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 __be32 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);
1100 /* when bonding does not set the slave MAC address, the bond MAC
1101 * address is the one of the active slave.
1103 if (new_active && !bond->do_set_mac_addr)
1104 memcpy(bond->dev->dev_addr, new_active->dev->dev_addr,
1105 new_active->dev->addr_len);
1107 bond_send_gratuitous_arp(bond);
1112 * bond_select_active_slave - select a new active slave, if needed
1113 * @bond: our bonding struct
1115 * This functions shoud be called when one of the following occurs:
1116 * - The old curr_active_slave has been released or lost its link.
1117 * - The primary_slave has got its link back.
1118 * - A slave has got its link back and there's no old curr_active_slave.
1120 * Warning: Caller must hold curr_slave_lock for writing.
1122 void bond_select_active_slave(struct bonding *bond)
1124 struct slave *best_slave;
1127 best_slave = bond_find_best_slave(bond);
1128 if (best_slave != bond->curr_active_slave) {
1129 bond_change_active_slave(bond, best_slave);
1130 rv = bond_set_carrier(bond);
1134 if (netif_carrier_ok(bond->dev)) {
1135 printk(KERN_INFO DRV_NAME
1136 ": %s: first active interface up!\n",
1139 printk(KERN_INFO DRV_NAME ": %s: "
1140 "now running without any active interface !\n",
1146 /*--------------------------- slave list handling ---------------------------*/
1149 * This function attaches the slave to the end of list.
1151 * bond->lock held for writing by caller.
1153 static void bond_attach_slave(struct bonding *bond, struct slave *new_slave)
1155 if (bond->first_slave == NULL) { /* attaching the first slave */
1156 new_slave->next = new_slave;
1157 new_slave->prev = new_slave;
1158 bond->first_slave = new_slave;
1160 new_slave->next = bond->first_slave;
1161 new_slave->prev = bond->first_slave->prev;
1162 new_slave->next->prev = new_slave;
1163 new_slave->prev->next = new_slave;
1170 * This function detaches the slave from the list.
1171 * WARNING: no check is made to verify if the slave effectively
1172 * belongs to <bond>.
1173 * Nothing is freed on return, structures are just unchained.
1174 * If any slave pointer in bond was pointing to <slave>,
1175 * it should be changed by the calling function.
1177 * bond->lock held for writing by caller.
1179 static void bond_detach_slave(struct bonding *bond, struct slave *slave)
1182 slave->next->prev = slave->prev;
1186 slave->prev->next = slave->next;
1189 if (bond->first_slave == slave) { /* slave is the first slave */
1190 if (bond->slave_cnt > 1) { /* there are more slave */
1191 bond->first_slave = slave->next;
1193 bond->first_slave = NULL; /* slave was the last one */
1202 /*---------------------------------- IOCTL ----------------------------------*/
1204 static int bond_sethwaddr(struct net_device *bond_dev,
1205 struct net_device *slave_dev)
1207 dprintk("bond_dev=%p\n", bond_dev);
1208 dprintk("slave_dev=%p\n", slave_dev);
1209 dprintk("slave_dev->addr_len=%d\n", slave_dev->addr_len);
1210 memcpy(bond_dev->dev_addr, slave_dev->dev_addr, slave_dev->addr_len);
1214 #define BOND_VLAN_FEATURES \
1215 (NETIF_F_VLAN_CHALLENGED | NETIF_F_HW_VLAN_RX | NETIF_F_HW_VLAN_TX | \
1216 NETIF_F_HW_VLAN_FILTER)
1219 * Compute the common dev->feature set available to all slaves. Some
1220 * feature bits are managed elsewhere, so preserve those feature bits
1221 * on the master device.
1223 static int bond_compute_features(struct bonding *bond)
1225 struct slave *slave;
1226 struct net_device *bond_dev = bond->dev;
1227 unsigned long features = bond_dev->features;
1228 unsigned short max_hard_header_len = ETH_HLEN;
1231 features &= ~(NETIF_F_ALL_CSUM | BOND_VLAN_FEATURES);
1232 features |= NETIF_F_SG | NETIF_F_FRAGLIST | NETIF_F_HIGHDMA |
1233 NETIF_F_GSO_MASK | NETIF_F_NO_CSUM;
1235 bond_for_each_slave(bond, slave, i) {
1236 features = netdev_compute_features(features,
1237 slave->dev->features);
1238 if (slave->dev->hard_header_len > max_hard_header_len)
1239 max_hard_header_len = slave->dev->hard_header_len;
1242 features |= (bond_dev->features & BOND_VLAN_FEATURES);
1243 bond_dev->features = features;
1244 bond_dev->hard_header_len = max_hard_header_len;
1250 static void bond_setup_by_slave(struct net_device *bond_dev,
1251 struct net_device *slave_dev)
1253 bond_dev->neigh_setup = slave_dev->neigh_setup;
1255 bond_dev->type = slave_dev->type;
1256 bond_dev->hard_header_len = slave_dev->hard_header_len;
1257 bond_dev->addr_len = slave_dev->addr_len;
1259 memcpy(bond_dev->broadcast, slave_dev->broadcast,
1260 slave_dev->addr_len);
1263 /* enslave device <slave> to bond device <master> */
1264 int bond_enslave(struct net_device *bond_dev, struct net_device *slave_dev)
1266 struct bonding *bond = bond_dev->priv;
1267 struct slave *new_slave = NULL;
1268 struct dev_mc_list *dmi;
1269 struct sockaddr addr;
1271 int old_features = bond_dev->features;
1274 if (!bond->params.use_carrier && slave_dev->ethtool_ops == NULL &&
1275 slave_dev->do_ioctl == NULL) {
1276 printk(KERN_WARNING DRV_NAME
1277 ": %s: Warning: no link monitoring support for %s\n",
1278 bond_dev->name, slave_dev->name);
1281 /* bond must be initialized by bond_open() before enslaving */
1282 if (!(bond_dev->flags & IFF_UP)) {
1283 printk(KERN_WARNING DRV_NAME
1284 " %s: master_dev is not up in bond_enslave\n",
1288 /* already enslaved */
1289 if (slave_dev->flags & IFF_SLAVE) {
1290 dprintk("Error, Device was already enslaved\n");
1294 /* vlan challenged mutual exclusion */
1295 /* no need to lock since we're protected by rtnl_lock */
1296 if (slave_dev->features & NETIF_F_VLAN_CHALLENGED) {
1297 dprintk("%s: NETIF_F_VLAN_CHALLENGED\n", slave_dev->name);
1298 if (!list_empty(&bond->vlan_list)) {
1299 printk(KERN_ERR DRV_NAME
1300 ": %s: Error: cannot enslave VLAN "
1301 "challenged slave %s on VLAN enabled "
1302 "bond %s\n", bond_dev->name, slave_dev->name,
1306 printk(KERN_WARNING DRV_NAME
1307 ": %s: Warning: enslaved VLAN challenged "
1308 "slave %s. Adding VLANs will be blocked as "
1309 "long as %s is part of bond %s\n",
1310 bond_dev->name, slave_dev->name, slave_dev->name,
1312 bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
1315 dprintk("%s: ! NETIF_F_VLAN_CHALLENGED\n", slave_dev->name);
1316 if (bond->slave_cnt == 0) {
1317 /* First slave, and it is not VLAN challenged,
1318 * so remove the block of adding VLANs over the bond.
1320 bond_dev->features &= ~NETIF_F_VLAN_CHALLENGED;
1325 * Old ifenslave binaries are no longer supported. These can
1326 * be identified with moderate accurary by the state of the slave:
1327 * the current ifenslave will set the interface down prior to
1328 * enslaving it; the old ifenslave will not.
1330 if ((slave_dev->flags & IFF_UP)) {
1331 printk(KERN_ERR DRV_NAME ": %s is up. "
1332 "This may be due to an out of date ifenslave.\n",
1335 goto err_undo_flags;
1338 /* set bonding device ether type by slave - bonding netdevices are
1339 * created with ether_setup, so when the slave type is not ARPHRD_ETHER
1340 * there is a need to override some of the type dependent attribs/funcs.
1342 * bond ether type mutual exclusion - don't allow slaves of dissimilar
1343 * ether type (eg ARPHRD_ETHER and ARPHRD_INFINIBAND) share the same bond
1345 if (bond->slave_cnt == 0) {
1346 if (slave_dev->type != ARPHRD_ETHER)
1347 bond_setup_by_slave(bond_dev, slave_dev);
1348 } else if (bond_dev->type != slave_dev->type) {
1349 printk(KERN_ERR DRV_NAME ": %s ether type (%d) is different "
1350 "from other slaves (%d), can not enslave it.\n",
1352 slave_dev->type, bond_dev->type);
1354 goto err_undo_flags;
1357 if (slave_dev->set_mac_address == NULL) {
1358 if (bond->slave_cnt == 0) {
1359 printk(KERN_WARNING DRV_NAME
1360 ": %s: Warning: The first slave device you "
1361 "specified does not support setting the MAC "
1362 "address. This bond MAC address would be that "
1363 "of the active slave.\n", bond_dev->name);
1364 bond->do_set_mac_addr = 0;
1365 } else if (bond->do_set_mac_addr) {
1366 printk(KERN_ERR DRV_NAME
1367 ": %s: Error: The slave device you specified "
1368 "does not support setting the MAC addres,."
1369 "but this bond uses this practice. \n"
1372 goto err_undo_flags;
1376 new_slave = kzalloc(sizeof(struct slave), GFP_KERNEL);
1379 goto err_undo_flags;
1382 /* save slave's original flags before calling
1383 * netdev_set_master and dev_open
1385 new_slave->original_flags = slave_dev->flags;
1388 * Save slave's original ("permanent") mac address for modes
1389 * that need it, and for restoring it upon release, and then
1390 * set it to the master's address
1392 memcpy(new_slave->perm_hwaddr, slave_dev->dev_addr, ETH_ALEN);
1394 if (bond->do_set_mac_addr) {
1396 * Set slave to master's mac address. The application already
1397 * set the master's mac address to that of the first slave
1399 memcpy(addr.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
1400 addr.sa_family = slave_dev->type;
1401 res = dev_set_mac_address(slave_dev, &addr);
1403 dprintk("Error %d calling set_mac_address\n", res);
1408 res = netdev_set_master(slave_dev, bond_dev);
1410 dprintk("Error %d calling netdev_set_master\n", res);
1413 /* open the slave since the application closed it */
1414 res = dev_open(slave_dev);
1416 dprintk("Openning slave %s failed\n", slave_dev->name);
1417 goto err_restore_mac;
1420 new_slave->dev = slave_dev;
1421 slave_dev->priv_flags |= IFF_BONDING;
1423 if ((bond->params.mode == BOND_MODE_TLB) ||
1424 (bond->params.mode == BOND_MODE_ALB)) {
1425 /* bond_alb_init_slave() must be called before all other stages since
1426 * it might fail and we do not want to have to undo everything
1428 res = bond_alb_init_slave(bond, new_slave);
1430 goto err_unset_master;
1434 /* If the mode USES_PRIMARY, then the new slave gets the
1435 * master's promisc (and mc) settings only if it becomes the
1436 * curr_active_slave, and that is taken care of later when calling
1437 * bond_change_active()
1439 if (!USES_PRIMARY(bond->params.mode)) {
1440 /* set promiscuity level to new slave */
1441 if (bond_dev->flags & IFF_PROMISC) {
1442 dev_set_promiscuity(slave_dev, 1);
1445 /* set allmulti level to new slave */
1446 if (bond_dev->flags & IFF_ALLMULTI) {
1447 dev_set_allmulti(slave_dev, 1);
1450 /* upload master's mc_list to new slave */
1451 for (dmi = bond_dev->mc_list; dmi; dmi = dmi->next) {
1452 dev_mc_add (slave_dev, dmi->dmi_addr, dmi->dmi_addrlen, 0);
1456 if (bond->params.mode == BOND_MODE_8023AD) {
1457 /* add lacpdu mc addr to mc list */
1458 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
1460 dev_mc_add(slave_dev, lacpdu_multicast, ETH_ALEN, 0);
1463 bond_add_vlans_on_slave(bond, slave_dev);
1465 write_lock_bh(&bond->lock);
1467 bond_attach_slave(bond, new_slave);
1469 new_slave->delay = 0;
1470 new_slave->link_failure_count = 0;
1472 bond_compute_features(bond);
1474 new_slave->last_arp_rx = jiffies;
1476 if (bond->params.miimon && !bond->params.use_carrier) {
1477 link_reporting = bond_check_dev_link(bond, slave_dev, 1);
1479 if ((link_reporting == -1) && !bond->params.arp_interval) {
1481 * miimon is set but a bonded network driver
1482 * does not support ETHTOOL/MII and
1483 * arp_interval is not set. Note: if
1484 * use_carrier is enabled, we will never go
1485 * here (because netif_carrier is always
1486 * supported); thus, we don't need to change
1487 * the messages for netif_carrier.
1489 printk(KERN_WARNING DRV_NAME
1490 ": %s: Warning: MII and ETHTOOL support not "
1491 "available for interface %s, and "
1492 "arp_interval/arp_ip_target module parameters "
1493 "not specified, thus bonding will not detect "
1494 "link failures! see bonding.txt for details.\n",
1495 bond_dev->name, slave_dev->name);
1496 } else if (link_reporting == -1) {
1497 /* unable get link status using mii/ethtool */
1498 printk(KERN_WARNING DRV_NAME
1499 ": %s: Warning: can't get link status from "
1500 "interface %s; the network driver associated "
1501 "with this interface does not support MII or "
1502 "ETHTOOL link status reporting, thus miimon "
1503 "has no effect on this interface.\n",
1504 bond_dev->name, slave_dev->name);
1508 /* check for initial state */
1509 if (!bond->params.miimon ||
1510 (bond_check_dev_link(bond, slave_dev, 0) == BMSR_LSTATUS)) {
1511 if (bond->params.updelay) {
1512 dprintk("Initial state of slave_dev is "
1513 "BOND_LINK_BACK\n");
1514 new_slave->link = BOND_LINK_BACK;
1515 new_slave->delay = bond->params.updelay;
1517 dprintk("Initial state of slave_dev is "
1519 new_slave->link = BOND_LINK_UP;
1521 new_slave->jiffies = jiffies;
1523 dprintk("Initial state of slave_dev is "
1524 "BOND_LINK_DOWN\n");
1525 new_slave->link = BOND_LINK_DOWN;
1528 if (bond_update_speed_duplex(new_slave) &&
1529 (new_slave->link != BOND_LINK_DOWN)) {
1530 printk(KERN_WARNING DRV_NAME
1531 ": %s: Warning: failed to get speed and duplex from %s, "
1532 "assumed to be 100Mb/sec and Full.\n",
1533 bond_dev->name, new_slave->dev->name);
1535 if (bond->params.mode == BOND_MODE_8023AD) {
1536 printk(KERN_WARNING DRV_NAME
1537 ": %s: Warning: Operation of 802.3ad mode requires ETHTOOL "
1538 "support in base driver for proper aggregator "
1539 "selection.\n", bond_dev->name);
1543 if (USES_PRIMARY(bond->params.mode) && bond->params.primary[0]) {
1544 /* if there is a primary slave, remember it */
1545 if (strcmp(bond->params.primary, new_slave->dev->name) == 0) {
1546 bond->primary_slave = new_slave;
1550 switch (bond->params.mode) {
1551 case BOND_MODE_ACTIVEBACKUP:
1552 bond_set_slave_inactive_flags(new_slave);
1553 bond_select_active_slave(bond);
1555 case BOND_MODE_8023AD:
1556 /* in 802.3ad mode, the internal mechanism
1557 * will activate the slaves in the selected
1560 bond_set_slave_inactive_flags(new_slave);
1561 /* if this is the first slave */
1562 if (bond->slave_cnt == 1) {
1563 SLAVE_AD_INFO(new_slave).id = 1;
1564 /* Initialize AD with the number of times that the AD timer is called in 1 second
1565 * can be called only after the mac address of the bond is set
1567 bond_3ad_initialize(bond, 1000/AD_TIMER_INTERVAL,
1568 bond->params.lacp_fast);
1570 SLAVE_AD_INFO(new_slave).id =
1571 SLAVE_AD_INFO(new_slave->prev).id + 1;
1574 bond_3ad_bind_slave(new_slave);
1578 new_slave->state = BOND_STATE_ACTIVE;
1579 if ((!bond->curr_active_slave) &&
1580 (new_slave->link != BOND_LINK_DOWN)) {
1581 /* first slave or no active slave yet, and this link
1582 * is OK, so make this interface the active one
1584 bond_change_active_slave(bond, new_slave);
1586 bond_set_slave_inactive_flags(new_slave);
1590 dprintk("This slave is always active in trunk mode\n");
1592 /* always active in trunk mode */
1593 new_slave->state = BOND_STATE_ACTIVE;
1595 /* In trunking mode there is little meaning to curr_active_slave
1596 * anyway (it holds no special properties of the bond device),
1597 * so we can change it without calling change_active_interface()
1599 if (!bond->curr_active_slave) {
1600 bond->curr_active_slave = new_slave;
1603 } /* switch(bond_mode) */
1605 bond_set_carrier(bond);
1607 write_unlock_bh(&bond->lock);
1609 res = bond_create_slave_symlinks(bond_dev, slave_dev);
1611 goto err_unset_master;
1613 printk(KERN_INFO DRV_NAME
1614 ": %s: enslaving %s as a%s interface with a%s link.\n",
1615 bond_dev->name, slave_dev->name,
1616 new_slave->state == BOND_STATE_ACTIVE ? "n active" : " backup",
1617 new_slave->link != BOND_LINK_DOWN ? "n up" : " down");
1619 /* enslave is successful */
1622 /* Undo stages on error */
1624 netdev_set_master(slave_dev, NULL);
1627 dev_close(slave_dev);
1630 if (bond->do_set_mac_addr) {
1631 memcpy(addr.sa_data, new_slave->perm_hwaddr, ETH_ALEN);
1632 addr.sa_family = slave_dev->type;
1633 dev_set_mac_address(slave_dev, &addr);
1640 bond_dev->features = old_features;
1646 * Try to release the slave device <slave> from the bond device <master>
1647 * It is legal to access curr_active_slave without a lock because all the function
1650 * The rules for slave state should be:
1651 * for Active/Backup:
1652 * Active stays on all backups go down
1653 * for Bonded connections:
1654 * The first up interface should be left on and all others downed.
1656 int bond_release(struct net_device *bond_dev, struct net_device *slave_dev)
1658 struct bonding *bond = bond_dev->priv;
1659 struct slave *slave, *oldcurrent;
1660 struct sockaddr addr;
1661 int mac_addr_differ;
1662 DECLARE_MAC_BUF(mac);
1664 /* slave is not a slave or master is not master of this slave */
1665 if (!(slave_dev->flags & IFF_SLAVE) ||
1666 (slave_dev->master != bond_dev)) {
1667 printk(KERN_ERR DRV_NAME
1668 ": %s: Error: cannot release %s.\n",
1669 bond_dev->name, slave_dev->name);
1673 write_lock_bh(&bond->lock);
1675 slave = bond_get_slave_by_dev(bond, slave_dev);
1677 /* not a slave of this bond */
1678 printk(KERN_INFO DRV_NAME
1679 ": %s: %s not enslaved\n",
1680 bond_dev->name, slave_dev->name);
1681 write_unlock_bh(&bond->lock);
1685 mac_addr_differ = memcmp(bond_dev->dev_addr,
1688 if (!mac_addr_differ && (bond->slave_cnt > 1)) {
1689 printk(KERN_WARNING DRV_NAME
1690 ": %s: Warning: the permanent HWaddr of %s - "
1691 "%s - is still in use by %s. "
1692 "Set the HWaddr of %s to a different address "
1693 "to avoid conflicts.\n",
1696 print_mac(mac, slave->perm_hwaddr),
1701 /* Inform AD package of unbinding of slave. */
1702 if (bond->params.mode == BOND_MODE_8023AD) {
1703 /* must be called before the slave is
1704 * detached from the list
1706 bond_3ad_unbind_slave(slave);
1709 printk(KERN_INFO DRV_NAME
1710 ": %s: releasing %s interface %s\n",
1712 (slave->state == BOND_STATE_ACTIVE)
1713 ? "active" : "backup",
1716 oldcurrent = bond->curr_active_slave;
1718 bond->current_arp_slave = NULL;
1720 /* release the slave from its bond */
1721 bond_detach_slave(bond, slave);
1723 bond_compute_features(bond);
1725 if (bond->primary_slave == slave) {
1726 bond->primary_slave = NULL;
1729 if (oldcurrent == slave) {
1730 bond_change_active_slave(bond, NULL);
1733 if ((bond->params.mode == BOND_MODE_TLB) ||
1734 (bond->params.mode == BOND_MODE_ALB)) {
1735 /* Must be called only after the slave has been
1736 * detached from the list and the curr_active_slave
1737 * has been cleared (if our_slave == old_current),
1738 * but before a new active slave is selected.
1740 bond_alb_deinit_slave(bond, slave);
1743 if (oldcurrent == slave)
1744 bond_select_active_slave(bond);
1746 if (bond->slave_cnt == 0) {
1747 bond_set_carrier(bond);
1749 /* if the last slave was removed, zero the mac address
1750 * of the master so it will be set by the application
1751 * to the mac address of the first slave
1753 memset(bond_dev->dev_addr, 0, bond_dev->addr_len);
1755 if (list_empty(&bond->vlan_list)) {
1756 bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
1758 printk(KERN_WARNING DRV_NAME
1759 ": %s: Warning: clearing HW address of %s while it "
1760 "still has VLANs.\n",
1761 bond_dev->name, bond_dev->name);
1762 printk(KERN_WARNING DRV_NAME
1763 ": %s: When re-adding slaves, make sure the bond's "
1764 "HW address matches its VLANs'.\n",
1767 } else if ((bond_dev->features & NETIF_F_VLAN_CHALLENGED) &&
1768 !bond_has_challenged_slaves(bond)) {
1769 printk(KERN_INFO DRV_NAME
1770 ": %s: last VLAN challenged slave %s "
1771 "left bond %s. VLAN blocking is removed\n",
1772 bond_dev->name, slave_dev->name, bond_dev->name);
1773 bond_dev->features &= ~NETIF_F_VLAN_CHALLENGED;
1776 write_unlock_bh(&bond->lock);
1778 /* must do this from outside any spinlocks */
1779 bond_destroy_slave_symlinks(bond_dev, slave_dev);
1781 bond_del_vlans_from_slave(bond, slave_dev);
1783 /* If the mode USES_PRIMARY, then we should only remove its
1784 * promisc and mc settings if it was the curr_active_slave, but that was
1785 * already taken care of above when we detached the slave
1787 if (!USES_PRIMARY(bond->params.mode)) {
1788 /* unset promiscuity level from slave */
1789 if (bond_dev->flags & IFF_PROMISC) {
1790 dev_set_promiscuity(slave_dev, -1);
1793 /* unset allmulti level from slave */
1794 if (bond_dev->flags & IFF_ALLMULTI) {
1795 dev_set_allmulti(slave_dev, -1);
1798 /* flush master's mc_list from slave */
1799 bond_mc_list_flush(bond_dev, slave_dev);
1802 netdev_set_master(slave_dev, NULL);
1804 /* close slave before restoring its mac address */
1805 dev_close(slave_dev);
1807 if (bond->do_set_mac_addr) {
1808 /* restore original ("permanent") mac address */
1809 memcpy(addr.sa_data, slave->perm_hwaddr, ETH_ALEN);
1810 addr.sa_family = slave_dev->type;
1811 dev_set_mac_address(slave_dev, &addr);
1814 slave_dev->priv_flags &= ~(IFF_MASTER_8023AD | IFF_MASTER_ALB |
1815 IFF_SLAVE_INACTIVE | IFF_BONDING |
1820 return 0; /* deletion OK */
1824 * This function releases all slaves.
1826 static int bond_release_all(struct net_device *bond_dev)
1828 struct bonding *bond = bond_dev->priv;
1829 struct slave *slave;
1830 struct net_device *slave_dev;
1831 struct sockaddr addr;
1833 write_lock_bh(&bond->lock);
1835 netif_carrier_off(bond_dev);
1837 if (bond->slave_cnt == 0) {
1841 bond->current_arp_slave = NULL;
1842 bond->primary_slave = NULL;
1843 bond_change_active_slave(bond, NULL);
1845 while ((slave = bond->first_slave) != NULL) {
1846 /* Inform AD package of unbinding of slave
1847 * before slave is detached from the list.
1849 if (bond->params.mode == BOND_MODE_8023AD) {
1850 bond_3ad_unbind_slave(slave);
1853 slave_dev = slave->dev;
1854 bond_detach_slave(bond, slave);
1856 if ((bond->params.mode == BOND_MODE_TLB) ||
1857 (bond->params.mode == BOND_MODE_ALB)) {
1858 /* must be called only after the slave
1859 * has been detached from the list
1861 bond_alb_deinit_slave(bond, slave);
1864 bond_compute_features(bond);
1866 /* now that the slave is detached, unlock and perform
1867 * all the undo steps that should not be called from
1870 write_unlock_bh(&bond->lock);
1872 bond_destroy_slave_symlinks(bond_dev, slave_dev);
1873 bond_del_vlans_from_slave(bond, slave_dev);
1875 /* If the mode USES_PRIMARY, then we should only remove its
1876 * promisc and mc settings if it was the curr_active_slave, but that was
1877 * already taken care of above when we detached the slave
1879 if (!USES_PRIMARY(bond->params.mode)) {
1880 /* unset promiscuity level from slave */
1881 if (bond_dev->flags & IFF_PROMISC) {
1882 dev_set_promiscuity(slave_dev, -1);
1885 /* unset allmulti level from slave */
1886 if (bond_dev->flags & IFF_ALLMULTI) {
1887 dev_set_allmulti(slave_dev, -1);
1890 /* flush master's mc_list from slave */
1891 bond_mc_list_flush(bond_dev, slave_dev);
1894 netdev_set_master(slave_dev, NULL);
1896 /* close slave before restoring its mac address */
1897 dev_close(slave_dev);
1899 if (bond->do_set_mac_addr) {
1900 /* restore original ("permanent") mac address*/
1901 memcpy(addr.sa_data, slave->perm_hwaddr, ETH_ALEN);
1902 addr.sa_family = slave_dev->type;
1903 dev_set_mac_address(slave_dev, &addr);
1906 slave_dev->priv_flags &= ~(IFF_MASTER_8023AD | IFF_MASTER_ALB |
1907 IFF_SLAVE_INACTIVE);
1911 /* re-acquire the lock before getting the next slave */
1912 write_lock_bh(&bond->lock);
1915 /* zero the mac address of the master so it will be
1916 * set by the application to the mac address of the
1919 memset(bond_dev->dev_addr, 0, bond_dev->addr_len);
1921 if (list_empty(&bond->vlan_list)) {
1922 bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
1924 printk(KERN_WARNING DRV_NAME
1925 ": %s: Warning: clearing HW address of %s while it "
1926 "still has VLANs.\n",
1927 bond_dev->name, bond_dev->name);
1928 printk(KERN_WARNING DRV_NAME
1929 ": %s: When re-adding slaves, make sure the bond's "
1930 "HW address matches its VLANs'.\n",
1934 printk(KERN_INFO DRV_NAME
1935 ": %s: released all slaves\n",
1939 write_unlock_bh(&bond->lock);
1945 * This function changes the active slave to slave <slave_dev>.
1946 * It returns -EINVAL in the following cases.
1947 * - <slave_dev> is not found in the list.
1948 * - There is not active slave now.
1949 * - <slave_dev> is already active.
1950 * - The link state of <slave_dev> is not BOND_LINK_UP.
1951 * - <slave_dev> is not running.
1952 * In these cases, this fuction does nothing.
1953 * In the other cases, currnt_slave pointer is changed and 0 is returned.
1955 static int bond_ioctl_change_active(struct net_device *bond_dev, struct net_device *slave_dev)
1957 struct bonding *bond = bond_dev->priv;
1958 struct slave *old_active = NULL;
1959 struct slave *new_active = NULL;
1962 if (!USES_PRIMARY(bond->params.mode)) {
1966 /* Verify that master_dev is indeed the master of slave_dev */
1967 if (!(slave_dev->flags & IFF_SLAVE) ||
1968 (slave_dev->master != bond_dev)) {
1972 write_lock_bh(&bond->lock);
1974 old_active = bond->curr_active_slave;
1975 new_active = bond_get_slave_by_dev(bond, slave_dev);
1978 * Changing to the current active: do nothing; return success.
1980 if (new_active && (new_active == old_active)) {
1981 write_unlock_bh(&bond->lock);
1987 (new_active->link == BOND_LINK_UP) &&
1988 IS_UP(new_active->dev)) {
1989 bond_change_active_slave(bond, new_active);
1994 write_unlock_bh(&bond->lock);
1999 static int bond_info_query(struct net_device *bond_dev, struct ifbond *info)
2001 struct bonding *bond = bond_dev->priv;
2003 info->bond_mode = bond->params.mode;
2004 info->miimon = bond->params.miimon;
2006 read_lock_bh(&bond->lock);
2007 info->num_slaves = bond->slave_cnt;
2008 read_unlock_bh(&bond->lock);
2013 static int bond_slave_info_query(struct net_device *bond_dev, struct ifslave *info)
2015 struct bonding *bond = bond_dev->priv;
2016 struct slave *slave;
2019 if (info->slave_id < 0) {
2023 read_lock_bh(&bond->lock);
2025 bond_for_each_slave(bond, slave, i) {
2026 if (i == (int)info->slave_id) {
2032 read_unlock_bh(&bond->lock);
2035 strcpy(info->slave_name, slave->dev->name);
2036 info->link = slave->link;
2037 info->state = slave->state;
2038 info->link_failure_count = slave->link_failure_count;
2046 /*-------------------------------- Monitoring -------------------------------*/
2048 /* this function is called regularly to monitor each slave's link. */
2049 void bond_mii_monitor(struct net_device *bond_dev)
2051 struct bonding *bond = bond_dev->priv;
2052 struct slave *slave, *oldcurrent;
2053 int do_failover = 0;
2057 read_lock(&bond->lock);
2059 delta_in_ticks = (bond->params.miimon * HZ) / 1000;
2061 if (bond->kill_timers) {
2065 if (bond->slave_cnt == 0) {
2069 /* we will try to read the link status of each of our slaves, and
2070 * set their IFF_RUNNING flag appropriately. For each slave not
2071 * supporting MII status, we won't do anything so that a user-space
2072 * program could monitor the link itself if needed.
2075 read_lock(&bond->curr_slave_lock);
2076 oldcurrent = bond->curr_active_slave;
2077 read_unlock(&bond->curr_slave_lock);
2079 bond_for_each_slave(bond, slave, i) {
2080 struct net_device *slave_dev = slave->dev;
2082 u16 old_speed = slave->speed;
2083 u8 old_duplex = slave->duplex;
2085 link_state = bond_check_dev_link(bond, slave_dev, 0);
2087 switch (slave->link) {
2088 case BOND_LINK_UP: /* the link was up */
2089 if (link_state == BMSR_LSTATUS) {
2090 /* link stays up, nothing more to do */
2092 } else { /* link going down */
2093 slave->link = BOND_LINK_FAIL;
2094 slave->delay = bond->params.downdelay;
2096 if (slave->link_failure_count < UINT_MAX) {
2097 slave->link_failure_count++;
2100 if (bond->params.downdelay) {
2101 printk(KERN_INFO DRV_NAME
2102 ": %s: link status down for %s "
2103 "interface %s, disabling it in "
2107 ? ((bond->params.mode == BOND_MODE_ACTIVEBACKUP)
2108 ? ((slave == oldcurrent)
2109 ? "active " : "backup ")
2113 bond->params.downdelay * bond->params.miimon);
2116 /* no break ! fall through the BOND_LINK_FAIL test to
2117 ensure proper action to be taken
2119 case BOND_LINK_FAIL: /* the link has just gone down */
2120 if (link_state != BMSR_LSTATUS) {
2121 /* link stays down */
2122 if (slave->delay <= 0) {
2123 /* link down for too long time */
2124 slave->link = BOND_LINK_DOWN;
2126 /* in active/backup mode, we must
2127 * completely disable this interface
2129 if ((bond->params.mode == BOND_MODE_ACTIVEBACKUP) ||
2130 (bond->params.mode == BOND_MODE_8023AD)) {
2131 bond_set_slave_inactive_flags(slave);
2134 printk(KERN_INFO DRV_NAME
2135 ": %s: link status definitely "
2136 "down for interface %s, "
2141 /* notify ad that the link status has changed */
2142 if (bond->params.mode == BOND_MODE_8023AD) {
2143 bond_3ad_handle_link_change(slave, BOND_LINK_DOWN);
2146 if ((bond->params.mode == BOND_MODE_TLB) ||
2147 (bond->params.mode == BOND_MODE_ALB)) {
2148 bond_alb_handle_link_change(bond, slave, BOND_LINK_DOWN);
2151 if (slave == oldcurrent) {
2159 slave->link = BOND_LINK_UP;
2160 slave->jiffies = jiffies;
2161 printk(KERN_INFO DRV_NAME
2162 ": %s: link status up again after %d "
2163 "ms for interface %s.\n",
2165 (bond->params.downdelay - slave->delay) * bond->params.miimon,
2169 case BOND_LINK_DOWN: /* the link was down */
2170 if (link_state != BMSR_LSTATUS) {
2171 /* the link stays down, nothing more to do */
2173 } else { /* link going up */
2174 slave->link = BOND_LINK_BACK;
2175 slave->delay = bond->params.updelay;
2177 if (bond->params.updelay) {
2178 /* if updelay == 0, no need to
2179 advertise about a 0 ms delay */
2180 printk(KERN_INFO DRV_NAME
2181 ": %s: link status up for "
2182 "interface %s, enabling it "
2186 bond->params.updelay * bond->params.miimon);
2189 /* no break ! fall through the BOND_LINK_BACK state in
2190 case there's something to do.
2192 case BOND_LINK_BACK: /* the link has just come back */
2193 if (link_state != BMSR_LSTATUS) {
2194 /* link down again */
2195 slave->link = BOND_LINK_DOWN;
2197 printk(KERN_INFO DRV_NAME
2198 ": %s: link status down again after %d "
2199 "ms for interface %s.\n",
2201 (bond->params.updelay - slave->delay) * bond->params.miimon,
2205 if (slave->delay == 0) {
2206 /* now the link has been up for long time enough */
2207 slave->link = BOND_LINK_UP;
2208 slave->jiffies = jiffies;
2210 if (bond->params.mode == BOND_MODE_8023AD) {
2211 /* prevent it from being the active one */
2212 slave->state = BOND_STATE_BACKUP;
2213 } else if (bond->params.mode != BOND_MODE_ACTIVEBACKUP) {
2214 /* make it immediately active */
2215 slave->state = BOND_STATE_ACTIVE;
2216 } else if (slave != bond->primary_slave) {
2217 /* prevent it from being the active one */
2218 slave->state = BOND_STATE_BACKUP;
2221 printk(KERN_INFO DRV_NAME
2222 ": %s: link status definitely "
2223 "up for interface %s.\n",
2227 /* notify ad that the link status has changed */
2228 if (bond->params.mode == BOND_MODE_8023AD) {
2229 bond_3ad_handle_link_change(slave, BOND_LINK_UP);
2232 if ((bond->params.mode == BOND_MODE_TLB) ||
2233 (bond->params.mode == BOND_MODE_ALB)) {
2234 bond_alb_handle_link_change(bond, slave, BOND_LINK_UP);
2237 if ((!oldcurrent) ||
2238 (slave == bond->primary_slave)) {
2247 /* Should not happen */
2248 printk(KERN_ERR DRV_NAME
2249 ": %s: Error: %s Illegal value (link=%d)\n",
2254 } /* end of switch (slave->link) */
2256 bond_update_speed_duplex(slave);
2258 if (bond->params.mode == BOND_MODE_8023AD) {
2259 if (old_speed != slave->speed) {
2260 bond_3ad_adapter_speed_changed(slave);
2263 if (old_duplex != slave->duplex) {
2264 bond_3ad_adapter_duplex_changed(slave);
2271 write_lock(&bond->curr_slave_lock);
2273 bond_select_active_slave(bond);
2275 write_unlock(&bond->curr_slave_lock);
2277 bond_set_carrier(bond);
2280 if (bond->params.miimon) {
2281 mod_timer(&bond->mii_timer, jiffies + delta_in_ticks);
2284 read_unlock(&bond->lock);
2288 static __be32 bond_glean_dev_ip(struct net_device *dev)
2290 struct in_device *idev;
2291 struct in_ifaddr *ifa;
2298 idev = __in_dev_get_rcu(dev);
2302 ifa = idev->ifa_list;
2306 addr = ifa->ifa_local;
2312 static int bond_has_ip(struct bonding *bond)
2314 struct vlan_entry *vlan, *vlan_next;
2316 if (bond->master_ip)
2319 if (list_empty(&bond->vlan_list))
2322 list_for_each_entry_safe(vlan, vlan_next, &bond->vlan_list,
2331 static int bond_has_this_ip(struct bonding *bond, __be32 ip)
2333 struct vlan_entry *vlan, *vlan_next;
2335 if (ip == bond->master_ip)
2338 if (list_empty(&bond->vlan_list))
2341 list_for_each_entry_safe(vlan, vlan_next, &bond->vlan_list,
2343 if (ip == vlan->vlan_ip)
2351 * We go to the (large) trouble of VLAN tagging ARP frames because
2352 * switches in VLAN mode (especially if ports are configured as
2353 * "native" to a VLAN) might not pass non-tagged frames.
2355 static void bond_arp_send(struct net_device *slave_dev, int arp_op, __be32 dest_ip, __be32 src_ip, unsigned short vlan_id)
2357 struct sk_buff *skb;
2359 dprintk("arp %d on slave %s: dst %x src %x vid %d\n", arp_op,
2360 slave_dev->name, dest_ip, src_ip, vlan_id);
2362 skb = arp_create(arp_op, ETH_P_ARP, dest_ip, slave_dev, src_ip,
2363 NULL, slave_dev->dev_addr, NULL);
2366 printk(KERN_ERR DRV_NAME ": ARP packet allocation failed\n");
2370 skb = vlan_put_tag(skb, vlan_id);
2372 printk(KERN_ERR DRV_NAME ": failed to insert VLAN tag\n");
2380 static void bond_arp_send_all(struct bonding *bond, struct slave *slave)
2383 __be32 *targets = bond->params.arp_targets;
2384 struct vlan_entry *vlan, *vlan_next;
2385 struct net_device *vlan_dev;
2389 for (i = 0; (i < BOND_MAX_ARP_TARGETS); i++) {
2392 dprintk("basa: target %x\n", targets[i]);
2393 if (list_empty(&bond->vlan_list)) {
2394 dprintk("basa: empty vlan: arp_send\n");
2395 bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2396 bond->master_ip, 0);
2401 * If VLANs are configured, we do a route lookup to
2402 * determine which VLAN interface would be used, so we
2403 * can tag the ARP with the proper VLAN tag.
2405 memset(&fl, 0, sizeof(fl));
2406 fl.fl4_dst = targets[i];
2407 fl.fl4_tos = RTO_ONLINK;
2409 rv = ip_route_output_key(&rt, &fl);
2411 if (net_ratelimit()) {
2412 printk(KERN_WARNING DRV_NAME
2413 ": %s: no route to arp_ip_target %u.%u.%u.%u\n",
2414 bond->dev->name, NIPQUAD(fl.fl4_dst));
2420 * This target is not on a VLAN
2422 if (rt->u.dst.dev == bond->dev) {
2424 dprintk("basa: rtdev == bond->dev: arp_send\n");
2425 bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2426 bond->master_ip, 0);
2431 list_for_each_entry_safe(vlan, vlan_next, &bond->vlan_list,
2433 vlan_dev = vlan_group_get_device(bond->vlgrp, vlan->vlan_id);
2434 if (vlan_dev == rt->u.dst.dev) {
2435 vlan_id = vlan->vlan_id;
2436 dprintk("basa: vlan match on %s %d\n",
2437 vlan_dev->name, vlan_id);
2444 bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2445 vlan->vlan_ip, vlan_id);
2449 if (net_ratelimit()) {
2450 printk(KERN_WARNING DRV_NAME
2451 ": %s: no path to arp_ip_target %u.%u.%u.%u via rt.dev %s\n",
2452 bond->dev->name, NIPQUAD(fl.fl4_dst),
2453 rt->u.dst.dev ? rt->u.dst.dev->name : "NULL");
2460 * Kick out a gratuitous ARP for an IP on the bonding master plus one
2461 * for each VLAN above us.
2463 static void bond_send_gratuitous_arp(struct bonding *bond)
2465 struct slave *slave = bond->curr_active_slave;
2466 struct vlan_entry *vlan;
2467 struct net_device *vlan_dev;
2469 dprintk("bond_send_grat_arp: bond %s slave %s\n", bond->dev->name,
2470 slave ? slave->dev->name : "NULL");
2474 if (bond->master_ip) {
2475 bond_arp_send(slave->dev, ARPOP_REPLY, bond->master_ip,
2476 bond->master_ip, 0);
2479 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
2480 vlan_dev = vlan_group_get_device(bond->vlgrp, vlan->vlan_id);
2481 if (vlan->vlan_ip) {
2482 bond_arp_send(slave->dev, ARPOP_REPLY, vlan->vlan_ip,
2483 vlan->vlan_ip, vlan->vlan_id);
2488 static void bond_validate_arp(struct bonding *bond, struct slave *slave, __be32 sip, __be32 tip)
2491 __be32 *targets = bond->params.arp_targets;
2493 targets = bond->params.arp_targets;
2494 for (i = 0; (i < BOND_MAX_ARP_TARGETS) && targets[i]; i++) {
2495 dprintk("bva: sip %u.%u.%u.%u tip %u.%u.%u.%u t[%d] "
2496 "%u.%u.%u.%u bhti(tip) %d\n",
2497 NIPQUAD(sip), NIPQUAD(tip), i, NIPQUAD(targets[i]),
2498 bond_has_this_ip(bond, tip));
2499 if (sip == targets[i]) {
2500 if (bond_has_this_ip(bond, tip))
2501 slave->last_arp_rx = jiffies;
2507 static int bond_arp_rcv(struct sk_buff *skb, struct net_device *dev, struct packet_type *pt, struct net_device *orig_dev)
2510 struct slave *slave;
2511 struct bonding *bond;
2512 unsigned char *arp_ptr;
2515 if (dev->nd_net != &init_net)
2518 if (!(dev->priv_flags & IFF_BONDING) || !(dev->flags & IFF_MASTER))
2522 read_lock(&bond->lock);
2524 dprintk("bond_arp_rcv: bond %s skb->dev %s orig_dev %s\n",
2525 bond->dev->name, skb->dev ? skb->dev->name : "NULL",
2526 orig_dev ? orig_dev->name : "NULL");
2528 slave = bond_get_slave_by_dev(bond, orig_dev);
2529 if (!slave || !slave_do_arp_validate(bond, slave))
2532 /* ARP header, plus 2 device addresses, plus 2 IP addresses. */
2533 if (!pskb_may_pull(skb, (sizeof(struct arphdr) +
2534 (2 * dev->addr_len) +
2535 (2 * sizeof(u32)))))
2539 if (arp->ar_hln != dev->addr_len ||
2540 skb->pkt_type == PACKET_OTHERHOST ||
2541 skb->pkt_type == PACKET_LOOPBACK ||
2542 arp->ar_hrd != htons(ARPHRD_ETHER) ||
2543 arp->ar_pro != htons(ETH_P_IP) ||
2547 arp_ptr = (unsigned char *)(arp + 1);
2548 arp_ptr += dev->addr_len;
2549 memcpy(&sip, arp_ptr, 4);
2550 arp_ptr += 4 + dev->addr_len;
2551 memcpy(&tip, arp_ptr, 4);
2553 dprintk("bond_arp_rcv: %s %s/%d av %d sv %d sip %u.%u.%u.%u"
2554 " tip %u.%u.%u.%u\n", bond->dev->name, slave->dev->name,
2555 slave->state, bond->params.arp_validate,
2556 slave_do_arp_validate(bond, slave), NIPQUAD(sip), NIPQUAD(tip));
2559 * Backup slaves won't see the ARP reply, but do come through
2560 * here for each ARP probe (so we swap the sip/tip to validate
2561 * the probe). In a "redundant switch, common router" type of
2562 * configuration, the ARP probe will (hopefully) travel from
2563 * the active, through one switch, the router, then the other
2564 * switch before reaching the backup.
2566 if (slave->state == BOND_STATE_ACTIVE)
2567 bond_validate_arp(bond, slave, sip, tip);
2569 bond_validate_arp(bond, slave, tip, sip);
2572 read_unlock(&bond->lock);
2575 return NET_RX_SUCCESS;
2579 * this function is called regularly to monitor each slave's link
2580 * ensuring that traffic is being sent and received when arp monitoring
2581 * is used in load-balancing mode. if the adapter has been dormant, then an
2582 * arp is transmitted to generate traffic. see activebackup_arp_monitor for
2583 * arp monitoring in active backup mode.
2585 void bond_loadbalance_arp_mon(struct net_device *bond_dev)
2587 struct bonding *bond = bond_dev->priv;
2588 struct slave *slave, *oldcurrent;
2589 int do_failover = 0;
2593 read_lock(&bond->lock);
2595 delta_in_ticks = (bond->params.arp_interval * HZ) / 1000;
2597 if (bond->kill_timers) {
2601 if (bond->slave_cnt == 0) {
2605 read_lock(&bond->curr_slave_lock);
2606 oldcurrent = bond->curr_active_slave;
2607 read_unlock(&bond->curr_slave_lock);
2609 /* see if any of the previous devices are up now (i.e. they have
2610 * xmt and rcv traffic). the curr_active_slave does not come into
2611 * the picture unless it is null. also, slave->jiffies is not needed
2612 * here because we send an arp on each slave and give a slave as
2613 * long as it needs to get the tx/rx within the delta.
2614 * TODO: what about up/down delay in arp mode? it wasn't here before
2617 bond_for_each_slave(bond, slave, i) {
2618 if (slave->link != BOND_LINK_UP) {
2619 if (((jiffies - slave->dev->trans_start) <= delta_in_ticks) &&
2620 ((jiffies - slave->dev->last_rx) <= delta_in_ticks)) {
2622 slave->link = BOND_LINK_UP;
2623 slave->state = BOND_STATE_ACTIVE;
2625 /* primary_slave has no meaning in round-robin
2626 * mode. the window of a slave being up and
2627 * curr_active_slave being null after enslaving
2631 printk(KERN_INFO DRV_NAME
2632 ": %s: link status definitely "
2633 "up for interface %s, ",
2638 printk(KERN_INFO DRV_NAME
2639 ": %s: interface %s is now up\n",
2645 /* slave->link == BOND_LINK_UP */
2647 /* not all switches will respond to an arp request
2648 * when the source ip is 0, so don't take the link down
2649 * if we don't know our ip yet
2651 if (((jiffies - slave->dev->trans_start) >= (2*delta_in_ticks)) ||
2652 (((jiffies - slave->dev->last_rx) >= (2*delta_in_ticks)) &&
2653 bond_has_ip(bond))) {
2655 slave->link = BOND_LINK_DOWN;
2656 slave->state = BOND_STATE_BACKUP;
2658 if (slave->link_failure_count < UINT_MAX) {
2659 slave->link_failure_count++;
2662 printk(KERN_INFO DRV_NAME
2663 ": %s: interface %s is now down.\n",
2667 if (slave == oldcurrent) {
2673 /* note: if switch is in round-robin mode, all links
2674 * must tx arp to ensure all links rx an arp - otherwise
2675 * links may oscillate or not come up at all; if switch is
2676 * in something like xor mode, there is nothing we can
2677 * do - all replies will be rx'ed on same link causing slaves
2678 * to be unstable during low/no traffic periods
2680 if (IS_UP(slave->dev)) {
2681 bond_arp_send_all(bond, slave);
2686 write_lock(&bond->curr_slave_lock);
2688 bond_select_active_slave(bond);
2690 write_unlock(&bond->curr_slave_lock);
2694 if (bond->params.arp_interval) {
2695 mod_timer(&bond->arp_timer, jiffies + delta_in_ticks);
2698 read_unlock(&bond->lock);
2702 * When using arp monitoring in active-backup mode, this function is
2703 * called to determine if any backup slaves have went down or a new
2704 * current slave needs to be found.
2705 * The backup slaves never generate traffic, they are considered up by merely
2706 * receiving traffic. If the current slave goes down, each backup slave will
2707 * be given the opportunity to tx/rx an arp before being taken down - this
2708 * prevents all slaves from being taken down due to the current slave not
2709 * sending any traffic for the backups to receive. The arps are not necessarily
2710 * necessary, any tx and rx traffic will keep the current slave up. While any
2711 * rx traffic will keep the backup slaves up, the current slave is responsible
2712 * for generating traffic to keep them up regardless of any other traffic they
2713 * may have received.
2714 * see loadbalance_arp_monitor for arp monitoring in load balancing mode
2716 void bond_activebackup_arp_mon(struct net_device *bond_dev)
2718 struct bonding *bond = bond_dev->priv;
2719 struct slave *slave;
2723 read_lock(&bond->lock);
2725 delta_in_ticks = (bond->params.arp_interval * HZ) / 1000;
2727 if (bond->kill_timers) {
2731 if (bond->slave_cnt == 0) {
2735 /* determine if any slave has come up or any backup slave has
2737 * TODO: what about up/down delay in arp mode? it wasn't here before
2740 bond_for_each_slave(bond, slave, i) {
2741 if (slave->link != BOND_LINK_UP) {
2742 if ((jiffies - slave_last_rx(bond, slave)) <=
2745 slave->link = BOND_LINK_UP;
2747 write_lock(&bond->curr_slave_lock);
2749 if ((!bond->curr_active_slave) &&
2750 ((jiffies - slave->dev->trans_start) <= delta_in_ticks)) {
2751 bond_change_active_slave(bond, slave);
2752 bond->current_arp_slave = NULL;
2753 } else if (bond->curr_active_slave != slave) {
2754 /* this slave has just come up but we
2755 * already have a current slave; this
2756 * can also happen if bond_enslave adds
2757 * a new slave that is up while we are
2758 * searching for a new slave
2760 bond_set_slave_inactive_flags(slave);
2761 bond->current_arp_slave = NULL;
2764 bond_set_carrier(bond);
2766 if (slave == bond->curr_active_slave) {
2767 printk(KERN_INFO DRV_NAME
2768 ": %s: %s is up and now the "
2769 "active interface\n",
2772 netif_carrier_on(bond->dev);
2774 printk(KERN_INFO DRV_NAME
2775 ": %s: backup interface %s is "
2781 write_unlock(&bond->curr_slave_lock);
2784 read_lock(&bond->curr_slave_lock);
2786 if ((slave != bond->curr_active_slave) &&
2787 (!bond->current_arp_slave) &&
2788 (((jiffies - slave_last_rx(bond, slave)) >= 3*delta_in_ticks) &&
2789 bond_has_ip(bond))) {
2790 /* a backup slave has gone down; three times
2791 * the delta allows the current slave to be
2792 * taken out before the backup slave.
2793 * note: a non-null current_arp_slave indicates
2794 * the curr_active_slave went down and we are
2795 * searching for a new one; under this
2796 * condition we only take the curr_active_slave
2797 * down - this gives each slave a chance to
2798 * tx/rx traffic before being taken out
2801 read_unlock(&bond->curr_slave_lock);
2803 slave->link = BOND_LINK_DOWN;
2805 if (slave->link_failure_count < UINT_MAX) {
2806 slave->link_failure_count++;
2809 bond_set_slave_inactive_flags(slave);
2811 printk(KERN_INFO DRV_NAME
2812 ": %s: backup interface %s is now down\n",
2816 read_unlock(&bond->curr_slave_lock);
2821 read_lock(&bond->curr_slave_lock);
2822 slave = bond->curr_active_slave;
2823 read_unlock(&bond->curr_slave_lock);
2826 /* if we have sent traffic in the past 2*arp_intervals but
2827 * haven't xmit and rx traffic in that time interval, select
2828 * a different slave. slave->jiffies is only updated when
2829 * a slave first becomes the curr_active_slave - not necessarily
2830 * after every arp; this ensures the slave has a full 2*delta
2831 * before being taken out. if a primary is being used, check
2832 * if it is up and needs to take over as the curr_active_slave
2834 if ((((jiffies - slave->dev->trans_start) >= (2*delta_in_ticks)) ||
2835 (((jiffies - slave_last_rx(bond, slave)) >= (2*delta_in_ticks)) &&
2836 bond_has_ip(bond))) &&
2837 ((jiffies - slave->jiffies) >= 2*delta_in_ticks)) {
2839 slave->link = BOND_LINK_DOWN;
2841 if (slave->link_failure_count < UINT_MAX) {
2842 slave->link_failure_count++;
2845 printk(KERN_INFO DRV_NAME
2846 ": %s: link status down for active interface "
2847 "%s, disabling it\n",
2851 write_lock(&bond->curr_slave_lock);
2853 bond_select_active_slave(bond);
2854 slave = bond->curr_active_slave;
2856 write_unlock(&bond->curr_slave_lock);
2858 bond->current_arp_slave = slave;
2861 slave->jiffies = jiffies;
2863 } else if ((bond->primary_slave) &&
2864 (bond->primary_slave != slave) &&
2865 (bond->primary_slave->link == BOND_LINK_UP)) {
2866 /* at this point, slave is the curr_active_slave */
2867 printk(KERN_INFO DRV_NAME
2868 ": %s: changing from interface %s to primary "
2872 bond->primary_slave->dev->name);
2874 /* primary is up so switch to it */
2875 write_lock(&bond->curr_slave_lock);
2876 bond_change_active_slave(bond, bond->primary_slave);
2877 write_unlock(&bond->curr_slave_lock);
2879 slave = bond->primary_slave;
2880 slave->jiffies = jiffies;
2882 bond->current_arp_slave = NULL;
2885 /* the current slave must tx an arp to ensure backup slaves
2888 if (slave && bond_has_ip(bond)) {
2889 bond_arp_send_all(bond, slave);
2893 /* if we don't have a curr_active_slave, search for the next available
2894 * backup slave from the current_arp_slave and make it the candidate
2895 * for becoming the curr_active_slave
2898 if (!bond->current_arp_slave) {
2899 bond->current_arp_slave = bond->first_slave;
2902 if (bond->current_arp_slave) {
2903 bond_set_slave_inactive_flags(bond->current_arp_slave);
2905 /* search for next candidate */
2906 bond_for_each_slave_from(bond, slave, i, bond->current_arp_slave->next) {
2907 if (IS_UP(slave->dev)) {
2908 slave->link = BOND_LINK_BACK;
2909 bond_set_slave_active_flags(slave);
2910 bond_arp_send_all(bond, slave);
2911 slave->jiffies = jiffies;
2912 bond->current_arp_slave = slave;
2916 /* if the link state is up at this point, we
2917 * mark it down - this can happen if we have
2918 * simultaneous link failures and
2919 * reselect_active_interface doesn't make this
2920 * one the current slave so it is still marked
2921 * up when it is actually down
2923 if (slave->link == BOND_LINK_UP) {
2924 slave->link = BOND_LINK_DOWN;
2925 if (slave->link_failure_count < UINT_MAX) {
2926 slave->link_failure_count++;
2929 bond_set_slave_inactive_flags(slave);
2931 printk(KERN_INFO DRV_NAME
2932 ": %s: backup interface %s is "
2942 if (bond->params.arp_interval) {
2943 mod_timer(&bond->arp_timer, jiffies + delta_in_ticks);
2946 read_unlock(&bond->lock);
2949 /*------------------------------ proc/seq_file-------------------------------*/
2951 #ifdef CONFIG_PROC_FS
2953 #define SEQ_START_TOKEN ((void *)1)
2955 static void *bond_info_seq_start(struct seq_file *seq, loff_t *pos)
2957 struct bonding *bond = seq->private;
2959 struct slave *slave;
2962 /* make sure the bond won't be taken away */
2963 read_lock(&dev_base_lock);
2964 read_lock_bh(&bond->lock);
2967 return SEQ_START_TOKEN;
2970 bond_for_each_slave(bond, slave, i) {
2971 if (++off == *pos) {
2979 static void *bond_info_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2981 struct bonding *bond = seq->private;
2982 struct slave *slave = v;
2985 if (v == SEQ_START_TOKEN) {
2986 return bond->first_slave;
2989 slave = slave->next;
2991 return (slave == bond->first_slave) ? NULL : slave;
2994 static void bond_info_seq_stop(struct seq_file *seq, void *v)
2996 struct bonding *bond = seq->private;
2998 read_unlock_bh(&bond->lock);
2999 read_unlock(&dev_base_lock);
3002 static void bond_info_show_master(struct seq_file *seq)
3004 struct bonding *bond = seq->private;
3009 read_lock(&bond->curr_slave_lock);
3010 curr = bond->curr_active_slave;
3011 read_unlock(&bond->curr_slave_lock);
3013 seq_printf(seq, "Bonding Mode: %s\n",
3014 bond_mode_name(bond->params.mode));
3016 if (bond->params.mode == BOND_MODE_XOR ||
3017 bond->params.mode == BOND_MODE_8023AD) {
3018 seq_printf(seq, "Transmit Hash Policy: %s (%d)\n",
3019 xmit_hashtype_tbl[bond->params.xmit_policy].modename,
3020 bond->params.xmit_policy);
3023 if (USES_PRIMARY(bond->params.mode)) {
3024 seq_printf(seq, "Primary Slave: %s\n",
3025 (bond->primary_slave) ?
3026 bond->primary_slave->dev->name : "None");
3028 seq_printf(seq, "Currently Active Slave: %s\n",
3029 (curr) ? curr->dev->name : "None");
3032 seq_printf(seq, "MII Status: %s\n", netif_carrier_ok(bond->dev) ?
3034 seq_printf(seq, "MII Polling Interval (ms): %d\n", bond->params.miimon);
3035 seq_printf(seq, "Up Delay (ms): %d\n",
3036 bond->params.updelay * bond->params.miimon);
3037 seq_printf(seq, "Down Delay (ms): %d\n",
3038 bond->params.downdelay * bond->params.miimon);
3041 /* ARP information */
3042 if(bond->params.arp_interval > 0) {
3044 seq_printf(seq, "ARP Polling Interval (ms): %d\n",
3045 bond->params.arp_interval);
3047 seq_printf(seq, "ARP IP target/s (n.n.n.n form):");
3049 for(i = 0; (i < BOND_MAX_ARP_TARGETS) ;i++) {
3050 if (!bond->params.arp_targets[i])
3053 seq_printf(seq, ",");
3054 target = ntohl(bond->params.arp_targets[i]);
3055 seq_printf(seq, " %d.%d.%d.%d", HIPQUAD(target));
3058 seq_printf(seq, "\n");
3061 if (bond->params.mode == BOND_MODE_8023AD) {
3062 struct ad_info ad_info;
3063 DECLARE_MAC_BUF(mac);
3065 seq_puts(seq, "\n802.3ad info\n");
3066 seq_printf(seq, "LACP rate: %s\n",
3067 (bond->params.lacp_fast) ? "fast" : "slow");
3069 if (bond_3ad_get_active_agg_info(bond, &ad_info)) {
3070 seq_printf(seq, "bond %s has no active aggregator\n",
3073 seq_printf(seq, "Active Aggregator Info:\n");
3075 seq_printf(seq, "\tAggregator ID: %d\n",
3076 ad_info.aggregator_id);
3077 seq_printf(seq, "\tNumber of ports: %d\n",
3079 seq_printf(seq, "\tActor Key: %d\n",
3081 seq_printf(seq, "\tPartner Key: %d\n",
3082 ad_info.partner_key);
3083 seq_printf(seq, "\tPartner Mac Address: %s\n",
3084 print_mac(mac, ad_info.partner_system));
3089 static void bond_info_show_slave(struct seq_file *seq, const struct slave *slave)
3091 struct bonding *bond = seq->private;
3092 DECLARE_MAC_BUF(mac);
3094 seq_printf(seq, "\nSlave Interface: %s\n", slave->dev->name);
3095 seq_printf(seq, "MII Status: %s\n",
3096 (slave->link == BOND_LINK_UP) ? "up" : "down");
3097 seq_printf(seq, "Link Failure Count: %u\n",
3098 slave->link_failure_count);
3101 "Permanent HW addr: %s\n",
3102 print_mac(mac, slave->perm_hwaddr));
3104 if (bond->params.mode == BOND_MODE_8023AD) {
3105 const struct aggregator *agg
3106 = SLAVE_AD_INFO(slave).port.aggregator;
3109 seq_printf(seq, "Aggregator ID: %d\n",
3110 agg->aggregator_identifier);
3112 seq_puts(seq, "Aggregator ID: N/A\n");
3117 static int bond_info_seq_show(struct seq_file *seq, void *v)
3119 if (v == SEQ_START_TOKEN) {
3120 seq_printf(seq, "%s\n", version);
3121 bond_info_show_master(seq);
3123 bond_info_show_slave(seq, v);
3129 static struct seq_operations bond_info_seq_ops = {
3130 .start = bond_info_seq_start,
3131 .next = bond_info_seq_next,
3132 .stop = bond_info_seq_stop,
3133 .show = bond_info_seq_show,
3136 static int bond_info_open(struct inode *inode, struct file *file)
3138 struct seq_file *seq;
3139 struct proc_dir_entry *proc;
3142 res = seq_open(file, &bond_info_seq_ops);
3144 /* recover the pointer buried in proc_dir_entry data */
3145 seq = file->private_data;
3147 seq->private = proc->data;
3153 static const struct file_operations bond_info_fops = {
3154 .owner = THIS_MODULE,
3155 .open = bond_info_open,
3157 .llseek = seq_lseek,
3158 .release = seq_release,
3161 static int bond_create_proc_entry(struct bonding *bond)
3163 struct net_device *bond_dev = bond->dev;
3165 if (bond_proc_dir) {
3166 bond->proc_entry = create_proc_entry(bond_dev->name,
3169 if (bond->proc_entry == NULL) {
3170 printk(KERN_WARNING DRV_NAME
3171 ": Warning: Cannot create /proc/net/%s/%s\n",
3172 DRV_NAME, bond_dev->name);
3174 bond->proc_entry->data = bond;
3175 bond->proc_entry->proc_fops = &bond_info_fops;
3176 bond->proc_entry->owner = THIS_MODULE;
3177 memcpy(bond->proc_file_name, bond_dev->name, IFNAMSIZ);
3184 static void bond_remove_proc_entry(struct bonding *bond)
3186 if (bond_proc_dir && bond->proc_entry) {
3187 remove_proc_entry(bond->proc_file_name, bond_proc_dir);
3188 memset(bond->proc_file_name, 0, IFNAMSIZ);
3189 bond->proc_entry = NULL;
3193 /* Create the bonding directory under /proc/net, if doesn't exist yet.
3194 * Caller must hold rtnl_lock.
3196 static void bond_create_proc_dir(void)
3198 int len = strlen(DRV_NAME);
3200 for (bond_proc_dir = init_net.proc_net->subdir; bond_proc_dir;
3201 bond_proc_dir = bond_proc_dir->next) {
3202 if ((bond_proc_dir->namelen == len) &&
3203 !memcmp(bond_proc_dir->name, DRV_NAME, len)) {
3208 if (!bond_proc_dir) {
3209 bond_proc_dir = proc_mkdir(DRV_NAME, init_net.proc_net);
3210 if (bond_proc_dir) {
3211 bond_proc_dir->owner = THIS_MODULE;
3213 printk(KERN_WARNING DRV_NAME
3214 ": Warning: cannot create /proc/net/%s\n",
3220 /* Destroy the bonding directory under /proc/net, if empty.
3221 * Caller must hold rtnl_lock.
3223 static void bond_destroy_proc_dir(void)
3225 struct proc_dir_entry *de;
3227 if (!bond_proc_dir) {
3231 /* verify that the /proc dir is empty */
3232 for (de = bond_proc_dir->subdir; de; de = de->next) {
3233 /* ignore . and .. */
3234 if (*(de->name) != '.') {
3240 if (bond_proc_dir->owner == THIS_MODULE) {
3241 bond_proc_dir->owner = NULL;
3244 remove_proc_entry(DRV_NAME, init_net.proc_net);
3245 bond_proc_dir = NULL;
3248 #endif /* CONFIG_PROC_FS */
3250 /*-------------------------- netdev event handling --------------------------*/
3253 * Change device name
3255 static int bond_event_changename(struct bonding *bond)
3257 #ifdef CONFIG_PROC_FS
3258 bond_remove_proc_entry(bond);
3259 bond_create_proc_entry(bond);
3261 down_write(&(bonding_rwsem));
3262 bond_destroy_sysfs_entry(bond);
3263 bond_create_sysfs_entry(bond);
3264 up_write(&(bonding_rwsem));
3268 static int bond_master_netdev_event(unsigned long event, struct net_device *bond_dev)
3270 struct bonding *event_bond = bond_dev->priv;
3273 case NETDEV_CHANGENAME:
3274 return bond_event_changename(event_bond);
3275 case NETDEV_UNREGISTER:
3277 * TODO: remove a bond from the list?
3287 static int bond_slave_netdev_event(unsigned long event, struct net_device *slave_dev)
3289 struct net_device *bond_dev = slave_dev->master;
3290 struct bonding *bond = bond_dev->priv;
3293 case NETDEV_UNREGISTER:
3295 bond_release(bond_dev, slave_dev);
3300 * TODO: is this what we get if somebody
3301 * sets up a hierarchical bond, then rmmod's
3302 * one of the slave bonding devices?
3307 * ... Or is it this?
3310 case NETDEV_CHANGEMTU:
3312 * TODO: Should slaves be allowed to
3313 * independently alter their MTU? For
3314 * an active-backup bond, slaves need
3315 * not be the same type of device, so
3316 * MTUs may vary. For other modes,
3317 * slaves arguably should have the
3318 * same MTUs. To do this, we'd need to
3319 * take over the slave's change_mtu
3320 * function for the duration of their
3324 case NETDEV_CHANGENAME:
3326 * TODO: handle changing the primary's name
3329 case NETDEV_FEAT_CHANGE:
3330 bond_compute_features(bond);
3340 * bond_netdev_event: handle netdev notifier chain events.
3342 * This function receives events for the netdev chain. The caller (an
3343 * ioctl handler calling blocking_notifier_call_chain) holds the necessary
3344 * locks for us to safely manipulate the slave devices (RTNL lock,
3347 static int bond_netdev_event(struct notifier_block *this, unsigned long event, void *ptr)
3349 struct net_device *event_dev = (struct net_device *)ptr;
3351 if (event_dev->nd_net != &init_net)
3354 dprintk("event_dev: %s, event: %lx\n",
3355 (event_dev ? event_dev->name : "None"),
3358 if (!(event_dev->priv_flags & IFF_BONDING))
3361 if (event_dev->flags & IFF_MASTER) {
3362 dprintk("IFF_MASTER\n");
3363 return bond_master_netdev_event(event, event_dev);
3366 if (event_dev->flags & IFF_SLAVE) {
3367 dprintk("IFF_SLAVE\n");
3368 return bond_slave_netdev_event(event, event_dev);
3375 * bond_inetaddr_event: handle inetaddr notifier chain events.
3377 * We keep track of device IPs primarily to use as source addresses in
3378 * ARP monitor probes (rather than spewing out broadcasts all the time).
3380 * We track one IP for the main device (if it has one), plus one per VLAN.
3382 static int bond_inetaddr_event(struct notifier_block *this, unsigned long event, void *ptr)
3384 struct in_ifaddr *ifa = ptr;
3385 struct net_device *vlan_dev, *event_dev = ifa->ifa_dev->dev;
3386 struct bonding *bond, *bond_next;
3387 struct vlan_entry *vlan, *vlan_next;
3389 list_for_each_entry_safe(bond, bond_next, &bond_dev_list, bond_list) {
3390 if (bond->dev == event_dev) {
3393 bond->master_ip = ifa->ifa_local;
3396 bond->master_ip = bond_glean_dev_ip(bond->dev);
3403 if (list_empty(&bond->vlan_list))
3406 list_for_each_entry_safe(vlan, vlan_next, &bond->vlan_list,
3408 vlan_dev = vlan_group_get_device(bond->vlgrp, vlan->vlan_id);
3409 if (vlan_dev == event_dev) {
3412 vlan->vlan_ip = ifa->ifa_local;
3416 bond_glean_dev_ip(vlan_dev);
3427 static struct notifier_block bond_netdev_notifier = {
3428 .notifier_call = bond_netdev_event,
3431 static struct notifier_block bond_inetaddr_notifier = {
3432 .notifier_call = bond_inetaddr_event,
3435 /*-------------------------- Packet type handling ---------------------------*/
3437 /* register to receive lacpdus on a bond */
3438 static void bond_register_lacpdu(struct bonding *bond)
3440 struct packet_type *pk_type = &(BOND_AD_INFO(bond).ad_pkt_type);
3442 /* initialize packet type */
3443 pk_type->type = PKT_TYPE_LACPDU;
3444 pk_type->dev = bond->dev;
3445 pk_type->func = bond_3ad_lacpdu_recv;
3447 dev_add_pack(pk_type);
3450 /* unregister to receive lacpdus on a bond */
3451 static void bond_unregister_lacpdu(struct bonding *bond)
3453 dev_remove_pack(&(BOND_AD_INFO(bond).ad_pkt_type));
3456 void bond_register_arp(struct bonding *bond)
3458 struct packet_type *pt = &bond->arp_mon_pt;
3463 pt->type = htons(ETH_P_ARP);
3464 pt->dev = bond->dev;
3465 pt->func = bond_arp_rcv;
3469 void bond_unregister_arp(struct bonding *bond)
3471 struct packet_type *pt = &bond->arp_mon_pt;
3473 dev_remove_pack(pt);
3477 /*---------------------------- Hashing Policies -----------------------------*/
3480 * Hash for the output device based upon layer 3 and layer 4 data. If
3481 * the packet is a frag or not TCP or UDP, just use layer 3 data. If it is
3482 * altogether not IP, mimic bond_xmit_hash_policy_l2()
3484 static int bond_xmit_hash_policy_l34(struct sk_buff *skb,
3485 struct net_device *bond_dev, int count)
3487 struct ethhdr *data = (struct ethhdr *)skb->data;
3488 struct iphdr *iph = ip_hdr(skb);
3489 __be16 *layer4hdr = (__be16 *)((u32 *)iph + iph->ihl);
3492 if (skb->protocol == __constant_htons(ETH_P_IP)) {
3493 if (!(iph->frag_off & __constant_htons(IP_MF|IP_OFFSET)) &&
3494 (iph->protocol == IPPROTO_TCP ||
3495 iph->protocol == IPPROTO_UDP)) {
3496 layer4_xor = ntohs((*layer4hdr ^ *(layer4hdr + 1)));
3498 return (layer4_xor ^
3499 ((ntohl(iph->saddr ^ iph->daddr)) & 0xffff)) % count;
3503 return (data->h_dest[5] ^ bond_dev->dev_addr[5]) % count;
3507 * Hash for the output device based upon layer 2 data
3509 static int bond_xmit_hash_policy_l2(struct sk_buff *skb,
3510 struct net_device *bond_dev, int count)
3512 struct ethhdr *data = (struct ethhdr *)skb->data;
3514 return (data->h_dest[5] ^ bond_dev->dev_addr[5]) % count;
3517 /*-------------------------- Device entry points ----------------------------*/
3519 static int bond_open(struct net_device *bond_dev)
3521 struct bonding *bond = bond_dev->priv;
3522 struct timer_list *mii_timer = &bond->mii_timer;
3523 struct timer_list *arp_timer = &bond->arp_timer;
3525 bond->kill_timers = 0;
3527 if ((bond->params.mode == BOND_MODE_TLB) ||
3528 (bond->params.mode == BOND_MODE_ALB)) {
3529 struct timer_list *alb_timer = &(BOND_ALB_INFO(bond).alb_timer);
3531 /* bond_alb_initialize must be called before the timer
3534 if (bond_alb_initialize(bond, (bond->params.mode == BOND_MODE_ALB))) {
3535 /* something went wrong - fail the open operation */
3539 init_timer(alb_timer);
3540 alb_timer->expires = jiffies + 1;
3541 alb_timer->data = (unsigned long)bond;
3542 alb_timer->function = (void *)&bond_alb_monitor;
3543 add_timer(alb_timer);
3546 if (bond->params.miimon) { /* link check interval, in milliseconds. */
3547 init_timer(mii_timer);
3548 mii_timer->expires = jiffies + 1;
3549 mii_timer->data = (unsigned long)bond_dev;
3550 mii_timer->function = (void *)&bond_mii_monitor;
3551 add_timer(mii_timer);
3554 if (bond->params.arp_interval) { /* arp interval, in milliseconds. */
3555 init_timer(arp_timer);
3556 arp_timer->expires = jiffies + 1;
3557 arp_timer->data = (unsigned long)bond_dev;
3558 if (bond->params.mode == BOND_MODE_ACTIVEBACKUP) {
3559 arp_timer->function = (void *)&bond_activebackup_arp_mon;
3561 arp_timer->function = (void *)&bond_loadbalance_arp_mon;
3563 if (bond->params.arp_validate)
3564 bond_register_arp(bond);
3566 add_timer(arp_timer);
3569 if (bond->params.mode == BOND_MODE_8023AD) {
3570 struct timer_list *ad_timer = &(BOND_AD_INFO(bond).ad_timer);
3571 init_timer(ad_timer);
3572 ad_timer->expires = jiffies + 1;
3573 ad_timer->data = (unsigned long)bond;
3574 ad_timer->function = (void *)&bond_3ad_state_machine_handler;
3575 add_timer(ad_timer);
3577 /* register to receive LACPDUs */
3578 bond_register_lacpdu(bond);
3584 static int bond_close(struct net_device *bond_dev)
3586 struct bonding *bond = bond_dev->priv;
3588 if (bond->params.mode == BOND_MODE_8023AD) {
3589 /* Unregister the receive of LACPDUs */
3590 bond_unregister_lacpdu(bond);
3593 if (bond->params.arp_validate)
3594 bond_unregister_arp(bond);
3596 write_lock_bh(&bond->lock);
3599 /* signal timers not to re-arm */
3600 bond->kill_timers = 1;
3602 write_unlock_bh(&bond->lock);
3604 /* del_timer_sync must run without holding the bond->lock
3605 * because a running timer might be trying to hold it too
3608 if (bond->params.miimon) { /* link check interval, in milliseconds. */
3609 del_timer_sync(&bond->mii_timer);
3612 if (bond->params.arp_interval) { /* arp interval, in milliseconds. */
3613 del_timer_sync(&bond->arp_timer);
3616 switch (bond->params.mode) {
3617 case BOND_MODE_8023AD:
3618 del_timer_sync(&(BOND_AD_INFO(bond).ad_timer));
3622 del_timer_sync(&(BOND_ALB_INFO(bond).alb_timer));
3629 if ((bond->params.mode == BOND_MODE_TLB) ||
3630 (bond->params.mode == BOND_MODE_ALB)) {
3631 /* Must be called only after all
3632 * slaves have been released
3634 bond_alb_deinitialize(bond);
3640 static struct net_device_stats *bond_get_stats(struct net_device *bond_dev)
3642 struct bonding *bond = bond_dev->priv;
3643 struct net_device_stats *stats = &(bond->stats), *sstats;
3644 struct slave *slave;
3647 memset(stats, 0, sizeof(struct net_device_stats));
3649 read_lock_bh(&bond->lock);
3651 bond_for_each_slave(bond, slave, i) {
3652 sstats = slave->dev->get_stats(slave->dev);
3653 stats->rx_packets += sstats->rx_packets;
3654 stats->rx_bytes += sstats->rx_bytes;
3655 stats->rx_errors += sstats->rx_errors;
3656 stats->rx_dropped += sstats->rx_dropped;
3658 stats->tx_packets += sstats->tx_packets;
3659 stats->tx_bytes += sstats->tx_bytes;
3660 stats->tx_errors += sstats->tx_errors;
3661 stats->tx_dropped += sstats->tx_dropped;
3663 stats->multicast += sstats->multicast;
3664 stats->collisions += sstats->collisions;
3666 stats->rx_length_errors += sstats->rx_length_errors;
3667 stats->rx_over_errors += sstats->rx_over_errors;
3668 stats->rx_crc_errors += sstats->rx_crc_errors;
3669 stats->rx_frame_errors += sstats->rx_frame_errors;
3670 stats->rx_fifo_errors += sstats->rx_fifo_errors;
3671 stats->rx_missed_errors += sstats->rx_missed_errors;
3673 stats->tx_aborted_errors += sstats->tx_aborted_errors;
3674 stats->tx_carrier_errors += sstats->tx_carrier_errors;
3675 stats->tx_fifo_errors += sstats->tx_fifo_errors;
3676 stats->tx_heartbeat_errors += sstats->tx_heartbeat_errors;
3677 stats->tx_window_errors += sstats->tx_window_errors;
3680 read_unlock_bh(&bond->lock);
3685 static int bond_do_ioctl(struct net_device *bond_dev, struct ifreq *ifr, int cmd)
3687 struct net_device *slave_dev = NULL;
3688 struct ifbond k_binfo;
3689 struct ifbond __user *u_binfo = NULL;
3690 struct ifslave k_sinfo;
3691 struct ifslave __user *u_sinfo = NULL;
3692 struct mii_ioctl_data *mii = NULL;
3695 dprintk("bond_ioctl: master=%s, cmd=%d\n",
3696 bond_dev->name, cmd);
3708 * We do this again just in case we were called by SIOCGMIIREG
3709 * instead of SIOCGMIIPHY.
3716 if (mii->reg_num == 1) {
3717 struct bonding *bond = bond_dev->priv;
3719 read_lock_bh(&bond->lock);
3720 read_lock(&bond->curr_slave_lock);
3721 if (netif_carrier_ok(bond->dev)) {
3722 mii->val_out = BMSR_LSTATUS;
3724 read_unlock(&bond->curr_slave_lock);
3725 read_unlock_bh(&bond->lock);
3729 case BOND_INFO_QUERY_OLD:
3730 case SIOCBONDINFOQUERY:
3731 u_binfo = (struct ifbond __user *)ifr->ifr_data;
3733 if (copy_from_user(&k_binfo, u_binfo, sizeof(ifbond))) {
3737 res = bond_info_query(bond_dev, &k_binfo);
3739 if (copy_to_user(u_binfo, &k_binfo, sizeof(ifbond))) {
3745 case BOND_SLAVE_INFO_QUERY_OLD:
3746 case SIOCBONDSLAVEINFOQUERY:
3747 u_sinfo = (struct ifslave __user *)ifr->ifr_data;
3749 if (copy_from_user(&k_sinfo, u_sinfo, sizeof(ifslave))) {
3753 res = bond_slave_info_query(bond_dev, &k_sinfo);
3755 if (copy_to_user(u_sinfo, &k_sinfo, sizeof(ifslave))) {
3766 if (!capable(CAP_NET_ADMIN)) {
3770 down_write(&(bonding_rwsem));
3771 slave_dev = dev_get_by_name(&init_net, ifr->ifr_slave);
3773 dprintk("slave_dev=%p: \n", slave_dev);
3778 dprintk("slave_dev->name=%s: \n", slave_dev->name);
3780 case BOND_ENSLAVE_OLD:
3781 case SIOCBONDENSLAVE:
3782 res = bond_enslave(bond_dev, slave_dev);
3784 case BOND_RELEASE_OLD:
3785 case SIOCBONDRELEASE:
3786 res = bond_release(bond_dev, slave_dev);
3788 case BOND_SETHWADDR_OLD:
3789 case SIOCBONDSETHWADDR:
3790 res = bond_sethwaddr(bond_dev, slave_dev);
3792 case BOND_CHANGE_ACTIVE_OLD:
3793 case SIOCBONDCHANGEACTIVE:
3794 res = bond_ioctl_change_active(bond_dev, slave_dev);
3803 up_write(&(bonding_rwsem));
3807 static void bond_set_multicast_list(struct net_device *bond_dev)
3809 struct bonding *bond = bond_dev->priv;
3810 struct dev_mc_list *dmi;
3812 write_lock_bh(&bond->lock);
3815 * Do promisc before checking multicast_mode
3817 if ((bond_dev->flags & IFF_PROMISC) && !(bond->flags & IFF_PROMISC)) {
3818 bond_set_promiscuity(bond, 1);
3821 if (!(bond_dev->flags & IFF_PROMISC) && (bond->flags & IFF_PROMISC)) {
3822 bond_set_promiscuity(bond, -1);
3825 /* set allmulti flag to slaves */
3826 if ((bond_dev->flags & IFF_ALLMULTI) && !(bond->flags & IFF_ALLMULTI)) {
3827 bond_set_allmulti(bond, 1);
3830 if (!(bond_dev->flags & IFF_ALLMULTI) && (bond->flags & IFF_ALLMULTI)) {
3831 bond_set_allmulti(bond, -1);
3834 bond->flags = bond_dev->flags;
3836 /* looking for addresses to add to slaves' mc list */
3837 for (dmi = bond_dev->mc_list; dmi; dmi = dmi->next) {
3838 if (!bond_mc_list_find_dmi(dmi, bond->mc_list)) {
3839 bond_mc_add(bond, dmi->dmi_addr, dmi->dmi_addrlen);
3843 /* looking for addresses to delete from slaves' list */
3844 for (dmi = bond->mc_list; dmi; dmi = dmi->next) {
3845 if (!bond_mc_list_find_dmi(dmi, bond_dev->mc_list)) {
3846 bond_mc_delete(bond, dmi->dmi_addr, dmi->dmi_addrlen);
3850 /* save master's multicast list */
3851 bond_mc_list_destroy(bond);
3852 bond_mc_list_copy(bond_dev->mc_list, bond, GFP_ATOMIC);
3854 write_unlock_bh(&bond->lock);
3858 * Change the MTU of all of a master's slaves to match the master
3860 static int bond_change_mtu(struct net_device *bond_dev, int new_mtu)
3862 struct bonding *bond = bond_dev->priv;
3863 struct slave *slave, *stop_at;
3867 dprintk("bond=%p, name=%s, new_mtu=%d\n", bond,
3868 (bond_dev ? bond_dev->name : "None"), new_mtu);
3870 /* Can't hold bond->lock with bh disabled here since
3871 * some base drivers panic. On the other hand we can't
3872 * hold bond->lock without bh disabled because we'll
3873 * deadlock. The only solution is to rely on the fact
3874 * that we're under rtnl_lock here, and the slaves
3875 * list won't change. This doesn't solve the problem
3876 * of setting the slave's MTU while it is
3877 * transmitting, but the assumption is that the base
3878 * driver can handle that.
3880 * TODO: figure out a way to safely iterate the slaves
3881 * list, but without holding a lock around the actual
3882 * call to the base driver.
3885 bond_for_each_slave(bond, slave, i) {
3886 dprintk("s %p s->p %p c_m %p\n", slave,
3887 slave->prev, slave->dev->change_mtu);
3889 res = dev_set_mtu(slave->dev, new_mtu);
3892 /* If we failed to set the slave's mtu to the new value
3893 * we must abort the operation even in ACTIVE_BACKUP
3894 * mode, because if we allow the backup slaves to have
3895 * different mtu values than the active slave we'll
3896 * need to change their mtu when doing a failover. That
3897 * means changing their mtu from timer context, which
3898 * is probably not a good idea.
3900 dprintk("err %d %s\n", res, slave->dev->name);
3905 bond_dev->mtu = new_mtu;
3910 /* unwind from head to the slave that failed */
3912 bond_for_each_slave_from_to(bond, slave, i, bond->first_slave, stop_at) {
3915 tmp_res = dev_set_mtu(slave->dev, bond_dev->mtu);
3917 dprintk("unwind err %d dev %s\n", tmp_res,
3928 * Note that many devices must be down to change the HW address, and
3929 * downing the master releases all slaves. We can make bonds full of
3930 * bonding devices to test this, however.
3932 static int bond_set_mac_address(struct net_device *bond_dev, void *addr)
3934 struct bonding *bond = bond_dev->priv;
3935 struct sockaddr *sa = addr, tmp_sa;
3936 struct slave *slave, *stop_at;
3940 dprintk("bond=%p, name=%s\n", bond, (bond_dev ? bond_dev->name : "None"));
3942 if (!bond->do_set_mac_addr)
3945 if (!is_valid_ether_addr(sa->sa_data)) {
3946 return -EADDRNOTAVAIL;
3949 /* Can't hold bond->lock with bh disabled here since
3950 * some base drivers panic. On the other hand we can't
3951 * hold bond->lock without bh disabled because we'll
3952 * deadlock. The only solution is to rely on the fact
3953 * that we're under rtnl_lock here, and the slaves
3954 * list won't change. This doesn't solve the problem
3955 * of setting the slave's hw address while it is
3956 * transmitting, but the assumption is that the base
3957 * driver can handle that.
3959 * TODO: figure out a way to safely iterate the slaves
3960 * list, but without holding a lock around the actual
3961 * call to the base driver.
3964 bond_for_each_slave(bond, slave, i) {
3965 dprintk("slave %p %s\n", slave, slave->dev->name);
3967 if (slave->dev->set_mac_address == NULL) {
3969 dprintk("EOPNOTSUPP %s\n", slave->dev->name);
3973 res = dev_set_mac_address(slave->dev, addr);
3975 /* TODO: consider downing the slave
3977 * User should expect communications
3978 * breakage anyway until ARP finish
3981 dprintk("err %d %s\n", res, slave->dev->name);
3987 memcpy(bond_dev->dev_addr, sa->sa_data, bond_dev->addr_len);
3991 memcpy(tmp_sa.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
3992 tmp_sa.sa_family = bond_dev->type;
3994 /* unwind from head to the slave that failed */
3996 bond_for_each_slave_from_to(bond, slave, i, bond->first_slave, stop_at) {
3999 tmp_res = dev_set_mac_address(slave->dev, &tmp_sa);
4001 dprintk("unwind err %d dev %s\n", tmp_res,
4009 static int bond_xmit_roundrobin(struct sk_buff *skb, struct net_device *bond_dev)
4011 struct bonding *bond = bond_dev->priv;
4012 struct slave *slave, *start_at;
4016 read_lock(&bond->lock);
4018 if (!BOND_IS_OK(bond)) {
4022 read_lock(&bond->curr_slave_lock);
4023 slave = start_at = bond->curr_active_slave;
4024 read_unlock(&bond->curr_slave_lock);
4030 bond_for_each_slave_from(bond, slave, i, start_at) {
4031 if (IS_UP(slave->dev) &&
4032 (slave->link == BOND_LINK_UP) &&
4033 (slave->state == BOND_STATE_ACTIVE)) {
4034 res = bond_dev_queue_xmit(bond, skb, slave->dev);
4036 write_lock(&bond->curr_slave_lock);
4037 bond->curr_active_slave = slave->next;
4038 write_unlock(&bond->curr_slave_lock);
4047 /* no suitable interface, frame not sent */
4050 read_unlock(&bond->lock);
4056 * in active-backup mode, we know that bond->curr_active_slave is always valid if
4057 * the bond has a usable interface.
4059 static int bond_xmit_activebackup(struct sk_buff *skb, struct net_device *bond_dev)
4061 struct bonding *bond = bond_dev->priv;
4064 read_lock(&bond->lock);
4065 read_lock(&bond->curr_slave_lock);
4067 if (!BOND_IS_OK(bond)) {
4071 if (!bond->curr_active_slave)
4074 res = bond_dev_queue_xmit(bond, skb, bond->curr_active_slave->dev);
4078 /* no suitable interface, frame not sent */
4081 read_unlock(&bond->curr_slave_lock);
4082 read_unlock(&bond->lock);
4087 * In bond_xmit_xor() , we determine the output device by using a pre-
4088 * determined xmit_hash_policy(), If the selected device is not enabled,
4089 * find the next active slave.
4091 static int bond_xmit_xor(struct sk_buff *skb, struct net_device *bond_dev)
4093 struct bonding *bond = bond_dev->priv;
4094 struct slave *slave, *start_at;
4099 read_lock(&bond->lock);
4101 if (!BOND_IS_OK(bond)) {
4105 slave_no = bond->xmit_hash_policy(skb, bond_dev, bond->slave_cnt);
4107 bond_for_each_slave(bond, slave, i) {
4116 bond_for_each_slave_from(bond, slave, i, start_at) {
4117 if (IS_UP(slave->dev) &&
4118 (slave->link == BOND_LINK_UP) &&
4119 (slave->state == BOND_STATE_ACTIVE)) {
4120 res = bond_dev_queue_xmit(bond, skb, slave->dev);
4127 /* no suitable interface, frame not sent */
4130 read_unlock(&bond->lock);
4135 * in broadcast mode, we send everything to all usable interfaces.
4137 static int bond_xmit_broadcast(struct sk_buff *skb, struct net_device *bond_dev)
4139 struct bonding *bond = bond_dev->priv;
4140 struct slave *slave, *start_at;
4141 struct net_device *tx_dev = NULL;
4145 read_lock(&bond->lock);
4147 if (!BOND_IS_OK(bond)) {
4151 read_lock(&bond->curr_slave_lock);
4152 start_at = bond->curr_active_slave;
4153 read_unlock(&bond->curr_slave_lock);
4159 bond_for_each_slave_from(bond, slave, i, start_at) {
4160 if (IS_UP(slave->dev) &&
4161 (slave->link == BOND_LINK_UP) &&
4162 (slave->state == BOND_STATE_ACTIVE)) {
4164 struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
4166 printk(KERN_ERR DRV_NAME
4167 ": %s: Error: bond_xmit_broadcast(): "
4168 "skb_clone() failed\n",
4173 res = bond_dev_queue_xmit(bond, skb2, tx_dev);
4175 dev_kfree_skb(skb2);
4179 tx_dev = slave->dev;
4184 res = bond_dev_queue_xmit(bond, skb, tx_dev);
4189 /* no suitable interface, frame not sent */
4192 /* frame sent to all suitable interfaces */
4193 read_unlock(&bond->lock);
4197 /*------------------------- Device initialization ---------------------------*/
4200 * set bond mode specific net device operations
4202 void bond_set_mode_ops(struct bonding *bond, int mode)
4204 struct net_device *bond_dev = bond->dev;
4207 case BOND_MODE_ROUNDROBIN:
4208 bond_dev->hard_start_xmit = bond_xmit_roundrobin;
4210 case BOND_MODE_ACTIVEBACKUP:
4211 bond_dev->hard_start_xmit = bond_xmit_activebackup;
4214 bond_dev->hard_start_xmit = bond_xmit_xor;
4215 if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER34)
4216 bond->xmit_hash_policy = bond_xmit_hash_policy_l34;
4218 bond->xmit_hash_policy = bond_xmit_hash_policy_l2;
4220 case BOND_MODE_BROADCAST:
4221 bond_dev->hard_start_xmit = bond_xmit_broadcast;
4223 case BOND_MODE_8023AD:
4224 bond_set_master_3ad_flags(bond);
4225 bond_dev->hard_start_xmit = bond_3ad_xmit_xor;
4226 if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER34)
4227 bond->xmit_hash_policy = bond_xmit_hash_policy_l34;
4229 bond->xmit_hash_policy = bond_xmit_hash_policy_l2;
4232 bond_set_master_alb_flags(bond);
4235 bond_dev->hard_start_xmit = bond_alb_xmit;
4236 bond_dev->set_mac_address = bond_alb_set_mac_address;
4239 /* Should never happen, mode already checked */
4240 printk(KERN_ERR DRV_NAME
4241 ": %s: Error: Unknown bonding mode %d\n",
4248 static void bond_ethtool_get_drvinfo(struct net_device *bond_dev,
4249 struct ethtool_drvinfo *drvinfo)
4251 strncpy(drvinfo->driver, DRV_NAME, 32);
4252 strncpy(drvinfo->version, DRV_VERSION, 32);
4253 snprintf(drvinfo->fw_version, 32, "%d", BOND_ABI_VERSION);
4256 static const struct ethtool_ops bond_ethtool_ops = {
4257 .get_drvinfo = bond_ethtool_get_drvinfo,
4261 * Does not allocate but creates a /proc entry.
4264 static int bond_init(struct net_device *bond_dev, struct bond_params *params)
4266 struct bonding *bond = bond_dev->priv;
4268 dprintk("Begin bond_init for %s\n", bond_dev->name);
4270 /* initialize rwlocks */
4271 rwlock_init(&bond->lock);
4272 rwlock_init(&bond->curr_slave_lock);
4274 bond->params = *params; /* copy params struct */
4276 /* Initialize pointers */
4277 bond->first_slave = NULL;
4278 bond->curr_active_slave = NULL;
4279 bond->current_arp_slave = NULL;
4280 bond->primary_slave = NULL;
4281 bond->dev = bond_dev;
4282 INIT_LIST_HEAD(&bond->vlan_list);
4284 /* Initialize the device entry points */
4285 bond_dev->open = bond_open;
4286 bond_dev->stop = bond_close;
4287 bond_dev->get_stats = bond_get_stats;
4288 bond_dev->do_ioctl = bond_do_ioctl;
4289 bond_dev->ethtool_ops = &bond_ethtool_ops;
4290 bond_dev->set_multicast_list = bond_set_multicast_list;
4291 bond_dev->change_mtu = bond_change_mtu;
4292 bond_dev->set_mac_address = bond_set_mac_address;
4294 bond_set_mode_ops(bond, bond->params.mode);
4296 bond_dev->destructor = free_netdev;
4298 /* Initialize the device options */
4299 bond_dev->tx_queue_len = 0;
4300 bond_dev->flags |= IFF_MASTER|IFF_MULTICAST;
4301 bond_dev->priv_flags |= IFF_BONDING;
4303 /* At first, we block adding VLANs. That's the only way to
4304 * prevent problems that occur when adding VLANs over an
4305 * empty bond. The block will be removed once non-challenged
4306 * slaves are enslaved.
4308 bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
4310 /* don't acquire bond device's netif_tx_lock when
4312 bond_dev->features |= NETIF_F_LLTX;
4314 /* By default, we declare the bond to be fully
4315 * VLAN hardware accelerated capable. Special
4316 * care is taken in the various xmit functions
4317 * when there are slaves that are not hw accel
4320 bond_dev->vlan_rx_register = bond_vlan_rx_register;
4321 bond_dev->vlan_rx_add_vid = bond_vlan_rx_add_vid;
4322 bond_dev->vlan_rx_kill_vid = bond_vlan_rx_kill_vid;
4323 bond_dev->features |= (NETIF_F_HW_VLAN_TX |
4324 NETIF_F_HW_VLAN_RX |
4325 NETIF_F_HW_VLAN_FILTER);
4327 #ifdef CONFIG_PROC_FS
4328 bond_create_proc_entry(bond);
4331 /* set do_set_mac_addr to true on startup */
4332 bond->do_set_mac_addr = 1;
4334 list_add_tail(&bond->bond_list, &bond_dev_list);
4339 /* De-initialize device specific data.
4340 * Caller must hold rtnl_lock.
4342 void bond_deinit(struct net_device *bond_dev)
4344 struct bonding *bond = bond_dev->priv;
4346 list_del(&bond->bond_list);
4348 #ifdef CONFIG_PROC_FS
4349 bond_remove_proc_entry(bond);
4353 /* Unregister and free all bond devices.
4354 * Caller must hold rtnl_lock.
4356 static void bond_free_all(void)
4358 struct bonding *bond, *nxt;
4360 list_for_each_entry_safe(bond, nxt, &bond_dev_list, bond_list) {
4361 struct net_device *bond_dev = bond->dev;
4363 bond_mc_list_destroy(bond);
4364 /* Release the bonded slaves */
4365 bond_release_all(bond_dev);
4366 bond_deinit(bond_dev);
4367 unregister_netdevice(bond_dev);
4370 #ifdef CONFIG_PROC_FS
4371 bond_destroy_proc_dir();
4375 /*------------------------- Module initialization ---------------------------*/
4378 * Convert string input module parms. Accept either the
4379 * number of the mode or its string name.
4381 int bond_parse_parm(char *mode_arg, struct bond_parm_tbl *tbl)
4385 for (i = 0; tbl[i].modename; i++) {
4386 if ((isdigit(*mode_arg) &&
4387 tbl[i].mode == simple_strtol(mode_arg, NULL, 0)) ||
4388 (strncmp(mode_arg, tbl[i].modename,
4389 strlen(tbl[i].modename)) == 0)) {
4397 static int bond_check_params(struct bond_params *params)
4399 int arp_validate_value;
4402 * Convert string parameters.
4405 bond_mode = bond_parse_parm(mode, bond_mode_tbl);
4406 if (bond_mode == -1) {
4407 printk(KERN_ERR DRV_NAME
4408 ": Error: Invalid bonding mode \"%s\"\n",
4409 mode == NULL ? "NULL" : mode);
4414 if (xmit_hash_policy) {
4415 if ((bond_mode != BOND_MODE_XOR) &&
4416 (bond_mode != BOND_MODE_8023AD)) {
4417 printk(KERN_INFO DRV_NAME
4418 ": xor_mode param is irrelevant in mode %s\n",
4419 bond_mode_name(bond_mode));
4421 xmit_hashtype = bond_parse_parm(xmit_hash_policy,
4423 if (xmit_hashtype == -1) {
4424 printk(KERN_ERR DRV_NAME
4425 ": Error: Invalid xmit_hash_policy \"%s\"\n",
4426 xmit_hash_policy == NULL ? "NULL" :
4434 if (bond_mode != BOND_MODE_8023AD) {
4435 printk(KERN_INFO DRV_NAME
4436 ": lacp_rate param is irrelevant in mode %s\n",
4437 bond_mode_name(bond_mode));
4439 lacp_fast = bond_parse_parm(lacp_rate, bond_lacp_tbl);
4440 if (lacp_fast == -1) {
4441 printk(KERN_ERR DRV_NAME
4442 ": Error: Invalid lacp rate \"%s\"\n",
4443 lacp_rate == NULL ? "NULL" : lacp_rate);
4449 if (max_bonds < 1 || max_bonds > INT_MAX) {
4450 printk(KERN_WARNING DRV_NAME
4451 ": Warning: max_bonds (%d) not in range %d-%d, so it "
4452 "was reset to BOND_DEFAULT_MAX_BONDS (%d)\n",
4453 max_bonds, 1, INT_MAX, BOND_DEFAULT_MAX_BONDS);
4454 max_bonds = BOND_DEFAULT_MAX_BONDS;
4458 printk(KERN_WARNING DRV_NAME
4459 ": Warning: miimon module parameter (%d), "
4460 "not in range 0-%d, so it was reset to %d\n",
4461 miimon, INT_MAX, BOND_LINK_MON_INTERV);
4462 miimon = BOND_LINK_MON_INTERV;
4466 printk(KERN_WARNING DRV_NAME
4467 ": Warning: updelay module parameter (%d), "
4468 "not in range 0-%d, so it was reset to 0\n",
4473 if (downdelay < 0) {
4474 printk(KERN_WARNING DRV_NAME
4475 ": Warning: downdelay module parameter (%d), "
4476 "not in range 0-%d, so it was reset to 0\n",
4477 downdelay, INT_MAX);
4481 if ((use_carrier != 0) && (use_carrier != 1)) {
4482 printk(KERN_WARNING DRV_NAME
4483 ": Warning: use_carrier module parameter (%d), "
4484 "not of valid value (0/1), so it was set to 1\n",
4489 /* reset values for 802.3ad */
4490 if (bond_mode == BOND_MODE_8023AD) {
4492 printk(KERN_WARNING DRV_NAME
4493 ": Warning: miimon must be specified, "
4494 "otherwise bonding will not detect link "
4495 "failure, speed and duplex which are "
4496 "essential for 802.3ad operation\n");
4497 printk(KERN_WARNING "Forcing miimon to 100msec\n");
4502 /* reset values for TLB/ALB */
4503 if ((bond_mode == BOND_MODE_TLB) ||
4504 (bond_mode == BOND_MODE_ALB)) {
4506 printk(KERN_WARNING DRV_NAME
4507 ": Warning: miimon must be specified, "
4508 "otherwise bonding will not detect link "
4509 "failure and link speed which are essential "
4510 "for TLB/ALB load balancing\n");
4511 printk(KERN_WARNING "Forcing miimon to 100msec\n");
4516 if (bond_mode == BOND_MODE_ALB) {
4517 printk(KERN_NOTICE DRV_NAME
4518 ": In ALB mode you might experience client "
4519 "disconnections upon reconnection of a link if the "
4520 "bonding module updelay parameter (%d msec) is "
4521 "incompatible with the forwarding delay time of the "
4527 if (updelay || downdelay) {
4528 /* just warn the user the up/down delay will have
4529 * no effect since miimon is zero...
4531 printk(KERN_WARNING DRV_NAME
4532 ": Warning: miimon module parameter not set "
4533 "and updelay (%d) or downdelay (%d) module "
4534 "parameter is set; updelay and downdelay have "
4535 "no effect unless miimon is set\n",
4536 updelay, downdelay);
4539 /* don't allow arp monitoring */
4541 printk(KERN_WARNING DRV_NAME
4542 ": Warning: miimon (%d) and arp_interval (%d) "
4543 "can't be used simultaneously, disabling ARP "
4545 miimon, arp_interval);
4549 if ((updelay % miimon) != 0) {
4550 printk(KERN_WARNING DRV_NAME
4551 ": Warning: updelay (%d) is not a multiple "
4552 "of miimon (%d), updelay rounded to %d ms\n",
4553 updelay, miimon, (updelay / miimon) * miimon);
4558 if ((downdelay % miimon) != 0) {
4559 printk(KERN_WARNING DRV_NAME
4560 ": Warning: downdelay (%d) is not a multiple "
4561 "of miimon (%d), downdelay rounded to %d ms\n",
4563 (downdelay / miimon) * miimon);
4566 downdelay /= miimon;
4569 if (arp_interval < 0) {
4570 printk(KERN_WARNING DRV_NAME
4571 ": Warning: arp_interval module parameter (%d) "
4572 ", not in range 0-%d, so it was reset to %d\n",
4573 arp_interval, INT_MAX, BOND_LINK_ARP_INTERV);
4574 arp_interval = BOND_LINK_ARP_INTERV;
4577 for (arp_ip_count = 0;
4578 (arp_ip_count < BOND_MAX_ARP_TARGETS) && arp_ip_target[arp_ip_count];
4580 /* not complete check, but should be good enough to
4582 if (!isdigit(arp_ip_target[arp_ip_count][0])) {
4583 printk(KERN_WARNING DRV_NAME
4584 ": Warning: bad arp_ip_target module parameter "
4585 "(%s), ARP monitoring will not be performed\n",
4586 arp_ip_target[arp_ip_count]);
4589 __be32 ip = in_aton(arp_ip_target[arp_ip_count]);
4590 arp_target[arp_ip_count] = ip;
4594 if (arp_interval && !arp_ip_count) {
4595 /* don't allow arping if no arp_ip_target given... */
4596 printk(KERN_WARNING DRV_NAME
4597 ": Warning: arp_interval module parameter (%d) "
4598 "specified without providing an arp_ip_target "
4599 "parameter, arp_interval was reset to 0\n",
4605 if (bond_mode != BOND_MODE_ACTIVEBACKUP) {
4606 printk(KERN_ERR DRV_NAME
4607 ": arp_validate only supported in active-backup mode\n");
4610 if (!arp_interval) {
4611 printk(KERN_ERR DRV_NAME
4612 ": arp_validate requires arp_interval\n");
4616 arp_validate_value = bond_parse_parm(arp_validate,
4618 if (arp_validate_value == -1) {
4619 printk(KERN_ERR DRV_NAME
4620 ": Error: invalid arp_validate \"%s\"\n",
4621 arp_validate == NULL ? "NULL" : arp_validate);
4625 arp_validate_value = 0;
4628 printk(KERN_INFO DRV_NAME
4629 ": MII link monitoring set to %d ms\n",
4631 } else if (arp_interval) {
4634 printk(KERN_INFO DRV_NAME
4635 ": ARP monitoring set to %d ms, validate %s, with %d target(s):",
4637 arp_validate_tbl[arp_validate_value].modename,
4640 for (i = 0; i < arp_ip_count; i++)
4641 printk (" %s", arp_ip_target[i]);
4646 /* miimon and arp_interval not set, we need one so things
4647 * work as expected, see bonding.txt for details
4649 printk(KERN_WARNING DRV_NAME
4650 ": Warning: either miimon or arp_interval and "
4651 "arp_ip_target module parameters must be specified, "
4652 "otherwise bonding will not detect link failures! see "
4653 "bonding.txt for details.\n");
4656 if (primary && !USES_PRIMARY(bond_mode)) {
4657 /* currently, using a primary only makes sense
4658 * in active backup, TLB or ALB modes
4660 printk(KERN_WARNING DRV_NAME
4661 ": Warning: %s primary device specified but has no "
4662 "effect in %s mode\n",
4663 primary, bond_mode_name(bond_mode));
4667 /* fill params struct with the proper values */
4668 params->mode = bond_mode;
4669 params->xmit_policy = xmit_hashtype;
4670 params->miimon = miimon;
4671 params->arp_interval = arp_interval;
4672 params->arp_validate = arp_validate_value;
4673 params->updelay = updelay;
4674 params->downdelay = downdelay;
4675 params->use_carrier = use_carrier;
4676 params->lacp_fast = lacp_fast;
4677 params->primary[0] = 0;
4680 strncpy(params->primary, primary, IFNAMSIZ);
4681 params->primary[IFNAMSIZ - 1] = 0;
4684 memcpy(params->arp_targets, arp_target, sizeof(arp_target));
4689 static struct lock_class_key bonding_netdev_xmit_lock_key;
4691 /* Create a new bond based on the specified name and bonding parameters.
4692 * If name is NULL, obtain a suitable "bond%d" name for us.
4693 * Caller must NOT hold rtnl_lock; we need to release it here before we
4694 * set up our sysfs entries.
4696 int bond_create(char *name, struct bond_params *params, struct bonding **newbond)
4698 struct net_device *bond_dev;
4702 bond_dev = alloc_netdev(sizeof(struct bonding), name ? name : "",
4705 printk(KERN_ERR DRV_NAME
4706 ": %s: eek! can't alloc netdev!\n",
4713 res = dev_alloc_name(bond_dev, "bond%d");
4718 /* bond_init() must be called after dev_alloc_name() (for the
4719 * /proc files), but before register_netdevice(), because we
4720 * need to set function pointers.
4723 res = bond_init(bond_dev, params);
4728 res = register_netdevice(bond_dev);
4733 lockdep_set_class(&bond_dev->_xmit_lock, &bonding_netdev_xmit_lock_key);
4736 *newbond = bond_dev->priv;
4738 netif_carrier_off(bond_dev);
4740 rtnl_unlock(); /* allows sysfs registration of net device */
4741 res = bond_create_sysfs_entry(bond_dev->priv);
4750 bond_deinit(bond_dev);
4752 free_netdev(bond_dev);
4758 static int __init bonding_init(void)
4763 printk(KERN_INFO "%s", version);
4765 res = bond_check_params(&bonding_defaults);
4770 #ifdef CONFIG_PROC_FS
4771 bond_create_proc_dir();
4773 for (i = 0; i < max_bonds; i++) {
4774 res = bond_create(NULL, &bonding_defaults, NULL);
4779 res = bond_create_sysfs();
4783 register_netdevice_notifier(&bond_netdev_notifier);
4784 register_inetaddr_notifier(&bond_inetaddr_notifier);
4790 bond_destroy_sysfs();
4797 static void __exit bonding_exit(void)
4799 unregister_netdevice_notifier(&bond_netdev_notifier);
4800 unregister_inetaddr_notifier(&bond_inetaddr_notifier);
4804 bond_destroy_sysfs();
4808 module_init(bonding_init);
4809 module_exit(bonding_exit);
4810 MODULE_LICENSE("GPL");
4811 MODULE_VERSION(DRV_VERSION);
4812 MODULE_DESCRIPTION(DRV_DESCRIPTION ", v" DRV_VERSION);
4813 MODULE_AUTHOR("Thomas Davis, tadavis@lbl.gov and many others");
4814 MODULE_SUPPORTED_DEVICE("most ethernet devices");