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);
1099 bond_send_gratuitous_arp(bond);
1104 * bond_select_active_slave - select a new active slave, if needed
1105 * @bond: our bonding struct
1107 * This functions shoud be called when one of the following occurs:
1108 * - The old curr_active_slave has been released or lost its link.
1109 * - The primary_slave has got its link back.
1110 * - A slave has got its link back and there's no old curr_active_slave.
1112 * Warning: Caller must hold curr_slave_lock for writing.
1114 void bond_select_active_slave(struct bonding *bond)
1116 struct slave *best_slave;
1119 best_slave = bond_find_best_slave(bond);
1120 if (best_slave != bond->curr_active_slave) {
1121 bond_change_active_slave(bond, best_slave);
1122 rv = bond_set_carrier(bond);
1126 if (netif_carrier_ok(bond->dev)) {
1127 printk(KERN_INFO DRV_NAME
1128 ": %s: first active interface up!\n",
1131 printk(KERN_INFO DRV_NAME ": %s: "
1132 "now running without any active interface !\n",
1138 /*--------------------------- slave list handling ---------------------------*/
1141 * This function attaches the slave to the end of list.
1143 * bond->lock held for writing by caller.
1145 static void bond_attach_slave(struct bonding *bond, struct slave *new_slave)
1147 if (bond->first_slave == NULL) { /* attaching the first slave */
1148 new_slave->next = new_slave;
1149 new_slave->prev = new_slave;
1150 bond->first_slave = new_slave;
1152 new_slave->next = bond->first_slave;
1153 new_slave->prev = bond->first_slave->prev;
1154 new_slave->next->prev = new_slave;
1155 new_slave->prev->next = new_slave;
1162 * This function detaches the slave from the list.
1163 * WARNING: no check is made to verify if the slave effectively
1164 * belongs to <bond>.
1165 * Nothing is freed on return, structures are just unchained.
1166 * If any slave pointer in bond was pointing to <slave>,
1167 * it should be changed by the calling function.
1169 * bond->lock held for writing by caller.
1171 static void bond_detach_slave(struct bonding *bond, struct slave *slave)
1174 slave->next->prev = slave->prev;
1178 slave->prev->next = slave->next;
1181 if (bond->first_slave == slave) { /* slave is the first slave */
1182 if (bond->slave_cnt > 1) { /* there are more slave */
1183 bond->first_slave = slave->next;
1185 bond->first_slave = NULL; /* slave was the last one */
1194 /*---------------------------------- IOCTL ----------------------------------*/
1196 static int bond_sethwaddr(struct net_device *bond_dev,
1197 struct net_device *slave_dev)
1199 dprintk("bond_dev=%p\n", bond_dev);
1200 dprintk("slave_dev=%p\n", slave_dev);
1201 dprintk("slave_dev->addr_len=%d\n", slave_dev->addr_len);
1202 memcpy(bond_dev->dev_addr, slave_dev->dev_addr, slave_dev->addr_len);
1206 #define BOND_VLAN_FEATURES \
1207 (NETIF_F_VLAN_CHALLENGED | NETIF_F_HW_VLAN_RX | NETIF_F_HW_VLAN_TX | \
1208 NETIF_F_HW_VLAN_FILTER)
1211 * Compute the common dev->feature set available to all slaves. Some
1212 * feature bits are managed elsewhere, so preserve those feature bits
1213 * on the master device.
1215 static int bond_compute_features(struct bonding *bond)
1217 struct slave *slave;
1218 struct net_device *bond_dev = bond->dev;
1219 unsigned long features = bond_dev->features;
1220 unsigned short max_hard_header_len = ETH_HLEN;
1223 features &= ~(NETIF_F_ALL_CSUM | BOND_VLAN_FEATURES);
1224 features |= NETIF_F_SG | NETIF_F_FRAGLIST | NETIF_F_HIGHDMA |
1225 NETIF_F_GSO_MASK | NETIF_F_NO_CSUM;
1227 bond_for_each_slave(bond, slave, i) {
1228 features = netdev_compute_features(features,
1229 slave->dev->features);
1230 if (slave->dev->hard_header_len > max_hard_header_len)
1231 max_hard_header_len = slave->dev->hard_header_len;
1234 features |= (bond_dev->features & BOND_VLAN_FEATURES);
1235 bond_dev->features = features;
1236 bond_dev->hard_header_len = max_hard_header_len;
1242 static void bond_setup_by_slave(struct net_device *bond_dev,
1243 struct net_device *slave_dev)
1245 bond_dev->neigh_setup = slave_dev->neigh_setup;
1247 bond_dev->type = slave_dev->type;
1248 bond_dev->hard_header_len = slave_dev->hard_header_len;
1249 bond_dev->addr_len = slave_dev->addr_len;
1251 memcpy(bond_dev->broadcast, slave_dev->broadcast,
1252 slave_dev->addr_len);
1255 /* enslave device <slave> to bond device <master> */
1256 int bond_enslave(struct net_device *bond_dev, struct net_device *slave_dev)
1258 struct bonding *bond = bond_dev->priv;
1259 struct slave *new_slave = NULL;
1260 struct dev_mc_list *dmi;
1261 struct sockaddr addr;
1263 int old_features = bond_dev->features;
1266 if (!bond->params.use_carrier && slave_dev->ethtool_ops == NULL &&
1267 slave_dev->do_ioctl == NULL) {
1268 printk(KERN_WARNING DRV_NAME
1269 ": %s: Warning: no link monitoring support for %s\n",
1270 bond_dev->name, slave_dev->name);
1273 /* bond must be initialized by bond_open() before enslaving */
1274 if (!(bond_dev->flags & IFF_UP)) {
1275 dprintk("Error, master_dev is not up\n");
1279 /* already enslaved */
1280 if (slave_dev->flags & IFF_SLAVE) {
1281 dprintk("Error, Device was already enslaved\n");
1285 /* vlan challenged mutual exclusion */
1286 /* no need to lock since we're protected by rtnl_lock */
1287 if (slave_dev->features & NETIF_F_VLAN_CHALLENGED) {
1288 dprintk("%s: NETIF_F_VLAN_CHALLENGED\n", slave_dev->name);
1289 if (!list_empty(&bond->vlan_list)) {
1290 printk(KERN_ERR DRV_NAME
1291 ": %s: Error: cannot enslave VLAN "
1292 "challenged slave %s on VLAN enabled "
1293 "bond %s\n", bond_dev->name, slave_dev->name,
1297 printk(KERN_WARNING DRV_NAME
1298 ": %s: Warning: enslaved VLAN challenged "
1299 "slave %s. Adding VLANs will be blocked as "
1300 "long as %s is part of bond %s\n",
1301 bond_dev->name, slave_dev->name, slave_dev->name,
1303 bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
1306 dprintk("%s: ! NETIF_F_VLAN_CHALLENGED\n", slave_dev->name);
1307 if (bond->slave_cnt == 0) {
1308 /* First slave, and it is not VLAN challenged,
1309 * so remove the block of adding VLANs over the bond.
1311 bond_dev->features &= ~NETIF_F_VLAN_CHALLENGED;
1316 * Old ifenslave binaries are no longer supported. These can
1317 * be identified with moderate accurary by the state of the slave:
1318 * the current ifenslave will set the interface down prior to
1319 * enslaving it; the old ifenslave will not.
1321 if ((slave_dev->flags & IFF_UP)) {
1322 printk(KERN_ERR DRV_NAME ": %s is up. "
1323 "This may be due to an out of date ifenslave.\n",
1326 goto err_undo_flags;
1329 /* set bonding device ether type by slave - bonding netdevices are
1330 * created with ether_setup, so when the slave type is not ARPHRD_ETHER
1331 * there is a need to override some of the type dependent attribs/funcs.
1333 * bond ether type mutual exclusion - don't allow slaves of dissimilar
1334 * ether type (eg ARPHRD_ETHER and ARPHRD_INFINIBAND) share the same bond
1336 if (bond->slave_cnt == 0) {
1337 if (slave_dev->type != ARPHRD_ETHER)
1338 bond_setup_by_slave(bond_dev, slave_dev);
1339 } else if (bond_dev->type != slave_dev->type) {
1340 printk(KERN_ERR DRV_NAME ": %s ether type (%d) is different "
1341 "from other slaves (%d), can not enslave it.\n",
1343 slave_dev->type, bond_dev->type);
1345 goto err_undo_flags;
1348 if (slave_dev->set_mac_address == NULL) {
1349 printk(KERN_ERR DRV_NAME
1350 ": %s: Error: The slave device you specified does "
1351 "not support setting the MAC address. "
1352 "Your kernel likely does not support slave "
1353 "devices.\n", bond_dev->name);
1355 goto err_undo_flags;
1358 new_slave = kzalloc(sizeof(struct slave), GFP_KERNEL);
1361 goto err_undo_flags;
1364 /* save slave's original flags before calling
1365 * netdev_set_master and dev_open
1367 new_slave->original_flags = slave_dev->flags;
1370 * Save slave's original ("permanent") mac address for modes
1371 * that need it, and for restoring it upon release, and then
1372 * set it to the master's address
1374 memcpy(new_slave->perm_hwaddr, slave_dev->dev_addr, ETH_ALEN);
1377 * Set slave to master's mac address. The application already
1378 * set the master's mac address to that of the first slave
1380 memcpy(addr.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
1381 addr.sa_family = slave_dev->type;
1382 res = dev_set_mac_address(slave_dev, &addr);
1384 dprintk("Error %d calling set_mac_address\n", res);
1388 res = netdev_set_master(slave_dev, bond_dev);
1390 dprintk("Error %d calling netdev_set_master\n", res);
1393 /* open the slave since the application closed it */
1394 res = dev_open(slave_dev);
1396 dprintk("Openning slave %s failed\n", slave_dev->name);
1397 goto err_restore_mac;
1400 new_slave->dev = slave_dev;
1401 slave_dev->priv_flags |= IFF_BONDING;
1403 if ((bond->params.mode == BOND_MODE_TLB) ||
1404 (bond->params.mode == BOND_MODE_ALB)) {
1405 /* bond_alb_init_slave() must be called before all other stages since
1406 * it might fail and we do not want to have to undo everything
1408 res = bond_alb_init_slave(bond, new_slave);
1410 goto err_unset_master;
1414 /* If the mode USES_PRIMARY, then the new slave gets the
1415 * master's promisc (and mc) settings only if it becomes the
1416 * curr_active_slave, and that is taken care of later when calling
1417 * bond_change_active()
1419 if (!USES_PRIMARY(bond->params.mode)) {
1420 /* set promiscuity level to new slave */
1421 if (bond_dev->flags & IFF_PROMISC) {
1422 dev_set_promiscuity(slave_dev, 1);
1425 /* set allmulti level to new slave */
1426 if (bond_dev->flags & IFF_ALLMULTI) {
1427 dev_set_allmulti(slave_dev, 1);
1430 /* upload master's mc_list to new slave */
1431 for (dmi = bond_dev->mc_list; dmi; dmi = dmi->next) {
1432 dev_mc_add (slave_dev, dmi->dmi_addr, dmi->dmi_addrlen, 0);
1436 if (bond->params.mode == BOND_MODE_8023AD) {
1437 /* add lacpdu mc addr to mc list */
1438 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
1440 dev_mc_add(slave_dev, lacpdu_multicast, ETH_ALEN, 0);
1443 bond_add_vlans_on_slave(bond, slave_dev);
1445 write_lock_bh(&bond->lock);
1447 bond_attach_slave(bond, new_slave);
1449 new_slave->delay = 0;
1450 new_slave->link_failure_count = 0;
1452 bond_compute_features(bond);
1454 new_slave->last_arp_rx = jiffies;
1456 if (bond->params.miimon && !bond->params.use_carrier) {
1457 link_reporting = bond_check_dev_link(bond, slave_dev, 1);
1459 if ((link_reporting == -1) && !bond->params.arp_interval) {
1461 * miimon is set but a bonded network driver
1462 * does not support ETHTOOL/MII and
1463 * arp_interval is not set. Note: if
1464 * use_carrier is enabled, we will never go
1465 * here (because netif_carrier is always
1466 * supported); thus, we don't need to change
1467 * the messages for netif_carrier.
1469 printk(KERN_WARNING DRV_NAME
1470 ": %s: Warning: MII and ETHTOOL support not "
1471 "available for interface %s, and "
1472 "arp_interval/arp_ip_target module parameters "
1473 "not specified, thus bonding will not detect "
1474 "link failures! see bonding.txt for details.\n",
1475 bond_dev->name, slave_dev->name);
1476 } else if (link_reporting == -1) {
1477 /* unable get link status using mii/ethtool */
1478 printk(KERN_WARNING DRV_NAME
1479 ": %s: Warning: can't get link status from "
1480 "interface %s; the network driver associated "
1481 "with this interface does not support MII or "
1482 "ETHTOOL link status reporting, thus miimon "
1483 "has no effect on this interface.\n",
1484 bond_dev->name, slave_dev->name);
1488 /* check for initial state */
1489 if (!bond->params.miimon ||
1490 (bond_check_dev_link(bond, slave_dev, 0) == BMSR_LSTATUS)) {
1491 if (bond->params.updelay) {
1492 dprintk("Initial state of slave_dev is "
1493 "BOND_LINK_BACK\n");
1494 new_slave->link = BOND_LINK_BACK;
1495 new_slave->delay = bond->params.updelay;
1497 dprintk("Initial state of slave_dev is "
1499 new_slave->link = BOND_LINK_UP;
1501 new_slave->jiffies = jiffies;
1503 dprintk("Initial state of slave_dev is "
1504 "BOND_LINK_DOWN\n");
1505 new_slave->link = BOND_LINK_DOWN;
1508 if (bond_update_speed_duplex(new_slave) &&
1509 (new_slave->link != BOND_LINK_DOWN)) {
1510 printk(KERN_WARNING DRV_NAME
1511 ": %s: Warning: failed to get speed and duplex from %s, "
1512 "assumed to be 100Mb/sec and Full.\n",
1513 bond_dev->name, new_slave->dev->name);
1515 if (bond->params.mode == BOND_MODE_8023AD) {
1516 printk(KERN_WARNING DRV_NAME
1517 ": %s: Warning: Operation of 802.3ad mode requires ETHTOOL "
1518 "support in base driver for proper aggregator "
1519 "selection.\n", bond_dev->name);
1523 if (USES_PRIMARY(bond->params.mode) && bond->params.primary[0]) {
1524 /* if there is a primary slave, remember it */
1525 if (strcmp(bond->params.primary, new_slave->dev->name) == 0) {
1526 bond->primary_slave = new_slave;
1530 switch (bond->params.mode) {
1531 case BOND_MODE_ACTIVEBACKUP:
1532 bond_set_slave_inactive_flags(new_slave);
1533 bond_select_active_slave(bond);
1535 case BOND_MODE_8023AD:
1536 /* in 802.3ad mode, the internal mechanism
1537 * will activate the slaves in the selected
1540 bond_set_slave_inactive_flags(new_slave);
1541 /* if this is the first slave */
1542 if (bond->slave_cnt == 1) {
1543 SLAVE_AD_INFO(new_slave).id = 1;
1544 /* Initialize AD with the number of times that the AD timer is called in 1 second
1545 * can be called only after the mac address of the bond is set
1547 bond_3ad_initialize(bond, 1000/AD_TIMER_INTERVAL,
1548 bond->params.lacp_fast);
1550 SLAVE_AD_INFO(new_slave).id =
1551 SLAVE_AD_INFO(new_slave->prev).id + 1;
1554 bond_3ad_bind_slave(new_slave);
1558 new_slave->state = BOND_STATE_ACTIVE;
1559 if ((!bond->curr_active_slave) &&
1560 (new_slave->link != BOND_LINK_DOWN)) {
1561 /* first slave or no active slave yet, and this link
1562 * is OK, so make this interface the active one
1564 bond_change_active_slave(bond, new_slave);
1566 bond_set_slave_inactive_flags(new_slave);
1570 dprintk("This slave is always active in trunk mode\n");
1572 /* always active in trunk mode */
1573 new_slave->state = BOND_STATE_ACTIVE;
1575 /* In trunking mode there is little meaning to curr_active_slave
1576 * anyway (it holds no special properties of the bond device),
1577 * so we can change it without calling change_active_interface()
1579 if (!bond->curr_active_slave) {
1580 bond->curr_active_slave = new_slave;
1583 } /* switch(bond_mode) */
1585 bond_set_carrier(bond);
1587 write_unlock_bh(&bond->lock);
1589 res = bond_create_slave_symlinks(bond_dev, slave_dev);
1591 goto err_unset_master;
1593 printk(KERN_INFO DRV_NAME
1594 ": %s: enslaving %s as a%s interface with a%s link.\n",
1595 bond_dev->name, slave_dev->name,
1596 new_slave->state == BOND_STATE_ACTIVE ? "n active" : " backup",
1597 new_slave->link != BOND_LINK_DOWN ? "n up" : " down");
1599 /* enslave is successful */
1602 /* Undo stages on error */
1604 netdev_set_master(slave_dev, NULL);
1607 dev_close(slave_dev);
1610 memcpy(addr.sa_data, new_slave->perm_hwaddr, ETH_ALEN);
1611 addr.sa_family = slave_dev->type;
1612 dev_set_mac_address(slave_dev, &addr);
1618 bond_dev->features = old_features;
1624 * Try to release the slave device <slave> from the bond device <master>
1625 * It is legal to access curr_active_slave without a lock because all the function
1628 * The rules for slave state should be:
1629 * for Active/Backup:
1630 * Active stays on all backups go down
1631 * for Bonded connections:
1632 * The first up interface should be left on and all others downed.
1634 int bond_release(struct net_device *bond_dev, struct net_device *slave_dev)
1636 struct bonding *bond = bond_dev->priv;
1637 struct slave *slave, *oldcurrent;
1638 struct sockaddr addr;
1639 int mac_addr_differ;
1640 DECLARE_MAC_BUF(mac);
1642 /* slave is not a slave or master is not master of this slave */
1643 if (!(slave_dev->flags & IFF_SLAVE) ||
1644 (slave_dev->master != bond_dev)) {
1645 printk(KERN_ERR DRV_NAME
1646 ": %s: Error: cannot release %s.\n",
1647 bond_dev->name, slave_dev->name);
1651 write_lock_bh(&bond->lock);
1653 slave = bond_get_slave_by_dev(bond, slave_dev);
1655 /* not a slave of this bond */
1656 printk(KERN_INFO DRV_NAME
1657 ": %s: %s not enslaved\n",
1658 bond_dev->name, slave_dev->name);
1659 write_unlock_bh(&bond->lock);
1663 mac_addr_differ = memcmp(bond_dev->dev_addr,
1666 if (!mac_addr_differ && (bond->slave_cnt > 1)) {
1667 printk(KERN_WARNING DRV_NAME
1668 ": %s: Warning: the permanent HWaddr of %s - "
1669 "%s - is still in use by %s. "
1670 "Set the HWaddr of %s to a different address "
1671 "to avoid conflicts.\n",
1674 print_mac(mac, slave->perm_hwaddr),
1679 /* Inform AD package of unbinding of slave. */
1680 if (bond->params.mode == BOND_MODE_8023AD) {
1681 /* must be called before the slave is
1682 * detached from the list
1684 bond_3ad_unbind_slave(slave);
1687 printk(KERN_INFO DRV_NAME
1688 ": %s: releasing %s interface %s\n",
1690 (slave->state == BOND_STATE_ACTIVE)
1691 ? "active" : "backup",
1694 oldcurrent = bond->curr_active_slave;
1696 bond->current_arp_slave = NULL;
1698 /* release the slave from its bond */
1699 bond_detach_slave(bond, slave);
1701 bond_compute_features(bond);
1703 if (bond->primary_slave == slave) {
1704 bond->primary_slave = NULL;
1707 if (oldcurrent == slave) {
1708 bond_change_active_slave(bond, NULL);
1711 if ((bond->params.mode == BOND_MODE_TLB) ||
1712 (bond->params.mode == BOND_MODE_ALB)) {
1713 /* Must be called only after the slave has been
1714 * detached from the list and the curr_active_slave
1715 * has been cleared (if our_slave == old_current),
1716 * but before a new active slave is selected.
1718 bond_alb_deinit_slave(bond, slave);
1721 if (oldcurrent == slave)
1722 bond_select_active_slave(bond);
1724 if (bond->slave_cnt == 0) {
1725 bond_set_carrier(bond);
1727 /* if the last slave was removed, zero the mac address
1728 * of the master so it will be set by the application
1729 * to the mac address of the first slave
1731 memset(bond_dev->dev_addr, 0, bond_dev->addr_len);
1733 if (list_empty(&bond->vlan_list)) {
1734 bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
1736 printk(KERN_WARNING DRV_NAME
1737 ": %s: Warning: clearing HW address of %s while it "
1738 "still has VLANs.\n",
1739 bond_dev->name, bond_dev->name);
1740 printk(KERN_WARNING DRV_NAME
1741 ": %s: When re-adding slaves, make sure the bond's "
1742 "HW address matches its VLANs'.\n",
1745 } else if ((bond_dev->features & NETIF_F_VLAN_CHALLENGED) &&
1746 !bond_has_challenged_slaves(bond)) {
1747 printk(KERN_INFO DRV_NAME
1748 ": %s: last VLAN challenged slave %s "
1749 "left bond %s. VLAN blocking is removed\n",
1750 bond_dev->name, slave_dev->name, bond_dev->name);
1751 bond_dev->features &= ~NETIF_F_VLAN_CHALLENGED;
1754 write_unlock_bh(&bond->lock);
1756 /* must do this from outside any spinlocks */
1757 bond_destroy_slave_symlinks(bond_dev, slave_dev);
1759 bond_del_vlans_from_slave(bond, slave_dev);
1761 /* If the mode USES_PRIMARY, then we should only remove its
1762 * promisc and mc settings if it was the curr_active_slave, but that was
1763 * already taken care of above when we detached the slave
1765 if (!USES_PRIMARY(bond->params.mode)) {
1766 /* unset promiscuity level from slave */
1767 if (bond_dev->flags & IFF_PROMISC) {
1768 dev_set_promiscuity(slave_dev, -1);
1771 /* unset allmulti level from slave */
1772 if (bond_dev->flags & IFF_ALLMULTI) {
1773 dev_set_allmulti(slave_dev, -1);
1776 /* flush master's mc_list from slave */
1777 bond_mc_list_flush(bond_dev, slave_dev);
1780 netdev_set_master(slave_dev, NULL);
1782 /* close slave before restoring its mac address */
1783 dev_close(slave_dev);
1785 /* restore original ("permanent") mac address */
1786 memcpy(addr.sa_data, slave->perm_hwaddr, ETH_ALEN);
1787 addr.sa_family = slave_dev->type;
1788 dev_set_mac_address(slave_dev, &addr);
1790 slave_dev->priv_flags &= ~(IFF_MASTER_8023AD | IFF_MASTER_ALB |
1791 IFF_SLAVE_INACTIVE | IFF_BONDING |
1796 return 0; /* deletion OK */
1800 * This function releases all slaves.
1802 static int bond_release_all(struct net_device *bond_dev)
1804 struct bonding *bond = bond_dev->priv;
1805 struct slave *slave;
1806 struct net_device *slave_dev;
1807 struct sockaddr addr;
1809 write_lock_bh(&bond->lock);
1811 netif_carrier_off(bond_dev);
1813 if (bond->slave_cnt == 0) {
1817 bond->current_arp_slave = NULL;
1818 bond->primary_slave = NULL;
1819 bond_change_active_slave(bond, NULL);
1821 while ((slave = bond->first_slave) != NULL) {
1822 /* Inform AD package of unbinding of slave
1823 * before slave is detached from the list.
1825 if (bond->params.mode == BOND_MODE_8023AD) {
1826 bond_3ad_unbind_slave(slave);
1829 slave_dev = slave->dev;
1830 bond_detach_slave(bond, slave);
1832 if ((bond->params.mode == BOND_MODE_TLB) ||
1833 (bond->params.mode == BOND_MODE_ALB)) {
1834 /* must be called only after the slave
1835 * has been detached from the list
1837 bond_alb_deinit_slave(bond, slave);
1840 bond_compute_features(bond);
1842 /* now that the slave is detached, unlock and perform
1843 * all the undo steps that should not be called from
1846 write_unlock_bh(&bond->lock);
1848 bond_destroy_slave_symlinks(bond_dev, slave_dev);
1849 bond_del_vlans_from_slave(bond, slave_dev);
1851 /* If the mode USES_PRIMARY, then we should only remove its
1852 * promisc and mc settings if it was the curr_active_slave, but that was
1853 * already taken care of above when we detached the slave
1855 if (!USES_PRIMARY(bond->params.mode)) {
1856 /* unset promiscuity level from slave */
1857 if (bond_dev->flags & IFF_PROMISC) {
1858 dev_set_promiscuity(slave_dev, -1);
1861 /* unset allmulti level from slave */
1862 if (bond_dev->flags & IFF_ALLMULTI) {
1863 dev_set_allmulti(slave_dev, -1);
1866 /* flush master's mc_list from slave */
1867 bond_mc_list_flush(bond_dev, slave_dev);
1870 netdev_set_master(slave_dev, NULL);
1872 /* close slave before restoring its mac address */
1873 dev_close(slave_dev);
1875 /* restore original ("permanent") mac address*/
1876 memcpy(addr.sa_data, slave->perm_hwaddr, ETH_ALEN);
1877 addr.sa_family = slave_dev->type;
1878 dev_set_mac_address(slave_dev, &addr);
1880 slave_dev->priv_flags &= ~(IFF_MASTER_8023AD | IFF_MASTER_ALB |
1881 IFF_SLAVE_INACTIVE);
1885 /* re-acquire the lock before getting the next slave */
1886 write_lock_bh(&bond->lock);
1889 /* zero the mac address of the master so it will be
1890 * set by the application to the mac address of the
1893 memset(bond_dev->dev_addr, 0, bond_dev->addr_len);
1895 if (list_empty(&bond->vlan_list)) {
1896 bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
1898 printk(KERN_WARNING DRV_NAME
1899 ": %s: Warning: clearing HW address of %s while it "
1900 "still has VLANs.\n",
1901 bond_dev->name, bond_dev->name);
1902 printk(KERN_WARNING DRV_NAME
1903 ": %s: When re-adding slaves, make sure the bond's "
1904 "HW address matches its VLANs'.\n",
1908 printk(KERN_INFO DRV_NAME
1909 ": %s: released all slaves\n",
1913 write_unlock_bh(&bond->lock);
1919 * This function changes the active slave to slave <slave_dev>.
1920 * It returns -EINVAL in the following cases.
1921 * - <slave_dev> is not found in the list.
1922 * - There is not active slave now.
1923 * - <slave_dev> is already active.
1924 * - The link state of <slave_dev> is not BOND_LINK_UP.
1925 * - <slave_dev> is not running.
1926 * In these cases, this fuction does nothing.
1927 * In the other cases, currnt_slave pointer is changed and 0 is returned.
1929 static int bond_ioctl_change_active(struct net_device *bond_dev, struct net_device *slave_dev)
1931 struct bonding *bond = bond_dev->priv;
1932 struct slave *old_active = NULL;
1933 struct slave *new_active = NULL;
1936 if (!USES_PRIMARY(bond->params.mode)) {
1940 /* Verify that master_dev is indeed the master of slave_dev */
1941 if (!(slave_dev->flags & IFF_SLAVE) ||
1942 (slave_dev->master != bond_dev)) {
1946 write_lock_bh(&bond->lock);
1948 old_active = bond->curr_active_slave;
1949 new_active = bond_get_slave_by_dev(bond, slave_dev);
1952 * Changing to the current active: do nothing; return success.
1954 if (new_active && (new_active == old_active)) {
1955 write_unlock_bh(&bond->lock);
1961 (new_active->link == BOND_LINK_UP) &&
1962 IS_UP(new_active->dev)) {
1963 bond_change_active_slave(bond, new_active);
1968 write_unlock_bh(&bond->lock);
1973 static int bond_info_query(struct net_device *bond_dev, struct ifbond *info)
1975 struct bonding *bond = bond_dev->priv;
1977 info->bond_mode = bond->params.mode;
1978 info->miimon = bond->params.miimon;
1980 read_lock_bh(&bond->lock);
1981 info->num_slaves = bond->slave_cnt;
1982 read_unlock_bh(&bond->lock);
1987 static int bond_slave_info_query(struct net_device *bond_dev, struct ifslave *info)
1989 struct bonding *bond = bond_dev->priv;
1990 struct slave *slave;
1993 if (info->slave_id < 0) {
1997 read_lock_bh(&bond->lock);
1999 bond_for_each_slave(bond, slave, i) {
2000 if (i == (int)info->slave_id) {
2006 read_unlock_bh(&bond->lock);
2009 strcpy(info->slave_name, slave->dev->name);
2010 info->link = slave->link;
2011 info->state = slave->state;
2012 info->link_failure_count = slave->link_failure_count;
2020 /*-------------------------------- Monitoring -------------------------------*/
2022 /* this function is called regularly to monitor each slave's link. */
2023 void bond_mii_monitor(struct net_device *bond_dev)
2025 struct bonding *bond = bond_dev->priv;
2026 struct slave *slave, *oldcurrent;
2027 int do_failover = 0;
2031 read_lock(&bond->lock);
2033 delta_in_ticks = (bond->params.miimon * HZ) / 1000;
2035 if (bond->kill_timers) {
2039 if (bond->slave_cnt == 0) {
2043 /* we will try to read the link status of each of our slaves, and
2044 * set their IFF_RUNNING flag appropriately. For each slave not
2045 * supporting MII status, we won't do anything so that a user-space
2046 * program could monitor the link itself if needed.
2049 read_lock(&bond->curr_slave_lock);
2050 oldcurrent = bond->curr_active_slave;
2051 read_unlock(&bond->curr_slave_lock);
2053 bond_for_each_slave(bond, slave, i) {
2054 struct net_device *slave_dev = slave->dev;
2056 u16 old_speed = slave->speed;
2057 u8 old_duplex = slave->duplex;
2059 link_state = bond_check_dev_link(bond, slave_dev, 0);
2061 switch (slave->link) {
2062 case BOND_LINK_UP: /* the link was up */
2063 if (link_state == BMSR_LSTATUS) {
2064 /* link stays up, nothing more to do */
2066 } else { /* link going down */
2067 slave->link = BOND_LINK_FAIL;
2068 slave->delay = bond->params.downdelay;
2070 if (slave->link_failure_count < UINT_MAX) {
2071 slave->link_failure_count++;
2074 if (bond->params.downdelay) {
2075 printk(KERN_INFO DRV_NAME
2076 ": %s: link status down for %s "
2077 "interface %s, disabling it in "
2081 ? ((bond->params.mode == BOND_MODE_ACTIVEBACKUP)
2082 ? ((slave == oldcurrent)
2083 ? "active " : "backup ")
2087 bond->params.downdelay * bond->params.miimon);
2090 /* no break ! fall through the BOND_LINK_FAIL test to
2091 ensure proper action to be taken
2093 case BOND_LINK_FAIL: /* the link has just gone down */
2094 if (link_state != BMSR_LSTATUS) {
2095 /* link stays down */
2096 if (slave->delay <= 0) {
2097 /* link down for too long time */
2098 slave->link = BOND_LINK_DOWN;
2100 /* in active/backup mode, we must
2101 * completely disable this interface
2103 if ((bond->params.mode == BOND_MODE_ACTIVEBACKUP) ||
2104 (bond->params.mode == BOND_MODE_8023AD)) {
2105 bond_set_slave_inactive_flags(slave);
2108 printk(KERN_INFO DRV_NAME
2109 ": %s: link status definitely "
2110 "down for interface %s, "
2115 /* notify ad that the link status has changed */
2116 if (bond->params.mode == BOND_MODE_8023AD) {
2117 bond_3ad_handle_link_change(slave, BOND_LINK_DOWN);
2120 if ((bond->params.mode == BOND_MODE_TLB) ||
2121 (bond->params.mode == BOND_MODE_ALB)) {
2122 bond_alb_handle_link_change(bond, slave, BOND_LINK_DOWN);
2125 if (slave == oldcurrent) {
2133 slave->link = BOND_LINK_UP;
2134 slave->jiffies = jiffies;
2135 printk(KERN_INFO DRV_NAME
2136 ": %s: link status up again after %d "
2137 "ms for interface %s.\n",
2139 (bond->params.downdelay - slave->delay) * bond->params.miimon,
2143 case BOND_LINK_DOWN: /* the link was down */
2144 if (link_state != BMSR_LSTATUS) {
2145 /* the link stays down, nothing more to do */
2147 } else { /* link going up */
2148 slave->link = BOND_LINK_BACK;
2149 slave->delay = bond->params.updelay;
2151 if (bond->params.updelay) {
2152 /* if updelay == 0, no need to
2153 advertise about a 0 ms delay */
2154 printk(KERN_INFO DRV_NAME
2155 ": %s: link status up for "
2156 "interface %s, enabling it "
2160 bond->params.updelay * bond->params.miimon);
2163 /* no break ! fall through the BOND_LINK_BACK state in
2164 case there's something to do.
2166 case BOND_LINK_BACK: /* the link has just come back */
2167 if (link_state != BMSR_LSTATUS) {
2168 /* link down again */
2169 slave->link = BOND_LINK_DOWN;
2171 printk(KERN_INFO DRV_NAME
2172 ": %s: link status down again after %d "
2173 "ms for interface %s.\n",
2175 (bond->params.updelay - slave->delay) * bond->params.miimon,
2179 if (slave->delay == 0) {
2180 /* now the link has been up for long time enough */
2181 slave->link = BOND_LINK_UP;
2182 slave->jiffies = jiffies;
2184 if (bond->params.mode == BOND_MODE_8023AD) {
2185 /* prevent it from being the active one */
2186 slave->state = BOND_STATE_BACKUP;
2187 } else if (bond->params.mode != BOND_MODE_ACTIVEBACKUP) {
2188 /* make it immediately active */
2189 slave->state = BOND_STATE_ACTIVE;
2190 } else if (slave != bond->primary_slave) {
2191 /* prevent it from being the active one */
2192 slave->state = BOND_STATE_BACKUP;
2195 printk(KERN_INFO DRV_NAME
2196 ": %s: link status definitely "
2197 "up for interface %s.\n",
2201 /* notify ad that the link status has changed */
2202 if (bond->params.mode == BOND_MODE_8023AD) {
2203 bond_3ad_handle_link_change(slave, BOND_LINK_UP);
2206 if ((bond->params.mode == BOND_MODE_TLB) ||
2207 (bond->params.mode == BOND_MODE_ALB)) {
2208 bond_alb_handle_link_change(bond, slave, BOND_LINK_UP);
2211 if ((!oldcurrent) ||
2212 (slave == bond->primary_slave)) {
2221 /* Should not happen */
2222 printk(KERN_ERR DRV_NAME
2223 ": %s: Error: %s Illegal value (link=%d)\n",
2228 } /* end of switch (slave->link) */
2230 bond_update_speed_duplex(slave);
2232 if (bond->params.mode == BOND_MODE_8023AD) {
2233 if (old_speed != slave->speed) {
2234 bond_3ad_adapter_speed_changed(slave);
2237 if (old_duplex != slave->duplex) {
2238 bond_3ad_adapter_duplex_changed(slave);
2245 write_lock(&bond->curr_slave_lock);
2247 bond_select_active_slave(bond);
2249 write_unlock(&bond->curr_slave_lock);
2251 bond_set_carrier(bond);
2254 if (bond->params.miimon) {
2255 mod_timer(&bond->mii_timer, jiffies + delta_in_ticks);
2258 read_unlock(&bond->lock);
2262 static __be32 bond_glean_dev_ip(struct net_device *dev)
2264 struct in_device *idev;
2265 struct in_ifaddr *ifa;
2272 idev = __in_dev_get_rcu(dev);
2276 ifa = idev->ifa_list;
2280 addr = ifa->ifa_local;
2286 static int bond_has_ip(struct bonding *bond)
2288 struct vlan_entry *vlan, *vlan_next;
2290 if (bond->master_ip)
2293 if (list_empty(&bond->vlan_list))
2296 list_for_each_entry_safe(vlan, vlan_next, &bond->vlan_list,
2305 static int bond_has_this_ip(struct bonding *bond, __be32 ip)
2307 struct vlan_entry *vlan, *vlan_next;
2309 if (ip == bond->master_ip)
2312 if (list_empty(&bond->vlan_list))
2315 list_for_each_entry_safe(vlan, vlan_next, &bond->vlan_list,
2317 if (ip == vlan->vlan_ip)
2325 * We go to the (large) trouble of VLAN tagging ARP frames because
2326 * switches in VLAN mode (especially if ports are configured as
2327 * "native" to a VLAN) might not pass non-tagged frames.
2329 static void bond_arp_send(struct net_device *slave_dev, int arp_op, __be32 dest_ip, __be32 src_ip, unsigned short vlan_id)
2331 struct sk_buff *skb;
2333 dprintk("arp %d on slave %s: dst %x src %x vid %d\n", arp_op,
2334 slave_dev->name, dest_ip, src_ip, vlan_id);
2336 skb = arp_create(arp_op, ETH_P_ARP, dest_ip, slave_dev, src_ip,
2337 NULL, slave_dev->dev_addr, NULL);
2340 printk(KERN_ERR DRV_NAME ": ARP packet allocation failed\n");
2344 skb = vlan_put_tag(skb, vlan_id);
2346 printk(KERN_ERR DRV_NAME ": failed to insert VLAN tag\n");
2354 static void bond_arp_send_all(struct bonding *bond, struct slave *slave)
2357 __be32 *targets = bond->params.arp_targets;
2358 struct vlan_entry *vlan, *vlan_next;
2359 struct net_device *vlan_dev;
2363 for (i = 0; (i < BOND_MAX_ARP_TARGETS); i++) {
2366 dprintk("basa: target %x\n", targets[i]);
2367 if (list_empty(&bond->vlan_list)) {
2368 dprintk("basa: empty vlan: arp_send\n");
2369 bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2370 bond->master_ip, 0);
2375 * If VLANs are configured, we do a route lookup to
2376 * determine which VLAN interface would be used, so we
2377 * can tag the ARP with the proper VLAN tag.
2379 memset(&fl, 0, sizeof(fl));
2380 fl.fl4_dst = targets[i];
2381 fl.fl4_tos = RTO_ONLINK;
2383 rv = ip_route_output_key(&rt, &fl);
2385 if (net_ratelimit()) {
2386 printk(KERN_WARNING DRV_NAME
2387 ": %s: no route to arp_ip_target %u.%u.%u.%u\n",
2388 bond->dev->name, NIPQUAD(fl.fl4_dst));
2394 * This target is not on a VLAN
2396 if (rt->u.dst.dev == bond->dev) {
2398 dprintk("basa: rtdev == bond->dev: arp_send\n");
2399 bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2400 bond->master_ip, 0);
2405 list_for_each_entry_safe(vlan, vlan_next, &bond->vlan_list,
2407 vlan_dev = vlan_group_get_device(bond->vlgrp, vlan->vlan_id);
2408 if (vlan_dev == rt->u.dst.dev) {
2409 vlan_id = vlan->vlan_id;
2410 dprintk("basa: vlan match on %s %d\n",
2411 vlan_dev->name, vlan_id);
2418 bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2419 vlan->vlan_ip, vlan_id);
2423 if (net_ratelimit()) {
2424 printk(KERN_WARNING DRV_NAME
2425 ": %s: no path to arp_ip_target %u.%u.%u.%u via rt.dev %s\n",
2426 bond->dev->name, NIPQUAD(fl.fl4_dst),
2427 rt->u.dst.dev ? rt->u.dst.dev->name : "NULL");
2434 * Kick out a gratuitous ARP for an IP on the bonding master plus one
2435 * for each VLAN above us.
2437 static void bond_send_gratuitous_arp(struct bonding *bond)
2439 struct slave *slave = bond->curr_active_slave;
2440 struct vlan_entry *vlan;
2441 struct net_device *vlan_dev;
2443 dprintk("bond_send_grat_arp: bond %s slave %s\n", bond->dev->name,
2444 slave ? slave->dev->name : "NULL");
2448 if (bond->master_ip) {
2449 bond_arp_send(slave->dev, ARPOP_REPLY, bond->master_ip,
2450 bond->master_ip, 0);
2453 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
2454 vlan_dev = vlan_group_get_device(bond->vlgrp, vlan->vlan_id);
2455 if (vlan->vlan_ip) {
2456 bond_arp_send(slave->dev, ARPOP_REPLY, vlan->vlan_ip,
2457 vlan->vlan_ip, vlan->vlan_id);
2462 static void bond_validate_arp(struct bonding *bond, struct slave *slave, __be32 sip, __be32 tip)
2465 __be32 *targets = bond->params.arp_targets;
2467 targets = bond->params.arp_targets;
2468 for (i = 0; (i < BOND_MAX_ARP_TARGETS) && targets[i]; i++) {
2469 dprintk("bva: sip %u.%u.%u.%u tip %u.%u.%u.%u t[%d] "
2470 "%u.%u.%u.%u bhti(tip) %d\n",
2471 NIPQUAD(sip), NIPQUAD(tip), i, NIPQUAD(targets[i]),
2472 bond_has_this_ip(bond, tip));
2473 if (sip == targets[i]) {
2474 if (bond_has_this_ip(bond, tip))
2475 slave->last_arp_rx = jiffies;
2481 static int bond_arp_rcv(struct sk_buff *skb, struct net_device *dev, struct packet_type *pt, struct net_device *orig_dev)
2484 struct slave *slave;
2485 struct bonding *bond;
2486 unsigned char *arp_ptr;
2489 if (dev->nd_net != &init_net)
2492 if (!(dev->priv_flags & IFF_BONDING) || !(dev->flags & IFF_MASTER))
2496 read_lock(&bond->lock);
2498 dprintk("bond_arp_rcv: bond %s skb->dev %s orig_dev %s\n",
2499 bond->dev->name, skb->dev ? skb->dev->name : "NULL",
2500 orig_dev ? orig_dev->name : "NULL");
2502 slave = bond_get_slave_by_dev(bond, orig_dev);
2503 if (!slave || !slave_do_arp_validate(bond, slave))
2506 /* ARP header, plus 2 device addresses, plus 2 IP addresses. */
2507 if (!pskb_may_pull(skb, (sizeof(struct arphdr) +
2508 (2 * dev->addr_len) +
2509 (2 * sizeof(u32)))))
2513 if (arp->ar_hln != dev->addr_len ||
2514 skb->pkt_type == PACKET_OTHERHOST ||
2515 skb->pkt_type == PACKET_LOOPBACK ||
2516 arp->ar_hrd != htons(ARPHRD_ETHER) ||
2517 arp->ar_pro != htons(ETH_P_IP) ||
2521 arp_ptr = (unsigned char *)(arp + 1);
2522 arp_ptr += dev->addr_len;
2523 memcpy(&sip, arp_ptr, 4);
2524 arp_ptr += 4 + dev->addr_len;
2525 memcpy(&tip, arp_ptr, 4);
2527 dprintk("bond_arp_rcv: %s %s/%d av %d sv %d sip %u.%u.%u.%u"
2528 " tip %u.%u.%u.%u\n", bond->dev->name, slave->dev->name,
2529 slave->state, bond->params.arp_validate,
2530 slave_do_arp_validate(bond, slave), NIPQUAD(sip), NIPQUAD(tip));
2533 * Backup slaves won't see the ARP reply, but do come through
2534 * here for each ARP probe (so we swap the sip/tip to validate
2535 * the probe). In a "redundant switch, common router" type of
2536 * configuration, the ARP probe will (hopefully) travel from
2537 * the active, through one switch, the router, then the other
2538 * switch before reaching the backup.
2540 if (slave->state == BOND_STATE_ACTIVE)
2541 bond_validate_arp(bond, slave, sip, tip);
2543 bond_validate_arp(bond, slave, tip, sip);
2546 read_unlock(&bond->lock);
2549 return NET_RX_SUCCESS;
2553 * this function is called regularly to monitor each slave's link
2554 * ensuring that traffic is being sent and received when arp monitoring
2555 * is used in load-balancing mode. if the adapter has been dormant, then an
2556 * arp is transmitted to generate traffic. see activebackup_arp_monitor for
2557 * arp monitoring in active backup mode.
2559 void bond_loadbalance_arp_mon(struct net_device *bond_dev)
2561 struct bonding *bond = bond_dev->priv;
2562 struct slave *slave, *oldcurrent;
2563 int do_failover = 0;
2567 read_lock(&bond->lock);
2569 delta_in_ticks = (bond->params.arp_interval * HZ) / 1000;
2571 if (bond->kill_timers) {
2575 if (bond->slave_cnt == 0) {
2579 read_lock(&bond->curr_slave_lock);
2580 oldcurrent = bond->curr_active_slave;
2581 read_unlock(&bond->curr_slave_lock);
2583 /* see if any of the previous devices are up now (i.e. they have
2584 * xmt and rcv traffic). the curr_active_slave does not come into
2585 * the picture unless it is null. also, slave->jiffies is not needed
2586 * here because we send an arp on each slave and give a slave as
2587 * long as it needs to get the tx/rx within the delta.
2588 * TODO: what about up/down delay in arp mode? it wasn't here before
2591 bond_for_each_slave(bond, slave, i) {
2592 if (slave->link != BOND_LINK_UP) {
2593 if (((jiffies - slave->dev->trans_start) <= delta_in_ticks) &&
2594 ((jiffies - slave->dev->last_rx) <= delta_in_ticks)) {
2596 slave->link = BOND_LINK_UP;
2597 slave->state = BOND_STATE_ACTIVE;
2599 /* primary_slave has no meaning in round-robin
2600 * mode. the window of a slave being up and
2601 * curr_active_slave being null after enslaving
2605 printk(KERN_INFO DRV_NAME
2606 ": %s: link status definitely "
2607 "up for interface %s, ",
2612 printk(KERN_INFO DRV_NAME
2613 ": %s: interface %s is now up\n",
2619 /* slave->link == BOND_LINK_UP */
2621 /* not all switches will respond to an arp request
2622 * when the source ip is 0, so don't take the link down
2623 * if we don't know our ip yet
2625 if (((jiffies - slave->dev->trans_start) >= (2*delta_in_ticks)) ||
2626 (((jiffies - slave->dev->last_rx) >= (2*delta_in_ticks)) &&
2627 bond_has_ip(bond))) {
2629 slave->link = BOND_LINK_DOWN;
2630 slave->state = BOND_STATE_BACKUP;
2632 if (slave->link_failure_count < UINT_MAX) {
2633 slave->link_failure_count++;
2636 printk(KERN_INFO DRV_NAME
2637 ": %s: interface %s is now down.\n",
2641 if (slave == oldcurrent) {
2647 /* note: if switch is in round-robin mode, all links
2648 * must tx arp to ensure all links rx an arp - otherwise
2649 * links may oscillate or not come up at all; if switch is
2650 * in something like xor mode, there is nothing we can
2651 * do - all replies will be rx'ed on same link causing slaves
2652 * to be unstable during low/no traffic periods
2654 if (IS_UP(slave->dev)) {
2655 bond_arp_send_all(bond, slave);
2660 write_lock(&bond->curr_slave_lock);
2662 bond_select_active_slave(bond);
2664 write_unlock(&bond->curr_slave_lock);
2668 if (bond->params.arp_interval) {
2669 mod_timer(&bond->arp_timer, jiffies + delta_in_ticks);
2672 read_unlock(&bond->lock);
2676 * When using arp monitoring in active-backup mode, this function is
2677 * called to determine if any backup slaves have went down or a new
2678 * current slave needs to be found.
2679 * The backup slaves never generate traffic, they are considered up by merely
2680 * receiving traffic. If the current slave goes down, each backup slave will
2681 * be given the opportunity to tx/rx an arp before being taken down - this
2682 * prevents all slaves from being taken down due to the current slave not
2683 * sending any traffic for the backups to receive. The arps are not necessarily
2684 * necessary, any tx and rx traffic will keep the current slave up. While any
2685 * rx traffic will keep the backup slaves up, the current slave is responsible
2686 * for generating traffic to keep them up regardless of any other traffic they
2687 * may have received.
2688 * see loadbalance_arp_monitor for arp monitoring in load balancing mode
2690 void bond_activebackup_arp_mon(struct net_device *bond_dev)
2692 struct bonding *bond = bond_dev->priv;
2693 struct slave *slave;
2697 read_lock(&bond->lock);
2699 delta_in_ticks = (bond->params.arp_interval * HZ) / 1000;
2701 if (bond->kill_timers) {
2705 if (bond->slave_cnt == 0) {
2709 /* determine if any slave has come up or any backup slave has
2711 * TODO: what about up/down delay in arp mode? it wasn't here before
2714 bond_for_each_slave(bond, slave, i) {
2715 if (slave->link != BOND_LINK_UP) {
2716 if ((jiffies - slave_last_rx(bond, slave)) <=
2719 slave->link = BOND_LINK_UP;
2721 write_lock(&bond->curr_slave_lock);
2723 if ((!bond->curr_active_slave) &&
2724 ((jiffies - slave->dev->trans_start) <= delta_in_ticks)) {
2725 bond_change_active_slave(bond, slave);
2726 bond->current_arp_slave = NULL;
2727 } else if (bond->curr_active_slave != slave) {
2728 /* this slave has just come up but we
2729 * already have a current slave; this
2730 * can also happen if bond_enslave adds
2731 * a new slave that is up while we are
2732 * searching for a new slave
2734 bond_set_slave_inactive_flags(slave);
2735 bond->current_arp_slave = NULL;
2738 bond_set_carrier(bond);
2740 if (slave == bond->curr_active_slave) {
2741 printk(KERN_INFO DRV_NAME
2742 ": %s: %s is up and now the "
2743 "active interface\n",
2746 netif_carrier_on(bond->dev);
2748 printk(KERN_INFO DRV_NAME
2749 ": %s: backup interface %s is "
2755 write_unlock(&bond->curr_slave_lock);
2758 read_lock(&bond->curr_slave_lock);
2760 if ((slave != bond->curr_active_slave) &&
2761 (!bond->current_arp_slave) &&
2762 (((jiffies - slave_last_rx(bond, slave)) >= 3*delta_in_ticks) &&
2763 bond_has_ip(bond))) {
2764 /* a backup slave has gone down; three times
2765 * the delta allows the current slave to be
2766 * taken out before the backup slave.
2767 * note: a non-null current_arp_slave indicates
2768 * the curr_active_slave went down and we are
2769 * searching for a new one; under this
2770 * condition we only take the curr_active_slave
2771 * down - this gives each slave a chance to
2772 * tx/rx traffic before being taken out
2775 read_unlock(&bond->curr_slave_lock);
2777 slave->link = BOND_LINK_DOWN;
2779 if (slave->link_failure_count < UINT_MAX) {
2780 slave->link_failure_count++;
2783 bond_set_slave_inactive_flags(slave);
2785 printk(KERN_INFO DRV_NAME
2786 ": %s: backup interface %s is now down\n",
2790 read_unlock(&bond->curr_slave_lock);
2795 read_lock(&bond->curr_slave_lock);
2796 slave = bond->curr_active_slave;
2797 read_unlock(&bond->curr_slave_lock);
2800 /* if we have sent traffic in the past 2*arp_intervals but
2801 * haven't xmit and rx traffic in that time interval, select
2802 * a different slave. slave->jiffies is only updated when
2803 * a slave first becomes the curr_active_slave - not necessarily
2804 * after every arp; this ensures the slave has a full 2*delta
2805 * before being taken out. if a primary is being used, check
2806 * if it is up and needs to take over as the curr_active_slave
2808 if ((((jiffies - slave->dev->trans_start) >= (2*delta_in_ticks)) ||
2809 (((jiffies - slave_last_rx(bond, slave)) >= (2*delta_in_ticks)) &&
2810 bond_has_ip(bond))) &&
2811 ((jiffies - slave->jiffies) >= 2*delta_in_ticks)) {
2813 slave->link = BOND_LINK_DOWN;
2815 if (slave->link_failure_count < UINT_MAX) {
2816 slave->link_failure_count++;
2819 printk(KERN_INFO DRV_NAME
2820 ": %s: link status down for active interface "
2821 "%s, disabling it\n",
2825 write_lock(&bond->curr_slave_lock);
2827 bond_select_active_slave(bond);
2828 slave = bond->curr_active_slave;
2830 write_unlock(&bond->curr_slave_lock);
2832 bond->current_arp_slave = slave;
2835 slave->jiffies = jiffies;
2837 } else if ((bond->primary_slave) &&
2838 (bond->primary_slave != slave) &&
2839 (bond->primary_slave->link == BOND_LINK_UP)) {
2840 /* at this point, slave is the curr_active_slave */
2841 printk(KERN_INFO DRV_NAME
2842 ": %s: changing from interface %s to primary "
2846 bond->primary_slave->dev->name);
2848 /* primary is up so switch to it */
2849 write_lock(&bond->curr_slave_lock);
2850 bond_change_active_slave(bond, bond->primary_slave);
2851 write_unlock(&bond->curr_slave_lock);
2853 slave = bond->primary_slave;
2854 slave->jiffies = jiffies;
2856 bond->current_arp_slave = NULL;
2859 /* the current slave must tx an arp to ensure backup slaves
2862 if (slave && bond_has_ip(bond)) {
2863 bond_arp_send_all(bond, slave);
2867 /* if we don't have a curr_active_slave, search for the next available
2868 * backup slave from the current_arp_slave and make it the candidate
2869 * for becoming the curr_active_slave
2872 if (!bond->current_arp_slave) {
2873 bond->current_arp_slave = bond->first_slave;
2876 if (bond->current_arp_slave) {
2877 bond_set_slave_inactive_flags(bond->current_arp_slave);
2879 /* search for next candidate */
2880 bond_for_each_slave_from(bond, slave, i, bond->current_arp_slave->next) {
2881 if (IS_UP(slave->dev)) {
2882 slave->link = BOND_LINK_BACK;
2883 bond_set_slave_active_flags(slave);
2884 bond_arp_send_all(bond, slave);
2885 slave->jiffies = jiffies;
2886 bond->current_arp_slave = slave;
2890 /* if the link state is up at this point, we
2891 * mark it down - this can happen if we have
2892 * simultaneous link failures and
2893 * reselect_active_interface doesn't make this
2894 * one the current slave so it is still marked
2895 * up when it is actually down
2897 if (slave->link == BOND_LINK_UP) {
2898 slave->link = BOND_LINK_DOWN;
2899 if (slave->link_failure_count < UINT_MAX) {
2900 slave->link_failure_count++;
2903 bond_set_slave_inactive_flags(slave);
2905 printk(KERN_INFO DRV_NAME
2906 ": %s: backup interface %s is "
2916 if (bond->params.arp_interval) {
2917 mod_timer(&bond->arp_timer, jiffies + delta_in_ticks);
2920 read_unlock(&bond->lock);
2923 /*------------------------------ proc/seq_file-------------------------------*/
2925 #ifdef CONFIG_PROC_FS
2927 #define SEQ_START_TOKEN ((void *)1)
2929 static void *bond_info_seq_start(struct seq_file *seq, loff_t *pos)
2931 struct bonding *bond = seq->private;
2933 struct slave *slave;
2936 /* make sure the bond won't be taken away */
2937 read_lock(&dev_base_lock);
2938 read_lock_bh(&bond->lock);
2941 return SEQ_START_TOKEN;
2944 bond_for_each_slave(bond, slave, i) {
2945 if (++off == *pos) {
2953 static void *bond_info_seq_next(struct seq_file *seq, void *v, loff_t *pos)
2955 struct bonding *bond = seq->private;
2956 struct slave *slave = v;
2959 if (v == SEQ_START_TOKEN) {
2960 return bond->first_slave;
2963 slave = slave->next;
2965 return (slave == bond->first_slave) ? NULL : slave;
2968 static void bond_info_seq_stop(struct seq_file *seq, void *v)
2970 struct bonding *bond = seq->private;
2972 read_unlock_bh(&bond->lock);
2973 read_unlock(&dev_base_lock);
2976 static void bond_info_show_master(struct seq_file *seq)
2978 struct bonding *bond = seq->private;
2983 read_lock(&bond->curr_slave_lock);
2984 curr = bond->curr_active_slave;
2985 read_unlock(&bond->curr_slave_lock);
2987 seq_printf(seq, "Bonding Mode: %s\n",
2988 bond_mode_name(bond->params.mode));
2990 if (bond->params.mode == BOND_MODE_XOR ||
2991 bond->params.mode == BOND_MODE_8023AD) {
2992 seq_printf(seq, "Transmit Hash Policy: %s (%d)\n",
2993 xmit_hashtype_tbl[bond->params.xmit_policy].modename,
2994 bond->params.xmit_policy);
2997 if (USES_PRIMARY(bond->params.mode)) {
2998 seq_printf(seq, "Primary Slave: %s\n",
2999 (bond->primary_slave) ?
3000 bond->primary_slave->dev->name : "None");
3002 seq_printf(seq, "Currently Active Slave: %s\n",
3003 (curr) ? curr->dev->name : "None");
3006 seq_printf(seq, "MII Status: %s\n", netif_carrier_ok(bond->dev) ?
3008 seq_printf(seq, "MII Polling Interval (ms): %d\n", bond->params.miimon);
3009 seq_printf(seq, "Up Delay (ms): %d\n",
3010 bond->params.updelay * bond->params.miimon);
3011 seq_printf(seq, "Down Delay (ms): %d\n",
3012 bond->params.downdelay * bond->params.miimon);
3015 /* ARP information */
3016 if(bond->params.arp_interval > 0) {
3018 seq_printf(seq, "ARP Polling Interval (ms): %d\n",
3019 bond->params.arp_interval);
3021 seq_printf(seq, "ARP IP target/s (n.n.n.n form):");
3023 for(i = 0; (i < BOND_MAX_ARP_TARGETS) ;i++) {
3024 if (!bond->params.arp_targets[i])
3027 seq_printf(seq, ",");
3028 target = ntohl(bond->params.arp_targets[i]);
3029 seq_printf(seq, " %d.%d.%d.%d", HIPQUAD(target));
3032 seq_printf(seq, "\n");
3035 if (bond->params.mode == BOND_MODE_8023AD) {
3036 struct ad_info ad_info;
3037 DECLARE_MAC_BUF(mac);
3039 seq_puts(seq, "\n802.3ad info\n");
3040 seq_printf(seq, "LACP rate: %s\n",
3041 (bond->params.lacp_fast) ? "fast" : "slow");
3043 if (bond_3ad_get_active_agg_info(bond, &ad_info)) {
3044 seq_printf(seq, "bond %s has no active aggregator\n",
3047 seq_printf(seq, "Active Aggregator Info:\n");
3049 seq_printf(seq, "\tAggregator ID: %d\n",
3050 ad_info.aggregator_id);
3051 seq_printf(seq, "\tNumber of ports: %d\n",
3053 seq_printf(seq, "\tActor Key: %d\n",
3055 seq_printf(seq, "\tPartner Key: %d\n",
3056 ad_info.partner_key);
3057 seq_printf(seq, "\tPartner Mac Address: %s\n",
3058 print_mac(mac, ad_info.partner_system));
3063 static void bond_info_show_slave(struct seq_file *seq, const struct slave *slave)
3065 struct bonding *bond = seq->private;
3066 DECLARE_MAC_BUF(mac);
3068 seq_printf(seq, "\nSlave Interface: %s\n", slave->dev->name);
3069 seq_printf(seq, "MII Status: %s\n",
3070 (slave->link == BOND_LINK_UP) ? "up" : "down");
3071 seq_printf(seq, "Link Failure Count: %u\n",
3072 slave->link_failure_count);
3075 "Permanent HW addr: %s\n",
3076 print_mac(mac, slave->perm_hwaddr));
3078 if (bond->params.mode == BOND_MODE_8023AD) {
3079 const struct aggregator *agg
3080 = SLAVE_AD_INFO(slave).port.aggregator;
3083 seq_printf(seq, "Aggregator ID: %d\n",
3084 agg->aggregator_identifier);
3086 seq_puts(seq, "Aggregator ID: N/A\n");
3091 static int bond_info_seq_show(struct seq_file *seq, void *v)
3093 if (v == SEQ_START_TOKEN) {
3094 seq_printf(seq, "%s\n", version);
3095 bond_info_show_master(seq);
3097 bond_info_show_slave(seq, v);
3103 static struct seq_operations bond_info_seq_ops = {
3104 .start = bond_info_seq_start,
3105 .next = bond_info_seq_next,
3106 .stop = bond_info_seq_stop,
3107 .show = bond_info_seq_show,
3110 static int bond_info_open(struct inode *inode, struct file *file)
3112 struct seq_file *seq;
3113 struct proc_dir_entry *proc;
3116 res = seq_open(file, &bond_info_seq_ops);
3118 /* recover the pointer buried in proc_dir_entry data */
3119 seq = file->private_data;
3121 seq->private = proc->data;
3127 static const struct file_operations bond_info_fops = {
3128 .owner = THIS_MODULE,
3129 .open = bond_info_open,
3131 .llseek = seq_lseek,
3132 .release = seq_release,
3135 static int bond_create_proc_entry(struct bonding *bond)
3137 struct net_device *bond_dev = bond->dev;
3139 if (bond_proc_dir) {
3140 bond->proc_entry = create_proc_entry(bond_dev->name,
3143 if (bond->proc_entry == NULL) {
3144 printk(KERN_WARNING DRV_NAME
3145 ": Warning: Cannot create /proc/net/%s/%s\n",
3146 DRV_NAME, bond_dev->name);
3148 bond->proc_entry->data = bond;
3149 bond->proc_entry->proc_fops = &bond_info_fops;
3150 bond->proc_entry->owner = THIS_MODULE;
3151 memcpy(bond->proc_file_name, bond_dev->name, IFNAMSIZ);
3158 static void bond_remove_proc_entry(struct bonding *bond)
3160 if (bond_proc_dir && bond->proc_entry) {
3161 remove_proc_entry(bond->proc_file_name, bond_proc_dir);
3162 memset(bond->proc_file_name, 0, IFNAMSIZ);
3163 bond->proc_entry = NULL;
3167 /* Create the bonding directory under /proc/net, if doesn't exist yet.
3168 * Caller must hold rtnl_lock.
3170 static void bond_create_proc_dir(void)
3172 int len = strlen(DRV_NAME);
3174 for (bond_proc_dir = init_net.proc_net->subdir; bond_proc_dir;
3175 bond_proc_dir = bond_proc_dir->next) {
3176 if ((bond_proc_dir->namelen == len) &&
3177 !memcmp(bond_proc_dir->name, DRV_NAME, len)) {
3182 if (!bond_proc_dir) {
3183 bond_proc_dir = proc_mkdir(DRV_NAME, init_net.proc_net);
3184 if (bond_proc_dir) {
3185 bond_proc_dir->owner = THIS_MODULE;
3187 printk(KERN_WARNING DRV_NAME
3188 ": Warning: cannot create /proc/net/%s\n",
3194 /* Destroy the bonding directory under /proc/net, if empty.
3195 * Caller must hold rtnl_lock.
3197 static void bond_destroy_proc_dir(void)
3199 struct proc_dir_entry *de;
3201 if (!bond_proc_dir) {
3205 /* verify that the /proc dir is empty */
3206 for (de = bond_proc_dir->subdir; de; de = de->next) {
3207 /* ignore . and .. */
3208 if (*(de->name) != '.') {
3214 if (bond_proc_dir->owner == THIS_MODULE) {
3215 bond_proc_dir->owner = NULL;
3218 remove_proc_entry(DRV_NAME, init_net.proc_net);
3219 bond_proc_dir = NULL;
3222 #endif /* CONFIG_PROC_FS */
3224 /*-------------------------- netdev event handling --------------------------*/
3227 * Change device name
3229 static int bond_event_changename(struct bonding *bond)
3231 #ifdef CONFIG_PROC_FS
3232 bond_remove_proc_entry(bond);
3233 bond_create_proc_entry(bond);
3235 down_write(&(bonding_rwsem));
3236 bond_destroy_sysfs_entry(bond);
3237 bond_create_sysfs_entry(bond);
3238 up_write(&(bonding_rwsem));
3242 static int bond_master_netdev_event(unsigned long event, struct net_device *bond_dev)
3244 struct bonding *event_bond = bond_dev->priv;
3247 case NETDEV_CHANGENAME:
3248 return bond_event_changename(event_bond);
3249 case NETDEV_UNREGISTER:
3251 * TODO: remove a bond from the list?
3261 static int bond_slave_netdev_event(unsigned long event, struct net_device *slave_dev)
3263 struct net_device *bond_dev = slave_dev->master;
3264 struct bonding *bond = bond_dev->priv;
3267 case NETDEV_UNREGISTER:
3269 bond_release(bond_dev, slave_dev);
3274 * TODO: is this what we get if somebody
3275 * sets up a hierarchical bond, then rmmod's
3276 * one of the slave bonding devices?
3281 * ... Or is it this?
3284 case NETDEV_CHANGEMTU:
3286 * TODO: Should slaves be allowed to
3287 * independently alter their MTU? For
3288 * an active-backup bond, slaves need
3289 * not be the same type of device, so
3290 * MTUs may vary. For other modes,
3291 * slaves arguably should have the
3292 * same MTUs. To do this, we'd need to
3293 * take over the slave's change_mtu
3294 * function for the duration of their
3298 case NETDEV_CHANGENAME:
3300 * TODO: handle changing the primary's name
3303 case NETDEV_FEAT_CHANGE:
3304 bond_compute_features(bond);
3314 * bond_netdev_event: handle netdev notifier chain events.
3316 * This function receives events for the netdev chain. The caller (an
3317 * ioctl handler calling blocking_notifier_call_chain) holds the necessary
3318 * locks for us to safely manipulate the slave devices (RTNL lock,
3321 static int bond_netdev_event(struct notifier_block *this, unsigned long event, void *ptr)
3323 struct net_device *event_dev = (struct net_device *)ptr;
3325 if (event_dev->nd_net != &init_net)
3328 dprintk("event_dev: %s, event: %lx\n",
3329 (event_dev ? event_dev->name : "None"),
3332 if (!(event_dev->priv_flags & IFF_BONDING))
3335 if (event_dev->flags & IFF_MASTER) {
3336 dprintk("IFF_MASTER\n");
3337 return bond_master_netdev_event(event, event_dev);
3340 if (event_dev->flags & IFF_SLAVE) {
3341 dprintk("IFF_SLAVE\n");
3342 return bond_slave_netdev_event(event, event_dev);
3349 * bond_inetaddr_event: handle inetaddr notifier chain events.
3351 * We keep track of device IPs primarily to use as source addresses in
3352 * ARP monitor probes (rather than spewing out broadcasts all the time).
3354 * We track one IP for the main device (if it has one), plus one per VLAN.
3356 static int bond_inetaddr_event(struct notifier_block *this, unsigned long event, void *ptr)
3358 struct in_ifaddr *ifa = ptr;
3359 struct net_device *vlan_dev, *event_dev = ifa->ifa_dev->dev;
3360 struct bonding *bond, *bond_next;
3361 struct vlan_entry *vlan, *vlan_next;
3363 list_for_each_entry_safe(bond, bond_next, &bond_dev_list, bond_list) {
3364 if (bond->dev == event_dev) {
3367 bond->master_ip = ifa->ifa_local;
3370 bond->master_ip = bond_glean_dev_ip(bond->dev);
3377 if (list_empty(&bond->vlan_list))
3380 list_for_each_entry_safe(vlan, vlan_next, &bond->vlan_list,
3382 vlan_dev = vlan_group_get_device(bond->vlgrp, vlan->vlan_id);
3383 if (vlan_dev == event_dev) {
3386 vlan->vlan_ip = ifa->ifa_local;
3390 bond_glean_dev_ip(vlan_dev);
3401 static struct notifier_block bond_netdev_notifier = {
3402 .notifier_call = bond_netdev_event,
3405 static struct notifier_block bond_inetaddr_notifier = {
3406 .notifier_call = bond_inetaddr_event,
3409 /*-------------------------- Packet type handling ---------------------------*/
3411 /* register to receive lacpdus on a bond */
3412 static void bond_register_lacpdu(struct bonding *bond)
3414 struct packet_type *pk_type = &(BOND_AD_INFO(bond).ad_pkt_type);
3416 /* initialize packet type */
3417 pk_type->type = PKT_TYPE_LACPDU;
3418 pk_type->dev = bond->dev;
3419 pk_type->func = bond_3ad_lacpdu_recv;
3421 dev_add_pack(pk_type);
3424 /* unregister to receive lacpdus on a bond */
3425 static void bond_unregister_lacpdu(struct bonding *bond)
3427 dev_remove_pack(&(BOND_AD_INFO(bond).ad_pkt_type));
3430 void bond_register_arp(struct bonding *bond)
3432 struct packet_type *pt = &bond->arp_mon_pt;
3437 pt->type = htons(ETH_P_ARP);
3438 pt->dev = bond->dev;
3439 pt->func = bond_arp_rcv;
3443 void bond_unregister_arp(struct bonding *bond)
3445 struct packet_type *pt = &bond->arp_mon_pt;
3447 dev_remove_pack(pt);
3451 /*---------------------------- Hashing Policies -----------------------------*/
3454 * Hash for the output device based upon layer 3 and layer 4 data. If
3455 * the packet is a frag or not TCP or UDP, just use layer 3 data. If it is
3456 * altogether not IP, mimic bond_xmit_hash_policy_l2()
3458 static int bond_xmit_hash_policy_l34(struct sk_buff *skb,
3459 struct net_device *bond_dev, int count)
3461 struct ethhdr *data = (struct ethhdr *)skb->data;
3462 struct iphdr *iph = ip_hdr(skb);
3463 __be16 *layer4hdr = (__be16 *)((u32 *)iph + iph->ihl);
3466 if (skb->protocol == __constant_htons(ETH_P_IP)) {
3467 if (!(iph->frag_off & __constant_htons(IP_MF|IP_OFFSET)) &&
3468 (iph->protocol == IPPROTO_TCP ||
3469 iph->protocol == IPPROTO_UDP)) {
3470 layer4_xor = ntohs((*layer4hdr ^ *(layer4hdr + 1)));
3472 return (layer4_xor ^
3473 ((ntohl(iph->saddr ^ iph->daddr)) & 0xffff)) % count;
3477 return (data->h_dest[5] ^ bond_dev->dev_addr[5]) % count;
3481 * Hash for the output device based upon layer 2 data
3483 static int bond_xmit_hash_policy_l2(struct sk_buff *skb,
3484 struct net_device *bond_dev, int count)
3486 struct ethhdr *data = (struct ethhdr *)skb->data;
3488 return (data->h_dest[5] ^ bond_dev->dev_addr[5]) % count;
3491 /*-------------------------- Device entry points ----------------------------*/
3493 static int bond_open(struct net_device *bond_dev)
3495 struct bonding *bond = bond_dev->priv;
3496 struct timer_list *mii_timer = &bond->mii_timer;
3497 struct timer_list *arp_timer = &bond->arp_timer;
3499 bond->kill_timers = 0;
3501 if ((bond->params.mode == BOND_MODE_TLB) ||
3502 (bond->params.mode == BOND_MODE_ALB)) {
3503 struct timer_list *alb_timer = &(BOND_ALB_INFO(bond).alb_timer);
3505 /* bond_alb_initialize must be called before the timer
3508 if (bond_alb_initialize(bond, (bond->params.mode == BOND_MODE_ALB))) {
3509 /* something went wrong - fail the open operation */
3513 init_timer(alb_timer);
3514 alb_timer->expires = jiffies + 1;
3515 alb_timer->data = (unsigned long)bond;
3516 alb_timer->function = (void *)&bond_alb_monitor;
3517 add_timer(alb_timer);
3520 if (bond->params.miimon) { /* link check interval, in milliseconds. */
3521 init_timer(mii_timer);
3522 mii_timer->expires = jiffies + 1;
3523 mii_timer->data = (unsigned long)bond_dev;
3524 mii_timer->function = (void *)&bond_mii_monitor;
3525 add_timer(mii_timer);
3528 if (bond->params.arp_interval) { /* arp interval, in milliseconds. */
3529 init_timer(arp_timer);
3530 arp_timer->expires = jiffies + 1;
3531 arp_timer->data = (unsigned long)bond_dev;
3532 if (bond->params.mode == BOND_MODE_ACTIVEBACKUP) {
3533 arp_timer->function = (void *)&bond_activebackup_arp_mon;
3535 arp_timer->function = (void *)&bond_loadbalance_arp_mon;
3537 if (bond->params.arp_validate)
3538 bond_register_arp(bond);
3540 add_timer(arp_timer);
3543 if (bond->params.mode == BOND_MODE_8023AD) {
3544 struct timer_list *ad_timer = &(BOND_AD_INFO(bond).ad_timer);
3545 init_timer(ad_timer);
3546 ad_timer->expires = jiffies + 1;
3547 ad_timer->data = (unsigned long)bond;
3548 ad_timer->function = (void *)&bond_3ad_state_machine_handler;
3549 add_timer(ad_timer);
3551 /* register to receive LACPDUs */
3552 bond_register_lacpdu(bond);
3558 static int bond_close(struct net_device *bond_dev)
3560 struct bonding *bond = bond_dev->priv;
3562 if (bond->params.mode == BOND_MODE_8023AD) {
3563 /* Unregister the receive of LACPDUs */
3564 bond_unregister_lacpdu(bond);
3567 if (bond->params.arp_validate)
3568 bond_unregister_arp(bond);
3570 write_lock_bh(&bond->lock);
3573 /* signal timers not to re-arm */
3574 bond->kill_timers = 1;
3576 write_unlock_bh(&bond->lock);
3578 /* del_timer_sync must run without holding the bond->lock
3579 * because a running timer might be trying to hold it too
3582 if (bond->params.miimon) { /* link check interval, in milliseconds. */
3583 del_timer_sync(&bond->mii_timer);
3586 if (bond->params.arp_interval) { /* arp interval, in milliseconds. */
3587 del_timer_sync(&bond->arp_timer);
3590 switch (bond->params.mode) {
3591 case BOND_MODE_8023AD:
3592 del_timer_sync(&(BOND_AD_INFO(bond).ad_timer));
3596 del_timer_sync(&(BOND_ALB_INFO(bond).alb_timer));
3603 if ((bond->params.mode == BOND_MODE_TLB) ||
3604 (bond->params.mode == BOND_MODE_ALB)) {
3605 /* Must be called only after all
3606 * slaves have been released
3608 bond_alb_deinitialize(bond);
3614 static struct net_device_stats *bond_get_stats(struct net_device *bond_dev)
3616 struct bonding *bond = bond_dev->priv;
3617 struct net_device_stats *stats = &(bond->stats), *sstats;
3618 struct slave *slave;
3621 memset(stats, 0, sizeof(struct net_device_stats));
3623 read_lock_bh(&bond->lock);
3625 bond_for_each_slave(bond, slave, i) {
3626 sstats = slave->dev->get_stats(slave->dev);
3627 stats->rx_packets += sstats->rx_packets;
3628 stats->rx_bytes += sstats->rx_bytes;
3629 stats->rx_errors += sstats->rx_errors;
3630 stats->rx_dropped += sstats->rx_dropped;
3632 stats->tx_packets += sstats->tx_packets;
3633 stats->tx_bytes += sstats->tx_bytes;
3634 stats->tx_errors += sstats->tx_errors;
3635 stats->tx_dropped += sstats->tx_dropped;
3637 stats->multicast += sstats->multicast;
3638 stats->collisions += sstats->collisions;
3640 stats->rx_length_errors += sstats->rx_length_errors;
3641 stats->rx_over_errors += sstats->rx_over_errors;
3642 stats->rx_crc_errors += sstats->rx_crc_errors;
3643 stats->rx_frame_errors += sstats->rx_frame_errors;
3644 stats->rx_fifo_errors += sstats->rx_fifo_errors;
3645 stats->rx_missed_errors += sstats->rx_missed_errors;
3647 stats->tx_aborted_errors += sstats->tx_aborted_errors;
3648 stats->tx_carrier_errors += sstats->tx_carrier_errors;
3649 stats->tx_fifo_errors += sstats->tx_fifo_errors;
3650 stats->tx_heartbeat_errors += sstats->tx_heartbeat_errors;
3651 stats->tx_window_errors += sstats->tx_window_errors;
3654 read_unlock_bh(&bond->lock);
3659 static int bond_do_ioctl(struct net_device *bond_dev, struct ifreq *ifr, int cmd)
3661 struct net_device *slave_dev = NULL;
3662 struct ifbond k_binfo;
3663 struct ifbond __user *u_binfo = NULL;
3664 struct ifslave k_sinfo;
3665 struct ifslave __user *u_sinfo = NULL;
3666 struct mii_ioctl_data *mii = NULL;
3669 dprintk("bond_ioctl: master=%s, cmd=%d\n",
3670 bond_dev->name, cmd);
3682 * We do this again just in case we were called by SIOCGMIIREG
3683 * instead of SIOCGMIIPHY.
3690 if (mii->reg_num == 1) {
3691 struct bonding *bond = bond_dev->priv;
3693 read_lock_bh(&bond->lock);
3694 read_lock(&bond->curr_slave_lock);
3695 if (netif_carrier_ok(bond->dev)) {
3696 mii->val_out = BMSR_LSTATUS;
3698 read_unlock(&bond->curr_slave_lock);
3699 read_unlock_bh(&bond->lock);
3703 case BOND_INFO_QUERY_OLD:
3704 case SIOCBONDINFOQUERY:
3705 u_binfo = (struct ifbond __user *)ifr->ifr_data;
3707 if (copy_from_user(&k_binfo, u_binfo, sizeof(ifbond))) {
3711 res = bond_info_query(bond_dev, &k_binfo);
3713 if (copy_to_user(u_binfo, &k_binfo, sizeof(ifbond))) {
3719 case BOND_SLAVE_INFO_QUERY_OLD:
3720 case SIOCBONDSLAVEINFOQUERY:
3721 u_sinfo = (struct ifslave __user *)ifr->ifr_data;
3723 if (copy_from_user(&k_sinfo, u_sinfo, sizeof(ifslave))) {
3727 res = bond_slave_info_query(bond_dev, &k_sinfo);
3729 if (copy_to_user(u_sinfo, &k_sinfo, sizeof(ifslave))) {
3740 if (!capable(CAP_NET_ADMIN)) {
3744 down_write(&(bonding_rwsem));
3745 slave_dev = dev_get_by_name(&init_net, ifr->ifr_slave);
3747 dprintk("slave_dev=%p: \n", slave_dev);
3752 dprintk("slave_dev->name=%s: \n", slave_dev->name);
3754 case BOND_ENSLAVE_OLD:
3755 case SIOCBONDENSLAVE:
3756 res = bond_enslave(bond_dev, slave_dev);
3758 case BOND_RELEASE_OLD:
3759 case SIOCBONDRELEASE:
3760 res = bond_release(bond_dev, slave_dev);
3762 case BOND_SETHWADDR_OLD:
3763 case SIOCBONDSETHWADDR:
3764 res = bond_sethwaddr(bond_dev, slave_dev);
3766 case BOND_CHANGE_ACTIVE_OLD:
3767 case SIOCBONDCHANGEACTIVE:
3768 res = bond_ioctl_change_active(bond_dev, slave_dev);
3777 up_write(&(bonding_rwsem));
3781 static void bond_set_multicast_list(struct net_device *bond_dev)
3783 struct bonding *bond = bond_dev->priv;
3784 struct dev_mc_list *dmi;
3786 write_lock_bh(&bond->lock);
3789 * Do promisc before checking multicast_mode
3791 if ((bond_dev->flags & IFF_PROMISC) && !(bond->flags & IFF_PROMISC)) {
3792 bond_set_promiscuity(bond, 1);
3795 if (!(bond_dev->flags & IFF_PROMISC) && (bond->flags & IFF_PROMISC)) {
3796 bond_set_promiscuity(bond, -1);
3799 /* set allmulti flag to slaves */
3800 if ((bond_dev->flags & IFF_ALLMULTI) && !(bond->flags & IFF_ALLMULTI)) {
3801 bond_set_allmulti(bond, 1);
3804 if (!(bond_dev->flags & IFF_ALLMULTI) && (bond->flags & IFF_ALLMULTI)) {
3805 bond_set_allmulti(bond, -1);
3808 bond->flags = bond_dev->flags;
3810 /* looking for addresses to add to slaves' mc list */
3811 for (dmi = bond_dev->mc_list; dmi; dmi = dmi->next) {
3812 if (!bond_mc_list_find_dmi(dmi, bond->mc_list)) {
3813 bond_mc_add(bond, dmi->dmi_addr, dmi->dmi_addrlen);
3817 /* looking for addresses to delete from slaves' list */
3818 for (dmi = bond->mc_list; dmi; dmi = dmi->next) {
3819 if (!bond_mc_list_find_dmi(dmi, bond_dev->mc_list)) {
3820 bond_mc_delete(bond, dmi->dmi_addr, dmi->dmi_addrlen);
3824 /* save master's multicast list */
3825 bond_mc_list_destroy(bond);
3826 bond_mc_list_copy(bond_dev->mc_list, bond, GFP_ATOMIC);
3828 write_unlock_bh(&bond->lock);
3832 * Change the MTU of all of a master's slaves to match the master
3834 static int bond_change_mtu(struct net_device *bond_dev, int new_mtu)
3836 struct bonding *bond = bond_dev->priv;
3837 struct slave *slave, *stop_at;
3841 dprintk("bond=%p, name=%s, new_mtu=%d\n", bond,
3842 (bond_dev ? bond_dev->name : "None"), new_mtu);
3844 /* Can't hold bond->lock with bh disabled here since
3845 * some base drivers panic. On the other hand we can't
3846 * hold bond->lock without bh disabled because we'll
3847 * deadlock. The only solution is to rely on the fact
3848 * that we're under rtnl_lock here, and the slaves
3849 * list won't change. This doesn't solve the problem
3850 * of setting the slave's MTU while it is
3851 * transmitting, but the assumption is that the base
3852 * driver can handle that.
3854 * TODO: figure out a way to safely iterate the slaves
3855 * list, but without holding a lock around the actual
3856 * call to the base driver.
3859 bond_for_each_slave(bond, slave, i) {
3860 dprintk("s %p s->p %p c_m %p\n", slave,
3861 slave->prev, slave->dev->change_mtu);
3863 res = dev_set_mtu(slave->dev, new_mtu);
3866 /* If we failed to set the slave's mtu to the new value
3867 * we must abort the operation even in ACTIVE_BACKUP
3868 * mode, because if we allow the backup slaves to have
3869 * different mtu values than the active slave we'll
3870 * need to change their mtu when doing a failover. That
3871 * means changing their mtu from timer context, which
3872 * is probably not a good idea.
3874 dprintk("err %d %s\n", res, slave->dev->name);
3879 bond_dev->mtu = new_mtu;
3884 /* unwind from head to the slave that failed */
3886 bond_for_each_slave_from_to(bond, slave, i, bond->first_slave, stop_at) {
3889 tmp_res = dev_set_mtu(slave->dev, bond_dev->mtu);
3891 dprintk("unwind err %d dev %s\n", tmp_res,
3902 * Note that many devices must be down to change the HW address, and
3903 * downing the master releases all slaves. We can make bonds full of
3904 * bonding devices to test this, however.
3906 static int bond_set_mac_address(struct net_device *bond_dev, void *addr)
3908 struct bonding *bond = bond_dev->priv;
3909 struct sockaddr *sa = addr, tmp_sa;
3910 struct slave *slave, *stop_at;
3914 dprintk("bond=%p, name=%s\n", bond, (bond_dev ? bond_dev->name : "None"));
3916 if (!is_valid_ether_addr(sa->sa_data)) {
3917 return -EADDRNOTAVAIL;
3920 /* Can't hold bond->lock with bh disabled here since
3921 * some base drivers panic. On the other hand we can't
3922 * hold bond->lock without bh disabled because we'll
3923 * deadlock. The only solution is to rely on the fact
3924 * that we're under rtnl_lock here, and the slaves
3925 * list won't change. This doesn't solve the problem
3926 * of setting the slave's hw address while it is
3927 * transmitting, but the assumption is that the base
3928 * driver can handle that.
3930 * TODO: figure out a way to safely iterate the slaves
3931 * list, but without holding a lock around the actual
3932 * call to the base driver.
3935 bond_for_each_slave(bond, slave, i) {
3936 dprintk("slave %p %s\n", slave, slave->dev->name);
3938 if (slave->dev->set_mac_address == NULL) {
3940 dprintk("EOPNOTSUPP %s\n", slave->dev->name);
3944 res = dev_set_mac_address(slave->dev, addr);
3946 /* TODO: consider downing the slave
3948 * User should expect communications
3949 * breakage anyway until ARP finish
3952 dprintk("err %d %s\n", res, slave->dev->name);
3958 memcpy(bond_dev->dev_addr, sa->sa_data, bond_dev->addr_len);
3962 memcpy(tmp_sa.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
3963 tmp_sa.sa_family = bond_dev->type;
3965 /* unwind from head to the slave that failed */
3967 bond_for_each_slave_from_to(bond, slave, i, bond->first_slave, stop_at) {
3970 tmp_res = dev_set_mac_address(slave->dev, &tmp_sa);
3972 dprintk("unwind err %d dev %s\n", tmp_res,
3980 static int bond_xmit_roundrobin(struct sk_buff *skb, struct net_device *bond_dev)
3982 struct bonding *bond = bond_dev->priv;
3983 struct slave *slave, *start_at;
3987 read_lock(&bond->lock);
3989 if (!BOND_IS_OK(bond)) {
3993 read_lock(&bond->curr_slave_lock);
3994 slave = start_at = bond->curr_active_slave;
3995 read_unlock(&bond->curr_slave_lock);
4001 bond_for_each_slave_from(bond, slave, i, start_at) {
4002 if (IS_UP(slave->dev) &&
4003 (slave->link == BOND_LINK_UP) &&
4004 (slave->state == BOND_STATE_ACTIVE)) {
4005 res = bond_dev_queue_xmit(bond, skb, slave->dev);
4007 write_lock(&bond->curr_slave_lock);
4008 bond->curr_active_slave = slave->next;
4009 write_unlock(&bond->curr_slave_lock);
4018 /* no suitable interface, frame not sent */
4021 read_unlock(&bond->lock);
4027 * in active-backup mode, we know that bond->curr_active_slave is always valid if
4028 * the bond has a usable interface.
4030 static int bond_xmit_activebackup(struct sk_buff *skb, struct net_device *bond_dev)
4032 struct bonding *bond = bond_dev->priv;
4035 read_lock(&bond->lock);
4036 read_lock(&bond->curr_slave_lock);
4038 if (!BOND_IS_OK(bond)) {
4042 if (!bond->curr_active_slave)
4045 res = bond_dev_queue_xmit(bond, skb, bond->curr_active_slave->dev);
4049 /* no suitable interface, frame not sent */
4052 read_unlock(&bond->curr_slave_lock);
4053 read_unlock(&bond->lock);
4058 * In bond_xmit_xor() , we determine the output device by using a pre-
4059 * determined xmit_hash_policy(), If the selected device is not enabled,
4060 * find the next active slave.
4062 static int bond_xmit_xor(struct sk_buff *skb, struct net_device *bond_dev)
4064 struct bonding *bond = bond_dev->priv;
4065 struct slave *slave, *start_at;
4070 read_lock(&bond->lock);
4072 if (!BOND_IS_OK(bond)) {
4076 slave_no = bond->xmit_hash_policy(skb, bond_dev, bond->slave_cnt);
4078 bond_for_each_slave(bond, slave, i) {
4087 bond_for_each_slave_from(bond, slave, i, start_at) {
4088 if (IS_UP(slave->dev) &&
4089 (slave->link == BOND_LINK_UP) &&
4090 (slave->state == BOND_STATE_ACTIVE)) {
4091 res = bond_dev_queue_xmit(bond, skb, slave->dev);
4098 /* no suitable interface, frame not sent */
4101 read_unlock(&bond->lock);
4106 * in broadcast mode, we send everything to all usable interfaces.
4108 static int bond_xmit_broadcast(struct sk_buff *skb, struct net_device *bond_dev)
4110 struct bonding *bond = bond_dev->priv;
4111 struct slave *slave, *start_at;
4112 struct net_device *tx_dev = NULL;
4116 read_lock(&bond->lock);
4118 if (!BOND_IS_OK(bond)) {
4122 read_lock(&bond->curr_slave_lock);
4123 start_at = bond->curr_active_slave;
4124 read_unlock(&bond->curr_slave_lock);
4130 bond_for_each_slave_from(bond, slave, i, start_at) {
4131 if (IS_UP(slave->dev) &&
4132 (slave->link == BOND_LINK_UP) &&
4133 (slave->state == BOND_STATE_ACTIVE)) {
4135 struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
4137 printk(KERN_ERR DRV_NAME
4138 ": %s: Error: bond_xmit_broadcast(): "
4139 "skb_clone() failed\n",
4144 res = bond_dev_queue_xmit(bond, skb2, tx_dev);
4146 dev_kfree_skb(skb2);
4150 tx_dev = slave->dev;
4155 res = bond_dev_queue_xmit(bond, skb, tx_dev);
4160 /* no suitable interface, frame not sent */
4163 /* frame sent to all suitable interfaces */
4164 read_unlock(&bond->lock);
4168 /*------------------------- Device initialization ---------------------------*/
4171 * set bond mode specific net device operations
4173 void bond_set_mode_ops(struct bonding *bond, int mode)
4175 struct net_device *bond_dev = bond->dev;
4178 case BOND_MODE_ROUNDROBIN:
4179 bond_dev->hard_start_xmit = bond_xmit_roundrobin;
4181 case BOND_MODE_ACTIVEBACKUP:
4182 bond_dev->hard_start_xmit = bond_xmit_activebackup;
4185 bond_dev->hard_start_xmit = bond_xmit_xor;
4186 if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER34)
4187 bond->xmit_hash_policy = bond_xmit_hash_policy_l34;
4189 bond->xmit_hash_policy = bond_xmit_hash_policy_l2;
4191 case BOND_MODE_BROADCAST:
4192 bond_dev->hard_start_xmit = bond_xmit_broadcast;
4194 case BOND_MODE_8023AD:
4195 bond_set_master_3ad_flags(bond);
4196 bond_dev->hard_start_xmit = bond_3ad_xmit_xor;
4197 if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER34)
4198 bond->xmit_hash_policy = bond_xmit_hash_policy_l34;
4200 bond->xmit_hash_policy = bond_xmit_hash_policy_l2;
4203 bond_set_master_alb_flags(bond);
4206 bond_dev->hard_start_xmit = bond_alb_xmit;
4207 bond_dev->set_mac_address = bond_alb_set_mac_address;
4210 /* Should never happen, mode already checked */
4211 printk(KERN_ERR DRV_NAME
4212 ": %s: Error: Unknown bonding mode %d\n",
4219 static void bond_ethtool_get_drvinfo(struct net_device *bond_dev,
4220 struct ethtool_drvinfo *drvinfo)
4222 strncpy(drvinfo->driver, DRV_NAME, 32);
4223 strncpy(drvinfo->version, DRV_VERSION, 32);
4224 snprintf(drvinfo->fw_version, 32, "%d", BOND_ABI_VERSION);
4227 static const struct ethtool_ops bond_ethtool_ops = {
4228 .get_drvinfo = bond_ethtool_get_drvinfo,
4232 * Does not allocate but creates a /proc entry.
4235 static int bond_init(struct net_device *bond_dev, struct bond_params *params)
4237 struct bonding *bond = bond_dev->priv;
4239 dprintk("Begin bond_init for %s\n", bond_dev->name);
4241 /* initialize rwlocks */
4242 rwlock_init(&bond->lock);
4243 rwlock_init(&bond->curr_slave_lock);
4245 bond->params = *params; /* copy params struct */
4247 /* Initialize pointers */
4248 bond->first_slave = NULL;
4249 bond->curr_active_slave = NULL;
4250 bond->current_arp_slave = NULL;
4251 bond->primary_slave = NULL;
4252 bond->dev = bond_dev;
4253 INIT_LIST_HEAD(&bond->vlan_list);
4255 /* Initialize the device entry points */
4256 bond_dev->open = bond_open;
4257 bond_dev->stop = bond_close;
4258 bond_dev->get_stats = bond_get_stats;
4259 bond_dev->do_ioctl = bond_do_ioctl;
4260 bond_dev->ethtool_ops = &bond_ethtool_ops;
4261 bond_dev->set_multicast_list = bond_set_multicast_list;
4262 bond_dev->change_mtu = bond_change_mtu;
4263 bond_dev->set_mac_address = bond_set_mac_address;
4265 bond_set_mode_ops(bond, bond->params.mode);
4267 bond_dev->destructor = free_netdev;
4269 /* Initialize the device options */
4270 bond_dev->tx_queue_len = 0;
4271 bond_dev->flags |= IFF_MASTER|IFF_MULTICAST;
4272 bond_dev->priv_flags |= IFF_BONDING;
4274 /* At first, we block adding VLANs. That's the only way to
4275 * prevent problems that occur when adding VLANs over an
4276 * empty bond. The block will be removed once non-challenged
4277 * slaves are enslaved.
4279 bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
4281 /* don't acquire bond device's netif_tx_lock when
4283 bond_dev->features |= NETIF_F_LLTX;
4285 /* By default, we declare the bond to be fully
4286 * VLAN hardware accelerated capable. Special
4287 * care is taken in the various xmit functions
4288 * when there are slaves that are not hw accel
4291 bond_dev->vlan_rx_register = bond_vlan_rx_register;
4292 bond_dev->vlan_rx_add_vid = bond_vlan_rx_add_vid;
4293 bond_dev->vlan_rx_kill_vid = bond_vlan_rx_kill_vid;
4294 bond_dev->features |= (NETIF_F_HW_VLAN_TX |
4295 NETIF_F_HW_VLAN_RX |
4296 NETIF_F_HW_VLAN_FILTER);
4298 #ifdef CONFIG_PROC_FS
4299 bond_create_proc_entry(bond);
4302 list_add_tail(&bond->bond_list, &bond_dev_list);
4307 /* De-initialize device specific data.
4308 * Caller must hold rtnl_lock.
4310 void bond_deinit(struct net_device *bond_dev)
4312 struct bonding *bond = bond_dev->priv;
4314 list_del(&bond->bond_list);
4316 #ifdef CONFIG_PROC_FS
4317 bond_remove_proc_entry(bond);
4321 /* Unregister and free all bond devices.
4322 * Caller must hold rtnl_lock.
4324 static void bond_free_all(void)
4326 struct bonding *bond, *nxt;
4328 list_for_each_entry_safe(bond, nxt, &bond_dev_list, bond_list) {
4329 struct net_device *bond_dev = bond->dev;
4331 bond_mc_list_destroy(bond);
4332 /* Release the bonded slaves */
4333 bond_release_all(bond_dev);
4334 bond_deinit(bond_dev);
4335 unregister_netdevice(bond_dev);
4338 #ifdef CONFIG_PROC_FS
4339 bond_destroy_proc_dir();
4343 /*------------------------- Module initialization ---------------------------*/
4346 * Convert string input module parms. Accept either the
4347 * number of the mode or its string name.
4349 int bond_parse_parm(char *mode_arg, struct bond_parm_tbl *tbl)
4353 for (i = 0; tbl[i].modename; i++) {
4354 if ((isdigit(*mode_arg) &&
4355 tbl[i].mode == simple_strtol(mode_arg, NULL, 0)) ||
4356 (strncmp(mode_arg, tbl[i].modename,
4357 strlen(tbl[i].modename)) == 0)) {
4365 static int bond_check_params(struct bond_params *params)
4367 int arp_validate_value;
4370 * Convert string parameters.
4373 bond_mode = bond_parse_parm(mode, bond_mode_tbl);
4374 if (bond_mode == -1) {
4375 printk(KERN_ERR DRV_NAME
4376 ": Error: Invalid bonding mode \"%s\"\n",
4377 mode == NULL ? "NULL" : mode);
4382 if (xmit_hash_policy) {
4383 if ((bond_mode != BOND_MODE_XOR) &&
4384 (bond_mode != BOND_MODE_8023AD)) {
4385 printk(KERN_INFO DRV_NAME
4386 ": xor_mode param is irrelevant in mode %s\n",
4387 bond_mode_name(bond_mode));
4389 xmit_hashtype = bond_parse_parm(xmit_hash_policy,
4391 if (xmit_hashtype == -1) {
4392 printk(KERN_ERR DRV_NAME
4393 ": Error: Invalid xmit_hash_policy \"%s\"\n",
4394 xmit_hash_policy == NULL ? "NULL" :
4402 if (bond_mode != BOND_MODE_8023AD) {
4403 printk(KERN_INFO DRV_NAME
4404 ": lacp_rate param is irrelevant in mode %s\n",
4405 bond_mode_name(bond_mode));
4407 lacp_fast = bond_parse_parm(lacp_rate, bond_lacp_tbl);
4408 if (lacp_fast == -1) {
4409 printk(KERN_ERR DRV_NAME
4410 ": Error: Invalid lacp rate \"%s\"\n",
4411 lacp_rate == NULL ? "NULL" : lacp_rate);
4417 if (max_bonds < 1 || max_bonds > INT_MAX) {
4418 printk(KERN_WARNING DRV_NAME
4419 ": Warning: max_bonds (%d) not in range %d-%d, so it "
4420 "was reset to BOND_DEFAULT_MAX_BONDS (%d)\n",
4421 max_bonds, 1, INT_MAX, BOND_DEFAULT_MAX_BONDS);
4422 max_bonds = BOND_DEFAULT_MAX_BONDS;
4426 printk(KERN_WARNING DRV_NAME
4427 ": Warning: miimon module parameter (%d), "
4428 "not in range 0-%d, so it was reset to %d\n",
4429 miimon, INT_MAX, BOND_LINK_MON_INTERV);
4430 miimon = BOND_LINK_MON_INTERV;
4434 printk(KERN_WARNING DRV_NAME
4435 ": Warning: updelay module parameter (%d), "
4436 "not in range 0-%d, so it was reset to 0\n",
4441 if (downdelay < 0) {
4442 printk(KERN_WARNING DRV_NAME
4443 ": Warning: downdelay module parameter (%d), "
4444 "not in range 0-%d, so it was reset to 0\n",
4445 downdelay, INT_MAX);
4449 if ((use_carrier != 0) && (use_carrier != 1)) {
4450 printk(KERN_WARNING DRV_NAME
4451 ": Warning: use_carrier module parameter (%d), "
4452 "not of valid value (0/1), so it was set to 1\n",
4457 /* reset values for 802.3ad */
4458 if (bond_mode == BOND_MODE_8023AD) {
4460 printk(KERN_WARNING DRV_NAME
4461 ": Warning: miimon must be specified, "
4462 "otherwise bonding will not detect link "
4463 "failure, speed and duplex which are "
4464 "essential for 802.3ad operation\n");
4465 printk(KERN_WARNING "Forcing miimon to 100msec\n");
4470 /* reset values for TLB/ALB */
4471 if ((bond_mode == BOND_MODE_TLB) ||
4472 (bond_mode == BOND_MODE_ALB)) {
4474 printk(KERN_WARNING DRV_NAME
4475 ": Warning: miimon must be specified, "
4476 "otherwise bonding will not detect link "
4477 "failure and link speed which are essential "
4478 "for TLB/ALB load balancing\n");
4479 printk(KERN_WARNING "Forcing miimon to 100msec\n");
4484 if (bond_mode == BOND_MODE_ALB) {
4485 printk(KERN_NOTICE DRV_NAME
4486 ": In ALB mode you might experience client "
4487 "disconnections upon reconnection of a link if the "
4488 "bonding module updelay parameter (%d msec) is "
4489 "incompatible with the forwarding delay time of the "
4495 if (updelay || downdelay) {
4496 /* just warn the user the up/down delay will have
4497 * no effect since miimon is zero...
4499 printk(KERN_WARNING DRV_NAME
4500 ": Warning: miimon module parameter not set "
4501 "and updelay (%d) or downdelay (%d) module "
4502 "parameter is set; updelay and downdelay have "
4503 "no effect unless miimon is set\n",
4504 updelay, downdelay);
4507 /* don't allow arp monitoring */
4509 printk(KERN_WARNING DRV_NAME
4510 ": Warning: miimon (%d) and arp_interval (%d) "
4511 "can't be used simultaneously, disabling ARP "
4513 miimon, arp_interval);
4517 if ((updelay % miimon) != 0) {
4518 printk(KERN_WARNING DRV_NAME
4519 ": Warning: updelay (%d) is not a multiple "
4520 "of miimon (%d), updelay rounded to %d ms\n",
4521 updelay, miimon, (updelay / miimon) * miimon);
4526 if ((downdelay % miimon) != 0) {
4527 printk(KERN_WARNING DRV_NAME
4528 ": Warning: downdelay (%d) is not a multiple "
4529 "of miimon (%d), downdelay rounded to %d ms\n",
4531 (downdelay / miimon) * miimon);
4534 downdelay /= miimon;
4537 if (arp_interval < 0) {
4538 printk(KERN_WARNING DRV_NAME
4539 ": Warning: arp_interval module parameter (%d) "
4540 ", not in range 0-%d, so it was reset to %d\n",
4541 arp_interval, INT_MAX, BOND_LINK_ARP_INTERV);
4542 arp_interval = BOND_LINK_ARP_INTERV;
4545 for (arp_ip_count = 0;
4546 (arp_ip_count < BOND_MAX_ARP_TARGETS) && arp_ip_target[arp_ip_count];
4548 /* not complete check, but should be good enough to
4550 if (!isdigit(arp_ip_target[arp_ip_count][0])) {
4551 printk(KERN_WARNING DRV_NAME
4552 ": Warning: bad arp_ip_target module parameter "
4553 "(%s), ARP monitoring will not be performed\n",
4554 arp_ip_target[arp_ip_count]);
4557 __be32 ip = in_aton(arp_ip_target[arp_ip_count]);
4558 arp_target[arp_ip_count] = ip;
4562 if (arp_interval && !arp_ip_count) {
4563 /* don't allow arping if no arp_ip_target given... */
4564 printk(KERN_WARNING DRV_NAME
4565 ": Warning: arp_interval module parameter (%d) "
4566 "specified without providing an arp_ip_target "
4567 "parameter, arp_interval was reset to 0\n",
4573 if (bond_mode != BOND_MODE_ACTIVEBACKUP) {
4574 printk(KERN_ERR DRV_NAME
4575 ": arp_validate only supported in active-backup mode\n");
4578 if (!arp_interval) {
4579 printk(KERN_ERR DRV_NAME
4580 ": arp_validate requires arp_interval\n");
4584 arp_validate_value = bond_parse_parm(arp_validate,
4586 if (arp_validate_value == -1) {
4587 printk(KERN_ERR DRV_NAME
4588 ": Error: invalid arp_validate \"%s\"\n",
4589 arp_validate == NULL ? "NULL" : arp_validate);
4593 arp_validate_value = 0;
4596 printk(KERN_INFO DRV_NAME
4597 ": MII link monitoring set to %d ms\n",
4599 } else if (arp_interval) {
4602 printk(KERN_INFO DRV_NAME
4603 ": ARP monitoring set to %d ms, validate %s, with %d target(s):",
4605 arp_validate_tbl[arp_validate_value].modename,
4608 for (i = 0; i < arp_ip_count; i++)
4609 printk (" %s", arp_ip_target[i]);
4614 /* miimon and arp_interval not set, we need one so things
4615 * work as expected, see bonding.txt for details
4617 printk(KERN_WARNING DRV_NAME
4618 ": Warning: either miimon or arp_interval and "
4619 "arp_ip_target module parameters must be specified, "
4620 "otherwise bonding will not detect link failures! see "
4621 "bonding.txt for details.\n");
4624 if (primary && !USES_PRIMARY(bond_mode)) {
4625 /* currently, using a primary only makes sense
4626 * in active backup, TLB or ALB modes
4628 printk(KERN_WARNING DRV_NAME
4629 ": Warning: %s primary device specified but has no "
4630 "effect in %s mode\n",
4631 primary, bond_mode_name(bond_mode));
4635 /* fill params struct with the proper values */
4636 params->mode = bond_mode;
4637 params->xmit_policy = xmit_hashtype;
4638 params->miimon = miimon;
4639 params->arp_interval = arp_interval;
4640 params->arp_validate = arp_validate_value;
4641 params->updelay = updelay;
4642 params->downdelay = downdelay;
4643 params->use_carrier = use_carrier;
4644 params->lacp_fast = lacp_fast;
4645 params->primary[0] = 0;
4648 strncpy(params->primary, primary, IFNAMSIZ);
4649 params->primary[IFNAMSIZ - 1] = 0;
4652 memcpy(params->arp_targets, arp_target, sizeof(arp_target));
4657 static struct lock_class_key bonding_netdev_xmit_lock_key;
4659 /* Create a new bond based on the specified name and bonding parameters.
4660 * If name is NULL, obtain a suitable "bond%d" name for us.
4661 * Caller must NOT hold rtnl_lock; we need to release it here before we
4662 * set up our sysfs entries.
4664 int bond_create(char *name, struct bond_params *params, struct bonding **newbond)
4666 struct net_device *bond_dev;
4670 bond_dev = alloc_netdev(sizeof(struct bonding), name ? name : "",
4673 printk(KERN_ERR DRV_NAME
4674 ": %s: eek! can't alloc netdev!\n",
4681 res = dev_alloc_name(bond_dev, "bond%d");
4686 /* bond_init() must be called after dev_alloc_name() (for the
4687 * /proc files), but before register_netdevice(), because we
4688 * need to set function pointers.
4691 res = bond_init(bond_dev, params);
4696 res = register_netdevice(bond_dev);
4701 lockdep_set_class(&bond_dev->_xmit_lock, &bonding_netdev_xmit_lock_key);
4704 *newbond = bond_dev->priv;
4706 netif_carrier_off(bond_dev);
4708 rtnl_unlock(); /* allows sysfs registration of net device */
4709 res = bond_create_sysfs_entry(bond_dev->priv);
4718 bond_deinit(bond_dev);
4720 free_netdev(bond_dev);
4726 static int __init bonding_init(void)
4731 printk(KERN_INFO "%s", version);
4733 res = bond_check_params(&bonding_defaults);
4738 #ifdef CONFIG_PROC_FS
4739 bond_create_proc_dir();
4741 for (i = 0; i < max_bonds; i++) {
4742 res = bond_create(NULL, &bonding_defaults, NULL);
4747 res = bond_create_sysfs();
4751 register_netdevice_notifier(&bond_netdev_notifier);
4752 register_inetaddr_notifier(&bond_inetaddr_notifier);
4758 bond_destroy_sysfs();
4765 static void __exit bonding_exit(void)
4767 unregister_netdevice_notifier(&bond_netdev_notifier);
4768 unregister_inetaddr_notifier(&bond_inetaddr_notifier);
4772 bond_destroy_sysfs();
4776 module_init(bonding_init);
4777 module_exit(bonding_exit);
4778 MODULE_LICENSE("GPL");
4779 MODULE_VERSION(DRV_VERSION);
4780 MODULE_DESCRIPTION(DRV_DESCRIPTION ", v" DRV_VERSION);
4781 MODULE_AUTHOR("Thomas Davis, tadavis@lbl.gov and many others");
4782 MODULE_SUPPORTED_DEVICE("most ethernet devices");