2 * originally based on the dummy device.
4 * Copyright 1999, Thomas Davis, tadavis@lbl.gov.
5 * Licensed under the GPL. Based on dummy.c, and eql.c devices.
7 * bonding.c: an Ethernet Bonding driver
9 * This is useful to talk to a Cisco EtherChannel compatible equipment:
11 * Sun Trunking (Solaris)
12 * Alteon AceDirector Trunks
14 * and probably many L2 switches ...
17 * ifconfig bond0 ipaddress netmask up
18 * will setup a network device, with an ip address. No mac address
19 * will be assigned at this time. The hw mac address will come from
20 * the first slave bonded to the channel. All slaves will then use
21 * this hw mac address.
24 * will release all slaves, marking them as down.
26 * ifenslave bond0 eth0
27 * will attach eth0 to bond0 as a slave. eth0 hw mac address will either
28 * a: be used as initial mac address
29 * b: if a hw mac address already is there, eth0's hw mac address
30 * will then be set from bond0.
34 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
36 #include <linux/kernel.h>
37 #include <linux/module.h>
38 #include <linux/types.h>
39 #include <linux/fcntl.h>
40 #include <linux/interrupt.h>
41 #include <linux/ptrace.h>
42 #include <linux/ioport.h>
46 #include <linux/tcp.h>
47 #include <linux/udp.h>
48 #include <linux/slab.h>
49 #include <linux/string.h>
50 #include <linux/init.h>
51 #include <linux/timer.h>
52 #include <linux/socket.h>
53 #include <linux/ctype.h>
54 #include <linux/inet.h>
55 #include <linux/bitops.h>
57 #include <asm/system.h>
59 #include <linux/uaccess.h>
60 #include <linux/errno.h>
61 #include <linux/netdevice.h>
62 #include <linux/netpoll.h>
63 #include <linux/inetdevice.h>
64 #include <linux/igmp.h>
65 #include <linux/etherdevice.h>
66 #include <linux/skbuff.h>
68 #include <linux/rtnetlink.h>
69 #include <linux/proc_fs.h>
70 #include <linux/seq_file.h>
71 #include <linux/smp.h>
72 #include <linux/if_ether.h>
74 #include <linux/mii.h>
75 #include <linux/ethtool.h>
76 #include <linux/if_vlan.h>
77 #include <linux/if_bonding.h>
78 #include <linux/jiffies.h>
79 #include <net/route.h>
80 #include <net/net_namespace.h>
81 #include <net/netns/generic.h>
86 /*---------------------------- Module parameters ----------------------------*/
88 /* monitor all links that often (in milliseconds). <=0 disables monitoring */
89 #define BOND_LINK_MON_INTERV 0
90 #define BOND_LINK_ARP_INTERV 0
92 static int max_bonds = BOND_DEFAULT_MAX_BONDS;
93 static int tx_queues = BOND_DEFAULT_TX_QUEUES;
94 static int num_grat_arp = 1;
95 static int num_unsol_na = 1;
96 static int miimon = BOND_LINK_MON_INTERV;
99 static int use_carrier = 1;
101 static char *primary;
102 static char *primary_reselect;
103 static char *lacp_rate;
104 static char *ad_select;
105 static char *xmit_hash_policy;
106 static int arp_interval = BOND_LINK_ARP_INTERV;
107 static char *arp_ip_target[BOND_MAX_ARP_TARGETS];
108 static char *arp_validate;
109 static char *fail_over_mac;
110 static int all_slaves_active = 0;
111 static struct bond_params bonding_defaults;
113 module_param(max_bonds, int, 0);
114 MODULE_PARM_DESC(max_bonds, "Max number of bonded devices");
115 module_param(tx_queues, int, 0);
116 MODULE_PARM_DESC(tx_queues, "Max number of transmit queues (default = 16)");
117 module_param(num_grat_arp, int, 0644);
118 MODULE_PARM_DESC(num_grat_arp, "Number of gratuitous ARP packets to send on failover event");
119 module_param(num_unsol_na, int, 0644);
120 MODULE_PARM_DESC(num_unsol_na, "Number of unsolicited IPv6 Neighbor Advertisements packets to send on failover event");
121 module_param(miimon, int, 0);
122 MODULE_PARM_DESC(miimon, "Link check interval in milliseconds");
123 module_param(updelay, int, 0);
124 MODULE_PARM_DESC(updelay, "Delay before considering link up, in milliseconds");
125 module_param(downdelay, int, 0);
126 MODULE_PARM_DESC(downdelay, "Delay before considering link down, "
128 module_param(use_carrier, int, 0);
129 MODULE_PARM_DESC(use_carrier, "Use netif_carrier_ok (vs MII ioctls) in miimon; "
130 "0 for off, 1 for on (default)");
131 module_param(mode, charp, 0);
132 MODULE_PARM_DESC(mode, "Mode of operation : 0 for balance-rr, "
133 "1 for active-backup, 2 for balance-xor, "
134 "3 for broadcast, 4 for 802.3ad, 5 for balance-tlb, "
135 "6 for balance-alb");
136 module_param(primary, charp, 0);
137 MODULE_PARM_DESC(primary, "Primary network device to use");
138 module_param(primary_reselect, charp, 0);
139 MODULE_PARM_DESC(primary_reselect, "Reselect primary slave "
141 "0 for always (default), "
142 "1 for only if speed of primary is "
144 "2 for only on active slave "
146 module_param(lacp_rate, charp, 0);
147 MODULE_PARM_DESC(lacp_rate, "LACPDU tx rate to request from 802.3ad partner "
149 module_param(ad_select, charp, 0);
150 MODULE_PARM_DESC(ad_select, "803.ad aggregation selection logic: stable (0, default), bandwidth (1), count (2)");
151 module_param(xmit_hash_policy, charp, 0);
152 MODULE_PARM_DESC(xmit_hash_policy, "XOR hashing method: 0 for layer 2 (default)"
153 ", 1 for layer 3+4");
154 module_param(arp_interval, int, 0);
155 MODULE_PARM_DESC(arp_interval, "arp interval in milliseconds");
156 module_param_array(arp_ip_target, charp, NULL, 0);
157 MODULE_PARM_DESC(arp_ip_target, "arp targets in n.n.n.n form");
158 module_param(arp_validate, charp, 0);
159 MODULE_PARM_DESC(arp_validate, "validate src/dst of ARP probes: none (default), active, backup or all");
160 module_param(fail_over_mac, charp, 0);
161 MODULE_PARM_DESC(fail_over_mac, "For active-backup, do not set all slaves to the same MAC. none (default), active or follow");
162 module_param(all_slaves_active, int, 0);
163 MODULE_PARM_DESC(all_slaves_active, "Keep all frames received on an interface"
164 "by setting active flag for all slaves. "
165 "0 for never (default), 1 for always.");
167 /*----------------------------- Global variables ----------------------------*/
169 static const char * const version =
170 DRV_DESCRIPTION ": v" DRV_VERSION " (" DRV_RELDATE ")\n";
172 int bond_net_id __read_mostly;
174 static __be32 arp_target[BOND_MAX_ARP_TARGETS];
175 static int arp_ip_count;
176 static int bond_mode = BOND_MODE_ROUNDROBIN;
177 static int xmit_hashtype = BOND_XMIT_POLICY_LAYER2;
178 static int lacp_fast;
179 #ifdef CONFIG_NET_POLL_CONTROLLER
180 static int disable_netpoll = 1;
183 const struct bond_parm_tbl bond_lacp_tbl[] = {
184 { "slow", AD_LACP_SLOW},
185 { "fast", AD_LACP_FAST},
189 const struct bond_parm_tbl bond_mode_tbl[] = {
190 { "balance-rr", BOND_MODE_ROUNDROBIN},
191 { "active-backup", BOND_MODE_ACTIVEBACKUP},
192 { "balance-xor", BOND_MODE_XOR},
193 { "broadcast", BOND_MODE_BROADCAST},
194 { "802.3ad", BOND_MODE_8023AD},
195 { "balance-tlb", BOND_MODE_TLB},
196 { "balance-alb", BOND_MODE_ALB},
200 const struct bond_parm_tbl xmit_hashtype_tbl[] = {
201 { "layer2", BOND_XMIT_POLICY_LAYER2},
202 { "layer3+4", BOND_XMIT_POLICY_LAYER34},
203 { "layer2+3", BOND_XMIT_POLICY_LAYER23},
207 const struct bond_parm_tbl arp_validate_tbl[] = {
208 { "none", BOND_ARP_VALIDATE_NONE},
209 { "active", BOND_ARP_VALIDATE_ACTIVE},
210 { "backup", BOND_ARP_VALIDATE_BACKUP},
211 { "all", BOND_ARP_VALIDATE_ALL},
215 const struct bond_parm_tbl fail_over_mac_tbl[] = {
216 { "none", BOND_FOM_NONE},
217 { "active", BOND_FOM_ACTIVE},
218 { "follow", BOND_FOM_FOLLOW},
222 const struct bond_parm_tbl pri_reselect_tbl[] = {
223 { "always", BOND_PRI_RESELECT_ALWAYS},
224 { "better", BOND_PRI_RESELECT_BETTER},
225 { "failure", BOND_PRI_RESELECT_FAILURE},
229 struct bond_parm_tbl ad_select_tbl[] = {
230 { "stable", BOND_AD_STABLE},
231 { "bandwidth", BOND_AD_BANDWIDTH},
232 { "count", BOND_AD_COUNT},
236 /*-------------------------- Forward declarations ---------------------------*/
238 static void bond_send_gratuitous_arp(struct bonding *bond);
239 static int bond_init(struct net_device *bond_dev);
240 static void bond_uninit(struct net_device *bond_dev);
242 /*---------------------------- General routines -----------------------------*/
244 static const char *bond_mode_name(int mode)
246 static const char *names[] = {
247 [BOND_MODE_ROUNDROBIN] = "load balancing (round-robin)",
248 [BOND_MODE_ACTIVEBACKUP] = "fault-tolerance (active-backup)",
249 [BOND_MODE_XOR] = "load balancing (xor)",
250 [BOND_MODE_BROADCAST] = "fault-tolerance (broadcast)",
251 [BOND_MODE_8023AD] = "IEEE 802.3ad Dynamic link aggregation",
252 [BOND_MODE_TLB] = "transmit load balancing",
253 [BOND_MODE_ALB] = "adaptive load balancing",
256 if (mode < 0 || mode > BOND_MODE_ALB)
262 /*---------------------------------- VLAN -----------------------------------*/
265 * bond_add_vlan - add a new vlan id on bond
266 * @bond: bond that got the notification
267 * @vlan_id: the vlan id to add
269 * Returns -ENOMEM if allocation failed.
271 static int bond_add_vlan(struct bonding *bond, unsigned short vlan_id)
273 struct vlan_entry *vlan;
275 pr_debug("bond: %s, vlan id %d\n",
276 (bond ? bond->dev->name : "None"), vlan_id);
278 vlan = kzalloc(sizeof(struct vlan_entry), GFP_KERNEL);
282 INIT_LIST_HEAD(&vlan->vlan_list);
283 vlan->vlan_id = vlan_id;
285 write_lock_bh(&bond->lock);
287 list_add_tail(&vlan->vlan_list, &bond->vlan_list);
289 write_unlock_bh(&bond->lock);
291 pr_debug("added VLAN ID %d on bond %s\n", vlan_id, bond->dev->name);
297 * bond_del_vlan - delete a vlan id from bond
298 * @bond: bond that got the notification
299 * @vlan_id: the vlan id to delete
301 * returns -ENODEV if @vlan_id was not found in @bond.
303 static int bond_del_vlan(struct bonding *bond, unsigned short vlan_id)
305 struct vlan_entry *vlan;
308 pr_debug("bond: %s, vlan id %d\n", bond->dev->name, vlan_id);
310 write_lock_bh(&bond->lock);
312 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
313 if (vlan->vlan_id == vlan_id) {
314 list_del(&vlan->vlan_list);
316 if (bond_is_lb(bond))
317 bond_alb_clear_vlan(bond, vlan_id);
319 pr_debug("removed VLAN ID %d from bond %s\n",
320 vlan_id, bond->dev->name);
324 if (list_empty(&bond->vlan_list) &&
325 (bond->slave_cnt == 0)) {
326 /* Last VLAN removed and no slaves, so
327 * restore block on adding VLANs. This will
328 * be removed once new slaves that are not
329 * VLAN challenged will be added.
331 bond->dev->features |= NETIF_F_VLAN_CHALLENGED;
339 pr_debug("couldn't find VLAN ID %d in bond %s\n",
340 vlan_id, bond->dev->name);
343 write_unlock_bh(&bond->lock);
348 * bond_has_challenged_slaves
349 * @bond: the bond we're working on
351 * Searches the slave list. Returns 1 if a vlan challenged slave
352 * was found, 0 otherwise.
354 * Assumes bond->lock is held.
356 static int bond_has_challenged_slaves(struct bonding *bond)
361 bond_for_each_slave(bond, slave, i) {
362 if (slave->dev->features & NETIF_F_VLAN_CHALLENGED) {
363 pr_debug("found VLAN challenged slave - %s\n",
369 pr_debug("no VLAN challenged slaves found\n");
374 * bond_next_vlan - safely skip to the next item in the vlans list.
375 * @bond: the bond we're working on
376 * @curr: item we're advancing from
378 * Returns %NULL if list is empty, bond->next_vlan if @curr is %NULL,
379 * or @curr->next otherwise (even if it is @curr itself again).
381 * Caller must hold bond->lock
383 struct vlan_entry *bond_next_vlan(struct bonding *bond, struct vlan_entry *curr)
385 struct vlan_entry *next, *last;
387 if (list_empty(&bond->vlan_list))
391 next = list_entry(bond->vlan_list.next,
392 struct vlan_entry, vlan_list);
394 last = list_entry(bond->vlan_list.prev,
395 struct vlan_entry, vlan_list);
397 next = list_entry(bond->vlan_list.next,
398 struct vlan_entry, vlan_list);
400 next = list_entry(curr->vlan_list.next,
401 struct vlan_entry, vlan_list);
409 * bond_dev_queue_xmit - Prepare skb for xmit.
411 * @bond: bond device that got this skb for tx.
412 * @skb: hw accel VLAN tagged skb to transmit
413 * @slave_dev: slave that is supposed to xmit this skbuff
415 * When the bond gets an skb to transmit that is
416 * already hardware accelerated VLAN tagged, and it
417 * needs to relay this skb to a slave that is not
418 * hw accel capable, the skb needs to be "unaccelerated",
419 * i.e. strip the hwaccel tag and re-insert it as part
422 int bond_dev_queue_xmit(struct bonding *bond, struct sk_buff *skb,
423 struct net_device *slave_dev)
425 unsigned short uninitialized_var(vlan_id);
427 if (!list_empty(&bond->vlan_list) &&
428 !(slave_dev->features & NETIF_F_HW_VLAN_TX) &&
429 vlan_get_tag(skb, &vlan_id) == 0) {
430 skb->dev = slave_dev;
431 skb = vlan_put_tag(skb, vlan_id);
433 /* vlan_put_tag() frees the skb in case of error,
434 * so return success here so the calling functions
435 * won't attempt to free is again.
440 skb->dev = slave_dev;
444 #ifdef CONFIG_NET_POLL_CONTROLLER
445 if (unlikely(bond->dev->priv_flags & IFF_IN_NETPOLL)) {
446 struct netpoll *np = bond->dev->npinfo->netpoll;
447 slave_dev->npinfo = bond->dev->npinfo;
448 np->real_dev = np->dev = skb->dev;
449 slave_dev->priv_flags |= IFF_IN_NETPOLL;
450 netpoll_send_skb(np, skb);
451 slave_dev->priv_flags &= ~IFF_IN_NETPOLL;
461 * In the following 3 functions, bond_vlan_rx_register(), bond_vlan_rx_add_vid
462 * and bond_vlan_rx_kill_vid, We don't protect the slave list iteration with a
464 * a. This operation is performed in IOCTL context,
465 * b. The operation is protected by the RTNL semaphore in the 8021q code,
466 * c. Holding a lock with BH disabled while directly calling a base driver
467 * entry point is generally a BAD idea.
469 * The design of synchronization/protection for this operation in the 8021q
470 * module is good for one or more VLAN devices over a single physical device
471 * and cannot be extended for a teaming solution like bonding, so there is a
472 * potential race condition here where a net device from the vlan group might
473 * be referenced (either by a base driver or the 8021q code) while it is being
474 * removed from the system. However, it turns out we're not making matters
475 * worse, and if it works for regular VLAN usage it will work here too.
479 * bond_vlan_rx_register - Propagates registration to slaves
480 * @bond_dev: bonding net device that got called
481 * @grp: vlan group being registered
483 static void bond_vlan_rx_register(struct net_device *bond_dev,
484 struct vlan_group *grp)
486 struct bonding *bond = netdev_priv(bond_dev);
492 bond_for_each_slave(bond, slave, i) {
493 struct net_device *slave_dev = slave->dev;
494 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
496 if ((slave_dev->features & NETIF_F_HW_VLAN_RX) &&
497 slave_ops->ndo_vlan_rx_register) {
498 slave_ops->ndo_vlan_rx_register(slave_dev, grp);
504 * bond_vlan_rx_add_vid - Propagates adding an id to slaves
505 * @bond_dev: bonding net device that got called
506 * @vid: vlan id being added
508 static void bond_vlan_rx_add_vid(struct net_device *bond_dev, uint16_t vid)
510 struct bonding *bond = netdev_priv(bond_dev);
514 bond_for_each_slave(bond, slave, i) {
515 struct net_device *slave_dev = slave->dev;
516 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
518 if ((slave_dev->features & NETIF_F_HW_VLAN_FILTER) &&
519 slave_ops->ndo_vlan_rx_add_vid) {
520 slave_ops->ndo_vlan_rx_add_vid(slave_dev, vid);
524 res = bond_add_vlan(bond, vid);
526 pr_err("%s: Error: Failed to add vlan id %d\n",
527 bond_dev->name, vid);
532 * bond_vlan_rx_kill_vid - Propagates deleting an id to slaves
533 * @bond_dev: bonding net device that got called
534 * @vid: vlan id being removed
536 static void bond_vlan_rx_kill_vid(struct net_device *bond_dev, uint16_t vid)
538 struct bonding *bond = netdev_priv(bond_dev);
540 struct net_device *vlan_dev;
543 bond_for_each_slave(bond, slave, i) {
544 struct net_device *slave_dev = slave->dev;
545 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
547 if ((slave_dev->features & NETIF_F_HW_VLAN_FILTER) &&
548 slave_ops->ndo_vlan_rx_kill_vid) {
549 /* Save and then restore vlan_dev in the grp array,
550 * since the slave's driver might clear it.
552 vlan_dev = vlan_group_get_device(bond->vlgrp, vid);
553 slave_ops->ndo_vlan_rx_kill_vid(slave_dev, vid);
554 vlan_group_set_device(bond->vlgrp, vid, vlan_dev);
558 res = bond_del_vlan(bond, vid);
560 pr_err("%s: Error: Failed to remove vlan id %d\n",
561 bond_dev->name, vid);
565 static void bond_add_vlans_on_slave(struct bonding *bond, struct net_device *slave_dev)
567 struct vlan_entry *vlan;
568 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
570 write_lock_bh(&bond->lock);
572 if (list_empty(&bond->vlan_list))
575 if ((slave_dev->features & NETIF_F_HW_VLAN_RX) &&
576 slave_ops->ndo_vlan_rx_register)
577 slave_ops->ndo_vlan_rx_register(slave_dev, bond->vlgrp);
579 if (!(slave_dev->features & NETIF_F_HW_VLAN_FILTER) ||
580 !(slave_ops->ndo_vlan_rx_add_vid))
583 list_for_each_entry(vlan, &bond->vlan_list, vlan_list)
584 slave_ops->ndo_vlan_rx_add_vid(slave_dev, vlan->vlan_id);
587 write_unlock_bh(&bond->lock);
590 static void bond_del_vlans_from_slave(struct bonding *bond,
591 struct net_device *slave_dev)
593 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
594 struct vlan_entry *vlan;
595 struct net_device *vlan_dev;
597 write_lock_bh(&bond->lock);
599 if (list_empty(&bond->vlan_list))
602 if (!(slave_dev->features & NETIF_F_HW_VLAN_FILTER) ||
603 !(slave_ops->ndo_vlan_rx_kill_vid))
606 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
607 /* Save and then restore vlan_dev in the grp array,
608 * since the slave's driver might clear it.
610 vlan_dev = vlan_group_get_device(bond->vlgrp, vlan->vlan_id);
611 slave_ops->ndo_vlan_rx_kill_vid(slave_dev, vlan->vlan_id);
612 vlan_group_set_device(bond->vlgrp, vlan->vlan_id, vlan_dev);
616 if ((slave_dev->features & NETIF_F_HW_VLAN_RX) &&
617 slave_ops->ndo_vlan_rx_register)
618 slave_ops->ndo_vlan_rx_register(slave_dev, NULL);
621 write_unlock_bh(&bond->lock);
624 /*------------------------------- Link status -------------------------------*/
627 * Set the carrier state for the master according to the state of its
628 * slaves. If any slaves are up, the master is up. In 802.3ad mode,
629 * do special 802.3ad magic.
631 * Returns zero if carrier state does not change, nonzero if it does.
633 static int bond_set_carrier(struct bonding *bond)
638 if (bond->slave_cnt == 0)
641 if (bond->params.mode == BOND_MODE_8023AD)
642 return bond_3ad_set_carrier(bond);
644 bond_for_each_slave(bond, slave, i) {
645 if (slave->link == BOND_LINK_UP) {
646 if (!netif_carrier_ok(bond->dev)) {
647 netif_carrier_on(bond->dev);
655 if (netif_carrier_ok(bond->dev)) {
656 netif_carrier_off(bond->dev);
663 * Get link speed and duplex from the slave's base driver
664 * using ethtool. If for some reason the call fails or the
665 * values are invalid, fake speed and duplex to 100/Full
668 static int bond_update_speed_duplex(struct slave *slave)
670 struct net_device *slave_dev = slave->dev;
671 struct ethtool_cmd etool;
674 /* Fake speed and duplex */
675 slave->speed = SPEED_100;
676 slave->duplex = DUPLEX_FULL;
678 if (!slave_dev->ethtool_ops || !slave_dev->ethtool_ops->get_settings)
681 res = slave_dev->ethtool_ops->get_settings(slave_dev, &etool);
685 switch (etool.speed) {
695 switch (etool.duplex) {
703 slave->speed = etool.speed;
704 slave->duplex = etool.duplex;
710 * if <dev> supports MII link status reporting, check its link status.
712 * We either do MII/ETHTOOL ioctls, or check netif_carrier_ok(),
713 * depending upon the setting of the use_carrier parameter.
715 * Return either BMSR_LSTATUS, meaning that the link is up (or we
716 * can't tell and just pretend it is), or 0, meaning that the link is
719 * If reporting is non-zero, instead of faking link up, return -1 if
720 * both ETHTOOL and MII ioctls fail (meaning the device does not
721 * support them). If use_carrier is set, return whatever it says.
722 * It'd be nice if there was a good way to tell if a driver supports
723 * netif_carrier, but there really isn't.
725 static int bond_check_dev_link(struct bonding *bond,
726 struct net_device *slave_dev, int reporting)
728 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
729 int (*ioctl)(struct net_device *, struct ifreq *, int);
731 struct mii_ioctl_data *mii;
733 if (!reporting && !netif_running(slave_dev))
736 if (bond->params.use_carrier)
737 return netif_carrier_ok(slave_dev) ? BMSR_LSTATUS : 0;
739 /* Try to get link status using Ethtool first. */
740 if (slave_dev->ethtool_ops) {
741 if (slave_dev->ethtool_ops->get_link) {
744 link = slave_dev->ethtool_ops->get_link(slave_dev);
746 return link ? BMSR_LSTATUS : 0;
750 /* Ethtool can't be used, fallback to MII ioctls. */
751 ioctl = slave_ops->ndo_do_ioctl;
753 /* TODO: set pointer to correct ioctl on a per team member */
754 /* bases to make this more efficient. that is, once */
755 /* we determine the correct ioctl, we will always */
756 /* call it and not the others for that team */
760 * We cannot assume that SIOCGMIIPHY will also read a
761 * register; not all network drivers (e.g., e100)
765 /* Yes, the mii is overlaid on the ifreq.ifr_ifru */
766 strncpy(ifr.ifr_name, slave_dev->name, IFNAMSIZ);
768 if (IOCTL(slave_dev, &ifr, SIOCGMIIPHY) == 0) {
769 mii->reg_num = MII_BMSR;
770 if (IOCTL(slave_dev, &ifr, SIOCGMIIREG) == 0)
771 return mii->val_out & BMSR_LSTATUS;
776 * If reporting, report that either there's no dev->do_ioctl,
777 * or both SIOCGMIIREG and get_link failed (meaning that we
778 * cannot report link status). If not reporting, pretend
781 return reporting ? -1 : BMSR_LSTATUS;
784 /*----------------------------- Multicast list ------------------------------*/
787 * Push the promiscuity flag down to appropriate slaves
789 static int bond_set_promiscuity(struct bonding *bond, int inc)
792 if (USES_PRIMARY(bond->params.mode)) {
793 /* write lock already acquired */
794 if (bond->curr_active_slave) {
795 err = dev_set_promiscuity(bond->curr_active_slave->dev,
801 bond_for_each_slave(bond, slave, i) {
802 err = dev_set_promiscuity(slave->dev, inc);
811 * Push the allmulti flag down to all slaves
813 static int bond_set_allmulti(struct bonding *bond, int inc)
816 if (USES_PRIMARY(bond->params.mode)) {
817 /* write lock already acquired */
818 if (bond->curr_active_slave) {
819 err = dev_set_allmulti(bond->curr_active_slave->dev,
825 bond_for_each_slave(bond, slave, i) {
826 err = dev_set_allmulti(slave->dev, inc);
835 * Add a Multicast address to slaves
838 static void bond_mc_add(struct bonding *bond, void *addr)
840 if (USES_PRIMARY(bond->params.mode)) {
841 /* write lock already acquired */
842 if (bond->curr_active_slave)
843 dev_mc_add(bond->curr_active_slave->dev, addr);
848 bond_for_each_slave(bond, slave, i)
849 dev_mc_add(slave->dev, addr);
854 * Remove a multicast address from slave
857 static void bond_mc_del(struct bonding *bond, void *addr)
859 if (USES_PRIMARY(bond->params.mode)) {
860 /* write lock already acquired */
861 if (bond->curr_active_slave)
862 dev_mc_del(bond->curr_active_slave->dev, addr);
866 bond_for_each_slave(bond, slave, i) {
867 dev_mc_del(slave->dev, addr);
874 * Retrieve the list of registered multicast addresses for the bonding
875 * device and retransmit an IGMP JOIN request to the current active
878 static void bond_resend_igmp_join_requests(struct bonding *bond)
880 struct in_device *in_dev;
881 struct ip_mc_list *im;
884 in_dev = __in_dev_get_rcu(bond->dev);
886 for (im = in_dev->mc_list; im; im = im->next)
887 ip_mc_rejoin_group(im);
894 * flush all members of flush->mc_list from device dev->mc_list
896 static void bond_mc_list_flush(struct net_device *bond_dev,
897 struct net_device *slave_dev)
899 struct bonding *bond = netdev_priv(bond_dev);
900 struct netdev_hw_addr *ha;
902 netdev_for_each_mc_addr(ha, bond_dev)
903 dev_mc_del(slave_dev, ha->addr);
905 if (bond->params.mode == BOND_MODE_8023AD) {
906 /* del lacpdu mc addr from mc list */
907 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
909 dev_mc_del(slave_dev, lacpdu_multicast);
913 /*--------------------------- Active slave change ---------------------------*/
916 * Update the mc list and multicast-related flags for the new and
917 * old active slaves (if any) according to the multicast mode, and
918 * promiscuous flags unconditionally.
920 static void bond_mc_swap(struct bonding *bond, struct slave *new_active,
921 struct slave *old_active)
923 struct netdev_hw_addr *ha;
925 if (!USES_PRIMARY(bond->params.mode))
926 /* nothing to do - mc list is already up-to-date on
932 if (bond->dev->flags & IFF_PROMISC)
933 dev_set_promiscuity(old_active->dev, -1);
935 if (bond->dev->flags & IFF_ALLMULTI)
936 dev_set_allmulti(old_active->dev, -1);
938 netdev_for_each_mc_addr(ha, bond->dev)
939 dev_mc_del(old_active->dev, ha->addr);
943 /* FIXME: Signal errors upstream. */
944 if (bond->dev->flags & IFF_PROMISC)
945 dev_set_promiscuity(new_active->dev, 1);
947 if (bond->dev->flags & IFF_ALLMULTI)
948 dev_set_allmulti(new_active->dev, 1);
950 netdev_for_each_mc_addr(ha, bond->dev)
951 dev_mc_add(new_active->dev, ha->addr);
952 bond_resend_igmp_join_requests(bond);
957 * bond_do_fail_over_mac
959 * Perform special MAC address swapping for fail_over_mac settings
961 * Called with RTNL, bond->lock for read, curr_slave_lock for write_bh.
963 static void bond_do_fail_over_mac(struct bonding *bond,
964 struct slave *new_active,
965 struct slave *old_active)
966 __releases(&bond->curr_slave_lock)
967 __releases(&bond->lock)
968 __acquires(&bond->lock)
969 __acquires(&bond->curr_slave_lock)
971 u8 tmp_mac[ETH_ALEN];
972 struct sockaddr saddr;
975 switch (bond->params.fail_over_mac) {
976 case BOND_FOM_ACTIVE:
978 memcpy(bond->dev->dev_addr, new_active->dev->dev_addr,
979 new_active->dev->addr_len);
981 case BOND_FOM_FOLLOW:
983 * if new_active && old_active, swap them
984 * if just old_active, do nothing (going to no active slave)
985 * if just new_active, set new_active to bond's MAC
990 write_unlock_bh(&bond->curr_slave_lock);
991 read_unlock(&bond->lock);
994 memcpy(tmp_mac, new_active->dev->dev_addr, ETH_ALEN);
995 memcpy(saddr.sa_data, old_active->dev->dev_addr,
997 saddr.sa_family = new_active->dev->type;
999 memcpy(saddr.sa_data, bond->dev->dev_addr, ETH_ALEN);
1000 saddr.sa_family = bond->dev->type;
1003 rv = dev_set_mac_address(new_active->dev, &saddr);
1005 pr_err("%s: Error %d setting MAC of slave %s\n",
1006 bond->dev->name, -rv, new_active->dev->name);
1013 memcpy(saddr.sa_data, tmp_mac, ETH_ALEN);
1014 saddr.sa_family = old_active->dev->type;
1016 rv = dev_set_mac_address(old_active->dev, &saddr);
1018 pr_err("%s: Error %d setting MAC of slave %s\n",
1019 bond->dev->name, -rv, new_active->dev->name);
1021 read_lock(&bond->lock);
1022 write_lock_bh(&bond->curr_slave_lock);
1025 pr_err("%s: bond_do_fail_over_mac impossible: bad policy %d\n",
1026 bond->dev->name, bond->params.fail_over_mac);
1032 static bool bond_should_change_active(struct bonding *bond)
1034 struct slave *prim = bond->primary_slave;
1035 struct slave *curr = bond->curr_active_slave;
1037 if (!prim || !curr || curr->link != BOND_LINK_UP)
1039 if (bond->force_primary) {
1040 bond->force_primary = false;
1043 if (bond->params.primary_reselect == BOND_PRI_RESELECT_BETTER &&
1044 (prim->speed < curr->speed ||
1045 (prim->speed == curr->speed && prim->duplex <= curr->duplex)))
1047 if (bond->params.primary_reselect == BOND_PRI_RESELECT_FAILURE)
1053 * find_best_interface - select the best available slave to be the active one
1054 * @bond: our bonding struct
1056 * Warning: Caller must hold curr_slave_lock for writing.
1058 static struct slave *bond_find_best_slave(struct bonding *bond)
1060 struct slave *new_active, *old_active;
1061 struct slave *bestslave = NULL;
1062 int mintime = bond->params.updelay;
1065 new_active = bond->curr_active_slave;
1067 if (!new_active) { /* there were no active slaves left */
1068 if (bond->slave_cnt > 0) /* found one slave */
1069 new_active = bond->first_slave;
1071 return NULL; /* still no slave, return NULL */
1074 if ((bond->primary_slave) &&
1075 bond->primary_slave->link == BOND_LINK_UP &&
1076 bond_should_change_active(bond)) {
1077 new_active = bond->primary_slave;
1080 /* remember where to stop iterating over the slaves */
1081 old_active = new_active;
1083 bond_for_each_slave_from(bond, new_active, i, old_active) {
1084 if (new_active->link == BOND_LINK_UP) {
1086 } else if (new_active->link == BOND_LINK_BACK &&
1087 IS_UP(new_active->dev)) {
1088 /* link up, but waiting for stabilization */
1089 if (new_active->delay < mintime) {
1090 mintime = new_active->delay;
1091 bestslave = new_active;
1100 * change_active_interface - change the active slave into the specified one
1101 * @bond: our bonding struct
1102 * @new: the new slave to make the active one
1104 * Set the new slave to the bond's settings and unset them on the old
1105 * curr_active_slave.
1106 * Setting include flags, mc-list, promiscuity, allmulti, etc.
1108 * If @new's link state is %BOND_LINK_BACK we'll set it to %BOND_LINK_UP,
1109 * because it is apparently the best available slave we have, even though its
1110 * updelay hasn't timed out yet.
1112 * If new_active is not NULL, caller must hold bond->lock for read and
1113 * curr_slave_lock for write_bh.
1115 void bond_change_active_slave(struct bonding *bond, struct slave *new_active)
1117 struct slave *old_active = bond->curr_active_slave;
1119 if (old_active == new_active)
1123 new_active->jiffies = jiffies;
1125 if (new_active->link == BOND_LINK_BACK) {
1126 if (USES_PRIMARY(bond->params.mode)) {
1127 pr_info("%s: making interface %s the new active one %d ms earlier.\n",
1128 bond->dev->name, new_active->dev->name,
1129 (bond->params.updelay - new_active->delay) * bond->params.miimon);
1132 new_active->delay = 0;
1133 new_active->link = BOND_LINK_UP;
1135 if (bond->params.mode == BOND_MODE_8023AD)
1136 bond_3ad_handle_link_change(new_active, BOND_LINK_UP);
1138 if (bond_is_lb(bond))
1139 bond_alb_handle_link_change(bond, new_active, BOND_LINK_UP);
1141 if (USES_PRIMARY(bond->params.mode)) {
1142 pr_info("%s: making interface %s the new active one.\n",
1143 bond->dev->name, new_active->dev->name);
1148 if (USES_PRIMARY(bond->params.mode))
1149 bond_mc_swap(bond, new_active, old_active);
1151 if (bond_is_lb(bond)) {
1152 bond_alb_handle_active_change(bond, new_active);
1154 bond_set_slave_inactive_flags(old_active);
1156 bond_set_slave_active_flags(new_active);
1158 bond->curr_active_slave = new_active;
1161 if (bond->params.mode == BOND_MODE_ACTIVEBACKUP) {
1163 bond_set_slave_inactive_flags(old_active);
1166 bond_set_slave_active_flags(new_active);
1168 if (bond->params.fail_over_mac)
1169 bond_do_fail_over_mac(bond, new_active,
1172 bond->send_grat_arp = bond->params.num_grat_arp;
1173 bond_send_gratuitous_arp(bond);
1175 bond->send_unsol_na = bond->params.num_unsol_na;
1176 bond_send_unsolicited_na(bond);
1178 write_unlock_bh(&bond->curr_slave_lock);
1179 read_unlock(&bond->lock);
1181 netdev_bonding_change(bond->dev, NETDEV_BONDING_FAILOVER);
1183 read_lock(&bond->lock);
1184 write_lock_bh(&bond->curr_slave_lock);
1188 /* resend IGMP joins since all were sent on curr_active_slave */
1189 if (bond->params.mode == BOND_MODE_ROUNDROBIN) {
1190 bond_resend_igmp_join_requests(bond);
1195 * bond_select_active_slave - select a new active slave, if needed
1196 * @bond: our bonding struct
1198 * This functions should be called when one of the following occurs:
1199 * - The old curr_active_slave has been released or lost its link.
1200 * - The primary_slave has got its link back.
1201 * - A slave has got its link back and there's no old curr_active_slave.
1203 * Caller must hold bond->lock for read and curr_slave_lock for write_bh.
1205 void bond_select_active_slave(struct bonding *bond)
1207 struct slave *best_slave;
1210 best_slave = bond_find_best_slave(bond);
1211 if (best_slave != bond->curr_active_slave) {
1212 bond_change_active_slave(bond, best_slave);
1213 rv = bond_set_carrier(bond);
1217 if (netif_carrier_ok(bond->dev)) {
1218 pr_info("%s: first active interface up!\n",
1221 pr_info("%s: now running without any active interface !\n",
1227 /*--------------------------- slave list handling ---------------------------*/
1230 * This function attaches the slave to the end of list.
1232 * bond->lock held for writing by caller.
1234 static void bond_attach_slave(struct bonding *bond, struct slave *new_slave)
1236 if (bond->first_slave == NULL) { /* attaching the first slave */
1237 new_slave->next = new_slave;
1238 new_slave->prev = new_slave;
1239 bond->first_slave = new_slave;
1241 new_slave->next = bond->first_slave;
1242 new_slave->prev = bond->first_slave->prev;
1243 new_slave->next->prev = new_slave;
1244 new_slave->prev->next = new_slave;
1251 * This function detaches the slave from the list.
1252 * WARNING: no check is made to verify if the slave effectively
1253 * belongs to <bond>.
1254 * Nothing is freed on return, structures are just unchained.
1255 * If any slave pointer in bond was pointing to <slave>,
1256 * it should be changed by the calling function.
1258 * bond->lock held for writing by caller.
1260 static void bond_detach_slave(struct bonding *bond, struct slave *slave)
1263 slave->next->prev = slave->prev;
1266 slave->prev->next = slave->next;
1268 if (bond->first_slave == slave) { /* slave is the first slave */
1269 if (bond->slave_cnt > 1) { /* there are more slave */
1270 bond->first_slave = slave->next;
1272 bond->first_slave = NULL; /* slave was the last one */
1281 #ifdef CONFIG_NET_POLL_CONTROLLER
1283 * You must hold read lock on bond->lock before calling this.
1285 static bool slaves_support_netpoll(struct net_device *bond_dev)
1287 struct bonding *bond = netdev_priv(bond_dev);
1288 struct slave *slave;
1292 bond_for_each_slave(bond, slave, i) {
1293 if ((slave->dev->priv_flags & IFF_DISABLE_NETPOLL) ||
1294 !slave->dev->netdev_ops->ndo_poll_controller)
1297 return i != 0 && ret;
1300 static void bond_poll_controller(struct net_device *bond_dev)
1302 struct net_device *dev = bond_dev->npinfo->netpoll->real_dev;
1303 if (dev != bond_dev)
1304 netpoll_poll_dev(dev);
1307 static void bond_netpoll_cleanup(struct net_device *bond_dev)
1309 struct bonding *bond = netdev_priv(bond_dev);
1310 struct slave *slave;
1311 const struct net_device_ops *ops;
1314 read_lock(&bond->lock);
1315 bond_dev->npinfo = NULL;
1316 bond_for_each_slave(bond, slave, i) {
1318 ops = slave->dev->netdev_ops;
1319 if (ops->ndo_netpoll_cleanup)
1320 ops->ndo_netpoll_cleanup(slave->dev);
1322 slave->dev->npinfo = NULL;
1325 read_unlock(&bond->lock);
1330 static void bond_netpoll_cleanup(struct net_device *bond_dev)
1336 /*---------------------------------- IOCTL ----------------------------------*/
1338 static int bond_sethwaddr(struct net_device *bond_dev,
1339 struct net_device *slave_dev)
1341 pr_debug("bond_dev=%p\n", bond_dev);
1342 pr_debug("slave_dev=%p\n", slave_dev);
1343 pr_debug("slave_dev->addr_len=%d\n", slave_dev->addr_len);
1344 memcpy(bond_dev->dev_addr, slave_dev->dev_addr, slave_dev->addr_len);
1348 #define BOND_VLAN_FEATURES \
1349 (NETIF_F_VLAN_CHALLENGED | NETIF_F_HW_VLAN_RX | NETIF_F_HW_VLAN_TX | \
1350 NETIF_F_HW_VLAN_FILTER)
1353 * Compute the common dev->feature set available to all slaves. Some
1354 * feature bits are managed elsewhere, so preserve those feature bits
1355 * on the master device.
1357 static int bond_compute_features(struct bonding *bond)
1359 struct slave *slave;
1360 struct net_device *bond_dev = bond->dev;
1361 unsigned long features = bond_dev->features;
1362 unsigned long vlan_features = 0;
1363 unsigned short max_hard_header_len = max((u16)ETH_HLEN,
1364 bond_dev->hard_header_len);
1367 features &= ~(NETIF_F_ALL_CSUM | BOND_VLAN_FEATURES);
1368 features |= NETIF_F_GSO_MASK | NETIF_F_NO_CSUM;
1370 if (!bond->first_slave)
1373 features &= ~NETIF_F_ONE_FOR_ALL;
1375 vlan_features = bond->first_slave->dev->vlan_features;
1376 bond_for_each_slave(bond, slave, i) {
1377 features = netdev_increment_features(features,
1378 slave->dev->features,
1379 NETIF_F_ONE_FOR_ALL);
1380 vlan_features = netdev_increment_features(vlan_features,
1381 slave->dev->vlan_features,
1382 NETIF_F_ONE_FOR_ALL);
1383 if (slave->dev->hard_header_len > max_hard_header_len)
1384 max_hard_header_len = slave->dev->hard_header_len;
1388 features |= (bond_dev->features & BOND_VLAN_FEATURES);
1389 bond_dev->features = netdev_fix_features(features, NULL);
1390 bond_dev->vlan_features = netdev_fix_features(vlan_features, NULL);
1391 bond_dev->hard_header_len = max_hard_header_len;
1396 static void bond_setup_by_slave(struct net_device *bond_dev,
1397 struct net_device *slave_dev)
1399 struct bonding *bond = netdev_priv(bond_dev);
1401 bond_dev->header_ops = slave_dev->header_ops;
1403 bond_dev->type = slave_dev->type;
1404 bond_dev->hard_header_len = slave_dev->hard_header_len;
1405 bond_dev->addr_len = slave_dev->addr_len;
1407 memcpy(bond_dev->broadcast, slave_dev->broadcast,
1408 slave_dev->addr_len);
1409 bond->setup_by_slave = 1;
1412 /* enslave device <slave> to bond device <master> */
1413 int bond_enslave(struct net_device *bond_dev, struct net_device *slave_dev)
1415 struct bonding *bond = netdev_priv(bond_dev);
1416 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
1417 struct slave *new_slave = NULL;
1418 struct netdev_hw_addr *ha;
1419 struct sockaddr addr;
1421 int old_features = bond_dev->features;
1424 if (!bond->params.use_carrier && slave_dev->ethtool_ops == NULL &&
1425 slave_ops->ndo_do_ioctl == NULL) {
1426 pr_warning("%s: Warning: no link monitoring support for %s\n",
1427 bond_dev->name, slave_dev->name);
1430 /* bond must be initialized by bond_open() before enslaving */
1431 if (!(bond_dev->flags & IFF_UP)) {
1432 pr_warning("%s: master_dev is not up in bond_enslave\n",
1436 /* already enslaved */
1437 if (slave_dev->flags & IFF_SLAVE) {
1438 pr_debug("Error, Device was already enslaved\n");
1442 /* vlan challenged mutual exclusion */
1443 /* no need to lock since we're protected by rtnl_lock */
1444 if (slave_dev->features & NETIF_F_VLAN_CHALLENGED) {
1445 pr_debug("%s: NETIF_F_VLAN_CHALLENGED\n", slave_dev->name);
1446 if (!list_empty(&bond->vlan_list)) {
1447 pr_err("%s: Error: cannot enslave VLAN challenged slave %s on VLAN enabled bond %s\n",
1448 bond_dev->name, slave_dev->name, bond_dev->name);
1451 pr_warning("%s: Warning: enslaved VLAN challenged slave %s. Adding VLANs will be blocked as long as %s is part of bond %s\n",
1452 bond_dev->name, slave_dev->name,
1453 slave_dev->name, bond_dev->name);
1454 bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
1457 pr_debug("%s: ! NETIF_F_VLAN_CHALLENGED\n", slave_dev->name);
1458 if (bond->slave_cnt == 0) {
1459 /* First slave, and it is not VLAN challenged,
1460 * so remove the block of adding VLANs over the bond.
1462 bond_dev->features &= ~NETIF_F_VLAN_CHALLENGED;
1467 * Old ifenslave binaries are no longer supported. These can
1468 * be identified with moderate accuracy by the state of the slave:
1469 * the current ifenslave will set the interface down prior to
1470 * enslaving it; the old ifenslave will not.
1472 if ((slave_dev->flags & IFF_UP)) {
1473 pr_err("%s is up. This may be due to an out of date ifenslave.\n",
1476 goto err_undo_flags;
1479 /* set bonding device ether type by slave - bonding netdevices are
1480 * created with ether_setup, so when the slave type is not ARPHRD_ETHER
1481 * there is a need to override some of the type dependent attribs/funcs.
1483 * bond ether type mutual exclusion - don't allow slaves of dissimilar
1484 * ether type (eg ARPHRD_ETHER and ARPHRD_INFINIBAND) share the same bond
1486 if (bond->slave_cnt == 0) {
1487 if (bond_dev->type != slave_dev->type) {
1488 pr_debug("%s: change device type from %d to %d\n",
1490 bond_dev->type, slave_dev->type);
1492 res = netdev_bonding_change(bond_dev,
1493 NETDEV_PRE_TYPE_CHANGE);
1494 res = notifier_to_errno(res);
1496 pr_err("%s: refused to change device type\n",
1499 goto err_undo_flags;
1502 /* Flush unicast and multicast addresses */
1503 dev_uc_flush(bond_dev);
1504 dev_mc_flush(bond_dev);
1506 if (slave_dev->type != ARPHRD_ETHER)
1507 bond_setup_by_slave(bond_dev, slave_dev);
1509 ether_setup(bond_dev);
1511 netdev_bonding_change(bond_dev,
1512 NETDEV_POST_TYPE_CHANGE);
1514 } else if (bond_dev->type != slave_dev->type) {
1515 pr_err("%s ether type (%d) is different from other slaves (%d), can not enslave it.\n",
1517 slave_dev->type, bond_dev->type);
1519 goto err_undo_flags;
1522 if (slave_ops->ndo_set_mac_address == NULL) {
1523 if (bond->slave_cnt == 0) {
1524 pr_warning("%s: Warning: The first slave device specified does not support setting the MAC address. Setting fail_over_mac to active.",
1526 bond->params.fail_over_mac = BOND_FOM_ACTIVE;
1527 } else if (bond->params.fail_over_mac != BOND_FOM_ACTIVE) {
1528 pr_err("%s: Error: The slave device specified does not support setting the MAC address, but fail_over_mac is not set to active.\n",
1531 goto err_undo_flags;
1535 /* If this is the first slave, then we need to set the master's hardware
1536 * address to be the same as the slave's. */
1537 if (bond->slave_cnt == 0)
1538 memcpy(bond->dev->dev_addr, slave_dev->dev_addr,
1539 slave_dev->addr_len);
1542 new_slave = kzalloc(sizeof(struct slave), GFP_KERNEL);
1545 goto err_undo_flags;
1549 * Set the new_slave's queue_id to be zero. Queue ID mapping
1550 * is set via sysfs or module option if desired.
1552 new_slave->queue_id = 0;
1554 /* Save slave's original mtu and then set it to match the bond */
1555 new_slave->original_mtu = slave_dev->mtu;
1556 res = dev_set_mtu(slave_dev, bond->dev->mtu);
1558 pr_debug("Error %d calling dev_set_mtu\n", res);
1563 * Save slave's original ("permanent") mac address for modes
1564 * that need it, and for restoring it upon release, and then
1565 * set it to the master's address
1567 memcpy(new_slave->perm_hwaddr, slave_dev->dev_addr, ETH_ALEN);
1569 if (!bond->params.fail_over_mac) {
1571 * Set slave to master's mac address. The application already
1572 * set the master's mac address to that of the first slave
1574 memcpy(addr.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
1575 addr.sa_family = slave_dev->type;
1576 res = dev_set_mac_address(slave_dev, &addr);
1578 pr_debug("Error %d calling set_mac_address\n", res);
1579 goto err_restore_mtu;
1583 res = netdev_set_master(slave_dev, bond_dev);
1585 pr_debug("Error %d calling netdev_set_master\n", res);
1586 goto err_restore_mac;
1588 /* open the slave since the application closed it */
1589 res = dev_open(slave_dev);
1591 pr_debug("Opening slave %s failed\n", slave_dev->name);
1592 goto err_unset_master;
1595 new_slave->dev = slave_dev;
1596 slave_dev->priv_flags |= IFF_BONDING;
1598 if (bond_is_lb(bond)) {
1599 /* bond_alb_init_slave() must be called before all other stages since
1600 * it might fail and we do not want to have to undo everything
1602 res = bond_alb_init_slave(bond, new_slave);
1607 /* If the mode USES_PRIMARY, then the new slave gets the
1608 * master's promisc (and mc) settings only if it becomes the
1609 * curr_active_slave, and that is taken care of later when calling
1610 * bond_change_active()
1612 if (!USES_PRIMARY(bond->params.mode)) {
1613 /* set promiscuity level to new slave */
1614 if (bond_dev->flags & IFF_PROMISC) {
1615 res = dev_set_promiscuity(slave_dev, 1);
1620 /* set allmulti level to new slave */
1621 if (bond_dev->flags & IFF_ALLMULTI) {
1622 res = dev_set_allmulti(slave_dev, 1);
1627 netif_addr_lock_bh(bond_dev);
1628 /* upload master's mc_list to new slave */
1629 netdev_for_each_mc_addr(ha, bond_dev)
1630 dev_mc_add(slave_dev, ha->addr);
1631 netif_addr_unlock_bh(bond_dev);
1634 if (bond->params.mode == BOND_MODE_8023AD) {
1635 /* add lacpdu mc addr to mc list */
1636 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
1638 dev_mc_add(slave_dev, lacpdu_multicast);
1641 bond_add_vlans_on_slave(bond, slave_dev);
1643 write_lock_bh(&bond->lock);
1645 bond_attach_slave(bond, new_slave);
1647 new_slave->delay = 0;
1648 new_slave->link_failure_count = 0;
1650 bond_compute_features(bond);
1652 write_unlock_bh(&bond->lock);
1654 read_lock(&bond->lock);
1656 new_slave->last_arp_rx = jiffies;
1658 if (bond->params.miimon && !bond->params.use_carrier) {
1659 link_reporting = bond_check_dev_link(bond, slave_dev, 1);
1661 if ((link_reporting == -1) && !bond->params.arp_interval) {
1663 * miimon is set but a bonded network driver
1664 * does not support ETHTOOL/MII and
1665 * arp_interval is not set. Note: if
1666 * use_carrier is enabled, we will never go
1667 * here (because netif_carrier is always
1668 * supported); thus, we don't need to change
1669 * the messages for netif_carrier.
1671 pr_warning("%s: Warning: MII and ETHTOOL support not available for interface %s, and arp_interval/arp_ip_target module parameters not specified, thus bonding will not detect link failures! see bonding.txt for details.\n",
1672 bond_dev->name, slave_dev->name);
1673 } else if (link_reporting == -1) {
1674 /* unable get link status using mii/ethtool */
1675 pr_warning("%s: Warning: can't get link status from interface %s; the network driver associated with this interface does not support MII or ETHTOOL link status reporting, thus miimon has no effect on this interface.\n",
1676 bond_dev->name, slave_dev->name);
1680 /* check for initial state */
1681 if (!bond->params.miimon ||
1682 (bond_check_dev_link(bond, slave_dev, 0) == BMSR_LSTATUS)) {
1683 if (bond->params.updelay) {
1684 pr_debug("Initial state of slave_dev is BOND_LINK_BACK\n");
1685 new_slave->link = BOND_LINK_BACK;
1686 new_slave->delay = bond->params.updelay;
1688 pr_debug("Initial state of slave_dev is BOND_LINK_UP\n");
1689 new_slave->link = BOND_LINK_UP;
1691 new_slave->jiffies = jiffies;
1693 pr_debug("Initial state of slave_dev is BOND_LINK_DOWN\n");
1694 new_slave->link = BOND_LINK_DOWN;
1697 if (bond_update_speed_duplex(new_slave) &&
1698 (new_slave->link != BOND_LINK_DOWN)) {
1699 pr_warning("%s: Warning: failed to get speed and duplex from %s, assumed to be 100Mb/sec and Full.\n",
1700 bond_dev->name, new_slave->dev->name);
1702 if (bond->params.mode == BOND_MODE_8023AD) {
1703 pr_warning("%s: Warning: Operation of 802.3ad mode requires ETHTOOL support in base driver for proper aggregator selection.\n",
1708 if (USES_PRIMARY(bond->params.mode) && bond->params.primary[0]) {
1709 /* if there is a primary slave, remember it */
1710 if (strcmp(bond->params.primary, new_slave->dev->name) == 0) {
1711 bond->primary_slave = new_slave;
1712 bond->force_primary = true;
1716 write_lock_bh(&bond->curr_slave_lock);
1718 switch (bond->params.mode) {
1719 case BOND_MODE_ACTIVEBACKUP:
1720 bond_set_slave_inactive_flags(new_slave);
1721 bond_select_active_slave(bond);
1723 case BOND_MODE_8023AD:
1724 /* in 802.3ad mode, the internal mechanism
1725 * will activate the slaves in the selected
1728 bond_set_slave_inactive_flags(new_slave);
1729 /* if this is the first slave */
1730 if (bond->slave_cnt == 1) {
1731 SLAVE_AD_INFO(new_slave).id = 1;
1732 /* Initialize AD with the number of times that the AD timer is called in 1 second
1733 * can be called only after the mac address of the bond is set
1735 bond_3ad_initialize(bond, 1000/AD_TIMER_INTERVAL,
1736 bond->params.lacp_fast);
1738 SLAVE_AD_INFO(new_slave).id =
1739 SLAVE_AD_INFO(new_slave->prev).id + 1;
1742 bond_3ad_bind_slave(new_slave);
1746 new_slave->state = BOND_STATE_ACTIVE;
1747 bond_set_slave_inactive_flags(new_slave);
1748 bond_select_active_slave(bond);
1751 pr_debug("This slave is always active in trunk mode\n");
1753 /* always active in trunk mode */
1754 new_slave->state = BOND_STATE_ACTIVE;
1756 /* In trunking mode there is little meaning to curr_active_slave
1757 * anyway (it holds no special properties of the bond device),
1758 * so we can change it without calling change_active_interface()
1760 if (!bond->curr_active_slave)
1761 bond->curr_active_slave = new_slave;
1764 } /* switch(bond_mode) */
1766 write_unlock_bh(&bond->curr_slave_lock);
1768 bond_set_carrier(bond);
1770 #ifdef CONFIG_NET_POLL_CONTROLLER
1772 * Netpoll and bonding is broken, make sure it is not initialized
1773 * until it is fixed.
1775 if (disable_netpoll) {
1776 bond_dev->priv_flags |= IFF_DISABLE_NETPOLL;
1778 if (slaves_support_netpoll(bond_dev)) {
1779 bond_dev->priv_flags &= ~IFF_DISABLE_NETPOLL;
1780 if (bond_dev->npinfo)
1781 slave_dev->npinfo = bond_dev->npinfo;
1782 } else if (!(bond_dev->priv_flags & IFF_DISABLE_NETPOLL)) {
1783 bond_dev->priv_flags |= IFF_DISABLE_NETPOLL;
1784 pr_info("New slave device %s does not support netpoll\n",
1786 pr_info("Disabling netpoll support for %s\n", bond_dev->name);
1790 read_unlock(&bond->lock);
1792 res = bond_create_slave_symlinks(bond_dev, slave_dev);
1796 pr_info("%s: enslaving %s as a%s interface with a%s link.\n",
1797 bond_dev->name, slave_dev->name,
1798 new_slave->state == BOND_STATE_ACTIVE ? "n active" : " backup",
1799 new_slave->link != BOND_LINK_DOWN ? "n up" : " down");
1801 /* enslave is successful */
1804 /* Undo stages on error */
1806 dev_close(slave_dev);
1809 netdev_set_master(slave_dev, NULL);
1812 if (!bond->params.fail_over_mac) {
1813 /* XXX TODO - fom follow mode needs to change master's
1814 * MAC if this slave's MAC is in use by the bond, or at
1815 * least print a warning.
1817 memcpy(addr.sa_data, new_slave->perm_hwaddr, ETH_ALEN);
1818 addr.sa_family = slave_dev->type;
1819 dev_set_mac_address(slave_dev, &addr);
1823 dev_set_mtu(slave_dev, new_slave->original_mtu);
1829 bond_dev->features = old_features;
1835 * Try to release the slave device <slave> from the bond device <master>
1836 * It is legal to access curr_active_slave without a lock because all the function
1839 * The rules for slave state should be:
1840 * for Active/Backup:
1841 * Active stays on all backups go down
1842 * for Bonded connections:
1843 * The first up interface should be left on and all others downed.
1845 int bond_release(struct net_device *bond_dev, struct net_device *slave_dev)
1847 struct bonding *bond = netdev_priv(bond_dev);
1848 struct slave *slave, *oldcurrent;
1849 struct sockaddr addr;
1851 /* slave is not a slave or master is not master of this slave */
1852 if (!(slave_dev->flags & IFF_SLAVE) ||
1853 (slave_dev->master != bond_dev)) {
1854 pr_err("%s: Error: cannot release %s.\n",
1855 bond_dev->name, slave_dev->name);
1859 netdev_bonding_change(bond_dev, NETDEV_BONDING_DESLAVE);
1860 write_lock_bh(&bond->lock);
1862 slave = bond_get_slave_by_dev(bond, slave_dev);
1864 /* not a slave of this bond */
1865 pr_info("%s: %s not enslaved\n",
1866 bond_dev->name, slave_dev->name);
1867 write_unlock_bh(&bond->lock);
1871 if (!bond->params.fail_over_mac) {
1872 if (!compare_ether_addr(bond_dev->dev_addr, slave->perm_hwaddr) &&
1873 bond->slave_cnt > 1)
1874 pr_warning("%s: Warning: the permanent HWaddr of %s - %pM - is still in use by %s. Set the HWaddr of %s to a different address to avoid conflicts.\n",
1875 bond_dev->name, slave_dev->name,
1877 bond_dev->name, slave_dev->name);
1880 /* Inform AD package of unbinding of slave. */
1881 if (bond->params.mode == BOND_MODE_8023AD) {
1882 /* must be called before the slave is
1883 * detached from the list
1885 bond_3ad_unbind_slave(slave);
1888 pr_info("%s: releasing %s interface %s\n",
1890 (slave->state == BOND_STATE_ACTIVE) ? "active" : "backup",
1893 oldcurrent = bond->curr_active_slave;
1895 bond->current_arp_slave = NULL;
1897 /* release the slave from its bond */
1898 bond_detach_slave(bond, slave);
1900 bond_compute_features(bond);
1902 if (bond->primary_slave == slave)
1903 bond->primary_slave = NULL;
1905 if (oldcurrent == slave)
1906 bond_change_active_slave(bond, NULL);
1908 if (bond_is_lb(bond)) {
1909 /* Must be called only after the slave has been
1910 * detached from the list and the curr_active_slave
1911 * has been cleared (if our_slave == old_current),
1912 * but before a new active slave is selected.
1914 write_unlock_bh(&bond->lock);
1915 bond_alb_deinit_slave(bond, slave);
1916 write_lock_bh(&bond->lock);
1919 if (oldcurrent == slave) {
1921 * Note that we hold RTNL over this sequence, so there
1922 * is no concern that another slave add/remove event
1925 write_unlock_bh(&bond->lock);
1926 read_lock(&bond->lock);
1927 write_lock_bh(&bond->curr_slave_lock);
1929 bond_select_active_slave(bond);
1931 write_unlock_bh(&bond->curr_slave_lock);
1932 read_unlock(&bond->lock);
1933 write_lock_bh(&bond->lock);
1936 if (bond->slave_cnt == 0) {
1937 bond_set_carrier(bond);
1939 /* if the last slave was removed, zero the mac address
1940 * of the master so it will be set by the application
1941 * to the mac address of the first slave
1943 memset(bond_dev->dev_addr, 0, bond_dev->addr_len);
1945 if (list_empty(&bond->vlan_list)) {
1946 bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
1948 pr_warning("%s: Warning: clearing HW address of %s while it still has VLANs.\n",
1949 bond_dev->name, bond_dev->name);
1950 pr_warning("%s: When re-adding slaves, make sure the bond's HW address matches its VLANs'.\n",
1953 } else if ((bond_dev->features & NETIF_F_VLAN_CHALLENGED) &&
1954 !bond_has_challenged_slaves(bond)) {
1955 pr_info("%s: last VLAN challenged slave %s left bond %s. VLAN blocking is removed\n",
1956 bond_dev->name, slave_dev->name, bond_dev->name);
1957 bond_dev->features &= ~NETIF_F_VLAN_CHALLENGED;
1960 write_unlock_bh(&bond->lock);
1962 /* must do this from outside any spinlocks */
1963 bond_destroy_slave_symlinks(bond_dev, slave_dev);
1965 bond_del_vlans_from_slave(bond, slave_dev);
1967 /* If the mode USES_PRIMARY, then we should only remove its
1968 * promisc and mc settings if it was the curr_active_slave, but that was
1969 * already taken care of above when we detached the slave
1971 if (!USES_PRIMARY(bond->params.mode)) {
1972 /* unset promiscuity level from slave */
1973 if (bond_dev->flags & IFF_PROMISC)
1974 dev_set_promiscuity(slave_dev, -1);
1976 /* unset allmulti level from slave */
1977 if (bond_dev->flags & IFF_ALLMULTI)
1978 dev_set_allmulti(slave_dev, -1);
1980 /* flush master's mc_list from slave */
1981 netif_addr_lock_bh(bond_dev);
1982 bond_mc_list_flush(bond_dev, slave_dev);
1983 netif_addr_unlock_bh(bond_dev);
1986 netdev_set_master(slave_dev, NULL);
1988 #ifdef CONFIG_NET_POLL_CONTROLLER
1989 read_lock_bh(&bond->lock);
1991 /* Make sure netpoll over stays disabled until fixed. */
1992 if (!disable_netpoll)
1993 if (slaves_support_netpoll(bond_dev))
1994 bond_dev->priv_flags &= ~IFF_DISABLE_NETPOLL;
1995 read_unlock_bh(&bond->lock);
1996 if (slave_dev->netdev_ops->ndo_netpoll_cleanup)
1997 slave_dev->netdev_ops->ndo_netpoll_cleanup(slave_dev);
1999 slave_dev->npinfo = NULL;
2002 /* close slave before restoring its mac address */
2003 dev_close(slave_dev);
2005 if (bond->params.fail_over_mac != BOND_FOM_ACTIVE) {
2006 /* restore original ("permanent") mac address */
2007 memcpy(addr.sa_data, slave->perm_hwaddr, ETH_ALEN);
2008 addr.sa_family = slave_dev->type;
2009 dev_set_mac_address(slave_dev, &addr);
2012 dev_set_mtu(slave_dev, slave->original_mtu);
2014 slave_dev->priv_flags &= ~(IFF_MASTER_8023AD | IFF_MASTER_ALB |
2015 IFF_SLAVE_INACTIVE | IFF_BONDING |
2020 return 0; /* deletion OK */
2024 * First release a slave and than destroy the bond if no more slaves are left.
2025 * Must be under rtnl_lock when this function is called.
2027 int bond_release_and_destroy(struct net_device *bond_dev,
2028 struct net_device *slave_dev)
2030 struct bonding *bond = netdev_priv(bond_dev);
2033 ret = bond_release(bond_dev, slave_dev);
2034 if ((ret == 0) && (bond->slave_cnt == 0)) {
2035 pr_info("%s: destroying bond %s.\n",
2036 bond_dev->name, bond_dev->name);
2037 unregister_netdevice(bond_dev);
2043 * This function releases all slaves.
2045 static int bond_release_all(struct net_device *bond_dev)
2047 struct bonding *bond = netdev_priv(bond_dev);
2048 struct slave *slave;
2049 struct net_device *slave_dev;
2050 struct sockaddr addr;
2052 write_lock_bh(&bond->lock);
2054 netif_carrier_off(bond_dev);
2056 if (bond->slave_cnt == 0)
2059 bond->current_arp_slave = NULL;
2060 bond->primary_slave = NULL;
2061 bond_change_active_slave(bond, NULL);
2063 while ((slave = bond->first_slave) != NULL) {
2064 /* Inform AD package of unbinding of slave
2065 * before slave is detached from the list.
2067 if (bond->params.mode == BOND_MODE_8023AD)
2068 bond_3ad_unbind_slave(slave);
2070 slave_dev = slave->dev;
2071 bond_detach_slave(bond, slave);
2073 /* now that the slave is detached, unlock and perform
2074 * all the undo steps that should not be called from
2077 write_unlock_bh(&bond->lock);
2079 if (bond_is_lb(bond)) {
2080 /* must be called only after the slave
2081 * has been detached from the list
2083 bond_alb_deinit_slave(bond, slave);
2086 bond_compute_features(bond);
2088 bond_destroy_slave_symlinks(bond_dev, slave_dev);
2089 bond_del_vlans_from_slave(bond, slave_dev);
2091 /* If the mode USES_PRIMARY, then we should only remove its
2092 * promisc and mc settings if it was the curr_active_slave, but that was
2093 * already taken care of above when we detached the slave
2095 if (!USES_PRIMARY(bond->params.mode)) {
2096 /* unset promiscuity level from slave */
2097 if (bond_dev->flags & IFF_PROMISC)
2098 dev_set_promiscuity(slave_dev, -1);
2100 /* unset allmulti level from slave */
2101 if (bond_dev->flags & IFF_ALLMULTI)
2102 dev_set_allmulti(slave_dev, -1);
2104 /* flush master's mc_list from slave */
2105 netif_addr_lock_bh(bond_dev);
2106 bond_mc_list_flush(bond_dev, slave_dev);
2107 netif_addr_unlock_bh(bond_dev);
2110 netdev_set_master(slave_dev, NULL);
2112 /* close slave before restoring its mac address */
2113 dev_close(slave_dev);
2115 if (!bond->params.fail_over_mac) {
2116 /* restore original ("permanent") mac address*/
2117 memcpy(addr.sa_data, slave->perm_hwaddr, ETH_ALEN);
2118 addr.sa_family = slave_dev->type;
2119 dev_set_mac_address(slave_dev, &addr);
2122 slave_dev->priv_flags &= ~(IFF_MASTER_8023AD | IFF_MASTER_ALB |
2123 IFF_SLAVE_INACTIVE);
2127 /* re-acquire the lock before getting the next slave */
2128 write_lock_bh(&bond->lock);
2131 /* zero the mac address of the master so it will be
2132 * set by the application to the mac address of the
2135 memset(bond_dev->dev_addr, 0, bond_dev->addr_len);
2137 if (list_empty(&bond->vlan_list))
2138 bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
2140 pr_warning("%s: Warning: clearing HW address of %s while it still has VLANs.\n",
2141 bond_dev->name, bond_dev->name);
2142 pr_warning("%s: When re-adding slaves, make sure the bond's HW address matches its VLANs'.\n",
2146 pr_info("%s: released all slaves\n", bond_dev->name);
2149 write_unlock_bh(&bond->lock);
2155 * This function changes the active slave to slave <slave_dev>.
2156 * It returns -EINVAL in the following cases.
2157 * - <slave_dev> is not found in the list.
2158 * - There is not active slave now.
2159 * - <slave_dev> is already active.
2160 * - The link state of <slave_dev> is not BOND_LINK_UP.
2161 * - <slave_dev> is not running.
2162 * In these cases, this function does nothing.
2163 * In the other cases, current_slave pointer is changed and 0 is returned.
2165 static int bond_ioctl_change_active(struct net_device *bond_dev, struct net_device *slave_dev)
2167 struct bonding *bond = netdev_priv(bond_dev);
2168 struct slave *old_active = NULL;
2169 struct slave *new_active = NULL;
2172 if (!USES_PRIMARY(bond->params.mode))
2175 /* Verify that master_dev is indeed the master of slave_dev */
2176 if (!(slave_dev->flags & IFF_SLAVE) || (slave_dev->master != bond_dev))
2179 read_lock(&bond->lock);
2181 read_lock(&bond->curr_slave_lock);
2182 old_active = bond->curr_active_slave;
2183 read_unlock(&bond->curr_slave_lock);
2185 new_active = bond_get_slave_by_dev(bond, slave_dev);
2188 * Changing to the current active: do nothing; return success.
2190 if (new_active && (new_active == old_active)) {
2191 read_unlock(&bond->lock);
2197 (new_active->link == BOND_LINK_UP) &&
2198 IS_UP(new_active->dev)) {
2199 write_lock_bh(&bond->curr_slave_lock);
2200 bond_change_active_slave(bond, new_active);
2201 write_unlock_bh(&bond->curr_slave_lock);
2205 read_unlock(&bond->lock);
2210 static int bond_info_query(struct net_device *bond_dev, struct ifbond *info)
2212 struct bonding *bond = netdev_priv(bond_dev);
2214 info->bond_mode = bond->params.mode;
2215 info->miimon = bond->params.miimon;
2217 read_lock(&bond->lock);
2218 info->num_slaves = bond->slave_cnt;
2219 read_unlock(&bond->lock);
2224 static int bond_slave_info_query(struct net_device *bond_dev, struct ifslave *info)
2226 struct bonding *bond = netdev_priv(bond_dev);
2227 struct slave *slave;
2228 int i, res = -ENODEV;
2230 read_lock(&bond->lock);
2232 bond_for_each_slave(bond, slave, i) {
2233 if (i == (int)info->slave_id) {
2235 strcpy(info->slave_name, slave->dev->name);
2236 info->link = slave->link;
2237 info->state = slave->state;
2238 info->link_failure_count = slave->link_failure_count;
2243 read_unlock(&bond->lock);
2248 /*-------------------------------- Monitoring -------------------------------*/
2251 static int bond_miimon_inspect(struct bonding *bond)
2253 struct slave *slave;
2254 int i, link_state, commit = 0;
2255 bool ignore_updelay;
2257 ignore_updelay = !bond->curr_active_slave ? true : false;
2259 bond_for_each_slave(bond, slave, i) {
2260 slave->new_link = BOND_LINK_NOCHANGE;
2262 link_state = bond_check_dev_link(bond, slave->dev, 0);
2264 switch (slave->link) {
2269 slave->link = BOND_LINK_FAIL;
2270 slave->delay = bond->params.downdelay;
2272 pr_info("%s: link status down for %sinterface %s, disabling it in %d ms.\n",
2274 (bond->params.mode ==
2275 BOND_MODE_ACTIVEBACKUP) ?
2276 ((slave->state == BOND_STATE_ACTIVE) ?
2277 "active " : "backup ") : "",
2279 bond->params.downdelay * bond->params.miimon);
2282 case BOND_LINK_FAIL:
2285 * recovered before downdelay expired
2287 slave->link = BOND_LINK_UP;
2288 slave->jiffies = jiffies;
2289 pr_info("%s: link status up again after %d ms for interface %s.\n",
2291 (bond->params.downdelay - slave->delay) *
2292 bond->params.miimon,
2297 if (slave->delay <= 0) {
2298 slave->new_link = BOND_LINK_DOWN;
2306 case BOND_LINK_DOWN:
2310 slave->link = BOND_LINK_BACK;
2311 slave->delay = bond->params.updelay;
2314 pr_info("%s: link status up for interface %s, enabling it in %d ms.\n",
2315 bond->dev->name, slave->dev->name,
2316 ignore_updelay ? 0 :
2317 bond->params.updelay *
2318 bond->params.miimon);
2321 case BOND_LINK_BACK:
2323 slave->link = BOND_LINK_DOWN;
2324 pr_info("%s: link status down again after %d ms for interface %s.\n",
2326 (bond->params.updelay - slave->delay) *
2327 bond->params.miimon,
2336 if (slave->delay <= 0) {
2337 slave->new_link = BOND_LINK_UP;
2339 ignore_updelay = false;
2351 static void bond_miimon_commit(struct bonding *bond)
2353 struct slave *slave;
2356 bond_for_each_slave(bond, slave, i) {
2357 switch (slave->new_link) {
2358 case BOND_LINK_NOCHANGE:
2362 slave->link = BOND_LINK_UP;
2363 slave->jiffies = jiffies;
2365 if (bond->params.mode == BOND_MODE_8023AD) {
2366 /* prevent it from being the active one */
2367 slave->state = BOND_STATE_BACKUP;
2368 } else if (bond->params.mode != BOND_MODE_ACTIVEBACKUP) {
2369 /* make it immediately active */
2370 slave->state = BOND_STATE_ACTIVE;
2371 } else if (slave != bond->primary_slave) {
2372 /* prevent it from being the active one */
2373 slave->state = BOND_STATE_BACKUP;
2376 pr_info("%s: link status definitely up for interface %s.\n",
2377 bond->dev->name, slave->dev->name);
2379 /* notify ad that the link status has changed */
2380 if (bond->params.mode == BOND_MODE_8023AD)
2381 bond_3ad_handle_link_change(slave, BOND_LINK_UP);
2383 if (bond_is_lb(bond))
2384 bond_alb_handle_link_change(bond, slave,
2387 if (!bond->curr_active_slave ||
2388 (slave == bond->primary_slave))
2393 case BOND_LINK_DOWN:
2394 if (slave->link_failure_count < UINT_MAX)
2395 slave->link_failure_count++;
2397 slave->link = BOND_LINK_DOWN;
2399 if (bond->params.mode == BOND_MODE_ACTIVEBACKUP ||
2400 bond->params.mode == BOND_MODE_8023AD)
2401 bond_set_slave_inactive_flags(slave);
2403 pr_info("%s: link status definitely down for interface %s, disabling it\n",
2404 bond->dev->name, slave->dev->name);
2406 if (bond->params.mode == BOND_MODE_8023AD)
2407 bond_3ad_handle_link_change(slave,
2410 if (bond_is_lb(bond))
2411 bond_alb_handle_link_change(bond, slave,
2414 if (slave == bond->curr_active_slave)
2420 pr_err("%s: invalid new link %d on slave %s\n",
2421 bond->dev->name, slave->new_link,
2423 slave->new_link = BOND_LINK_NOCHANGE;
2430 write_lock_bh(&bond->curr_slave_lock);
2431 bond_select_active_slave(bond);
2432 write_unlock_bh(&bond->curr_slave_lock);
2435 bond_set_carrier(bond);
2441 * Really a wrapper that splits the mii monitor into two phases: an
2442 * inspection, then (if inspection indicates something needs to be done)
2443 * an acquisition of appropriate locks followed by a commit phase to
2444 * implement whatever link state changes are indicated.
2446 void bond_mii_monitor(struct work_struct *work)
2448 struct bonding *bond = container_of(work, struct bonding,
2451 read_lock(&bond->lock);
2452 if (bond->kill_timers)
2455 if (bond->slave_cnt == 0)
2458 if (bond->send_grat_arp) {
2459 read_lock(&bond->curr_slave_lock);
2460 bond_send_gratuitous_arp(bond);
2461 read_unlock(&bond->curr_slave_lock);
2464 if (bond->send_unsol_na) {
2465 read_lock(&bond->curr_slave_lock);
2466 bond_send_unsolicited_na(bond);
2467 read_unlock(&bond->curr_slave_lock);
2470 if (bond_miimon_inspect(bond)) {
2471 read_unlock(&bond->lock);
2473 read_lock(&bond->lock);
2475 bond_miimon_commit(bond);
2477 read_unlock(&bond->lock);
2478 rtnl_unlock(); /* might sleep, hold no other locks */
2479 read_lock(&bond->lock);
2483 if (bond->params.miimon)
2484 queue_delayed_work(bond->wq, &bond->mii_work,
2485 msecs_to_jiffies(bond->params.miimon));
2487 read_unlock(&bond->lock);
2490 static __be32 bond_glean_dev_ip(struct net_device *dev)
2492 struct in_device *idev;
2493 struct in_ifaddr *ifa;
2500 idev = __in_dev_get_rcu(dev);
2504 ifa = idev->ifa_list;
2508 addr = ifa->ifa_local;
2514 static int bond_has_this_ip(struct bonding *bond, __be32 ip)
2516 struct vlan_entry *vlan;
2518 if (ip == bond->master_ip)
2521 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
2522 if (ip == vlan->vlan_ip)
2530 * We go to the (large) trouble of VLAN tagging ARP frames because
2531 * switches in VLAN mode (especially if ports are configured as
2532 * "native" to a VLAN) might not pass non-tagged frames.
2534 static void bond_arp_send(struct net_device *slave_dev, int arp_op, __be32 dest_ip, __be32 src_ip, unsigned short vlan_id)
2536 struct sk_buff *skb;
2538 pr_debug("arp %d on slave %s: dst %x src %x vid %d\n", arp_op,
2539 slave_dev->name, dest_ip, src_ip, vlan_id);
2541 skb = arp_create(arp_op, ETH_P_ARP, dest_ip, slave_dev, src_ip,
2542 NULL, slave_dev->dev_addr, NULL);
2545 pr_err("ARP packet allocation failed\n");
2549 skb = vlan_put_tag(skb, vlan_id);
2551 pr_err("failed to insert VLAN tag\n");
2559 static void bond_arp_send_all(struct bonding *bond, struct slave *slave)
2562 __be32 *targets = bond->params.arp_targets;
2563 struct vlan_entry *vlan;
2564 struct net_device *vlan_dev;
2568 for (i = 0; (i < BOND_MAX_ARP_TARGETS); i++) {
2571 pr_debug("basa: target %x\n", targets[i]);
2572 if (list_empty(&bond->vlan_list)) {
2573 pr_debug("basa: empty vlan: arp_send\n");
2574 bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2575 bond->master_ip, 0);
2580 * If VLANs are configured, we do a route lookup to
2581 * determine which VLAN interface would be used, so we
2582 * can tag the ARP with the proper VLAN tag.
2584 memset(&fl, 0, sizeof(fl));
2585 fl.fl4_dst = targets[i];
2586 fl.fl4_tos = RTO_ONLINK;
2588 rv = ip_route_output_key(dev_net(bond->dev), &rt, &fl);
2590 if (net_ratelimit()) {
2591 pr_warning("%s: no route to arp_ip_target %pI4\n",
2592 bond->dev->name, &fl.fl4_dst);
2598 * This target is not on a VLAN
2600 if (rt->dst.dev == bond->dev) {
2602 pr_debug("basa: rtdev == bond->dev: arp_send\n");
2603 bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2604 bond->master_ip, 0);
2609 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
2610 vlan_dev = vlan_group_get_device(bond->vlgrp, vlan->vlan_id);
2611 if (vlan_dev == rt->dst.dev) {
2612 vlan_id = vlan->vlan_id;
2613 pr_debug("basa: vlan match on %s %d\n",
2614 vlan_dev->name, vlan_id);
2621 bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2622 vlan->vlan_ip, vlan_id);
2626 if (net_ratelimit()) {
2627 pr_warning("%s: no path to arp_ip_target %pI4 via rt.dev %s\n",
2628 bond->dev->name, &fl.fl4_dst,
2629 rt->dst.dev ? rt->dst.dev->name : "NULL");
2636 * Kick out a gratuitous ARP for an IP on the bonding master plus one
2637 * for each VLAN above us.
2639 * Caller must hold curr_slave_lock for read or better
2641 static void bond_send_gratuitous_arp(struct bonding *bond)
2643 struct slave *slave = bond->curr_active_slave;
2644 struct vlan_entry *vlan;
2645 struct net_device *vlan_dev;
2647 pr_debug("bond_send_grat_arp: bond %s slave %s\n",
2648 bond->dev->name, slave ? slave->dev->name : "NULL");
2650 if (!slave || !bond->send_grat_arp ||
2651 test_bit(__LINK_STATE_LINKWATCH_PENDING, &slave->dev->state))
2654 bond->send_grat_arp--;
2656 if (bond->master_ip) {
2657 bond_arp_send(slave->dev, ARPOP_REPLY, bond->master_ip,
2658 bond->master_ip, 0);
2661 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
2662 vlan_dev = vlan_group_get_device(bond->vlgrp, vlan->vlan_id);
2663 if (vlan->vlan_ip) {
2664 bond_arp_send(slave->dev, ARPOP_REPLY, vlan->vlan_ip,
2665 vlan->vlan_ip, vlan->vlan_id);
2670 static void bond_validate_arp(struct bonding *bond, struct slave *slave, __be32 sip, __be32 tip)
2673 __be32 *targets = bond->params.arp_targets;
2675 for (i = 0; (i < BOND_MAX_ARP_TARGETS) && targets[i]; i++) {
2676 pr_debug("bva: sip %pI4 tip %pI4 t[%d] %pI4 bhti(tip) %d\n",
2677 &sip, &tip, i, &targets[i],
2678 bond_has_this_ip(bond, tip));
2679 if (sip == targets[i]) {
2680 if (bond_has_this_ip(bond, tip))
2681 slave->last_arp_rx = jiffies;
2687 static int bond_arp_rcv(struct sk_buff *skb, struct net_device *dev, struct packet_type *pt, struct net_device *orig_dev)
2690 struct slave *slave;
2691 struct bonding *bond;
2692 unsigned char *arp_ptr;
2695 if (dev->priv_flags & IFF_802_1Q_VLAN) {
2697 * When using VLANS and bonding, dev and oriv_dev may be
2698 * incorrect if the physical interface supports VLAN
2699 * acceleration. With this change ARP validation now
2700 * works for hosts only reachable on the VLAN interface.
2702 dev = vlan_dev_real_dev(dev);
2703 orig_dev = dev_get_by_index_rcu(dev_net(skb->dev),skb->skb_iif);
2706 if (!(dev->priv_flags & IFF_BONDING) || !(dev->flags & IFF_MASTER))
2709 bond = netdev_priv(dev);
2710 read_lock(&bond->lock);
2712 pr_debug("bond_arp_rcv: bond %s skb->dev %s orig_dev %s\n",
2713 bond->dev->name, skb->dev ? skb->dev->name : "NULL",
2714 orig_dev ? orig_dev->name : "NULL");
2716 slave = bond_get_slave_by_dev(bond, orig_dev);
2717 if (!slave || !slave_do_arp_validate(bond, slave))
2720 if (!pskb_may_pull(skb, arp_hdr_len(dev)))
2724 if (arp->ar_hln != dev->addr_len ||
2725 skb->pkt_type == PACKET_OTHERHOST ||
2726 skb->pkt_type == PACKET_LOOPBACK ||
2727 arp->ar_hrd != htons(ARPHRD_ETHER) ||
2728 arp->ar_pro != htons(ETH_P_IP) ||
2732 arp_ptr = (unsigned char *)(arp + 1);
2733 arp_ptr += dev->addr_len;
2734 memcpy(&sip, arp_ptr, 4);
2735 arp_ptr += 4 + dev->addr_len;
2736 memcpy(&tip, arp_ptr, 4);
2738 pr_debug("bond_arp_rcv: %s %s/%d av %d sv %d sip %pI4 tip %pI4\n",
2739 bond->dev->name, slave->dev->name, slave->state,
2740 bond->params.arp_validate, slave_do_arp_validate(bond, slave),
2744 * Backup slaves won't see the ARP reply, but do come through
2745 * here for each ARP probe (so we swap the sip/tip to validate
2746 * the probe). In a "redundant switch, common router" type of
2747 * configuration, the ARP probe will (hopefully) travel from
2748 * the active, through one switch, the router, then the other
2749 * switch before reaching the backup.
2751 if (slave->state == BOND_STATE_ACTIVE)
2752 bond_validate_arp(bond, slave, sip, tip);
2754 bond_validate_arp(bond, slave, tip, sip);
2757 read_unlock(&bond->lock);
2760 return NET_RX_SUCCESS;
2764 * this function is called regularly to monitor each slave's link
2765 * ensuring that traffic is being sent and received when arp monitoring
2766 * is used in load-balancing mode. if the adapter has been dormant, then an
2767 * arp is transmitted to generate traffic. see activebackup_arp_monitor for
2768 * arp monitoring in active backup mode.
2770 void bond_loadbalance_arp_mon(struct work_struct *work)
2772 struct bonding *bond = container_of(work, struct bonding,
2774 struct slave *slave, *oldcurrent;
2775 int do_failover = 0;
2779 read_lock(&bond->lock);
2781 delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
2783 if (bond->kill_timers)
2786 if (bond->slave_cnt == 0)
2789 read_lock(&bond->curr_slave_lock);
2790 oldcurrent = bond->curr_active_slave;
2791 read_unlock(&bond->curr_slave_lock);
2793 /* see if any of the previous devices are up now (i.e. they have
2794 * xmt and rcv traffic). the curr_active_slave does not come into
2795 * the picture unless it is null. also, slave->jiffies is not needed
2796 * here because we send an arp on each slave and give a slave as
2797 * long as it needs to get the tx/rx within the delta.
2798 * TODO: what about up/down delay in arp mode? it wasn't here before
2801 bond_for_each_slave(bond, slave, i) {
2802 if (slave->link != BOND_LINK_UP) {
2803 if (time_before_eq(jiffies, dev_trans_start(slave->dev) + delta_in_ticks) &&
2804 time_before_eq(jiffies, slave->dev->last_rx + delta_in_ticks)) {
2806 slave->link = BOND_LINK_UP;
2807 slave->state = BOND_STATE_ACTIVE;
2809 /* primary_slave has no meaning in round-robin
2810 * mode. the window of a slave being up and
2811 * curr_active_slave being null after enslaving
2815 pr_info("%s: link status definitely up for interface %s, ",
2820 pr_info("%s: interface %s is now up\n",
2826 /* slave->link == BOND_LINK_UP */
2828 /* not all switches will respond to an arp request
2829 * when the source ip is 0, so don't take the link down
2830 * if we don't know our ip yet
2832 if (time_after_eq(jiffies, dev_trans_start(slave->dev) + 2*delta_in_ticks) ||
2833 (time_after_eq(jiffies, slave->dev->last_rx + 2*delta_in_ticks))) {
2835 slave->link = BOND_LINK_DOWN;
2836 slave->state = BOND_STATE_BACKUP;
2838 if (slave->link_failure_count < UINT_MAX)
2839 slave->link_failure_count++;
2841 pr_info("%s: interface %s is now down.\n",
2845 if (slave == oldcurrent)
2850 /* note: if switch is in round-robin mode, all links
2851 * must tx arp to ensure all links rx an arp - otherwise
2852 * links may oscillate or not come up at all; if switch is
2853 * in something like xor mode, there is nothing we can
2854 * do - all replies will be rx'ed on same link causing slaves
2855 * to be unstable during low/no traffic periods
2857 if (IS_UP(slave->dev))
2858 bond_arp_send_all(bond, slave);
2862 write_lock_bh(&bond->curr_slave_lock);
2864 bond_select_active_slave(bond);
2866 write_unlock_bh(&bond->curr_slave_lock);
2870 if (bond->params.arp_interval)
2871 queue_delayed_work(bond->wq, &bond->arp_work, delta_in_ticks);
2873 read_unlock(&bond->lock);
2877 * Called to inspect slaves for active-backup mode ARP monitor link state
2878 * changes. Sets new_link in slaves to specify what action should take
2879 * place for the slave. Returns 0 if no changes are found, >0 if changes
2880 * to link states must be committed.
2882 * Called with bond->lock held for read.
2884 static int bond_ab_arp_inspect(struct bonding *bond, int delta_in_ticks)
2886 struct slave *slave;
2889 bond_for_each_slave(bond, slave, i) {
2890 slave->new_link = BOND_LINK_NOCHANGE;
2892 if (slave->link != BOND_LINK_UP) {
2893 if (time_before_eq(jiffies, slave_last_rx(bond, slave) +
2895 slave->new_link = BOND_LINK_UP;
2903 * Give slaves 2*delta after being enslaved or made
2904 * active. This avoids bouncing, as the last receive
2905 * times need a full ARP monitor cycle to be updated.
2907 if (!time_after_eq(jiffies, slave->jiffies +
2908 2 * delta_in_ticks))
2912 * Backup slave is down if:
2913 * - No current_arp_slave AND
2914 * - more than 3*delta since last receive AND
2915 * - the bond has an IP address
2917 * Note: a non-null current_arp_slave indicates
2918 * the curr_active_slave went down and we are
2919 * searching for a new one; under this condition
2920 * we only take the curr_active_slave down - this
2921 * gives each slave a chance to tx/rx traffic
2922 * before being taken out
2924 if (slave->state == BOND_STATE_BACKUP &&
2925 !bond->current_arp_slave &&
2926 time_after(jiffies, slave_last_rx(bond, slave) +
2927 3 * delta_in_ticks)) {
2928 slave->new_link = BOND_LINK_DOWN;
2933 * Active slave is down if:
2934 * - more than 2*delta since transmitting OR
2935 * - (more than 2*delta since receive AND
2936 * the bond has an IP address)
2938 if ((slave->state == BOND_STATE_ACTIVE) &&
2939 (time_after_eq(jiffies, dev_trans_start(slave->dev) +
2940 2 * delta_in_ticks) ||
2941 (time_after_eq(jiffies, slave_last_rx(bond, slave)
2942 + 2 * delta_in_ticks)))) {
2943 slave->new_link = BOND_LINK_DOWN;
2952 * Called to commit link state changes noted by inspection step of
2953 * active-backup mode ARP monitor.
2955 * Called with RTNL and bond->lock for read.
2957 static void bond_ab_arp_commit(struct bonding *bond, int delta_in_ticks)
2959 struct slave *slave;
2962 bond_for_each_slave(bond, slave, i) {
2963 switch (slave->new_link) {
2964 case BOND_LINK_NOCHANGE:
2968 if ((!bond->curr_active_slave &&
2969 time_before_eq(jiffies,
2970 dev_trans_start(slave->dev) +
2972 bond->curr_active_slave != slave) {
2973 slave->link = BOND_LINK_UP;
2974 bond->current_arp_slave = NULL;
2976 pr_info("%s: link status definitely up for interface %s.\n",
2977 bond->dev->name, slave->dev->name);
2979 if (!bond->curr_active_slave ||
2980 (slave == bond->primary_slave))
2987 case BOND_LINK_DOWN:
2988 if (slave->link_failure_count < UINT_MAX)
2989 slave->link_failure_count++;
2991 slave->link = BOND_LINK_DOWN;
2992 bond_set_slave_inactive_flags(slave);
2994 pr_info("%s: link status definitely down for interface %s, disabling it\n",
2995 bond->dev->name, slave->dev->name);
2997 if (slave == bond->curr_active_slave) {
2998 bond->current_arp_slave = NULL;
3005 pr_err("%s: impossible: new_link %d on slave %s\n",
3006 bond->dev->name, slave->new_link,
3013 write_lock_bh(&bond->curr_slave_lock);
3014 bond_select_active_slave(bond);
3015 write_unlock_bh(&bond->curr_slave_lock);
3018 bond_set_carrier(bond);
3022 * Send ARP probes for active-backup mode ARP monitor.
3024 * Called with bond->lock held for read.
3026 static void bond_ab_arp_probe(struct bonding *bond)
3028 struct slave *slave;
3031 read_lock(&bond->curr_slave_lock);
3033 if (bond->current_arp_slave && bond->curr_active_slave)
3034 pr_info("PROBE: c_arp %s && cas %s BAD\n",
3035 bond->current_arp_slave->dev->name,
3036 bond->curr_active_slave->dev->name);
3038 if (bond->curr_active_slave) {
3039 bond_arp_send_all(bond, bond->curr_active_slave);
3040 read_unlock(&bond->curr_slave_lock);
3044 read_unlock(&bond->curr_slave_lock);
3046 /* if we don't have a curr_active_slave, search for the next available
3047 * backup slave from the current_arp_slave and make it the candidate
3048 * for becoming the curr_active_slave
3051 if (!bond->current_arp_slave) {
3052 bond->current_arp_slave = bond->first_slave;
3053 if (!bond->current_arp_slave)
3057 bond_set_slave_inactive_flags(bond->current_arp_slave);
3059 /* search for next candidate */
3060 bond_for_each_slave_from(bond, slave, i, bond->current_arp_slave->next) {
3061 if (IS_UP(slave->dev)) {
3062 slave->link = BOND_LINK_BACK;
3063 bond_set_slave_active_flags(slave);
3064 bond_arp_send_all(bond, slave);
3065 slave->jiffies = jiffies;
3066 bond->current_arp_slave = slave;
3070 /* if the link state is up at this point, we
3071 * mark it down - this can happen if we have
3072 * simultaneous link failures and
3073 * reselect_active_interface doesn't make this
3074 * one the current slave so it is still marked
3075 * up when it is actually down
3077 if (slave->link == BOND_LINK_UP) {
3078 slave->link = BOND_LINK_DOWN;
3079 if (slave->link_failure_count < UINT_MAX)
3080 slave->link_failure_count++;
3082 bond_set_slave_inactive_flags(slave);
3084 pr_info("%s: backup interface %s is now down.\n",
3085 bond->dev->name, slave->dev->name);
3090 void bond_activebackup_arp_mon(struct work_struct *work)
3092 struct bonding *bond = container_of(work, struct bonding,
3096 read_lock(&bond->lock);
3098 if (bond->kill_timers)
3101 delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
3103 if (bond->slave_cnt == 0)
3106 if (bond->send_grat_arp) {
3107 read_lock(&bond->curr_slave_lock);
3108 bond_send_gratuitous_arp(bond);
3109 read_unlock(&bond->curr_slave_lock);
3112 if (bond->send_unsol_na) {
3113 read_lock(&bond->curr_slave_lock);
3114 bond_send_unsolicited_na(bond);
3115 read_unlock(&bond->curr_slave_lock);
3118 if (bond_ab_arp_inspect(bond, delta_in_ticks)) {
3119 read_unlock(&bond->lock);
3121 read_lock(&bond->lock);
3123 bond_ab_arp_commit(bond, delta_in_ticks);
3125 read_unlock(&bond->lock);
3127 read_lock(&bond->lock);
3130 bond_ab_arp_probe(bond);
3133 if (bond->params.arp_interval)
3134 queue_delayed_work(bond->wq, &bond->arp_work, delta_in_ticks);
3136 read_unlock(&bond->lock);
3139 /*------------------------------ proc/seq_file-------------------------------*/
3141 #ifdef CONFIG_PROC_FS
3143 static void *bond_info_seq_start(struct seq_file *seq, loff_t *pos)
3144 __acquires(&dev_base_lock)
3145 __acquires(&bond->lock)
3147 struct bonding *bond = seq->private;
3149 struct slave *slave;
3152 /* make sure the bond won't be taken away */
3153 read_lock(&dev_base_lock);
3154 read_lock(&bond->lock);
3157 return SEQ_START_TOKEN;
3159 bond_for_each_slave(bond, slave, i) {
3167 static void *bond_info_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3169 struct bonding *bond = seq->private;
3170 struct slave *slave = v;
3173 if (v == SEQ_START_TOKEN)
3174 return bond->first_slave;
3176 slave = slave->next;
3178 return (slave == bond->first_slave) ? NULL : slave;
3181 static void bond_info_seq_stop(struct seq_file *seq, void *v)
3182 __releases(&bond->lock)
3183 __releases(&dev_base_lock)
3185 struct bonding *bond = seq->private;
3187 read_unlock(&bond->lock);
3188 read_unlock(&dev_base_lock);
3191 static void bond_info_show_master(struct seq_file *seq)
3193 struct bonding *bond = seq->private;
3197 read_lock(&bond->curr_slave_lock);
3198 curr = bond->curr_active_slave;
3199 read_unlock(&bond->curr_slave_lock);
3201 seq_printf(seq, "Bonding Mode: %s",
3202 bond_mode_name(bond->params.mode));
3204 if (bond->params.mode == BOND_MODE_ACTIVEBACKUP &&
3205 bond->params.fail_over_mac)
3206 seq_printf(seq, " (fail_over_mac %s)",
3207 fail_over_mac_tbl[bond->params.fail_over_mac].modename);
3209 seq_printf(seq, "\n");
3211 if (bond->params.mode == BOND_MODE_XOR ||
3212 bond->params.mode == BOND_MODE_8023AD) {
3213 seq_printf(seq, "Transmit Hash Policy: %s (%d)\n",
3214 xmit_hashtype_tbl[bond->params.xmit_policy].modename,
3215 bond->params.xmit_policy);
3218 if (USES_PRIMARY(bond->params.mode)) {
3219 seq_printf(seq, "Primary Slave: %s",
3220 (bond->primary_slave) ?
3221 bond->primary_slave->dev->name : "None");
3222 if (bond->primary_slave)
3223 seq_printf(seq, " (primary_reselect %s)",
3224 pri_reselect_tbl[bond->params.primary_reselect].modename);
3226 seq_printf(seq, "\nCurrently Active Slave: %s\n",
3227 (curr) ? curr->dev->name : "None");
3230 seq_printf(seq, "MII Status: %s\n", netif_carrier_ok(bond->dev) ?
3232 seq_printf(seq, "MII Polling Interval (ms): %d\n", bond->params.miimon);
3233 seq_printf(seq, "Up Delay (ms): %d\n",
3234 bond->params.updelay * bond->params.miimon);
3235 seq_printf(seq, "Down Delay (ms): %d\n",
3236 bond->params.downdelay * bond->params.miimon);
3239 /* ARP information */
3240 if (bond->params.arp_interval > 0) {
3242 seq_printf(seq, "ARP Polling Interval (ms): %d\n",
3243 bond->params.arp_interval);
3245 seq_printf(seq, "ARP IP target/s (n.n.n.n form):");
3247 for (i = 0; (i < BOND_MAX_ARP_TARGETS); i++) {
3248 if (!bond->params.arp_targets[i])
3251 seq_printf(seq, ",");
3252 seq_printf(seq, " %pI4", &bond->params.arp_targets[i]);
3255 seq_printf(seq, "\n");
3258 if (bond->params.mode == BOND_MODE_8023AD) {
3259 struct ad_info ad_info;
3261 seq_puts(seq, "\n802.3ad info\n");
3262 seq_printf(seq, "LACP rate: %s\n",
3263 (bond->params.lacp_fast) ? "fast" : "slow");
3264 seq_printf(seq, "Aggregator selection policy (ad_select): %s\n",
3265 ad_select_tbl[bond->params.ad_select].modename);
3267 if (bond_3ad_get_active_agg_info(bond, &ad_info)) {
3268 seq_printf(seq, "bond %s has no active aggregator\n",
3271 seq_printf(seq, "Active Aggregator Info:\n");
3273 seq_printf(seq, "\tAggregator ID: %d\n",
3274 ad_info.aggregator_id);
3275 seq_printf(seq, "\tNumber of ports: %d\n",
3277 seq_printf(seq, "\tActor Key: %d\n",
3279 seq_printf(seq, "\tPartner Key: %d\n",
3280 ad_info.partner_key);
3281 seq_printf(seq, "\tPartner Mac Address: %pM\n",
3282 ad_info.partner_system);
3287 static void bond_info_show_slave(struct seq_file *seq,
3288 const struct slave *slave)
3290 struct bonding *bond = seq->private;
3292 seq_printf(seq, "\nSlave Interface: %s\n", slave->dev->name);
3293 seq_printf(seq, "MII Status: %s\n",
3294 (slave->link == BOND_LINK_UP) ? "up" : "down");
3295 seq_printf(seq, "Link Failure Count: %u\n",
3296 slave->link_failure_count);
3298 seq_printf(seq, "Permanent HW addr: %pM\n", slave->perm_hwaddr);
3300 if (bond->params.mode == BOND_MODE_8023AD) {
3301 const struct aggregator *agg
3302 = SLAVE_AD_INFO(slave).port.aggregator;
3305 seq_printf(seq, "Aggregator ID: %d\n",
3306 agg->aggregator_identifier);
3308 seq_puts(seq, "Aggregator ID: N/A\n");
3310 seq_printf(seq, "Slave queue ID: %d\n", slave->queue_id);
3313 static int bond_info_seq_show(struct seq_file *seq, void *v)
3315 if (v == SEQ_START_TOKEN) {
3316 seq_printf(seq, "%s\n", version);
3317 bond_info_show_master(seq);
3319 bond_info_show_slave(seq, v);
3324 static const struct seq_operations bond_info_seq_ops = {
3325 .start = bond_info_seq_start,
3326 .next = bond_info_seq_next,
3327 .stop = bond_info_seq_stop,
3328 .show = bond_info_seq_show,
3331 static int bond_info_open(struct inode *inode, struct file *file)
3333 struct seq_file *seq;
3334 struct proc_dir_entry *proc;
3337 res = seq_open(file, &bond_info_seq_ops);
3339 /* recover the pointer buried in proc_dir_entry data */
3340 seq = file->private_data;
3342 seq->private = proc->data;
3348 static const struct file_operations bond_info_fops = {
3349 .owner = THIS_MODULE,
3350 .open = bond_info_open,
3352 .llseek = seq_lseek,
3353 .release = seq_release,
3356 static void bond_create_proc_entry(struct bonding *bond)
3358 struct net_device *bond_dev = bond->dev;
3359 struct bond_net *bn = net_generic(dev_net(bond_dev), bond_net_id);
3362 bond->proc_entry = proc_create_data(bond_dev->name,
3363 S_IRUGO, bn->proc_dir,
3364 &bond_info_fops, bond);
3365 if (bond->proc_entry == NULL)
3366 pr_warning("Warning: Cannot create /proc/net/%s/%s\n",
3367 DRV_NAME, bond_dev->name);
3369 memcpy(bond->proc_file_name, bond_dev->name, IFNAMSIZ);
3373 static void bond_remove_proc_entry(struct bonding *bond)
3375 struct net_device *bond_dev = bond->dev;
3376 struct bond_net *bn = net_generic(dev_net(bond_dev), bond_net_id);
3378 if (bn->proc_dir && bond->proc_entry) {
3379 remove_proc_entry(bond->proc_file_name, bn->proc_dir);
3380 memset(bond->proc_file_name, 0, IFNAMSIZ);
3381 bond->proc_entry = NULL;
3385 /* Create the bonding directory under /proc/net, if doesn't exist yet.
3386 * Caller must hold rtnl_lock.
3388 static void __net_init bond_create_proc_dir(struct bond_net *bn)
3390 if (!bn->proc_dir) {
3391 bn->proc_dir = proc_mkdir(DRV_NAME, bn->net->proc_net);
3393 pr_warning("Warning: cannot create /proc/net/%s\n",
3398 /* Destroy the bonding directory under /proc/net, if empty.
3399 * Caller must hold rtnl_lock.
3401 static void __net_exit bond_destroy_proc_dir(struct bond_net *bn)
3404 remove_proc_entry(DRV_NAME, bn->net->proc_net);
3405 bn->proc_dir = NULL;
3409 #else /* !CONFIG_PROC_FS */
3411 static void bond_create_proc_entry(struct bonding *bond)
3415 static void bond_remove_proc_entry(struct bonding *bond)
3419 static inline void bond_create_proc_dir(struct bond_net *bn)
3423 static inline void bond_destroy_proc_dir(struct bond_net *bn)
3427 #endif /* CONFIG_PROC_FS */
3430 /*-------------------------- netdev event handling --------------------------*/
3433 * Change device name
3435 static int bond_event_changename(struct bonding *bond)
3437 bond_remove_proc_entry(bond);
3438 bond_create_proc_entry(bond);
3443 static int bond_master_netdev_event(unsigned long event,
3444 struct net_device *bond_dev)
3446 struct bonding *event_bond = netdev_priv(bond_dev);
3449 case NETDEV_CHANGENAME:
3450 return bond_event_changename(event_bond);
3458 static int bond_slave_netdev_event(unsigned long event,
3459 struct net_device *slave_dev)
3461 struct net_device *bond_dev = slave_dev->master;
3462 struct bonding *bond = netdev_priv(bond_dev);
3465 case NETDEV_UNREGISTER:
3467 if (bond->setup_by_slave)
3468 bond_release_and_destroy(bond_dev, slave_dev);
3470 bond_release(bond_dev, slave_dev);
3474 if (bond->params.mode == BOND_MODE_8023AD || bond_is_lb(bond)) {
3475 struct slave *slave;
3477 slave = bond_get_slave_by_dev(bond, slave_dev);
3479 u16 old_speed = slave->speed;
3480 u16 old_duplex = slave->duplex;
3482 bond_update_speed_duplex(slave);
3484 if (bond_is_lb(bond))
3487 if (old_speed != slave->speed)
3488 bond_3ad_adapter_speed_changed(slave);
3489 if (old_duplex != slave->duplex)
3490 bond_3ad_adapter_duplex_changed(slave);
3497 * ... Or is it this?
3500 case NETDEV_CHANGEMTU:
3502 * TODO: Should slaves be allowed to
3503 * independently alter their MTU? For
3504 * an active-backup bond, slaves need
3505 * not be the same type of device, so
3506 * MTUs may vary. For other modes,
3507 * slaves arguably should have the
3508 * same MTUs. To do this, we'd need to
3509 * take over the slave's change_mtu
3510 * function for the duration of their
3514 case NETDEV_CHANGENAME:
3516 * TODO: handle changing the primary's name
3519 case NETDEV_FEAT_CHANGE:
3520 bond_compute_features(bond);
3530 * bond_netdev_event: handle netdev notifier chain events.
3532 * This function receives events for the netdev chain. The caller (an
3533 * ioctl handler calling blocking_notifier_call_chain) holds the necessary
3534 * locks for us to safely manipulate the slave devices (RTNL lock,
3537 static int bond_netdev_event(struct notifier_block *this,
3538 unsigned long event, void *ptr)
3540 struct net_device *event_dev = (struct net_device *)ptr;
3542 pr_debug("event_dev: %s, event: %lx\n",
3543 event_dev ? event_dev->name : "None",
3546 if (!(event_dev->priv_flags & IFF_BONDING))
3549 if (event_dev->flags & IFF_MASTER) {
3550 pr_debug("IFF_MASTER\n");
3551 return bond_master_netdev_event(event, event_dev);
3554 if (event_dev->flags & IFF_SLAVE) {
3555 pr_debug("IFF_SLAVE\n");
3556 return bond_slave_netdev_event(event, event_dev);
3563 * bond_inetaddr_event: handle inetaddr notifier chain events.
3565 * We keep track of device IPs primarily to use as source addresses in
3566 * ARP monitor probes (rather than spewing out broadcasts all the time).
3568 * We track one IP for the main device (if it has one), plus one per VLAN.
3570 static int bond_inetaddr_event(struct notifier_block *this, unsigned long event, void *ptr)
3572 struct in_ifaddr *ifa = ptr;
3573 struct net_device *vlan_dev, *event_dev = ifa->ifa_dev->dev;
3574 struct bond_net *bn = net_generic(dev_net(event_dev), bond_net_id);
3575 struct bonding *bond;
3576 struct vlan_entry *vlan;
3578 list_for_each_entry(bond, &bn->dev_list, bond_list) {
3579 if (bond->dev == event_dev) {
3582 bond->master_ip = ifa->ifa_local;
3585 bond->master_ip = bond_glean_dev_ip(bond->dev);
3592 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
3593 vlan_dev = vlan_group_get_device(bond->vlgrp, vlan->vlan_id);
3594 if (vlan_dev == event_dev) {
3597 vlan->vlan_ip = ifa->ifa_local;
3601 bond_glean_dev_ip(vlan_dev);
3612 static struct notifier_block bond_netdev_notifier = {
3613 .notifier_call = bond_netdev_event,
3616 static struct notifier_block bond_inetaddr_notifier = {
3617 .notifier_call = bond_inetaddr_event,
3620 /*-------------------------- Packet type handling ---------------------------*/
3622 /* register to receive lacpdus on a bond */
3623 static void bond_register_lacpdu(struct bonding *bond)
3625 struct packet_type *pk_type = &(BOND_AD_INFO(bond).ad_pkt_type);
3627 /* initialize packet type */
3628 pk_type->type = PKT_TYPE_LACPDU;
3629 pk_type->dev = bond->dev;
3630 pk_type->func = bond_3ad_lacpdu_recv;
3632 dev_add_pack(pk_type);
3635 /* unregister to receive lacpdus on a bond */
3636 static void bond_unregister_lacpdu(struct bonding *bond)
3638 dev_remove_pack(&(BOND_AD_INFO(bond).ad_pkt_type));
3641 void bond_register_arp(struct bonding *bond)
3643 struct packet_type *pt = &bond->arp_mon_pt;
3648 pt->type = htons(ETH_P_ARP);
3649 pt->dev = bond->dev;
3650 pt->func = bond_arp_rcv;
3654 void bond_unregister_arp(struct bonding *bond)
3656 struct packet_type *pt = &bond->arp_mon_pt;
3658 dev_remove_pack(pt);
3662 /*---------------------------- Hashing Policies -----------------------------*/
3665 * Hash for the output device based upon layer 2 and layer 3 data. If
3666 * the packet is not IP mimic bond_xmit_hash_policy_l2()
3668 static int bond_xmit_hash_policy_l23(struct sk_buff *skb, int count)
3670 struct ethhdr *data = (struct ethhdr *)skb->data;
3671 struct iphdr *iph = ip_hdr(skb);
3673 if (skb->protocol == htons(ETH_P_IP)) {
3674 return ((ntohl(iph->saddr ^ iph->daddr) & 0xffff) ^
3675 (data->h_dest[5] ^ data->h_source[5])) % count;
3678 return (data->h_dest[5] ^ data->h_source[5]) % count;
3682 * Hash for the output device based upon layer 3 and layer 4 data. If
3683 * the packet is a frag or not TCP or UDP, just use layer 3 data. If it is
3684 * altogether not IP, mimic bond_xmit_hash_policy_l2()
3686 static int bond_xmit_hash_policy_l34(struct sk_buff *skb, int count)
3688 struct ethhdr *data = (struct ethhdr *)skb->data;
3689 struct iphdr *iph = ip_hdr(skb);
3690 __be16 *layer4hdr = (__be16 *)((u32 *)iph + iph->ihl);
3693 if (skb->protocol == htons(ETH_P_IP)) {
3694 if (!(iph->frag_off & htons(IP_MF|IP_OFFSET)) &&
3695 (iph->protocol == IPPROTO_TCP ||
3696 iph->protocol == IPPROTO_UDP)) {
3697 layer4_xor = ntohs((*layer4hdr ^ *(layer4hdr + 1)));
3699 return (layer4_xor ^
3700 ((ntohl(iph->saddr ^ iph->daddr)) & 0xffff)) % count;
3704 return (data->h_dest[5] ^ data->h_source[5]) % count;
3708 * Hash for the output device based upon layer 2 data
3710 static int bond_xmit_hash_policy_l2(struct sk_buff *skb, int count)
3712 struct ethhdr *data = (struct ethhdr *)skb->data;
3714 return (data->h_dest[5] ^ data->h_source[5]) % count;
3717 /*-------------------------- Device entry points ----------------------------*/
3719 static int bond_open(struct net_device *bond_dev)
3721 struct bonding *bond = netdev_priv(bond_dev);
3723 bond->kill_timers = 0;
3725 if (bond_is_lb(bond)) {
3726 /* bond_alb_initialize must be called before the timer
3729 if (bond_alb_initialize(bond, (bond->params.mode == BOND_MODE_ALB))) {
3730 /* something went wrong - fail the open operation */
3734 INIT_DELAYED_WORK(&bond->alb_work, bond_alb_monitor);
3735 queue_delayed_work(bond->wq, &bond->alb_work, 0);
3738 if (bond->params.miimon) { /* link check interval, in milliseconds. */
3739 INIT_DELAYED_WORK(&bond->mii_work, bond_mii_monitor);
3740 queue_delayed_work(bond->wq, &bond->mii_work, 0);
3743 if (bond->params.arp_interval) { /* arp interval, in milliseconds. */
3744 if (bond->params.mode == BOND_MODE_ACTIVEBACKUP)
3745 INIT_DELAYED_WORK(&bond->arp_work,
3746 bond_activebackup_arp_mon);
3748 INIT_DELAYED_WORK(&bond->arp_work,
3749 bond_loadbalance_arp_mon);
3751 queue_delayed_work(bond->wq, &bond->arp_work, 0);
3752 if (bond->params.arp_validate)
3753 bond_register_arp(bond);
3756 if (bond->params.mode == BOND_MODE_8023AD) {
3757 INIT_DELAYED_WORK(&bond->ad_work, bond_3ad_state_machine_handler);
3758 queue_delayed_work(bond->wq, &bond->ad_work, 0);
3759 /* register to receive LACPDUs */
3760 bond_register_lacpdu(bond);
3761 bond_3ad_initiate_agg_selection(bond, 1);
3767 static int bond_close(struct net_device *bond_dev)
3769 struct bonding *bond = netdev_priv(bond_dev);
3771 if (bond->params.mode == BOND_MODE_8023AD) {
3772 /* Unregister the receive of LACPDUs */
3773 bond_unregister_lacpdu(bond);
3776 if (bond->params.arp_validate)
3777 bond_unregister_arp(bond);
3779 write_lock_bh(&bond->lock);
3781 bond->send_grat_arp = 0;
3782 bond->send_unsol_na = 0;
3784 /* signal timers not to re-arm */
3785 bond->kill_timers = 1;
3787 write_unlock_bh(&bond->lock);
3789 if (bond->params.miimon) { /* link check interval, in milliseconds. */
3790 cancel_delayed_work(&bond->mii_work);
3793 if (bond->params.arp_interval) { /* arp interval, in milliseconds. */
3794 cancel_delayed_work(&bond->arp_work);
3797 switch (bond->params.mode) {
3798 case BOND_MODE_8023AD:
3799 cancel_delayed_work(&bond->ad_work);
3803 cancel_delayed_work(&bond->alb_work);
3810 if (bond_is_lb(bond)) {
3811 /* Must be called only after all
3812 * slaves have been released
3814 bond_alb_deinitialize(bond);
3820 static struct rtnl_link_stats64 *bond_get_stats(struct net_device *bond_dev,
3821 struct rtnl_link_stats64 *stats)
3823 struct bonding *bond = netdev_priv(bond_dev);
3824 struct rtnl_link_stats64 temp;
3825 struct slave *slave;
3828 memset(stats, 0, sizeof(*stats));
3830 read_lock_bh(&bond->lock);
3832 bond_for_each_slave(bond, slave, i) {
3833 const struct rtnl_link_stats64 *sstats =
3834 dev_get_stats(slave->dev, &temp);
3836 stats->rx_packets += sstats->rx_packets;
3837 stats->rx_bytes += sstats->rx_bytes;
3838 stats->rx_errors += sstats->rx_errors;
3839 stats->rx_dropped += sstats->rx_dropped;
3841 stats->tx_packets += sstats->tx_packets;
3842 stats->tx_bytes += sstats->tx_bytes;
3843 stats->tx_errors += sstats->tx_errors;
3844 stats->tx_dropped += sstats->tx_dropped;
3846 stats->multicast += sstats->multicast;
3847 stats->collisions += sstats->collisions;
3849 stats->rx_length_errors += sstats->rx_length_errors;
3850 stats->rx_over_errors += sstats->rx_over_errors;
3851 stats->rx_crc_errors += sstats->rx_crc_errors;
3852 stats->rx_frame_errors += sstats->rx_frame_errors;
3853 stats->rx_fifo_errors += sstats->rx_fifo_errors;
3854 stats->rx_missed_errors += sstats->rx_missed_errors;
3856 stats->tx_aborted_errors += sstats->tx_aborted_errors;
3857 stats->tx_carrier_errors += sstats->tx_carrier_errors;
3858 stats->tx_fifo_errors += sstats->tx_fifo_errors;
3859 stats->tx_heartbeat_errors += sstats->tx_heartbeat_errors;
3860 stats->tx_window_errors += sstats->tx_window_errors;
3863 read_unlock_bh(&bond->lock);
3868 static int bond_do_ioctl(struct net_device *bond_dev, struct ifreq *ifr, int cmd)
3870 struct net_device *slave_dev = NULL;
3871 struct ifbond k_binfo;
3872 struct ifbond __user *u_binfo = NULL;
3873 struct ifslave k_sinfo;
3874 struct ifslave __user *u_sinfo = NULL;
3875 struct mii_ioctl_data *mii = NULL;
3878 pr_debug("bond_ioctl: master=%s, cmd=%d\n", bond_dev->name, cmd);
3890 * We do this again just in case we were called by SIOCGMIIREG
3891 * instead of SIOCGMIIPHY.
3898 if (mii->reg_num == 1) {
3899 struct bonding *bond = netdev_priv(bond_dev);
3901 read_lock(&bond->lock);
3902 read_lock(&bond->curr_slave_lock);
3903 if (netif_carrier_ok(bond->dev))
3904 mii->val_out = BMSR_LSTATUS;
3906 read_unlock(&bond->curr_slave_lock);
3907 read_unlock(&bond->lock);
3911 case BOND_INFO_QUERY_OLD:
3912 case SIOCBONDINFOQUERY:
3913 u_binfo = (struct ifbond __user *)ifr->ifr_data;
3915 if (copy_from_user(&k_binfo, u_binfo, sizeof(ifbond)))
3918 res = bond_info_query(bond_dev, &k_binfo);
3920 copy_to_user(u_binfo, &k_binfo, sizeof(ifbond)))
3924 case BOND_SLAVE_INFO_QUERY_OLD:
3925 case SIOCBONDSLAVEINFOQUERY:
3926 u_sinfo = (struct ifslave __user *)ifr->ifr_data;
3928 if (copy_from_user(&k_sinfo, u_sinfo, sizeof(ifslave)))
3931 res = bond_slave_info_query(bond_dev, &k_sinfo);
3933 copy_to_user(u_sinfo, &k_sinfo, sizeof(ifslave)))
3942 if (!capable(CAP_NET_ADMIN))
3945 slave_dev = dev_get_by_name(dev_net(bond_dev), ifr->ifr_slave);
3947 pr_debug("slave_dev=%p:\n", slave_dev);
3952 pr_debug("slave_dev->name=%s:\n", slave_dev->name);
3954 case BOND_ENSLAVE_OLD:
3955 case SIOCBONDENSLAVE:
3956 res = bond_enslave(bond_dev, slave_dev);
3958 case BOND_RELEASE_OLD:
3959 case SIOCBONDRELEASE:
3960 res = bond_release(bond_dev, slave_dev);
3962 case BOND_SETHWADDR_OLD:
3963 case SIOCBONDSETHWADDR:
3964 res = bond_sethwaddr(bond_dev, slave_dev);
3966 case BOND_CHANGE_ACTIVE_OLD:
3967 case SIOCBONDCHANGEACTIVE:
3968 res = bond_ioctl_change_active(bond_dev, slave_dev);
3980 static bool bond_addr_in_mc_list(unsigned char *addr,
3981 struct netdev_hw_addr_list *list,
3984 struct netdev_hw_addr *ha;
3986 netdev_hw_addr_list_for_each(ha, list)
3987 if (!memcmp(ha->addr, addr, addrlen))
3993 static void bond_set_multicast_list(struct net_device *bond_dev)
3995 struct bonding *bond = netdev_priv(bond_dev);
3996 struct netdev_hw_addr *ha;
4000 * Do promisc before checking multicast_mode
4002 if ((bond_dev->flags & IFF_PROMISC) && !(bond->flags & IFF_PROMISC))
4004 * FIXME: Need to handle the error when one of the multi-slaves
4007 bond_set_promiscuity(bond, 1);
4010 if (!(bond_dev->flags & IFF_PROMISC) && (bond->flags & IFF_PROMISC))
4011 bond_set_promiscuity(bond, -1);
4014 /* set allmulti flag to slaves */
4015 if ((bond_dev->flags & IFF_ALLMULTI) && !(bond->flags & IFF_ALLMULTI))
4017 * FIXME: Need to handle the error when one of the multi-slaves
4020 bond_set_allmulti(bond, 1);
4023 if (!(bond_dev->flags & IFF_ALLMULTI) && (bond->flags & IFF_ALLMULTI))
4024 bond_set_allmulti(bond, -1);
4027 read_lock(&bond->lock);
4029 bond->flags = bond_dev->flags;
4031 /* looking for addresses to add to slaves' mc list */
4032 netdev_for_each_mc_addr(ha, bond_dev) {
4033 found = bond_addr_in_mc_list(ha->addr, &bond->mc_list,
4034 bond_dev->addr_len);
4036 bond_mc_add(bond, ha->addr);
4039 /* looking for addresses to delete from slaves' list */
4040 netdev_hw_addr_list_for_each(ha, &bond->mc_list) {
4041 found = bond_addr_in_mc_list(ha->addr, &bond_dev->mc,
4042 bond_dev->addr_len);
4044 bond_mc_del(bond, ha->addr);
4047 /* save master's multicast list */
4048 __hw_addr_flush(&bond->mc_list);
4049 __hw_addr_add_multiple(&bond->mc_list, &bond_dev->mc,
4050 bond_dev->addr_len, NETDEV_HW_ADDR_T_MULTICAST);
4052 read_unlock(&bond->lock);
4055 static int bond_neigh_setup(struct net_device *dev, struct neigh_parms *parms)
4057 struct bonding *bond = netdev_priv(dev);
4058 struct slave *slave = bond->first_slave;
4061 const struct net_device_ops *slave_ops
4062 = slave->dev->netdev_ops;
4063 if (slave_ops->ndo_neigh_setup)
4064 return slave_ops->ndo_neigh_setup(slave->dev, parms);
4070 * Change the MTU of all of a master's slaves to match the master
4072 static int bond_change_mtu(struct net_device *bond_dev, int new_mtu)
4074 struct bonding *bond = netdev_priv(bond_dev);
4075 struct slave *slave, *stop_at;
4079 pr_debug("bond=%p, name=%s, new_mtu=%d\n", bond,
4080 (bond_dev ? bond_dev->name : "None"), new_mtu);
4082 /* Can't hold bond->lock with bh disabled here since
4083 * some base drivers panic. On the other hand we can't
4084 * hold bond->lock without bh disabled because we'll
4085 * deadlock. The only solution is to rely on the fact
4086 * that we're under rtnl_lock here, and the slaves
4087 * list won't change. This doesn't solve the problem
4088 * of setting the slave's MTU while it is
4089 * transmitting, but the assumption is that the base
4090 * driver can handle that.
4092 * TODO: figure out a way to safely iterate the slaves
4093 * list, but without holding a lock around the actual
4094 * call to the base driver.
4097 bond_for_each_slave(bond, slave, i) {
4098 pr_debug("s %p s->p %p c_m %p\n",
4101 slave->dev->netdev_ops->ndo_change_mtu);
4103 res = dev_set_mtu(slave->dev, new_mtu);
4106 /* If we failed to set the slave's mtu to the new value
4107 * we must abort the operation even in ACTIVE_BACKUP
4108 * mode, because if we allow the backup slaves to have
4109 * different mtu values than the active slave we'll
4110 * need to change their mtu when doing a failover. That
4111 * means changing their mtu from timer context, which
4112 * is probably not a good idea.
4114 pr_debug("err %d %s\n", res, slave->dev->name);
4119 bond_dev->mtu = new_mtu;
4124 /* unwind from head to the slave that failed */
4126 bond_for_each_slave_from_to(bond, slave, i, bond->first_slave, stop_at) {
4129 tmp_res = dev_set_mtu(slave->dev, bond_dev->mtu);
4131 pr_debug("unwind err %d dev %s\n",
4132 tmp_res, slave->dev->name);
4142 * Note that many devices must be down to change the HW address, and
4143 * downing the master releases all slaves. We can make bonds full of
4144 * bonding devices to test this, however.
4146 static int bond_set_mac_address(struct net_device *bond_dev, void *addr)
4148 struct bonding *bond = netdev_priv(bond_dev);
4149 struct sockaddr *sa = addr, tmp_sa;
4150 struct slave *slave, *stop_at;
4154 if (bond->params.mode == BOND_MODE_ALB)
4155 return bond_alb_set_mac_address(bond_dev, addr);
4158 pr_debug("bond=%p, name=%s\n",
4159 bond, bond_dev ? bond_dev->name : "None");
4162 * If fail_over_mac is set to active, do nothing and return
4163 * success. Returning an error causes ifenslave to fail.
4165 if (bond->params.fail_over_mac == BOND_FOM_ACTIVE)
4168 if (!is_valid_ether_addr(sa->sa_data))
4169 return -EADDRNOTAVAIL;
4171 /* Can't hold bond->lock with bh disabled here since
4172 * some base drivers panic. On the other hand we can't
4173 * hold bond->lock without bh disabled because we'll
4174 * deadlock. The only solution is to rely on the fact
4175 * that we're under rtnl_lock here, and the slaves
4176 * list won't change. This doesn't solve the problem
4177 * of setting the slave's hw address while it is
4178 * transmitting, but the assumption is that the base
4179 * driver can handle that.
4181 * TODO: figure out a way to safely iterate the slaves
4182 * list, but without holding a lock around the actual
4183 * call to the base driver.
4186 bond_for_each_slave(bond, slave, i) {
4187 const struct net_device_ops *slave_ops = slave->dev->netdev_ops;
4188 pr_debug("slave %p %s\n", slave, slave->dev->name);
4190 if (slave_ops->ndo_set_mac_address == NULL) {
4192 pr_debug("EOPNOTSUPP %s\n", slave->dev->name);
4196 res = dev_set_mac_address(slave->dev, addr);
4198 /* TODO: consider downing the slave
4200 * User should expect communications
4201 * breakage anyway until ARP finish
4204 pr_debug("err %d %s\n", res, slave->dev->name);
4210 memcpy(bond_dev->dev_addr, sa->sa_data, bond_dev->addr_len);
4214 memcpy(tmp_sa.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
4215 tmp_sa.sa_family = bond_dev->type;
4217 /* unwind from head to the slave that failed */
4219 bond_for_each_slave_from_to(bond, slave, i, bond->first_slave, stop_at) {
4222 tmp_res = dev_set_mac_address(slave->dev, &tmp_sa);
4224 pr_debug("unwind err %d dev %s\n",
4225 tmp_res, slave->dev->name);
4232 static int bond_xmit_roundrobin(struct sk_buff *skb, struct net_device *bond_dev)
4234 struct bonding *bond = netdev_priv(bond_dev);
4235 struct slave *slave, *start_at;
4236 int i, slave_no, res = 1;
4237 struct iphdr *iph = ip_hdr(skb);
4239 read_lock(&bond->lock);
4241 if (!BOND_IS_OK(bond))
4244 * Start with the curr_active_slave that joined the bond as the
4245 * default for sending IGMP traffic. For failover purposes one
4246 * needs to maintain some consistency for the interface that will
4247 * send the join/membership reports. The curr_active_slave found
4248 * will send all of this type of traffic.
4250 if ((iph->protocol == IPPROTO_IGMP) &&
4251 (skb->protocol == htons(ETH_P_IP))) {
4253 read_lock(&bond->curr_slave_lock);
4254 slave = bond->curr_active_slave;
4255 read_unlock(&bond->curr_slave_lock);
4261 * Concurrent TX may collide on rr_tx_counter; we accept
4262 * that as being rare enough not to justify using an
4265 slave_no = bond->rr_tx_counter++ % bond->slave_cnt;
4267 bond_for_each_slave(bond, slave, i) {
4275 bond_for_each_slave_from(bond, slave, i, start_at) {
4276 if (IS_UP(slave->dev) &&
4277 (slave->link == BOND_LINK_UP) &&
4278 (slave->state == BOND_STATE_ACTIVE)) {
4279 res = bond_dev_queue_xmit(bond, skb, slave->dev);
4286 /* no suitable interface, frame not sent */
4289 read_unlock(&bond->lock);
4290 return NETDEV_TX_OK;
4295 * in active-backup mode, we know that bond->curr_active_slave is always valid if
4296 * the bond has a usable interface.
4298 static int bond_xmit_activebackup(struct sk_buff *skb, struct net_device *bond_dev)
4300 struct bonding *bond = netdev_priv(bond_dev);
4303 read_lock(&bond->lock);
4304 read_lock(&bond->curr_slave_lock);
4306 if (!BOND_IS_OK(bond))
4309 if (!bond->curr_active_slave)
4312 res = bond_dev_queue_xmit(bond, skb, bond->curr_active_slave->dev);
4316 /* no suitable interface, frame not sent */
4319 read_unlock(&bond->curr_slave_lock);
4320 read_unlock(&bond->lock);
4321 return NETDEV_TX_OK;
4325 * In bond_xmit_xor() , we determine the output device by using a pre-
4326 * determined xmit_hash_policy(), If the selected device is not enabled,
4327 * find the next active slave.
4329 static int bond_xmit_xor(struct sk_buff *skb, struct net_device *bond_dev)
4331 struct bonding *bond = netdev_priv(bond_dev);
4332 struct slave *slave, *start_at;
4337 read_lock(&bond->lock);
4339 if (!BOND_IS_OK(bond))
4342 slave_no = bond->xmit_hash_policy(skb, bond->slave_cnt);
4344 bond_for_each_slave(bond, slave, i) {
4352 bond_for_each_slave_from(bond, slave, i, start_at) {
4353 if (IS_UP(slave->dev) &&
4354 (slave->link == BOND_LINK_UP) &&
4355 (slave->state == BOND_STATE_ACTIVE)) {
4356 res = bond_dev_queue_xmit(bond, skb, slave->dev);
4363 /* no suitable interface, frame not sent */
4366 read_unlock(&bond->lock);
4367 return NETDEV_TX_OK;
4371 * in broadcast mode, we send everything to all usable interfaces.
4373 static int bond_xmit_broadcast(struct sk_buff *skb, struct net_device *bond_dev)
4375 struct bonding *bond = netdev_priv(bond_dev);
4376 struct slave *slave, *start_at;
4377 struct net_device *tx_dev = NULL;
4381 read_lock(&bond->lock);
4383 if (!BOND_IS_OK(bond))
4386 read_lock(&bond->curr_slave_lock);
4387 start_at = bond->curr_active_slave;
4388 read_unlock(&bond->curr_slave_lock);
4393 bond_for_each_slave_from(bond, slave, i, start_at) {
4394 if (IS_UP(slave->dev) &&
4395 (slave->link == BOND_LINK_UP) &&
4396 (slave->state == BOND_STATE_ACTIVE)) {
4398 struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
4400 pr_err("%s: Error: bond_xmit_broadcast(): skb_clone() failed\n",
4405 res = bond_dev_queue_xmit(bond, skb2, tx_dev);
4407 dev_kfree_skb(skb2);
4411 tx_dev = slave->dev;
4416 res = bond_dev_queue_xmit(bond, skb, tx_dev);
4420 /* no suitable interface, frame not sent */
4423 /* frame sent to all suitable interfaces */
4424 read_unlock(&bond->lock);
4425 return NETDEV_TX_OK;
4428 /*------------------------- Device initialization ---------------------------*/
4430 static void bond_set_xmit_hash_policy(struct bonding *bond)
4432 switch (bond->params.xmit_policy) {
4433 case BOND_XMIT_POLICY_LAYER23:
4434 bond->xmit_hash_policy = bond_xmit_hash_policy_l23;
4436 case BOND_XMIT_POLICY_LAYER34:
4437 bond->xmit_hash_policy = bond_xmit_hash_policy_l34;
4439 case BOND_XMIT_POLICY_LAYER2:
4441 bond->xmit_hash_policy = bond_xmit_hash_policy_l2;
4447 * Lookup the slave that corresponds to a qid
4449 static inline int bond_slave_override(struct bonding *bond,
4450 struct sk_buff *skb)
4453 struct slave *slave = NULL;
4454 struct slave *check_slave;
4456 read_lock(&bond->lock);
4458 if (!BOND_IS_OK(bond) || !skb->queue_mapping)
4461 /* Find out if any slaves have the same mapping as this skb. */
4462 bond_for_each_slave(bond, check_slave, i) {
4463 if (check_slave->queue_id == skb->queue_mapping) {
4464 slave = check_slave;
4469 /* If the slave isn't UP, use default transmit policy. */
4470 if (slave && slave->queue_id && IS_UP(slave->dev) &&
4471 (slave->link == BOND_LINK_UP)) {
4472 res = bond_dev_queue_xmit(bond, skb, slave->dev);
4476 read_unlock(&bond->lock);
4480 static u16 bond_select_queue(struct net_device *dev, struct sk_buff *skb)
4483 * This helper function exists to help dev_pick_tx get the correct
4484 * destination queue. Using a helper function skips the a call to
4485 * skb_tx_hash and will put the skbs in the queue we expect on their
4486 * way down to the bonding driver.
4488 return skb->queue_mapping;
4491 static netdev_tx_t bond_start_xmit(struct sk_buff *skb, struct net_device *dev)
4493 struct bonding *bond = netdev_priv(dev);
4495 if (TX_QUEUE_OVERRIDE(bond->params.mode)) {
4496 if (!bond_slave_override(bond, skb))
4497 return NETDEV_TX_OK;
4500 switch (bond->params.mode) {
4501 case BOND_MODE_ROUNDROBIN:
4502 return bond_xmit_roundrobin(skb, dev);
4503 case BOND_MODE_ACTIVEBACKUP:
4504 return bond_xmit_activebackup(skb, dev);
4506 return bond_xmit_xor(skb, dev);
4507 case BOND_MODE_BROADCAST:
4508 return bond_xmit_broadcast(skb, dev);
4509 case BOND_MODE_8023AD:
4510 return bond_3ad_xmit_xor(skb, dev);
4513 return bond_alb_xmit(skb, dev);
4515 /* Should never happen, mode already checked */
4516 pr_err("%s: Error: Unknown bonding mode %d\n",
4517 dev->name, bond->params.mode);
4520 return NETDEV_TX_OK;
4526 * set bond mode specific net device operations
4528 void bond_set_mode_ops(struct bonding *bond, int mode)
4530 struct net_device *bond_dev = bond->dev;
4533 case BOND_MODE_ROUNDROBIN:
4535 case BOND_MODE_ACTIVEBACKUP:
4538 bond_set_xmit_hash_policy(bond);
4540 case BOND_MODE_BROADCAST:
4542 case BOND_MODE_8023AD:
4543 bond_set_master_3ad_flags(bond);
4544 bond_set_xmit_hash_policy(bond);
4547 bond_set_master_alb_flags(bond);
4552 /* Should never happen, mode already checked */
4553 pr_err("%s: Error: Unknown bonding mode %d\n",
4554 bond_dev->name, mode);
4559 static void bond_ethtool_get_drvinfo(struct net_device *bond_dev,
4560 struct ethtool_drvinfo *drvinfo)
4562 strncpy(drvinfo->driver, DRV_NAME, 32);
4563 strncpy(drvinfo->version, DRV_VERSION, 32);
4564 snprintf(drvinfo->fw_version, 32, "%d", BOND_ABI_VERSION);
4567 static const struct ethtool_ops bond_ethtool_ops = {
4568 .get_drvinfo = bond_ethtool_get_drvinfo,
4569 .get_link = ethtool_op_get_link,
4570 .get_tx_csum = ethtool_op_get_tx_csum,
4571 .get_sg = ethtool_op_get_sg,
4572 .get_tso = ethtool_op_get_tso,
4573 .get_ufo = ethtool_op_get_ufo,
4574 .get_flags = ethtool_op_get_flags,
4577 static const struct net_device_ops bond_netdev_ops = {
4578 .ndo_init = bond_init,
4579 .ndo_uninit = bond_uninit,
4580 .ndo_open = bond_open,
4581 .ndo_stop = bond_close,
4582 .ndo_start_xmit = bond_start_xmit,
4583 .ndo_select_queue = bond_select_queue,
4584 .ndo_get_stats64 = bond_get_stats,
4585 .ndo_do_ioctl = bond_do_ioctl,
4586 .ndo_set_multicast_list = bond_set_multicast_list,
4587 .ndo_change_mtu = bond_change_mtu,
4588 .ndo_set_mac_address = bond_set_mac_address,
4589 .ndo_neigh_setup = bond_neigh_setup,
4590 .ndo_vlan_rx_register = bond_vlan_rx_register,
4591 .ndo_vlan_rx_add_vid = bond_vlan_rx_add_vid,
4592 .ndo_vlan_rx_kill_vid = bond_vlan_rx_kill_vid,
4593 #ifdef CONFIG_NET_POLL_CONTROLLER
4594 .ndo_netpoll_cleanup = bond_netpoll_cleanup,
4595 .ndo_poll_controller = bond_poll_controller,
4599 static void bond_destructor(struct net_device *bond_dev)
4601 struct bonding *bond = netdev_priv(bond_dev);
4603 destroy_workqueue(bond->wq);
4604 free_netdev(bond_dev);
4607 static void bond_setup(struct net_device *bond_dev)
4609 struct bonding *bond = netdev_priv(bond_dev);
4611 /* initialize rwlocks */
4612 rwlock_init(&bond->lock);
4613 rwlock_init(&bond->curr_slave_lock);
4615 bond->params = bonding_defaults;
4617 /* Initialize pointers */
4618 bond->dev = bond_dev;
4619 INIT_LIST_HEAD(&bond->vlan_list);
4621 /* Initialize the device entry points */
4622 ether_setup(bond_dev);
4623 bond_dev->netdev_ops = &bond_netdev_ops;
4624 bond_dev->ethtool_ops = &bond_ethtool_ops;
4625 bond_set_mode_ops(bond, bond->params.mode);
4627 bond_dev->destructor = bond_destructor;
4629 /* Initialize the device options */
4630 bond_dev->tx_queue_len = 0;
4631 bond_dev->flags |= IFF_MASTER|IFF_MULTICAST;
4632 bond_dev->priv_flags |= IFF_BONDING;
4633 bond_dev->priv_flags &= ~IFF_XMIT_DST_RELEASE;
4635 if (bond->params.arp_interval)
4636 bond_dev->priv_flags |= IFF_MASTER_ARPMON;
4638 /* At first, we block adding VLANs. That's the only way to
4639 * prevent problems that occur when adding VLANs over an
4640 * empty bond. The block will be removed once non-challenged
4641 * slaves are enslaved.
4643 bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
4645 /* don't acquire bond device's netif_tx_lock when
4647 bond_dev->features |= NETIF_F_LLTX;
4649 /* By default, we declare the bond to be fully
4650 * VLAN hardware accelerated capable. Special
4651 * care is taken in the various xmit functions
4652 * when there are slaves that are not hw accel
4655 bond_dev->features |= (NETIF_F_HW_VLAN_TX |
4656 NETIF_F_HW_VLAN_RX |
4657 NETIF_F_HW_VLAN_FILTER);
4661 static void bond_work_cancel_all(struct bonding *bond)
4663 write_lock_bh(&bond->lock);
4664 bond->kill_timers = 1;
4665 write_unlock_bh(&bond->lock);
4667 if (bond->params.miimon && delayed_work_pending(&bond->mii_work))
4668 cancel_delayed_work(&bond->mii_work);
4670 if (bond->params.arp_interval && delayed_work_pending(&bond->arp_work))
4671 cancel_delayed_work(&bond->arp_work);
4673 if (bond->params.mode == BOND_MODE_ALB &&
4674 delayed_work_pending(&bond->alb_work))
4675 cancel_delayed_work(&bond->alb_work);
4677 if (bond->params.mode == BOND_MODE_8023AD &&
4678 delayed_work_pending(&bond->ad_work))
4679 cancel_delayed_work(&bond->ad_work);
4683 * Destroy a bonding device.
4684 * Must be under rtnl_lock when this function is called.
4686 static void bond_uninit(struct net_device *bond_dev)
4688 struct bonding *bond = netdev_priv(bond_dev);
4690 bond_netpoll_cleanup(bond_dev);
4692 /* Release the bonded slaves */
4693 bond_release_all(bond_dev);
4695 list_del(&bond->bond_list);
4697 bond_work_cancel_all(bond);
4699 bond_remove_proc_entry(bond);
4701 __hw_addr_flush(&bond->mc_list);
4704 /*------------------------- Module initialization ---------------------------*/
4707 * Convert string input module parms. Accept either the
4708 * number of the mode or its string name. A bit complicated because
4709 * some mode names are substrings of other names, and calls from sysfs
4710 * may have whitespace in the name (trailing newlines, for example).
4712 int bond_parse_parm(const char *buf, const struct bond_parm_tbl *tbl)
4714 int modeint = -1, i, rv;
4715 char *p, modestr[BOND_MAX_MODENAME_LEN + 1] = { 0, };
4717 for (p = (char *)buf; *p; p++)
4718 if (!(isdigit(*p) || isspace(*p)))
4722 rv = sscanf(buf, "%20s", modestr);
4724 rv = sscanf(buf, "%d", &modeint);
4729 for (i = 0; tbl[i].modename; i++) {
4730 if (modeint == tbl[i].mode)
4732 if (strcmp(modestr, tbl[i].modename) == 0)
4739 static int bond_check_params(struct bond_params *params)
4741 int arp_validate_value, fail_over_mac_value, primary_reselect_value;
4744 * Convert string parameters.
4747 bond_mode = bond_parse_parm(mode, bond_mode_tbl);
4748 if (bond_mode == -1) {
4749 pr_err("Error: Invalid bonding mode \"%s\"\n",
4750 mode == NULL ? "NULL" : mode);
4755 if (xmit_hash_policy) {
4756 if ((bond_mode != BOND_MODE_XOR) &&
4757 (bond_mode != BOND_MODE_8023AD)) {
4758 pr_info("xmit_hash_policy param is irrelevant in mode %s\n",
4759 bond_mode_name(bond_mode));
4761 xmit_hashtype = bond_parse_parm(xmit_hash_policy,
4763 if (xmit_hashtype == -1) {
4764 pr_err("Error: Invalid xmit_hash_policy \"%s\"\n",
4765 xmit_hash_policy == NULL ? "NULL" :
4773 if (bond_mode != BOND_MODE_8023AD) {
4774 pr_info("lacp_rate param is irrelevant in mode %s\n",
4775 bond_mode_name(bond_mode));
4777 lacp_fast = bond_parse_parm(lacp_rate, bond_lacp_tbl);
4778 if (lacp_fast == -1) {
4779 pr_err("Error: Invalid lacp rate \"%s\"\n",
4780 lacp_rate == NULL ? "NULL" : lacp_rate);
4787 params->ad_select = bond_parse_parm(ad_select, ad_select_tbl);
4788 if (params->ad_select == -1) {
4789 pr_err("Error: Invalid ad_select \"%s\"\n",
4790 ad_select == NULL ? "NULL" : ad_select);
4794 if (bond_mode != BOND_MODE_8023AD) {
4795 pr_warning("ad_select param only affects 802.3ad mode\n");
4798 params->ad_select = BOND_AD_STABLE;
4801 if (max_bonds < 0) {
4802 pr_warning("Warning: max_bonds (%d) not in range %d-%d, so it was reset to BOND_DEFAULT_MAX_BONDS (%d)\n",
4803 max_bonds, 0, INT_MAX, BOND_DEFAULT_MAX_BONDS);
4804 max_bonds = BOND_DEFAULT_MAX_BONDS;
4808 pr_warning("Warning: miimon module parameter (%d), not in range 0-%d, so it was reset to %d\n",
4809 miimon, INT_MAX, BOND_LINK_MON_INTERV);
4810 miimon = BOND_LINK_MON_INTERV;
4814 pr_warning("Warning: updelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4819 if (downdelay < 0) {
4820 pr_warning("Warning: downdelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4821 downdelay, INT_MAX);
4825 if ((use_carrier != 0) && (use_carrier != 1)) {
4826 pr_warning("Warning: use_carrier module parameter (%d), not of valid value (0/1), so it was set to 1\n",
4831 if (num_grat_arp < 0 || num_grat_arp > 255) {
4832 pr_warning("Warning: num_grat_arp (%d) not in range 0-255 so it was reset to 1\n",
4837 if (num_unsol_na < 0 || num_unsol_na > 255) {
4838 pr_warning("Warning: num_unsol_na (%d) not in range 0-255 so it was reset to 1\n",
4843 /* reset values for 802.3ad */
4844 if (bond_mode == BOND_MODE_8023AD) {
4846 pr_warning("Warning: miimon must be specified, otherwise bonding will not detect link failure, speed and duplex which are essential for 802.3ad operation\n");
4847 pr_warning("Forcing miimon to 100msec\n");
4852 if (tx_queues < 1 || tx_queues > 255) {
4853 pr_warning("Warning: tx_queues (%d) should be between "
4854 "1 and 255, resetting to %d\n",
4855 tx_queues, BOND_DEFAULT_TX_QUEUES);
4856 tx_queues = BOND_DEFAULT_TX_QUEUES;
4859 if ((all_slaves_active != 0) && (all_slaves_active != 1)) {
4860 pr_warning("Warning: all_slaves_active module parameter (%d), "
4861 "not of valid value (0/1), so it was set to "
4862 "0\n", all_slaves_active);
4863 all_slaves_active = 0;
4866 /* reset values for TLB/ALB */
4867 if ((bond_mode == BOND_MODE_TLB) ||
4868 (bond_mode == BOND_MODE_ALB)) {
4870 pr_warning("Warning: miimon must be specified, otherwise bonding will not detect link failure and link speed which are essential for TLB/ALB load balancing\n");
4871 pr_warning("Forcing miimon to 100msec\n");
4876 if (bond_mode == BOND_MODE_ALB) {
4877 pr_notice("In ALB mode you might experience client disconnections upon reconnection of a link if the bonding module updelay parameter (%d msec) is incompatible with the forwarding delay time of the switch\n",
4882 if (updelay || downdelay) {
4883 /* just warn the user the up/down delay will have
4884 * no effect since miimon is zero...
4886 pr_warning("Warning: miimon module parameter not set and updelay (%d) or downdelay (%d) module parameter is set; updelay and downdelay have no effect unless miimon is set\n",
4887 updelay, downdelay);
4890 /* don't allow arp monitoring */
4892 pr_warning("Warning: miimon (%d) and arp_interval (%d) can't be used simultaneously, disabling ARP monitoring\n",
4893 miimon, arp_interval);
4897 if ((updelay % miimon) != 0) {
4898 pr_warning("Warning: updelay (%d) is not a multiple of miimon (%d), updelay rounded to %d ms\n",
4900 (updelay / miimon) * miimon);
4905 if ((downdelay % miimon) != 0) {
4906 pr_warning("Warning: downdelay (%d) is not a multiple of miimon (%d), downdelay rounded to %d ms\n",
4908 (downdelay / miimon) * miimon);
4911 downdelay /= miimon;
4914 if (arp_interval < 0) {
4915 pr_warning("Warning: arp_interval module parameter (%d) , not in range 0-%d, so it was reset to %d\n",
4916 arp_interval, INT_MAX, BOND_LINK_ARP_INTERV);
4917 arp_interval = BOND_LINK_ARP_INTERV;
4920 for (arp_ip_count = 0;
4921 (arp_ip_count < BOND_MAX_ARP_TARGETS) && arp_ip_target[arp_ip_count];
4923 /* not complete check, but should be good enough to
4925 if (!isdigit(arp_ip_target[arp_ip_count][0])) {
4926 pr_warning("Warning: bad arp_ip_target module parameter (%s), ARP monitoring will not be performed\n",
4927 arp_ip_target[arp_ip_count]);
4930 __be32 ip = in_aton(arp_ip_target[arp_ip_count]);
4931 arp_target[arp_ip_count] = ip;
4935 if (arp_interval && !arp_ip_count) {
4936 /* don't allow arping if no arp_ip_target given... */
4937 pr_warning("Warning: arp_interval module parameter (%d) specified without providing an arp_ip_target parameter, arp_interval was reset to 0\n",
4943 if (bond_mode != BOND_MODE_ACTIVEBACKUP) {
4944 pr_err("arp_validate only supported in active-backup mode\n");
4947 if (!arp_interval) {
4948 pr_err("arp_validate requires arp_interval\n");
4952 arp_validate_value = bond_parse_parm(arp_validate,
4954 if (arp_validate_value == -1) {
4955 pr_err("Error: invalid arp_validate \"%s\"\n",
4956 arp_validate == NULL ? "NULL" : arp_validate);
4960 arp_validate_value = 0;
4963 pr_info("MII link monitoring set to %d ms\n", miimon);
4964 } else if (arp_interval) {
4967 pr_info("ARP monitoring set to %d ms, validate %s, with %d target(s):",
4969 arp_validate_tbl[arp_validate_value].modename,
4972 for (i = 0; i < arp_ip_count; i++)
4973 pr_info(" %s", arp_ip_target[i]);
4977 } else if (max_bonds) {
4978 /* miimon and arp_interval not set, we need one so things
4979 * work as expected, see bonding.txt for details
4981 pr_warning("Warning: either miimon or arp_interval and arp_ip_target module parameters must be specified, otherwise bonding will not detect link failures! see bonding.txt for details.\n");
4984 if (primary && !USES_PRIMARY(bond_mode)) {
4985 /* currently, using a primary only makes sense
4986 * in active backup, TLB or ALB modes
4988 pr_warning("Warning: %s primary device specified but has no effect in %s mode\n",
4989 primary, bond_mode_name(bond_mode));
4993 if (primary && primary_reselect) {
4994 primary_reselect_value = bond_parse_parm(primary_reselect,
4996 if (primary_reselect_value == -1) {
4997 pr_err("Error: Invalid primary_reselect \"%s\"\n",
4999 NULL ? "NULL" : primary_reselect);
5003 primary_reselect_value = BOND_PRI_RESELECT_ALWAYS;
5006 if (fail_over_mac) {
5007 fail_over_mac_value = bond_parse_parm(fail_over_mac,
5009 if (fail_over_mac_value == -1) {
5010 pr_err("Error: invalid fail_over_mac \"%s\"\n",
5011 arp_validate == NULL ? "NULL" : arp_validate);
5015 if (bond_mode != BOND_MODE_ACTIVEBACKUP)
5016 pr_warning("Warning: fail_over_mac only affects active-backup mode.\n");
5018 fail_over_mac_value = BOND_FOM_NONE;
5021 /* fill params struct with the proper values */
5022 params->mode = bond_mode;
5023 params->xmit_policy = xmit_hashtype;
5024 params->miimon = miimon;
5025 params->num_grat_arp = num_grat_arp;
5026 params->num_unsol_na = num_unsol_na;
5027 params->arp_interval = arp_interval;
5028 params->arp_validate = arp_validate_value;
5029 params->updelay = updelay;
5030 params->downdelay = downdelay;
5031 params->use_carrier = use_carrier;
5032 params->lacp_fast = lacp_fast;
5033 params->primary[0] = 0;
5034 params->primary_reselect = primary_reselect_value;
5035 params->fail_over_mac = fail_over_mac_value;
5036 params->tx_queues = tx_queues;
5037 params->all_slaves_active = all_slaves_active;
5040 strncpy(params->primary, primary, IFNAMSIZ);
5041 params->primary[IFNAMSIZ - 1] = 0;
5044 memcpy(params->arp_targets, arp_target, sizeof(arp_target));
5049 static struct lock_class_key bonding_netdev_xmit_lock_key;
5050 static struct lock_class_key bonding_netdev_addr_lock_key;
5052 static void bond_set_lockdep_class_one(struct net_device *dev,
5053 struct netdev_queue *txq,
5056 lockdep_set_class(&txq->_xmit_lock,
5057 &bonding_netdev_xmit_lock_key);
5060 static void bond_set_lockdep_class(struct net_device *dev)
5062 lockdep_set_class(&dev->addr_list_lock,
5063 &bonding_netdev_addr_lock_key);
5064 netdev_for_each_tx_queue(dev, bond_set_lockdep_class_one, NULL);
5068 * Called from registration process
5070 static int bond_init(struct net_device *bond_dev)
5072 struct bonding *bond = netdev_priv(bond_dev);
5073 struct bond_net *bn = net_generic(dev_net(bond_dev), bond_net_id);
5075 pr_debug("Begin bond_init for %s\n", bond_dev->name);
5077 bond->wq = create_singlethread_workqueue(bond_dev->name);
5081 bond_set_lockdep_class(bond_dev);
5083 netif_carrier_off(bond_dev);
5085 bond_create_proc_entry(bond);
5086 list_add_tail(&bond->bond_list, &bn->dev_list);
5088 bond_prepare_sysfs_group(bond);
5090 __hw_addr_init(&bond->mc_list);
5094 static int bond_validate(struct nlattr *tb[], struct nlattr *data[])
5096 if (tb[IFLA_ADDRESS]) {
5097 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN)
5099 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS])))
5100 return -EADDRNOTAVAIL;
5105 static struct rtnl_link_ops bond_link_ops __read_mostly = {
5107 .priv_size = sizeof(struct bonding),
5108 .setup = bond_setup,
5109 .validate = bond_validate,
5112 /* Create a new bond based on the specified name and bonding parameters.
5113 * If name is NULL, obtain a suitable "bond%d" name for us.
5114 * Caller must NOT hold rtnl_lock; we need to release it here before we
5115 * set up our sysfs entries.
5117 int bond_create(struct net *net, const char *name)
5119 struct net_device *bond_dev;
5124 bond_dev = alloc_netdev_mq(sizeof(struct bonding), name ? name : "",
5125 bond_setup, tx_queues);
5127 pr_err("%s: eek! can't alloc netdev!\n", name);
5132 dev_net_set(bond_dev, net);
5133 bond_dev->rtnl_link_ops = &bond_link_ops;
5136 res = dev_alloc_name(bond_dev, "bond%d");
5141 res = register_netdevice(bond_dev);
5146 bond_destructor(bond_dev);
5150 static int __net_init bond_net_init(struct net *net)
5152 struct bond_net *bn = net_generic(net, bond_net_id);
5155 INIT_LIST_HEAD(&bn->dev_list);
5157 bond_create_proc_dir(bn);
5162 static void __net_exit bond_net_exit(struct net *net)
5164 struct bond_net *bn = net_generic(net, bond_net_id);
5166 bond_destroy_proc_dir(bn);
5169 static struct pernet_operations bond_net_ops = {
5170 .init = bond_net_init,
5171 .exit = bond_net_exit,
5173 .size = sizeof(struct bond_net),
5176 static int __init bonding_init(void)
5181 pr_info("%s", version);
5183 res = bond_check_params(&bonding_defaults);
5187 res = register_pernet_subsys(&bond_net_ops);
5191 res = rtnl_link_register(&bond_link_ops);
5195 for (i = 0; i < max_bonds; i++) {
5196 res = bond_create(&init_net, NULL);
5201 res = bond_create_sysfs();
5205 register_netdevice_notifier(&bond_netdev_notifier);
5206 register_inetaddr_notifier(&bond_inetaddr_notifier);
5207 bond_register_ipv6_notifier();
5211 rtnl_link_unregister(&bond_link_ops);
5213 unregister_pernet_subsys(&bond_net_ops);
5218 static void __exit bonding_exit(void)
5220 unregister_netdevice_notifier(&bond_netdev_notifier);
5221 unregister_inetaddr_notifier(&bond_inetaddr_notifier);
5222 bond_unregister_ipv6_notifier();
5224 bond_destroy_sysfs();
5226 rtnl_link_unregister(&bond_link_ops);
5227 unregister_pernet_subsys(&bond_net_ops);
5230 module_init(bonding_init);
5231 module_exit(bonding_exit);
5232 MODULE_LICENSE("GPL");
5233 MODULE_VERSION(DRV_VERSION);
5234 MODULE_DESCRIPTION(DRV_DESCRIPTION ", v" DRV_VERSION);
5235 MODULE_AUTHOR("Thomas Davis, tadavis@lbl.gov and many others");
5236 MODULE_ALIAS_RTNL_LINK("bond");