (vlan->vlan_id == vlan_id) {
256 list_del(&vlan->vlan_list);
258 if ((bond->params.mode == BOND_MODE_TLB) ||
259 (bond->params.mode == BOND_MODE_ALB)) {
260 bond_alb_clear_vlan(bond, vlan_id);
263 dprintk("removed VLAN ID %d from bond %s\n", vlan_id,
268 if (list_empty(&bond->vlan_list) &&
269 (bond->slave_cnt == 0)) {
270 /* Last VLAN removed and no slaves, so
271 * restore block on adding VLANs. This will
272 * be removed once new slaves that are not
273 * VLAN challenged will be added.
275 bond->dev->features |= NETIF_F_VLAN_CHALLENGED;
283 dprintk("couldn't find VLAN ID %d in bond %s\n", vlan_id,
287 write_unlock_bh(&bond->lock);
292 * bond_has_challenged_slaves
293 * @bond: the bond we're working on
295 * Searches the slave list. Returns 1 if a vlan challenged slave
296 * was found, 0 otherwise.
298 * Assumes bond->lock is held.
300 static int bond_has_challenged_slaves(struct bonding *bond)
305 bond_for_each_slave(bond, slave, i) {
306 if (slave->dev->features & NETIF_F_VLAN_CHALLENGED) {
307 dprintk("found VLAN challenged slave - %s\n",
313 dprintk("no VLAN challenged slaves found\n");
318 * bond_next_vlan - safely skip to the next item in the vlans list.
319 * @bond: the bond we're working on
320 * @curr: item we're advancing from
322 * Returns %NULL if list is empty, bond->next_vlan if @curr is %NULL,
323 * or @curr->next otherwise (even if it is @curr itself again).
325 * Caller must hold bond->lock
327 struct vlan_entry *bond_next_vlan(struct bonding *bond, struct vlan_entry *curr)
329 struct vlan_entry *next, *last;
331 if (list_empty(&bond->vlan_list)) {
336 next = list_entry(bond->vlan_list.next,
337 struct vlan_entry, vlan_list);
339 last = list_entry(bond->vlan_list.prev,
340 struct vlan_entry, vlan_list);
342 next = list_entry(bond->vlan_list.next,
343 struct vlan_entry, vlan_list);
345 next = list_entry(curr->vlan_list.next,
346 struct vlan_entry, vlan_list);
354 * bond_dev_queue_xmit - Prepare skb for xmit.
356 * @bond: bond device that got this skb for tx.
357 * @skb: hw accel VLAN tagged skb to transmit
358 * @slave_dev: slave that is supposed to xmit this skbuff
360 * When the bond gets an skb to transmit that is
361 * already hardware accelerated VLAN tagged, and it
362 * needs to relay this skb to a slave that is not
363 * hw accel capable, the skb needs to be "unaccelerated",
364 * i.e. strip the hwaccel tag and re-insert it as part
367 int bond_dev_queue_xmit(struct bonding *bond, struct sk_buff *skb, struct net_device *slave_dev)
369 unsigned short vlan_id;
371 if (!list_empty(&bond->vlan_list) &&
372 !(slave_dev->features & NETIF_F_HW_VLAN_TX) &&
373 vlan_get_tag(skb, &vlan_id) == 0) {
374 skb->dev = slave_dev;
375 skb = vlan_put_tag(skb, vlan_id);
377 /* vlan_put_tag() frees the skb in case of error,
378 * so return success here so the calling functions
379 * won't attempt to free is again.
384 skb->dev = slave_dev;
394 * In the following 3 functions, bond_vlan_rx_register(), bond_vlan_rx_add_vid
395 * and bond_vlan_rx_kill_vid, We don't protect the slave list iteration with a
397 * a. This operation is performed in IOCTL context,
398 * b. The operation is protected by the RTNL semaphore in the 8021q code,
399 * c. Holding a lock with BH disabled while directly calling a base driver
400 * entry point is generally a BAD idea.
402 * The design of synchronization/protection for this operation in the 8021q
403 * module is good for one or more VLAN devices over a single physical device
404 * and cannot be extended for a teaming solution like bonding, so there is a
405 * potential race condition here where a net device from the vlan group might
406 * be referenced (either by a base driver or the 8021q code) while it is being
407 * removed from the system. However, it turns out we're not making matters
408 * worse, and if it works for regular VLAN usage it will work here too.
412 * bond_vlan_rx_register - Propagates registration to slaves
413 * @bond_dev: bonding net device that got called
414 * @grp: vlan group being registered
416 static void bond_vlan_rx_register(struct net_device *bond_dev, struct vlan_group *grp)
418 struct bonding *bond = bond_dev->priv;
424 bond_for_each_slave(bond, slave, i) {
425 struct net_device *slave_dev = slave->dev;
427 if ((slave_dev->features & NETIF_F_HW_VLAN_RX) &&
428 slave_dev->vlan_rx_register) {
429 slave_dev->vlan_rx_register(slave_dev, grp);
435 * bond_vlan_rx_add_vid - Propagates adding an id to slaves
436 * @bond_dev: bonding net device that got called
437 * @vid: vlan id being added
439 static void bond_vlan_rx_add_vid(struct net_device *bond_dev, uint16_t vid)
441 struct bonding *bond = bond_dev->priv;
445 bond_for_each_slave(bond, slave, i) {
446 struct net_device *slave_dev = slave->dev;
448 if ((slave_dev->features & NETIF_F_HW_VLAN_FILTER) &&
449 slave_dev->vlan_rx_add_vid) {
450 slave_dev->vlan_rx_add_vid(slave_dev, vid);
454 res = bond_add_vlan(bond, vid);
456 printk(KERN_ERR DRV_NAME
457 ": %s: Error: Failed to add vlan id %d\n",
458 bond_dev->name, vid);
463 * bond_vlan_rx_kill_vid - Propagates deleting an id to slaves
464 * @bond_dev: bonding net device that got called
465 * @vid: vlan id being removed
467 static void bond_vlan_rx_kill_vid(struct net_device *bond_dev, uint16_t vid)
469 struct bonding *bond = bond_dev->priv;
471 struct net_device *vlan_dev;
474 bond_for_each_slave(bond, slave, i) {
475 struct net_device *slave_dev = slave->dev;
477 if ((slave_dev->features & NETIF_F_HW_VLAN_FILTER) &&
478 slave_dev->vlan_rx_kill_vid) {
479 /* Save and then restore vlan_dev in the grp array,
480 * since the slave's driver might clear it.
482 vlan_dev = bond->vlgrp->vlan_devices[vid];
483 slave_dev->vlan_rx_kill_vid(slave_dev, vid);
484 bond->vlgrp->vlan_devices[vid] = vlan_dev;
488 res = bond_del_vlan(bond, vid);
490 printk(KERN_ERR DRV_NAME
491 ": %s: Error: Failed to remove vlan id %d\n",
492 bond_dev->name, vid);
496 static void bond_add_vlans_on_slave(struct bonding *bond, struct net_device *slave_dev)
498 struct vlan_entry *vlan;
500 write_lock_bh(&bond->lock);
502 if (list_empty(&bond->vlan_list)) {
506 if ((slave_dev->features & NETIF_F_HW_VLAN_RX) &&
507 slave_dev->vlan_rx_register) {
508 slave_dev->vlan_rx_register(slave_dev, bond->vlgrp);
511 if (!(slave_dev->features & NETIF_F_HW_VLAN_FILTER) ||
512 !(slave_dev->vlan_rx_add_vid)) {
516 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
517 slave_dev->vlan_rx_add_vid(slave_dev, vlan->vlan_id);
521 write_unlock_bh(&bond->lock);
524 static void bond_del_vlans_from_slave(struct bonding *bond, struct net_device *slave_dev)
526 struct vlan_entry *vlan;
527 struct net_device *vlan_dev;
529 write_lock_bh(&bond->lock);
531 if (list_empty(&bond->vlan_list)) {
535 if (!(slave_dev->features & NETIF_F_HW_VLAN_FILTER) ||
536 !(slave_dev->vlan_rx_kill_vid)) {
540 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
541 /* Save and then restore vlan_dev in the grp array,
542 * since the slave's driver might clear it.
544 vlan_dev = bond->vlgrp->vlan_devices[vlan->vlan_id];
545 slave_dev->vlan_rx_kill_vid(slave_dev, vlan->vlan_id);
546 bond->vlgrp->vlan_devices[vlan->vlan_id] = vlan_dev;
550 if ((slave_dev->features & NETIF_F_HW_VLAN_RX) &&
551 slave_dev->vlan_rx_register) {
552 slave_dev->vlan_rx_register(slave_dev, NULL);
556 write_unlock_bh(&bond->lock);
559 /*------------------------------- Link status -------------------------------*/
562 * Set the carrier state for the master according to the state of its
563 * slaves. If any slaves are up, the master is up. In 802.3ad mode,
564 * do special 802.3ad magic.
566 * Returns zero if carrier state does not change, nonzero if it does.
568 static int bond_set_carrier(struct bonding *bond)
573 if (bond->slave_cnt == 0)
576 if (bond->params.mode == BOND_MODE_8023AD)
577 return bond_3ad_set_carrier(bond);
579 bond_for_each_slave(bond, slave, i) {
580 if (slave->link == BOND_LINK_UP) {
581 if (!netif_carrier_ok(bond->dev)) {
582 netif_carrier_on(bond->dev);
590 if (netif_carrier_ok(bond->dev)) {
591 netif_carrier_off(bond->dev);
598 * Get link speed and duplex from the slave's base driver
599 * using ethtool. If for some reason the call fails or the
600 * values are invalid, fake speed and duplex to 100/Full
603 static int bond_update_speed_duplex(struct slave *slave)
605 struct net_device *slave_dev = slave->dev;
606 static int (* ioctl)(struct net_device *, struct ifreq *, int);
608 struct ethtool_cmd etool;
610 /* Fake speed and duplex */
611 slave->speed = SPEED_100;
612 slave->duplex = DUPLEX_FULL;
614 if (slave_dev->ethtool_ops) {
617 if (!slave_dev->ethtool_ops->get_settings) {
621 res = slave_dev->ethtool_ops->get_settings(slave_dev, &etool);
629 ioctl = slave_dev->do_ioctl;
630 strncpy(ifr.ifr_name, slave_dev->name, IFNAMSIZ);
631 etool.cmd = ETHTOOL_GSET;
632 ifr.ifr_data = (char*)&etool;
633 if (!ioctl || (IOCTL(slave_dev, &ifr, SIOCETHTOOL) < 0)) {
638 switch (etool.speed) {
647 switch (etool.duplex) {
655 slave->speed = etool.speed;
656 slave->duplex = etool.duplex;
662 * if <dev> supports MII link status reporting, check its link status.
664 * We either do MII/ETHTOOL ioctls, or check netif_carrier_ok(),
665 * depening upon the setting of the use_carrier parameter.
667 * Return either BMSR_LSTATUS, meaning that the link is up (or we
668 * can't tell and just pretend it is), or 0, meaning that the link is
671 * If reporting is non-zero, instead of faking link up, return -1 if
672 * both ETHTOOL and MII ioctls fail (meaning the device does not
673 * support them). If use_carrier is set, return whatever it says.
674 * It'd be nice if there was a good way to tell if a driver supports
675 * netif_carrier, but there really isn't.
677 static int bond_check_dev_link(struct bonding *bond, struct net_device *slave_dev, int reporting)
679 static int (* ioctl)(struct net_device *, struct ifreq *, int);
681 struct mii_ioctl_data *mii;
682 struct ethtool_value etool;
684 if (bond->params.use_carrier) {
685 return netif_carrier_ok(slave_dev) ? BMSR_LSTATUS : 0;
688 ioctl = slave_dev->do_ioctl;
690 /* TODO: set pointer to correct ioctl on a per team member */
691 /* bases to make this more efficient. that is, once */
692 /* we determine the correct ioctl, we will always */
693 /* call it and not the others for that team */
697 * We cannot assume that SIOCGMIIPHY will also read a
698 * register; not all network drivers (e.g., e100)
702 /* Yes, the mii is overlaid on the ifreq.ifr_ifru */
703 strncpy(ifr.ifr_name, slave_dev->name, IFNAMSIZ);
705 if (IOCTL(slave_dev, &ifr, SIOCGMIIPHY) == 0) {
706 mii->reg_num = MII_BMSR;
707 if (IOCTL(slave_dev, &ifr, SIOCGMIIREG) == 0) {
708 return (mii->val_out & BMSR_LSTATUS);
713 /* try SIOCETHTOOL ioctl, some drivers cache ETHTOOL_GLINK */
714 /* for a period of time so we attempt to get link status */
715 /* from it last if the above MII ioctls fail... */
716 if (slave_dev->ethtool_ops) {
717 if (slave_dev->ethtool_ops->get_link) {
720 link = slave_dev->ethtool_ops->get_link(slave_dev);
722 return link ? BMSR_LSTATUS : 0;
727 strncpy(ifr.ifr_name, slave_dev->name, IFNAMSIZ);
728 etool.cmd = ETHTOOL_GLINK;
729 ifr.ifr_data = (char*)&etool;
730 if (IOCTL(slave_dev, &ifr, SIOCETHTOOL) == 0) {
731 if (etool.data == 1) {
734 dprintk("SIOCETHTOOL shows link down\n");
741 * If reporting, report that either there's no dev->do_ioctl,
742 * or both SIOCGMIIREG and SIOCETHTOOL failed (meaning that we
743 * cannot report link status). If not reporting, pretend
746 return (reporting ? -1 : BMSR_LSTATUS);
749 /*----------------------------- Multicast list ------------------------------*/
752 * Returns 0 if dmi1 and dmi2 are the same, non-0 otherwise
754 static inline int bond_is_dmi_same(struct dev_mc_list *dmi1, struct dev_mc_list *dmi2)
756 return memcmp(dmi1->dmi_addr, dmi2->dmi_addr, dmi1->dmi_addrlen) == 0 &&
757 dmi1->dmi_addrlen == dmi2->dmi_addrlen;
761 * returns dmi entry if found, NULL otherwise
763 static struct dev_mc_list *bond_mc_list_find_dmi(struct dev_mc_list *dmi, struct dev_mc_list *mc_list)
765 struct dev_mc_list *idmi;
767 for (idmi = mc_list; idmi; idmi = idmi->next) {
768 if (bond_is_dmi_same(dmi, idmi)) {
777 * Push the promiscuity flag down to appropriate slaves
779 static void bond_set_promiscuity(struct bonding *bond, int inc)
781 if (USES_PRIMARY(bond->params.mode)) {
782 /* write lock already acquired */
783 if (bond->curr_active_slave) {
784 dev_set_promiscuity(bond->curr_active_slave->dev, inc);
789 bond_for_each_slave(bond, slave, i) {
790 dev_set_promiscuity(slave->dev, inc);
796 * Push the allmulti flag down to all slaves
798 static void bond_set_allmulti(struct bonding *bond, int inc)
800 if (USES_PRIMARY(bond->params.mode)) {
801 /* write lock already acquired */
802 if (bond->curr_active_slave) {
803 dev_set_allmulti(bond->curr_active_slave->dev, inc);
808 bond_for_each_slave(bond, slave, i) {
809 dev_set_allmulti(slave->dev, inc);
815 * Add a Multicast address to slaves
818 static void bond_mc_add(struct bonding *bond, void *addr, int alen)
820 if (USES_PRIMARY(bond->params.mode)) {
821 /* write lock already acquired */
822 if (bond->curr_active_slave) {
823 dev_mc_add(bond->curr_active_slave->dev, addr, alen, 0);
828 bond_for_each_slave(bond, slave, i) {
829 dev_mc_add(slave->dev, addr, alen, 0);
835 * Remove a multicast address from slave
838 static void bond_mc_delete(struct bonding *bond, void *addr, int alen)
840 if (USES_PRIMARY(bond->params.mode)) {
841 /* write lock already acquired */
842 if (bond->curr_active_slave) {
843 dev_mc_delete(bond->curr_active_slave->dev, addr, alen, 0);
848 bond_for_each_slave(bond, slave, i) {
849 dev_mc_delete(slave->dev, addr, alen, 0);
855 * Totally destroys the mc_list in bond
857 static void bond_mc_list_destroy(struct bonding *bond)
859 struct dev_mc_list *dmi;
863 bond->mc_list = dmi->next;
870 * Copy all the Multicast addresses from src to the bonding device dst
872 static int bond_mc_list_copy(struct dev_mc_list *mc_list, struct bonding *bond,
875 struct dev_mc_list *dmi, *new_dmi;
877 for (dmi = mc_list; dmi; dmi = dmi->next) {
878 new_dmi = kmalloc(sizeof(struct dev_mc_list), gfp_flag);
881 /* FIXME: Potential memory leak !!! */
885 new_dmi->next = bond->mc_list;
886 bond->mc_list = new_dmi;
887 new_dmi->dmi_addrlen = dmi->dmi_addrlen;
888 memcpy(new_dmi->dmi_addr, dmi->dmi_addr, dmi->dmi_addrlen);
889 new_dmi->dmi_users = dmi->dmi_users;
890 new_dmi->dmi_gusers = dmi->dmi_gusers;
897 * flush all members of flush->mc_list from device dev->mc_list
899 static void bond_mc_list_flush(struct net_device *bond_dev, struct net_device *slave_dev)
901 struct bonding *bond = bond_dev->priv;
902 struct dev_mc_list *dmi;
904 for (dmi = bond_dev->mc_list; dmi; dmi = dmi->next) {
905 dev_mc_delete(slave_dev, dmi->dmi_addr, dmi->dmi_addrlen, 0);
908 if (bond->params.mode == BOND_MODE_8023AD) {
909 /* del lacpdu mc addr from mc list */
910 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
912 dev_mc_delete(slave_dev, lacpdu_multicast, ETH_ALEN, 0);
916 /*--------------------------- Active slave change ---------------------------*/
919 * Update the mc list and multicast-related flags for the new and
920 * old active slaves (if any) according to the multicast mode, and
921 * promiscuous flags unconditionally.
923 static void bond_mc_swap(struct bonding *bond, struct slave *new_active, struct slave *old_active)
925 struct dev_mc_list *dmi;
927 if (!USES_PRIMARY(bond->params.mode)) {
928 /* nothing to do - mc list is already up-to-date on
935 if (bond->dev->flags & IFF_PROMISC) {
936 dev_set_promiscuity(old_active->dev, -1);
939 if (bond->dev->flags & IFF_ALLMULTI) {
940 dev_set_allmulti(old_active->dev, -1);
943 for (dmi = bond->dev->mc_list; dmi; dmi = dmi->next) {
944 dev_mc_delete(old_active->dev, dmi->dmi_addr, dmi->dmi_addrlen, 0);
949 if (bond->dev->flags & IFF_PROMISC) {
950 dev_set_promiscuity(new_active->dev, 1);
953 if (bond->dev->flags & IFF_ALLMULTI) {
954 dev_set_allmulti(new_active->dev, 1);
957 for (dmi = bond->dev->mc_list; dmi; dmi = dmi->next) {
958 dev_mc_add(new_active->dev, dmi->dmi_addr, dmi->dmi_addrlen, 0);
964 * find_best_interface - select the best available slave to be the active one
965 * @bond: our bonding struct
967 * Warning: Caller must hold curr_slave_lock for writing.
969 static struct slave *bond_find_best_slave(struct bonding *bond)
971 struct slave *new_active, *old_active;
972 struct slave *bestslave = NULL;
973 int mintime = bond->params.updelay;
976 new_active = old_active = bond->curr_active_slave;
978 if (!new_active) { /* there were no active slaves left */
979 if (bond->slave_cnt > 0) { /* found one slave */
980 new_active = bond->first_slave;
982 return NULL; /* still no slave, return NULL */
986 /* first try the primary link; if arping, a link must tx/rx traffic
987 * before it can be considered the curr_active_slave - also, we would skip
988 * slaves between the curr_active_slave and primary_slave that may be up
991 if ((bond->primary_slave) &&
992 (!bond->params.arp_interval) &&
993 (IS_UP(bond->primary_slave->dev))) {
994 new_active = bond->primary_slave;
997 /* remember where to stop iterating over the slaves */
998 old_active = new_active;
1000 bond_for_each_slave_from(bond, new_active, i, old_active) {
1001 if (IS_UP(new_active->dev)) {
1002 if (new_active->link == BOND_LINK_UP) {
1004 } else if (new_active->link == BOND_LINK_BACK) {
1005 /* link up, but waiting for stabilization */
1006 if (new_active->delay < mintime) {
1007 mintime = new_active->delay;
1008 bestslave = new_active;
1018 * change_active_interface - change the active slave into the specified one
1019 * @bond: our bonding struct
1020 * @new: the new slave to make the active one
1022 * Set the new slave to the bond's settings and unset them on the old
1023 * curr_active_slave.
1024 * Setting include flags, mc-list, promiscuity, allmulti, etc.
1026 * If @new's link state is %BOND_LINK_BACK we'll set it to %BOND_LINK_UP,
1027 * because it is apparently the best available slave we have, even though its
1028 * updelay hasn't timed out yet.
1030 * Warning: Caller must hold curr_slave_lock for writing.
1032 void bond_change_active_slave(struct bonding *bond, struct slave *new_active)
1034 struct slave *old_active = bond->curr_active_slave;
1036 if (old_active == new_active) {
1041 if (new_active->link == BOND_LINK_BACK) {
1042 if (USES_PRIMARY(bond->params.mode)) {
1043 printk(KERN_INFO DRV_NAME
1044 ": %s: making interface %s the new "
1045 "active one %d ms earlier.\n",
1046 bond->dev->name, new_active->dev->name,
1047 (bond->params.updelay - new_active->delay) * bond->params.miimon);
1050 new_active->delay = 0;
1051 new_active->link = BOND_LINK_UP;
1052 new_active->jiffies = jiffies;
1054 if (bond->params.mode == BOND_MODE_8023AD) {
1055 bond_3ad_handle_link_change(new_active, BOND_LINK_UP);
1058 if ((bond->params.mode == BOND_MODE_TLB) ||
1059 (bond->params.mode == BOND_MODE_ALB)) {
1060 bond_alb_handle_link_change(bond, new_active, BOND_LINK_UP);
1063 if (USES_PRIMARY(bond->params.mode)) {
1064 printk(KERN_INFO DRV_NAME
1065 ": %s: making interface %s the new "
1067 bond->dev->name, new_active->dev->name);
1072 if (USES_PRIMARY(bond->params.mode)) {
1073 bond_mc_swap(bond, new_active, old_active);
1076 if ((bond->params.mode == BOND_MODE_TLB) ||
1077 (bond->params.mode == BOND_MODE_ALB)) {
1078 bond_alb_handle_active_change(bond, new_active);
1080 bond_set_slave_inactive_flags(old_active);
1082 bond_set_slave_active_flags(new_active);
1084 bond->curr_active_slave = new_active;
1087 if (bond->params.mode == BOND_MODE_ACTIVEBACKUP) {
1089 bond_set_slave_inactive_flags(old_active);
1093 bond_set_slave_active_flags(new_active);
1095 bond_send_gratuitous_arp(bond);
1100 * bond_select_active_slave - select a new active slave, if needed
1101 * @bond: our bonding struct
1103 * This functions shoud be called when one of the following occurs:
1104 * - The old curr_active_slave has been released or lost its link.
1105 * - The primary_slave has got its link back.
1106 * - A slave has got its link back and there's no old curr_active_slave.
1108 * Warning: Caller must hold curr_slave_lock for writing.
1110 void bond_select_active_slave(struct bonding *bond)
1112 struct slave *best_slave;
1115 best_slave = bond_find_best_slave(bond);
1116 if (best_slave != bond->curr_active_slave) {
1117 bond_change_active_slave(bond, best_slave);
1118 rv = bond_set_carrier(bond);
1122 if (netif_carrier_ok(bond->dev)) {
1123 printk(KERN_INFO DRV_NAME
1124 ": %s: first active interface up!\n",
1127 printk(KERN_INFO DRV_NAME ": %s: "
1128 "now running without any active interface !\n",
1134 /*--------------------------- slave list handling ---------------------------*/
1137 * This function attaches the slave to the end of list.
1139 * bond->lock held for writing by caller.
1141 static void bond_attach_slave(struct bonding *bond, struct slave *new_slave)
1143 if (bond->first_slave == NULL) { /* attaching the first slave */
1144 new_slave->next = new_slave;
1145 new_slave->prev = new_slave;
1146 bond->first_slave = new_slave;
1148 new_slave->next = bond->first_slave;
1149 new_slave->prev = bond->first_slave->prev;
1150 new_slave->next->prev = new_slave;
1151 new_slave->prev->next = new_slave;
1158 * This function detaches the slave from the list.
1159 * WARNING: no check is made to verify if the slave effectively
1160 * belongs to <bond>.
1161 * Nothing is freed on return, structures are just unchained.
1162 * If any slave pointer in bond was pointing to <slave>,
1163 * it should be changed by the calling function.
1165 * bond->lock held for writing by caller.
1167 static void bond_detach_slave(struct bonding *bond, struct slave *slave)
1170 slave->next->prev = slave->prev;
1174 slave->prev->next = slave->next;
1177 if (bond->first_slave == slave) { /* slave is the first slave */
1178 if (bond->slave_cnt > 1) { /* there are more slave */
1179 bond->first_slave = slave->next;
1181 bond->first_slave = NULL; /* slave was the last one */
1190 /*---------------------------------- IOCTL ----------------------------------*/
1192 int bond_sethwaddr(struct net_device *bond_dev, struct net_device *slave_dev)
1194 dprintk("bond_dev=%p\n", bond_dev);
1195 dprintk("slave_dev=%p\n", slave_dev);
1196 dprintk("slave_dev->addr_len=%d\n", slave_dev->addr_len);
1197 memcpy(bond_dev->dev_addr, slave_dev->dev_addr, slave_dev->addr_len);
1201 #define BOND_INTERSECT_FEATURES \
1202 (NETIF_F_SG | NETIF_F_ALL_CSUM | NETIF_F_TSO | NETIF_F_UFO)
1205 * Compute the common dev->feature set available to all slaves. Some
1206 * feature bits are managed elsewhere, so preserve feature bits set on
1207 * master device that are not part of the examined set.
1209 static int bond_compute_features(struct bonding *bond)
1211 unsigned long features = BOND_INTERSECT_FEATURES;
1212 struct slave *slave;
1213 struct net_device *bond_dev = bond->dev;
1216 bond_for_each_slave(bond, slave, i)
1217 features &= (slave->dev->features & BOND_INTERSECT_FEATURES);
1219 if ((features & NETIF_F_SG) &&
1220 !(features & NETIF_F_ALL_CSUM))
1221 features &= ~NETIF_F_SG;
1224 * features will include NETIF_F_TSO (NETIF_F_UFO) iff all
1225 * slave devices support NETIF_F_TSO (NETIF_F_UFO), which
1226 * implies that all slaves also support scatter-gather
1227 * (NETIF_F_SG), which implies that features also includes
1228 * NETIF_F_SG. So no need to check whether we have an
1229 * illegal combination of NETIF_F_{TSO,UFO} and
1233 features |= (bond_dev->features & ~BOND_INTERSECT_FEATURES);
1234 bond_dev->features = features;
1239 /* enslave device <slave> to bond device <master> */
1240 int bond_enslave(struct net_device *bond_dev, struct net_device *slave_dev)
1242 struct bonding *bond = bond_dev->priv;
1243 struct slave *new_slave = NULL;
1244 struct dev_mc_list *dmi;
1245 struct sockaddr addr;
1247 int old_features = bond_dev->features;
1250 if (!bond->params.use_carrier && slave_dev->ethtool_ops == NULL &&
1251 slave_dev->do_ioctl == NULL) {
1252 printk(KERN_WARNING DRV_NAME
1253 ": %s: Warning: no link monitoring support for %s\n",
1254 bond_dev->name, slave_dev->name);
1257 /* bond must be initialized by bond_open() before enslaving */
1258 if (!(bond_dev->flags & IFF_UP)) {
1259 dprintk("Error, master_dev is not up\n");
1263 /* already enslaved */
1264 if (slave_dev->flags & IFF_SLAVE) {
1265 dprintk("Error, Device was already enslaved\n");
1269 /* vlan challenged mutual exclusion */
1270 /* no need to lock since we're protected by rtnl_lock */
1271 if (slave_dev->features & NETIF_F_VLAN_CHALLENGED) {
1272 dprintk("%s: NETIF_F_VLAN_CHALLENGED\n", slave_dev->name);
1273 if (!list_empty(&bond->vlan_list)) {
1274 printk(KERN_ERR DRV_NAME
1275 ": %s: Error: cannot enslave VLAN "
1276 "challenged slave %s on VLAN enabled "
1277 "bond %s\n", bond_dev->name, slave_dev->name,
1281 printk(KERN_WARNING DRV_NAME
1282 ": %s: Warning: enslaved VLAN challenged "
1283 "slave %s. Adding VLANs will be blocked as "
1284 "long as %s is part of bond %s\n",
1285 bond_dev->name, slave_dev->name, slave_dev->name,
1287 bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
1290 dprintk("%s: ! NETIF_F_VLAN_CHALLENGED\n", slave_dev->name);
1291 if (bond->slave_cnt == 0) {
1292 /* First slave, and it is not VLAN challenged,
1293 * so remove the block of adding VLANs over the bond.
1295 bond_dev->features &= ~NETIF_F_VLAN_CHALLENGED;
1300 * Old ifenslave binaries are no longer supported. These can
1301 * be identified with moderate accurary by the state of the slave:
1302 * the current ifenslave will set the interface down prior to
1303 * enslaving it; the old ifenslave will not.
1305 if ((slave_dev->flags & IFF_UP)) {
1306 printk(KERN_ERR DRV_NAME ": %s is up. "
1307 "This may be due to an out of date ifenslave.\n",
1310 goto err_undo_flags;
1313 if (slave_dev->set_mac_address == NULL) {
1314 printk(KERN_ERR DRV_NAME
1315 ": %s: Error: The slave device you specified does "
1316 "not support setting the MAC address. "
1317 "Your kernel likely does not support slave "
1318 "devices.\n", bond_dev->name);
1320 goto err_undo_flags;
1323 new_slave = kmalloc(sizeof(struct slave), GFP_KERNEL);
1326 goto err_undo_flags;
1329 memset(new_slave, 0, sizeof(struct slave));
1331 /* save slave's original flags before calling
1332 * netdev_set_master and dev_open
1334 new_slave->original_flags = slave_dev->flags;
1337 * Save slave's original ("permanent") mac address for modes
1338 * that need it, and for restoring it upon release, and then
1339 * set it to the master's address
1341 memcpy(new_slave->perm_hwaddr, slave_dev->dev_addr, ETH_ALEN);
1344 * Set slave to master's mac address. The application already
1345 * set the master's mac address to that of the first slave
1347 memcpy(addr.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
1348 addr.sa_family = slave_dev->type;
1349 res = dev_set_mac_address(slave_dev, &addr);
1351 dprintk("Error %d calling set_mac_address\n", res);
1355 /* open the slave since the application closed it */
1356 res = dev_open(slave_dev);
1358 dprintk("Openning slave %s failed\n", slave_dev->name);
1359 goto err_restore_mac;
1362 res = netdev_set_master(slave_dev, bond_dev);
1364 dprintk("Error %d calling netdev_set_master\n", res);
1368 new_slave->dev = slave_dev;
1370 if ((bond->params.mode == BOND_MODE_TLB) ||
1371 (bond->params.mode == BOND_MODE_ALB)) {
1372 /* bond_alb_init_slave() must be called before all other stages since
1373 * it might fail and we do not want to have to undo everything
1375 res = bond_alb_init_slave(bond, new_slave);
1377 goto err_unset_master;
1381 /* If the mode USES_PRIMARY, then the new slave gets the
1382 * master's promisc (and mc) settings only if it becomes the
1383 * curr_active_slave, and that is taken care of later when calling
1384 * bond_change_active()
1386 if (!USES_PRIMARY(bond->params.mode)) {
1387 /* set promiscuity level to new slave */
1388 if (bond_dev->flags & IFF_PROMISC) {
1389 dev_set_promiscuity(slave_dev, 1);
1392 /* set allmulti level to new slave */
1393 if (bond_dev->flags & IFF_ALLMULTI) {
1394 dev_set_allmulti(slave_dev, 1);
1397 /* upload master's mc_list to new slave */
1398 for (dmi = bond_dev->mc_list; dmi; dmi = dmi->next) {
1399 dev_mc_add (slave_dev, dmi->dmi_addr, dmi->dmi_addrlen, 0);
1403 if (bond->params.mode == BOND_MODE_8023AD) {
1404 /* add lacpdu mc addr to mc list */
1405 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
1407 dev_mc_add(slave_dev, lacpdu_multicast, ETH_ALEN, 0);
1410 bond_add_vlans_on_slave(bond, slave_dev);
1412 write_lock_bh(&bond->lock);
1414 bond_attach_slave(bond, new_slave);
1416 new_slave->delay = 0;
1417 new_slave->link_failure_count = 0;
1419 bond_compute_features(bond);
1421 if (bond->params.miimon && !bond->params.use_carrier) {
1422 link_reporting = bond_check_dev_link(bond, slave_dev, 1);
1424 if ((link_reporting == -1) && !bond->params.arp_interval) {
1426 * miimon is set but a bonded network driver
1427 * does not support ETHTOOL/MII and
1428 * arp_interval is not set. Note: if
1429 * use_carrier is enabled, we will never go
1430 * here (because netif_carrier is always
1431 * supported); thus, we don't need to change
1432 * the messages for netif_carrier.
1434 printk(KERN_WARNING DRV_NAME
1435 ": %s: Warning: MII and ETHTOOL support not "
1436 "available for interface %s, and "
1437 "arp_interval/arp_ip_target module parameters "
1438 "not specified, thus bonding will not detect "
1439 "link failures! see bonding.txt for details.\n",
1440 bond_dev->name, slave_dev->name);
1441 } else if (link_reporting == -1) {
1442 /* unable get link status using mii/ethtool */
1443 printk(KERN_WARNING DRV_NAME
1444 ": %s: Warning: can't get link status from "
1445 "interface %s; the network driver associated "
1446 "with this interface does not support MII or "
1447 "ETHTOOL link status reporting, thus miimon "
1448 "has no effect on this interface.\n",
1449 bond_dev->name, slave_dev->name);
1453 /* check for initial state */
1454 if (!bond->params.miimon ||
1455 (bond_check_dev_link(bond, slave_dev, 0) == BMSR_LSTATUS)) {
1456 if (bond->params.updelay) {
1457 dprintk("Initial state of slave_dev is "
1458 "BOND_LINK_BACK\n");
1459 new_slave->link = BOND_LINK_BACK;
1460 new_slave->delay = bond->params.updelay;
1462 dprintk("Initial state of slave_dev is "
1464 new_slave->link = BOND_LINK_UP;
1466 new_slave->jiffies = jiffies;
1468 dprintk("Initial state of slave_dev is "
1469 "BOND_LINK_DOWN\n");
1470 new_slave->link = BOND_LINK_DOWN;
1473 if (bond_update_speed_duplex(new_slave) &&
1474 (new_slave->link != BOND_LINK_DOWN)) {
1475 printk(KERN_WARNING DRV_NAME
1476 ": %s: Warning: failed to get speed and duplex from %s, "
1477 "assumed to be 100Mb/sec and Full.\n",
1478 bond_dev->name, new_slave->dev->name);
1480 if (bond->params.mode == BOND_MODE_8023AD) {
1481 printk(KERN_WARNING DRV_NAME
1482 ": %s: Warning: Operation of 802.3ad mode requires ETHTOOL "
1483 "support in base driver for proper aggregator "
1484 "selection.\n", bond_dev->name);
1488 if (USES_PRIMARY(bond->params.mode) && bond->params.primary[0]) {
1489 /* if there is a primary slave, remember it */
1490 if (strcmp(bond->params.primary, new_slave->dev->name) == 0) {
1491 bond->primary_slave = new_slave;
1495 switch (bond->params.mode) {
1496 case BOND_MODE_ACTIVEBACKUP:
1497 /* if we're in active-backup mode, we need one and
1498 * only one active interface. The backup interfaces
1499 * will have their SLAVE_INACTIVE flag set because we
1500 * need them to be drop all packets. Thus, since we
1501 * guarantee that curr_active_slave always point to
1502 * the last usable interface, we just have to verify
1503 * this interface's flag.
1505 if (((!bond->curr_active_slave) ||
1506 (bond->curr_active_slave->dev->priv_flags & IFF_SLAVE_INACTIVE)) &&
1507 (new_slave->link != BOND_LINK_DOWN)) {
1508 /* first slave or no active slave yet, and this link
1509 is OK, so make this interface the active one */
1510 bond_change_active_slave(bond, new_slave);
1511 printk(KERN_INFO DRV_NAME
1512 ": %s: first active interface up!\n",
1514 netif_carrier_on(bond->dev);
1517 dprintk("This is just a backup slave\n");
1518 bond_set_slave_inactive_flags(new_slave);
1521 case BOND_MODE_8023AD:
1522 /* in 802.3ad mode, the internal mechanism
1523 * will activate the slaves in the selected
1526 bond_set_slave_inactive_flags(new_slave);
1527 /* if this is the first slave */
1528 if (bond->slave_cnt == 1) {
1529 SLAVE_AD_INFO(new_slave).id = 1;
1530 /* Initialize AD with the number of times that the AD timer is called in 1 second
1531 * can be called only after the mac address of the bond is set
1533 bond_3ad_initialize(bond, 1000/AD_TIMER_INTERVAL,
1534 bond->params.lacp_fast);
1536 SLAVE_AD_INFO(new_slave).id =
1537 SLAVE_AD_INFO(new_slave->prev).id + 1;
1540 bond_3ad_bind_slave(new_slave);
1544 new_slave->state = BOND_STATE_ACTIVE;
1545 if ((!bond->curr_active_slave) &&
1546 (new_slave->link != BOND_LINK_DOWN)) {
1547 /* first slave or no active slave yet, and this link
1548 * is OK, so make this interface the active one
1550 bond_change_active_slave(bond, new_slave);
1552 bond_set_slave_inactive_flags(new_slave);
1556 dprintk("This slave is always active in trunk mode\n");
1558 /* always active in trunk mode */
1559 new_slave->state = BOND_STATE_ACTIVE;
1561 /* In trunking mode there is little meaning to curr_active_slave
1562 * anyway (it holds no special properties of the bond device),
1563 * so we can change it without calling change_active_interface()
1565 if (!bond->curr_active_slave) {
1566 bond->curr_active_slave = new_slave;
1569 } /* switch(bond_mode) */
1571 bond_set_carrier(bond);
1573 write_unlock_bh(&bond->lock);
1575 res = bond_create_slave_symlinks(bond_dev, slave_dev);
1577 goto err_unset_master;
1579 printk(KERN_INFO DRV_NAME
1580 ": %s: enslaving %s as a%s interface with a%s link.\n",
1581 bond_dev->name, slave_dev->name,
1582 new_slave->state == BOND_STATE_ACTIVE ? "n active" : " backup",
1583 new_slave->link != BOND_LINK_DOWN ? "n up" : " down");
1585 /* enslave is successful */
1588 /* Undo stages on error */
1590 netdev_set_master(slave_dev, NULL);
1593 dev_close(slave_dev);
1596 memcpy(addr.sa_data, new_slave->perm_hwaddr, ETH_ALEN);
1597 addr.sa_family = slave_dev->type;
1598 dev_set_mac_address(slave_dev, &addr);
1604 bond_dev->features = old_features;
1610 * Try to release the slave device <slave> from the bond device <master>
1611 * It is legal to access curr_active_slave without a lock because all the function
1614 * The rules for slave state should be:
1615 * for Active/Backup:
1616 * Active stays on all backups go down
1617 * for Bonded connections:
1618 * The first up interface should be left on and all others downed.
1620 int bond_release(struct net_device *bond_dev, struct net_device *slave_dev)
1622 struct bonding *bond = bond_dev->priv;
1623 struct slave *slave, *oldcurrent;
1624 struct sockaddr addr;
1625 int mac_addr_differ;
1627 /* slave is not a slave or master is not master of this slave */
1628 if (!(slave_dev->flags & IFF_SLAVE) ||
1629 (slave_dev->master != bond_dev)) {
1630 printk(KERN_ERR DRV_NAME
1631 ": %s: Error: cannot release %s.\n",
1632 bond_dev->name, slave_dev->name);
1636 write_lock_bh(&bond->lock);
1638 slave = bond_get_slave_by_dev(bond, slave_dev);
1640 /* not a slave of this bond */
1641 printk(KERN_INFO DRV_NAME
1642 ": %s: %s not enslaved\n",
1643 bond_dev->name, slave_dev->name);
1644 write_unlock_bh(&bond->lock);
1648 mac_addr_differ = memcmp(bond_dev->dev_addr,
1651 if (!mac_addr_differ && (bond->slave_cnt > 1)) {
1652 printk(KERN_WARNING DRV_NAME
1653 ": %s: Warning: the permanent HWaddr of %s "
1654 "- %02X:%02X:%02X:%02X:%02X:%02X - is "
1655 "still in use by %s. Set the HWaddr of "
1656 "%s to a different address to avoid "
1660 slave->perm_hwaddr[0],
1661 slave->perm_hwaddr[1],
1662 slave->perm_hwaddr[2],
1663 slave->perm_hwaddr[3],
1664 slave->perm_hwaddr[4],
1665 slave->perm_hwaddr[5],
1670 /* Inform AD package of unbinding of slave. */
1671 if (bond->params.mode == BOND_MODE_8023AD) {
1672 /* must be called before the slave is
1673 * detached from the list
1675 bond_3ad_unbind_slave(slave);
1678 printk(KERN_INFO DRV_NAME
1679 ": %s: releasing %s interface %s\n",
1681 (slave->state == BOND_STATE_ACTIVE)
1682 ? "active" : "backup",
1685 oldcurrent = bond->curr_active_slave;
1687 bond->current_arp_slave = NULL;
1689 /* release the slave from its bond */
1690 bond_detach_slave(bond, slave);
1692 bond_compute_features(bond);
1694 if (bond->primary_slave == slave) {
1695 bond->primary_slave = NULL;
1698 if (oldcurrent == slave) {
1699 bond_change_active_slave(bond, NULL);
1702 if ((bond->params.mode == BOND_MODE_TLB) ||
1703 (bond->params.mode == BOND_MODE_ALB)) {
1704 /* Must be called only after the slave has been
1705 * detached from the list and the curr_active_slave
1706 * has been cleared (if our_slave == old_current),
1707 * but before a new active slave is selected.
1709 bond_alb_deinit_slave(bond, slave);
1712 if (oldcurrent == slave)
1713 bond_select_active_slave(bond);
1715 if (bond->slave_cnt == 0) {
1716 bond_set_carrier(bond);
1718 /* if the last slave was removed, zero the mac address
1719 * of the master so it will be set by the application
1720 * to the mac address of the first slave
1722 memset(bond_dev->dev_addr, 0, bond_dev->addr_len);
1724 if (list_empty(&bond->vlan_list)) {
1725 bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
1727 printk(KERN_WARNING DRV_NAME
1728 ": %s: Warning: clearing HW address of %s while it "
1729 "still has VLANs.\n",
1730 bond_dev->name, bond_dev->name);
1731 printk(KERN_WARNING DRV_NAME
1732 ": %s: When re-adding slaves, make sure the bond's "
1733 "HW address matches its VLANs'.\n",
1736 } else if ((bond_dev->features & NETIF_F_VLAN_CHALLENGED) &&
1737 !bond_has_challenged_slaves(bond)) {
1738 printk(KERN_INFO DRV_NAME
1739 ": %s: last VLAN challenged slave %s "
1740 "left bond %s. VLAN blocking is removed\n",
1741 bond_dev->name, slave_dev->name, bond_dev->name);
1742 bond_dev->features &= ~NETIF_F_VLAN_CHALLENGED;
1745 write_unlock_bh(&bond->lock);
1747 /* must do this from outside any spinlocks */
1748 bond_destroy_slave_symlinks(bond_dev, slave_dev);
1750 bond_del_vlans_from_slave(bond, slave_dev);
1752 /* If the mode USES_PRIMARY, then we should only remove its
1753 * promisc and mc settings if it was the curr_active_slave, but that was
1754 * already taken care of above when we detached the slave
1756 if (!USES_PRIMARY(bond->params.mode)) {
1757 /* unset promiscuity level from slave */
1758 if (bond_dev->flags & IFF_PROMISC) {
1759 dev_set_promiscuity(slave_dev, -1);
1762 /* unset allmulti level from slave */
1763 if (bond_dev->flags & IFF_ALLMULTI) {
1764 dev_set_allmulti(slave_dev, -1);
1767 /* flush master's mc_list from slave */
1768 bond_mc_list_flush(bond_dev, slave_dev);
1771 netdev_set_master(slave_dev, NULL);
1773 /* close slave before restoring its mac address */
1774 dev_close(slave_dev);
1776 /* restore original ("permanent") mac address */
1777 memcpy(addr.sa_data, slave->perm_hwaddr, ETH_ALEN);
1778 addr.sa_family = slave_dev->type;
1779 dev_set_mac_address(slave_dev, &addr);
1781 slave_dev->priv_flags &= ~(IFF_MASTER_8023AD | IFF_MASTER_ALB |
1782 IFF_SLAVE_INACTIVE);
1786 return 0; /* deletion OK */
1790 * This function releases all slaves.
1792 static int bond_release_all(struct net_device *bond_dev)
1794 struct bonding *bond = bond_dev->priv;
1795 struct slave *slave;
1796 struct net_device *slave_dev;
1797 struct sockaddr addr;
1799 write_lock_bh(&bond->lock);
1801 netif_carrier_off(bond_dev);
1803 if (bond->slave_cnt == 0) {
1807 bond->current_arp_slave = NULL;
1808 bond->primary_slave = NULL;
1809 bond_change_active_slave(bond, NULL);
1811 while ((slave = bond->first_slave) != NULL) {
1812 /* Inform AD package of unbinding of slave
1813 * before slave is detached from the list.
1815 if (bond->params.mode == BOND_MODE_8023AD) {
1816 bond_3ad_unbind_slave(slave);
1819 slave_dev = slave->dev;
1820 bond_detach_slave(bond, slave);
1822 if ((bond->params.mode == BOND_MODE_TLB) ||
1823 (bond->params.mode == BOND_MODE_ALB)) {
1824 /* must be called only after the slave
1825 * has been detached from the list
1827 bond_alb_deinit_slave(bond, slave);
1830 bond_compute_features(bond);
1832 /* now that the slave is detached, unlock and perform
1833 * all the undo steps that should not be called from
1836 write_unlock_bh(&bond->lock);
1838 bond_destroy_slave_symlinks(bond_dev, slave_dev);
1839 bond_del_vlans_from_slave(bond, slave_dev);
1841 /* If the mode USES_PRIMARY, then we should only remove its
1842 * promisc and mc settings if it was the curr_active_slave, but that was
1843 * already taken care of above when we detached the slave
1845 if (!USES_PRIMARY(bond->params.mode)) {
1846 /* unset promiscuity level from slave */
1847 if (bond_dev->flags & IFF_PROMISC) {
1848 dev_set_promiscuity(slave_dev, -1);
1851 /* unset allmulti level from slave */
1852 if (bond_dev->flags & IFF_ALLMULTI) {
1853 dev_set_allmulti(slave_dev, -1);
1856 /* flush master's mc_list from slave */
1857 bond_mc_list_flush(bond_dev, slave_dev);
1860 netdev_set_master(slave_dev, NULL);
1862 /* close slave before restoring its mac address */
1863 dev_close(slave_dev);
1865 /* restore original ("permanent") mac address*/
1866 memcpy(addr.sa_data, slave->perm_hwaddr, ETH_ALEN);
1867 addr.sa_family = slave_dev->type;
1868 dev_set_mac_address(slave_dev, &addr);
1870 slave_dev->priv_flags &= ~(IFF_MASTER_8023AD | IFF_MASTER_ALB |
1871 IFF_SLAVE_INACTIVE);
1875 /* re-acquire the lock before getting the next slave */
1876 write_lock_bh(&bond->lock);
1879 /* zero the mac address of the master so it will be
1880 * set by the application to the mac address of the
1883 memset(bond_dev->dev_addr, 0, bond_dev->addr_len);
1885 if (list_empty(&bond->vlan_list)) {
1886 bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
1888 printk(KERN_WARNING DRV_NAME
1889 ": %s: Warning: clearing HW address of %s while it "
1890 "still has VLANs.\n",
1891 bond_dev->name, bond_dev->name);
1892 printk(KERN_WARNING DRV_NAME
1893 ": %s: When re-adding slaves, make sure the bond's "
1894 "HW address matches its VLANs'.\n",
1898 printk(KERN_INFO DRV_NAME
1899 ": %s: released all slaves\n",
1903 write_unlock_bh(&bond->lock);
1909 * This function changes the active slave to slave <slave_dev>.
1910 * It returns -EINVAL in the following cases.
1911 * - <slave_dev> is not found in the list.
1912 * - There is not active slave now.
1913 * - <slave_dev> is already active.
1914 * - The link state of <slave_dev> is not BOND_LINK_UP.
1915 * - <slave_dev> is not running.
1916 * In these cases, this fuction does nothing.
1917 * In the other cases, currnt_slave pointer is changed and 0 is returned.
1919 static int bond_ioctl_change_active(struct net_device *bond_dev, struct net_device *slave_dev)
1921 struct bonding *bond = bond_dev->priv;
1922 struct slave *old_active = NULL;
1923 struct slave *new_active = NULL;
1926 if (!USES_PRIMARY(bond->params.mode)) {
1930 /* Verify that master_dev is indeed the master of slave_dev */
1931 if (!(slave_dev->flags & IFF_SLAVE) ||
1932 (slave_dev->master != bond_dev)) {
1936 write_lock_bh(&bond->lock);
1938 old_active = bond->curr_active_slave;
1939 new_active = bond_get_slave_by_dev(bond, slave_dev);
1942 * Changing to the current active: do nothing; return success.
1944 if (new_active && (new_active == old_active)) {
1945 write_unlock_bh(&bond->lock);
1951 (new_active->link == BOND_LINK_UP) &&
1952 IS_UP(new_active->dev)) {
1953 bond_change_active_slave(bond, new_active);
1958 write_unlock_bh(&bond->lock);
1963 static int bond_info_query(struct net_device *bond_dev, struct ifbond *info)
1965 struct bonding *bond = bond_dev->priv;
1967 info->bond_mode = bond->params.mode;
1968 info->miimon = bond->params.miimon;
1970 read_lock_bh(&bond->lock);
1971 info->num_slaves = bond->slave_cnt;
1972 read_unlock_bh(&bond->lock);
1977 static int bond_slave_info_query(struct net_device *bond_dev, struct ifslave *info)
1979 struct bonding *bond = bond_dev->priv;
1980 struct slave *slave;
1983 if (info->slave_id < 0) {
1987 read_lock_bh(&bond->lock);
1989 bond_for_each_slave(bond, slave, i) {
1990 if (i == (int)info->slave_id) {
1996 read_unlock_bh(&bond->lock);
1999 strcpy(info->slave_name, slave->dev->name);
2000 info->link = slave->link;
2001 info->state = slave->state;
2002 info->link_failure_count = slave->link_failure_count;
2010 /*-------------------------------- Monitoring -------------------------------*/
2012 /* this function is called regularly to monitor each slave's link. */
2013 void bond_mii_monitor(struct net_device *bond_dev)
2015 struct bonding *bond = bond_dev->priv;
2016 struct slave *slave, *oldcurrent;
2017 int do_failover = 0;
2021 read_lock(&bond->lock);
2023 delta_in_ticks = (bond->params.miimon * HZ) / 1000;
2025 if (bond->kill_timers) {
2029 if (bond->slave_cnt == 0) {
2033 /* we will try to read the link status of each of our slaves, and
2034 * set their IFF_RUNNING flag appropriately. For each slave not
2035 * supporting MII status, we won't do anything so that a user-space
2036 * program could monitor the link itself if needed.
2039 read_lock(&bond->curr_slave_lock);
2040 oldcurrent = bond->curr_active_slave;
2041 read_unlock(&bond->curr_slave_lock);
2043 bond_for_each_slave(bond, slave, i) {
2044 struct net_device *slave_dev = slave->dev;
2046 u16 old_speed = slave->speed;
2047 u8 old_duplex = slave->duplex;
2049 link_state = bond_check_dev_link(bond, slave_dev, 0);
2051 switch (slave->link) {
2052 case BOND_LINK_UP: /* the link was up */
2053 if (link_state == BMSR_LSTATUS) {
2054 /* link stays up, nothing more to do */
2056 } else { /* link going down */
2057 slave->link = BOND_LINK_FAIL;
2058 slave->delay = bond->params.downdelay;
2060 if (slave->link_failure_count < UINT_MAX) {
2061 slave->link_failure_count++;
2064 if (bond->params.downdelay) {
2065 printk(KERN_INFO DRV_NAME
2066 ": %s: link status down for %s "
2067 "interface %s, disabling it in "
2071 ? ((bond->params.mode == BOND_MODE_ACTIVEBACKUP)
2072 ? ((slave == oldcurrent)
2073 ? "active " : "backup ")
2077 bond->params.downdelay * bond->params.miimon);
2080 /* no break ! fall through the BOND_LINK_FAIL test to
2081 ensure proper action to be taken
2083 case BOND_LINK_FAIL: /* the link has just gone down */
2084 if (link_state != BMSR_LSTATUS) {
2085 /* link stays down */
2086 if (slave->delay <= 0) {
2087 /* link down for too long time */
2088 slave->link = BOND_LINK_DOWN;
2090 /* in active/backup mode, we must
2091 * completely disable this interface
2093 if ((bond->params.mode == BOND_MODE_ACTIVEBACKUP) ||
2094 (bond->params.mode == BOND_MODE_8023AD)) {
2095 bond_set_slave_inactive_flags(slave);
2098 printk(KERN_INFO DRV_NAME
2099 ": %s: link status definitely "
2100 "down for interface %s, "
2105 /* notify ad that the link status has changed */
2106 if (bond->params.mode == BOND_MODE_8023AD) {
2107 bond_3ad_handle_link_change(slave, BOND_LINK_DOWN);
2110 if ((bond->params.mode == BOND_MODE_TLB) ||
2111 (bond->params.mode == BOND_MODE_ALB)) {
2112 bond_alb_handle_link_change(bond, slave, BOND_LINK_DOWN);
2115 if (slave == oldcurrent) {
2123 slave->link = BOND_LINK_UP;
2124 slave->jiffies = jiffies;
2125 printk(KERN_INFO DRV_NAME
2126 ": %s: link status up again after %d "
2127 "ms for interface %s.\n",
2129 (bond->params.downdelay - slave->delay) * bond->params.miimon,
2133 case BOND_LINK_DOWN: /* the link was down */
2134 if (link_state != BMSR_LSTATUS) {
2135 /* the link stays down, nothing more to do */
2137 } else { /* link going up */
2138 slave->link = BOND_LINK_BACK;
2139 slave->delay = bond->params.updelay;
2141 if (bond->params.updelay) {
2142 /* if updelay == 0, no need to
2143 advertise about a 0 ms delay */
2144 printk(KERN_INFO DRV_NAME
2145 ": %s: link status up for "
2146 "interface %s, enabling it "
2150 bond->params.updelay * bond->params.miimon);
2153 /* no break ! fall through the BOND_LINK_BACK state in
2154 case there's something to do.
2156 case BOND_LINK_BACK: /* the link has just come back */
2157 if (link_state != BMSR_LSTATUS) {
2158 /* link down again */
2159 slave->link = BOND_LINK_DOWN;
2161 printk(KERN_INFO DRV_NAME
2162 ": %s: link status down again after %d "
2163 "ms for interface %s.\n",
2165 (bond->params.updelay - slave->delay) * bond->params.miimon,
2169 if (slave->delay == 0) {
2170 /* now the link has been up for long time enough */
2171 slave->link = BOND_LINK_UP;
2172 slave->jiffies = jiffies;
2174 if (bond->params.mode == BOND_MODE_8023AD) {
2175 /* prevent it from being the active one */
2176 slave->state = BOND_STATE_BACKUP;
2177 } else if (bond->params.mode != BOND_MODE_ACTIVEBACKUP) {
2178 /* make it immediately active */
2179 slave->state = BOND_STATE_ACTIVE;
2180 } else if (slave != bond->primary_slave) {
2181 /* prevent it from being the active one */
2182 slave->state = BOND_STATE_BACKUP;
2185 printk(KERN_INFO DRV_NAME
2186 ": %s: link status definitely "
2187 "up for interface %s.\n",
2191 /* notify ad that the link status has changed */
2192 if (bond->params.mode == BOND_MODE_8023AD) {
2193 bond_3ad_handle_link_change(slave, BOND_LINK_UP);
2196 if ((bond->params.mode == BOND_MODE_TLB) ||
2197 (bond->params.mode == BOND_MODE_ALB)) {
2198 bond_alb_handle_link_change(bond, slave, BOND_LINK_UP);
2201 if ((!oldcurrent) ||
2202 (slave == bond->primary_slave)) {
2211 /* Should not happen */
2212 printk(KERN_ERR DRV_NAME
2213 ": %s: Error: %s Illegal value (link=%d)\n",
2218 } /* end of switch (slave->link) */
2220 bond_update_speed_duplex(slave);
2222 if (bond->params.mode == BOND_MODE_8023AD) {
2223 if (old_speed != slave->speed) {
2224 bond_3ad_adapter_speed_changed(slave);
2227 if (old_duplex != slave->duplex) {
2228 bond_3ad_adapter_duplex_changed(slave);
2235 write_lock(&bond->curr_slave_lock);
2237 bond_select_active_slave(bond);
2239 write_unlock(&bond->curr_slave_lock);
2241 bond_set_carrier(bond);
2244 if (bond->params.miimon) {
2245 mod_timer(&bond->mii_timer, jiffies + delta_in_ticks);
2248 read_unlock(&bond->lock);
2252 static u32 bond_glean_dev_ip(struct net_device *dev)
2254 struct in_device *idev;
2255 struct in_ifaddr *ifa;
2262 idev = __in_dev_get_rcu(dev);
2266 ifa = idev->ifa_list;
2270 addr = ifa->ifa_local;
2276 static int bond_has_ip(struct bonding *bond)
2278 struct vlan_entry *vlan, *vlan_next;
2280 if (bond->master_ip)
2283 if (list_empty(&bond->vlan_list))
2286 list_for_each_entry_safe(vlan, vlan_next, &bond->vlan_list,
2296 * We go to the (large) trouble of VLAN tagging ARP frames because
2297 * switches in VLAN mode (especially if ports are configured as
2298 * "native" to a VLAN) might not pass non-tagged frames.
2300 static void bond_arp_send(struct net_device *slave_dev, int arp_op, u32 dest_ip, u32 src_ip, unsigned short vlan_id)
2302 struct sk_buff *skb;
2304 dprintk("arp %d on slave %s: dst %x src %x vid %d\n", arp_op,
2305 slave_dev->name, dest_ip, src_ip, vlan_id);
2307 skb = arp_create(arp_op, ETH_P_ARP, dest_ip, slave_dev, src_ip,
2308 NULL, slave_dev->dev_addr, NULL);
2311 printk(KERN_ERR DRV_NAME ": ARP packet allocation failed\n");
2315 skb = vlan_put_tag(skb, vlan_id);
2317 printk(KERN_ERR DRV_NAME ": failed to insert VLAN tag\n");
2325 static void bond_arp_send_all(struct bonding *bond, struct slave *slave)
2328 u32 *targets = bond->params.arp_targets;
2329 struct vlan_entry *vlan, *vlan_next;
2330 struct net_device *vlan_dev;
2334 for (i = 0; (i < BOND_MAX_ARP_TARGETS); i++) {
2337 dprintk("basa: target %x\n", targets[i]);
2338 if (list_empty(&bond->vlan_list)) {
2339 dprintk("basa: empty vlan: arp_send\n");
2340 bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2341 bond->master_ip, 0);
2346 * If VLANs are configured, we do a route lookup to
2347 * determine which VLAN interface would be used, so we
2348 * can tag the ARP with the proper VLAN tag.
2350 memset(&fl, 0, sizeof(fl));
2351 fl.fl4_dst = targets[i];
2352 fl.fl4_tos = RTO_ONLINK;
2354 rv = ip_route_output_key(&rt, &fl);
2356 if (net_ratelimit()) {
2357 printk(KERN_WARNING DRV_NAME
2358 ": %s: no route to arp_ip_target %u.%u.%u.%u\n",
2359 bond->dev->name, NIPQUAD(fl.fl4_dst));
2365 * This target is not on a VLAN
2367 if (rt->u.dst.dev == bond->dev) {
2369 dprintk("basa: rtdev == bond->dev: arp_send\n");
2370 bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2371 bond->master_ip, 0);
2376 list_for_each_entry_safe(vlan, vlan_next, &bond->vlan_list,
2378 vlan_dev = bond->vlgrp->vlan_devices[vlan->vlan_id];
2379 if (vlan_dev == rt->u.dst.dev) {
2380 vlan_id = vlan->vlan_id;
2381 dprintk("basa: vlan match on %s %d\n",
2382 vlan_dev->name, vlan_id);
2389 bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2390 vlan->vlan_ip, vlan_id);
2394 if (net_ratelimit()) {
2395 printk(KERN_WARNING DRV_NAME
2396 ": %s: no path to arp_ip_target %u.%u.%u.%u via rt.dev %s\n",
2397 bond->dev->name, NIPQUAD(fl.fl4_dst),
2398 rt->u.dst.dev ? rt->u.dst.dev->name : "NULL");
2405 * Kick out a gratuitous ARP for an IP on the bonding master plus one
2406 * for each VLAN above us.
2408 static void bond_send_gratuitous_arp(struct bonding *bond)
2410 struct slave *slave = bond->curr_active_slave;
2411 struct vlan_entry *vlan;
2412 struct net_device *vlan_dev;
2414 dprintk("bond_send_grat_arp: bond %s slave %s\n", bond->dev->name,
2415 slave ? slave->dev->name : "NULL");
2419 if (bond->master_ip) {
2420 bond_arp_send(slave->dev, ARPOP_REPLY, bond->master_ip,
2421 bond->master_ip, 0);
2424 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
2425 vlan_dev = bond->vlgrp->vlan_devices[vlan->vlan_id];
2426 if (vlan->vlan_ip) {
2427 bond_arp_send(slave->dev, ARPOP_REPLY, vlan->vlan_ip,
2428 vlan->vlan_ip, vlan->vlan_id);
2434 * this function is called regularly to monitor each slave's link
2435 * ensuring that traffic is being sent and received when arp monitoring
2436 * is used in load-balancing mode. if the adapter has been dormant, then an
2437 * arp is transmitted to generate traffic. see activebackup_arp_monitor for
2438 * arp monitoring in active backup mode.
2440 void bond_loadbalance_arp_mon(struct net_device *bond_dev)
2442 struct bonding *bond = bond_dev->priv;
2443 struct slave *slave, *oldcurrent;
2444 int do_failover = 0;
2448 read_lock(&bond->lock);
2450 delta_in_ticks = (bond->params.arp_interval * HZ) / 1000;
2452 if (bond->kill_timers) {
2456 if (bond->slave_cnt == 0) {
2460 read_lock(&bond->curr_slave_lock);
2461 oldcurrent = bond->curr_active_slave;
2462 read_unlock(&bond->curr_slave_lock);
2464 /* see if any of the previous devices are up now (i.e. they have
2465 * xmt and rcv traffic). the curr_active_slave does not come into
2466 * the picture unless it is null. also, slave->jiffies is not needed
2467 * here because we send an arp on each slave and give a slave as
2468 * long as it needs to get the tx/rx within the delta.
2469 * TODO: what about up/down delay in arp mode? it wasn't here before
2472 bond_for_each_slave(bond, slave, i) {
2473 if (slave->link != BOND_LINK_UP) {
2474 if (((jiffies - slave->dev->trans_start) <= delta_in_ticks) &&
2475 ((jiffies - slave->dev->last_rx) <= delta_in_ticks)) {
2477 slave->link = BOND_LINK_UP;
2478 slave->state = BOND_STATE_ACTIVE;
2480 /* primary_slave has no meaning in round-robin
2481 * mode. the window of a slave being up and
2482 * curr_active_slave being null after enslaving
2486 printk(KERN_INFO DRV_NAME
2487 ": %s: link status definitely "
2488 "up for interface %s, ",
2493 printk(KERN_INFO DRV_NAME
2494 ": %s: interface %s is now up\n",
2500 /* slave->link == BOND_LINK_UP */
2502 /* not all switches will respond to an arp request
2503 * when the source ip is 0, so don't take the link down
2504 * if we don't know our ip yet
2506 if (((jiffies - slave->dev->trans_start) >= (2*delta_in_ticks)) ||
2507 (((jiffies - slave->dev->last_rx) >= (2*delta_in_ticks)) &&
2508 bond_has_ip(bond))) {
2510 slave->link = BOND_LINK_DOWN;
2511 slave->state = BOND_STATE_BACKUP;
2513 if (slave->link_failure_count < UINT_MAX) {
2514 slave->link_failure_count++;
2517 printk(KERN_INFO DRV_NAME
2518 ": %s: interface %s is now down.\n",
2522 if (slave == oldcurrent) {
2528 /* note: if switch is in round-robin mode, all links
2529 * must tx arp to ensure all links rx an arp - otherwise
2530 * links may oscillate or not come up at all; if switch is
2531 * in something like xor mode, there is nothing we can
2532 * do - all replies will be rx'ed on same link causing slaves
2533 * to be unstable during low/no traffic periods
2535 if (IS_UP(slave->dev)) {
2536 bond_arp_send_all(bond, slave);
2541 write_lock(&bond->curr_slave_lock);
2543 bond_select_active_slave(bond);
2545 write_unlock(&bond->curr_slave_lock);
2549 if (bond->params.arp_interval) {
2550 mod_timer(&bond->arp_timer, jiffies + delta_in_ticks);
2553 read_unlock(&bond->lock);
2557 * When using arp monitoring in active-backup mode, this function is
2558 * called to determine if any backup slaves have went down or a new
2559 * current slave needs to be found.
2560 * The backup slaves never generate traffic, they are considered up by merely
2561 * receiving traffic. If the current slave goes down, each backup slave will
2562 * be given the opportunity to tx/rx an arp before being taken down - this
2563 * prevents all slaves from being taken down due to the current slave not
2564 * sending any traffic for the backups to receive. The arps are not necessarily
2565 * necessary, any tx and rx traffic will keep the current slave up. While any
2566 * rx traffic will keep the backup slaves up, the current slave is responsible
2567 * for generating traffic to keep them up regardless of any other traffic they
2568 * may have received.
2569 * see loadbalance_arp_monitor for arp monitoring in load balancing mode
2571 void bond_activebackup_arp_mon(struct net_device *bond_dev)
2573 struct bonding *bond = bond_dev->priv;
2574 struct slave *slave;
2578 read_lock(&bond->lock);
2580 delta_in_ticks = (bond->params.arp_interval * HZ) / 1000;
2582 if (bond->kill_timers) {
2586 if (bond->slave_cnt == 0) {
2590 /* determine if any slave has come up or any backup slave has
2592 * TODO: what about up/down delay in arp mode? it wasn't here before
2595 bond_for_each_slave(bond, slave, i) {
2596 if (slave->link != BOND_LINK_UP) {
2597 if ((jiffies - slave->dev->last_rx) <= delta_in_ticks) {
2599 slave->link = BOND_LINK_UP;
2601 write_lock(&bond->curr_slave_lock);
2603 if ((!bond->curr_active_slave) &&
2604 ((jiffies - slave->dev->trans_start) <= delta_in_ticks)) {
2605 bond_change_active_slave(bond, slave);
2606 bond->current_arp_slave = NULL;
2607 } else if (bond->curr_active_slave != slave) {
2608 /* this slave has just come up but we
2609 * already have a current slave; this
2610 * can also happen if bond_enslave adds
2611 * a new slave that is up while we are
2612 * searching for a new slave
2614 bond_set_slave_inactive_flags(slave);
2615 bond->current_arp_slave = NULL;
2618 bond_set_carrier(bond);
2620 if (slave == bond->curr_active_slave) {
2621 printk(KERN_INFO DRV_NAME
2622 ": %s: %s is up and now the "
2623 "active interface\n",
2626 netif_carrier_on(bond->dev);
2628 printk(KERN_INFO DRV_NAME
2629 ": %s: backup interface %s is "
2635 write_unlock(&bond->curr_slave_lock);
2638 read_lock(&bond->curr_slave_lock);
2640 if ((slave != bond->curr_active_slave) &&
2641 (!bond->current_arp_slave) &&
2642 (((jiffies - slave->dev->last_rx) >= 3*delta_in_ticks) &&
2643 bond_has_ip(bond))) {
2644 /* a backup slave has gone down; three times
2645 * the delta allows the current slave to be
2646 * taken out before the backup slave.
2647 * note: a non-null current_arp_slave indicates
2648 * the curr_active_slave went down and we are
2649 * searching for a new one; under this
2650 * condition we only take the curr_active_slave
2651 * down - this gives each slave a chance to
2652 * tx/rx traffic before being taken out
2655 read_unlock(&bond->curr_slave_lock);
2657 slave->link = BOND_LINK_DOWN;
2659 if (slave->link_failure_count < UINT_MAX) {
2660 slave->link_failure_count++;
2663 bond_set_slave_inactive_flags(slave);
2665 printk(KERN_INFO DRV_NAME
2666 ": %s: backup interface %s is now down\n",
2670 read_unlock(&bond->curr_slave_lock);
2675 read_lock(&bond->curr_slave_lock);
2676 slave = bond->curr_active_slave;
2677 read_unlock(&bond->curr_slave_lock);
2680 /* if we have sent traffic in the past 2*arp_intervals but
2681 * haven't xmit and rx traffic in that time interval, select
2682 * a different slave. slave->jiffies is only updated when
2683 * a slave first becomes the curr_active_slave - not necessarily
2684 * after every arp; this ensures the slave has a full 2*delta
2685 * before being taken out. if a primary is being used, check
2686 * if it is up and needs to take over as the curr_active_slave
2688 if ((((jiffies - slave->dev->trans_start) >= (2*delta_in_ticks)) ||
2689 (((jiffies - slave->dev->last_rx) >= (2*delta_in_ticks)) &&
2690 bond_has_ip(bond))) &&
2691 ((jiffies - slave->jiffies) >= 2*delta_in_ticks)) {
2693 slave->link = BOND_LINK_DOWN;
2695 if (slave->link_failure_count < UINT_MAX) {
2696 slave->link_failure_count++;
2699 printk(KERN_INFO DRV_NAME
2700 ": %s: link status down for active interface "
2701 "%s, disabling it\n",
2705 write_lock(&bond->curr_slave_lock);
2707 bond_select_active_slave(bond);
2708 slave = bond->curr_active_slave;
2710 write_unlock(&bond->curr_slave_lock);
2712 bond->current_arp_slave = slave;
2715 slave->jiffies = jiffies;
2717 } else if ((bond->primary_slave) &&
2718 (bond->primary_slave != slave) &&
2719 (bond->primary_slave->link == BOND_LINK_UP)) {
2720 /* at this point, slave is the curr_active_slave */
2721 printk(KERN_INFO DRV_NAME
2722 ": %s: changing from interface %s to primary "
2726 bond->primary_slave->dev->name);
2728 /* primary is up so switch to it */
2729 write_lock(&bond->curr_slave_lock);
2730 bond_change_active_slave(bond, bond->primary_slave);
2731 write_unlock(&bond->curr_slave_lock);
2733 slave = bond->primary_slave;
2734 slave->jiffies = jiffies;
2736 bond->current_arp_slave = NULL;
2739 /* the current slave must tx an arp to ensure backup slaves
2742 if (slave && bond_has_ip(bond)) {
2743 bond_arp_send_all(bond, slave);
2747 /* if we don't have a curr_active_slave, search for the next available
2748 * backup slave from the current_arp_slave and make it the candidate
2749 * for becoming the curr_active_slave
2752 if (!bond->current_arp_slave) {
2753 bond->current_arp_slave = bond->first_slave;
2756 if (bond->current_arp_slave) {
2757 bond_set_slave_inactive_flags(bond->current_arp_slave);
2759 /* search for next candidate */
2760 bond_for_each_slave_from(bond, slave, i, bond->current_arp_slave->next) {
2761 if (IS_UP(slave->dev)) {
2762 slave->link = BOND_LINK_BACK;
2763 bond_set_slave_active_flags(slave);
2764 bond_arp_send_all(bond, slave);
2765 slave->jiffies = jiffies;
2766 bond->current_arp_slave = slave;
2770 /* if the link state is up at this point, we
2771 * mark it down - this can happen if we have
2772 * simultaneous link failures and
2773 * reselect_active_interface doesn't make this
2774 * one the current slave so it is still marked
2775 * up when it is actually down
2777 if (slave->link == BOND_LINK_UP) {
2778 slave->link = BOND_LINK_DOWN;
2779 if (slave->link_failure_count < UINT_MAX) {
2780 slave->link_failure_count++;
2783 bond_set_slave_inactive_flags(slave);
2785 printk(KERN_INFO DRV_NAME
2786 ": %s: backup interface %s is "
2796 if (bond->params.arp_interval) {
2797 mod_timer(&bond->arp_timer, jiffies + delta_in_ticks);
2800 read_unlock(&bond->lock);
2803 /*------------------------------ proc/seq_file-------------------------------*/
2805 #ifdef CONFIG_PROC_FS
2807 #define SEQ_START_TOKEN ((void *)1)
2809 static void *bond_info_seq_start(struct seq_file *seq, loff_t *pos)
2811 struct bonding *bond = seq->private;
2813 struct slave *slave;
2816 /* make sure the bond won't be taken away */
2817 read_lock(&dev_base_lock);
2818 read_lock_bh(&bond->lock);
2821 return SEQ_START_TOKEN;
2824 bond_for_each_slave(bond, slave, i) {
2825 if (++off == *pos) {
2833 static void *bond_info_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2835 struct bonding *bond = seq->private;
2836 struct slave *slave = v;
2839 if (v == SEQ_START_TOKEN) {
2840 return bond->first_slave;
2843 slave = slave->next;
2845 return (slave == bond->first_slave) ? NULL : slave;
2848 static void bond_info_seq_stop(struct seq_file *seq, void *v)
2850 struct bonding *bond = seq->private;
2852 read_unlock_bh(&bond->lock);
2853 read_unlock(&dev_base_lock);
2856 static void bond_info_show_master(struct seq_file *seq)
2858 struct bonding *bond = seq->private;
2863 read_lock(&bond->curr_slave_lock);
2864 curr = bond->curr_active_slave;
2865 read_unlock(&bond->curr_slave_lock);
2867 seq_printf(seq, "Bonding Mode: %s\n",
2868 bond_mode_name(bond->params.mode));
2870 if (bond->params.mode == BOND_MODE_XOR ||
2871 bond->params.mode == BOND_MODE_8023AD) {
2872 seq_printf(seq, "Transmit Hash Policy: %s (%d)\n",
2873 xmit_hashtype_tbl[bond->params.xmit_policy].modename,
2874 bond->params.xmit_policy);
2877 if (USES_PRIMARY(bond->params.mode)) {
2878 seq_printf(seq, "Primary Slave: %s\n",
2879 (bond->primary_slave) ?
2880 bond->primary_slave->dev->name : "None");
2882 seq_printf(seq, "Currently Active Slave: %s\n",
2883 (curr) ? curr->dev->name : "None");
2886 seq_printf(seq, "MII Status: %s\n", netif_carrier_ok(bond->dev) ?
2888 seq_printf(seq, "MII Polling Interval (ms): %d\n", bond->params.miimon);
2889 seq_printf(seq, "Up Delay (ms): %d\n",
2890 bond->params.updelay * bond->params.miimon);
2891 seq_printf(seq, "Down Delay (ms): %d\n",
2892 bond->params.downdelay * bond->params.miimon);
2895 /* ARP information */
2896 if(bond->params.arp_interval > 0) {
2898 seq_printf(seq, "ARP Polling Interval (ms): %d\n",
2899 bond->params.arp_interval);
2901 seq_printf(seq, "ARP IP target/s (n.n.n.n form):");
2903 for(i = 0; (i < BOND_MAX_ARP_TARGETS) ;i++) {
2904 if (!bond->params.arp_targets[i])
2907 seq_printf(seq, ",");
2908 target = ntohl(bond->params.arp_targets[i]);
2909 seq_printf(seq, " %d.%d.%d.%d", HIPQUAD(target));
2912 seq_printf(seq, "\n");
2915 if (bond->params.mode == BOND_MODE_8023AD) {
2916 struct ad_info ad_info;
2918 seq_puts(seq, "\n802.3ad info\n");
2919 seq_printf(seq, "LACP rate: %s\n",
2920 (bond->params.lacp_fast) ? "fast" : "slow");
2922 if (bond_3ad_get_active_agg_info(bond, &ad_info)) {
2923 seq_printf(seq, "bond %s has no active aggregator\n",
2926 seq_printf(seq, "Active Aggregator Info:\n");
2928 seq_printf(seq, "\tAggregator ID: %d\n",
2929 ad_info.aggregator_id);
2930 seq_printf(seq, "\tNumber of ports: %d\n",
2932 seq_printf(seq, "\tActor Key: %d\n",
2934 seq_printf(seq, "\tPartner Key: %d\n",
2935 ad_info.partner_key);
2936 seq_printf(seq, "\tPartner Mac Address: %02x:%02x:%02x:%02x:%02x:%02x\n",
2937 ad_info.partner_system[0],
2938 ad_info.partner_system[1],
2939 ad_info.partner_system[2],
2940 ad_info.partner_system[3],
2941 ad_info.partner_system[4],
2942 ad_info.partner_system[5]);
2947 static void bond_info_show_slave(struct seq_file *seq, const struct slave *slave)
2949 struct bonding *bond = seq->private;
2951 seq_printf(seq, "\nSlave Interface: %s\n", slave->dev->name);
2952 seq_printf(seq, "MII Status: %s\n",
2953 (slave->link == BOND_LINK_UP) ? "up" : "down");
2954 seq_printf(seq, "Link Failure Count: %d\n",
2955 slave->link_failure_count);
2958 "Permanent HW addr: %02x:%02x:%02x:%02x:%02x:%02x\n",
2959 slave->perm_hwaddr[0], slave->perm_hwaddr[1],
2960 slave->perm_hwaddr[2], slave->perm_hwaddr[3],
2961 slave->perm_hwaddr[4], slave->perm_hwaddr[5]);
2963 if (bond->params.mode == BOND_MODE_8023AD) {
2964 const struct aggregator *agg
2965 = SLAVE_AD_INFO(slave).port.aggregator;
2968 seq_printf(seq, "Aggregator ID: %d\n",
2969 agg->aggregator_identifier);
2971 seq_puts(seq, "Aggregator ID: N/A\n");
2976 static int bond_info_seq_show(struct seq_file *seq, void *v)
2978 if (v == SEQ_START_TOKEN) {
2979 seq_printf(seq, "%s\n", version);
2980 bond_info_show_master(seq);
2982 bond_info_show_slave(seq, v);
2988 static struct seq_operations bond_info_seq_ops = {
2989 .start = bond_info_seq_start,
2990 .next = bond_info_seq_next,
2991 .stop = bond_info_seq_stop,
2992 .show = bond_info_seq_show,
2995 static int bond_info_open(struct inode *inode, struct file *file)
2997 struct seq_file *seq;
2998 struct proc_dir_entry *proc;
3001 res = seq_open(file, &bond_info_seq_ops);
3003 /* recover the pointer buried in proc_dir_entry data */
3004 seq = file->private_data;
3006 seq->private = proc->data;
3012 static struct file_operations bond_info_fops = {
3013 .owner = THIS_MODULE,
3014 .open = bond_info_open,
3016 .llseek = seq_lseek,
3017 .release = seq_release,
3020 static int bond_create_proc_entry(struct bonding *bond)
3022 struct net_device *bond_dev = bond->dev;
3024 if (bond_proc_dir) {
3025 bond->proc_entry = create_proc_entry(bond_dev->name,
3028 if (bond->proc_entry == NULL) {
3029 printk(KERN_WARNING DRV_NAME
3030 ": Warning: Cannot create /proc/net/%s/%s\n",
3031 DRV_NAME, bond_dev->name);
3033 bond->proc_entry->data = bond;
3034 bond->proc_entry->proc_fops = &bond_info_fops;
3035 bond->proc_entry->owner = THIS_MODULE;
3036 memcpy(bond->proc_file_name, bond_dev->name, IFNAMSIZ);
3043 static void bond_remove_proc_entry(struct bonding *bond)
3045 if (bond_proc_dir && bond->proc_entry) {
3046 remove_proc_entry(bond->proc_file_name, bond_proc_dir);
3047 memset(bond->proc_file_name, 0, IFNAMSIZ);
3048 bond->proc_entry = NULL;
3052 /* Create the bonding directory under /proc/net, if doesn't exist yet.
3053 * Caller must hold rtnl_lock.
3055 static void bond_create_proc_dir(void)
3057 int len = strlen(DRV_NAME);
3059 for (bond_proc_dir = proc_net->subdir; bond_proc_dir;
3060 bond_proc_dir = bond_proc_dir->next) {
3061 if ((bond_proc_dir->namelen == len) &&
3062 !memcmp(bond_proc_dir->name, DRV_NAME, len)) {
3067 if (!bond_proc_dir) {
3068 bond_proc_dir = proc_mkdir(DRV_NAME, proc_net);
3069 if (bond_proc_dir) {
3070 bond_proc_dir->owner = THIS_MODULE;
3072 printk(KERN_WARNING DRV_NAME
3073 ": Warning: cannot create /proc/net/%s\n",
3079 /* Destroy the bonding directory under /proc/net, if empty.
3080 * Caller must hold rtnl_lock.
3082 static void bond_destroy_proc_dir(void)
3084 struct proc_dir_entry *de;
3086 if (!bond_proc_dir) {
3090 /* verify that the /proc dir is empty */
3091 for (de = bond_proc_dir->subdir; de; de = de->next) {
3092 /* ignore . and .. */
3093 if (*(de->name) != '.') {
3099 if (bond_proc_dir->owner == THIS_MODULE) {
3100 bond_proc_dir->owner = NULL;
3103 remove_proc_entry(DRV_NAME, proc_net);
3104 bond_proc_dir = NULL;
3107 #endif /* CONFIG_PROC_FS */
3109 /*-------------------------- netdev event handling --------------------------*/
3112 * Change device name
3114 static int bond_event_changename(struct bonding *bond)
3116 #ifdef CONFIG_PROC_FS
3117 bond_remove_proc_entry(bond);
3118 bond_create_proc_entry(bond);
3120 down_write(&(bonding_rwsem));
3121 bond_destroy_sysfs_entry(bond);
3122 bond_create_sysfs_entry(bond);
3123 up_write(&(bonding_rwsem));
3127 static int bond_master_netdev_event(unsigned long event, struct net_device *bond_dev)
3129 struct bonding *event_bond = bond_dev->priv;
3132 case NETDEV_CHANGENAME:
3133 return bond_event_changename(event_bond);
3134 case NETDEV_UNREGISTER:
3136 * TODO: remove a bond from the list?
3146 static int bond_slave_netdev_event(unsigned long event, struct net_device *slave_dev)
3148 struct net_device *bond_dev = slave_dev->master;
3149 struct bonding *bond = bond_dev->priv;
3152 case NETDEV_UNREGISTER:
3154 bond_release(bond_dev, slave_dev);
3159 * TODO: is this what we get if somebody
3160 * sets up a hierarchical bond, then rmmod's
3161 * one of the slave bonding devices?
3166 * ... Or is it this?
3169 case NETDEV_CHANGEMTU:
3171 * TODO: Should slaves be allowed to
3172 * independently alter their MTU? For
3173 * an active-backup bond, slaves need
3174 * not be the same type of device, so
3175 * MTUs may vary. For other modes,
3176 * slaves arguably should have the
3177 * same MTUs. To do this, we'd need to
3178 * take over the slave's change_mtu
3179 * function for the duration of their
3183 case NETDEV_CHANGENAME:
3185 * TODO: handle changing the primary's name
3188 case NETDEV_FEAT_CHANGE:
3189 bond_compute_features(bond);
3199 * bond_netdev_event: handle netdev notifier chain events.
3201 * This function receives events for the netdev chain. The caller (an
3202 * ioctl handler calling blocking_notifier_call_chain) holds the necessary
3203 * locks for us to safely manipulate the slave devices (RTNL lock,
3206 static int bond_netdev_event(struct notifier_block *this, unsigned long event, void *ptr)
3208 struct net_device *event_dev = (struct net_device *)ptr;
3210 dprintk("event_dev: %s, event: %lx\n",
3211 (event_dev ? event_dev->name : "None"),
3214 if (event_dev->flags & IFF_MASTER) {
3215 dprintk("IFF_MASTER\n");
3216 return bond_master_netdev_event(event, event_dev);
3219 if (event_dev->flags & IFF_SLAVE) {
3220 dprintk("IFF_SLAVE\n");
3221 return bond_slave_netdev_event(event, event_dev);
3228 * bond_inetaddr_event: handle inetaddr notifier chain events.
3230 * We keep track of device IPs primarily to use as source addresses in
3231 * ARP monitor probes (rather than spewing out broadcasts all the time).
3233 * We track one IP for the main device (if it has one), plus one per VLAN.
3235 static int bond_inetaddr_event(struct notifier_block *this, unsigned long event, void *ptr)
3237 struct in_ifaddr *ifa = ptr;
3238 struct net_device *vlan_dev, *event_dev = ifa->ifa_dev->dev;
3239 struct bonding *bond, *bond_next;
3240 struct vlan_entry *vlan, *vlan_next;
3242 list_for_each_entry_safe(bond, bond_next, &bond_dev_list, bond_list) {
3243 if (bond->dev == event_dev) {
3246 bond->master_ip = ifa->ifa_local;
3249 bond->master_ip = bond_glean_dev_ip(bond->dev);
3256 if (list_empty(&bond->vlan_list))
3259 list_for_each_entry_safe(vlan, vlan_next, &bond->vlan_list,
3261 vlan_dev = bond->vlgrp->vlan_devices[vlan->vlan_id];
3262 if (vlan_dev == event_dev) {
3265 vlan->vlan_ip = ifa->ifa_local;
3269 bond_glean_dev_ip(vlan_dev);
3280 static struct notifier_block bond_netdev_notifier = {
3281 .notifier_call = bond_netdev_event,
3284 static struct notifier_block bond_inetaddr_notifier = {
3285 .notifier_call = bond_inetaddr_event,
3288 /*-------------------------- Packet type handling ---------------------------*/
3290 /* register to receive lacpdus on a bond */
3291 static void bond_register_lacpdu(struct bonding *bond)
3293 struct packet_type *pk_type = &(BOND_AD_INFO(bond).ad_pkt_type);
3295 /* initialize packet type */
3296 pk_type->type = PKT_TYPE_LACPDU;
3297 pk_type->dev = bond->dev;
3298 pk_type->func = bond_3ad_lacpdu_recv;
3300 dev_add_pack(pk_type);
3303 /* unregister to receive lacpdus on a bond */
3304 static void bond_unregister_lacpdu(struct bonding *bond)
3306 dev_remove_pack(&(BOND_AD_INFO(bond).ad_pkt_type));
3309 /*---------------------------- Hashing Policies -----------------------------*/
3312 * Hash for the the output device based upon layer 3 and layer 4 data. If
3313 * the packet is a frag or not TCP or UDP, just use layer 3 data. If it is
3314 * altogether not IP, mimic bond_xmit_hash_policy_l2()
3316 static int bond_xmit_hash_policy_l34(struct sk_buff *skb,
3317 struct net_device *bond_dev, int count)
3319 struct ethhdr *data = (struct ethhdr *)skb->data;
3320 struct iphdr *iph = skb->nh.iph;
3321 u16 *layer4hdr = (u16 *)((u32 *)iph + iph->ihl);
3324 if (skb->protocol == __constant_htons(ETH_P_IP)) {
3325 if (!(iph->frag_off & __constant_htons(IP_MF|IP_OFFSET)) &&
3326 (iph->protocol == IPPROTO_TCP ||
3327 iph->protocol == IPPROTO_UDP)) {
3328 layer4_xor = htons((*layer4hdr ^ *(layer4hdr + 1)));
3330 return (layer4_xor ^
3331 ((ntohl(iph->saddr ^ iph->daddr)) & 0xffff)) % count;
3335 return (data->h_dest[5] ^ bond_dev->dev_addr[5]) % count;
3339 * Hash for the output device based upon layer 2 data
3341 static int bond_xmit_hash_policy_l2(struct sk_buff *skb,
3342 struct net_device *bond_dev, int count)
3344 struct ethhdr *data = (struct ethhdr *)skb->data;
3346 return (data->h_dest[5] ^ bond_dev->dev_addr[5]) % count;
3349 /*-------------------------- Device entry points ----------------------------*/
3351 static int bond_open(struct net_device *bond_dev)
3353 struct bonding *bond = bond_dev->priv;
3354 struct timer_list *mii_timer = &bond->mii_timer;
3355 struct timer_list *arp_timer = &bond->arp_timer;
3357 bond->kill_timers = 0;
3359 if ((bond->params.mode == BOND_MODE_TLB) ||
3360 (bond->params.mode == BOND_MODE_ALB)) {
3361 struct timer_list *alb_timer = &(BOND_ALB_INFO(bond).alb_timer);
3363 /* bond_alb_initialize must be called before the timer
3366 if (bond_alb_initialize(bond, (bond->params.mode == BOND_MODE_ALB))) {
3367 /* something went wrong - fail the open operation */
3371 init_timer(alb_timer);
3372 alb_timer->expires = jiffies + 1;
3373 alb_timer->data = (unsigned long)bond;
3374 alb_timer->function = (void *)&bond_alb_monitor;
3375 add_timer(alb_timer);
3378 if (bond->params.miimon) { /* link check interval, in milliseconds. */
3379 init_timer(mii_timer);
3380 mii_timer->expires = jiffies + 1;
3381 mii_timer->data = (unsigned long)bond_dev;
3382 mii_timer->function = (void *)&bond_mii_monitor;
3383 add_timer(mii_timer);
3386 if (bond->params.arp_interval) { /* arp interval, in milliseconds. */
3387 init_timer(arp_timer);
3388 arp_timer->expires = jiffies + 1;
3389 arp_timer->data = (unsigned long)bond_dev;
3390 if (bond->params.mode == BOND_MODE_ACTIVEBACKUP) {
3391 arp_timer->function = (void *)&bond_activebackup_arp_mon;
3393 arp_timer->function = (void *)&bond_loadbalance_arp_mon;
3395 add_timer(arp_timer);
3398 if (bond->params.mode == BOND_MODE_8023AD) {
3399 struct timer_list *ad_timer = &(BOND_AD_INFO(bond).ad_timer);
3400 init_timer(ad_timer);
3401 ad_timer->expires = jiffies + 1;
3402 ad_timer->data = (unsigned long)bond;
3403 ad_timer->function = (void *)&bond_3ad_state_machine_handler;
3404 add_timer(ad_timer);
3406 /* register to receive LACPDUs */
3407 bond_register_lacpdu(bond);
3413 static int bond_close(struct net_device *bond_dev)
3415 struct bonding *bond = bond_dev->priv;
3417 if (bond->params.mode == BOND_MODE_8023AD) {
3418 /* Unregister the receive of LACPDUs */
3419 bond_unregister_lacpdu(bond);
3422 write_lock_bh(&bond->lock);
3424 bond_mc_list_destroy(bond);
3426 /* signal timers not to re-arm */
3427 bond->kill_timers = 1;
3429 write_unlock_bh(&bond->lock);
3431 /* del_timer_sync must run without holding the bond->lock
3432 * because a running timer might be trying to hold it too
3435 if (bond->params.miimon) { /* link check interval, in milliseconds. */
3436 del_timer_sync(&bond->mii_timer);
3439 if (bond->params.arp_interval) { /* arp interval, in milliseconds. */
3440 del_timer_sync(&bond->arp_timer);
3443 switch (bond->params.mode) {
3444 case BOND_MODE_8023AD:
3445 del_timer_sync(&(BOND_AD_INFO(bond).ad_timer));
3449 del_timer_sync(&(BOND_ALB_INFO(bond).alb_timer));
3455 /* Release the bonded slaves */
3456 bond_release_all(bond_dev);
3458 if ((bond->params.mode == BOND_MODE_TLB) ||
3459 (bond->params.mode == BOND_MODE_ALB)) {
3460 /* Must be called only after all
3461 * slaves have been released
3463 bond_alb_deinitialize(bond);
3469 static struct net_device_stats *bond_get_stats(struct net_device *bond_dev)
3471 struct bonding *bond = bond_dev->priv;
3472 struct net_device_stats *stats = &(bond->stats), *sstats;
3473 struct slave *slave;
3476 memset(stats, 0, sizeof(struct net_device_stats));
3478 read_lock_bh(&bond->lock);
3480 bond_for_each_slave(bond, slave, i) {
3481 sstats = slave->dev->get_stats(slave->dev);
3483 stats->rx_packets += sstats->rx_packets;
3484 stats->rx_bytes += sstats->rx_bytes;
3485 stats->rx_errors += sstats->rx_errors;
3486 stats->rx_dropped += sstats->rx_dropped;
3488 stats->tx_packets += sstats->tx_packets;
3489 stats->tx_bytes += sstats->tx_bytes;
3490 stats->tx_errors += sstats->tx_errors;
3491 stats->tx_dropped += sstats->tx_dropped;
3493 stats->multicast += sstats->multicast;
3494 stats->collisions += sstats->collisions;
3496 stats->rx_length_errors += sstats->rx_length_errors;
3497 stats->rx_over_errors += sstats->rx_over_errors;
3498 stats->rx_crc_errors += sstats->rx_crc_errors;
3499 stats->rx_frame_errors += sstats->rx_frame_errors;
3500 stats->rx_fifo_errors += sstats->rx_fifo_errors;
3501 stats->rx_missed_errors += sstats->rx_missed_errors;
3503 stats->tx_aborted_errors += sstats->tx_aborted_errors;
3504 stats->tx_carrier_errors += sstats->tx_carrier_errors;
3505 stats->tx_fifo_errors += sstats->tx_fifo_errors;
3506 stats->tx_heartbeat_errors += sstats->tx_heartbeat_errors;
3507 stats->tx_window_errors += sstats->tx_window_errors;
3510 read_unlock_bh(&bond->lock);
3515 static int bond_do_ioctl(struct net_device *bond_dev, struct ifreq *ifr, int cmd)
3517 struct net_device *slave_dev = NULL;
3518 struct ifbond k_binfo;
3519 struct ifbond __user *u_binfo = NULL;
3520 struct ifslave k_sinfo;
3521 struct ifslave __user *u_sinfo = NULL;
3522 struct mii_ioctl_data *mii = NULL;
3525 dprintk("bond_ioctl: master=%s, cmd=%d\n",
3526 bond_dev->name, cmd);
3538 * We do this again just in case we were called by SIOCGMIIREG
3539 * instead of SIOCGMIIPHY.
3546 if (mii->reg_num == 1) {
3547 struct bonding *bond = bond_dev->priv;
3549 read_lock_bh(&bond->lock);
3550 read_lock(&bond->curr_slave_lock);
3551 if (bond->curr_active_slave) {
3552 mii->val_out = BMSR_LSTATUS;
3554 read_unlock(&bond->curr_slave_lock);
3555 read_unlock_bh(&bond->lock);
3559 case BOND_INFO_QUERY_OLD:
3560 case SIOCBONDINFOQUERY:
3561 u_binfo = (struct ifbond __user *)ifr->ifr_data;
3563 if (copy_from_user(&k_binfo, u_binfo, sizeof(ifbond))) {
3567 res = bond_info_query(bond_dev, &k_binfo);
3569 if (copy_to_user(u_binfo, &k_binfo, sizeof(ifbond))) {
3575 case BOND_SLAVE_INFO_QUERY_OLD:
3576 case SIOCBONDSLAVEINFOQUERY:
3577 u_sinfo = (struct ifslave __user *)ifr->ifr_data;
3579 if (copy_from_user(&k_sinfo, u_sinfo, sizeof(ifslave))) {
3583 res = bond_slave_info_query(bond_dev, &k_sinfo);
3585 if (copy_to_user(u_sinfo, &k_sinfo, sizeof(ifslave))) {
3596 if (!capable(CAP_NET_ADMIN)) {
3600 down_write(&(bonding_rwsem));
3601 slave_dev = dev_get_by_name(ifr->ifr_slave);
3603 dprintk("slave_dev=%p: \n", slave_dev);
3608 dprintk("slave_dev->name=%s: \n", slave_dev->name);
3610 case BOND_ENSLAVE_OLD:
3611 case SIOCBONDENSLAVE:
3612 res = bond_enslave(bond_dev, slave_dev);
3614 case BOND_RELEASE_OLD:
3615 case SIOCBONDRELEASE:
3616 res = bond_release(bond_dev, slave_dev);
3618 case BOND_SETHWADDR_OLD:
3619 case SIOCBONDSETHWADDR:
3620 res = bond_sethwaddr(bond_dev, slave_dev);
3622 case BOND_CHANGE_ACTIVE_OLD:
3623 case SIOCBONDCHANGEACTIVE:
3624 res = bond_ioctl_change_active(bond_dev, slave_dev);
3633 up_write(&(bonding_rwsem));
3637 static void bond_set_multicast_list(struct net_device *bond_dev)
3639 struct bonding *bond = bond_dev->priv;
3640 struct dev_mc_list *dmi;
3642 write_lock_bh(&bond->lock);
3645 * Do promisc before checking multicast_mode
3647 if ((bond_dev->flags & IFF_PROMISC) && !(bond->flags & IFF_PROMISC)) {
3648 bond_set_promiscuity(bond, 1);
3651 if (!(bond_dev->flags & IFF_PROMISC) && (bond->flags & IFF_PROMISC)) {
3652 bond_set_promiscuity(bond, -1);
3655 /* set allmulti flag to slaves */
3656 if ((bond_dev->flags & IFF_ALLMULTI) && !(bond->flags & IFF_ALLMULTI)) {
3657 bond_set_allmulti(bond, 1);
3660 if (!(bond_dev->flags & IFF_ALLMULTI) && (bond->flags & IFF_ALLMULTI)) {
3661 bond_set_allmulti(bond, -1);
3664 bond->flags = bond_dev->flags;
3666 /* looking for addresses to add to slaves' mc list */
3667 for (dmi = bond_dev->mc_list; dmi; dmi = dmi->next) {
3668 if (!bond_mc_list_find_dmi(dmi, bond->mc_list)) {
3669 bond_mc_add(bond, dmi->dmi_addr, dmi->dmi_addrlen);
3673 /* looking for addresses to delete from slaves' list */
3674 for (dmi = bond->mc_list; dmi; dmi = dmi->next) {
3675 if (!bond_mc_list_find_dmi(dmi, bond_dev->mc_list)) {
3676 bond_mc_delete(bond, dmi->dmi_addr, dmi->dmi_addrlen);
3680 /* save master's multicast list */
3681 bond_mc_list_destroy(bond);
3682 bond_mc_list_copy(bond_dev->mc_list, bond, GFP_ATOMIC);
3684 write_unlock_bh(&bond->lock);
3688 * Change the MTU of all of a master's slaves to match the master
3690 static int bond_change_mtu(struct net_device *bond_dev, int new_mtu)
3692 struct bonding *bond = bond_dev->priv;
3693 struct slave *slave, *stop_at;
3697 dprintk("bond=%p, name=%s, new_mtu=%d\n", bond,
3698 (bond_dev ? bond_dev->name : "None"), new_mtu);
3700 /* Can't hold bond->lock with bh disabled here since
3701 * some base drivers panic. On the other hand we can't
3702 * hold bond->lock without bh disabled because we'll
3703 * deadlock. The only solution is to rely on the fact
3704 * that we're under rtnl_lock here, and the slaves
3705 * list won't change. This doesn't solve the problem
3706 * of setting the slave's MTU while it is
3707 * transmitting, but the assumption is that the base
3708 * driver can handle that.
3710 * TODO: figure out a way to safely iterate the slaves
3711 * list, but without holding a lock around the actual
3712 * call to the base driver.
3715 bond_for_each_slave(bond, slave, i) {
3716 dprintk("s %p s->p %p c_m %p\n", slave,
3717 slave->prev, slave->dev->change_mtu);
3719 res = dev_set_mtu(slave->dev, new_mtu);
3722 /* If we failed to set the slave's mtu to the new value
3723 * we must abort the operation even in ACTIVE_BACKUP
3724 * mode, because if we allow the backup slaves to have
3725 * different mtu values than the active slave we'll
3726 * need to change their mtu when doing a failover. That
3727 * means changing their mtu from timer context, which
3728 * is probably not a good idea.
3730 dprintk("err %d %s\n", res, slave->dev->name);
3735 bond_dev->mtu = new_mtu;
3740 /* unwind from head to the slave that failed */
3742 bond_for_each_slave_from_to(bond, slave, i, bond->first_slave, stop_at) {
3745 tmp_res = dev_set_mtu(slave->dev, bond_dev->mtu);
3747 dprintk("unwind err %d dev %s\n", tmp_res,
3758 * Note that many devices must be down to change the HW address, and
3759 * downing the master releases all slaves. We can make bonds full of
3760 * bonding devices to test this, however.
3762 static int bond_set_mac_address(struct net_device *bond_dev, void *addr)
3764 struct bonding *bond = bond_dev->priv;
3765 struct sockaddr *sa = addr, tmp_sa;
3766 struct slave *slave, *stop_at;
3770 dprintk("bond=%p, name=%s\n", bond, (bond_dev ? bond_dev->name : "None"));
3772 if (!is_valid_ether_addr(sa->sa_data)) {
3773 return -EADDRNOTAVAIL;
3776 /* Can't hold bond->lock with bh disabled here since
3777 * some base drivers panic. On the other hand we can't
3778 * hold bond->lock without bh disabled because we'll
3779 * deadlock. The only solution is to rely on the fact
3780 * that we're under rtnl_lock here, and the slaves
3781 * list won't change. This doesn't solve the problem
3782 * of setting the slave's hw address while it is
3783 * transmitting, but the assumption is that the base
3784 * driver can handle that.
3786 * TODO: figure out a way to safely iterate the slaves
3787 * list, but without holding a lock around the actual
3788 * call to the base driver.
3791 bond_for_each_slave(bond, slave, i) {
3792 dprintk("slave %p %s\n", slave, slave->dev->name);
3794 if (slave->dev->set_mac_address == NULL) {
3796 dprintk("EOPNOTSUPP %s\n", slave->dev->name);
3800 res = dev_set_mac_address(slave->dev, addr);
3802 /* TODO: consider downing the slave
3804 * User should expect communications
3805 * breakage anyway until ARP finish
3808 dprintk("err %d %s\n", res, slave->dev->name);
3814 memcpy(bond_dev->dev_addr, sa->sa_data, bond_dev->addr_len);
3818 memcpy(tmp_sa.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
3819 tmp_sa.sa_family = bond_dev->type;
3821 /* unwind from head to the slave that failed */
3823 bond_for_each_slave_from_to(bond, slave, i, bond->first_slave, stop_at) {
3826 tmp_res = dev_set_mac_address(slave->dev, &tmp_sa);
3828 dprintk("unwind err %d dev %s\n", tmp_res,
3836 static int bond_xmit_roundrobin(struct sk_buff *skb, struct net_device *bond_dev)
3838 struct bonding *bond = bond_dev->priv;
3839 struct slave *slave, *start_at;
3843 read_lock(&bond->lock);
3845 if (!BOND_IS_OK(bond)) {
3849 read_lock(&bond->curr_slave_lock);
3850 slave = start_at = bond->curr_active_slave;
3851 read_unlock(&bond->curr_slave_lock);
3857 bond_for_each_slave_from(bond, slave, i, start_at) {
3858 if (IS_UP(slave->dev) &&
3859 (slave->link == BOND_LINK_UP) &&
3860 (slave->state == BOND_STATE_ACTIVE)) {
3861 res = bond_dev_queue_xmit(bond, skb, slave->dev);
3863 write_lock(&bond->curr_slave_lock);
3864 bond->curr_active_slave = slave->next;
3865 write_unlock(&bond->curr_slave_lock);
3874 /* no suitable interface, frame not sent */
3877 read_unlock(&bond->lock);
3881 static void bond_activebackup_xmit_copy(struct sk_buff *skb,
3882 struct bonding *bond,
3883 struct slave *slave)
3885 struct sk_buff *skb2 = skb_copy(skb, GFP_ATOMIC);
3886 struct ethhdr *eth_data;
3891 printk(KERN_ERR DRV_NAME ": Error: "
3892 "bond_activebackup_xmit_copy(): skb_copy() failed\n");
3896 skb2->mac.raw = (unsigned char *)skb2->data;
3897 eth_data = eth_hdr(skb2);
3899 /* Pick an appropriate source MAC address
3900 * -- use slave's perm MAC addr, unless used by bond
3901 * -- otherwise, borrow active slave's perm MAC addr
3902 * since that will not be used
3904 hwaddr = slave->perm_hwaddr;
3905 if (!memcmp(eth_data->h_source, hwaddr, ETH_ALEN))
3906 hwaddr = bond->curr_active_slave->perm_hwaddr;
3908 /* Set source MAC address appropriately */
3909 memcpy(eth_data->h_source, hwaddr, ETH_ALEN);
3911 res = bond_dev_queue_xmit(bond, skb2, slave->dev);
3913 dev_kfree_skb(skb2);
3919 * in active-backup mode, we know that bond->curr_active_slave is always valid if
3920 * the bond has a usable interface.
3922 static int bond_xmit_activebackup(struct sk_buff *skb, struct net_device *bond_dev)
3924 struct bonding *bond = bond_dev->priv;
3927 read_lock(&bond->lock);
3928 read_lock(&bond->curr_slave_lock);
3930 if (!BOND_IS_OK(bond)) {
3934 if (!bond->curr_active_slave)
3937 /* Xmit IGMP frames on all slaves to ensure rapid fail-over
3938 for multicast traffic on snooping switches */
3939 if (skb->protocol == __constant_htons(ETH_P_IP) &&
3940 skb->nh.iph->protocol == IPPROTO_IGMP) {
3941 struct slave *slave, *active_slave;
3944 active_slave = bond->curr_active_slave;
3945 bond_for_each_slave_from_to(bond, slave, i, active_slave->next,
3947 if (IS_UP(slave->dev) &&
3948 (slave->link == BOND_LINK_UP))
3949 bond_activebackup_xmit_copy(skb, bond, slave);
3952 res = bond_dev_queue_xmit(bond, skb, bond->curr_active_slave->dev);
3956 /* no suitable interface, frame not sent */
3959 read_unlock(&bond->curr_slave_lock);
3960 read_unlock(&bond->lock);
3965 * In bond_xmit_xor() , we determine the output device by using a pre-
3966 * determined xmit_hash_policy(), If the selected device is not enabled,
3967 * find the next active slave.
3969 static int bond_xmit_xor(struct sk_buff *skb, struct net_device *bond_dev)
3971 struct bonding *bond = bond_dev->priv;
3972 struct slave *slave, *start_at;
3977 read_lock(&bond->lock);
3979 if (!BOND_IS_OK(bond)) {
3983 slave_no = bond->xmit_hash_policy(skb, bond_dev, bond->slave_cnt);
3985 bond_for_each_slave(bond, slave, i) {
3994 bond_for_each_slave_from(bond, slave, i, start_at) {
3995 if (IS_UP(slave->dev) &&
3996 (slave->link == BOND_LINK_UP) &&
3997 (slave->state == BOND_STATE_ACTIVE)) {
3998 res = bond_dev_queue_xmit(bond, skb, slave->dev);
4005 /* no suitable interface, frame not sent */
4008 read_unlock(&bond->lock);
4013 * in broadcast mode, we send everything to all usable interfaces.
4015 static int bond_xmit_broadcast(struct sk_buff *skb, struct net_device *bond_dev)
4017 struct bonding *bond = bond_dev->priv;
4018 struct slave *slave, *start_at;
4019 struct net_device *tx_dev = NULL;
4023 read_lock(&bond->lock);
4025 if (!BOND_IS_OK(bond)) {
4029 read_lock(&bond->curr_slave_lock);
4030 start_at = bond->curr_active_slave;
4031 read_unlock(&bond->curr_slave_lock);
4037 bond_for_each_slave_from(bond, slave, i, start_at) {
4038 if (IS_UP(slave->dev) &&
4039 (slave->link == BOND_LINK_UP) &&
4040 (slave->state == BOND_STATE_ACTIVE)) {
4042 struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
4044 printk(KERN_ERR DRV_NAME
4045 ": %s: Error: bond_xmit_broadcast(): "
4046 "skb_clone() failed\n",
4051 res = bond_dev_queue_xmit(bond, skb2, tx_dev);
4053 dev_kfree_skb(skb2);
4057 tx_dev = slave->dev;
4062 res = bond_dev_queue_xmit(bond, skb, tx_dev);
4067 /* no suitable interface, frame not sent */
4070 /* frame sent to all suitable interfaces */
4071 read_unlock(&bond->lock);
4075 /*------------------------- Device initialization ---------------------------*/
4078 * set bond mode specific net device operations
4080 void bond_set_mode_ops(struct bonding *bond, int mode)
4082 struct net_device *bond_dev = bond->dev;
4085 case BOND_MODE_ROUNDROBIN:
4086 bond_dev->hard_start_xmit = bond_xmit_roundrobin;
4088 case BOND_MODE_ACTIVEBACKUP:
4089 bond_dev->hard_start_xmit = bond_xmit_activebackup;
4092 bond_dev->hard_start_xmit = bond_xmit_xor;
4093 if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER34)
4094 bond->xmit_hash_policy = bond_xmit_hash_policy_l34;
4096 bond->xmit_hash_policy = bond_xmit_hash_policy_l2;
4098 case BOND_MODE_BROADCAST:
4099 bond_dev->hard_start_xmit = bond_xmit_broadcast;
4101 case BOND_MODE_8023AD:
4102 bond_set_master_3ad_flags(bond);
4103 bond_dev->hard_start_xmit = bond_3ad_xmit_xor;
4104 if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER34)
4105 bond->xmit_hash_policy = bond_xmit_hash_policy_l34;
4107 bond->xmit_hash_policy = bond_xmit_hash_policy_l2;
4110 bond_set_master_alb_flags(bond);
4113 bond_dev->hard_start_xmit = bond_alb_xmit;
4114 bond_dev->set_mac_address = bond_alb_set_mac_address;
4117 /* Should never happen, mode already checked */
4118 printk(KERN_ERR DRV_NAME
4119 ": %s: Error: Unknown bonding mode %d\n",
4126 static void bond_ethtool_get_drvinfo(struct net_device *bond_dev,
4127 struct ethtool_drvinfo *drvinfo)
4129 strncpy(drvinfo->driver, DRV_NAME, 32);
4130 strncpy(drvinfo->version, DRV_VERSION, 32);
4131 snprintf(drvinfo->fw_version, 32, "%d", BOND_ABI_VERSION);
4134 static struct ethtool_ops bond_ethtool_ops = {
4135 .get_tx_csum = ethtool_op_get_tx_csum,
4136 .get_tso = ethtool_op_get_tso,
4137 .get_ufo = ethtool_op_get_ufo,
4138 .get_sg = ethtool_op_get_sg,
4139 .get_drvinfo = bond_ethtool_get_drvinfo,
4143 * Does not allocate but creates a /proc entry.
4146 static int bond_init(struct net_device *bond_dev, struct bond_params *params)
4148 struct bonding *bond = bond_dev->priv;
4150 dprintk("Begin bond_init for %s\n", bond_dev->name);
4152 /* initialize rwlocks */
4153 rwlock_init(&bond->lock);
4154 rwlock_init(&bond->curr_slave_lock);
4156 bond->params = *params; /* copy params struct */
4158 /* Initialize pointers */
4159 bond->first_slave = NULL;
4160 bond->curr_active_slave = NULL;
4161 bond->current_arp_slave = NULL;
4162 bond->primary_slave = NULL;
4163 bond->dev = bond_dev;
4164 INIT_LIST_HEAD(&bond->vlan_list);
4166 /* Initialize the device entry points */
4167 bond_dev->open = bond_open;
4168 bond_dev->stop = bond_close;
4169 bond_dev->get_stats = bond_get_stats;
4170 bond_dev->do_ioctl = bond_do_ioctl;
4171 bond_dev->ethtool_ops = &bond_ethtool_ops;
4172 bond_dev->set_multicast_list = bond_set_multicast_list;
4173 bond_dev->change_mtu = bond_change_mtu;
4174 bond_dev->set_mac_address = bond_set_mac_address;
4176 bond_set_mode_ops(bond, bond->params.mode);
4178 bond_dev->destructor = free_netdev;
4180 /* Initialize the device options */
4181 bond_dev->tx_queue_len = 0;
4182 bond_dev->flags |= IFF_MASTER|IFF_MULTICAST;
4184 /* At first, we block adding VLANs. That's the only way to
4185 * prevent problems that occur when adding VLANs over an
4186 * empty bond. The block will be removed once non-challenged
4187 * slaves are enslaved.
4189 bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
4191 /* don't acquire bond device's netif_tx_lock when
4193 bond_dev->features |= NETIF_F_LLTX;
4195 /* By default, we declare the bond to be fully
4196 * VLAN hardware accelerated capable. Special
4197 * care is taken in the various xmit functions
4198 * when there are slaves that are not hw accel
4201 bond_dev->vlan_rx_register = bond_vlan_rx_register;
4202 bond_dev->vlan_rx_add_vid = bond_vlan_rx_add_vid;
4203 bond_dev->vlan_rx_kill_vid = bond_vlan_rx_kill_vid;
4204 bond_dev->features |= (NETIF_F_HW_VLAN_TX |
4205 NETIF_F_HW_VLAN_RX |
4206 NETIF_F_HW_VLAN_FILTER);
4208 #ifdef CONFIG_PROC_FS
4209 bond_create_proc_entry(bond);
4212 list_add_tail(&bond->bond_list, &bond_dev_list);
4217 /* De-initialize device specific data.
4218 * Caller must hold rtnl_lock.
4220 void bond_deinit(struct net_device *bond_dev)
4222 struct bonding *bond = bond_dev->priv;
4224 list_del(&bond->bond_list);
4226 #ifdef CONFIG_PROC_FS
4227 bond_remove_proc_entry(bond);
4231 /* Unregister and free all bond devices.
4232 * Caller must hold rtnl_lock.
4234 static void bond_free_all(void)
4236 struct bonding *bond, *nxt;
4238 list_for_each_entry_safe(bond, nxt, &bond_dev_list, bond_list) {
4239 struct net_device *bond_dev = bond->dev;
4241 unregister_netdevice(bond_dev);
4242 bond_deinit(bond_dev);
4245 #ifdef CONFIG_PROC_FS
4246 bond_destroy_proc_dir();
4250 /*------------------------- Module initialization ---------------------------*/
4253 * Convert string input module parms. Accept either the
4254 * number of the mode or its string name.
4256 int bond_parse_parm(char *mode_arg, struct bond_parm_tbl *tbl)
4260 for (i = 0; tbl[i].modename; i++) {
4261 if ((isdigit(*mode_arg) &&
4262 tbl[i].mode == simple_strtol(mode_arg, NULL, 0)) ||
4263 (strncmp(mode_arg, tbl[i].modename,
4264 strlen(tbl[i].modename)) == 0)) {
4272 static int bond_check_params(struct bond_params *params)
4275 * Convert string parameters.
4278 bond_mode = bond_parse_parm(mode, bond_mode_tbl);
4279 if (bond_mode == -1) {
4280 printk(KERN_ERR DRV_NAME
4281 ": Error: Invalid bonding mode \"%s\"\n",
4282 mode == NULL ? "NULL" : mode);
4287 if (xmit_hash_policy) {
4288 if ((bond_mode != BOND_MODE_XOR) &&
4289 (bond_mode != BOND_MODE_8023AD)) {
4290 printk(KERN_INFO DRV_NAME
4291 ": xor_mode param is irrelevant in mode %s\n",
4292 bond_mode_name(bond_mode));
4294 xmit_hashtype = bond_parse_parm(xmit_hash_policy,
4296 if (xmit_hashtype == -1) {
4297 printk(KERN_ERR DRV_NAME
4298 ": Error: Invalid xmit_hash_policy \"%s\"\n",
4299 xmit_hash_policy == NULL ? "NULL" :
4307 if (bond_mode != BOND_MODE_8023AD) {
4308 printk(KERN_INFO DRV_NAME
4309 ": lacp_rate param is irrelevant in mode %s\n",
4310 bond_mode_name(bond_mode));
4312 lacp_fast = bond_parse_parm(lacp_rate, bond_lacp_tbl);
4313 if (lacp_fast == -1) {
4314 printk(KERN_ERR DRV_NAME
4315 ": Error: Invalid lacp rate \"%s\"\n",
4316 lacp_rate == NULL ? "NULL" : lacp_rate);
4322 if (max_bonds < 1 || max_bonds > INT_MAX) {
4323 printk(KERN_WARNING DRV_NAME
4324 ": Warning: max_bonds (%d) not in range %d-%d, so it "
4325 "was reset to BOND_DEFAULT_MAX_BONDS (%d)\n",
4326 max_bonds, 1, INT_MAX, BOND_DEFAULT_MAX_BONDS);
4327 max_bonds = BOND_DEFAULT_MAX_BONDS;
4331 printk(KERN_WARNING DRV_NAME
4332 ": Warning: miimon module parameter (%d), "
4333 "not in range 0-%d, so it was reset to %d\n",
4334 miimon, INT_MAX, BOND_LINK_MON_INTERV);
4335 miimon = BOND_LINK_MON_INTERV;
4339 printk(KERN_WARNING DRV_NAME
4340 ": Warning: updelay module parameter (%d), "
4341 "not in range 0-%d, so it was reset to 0\n",
4346 if (downdelay < 0) {
4347 printk(KERN_WARNING DRV_NAME
4348 ": Warning: downdelay module parameter (%d), "
4349 "not in range 0-%d, so it was reset to 0\n",
4350 downdelay, INT_MAX);
4354 if ((use_carrier != 0) && (use_carrier != 1)) {
4355 printk(KERN_WARNING DRV_NAME
4356 ": Warning: use_carrier module parameter (%d), "
4357 "not of valid value (0/1), so it was set to 1\n",
4362 /* reset values for 802.3ad */
4363 if (bond_mode == BOND_MODE_8023AD) {
4365 printk(KERN_WARNING DRV_NAME
4366 ": Warning: miimon must be specified, "
4367 "otherwise bonding will not detect link "
4368 "failure, speed and duplex which are "
4369 "essential for 802.3ad operation\n");
4370 printk(KERN_WARNING "Forcing miimon to 100msec\n");
4375 /* reset values for TLB/ALB */
4376 if ((bond_mode == BOND_MODE_TLB) ||
4377 (bond_mode == BOND_MODE_ALB)) {
4379 printk(KERN_WARNING DRV_NAME
4380 ": Warning: miimon must be specified, "
4381 "otherwise bonding will not detect link "
4382 "failure and link speed which are essential "
4383 "for TLB/ALB load balancing\n");
4384 printk(KERN_WARNING "Forcing miimon to 100msec\n");
4389 if (bond_mode == BOND_MODE_ALB) {
4390 printk(KERN_NOTICE DRV_NAME
4391 ": In ALB mode you might experience client "
4392 "disconnections upon reconnection of a link if the "
4393 "bonding module updelay parameter (%d msec) is "
4394 "incompatible with the forwarding delay time of the "
4400 if (updelay || downdelay) {
4401 /* just warn the user the up/down delay will have
4402 * no effect since miimon is zero...
4404 printk(KERN_WARNING DRV_NAME
4405 ": Warning: miimon module parameter not set "
4406 "and updelay (%d) or downdelay (%d) module "
4407 "parameter is set; updelay and downdelay have "
4408 "no effect unless miimon is set\n",
4409 updelay, downdelay);
4412 /* don't allow arp monitoring */
4414 printk(KERN_WARNING DRV_NAME
4415 ": Warning: miimon (%d) and arp_interval (%d) "
4416 "can't be used simultaneously, disabling ARP "
4418 miimon, arp_interval);
4422 if ((updelay % miimon) != 0) {
4423 printk(KERN_WARNING DRV_NAME
4424 ": Warning: updelay (%d) is not a multiple "
4425 "of miimon (%d), updelay rounded to %d ms\n",
4426 updelay, miimon, (updelay / miimon) * miimon);
4431 if ((downdelay % miimon) != 0) {
4432 printk(KERN_WARNING DRV_NAME
4433 ": Warning: downdelay (%d) is not a multiple "
4434 "of miimon (%d), downdelay rounded to %d ms\n",
4436 (downdelay / miimon) * miimon);
4439 downdelay /= miimon;
4442 if (arp_interval < 0) {
4443 printk(KERN_WARNING DRV_NAME
4444 ": Warning: arp_interval module parameter (%d) "
4445 ", not in range 0-%d, so it was reset to %d\n",
4446 arp_interval, INT_MAX, BOND_LINK_ARP_INTERV);
4447 arp_interval = BOND_LINK_ARP_INTERV;
4450 for (arp_ip_count = 0;
4451 (arp_ip_count < BOND_MAX_ARP_TARGETS) && arp_ip_target[arp_ip_count];
4453 /* not complete check, but should be good enough to
4455 if (!isdigit(arp_ip_target[arp_ip_count][0])) {
4456 printk(KERN_WARNING DRV_NAME
4457 ": Warning: bad arp_ip_target module parameter "
4458 "(%s), ARP monitoring will not be performed\n",
4459 arp_ip_target[arp_ip_count]);
4462 u32 ip = in_aton(arp_ip_target[arp_ip_count]);
4463 arp_target[arp_ip_count] = ip;
4467 if (arp_interval && !arp_ip_count) {
4468 /* don't allow arping if no arp_ip_target given... */
4469 printk(KERN_WARNING DRV_NAME
4470 ": Warning: arp_interval module parameter (%d) "
4471 "specified without providing an arp_ip_target "
4472 "parameter, arp_interval was reset to 0\n",
4478 printk(KERN_INFO DRV_NAME
4479 ": MII link monitoring set to %d ms\n",
4481 } else if (arp_interval) {
4484 printk(KERN_INFO DRV_NAME
4485 ": ARP monitoring set to %d ms with %d target(s):",
4486 arp_interval, arp_ip_count);
4488 for (i = 0; i < arp_ip_count; i++)
4489 printk (" %s", arp_ip_target[i]);
4494 /* miimon and arp_interval not set, we need one so things
4495 * work as expected, see bonding.txt for details
4497 printk(KERN_WARNING DRV_NAME
4498 ": Warning: either miimon or arp_interval and "
4499 "arp_ip_target module parameters must be specified, "
4500 "otherwise bonding will not detect link failures! see "
4501 "bonding.txt for details.\n");
4504 if (primary && !USES_PRIMARY(bond_mode)) {
4505 /* currently, using a primary only makes sense
4506 * in active backup, TLB or ALB modes
4508 printk(KERN_WARNING DRV_NAME
4509 ": Warning: %s primary device specified but has no "
4510 "effect in %s mode\n",
4511 primary, bond_mode_name(bond_mode));
4515 /* fill params struct with the proper values */
4516 params->mode = bond_mode;
4517 params->xmit_policy = xmit_hashtype;
4518 params->miimon = miimon;
4519 params->arp_interval = arp_interval;
4520 params->updelay = updelay;
4521 params->downdelay = downdelay;
4522 params->use_carrier = use_carrier;
4523 params->lacp_fast = lacp_fast;
4524 params->primary[0] = 0;
4527 strncpy(params->primary, primary, IFNAMSIZ);
4528 params->primary[IFNAMSIZ - 1] = 0;
4531 memcpy(params->arp_targets, arp_target, sizeof(arp_target));
4536 /* Create a new bond based on the specified name and bonding parameters.
4537 * Caller must NOT hold rtnl_lock; we need to release it here before we
4538 * set up our sysfs entries.
4540 int bond_create(char *name, struct bond_params *params, struct bonding **newbond)
4542 struct net_device *bond_dev;
4546 bond_dev = alloc_netdev(sizeof(struct bonding), name, ether_setup);
4548 printk(KERN_ERR DRV_NAME
4549 ": %s: eek! can't alloc netdev!\n",
4555 /* bond_init() must be called after dev_alloc_name() (for the
4556 * /proc files), but before register_netdevice(), because we
4557 * need to set function pointers.
4560 res = bond_init(bond_dev, params);
4565 SET_MODULE_OWNER(bond_dev);
4567 res = register_netdevice(bond_dev);
4572 *newbond = bond_dev->priv;
4574 netif_carrier_off(bond_dev);
4576 rtnl_unlock(); /* allows sysfs registration of net device */
4577 res = bond_create_sysfs_entry(bond_dev->priv);
4580 bond_deinit(bond_dev);
4582 free_netdev(bond_dev);
4589 static int __init bonding_init(void)
4593 char new_bond_name[8]; /* Enough room for 999 bonds at init. */
4595 printk(KERN_INFO "%s", version);
4597 res = bond_check_params(&bonding_defaults);
4602 #ifdef CONFIG_PROC_FS
4603 bond_create_proc_dir();
4605 for (i = 0; i < max_bonds; i++) {
4606 sprintf(new_bond_name, "bond%d",i);
4607 res = bond_create(new_bond_name,&bonding_defaults, NULL);
4612 res = bond_create_sysfs();
4616 register_netdevice_notifier(&bond_netdev_notifier);
4617 register_inetaddr_notifier(&bond_inetaddr_notifier);
4623 bond_destroy_sysfs();
4630 static void __exit bonding_exit(void)
4632 unregister_netdevice_notifier(&bond_netdev_notifier);
4633 unregister_inetaddr_notifier(&bond_inetaddr_notifier);
4637 bond_destroy_sysfs();
4641 module_init(bonding_init);
4642 module_exit(bonding_exit);
4643 MODULE_LICENSE("GPL");
4644 MODULE_VERSION(DRV_VERSION);
4645 MODULE_DESCRIPTION(DRV_DESCRIPTION ", v" DRV_VERSION);
4646 MODULE_AUTHOR("Thomas Davis, tadavis@lbl.gov and many others");
4647 MODULE_SUPPORTED_DEVICE("most ethernet devices");