2 * originally based on the dummy device.
4 * Copyright 1999, Thomas Davis, tadavis@lbl.gov.
5 * Licensed under the GPL. Based on dummy.c, and eql.c devices.
7 * bonding.c: an Ethernet Bonding driver
9 * This is useful to talk to a Cisco EtherChannel compatible equipment:
11 * Sun Trunking (Solaris)
12 * Alteon AceDirector Trunks
14 * and probably many L2 switches ...
17 * ifconfig bond0 ipaddress netmask up
18 * will setup a network device, with an ip address. No mac address
19 * will be assigned at this time. The hw mac address will come from
20 * the first slave bonded to the channel. All slaves will then use
21 * this hw mac address.
24 * will release all slaves, marking them as down.
26 * ifenslave bond0 eth0
27 * will attach eth0 to bond0 as a slave. eth0 hw mac address will either
28 * a: be used as initial mac address
29 * b: if a hw mac address already is there, eth0's hw mac address
30 * will then be set from bond0.
34 //#define BONDING_DEBUG 1
36 #include <linux/kernel.h>
37 #include <linux/module.h>
38 #include <linux/types.h>
39 #include <linux/fcntl.h>
40 #include <linux/interrupt.h>
41 #include <linux/ptrace.h>
42 #include <linux/ioport.h>
46 #include <linux/tcp.h>
47 #include <linux/udp.h>
48 #include <linux/slab.h>
49 #include <linux/string.h>
50 #include <linux/init.h>
51 #include <linux/timer.h>
52 #include <linux/socket.h>
53 #include <linux/ctype.h>
54 #include <linux/inet.h>
55 #include <linux/bitops.h>
56 #include <asm/system.h>
59 #include <asm/uaccess.h>
60 #include <linux/errno.h>
61 #include <linux/netdevice.h>
62 #include <linux/inetdevice.h>
63 #include <linux/igmp.h>
64 #include <linux/etherdevice.h>
65 #include <linux/skbuff.h>
67 #include <linux/rtnetlink.h>
68 #include <linux/proc_fs.h>
69 #include <linux/seq_file.h>
70 #include <linux/smp.h>
71 #include <linux/if_ether.h>
73 #include <linux/mii.h>
74 #include <linux/ethtool.h>
75 #include <linux/if_vlan.h>
76 #include <linux/if_bonding.h>
77 #include <net/route.h>
78 #include <net/net_namespace.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 static char *arp_validate = NULL;
101 struct bond_params bonding_defaults;
103 module_param(max_bonds, int, 0);
104 MODULE_PARM_DESC(max_bonds, "Max number of bonded devices");
105 module_param(miimon, int, 0);
106 MODULE_PARM_DESC(miimon, "Link check interval in milliseconds");
107 module_param(updelay, int, 0);
108 MODULE_PARM_DESC(updelay, "Delay before considering link up, in milliseconds");
109 module_param(downdelay, int, 0);
110 MODULE_PARM_DESC(downdelay, "Delay before considering link down, "
112 module_param(use_carrier, int, 0);
113 MODULE_PARM_DESC(use_carrier, "Use netif_carrier_ok (vs MII ioctls) in miimon; "
114 "0 for off, 1 for on (default)");
115 module_param(mode, charp, 0);
116 MODULE_PARM_DESC(mode, "Mode of operation : 0 for balance-rr, "
117 "1 for active-backup, 2 for balance-xor, "
118 "3 for broadcast, 4 for 802.3ad, 5 for balance-tlb, "
119 "6 for balance-alb");
120 module_param(primary, charp, 0);
121 MODULE_PARM_DESC(primary, "Primary network device to use");
122 module_param(lacp_rate, charp, 0);
123 MODULE_PARM_DESC(lacp_rate, "LACPDU tx rate to request from 802.3ad partner "
125 module_param(xmit_hash_policy, charp, 0);
126 MODULE_PARM_DESC(xmit_hash_policy, "XOR hashing method: 0 for layer 2 (default)"
127 ", 1 for layer 3+4");
128 module_param(arp_interval, int, 0);
129 MODULE_PARM_DESC(arp_interval, "arp interval in milliseconds");
130 module_param_array(arp_ip_target, charp, NULL, 0);
131 MODULE_PARM_DESC(arp_ip_target, "arp targets in n.n.n.n form");
132 module_param(arp_validate, charp, 0);
133 MODULE_PARM_DESC(arp_validate, "validate src/dst of ARP probes: none (default), active, backup or all");
135 /*----------------------------- Global variables ----------------------------*/
137 static const char * const version =
138 DRV_DESCRIPTION ": v" DRV_VERSION " (" DRV_RELDATE ")\n";
140 LIST_HEAD(bond_dev_list);
142 #ifdef CONFIG_PROC_FS
143 static struct proc_dir_entry *bond_proc_dir = NULL;
146 extern struct rw_semaphore bonding_rwsem;
147 static u32 arp_target[BOND_MAX_ARP_TARGETS] = { 0, } ;
148 static int arp_ip_count = 0;
149 static int bond_mode = BOND_MODE_ROUNDROBIN;
150 static int xmit_hashtype= BOND_XMIT_POLICY_LAYER2;
151 static int lacp_fast = 0;
154 struct bond_parm_tbl bond_lacp_tbl[] = {
155 { "slow", AD_LACP_SLOW},
156 { "fast", AD_LACP_FAST},
160 struct bond_parm_tbl bond_mode_tbl[] = {
161 { "balance-rr", BOND_MODE_ROUNDROBIN},
162 { "active-backup", BOND_MODE_ACTIVEBACKUP},
163 { "balance-xor", BOND_MODE_XOR},
164 { "broadcast", BOND_MODE_BROADCAST},
165 { "802.3ad", BOND_MODE_8023AD},
166 { "balance-tlb", BOND_MODE_TLB},
167 { "balance-alb", BOND_MODE_ALB},
171 struct bond_parm_tbl xmit_hashtype_tbl[] = {
172 { "layer2", BOND_XMIT_POLICY_LAYER2},
173 { "layer3+4", BOND_XMIT_POLICY_LAYER34},
177 struct bond_parm_tbl arp_validate_tbl[] = {
178 { "none", BOND_ARP_VALIDATE_NONE},
179 { "active", BOND_ARP_VALIDATE_ACTIVE},
180 { "backup", BOND_ARP_VALIDATE_BACKUP},
181 { "all", BOND_ARP_VALIDATE_ALL},
185 /*-------------------------- Forward declarations ---------------------------*/
187 static void bond_send_gratuitous_arp(struct bonding *bond);
189 /*---------------------------- General routines -----------------------------*/
191 static const char *bond_mode_name(int mode)
194 case BOND_MODE_ROUNDROBIN :
195 return "load balancing (round-robin)";
196 case BOND_MODE_ACTIVEBACKUP :
197 return "fault-tolerance (active-backup)";
199 return "load balancing (xor)";
200 case BOND_MODE_BROADCAST :
201 return "fault-tolerance (broadcast)";
202 case BOND_MODE_8023AD:
203 return "IEEE 802.3ad Dynamic link aggregation";
205 return "transmit load balancing";
207 return "adaptive load balancing";
213 /*---------------------------------- VLAN -----------------------------------*/
216 * bond_add_vlan - add a new vlan id on bond
217 * @bond: bond that got the notification
218 * @vlan_id: the vlan id to add
220 * Returns -ENOMEM if allocation failed.
222 static int bond_add_vlan(struct bonding *bond, unsigned short vlan_id)
224 struct vlan_entry *vlan;
226 dprintk("bond: %s, vlan id %d\n",
227 (bond ? bond->dev->name: "None"), vlan_id);
229 vlan = kmalloc(sizeof(struct vlan_entry), GFP_KERNEL);
234 INIT_LIST_HEAD(&vlan->vlan_list);
235 vlan->vlan_id = vlan_id;
238 write_lock_bh(&bond->lock);
240 list_add_tail(&vlan->vlan_list, &bond->vlan_list);
242 write_unlock_bh(&bond->lock);
244 dprintk("added VLAN ID %d on bond %s\n", vlan_id, bond->dev->name);
250 * bond_del_vlan - delete a vlan id from bond
251 * @bond: bond that got the notification
252 * @vlan_id: the vlan id to delete
254 * returns -ENODEV if @vlan_id was not found in @bond.
256 static int bond_del_vlan(struct bonding *bond, unsigned short vlan_id)
258 struct vlan_entry *vlan, *next;
261 dprintk("bond: %s, vlan id %d\n", bond->dev->name, vlan_id);
263 write_lock_bh(&bond->lock);
265 list_for_each_entry_safe(vlan, next, &bond->vlan_list, vlan_list) {
266 if (vlan->vlan_id == vlan_id) {
267 list_del(&vlan->vlan_list);
269 if ((bond->params.mode == BOND_MODE_TLB) ||
270 (bond->params.mode == BOND_MODE_ALB)) {
271 bond_alb_clear_vlan(bond, vlan_id);
274 dprintk("removed VLAN ID %d from bond %s\n", vlan_id,
279 if (list_empty(&bond->vlan_list) &&
280 (bond->slave_cnt == 0)) {
281 /* Last VLAN removed and no slaves, so
282 * restore block on adding VLANs. This will
283 * be removed once new slaves that are not
284 * VLAN challenged will be added.
286 bond->dev->features |= NETIF_F_VLAN_CHALLENGED;
294 dprintk("couldn't find VLAN ID %d in bond %s\n", vlan_id,
298 write_unlock_bh(&bond->lock);
303 * bond_has_challenged_slaves
304 * @bond: the bond we're working on
306 * Searches the slave list. Returns 1 if a vlan challenged slave
307 * was found, 0 otherwise.
309 * Assumes bond->lock is held.
311 static int bond_has_challenged_slaves(struct bonding *bond)
316 bond_for_each_slave(bond, slave, i) {
317 if (slave->dev->features & NETIF_F_VLAN_CHALLENGED) {
318 dprintk("found VLAN challenged slave - %s\n",
324 dprintk("no VLAN challenged slaves found\n");
329 * bond_next_vlan - safely skip to the next item in the vlans list.
330 * @bond: the bond we're working on
331 * @curr: item we're advancing from
333 * Returns %NULL if list is empty, bond->next_vlan if @curr is %NULL,
334 * or @curr->next otherwise (even if it is @curr itself again).
336 * Caller must hold bond->lock
338 struct vlan_entry *bond_next_vlan(struct bonding *bond, struct vlan_entry *curr)
340 struct vlan_entry *next, *last;
342 if (list_empty(&bond->vlan_list)) {
347 next = list_entry(bond->vlan_list.next,
348 struct vlan_entry, vlan_list);
350 last = list_entry(bond->vlan_list.prev,
351 struct vlan_entry, vlan_list);
353 next = list_entry(bond->vlan_list.next,
354 struct vlan_entry, vlan_list);
356 next = list_entry(curr->vlan_list.next,
357 struct vlan_entry, vlan_list);
365 * bond_dev_queue_xmit - Prepare skb for xmit.
367 * @bond: bond device that got this skb for tx.
368 * @skb: hw accel VLAN tagged skb to transmit
369 * @slave_dev: slave that is supposed to xmit this skbuff
371 * When the bond gets an skb to transmit that is
372 * already hardware accelerated VLAN tagged, and it
373 * needs to relay this skb to a slave that is not
374 * hw accel capable, the skb needs to be "unaccelerated",
375 * i.e. strip the hwaccel tag and re-insert it as part
378 int bond_dev_queue_xmit(struct bonding *bond, struct sk_buff *skb, struct net_device *slave_dev)
380 unsigned short vlan_id;
382 if (!list_empty(&bond->vlan_list) &&
383 !(slave_dev->features & NETIF_F_HW_VLAN_TX) &&
384 vlan_get_tag(skb, &vlan_id) == 0) {
385 skb->dev = slave_dev;
386 skb = vlan_put_tag(skb, vlan_id);
388 /* vlan_put_tag() frees the skb in case of error,
389 * so return success here so the calling functions
390 * won't attempt to free is again.
395 skb->dev = slave_dev;
405 * In the following 3 functions, bond_vlan_rx_register(), bond_vlan_rx_add_vid
406 * and bond_vlan_rx_kill_vid, We don't protect the slave list iteration with a
408 * a. This operation is performed in IOCTL context,
409 * b. The operation is protected by the RTNL semaphore in the 8021q code,
410 * c. Holding a lock with BH disabled while directly calling a base driver
411 * entry point is generally a BAD idea.
413 * The design of synchronization/protection for this operation in the 8021q
414 * module is good for one or more VLAN devices over a single physical device
415 * and cannot be extended for a teaming solution like bonding, so there is a
416 * potential race condition here where a net device from the vlan group might
417 * be referenced (either by a base driver or the 8021q code) while it is being
418 * removed from the system. However, it turns out we're not making matters
419 * worse, and if it works for regular VLAN usage it will work here too.
423 * bond_vlan_rx_register - Propagates registration to slaves
424 * @bond_dev: bonding net device that got called
425 * @grp: vlan group being registered
427 static void bond_vlan_rx_register(struct net_device *bond_dev, struct vlan_group *grp)
429 struct bonding *bond = bond_dev->priv;
435 bond_for_each_slave(bond, slave, i) {
436 struct net_device *slave_dev = slave->dev;
438 if ((slave_dev->features & NETIF_F_HW_VLAN_RX) &&
439 slave_dev->vlan_rx_register) {
440 slave_dev->vlan_rx_register(slave_dev, grp);
446 * bond_vlan_rx_add_vid - Propagates adding an id to slaves
447 * @bond_dev: bonding net device that got called
448 * @vid: vlan id being added
450 static void bond_vlan_rx_add_vid(struct net_device *bond_dev, uint16_t vid)
452 struct bonding *bond = bond_dev->priv;
456 bond_for_each_slave(bond, slave, i) {
457 struct net_device *slave_dev = slave->dev;
459 if ((slave_dev->features & NETIF_F_HW_VLAN_FILTER) &&
460 slave_dev->vlan_rx_add_vid) {
461 slave_dev->vlan_rx_add_vid(slave_dev, vid);
465 res = bond_add_vlan(bond, vid);
467 printk(KERN_ERR DRV_NAME
468 ": %s: Error: Failed to add vlan id %d\n",
469 bond_dev->name, vid);
474 * bond_vlan_rx_kill_vid - Propagates deleting an id to slaves
475 * @bond_dev: bonding net device that got called
476 * @vid: vlan id being removed
478 static void bond_vlan_rx_kill_vid(struct net_device *bond_dev, uint16_t vid)
480 struct bonding *bond = bond_dev->priv;
482 struct net_device *vlan_dev;
485 bond_for_each_slave(bond, slave, i) {
486 struct net_device *slave_dev = slave->dev;
488 if ((slave_dev->features & NETIF_F_HW_VLAN_FILTER) &&
489 slave_dev->vlan_rx_kill_vid) {
490 /* Save and then restore vlan_dev in the grp array,
491 * since the slave's driver might clear it.
493 vlan_dev = vlan_group_get_device(bond->vlgrp, vid);
494 slave_dev->vlan_rx_kill_vid(slave_dev, vid);
495 vlan_group_set_device(bond->vlgrp, vid, vlan_dev);
499 res = bond_del_vlan(bond, vid);
501 printk(KERN_ERR DRV_NAME
502 ": %s: Error: Failed to remove vlan id %d\n",
503 bond_dev->name, vid);
507 static void bond_add_vlans_on_slave(struct bonding *bond, struct net_device *slave_dev)
509 struct vlan_entry *vlan;
511 write_lock_bh(&bond->lock);
513 if (list_empty(&bond->vlan_list)) {
517 if ((slave_dev->features & NETIF_F_HW_VLAN_RX) &&
518 slave_dev->vlan_rx_register) {
519 slave_dev->vlan_rx_register(slave_dev, bond->vlgrp);
522 if (!(slave_dev->features & NETIF_F_HW_VLAN_FILTER) ||
523 !(slave_dev->vlan_rx_add_vid)) {
527 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
528 slave_dev->vlan_rx_add_vid(slave_dev, vlan->vlan_id);
532 write_unlock_bh(&bond->lock);
535 static void bond_del_vlans_from_slave(struct bonding *bond, struct net_device *slave_dev)
537 struct vlan_entry *vlan;
538 struct net_device *vlan_dev;
540 write_lock_bh(&bond->lock);
542 if (list_empty(&bond->vlan_list)) {
546 if (!(slave_dev->features & NETIF_F_HW_VLAN_FILTER) ||
547 !(slave_dev->vlan_rx_kill_vid)) {
551 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
552 /* Save and then restore vlan_dev in the grp array,
553 * since the slave's driver might clear it.
555 vlan_dev = vlan_group_get_device(bond->vlgrp, vlan->vlan_id);
556 slave_dev->vlan_rx_kill_vid(slave_dev, vlan->vlan_id);
557 vlan_group_set_device(bond->vlgrp, vlan->vlan_id, vlan_dev);
561 if ((slave_dev->features & NETIF_F_HW_VLAN_RX) &&
562 slave_dev->vlan_rx_register) {
563 slave_dev->vlan_rx_register(slave_dev, NULL);
567 write_unlock_bh(&bond->lock);
570 /*------------------------------- Link status -------------------------------*/
573 * Set the carrier state for the master according to the state of its
574 * slaves. If any slaves are up, the master is up. In 802.3ad mode,
575 * do special 802.3ad magic.
577 * Returns zero if carrier state does not change, nonzero if it does.
579 static int bond_set_carrier(struct bonding *bond)
584 if (bond->slave_cnt == 0)
587 if (bond->params.mode == BOND_MODE_8023AD)
588 return bond_3ad_set_carrier(bond);
590 bond_for_each_slave(bond, slave, i) {
591 if (slave->link == BOND_LINK_UP) {
592 if (!netif_carrier_ok(bond->dev)) {
593 netif_carrier_on(bond->dev);
601 if (netif_carrier_ok(bond->dev)) {
602 netif_carrier_off(bond->dev);
609 * Get link speed and duplex from the slave's base driver
610 * using ethtool. If for some reason the call fails or the
611 * values are invalid, fake speed and duplex to 100/Full
614 static int bond_update_speed_duplex(struct slave *slave)
616 struct net_device *slave_dev = slave->dev;
617 struct ethtool_cmd etool;
620 /* Fake speed and duplex */
621 slave->speed = SPEED_100;
622 slave->duplex = DUPLEX_FULL;
624 if (!slave_dev->ethtool_ops || !slave_dev->ethtool_ops->get_settings)
627 res = slave_dev->ethtool_ops->get_settings(slave_dev, &etool);
631 switch (etool.speed) {
641 switch (etool.duplex) {
649 slave->speed = etool.speed;
650 slave->duplex = etool.duplex;
656 * if <dev> supports MII link status reporting, check its link status.
658 * We either do MII/ETHTOOL ioctls, or check netif_carrier_ok(),
659 * depening upon the setting of the use_carrier parameter.
661 * Return either BMSR_LSTATUS, meaning that the link is up (or we
662 * can't tell and just pretend it is), or 0, meaning that the link is
665 * If reporting is non-zero, instead of faking link up, return -1 if
666 * both ETHTOOL and MII ioctls fail (meaning the device does not
667 * support them). If use_carrier is set, return whatever it says.
668 * It'd be nice if there was a good way to tell if a driver supports
669 * netif_carrier, but there really isn't.
671 static int bond_check_dev_link(struct bonding *bond, struct net_device *slave_dev, int reporting)
673 static int (* ioctl)(struct net_device *, struct ifreq *, int);
675 struct mii_ioctl_data *mii;
677 if (bond->params.use_carrier) {
678 return netif_carrier_ok(slave_dev) ? BMSR_LSTATUS : 0;
681 ioctl = slave_dev->do_ioctl;
683 /* TODO: set pointer to correct ioctl on a per team member */
684 /* bases to make this more efficient. that is, once */
685 /* we determine the correct ioctl, we will always */
686 /* call it and not the others for that team */
690 * We cannot assume that SIOCGMIIPHY will also read a
691 * register; not all network drivers (e.g., e100)
695 /* Yes, the mii is overlaid on the ifreq.ifr_ifru */
696 strncpy(ifr.ifr_name, slave_dev->name, IFNAMSIZ);
698 if (IOCTL(slave_dev, &ifr, SIOCGMIIPHY) == 0) {
699 mii->reg_num = MII_BMSR;
700 if (IOCTL(slave_dev, &ifr, SIOCGMIIREG) == 0) {
701 return (mii->val_out & BMSR_LSTATUS);
707 * Some drivers cache ETHTOOL_GLINK for a period of time so we only
708 * attempt to get link status from it if the above MII ioctls fail.
710 if (slave_dev->ethtool_ops) {
711 if (slave_dev->ethtool_ops->get_link) {
714 link = slave_dev->ethtool_ops->get_link(slave_dev);
716 return link ? BMSR_LSTATUS : 0;
721 * If reporting, report that either there's no dev->do_ioctl,
722 * or both SIOCGMIIREG and get_link failed (meaning that we
723 * cannot report link status). If not reporting, pretend
726 return (reporting ? -1 : BMSR_LSTATUS);
729 /*----------------------------- Multicast list ------------------------------*/
732 * Returns 0 if dmi1 and dmi2 are the same, non-0 otherwise
734 static inline int bond_is_dmi_same(struct dev_mc_list *dmi1, struct dev_mc_list *dmi2)
736 return memcmp(dmi1->dmi_addr, dmi2->dmi_addr, dmi1->dmi_addrlen) == 0 &&
737 dmi1->dmi_addrlen == dmi2->dmi_addrlen;
741 * returns dmi entry if found, NULL otherwise
743 static struct dev_mc_list *bond_mc_list_find_dmi(struct dev_mc_list *dmi, struct dev_mc_list *mc_list)
745 struct dev_mc_list *idmi;
747 for (idmi = mc_list; idmi; idmi = idmi->next) {
748 if (bond_is_dmi_same(dmi, idmi)) {
757 * Push the promiscuity flag down to appropriate slaves
759 static void bond_set_promiscuity(struct bonding *bond, int inc)
761 if (USES_PRIMARY(bond->params.mode)) {
762 /* write lock already acquired */
763 if (bond->curr_active_slave) {
764 dev_set_promiscuity(bond->curr_active_slave->dev, inc);
769 bond_for_each_slave(bond, slave, i) {
770 dev_set_promiscuity(slave->dev, inc);
776 * Push the allmulti flag down to all slaves
778 static void bond_set_allmulti(struct bonding *bond, int inc)
780 if (USES_PRIMARY(bond->params.mode)) {
781 /* write lock already acquired */
782 if (bond->curr_active_slave) {
783 dev_set_allmulti(bond->curr_active_slave->dev, inc);
788 bond_for_each_slave(bond, slave, i) {
789 dev_set_allmulti(slave->dev, inc);
795 * Add a Multicast address to slaves
798 static void bond_mc_add(struct bonding *bond, void *addr, int alen)
800 if (USES_PRIMARY(bond->params.mode)) {
801 /* write lock already acquired */
802 if (bond->curr_active_slave) {
803 dev_mc_add(bond->curr_active_slave->dev, addr, alen, 0);
808 bond_for_each_slave(bond, slave, i) {
809 dev_mc_add(slave->dev, addr, alen, 0);
815 * Remove a multicast address from slave
818 static void bond_mc_delete(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_delete(bond->curr_active_slave->dev, addr, alen, 0);
828 bond_for_each_slave(bond, slave, i) {
829 dev_mc_delete(slave->dev, addr, alen, 0);
836 * Retrieve the list of registered multicast addresses for the bonding
837 * device and retransmit an IGMP JOIN request to the current active
840 static void bond_resend_igmp_join_requests(struct bonding *bond)
842 struct in_device *in_dev;
843 struct ip_mc_list *im;
846 in_dev = __in_dev_get_rcu(bond->dev);
848 for (im = in_dev->mc_list; im; im = im->next) {
849 ip_mc_rejoin_group(im);
857 * Totally destroys the mc_list in bond
859 static void bond_mc_list_destroy(struct bonding *bond)
861 struct dev_mc_list *dmi;
865 bond->mc_list = dmi->next;
869 bond->mc_list = NULL;
873 * Copy all the Multicast addresses from src to the bonding device dst
875 static int bond_mc_list_copy(struct dev_mc_list *mc_list, struct bonding *bond,
878 struct dev_mc_list *dmi, *new_dmi;
880 for (dmi = mc_list; dmi; dmi = dmi->next) {
881 new_dmi = kmalloc(sizeof(struct dev_mc_list), gfp_flag);
884 /* FIXME: Potential memory leak !!! */
888 new_dmi->next = bond->mc_list;
889 bond->mc_list = new_dmi;
890 new_dmi->dmi_addrlen = dmi->dmi_addrlen;
891 memcpy(new_dmi->dmi_addr, dmi->dmi_addr, dmi->dmi_addrlen);
892 new_dmi->dmi_users = dmi->dmi_users;
893 new_dmi->dmi_gusers = dmi->dmi_gusers;
900 * flush all members of flush->mc_list from device dev->mc_list
902 static void bond_mc_list_flush(struct net_device *bond_dev, struct net_device *slave_dev)
904 struct bonding *bond = bond_dev->priv;
905 struct dev_mc_list *dmi;
907 for (dmi = bond_dev->mc_list; dmi; dmi = dmi->next) {
908 dev_mc_delete(slave_dev, dmi->dmi_addr, dmi->dmi_addrlen, 0);
911 if (bond->params.mode == BOND_MODE_8023AD) {
912 /* del lacpdu mc addr from mc list */
913 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
915 dev_mc_delete(slave_dev, lacpdu_multicast, ETH_ALEN, 0);
919 /*--------------------------- Active slave change ---------------------------*/
922 * Update the mc list and multicast-related flags for the new and
923 * old active slaves (if any) according to the multicast mode, and
924 * promiscuous flags unconditionally.
926 static void bond_mc_swap(struct bonding *bond, struct slave *new_active, struct slave *old_active)
928 struct dev_mc_list *dmi;
930 if (!USES_PRIMARY(bond->params.mode)) {
931 /* nothing to do - mc list is already up-to-date on
938 if (bond->dev->flags & IFF_PROMISC) {
939 dev_set_promiscuity(old_active->dev, -1);
942 if (bond->dev->flags & IFF_ALLMULTI) {
943 dev_set_allmulti(old_active->dev, -1);
946 for (dmi = bond->dev->mc_list; dmi; dmi = dmi->next) {
947 dev_mc_delete(old_active->dev, dmi->dmi_addr, dmi->dmi_addrlen, 0);
952 if (bond->dev->flags & IFF_PROMISC) {
953 dev_set_promiscuity(new_active->dev, 1);
956 if (bond->dev->flags & IFF_ALLMULTI) {
957 dev_set_allmulti(new_active->dev, 1);
960 for (dmi = bond->dev->mc_list; dmi; dmi = dmi->next) {
961 dev_mc_add(new_active->dev, dmi->dmi_addr, dmi->dmi_addrlen, 0);
963 bond_resend_igmp_join_requests(bond);
968 * find_best_interface - select the best available slave to be the active one
969 * @bond: our bonding struct
971 * Warning: Caller must hold curr_slave_lock for writing.
973 static struct slave *bond_find_best_slave(struct bonding *bond)
975 struct slave *new_active, *old_active;
976 struct slave *bestslave = NULL;
977 int mintime = bond->params.updelay;
980 new_active = old_active = bond->curr_active_slave;
982 if (!new_active) { /* there were no active slaves left */
983 if (bond->slave_cnt > 0) { /* found one slave */
984 new_active = bond->first_slave;
986 return NULL; /* still no slave, return NULL */
990 /* first try the primary link; if arping, a link must tx/rx traffic
991 * before it can be considered the curr_active_slave - also, we would skip
992 * slaves between the curr_active_slave and primary_slave that may be up
995 if ((bond->primary_slave) &&
996 (!bond->params.arp_interval) &&
997 (IS_UP(bond->primary_slave->dev))) {
998 new_active = bond->primary_slave;
1001 /* remember where to stop iterating over the slaves */
1002 old_active = new_active;
1004 bond_for_each_slave_from(bond, new_active, i, old_active) {
1005 if (IS_UP(new_active->dev)) {
1006 if (new_active->link == BOND_LINK_UP) {
1008 } else if (new_active->link == BOND_LINK_BACK) {
1009 /* link up, but waiting for stabilization */
1010 if (new_active->delay < mintime) {
1011 mintime = new_active->delay;
1012 bestslave = new_active;
1022 * change_active_interface - change the active slave into the specified one
1023 * @bond: our bonding struct
1024 * @new: the new slave to make the active one
1026 * Set the new slave to the bond's settings and unset them on the old
1027 * curr_active_slave.
1028 * Setting include flags, mc-list, promiscuity, allmulti, etc.
1030 * If @new's link state is %BOND_LINK_BACK we'll set it to %BOND_LINK_UP,
1031 * because it is apparently the best available slave we have, even though its
1032 * updelay hasn't timed out yet.
1034 * Warning: Caller must hold curr_slave_lock for writing.
1036 void bond_change_active_slave(struct bonding *bond, struct slave *new_active)
1038 struct slave *old_active = bond->curr_active_slave;
1040 if (old_active == new_active) {
1045 if (new_active->link == BOND_LINK_BACK) {
1046 if (USES_PRIMARY(bond->params.mode)) {
1047 printk(KERN_INFO DRV_NAME
1048 ": %s: making interface %s the new "
1049 "active one %d ms earlier.\n",
1050 bond->dev->name, new_active->dev->name,
1051 (bond->params.updelay - new_active->delay) * bond->params.miimon);
1054 new_active->delay = 0;
1055 new_active->link = BOND_LINK_UP;
1056 new_active->jiffies = jiffies;
1058 if (bond->params.mode == BOND_MODE_8023AD) {
1059 bond_3ad_handle_link_change(new_active, BOND_LINK_UP);
1062 if ((bond->params.mode == BOND_MODE_TLB) ||
1063 (bond->params.mode == BOND_MODE_ALB)) {
1064 bond_alb_handle_link_change(bond, new_active, BOND_LINK_UP);
1067 if (USES_PRIMARY(bond->params.mode)) {
1068 printk(KERN_INFO DRV_NAME
1069 ": %s: making interface %s the new "
1071 bond->dev->name, new_active->dev->name);
1076 if (USES_PRIMARY(bond->params.mode)) {
1077 bond_mc_swap(bond, new_active, old_active);
1080 if ((bond->params.mode == BOND_MODE_TLB) ||
1081 (bond->params.mode == BOND_MODE_ALB)) {
1082 bond_alb_handle_active_change(bond, new_active);
1084 bond_set_slave_inactive_flags(old_active);
1086 bond_set_slave_active_flags(new_active);
1088 bond->curr_active_slave = new_active;
1091 if (bond->params.mode == BOND_MODE_ACTIVEBACKUP) {
1093 bond_set_slave_inactive_flags(old_active);
1097 bond_set_slave_active_flags(new_active);
1099 bond_send_gratuitous_arp(bond);
1104 * bond_select_active_slave - select a new active slave, if needed
1105 * @bond: our bonding struct
1107 * This functions shoud be called when one of the following occurs:
1108 * - The old curr_active_slave has been released or lost its link.
1109 * - The primary_slave has got its link back.
1110 * - A slave has got its link back and there's no old curr_active_slave.
1112 * Warning: Caller must hold curr_slave_lock for writing.
1114 void bond_select_active_slave(struct bonding *bond)
1116 struct slave *best_slave;
1119 best_slave = bond_find_best_slave(bond);
1120 if (best_slave != bond->curr_active_slave) {
1121 bond_change_active_slave(bond, best_slave);
1122 rv = bond_set_carrier(bond);
1126 if (netif_carrier_ok(bond->dev)) {
1127 printk(KERN_INFO DRV_NAME
1128 ": %s: first active interface up!\n",
1131 printk(KERN_INFO DRV_NAME ": %s: "
1132 "now running without any active interface !\n",
1138 /*--------------------------- slave list handling ---------------------------*/
1141 * This function attaches the slave to the end of list.
1143 * bond->lock held for writing by caller.
1145 static void bond_attach_slave(struct bonding *bond, struct slave *new_slave)
1147 if (bond->first_slave == NULL) { /* attaching the first slave */
1148 new_slave->next = new_slave;
1149 new_slave->prev = new_slave;
1150 bond->first_slave = new_slave;
1152 new_slave->next = bond->first_slave;
1153 new_slave->prev = bond->first_slave->prev;
1154 new_slave->next->prev = new_slave;
1155 new_slave->prev->next = new_slave;
1162 * This function detaches the slave from the list.
1163 * WARNING: no check is made to verify if the slave effectively
1164 * belongs to <bond>.
1165 * Nothing is freed on return, structures are just unchained.
1166 * If any slave pointer in bond was pointing to <slave>,
1167 * it should be changed by the calling function.
1169 * bond->lock held for writing by caller.
1171 static void bond_detach_slave(struct bonding *bond, struct slave *slave)
1174 slave->next->prev = slave->prev;
1178 slave->prev->next = slave->next;
1181 if (bond->first_slave == slave) { /* slave is the first slave */
1182 if (bond->slave_cnt > 1) { /* there are more slave */
1183 bond->first_slave = slave->next;
1185 bond->first_slave = NULL; /* slave was the last one */
1194 /*---------------------------------- IOCTL ----------------------------------*/
1196 static int bond_sethwaddr(struct net_device *bond_dev,
1197 struct net_device *slave_dev)
1199 dprintk("bond_dev=%p\n", bond_dev);
1200 dprintk("slave_dev=%p\n", slave_dev);
1201 dprintk("slave_dev->addr_len=%d\n", slave_dev->addr_len);
1202 memcpy(bond_dev->dev_addr, slave_dev->dev_addr, slave_dev->addr_len);
1206 #define BOND_VLAN_FEATURES \
1207 (NETIF_F_VLAN_CHALLENGED | NETIF_F_HW_VLAN_RX | NETIF_F_HW_VLAN_TX | \
1208 NETIF_F_HW_VLAN_FILTER)
1211 * Compute the common dev->feature set available to all slaves. Some
1212 * feature bits are managed elsewhere, so preserve those feature bits
1213 * on the master device.
1215 static int bond_compute_features(struct bonding *bond)
1217 struct slave *slave;
1218 struct net_device *bond_dev = bond->dev;
1219 unsigned long features = bond_dev->features;
1220 unsigned short max_hard_header_len = ETH_HLEN;
1223 features &= ~(NETIF_F_ALL_CSUM | BOND_VLAN_FEATURES);
1224 features |= NETIF_F_SG | NETIF_F_FRAGLIST | NETIF_F_HIGHDMA |
1225 NETIF_F_GSO_MASK | NETIF_F_NO_CSUM;
1227 bond_for_each_slave(bond, slave, i) {
1228 features = netdev_compute_features(features,
1229 slave->dev->features);
1230 if (slave->dev->hard_header_len > max_hard_header_len)
1231 max_hard_header_len = slave->dev->hard_header_len;
1234 features |= (bond_dev->features & BOND_VLAN_FEATURES);
1235 bond_dev->features = features;
1236 bond_dev->hard_header_len = max_hard_header_len;
1241 /* enslave device <slave> to bond device <master> */
1242 int bond_enslave(struct net_device *bond_dev, struct net_device *slave_dev)
1244 struct bonding *bond = bond_dev->priv;
1245 struct slave *new_slave = NULL;
1246 struct dev_mc_list *dmi;
1247 struct sockaddr addr;
1249 int old_features = bond_dev->features;
1252 if (!bond->params.use_carrier && slave_dev->ethtool_ops == NULL &&
1253 slave_dev->do_ioctl == NULL) {
1254 printk(KERN_WARNING DRV_NAME
1255 ": %s: Warning: no link monitoring support for %s\n",
1256 bond_dev->name, slave_dev->name);
1259 /* bond must be initialized by bond_open() before enslaving */
1260 if (!(bond_dev->flags & IFF_UP)) {
1261 dprintk("Error, master_dev is not up\n");
1265 /* already enslaved */
1266 if (slave_dev->flags & IFF_SLAVE) {
1267 dprintk("Error, Device was already enslaved\n");
1271 /* vlan challenged mutual exclusion */
1272 /* no need to lock since we're protected by rtnl_lock */
1273 if (slave_dev->features & NETIF_F_VLAN_CHALLENGED) {
1274 dprintk("%s: NETIF_F_VLAN_CHALLENGED\n", slave_dev->name);
1275 if (!list_empty(&bond->vlan_list)) {
1276 printk(KERN_ERR DRV_NAME
1277 ": %s: Error: cannot enslave VLAN "
1278 "challenged slave %s on VLAN enabled "
1279 "bond %s\n", bond_dev->name, slave_dev->name,
1283 printk(KERN_WARNING DRV_NAME
1284 ": %s: Warning: enslaved VLAN challenged "
1285 "slave %s. Adding VLANs will be blocked as "
1286 "long as %s is part of bond %s\n",
1287 bond_dev->name, slave_dev->name, slave_dev->name,
1289 bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
1292 dprintk("%s: ! NETIF_F_VLAN_CHALLENGED\n", slave_dev->name);
1293 if (bond->slave_cnt == 0) {
1294 /* First slave, and it is not VLAN challenged,
1295 * so remove the block of adding VLANs over the bond.
1297 bond_dev->features &= ~NETIF_F_VLAN_CHALLENGED;
1302 * Old ifenslave binaries are no longer supported. These can
1303 * be identified with moderate accurary by the state of the slave:
1304 * the current ifenslave will set the interface down prior to
1305 * enslaving it; the old ifenslave will not.
1307 if ((slave_dev->flags & IFF_UP)) {
1308 printk(KERN_ERR DRV_NAME ": %s is up. "
1309 "This may be due to an out of date ifenslave.\n",
1312 goto err_undo_flags;
1315 if (slave_dev->set_mac_address == NULL) {
1316 printk(KERN_ERR DRV_NAME
1317 ": %s: Error: The slave device you specified does "
1318 "not support setting the MAC address. "
1319 "Your kernel likely does not support slave "
1320 "devices.\n", bond_dev->name);
1322 goto err_undo_flags;
1325 new_slave = kzalloc(sizeof(struct slave), GFP_KERNEL);
1328 goto err_undo_flags;
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 res = netdev_set_master(slave_dev, bond_dev);
1357 dprintk("Error %d calling netdev_set_master\n", res);
1360 /* open the slave since the application closed it */
1361 res = dev_open(slave_dev);
1363 dprintk("Openning slave %s failed\n", slave_dev->name);
1364 goto err_restore_mac;
1367 new_slave->dev = slave_dev;
1368 slave_dev->priv_flags |= IFF_BONDING;
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 new_slave->last_arp_rx = jiffies;
1423 if (bond->params.miimon && !bond->params.use_carrier) {
1424 link_reporting = bond_check_dev_link(bond, slave_dev, 1);
1426 if ((link_reporting == -1) && !bond->params.arp_interval) {
1428 * miimon is set but a bonded network driver
1429 * does not support ETHTOOL/MII and
1430 * arp_interval is not set. Note: if
1431 * use_carrier is enabled, we will never go
1432 * here (because netif_carrier is always
1433 * supported); thus, we don't need to change
1434 * the messages for netif_carrier.
1436 printk(KERN_WARNING DRV_NAME
1437 ": %s: Warning: MII and ETHTOOL support not "
1438 "available for interface %s, and "
1439 "arp_interval/arp_ip_target module parameters "
1440 "not specified, thus bonding will not detect "
1441 "link failures! see bonding.txt for details.\n",
1442 bond_dev->name, slave_dev->name);
1443 } else if (link_reporting == -1) {
1444 /* unable get link status using mii/ethtool */
1445 printk(KERN_WARNING DRV_NAME
1446 ": %s: Warning: can't get link status from "
1447 "interface %s; the network driver associated "
1448 "with this interface does not support MII or "
1449 "ETHTOOL link status reporting, thus miimon "
1450 "has no effect on this interface.\n",
1451 bond_dev->name, slave_dev->name);
1455 /* check for initial state */
1456 if (!bond->params.miimon ||
1457 (bond_check_dev_link(bond, slave_dev, 0) == BMSR_LSTATUS)) {
1458 if (bond->params.updelay) {
1459 dprintk("Initial state of slave_dev is "
1460 "BOND_LINK_BACK\n");
1461 new_slave->link = BOND_LINK_BACK;
1462 new_slave->delay = bond->params.updelay;
1464 dprintk("Initial state of slave_dev is "
1466 new_slave->link = BOND_LINK_UP;
1468 new_slave->jiffies = jiffies;
1470 dprintk("Initial state of slave_dev is "
1471 "BOND_LINK_DOWN\n");
1472 new_slave->link = BOND_LINK_DOWN;
1475 if (bond_update_speed_duplex(new_slave) &&
1476 (new_slave->link != BOND_LINK_DOWN)) {
1477 printk(KERN_WARNING DRV_NAME
1478 ": %s: Warning: failed to get speed and duplex from %s, "
1479 "assumed to be 100Mb/sec and Full.\n",
1480 bond_dev->name, new_slave->dev->name);
1482 if (bond->params.mode == BOND_MODE_8023AD) {
1483 printk(KERN_WARNING DRV_NAME
1484 ": %s: Warning: Operation of 802.3ad mode requires ETHTOOL "
1485 "support in base driver for proper aggregator "
1486 "selection.\n", bond_dev->name);
1490 if (USES_PRIMARY(bond->params.mode) && bond->params.primary[0]) {
1491 /* if there is a primary slave, remember it */
1492 if (strcmp(bond->params.primary, new_slave->dev->name) == 0) {
1493 bond->primary_slave = new_slave;
1497 switch (bond->params.mode) {
1498 case BOND_MODE_ACTIVEBACKUP:
1499 bond_set_slave_inactive_flags(new_slave);
1500 bond_select_active_slave(bond);
1502 case BOND_MODE_8023AD:
1503 /* in 802.3ad mode, the internal mechanism
1504 * will activate the slaves in the selected
1507 bond_set_slave_inactive_flags(new_slave);
1508 /* if this is the first slave */
1509 if (bond->slave_cnt == 1) {
1510 SLAVE_AD_INFO(new_slave).id = 1;
1511 /* Initialize AD with the number of times that the AD timer is called in 1 second
1512 * can be called only after the mac address of the bond is set
1514 bond_3ad_initialize(bond, 1000/AD_TIMER_INTERVAL,
1515 bond->params.lacp_fast);
1517 SLAVE_AD_INFO(new_slave).id =
1518 SLAVE_AD_INFO(new_slave->prev).id + 1;
1521 bond_3ad_bind_slave(new_slave);
1525 new_slave->state = BOND_STATE_ACTIVE;
1526 if ((!bond->curr_active_slave) &&
1527 (new_slave->link != BOND_LINK_DOWN)) {
1528 /* first slave or no active slave yet, and this link
1529 * is OK, so make this interface the active one
1531 bond_change_active_slave(bond, new_slave);
1533 bond_set_slave_inactive_flags(new_slave);
1537 dprintk("This slave is always active in trunk mode\n");
1539 /* always active in trunk mode */
1540 new_slave->state = BOND_STATE_ACTIVE;
1542 /* In trunking mode there is little meaning to curr_active_slave
1543 * anyway (it holds no special properties of the bond device),
1544 * so we can change it without calling change_active_interface()
1546 if (!bond->curr_active_slave) {
1547 bond->curr_active_slave = new_slave;
1550 } /* switch(bond_mode) */
1552 bond_set_carrier(bond);
1554 write_unlock_bh(&bond->lock);
1556 res = bond_create_slave_symlinks(bond_dev, slave_dev);
1558 goto err_unset_master;
1560 printk(KERN_INFO DRV_NAME
1561 ": %s: enslaving %s as a%s interface with a%s link.\n",
1562 bond_dev->name, slave_dev->name,
1563 new_slave->state == BOND_STATE_ACTIVE ? "n active" : " backup",
1564 new_slave->link != BOND_LINK_DOWN ? "n up" : " down");
1566 /* enslave is successful */
1569 /* Undo stages on error */
1571 netdev_set_master(slave_dev, NULL);
1574 dev_close(slave_dev);
1577 memcpy(addr.sa_data, new_slave->perm_hwaddr, ETH_ALEN);
1578 addr.sa_family = slave_dev->type;
1579 dev_set_mac_address(slave_dev, &addr);
1585 bond_dev->features = old_features;
1591 * Try to release the slave device <slave> from the bond device <master>
1592 * It is legal to access curr_active_slave without a lock because all the function
1595 * The rules for slave state should be:
1596 * for Active/Backup:
1597 * Active stays on all backups go down
1598 * for Bonded connections:
1599 * The first up interface should be left on and all others downed.
1601 int bond_release(struct net_device *bond_dev, struct net_device *slave_dev)
1603 struct bonding *bond = bond_dev->priv;
1604 struct slave *slave, *oldcurrent;
1605 struct sockaddr addr;
1606 int mac_addr_differ;
1608 /* slave is not a slave or master is not master of this slave */
1609 if (!(slave_dev->flags & IFF_SLAVE) ||
1610 (slave_dev->master != bond_dev)) {
1611 printk(KERN_ERR DRV_NAME
1612 ": %s: Error: cannot release %s.\n",
1613 bond_dev->name, slave_dev->name);
1617 write_lock_bh(&bond->lock);
1619 slave = bond_get_slave_by_dev(bond, slave_dev);
1621 /* not a slave of this bond */
1622 printk(KERN_INFO DRV_NAME
1623 ": %s: %s not enslaved\n",
1624 bond_dev->name, slave_dev->name);
1625 write_unlock_bh(&bond->lock);
1629 mac_addr_differ = memcmp(bond_dev->dev_addr,
1632 if (!mac_addr_differ && (bond->slave_cnt > 1)) {
1633 printk(KERN_WARNING DRV_NAME
1634 ": %s: Warning: the permanent HWaddr of %s "
1635 "- %02X:%02X:%02X:%02X:%02X:%02X - is "
1636 "still in use by %s. Set the HWaddr of "
1637 "%s to a different address to avoid "
1641 slave->perm_hwaddr[0],
1642 slave->perm_hwaddr[1],
1643 slave->perm_hwaddr[2],
1644 slave->perm_hwaddr[3],
1645 slave->perm_hwaddr[4],
1646 slave->perm_hwaddr[5],
1651 /* Inform AD package of unbinding of slave. */
1652 if (bond->params.mode == BOND_MODE_8023AD) {
1653 /* must be called before the slave is
1654 * detached from the list
1656 bond_3ad_unbind_slave(slave);
1659 printk(KERN_INFO DRV_NAME
1660 ": %s: releasing %s interface %s\n",
1662 (slave->state == BOND_STATE_ACTIVE)
1663 ? "active" : "backup",
1666 oldcurrent = bond->curr_active_slave;
1668 bond->current_arp_slave = NULL;
1670 /* release the slave from its bond */
1671 bond_detach_slave(bond, slave);
1673 bond_compute_features(bond);
1675 if (bond->primary_slave == slave) {
1676 bond->primary_slave = NULL;
1679 if (oldcurrent == slave) {
1680 bond_change_active_slave(bond, NULL);
1683 if ((bond->params.mode == BOND_MODE_TLB) ||
1684 (bond->params.mode == BOND_MODE_ALB)) {
1685 /* Must be called only after the slave has been
1686 * detached from the list and the curr_active_slave
1687 * has been cleared (if our_slave == old_current),
1688 * but before a new active slave is selected.
1690 bond_alb_deinit_slave(bond, slave);
1693 if (oldcurrent == slave)
1694 bond_select_active_slave(bond);
1696 if (bond->slave_cnt == 0) {
1697 bond_set_carrier(bond);
1699 /* if the last slave was removed, zero the mac address
1700 * of the master so it will be set by the application
1701 * to the mac address of the first slave
1703 memset(bond_dev->dev_addr, 0, bond_dev->addr_len);
1705 if (list_empty(&bond->vlan_list)) {
1706 bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
1708 printk(KERN_WARNING DRV_NAME
1709 ": %s: Warning: clearing HW address of %s while it "
1710 "still has VLANs.\n",
1711 bond_dev->name, bond_dev->name);
1712 printk(KERN_WARNING DRV_NAME
1713 ": %s: When re-adding slaves, make sure the bond's "
1714 "HW address matches its VLANs'.\n",
1717 } else if ((bond_dev->features & NETIF_F_VLAN_CHALLENGED) &&
1718 !bond_has_challenged_slaves(bond)) {
1719 printk(KERN_INFO DRV_NAME
1720 ": %s: last VLAN challenged slave %s "
1721 "left bond %s. VLAN blocking is removed\n",
1722 bond_dev->name, slave_dev->name, bond_dev->name);
1723 bond_dev->features &= ~NETIF_F_VLAN_CHALLENGED;
1726 write_unlock_bh(&bond->lock);
1728 /* must do this from outside any spinlocks */
1729 bond_destroy_slave_symlinks(bond_dev, slave_dev);
1731 bond_del_vlans_from_slave(bond, slave_dev);
1733 /* If the mode USES_PRIMARY, then we should only remove its
1734 * promisc and mc settings if it was the curr_active_slave, but that was
1735 * already taken care of above when we detached the slave
1737 if (!USES_PRIMARY(bond->params.mode)) {
1738 /* unset promiscuity level from slave */
1739 if (bond_dev->flags & IFF_PROMISC) {
1740 dev_set_promiscuity(slave_dev, -1);
1743 /* unset allmulti level from slave */
1744 if (bond_dev->flags & IFF_ALLMULTI) {
1745 dev_set_allmulti(slave_dev, -1);
1748 /* flush master's mc_list from slave */
1749 bond_mc_list_flush(bond_dev, slave_dev);
1752 netdev_set_master(slave_dev, NULL);
1754 /* close slave before restoring its mac address */
1755 dev_close(slave_dev);
1757 /* restore original ("permanent") mac address */
1758 memcpy(addr.sa_data, slave->perm_hwaddr, ETH_ALEN);
1759 addr.sa_family = slave_dev->type;
1760 dev_set_mac_address(slave_dev, &addr);
1762 slave_dev->priv_flags &= ~(IFF_MASTER_8023AD | IFF_MASTER_ALB |
1763 IFF_SLAVE_INACTIVE | IFF_BONDING |
1768 return 0; /* deletion OK */
1772 * This function releases all slaves.
1774 static int bond_release_all(struct net_device *bond_dev)
1776 struct bonding *bond = bond_dev->priv;
1777 struct slave *slave;
1778 struct net_device *slave_dev;
1779 struct sockaddr addr;
1781 write_lock_bh(&bond->lock);
1783 netif_carrier_off(bond_dev);
1785 if (bond->slave_cnt == 0) {
1789 bond->current_arp_slave = NULL;
1790 bond->primary_slave = NULL;
1791 bond_change_active_slave(bond, NULL);
1793 while ((slave = bond->first_slave) != NULL) {
1794 /* Inform AD package of unbinding of slave
1795 * before slave is detached from the list.
1797 if (bond->params.mode == BOND_MODE_8023AD) {
1798 bond_3ad_unbind_slave(slave);
1801 slave_dev = slave->dev;
1802 bond_detach_slave(bond, slave);
1804 if ((bond->params.mode == BOND_MODE_TLB) ||
1805 (bond->params.mode == BOND_MODE_ALB)) {
1806 /* must be called only after the slave
1807 * has been detached from the list
1809 bond_alb_deinit_slave(bond, slave);
1812 bond_compute_features(bond);
1814 /* now that the slave is detached, unlock and perform
1815 * all the undo steps that should not be called from
1818 write_unlock_bh(&bond->lock);
1820 bond_destroy_slave_symlinks(bond_dev, slave_dev);
1821 bond_del_vlans_from_slave(bond, slave_dev);
1823 /* If the mode USES_PRIMARY, then we should only remove its
1824 * promisc and mc settings if it was the curr_active_slave, but that was
1825 * already taken care of above when we detached the slave
1827 if (!USES_PRIMARY(bond->params.mode)) {
1828 /* unset promiscuity level from slave */
1829 if (bond_dev->flags & IFF_PROMISC) {
1830 dev_set_promiscuity(slave_dev, -1);
1833 /* unset allmulti level from slave */
1834 if (bond_dev->flags & IFF_ALLMULTI) {
1835 dev_set_allmulti(slave_dev, -1);
1838 /* flush master's mc_list from slave */
1839 bond_mc_list_flush(bond_dev, slave_dev);
1842 netdev_set_master(slave_dev, NULL);
1844 /* close slave before restoring its mac address */
1845 dev_close(slave_dev);
1847 /* restore original ("permanent") mac address*/
1848 memcpy(addr.sa_data, slave->perm_hwaddr, ETH_ALEN);
1849 addr.sa_family = slave_dev->type;
1850 dev_set_mac_address(slave_dev, &addr);
1852 slave_dev->priv_flags &= ~(IFF_MASTER_8023AD | IFF_MASTER_ALB |
1853 IFF_SLAVE_INACTIVE);
1857 /* re-acquire the lock before getting the next slave */
1858 write_lock_bh(&bond->lock);
1861 /* zero the mac address of the master so it will be
1862 * set by the application to the mac address of the
1865 memset(bond_dev->dev_addr, 0, bond_dev->addr_len);
1867 if (list_empty(&bond->vlan_list)) {
1868 bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
1870 printk(KERN_WARNING DRV_NAME
1871 ": %s: Warning: clearing HW address of %s while it "
1872 "still has VLANs.\n",
1873 bond_dev->name, bond_dev->name);
1874 printk(KERN_WARNING DRV_NAME
1875 ": %s: When re-adding slaves, make sure the bond's "
1876 "HW address matches its VLANs'.\n",
1880 printk(KERN_INFO DRV_NAME
1881 ": %s: released all slaves\n",
1885 write_unlock_bh(&bond->lock);
1891 * This function changes the active slave to slave <slave_dev>.
1892 * It returns -EINVAL in the following cases.
1893 * - <slave_dev> is not found in the list.
1894 * - There is not active slave now.
1895 * - <slave_dev> is already active.
1896 * - The link state of <slave_dev> is not BOND_LINK_UP.
1897 * - <slave_dev> is not running.
1898 * In these cases, this fuction does nothing.
1899 * In the other cases, currnt_slave pointer is changed and 0 is returned.
1901 static int bond_ioctl_change_active(struct net_device *bond_dev, struct net_device *slave_dev)
1903 struct bonding *bond = bond_dev->priv;
1904 struct slave *old_active = NULL;
1905 struct slave *new_active = NULL;
1908 if (!USES_PRIMARY(bond->params.mode)) {
1912 /* Verify that master_dev is indeed the master of slave_dev */
1913 if (!(slave_dev->flags & IFF_SLAVE) ||
1914 (slave_dev->master != bond_dev)) {
1918 write_lock_bh(&bond->lock);
1920 old_active = bond->curr_active_slave;
1921 new_active = bond_get_slave_by_dev(bond, slave_dev);
1924 * Changing to the current active: do nothing; return success.
1926 if (new_active && (new_active == old_active)) {
1927 write_unlock_bh(&bond->lock);
1933 (new_active->link == BOND_LINK_UP) &&
1934 IS_UP(new_active->dev)) {
1935 bond_change_active_slave(bond, new_active);
1940 write_unlock_bh(&bond->lock);
1945 static int bond_info_query(struct net_device *bond_dev, struct ifbond *info)
1947 struct bonding *bond = bond_dev->priv;
1949 info->bond_mode = bond->params.mode;
1950 info->miimon = bond->params.miimon;
1952 read_lock_bh(&bond->lock);
1953 info->num_slaves = bond->slave_cnt;
1954 read_unlock_bh(&bond->lock);
1959 static int bond_slave_info_query(struct net_device *bond_dev, struct ifslave *info)
1961 struct bonding *bond = bond_dev->priv;
1962 struct slave *slave;
1965 if (info->slave_id < 0) {
1969 read_lock_bh(&bond->lock);
1971 bond_for_each_slave(bond, slave, i) {
1972 if (i == (int)info->slave_id) {
1978 read_unlock_bh(&bond->lock);
1981 strcpy(info->slave_name, slave->dev->name);
1982 info->link = slave->link;
1983 info->state = slave->state;
1984 info->link_failure_count = slave->link_failure_count;
1992 /*-------------------------------- Monitoring -------------------------------*/
1994 /* this function is called regularly to monitor each slave's link. */
1995 void bond_mii_monitor(struct net_device *bond_dev)
1997 struct bonding *bond = bond_dev->priv;
1998 struct slave *slave, *oldcurrent;
1999 int do_failover = 0;
2003 read_lock(&bond->lock);
2005 delta_in_ticks = (bond->params.miimon * HZ) / 1000;
2007 if (bond->kill_timers) {
2011 if (bond->slave_cnt == 0) {
2015 /* we will try to read the link status of each of our slaves, and
2016 * set their IFF_RUNNING flag appropriately. For each slave not
2017 * supporting MII status, we won't do anything so that a user-space
2018 * program could monitor the link itself if needed.
2021 read_lock(&bond->curr_slave_lock);
2022 oldcurrent = bond->curr_active_slave;
2023 read_unlock(&bond->curr_slave_lock);
2025 bond_for_each_slave(bond, slave, i) {
2026 struct net_device *slave_dev = slave->dev;
2028 u16 old_speed = slave->speed;
2029 u8 old_duplex = slave->duplex;
2031 link_state = bond_check_dev_link(bond, slave_dev, 0);
2033 switch (slave->link) {
2034 case BOND_LINK_UP: /* the link was up */
2035 if (link_state == BMSR_LSTATUS) {
2036 /* link stays up, nothing more to do */
2038 } else { /* link going down */
2039 slave->link = BOND_LINK_FAIL;
2040 slave->delay = bond->params.downdelay;
2042 if (slave->link_failure_count < UINT_MAX) {
2043 slave->link_failure_count++;
2046 if (bond->params.downdelay) {
2047 printk(KERN_INFO DRV_NAME
2048 ": %s: link status down for %s "
2049 "interface %s, disabling it in "
2053 ? ((bond->params.mode == BOND_MODE_ACTIVEBACKUP)
2054 ? ((slave == oldcurrent)
2055 ? "active " : "backup ")
2059 bond->params.downdelay * bond->params.miimon);
2062 /* no break ! fall through the BOND_LINK_FAIL test to
2063 ensure proper action to be taken
2065 case BOND_LINK_FAIL: /* the link has just gone down */
2066 if (link_state != BMSR_LSTATUS) {
2067 /* link stays down */
2068 if (slave->delay <= 0) {
2069 /* link down for too long time */
2070 slave->link = BOND_LINK_DOWN;
2072 /* in active/backup mode, we must
2073 * completely disable this interface
2075 if ((bond->params.mode == BOND_MODE_ACTIVEBACKUP) ||
2076 (bond->params.mode == BOND_MODE_8023AD)) {
2077 bond_set_slave_inactive_flags(slave);
2080 printk(KERN_INFO DRV_NAME
2081 ": %s: link status definitely "
2082 "down for interface %s, "
2087 /* notify ad that the link status has changed */
2088 if (bond->params.mode == BOND_MODE_8023AD) {
2089 bond_3ad_handle_link_change(slave, BOND_LINK_DOWN);
2092 if ((bond->params.mode == BOND_MODE_TLB) ||
2093 (bond->params.mode == BOND_MODE_ALB)) {
2094 bond_alb_handle_link_change(bond, slave, BOND_LINK_DOWN);
2097 if (slave == oldcurrent) {
2105 slave->link = BOND_LINK_UP;
2106 slave->jiffies = jiffies;
2107 printk(KERN_INFO DRV_NAME
2108 ": %s: link status up again after %d "
2109 "ms for interface %s.\n",
2111 (bond->params.downdelay - slave->delay) * bond->params.miimon,
2115 case BOND_LINK_DOWN: /* the link was down */
2116 if (link_state != BMSR_LSTATUS) {
2117 /* the link stays down, nothing more to do */
2119 } else { /* link going up */
2120 slave->link = BOND_LINK_BACK;
2121 slave->delay = bond->params.updelay;
2123 if (bond->params.updelay) {
2124 /* if updelay == 0, no need to
2125 advertise about a 0 ms delay */
2126 printk(KERN_INFO DRV_NAME
2127 ": %s: link status up for "
2128 "interface %s, enabling it "
2132 bond->params.updelay * bond->params.miimon);
2135 /* no break ! fall through the BOND_LINK_BACK state in
2136 case there's something to do.
2138 case BOND_LINK_BACK: /* the link has just come back */
2139 if (link_state != BMSR_LSTATUS) {
2140 /* link down again */
2141 slave->link = BOND_LINK_DOWN;
2143 printk(KERN_INFO DRV_NAME
2144 ": %s: link status down again after %d "
2145 "ms for interface %s.\n",
2147 (bond->params.updelay - slave->delay) * bond->params.miimon,
2151 if (slave->delay == 0) {
2152 /* now the link has been up for long time enough */
2153 slave->link = BOND_LINK_UP;
2154 slave->jiffies = jiffies;
2156 if (bond->params.mode == BOND_MODE_8023AD) {
2157 /* prevent it from being the active one */
2158 slave->state = BOND_STATE_BACKUP;
2159 } else if (bond->params.mode != BOND_MODE_ACTIVEBACKUP) {
2160 /* make it immediately active */
2161 slave->state = BOND_STATE_ACTIVE;
2162 } else if (slave != bond->primary_slave) {
2163 /* prevent it from being the active one */
2164 slave->state = BOND_STATE_BACKUP;
2167 printk(KERN_INFO DRV_NAME
2168 ": %s: link status definitely "
2169 "up for interface %s.\n",
2173 /* notify ad that the link status has changed */
2174 if (bond->params.mode == BOND_MODE_8023AD) {
2175 bond_3ad_handle_link_change(slave, BOND_LINK_UP);
2178 if ((bond->params.mode == BOND_MODE_TLB) ||
2179 (bond->params.mode == BOND_MODE_ALB)) {
2180 bond_alb_handle_link_change(bond, slave, BOND_LINK_UP);
2183 if ((!oldcurrent) ||
2184 (slave == bond->primary_slave)) {
2193 /* Should not happen */
2194 printk(KERN_ERR DRV_NAME
2195 ": %s: Error: %s Illegal value (link=%d)\n",
2200 } /* end of switch (slave->link) */
2202 bond_update_speed_duplex(slave);
2204 if (bond->params.mode == BOND_MODE_8023AD) {
2205 if (old_speed != slave->speed) {
2206 bond_3ad_adapter_speed_changed(slave);
2209 if (old_duplex != slave->duplex) {
2210 bond_3ad_adapter_duplex_changed(slave);
2217 write_lock(&bond->curr_slave_lock);
2219 bond_select_active_slave(bond);
2221 write_unlock(&bond->curr_slave_lock);
2223 bond_set_carrier(bond);
2226 if (bond->params.miimon) {
2227 mod_timer(&bond->mii_timer, jiffies + delta_in_ticks);
2230 read_unlock(&bond->lock);
2234 static u32 bond_glean_dev_ip(struct net_device *dev)
2236 struct in_device *idev;
2237 struct in_ifaddr *ifa;
2244 idev = __in_dev_get_rcu(dev);
2248 ifa = idev->ifa_list;
2252 addr = ifa->ifa_local;
2258 static int bond_has_ip(struct bonding *bond)
2260 struct vlan_entry *vlan, *vlan_next;
2262 if (bond->master_ip)
2265 if (list_empty(&bond->vlan_list))
2268 list_for_each_entry_safe(vlan, vlan_next, &bond->vlan_list,
2277 static int bond_has_this_ip(struct bonding *bond, u32 ip)
2279 struct vlan_entry *vlan, *vlan_next;
2281 if (ip == bond->master_ip)
2284 if (list_empty(&bond->vlan_list))
2287 list_for_each_entry_safe(vlan, vlan_next, &bond->vlan_list,
2289 if (ip == vlan->vlan_ip)
2297 * We go to the (large) trouble of VLAN tagging ARP frames because
2298 * switches in VLAN mode (especially if ports are configured as
2299 * "native" to a VLAN) might not pass non-tagged frames.
2301 static void bond_arp_send(struct net_device *slave_dev, int arp_op, u32 dest_ip, u32 src_ip, unsigned short vlan_id)
2303 struct sk_buff *skb;
2305 dprintk("arp %d on slave %s: dst %x src %x vid %d\n", arp_op,
2306 slave_dev->name, dest_ip, src_ip, vlan_id);
2308 skb = arp_create(arp_op, ETH_P_ARP, dest_ip, slave_dev, src_ip,
2309 NULL, slave_dev->dev_addr, NULL);
2312 printk(KERN_ERR DRV_NAME ": ARP packet allocation failed\n");
2316 skb = vlan_put_tag(skb, vlan_id);
2318 printk(KERN_ERR DRV_NAME ": failed to insert VLAN tag\n");
2326 static void bond_arp_send_all(struct bonding *bond, struct slave *slave)
2329 u32 *targets = bond->params.arp_targets;
2330 struct vlan_entry *vlan, *vlan_next;
2331 struct net_device *vlan_dev;
2335 for (i = 0; (i < BOND_MAX_ARP_TARGETS); i++) {
2338 dprintk("basa: target %x\n", targets[i]);
2339 if (list_empty(&bond->vlan_list)) {
2340 dprintk("basa: empty vlan: arp_send\n");
2341 bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2342 bond->master_ip, 0);
2347 * If VLANs are configured, we do a route lookup to
2348 * determine which VLAN interface would be used, so we
2349 * can tag the ARP with the proper VLAN tag.
2351 memset(&fl, 0, sizeof(fl));
2352 fl.fl4_dst = targets[i];
2353 fl.fl4_tos = RTO_ONLINK;
2355 rv = ip_route_output_key(&rt, &fl);
2357 if (net_ratelimit()) {
2358 printk(KERN_WARNING DRV_NAME
2359 ": %s: no route to arp_ip_target %u.%u.%u.%u\n",
2360 bond->dev->name, NIPQUAD(fl.fl4_dst));
2366 * This target is not on a VLAN
2368 if (rt->u.dst.dev == bond->dev) {
2370 dprintk("basa: rtdev == bond->dev: arp_send\n");
2371 bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2372 bond->master_ip, 0);
2377 list_for_each_entry_safe(vlan, vlan_next, &bond->vlan_list,
2379 vlan_dev = vlan_group_get_device(bond->vlgrp, vlan->vlan_id);
2380 if (vlan_dev == rt->u.dst.dev) {
2381 vlan_id = vlan->vlan_id;
2382 dprintk("basa: vlan match on %s %d\n",
2383 vlan_dev->name, vlan_id);
2390 bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2391 vlan->vlan_ip, vlan_id);
2395 if (net_ratelimit()) {
2396 printk(KERN_WARNING DRV_NAME
2397 ": %s: no path to arp_ip_target %u.%u.%u.%u via rt.dev %s\n",
2398 bond->dev->name, NIPQUAD(fl.fl4_dst),
2399 rt->u.dst.dev ? rt->u.dst.dev->name : "NULL");
2406 * Kick out a gratuitous ARP for an IP on the bonding master plus one
2407 * for each VLAN above us.
2409 static void bond_send_gratuitous_arp(struct bonding *bond)
2411 struct slave *slave = bond->curr_active_slave;
2412 struct vlan_entry *vlan;
2413 struct net_device *vlan_dev;
2415 dprintk("bond_send_grat_arp: bond %s slave %s\n", bond->dev->name,
2416 slave ? slave->dev->name : "NULL");
2420 if (bond->master_ip) {
2421 bond_arp_send(slave->dev, ARPOP_REPLY, bond->master_ip,
2422 bond->master_ip, 0);
2425 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
2426 vlan_dev = vlan_group_get_device(bond->vlgrp, vlan->vlan_id);
2427 if (vlan->vlan_ip) {
2428 bond_arp_send(slave->dev, ARPOP_REPLY, vlan->vlan_ip,
2429 vlan->vlan_ip, vlan->vlan_id);
2434 static void bond_validate_arp(struct bonding *bond, struct slave *slave, u32 sip, u32 tip)
2437 u32 *targets = bond->params.arp_targets;
2439 targets = bond->params.arp_targets;
2440 for (i = 0; (i < BOND_MAX_ARP_TARGETS) && targets[i]; i++) {
2441 dprintk("bva: sip %u.%u.%u.%u tip %u.%u.%u.%u t[%d] "
2442 "%u.%u.%u.%u bhti(tip) %d\n",
2443 NIPQUAD(sip), NIPQUAD(tip), i, NIPQUAD(targets[i]),
2444 bond_has_this_ip(bond, tip));
2445 if (sip == targets[i]) {
2446 if (bond_has_this_ip(bond, tip))
2447 slave->last_arp_rx = jiffies;
2453 static int bond_arp_rcv(struct sk_buff *skb, struct net_device *dev, struct packet_type *pt, struct net_device *orig_dev)
2456 struct slave *slave;
2457 struct bonding *bond;
2458 unsigned char *arp_ptr;
2461 if (dev->nd_net != &init_net)
2464 if (!(dev->priv_flags & IFF_BONDING) || !(dev->flags & IFF_MASTER))
2468 read_lock(&bond->lock);
2470 dprintk("bond_arp_rcv: bond %s skb->dev %s orig_dev %s\n",
2471 bond->dev->name, skb->dev ? skb->dev->name : "NULL",
2472 orig_dev ? orig_dev->name : "NULL");
2474 slave = bond_get_slave_by_dev(bond, orig_dev);
2475 if (!slave || !slave_do_arp_validate(bond, slave))
2478 /* ARP header, plus 2 device addresses, plus 2 IP addresses. */
2479 if (!pskb_may_pull(skb, (sizeof(struct arphdr) +
2480 (2 * dev->addr_len) +
2481 (2 * sizeof(u32)))))
2485 if (arp->ar_hln != dev->addr_len ||
2486 skb->pkt_type == PACKET_OTHERHOST ||
2487 skb->pkt_type == PACKET_LOOPBACK ||
2488 arp->ar_hrd != htons(ARPHRD_ETHER) ||
2489 arp->ar_pro != htons(ETH_P_IP) ||
2493 arp_ptr = (unsigned char *)(arp + 1);
2494 arp_ptr += dev->addr_len;
2495 memcpy(&sip, arp_ptr, 4);
2496 arp_ptr += 4 + dev->addr_len;
2497 memcpy(&tip, arp_ptr, 4);
2499 dprintk("bond_arp_rcv: %s %s/%d av %d sv %d sip %u.%u.%u.%u"
2500 " tip %u.%u.%u.%u\n", bond->dev->name, slave->dev->name,
2501 slave->state, bond->params.arp_validate,
2502 slave_do_arp_validate(bond, slave), NIPQUAD(sip), NIPQUAD(tip));
2505 * Backup slaves won't see the ARP reply, but do come through
2506 * here for each ARP probe (so we swap the sip/tip to validate
2507 * the probe). In a "redundant switch, common router" type of
2508 * configuration, the ARP probe will (hopefully) travel from
2509 * the active, through one switch, the router, then the other
2510 * switch before reaching the backup.
2512 if (slave->state == BOND_STATE_ACTIVE)
2513 bond_validate_arp(bond, slave, sip, tip);
2515 bond_validate_arp(bond, slave, tip, sip);
2518 read_unlock(&bond->lock);
2521 return NET_RX_SUCCESS;
2525 * this function is called regularly to monitor each slave's link
2526 * ensuring that traffic is being sent and received when arp monitoring
2527 * is used in load-balancing mode. if the adapter has been dormant, then an
2528 * arp is transmitted to generate traffic. see activebackup_arp_monitor for
2529 * arp monitoring in active backup mode.
2531 void bond_loadbalance_arp_mon(struct net_device *bond_dev)
2533 struct bonding *bond = bond_dev->priv;
2534 struct slave *slave, *oldcurrent;
2535 int do_failover = 0;
2539 read_lock(&bond->lock);
2541 delta_in_ticks = (bond->params.arp_interval * HZ) / 1000;
2543 if (bond->kill_timers) {
2547 if (bond->slave_cnt == 0) {
2551 read_lock(&bond->curr_slave_lock);
2552 oldcurrent = bond->curr_active_slave;
2553 read_unlock(&bond->curr_slave_lock);
2555 /* see if any of the previous devices are up now (i.e. they have
2556 * xmt and rcv traffic). the curr_active_slave does not come into
2557 * the picture unless it is null. also, slave->jiffies is not needed
2558 * here because we send an arp on each slave and give a slave as
2559 * long as it needs to get the tx/rx within the delta.
2560 * TODO: what about up/down delay in arp mode? it wasn't here before
2563 bond_for_each_slave(bond, slave, i) {
2564 if (slave->link != BOND_LINK_UP) {
2565 if (((jiffies - slave->dev->trans_start) <= delta_in_ticks) &&
2566 ((jiffies - slave->dev->last_rx) <= delta_in_ticks)) {
2568 slave->link = BOND_LINK_UP;
2569 slave->state = BOND_STATE_ACTIVE;
2571 /* primary_slave has no meaning in round-robin
2572 * mode. the window of a slave being up and
2573 * curr_active_slave being null after enslaving
2577 printk(KERN_INFO DRV_NAME
2578 ": %s: link status definitely "
2579 "up for interface %s, ",
2584 printk(KERN_INFO DRV_NAME
2585 ": %s: interface %s is now up\n",
2591 /* slave->link == BOND_LINK_UP */
2593 /* not all switches will respond to an arp request
2594 * when the source ip is 0, so don't take the link down
2595 * if we don't know our ip yet
2597 if (((jiffies - slave->dev->trans_start) >= (2*delta_in_ticks)) ||
2598 (((jiffies - slave->dev->last_rx) >= (2*delta_in_ticks)) &&
2599 bond_has_ip(bond))) {
2601 slave->link = BOND_LINK_DOWN;
2602 slave->state = BOND_STATE_BACKUP;
2604 if (slave->link_failure_count < UINT_MAX) {
2605 slave->link_failure_count++;
2608 printk(KERN_INFO DRV_NAME
2609 ": %s: interface %s is now down.\n",
2613 if (slave == oldcurrent) {
2619 /* note: if switch is in round-robin mode, all links
2620 * must tx arp to ensure all links rx an arp - otherwise
2621 * links may oscillate or not come up at all; if switch is
2622 * in something like xor mode, there is nothing we can
2623 * do - all replies will be rx'ed on same link causing slaves
2624 * to be unstable during low/no traffic periods
2626 if (IS_UP(slave->dev)) {
2627 bond_arp_send_all(bond, slave);
2632 write_lock(&bond->curr_slave_lock);
2634 bond_select_active_slave(bond);
2636 write_unlock(&bond->curr_slave_lock);
2640 if (bond->params.arp_interval) {
2641 mod_timer(&bond->arp_timer, jiffies + delta_in_ticks);
2644 read_unlock(&bond->lock);
2648 * When using arp monitoring in active-backup mode, this function is
2649 * called to determine if any backup slaves have went down or a new
2650 * current slave needs to be found.
2651 * The backup slaves never generate traffic, they are considered up by merely
2652 * receiving traffic. If the current slave goes down, each backup slave will
2653 * be given the opportunity to tx/rx an arp before being taken down - this
2654 * prevents all slaves from being taken down due to the current slave not
2655 * sending any traffic for the backups to receive. The arps are not necessarily
2656 * necessary, any tx and rx traffic will keep the current slave up. While any
2657 * rx traffic will keep the backup slaves up, the current slave is responsible
2658 * for generating traffic to keep them up regardless of any other traffic they
2659 * may have received.
2660 * see loadbalance_arp_monitor for arp monitoring in load balancing mode
2662 void bond_activebackup_arp_mon(struct net_device *bond_dev)
2664 struct bonding *bond = bond_dev->priv;
2665 struct slave *slave;
2669 read_lock(&bond->lock);
2671 delta_in_ticks = (bond->params.arp_interval * HZ) / 1000;
2673 if (bond->kill_timers) {
2677 if (bond->slave_cnt == 0) {
2681 /* determine if any slave has come up or any backup slave has
2683 * TODO: what about up/down delay in arp mode? it wasn't here before
2686 bond_for_each_slave(bond, slave, i) {
2687 if (slave->link != BOND_LINK_UP) {
2688 if ((jiffies - slave_last_rx(bond, slave)) <=
2691 slave->link = BOND_LINK_UP;
2693 write_lock(&bond->curr_slave_lock);
2695 if ((!bond->curr_active_slave) &&
2696 ((jiffies - slave->dev->trans_start) <= delta_in_ticks)) {
2697 bond_change_active_slave(bond, slave);
2698 bond->current_arp_slave = NULL;
2699 } else if (bond->curr_active_slave != slave) {
2700 /* this slave has just come up but we
2701 * already have a current slave; this
2702 * can also happen if bond_enslave adds
2703 * a new slave that is up while we are
2704 * searching for a new slave
2706 bond_set_slave_inactive_flags(slave);
2707 bond->current_arp_slave = NULL;
2710 bond_set_carrier(bond);
2712 if (slave == bond->curr_active_slave) {
2713 printk(KERN_INFO DRV_NAME
2714 ": %s: %s is up and now the "
2715 "active interface\n",
2718 netif_carrier_on(bond->dev);
2720 printk(KERN_INFO DRV_NAME
2721 ": %s: backup interface %s is "
2727 write_unlock(&bond->curr_slave_lock);
2730 read_lock(&bond->curr_slave_lock);
2732 if ((slave != bond->curr_active_slave) &&
2733 (!bond->current_arp_slave) &&
2734 (((jiffies - slave_last_rx(bond, slave)) >= 3*delta_in_ticks) &&
2735 bond_has_ip(bond))) {
2736 /* a backup slave has gone down; three times
2737 * the delta allows the current slave to be
2738 * taken out before the backup slave.
2739 * note: a non-null current_arp_slave indicates
2740 * the curr_active_slave went down and we are
2741 * searching for a new one; under this
2742 * condition we only take the curr_active_slave
2743 * down - this gives each slave a chance to
2744 * tx/rx traffic before being taken out
2747 read_unlock(&bond->curr_slave_lock);
2749 slave->link = BOND_LINK_DOWN;
2751 if (slave->link_failure_count < UINT_MAX) {
2752 slave->link_failure_count++;
2755 bond_set_slave_inactive_flags(slave);
2757 printk(KERN_INFO DRV_NAME
2758 ": %s: backup interface %s is now down\n",
2762 read_unlock(&bond->curr_slave_lock);
2767 read_lock(&bond->curr_slave_lock);
2768 slave = bond->curr_active_slave;
2769 read_unlock(&bond->curr_slave_lock);
2772 /* if we have sent traffic in the past 2*arp_intervals but
2773 * haven't xmit and rx traffic in that time interval, select
2774 * a different slave. slave->jiffies is only updated when
2775 * a slave first becomes the curr_active_slave - not necessarily
2776 * after every arp; this ensures the slave has a full 2*delta
2777 * before being taken out. if a primary is being used, check
2778 * if it is up and needs to take over as the curr_active_slave
2780 if ((((jiffies - slave->dev->trans_start) >= (2*delta_in_ticks)) ||
2781 (((jiffies - slave_last_rx(bond, slave)) >= (2*delta_in_ticks)) &&
2782 bond_has_ip(bond))) &&
2783 ((jiffies - slave->jiffies) >= 2*delta_in_ticks)) {
2785 slave->link = BOND_LINK_DOWN;
2787 if (slave->link_failure_count < UINT_MAX) {
2788 slave->link_failure_count++;
2791 printk(KERN_INFO DRV_NAME
2792 ": %s: link status down for active interface "
2793 "%s, disabling it\n",
2797 write_lock(&bond->curr_slave_lock);
2799 bond_select_active_slave(bond);
2800 slave = bond->curr_active_slave;
2802 write_unlock(&bond->curr_slave_lock);
2804 bond->current_arp_slave = slave;
2807 slave->jiffies = jiffies;
2809 } else if ((bond->primary_slave) &&
2810 (bond->primary_slave != slave) &&
2811 (bond->primary_slave->link == BOND_LINK_UP)) {
2812 /* at this point, slave is the curr_active_slave */
2813 printk(KERN_INFO DRV_NAME
2814 ": %s: changing from interface %s to primary "
2818 bond->primary_slave->dev->name);
2820 /* primary is up so switch to it */
2821 write_lock(&bond->curr_slave_lock);
2822 bond_change_active_slave(bond, bond->primary_slave);
2823 write_unlock(&bond->curr_slave_lock);
2825 slave = bond->primary_slave;
2826 slave->jiffies = jiffies;
2828 bond->current_arp_slave = NULL;
2831 /* the current slave must tx an arp to ensure backup slaves
2834 if (slave && bond_has_ip(bond)) {
2835 bond_arp_send_all(bond, slave);
2839 /* if we don't have a curr_active_slave, search for the next available
2840 * backup slave from the current_arp_slave and make it the candidate
2841 * for becoming the curr_active_slave
2844 if (!bond->current_arp_slave) {
2845 bond->current_arp_slave = bond->first_slave;
2848 if (bond->current_arp_slave) {
2849 bond_set_slave_inactive_flags(bond->current_arp_slave);
2851 /* search for next candidate */
2852 bond_for_each_slave_from(bond, slave, i, bond->current_arp_slave->next) {
2853 if (IS_UP(slave->dev)) {
2854 slave->link = BOND_LINK_BACK;
2855 bond_set_slave_active_flags(slave);
2856 bond_arp_send_all(bond, slave);
2857 slave->jiffies = jiffies;
2858 bond->current_arp_slave = slave;
2862 /* if the link state is up at this point, we
2863 * mark it down - this can happen if we have
2864 * simultaneous link failures and
2865 * reselect_active_interface doesn't make this
2866 * one the current slave so it is still marked
2867 * up when it is actually down
2869 if (slave->link == BOND_LINK_UP) {
2870 slave->link = BOND_LINK_DOWN;
2871 if (slave->link_failure_count < UINT_MAX) {
2872 slave->link_failure_count++;
2875 bond_set_slave_inactive_flags(slave);
2877 printk(KERN_INFO DRV_NAME
2878 ": %s: backup interface %s is "
2888 if (bond->params.arp_interval) {
2889 mod_timer(&bond->arp_timer, jiffies + delta_in_ticks);
2892 read_unlock(&bond->lock);
2895 /*------------------------------ proc/seq_file-------------------------------*/
2897 #ifdef CONFIG_PROC_FS
2899 #define SEQ_START_TOKEN ((void *)1)
2901 static void *bond_info_seq_start(struct seq_file *seq, loff_t *pos)
2903 struct bonding *bond = seq->private;
2905 struct slave *slave;
2908 /* make sure the bond won't be taken away */
2909 read_lock(&dev_base_lock);
2910 read_lock_bh(&bond->lock);
2913 return SEQ_START_TOKEN;
2916 bond_for_each_slave(bond, slave, i) {
2917 if (++off == *pos) {
2925 static void *bond_info_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2927 struct bonding *bond = seq->private;
2928 struct slave *slave = v;
2931 if (v == SEQ_START_TOKEN) {
2932 return bond->first_slave;
2935 slave = slave->next;
2937 return (slave == bond->first_slave) ? NULL : slave;
2940 static void bond_info_seq_stop(struct seq_file *seq, void *v)
2942 struct bonding *bond = seq->private;
2944 read_unlock_bh(&bond->lock);
2945 read_unlock(&dev_base_lock);
2948 static void bond_info_show_master(struct seq_file *seq)
2950 struct bonding *bond = seq->private;
2955 read_lock(&bond->curr_slave_lock);
2956 curr = bond->curr_active_slave;
2957 read_unlock(&bond->curr_slave_lock);
2959 seq_printf(seq, "Bonding Mode: %s\n",
2960 bond_mode_name(bond->params.mode));
2962 if (bond->params.mode == BOND_MODE_XOR ||
2963 bond->params.mode == BOND_MODE_8023AD) {
2964 seq_printf(seq, "Transmit Hash Policy: %s (%d)\n",
2965 xmit_hashtype_tbl[bond->params.xmit_policy].modename,
2966 bond->params.xmit_policy);
2969 if (USES_PRIMARY(bond->params.mode)) {
2970 seq_printf(seq, "Primary Slave: %s\n",
2971 (bond->primary_slave) ?
2972 bond->primary_slave->dev->name : "None");
2974 seq_printf(seq, "Currently Active Slave: %s\n",
2975 (curr) ? curr->dev->name : "None");
2978 seq_printf(seq, "MII Status: %s\n", netif_carrier_ok(bond->dev) ?
2980 seq_printf(seq, "MII Polling Interval (ms): %d\n", bond->params.miimon);
2981 seq_printf(seq, "Up Delay (ms): %d\n",
2982 bond->params.updelay * bond->params.miimon);
2983 seq_printf(seq, "Down Delay (ms): %d\n",
2984 bond->params.downdelay * bond->params.miimon);
2987 /* ARP information */
2988 if(bond->params.arp_interval > 0) {
2990 seq_printf(seq, "ARP Polling Interval (ms): %d\n",
2991 bond->params.arp_interval);
2993 seq_printf(seq, "ARP IP target/s (n.n.n.n form):");
2995 for(i = 0; (i < BOND_MAX_ARP_TARGETS) ;i++) {
2996 if (!bond->params.arp_targets[i])
2999 seq_printf(seq, ",");
3000 target = ntohl(bond->params.arp_targets[i]);
3001 seq_printf(seq, " %d.%d.%d.%d", HIPQUAD(target));
3004 seq_printf(seq, "\n");
3007 if (bond->params.mode == BOND_MODE_8023AD) {
3008 struct ad_info ad_info;
3010 seq_puts(seq, "\n802.3ad info\n");
3011 seq_printf(seq, "LACP rate: %s\n",
3012 (bond->params.lacp_fast) ? "fast" : "slow");
3014 if (bond_3ad_get_active_agg_info(bond, &ad_info)) {
3015 seq_printf(seq, "bond %s has no active aggregator\n",
3018 seq_printf(seq, "Active Aggregator Info:\n");
3020 seq_printf(seq, "\tAggregator ID: %d\n",
3021 ad_info.aggregator_id);
3022 seq_printf(seq, "\tNumber of ports: %d\n",
3024 seq_printf(seq, "\tActor Key: %d\n",
3026 seq_printf(seq, "\tPartner Key: %d\n",
3027 ad_info.partner_key);
3028 seq_printf(seq, "\tPartner Mac Address: %02x:%02x:%02x:%02x:%02x:%02x\n",
3029 ad_info.partner_system[0],
3030 ad_info.partner_system[1],
3031 ad_info.partner_system[2],
3032 ad_info.partner_system[3],
3033 ad_info.partner_system[4],
3034 ad_info.partner_system[5]);
3039 static void bond_info_show_slave(struct seq_file *seq, const struct slave *slave)
3041 struct bonding *bond = seq->private;
3043 seq_printf(seq, "\nSlave Interface: %s\n", slave->dev->name);
3044 seq_printf(seq, "MII Status: %s\n",
3045 (slave->link == BOND_LINK_UP) ? "up" : "down");
3046 seq_printf(seq, "Link Failure Count: %u\n",
3047 slave->link_failure_count);
3050 "Permanent HW addr: %02x:%02x:%02x:%02x:%02x:%02x\n",
3051 slave->perm_hwaddr[0], slave->perm_hwaddr[1],
3052 slave->perm_hwaddr[2], slave->perm_hwaddr[3],
3053 slave->perm_hwaddr[4], slave->perm_hwaddr[5]);
3055 if (bond->params.mode == BOND_MODE_8023AD) {
3056 const struct aggregator *agg
3057 = SLAVE_AD_INFO(slave).port.aggregator;
3060 seq_printf(seq, "Aggregator ID: %d\n",
3061 agg->aggregator_identifier);
3063 seq_puts(seq, "Aggregator ID: N/A\n");
3068 static int bond_info_seq_show(struct seq_file *seq, void *v)
3070 if (v == SEQ_START_TOKEN) {
3071 seq_printf(seq, "%s\n", version);
3072 bond_info_show_master(seq);
3074 bond_info_show_slave(seq, v);
3080 static struct seq_operations bond_info_seq_ops = {
3081 .start = bond_info_seq_start,
3082 .next = bond_info_seq_next,
3083 .stop = bond_info_seq_stop,
3084 .show = bond_info_seq_show,
3087 static int bond_info_open(struct inode *inode, struct file *file)
3089 struct seq_file *seq;
3090 struct proc_dir_entry *proc;
3093 res = seq_open(file, &bond_info_seq_ops);
3095 /* recover the pointer buried in proc_dir_entry data */
3096 seq = file->private_data;
3098 seq->private = proc->data;
3104 static const struct file_operations bond_info_fops = {
3105 .owner = THIS_MODULE,
3106 .open = bond_info_open,
3108 .llseek = seq_lseek,
3109 .release = seq_release,
3112 static int bond_create_proc_entry(struct bonding *bond)
3114 struct net_device *bond_dev = bond->dev;
3116 if (bond_proc_dir) {
3117 bond->proc_entry = create_proc_entry(bond_dev->name,
3120 if (bond->proc_entry == NULL) {
3121 printk(KERN_WARNING DRV_NAME
3122 ": Warning: Cannot create /proc/net/%s/%s\n",
3123 DRV_NAME, bond_dev->name);
3125 bond->proc_entry->data = bond;
3126 bond->proc_entry->proc_fops = &bond_info_fops;
3127 bond->proc_entry->owner = THIS_MODULE;
3128 memcpy(bond->proc_file_name, bond_dev->name, IFNAMSIZ);
3135 static void bond_remove_proc_entry(struct bonding *bond)
3137 if (bond_proc_dir && bond->proc_entry) {
3138 remove_proc_entry(bond->proc_file_name, bond_proc_dir);
3139 memset(bond->proc_file_name, 0, IFNAMSIZ);
3140 bond->proc_entry = NULL;
3144 /* Create the bonding directory under /proc/net, if doesn't exist yet.
3145 * Caller must hold rtnl_lock.
3147 static void bond_create_proc_dir(void)
3149 int len = strlen(DRV_NAME);
3151 for (bond_proc_dir = init_net.proc_net->subdir; bond_proc_dir;
3152 bond_proc_dir = bond_proc_dir->next) {
3153 if ((bond_proc_dir->namelen == len) &&
3154 !memcmp(bond_proc_dir->name, DRV_NAME, len)) {
3159 if (!bond_proc_dir) {
3160 bond_proc_dir = proc_mkdir(DRV_NAME, init_net.proc_net);
3161 if (bond_proc_dir) {
3162 bond_proc_dir->owner = THIS_MODULE;
3164 printk(KERN_WARNING DRV_NAME
3165 ": Warning: cannot create /proc/net/%s\n",
3171 /* Destroy the bonding directory under /proc/net, if empty.
3172 * Caller must hold rtnl_lock.
3174 static void bond_destroy_proc_dir(void)
3176 struct proc_dir_entry *de;
3178 if (!bond_proc_dir) {
3182 /* verify that the /proc dir is empty */
3183 for (de = bond_proc_dir->subdir; de; de = de->next) {
3184 /* ignore . and .. */
3185 if (*(de->name) != '.') {
3191 if (bond_proc_dir->owner == THIS_MODULE) {
3192 bond_proc_dir->owner = NULL;
3195 remove_proc_entry(DRV_NAME, init_net.proc_net);
3196 bond_proc_dir = NULL;
3199 #endif /* CONFIG_PROC_FS */
3201 /*-------------------------- netdev event handling --------------------------*/
3204 * Change device name
3206 static int bond_event_changename(struct bonding *bond)
3208 #ifdef CONFIG_PROC_FS
3209 bond_remove_proc_entry(bond);
3210 bond_create_proc_entry(bond);
3212 down_write(&(bonding_rwsem));
3213 bond_destroy_sysfs_entry(bond);
3214 bond_create_sysfs_entry(bond);
3215 up_write(&(bonding_rwsem));
3219 static int bond_master_netdev_event(unsigned long event, struct net_device *bond_dev)
3221 struct bonding *event_bond = bond_dev->priv;
3224 case NETDEV_CHANGENAME:
3225 return bond_event_changename(event_bond);
3226 case NETDEV_UNREGISTER:
3228 * TODO: remove a bond from the list?
3238 static int bond_slave_netdev_event(unsigned long event, struct net_device *slave_dev)
3240 struct net_device *bond_dev = slave_dev->master;
3241 struct bonding *bond = bond_dev->priv;
3244 case NETDEV_UNREGISTER:
3246 bond_release(bond_dev, slave_dev);
3251 * TODO: is this what we get if somebody
3252 * sets up a hierarchical bond, then rmmod's
3253 * one of the slave bonding devices?
3258 * ... Or is it this?
3261 case NETDEV_CHANGEMTU:
3263 * TODO: Should slaves be allowed to
3264 * independently alter their MTU? For
3265 * an active-backup bond, slaves need
3266 * not be the same type of device, so
3267 * MTUs may vary. For other modes,
3268 * slaves arguably should have the
3269 * same MTUs. To do this, we'd need to
3270 * take over the slave's change_mtu
3271 * function for the duration of their
3275 case NETDEV_CHANGENAME:
3277 * TODO: handle changing the primary's name
3280 case NETDEV_FEAT_CHANGE:
3281 bond_compute_features(bond);
3291 * bond_netdev_event: handle netdev notifier chain events.
3293 * This function receives events for the netdev chain. The caller (an
3294 * ioctl handler calling blocking_notifier_call_chain) holds the necessary
3295 * locks for us to safely manipulate the slave devices (RTNL lock,
3298 static int bond_netdev_event(struct notifier_block *this, unsigned long event, void *ptr)
3300 struct net_device *event_dev = (struct net_device *)ptr;
3302 dprintk("event_dev: %s, event: %lx\n",
3303 (event_dev ? event_dev->name : "None"),
3306 if (!(event_dev->priv_flags & IFF_BONDING))
3309 if (event_dev->flags & IFF_MASTER) {
3310 dprintk("IFF_MASTER\n");
3311 return bond_master_netdev_event(event, event_dev);
3314 if (event_dev->flags & IFF_SLAVE) {
3315 dprintk("IFF_SLAVE\n");
3316 return bond_slave_netdev_event(event, event_dev);
3323 * bond_inetaddr_event: handle inetaddr notifier chain events.
3325 * We keep track of device IPs primarily to use as source addresses in
3326 * ARP monitor probes (rather than spewing out broadcasts all the time).
3328 * We track one IP for the main device (if it has one), plus one per VLAN.
3330 static int bond_inetaddr_event(struct notifier_block *this, unsigned long event, void *ptr)
3332 struct in_ifaddr *ifa = ptr;
3333 struct net_device *vlan_dev, *event_dev = ifa->ifa_dev->dev;
3334 struct bonding *bond, *bond_next;
3335 struct vlan_entry *vlan, *vlan_next;
3337 list_for_each_entry_safe(bond, bond_next, &bond_dev_list, bond_list) {
3338 if (bond->dev == event_dev) {
3341 bond->master_ip = ifa->ifa_local;
3344 bond->master_ip = bond_glean_dev_ip(bond->dev);
3351 if (list_empty(&bond->vlan_list))
3354 list_for_each_entry_safe(vlan, vlan_next, &bond->vlan_list,
3356 vlan_dev = vlan_group_get_device(bond->vlgrp, vlan->vlan_id);
3357 if (vlan_dev == event_dev) {
3360 vlan->vlan_ip = ifa->ifa_local;
3364 bond_glean_dev_ip(vlan_dev);
3375 static struct notifier_block bond_netdev_notifier = {
3376 .notifier_call = bond_netdev_event,
3379 static struct notifier_block bond_inetaddr_notifier = {
3380 .notifier_call = bond_inetaddr_event,
3383 /*-------------------------- Packet type handling ---------------------------*/
3385 /* register to receive lacpdus on a bond */
3386 static void bond_register_lacpdu(struct bonding *bond)
3388 struct packet_type *pk_type = &(BOND_AD_INFO(bond).ad_pkt_type);
3390 /* initialize packet type */
3391 pk_type->type = PKT_TYPE_LACPDU;
3392 pk_type->dev = bond->dev;
3393 pk_type->func = bond_3ad_lacpdu_recv;
3395 dev_add_pack(pk_type);
3398 /* unregister to receive lacpdus on a bond */
3399 static void bond_unregister_lacpdu(struct bonding *bond)
3401 dev_remove_pack(&(BOND_AD_INFO(bond).ad_pkt_type));
3404 void bond_register_arp(struct bonding *bond)
3406 struct packet_type *pt = &bond->arp_mon_pt;
3411 pt->type = htons(ETH_P_ARP);
3412 pt->dev = bond->dev;
3413 pt->func = bond_arp_rcv;
3417 void bond_unregister_arp(struct bonding *bond)
3419 struct packet_type *pt = &bond->arp_mon_pt;
3421 dev_remove_pack(pt);
3425 /*---------------------------- Hashing Policies -----------------------------*/
3428 * Hash for the output device based upon layer 3 and layer 4 data. If
3429 * the packet is a frag or not TCP or UDP, just use layer 3 data. If it is
3430 * altogether not IP, mimic bond_xmit_hash_policy_l2()
3432 static int bond_xmit_hash_policy_l34(struct sk_buff *skb,
3433 struct net_device *bond_dev, int count)
3435 struct ethhdr *data = (struct ethhdr *)skb->data;
3436 struct iphdr *iph = ip_hdr(skb);
3437 u16 *layer4hdr = (u16 *)((u32 *)iph + iph->ihl);
3440 if (skb->protocol == __constant_htons(ETH_P_IP)) {
3441 if (!(iph->frag_off & __constant_htons(IP_MF|IP_OFFSET)) &&
3442 (iph->protocol == IPPROTO_TCP ||
3443 iph->protocol == IPPROTO_UDP)) {
3444 layer4_xor = htons((*layer4hdr ^ *(layer4hdr + 1)));
3446 return (layer4_xor ^
3447 ((ntohl(iph->saddr ^ iph->daddr)) & 0xffff)) % count;
3451 return (data->h_dest[5] ^ bond_dev->dev_addr[5]) % count;
3455 * Hash for the output device based upon layer 2 data
3457 static int bond_xmit_hash_policy_l2(struct sk_buff *skb,
3458 struct net_device *bond_dev, int count)
3460 struct ethhdr *data = (struct ethhdr *)skb->data;
3462 return (data->h_dest[5] ^ bond_dev->dev_addr[5]) % count;
3465 /*-------------------------- Device entry points ----------------------------*/
3467 static int bond_open(struct net_device *bond_dev)
3469 struct bonding *bond = bond_dev->priv;
3470 struct timer_list *mii_timer = &bond->mii_timer;
3471 struct timer_list *arp_timer = &bond->arp_timer;
3473 bond->kill_timers = 0;
3475 if ((bond->params.mode == BOND_MODE_TLB) ||
3476 (bond->params.mode == BOND_MODE_ALB)) {
3477 struct timer_list *alb_timer = &(BOND_ALB_INFO(bond).alb_timer);
3479 /* bond_alb_initialize must be called before the timer
3482 if (bond_alb_initialize(bond, (bond->params.mode == BOND_MODE_ALB))) {
3483 /* something went wrong - fail the open operation */
3487 init_timer(alb_timer);
3488 alb_timer->expires = jiffies + 1;
3489 alb_timer->data = (unsigned long)bond;
3490 alb_timer->function = (void *)&bond_alb_monitor;
3491 add_timer(alb_timer);
3494 if (bond->params.miimon) { /* link check interval, in milliseconds. */
3495 init_timer(mii_timer);
3496 mii_timer->expires = jiffies + 1;
3497 mii_timer->data = (unsigned long)bond_dev;
3498 mii_timer->function = (void *)&bond_mii_monitor;
3499 add_timer(mii_timer);
3502 if (bond->params.arp_interval) { /* arp interval, in milliseconds. */
3503 init_timer(arp_timer);
3504 arp_timer->expires = jiffies + 1;
3505 arp_timer->data = (unsigned long)bond_dev;
3506 if (bond->params.mode == BOND_MODE_ACTIVEBACKUP) {
3507 arp_timer->function = (void *)&bond_activebackup_arp_mon;
3509 arp_timer->function = (void *)&bond_loadbalance_arp_mon;
3511 if (bond->params.arp_validate)
3512 bond_register_arp(bond);
3514 add_timer(arp_timer);
3517 if (bond->params.mode == BOND_MODE_8023AD) {
3518 struct timer_list *ad_timer = &(BOND_AD_INFO(bond).ad_timer);
3519 init_timer(ad_timer);
3520 ad_timer->expires = jiffies + 1;
3521 ad_timer->data = (unsigned long)bond;
3522 ad_timer->function = (void *)&bond_3ad_state_machine_handler;
3523 add_timer(ad_timer);
3525 /* register to receive LACPDUs */
3526 bond_register_lacpdu(bond);
3532 static int bond_close(struct net_device *bond_dev)
3534 struct bonding *bond = bond_dev->priv;
3536 if (bond->params.mode == BOND_MODE_8023AD) {
3537 /* Unregister the receive of LACPDUs */
3538 bond_unregister_lacpdu(bond);
3541 if (bond->params.arp_validate)
3542 bond_unregister_arp(bond);
3544 write_lock_bh(&bond->lock);
3547 /* signal timers not to re-arm */
3548 bond->kill_timers = 1;
3550 write_unlock_bh(&bond->lock);
3552 /* del_timer_sync must run without holding the bond->lock
3553 * because a running timer might be trying to hold it too
3556 if (bond->params.miimon) { /* link check interval, in milliseconds. */
3557 del_timer_sync(&bond->mii_timer);
3560 if (bond->params.arp_interval) { /* arp interval, in milliseconds. */
3561 del_timer_sync(&bond->arp_timer);
3564 switch (bond->params.mode) {
3565 case BOND_MODE_8023AD:
3566 del_timer_sync(&(BOND_AD_INFO(bond).ad_timer));
3570 del_timer_sync(&(BOND_ALB_INFO(bond).alb_timer));
3577 if ((bond->params.mode == BOND_MODE_TLB) ||
3578 (bond->params.mode == BOND_MODE_ALB)) {
3579 /* Must be called only after all
3580 * slaves have been released
3582 bond_alb_deinitialize(bond);
3588 static struct net_device_stats *bond_get_stats(struct net_device *bond_dev)
3590 struct bonding *bond = bond_dev->priv;
3591 struct net_device_stats *stats = &(bond->stats), *sstats;
3592 struct slave *slave;
3595 memset(stats, 0, sizeof(struct net_device_stats));
3597 read_lock_bh(&bond->lock);
3599 bond_for_each_slave(bond, slave, i) {
3600 sstats = slave->dev->get_stats(slave->dev);
3601 stats->rx_packets += sstats->rx_packets;
3602 stats->rx_bytes += sstats->rx_bytes;
3603 stats->rx_errors += sstats->rx_errors;
3604 stats->rx_dropped += sstats->rx_dropped;
3606 stats->tx_packets += sstats->tx_packets;
3607 stats->tx_bytes += sstats->tx_bytes;
3608 stats->tx_errors += sstats->tx_errors;
3609 stats->tx_dropped += sstats->tx_dropped;
3611 stats->multicast += sstats->multicast;
3612 stats->collisions += sstats->collisions;
3614 stats->rx_length_errors += sstats->rx_length_errors;
3615 stats->rx_over_errors += sstats->rx_over_errors;
3616 stats->rx_crc_errors += sstats->rx_crc_errors;
3617 stats->rx_frame_errors += sstats->rx_frame_errors;
3618 stats->rx_fifo_errors += sstats->rx_fifo_errors;
3619 stats->rx_missed_errors += sstats->rx_missed_errors;
3621 stats->tx_aborted_errors += sstats->tx_aborted_errors;
3622 stats->tx_carrier_errors += sstats->tx_carrier_errors;
3623 stats->tx_fifo_errors += sstats->tx_fifo_errors;
3624 stats->tx_heartbeat_errors += sstats->tx_heartbeat_errors;
3625 stats->tx_window_errors += sstats->tx_window_errors;
3628 read_unlock_bh(&bond->lock);
3633 static int bond_do_ioctl(struct net_device *bond_dev, struct ifreq *ifr, int cmd)
3635 struct net_device *slave_dev = NULL;
3636 struct ifbond k_binfo;
3637 struct ifbond __user *u_binfo = NULL;
3638 struct ifslave k_sinfo;
3639 struct ifslave __user *u_sinfo = NULL;
3640 struct mii_ioctl_data *mii = NULL;
3643 dprintk("bond_ioctl: master=%s, cmd=%d\n",
3644 bond_dev->name, cmd);
3656 * We do this again just in case we were called by SIOCGMIIREG
3657 * instead of SIOCGMIIPHY.
3664 if (mii->reg_num == 1) {
3665 struct bonding *bond = bond_dev->priv;
3667 read_lock_bh(&bond->lock);
3668 read_lock(&bond->curr_slave_lock);
3669 if (netif_carrier_ok(bond->dev)) {
3670 mii->val_out = BMSR_LSTATUS;
3672 read_unlock(&bond->curr_slave_lock);
3673 read_unlock_bh(&bond->lock);
3677 case BOND_INFO_QUERY_OLD:
3678 case SIOCBONDINFOQUERY:
3679 u_binfo = (struct ifbond __user *)ifr->ifr_data;
3681 if (copy_from_user(&k_binfo, u_binfo, sizeof(ifbond))) {
3685 res = bond_info_query(bond_dev, &k_binfo);
3687 if (copy_to_user(u_binfo, &k_binfo, sizeof(ifbond))) {
3693 case BOND_SLAVE_INFO_QUERY_OLD:
3694 case SIOCBONDSLAVEINFOQUERY:
3695 u_sinfo = (struct ifslave __user *)ifr->ifr_data;
3697 if (copy_from_user(&k_sinfo, u_sinfo, sizeof(ifslave))) {
3701 res = bond_slave_info_query(bond_dev, &k_sinfo);
3703 if (copy_to_user(u_sinfo, &k_sinfo, sizeof(ifslave))) {
3714 if (!capable(CAP_NET_ADMIN)) {
3718 down_write(&(bonding_rwsem));
3719 slave_dev = dev_get_by_name(ifr->ifr_slave);
3721 dprintk("slave_dev=%p: \n", slave_dev);
3726 dprintk("slave_dev->name=%s: \n", slave_dev->name);
3728 case BOND_ENSLAVE_OLD:
3729 case SIOCBONDENSLAVE:
3730 res = bond_enslave(bond_dev, slave_dev);
3732 case BOND_RELEASE_OLD:
3733 case SIOCBONDRELEASE:
3734 res = bond_release(bond_dev, slave_dev);
3736 case BOND_SETHWADDR_OLD:
3737 case SIOCBONDSETHWADDR:
3738 res = bond_sethwaddr(bond_dev, slave_dev);
3740 case BOND_CHANGE_ACTIVE_OLD:
3741 case SIOCBONDCHANGEACTIVE:
3742 res = bond_ioctl_change_active(bond_dev, slave_dev);
3751 up_write(&(bonding_rwsem));
3755 static void bond_set_multicast_list(struct net_device *bond_dev)
3757 struct bonding *bond = bond_dev->priv;
3758 struct dev_mc_list *dmi;
3760 write_lock_bh(&bond->lock);
3763 * Do promisc before checking multicast_mode
3765 if ((bond_dev->flags & IFF_PROMISC) && !(bond->flags & IFF_PROMISC)) {
3766 bond_set_promiscuity(bond, 1);
3769 if (!(bond_dev->flags & IFF_PROMISC) && (bond->flags & IFF_PROMISC)) {
3770 bond_set_promiscuity(bond, -1);
3773 /* set allmulti flag to slaves */
3774 if ((bond_dev->flags & IFF_ALLMULTI) && !(bond->flags & IFF_ALLMULTI)) {
3775 bond_set_allmulti(bond, 1);
3778 if (!(bond_dev->flags & IFF_ALLMULTI) && (bond->flags & IFF_ALLMULTI)) {
3779 bond_set_allmulti(bond, -1);
3782 bond->flags = bond_dev->flags;
3784 /* looking for addresses to add to slaves' mc list */
3785 for (dmi = bond_dev->mc_list; dmi; dmi = dmi->next) {
3786 if (!bond_mc_list_find_dmi(dmi, bond->mc_list)) {
3787 bond_mc_add(bond, dmi->dmi_addr, dmi->dmi_addrlen);
3791 /* looking for addresses to delete from slaves' list */
3792 for (dmi = bond->mc_list; dmi; dmi = dmi->next) {
3793 if (!bond_mc_list_find_dmi(dmi, bond_dev->mc_list)) {
3794 bond_mc_delete(bond, dmi->dmi_addr, dmi->dmi_addrlen);
3798 /* save master's multicast list */
3799 bond_mc_list_destroy(bond);
3800 bond_mc_list_copy(bond_dev->mc_list, bond, GFP_ATOMIC);
3802 write_unlock_bh(&bond->lock);
3806 * Change the MTU of all of a master's slaves to match the master
3808 static int bond_change_mtu(struct net_device *bond_dev, int new_mtu)
3810 struct bonding *bond = bond_dev->priv;
3811 struct slave *slave, *stop_at;
3815 dprintk("bond=%p, name=%s, new_mtu=%d\n", bond,
3816 (bond_dev ? bond_dev->name : "None"), new_mtu);
3818 /* Can't hold bond->lock with bh disabled here since
3819 * some base drivers panic. On the other hand we can't
3820 * hold bond->lock without bh disabled because we'll
3821 * deadlock. The only solution is to rely on the fact
3822 * that we're under rtnl_lock here, and the slaves
3823 * list won't change. This doesn't solve the problem
3824 * of setting the slave's MTU while it is
3825 * transmitting, but the assumption is that the base
3826 * driver can handle that.
3828 * TODO: figure out a way to safely iterate the slaves
3829 * list, but without holding a lock around the actual
3830 * call to the base driver.
3833 bond_for_each_slave(bond, slave, i) {
3834 dprintk("s %p s->p %p c_m %p\n", slave,
3835 slave->prev, slave->dev->change_mtu);
3837 res = dev_set_mtu(slave->dev, new_mtu);
3840 /* If we failed to set the slave's mtu to the new value
3841 * we must abort the operation even in ACTIVE_BACKUP
3842 * mode, because if we allow the backup slaves to have
3843 * different mtu values than the active slave we'll
3844 * need to change their mtu when doing a failover. That
3845 * means changing their mtu from timer context, which
3846 * is probably not a good idea.
3848 dprintk("err %d %s\n", res, slave->dev->name);
3853 bond_dev->mtu = new_mtu;
3858 /* unwind from head to the slave that failed */
3860 bond_for_each_slave_from_to(bond, slave, i, bond->first_slave, stop_at) {
3863 tmp_res = dev_set_mtu(slave->dev, bond_dev->mtu);
3865 dprintk("unwind err %d dev %s\n", tmp_res,
3876 * Note that many devices must be down to change the HW address, and
3877 * downing the master releases all slaves. We can make bonds full of
3878 * bonding devices to test this, however.
3880 static int bond_set_mac_address(struct net_device *bond_dev, void *addr)
3882 struct bonding *bond = bond_dev->priv;
3883 struct sockaddr *sa = addr, tmp_sa;
3884 struct slave *slave, *stop_at;
3888 dprintk("bond=%p, name=%s\n", bond, (bond_dev ? bond_dev->name : "None"));
3890 if (!is_valid_ether_addr(sa->sa_data)) {
3891 return -EADDRNOTAVAIL;
3894 /* Can't hold bond->lock with bh disabled here since
3895 * some base drivers panic. On the other hand we can't
3896 * hold bond->lock without bh disabled because we'll
3897 * deadlock. The only solution is to rely on the fact
3898 * that we're under rtnl_lock here, and the slaves
3899 * list won't change. This doesn't solve the problem
3900 * of setting the slave's hw address while it is
3901 * transmitting, but the assumption is that the base
3902 * driver can handle that.
3904 * TODO: figure out a way to safely iterate the slaves
3905 * list, but without holding a lock around the actual
3906 * call to the base driver.
3909 bond_for_each_slave(bond, slave, i) {
3910 dprintk("slave %p %s\n", slave, slave->dev->name);
3912 if (slave->dev->set_mac_address == NULL) {
3914 dprintk("EOPNOTSUPP %s\n", slave->dev->name);
3918 res = dev_set_mac_address(slave->dev, addr);
3920 /* TODO: consider downing the slave
3922 * User should expect communications
3923 * breakage anyway until ARP finish
3926 dprintk("err %d %s\n", res, slave->dev->name);
3932 memcpy(bond_dev->dev_addr, sa->sa_data, bond_dev->addr_len);
3936 memcpy(tmp_sa.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
3937 tmp_sa.sa_family = bond_dev->type;
3939 /* unwind from head to the slave that failed */
3941 bond_for_each_slave_from_to(bond, slave, i, bond->first_slave, stop_at) {
3944 tmp_res = dev_set_mac_address(slave->dev, &tmp_sa);
3946 dprintk("unwind err %d dev %s\n", tmp_res,
3954 static int bond_xmit_roundrobin(struct sk_buff *skb, struct net_device *bond_dev)
3956 struct bonding *bond = bond_dev->priv;
3957 struct slave *slave, *start_at;
3961 read_lock(&bond->lock);
3963 if (!BOND_IS_OK(bond)) {
3967 read_lock(&bond->curr_slave_lock);
3968 slave = start_at = bond->curr_active_slave;
3969 read_unlock(&bond->curr_slave_lock);
3975 bond_for_each_slave_from(bond, slave, i, start_at) {
3976 if (IS_UP(slave->dev) &&
3977 (slave->link == BOND_LINK_UP) &&
3978 (slave->state == BOND_STATE_ACTIVE)) {
3979 res = bond_dev_queue_xmit(bond, skb, slave->dev);
3981 write_lock(&bond->curr_slave_lock);
3982 bond->curr_active_slave = slave->next;
3983 write_unlock(&bond->curr_slave_lock);
3992 /* no suitable interface, frame not sent */
3995 read_unlock(&bond->lock);
4001 * in active-backup mode, we know that bond->curr_active_slave is always valid if
4002 * the bond has a usable interface.
4004 static int bond_xmit_activebackup(struct sk_buff *skb, struct net_device *bond_dev)
4006 struct bonding *bond = bond_dev->priv;
4009 read_lock(&bond->lock);
4010 read_lock(&bond->curr_slave_lock);
4012 if (!BOND_IS_OK(bond)) {
4016 if (!bond->curr_active_slave)
4019 res = bond_dev_queue_xmit(bond, skb, bond->curr_active_slave->dev);
4023 /* no suitable interface, frame not sent */
4026 read_unlock(&bond->curr_slave_lock);
4027 read_unlock(&bond->lock);
4032 * In bond_xmit_xor() , we determine the output device by using a pre-
4033 * determined xmit_hash_policy(), If the selected device is not enabled,
4034 * find the next active slave.
4036 static int bond_xmit_xor(struct sk_buff *skb, struct net_device *bond_dev)
4038 struct bonding *bond = bond_dev->priv;
4039 struct slave *slave, *start_at;
4044 read_lock(&bond->lock);
4046 if (!BOND_IS_OK(bond)) {
4050 slave_no = bond->xmit_hash_policy(skb, bond_dev, bond->slave_cnt);
4052 bond_for_each_slave(bond, slave, i) {
4061 bond_for_each_slave_from(bond, slave, i, start_at) {
4062 if (IS_UP(slave->dev) &&
4063 (slave->link == BOND_LINK_UP) &&
4064 (slave->state == BOND_STATE_ACTIVE)) {
4065 res = bond_dev_queue_xmit(bond, skb, slave->dev);
4072 /* no suitable interface, frame not sent */
4075 read_unlock(&bond->lock);
4080 * in broadcast mode, we send everything to all usable interfaces.
4082 static int bond_xmit_broadcast(struct sk_buff *skb, struct net_device *bond_dev)
4084 struct bonding *bond = bond_dev->priv;
4085 struct slave *slave, *start_at;
4086 struct net_device *tx_dev = NULL;
4090 read_lock(&bond->lock);
4092 if (!BOND_IS_OK(bond)) {
4096 read_lock(&bond->curr_slave_lock);
4097 start_at = bond->curr_active_slave;
4098 read_unlock(&bond->curr_slave_lock);
4104 bond_for_each_slave_from(bond, slave, i, start_at) {
4105 if (IS_UP(slave->dev) &&
4106 (slave->link == BOND_LINK_UP) &&
4107 (slave->state == BOND_STATE_ACTIVE)) {
4109 struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
4111 printk(KERN_ERR DRV_NAME
4112 ": %s: Error: bond_xmit_broadcast(): "
4113 "skb_clone() failed\n",
4118 res = bond_dev_queue_xmit(bond, skb2, tx_dev);
4120 dev_kfree_skb(skb2);
4124 tx_dev = slave->dev;
4129 res = bond_dev_queue_xmit(bond, skb, tx_dev);
4134 /* no suitable interface, frame not sent */
4137 /* frame sent to all suitable interfaces */
4138 read_unlock(&bond->lock);
4142 /*------------------------- Device initialization ---------------------------*/
4145 * set bond mode specific net device operations
4147 void bond_set_mode_ops(struct bonding *bond, int mode)
4149 struct net_device *bond_dev = bond->dev;
4152 case BOND_MODE_ROUNDROBIN:
4153 bond_dev->hard_start_xmit = bond_xmit_roundrobin;
4155 case BOND_MODE_ACTIVEBACKUP:
4156 bond_dev->hard_start_xmit = bond_xmit_activebackup;
4159 bond_dev->hard_start_xmit = bond_xmit_xor;
4160 if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER34)
4161 bond->xmit_hash_policy = bond_xmit_hash_policy_l34;
4163 bond->xmit_hash_policy = bond_xmit_hash_policy_l2;
4165 case BOND_MODE_BROADCAST:
4166 bond_dev->hard_start_xmit = bond_xmit_broadcast;
4168 case BOND_MODE_8023AD:
4169 bond_set_master_3ad_flags(bond);
4170 bond_dev->hard_start_xmit = bond_3ad_xmit_xor;
4171 if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER34)
4172 bond->xmit_hash_policy = bond_xmit_hash_policy_l34;
4174 bond->xmit_hash_policy = bond_xmit_hash_policy_l2;
4177 bond_set_master_alb_flags(bond);
4180 bond_dev->hard_start_xmit = bond_alb_xmit;
4181 bond_dev->set_mac_address = bond_alb_set_mac_address;
4184 /* Should never happen, mode already checked */
4185 printk(KERN_ERR DRV_NAME
4186 ": %s: Error: Unknown bonding mode %d\n",
4193 static void bond_ethtool_get_drvinfo(struct net_device *bond_dev,
4194 struct ethtool_drvinfo *drvinfo)
4196 strncpy(drvinfo->driver, DRV_NAME, 32);
4197 strncpy(drvinfo->version, DRV_VERSION, 32);
4198 snprintf(drvinfo->fw_version, 32, "%d", BOND_ABI_VERSION);
4201 static const struct ethtool_ops bond_ethtool_ops = {
4202 .get_tx_csum = ethtool_op_get_tx_csum,
4203 .get_tso = ethtool_op_get_tso,
4204 .get_ufo = ethtool_op_get_ufo,
4205 .get_sg = ethtool_op_get_sg,
4206 .get_drvinfo = bond_ethtool_get_drvinfo,
4210 * Does not allocate but creates a /proc entry.
4213 static int bond_init(struct net_device *bond_dev, struct bond_params *params)
4215 struct bonding *bond = bond_dev->priv;
4217 dprintk("Begin bond_init for %s\n", bond_dev->name);
4219 /* initialize rwlocks */
4220 rwlock_init(&bond->lock);
4221 rwlock_init(&bond->curr_slave_lock);
4223 bond->params = *params; /* copy params struct */
4225 /* Initialize pointers */
4226 bond->first_slave = NULL;
4227 bond->curr_active_slave = NULL;
4228 bond->current_arp_slave = NULL;
4229 bond->primary_slave = NULL;
4230 bond->dev = bond_dev;
4231 INIT_LIST_HEAD(&bond->vlan_list);
4233 /* Initialize the device entry points */
4234 bond_dev->open = bond_open;
4235 bond_dev->stop = bond_close;
4236 bond_dev->get_stats = bond_get_stats;
4237 bond_dev->do_ioctl = bond_do_ioctl;
4238 bond_dev->ethtool_ops = &bond_ethtool_ops;
4239 bond_dev->set_multicast_list = bond_set_multicast_list;
4240 bond_dev->change_mtu = bond_change_mtu;
4241 bond_dev->set_mac_address = bond_set_mac_address;
4243 bond_set_mode_ops(bond, bond->params.mode);
4245 bond_dev->destructor = free_netdev;
4247 /* Initialize the device options */
4248 bond_dev->tx_queue_len = 0;
4249 bond_dev->flags |= IFF_MASTER|IFF_MULTICAST;
4250 bond_dev->priv_flags |= IFF_BONDING;
4252 /* At first, we block adding VLANs. That's the only way to
4253 * prevent problems that occur when adding VLANs over an
4254 * empty bond. The block will be removed once non-challenged
4255 * slaves are enslaved.
4257 bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
4259 /* don't acquire bond device's netif_tx_lock when
4261 bond_dev->features |= NETIF_F_LLTX;
4263 /* By default, we declare the bond to be fully
4264 * VLAN hardware accelerated capable. Special
4265 * care is taken in the various xmit functions
4266 * when there are slaves that are not hw accel
4269 bond_dev->vlan_rx_register = bond_vlan_rx_register;
4270 bond_dev->vlan_rx_add_vid = bond_vlan_rx_add_vid;
4271 bond_dev->vlan_rx_kill_vid = bond_vlan_rx_kill_vid;
4272 bond_dev->features |= (NETIF_F_HW_VLAN_TX |
4273 NETIF_F_HW_VLAN_RX |
4274 NETIF_F_HW_VLAN_FILTER);
4276 #ifdef CONFIG_PROC_FS
4277 bond_create_proc_entry(bond);
4280 list_add_tail(&bond->bond_list, &bond_dev_list);
4285 /* De-initialize device specific data.
4286 * Caller must hold rtnl_lock.
4288 void bond_deinit(struct net_device *bond_dev)
4290 struct bonding *bond = bond_dev->priv;
4292 list_del(&bond->bond_list);
4294 #ifdef CONFIG_PROC_FS
4295 bond_remove_proc_entry(bond);
4299 /* Unregister and free all bond devices.
4300 * Caller must hold rtnl_lock.
4302 static void bond_free_all(void)
4304 struct bonding *bond, *nxt;
4306 list_for_each_entry_safe(bond, nxt, &bond_dev_list, bond_list) {
4307 struct net_device *bond_dev = bond->dev;
4309 bond_mc_list_destroy(bond);
4310 /* Release the bonded slaves */
4311 bond_release_all(bond_dev);
4312 bond_deinit(bond_dev);
4313 unregister_netdevice(bond_dev);
4316 #ifdef CONFIG_PROC_FS
4317 bond_destroy_proc_dir();
4321 /*------------------------- Module initialization ---------------------------*/
4324 * Convert string input module parms. Accept either the
4325 * number of the mode or its string name.
4327 int bond_parse_parm(char *mode_arg, struct bond_parm_tbl *tbl)
4331 for (i = 0; tbl[i].modename; i++) {
4332 if ((isdigit(*mode_arg) &&
4333 tbl[i].mode == simple_strtol(mode_arg, NULL, 0)) ||
4334 (strncmp(mode_arg, tbl[i].modename,
4335 strlen(tbl[i].modename)) == 0)) {
4343 static int bond_check_params(struct bond_params *params)
4345 int arp_validate_value;
4348 * Convert string parameters.
4351 bond_mode = bond_parse_parm(mode, bond_mode_tbl);
4352 if (bond_mode == -1) {
4353 printk(KERN_ERR DRV_NAME
4354 ": Error: Invalid bonding mode \"%s\"\n",
4355 mode == NULL ? "NULL" : mode);
4360 if (xmit_hash_policy) {
4361 if ((bond_mode != BOND_MODE_XOR) &&
4362 (bond_mode != BOND_MODE_8023AD)) {
4363 printk(KERN_INFO DRV_NAME
4364 ": xor_mode param is irrelevant in mode %s\n",
4365 bond_mode_name(bond_mode));
4367 xmit_hashtype = bond_parse_parm(xmit_hash_policy,
4369 if (xmit_hashtype == -1) {
4370 printk(KERN_ERR DRV_NAME
4371 ": Error: Invalid xmit_hash_policy \"%s\"\n",
4372 xmit_hash_policy == NULL ? "NULL" :
4380 if (bond_mode != BOND_MODE_8023AD) {
4381 printk(KERN_INFO DRV_NAME
4382 ": lacp_rate param is irrelevant in mode %s\n",
4383 bond_mode_name(bond_mode));
4385 lacp_fast = bond_parse_parm(lacp_rate, bond_lacp_tbl);
4386 if (lacp_fast == -1) {
4387 printk(KERN_ERR DRV_NAME
4388 ": Error: Invalid lacp rate \"%s\"\n",
4389 lacp_rate == NULL ? "NULL" : lacp_rate);
4395 if (max_bonds < 1 || max_bonds > INT_MAX) {
4396 printk(KERN_WARNING DRV_NAME
4397 ": Warning: max_bonds (%d) not in range %d-%d, so it "
4398 "was reset to BOND_DEFAULT_MAX_BONDS (%d)\n",
4399 max_bonds, 1, INT_MAX, BOND_DEFAULT_MAX_BONDS);
4400 max_bonds = BOND_DEFAULT_MAX_BONDS;
4404 printk(KERN_WARNING DRV_NAME
4405 ": Warning: miimon module parameter (%d), "
4406 "not in range 0-%d, so it was reset to %d\n",
4407 miimon, INT_MAX, BOND_LINK_MON_INTERV);
4408 miimon = BOND_LINK_MON_INTERV;
4412 printk(KERN_WARNING DRV_NAME
4413 ": Warning: updelay module parameter (%d), "
4414 "not in range 0-%d, so it was reset to 0\n",
4419 if (downdelay < 0) {
4420 printk(KERN_WARNING DRV_NAME
4421 ": Warning: downdelay module parameter (%d), "
4422 "not in range 0-%d, so it was reset to 0\n",
4423 downdelay, INT_MAX);
4427 if ((use_carrier != 0) && (use_carrier != 1)) {
4428 printk(KERN_WARNING DRV_NAME
4429 ": Warning: use_carrier module parameter (%d), "
4430 "not of valid value (0/1), so it was set to 1\n",
4435 /* reset values for 802.3ad */
4436 if (bond_mode == BOND_MODE_8023AD) {
4438 printk(KERN_WARNING DRV_NAME
4439 ": Warning: miimon must be specified, "
4440 "otherwise bonding will not detect link "
4441 "failure, speed and duplex which are "
4442 "essential for 802.3ad operation\n");
4443 printk(KERN_WARNING "Forcing miimon to 100msec\n");
4448 /* reset values for TLB/ALB */
4449 if ((bond_mode == BOND_MODE_TLB) ||
4450 (bond_mode == BOND_MODE_ALB)) {
4452 printk(KERN_WARNING DRV_NAME
4453 ": Warning: miimon must be specified, "
4454 "otherwise bonding will not detect link "
4455 "failure and link speed which are essential "
4456 "for TLB/ALB load balancing\n");
4457 printk(KERN_WARNING "Forcing miimon to 100msec\n");
4462 if (bond_mode == BOND_MODE_ALB) {
4463 printk(KERN_NOTICE DRV_NAME
4464 ": In ALB mode you might experience client "
4465 "disconnections upon reconnection of a link if the "
4466 "bonding module updelay parameter (%d msec) is "
4467 "incompatible with the forwarding delay time of the "
4473 if (updelay || downdelay) {
4474 /* just warn the user the up/down delay will have
4475 * no effect since miimon is zero...
4477 printk(KERN_WARNING DRV_NAME
4478 ": Warning: miimon module parameter not set "
4479 "and updelay (%d) or downdelay (%d) module "
4480 "parameter is set; updelay and downdelay have "
4481 "no effect unless miimon is set\n",
4482 updelay, downdelay);
4485 /* don't allow arp monitoring */
4487 printk(KERN_WARNING DRV_NAME
4488 ": Warning: miimon (%d) and arp_interval (%d) "
4489 "can't be used simultaneously, disabling ARP "
4491 miimon, arp_interval);
4495 if ((updelay % miimon) != 0) {
4496 printk(KERN_WARNING DRV_NAME
4497 ": Warning: updelay (%d) is not a multiple "
4498 "of miimon (%d), updelay rounded to %d ms\n",
4499 updelay, miimon, (updelay / miimon) * miimon);
4504 if ((downdelay % miimon) != 0) {
4505 printk(KERN_WARNING DRV_NAME
4506 ": Warning: downdelay (%d) is not a multiple "
4507 "of miimon (%d), downdelay rounded to %d ms\n",
4509 (downdelay / miimon) * miimon);
4512 downdelay /= miimon;
4515 if (arp_interval < 0) {
4516 printk(KERN_WARNING DRV_NAME
4517 ": Warning: arp_interval module parameter (%d) "
4518 ", not in range 0-%d, so it was reset to %d\n",
4519 arp_interval, INT_MAX, BOND_LINK_ARP_INTERV);
4520 arp_interval = BOND_LINK_ARP_INTERV;
4523 for (arp_ip_count = 0;
4524 (arp_ip_count < BOND_MAX_ARP_TARGETS) && arp_ip_target[arp_ip_count];
4526 /* not complete check, but should be good enough to
4528 if (!isdigit(arp_ip_target[arp_ip_count][0])) {
4529 printk(KERN_WARNING DRV_NAME
4530 ": Warning: bad arp_ip_target module parameter "
4531 "(%s), ARP monitoring will not be performed\n",
4532 arp_ip_target[arp_ip_count]);
4535 u32 ip = in_aton(arp_ip_target[arp_ip_count]);
4536 arp_target[arp_ip_count] = ip;
4540 if (arp_interval && !arp_ip_count) {
4541 /* don't allow arping if no arp_ip_target given... */
4542 printk(KERN_WARNING DRV_NAME
4543 ": Warning: arp_interval module parameter (%d) "
4544 "specified without providing an arp_ip_target "
4545 "parameter, arp_interval was reset to 0\n",
4551 if (bond_mode != BOND_MODE_ACTIVEBACKUP) {
4552 printk(KERN_ERR DRV_NAME
4553 ": arp_validate only supported in active-backup mode\n");
4556 if (!arp_interval) {
4557 printk(KERN_ERR DRV_NAME
4558 ": arp_validate requires arp_interval\n");
4562 arp_validate_value = bond_parse_parm(arp_validate,
4564 if (arp_validate_value == -1) {
4565 printk(KERN_ERR DRV_NAME
4566 ": Error: invalid arp_validate \"%s\"\n",
4567 arp_validate == NULL ? "NULL" : arp_validate);
4571 arp_validate_value = 0;
4574 printk(KERN_INFO DRV_NAME
4575 ": MII link monitoring set to %d ms\n",
4577 } else if (arp_interval) {
4580 printk(KERN_INFO DRV_NAME
4581 ": ARP monitoring set to %d ms, validate %s, with %d target(s):",
4583 arp_validate_tbl[arp_validate_value].modename,
4586 for (i = 0; i < arp_ip_count; i++)
4587 printk (" %s", arp_ip_target[i]);
4592 /* miimon and arp_interval not set, we need one so things
4593 * work as expected, see bonding.txt for details
4595 printk(KERN_WARNING DRV_NAME
4596 ": Warning: either miimon or arp_interval and "
4597 "arp_ip_target module parameters must be specified, "
4598 "otherwise bonding will not detect link failures! see "
4599 "bonding.txt for details.\n");
4602 if (primary && !USES_PRIMARY(bond_mode)) {
4603 /* currently, using a primary only makes sense
4604 * in active backup, TLB or ALB modes
4606 printk(KERN_WARNING DRV_NAME
4607 ": Warning: %s primary device specified but has no "
4608 "effect in %s mode\n",
4609 primary, bond_mode_name(bond_mode));
4613 /* fill params struct with the proper values */
4614 params->mode = bond_mode;
4615 params->xmit_policy = xmit_hashtype;
4616 params->miimon = miimon;
4617 params->arp_interval = arp_interval;
4618 params->arp_validate = arp_validate_value;
4619 params->updelay = updelay;
4620 params->downdelay = downdelay;
4621 params->use_carrier = use_carrier;
4622 params->lacp_fast = lacp_fast;
4623 params->primary[0] = 0;
4626 strncpy(params->primary, primary, IFNAMSIZ);
4627 params->primary[IFNAMSIZ - 1] = 0;
4630 memcpy(params->arp_targets, arp_target, sizeof(arp_target));
4635 static struct lock_class_key bonding_netdev_xmit_lock_key;
4637 /* Create a new bond based on the specified name and bonding parameters.
4638 * If name is NULL, obtain a suitable "bond%d" name for us.
4639 * Caller must NOT hold rtnl_lock; we need to release it here before we
4640 * set up our sysfs entries.
4642 int bond_create(char *name, struct bond_params *params, struct bonding **newbond)
4644 struct net_device *bond_dev;
4648 bond_dev = alloc_netdev(sizeof(struct bonding), name ? name : "",
4651 printk(KERN_ERR DRV_NAME
4652 ": %s: eek! can't alloc netdev!\n",
4659 res = dev_alloc_name(bond_dev, "bond%d");
4664 /* bond_init() must be called after dev_alloc_name() (for the
4665 * /proc files), but before register_netdevice(), because we
4666 * need to set function pointers.
4669 res = bond_init(bond_dev, params);
4674 SET_MODULE_OWNER(bond_dev);
4676 res = register_netdevice(bond_dev);
4681 lockdep_set_class(&bond_dev->_xmit_lock, &bonding_netdev_xmit_lock_key);
4684 *newbond = bond_dev->priv;
4686 netif_carrier_off(bond_dev);
4688 rtnl_unlock(); /* allows sysfs registration of net device */
4689 res = bond_create_sysfs_entry(bond_dev->priv);
4698 bond_deinit(bond_dev);
4700 free_netdev(bond_dev);
4706 static int __init bonding_init(void)
4711 printk(KERN_INFO "%s", version);
4713 res = bond_check_params(&bonding_defaults);
4718 #ifdef CONFIG_PROC_FS
4719 bond_create_proc_dir();
4721 for (i = 0; i < max_bonds; i++) {
4722 res = bond_create(NULL, &bonding_defaults, NULL);
4727 res = bond_create_sysfs();
4731 register_netdevice_notifier(&bond_netdev_notifier);
4732 register_inetaddr_notifier(&bond_inetaddr_notifier);
4738 bond_destroy_sysfs();
4745 static void __exit bonding_exit(void)
4747 unregister_netdevice_notifier(&bond_netdev_notifier);
4748 unregister_inetaddr_notifier(&bond_inetaddr_notifier);
4752 bond_destroy_sysfs();
4756 module_init(bonding_init);
4757 module_exit(bonding_exit);
4758 MODULE_LICENSE("GPL");
4759 MODULE_VERSION(DRV_VERSION);
4760 MODULE_DESCRIPTION(DRV_DESCRIPTION ", v" DRV_VERSION);
4761 MODULE_AUTHOR("Thomas Davis, tadavis@lbl.gov and many others");
4762 MODULE_SUPPORTED_DEVICE("most ethernet devices");