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 #include <linux/kernel.h>
35 #include <linux/module.h>
36 #include <linux/types.h>
37 #include <linux/fcntl.h>
38 #include <linux/interrupt.h>
39 #include <linux/ptrace.h>
40 #include <linux/ioport.h>
44 #include <linux/tcp.h>
45 #include <linux/udp.h>
46 #include <linux/slab.h>
47 #include <linux/string.h>
48 #include <linux/init.h>
49 #include <linux/timer.h>
50 #include <linux/socket.h>
51 #include <linux/ctype.h>
52 #include <linux/inet.h>
53 #include <linux/bitops.h>
55 #include <asm/system.h>
57 #include <linux/uaccess.h>
58 #include <linux/errno.h>
59 #include <linux/netdevice.h>
60 #include <linux/inetdevice.h>
61 #include <linux/igmp.h>
62 #include <linux/etherdevice.h>
63 #include <linux/skbuff.h>
65 #include <linux/rtnetlink.h>
66 #include <linux/proc_fs.h>
67 #include <linux/seq_file.h>
68 #include <linux/smp.h>
69 #include <linux/if_ether.h>
71 #include <linux/mii.h>
72 #include <linux/ethtool.h>
73 #include <linux/if_vlan.h>
74 #include <linux/if_bonding.h>
75 #include <linux/jiffies.h>
76 #include <net/route.h>
77 #include <net/net_namespace.h>
82 /*---------------------------- Module parameters ----------------------------*/
84 /* monitor all links that often (in milliseconds). <=0 disables monitoring */
85 #define BOND_LINK_MON_INTERV 0
86 #define BOND_LINK_ARP_INTERV 0
88 static int max_bonds = BOND_DEFAULT_MAX_BONDS;
89 static int num_grat_arp = 1;
90 static int num_unsol_na = 1;
91 static int miimon = BOND_LINK_MON_INTERV;
94 static int use_carrier = 1;
97 static char *lacp_rate;
98 static char *ad_select;
99 static char *xmit_hash_policy;
100 static int arp_interval = BOND_LINK_ARP_INTERV;
101 static char *arp_ip_target[BOND_MAX_ARP_TARGETS];
102 static char *arp_validate;
103 static char *fail_over_mac;
104 static struct bond_params bonding_defaults;
106 module_param(max_bonds, int, 0);
107 MODULE_PARM_DESC(max_bonds, "Max number of bonded devices");
108 module_param(num_grat_arp, int, 0644);
109 MODULE_PARM_DESC(num_grat_arp, "Number of gratuitous ARP packets to send on failover event");
110 module_param(num_unsol_na, int, 0644);
111 MODULE_PARM_DESC(num_unsol_na, "Number of unsolicited IPv6 Neighbor Advertisements packets to send on failover event");
112 module_param(miimon, int, 0);
113 MODULE_PARM_DESC(miimon, "Link check interval in milliseconds");
114 module_param(updelay, int, 0);
115 MODULE_PARM_DESC(updelay, "Delay before considering link up, in milliseconds");
116 module_param(downdelay, int, 0);
117 MODULE_PARM_DESC(downdelay, "Delay before considering link down, "
119 module_param(use_carrier, int, 0);
120 MODULE_PARM_DESC(use_carrier, "Use netif_carrier_ok (vs MII ioctls) in miimon; "
121 "0 for off, 1 for on (default)");
122 module_param(mode, charp, 0);
123 MODULE_PARM_DESC(mode, "Mode of operation : 0 for balance-rr, "
124 "1 for active-backup, 2 for balance-xor, "
125 "3 for broadcast, 4 for 802.3ad, 5 for balance-tlb, "
126 "6 for balance-alb");
127 module_param(primary, charp, 0);
128 MODULE_PARM_DESC(primary, "Primary network device to use");
129 module_param(lacp_rate, charp, 0);
130 MODULE_PARM_DESC(lacp_rate, "LACPDU tx rate to request from 802.3ad partner "
132 module_param(ad_select, charp, 0);
133 MODULE_PARM_DESC(ad_select, "803.ad aggregation selection logic: stable (0, default), bandwidth (1), count (2)");
134 module_param(xmit_hash_policy, charp, 0);
135 MODULE_PARM_DESC(xmit_hash_policy, "XOR hashing method: 0 for layer 2 (default)"
136 ", 1 for layer 3+4");
137 module_param(arp_interval, int, 0);
138 MODULE_PARM_DESC(arp_interval, "arp interval in milliseconds");
139 module_param_array(arp_ip_target, charp, NULL, 0);
140 MODULE_PARM_DESC(arp_ip_target, "arp targets in n.n.n.n form");
141 module_param(arp_validate, charp, 0);
142 MODULE_PARM_DESC(arp_validate, "validate src/dst of ARP probes: none (default), active, backup or all");
143 module_param(fail_over_mac, charp, 0);
144 MODULE_PARM_DESC(fail_over_mac, "For active-backup, do not set all slaves to the same MAC. none (default), active or follow");
146 /*----------------------------- Global variables ----------------------------*/
148 static const char * const version =
149 DRV_DESCRIPTION ": v" DRV_VERSION " (" DRV_RELDATE ")\n";
151 LIST_HEAD(bond_dev_list);
153 #ifdef CONFIG_PROC_FS
154 static struct proc_dir_entry *bond_proc_dir;
157 static __be32 arp_target[BOND_MAX_ARP_TARGETS];
158 static int arp_ip_count;
159 static int bond_mode = BOND_MODE_ROUNDROBIN;
160 static int xmit_hashtype = BOND_XMIT_POLICY_LAYER2;
161 static int lacp_fast;
164 const struct bond_parm_tbl bond_lacp_tbl[] = {
165 { "slow", AD_LACP_SLOW},
166 { "fast", AD_LACP_FAST},
170 const struct bond_parm_tbl bond_mode_tbl[] = {
171 { "balance-rr", BOND_MODE_ROUNDROBIN},
172 { "active-backup", BOND_MODE_ACTIVEBACKUP},
173 { "balance-xor", BOND_MODE_XOR},
174 { "broadcast", BOND_MODE_BROADCAST},
175 { "802.3ad", BOND_MODE_8023AD},
176 { "balance-tlb", BOND_MODE_TLB},
177 { "balance-alb", BOND_MODE_ALB},
181 const struct bond_parm_tbl xmit_hashtype_tbl[] = {
182 { "layer2", BOND_XMIT_POLICY_LAYER2},
183 { "layer3+4", BOND_XMIT_POLICY_LAYER34},
184 { "layer2+3", BOND_XMIT_POLICY_LAYER23},
188 const struct bond_parm_tbl arp_validate_tbl[] = {
189 { "none", BOND_ARP_VALIDATE_NONE},
190 { "active", BOND_ARP_VALIDATE_ACTIVE},
191 { "backup", BOND_ARP_VALIDATE_BACKUP},
192 { "all", BOND_ARP_VALIDATE_ALL},
196 const struct bond_parm_tbl fail_over_mac_tbl[] = {
197 { "none", BOND_FOM_NONE},
198 { "active", BOND_FOM_ACTIVE},
199 { "follow", BOND_FOM_FOLLOW},
203 struct bond_parm_tbl ad_select_tbl[] = {
204 { "stable", BOND_AD_STABLE},
205 { "bandwidth", BOND_AD_BANDWIDTH},
206 { "count", BOND_AD_COUNT},
210 /*-------------------------- Forward declarations ---------------------------*/
212 static void bond_send_gratuitous_arp(struct bonding *bond);
213 static int bond_init(struct net_device *bond_dev);
214 static void bond_deinit(struct net_device *bond_dev);
216 /*---------------------------- General routines -----------------------------*/
218 static const char *bond_mode_name(int mode)
220 static const char *names[] = {
221 [BOND_MODE_ROUNDROBIN] = "load balancing (round-robin)",
222 [BOND_MODE_ACTIVEBACKUP] = "fault-tolerance (active-backup)",
223 [BOND_MODE_XOR] = "load balancing (xor)",
224 [BOND_MODE_BROADCAST] = "fault-tolerance (broadcast)",
225 [BOND_MODE_8023AD] = "IEEE 802.3ad Dynamic link aggregation",
226 [BOND_MODE_TLB] = "transmit load balancing",
227 [BOND_MODE_ALB] = "adaptive load balancing",
230 if (mode < 0 || mode > BOND_MODE_ALB)
236 /*---------------------------------- VLAN -----------------------------------*/
239 * bond_add_vlan - add a new vlan id on bond
240 * @bond: bond that got the notification
241 * @vlan_id: the vlan id to add
243 * Returns -ENOMEM if allocation failed.
245 static int bond_add_vlan(struct bonding *bond, unsigned short vlan_id)
247 struct vlan_entry *vlan;
249 pr_debug("bond: %s, vlan id %d\n",
250 (bond ? bond->dev->name : "None"), vlan_id);
252 vlan = kzalloc(sizeof(struct vlan_entry), GFP_KERNEL);
256 INIT_LIST_HEAD(&vlan->vlan_list);
257 vlan->vlan_id = vlan_id;
259 write_lock_bh(&bond->lock);
261 list_add_tail(&vlan->vlan_list, &bond->vlan_list);
263 write_unlock_bh(&bond->lock);
265 pr_debug("added VLAN ID %d on bond %s\n", vlan_id, bond->dev->name);
271 * bond_del_vlan - delete a vlan id from bond
272 * @bond: bond that got the notification
273 * @vlan_id: the vlan id to delete
275 * returns -ENODEV if @vlan_id was not found in @bond.
277 static int bond_del_vlan(struct bonding *bond, unsigned short vlan_id)
279 struct vlan_entry *vlan;
282 pr_debug("bond: %s, vlan id %d\n", bond->dev->name, vlan_id);
284 write_lock_bh(&bond->lock);
286 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
287 if (vlan->vlan_id == vlan_id) {
288 list_del(&vlan->vlan_list);
290 if (bond_is_lb(bond))
291 bond_alb_clear_vlan(bond, vlan_id);
293 pr_debug("removed VLAN ID %d from bond %s\n", vlan_id,
298 if (list_empty(&bond->vlan_list) &&
299 (bond->slave_cnt == 0)) {
300 /* Last VLAN removed and no slaves, so
301 * restore block on adding VLANs. This will
302 * be removed once new slaves that are not
303 * VLAN challenged will be added.
305 bond->dev->features |= NETIF_F_VLAN_CHALLENGED;
313 pr_debug("couldn't find VLAN ID %d in bond %s\n", vlan_id,
317 write_unlock_bh(&bond->lock);
322 * bond_has_challenged_slaves
323 * @bond: the bond we're working on
325 * Searches the slave list. Returns 1 if a vlan challenged slave
326 * was found, 0 otherwise.
328 * Assumes bond->lock is held.
330 static int bond_has_challenged_slaves(struct bonding *bond)
335 bond_for_each_slave(bond, slave, i) {
336 if (slave->dev->features & NETIF_F_VLAN_CHALLENGED) {
337 pr_debug("found VLAN challenged slave - %s\n",
343 pr_debug("no VLAN challenged slaves found\n");
348 * bond_next_vlan - safely skip to the next item in the vlans list.
349 * @bond: the bond we're working on
350 * @curr: item we're advancing from
352 * Returns %NULL if list is empty, bond->next_vlan if @curr is %NULL,
353 * or @curr->next otherwise (even if it is @curr itself again).
355 * Caller must hold bond->lock
357 struct vlan_entry *bond_next_vlan(struct bonding *bond, struct vlan_entry *curr)
359 struct vlan_entry *next, *last;
361 if (list_empty(&bond->vlan_list))
365 next = list_entry(bond->vlan_list.next,
366 struct vlan_entry, vlan_list);
368 last = list_entry(bond->vlan_list.prev,
369 struct vlan_entry, vlan_list);
371 next = list_entry(bond->vlan_list.next,
372 struct vlan_entry, vlan_list);
374 next = list_entry(curr->vlan_list.next,
375 struct vlan_entry, vlan_list);
383 * bond_dev_queue_xmit - Prepare skb for xmit.
385 * @bond: bond device that got this skb for tx.
386 * @skb: hw accel VLAN tagged skb to transmit
387 * @slave_dev: slave that is supposed to xmit this skbuff
389 * When the bond gets an skb to transmit that is
390 * already hardware accelerated VLAN tagged, and it
391 * needs to relay this skb to a slave that is not
392 * hw accel capable, the skb needs to be "unaccelerated",
393 * i.e. strip the hwaccel tag and re-insert it as part
396 int bond_dev_queue_xmit(struct bonding *bond, struct sk_buff *skb,
397 struct net_device *slave_dev)
399 unsigned short uninitialized_var(vlan_id);
401 if (!list_empty(&bond->vlan_list) &&
402 !(slave_dev->features & NETIF_F_HW_VLAN_TX) &&
403 vlan_get_tag(skb, &vlan_id) == 0) {
404 skb->dev = slave_dev;
405 skb = vlan_put_tag(skb, vlan_id);
407 /* vlan_put_tag() frees the skb in case of error,
408 * so return success here so the calling functions
409 * won't attempt to free is again.
414 skb->dev = slave_dev;
424 * In the following 3 functions, bond_vlan_rx_register(), bond_vlan_rx_add_vid
425 * and bond_vlan_rx_kill_vid, We don't protect the slave list iteration with a
427 * a. This operation is performed in IOCTL context,
428 * b. The operation is protected by the RTNL semaphore in the 8021q code,
429 * c. Holding a lock with BH disabled while directly calling a base driver
430 * entry point is generally a BAD idea.
432 * The design of synchronization/protection for this operation in the 8021q
433 * module is good for one or more VLAN devices over a single physical device
434 * and cannot be extended for a teaming solution like bonding, so there is a
435 * potential race condition here where a net device from the vlan group might
436 * be referenced (either by a base driver or the 8021q code) while it is being
437 * removed from the system. However, it turns out we're not making matters
438 * worse, and if it works for regular VLAN usage it will work here too.
442 * bond_vlan_rx_register - Propagates registration to slaves
443 * @bond_dev: bonding net device that got called
444 * @grp: vlan group being registered
446 static void bond_vlan_rx_register(struct net_device *bond_dev,
447 struct vlan_group *grp)
449 struct bonding *bond = netdev_priv(bond_dev);
455 bond_for_each_slave(bond, slave, i) {
456 struct net_device *slave_dev = slave->dev;
457 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
459 if ((slave_dev->features & NETIF_F_HW_VLAN_RX) &&
460 slave_ops->ndo_vlan_rx_register) {
461 slave_ops->ndo_vlan_rx_register(slave_dev, grp);
467 * bond_vlan_rx_add_vid - Propagates adding an id to slaves
468 * @bond_dev: bonding net device that got called
469 * @vid: vlan id being added
471 static void bond_vlan_rx_add_vid(struct net_device *bond_dev, uint16_t vid)
473 struct bonding *bond = netdev_priv(bond_dev);
477 bond_for_each_slave(bond, slave, i) {
478 struct net_device *slave_dev = slave->dev;
479 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
481 if ((slave_dev->features & NETIF_F_HW_VLAN_FILTER) &&
482 slave_ops->ndo_vlan_rx_add_vid) {
483 slave_ops->ndo_vlan_rx_add_vid(slave_dev, vid);
487 res = bond_add_vlan(bond, vid);
490 ": %s: Error: Failed to add vlan id %d\n",
491 bond_dev->name, vid);
496 * bond_vlan_rx_kill_vid - Propagates deleting an id to slaves
497 * @bond_dev: bonding net device that got called
498 * @vid: vlan id being removed
500 static void bond_vlan_rx_kill_vid(struct net_device *bond_dev, uint16_t vid)
502 struct bonding *bond = netdev_priv(bond_dev);
504 struct net_device *vlan_dev;
507 bond_for_each_slave(bond, slave, i) {
508 struct net_device *slave_dev = slave->dev;
509 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
511 if ((slave_dev->features & NETIF_F_HW_VLAN_FILTER) &&
512 slave_ops->ndo_vlan_rx_kill_vid) {
513 /* Save and then restore vlan_dev in the grp array,
514 * since the slave's driver might clear it.
516 vlan_dev = vlan_group_get_device(bond->vlgrp, vid);
517 slave_ops->ndo_vlan_rx_kill_vid(slave_dev, vid);
518 vlan_group_set_device(bond->vlgrp, vid, vlan_dev);
522 res = bond_del_vlan(bond, vid);
525 ": %s: Error: Failed to remove vlan id %d\n",
526 bond_dev->name, vid);
530 static void bond_add_vlans_on_slave(struct bonding *bond, struct net_device *slave_dev)
532 struct vlan_entry *vlan;
533 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
535 write_lock_bh(&bond->lock);
537 if (list_empty(&bond->vlan_list))
540 if ((slave_dev->features & NETIF_F_HW_VLAN_RX) &&
541 slave_ops->ndo_vlan_rx_register)
542 slave_ops->ndo_vlan_rx_register(slave_dev, bond->vlgrp);
544 if (!(slave_dev->features & NETIF_F_HW_VLAN_FILTER) ||
545 !(slave_ops->ndo_vlan_rx_add_vid))
548 list_for_each_entry(vlan, &bond->vlan_list, vlan_list)
549 slave_ops->ndo_vlan_rx_add_vid(slave_dev, vlan->vlan_id);
552 write_unlock_bh(&bond->lock);
555 static void bond_del_vlans_from_slave(struct bonding *bond,
556 struct net_device *slave_dev)
558 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
559 struct vlan_entry *vlan;
560 struct net_device *vlan_dev;
562 write_lock_bh(&bond->lock);
564 if (list_empty(&bond->vlan_list))
567 if (!(slave_dev->features & NETIF_F_HW_VLAN_FILTER) ||
568 !(slave_ops->ndo_vlan_rx_kill_vid))
571 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
572 /* Save and then restore vlan_dev in the grp array,
573 * since the slave's driver might clear it.
575 vlan_dev = vlan_group_get_device(bond->vlgrp, vlan->vlan_id);
576 slave_ops->ndo_vlan_rx_kill_vid(slave_dev, vlan->vlan_id);
577 vlan_group_set_device(bond->vlgrp, vlan->vlan_id, vlan_dev);
581 if ((slave_dev->features & NETIF_F_HW_VLAN_RX) &&
582 slave_ops->ndo_vlan_rx_register)
583 slave_ops->ndo_vlan_rx_register(slave_dev, NULL);
586 write_unlock_bh(&bond->lock);
589 /*------------------------------- Link status -------------------------------*/
592 * Set the carrier state for the master according to the state of its
593 * slaves. If any slaves are up, the master is up. In 802.3ad mode,
594 * do special 802.3ad magic.
596 * Returns zero if carrier state does not change, nonzero if it does.
598 static int bond_set_carrier(struct bonding *bond)
603 if (bond->slave_cnt == 0)
606 if (bond->params.mode == BOND_MODE_8023AD)
607 return bond_3ad_set_carrier(bond);
609 bond_for_each_slave(bond, slave, i) {
610 if (slave->link == BOND_LINK_UP) {
611 if (!netif_carrier_ok(bond->dev)) {
612 netif_carrier_on(bond->dev);
620 if (netif_carrier_ok(bond->dev)) {
621 netif_carrier_off(bond->dev);
628 * Get link speed and duplex from the slave's base driver
629 * using ethtool. If for some reason the call fails or the
630 * values are invalid, fake speed and duplex to 100/Full
633 static int bond_update_speed_duplex(struct slave *slave)
635 struct net_device *slave_dev = slave->dev;
636 struct ethtool_cmd etool;
639 /* Fake speed and duplex */
640 slave->speed = SPEED_100;
641 slave->duplex = DUPLEX_FULL;
643 if (!slave_dev->ethtool_ops || !slave_dev->ethtool_ops->get_settings)
646 res = slave_dev->ethtool_ops->get_settings(slave_dev, &etool);
650 switch (etool.speed) {
660 switch (etool.duplex) {
668 slave->speed = etool.speed;
669 slave->duplex = etool.duplex;
675 * if <dev> supports MII link status reporting, check its link status.
677 * We either do MII/ETHTOOL ioctls, or check netif_carrier_ok(),
678 * depending upon the setting of the use_carrier parameter.
680 * Return either BMSR_LSTATUS, meaning that the link is up (or we
681 * can't tell and just pretend it is), or 0, meaning that the link is
684 * If reporting is non-zero, instead of faking link up, return -1 if
685 * both ETHTOOL and MII ioctls fail (meaning the device does not
686 * support them). If use_carrier is set, return whatever it says.
687 * It'd be nice if there was a good way to tell if a driver supports
688 * netif_carrier, but there really isn't.
690 static int bond_check_dev_link(struct bonding *bond,
691 struct net_device *slave_dev, int reporting)
693 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
694 static int (*ioctl)(struct net_device *, struct ifreq *, int);
696 struct mii_ioctl_data *mii;
698 if (bond->params.use_carrier)
699 return netif_carrier_ok(slave_dev) ? BMSR_LSTATUS : 0;
701 /* Try to get link status using Ethtool first. */
702 if (slave_dev->ethtool_ops) {
703 if (slave_dev->ethtool_ops->get_link) {
706 link = slave_dev->ethtool_ops->get_link(slave_dev);
708 return link ? BMSR_LSTATUS : 0;
712 /* Ethtool can't be used, fallback to MII ioctls. */
713 ioctl = slave_ops->ndo_do_ioctl;
715 /* TODO: set pointer to correct ioctl on a per team member */
716 /* bases to make this more efficient. that is, once */
717 /* we determine the correct ioctl, we will always */
718 /* call it and not the others for that team */
722 * We cannot assume that SIOCGMIIPHY will also read a
723 * register; not all network drivers (e.g., e100)
727 /* Yes, the mii is overlaid on the ifreq.ifr_ifru */
728 strncpy(ifr.ifr_name, slave_dev->name, IFNAMSIZ);
730 if (IOCTL(slave_dev, &ifr, SIOCGMIIPHY) == 0) {
731 mii->reg_num = MII_BMSR;
732 if (IOCTL(slave_dev, &ifr, SIOCGMIIREG) == 0)
733 return mii->val_out & BMSR_LSTATUS;
738 * If reporting, report that either there's no dev->do_ioctl,
739 * or both SIOCGMIIREG and get_link failed (meaning that we
740 * cannot report link status). If not reporting, pretend
743 return reporting ? -1 : BMSR_LSTATUS;
746 /*----------------------------- Multicast list ------------------------------*/
749 * Returns 0 if dmi1 and dmi2 are the same, non-0 otherwise
751 static inline int bond_is_dmi_same(const struct dev_mc_list *dmi1,
752 const struct dev_mc_list *dmi2)
754 return memcmp(dmi1->dmi_addr, dmi2->dmi_addr, dmi1->dmi_addrlen) == 0 &&
755 dmi1->dmi_addrlen == dmi2->dmi_addrlen;
759 * returns dmi entry if found, NULL otherwise
761 static struct dev_mc_list *bond_mc_list_find_dmi(struct dev_mc_list *dmi,
762 struct dev_mc_list *mc_list)
764 struct dev_mc_list *idmi;
766 for (idmi = mc_list; idmi; idmi = idmi->next) {
767 if (bond_is_dmi_same(dmi, idmi))
775 * Push the promiscuity flag down to appropriate slaves
777 static int bond_set_promiscuity(struct bonding *bond, int inc)
780 if (USES_PRIMARY(bond->params.mode)) {
781 /* write lock already acquired */
782 if (bond->curr_active_slave) {
783 err = dev_set_promiscuity(bond->curr_active_slave->dev,
789 bond_for_each_slave(bond, slave, i) {
790 err = dev_set_promiscuity(slave->dev, inc);
799 * Push the allmulti flag down to all slaves
801 static int bond_set_allmulti(struct bonding *bond, int inc)
804 if (USES_PRIMARY(bond->params.mode)) {
805 /* write lock already acquired */
806 if (bond->curr_active_slave) {
807 err = dev_set_allmulti(bond->curr_active_slave->dev,
813 bond_for_each_slave(bond, slave, i) {
814 err = dev_set_allmulti(slave->dev, inc);
823 * Add a Multicast address to slaves
826 static void bond_mc_add(struct bonding *bond, void *addr, int alen)
828 if (USES_PRIMARY(bond->params.mode)) {
829 /* write lock already acquired */
830 if (bond->curr_active_slave)
831 dev_mc_add(bond->curr_active_slave->dev, addr, alen, 0);
836 bond_for_each_slave(bond, slave, i)
837 dev_mc_add(slave->dev, addr, alen, 0);
842 * Remove a multicast address from slave
845 static void bond_mc_delete(struct bonding *bond, void *addr, int alen)
847 if (USES_PRIMARY(bond->params.mode)) {
848 /* write lock already acquired */
849 if (bond->curr_active_slave)
850 dev_mc_delete(bond->curr_active_slave->dev, addr,
855 bond_for_each_slave(bond, slave, i) {
856 dev_mc_delete(slave->dev, addr, alen, 0);
863 * Retrieve the list of registered multicast addresses for the bonding
864 * device and retransmit an IGMP JOIN request to the current active
867 static void bond_resend_igmp_join_requests(struct bonding *bond)
869 struct in_device *in_dev;
870 struct ip_mc_list *im;
873 in_dev = __in_dev_get_rcu(bond->dev);
875 for (im = in_dev->mc_list; im; im = im->next)
876 ip_mc_rejoin_group(im);
883 * Totally destroys the mc_list in bond
885 static void bond_mc_list_destroy(struct bonding *bond)
887 struct dev_mc_list *dmi;
891 bond->mc_list = dmi->next;
896 bond->mc_list = NULL;
900 * Copy all the Multicast addresses from src to the bonding device dst
902 static int bond_mc_list_copy(struct dev_mc_list *mc_list, struct bonding *bond,
905 struct dev_mc_list *dmi, *new_dmi;
907 for (dmi = mc_list; dmi; dmi = dmi->next) {
908 new_dmi = kmalloc(sizeof(struct dev_mc_list), gfp_flag);
911 /* FIXME: Potential memory leak !!! */
915 new_dmi->next = bond->mc_list;
916 bond->mc_list = new_dmi;
917 new_dmi->dmi_addrlen = dmi->dmi_addrlen;
918 memcpy(new_dmi->dmi_addr, dmi->dmi_addr, dmi->dmi_addrlen);
919 new_dmi->dmi_users = dmi->dmi_users;
920 new_dmi->dmi_gusers = dmi->dmi_gusers;
927 * flush all members of flush->mc_list from device dev->mc_list
929 static void bond_mc_list_flush(struct net_device *bond_dev,
930 struct net_device *slave_dev)
932 struct bonding *bond = netdev_priv(bond_dev);
933 struct dev_mc_list *dmi;
935 for (dmi = bond_dev->mc_list; dmi; dmi = dmi->next)
936 dev_mc_delete(slave_dev, dmi->dmi_addr, dmi->dmi_addrlen, 0);
938 if (bond->params.mode == BOND_MODE_8023AD) {
939 /* del lacpdu mc addr from mc list */
940 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
942 dev_mc_delete(slave_dev, lacpdu_multicast, ETH_ALEN, 0);
946 /*--------------------------- Active slave change ---------------------------*/
949 * Update the mc list and multicast-related flags for the new and
950 * old active slaves (if any) according to the multicast mode, and
951 * promiscuous flags unconditionally.
953 static void bond_mc_swap(struct bonding *bond, struct slave *new_active,
954 struct slave *old_active)
956 struct dev_mc_list *dmi;
958 if (!USES_PRIMARY(bond->params.mode))
959 /* nothing to do - mc list is already up-to-date on
965 if (bond->dev->flags & IFF_PROMISC)
966 dev_set_promiscuity(old_active->dev, -1);
968 if (bond->dev->flags & IFF_ALLMULTI)
969 dev_set_allmulti(old_active->dev, -1);
971 for (dmi = bond->dev->mc_list; dmi; dmi = dmi->next)
972 dev_mc_delete(old_active->dev, dmi->dmi_addr,
973 dmi->dmi_addrlen, 0);
977 /* FIXME: Signal errors upstream. */
978 if (bond->dev->flags & IFF_PROMISC)
979 dev_set_promiscuity(new_active->dev, 1);
981 if (bond->dev->flags & IFF_ALLMULTI)
982 dev_set_allmulti(new_active->dev, 1);
984 for (dmi = bond->dev->mc_list; dmi; dmi = dmi->next)
985 dev_mc_add(new_active->dev, dmi->dmi_addr,
986 dmi->dmi_addrlen, 0);
987 bond_resend_igmp_join_requests(bond);
992 * bond_do_fail_over_mac
994 * Perform special MAC address swapping for fail_over_mac settings
996 * Called with RTNL, bond->lock for read, curr_slave_lock for write_bh.
998 static void bond_do_fail_over_mac(struct bonding *bond,
999 struct slave *new_active,
1000 struct slave *old_active)
1001 __releases(&bond->curr_slave_lock)
1002 __releases(&bond->lock)
1003 __acquires(&bond->lock)
1004 __acquires(&bond->curr_slave_lock)
1006 u8 tmp_mac[ETH_ALEN];
1007 struct sockaddr saddr;
1010 switch (bond->params.fail_over_mac) {
1011 case BOND_FOM_ACTIVE:
1013 memcpy(bond->dev->dev_addr, new_active->dev->dev_addr,
1014 new_active->dev->addr_len);
1016 case BOND_FOM_FOLLOW:
1018 * if new_active && old_active, swap them
1019 * if just old_active, do nothing (going to no active slave)
1020 * if just new_active, set new_active to bond's MAC
1025 write_unlock_bh(&bond->curr_slave_lock);
1026 read_unlock(&bond->lock);
1029 memcpy(tmp_mac, new_active->dev->dev_addr, ETH_ALEN);
1030 memcpy(saddr.sa_data, old_active->dev->dev_addr,
1032 saddr.sa_family = new_active->dev->type;
1034 memcpy(saddr.sa_data, bond->dev->dev_addr, ETH_ALEN);
1035 saddr.sa_family = bond->dev->type;
1038 rv = dev_set_mac_address(new_active->dev, &saddr);
1041 ": %s: Error %d setting MAC of slave %s\n",
1042 bond->dev->name, -rv, new_active->dev->name);
1049 memcpy(saddr.sa_data, tmp_mac, ETH_ALEN);
1050 saddr.sa_family = old_active->dev->type;
1052 rv = dev_set_mac_address(old_active->dev, &saddr);
1055 ": %s: Error %d setting MAC of slave %s\n",
1056 bond->dev->name, -rv, new_active->dev->name);
1058 read_lock(&bond->lock);
1059 write_lock_bh(&bond->curr_slave_lock);
1063 ": %s: bond_do_fail_over_mac impossible: bad policy %d\n",
1064 bond->dev->name, bond->params.fail_over_mac);
1072 * find_best_interface - select the best available slave to be the active one
1073 * @bond: our bonding struct
1075 * Warning: Caller must hold curr_slave_lock for writing.
1077 static struct slave *bond_find_best_slave(struct bonding *bond)
1079 struct slave *new_active, *old_active;
1080 struct slave *bestslave = NULL;
1081 int mintime = bond->params.updelay;
1084 new_active = old_active = bond->curr_active_slave;
1086 if (!new_active) { /* there were no active slaves left */
1087 if (bond->slave_cnt > 0) /* found one slave */
1088 new_active = bond->first_slave;
1090 return NULL; /* still no slave, return NULL */
1094 * first try the primary link; if arping, a link must tx/rx
1095 * traffic before it can be considered the curr_active_slave.
1096 * also, we would skip slaves between the curr_active_slave
1097 * and primary_slave that may be up and able to arp
1099 if ((bond->primary_slave) &&
1100 (!bond->params.arp_interval) &&
1101 (IS_UP(bond->primary_slave->dev))) {
1102 new_active = bond->primary_slave;
1105 /* remember where to stop iterating over the slaves */
1106 old_active = new_active;
1108 bond_for_each_slave_from(bond, new_active, i, old_active) {
1109 if (IS_UP(new_active->dev)) {
1110 if (new_active->link == BOND_LINK_UP) {
1112 } else if (new_active->link == BOND_LINK_BACK) {
1113 /* link up, but waiting for stabilization */
1114 if (new_active->delay < mintime) {
1115 mintime = new_active->delay;
1116 bestslave = new_active;
1126 * change_active_interface - change the active slave into the specified one
1127 * @bond: our bonding struct
1128 * @new: the new slave to make the active one
1130 * Set the new slave to the bond's settings and unset them on the old
1131 * curr_active_slave.
1132 * Setting include flags, mc-list, promiscuity, allmulti, etc.
1134 * If @new's link state is %BOND_LINK_BACK we'll set it to %BOND_LINK_UP,
1135 * because it is apparently the best available slave we have, even though its
1136 * updelay hasn't timed out yet.
1138 * If new_active is not NULL, caller must hold bond->lock for read and
1139 * curr_slave_lock for write_bh.
1141 void bond_change_active_slave(struct bonding *bond, struct slave *new_active)
1143 struct slave *old_active = bond->curr_active_slave;
1145 if (old_active == new_active)
1149 new_active->jiffies = jiffies;
1151 if (new_active->link == BOND_LINK_BACK) {
1152 if (USES_PRIMARY(bond->params.mode)) {
1154 ": %s: making interface %s the new "
1155 "active one %d ms earlier.\n",
1156 bond->dev->name, new_active->dev->name,
1157 (bond->params.updelay - new_active->delay) * bond->params.miimon);
1160 new_active->delay = 0;
1161 new_active->link = BOND_LINK_UP;
1163 if (bond->params.mode == BOND_MODE_8023AD)
1164 bond_3ad_handle_link_change(new_active, BOND_LINK_UP);
1166 if (bond_is_lb(bond))
1167 bond_alb_handle_link_change(bond, new_active, BOND_LINK_UP);
1169 if (USES_PRIMARY(bond->params.mode)) {
1171 ": %s: making interface %s the new "
1173 bond->dev->name, new_active->dev->name);
1178 if (USES_PRIMARY(bond->params.mode))
1179 bond_mc_swap(bond, new_active, old_active);
1181 if (bond_is_lb(bond)) {
1182 bond_alb_handle_active_change(bond, new_active);
1184 bond_set_slave_inactive_flags(old_active);
1186 bond_set_slave_active_flags(new_active);
1188 bond->curr_active_slave = new_active;
1191 if (bond->params.mode == BOND_MODE_ACTIVEBACKUP) {
1193 bond_set_slave_inactive_flags(old_active);
1196 bond_set_slave_active_flags(new_active);
1198 if (bond->params.fail_over_mac)
1199 bond_do_fail_over_mac(bond, new_active,
1202 bond->send_grat_arp = bond->params.num_grat_arp;
1203 bond_send_gratuitous_arp(bond);
1205 bond->send_unsol_na = bond->params.num_unsol_na;
1206 bond_send_unsolicited_na(bond);
1208 write_unlock_bh(&bond->curr_slave_lock);
1209 read_unlock(&bond->lock);
1211 netdev_bonding_change(bond->dev);
1213 read_lock(&bond->lock);
1214 write_lock_bh(&bond->curr_slave_lock);
1220 * bond_select_active_slave - select a new active slave, if needed
1221 * @bond: our bonding struct
1223 * This functions should be called when one of the following occurs:
1224 * - The old curr_active_slave has been released or lost its link.
1225 * - The primary_slave has got its link back.
1226 * - A slave has got its link back and there's no old curr_active_slave.
1228 * Caller must hold bond->lock for read and curr_slave_lock for write_bh.
1230 void bond_select_active_slave(struct bonding *bond)
1232 struct slave *best_slave;
1235 best_slave = bond_find_best_slave(bond);
1236 if (best_slave != bond->curr_active_slave) {
1237 bond_change_active_slave(bond, best_slave);
1238 rv = bond_set_carrier(bond);
1242 if (netif_carrier_ok(bond->dev)) {
1244 ": %s: first active interface up!\n",
1247 pr_info(DRV_NAME ": %s: "
1248 "now running without any active interface !\n",
1254 /*--------------------------- slave list handling ---------------------------*/
1257 * This function attaches the slave to the end of list.
1259 * bond->lock held for writing by caller.
1261 static void bond_attach_slave(struct bonding *bond, struct slave *new_slave)
1263 if (bond->first_slave == NULL) { /* attaching the first slave */
1264 new_slave->next = new_slave;
1265 new_slave->prev = new_slave;
1266 bond->first_slave = new_slave;
1268 new_slave->next = bond->first_slave;
1269 new_slave->prev = bond->first_slave->prev;
1270 new_slave->next->prev = new_slave;
1271 new_slave->prev->next = new_slave;
1278 * This function detaches the slave from the list.
1279 * WARNING: no check is made to verify if the slave effectively
1280 * belongs to <bond>.
1281 * Nothing is freed on return, structures are just unchained.
1282 * If any slave pointer in bond was pointing to <slave>,
1283 * it should be changed by the calling function.
1285 * bond->lock held for writing by caller.
1287 static void bond_detach_slave(struct bonding *bond, struct slave *slave)
1290 slave->next->prev = slave->prev;
1293 slave->prev->next = slave->next;
1295 if (bond->first_slave == slave) { /* slave is the first slave */
1296 if (bond->slave_cnt > 1) { /* there are more slave */
1297 bond->first_slave = slave->next;
1299 bond->first_slave = NULL; /* slave was the last one */
1308 /*---------------------------------- IOCTL ----------------------------------*/
1310 static int bond_sethwaddr(struct net_device *bond_dev,
1311 struct net_device *slave_dev)
1313 pr_debug("bond_dev=%p\n", bond_dev);
1314 pr_debug("slave_dev=%p\n", slave_dev);
1315 pr_debug("slave_dev->addr_len=%d\n", slave_dev->addr_len);
1316 memcpy(bond_dev->dev_addr, slave_dev->dev_addr, slave_dev->addr_len);
1320 #define BOND_VLAN_FEATURES \
1321 (NETIF_F_VLAN_CHALLENGED | NETIF_F_HW_VLAN_RX | NETIF_F_HW_VLAN_TX | \
1322 NETIF_F_HW_VLAN_FILTER)
1325 * Compute the common dev->feature set available to all slaves. Some
1326 * feature bits are managed elsewhere, so preserve those feature bits
1327 * on the master device.
1329 static int bond_compute_features(struct bonding *bond)
1331 struct slave *slave;
1332 struct net_device *bond_dev = bond->dev;
1333 unsigned long features = bond_dev->features;
1334 unsigned long vlan_features = 0;
1335 unsigned short max_hard_header_len = max((u16)ETH_HLEN,
1336 bond_dev->hard_header_len);
1339 features &= ~(NETIF_F_ALL_CSUM | BOND_VLAN_FEATURES);
1340 features |= NETIF_F_GSO_MASK | NETIF_F_NO_CSUM;
1342 if (!bond->first_slave)
1345 features &= ~NETIF_F_ONE_FOR_ALL;
1347 vlan_features = bond->first_slave->dev->vlan_features;
1348 bond_for_each_slave(bond, slave, i) {
1349 features = netdev_increment_features(features,
1350 slave->dev->features,
1351 NETIF_F_ONE_FOR_ALL);
1352 vlan_features = netdev_increment_features(vlan_features,
1353 slave->dev->vlan_features,
1354 NETIF_F_ONE_FOR_ALL);
1355 if (slave->dev->hard_header_len > max_hard_header_len)
1356 max_hard_header_len = slave->dev->hard_header_len;
1360 features |= (bond_dev->features & BOND_VLAN_FEATURES);
1361 bond_dev->features = netdev_fix_features(features, NULL);
1362 bond_dev->vlan_features = netdev_fix_features(vlan_features, NULL);
1363 bond_dev->hard_header_len = max_hard_header_len;
1368 static void bond_setup_by_slave(struct net_device *bond_dev,
1369 struct net_device *slave_dev)
1371 struct bonding *bond = netdev_priv(bond_dev);
1373 bond_dev->header_ops = slave_dev->header_ops;
1375 bond_dev->type = slave_dev->type;
1376 bond_dev->hard_header_len = slave_dev->hard_header_len;
1377 bond_dev->addr_len = slave_dev->addr_len;
1379 memcpy(bond_dev->broadcast, slave_dev->broadcast,
1380 slave_dev->addr_len);
1381 bond->setup_by_slave = 1;
1384 /* enslave device <slave> to bond device <master> */
1385 int bond_enslave(struct net_device *bond_dev, struct net_device *slave_dev)
1387 struct bonding *bond = netdev_priv(bond_dev);
1388 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
1389 struct slave *new_slave = NULL;
1390 struct dev_mc_list *dmi;
1391 struct sockaddr addr;
1393 int old_features = bond_dev->features;
1396 if (!bond->params.use_carrier && slave_dev->ethtool_ops == NULL &&
1397 slave_ops->ndo_do_ioctl == NULL) {
1399 ": %s: Warning: no link monitoring support for %s\n",
1400 bond_dev->name, slave_dev->name);
1403 /* bond must be initialized by bond_open() before enslaving */
1404 if (!(bond_dev->flags & IFF_UP)) {
1406 " %s: master_dev is not up in bond_enslave\n",
1410 /* already enslaved */
1411 if (slave_dev->flags & IFF_SLAVE) {
1412 pr_debug("Error, Device was already enslaved\n");
1416 /* vlan challenged mutual exclusion */
1417 /* no need to lock since we're protected by rtnl_lock */
1418 if (slave_dev->features & NETIF_F_VLAN_CHALLENGED) {
1419 pr_debug("%s: NETIF_F_VLAN_CHALLENGED\n", slave_dev->name);
1420 if (!list_empty(&bond->vlan_list)) {
1422 ": %s: Error: cannot enslave VLAN "
1423 "challenged slave %s on VLAN enabled "
1424 "bond %s\n", bond_dev->name, slave_dev->name,
1429 ": %s: Warning: enslaved VLAN challenged "
1430 "slave %s. Adding VLANs will be blocked as "
1431 "long as %s is part of bond %s\n",
1432 bond_dev->name, slave_dev->name, slave_dev->name,
1434 bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
1437 pr_debug("%s: ! NETIF_F_VLAN_CHALLENGED\n", slave_dev->name);
1438 if (bond->slave_cnt == 0) {
1439 /* First slave, and it is not VLAN challenged,
1440 * so remove the block of adding VLANs over the bond.
1442 bond_dev->features &= ~NETIF_F_VLAN_CHALLENGED;
1447 * Old ifenslave binaries are no longer supported. These can
1448 * be identified with moderate accuracy by the state of the slave:
1449 * the current ifenslave will set the interface down prior to
1450 * enslaving it; the old ifenslave will not.
1452 if ((slave_dev->flags & IFF_UP)) {
1453 pr_err(DRV_NAME ": %s is up. "
1454 "This may be due to an out of date ifenslave.\n",
1457 goto err_undo_flags;
1460 /* set bonding device ether type by slave - bonding netdevices are
1461 * created with ether_setup, so when the slave type is not ARPHRD_ETHER
1462 * there is a need to override some of the type dependent attribs/funcs.
1464 * bond ether type mutual exclusion - don't allow slaves of dissimilar
1465 * ether type (eg ARPHRD_ETHER and ARPHRD_INFINIBAND) share the same bond
1467 if (bond->slave_cnt == 0) {
1468 if (bond_dev->type != slave_dev->type) {
1469 dev_close(bond_dev);
1470 pr_debug("%s: change device type from %d to %d\n",
1471 bond_dev->name, bond_dev->type, slave_dev->type);
1472 if (slave_dev->type != ARPHRD_ETHER)
1473 bond_setup_by_slave(bond_dev, slave_dev);
1475 ether_setup(bond_dev);
1478 } else if (bond_dev->type != slave_dev->type) {
1479 pr_err(DRV_NAME ": %s ether type (%d) is different "
1480 "from other slaves (%d), can not enslave it.\n",
1482 slave_dev->type, bond_dev->type);
1484 goto err_undo_flags;
1487 if (slave_ops->ndo_set_mac_address == NULL) {
1488 if (bond->slave_cnt == 0) {
1490 ": %s: Warning: The first slave device "
1491 "specified does not support setting the MAC "
1492 "address. Setting fail_over_mac to active.",
1494 bond->params.fail_over_mac = BOND_FOM_ACTIVE;
1495 } else if (bond->params.fail_over_mac != BOND_FOM_ACTIVE) {
1497 ": %s: Error: The slave device specified "
1498 "does not support setting the MAC address, "
1499 "but fail_over_mac is not set to active.\n"
1502 goto err_undo_flags;
1506 new_slave = kzalloc(sizeof(struct slave), GFP_KERNEL);
1509 goto err_undo_flags;
1512 /* save slave's original flags before calling
1513 * netdev_set_master and dev_open
1515 new_slave->original_flags = slave_dev->flags;
1518 * Save slave's original ("permanent") mac address for modes
1519 * that need it, and for restoring it upon release, and then
1520 * set it to the master's address
1522 memcpy(new_slave->perm_hwaddr, slave_dev->dev_addr, ETH_ALEN);
1524 if (!bond->params.fail_over_mac) {
1526 * Set slave to master's mac address. The application already
1527 * set the master's mac address to that of the first slave
1529 memcpy(addr.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
1530 addr.sa_family = slave_dev->type;
1531 res = dev_set_mac_address(slave_dev, &addr);
1533 pr_debug("Error %d calling set_mac_address\n", res);
1538 res = netdev_set_master(slave_dev, bond_dev);
1540 pr_debug("Error %d calling netdev_set_master\n", res);
1541 goto err_restore_mac;
1543 /* open the slave since the application closed it */
1544 res = dev_open(slave_dev);
1546 pr_debug("Opening slave %s failed\n", slave_dev->name);
1547 goto err_unset_master;
1550 new_slave->dev = slave_dev;
1551 slave_dev->priv_flags |= IFF_BONDING;
1553 if (bond_is_lb(bond)) {
1554 /* bond_alb_init_slave() must be called before all other stages since
1555 * it might fail and we do not want to have to undo everything
1557 res = bond_alb_init_slave(bond, new_slave);
1562 /* If the mode USES_PRIMARY, then the new slave gets the
1563 * master's promisc (and mc) settings only if it becomes the
1564 * curr_active_slave, and that is taken care of later when calling
1565 * bond_change_active()
1567 if (!USES_PRIMARY(bond->params.mode)) {
1568 /* set promiscuity level to new slave */
1569 if (bond_dev->flags & IFF_PROMISC) {
1570 res = dev_set_promiscuity(slave_dev, 1);
1575 /* set allmulti level to new slave */
1576 if (bond_dev->flags & IFF_ALLMULTI) {
1577 res = dev_set_allmulti(slave_dev, 1);
1582 netif_addr_lock_bh(bond_dev);
1583 /* upload master's mc_list to new slave */
1584 for (dmi = bond_dev->mc_list; dmi; dmi = dmi->next)
1585 dev_mc_add(slave_dev, dmi->dmi_addr,
1586 dmi->dmi_addrlen, 0);
1587 netif_addr_unlock_bh(bond_dev);
1590 if (bond->params.mode == BOND_MODE_8023AD) {
1591 /* add lacpdu mc addr to mc list */
1592 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
1594 dev_mc_add(slave_dev, lacpdu_multicast, ETH_ALEN, 0);
1597 bond_add_vlans_on_slave(bond, slave_dev);
1599 write_lock_bh(&bond->lock);
1601 bond_attach_slave(bond, new_slave);
1603 new_slave->delay = 0;
1604 new_slave->link_failure_count = 0;
1606 bond_compute_features(bond);
1608 write_unlock_bh(&bond->lock);
1610 read_lock(&bond->lock);
1612 new_slave->last_arp_rx = jiffies;
1614 if (bond->params.miimon && !bond->params.use_carrier) {
1615 link_reporting = bond_check_dev_link(bond, slave_dev, 1);
1617 if ((link_reporting == -1) && !bond->params.arp_interval) {
1619 * miimon is set but a bonded network driver
1620 * does not support ETHTOOL/MII and
1621 * arp_interval is not set. Note: if
1622 * use_carrier is enabled, we will never go
1623 * here (because netif_carrier is always
1624 * supported); thus, we don't need to change
1625 * the messages for netif_carrier.
1628 ": %s: Warning: MII and ETHTOOL support not "
1629 "available for interface %s, and "
1630 "arp_interval/arp_ip_target module parameters "
1631 "not specified, thus bonding will not detect "
1632 "link failures! see bonding.txt for details.\n",
1633 bond_dev->name, slave_dev->name);
1634 } else if (link_reporting == -1) {
1635 /* unable get link status using mii/ethtool */
1637 ": %s: Warning: can't get link status from "
1638 "interface %s; the network driver associated "
1639 "with this interface does not support MII or "
1640 "ETHTOOL link status reporting, thus miimon "
1641 "has no effect on this interface.\n",
1642 bond_dev->name, slave_dev->name);
1646 /* check for initial state */
1647 if (!bond->params.miimon ||
1648 (bond_check_dev_link(bond, slave_dev, 0) == BMSR_LSTATUS)) {
1649 if (bond->params.updelay) {
1650 pr_debug("Initial state of slave_dev is "
1651 "BOND_LINK_BACK\n");
1652 new_slave->link = BOND_LINK_BACK;
1653 new_slave->delay = bond->params.updelay;
1655 pr_debug("Initial state of slave_dev is "
1657 new_slave->link = BOND_LINK_UP;
1659 new_slave->jiffies = jiffies;
1661 pr_debug("Initial state of slave_dev is "
1662 "BOND_LINK_DOWN\n");
1663 new_slave->link = BOND_LINK_DOWN;
1666 if (bond_update_speed_duplex(new_slave) &&
1667 (new_slave->link != BOND_LINK_DOWN)) {
1669 ": %s: Warning: failed to get speed and duplex from %s, "
1670 "assumed to be 100Mb/sec and Full.\n",
1671 bond_dev->name, new_slave->dev->name);
1673 if (bond->params.mode == BOND_MODE_8023AD) {
1675 ": %s: Warning: Operation of 802.3ad mode requires ETHTOOL "
1676 "support in base driver for proper aggregator "
1677 "selection.\n", bond_dev->name);
1681 if (USES_PRIMARY(bond->params.mode) && bond->params.primary[0]) {
1682 /* if there is a primary slave, remember it */
1683 if (strcmp(bond->params.primary, new_slave->dev->name) == 0)
1684 bond->primary_slave = new_slave;
1687 write_lock_bh(&bond->curr_slave_lock);
1689 switch (bond->params.mode) {
1690 case BOND_MODE_ACTIVEBACKUP:
1691 bond_set_slave_inactive_flags(new_slave);
1692 bond_select_active_slave(bond);
1694 case BOND_MODE_8023AD:
1695 /* in 802.3ad mode, the internal mechanism
1696 * will activate the slaves in the selected
1699 bond_set_slave_inactive_flags(new_slave);
1700 /* if this is the first slave */
1701 if (bond->slave_cnt == 1) {
1702 SLAVE_AD_INFO(new_slave).id = 1;
1703 /* Initialize AD with the number of times that the AD timer is called in 1 second
1704 * can be called only after the mac address of the bond is set
1706 bond_3ad_initialize(bond, 1000/AD_TIMER_INTERVAL,
1707 bond->params.lacp_fast);
1709 SLAVE_AD_INFO(new_slave).id =
1710 SLAVE_AD_INFO(new_slave->prev).id + 1;
1713 bond_3ad_bind_slave(new_slave);
1717 new_slave->state = BOND_STATE_ACTIVE;
1718 bond_set_slave_inactive_flags(new_slave);
1719 bond_select_active_slave(bond);
1722 pr_debug("This slave is always active in trunk mode\n");
1724 /* always active in trunk mode */
1725 new_slave->state = BOND_STATE_ACTIVE;
1727 /* In trunking mode there is little meaning to curr_active_slave
1728 * anyway (it holds no special properties of the bond device),
1729 * so we can change it without calling change_active_interface()
1731 if (!bond->curr_active_slave)
1732 bond->curr_active_slave = new_slave;
1735 } /* switch(bond_mode) */
1737 write_unlock_bh(&bond->curr_slave_lock);
1739 bond_set_carrier(bond);
1741 read_unlock(&bond->lock);
1743 res = bond_create_slave_symlinks(bond_dev, slave_dev);
1748 ": %s: enslaving %s as a%s interface with a%s link.\n",
1749 bond_dev->name, slave_dev->name,
1750 new_slave->state == BOND_STATE_ACTIVE ? "n active" : " backup",
1751 new_slave->link != BOND_LINK_DOWN ? "n up" : " down");
1753 /* enslave is successful */
1756 /* Undo stages on error */
1758 dev_close(slave_dev);
1761 netdev_set_master(slave_dev, NULL);
1764 if (!bond->params.fail_over_mac) {
1765 /* XXX TODO - fom follow mode needs to change master's
1766 * MAC if this slave's MAC is in use by the bond, or at
1767 * least print a warning.
1769 memcpy(addr.sa_data, new_slave->perm_hwaddr, ETH_ALEN);
1770 addr.sa_family = slave_dev->type;
1771 dev_set_mac_address(slave_dev, &addr);
1778 bond_dev->features = old_features;
1784 * Try to release the slave device <slave> from the bond device <master>
1785 * It is legal to access curr_active_slave without a lock because all the function
1788 * The rules for slave state should be:
1789 * for Active/Backup:
1790 * Active stays on all backups go down
1791 * for Bonded connections:
1792 * The first up interface should be left on and all others downed.
1794 int bond_release(struct net_device *bond_dev, struct net_device *slave_dev)
1796 struct bonding *bond = netdev_priv(bond_dev);
1797 struct slave *slave, *oldcurrent;
1798 struct sockaddr addr;
1800 /* slave is not a slave or master is not master of this slave */
1801 if (!(slave_dev->flags & IFF_SLAVE) ||
1802 (slave_dev->master != bond_dev)) {
1804 ": %s: Error: cannot release %s.\n",
1805 bond_dev->name, slave_dev->name);
1809 write_lock_bh(&bond->lock);
1811 slave = bond_get_slave_by_dev(bond, slave_dev);
1813 /* not a slave of this bond */
1815 ": %s: %s not enslaved\n",
1816 bond_dev->name, slave_dev->name);
1817 write_unlock_bh(&bond->lock);
1821 if (!bond->params.fail_over_mac) {
1822 if (!compare_ether_addr(bond_dev->dev_addr, slave->perm_hwaddr)
1823 && bond->slave_cnt > 1)
1825 ": %s: Warning: the permanent HWaddr of %s - "
1826 "%pM - is still in use by %s. "
1827 "Set the HWaddr of %s to a different address "
1828 "to avoid conflicts.\n",
1829 bond_dev->name, slave_dev->name,
1831 bond_dev->name, slave_dev->name);
1834 /* Inform AD package of unbinding of slave. */
1835 if (bond->params.mode == BOND_MODE_8023AD) {
1836 /* must be called before the slave is
1837 * detached from the list
1839 bond_3ad_unbind_slave(slave);
1843 ": %s: releasing %s interface %s\n",
1845 (slave->state == BOND_STATE_ACTIVE)
1846 ? "active" : "backup",
1849 oldcurrent = bond->curr_active_slave;
1851 bond->current_arp_slave = NULL;
1853 /* release the slave from its bond */
1854 bond_detach_slave(bond, slave);
1856 bond_compute_features(bond);
1858 if (bond->primary_slave == slave)
1859 bond->primary_slave = NULL;
1861 if (oldcurrent == slave)
1862 bond_change_active_slave(bond, NULL);
1864 if (bond_is_lb(bond)) {
1865 /* Must be called only after the slave has been
1866 * detached from the list and the curr_active_slave
1867 * has been cleared (if our_slave == old_current),
1868 * but before a new active slave is selected.
1870 write_unlock_bh(&bond->lock);
1871 bond_alb_deinit_slave(bond, slave);
1872 write_lock_bh(&bond->lock);
1875 if (oldcurrent == slave) {
1877 * Note that we hold RTNL over this sequence, so there
1878 * is no concern that another slave add/remove event
1881 write_unlock_bh(&bond->lock);
1882 read_lock(&bond->lock);
1883 write_lock_bh(&bond->curr_slave_lock);
1885 bond_select_active_slave(bond);
1887 write_unlock_bh(&bond->curr_slave_lock);
1888 read_unlock(&bond->lock);
1889 write_lock_bh(&bond->lock);
1892 if (bond->slave_cnt == 0) {
1893 bond_set_carrier(bond);
1895 /* if the last slave was removed, zero the mac address
1896 * of the master so it will be set by the application
1897 * to the mac address of the first slave
1899 memset(bond_dev->dev_addr, 0, bond_dev->addr_len);
1901 if (list_empty(&bond->vlan_list)) {
1902 bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
1905 ": %s: Warning: clearing HW address of %s while it "
1906 "still has VLANs.\n",
1907 bond_dev->name, bond_dev->name);
1909 ": %s: When re-adding slaves, make sure the bond's "
1910 "HW address matches its VLANs'.\n",
1913 } else if ((bond_dev->features & NETIF_F_VLAN_CHALLENGED) &&
1914 !bond_has_challenged_slaves(bond)) {
1916 ": %s: last VLAN challenged slave %s "
1917 "left bond %s. VLAN blocking is removed\n",
1918 bond_dev->name, slave_dev->name, bond_dev->name);
1919 bond_dev->features &= ~NETIF_F_VLAN_CHALLENGED;
1922 write_unlock_bh(&bond->lock);
1924 /* must do this from outside any spinlocks */
1925 bond_destroy_slave_symlinks(bond_dev, slave_dev);
1927 bond_del_vlans_from_slave(bond, slave_dev);
1929 /* If the mode USES_PRIMARY, then we should only remove its
1930 * promisc and mc settings if it was the curr_active_slave, but that was
1931 * already taken care of above when we detached the slave
1933 if (!USES_PRIMARY(bond->params.mode)) {
1934 /* unset promiscuity level from slave */
1935 if (bond_dev->flags & IFF_PROMISC)
1936 dev_set_promiscuity(slave_dev, -1);
1938 /* unset allmulti level from slave */
1939 if (bond_dev->flags & IFF_ALLMULTI)
1940 dev_set_allmulti(slave_dev, -1);
1942 /* flush master's mc_list from slave */
1943 netif_addr_lock_bh(bond_dev);
1944 bond_mc_list_flush(bond_dev, slave_dev);
1945 netif_addr_unlock_bh(bond_dev);
1948 netdev_set_master(slave_dev, NULL);
1950 /* close slave before restoring its mac address */
1951 dev_close(slave_dev);
1953 if (bond->params.fail_over_mac != BOND_FOM_ACTIVE) {
1954 /* restore original ("permanent") mac address */
1955 memcpy(addr.sa_data, slave->perm_hwaddr, ETH_ALEN);
1956 addr.sa_family = slave_dev->type;
1957 dev_set_mac_address(slave_dev, &addr);
1960 slave_dev->priv_flags &= ~(IFF_MASTER_8023AD | IFF_MASTER_ALB |
1961 IFF_SLAVE_INACTIVE | IFF_BONDING |
1966 return 0; /* deletion OK */
1970 * Destroy a bonding device.
1971 * Must be under rtnl_lock when this function is called.
1973 static void bond_uninit(struct net_device *bond_dev)
1975 struct bonding *bond = netdev_priv(bond_dev);
1977 bond_deinit(bond_dev);
1978 bond_destroy_sysfs_entry(bond);
1981 destroy_workqueue(bond->wq);
1983 netif_addr_lock_bh(bond_dev);
1984 bond_mc_list_destroy(bond);
1985 netif_addr_unlock_bh(bond_dev);
1989 * First release a slave and than destroy the bond if no more slaves are left.
1990 * Must be under rtnl_lock when this function is called.
1992 int bond_release_and_destroy(struct net_device *bond_dev,
1993 struct net_device *slave_dev)
1995 struct bonding *bond = netdev_priv(bond_dev);
1998 ret = bond_release(bond_dev, slave_dev);
1999 if ((ret == 0) && (bond->slave_cnt == 0)) {
2000 pr_info(DRV_NAME ": %s: destroying bond %s.\n",
2001 bond_dev->name, bond_dev->name);
2002 unregister_netdevice(bond_dev);
2008 * This function releases all slaves.
2010 static int bond_release_all(struct net_device *bond_dev)
2012 struct bonding *bond = netdev_priv(bond_dev);
2013 struct slave *slave;
2014 struct net_device *slave_dev;
2015 struct sockaddr addr;
2017 write_lock_bh(&bond->lock);
2019 netif_carrier_off(bond_dev);
2021 if (bond->slave_cnt == 0)
2024 bond->current_arp_slave = NULL;
2025 bond->primary_slave = NULL;
2026 bond_change_active_slave(bond, NULL);
2028 while ((slave = bond->first_slave) != NULL) {
2029 /* Inform AD package of unbinding of slave
2030 * before slave is detached from the list.
2032 if (bond->params.mode == BOND_MODE_8023AD)
2033 bond_3ad_unbind_slave(slave);
2035 slave_dev = slave->dev;
2036 bond_detach_slave(bond, slave);
2038 /* now that the slave is detached, unlock and perform
2039 * all the undo steps that should not be called from
2042 write_unlock_bh(&bond->lock);
2044 if (bond_is_lb(bond)) {
2045 /* must be called only after the slave
2046 * has been detached from the list
2048 bond_alb_deinit_slave(bond, slave);
2051 bond_compute_features(bond);
2053 bond_destroy_slave_symlinks(bond_dev, slave_dev);
2054 bond_del_vlans_from_slave(bond, slave_dev);
2056 /* If the mode USES_PRIMARY, then we should only remove its
2057 * promisc and mc settings if it was the curr_active_slave, but that was
2058 * already taken care of above when we detached the slave
2060 if (!USES_PRIMARY(bond->params.mode)) {
2061 /* unset promiscuity level from slave */
2062 if (bond_dev->flags & IFF_PROMISC)
2063 dev_set_promiscuity(slave_dev, -1);
2065 /* unset allmulti level from slave */
2066 if (bond_dev->flags & IFF_ALLMULTI)
2067 dev_set_allmulti(slave_dev, -1);
2069 /* flush master's mc_list from slave */
2070 netif_addr_lock_bh(bond_dev);
2071 bond_mc_list_flush(bond_dev, slave_dev);
2072 netif_addr_unlock_bh(bond_dev);
2075 netdev_set_master(slave_dev, NULL);
2077 /* close slave before restoring its mac address */
2078 dev_close(slave_dev);
2080 if (!bond->params.fail_over_mac) {
2081 /* restore original ("permanent") mac address*/
2082 memcpy(addr.sa_data, slave->perm_hwaddr, ETH_ALEN);
2083 addr.sa_family = slave_dev->type;
2084 dev_set_mac_address(slave_dev, &addr);
2087 slave_dev->priv_flags &= ~(IFF_MASTER_8023AD | IFF_MASTER_ALB |
2088 IFF_SLAVE_INACTIVE);
2092 /* re-acquire the lock before getting the next slave */
2093 write_lock_bh(&bond->lock);
2096 /* zero the mac address of the master so it will be
2097 * set by the application to the mac address of the
2100 memset(bond_dev->dev_addr, 0, bond_dev->addr_len);
2102 if (list_empty(&bond->vlan_list))
2103 bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
2106 ": %s: Warning: clearing HW address of %s while it "
2107 "still has VLANs.\n",
2108 bond_dev->name, bond_dev->name);
2110 ": %s: When re-adding slaves, make sure the bond's "
2111 "HW address matches its VLANs'.\n",
2116 ": %s: released all slaves\n",
2120 write_unlock_bh(&bond->lock);
2126 * This function changes the active slave to slave <slave_dev>.
2127 * It returns -EINVAL in the following cases.
2128 * - <slave_dev> is not found in the list.
2129 * - There is not active slave now.
2130 * - <slave_dev> is already active.
2131 * - The link state of <slave_dev> is not BOND_LINK_UP.
2132 * - <slave_dev> is not running.
2133 * In these cases, this function does nothing.
2134 * In the other cases, current_slave pointer is changed and 0 is returned.
2136 static int bond_ioctl_change_active(struct net_device *bond_dev, struct net_device *slave_dev)
2138 struct bonding *bond = netdev_priv(bond_dev);
2139 struct slave *old_active = NULL;
2140 struct slave *new_active = NULL;
2143 if (!USES_PRIMARY(bond->params.mode))
2146 /* Verify that master_dev is indeed the master of slave_dev */
2147 if (!(slave_dev->flags & IFF_SLAVE) || (slave_dev->master != bond_dev))
2150 read_lock(&bond->lock);
2152 read_lock(&bond->curr_slave_lock);
2153 old_active = bond->curr_active_slave;
2154 read_unlock(&bond->curr_slave_lock);
2156 new_active = bond_get_slave_by_dev(bond, slave_dev);
2159 * Changing to the current active: do nothing; return success.
2161 if (new_active && (new_active == old_active)) {
2162 read_unlock(&bond->lock);
2168 (new_active->link == BOND_LINK_UP) &&
2169 IS_UP(new_active->dev)) {
2170 write_lock_bh(&bond->curr_slave_lock);
2171 bond_change_active_slave(bond, new_active);
2172 write_unlock_bh(&bond->curr_slave_lock);
2176 read_unlock(&bond->lock);
2181 static int bond_info_query(struct net_device *bond_dev, struct ifbond *info)
2183 struct bonding *bond = netdev_priv(bond_dev);
2185 info->bond_mode = bond->params.mode;
2186 info->miimon = bond->params.miimon;
2188 read_lock(&bond->lock);
2189 info->num_slaves = bond->slave_cnt;
2190 read_unlock(&bond->lock);
2195 static int bond_slave_info_query(struct net_device *bond_dev, struct ifslave *info)
2197 struct bonding *bond = netdev_priv(bond_dev);
2198 struct slave *slave;
2199 int i, res = -ENODEV;
2201 read_lock(&bond->lock);
2203 bond_for_each_slave(bond, slave, i) {
2204 if (i == (int)info->slave_id) {
2206 strcpy(info->slave_name, slave->dev->name);
2207 info->link = slave->link;
2208 info->state = slave->state;
2209 info->link_failure_count = slave->link_failure_count;
2214 read_unlock(&bond->lock);
2219 /*-------------------------------- Monitoring -------------------------------*/
2222 static int bond_miimon_inspect(struct bonding *bond)
2224 struct slave *slave;
2225 int i, link_state, commit = 0;
2226 bool ignore_updelay;
2228 ignore_updelay = !bond->curr_active_slave ? true : false;
2230 bond_for_each_slave(bond, slave, i) {
2231 slave->new_link = BOND_LINK_NOCHANGE;
2233 link_state = bond_check_dev_link(bond, slave->dev, 0);
2235 switch (slave->link) {
2240 slave->link = BOND_LINK_FAIL;
2241 slave->delay = bond->params.downdelay;
2244 ": %s: link status down for %s"
2245 "interface %s, disabling it in %d ms.\n",
2247 (bond->params.mode ==
2248 BOND_MODE_ACTIVEBACKUP) ?
2249 ((slave->state == BOND_STATE_ACTIVE) ?
2250 "active " : "backup ") : "",
2252 bond->params.downdelay * bond->params.miimon);
2255 case BOND_LINK_FAIL:
2258 * recovered before downdelay expired
2260 slave->link = BOND_LINK_UP;
2261 slave->jiffies = jiffies;
2263 ": %s: link status up again after %d "
2264 "ms for interface %s.\n",
2266 (bond->params.downdelay - slave->delay) *
2267 bond->params.miimon,
2272 if (slave->delay <= 0) {
2273 slave->new_link = BOND_LINK_DOWN;
2281 case BOND_LINK_DOWN:
2285 slave->link = BOND_LINK_BACK;
2286 slave->delay = bond->params.updelay;
2290 ": %s: link status up for "
2291 "interface %s, enabling it in %d ms.\n",
2292 bond->dev->name, slave->dev->name,
2293 ignore_updelay ? 0 :
2294 bond->params.updelay *
2295 bond->params.miimon);
2298 case BOND_LINK_BACK:
2300 slave->link = BOND_LINK_DOWN;
2302 ": %s: link status down again after %d "
2303 "ms for interface %s.\n",
2305 (bond->params.updelay - slave->delay) *
2306 bond->params.miimon,
2315 if (slave->delay <= 0) {
2316 slave->new_link = BOND_LINK_UP;
2318 ignore_updelay = false;
2330 static void bond_miimon_commit(struct bonding *bond)
2332 struct slave *slave;
2335 bond_for_each_slave(bond, slave, i) {
2336 switch (slave->new_link) {
2337 case BOND_LINK_NOCHANGE:
2341 slave->link = BOND_LINK_UP;
2342 slave->jiffies = jiffies;
2344 if (bond->params.mode == BOND_MODE_8023AD) {
2345 /* prevent it from being the active one */
2346 slave->state = BOND_STATE_BACKUP;
2347 } else if (bond->params.mode != BOND_MODE_ACTIVEBACKUP) {
2348 /* make it immediately active */
2349 slave->state = BOND_STATE_ACTIVE;
2350 } else if (slave != bond->primary_slave) {
2351 /* prevent it from being the active one */
2352 slave->state = BOND_STATE_BACKUP;
2356 ": %s: link status definitely "
2357 "up for interface %s.\n",
2358 bond->dev->name, slave->dev->name);
2360 /* notify ad that the link status has changed */
2361 if (bond->params.mode == BOND_MODE_8023AD)
2362 bond_3ad_handle_link_change(slave, BOND_LINK_UP);
2364 if (bond_is_lb(bond))
2365 bond_alb_handle_link_change(bond, slave,
2368 if (!bond->curr_active_slave ||
2369 (slave == bond->primary_slave))
2374 case BOND_LINK_DOWN:
2375 if (slave->link_failure_count < UINT_MAX)
2376 slave->link_failure_count++;
2378 slave->link = BOND_LINK_DOWN;
2380 if (bond->params.mode == BOND_MODE_ACTIVEBACKUP ||
2381 bond->params.mode == BOND_MODE_8023AD)
2382 bond_set_slave_inactive_flags(slave);
2385 ": %s: link status definitely down for "
2386 "interface %s, disabling it\n",
2387 bond->dev->name, slave->dev->name);
2389 if (bond->params.mode == BOND_MODE_8023AD)
2390 bond_3ad_handle_link_change(slave,
2393 if (bond_is_lb(bond))
2394 bond_alb_handle_link_change(bond, slave,
2397 if (slave == bond->curr_active_slave)
2404 ": %s: invalid new link %d on slave %s\n",
2405 bond->dev->name, slave->new_link,
2407 slave->new_link = BOND_LINK_NOCHANGE;
2414 write_lock_bh(&bond->curr_slave_lock);
2415 bond_select_active_slave(bond);
2416 write_unlock_bh(&bond->curr_slave_lock);
2419 bond_set_carrier(bond);
2425 * Really a wrapper that splits the mii monitor into two phases: an
2426 * inspection, then (if inspection indicates something needs to be done)
2427 * an acquisition of appropriate locks followed by a commit phase to
2428 * implement whatever link state changes are indicated.
2430 void bond_mii_monitor(struct work_struct *work)
2432 struct bonding *bond = container_of(work, struct bonding,
2435 read_lock(&bond->lock);
2436 if (bond->kill_timers)
2439 if (bond->slave_cnt == 0)
2442 if (bond->send_grat_arp) {
2443 read_lock(&bond->curr_slave_lock);
2444 bond_send_gratuitous_arp(bond);
2445 read_unlock(&bond->curr_slave_lock);
2448 if (bond->send_unsol_na) {
2449 read_lock(&bond->curr_slave_lock);
2450 bond_send_unsolicited_na(bond);
2451 read_unlock(&bond->curr_slave_lock);
2454 if (bond_miimon_inspect(bond)) {
2455 read_unlock(&bond->lock);
2457 read_lock(&bond->lock);
2459 bond_miimon_commit(bond);
2461 read_unlock(&bond->lock);
2462 rtnl_unlock(); /* might sleep, hold no other locks */
2463 read_lock(&bond->lock);
2467 if (bond->params.miimon)
2468 queue_delayed_work(bond->wq, &bond->mii_work,
2469 msecs_to_jiffies(bond->params.miimon));
2471 read_unlock(&bond->lock);
2474 static __be32 bond_glean_dev_ip(struct net_device *dev)
2476 struct in_device *idev;
2477 struct in_ifaddr *ifa;
2484 idev = __in_dev_get_rcu(dev);
2488 ifa = idev->ifa_list;
2492 addr = ifa->ifa_local;
2498 static int bond_has_this_ip(struct bonding *bond, __be32 ip)
2500 struct vlan_entry *vlan;
2502 if (ip == bond->master_ip)
2505 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
2506 if (ip == vlan->vlan_ip)
2514 * We go to the (large) trouble of VLAN tagging ARP frames because
2515 * switches in VLAN mode (especially if ports are configured as
2516 * "native" to a VLAN) might not pass non-tagged frames.
2518 static void bond_arp_send(struct net_device *slave_dev, int arp_op, __be32 dest_ip, __be32 src_ip, unsigned short vlan_id)
2520 struct sk_buff *skb;
2522 pr_debug("arp %d on slave %s: dst %x src %x vid %d\n", arp_op,
2523 slave_dev->name, dest_ip, src_ip, vlan_id);
2525 skb = arp_create(arp_op, ETH_P_ARP, dest_ip, slave_dev, src_ip,
2526 NULL, slave_dev->dev_addr, NULL);
2529 pr_err(DRV_NAME ": ARP packet allocation failed\n");
2533 skb = vlan_put_tag(skb, vlan_id);
2535 pr_err(DRV_NAME ": failed to insert VLAN tag\n");
2543 static void bond_arp_send_all(struct bonding *bond, struct slave *slave)
2546 __be32 *targets = bond->params.arp_targets;
2547 struct vlan_entry *vlan;
2548 struct net_device *vlan_dev;
2552 for (i = 0; (i < BOND_MAX_ARP_TARGETS); i++) {
2555 pr_debug("basa: target %x\n", targets[i]);
2556 if (list_empty(&bond->vlan_list)) {
2557 pr_debug("basa: empty vlan: arp_send\n");
2558 bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2559 bond->master_ip, 0);
2564 * If VLANs are configured, we do a route lookup to
2565 * determine which VLAN interface would be used, so we
2566 * can tag the ARP with the proper VLAN tag.
2568 memset(&fl, 0, sizeof(fl));
2569 fl.fl4_dst = targets[i];
2570 fl.fl4_tos = RTO_ONLINK;
2572 rv = ip_route_output_key(&init_net, &rt, &fl);
2574 if (net_ratelimit()) {
2576 ": %s: no route to arp_ip_target %pI4\n",
2577 bond->dev->name, &fl.fl4_dst);
2583 * This target is not on a VLAN
2585 if (rt->u.dst.dev == bond->dev) {
2587 pr_debug("basa: rtdev == bond->dev: arp_send\n");
2588 bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2589 bond->master_ip, 0);
2594 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
2595 vlan_dev = vlan_group_get_device(bond->vlgrp, vlan->vlan_id);
2596 if (vlan_dev == rt->u.dst.dev) {
2597 vlan_id = vlan->vlan_id;
2598 pr_debug("basa: vlan match on %s %d\n",
2599 vlan_dev->name, vlan_id);
2606 bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2607 vlan->vlan_ip, vlan_id);
2611 if (net_ratelimit()) {
2613 ": %s: no path to arp_ip_target %pI4 via rt.dev %s\n",
2614 bond->dev->name, &fl.fl4_dst,
2615 rt->u.dst.dev ? rt->u.dst.dev->name : "NULL");
2622 * Kick out a gratuitous ARP for an IP on the bonding master plus one
2623 * for each VLAN above us.
2625 * Caller must hold curr_slave_lock for read or better
2627 static void bond_send_gratuitous_arp(struct bonding *bond)
2629 struct slave *slave = bond->curr_active_slave;
2630 struct vlan_entry *vlan;
2631 struct net_device *vlan_dev;
2633 pr_debug("bond_send_grat_arp: bond %s slave %s\n", bond->dev->name,
2634 slave ? slave->dev->name : "NULL");
2636 if (!slave || !bond->send_grat_arp ||
2637 test_bit(__LINK_STATE_LINKWATCH_PENDING, &slave->dev->state))
2640 bond->send_grat_arp--;
2642 if (bond->master_ip) {
2643 bond_arp_send(slave->dev, ARPOP_REPLY, bond->master_ip,
2644 bond->master_ip, 0);
2647 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
2648 vlan_dev = vlan_group_get_device(bond->vlgrp, vlan->vlan_id);
2649 if (vlan->vlan_ip) {
2650 bond_arp_send(slave->dev, ARPOP_REPLY, vlan->vlan_ip,
2651 vlan->vlan_ip, vlan->vlan_id);
2656 static void bond_validate_arp(struct bonding *bond, struct slave *slave, __be32 sip, __be32 tip)
2659 __be32 *targets = bond->params.arp_targets;
2661 for (i = 0; (i < BOND_MAX_ARP_TARGETS) && targets[i]; i++) {
2662 pr_debug("bva: sip %pI4 tip %pI4 t[%d] %pI4 bhti(tip) %d\n",
2663 &sip, &tip, i, &targets[i], bond_has_this_ip(bond, tip));
2664 if (sip == targets[i]) {
2665 if (bond_has_this_ip(bond, tip))
2666 slave->last_arp_rx = jiffies;
2672 static int bond_arp_rcv(struct sk_buff *skb, struct net_device *dev, struct packet_type *pt, struct net_device *orig_dev)
2675 struct slave *slave;
2676 struct bonding *bond;
2677 unsigned char *arp_ptr;
2680 if (dev_net(dev) != &init_net)
2683 if (!(dev->priv_flags & IFF_BONDING) || !(dev->flags & IFF_MASTER))
2686 bond = netdev_priv(dev);
2687 read_lock(&bond->lock);
2689 pr_debug("bond_arp_rcv: bond %s skb->dev %s orig_dev %s\n",
2690 bond->dev->name, skb->dev ? skb->dev->name : "NULL",
2691 orig_dev ? orig_dev->name : "NULL");
2693 slave = bond_get_slave_by_dev(bond, orig_dev);
2694 if (!slave || !slave_do_arp_validate(bond, slave))
2697 if (!pskb_may_pull(skb, arp_hdr_len(dev)))
2701 if (arp->ar_hln != dev->addr_len ||
2702 skb->pkt_type == PACKET_OTHERHOST ||
2703 skb->pkt_type == PACKET_LOOPBACK ||
2704 arp->ar_hrd != htons(ARPHRD_ETHER) ||
2705 arp->ar_pro != htons(ETH_P_IP) ||
2709 arp_ptr = (unsigned char *)(arp + 1);
2710 arp_ptr += dev->addr_len;
2711 memcpy(&sip, arp_ptr, 4);
2712 arp_ptr += 4 + dev->addr_len;
2713 memcpy(&tip, arp_ptr, 4);
2715 pr_debug("bond_arp_rcv: %s %s/%d av %d sv %d sip %pI4 tip %pI4\n",
2716 bond->dev->name, slave->dev->name, slave->state,
2717 bond->params.arp_validate, slave_do_arp_validate(bond, slave),
2721 * Backup slaves won't see the ARP reply, but do come through
2722 * here for each ARP probe (so we swap the sip/tip to validate
2723 * the probe). In a "redundant switch, common router" type of
2724 * configuration, the ARP probe will (hopefully) travel from
2725 * the active, through one switch, the router, then the other
2726 * switch before reaching the backup.
2728 if (slave->state == BOND_STATE_ACTIVE)
2729 bond_validate_arp(bond, slave, sip, tip);
2731 bond_validate_arp(bond, slave, tip, sip);
2734 read_unlock(&bond->lock);
2737 return NET_RX_SUCCESS;
2741 * this function is called regularly to monitor each slave's link
2742 * ensuring that traffic is being sent and received when arp monitoring
2743 * is used in load-balancing mode. if the adapter has been dormant, then an
2744 * arp is transmitted to generate traffic. see activebackup_arp_monitor for
2745 * arp monitoring in active backup mode.
2747 void bond_loadbalance_arp_mon(struct work_struct *work)
2749 struct bonding *bond = container_of(work, struct bonding,
2751 struct slave *slave, *oldcurrent;
2752 int do_failover = 0;
2756 read_lock(&bond->lock);
2758 delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
2760 if (bond->kill_timers)
2763 if (bond->slave_cnt == 0)
2766 read_lock(&bond->curr_slave_lock);
2767 oldcurrent = bond->curr_active_slave;
2768 read_unlock(&bond->curr_slave_lock);
2770 /* see if any of the previous devices are up now (i.e. they have
2771 * xmt and rcv traffic). the curr_active_slave does not come into
2772 * the picture unless it is null. also, slave->jiffies is not needed
2773 * here because we send an arp on each slave and give a slave as
2774 * long as it needs to get the tx/rx within the delta.
2775 * TODO: what about up/down delay in arp mode? it wasn't here before
2778 bond_for_each_slave(bond, slave, i) {
2779 if (slave->link != BOND_LINK_UP) {
2780 if (time_before_eq(jiffies, dev_trans_start(slave->dev) + delta_in_ticks) &&
2781 time_before_eq(jiffies, slave->dev->last_rx + delta_in_ticks)) {
2783 slave->link = BOND_LINK_UP;
2784 slave->state = BOND_STATE_ACTIVE;
2786 /* primary_slave has no meaning in round-robin
2787 * mode. the window of a slave being up and
2788 * curr_active_slave being null after enslaving
2793 ": %s: link status definitely "
2794 "up for interface %s, ",
2800 ": %s: interface %s is now up\n",
2806 /* slave->link == BOND_LINK_UP */
2808 /* not all switches will respond to an arp request
2809 * when the source ip is 0, so don't take the link down
2810 * if we don't know our ip yet
2812 if (time_after_eq(jiffies, dev_trans_start(slave->dev) + 2*delta_in_ticks) ||
2813 (time_after_eq(jiffies, slave->dev->last_rx + 2*delta_in_ticks))) {
2815 slave->link = BOND_LINK_DOWN;
2816 slave->state = BOND_STATE_BACKUP;
2818 if (slave->link_failure_count < UINT_MAX)
2819 slave->link_failure_count++;
2822 ": %s: interface %s is now down.\n",
2826 if (slave == oldcurrent)
2831 /* note: if switch is in round-robin mode, all links
2832 * must tx arp to ensure all links rx an arp - otherwise
2833 * links may oscillate or not come up at all; if switch is
2834 * in something like xor mode, there is nothing we can
2835 * do - all replies will be rx'ed on same link causing slaves
2836 * to be unstable during low/no traffic periods
2838 if (IS_UP(slave->dev))
2839 bond_arp_send_all(bond, slave);
2843 write_lock_bh(&bond->curr_slave_lock);
2845 bond_select_active_slave(bond);
2847 write_unlock_bh(&bond->curr_slave_lock);
2851 if (bond->params.arp_interval)
2852 queue_delayed_work(bond->wq, &bond->arp_work, delta_in_ticks);
2854 read_unlock(&bond->lock);
2858 * Called to inspect slaves for active-backup mode ARP monitor link state
2859 * changes. Sets new_link in slaves to specify what action should take
2860 * place for the slave. Returns 0 if no changes are found, >0 if changes
2861 * to link states must be committed.
2863 * Called with bond->lock held for read.
2865 static int bond_ab_arp_inspect(struct bonding *bond, int delta_in_ticks)
2867 struct slave *slave;
2870 bond_for_each_slave(bond, slave, i) {
2871 slave->new_link = BOND_LINK_NOCHANGE;
2873 if (slave->link != BOND_LINK_UP) {
2874 if (time_before_eq(jiffies, slave_last_rx(bond, slave) +
2876 slave->new_link = BOND_LINK_UP;
2884 * Give slaves 2*delta after being enslaved or made
2885 * active. This avoids bouncing, as the last receive
2886 * times need a full ARP monitor cycle to be updated.
2888 if (!time_after_eq(jiffies, slave->jiffies +
2889 2 * delta_in_ticks))
2893 * Backup slave is down if:
2894 * - No current_arp_slave AND
2895 * - more than 3*delta since last receive AND
2896 * - the bond has an IP address
2898 * Note: a non-null current_arp_slave indicates
2899 * the curr_active_slave went down and we are
2900 * searching for a new one; under this condition
2901 * we only take the curr_active_slave down - this
2902 * gives each slave a chance to tx/rx traffic
2903 * before being taken out
2905 if (slave->state == BOND_STATE_BACKUP &&
2906 !bond->current_arp_slave &&
2907 time_after(jiffies, slave_last_rx(bond, slave) +
2908 3 * delta_in_ticks)) {
2909 slave->new_link = BOND_LINK_DOWN;
2914 * Active slave is down if:
2915 * - more than 2*delta since transmitting OR
2916 * - (more than 2*delta since receive AND
2917 * the bond has an IP address)
2919 if ((slave->state == BOND_STATE_ACTIVE) &&
2920 (time_after_eq(jiffies, dev_trans_start(slave->dev) +
2921 2 * delta_in_ticks) ||
2922 (time_after_eq(jiffies, slave_last_rx(bond, slave)
2923 + 2 * delta_in_ticks)))) {
2924 slave->new_link = BOND_LINK_DOWN;
2929 read_lock(&bond->curr_slave_lock);
2932 * Trigger a commit if the primary option setting has changed.
2934 if (bond->primary_slave &&
2935 (bond->primary_slave != bond->curr_active_slave) &&
2936 (bond->primary_slave->link == BOND_LINK_UP))
2939 read_unlock(&bond->curr_slave_lock);
2945 * Called to commit link state changes noted by inspection step of
2946 * active-backup mode ARP monitor.
2948 * Called with RTNL and bond->lock for read.
2950 static void bond_ab_arp_commit(struct bonding *bond, int delta_in_ticks)
2952 struct slave *slave;
2955 bond_for_each_slave(bond, slave, i) {
2956 switch (slave->new_link) {
2957 case BOND_LINK_NOCHANGE:
2961 write_lock_bh(&bond->curr_slave_lock);
2963 if (!bond->curr_active_slave &&
2964 time_before_eq(jiffies, dev_trans_start(slave->dev) +
2966 slave->link = BOND_LINK_UP;
2967 bond_change_active_slave(bond, slave);
2968 bond->current_arp_slave = NULL;
2971 ": %s: %s is up and now the "
2972 "active interface\n",
2973 bond->dev->name, slave->dev->name);
2975 } else if (bond->curr_active_slave != slave) {
2976 /* this slave has just come up but we
2977 * already have a current slave; this can
2978 * also happen if bond_enslave adds a new
2979 * slave that is up while we are searching
2982 slave->link = BOND_LINK_UP;
2983 bond_set_slave_inactive_flags(slave);
2984 bond->current_arp_slave = NULL;
2987 ": %s: backup interface %s is now up\n",
2988 bond->dev->name, slave->dev->name);
2991 write_unlock_bh(&bond->curr_slave_lock);
2995 case BOND_LINK_DOWN:
2996 if (slave->link_failure_count < UINT_MAX)
2997 slave->link_failure_count++;
2999 slave->link = BOND_LINK_DOWN;
3001 if (slave == bond->curr_active_slave) {
3003 ": %s: link status down for active "
3004 "interface %s, disabling it\n",
3005 bond->dev->name, slave->dev->name);
3007 bond_set_slave_inactive_flags(slave);
3009 write_lock_bh(&bond->curr_slave_lock);
3011 bond_select_active_slave(bond);
3012 if (bond->curr_active_slave)
3013 bond->curr_active_slave->jiffies =
3016 write_unlock_bh(&bond->curr_slave_lock);
3018 bond->current_arp_slave = NULL;
3020 } else if (slave->state == BOND_STATE_BACKUP) {
3022 ": %s: backup interface %s is now down\n",
3023 bond->dev->name, slave->dev->name);
3025 bond_set_slave_inactive_flags(slave);
3031 ": %s: impossible: new_link %d on slave %s\n",
3032 bond->dev->name, slave->new_link,
3038 * No race with changes to primary via sysfs, as we hold rtnl.
3040 if (bond->primary_slave &&
3041 (bond->primary_slave != bond->curr_active_slave) &&
3042 (bond->primary_slave->link == BOND_LINK_UP)) {
3043 write_lock_bh(&bond->curr_slave_lock);
3044 bond_change_active_slave(bond, bond->primary_slave);
3045 write_unlock_bh(&bond->curr_slave_lock);
3048 bond_set_carrier(bond);
3052 * Send ARP probes for active-backup mode ARP monitor.
3054 * Called with bond->lock held for read.
3056 static void bond_ab_arp_probe(struct bonding *bond)
3058 struct slave *slave;
3061 read_lock(&bond->curr_slave_lock);
3063 if (bond->current_arp_slave && bond->curr_active_slave)
3064 pr_info(DRV_NAME "PROBE: c_arp %s && cas %s BAD\n",
3065 bond->current_arp_slave->dev->name,
3066 bond->curr_active_slave->dev->name);
3068 if (bond->curr_active_slave) {
3069 bond_arp_send_all(bond, bond->curr_active_slave);
3070 read_unlock(&bond->curr_slave_lock);
3074 read_unlock(&bond->curr_slave_lock);
3076 /* if we don't have a curr_active_slave, search for the next available
3077 * backup slave from the current_arp_slave and make it the candidate
3078 * for becoming the curr_active_slave
3081 if (!bond->current_arp_slave) {
3082 bond->current_arp_slave = bond->first_slave;
3083 if (!bond->current_arp_slave)
3087 bond_set_slave_inactive_flags(bond->current_arp_slave);
3089 /* search for next candidate */
3090 bond_for_each_slave_from(bond, slave, i, bond->current_arp_slave->next) {
3091 if (IS_UP(slave->dev)) {
3092 slave->link = BOND_LINK_BACK;
3093 bond_set_slave_active_flags(slave);
3094 bond_arp_send_all(bond, slave);
3095 slave->jiffies = jiffies;
3096 bond->current_arp_slave = slave;
3100 /* if the link state is up at this point, we
3101 * mark it down - this can happen if we have
3102 * simultaneous link failures and
3103 * reselect_active_interface doesn't make this
3104 * one the current slave so it is still marked
3105 * up when it is actually down
3107 if (slave->link == BOND_LINK_UP) {
3108 slave->link = BOND_LINK_DOWN;
3109 if (slave->link_failure_count < UINT_MAX)
3110 slave->link_failure_count++;
3112 bond_set_slave_inactive_flags(slave);
3115 ": %s: backup interface %s is now down.\n",
3116 bond->dev->name, slave->dev->name);
3121 void bond_activebackup_arp_mon(struct work_struct *work)
3123 struct bonding *bond = container_of(work, struct bonding,
3127 read_lock(&bond->lock);
3129 if (bond->kill_timers)
3132 delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
3134 if (bond->slave_cnt == 0)
3137 if (bond->send_grat_arp) {
3138 read_lock(&bond->curr_slave_lock);
3139 bond_send_gratuitous_arp(bond);
3140 read_unlock(&bond->curr_slave_lock);
3143 if (bond->send_unsol_na) {
3144 read_lock(&bond->curr_slave_lock);
3145 bond_send_unsolicited_na(bond);
3146 read_unlock(&bond->curr_slave_lock);
3149 if (bond_ab_arp_inspect(bond, delta_in_ticks)) {
3150 read_unlock(&bond->lock);
3152 read_lock(&bond->lock);
3154 bond_ab_arp_commit(bond, delta_in_ticks);
3156 read_unlock(&bond->lock);
3158 read_lock(&bond->lock);
3161 bond_ab_arp_probe(bond);
3164 if (bond->params.arp_interval)
3165 queue_delayed_work(bond->wq, &bond->arp_work, delta_in_ticks);
3167 read_unlock(&bond->lock);
3170 /*------------------------------ proc/seq_file-------------------------------*/
3172 #ifdef CONFIG_PROC_FS
3174 static void *bond_info_seq_start(struct seq_file *seq, loff_t *pos)
3175 __acquires(&dev_base_lock)
3176 __acquires(&bond->lock)
3178 struct bonding *bond = seq->private;
3180 struct slave *slave;
3183 /* make sure the bond won't be taken away */
3184 read_lock(&dev_base_lock);
3185 read_lock(&bond->lock);
3188 return SEQ_START_TOKEN;
3190 bond_for_each_slave(bond, slave, i) {
3198 static void *bond_info_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3200 struct bonding *bond = seq->private;
3201 struct slave *slave = v;
3204 if (v == SEQ_START_TOKEN)
3205 return bond->first_slave;
3207 slave = slave->next;
3209 return (slave == bond->first_slave) ? NULL : slave;
3212 static void bond_info_seq_stop(struct seq_file *seq, void *v)
3213 __releases(&bond->lock)
3214 __releases(&dev_base_lock)
3216 struct bonding *bond = seq->private;
3218 read_unlock(&bond->lock);
3219 read_unlock(&dev_base_lock);
3222 static void bond_info_show_master(struct seq_file *seq)
3224 struct bonding *bond = seq->private;
3228 read_lock(&bond->curr_slave_lock);
3229 curr = bond->curr_active_slave;
3230 read_unlock(&bond->curr_slave_lock);
3232 seq_printf(seq, "Bonding Mode: %s",
3233 bond_mode_name(bond->params.mode));
3235 if (bond->params.mode == BOND_MODE_ACTIVEBACKUP &&
3236 bond->params.fail_over_mac)
3237 seq_printf(seq, " (fail_over_mac %s)",
3238 fail_over_mac_tbl[bond->params.fail_over_mac].modename);
3240 seq_printf(seq, "\n");
3242 if (bond->params.mode == BOND_MODE_XOR ||
3243 bond->params.mode == BOND_MODE_8023AD) {
3244 seq_printf(seq, "Transmit Hash Policy: %s (%d)\n",
3245 xmit_hashtype_tbl[bond->params.xmit_policy].modename,
3246 bond->params.xmit_policy);
3249 if (USES_PRIMARY(bond->params.mode)) {
3250 seq_printf(seq, "Primary Slave: %s\n",
3251 (bond->primary_slave) ?
3252 bond->primary_slave->dev->name : "None");
3254 seq_printf(seq, "Currently Active Slave: %s\n",
3255 (curr) ? curr->dev->name : "None");
3258 seq_printf(seq, "MII Status: %s\n", netif_carrier_ok(bond->dev) ?
3260 seq_printf(seq, "MII Polling Interval (ms): %d\n", bond->params.miimon);
3261 seq_printf(seq, "Up Delay (ms): %d\n",
3262 bond->params.updelay * bond->params.miimon);
3263 seq_printf(seq, "Down Delay (ms): %d\n",
3264 bond->params.downdelay * bond->params.miimon);
3267 /* ARP information */
3268 if (bond->params.arp_interval > 0) {
3270 seq_printf(seq, "ARP Polling Interval (ms): %d\n",
3271 bond->params.arp_interval);
3273 seq_printf(seq, "ARP IP target/s (n.n.n.n form):");
3275 for (i = 0; (i < BOND_MAX_ARP_TARGETS); i++) {
3276 if (!bond->params.arp_targets[i])
3279 seq_printf(seq, ",");
3280 seq_printf(seq, " %pI4", &bond->params.arp_targets[i]);
3283 seq_printf(seq, "\n");
3286 if (bond->params.mode == BOND_MODE_8023AD) {
3287 struct ad_info ad_info;
3289 seq_puts(seq, "\n802.3ad info\n");
3290 seq_printf(seq, "LACP rate: %s\n",
3291 (bond->params.lacp_fast) ? "fast" : "slow");
3292 seq_printf(seq, "Aggregator selection policy (ad_select): %s\n",
3293 ad_select_tbl[bond->params.ad_select].modename);
3295 if (bond_3ad_get_active_agg_info(bond, &ad_info)) {
3296 seq_printf(seq, "bond %s has no active aggregator\n",
3299 seq_printf(seq, "Active Aggregator Info:\n");
3301 seq_printf(seq, "\tAggregator ID: %d\n",
3302 ad_info.aggregator_id);
3303 seq_printf(seq, "\tNumber of ports: %d\n",
3305 seq_printf(seq, "\tActor Key: %d\n",
3307 seq_printf(seq, "\tPartner Key: %d\n",
3308 ad_info.partner_key);
3309 seq_printf(seq, "\tPartner Mac Address: %pM\n",
3310 ad_info.partner_system);
3315 static void bond_info_show_slave(struct seq_file *seq,
3316 const struct slave *slave)
3318 struct bonding *bond = seq->private;
3320 seq_printf(seq, "\nSlave Interface: %s\n", slave->dev->name);
3321 seq_printf(seq, "MII Status: %s\n",
3322 (slave->link == BOND_LINK_UP) ? "up" : "down");
3323 seq_printf(seq, "Link Failure Count: %u\n",
3324 slave->link_failure_count);
3326 seq_printf(seq, "Permanent HW addr: %pM\n", slave->perm_hwaddr);
3328 if (bond->params.mode == BOND_MODE_8023AD) {
3329 const struct aggregator *agg
3330 = SLAVE_AD_INFO(slave).port.aggregator;
3333 seq_printf(seq, "Aggregator ID: %d\n",
3334 agg->aggregator_identifier);
3336 seq_puts(seq, "Aggregator ID: N/A\n");
3340 static int bond_info_seq_show(struct seq_file *seq, void *v)
3342 if (v == SEQ_START_TOKEN) {
3343 seq_printf(seq, "%s\n", version);
3344 bond_info_show_master(seq);
3346 bond_info_show_slave(seq, v);
3351 static const struct seq_operations bond_info_seq_ops = {
3352 .start = bond_info_seq_start,
3353 .next = bond_info_seq_next,
3354 .stop = bond_info_seq_stop,
3355 .show = bond_info_seq_show,
3358 static int bond_info_open(struct inode *inode, struct file *file)
3360 struct seq_file *seq;
3361 struct proc_dir_entry *proc;
3364 res = seq_open(file, &bond_info_seq_ops);
3366 /* recover the pointer buried in proc_dir_entry data */
3367 seq = file->private_data;
3369 seq->private = proc->data;
3375 static const struct file_operations bond_info_fops = {
3376 .owner = THIS_MODULE,
3377 .open = bond_info_open,
3379 .llseek = seq_lseek,
3380 .release = seq_release,
3383 static int bond_create_proc_entry(struct bonding *bond)
3385 struct net_device *bond_dev = bond->dev;
3387 if (bond_proc_dir) {
3388 bond->proc_entry = proc_create_data(bond_dev->name,
3389 S_IRUGO, bond_proc_dir,
3390 &bond_info_fops, bond);
3391 if (bond->proc_entry == NULL)
3393 ": Warning: Cannot create /proc/net/%s/%s\n",
3394 DRV_NAME, bond_dev->name);
3396 memcpy(bond->proc_file_name, bond_dev->name, IFNAMSIZ);
3402 static void bond_remove_proc_entry(struct bonding *bond)
3404 if (bond_proc_dir && bond->proc_entry) {
3405 remove_proc_entry(bond->proc_file_name, bond_proc_dir);
3406 memset(bond->proc_file_name, 0, IFNAMSIZ);
3407 bond->proc_entry = NULL;
3411 /* Create the bonding directory under /proc/net, if doesn't exist yet.
3412 * Caller must hold rtnl_lock.
3414 static void bond_create_proc_dir(void)
3416 if (!bond_proc_dir) {
3417 bond_proc_dir = proc_mkdir(DRV_NAME, init_net.proc_net);
3420 ": Warning: cannot create /proc/net/%s\n",
3425 /* Destroy the bonding directory under /proc/net, if empty.
3426 * Caller must hold rtnl_lock.
3428 static void bond_destroy_proc_dir(void)
3430 if (bond_proc_dir) {
3431 remove_proc_entry(DRV_NAME, init_net.proc_net);
3432 bond_proc_dir = NULL;
3436 #else /* !CONFIG_PROC_FS */
3438 static int bond_create_proc_entry(struct bonding *bond)
3442 static void bond_remove_proc_entry(struct bonding *bond)
3446 static void bond_create_proc_dir(void)
3450 static void bond_destroy_proc_dir(void)
3454 #endif /* CONFIG_PROC_FS */
3457 /*-------------------------- netdev event handling --------------------------*/
3460 * Change device name
3462 static int bond_event_changename(struct bonding *bond)
3464 bond_remove_proc_entry(bond);
3465 bond_create_proc_entry(bond);
3467 bond_destroy_sysfs_entry(bond);
3468 bond_create_sysfs_entry(bond);
3473 static int bond_master_netdev_event(unsigned long event,
3474 struct net_device *bond_dev)
3476 struct bonding *event_bond = netdev_priv(bond_dev);
3479 case NETDEV_CHANGENAME:
3480 return bond_event_changename(event_bond);
3481 case NETDEV_UNREGISTER:
3482 bond_release_all(event_bond->dev);
3491 static int bond_slave_netdev_event(unsigned long event,
3492 struct net_device *slave_dev)
3494 struct net_device *bond_dev = slave_dev->master;
3495 struct bonding *bond = netdev_priv(bond_dev);
3498 case NETDEV_UNREGISTER:
3500 if (bond->setup_by_slave)
3501 bond_release_and_destroy(bond_dev, slave_dev);
3503 bond_release(bond_dev, slave_dev);
3507 if (bond->params.mode == BOND_MODE_8023AD || bond_is_lb(bond)) {
3508 struct slave *slave;
3510 slave = bond_get_slave_by_dev(bond, slave_dev);
3512 u16 old_speed = slave->speed;
3513 u16 old_duplex = slave->duplex;
3515 bond_update_speed_duplex(slave);
3517 if (bond_is_lb(bond))
3520 if (old_speed != slave->speed)
3521 bond_3ad_adapter_speed_changed(slave);
3522 if (old_duplex != slave->duplex)
3523 bond_3ad_adapter_duplex_changed(slave);
3530 * ... Or is it this?
3533 case NETDEV_CHANGEMTU:
3535 * TODO: Should slaves be allowed to
3536 * independently alter their MTU? For
3537 * an active-backup bond, slaves need
3538 * not be the same type of device, so
3539 * MTUs may vary. For other modes,
3540 * slaves arguably should have the
3541 * same MTUs. To do this, we'd need to
3542 * take over the slave's change_mtu
3543 * function for the duration of their
3547 case NETDEV_CHANGENAME:
3549 * TODO: handle changing the primary's name
3552 case NETDEV_FEAT_CHANGE:
3553 bond_compute_features(bond);
3563 * bond_netdev_event: handle netdev notifier chain events.
3565 * This function receives events for the netdev chain. The caller (an
3566 * ioctl handler calling blocking_notifier_call_chain) holds the necessary
3567 * locks for us to safely manipulate the slave devices (RTNL lock,
3570 static int bond_netdev_event(struct notifier_block *this,
3571 unsigned long event, void *ptr)
3573 struct net_device *event_dev = (struct net_device *)ptr;
3575 if (dev_net(event_dev) != &init_net)
3578 pr_debug("event_dev: %s, event: %lx\n",
3579 (event_dev ? event_dev->name : "None"),
3582 if (!(event_dev->priv_flags & IFF_BONDING))
3585 if (event_dev->flags & IFF_MASTER) {
3586 pr_debug("IFF_MASTER\n");
3587 return bond_master_netdev_event(event, event_dev);
3590 if (event_dev->flags & IFF_SLAVE) {
3591 pr_debug("IFF_SLAVE\n");
3592 return bond_slave_netdev_event(event, event_dev);
3599 * bond_inetaddr_event: handle inetaddr notifier chain events.
3601 * We keep track of device IPs primarily to use as source addresses in
3602 * ARP monitor probes (rather than spewing out broadcasts all the time).
3604 * We track one IP for the main device (if it has one), plus one per VLAN.
3606 static int bond_inetaddr_event(struct notifier_block *this, unsigned long event, void *ptr)
3608 struct in_ifaddr *ifa = ptr;
3609 struct net_device *vlan_dev, *event_dev = ifa->ifa_dev->dev;
3610 struct bonding *bond;
3611 struct vlan_entry *vlan;
3613 if (dev_net(ifa->ifa_dev->dev) != &init_net)
3616 list_for_each_entry(bond, &bond_dev_list, bond_list) {
3617 if (bond->dev == event_dev) {
3620 bond->master_ip = ifa->ifa_local;
3623 bond->master_ip = bond_glean_dev_ip(bond->dev);
3630 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
3631 vlan_dev = vlan_group_get_device(bond->vlgrp, vlan->vlan_id);
3632 if (vlan_dev == event_dev) {
3635 vlan->vlan_ip = ifa->ifa_local;
3639 bond_glean_dev_ip(vlan_dev);
3650 static struct notifier_block bond_netdev_notifier = {
3651 .notifier_call = bond_netdev_event,
3654 static struct notifier_block bond_inetaddr_notifier = {
3655 .notifier_call = bond_inetaddr_event,
3658 /*-------------------------- Packet type handling ---------------------------*/
3660 /* register to receive lacpdus on a bond */
3661 static void bond_register_lacpdu(struct bonding *bond)
3663 struct packet_type *pk_type = &(BOND_AD_INFO(bond).ad_pkt_type);
3665 /* initialize packet type */
3666 pk_type->type = PKT_TYPE_LACPDU;
3667 pk_type->dev = bond->dev;
3668 pk_type->func = bond_3ad_lacpdu_recv;
3670 dev_add_pack(pk_type);
3673 /* unregister to receive lacpdus on a bond */
3674 static void bond_unregister_lacpdu(struct bonding *bond)
3676 dev_remove_pack(&(BOND_AD_INFO(bond).ad_pkt_type));
3679 void bond_register_arp(struct bonding *bond)
3681 struct packet_type *pt = &bond->arp_mon_pt;
3686 pt->type = htons(ETH_P_ARP);
3687 pt->dev = bond->dev;
3688 pt->func = bond_arp_rcv;
3692 void bond_unregister_arp(struct bonding *bond)
3694 struct packet_type *pt = &bond->arp_mon_pt;
3696 dev_remove_pack(pt);
3700 /*---------------------------- Hashing Policies -----------------------------*/
3703 * Hash for the output device based upon layer 2 and layer 3 data. If
3704 * the packet is not IP mimic bond_xmit_hash_policy_l2()
3706 static int bond_xmit_hash_policy_l23(struct sk_buff *skb,
3707 struct net_device *bond_dev, int count)
3709 struct ethhdr *data = (struct ethhdr *)skb->data;
3710 struct iphdr *iph = ip_hdr(skb);
3712 if (skb->protocol == htons(ETH_P_IP)) {
3713 return ((ntohl(iph->saddr ^ iph->daddr) & 0xffff) ^
3714 (data->h_dest[5] ^ bond_dev->dev_addr[5])) % count;
3717 return (data->h_dest[5] ^ bond_dev->dev_addr[5]) % count;
3721 * Hash for the output device based upon layer 3 and layer 4 data. If
3722 * the packet is a frag or not TCP or UDP, just use layer 3 data. If it is
3723 * altogether not IP, mimic bond_xmit_hash_policy_l2()
3725 static int bond_xmit_hash_policy_l34(struct sk_buff *skb,
3726 struct net_device *bond_dev, int count)
3728 struct ethhdr *data = (struct ethhdr *)skb->data;
3729 struct iphdr *iph = ip_hdr(skb);
3730 __be16 *layer4hdr = (__be16 *)((u32 *)iph + iph->ihl);
3733 if (skb->protocol == htons(ETH_P_IP)) {
3734 if (!(iph->frag_off & htons(IP_MF|IP_OFFSET)) &&
3735 (iph->protocol == IPPROTO_TCP ||
3736 iph->protocol == IPPROTO_UDP)) {
3737 layer4_xor = ntohs((*layer4hdr ^ *(layer4hdr + 1)));
3739 return (layer4_xor ^
3740 ((ntohl(iph->saddr ^ iph->daddr)) & 0xffff)) % count;
3744 return (data->h_dest[5] ^ bond_dev->dev_addr[5]) % count;
3748 * Hash for the output device based upon layer 2 data
3750 static int bond_xmit_hash_policy_l2(struct sk_buff *skb,
3751 struct net_device *bond_dev, int count)
3753 struct ethhdr *data = (struct ethhdr *)skb->data;
3755 return (data->h_dest[5] ^ bond_dev->dev_addr[5]) % count;
3758 /*-------------------------- Device entry points ----------------------------*/
3760 static int bond_open(struct net_device *bond_dev)
3762 struct bonding *bond = netdev_priv(bond_dev);
3764 bond->kill_timers = 0;
3766 if (bond_is_lb(bond)) {
3767 /* bond_alb_initialize must be called before the timer
3770 if (bond_alb_initialize(bond, (bond->params.mode == BOND_MODE_ALB))) {
3771 /* something went wrong - fail the open operation */
3775 INIT_DELAYED_WORK(&bond->alb_work, bond_alb_monitor);
3776 queue_delayed_work(bond->wq, &bond->alb_work, 0);
3779 if (bond->params.miimon) { /* link check interval, in milliseconds. */
3780 INIT_DELAYED_WORK(&bond->mii_work, bond_mii_monitor);
3781 queue_delayed_work(bond->wq, &bond->mii_work, 0);
3784 if (bond->params.arp_interval) { /* arp interval, in milliseconds. */
3785 if (bond->params.mode == BOND_MODE_ACTIVEBACKUP)
3786 INIT_DELAYED_WORK(&bond->arp_work,
3787 bond_activebackup_arp_mon);
3789 INIT_DELAYED_WORK(&bond->arp_work,
3790 bond_loadbalance_arp_mon);
3792 queue_delayed_work(bond->wq, &bond->arp_work, 0);
3793 if (bond->params.arp_validate)
3794 bond_register_arp(bond);
3797 if (bond->params.mode == BOND_MODE_8023AD) {
3798 INIT_DELAYED_WORK(&bond->ad_work, bond_3ad_state_machine_handler);
3799 queue_delayed_work(bond->wq, &bond->ad_work, 0);
3800 /* register to receive LACPDUs */
3801 bond_register_lacpdu(bond);
3802 bond_3ad_initiate_agg_selection(bond, 1);
3808 static int bond_close(struct net_device *bond_dev)
3810 struct bonding *bond = netdev_priv(bond_dev);
3812 if (bond->params.mode == BOND_MODE_8023AD) {
3813 /* Unregister the receive of LACPDUs */
3814 bond_unregister_lacpdu(bond);
3817 if (bond->params.arp_validate)
3818 bond_unregister_arp(bond);
3820 write_lock_bh(&bond->lock);
3822 bond->send_grat_arp = 0;
3823 bond->send_unsol_na = 0;
3825 /* signal timers not to re-arm */
3826 bond->kill_timers = 1;
3828 write_unlock_bh(&bond->lock);
3830 if (bond->params.miimon) { /* link check interval, in milliseconds. */
3831 cancel_delayed_work(&bond->mii_work);
3834 if (bond->params.arp_interval) { /* arp interval, in milliseconds. */
3835 cancel_delayed_work(&bond->arp_work);
3838 switch (bond->params.mode) {
3839 case BOND_MODE_8023AD:
3840 cancel_delayed_work(&bond->ad_work);
3844 cancel_delayed_work(&bond->alb_work);
3851 if (bond_is_lb(bond)) {
3852 /* Must be called only after all
3853 * slaves have been released
3855 bond_alb_deinitialize(bond);
3861 static struct net_device_stats *bond_get_stats(struct net_device *bond_dev)
3863 struct bonding *bond = netdev_priv(bond_dev);
3864 struct net_device_stats *stats = &bond->stats;
3865 struct net_device_stats local_stats;
3866 struct slave *slave;
3869 memset(&local_stats, 0, sizeof(struct net_device_stats));
3871 read_lock_bh(&bond->lock);
3873 bond_for_each_slave(bond, slave, i) {
3874 const struct net_device_stats *sstats = dev_get_stats(slave->dev);
3876 local_stats.rx_packets += sstats->rx_packets;
3877 local_stats.rx_bytes += sstats->rx_bytes;
3878 local_stats.rx_errors += sstats->rx_errors;
3879 local_stats.rx_dropped += sstats->rx_dropped;
3881 local_stats.tx_packets += sstats->tx_packets;
3882 local_stats.tx_bytes += sstats->tx_bytes;
3883 local_stats.tx_errors += sstats->tx_errors;
3884 local_stats.tx_dropped += sstats->tx_dropped;
3886 local_stats.multicast += sstats->multicast;
3887 local_stats.collisions += sstats->collisions;
3889 local_stats.rx_length_errors += sstats->rx_length_errors;
3890 local_stats.rx_over_errors += sstats->rx_over_errors;
3891 local_stats.rx_crc_errors += sstats->rx_crc_errors;
3892 local_stats.rx_frame_errors += sstats->rx_frame_errors;
3893 local_stats.rx_fifo_errors += sstats->rx_fifo_errors;
3894 local_stats.rx_missed_errors += sstats->rx_missed_errors;
3896 local_stats.tx_aborted_errors += sstats->tx_aborted_errors;
3897 local_stats.tx_carrier_errors += sstats->tx_carrier_errors;
3898 local_stats.tx_fifo_errors += sstats->tx_fifo_errors;
3899 local_stats.tx_heartbeat_errors += sstats->tx_heartbeat_errors;
3900 local_stats.tx_window_errors += sstats->tx_window_errors;
3903 memcpy(stats, &local_stats, sizeof(struct net_device_stats));
3905 read_unlock_bh(&bond->lock);
3910 static int bond_do_ioctl(struct net_device *bond_dev, struct ifreq *ifr, int cmd)
3912 struct net_device *slave_dev = NULL;
3913 struct ifbond k_binfo;
3914 struct ifbond __user *u_binfo = NULL;
3915 struct ifslave k_sinfo;
3916 struct ifslave __user *u_sinfo = NULL;
3917 struct mii_ioctl_data *mii = NULL;
3920 pr_debug("bond_ioctl: master=%s, cmd=%d\n",
3921 bond_dev->name, cmd);
3933 * We do this again just in case we were called by SIOCGMIIREG
3934 * instead of SIOCGMIIPHY.
3941 if (mii->reg_num == 1) {
3942 struct bonding *bond = netdev_priv(bond_dev);
3944 read_lock(&bond->lock);
3945 read_lock(&bond->curr_slave_lock);
3946 if (netif_carrier_ok(bond->dev))
3947 mii->val_out = BMSR_LSTATUS;
3949 read_unlock(&bond->curr_slave_lock);
3950 read_unlock(&bond->lock);
3954 case BOND_INFO_QUERY_OLD:
3955 case SIOCBONDINFOQUERY:
3956 u_binfo = (struct ifbond __user *)ifr->ifr_data;
3958 if (copy_from_user(&k_binfo, u_binfo, sizeof(ifbond)))
3961 res = bond_info_query(bond_dev, &k_binfo);
3963 copy_to_user(u_binfo, &k_binfo, sizeof(ifbond)))
3967 case BOND_SLAVE_INFO_QUERY_OLD:
3968 case SIOCBONDSLAVEINFOQUERY:
3969 u_sinfo = (struct ifslave __user *)ifr->ifr_data;
3971 if (copy_from_user(&k_sinfo, u_sinfo, sizeof(ifslave)))
3974 res = bond_slave_info_query(bond_dev, &k_sinfo);
3976 copy_to_user(u_sinfo, &k_sinfo, sizeof(ifslave)))
3985 if (!capable(CAP_NET_ADMIN))
3988 slave_dev = dev_get_by_name(&init_net, ifr->ifr_slave);
3990 pr_debug("slave_dev=%p: \n", slave_dev);
3995 pr_debug("slave_dev->name=%s: \n", slave_dev->name);
3997 case BOND_ENSLAVE_OLD:
3998 case SIOCBONDENSLAVE:
3999 res = bond_enslave(bond_dev, slave_dev);
4001 case BOND_RELEASE_OLD:
4002 case SIOCBONDRELEASE:
4003 res = bond_release(bond_dev, slave_dev);
4005 case BOND_SETHWADDR_OLD:
4006 case SIOCBONDSETHWADDR:
4007 res = bond_sethwaddr(bond_dev, slave_dev);
4009 case BOND_CHANGE_ACTIVE_OLD:
4010 case SIOCBONDCHANGEACTIVE:
4011 res = bond_ioctl_change_active(bond_dev, slave_dev);
4023 static void bond_set_multicast_list(struct net_device *bond_dev)
4025 struct bonding *bond = netdev_priv(bond_dev);
4026 struct dev_mc_list *dmi;
4029 * Do promisc before checking multicast_mode
4031 if ((bond_dev->flags & IFF_PROMISC) && !(bond->flags & IFF_PROMISC))
4033 * FIXME: Need to handle the error when one of the multi-slaves
4036 bond_set_promiscuity(bond, 1);
4039 if (!(bond_dev->flags & IFF_PROMISC) && (bond->flags & IFF_PROMISC))
4040 bond_set_promiscuity(bond, -1);
4043 /* set allmulti flag to slaves */
4044 if ((bond_dev->flags & IFF_ALLMULTI) && !(bond->flags & IFF_ALLMULTI))
4046 * FIXME: Need to handle the error when one of the multi-slaves
4049 bond_set_allmulti(bond, 1);
4052 if (!(bond_dev->flags & IFF_ALLMULTI) && (bond->flags & IFF_ALLMULTI))
4053 bond_set_allmulti(bond, -1);
4056 read_lock(&bond->lock);
4058 bond->flags = bond_dev->flags;
4060 /* looking for addresses to add to slaves' mc list */
4061 for (dmi = bond_dev->mc_list; dmi; dmi = dmi->next) {
4062 if (!bond_mc_list_find_dmi(dmi, bond->mc_list))
4063 bond_mc_add(bond, dmi->dmi_addr, dmi->dmi_addrlen);
4066 /* looking for addresses to delete from slaves' list */
4067 for (dmi = bond->mc_list; dmi; dmi = dmi->next) {
4068 if (!bond_mc_list_find_dmi(dmi, bond_dev->mc_list))
4069 bond_mc_delete(bond, dmi->dmi_addr, dmi->dmi_addrlen);
4072 /* save master's multicast list */
4073 bond_mc_list_destroy(bond);
4074 bond_mc_list_copy(bond_dev->mc_list, bond, GFP_ATOMIC);
4076 read_unlock(&bond->lock);
4079 static int bond_neigh_setup(struct net_device *dev, struct neigh_parms *parms)
4081 struct bonding *bond = netdev_priv(dev);
4082 struct slave *slave = bond->first_slave;
4085 const struct net_device_ops *slave_ops
4086 = slave->dev->netdev_ops;
4087 if (slave_ops->ndo_neigh_setup)
4088 return slave_ops->ndo_neigh_setup(slave->dev, parms);
4094 * Change the MTU of all of a master's slaves to match the master
4096 static int bond_change_mtu(struct net_device *bond_dev, int new_mtu)
4098 struct bonding *bond = netdev_priv(bond_dev);
4099 struct slave *slave, *stop_at;
4103 pr_debug("bond=%p, name=%s, new_mtu=%d\n", bond,
4104 (bond_dev ? bond_dev->name : "None"), new_mtu);
4106 /* Can't hold bond->lock with bh disabled here since
4107 * some base drivers panic. On the other hand we can't
4108 * hold bond->lock without bh disabled because we'll
4109 * deadlock. The only solution is to rely on the fact
4110 * that we're under rtnl_lock here, and the slaves
4111 * list won't change. This doesn't solve the problem
4112 * of setting the slave's MTU while it is
4113 * transmitting, but the assumption is that the base
4114 * driver can handle that.
4116 * TODO: figure out a way to safely iterate the slaves
4117 * list, but without holding a lock around the actual
4118 * call to the base driver.
4121 bond_for_each_slave(bond, slave, i) {
4122 pr_debug("s %p s->p %p c_m %p\n", slave,
4123 slave->prev, slave->dev->netdev_ops->ndo_change_mtu);
4125 res = dev_set_mtu(slave->dev, new_mtu);
4128 /* If we failed to set the slave's mtu to the new value
4129 * we must abort the operation even in ACTIVE_BACKUP
4130 * mode, because if we allow the backup slaves to have
4131 * different mtu values than the active slave we'll
4132 * need to change their mtu when doing a failover. That
4133 * means changing their mtu from timer context, which
4134 * is probably not a good idea.
4136 pr_debug("err %d %s\n", res, slave->dev->name);
4141 bond_dev->mtu = new_mtu;
4146 /* unwind from head to the slave that failed */
4148 bond_for_each_slave_from_to(bond, slave, i, bond->first_slave, stop_at) {
4151 tmp_res = dev_set_mtu(slave->dev, bond_dev->mtu);
4153 pr_debug("unwind err %d dev %s\n", tmp_res,
4164 * Note that many devices must be down to change the HW address, and
4165 * downing the master releases all slaves. We can make bonds full of
4166 * bonding devices to test this, however.
4168 static int bond_set_mac_address(struct net_device *bond_dev, void *addr)
4170 struct bonding *bond = netdev_priv(bond_dev);
4171 struct sockaddr *sa = addr, tmp_sa;
4172 struct slave *slave, *stop_at;
4176 if (bond->params.mode == BOND_MODE_ALB)
4177 return bond_alb_set_mac_address(bond_dev, addr);
4180 pr_debug("bond=%p, name=%s\n", bond, (bond_dev ? bond_dev->name : "None"));
4183 * If fail_over_mac is set to active, do nothing and return
4184 * success. Returning an error causes ifenslave to fail.
4186 if (bond->params.fail_over_mac == BOND_FOM_ACTIVE)
4189 if (!is_valid_ether_addr(sa->sa_data))
4190 return -EADDRNOTAVAIL;
4192 /* Can't hold bond->lock with bh disabled here since
4193 * some base drivers panic. On the other hand we can't
4194 * hold bond->lock without bh disabled because we'll
4195 * deadlock. The only solution is to rely on the fact
4196 * that we're under rtnl_lock here, and the slaves
4197 * list won't change. This doesn't solve the problem
4198 * of setting the slave's hw address while it is
4199 * transmitting, but the assumption is that the base
4200 * driver can handle that.
4202 * TODO: figure out a way to safely iterate the slaves
4203 * list, but without holding a lock around the actual
4204 * call to the base driver.
4207 bond_for_each_slave(bond, slave, i) {
4208 const struct net_device_ops *slave_ops = slave->dev->netdev_ops;
4209 pr_debug("slave %p %s\n", slave, slave->dev->name);
4211 if (slave_ops->ndo_set_mac_address == NULL) {
4213 pr_debug("EOPNOTSUPP %s\n", slave->dev->name);
4217 res = dev_set_mac_address(slave->dev, addr);
4219 /* TODO: consider downing the slave
4221 * User should expect communications
4222 * breakage anyway until ARP finish
4225 pr_debug("err %d %s\n", res, slave->dev->name);
4231 memcpy(bond_dev->dev_addr, sa->sa_data, bond_dev->addr_len);
4235 memcpy(tmp_sa.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
4236 tmp_sa.sa_family = bond_dev->type;
4238 /* unwind from head to the slave that failed */
4240 bond_for_each_slave_from_to(bond, slave, i, bond->first_slave, stop_at) {
4243 tmp_res = dev_set_mac_address(slave->dev, &tmp_sa);
4245 pr_debug("unwind err %d dev %s\n", tmp_res,
4253 static int bond_xmit_roundrobin(struct sk_buff *skb, struct net_device *bond_dev)
4255 struct bonding *bond = netdev_priv(bond_dev);
4256 struct slave *slave, *start_at;
4257 int i, slave_no, res = 1;
4259 read_lock(&bond->lock);
4261 if (!BOND_IS_OK(bond))
4265 * Concurrent TX may collide on rr_tx_counter; we accept that
4266 * as being rare enough not to justify using an atomic op here
4268 slave_no = bond->rr_tx_counter++ % bond->slave_cnt;
4270 bond_for_each_slave(bond, slave, i) {
4277 bond_for_each_slave_from(bond, slave, i, start_at) {
4278 if (IS_UP(slave->dev) &&
4279 (slave->link == BOND_LINK_UP) &&
4280 (slave->state == BOND_STATE_ACTIVE)) {
4281 res = bond_dev_queue_xmit(bond, skb, slave->dev);
4288 /* no suitable interface, frame not sent */
4291 read_unlock(&bond->lock);
4292 return NETDEV_TX_OK;
4297 * in active-backup mode, we know that bond->curr_active_slave is always valid if
4298 * the bond has a usable interface.
4300 static int bond_xmit_activebackup(struct sk_buff *skb, struct net_device *bond_dev)
4302 struct bonding *bond = netdev_priv(bond_dev);
4305 read_lock(&bond->lock);
4306 read_lock(&bond->curr_slave_lock);
4308 if (!BOND_IS_OK(bond))
4311 if (!bond->curr_active_slave)
4314 res = bond_dev_queue_xmit(bond, skb, bond->curr_active_slave->dev);
4318 /* no suitable interface, frame not sent */
4321 read_unlock(&bond->curr_slave_lock);
4322 read_unlock(&bond->lock);
4323 return NETDEV_TX_OK;
4327 * In bond_xmit_xor() , we determine the output device by using a pre-
4328 * determined xmit_hash_policy(), If the selected device is not enabled,
4329 * find the next active slave.
4331 static int bond_xmit_xor(struct sk_buff *skb, struct net_device *bond_dev)
4333 struct bonding *bond = netdev_priv(bond_dev);
4334 struct slave *slave, *start_at;
4339 read_lock(&bond->lock);
4341 if (!BOND_IS_OK(bond))
4344 slave_no = bond->xmit_hash_policy(skb, bond_dev, bond->slave_cnt);
4346 bond_for_each_slave(bond, slave, i) {
4354 bond_for_each_slave_from(bond, slave, i, start_at) {
4355 if (IS_UP(slave->dev) &&
4356 (slave->link == BOND_LINK_UP) &&
4357 (slave->state == BOND_STATE_ACTIVE)) {
4358 res = bond_dev_queue_xmit(bond, skb, slave->dev);
4365 /* no suitable interface, frame not sent */
4368 read_unlock(&bond->lock);
4369 return NETDEV_TX_OK;
4373 * in broadcast mode, we send everything to all usable interfaces.
4375 static int bond_xmit_broadcast(struct sk_buff *skb, struct net_device *bond_dev)
4377 struct bonding *bond = netdev_priv(bond_dev);
4378 struct slave *slave, *start_at;
4379 struct net_device *tx_dev = NULL;
4383 read_lock(&bond->lock);
4385 if (!BOND_IS_OK(bond))
4388 read_lock(&bond->curr_slave_lock);
4389 start_at = bond->curr_active_slave;
4390 read_unlock(&bond->curr_slave_lock);
4395 bond_for_each_slave_from(bond, slave, i, start_at) {
4396 if (IS_UP(slave->dev) &&
4397 (slave->link == BOND_LINK_UP) &&
4398 (slave->state == BOND_STATE_ACTIVE)) {
4400 struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
4403 ": %s: Error: bond_xmit_broadcast(): "
4404 "skb_clone() failed\n",
4409 res = bond_dev_queue_xmit(bond, skb2, tx_dev);
4411 dev_kfree_skb(skb2);
4415 tx_dev = slave->dev;
4420 res = bond_dev_queue_xmit(bond, skb, tx_dev);
4424 /* no suitable interface, frame not sent */
4427 /* frame sent to all suitable interfaces */
4428 read_unlock(&bond->lock);
4429 return NETDEV_TX_OK;
4432 /*------------------------- Device initialization ---------------------------*/
4434 static void bond_set_xmit_hash_policy(struct bonding *bond)
4436 switch (bond->params.xmit_policy) {
4437 case BOND_XMIT_POLICY_LAYER23:
4438 bond->xmit_hash_policy = bond_xmit_hash_policy_l23;
4440 case BOND_XMIT_POLICY_LAYER34:
4441 bond->xmit_hash_policy = bond_xmit_hash_policy_l34;
4443 case BOND_XMIT_POLICY_LAYER2:
4445 bond->xmit_hash_policy = bond_xmit_hash_policy_l2;
4450 static int bond_start_xmit(struct sk_buff *skb, struct net_device *dev)
4452 const struct bonding *bond = netdev_priv(dev);
4454 switch (bond->params.mode) {
4455 case BOND_MODE_ROUNDROBIN:
4456 return bond_xmit_roundrobin(skb, dev);
4457 case BOND_MODE_ACTIVEBACKUP:
4458 return bond_xmit_activebackup(skb, dev);
4460 return bond_xmit_xor(skb, dev);
4461 case BOND_MODE_BROADCAST:
4462 return bond_xmit_broadcast(skb, dev);
4463 case BOND_MODE_8023AD:
4464 return bond_3ad_xmit_xor(skb, dev);
4467 return bond_alb_xmit(skb, dev);
4469 /* Should never happen, mode already checked */
4470 pr_err(DRV_NAME ": %s: Error: Unknown bonding mode %d\n",
4471 dev->name, bond->params.mode);
4474 return NETDEV_TX_OK;
4480 * set bond mode specific net device operations
4482 void bond_set_mode_ops(struct bonding *bond, int mode)
4484 struct net_device *bond_dev = bond->dev;
4487 case BOND_MODE_ROUNDROBIN:
4489 case BOND_MODE_ACTIVEBACKUP:
4492 bond_set_xmit_hash_policy(bond);
4494 case BOND_MODE_BROADCAST:
4496 case BOND_MODE_8023AD:
4497 bond_set_master_3ad_flags(bond);
4498 bond_set_xmit_hash_policy(bond);
4501 bond_set_master_alb_flags(bond);
4506 /* Should never happen, mode already checked */
4508 ": %s: Error: Unknown bonding mode %d\n",
4515 static void bond_ethtool_get_drvinfo(struct net_device *bond_dev,
4516 struct ethtool_drvinfo *drvinfo)
4518 strncpy(drvinfo->driver, DRV_NAME, 32);
4519 strncpy(drvinfo->version, DRV_VERSION, 32);
4520 snprintf(drvinfo->fw_version, 32, "%d", BOND_ABI_VERSION);
4523 static const struct ethtool_ops bond_ethtool_ops = {
4524 .get_drvinfo = bond_ethtool_get_drvinfo,
4525 .get_link = ethtool_op_get_link,
4526 .get_tx_csum = ethtool_op_get_tx_csum,
4527 .get_sg = ethtool_op_get_sg,
4528 .get_tso = ethtool_op_get_tso,
4529 .get_ufo = ethtool_op_get_ufo,
4530 .get_flags = ethtool_op_get_flags,
4533 static const struct net_device_ops bond_netdev_ops = {
4534 .ndo_init = bond_init,
4535 .ndo_uninit = bond_uninit,
4536 .ndo_open = bond_open,
4537 .ndo_stop = bond_close,
4538 .ndo_start_xmit = bond_start_xmit,
4539 .ndo_get_stats = bond_get_stats,
4540 .ndo_do_ioctl = bond_do_ioctl,
4541 .ndo_set_multicast_list = bond_set_multicast_list,
4542 .ndo_change_mtu = bond_change_mtu,
4543 .ndo_set_mac_address = bond_set_mac_address,
4544 .ndo_neigh_setup = bond_neigh_setup,
4545 .ndo_vlan_rx_register = bond_vlan_rx_register,
4546 .ndo_vlan_rx_add_vid = bond_vlan_rx_add_vid,
4547 .ndo_vlan_rx_kill_vid = bond_vlan_rx_kill_vid,
4550 static void bond_setup(struct net_device *bond_dev)
4552 struct bonding *bond = netdev_priv(bond_dev);
4554 /* initialize rwlocks */
4555 rwlock_init(&bond->lock);
4556 rwlock_init(&bond->curr_slave_lock);
4558 bond->params = bonding_defaults;
4560 /* Initialize pointers */
4561 bond->dev = bond_dev;
4562 INIT_LIST_HEAD(&bond->vlan_list);
4564 /* Initialize the device entry points */
4565 ether_setup(bond_dev);
4566 bond_dev->netdev_ops = &bond_netdev_ops;
4567 bond_dev->ethtool_ops = &bond_ethtool_ops;
4568 bond_set_mode_ops(bond, bond->params.mode);
4570 bond_dev->destructor = free_netdev;
4572 /* Initialize the device options */
4573 bond_dev->tx_queue_len = 0;
4574 bond_dev->flags |= IFF_MASTER|IFF_MULTICAST;
4575 bond_dev->priv_flags |= IFF_BONDING;
4576 bond_dev->priv_flags &= ~IFF_XMIT_DST_RELEASE;
4578 if (bond->params.arp_interval)
4579 bond_dev->priv_flags |= IFF_MASTER_ARPMON;
4581 /* At first, we block adding VLANs. That's the only way to
4582 * prevent problems that occur when adding VLANs over an
4583 * empty bond. The block will be removed once non-challenged
4584 * slaves are enslaved.
4586 bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
4588 /* don't acquire bond device's netif_tx_lock when
4590 bond_dev->features |= NETIF_F_LLTX;
4592 /* By default, we declare the bond to be fully
4593 * VLAN hardware accelerated capable. Special
4594 * care is taken in the various xmit functions
4595 * when there are slaves that are not hw accel
4598 bond_dev->features |= (NETIF_F_HW_VLAN_TX |
4599 NETIF_F_HW_VLAN_RX |
4600 NETIF_F_HW_VLAN_FILTER);
4604 static void bond_work_cancel_all(struct bonding *bond)
4606 write_lock_bh(&bond->lock);
4607 bond->kill_timers = 1;
4608 write_unlock_bh(&bond->lock);
4610 if (bond->params.miimon && delayed_work_pending(&bond->mii_work))
4611 cancel_delayed_work(&bond->mii_work);
4613 if (bond->params.arp_interval && delayed_work_pending(&bond->arp_work))
4614 cancel_delayed_work(&bond->arp_work);
4616 if (bond->params.mode == BOND_MODE_ALB &&
4617 delayed_work_pending(&bond->alb_work))
4618 cancel_delayed_work(&bond->alb_work);
4620 if (bond->params.mode == BOND_MODE_8023AD &&
4621 delayed_work_pending(&bond->ad_work))
4622 cancel_delayed_work(&bond->ad_work);
4625 /* De-initialize device specific data.
4626 * Caller must hold rtnl_lock.
4628 static void bond_deinit(struct net_device *bond_dev)
4630 struct bonding *bond = netdev_priv(bond_dev);
4632 list_del(&bond->bond_list);
4634 bond_work_cancel_all(bond);
4636 bond_remove_proc_entry(bond);
4639 /* Unregister and free all bond devices.
4640 * Caller must hold rtnl_lock.
4642 static void bond_free_all(void)
4644 struct bonding *bond, *nxt;
4646 list_for_each_entry_safe(bond, nxt, &bond_dev_list, bond_list) {
4647 struct net_device *bond_dev = bond->dev;
4649 bond_work_cancel_all(bond);
4650 /* Release the bonded slaves */
4651 bond_release_all(bond_dev);
4652 unregister_netdevice(bond_dev);
4655 bond_destroy_proc_dir();
4658 /*------------------------- Module initialization ---------------------------*/
4661 * Convert string input module parms. Accept either the
4662 * number of the mode or its string name. A bit complicated because
4663 * some mode names are substrings of other names, and calls from sysfs
4664 * may have whitespace in the name (trailing newlines, for example).
4666 int bond_parse_parm(const char *buf, const struct bond_parm_tbl *tbl)
4668 int modeint = -1, i, rv;
4669 char *p, modestr[BOND_MAX_MODENAME_LEN + 1] = { 0, };
4671 for (p = (char *)buf; *p; p++)
4672 if (!(isdigit(*p) || isspace(*p)))
4676 rv = sscanf(buf, "%20s", modestr);
4678 rv = sscanf(buf, "%d", &modeint);
4683 for (i = 0; tbl[i].modename; i++) {
4684 if (modeint == tbl[i].mode)
4686 if (strcmp(modestr, tbl[i].modename) == 0)
4693 static int bond_check_params(struct bond_params *params)
4695 int arp_validate_value, fail_over_mac_value;
4698 * Convert string parameters.
4701 bond_mode = bond_parse_parm(mode, bond_mode_tbl);
4702 if (bond_mode == -1) {
4704 ": Error: Invalid bonding mode \"%s\"\n",
4705 mode == NULL ? "NULL" : mode);
4710 if (xmit_hash_policy) {
4711 if ((bond_mode != BOND_MODE_XOR) &&
4712 (bond_mode != BOND_MODE_8023AD)) {
4714 ": xor_mode param is irrelevant in mode %s\n",
4715 bond_mode_name(bond_mode));
4717 xmit_hashtype = bond_parse_parm(xmit_hash_policy,
4719 if (xmit_hashtype == -1) {
4721 ": Error: Invalid xmit_hash_policy \"%s\"\n",
4722 xmit_hash_policy == NULL ? "NULL" :
4730 if (bond_mode != BOND_MODE_8023AD) {
4732 ": lacp_rate param is irrelevant in mode %s\n",
4733 bond_mode_name(bond_mode));
4735 lacp_fast = bond_parse_parm(lacp_rate, bond_lacp_tbl);
4736 if (lacp_fast == -1) {
4738 ": Error: Invalid lacp rate \"%s\"\n",
4739 lacp_rate == NULL ? "NULL" : lacp_rate);
4746 params->ad_select = bond_parse_parm(ad_select, ad_select_tbl);
4747 if (params->ad_select == -1) {
4749 ": Error: Invalid ad_select \"%s\"\n",
4750 ad_select == NULL ? "NULL" : ad_select);
4754 if (bond_mode != BOND_MODE_8023AD) {
4756 ": ad_select param only affects 802.3ad mode\n");
4759 params->ad_select = BOND_AD_STABLE;
4762 if (max_bonds < 0 || max_bonds > INT_MAX) {
4764 ": Warning: max_bonds (%d) not in range %d-%d, so it "
4765 "was reset to BOND_DEFAULT_MAX_BONDS (%d)\n",
4766 max_bonds, 0, INT_MAX, BOND_DEFAULT_MAX_BONDS);
4767 max_bonds = BOND_DEFAULT_MAX_BONDS;
4772 ": Warning: miimon module parameter (%d), "
4773 "not in range 0-%d, so it was reset to %d\n",
4774 miimon, INT_MAX, BOND_LINK_MON_INTERV);
4775 miimon = BOND_LINK_MON_INTERV;
4780 ": Warning: updelay module parameter (%d), "
4781 "not in range 0-%d, so it was reset to 0\n",
4786 if (downdelay < 0) {
4788 ": Warning: downdelay module parameter (%d), "
4789 "not in range 0-%d, so it was reset to 0\n",
4790 downdelay, INT_MAX);
4794 if ((use_carrier != 0) && (use_carrier != 1)) {
4796 ": Warning: use_carrier module parameter (%d), "
4797 "not of valid value (0/1), so it was set to 1\n",
4802 if (num_grat_arp < 0 || num_grat_arp > 255) {
4804 ": Warning: num_grat_arp (%d) not in range 0-255 so it "
4805 "was reset to 1 \n", num_grat_arp);
4809 if (num_unsol_na < 0 || num_unsol_na > 255) {
4811 ": Warning: num_unsol_na (%d) not in range 0-255 so it "
4812 "was reset to 1 \n", num_unsol_na);
4816 /* reset values for 802.3ad */
4817 if (bond_mode == BOND_MODE_8023AD) {
4820 ": Warning: miimon must be specified, "
4821 "otherwise bonding will not detect link "
4822 "failure, speed and duplex which are "
4823 "essential for 802.3ad operation\n");
4824 pr_warning("Forcing miimon to 100msec\n");
4829 /* reset values for TLB/ALB */
4830 if ((bond_mode == BOND_MODE_TLB) ||
4831 (bond_mode == BOND_MODE_ALB)) {
4834 ": Warning: miimon must be specified, "
4835 "otherwise bonding will not detect link "
4836 "failure and link speed which are essential "
4837 "for TLB/ALB load balancing\n");
4838 pr_warning("Forcing miimon to 100msec\n");
4843 if (bond_mode == BOND_MODE_ALB) {
4845 ": In ALB mode you might experience client "
4846 "disconnections upon reconnection of a link if the "
4847 "bonding module updelay parameter (%d msec) is "
4848 "incompatible with the forwarding delay time of the "
4854 if (updelay || downdelay) {
4855 /* just warn the user the up/down delay will have
4856 * no effect since miimon is zero...
4859 ": Warning: miimon module parameter not set "
4860 "and updelay (%d) or downdelay (%d) module "
4861 "parameter is set; updelay and downdelay have "
4862 "no effect unless miimon is set\n",
4863 updelay, downdelay);
4866 /* don't allow arp monitoring */
4869 ": Warning: miimon (%d) and arp_interval (%d) "
4870 "can't be used simultaneously, disabling ARP "
4872 miimon, arp_interval);
4876 if ((updelay % miimon) != 0) {
4878 ": Warning: updelay (%d) is not a multiple "
4879 "of miimon (%d), updelay rounded to %d ms\n",
4880 updelay, miimon, (updelay / miimon) * miimon);
4885 if ((downdelay % miimon) != 0) {
4887 ": Warning: downdelay (%d) is not a multiple "
4888 "of miimon (%d), downdelay rounded to %d ms\n",
4890 (downdelay / miimon) * miimon);
4893 downdelay /= miimon;
4896 if (arp_interval < 0) {
4898 ": Warning: arp_interval module parameter (%d) "
4899 ", not in range 0-%d, so it was reset to %d\n",
4900 arp_interval, INT_MAX, BOND_LINK_ARP_INTERV);
4901 arp_interval = BOND_LINK_ARP_INTERV;
4904 for (arp_ip_count = 0;
4905 (arp_ip_count < BOND_MAX_ARP_TARGETS) && arp_ip_target[arp_ip_count];
4907 /* not complete check, but should be good enough to
4909 if (!isdigit(arp_ip_target[arp_ip_count][0])) {
4911 ": Warning: bad arp_ip_target module parameter "
4912 "(%s), ARP monitoring will not be performed\n",
4913 arp_ip_target[arp_ip_count]);
4916 __be32 ip = in_aton(arp_ip_target[arp_ip_count]);
4917 arp_target[arp_ip_count] = ip;
4921 if (arp_interval && !arp_ip_count) {
4922 /* don't allow arping if no arp_ip_target given... */
4924 ": Warning: arp_interval module parameter (%d) "
4925 "specified without providing an arp_ip_target "
4926 "parameter, arp_interval was reset to 0\n",
4932 if (bond_mode != BOND_MODE_ACTIVEBACKUP) {
4934 ": arp_validate only supported in active-backup mode\n");
4937 if (!arp_interval) {
4939 ": arp_validate requires arp_interval\n");
4943 arp_validate_value = bond_parse_parm(arp_validate,
4945 if (arp_validate_value == -1) {
4947 ": Error: invalid arp_validate \"%s\"\n",
4948 arp_validate == NULL ? "NULL" : arp_validate);
4952 arp_validate_value = 0;
4956 ": MII link monitoring set to %d ms\n",
4958 } else if (arp_interval) {
4961 pr_info(DRV_NAME ": ARP monitoring set to %d ms,"
4962 " validate %s, with %d target(s):",
4964 arp_validate_tbl[arp_validate_value].modename,
4967 for (i = 0; i < arp_ip_count; i++)
4968 pr_info(" %s", arp_ip_target[i]);
4972 } else if (max_bonds) {
4973 /* miimon and arp_interval not set, we need one so things
4974 * work as expected, see bonding.txt for details
4977 ": Warning: either miimon or arp_interval and "
4978 "arp_ip_target module parameters must be specified, "
4979 "otherwise bonding will not detect link failures! see "
4980 "bonding.txt for details.\n");
4983 if (primary && !USES_PRIMARY(bond_mode)) {
4984 /* currently, using a primary only makes sense
4985 * in active backup, TLB or ALB modes
4988 ": Warning: %s primary device specified but has no "
4989 "effect in %s mode\n",
4990 primary, bond_mode_name(bond_mode));
4994 if (fail_over_mac) {
4995 fail_over_mac_value = bond_parse_parm(fail_over_mac,
4997 if (fail_over_mac_value == -1) {
4999 ": Error: invalid fail_over_mac \"%s\"\n",
5000 arp_validate == NULL ? "NULL" : arp_validate);
5004 if (bond_mode != BOND_MODE_ACTIVEBACKUP)
5006 ": Warning: fail_over_mac only affects "
5007 "active-backup mode.\n");
5009 fail_over_mac_value = BOND_FOM_NONE;
5012 /* fill params struct with the proper values */
5013 params->mode = bond_mode;
5014 params->xmit_policy = xmit_hashtype;
5015 params->miimon = miimon;
5016 params->num_grat_arp = num_grat_arp;
5017 params->num_unsol_na = num_unsol_na;
5018 params->arp_interval = arp_interval;
5019 params->arp_validate = arp_validate_value;
5020 params->updelay = updelay;
5021 params->downdelay = downdelay;
5022 params->use_carrier = use_carrier;
5023 params->lacp_fast = lacp_fast;
5024 params->primary[0] = 0;
5025 params->fail_over_mac = fail_over_mac_value;
5028 strncpy(params->primary, primary, IFNAMSIZ);
5029 params->primary[IFNAMSIZ - 1] = 0;
5032 memcpy(params->arp_targets, arp_target, sizeof(arp_target));
5037 static struct lock_class_key bonding_netdev_xmit_lock_key;
5038 static struct lock_class_key bonding_netdev_addr_lock_key;
5040 static void bond_set_lockdep_class_one(struct net_device *dev,
5041 struct netdev_queue *txq,
5044 lockdep_set_class(&txq->_xmit_lock,
5045 &bonding_netdev_xmit_lock_key);
5048 static void bond_set_lockdep_class(struct net_device *dev)
5050 lockdep_set_class(&dev->addr_list_lock,
5051 &bonding_netdev_addr_lock_key);
5052 netdev_for_each_tx_queue(dev, bond_set_lockdep_class_one, NULL);
5056 * Called from registration process
5058 static int bond_init(struct net_device *bond_dev)
5060 struct bonding *bond = netdev_priv(bond_dev);
5062 pr_debug("Begin bond_init for %s\n", bond_dev->name);
5064 bond->wq = create_singlethread_workqueue(bond_dev->name);
5068 bond_set_lockdep_class(bond_dev);
5070 netif_carrier_off(bond_dev);
5072 bond_create_proc_entry(bond);
5073 list_add_tail(&bond->bond_list, &bond_dev_list);
5078 /* Create a new bond based on the specified name and bonding parameters.
5079 * If name is NULL, obtain a suitable "bond%d" name for us.
5080 * Caller must NOT hold rtnl_lock; we need to release it here before we
5081 * set up our sysfs entries.
5083 int bond_create(const char *name)
5085 struct net_device *bond_dev;
5089 /* Check to see if the bond already exists. */
5090 /* FIXME: pass netns from caller */
5091 if (name && __dev_get_by_name(&init_net, name)) {
5092 pr_err(DRV_NAME ": cannot add bond %s; already exists\n",
5098 bond_dev = alloc_netdev(sizeof(struct bonding), name ? name : "",
5101 pr_err(DRV_NAME ": %s: eek! can't alloc netdev!\n",
5108 res = dev_alloc_name(bond_dev, "bond%d");
5113 res = register_netdevice(bond_dev);
5117 res = bond_create_sysfs_entry(netdev_priv(bond_dev));
5125 unregister_netdevice(bond_dev);
5127 bond_deinit(bond_dev);
5129 free_netdev(bond_dev);
5135 static int __init bonding_init(void)
5140 pr_info("%s", version);
5142 res = bond_check_params(&bonding_defaults);
5146 bond_create_proc_dir();
5148 for (i = 0; i < max_bonds; i++) {
5149 res = bond_create(NULL);
5154 res = bond_create_sysfs();
5158 register_netdevice_notifier(&bond_netdev_notifier);
5159 register_inetaddr_notifier(&bond_inetaddr_notifier);
5160 bond_register_ipv6_notifier();
5172 static void __exit bonding_exit(void)
5174 unregister_netdevice_notifier(&bond_netdev_notifier);
5175 unregister_inetaddr_notifier(&bond_inetaddr_notifier);
5176 bond_unregister_ipv6_notifier();
5178 bond_destroy_sysfs();
5185 module_init(bonding_init);
5186 module_exit(bonding_exit);
5187 MODULE_LICENSE("GPL");
5188 MODULE_VERSION(DRV_VERSION);
5189 MODULE_DESCRIPTION(DRV_DESCRIPTION ", v" DRV_VERSION);
5190 MODULE_AUTHOR("Thomas Davis, tadavis@lbl.gov and many others");