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 /* Test vlan_list not vlgrp to catch and handle 802.1p tags */
428 if (!list_empty(&bond->vlan_list) &&
429 !(slave_dev->features & NETIF_F_HW_VLAN_TX) &&
430 vlan_get_tag(skb, &vlan_id) == 0) {
431 skb->dev = slave_dev;
432 skb = vlan_put_tag(skb, vlan_id);
434 /* vlan_put_tag() frees the skb in case of error,
435 * so return success here so the calling functions
436 * won't attempt to free is again.
441 skb->dev = slave_dev;
445 #ifdef CONFIG_NET_POLL_CONTROLLER
446 if (unlikely(bond->dev->priv_flags & IFF_IN_NETPOLL)) {
447 struct netpoll *np = bond->dev->npinfo->netpoll;
448 slave_dev->npinfo = bond->dev->npinfo;
449 np->real_dev = np->dev = skb->dev;
450 slave_dev->priv_flags |= IFF_IN_NETPOLL;
451 netpoll_send_skb(np, skb);
452 slave_dev->priv_flags &= ~IFF_IN_NETPOLL;
462 * In the following 3 functions, bond_vlan_rx_register(), bond_vlan_rx_add_vid
463 * and bond_vlan_rx_kill_vid, We don't protect the slave list iteration with a
465 * a. This operation is performed in IOCTL context,
466 * b. The operation is protected by the RTNL semaphore in the 8021q code,
467 * c. Holding a lock with BH disabled while directly calling a base driver
468 * entry point is generally a BAD idea.
470 * The design of synchronization/protection for this operation in the 8021q
471 * module is good for one or more VLAN devices over a single physical device
472 * and cannot be extended for a teaming solution like bonding, so there is a
473 * potential race condition here where a net device from the vlan group might
474 * be referenced (either by a base driver or the 8021q code) while it is being
475 * removed from the system. However, it turns out we're not making matters
476 * worse, and if it works for regular VLAN usage it will work here too.
480 * bond_vlan_rx_register - Propagates registration to slaves
481 * @bond_dev: bonding net device that got called
482 * @grp: vlan group being registered
484 static void bond_vlan_rx_register(struct net_device *bond_dev,
485 struct vlan_group *grp)
487 struct bonding *bond = netdev_priv(bond_dev);
491 write_lock(&bond->lock);
493 write_unlock(&bond->lock);
495 bond_for_each_slave(bond, slave, i) {
496 struct net_device *slave_dev = slave->dev;
497 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
499 if ((slave_dev->features & NETIF_F_HW_VLAN_RX) &&
500 slave_ops->ndo_vlan_rx_register) {
501 slave_ops->ndo_vlan_rx_register(slave_dev, grp);
507 * bond_vlan_rx_add_vid - Propagates adding an id to slaves
508 * @bond_dev: bonding net device that got called
509 * @vid: vlan id being added
511 static void bond_vlan_rx_add_vid(struct net_device *bond_dev, uint16_t vid)
513 struct bonding *bond = netdev_priv(bond_dev);
517 bond_for_each_slave(bond, slave, i) {
518 struct net_device *slave_dev = slave->dev;
519 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
521 if ((slave_dev->features & NETIF_F_HW_VLAN_FILTER) &&
522 slave_ops->ndo_vlan_rx_add_vid) {
523 slave_ops->ndo_vlan_rx_add_vid(slave_dev, vid);
527 res = bond_add_vlan(bond, vid);
529 pr_err("%s: Error: Failed to add vlan id %d\n",
530 bond_dev->name, vid);
535 * bond_vlan_rx_kill_vid - Propagates deleting an id to slaves
536 * @bond_dev: bonding net device that got called
537 * @vid: vlan id being removed
539 static void bond_vlan_rx_kill_vid(struct net_device *bond_dev, uint16_t vid)
541 struct bonding *bond = netdev_priv(bond_dev);
543 struct net_device *vlan_dev;
546 bond_for_each_slave(bond, slave, i) {
547 struct net_device *slave_dev = slave->dev;
548 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
550 if ((slave_dev->features & NETIF_F_HW_VLAN_FILTER) &&
551 slave_ops->ndo_vlan_rx_kill_vid) {
552 /* Save and then restore vlan_dev in the grp array,
553 * since the slave's driver might clear it.
555 vlan_dev = vlan_group_get_device(bond->vlgrp, vid);
556 slave_ops->ndo_vlan_rx_kill_vid(slave_dev, vid);
557 vlan_group_set_device(bond->vlgrp, vid, vlan_dev);
561 res = bond_del_vlan(bond, vid);
563 pr_err("%s: Error: Failed to remove vlan id %d\n",
564 bond_dev->name, vid);
568 static void bond_add_vlans_on_slave(struct bonding *bond, struct net_device *slave_dev)
570 struct vlan_entry *vlan;
571 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
576 if ((slave_dev->features & NETIF_F_HW_VLAN_RX) &&
577 slave_ops->ndo_vlan_rx_register)
578 slave_ops->ndo_vlan_rx_register(slave_dev, bond->vlgrp);
580 if (!(slave_dev->features & NETIF_F_HW_VLAN_FILTER) ||
581 !(slave_ops->ndo_vlan_rx_add_vid))
584 list_for_each_entry(vlan, &bond->vlan_list, vlan_list)
585 slave_ops->ndo_vlan_rx_add_vid(slave_dev, vlan->vlan_id);
588 static void bond_del_vlans_from_slave(struct bonding *bond,
589 struct net_device *slave_dev)
591 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
592 struct vlan_entry *vlan;
593 struct net_device *vlan_dev;
598 if (!(slave_dev->features & NETIF_F_HW_VLAN_FILTER) ||
599 !(slave_ops->ndo_vlan_rx_kill_vid))
602 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
605 /* Save and then restore vlan_dev in the grp array,
606 * since the slave's driver might clear it.
608 vlan_dev = vlan_group_get_device(bond->vlgrp, vlan->vlan_id);
609 slave_ops->ndo_vlan_rx_kill_vid(slave_dev, vlan->vlan_id);
610 vlan_group_set_device(bond->vlgrp, vlan->vlan_id, vlan_dev);
614 if ((slave_dev->features & NETIF_F_HW_VLAN_RX) &&
615 slave_ops->ndo_vlan_rx_register)
616 slave_ops->ndo_vlan_rx_register(slave_dev, NULL);
619 /*------------------------------- Link status -------------------------------*/
622 * Set the carrier state for the master according to the state of its
623 * slaves. If any slaves are up, the master is up. In 802.3ad mode,
624 * do special 802.3ad magic.
626 * Returns zero if carrier state does not change, nonzero if it does.
628 static int bond_set_carrier(struct bonding *bond)
633 if (bond->slave_cnt == 0)
636 if (bond->params.mode == BOND_MODE_8023AD)
637 return bond_3ad_set_carrier(bond);
639 bond_for_each_slave(bond, slave, i) {
640 if (slave->link == BOND_LINK_UP) {
641 if (!netif_carrier_ok(bond->dev)) {
642 netif_carrier_on(bond->dev);
650 if (netif_carrier_ok(bond->dev)) {
651 netif_carrier_off(bond->dev);
658 * Get link speed and duplex from the slave's base driver
659 * using ethtool. If for some reason the call fails or the
660 * values are invalid, fake speed and duplex to 100/Full
663 static int bond_update_speed_duplex(struct slave *slave)
665 struct net_device *slave_dev = slave->dev;
666 struct ethtool_cmd etool;
669 /* Fake speed and duplex */
670 slave->speed = SPEED_100;
671 slave->duplex = DUPLEX_FULL;
673 if (!slave_dev->ethtool_ops || !slave_dev->ethtool_ops->get_settings)
676 res = slave_dev->ethtool_ops->get_settings(slave_dev, &etool);
680 switch (etool.speed) {
690 switch (etool.duplex) {
698 slave->speed = etool.speed;
699 slave->duplex = etool.duplex;
705 * if <dev> supports MII link status reporting, check its link status.
707 * We either do MII/ETHTOOL ioctls, or check netif_carrier_ok(),
708 * depending upon the setting of the use_carrier parameter.
710 * Return either BMSR_LSTATUS, meaning that the link is up (or we
711 * can't tell and just pretend it is), or 0, meaning that the link is
714 * If reporting is non-zero, instead of faking link up, return -1 if
715 * both ETHTOOL and MII ioctls fail (meaning the device does not
716 * support them). If use_carrier is set, return whatever it says.
717 * It'd be nice if there was a good way to tell if a driver supports
718 * netif_carrier, but there really isn't.
720 static int bond_check_dev_link(struct bonding *bond,
721 struct net_device *slave_dev, int reporting)
723 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
724 int (*ioctl)(struct net_device *, struct ifreq *, int);
726 struct mii_ioctl_data *mii;
728 if (!reporting && !netif_running(slave_dev))
731 if (bond->params.use_carrier)
732 return netif_carrier_ok(slave_dev) ? BMSR_LSTATUS : 0;
734 /* Try to get link status using Ethtool first. */
735 if (slave_dev->ethtool_ops) {
736 if (slave_dev->ethtool_ops->get_link) {
739 link = slave_dev->ethtool_ops->get_link(slave_dev);
741 return link ? BMSR_LSTATUS : 0;
745 /* Ethtool can't be used, fallback to MII ioctls. */
746 ioctl = slave_ops->ndo_do_ioctl;
748 /* TODO: set pointer to correct ioctl on a per team member */
749 /* bases to make this more efficient. that is, once */
750 /* we determine the correct ioctl, we will always */
751 /* call it and not the others for that team */
755 * We cannot assume that SIOCGMIIPHY will also read a
756 * register; not all network drivers (e.g., e100)
760 /* Yes, the mii is overlaid on the ifreq.ifr_ifru */
761 strncpy(ifr.ifr_name, slave_dev->name, IFNAMSIZ);
763 if (IOCTL(slave_dev, &ifr, SIOCGMIIPHY) == 0) {
764 mii->reg_num = MII_BMSR;
765 if (IOCTL(slave_dev, &ifr, SIOCGMIIREG) == 0)
766 return mii->val_out & BMSR_LSTATUS;
771 * If reporting, report that either there's no dev->do_ioctl,
772 * or both SIOCGMIIREG and get_link failed (meaning that we
773 * cannot report link status). If not reporting, pretend
776 return reporting ? -1 : BMSR_LSTATUS;
779 /*----------------------------- Multicast list ------------------------------*/
782 * Push the promiscuity flag down to appropriate slaves
784 static int bond_set_promiscuity(struct bonding *bond, int inc)
787 if (USES_PRIMARY(bond->params.mode)) {
788 /* write lock already acquired */
789 if (bond->curr_active_slave) {
790 err = dev_set_promiscuity(bond->curr_active_slave->dev,
796 bond_for_each_slave(bond, slave, i) {
797 err = dev_set_promiscuity(slave->dev, inc);
806 * Push the allmulti flag down to all slaves
808 static int bond_set_allmulti(struct bonding *bond, int inc)
811 if (USES_PRIMARY(bond->params.mode)) {
812 /* write lock already acquired */
813 if (bond->curr_active_slave) {
814 err = dev_set_allmulti(bond->curr_active_slave->dev,
820 bond_for_each_slave(bond, slave, i) {
821 err = dev_set_allmulti(slave->dev, inc);
830 * Add a Multicast address to slaves
833 static void bond_mc_add(struct bonding *bond, void *addr)
835 if (USES_PRIMARY(bond->params.mode)) {
836 /* write lock already acquired */
837 if (bond->curr_active_slave)
838 dev_mc_add(bond->curr_active_slave->dev, addr);
843 bond_for_each_slave(bond, slave, i)
844 dev_mc_add(slave->dev, addr);
849 * Remove a multicast address from slave
852 static void bond_mc_del(struct bonding *bond, void *addr)
854 if (USES_PRIMARY(bond->params.mode)) {
855 /* write lock already acquired */
856 if (bond->curr_active_slave)
857 dev_mc_del(bond->curr_active_slave->dev, addr);
861 bond_for_each_slave(bond, slave, i) {
862 dev_mc_del(slave->dev, addr);
868 static void __bond_resend_igmp_join_requests(struct net_device *dev)
870 struct in_device *in_dev;
871 struct ip_mc_list *im;
874 in_dev = __in_dev_get_rcu(dev);
876 for (im = in_dev->mc_list; im; im = im->next)
877 ip_mc_rejoin_group(im);
884 * Retrieve the list of registered multicast addresses for the bonding
885 * device and retransmit an IGMP JOIN request to the current active
888 static void bond_resend_igmp_join_requests(struct bonding *bond)
890 struct net_device *vlan_dev;
891 struct vlan_entry *vlan;
893 read_lock(&bond->lock);
895 /* rejoin all groups on bond device */
896 __bond_resend_igmp_join_requests(bond->dev);
898 /* rejoin all groups on vlan devices */
900 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
901 vlan_dev = vlan_group_get_device(bond->vlgrp,
904 __bond_resend_igmp_join_requests(vlan_dev);
908 read_unlock(&bond->lock);
911 void bond_resend_igmp_join_requests_delayed(struct work_struct *work)
913 struct bonding *bond = container_of(work, struct bonding,
915 bond_resend_igmp_join_requests(bond);
919 * flush all members of flush->mc_list from device dev->mc_list
921 static void bond_mc_list_flush(struct net_device *bond_dev,
922 struct net_device *slave_dev)
924 struct bonding *bond = netdev_priv(bond_dev);
925 struct netdev_hw_addr *ha;
927 netdev_for_each_mc_addr(ha, bond_dev)
928 dev_mc_del(slave_dev, ha->addr);
930 if (bond->params.mode == BOND_MODE_8023AD) {
931 /* del lacpdu mc addr from mc list */
932 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
934 dev_mc_del(slave_dev, lacpdu_multicast);
938 /*--------------------------- Active slave change ---------------------------*/
941 * Update the mc list and multicast-related flags for the new and
942 * old active slaves (if any) according to the multicast mode, and
943 * promiscuous flags unconditionally.
945 static void bond_mc_swap(struct bonding *bond, struct slave *new_active,
946 struct slave *old_active)
948 struct netdev_hw_addr *ha;
950 if (!USES_PRIMARY(bond->params.mode))
951 /* nothing to do - mc list is already up-to-date on
957 if (bond->dev->flags & IFF_PROMISC)
958 dev_set_promiscuity(old_active->dev, -1);
960 if (bond->dev->flags & IFF_ALLMULTI)
961 dev_set_allmulti(old_active->dev, -1);
963 netdev_for_each_mc_addr(ha, bond->dev)
964 dev_mc_del(old_active->dev, ha->addr);
968 /* FIXME: Signal errors upstream. */
969 if (bond->dev->flags & IFF_PROMISC)
970 dev_set_promiscuity(new_active->dev, 1);
972 if (bond->dev->flags & IFF_ALLMULTI)
973 dev_set_allmulti(new_active->dev, 1);
975 netdev_for_each_mc_addr(ha, bond->dev)
976 dev_mc_add(new_active->dev, ha->addr);
981 * bond_do_fail_over_mac
983 * Perform special MAC address swapping for fail_over_mac settings
985 * Called with RTNL, bond->lock for read, curr_slave_lock for write_bh.
987 static void bond_do_fail_over_mac(struct bonding *bond,
988 struct slave *new_active,
989 struct slave *old_active)
990 __releases(&bond->curr_slave_lock)
991 __releases(&bond->lock)
992 __acquires(&bond->lock)
993 __acquires(&bond->curr_slave_lock)
995 u8 tmp_mac[ETH_ALEN];
996 struct sockaddr saddr;
999 switch (bond->params.fail_over_mac) {
1000 case BOND_FOM_ACTIVE:
1002 memcpy(bond->dev->dev_addr, new_active->dev->dev_addr,
1003 new_active->dev->addr_len);
1005 case BOND_FOM_FOLLOW:
1007 * if new_active && old_active, swap them
1008 * if just old_active, do nothing (going to no active slave)
1009 * if just new_active, set new_active to bond's MAC
1014 write_unlock_bh(&bond->curr_slave_lock);
1015 read_unlock(&bond->lock);
1018 memcpy(tmp_mac, new_active->dev->dev_addr, ETH_ALEN);
1019 memcpy(saddr.sa_data, old_active->dev->dev_addr,
1021 saddr.sa_family = new_active->dev->type;
1023 memcpy(saddr.sa_data, bond->dev->dev_addr, ETH_ALEN);
1024 saddr.sa_family = bond->dev->type;
1027 rv = dev_set_mac_address(new_active->dev, &saddr);
1029 pr_err("%s: Error %d setting MAC of slave %s\n",
1030 bond->dev->name, -rv, new_active->dev->name);
1037 memcpy(saddr.sa_data, tmp_mac, ETH_ALEN);
1038 saddr.sa_family = old_active->dev->type;
1040 rv = dev_set_mac_address(old_active->dev, &saddr);
1042 pr_err("%s: Error %d setting MAC of slave %s\n",
1043 bond->dev->name, -rv, new_active->dev->name);
1045 read_lock(&bond->lock);
1046 write_lock_bh(&bond->curr_slave_lock);
1049 pr_err("%s: bond_do_fail_over_mac impossible: bad policy %d\n",
1050 bond->dev->name, bond->params.fail_over_mac);
1056 static bool bond_should_change_active(struct bonding *bond)
1058 struct slave *prim = bond->primary_slave;
1059 struct slave *curr = bond->curr_active_slave;
1061 if (!prim || !curr || curr->link != BOND_LINK_UP)
1063 if (bond->force_primary) {
1064 bond->force_primary = false;
1067 if (bond->params.primary_reselect == BOND_PRI_RESELECT_BETTER &&
1068 (prim->speed < curr->speed ||
1069 (prim->speed == curr->speed && prim->duplex <= curr->duplex)))
1071 if (bond->params.primary_reselect == BOND_PRI_RESELECT_FAILURE)
1077 * find_best_interface - select the best available slave to be the active one
1078 * @bond: our bonding struct
1080 * Warning: Caller must hold curr_slave_lock for writing.
1082 static struct slave *bond_find_best_slave(struct bonding *bond)
1084 struct slave *new_active, *old_active;
1085 struct slave *bestslave = NULL;
1086 int mintime = bond->params.updelay;
1089 new_active = bond->curr_active_slave;
1091 if (!new_active) { /* there were no active slaves left */
1092 if (bond->slave_cnt > 0) /* found one slave */
1093 new_active = bond->first_slave;
1095 return NULL; /* still no slave, return NULL */
1098 if ((bond->primary_slave) &&
1099 bond->primary_slave->link == BOND_LINK_UP &&
1100 bond_should_change_active(bond)) {
1101 new_active = bond->primary_slave;
1104 /* remember where to stop iterating over the slaves */
1105 old_active = new_active;
1107 bond_for_each_slave_from(bond, new_active, i, old_active) {
1108 if (new_active->link == BOND_LINK_UP) {
1110 } else if (new_active->link == BOND_LINK_BACK &&
1111 IS_UP(new_active->dev)) {
1112 /* link up, but waiting for stabilization */
1113 if (new_active->delay < mintime) {
1114 mintime = new_active->delay;
1115 bestslave = new_active;
1124 * change_active_interface - change the active slave into the specified one
1125 * @bond: our bonding struct
1126 * @new: the new slave to make the active one
1128 * Set the new slave to the bond's settings and unset them on the old
1129 * curr_active_slave.
1130 * Setting include flags, mc-list, promiscuity, allmulti, etc.
1132 * If @new's link state is %BOND_LINK_BACK we'll set it to %BOND_LINK_UP,
1133 * because it is apparently the best available slave we have, even though its
1134 * updelay hasn't timed out yet.
1136 * If new_active is not NULL, caller must hold bond->lock for read and
1137 * curr_slave_lock for write_bh.
1139 void bond_change_active_slave(struct bonding *bond, struct slave *new_active)
1141 struct slave *old_active = bond->curr_active_slave;
1143 if (old_active == new_active)
1147 new_active->jiffies = jiffies;
1149 if (new_active->link == BOND_LINK_BACK) {
1150 if (USES_PRIMARY(bond->params.mode)) {
1151 pr_info("%s: making interface %s the new active one %d ms earlier.\n",
1152 bond->dev->name, new_active->dev->name,
1153 (bond->params.updelay - new_active->delay) * bond->params.miimon);
1156 new_active->delay = 0;
1157 new_active->link = BOND_LINK_UP;
1159 if (bond->params.mode == BOND_MODE_8023AD)
1160 bond_3ad_handle_link_change(new_active, BOND_LINK_UP);
1162 if (bond_is_lb(bond))
1163 bond_alb_handle_link_change(bond, new_active, BOND_LINK_UP);
1165 if (USES_PRIMARY(bond->params.mode)) {
1166 pr_info("%s: making interface %s the new active one.\n",
1167 bond->dev->name, new_active->dev->name);
1172 if (USES_PRIMARY(bond->params.mode))
1173 bond_mc_swap(bond, new_active, old_active);
1175 if (bond_is_lb(bond)) {
1176 bond_alb_handle_active_change(bond, new_active);
1178 bond_set_slave_inactive_flags(old_active);
1180 bond_set_slave_active_flags(new_active);
1182 bond->curr_active_slave = new_active;
1185 if (bond->params.mode == BOND_MODE_ACTIVEBACKUP) {
1187 bond_set_slave_inactive_flags(old_active);
1190 bond_set_slave_active_flags(new_active);
1192 if (bond->params.fail_over_mac)
1193 bond_do_fail_over_mac(bond, new_active,
1196 bond->send_grat_arp = bond->params.num_grat_arp;
1197 bond_send_gratuitous_arp(bond);
1199 bond->send_unsol_na = bond->params.num_unsol_na;
1200 bond_send_unsolicited_na(bond);
1202 write_unlock_bh(&bond->curr_slave_lock);
1203 read_unlock(&bond->lock);
1205 netdev_bonding_change(bond->dev, NETDEV_BONDING_FAILOVER);
1207 read_lock(&bond->lock);
1208 write_lock_bh(&bond->curr_slave_lock);
1212 /* resend IGMP joins since active slave has changed or
1213 * all were sent on curr_active_slave */
1214 if ((USES_PRIMARY(bond->params.mode) && new_active) ||
1215 bond->params.mode == BOND_MODE_ROUNDROBIN) {
1216 queue_delayed_work(bond->wq, &bond->mcast_work, 0);
1221 * bond_select_active_slave - select a new active slave, if needed
1222 * @bond: our bonding struct
1224 * This functions should be called when one of the following occurs:
1225 * - The old curr_active_slave has been released or lost its link.
1226 * - The primary_slave has got its link back.
1227 * - A slave has got its link back and there's no old curr_active_slave.
1229 * Caller must hold bond->lock for read and curr_slave_lock for write_bh.
1231 void bond_select_active_slave(struct bonding *bond)
1233 struct slave *best_slave;
1236 best_slave = bond_find_best_slave(bond);
1237 if (best_slave != bond->curr_active_slave) {
1238 bond_change_active_slave(bond, best_slave);
1239 rv = bond_set_carrier(bond);
1243 if (netif_carrier_ok(bond->dev)) {
1244 pr_info("%s: first active interface up!\n",
1247 pr_info("%s: now running without any active interface !\n",
1253 /*--------------------------- slave list handling ---------------------------*/
1256 * This function attaches the slave to the end of list.
1258 * bond->lock held for writing by caller.
1260 static void bond_attach_slave(struct bonding *bond, struct slave *new_slave)
1262 if (bond->first_slave == NULL) { /* attaching the first slave */
1263 new_slave->next = new_slave;
1264 new_slave->prev = new_slave;
1265 bond->first_slave = new_slave;
1267 new_slave->next = bond->first_slave;
1268 new_slave->prev = bond->first_slave->prev;
1269 new_slave->next->prev = new_slave;
1270 new_slave->prev->next = new_slave;
1277 * This function detaches the slave from the list.
1278 * WARNING: no check is made to verify if the slave effectively
1279 * belongs to <bond>.
1280 * Nothing is freed on return, structures are just unchained.
1281 * If any slave pointer in bond was pointing to <slave>,
1282 * it should be changed by the calling function.
1284 * bond->lock held for writing by caller.
1286 static void bond_detach_slave(struct bonding *bond, struct slave *slave)
1289 slave->next->prev = slave->prev;
1292 slave->prev->next = slave->next;
1294 if (bond->first_slave == slave) { /* slave is the first slave */
1295 if (bond->slave_cnt > 1) { /* there are more slave */
1296 bond->first_slave = slave->next;
1298 bond->first_slave = NULL; /* slave was the last one */
1307 #ifdef CONFIG_NET_POLL_CONTROLLER
1309 * You must hold read lock on bond->lock before calling this.
1311 static bool slaves_support_netpoll(struct net_device *bond_dev)
1313 struct bonding *bond = netdev_priv(bond_dev);
1314 struct slave *slave;
1318 bond_for_each_slave(bond, slave, i) {
1319 if ((slave->dev->priv_flags & IFF_DISABLE_NETPOLL) ||
1320 !slave->dev->netdev_ops->ndo_poll_controller)
1323 return i != 0 && ret;
1326 static void bond_poll_controller(struct net_device *bond_dev)
1328 struct net_device *dev = bond_dev->npinfo->netpoll->real_dev;
1329 if (dev != bond_dev)
1330 netpoll_poll_dev(dev);
1333 static void bond_netpoll_cleanup(struct net_device *bond_dev)
1335 struct bonding *bond = netdev_priv(bond_dev);
1336 struct slave *slave;
1337 const struct net_device_ops *ops;
1340 read_lock(&bond->lock);
1341 bond_dev->npinfo = NULL;
1342 bond_for_each_slave(bond, slave, i) {
1344 ops = slave->dev->netdev_ops;
1345 if (ops->ndo_netpoll_cleanup)
1346 ops->ndo_netpoll_cleanup(slave->dev);
1348 slave->dev->npinfo = NULL;
1351 read_unlock(&bond->lock);
1356 static void bond_netpoll_cleanup(struct net_device *bond_dev)
1362 /*---------------------------------- IOCTL ----------------------------------*/
1364 static int bond_sethwaddr(struct net_device *bond_dev,
1365 struct net_device *slave_dev)
1367 pr_debug("bond_dev=%p\n", bond_dev);
1368 pr_debug("slave_dev=%p\n", slave_dev);
1369 pr_debug("slave_dev->addr_len=%d\n", slave_dev->addr_len);
1370 memcpy(bond_dev->dev_addr, slave_dev->dev_addr, slave_dev->addr_len);
1374 #define BOND_VLAN_FEATURES \
1375 (NETIF_F_VLAN_CHALLENGED | NETIF_F_HW_VLAN_RX | NETIF_F_HW_VLAN_TX | \
1376 NETIF_F_HW_VLAN_FILTER)
1379 * Compute the common dev->feature set available to all slaves. Some
1380 * feature bits are managed elsewhere, so preserve those feature bits
1381 * on the master device.
1383 static int bond_compute_features(struct bonding *bond)
1385 struct slave *slave;
1386 struct net_device *bond_dev = bond->dev;
1387 unsigned long features = bond_dev->features;
1388 unsigned long vlan_features = 0;
1389 unsigned short max_hard_header_len = max((u16)ETH_HLEN,
1390 bond_dev->hard_header_len);
1393 features &= ~(NETIF_F_ALL_CSUM | BOND_VLAN_FEATURES);
1394 features |= NETIF_F_GSO_MASK | NETIF_F_NO_CSUM;
1396 if (!bond->first_slave)
1399 features &= ~NETIF_F_ONE_FOR_ALL;
1401 vlan_features = bond->first_slave->dev->vlan_features;
1402 bond_for_each_slave(bond, slave, i) {
1403 features = netdev_increment_features(features,
1404 slave->dev->features,
1405 NETIF_F_ONE_FOR_ALL);
1406 vlan_features = netdev_increment_features(vlan_features,
1407 slave->dev->vlan_features,
1408 NETIF_F_ONE_FOR_ALL);
1409 if (slave->dev->hard_header_len > max_hard_header_len)
1410 max_hard_header_len = slave->dev->hard_header_len;
1414 features |= (bond_dev->features & BOND_VLAN_FEATURES);
1415 bond_dev->features = netdev_fix_features(features, NULL);
1416 bond_dev->vlan_features = netdev_fix_features(vlan_features, NULL);
1417 bond_dev->hard_header_len = max_hard_header_len;
1422 static void bond_setup_by_slave(struct net_device *bond_dev,
1423 struct net_device *slave_dev)
1425 struct bonding *bond = netdev_priv(bond_dev);
1427 bond_dev->header_ops = slave_dev->header_ops;
1429 bond_dev->type = slave_dev->type;
1430 bond_dev->hard_header_len = slave_dev->hard_header_len;
1431 bond_dev->addr_len = slave_dev->addr_len;
1433 memcpy(bond_dev->broadcast, slave_dev->broadcast,
1434 slave_dev->addr_len);
1435 bond->setup_by_slave = 1;
1438 /* enslave device <slave> to bond device <master> */
1439 int bond_enslave(struct net_device *bond_dev, struct net_device *slave_dev)
1441 struct bonding *bond = netdev_priv(bond_dev);
1442 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
1443 struct slave *new_slave = NULL;
1444 struct netdev_hw_addr *ha;
1445 struct sockaddr addr;
1447 int old_features = bond_dev->features;
1450 if (!bond->params.use_carrier && slave_dev->ethtool_ops == NULL &&
1451 slave_ops->ndo_do_ioctl == NULL) {
1452 pr_warning("%s: Warning: no link monitoring support for %s\n",
1453 bond_dev->name, slave_dev->name);
1456 /* bond must be initialized by bond_open() before enslaving */
1457 if (!(bond_dev->flags & IFF_UP)) {
1458 pr_warning("%s: master_dev is not up in bond_enslave\n",
1462 /* already enslaved */
1463 if (slave_dev->flags & IFF_SLAVE) {
1464 pr_debug("Error, Device was already enslaved\n");
1468 /* vlan challenged mutual exclusion */
1469 /* no need to lock since we're protected by rtnl_lock */
1470 if (slave_dev->features & NETIF_F_VLAN_CHALLENGED) {
1471 pr_debug("%s: NETIF_F_VLAN_CHALLENGED\n", slave_dev->name);
1473 pr_err("%s: Error: cannot enslave VLAN challenged slave %s on VLAN enabled bond %s\n",
1474 bond_dev->name, slave_dev->name, bond_dev->name);
1477 pr_warning("%s: Warning: enslaved VLAN challenged slave %s. Adding VLANs will be blocked as long as %s is part of bond %s\n",
1478 bond_dev->name, slave_dev->name,
1479 slave_dev->name, bond_dev->name);
1480 bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
1483 pr_debug("%s: ! NETIF_F_VLAN_CHALLENGED\n", slave_dev->name);
1484 if (bond->slave_cnt == 0) {
1485 /* First slave, and it is not VLAN challenged,
1486 * so remove the block of adding VLANs over the bond.
1488 bond_dev->features &= ~NETIF_F_VLAN_CHALLENGED;
1493 * Old ifenslave binaries are no longer supported. These can
1494 * be identified with moderate accuracy by the state of the slave:
1495 * the current ifenslave will set the interface down prior to
1496 * enslaving it; the old ifenslave will not.
1498 if ((slave_dev->flags & IFF_UP)) {
1499 pr_err("%s is up. This may be due to an out of date ifenslave.\n",
1502 goto err_undo_flags;
1505 /* set bonding device ether type by slave - bonding netdevices are
1506 * created with ether_setup, so when the slave type is not ARPHRD_ETHER
1507 * there is a need to override some of the type dependent attribs/funcs.
1509 * bond ether type mutual exclusion - don't allow slaves of dissimilar
1510 * ether type (eg ARPHRD_ETHER and ARPHRD_INFINIBAND) share the same bond
1512 if (bond->slave_cnt == 0) {
1513 if (bond_dev->type != slave_dev->type) {
1514 pr_debug("%s: change device type from %d to %d\n",
1516 bond_dev->type, slave_dev->type);
1518 res = netdev_bonding_change(bond_dev,
1519 NETDEV_PRE_TYPE_CHANGE);
1520 res = notifier_to_errno(res);
1522 pr_err("%s: refused to change device type\n",
1525 goto err_undo_flags;
1528 /* Flush unicast and multicast addresses */
1529 dev_uc_flush(bond_dev);
1530 dev_mc_flush(bond_dev);
1532 if (slave_dev->type != ARPHRD_ETHER)
1533 bond_setup_by_slave(bond_dev, slave_dev);
1535 ether_setup(bond_dev);
1537 netdev_bonding_change(bond_dev,
1538 NETDEV_POST_TYPE_CHANGE);
1540 } else if (bond_dev->type != slave_dev->type) {
1541 pr_err("%s ether type (%d) is different from other slaves (%d), can not enslave it.\n",
1543 slave_dev->type, bond_dev->type);
1545 goto err_undo_flags;
1548 if (slave_ops->ndo_set_mac_address == NULL) {
1549 if (bond->slave_cnt == 0) {
1550 pr_warning("%s: Warning: The first slave device specified does not support setting the MAC address. Setting fail_over_mac to active.",
1552 bond->params.fail_over_mac = BOND_FOM_ACTIVE;
1553 } else if (bond->params.fail_over_mac != BOND_FOM_ACTIVE) {
1554 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",
1557 goto err_undo_flags;
1561 /* If this is the first slave, then we need to set the master's hardware
1562 * address to be the same as the slave's. */
1563 if (bond->slave_cnt == 0)
1564 memcpy(bond->dev->dev_addr, slave_dev->dev_addr,
1565 slave_dev->addr_len);
1568 new_slave = kzalloc(sizeof(struct slave), GFP_KERNEL);
1571 goto err_undo_flags;
1575 * Set the new_slave's queue_id to be zero. Queue ID mapping
1576 * is set via sysfs or module option if desired.
1578 new_slave->queue_id = 0;
1580 /* Save slave's original mtu and then set it to match the bond */
1581 new_slave->original_mtu = slave_dev->mtu;
1582 res = dev_set_mtu(slave_dev, bond->dev->mtu);
1584 pr_debug("Error %d calling dev_set_mtu\n", res);
1589 * Save slave's original ("permanent") mac address for modes
1590 * that need it, and for restoring it upon release, and then
1591 * set it to the master's address
1593 memcpy(new_slave->perm_hwaddr, slave_dev->dev_addr, ETH_ALEN);
1595 if (!bond->params.fail_over_mac) {
1597 * Set slave to master's mac address. The application already
1598 * set the master's mac address to that of the first slave
1600 memcpy(addr.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
1601 addr.sa_family = slave_dev->type;
1602 res = dev_set_mac_address(slave_dev, &addr);
1604 pr_debug("Error %d calling set_mac_address\n", res);
1605 goto err_restore_mtu;
1609 res = netdev_set_master(slave_dev, bond_dev);
1611 pr_debug("Error %d calling netdev_set_master\n", res);
1612 goto err_restore_mac;
1614 /* open the slave since the application closed it */
1615 res = dev_open(slave_dev);
1617 pr_debug("Opening slave %s failed\n", slave_dev->name);
1618 goto err_unset_master;
1621 new_slave->dev = slave_dev;
1622 slave_dev->priv_flags |= IFF_BONDING;
1624 if (bond_is_lb(bond)) {
1625 /* bond_alb_init_slave() must be called before all other stages since
1626 * it might fail and we do not want to have to undo everything
1628 res = bond_alb_init_slave(bond, new_slave);
1633 /* If the mode USES_PRIMARY, then the new slave gets the
1634 * master's promisc (and mc) settings only if it becomes the
1635 * curr_active_slave, and that is taken care of later when calling
1636 * bond_change_active()
1638 if (!USES_PRIMARY(bond->params.mode)) {
1639 /* set promiscuity level to new slave */
1640 if (bond_dev->flags & IFF_PROMISC) {
1641 res = dev_set_promiscuity(slave_dev, 1);
1646 /* set allmulti level to new slave */
1647 if (bond_dev->flags & IFF_ALLMULTI) {
1648 res = dev_set_allmulti(slave_dev, 1);
1653 netif_addr_lock_bh(bond_dev);
1654 /* upload master's mc_list to new slave */
1655 netdev_for_each_mc_addr(ha, bond_dev)
1656 dev_mc_add(slave_dev, ha->addr);
1657 netif_addr_unlock_bh(bond_dev);
1660 if (bond->params.mode == BOND_MODE_8023AD) {
1661 /* add lacpdu mc addr to mc list */
1662 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
1664 dev_mc_add(slave_dev, lacpdu_multicast);
1667 bond_add_vlans_on_slave(bond, slave_dev);
1669 write_lock_bh(&bond->lock);
1671 bond_attach_slave(bond, new_slave);
1673 new_slave->delay = 0;
1674 new_slave->link_failure_count = 0;
1676 bond_compute_features(bond);
1678 write_unlock_bh(&bond->lock);
1680 read_lock(&bond->lock);
1682 new_slave->last_arp_rx = jiffies;
1684 if (bond->params.miimon && !bond->params.use_carrier) {
1685 link_reporting = bond_check_dev_link(bond, slave_dev, 1);
1687 if ((link_reporting == -1) && !bond->params.arp_interval) {
1689 * miimon is set but a bonded network driver
1690 * does not support ETHTOOL/MII and
1691 * arp_interval is not set. Note: if
1692 * use_carrier is enabled, we will never go
1693 * here (because netif_carrier is always
1694 * supported); thus, we don't need to change
1695 * the messages for netif_carrier.
1697 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",
1698 bond_dev->name, slave_dev->name);
1699 } else if (link_reporting == -1) {
1700 /* unable get link status using mii/ethtool */
1701 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",
1702 bond_dev->name, slave_dev->name);
1706 /* check for initial state */
1707 if (!bond->params.miimon ||
1708 (bond_check_dev_link(bond, slave_dev, 0) == BMSR_LSTATUS)) {
1709 if (bond->params.updelay) {
1710 pr_debug("Initial state of slave_dev is BOND_LINK_BACK\n");
1711 new_slave->link = BOND_LINK_BACK;
1712 new_slave->delay = bond->params.updelay;
1714 pr_debug("Initial state of slave_dev is BOND_LINK_UP\n");
1715 new_slave->link = BOND_LINK_UP;
1717 new_slave->jiffies = jiffies;
1719 pr_debug("Initial state of slave_dev is BOND_LINK_DOWN\n");
1720 new_slave->link = BOND_LINK_DOWN;
1723 if (bond_update_speed_duplex(new_slave) &&
1724 (new_slave->link != BOND_LINK_DOWN)) {
1725 pr_warning("%s: Warning: failed to get speed and duplex from %s, assumed to be 100Mb/sec and Full.\n",
1726 bond_dev->name, new_slave->dev->name);
1728 if (bond->params.mode == BOND_MODE_8023AD) {
1729 pr_warning("%s: Warning: Operation of 802.3ad mode requires ETHTOOL support in base driver for proper aggregator selection.\n",
1734 if (USES_PRIMARY(bond->params.mode) && bond->params.primary[0]) {
1735 /* if there is a primary slave, remember it */
1736 if (strcmp(bond->params.primary, new_slave->dev->name) == 0) {
1737 bond->primary_slave = new_slave;
1738 bond->force_primary = true;
1742 write_lock_bh(&bond->curr_slave_lock);
1744 switch (bond->params.mode) {
1745 case BOND_MODE_ACTIVEBACKUP:
1746 bond_set_slave_inactive_flags(new_slave);
1747 bond_select_active_slave(bond);
1749 case BOND_MODE_8023AD:
1750 /* in 802.3ad mode, the internal mechanism
1751 * will activate the slaves in the selected
1754 bond_set_slave_inactive_flags(new_slave);
1755 /* if this is the first slave */
1756 if (bond->slave_cnt == 1) {
1757 SLAVE_AD_INFO(new_slave).id = 1;
1758 /* Initialize AD with the number of times that the AD timer is called in 1 second
1759 * can be called only after the mac address of the bond is set
1761 bond_3ad_initialize(bond, 1000/AD_TIMER_INTERVAL,
1762 bond->params.lacp_fast);
1764 SLAVE_AD_INFO(new_slave).id =
1765 SLAVE_AD_INFO(new_slave->prev).id + 1;
1768 bond_3ad_bind_slave(new_slave);
1772 new_slave->state = BOND_STATE_ACTIVE;
1773 bond_set_slave_inactive_flags(new_slave);
1774 bond_select_active_slave(bond);
1777 pr_debug("This slave is always active in trunk mode\n");
1779 /* always active in trunk mode */
1780 new_slave->state = BOND_STATE_ACTIVE;
1782 /* In trunking mode there is little meaning to curr_active_slave
1783 * anyway (it holds no special properties of the bond device),
1784 * so we can change it without calling change_active_interface()
1786 if (!bond->curr_active_slave)
1787 bond->curr_active_slave = new_slave;
1790 } /* switch(bond_mode) */
1792 write_unlock_bh(&bond->curr_slave_lock);
1794 bond_set_carrier(bond);
1796 #ifdef CONFIG_NET_POLL_CONTROLLER
1798 * Netpoll and bonding is broken, make sure it is not initialized
1799 * until it is fixed.
1801 if (disable_netpoll) {
1802 bond_dev->priv_flags |= IFF_DISABLE_NETPOLL;
1804 if (slaves_support_netpoll(bond_dev)) {
1805 bond_dev->priv_flags &= ~IFF_DISABLE_NETPOLL;
1806 if (bond_dev->npinfo)
1807 slave_dev->npinfo = bond_dev->npinfo;
1808 } else if (!(bond_dev->priv_flags & IFF_DISABLE_NETPOLL)) {
1809 bond_dev->priv_flags |= IFF_DISABLE_NETPOLL;
1810 pr_info("New slave device %s does not support netpoll\n",
1812 pr_info("Disabling netpoll support for %s\n", bond_dev->name);
1816 read_unlock(&bond->lock);
1818 res = bond_create_slave_symlinks(bond_dev, slave_dev);
1822 pr_info("%s: enslaving %s as a%s interface with a%s link.\n",
1823 bond_dev->name, slave_dev->name,
1824 new_slave->state == BOND_STATE_ACTIVE ? "n active" : " backup",
1825 new_slave->link != BOND_LINK_DOWN ? "n up" : " down");
1827 /* enslave is successful */
1830 /* Undo stages on error */
1832 dev_close(slave_dev);
1835 netdev_set_master(slave_dev, NULL);
1838 if (!bond->params.fail_over_mac) {
1839 /* XXX TODO - fom follow mode needs to change master's
1840 * MAC if this slave's MAC is in use by the bond, or at
1841 * least print a warning.
1843 memcpy(addr.sa_data, new_slave->perm_hwaddr, ETH_ALEN);
1844 addr.sa_family = slave_dev->type;
1845 dev_set_mac_address(slave_dev, &addr);
1849 dev_set_mtu(slave_dev, new_slave->original_mtu);
1855 bond_dev->features = old_features;
1861 * Try to release the slave device <slave> from the bond device <master>
1862 * It is legal to access curr_active_slave without a lock because all the function
1865 * The rules for slave state should be:
1866 * for Active/Backup:
1867 * Active stays on all backups go down
1868 * for Bonded connections:
1869 * The first up interface should be left on and all others downed.
1871 int bond_release(struct net_device *bond_dev, struct net_device *slave_dev)
1873 struct bonding *bond = netdev_priv(bond_dev);
1874 struct slave *slave, *oldcurrent;
1875 struct sockaddr addr;
1877 /* slave is not a slave or master is not master of this slave */
1878 if (!(slave_dev->flags & IFF_SLAVE) ||
1879 (slave_dev->master != bond_dev)) {
1880 pr_err("%s: Error: cannot release %s.\n",
1881 bond_dev->name, slave_dev->name);
1885 netdev_bonding_change(bond_dev, NETDEV_BONDING_DESLAVE);
1886 write_lock_bh(&bond->lock);
1888 slave = bond_get_slave_by_dev(bond, slave_dev);
1890 /* not a slave of this bond */
1891 pr_info("%s: %s not enslaved\n",
1892 bond_dev->name, slave_dev->name);
1893 write_unlock_bh(&bond->lock);
1897 if (!bond->params.fail_over_mac) {
1898 if (!compare_ether_addr(bond_dev->dev_addr, slave->perm_hwaddr) &&
1899 bond->slave_cnt > 1)
1900 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",
1901 bond_dev->name, slave_dev->name,
1903 bond_dev->name, slave_dev->name);
1906 /* Inform AD package of unbinding of slave. */
1907 if (bond->params.mode == BOND_MODE_8023AD) {
1908 /* must be called before the slave is
1909 * detached from the list
1911 bond_3ad_unbind_slave(slave);
1914 pr_info("%s: releasing %s interface %s\n",
1916 (slave->state == BOND_STATE_ACTIVE) ? "active" : "backup",
1919 oldcurrent = bond->curr_active_slave;
1921 bond->current_arp_slave = NULL;
1923 /* release the slave from its bond */
1924 bond_detach_slave(bond, slave);
1926 bond_compute_features(bond);
1928 if (bond->primary_slave == slave)
1929 bond->primary_slave = NULL;
1931 if (oldcurrent == slave)
1932 bond_change_active_slave(bond, NULL);
1934 if (bond_is_lb(bond)) {
1935 /* Must be called only after the slave has been
1936 * detached from the list and the curr_active_slave
1937 * has been cleared (if our_slave == old_current),
1938 * but before a new active slave is selected.
1940 write_unlock_bh(&bond->lock);
1941 bond_alb_deinit_slave(bond, slave);
1942 write_lock_bh(&bond->lock);
1945 if (oldcurrent == slave) {
1947 * Note that we hold RTNL over this sequence, so there
1948 * is no concern that another slave add/remove event
1951 write_unlock_bh(&bond->lock);
1952 read_lock(&bond->lock);
1953 write_lock_bh(&bond->curr_slave_lock);
1955 bond_select_active_slave(bond);
1957 write_unlock_bh(&bond->curr_slave_lock);
1958 read_unlock(&bond->lock);
1959 write_lock_bh(&bond->lock);
1962 if (bond->slave_cnt == 0) {
1963 bond_set_carrier(bond);
1965 /* if the last slave was removed, zero the mac address
1966 * of the master so it will be set by the application
1967 * to the mac address of the first slave
1969 memset(bond_dev->dev_addr, 0, bond_dev->addr_len);
1972 bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
1974 pr_warning("%s: Warning: clearing HW address of %s while it still has VLANs.\n",
1975 bond_dev->name, bond_dev->name);
1976 pr_warning("%s: When re-adding slaves, make sure the bond's HW address matches its VLANs'.\n",
1979 } else if ((bond_dev->features & NETIF_F_VLAN_CHALLENGED) &&
1980 !bond_has_challenged_slaves(bond)) {
1981 pr_info("%s: last VLAN challenged slave %s left bond %s. VLAN blocking is removed\n",
1982 bond_dev->name, slave_dev->name, bond_dev->name);
1983 bond_dev->features &= ~NETIF_F_VLAN_CHALLENGED;
1986 write_unlock_bh(&bond->lock);
1988 /* must do this from outside any spinlocks */
1989 bond_destroy_slave_symlinks(bond_dev, slave_dev);
1991 bond_del_vlans_from_slave(bond, slave_dev);
1993 /* If the mode USES_PRIMARY, then we should only remove its
1994 * promisc and mc settings if it was the curr_active_slave, but that was
1995 * already taken care of above when we detached the slave
1997 if (!USES_PRIMARY(bond->params.mode)) {
1998 /* unset promiscuity level from slave */
1999 if (bond_dev->flags & IFF_PROMISC)
2000 dev_set_promiscuity(slave_dev, -1);
2002 /* unset allmulti level from slave */
2003 if (bond_dev->flags & IFF_ALLMULTI)
2004 dev_set_allmulti(slave_dev, -1);
2006 /* flush master's mc_list from slave */
2007 netif_addr_lock_bh(bond_dev);
2008 bond_mc_list_flush(bond_dev, slave_dev);
2009 netif_addr_unlock_bh(bond_dev);
2012 netdev_set_master(slave_dev, NULL);
2014 #ifdef CONFIG_NET_POLL_CONTROLLER
2015 read_lock_bh(&bond->lock);
2017 /* Make sure netpoll over stays disabled until fixed. */
2018 if (!disable_netpoll)
2019 if (slaves_support_netpoll(bond_dev))
2020 bond_dev->priv_flags &= ~IFF_DISABLE_NETPOLL;
2021 read_unlock_bh(&bond->lock);
2022 if (slave_dev->netdev_ops->ndo_netpoll_cleanup)
2023 slave_dev->netdev_ops->ndo_netpoll_cleanup(slave_dev);
2025 slave_dev->npinfo = NULL;
2028 /* close slave before restoring its mac address */
2029 dev_close(slave_dev);
2031 if (bond->params.fail_over_mac != BOND_FOM_ACTIVE) {
2032 /* restore original ("permanent") mac address */
2033 memcpy(addr.sa_data, slave->perm_hwaddr, ETH_ALEN);
2034 addr.sa_family = slave_dev->type;
2035 dev_set_mac_address(slave_dev, &addr);
2038 dev_set_mtu(slave_dev, slave->original_mtu);
2040 slave_dev->priv_flags &= ~(IFF_MASTER_8023AD | IFF_MASTER_ALB |
2041 IFF_SLAVE_INACTIVE | IFF_BONDING |
2046 return 0; /* deletion OK */
2050 * First release a slave and than destroy the bond if no more slaves are left.
2051 * Must be under rtnl_lock when this function is called.
2053 int bond_release_and_destroy(struct net_device *bond_dev,
2054 struct net_device *slave_dev)
2056 struct bonding *bond = netdev_priv(bond_dev);
2059 ret = bond_release(bond_dev, slave_dev);
2060 if ((ret == 0) && (bond->slave_cnt == 0)) {
2061 pr_info("%s: destroying bond %s.\n",
2062 bond_dev->name, bond_dev->name);
2063 unregister_netdevice(bond_dev);
2069 * This function releases all slaves.
2071 static int bond_release_all(struct net_device *bond_dev)
2073 struct bonding *bond = netdev_priv(bond_dev);
2074 struct slave *slave;
2075 struct net_device *slave_dev;
2076 struct sockaddr addr;
2078 write_lock_bh(&bond->lock);
2080 netif_carrier_off(bond_dev);
2082 if (bond->slave_cnt == 0)
2085 bond->current_arp_slave = NULL;
2086 bond->primary_slave = NULL;
2087 bond_change_active_slave(bond, NULL);
2089 while ((slave = bond->first_slave) != NULL) {
2090 /* Inform AD package of unbinding of slave
2091 * before slave is detached from the list.
2093 if (bond->params.mode == BOND_MODE_8023AD)
2094 bond_3ad_unbind_slave(slave);
2096 slave_dev = slave->dev;
2097 bond_detach_slave(bond, slave);
2099 /* now that the slave is detached, unlock and perform
2100 * all the undo steps that should not be called from
2103 write_unlock_bh(&bond->lock);
2105 if (bond_is_lb(bond)) {
2106 /* must be called only after the slave
2107 * has been detached from the list
2109 bond_alb_deinit_slave(bond, slave);
2112 bond_compute_features(bond);
2114 bond_destroy_slave_symlinks(bond_dev, slave_dev);
2115 bond_del_vlans_from_slave(bond, slave_dev);
2117 /* If the mode USES_PRIMARY, then we should only remove its
2118 * promisc and mc settings if it was the curr_active_slave, but that was
2119 * already taken care of above when we detached the slave
2121 if (!USES_PRIMARY(bond->params.mode)) {
2122 /* unset promiscuity level from slave */
2123 if (bond_dev->flags & IFF_PROMISC)
2124 dev_set_promiscuity(slave_dev, -1);
2126 /* unset allmulti level from slave */
2127 if (bond_dev->flags & IFF_ALLMULTI)
2128 dev_set_allmulti(slave_dev, -1);
2130 /* flush master's mc_list from slave */
2131 netif_addr_lock_bh(bond_dev);
2132 bond_mc_list_flush(bond_dev, slave_dev);
2133 netif_addr_unlock_bh(bond_dev);
2136 netdev_set_master(slave_dev, NULL);
2138 /* close slave before restoring its mac address */
2139 dev_close(slave_dev);
2141 if (!bond->params.fail_over_mac) {
2142 /* restore original ("permanent") mac address*/
2143 memcpy(addr.sa_data, slave->perm_hwaddr, ETH_ALEN);
2144 addr.sa_family = slave_dev->type;
2145 dev_set_mac_address(slave_dev, &addr);
2148 slave_dev->priv_flags &= ~(IFF_MASTER_8023AD | IFF_MASTER_ALB |
2149 IFF_SLAVE_INACTIVE);
2153 /* re-acquire the lock before getting the next slave */
2154 write_lock_bh(&bond->lock);
2157 /* zero the mac address of the master so it will be
2158 * set by the application to the mac address of the
2161 memset(bond_dev->dev_addr, 0, bond_dev->addr_len);
2164 bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
2166 pr_warning("%s: Warning: clearing HW address of %s while it still has VLANs.\n",
2167 bond_dev->name, bond_dev->name);
2168 pr_warning("%s: When re-adding slaves, make sure the bond's HW address matches its VLANs'.\n",
2172 pr_info("%s: released all slaves\n", bond_dev->name);
2175 write_unlock_bh(&bond->lock);
2181 * This function changes the active slave to slave <slave_dev>.
2182 * It returns -EINVAL in the following cases.
2183 * - <slave_dev> is not found in the list.
2184 * - There is not active slave now.
2185 * - <slave_dev> is already active.
2186 * - The link state of <slave_dev> is not BOND_LINK_UP.
2187 * - <slave_dev> is not running.
2188 * In these cases, this function does nothing.
2189 * In the other cases, current_slave pointer is changed and 0 is returned.
2191 static int bond_ioctl_change_active(struct net_device *bond_dev, struct net_device *slave_dev)
2193 struct bonding *bond = netdev_priv(bond_dev);
2194 struct slave *old_active = NULL;
2195 struct slave *new_active = NULL;
2198 if (!USES_PRIMARY(bond->params.mode))
2201 /* Verify that master_dev is indeed the master of slave_dev */
2202 if (!(slave_dev->flags & IFF_SLAVE) || (slave_dev->master != bond_dev))
2205 read_lock(&bond->lock);
2207 read_lock(&bond->curr_slave_lock);
2208 old_active = bond->curr_active_slave;
2209 read_unlock(&bond->curr_slave_lock);
2211 new_active = bond_get_slave_by_dev(bond, slave_dev);
2214 * Changing to the current active: do nothing; return success.
2216 if (new_active && (new_active == old_active)) {
2217 read_unlock(&bond->lock);
2223 (new_active->link == BOND_LINK_UP) &&
2224 IS_UP(new_active->dev)) {
2225 write_lock_bh(&bond->curr_slave_lock);
2226 bond_change_active_slave(bond, new_active);
2227 write_unlock_bh(&bond->curr_slave_lock);
2231 read_unlock(&bond->lock);
2236 static int bond_info_query(struct net_device *bond_dev, struct ifbond *info)
2238 struct bonding *bond = netdev_priv(bond_dev);
2240 info->bond_mode = bond->params.mode;
2241 info->miimon = bond->params.miimon;
2243 read_lock(&bond->lock);
2244 info->num_slaves = bond->slave_cnt;
2245 read_unlock(&bond->lock);
2250 static int bond_slave_info_query(struct net_device *bond_dev, struct ifslave *info)
2252 struct bonding *bond = netdev_priv(bond_dev);
2253 struct slave *slave;
2254 int i, res = -ENODEV;
2256 read_lock(&bond->lock);
2258 bond_for_each_slave(bond, slave, i) {
2259 if (i == (int)info->slave_id) {
2261 strcpy(info->slave_name, slave->dev->name);
2262 info->link = slave->link;
2263 info->state = slave->state;
2264 info->link_failure_count = slave->link_failure_count;
2269 read_unlock(&bond->lock);
2274 /*-------------------------------- Monitoring -------------------------------*/
2277 static int bond_miimon_inspect(struct bonding *bond)
2279 struct slave *slave;
2280 int i, link_state, commit = 0;
2281 bool ignore_updelay;
2283 ignore_updelay = !bond->curr_active_slave ? true : false;
2285 bond_for_each_slave(bond, slave, i) {
2286 slave->new_link = BOND_LINK_NOCHANGE;
2288 link_state = bond_check_dev_link(bond, slave->dev, 0);
2290 switch (slave->link) {
2295 slave->link = BOND_LINK_FAIL;
2296 slave->delay = bond->params.downdelay;
2298 pr_info("%s: link status down for %sinterface %s, disabling it in %d ms.\n",
2300 (bond->params.mode ==
2301 BOND_MODE_ACTIVEBACKUP) ?
2302 ((slave->state == BOND_STATE_ACTIVE) ?
2303 "active " : "backup ") : "",
2305 bond->params.downdelay * bond->params.miimon);
2308 case BOND_LINK_FAIL:
2311 * recovered before downdelay expired
2313 slave->link = BOND_LINK_UP;
2314 slave->jiffies = jiffies;
2315 pr_info("%s: link status up again after %d ms for interface %s.\n",
2317 (bond->params.downdelay - slave->delay) *
2318 bond->params.miimon,
2323 if (slave->delay <= 0) {
2324 slave->new_link = BOND_LINK_DOWN;
2332 case BOND_LINK_DOWN:
2336 slave->link = BOND_LINK_BACK;
2337 slave->delay = bond->params.updelay;
2340 pr_info("%s: link status up for interface %s, enabling it in %d ms.\n",
2341 bond->dev->name, slave->dev->name,
2342 ignore_updelay ? 0 :
2343 bond->params.updelay *
2344 bond->params.miimon);
2347 case BOND_LINK_BACK:
2349 slave->link = BOND_LINK_DOWN;
2350 pr_info("%s: link status down again after %d ms for interface %s.\n",
2352 (bond->params.updelay - slave->delay) *
2353 bond->params.miimon,
2362 if (slave->delay <= 0) {
2363 slave->new_link = BOND_LINK_UP;
2365 ignore_updelay = false;
2377 static void bond_miimon_commit(struct bonding *bond)
2379 struct slave *slave;
2382 bond_for_each_slave(bond, slave, i) {
2383 switch (slave->new_link) {
2384 case BOND_LINK_NOCHANGE:
2388 slave->link = BOND_LINK_UP;
2389 slave->jiffies = jiffies;
2391 if (bond->params.mode == BOND_MODE_8023AD) {
2392 /* prevent it from being the active one */
2393 slave->state = BOND_STATE_BACKUP;
2394 } else if (bond->params.mode != BOND_MODE_ACTIVEBACKUP) {
2395 /* make it immediately active */
2396 slave->state = BOND_STATE_ACTIVE;
2397 } else if (slave != bond->primary_slave) {
2398 /* prevent it from being the active one */
2399 slave->state = BOND_STATE_BACKUP;
2402 pr_info("%s: link status definitely up for interface %s.\n",
2403 bond->dev->name, slave->dev->name);
2405 /* notify ad that the link status has changed */
2406 if (bond->params.mode == BOND_MODE_8023AD)
2407 bond_3ad_handle_link_change(slave, BOND_LINK_UP);
2409 if (bond_is_lb(bond))
2410 bond_alb_handle_link_change(bond, slave,
2413 if (!bond->curr_active_slave ||
2414 (slave == bond->primary_slave))
2419 case BOND_LINK_DOWN:
2420 if (slave->link_failure_count < UINT_MAX)
2421 slave->link_failure_count++;
2423 slave->link = BOND_LINK_DOWN;
2425 if (bond->params.mode == BOND_MODE_ACTIVEBACKUP ||
2426 bond->params.mode == BOND_MODE_8023AD)
2427 bond_set_slave_inactive_flags(slave);
2429 pr_info("%s: link status definitely down for interface %s, disabling it\n",
2430 bond->dev->name, slave->dev->name);
2432 if (bond->params.mode == BOND_MODE_8023AD)
2433 bond_3ad_handle_link_change(slave,
2436 if (bond_is_lb(bond))
2437 bond_alb_handle_link_change(bond, slave,
2440 if (slave == bond->curr_active_slave)
2446 pr_err("%s: invalid new link %d on slave %s\n",
2447 bond->dev->name, slave->new_link,
2449 slave->new_link = BOND_LINK_NOCHANGE;
2456 write_lock_bh(&bond->curr_slave_lock);
2457 bond_select_active_slave(bond);
2458 write_unlock_bh(&bond->curr_slave_lock);
2461 bond_set_carrier(bond);
2467 * Really a wrapper that splits the mii monitor into two phases: an
2468 * inspection, then (if inspection indicates something needs to be done)
2469 * an acquisition of appropriate locks followed by a commit phase to
2470 * implement whatever link state changes are indicated.
2472 void bond_mii_monitor(struct work_struct *work)
2474 struct bonding *bond = container_of(work, struct bonding,
2477 read_lock(&bond->lock);
2478 if (bond->kill_timers)
2481 if (bond->slave_cnt == 0)
2484 if (bond->send_grat_arp) {
2485 read_lock(&bond->curr_slave_lock);
2486 bond_send_gratuitous_arp(bond);
2487 read_unlock(&bond->curr_slave_lock);
2490 if (bond->send_unsol_na) {
2491 read_lock(&bond->curr_slave_lock);
2492 bond_send_unsolicited_na(bond);
2493 read_unlock(&bond->curr_slave_lock);
2496 if (bond_miimon_inspect(bond)) {
2497 read_unlock(&bond->lock);
2499 read_lock(&bond->lock);
2501 bond_miimon_commit(bond);
2503 read_unlock(&bond->lock);
2504 rtnl_unlock(); /* might sleep, hold no other locks */
2505 read_lock(&bond->lock);
2509 if (bond->params.miimon)
2510 queue_delayed_work(bond->wq, &bond->mii_work,
2511 msecs_to_jiffies(bond->params.miimon));
2513 read_unlock(&bond->lock);
2516 static __be32 bond_glean_dev_ip(struct net_device *dev)
2518 struct in_device *idev;
2519 struct in_ifaddr *ifa;
2526 idev = __in_dev_get_rcu(dev);
2530 ifa = idev->ifa_list;
2534 addr = ifa->ifa_local;
2540 static int bond_has_this_ip(struct bonding *bond, __be32 ip)
2542 struct vlan_entry *vlan;
2544 if (ip == bond->master_ip)
2547 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
2548 if (ip == vlan->vlan_ip)
2556 * We go to the (large) trouble of VLAN tagging ARP frames because
2557 * switches in VLAN mode (especially if ports are configured as
2558 * "native" to a VLAN) might not pass non-tagged frames.
2560 static void bond_arp_send(struct net_device *slave_dev, int arp_op, __be32 dest_ip, __be32 src_ip, unsigned short vlan_id)
2562 struct sk_buff *skb;
2564 pr_debug("arp %d on slave %s: dst %x src %x vid %d\n", arp_op,
2565 slave_dev->name, dest_ip, src_ip, vlan_id);
2567 skb = arp_create(arp_op, ETH_P_ARP, dest_ip, slave_dev, src_ip,
2568 NULL, slave_dev->dev_addr, NULL);
2571 pr_err("ARP packet allocation failed\n");
2575 skb = vlan_put_tag(skb, vlan_id);
2577 pr_err("failed to insert VLAN tag\n");
2585 static void bond_arp_send_all(struct bonding *bond, struct slave *slave)
2588 __be32 *targets = bond->params.arp_targets;
2589 struct vlan_entry *vlan;
2590 struct net_device *vlan_dev;
2594 for (i = 0; (i < BOND_MAX_ARP_TARGETS); i++) {
2597 pr_debug("basa: target %x\n", targets[i]);
2599 pr_debug("basa: empty vlan: arp_send\n");
2600 bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2601 bond->master_ip, 0);
2606 * If VLANs are configured, we do a route lookup to
2607 * determine which VLAN interface would be used, so we
2608 * can tag the ARP with the proper VLAN tag.
2610 memset(&fl, 0, sizeof(fl));
2611 fl.fl4_dst = targets[i];
2612 fl.fl4_tos = RTO_ONLINK;
2614 rv = ip_route_output_key(dev_net(bond->dev), &rt, &fl);
2616 if (net_ratelimit()) {
2617 pr_warning("%s: no route to arp_ip_target %pI4\n",
2618 bond->dev->name, &fl.fl4_dst);
2624 * This target is not on a VLAN
2626 if (rt->dst.dev == bond->dev) {
2628 pr_debug("basa: rtdev == bond->dev: arp_send\n");
2629 bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2630 bond->master_ip, 0);
2635 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
2636 vlan_dev = vlan_group_get_device(bond->vlgrp, vlan->vlan_id);
2637 if (vlan_dev == rt->dst.dev) {
2638 vlan_id = vlan->vlan_id;
2639 pr_debug("basa: vlan match on %s %d\n",
2640 vlan_dev->name, vlan_id);
2647 bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2648 vlan->vlan_ip, vlan_id);
2652 if (net_ratelimit()) {
2653 pr_warning("%s: no path to arp_ip_target %pI4 via rt.dev %s\n",
2654 bond->dev->name, &fl.fl4_dst,
2655 rt->dst.dev ? rt->dst.dev->name : "NULL");
2662 * Kick out a gratuitous ARP for an IP on the bonding master plus one
2663 * for each VLAN above us.
2665 * Caller must hold curr_slave_lock for read or better
2667 static void bond_send_gratuitous_arp(struct bonding *bond)
2669 struct slave *slave = bond->curr_active_slave;
2670 struct vlan_entry *vlan;
2671 struct net_device *vlan_dev;
2673 pr_debug("bond_send_grat_arp: bond %s slave %s\n",
2674 bond->dev->name, slave ? slave->dev->name : "NULL");
2676 if (!slave || !bond->send_grat_arp ||
2677 test_bit(__LINK_STATE_LINKWATCH_PENDING, &slave->dev->state))
2680 bond->send_grat_arp--;
2682 if (bond->master_ip) {
2683 bond_arp_send(slave->dev, ARPOP_REPLY, bond->master_ip,
2684 bond->master_ip, 0);
2690 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
2691 vlan_dev = vlan_group_get_device(bond->vlgrp, vlan->vlan_id);
2692 if (vlan->vlan_ip) {
2693 bond_arp_send(slave->dev, ARPOP_REPLY, vlan->vlan_ip,
2694 vlan->vlan_ip, vlan->vlan_id);
2699 static void bond_validate_arp(struct bonding *bond, struct slave *slave, __be32 sip, __be32 tip)
2702 __be32 *targets = bond->params.arp_targets;
2704 for (i = 0; (i < BOND_MAX_ARP_TARGETS) && targets[i]; i++) {
2705 pr_debug("bva: sip %pI4 tip %pI4 t[%d] %pI4 bhti(tip) %d\n",
2706 &sip, &tip, i, &targets[i],
2707 bond_has_this_ip(bond, tip));
2708 if (sip == targets[i]) {
2709 if (bond_has_this_ip(bond, tip))
2710 slave->last_arp_rx = jiffies;
2716 static int bond_arp_rcv(struct sk_buff *skb, struct net_device *dev, struct packet_type *pt, struct net_device *orig_dev)
2719 struct slave *slave;
2720 struct bonding *bond;
2721 unsigned char *arp_ptr;
2724 if (dev->priv_flags & IFF_802_1Q_VLAN) {
2726 * When using VLANS and bonding, dev and oriv_dev may be
2727 * incorrect if the physical interface supports VLAN
2728 * acceleration. With this change ARP validation now
2729 * works for hosts only reachable on the VLAN interface.
2731 dev = vlan_dev_real_dev(dev);
2732 orig_dev = dev_get_by_index_rcu(dev_net(skb->dev),skb->skb_iif);
2735 if (!(dev->priv_flags & IFF_BONDING) || !(dev->flags & IFF_MASTER))
2738 bond = netdev_priv(dev);
2739 read_lock(&bond->lock);
2741 pr_debug("bond_arp_rcv: bond %s skb->dev %s orig_dev %s\n",
2742 bond->dev->name, skb->dev ? skb->dev->name : "NULL",
2743 orig_dev ? orig_dev->name : "NULL");
2745 slave = bond_get_slave_by_dev(bond, orig_dev);
2746 if (!slave || !slave_do_arp_validate(bond, slave))
2749 if (!pskb_may_pull(skb, arp_hdr_len(dev)))
2753 if (arp->ar_hln != dev->addr_len ||
2754 skb->pkt_type == PACKET_OTHERHOST ||
2755 skb->pkt_type == PACKET_LOOPBACK ||
2756 arp->ar_hrd != htons(ARPHRD_ETHER) ||
2757 arp->ar_pro != htons(ETH_P_IP) ||
2761 arp_ptr = (unsigned char *)(arp + 1);
2762 arp_ptr += dev->addr_len;
2763 memcpy(&sip, arp_ptr, 4);
2764 arp_ptr += 4 + dev->addr_len;
2765 memcpy(&tip, arp_ptr, 4);
2767 pr_debug("bond_arp_rcv: %s %s/%d av %d sv %d sip %pI4 tip %pI4\n",
2768 bond->dev->name, slave->dev->name, slave->state,
2769 bond->params.arp_validate, slave_do_arp_validate(bond, slave),
2773 * Backup slaves won't see the ARP reply, but do come through
2774 * here for each ARP probe (so we swap the sip/tip to validate
2775 * the probe). In a "redundant switch, common router" type of
2776 * configuration, the ARP probe will (hopefully) travel from
2777 * the active, through one switch, the router, then the other
2778 * switch before reaching the backup.
2780 if (slave->state == BOND_STATE_ACTIVE)
2781 bond_validate_arp(bond, slave, sip, tip);
2783 bond_validate_arp(bond, slave, tip, sip);
2786 read_unlock(&bond->lock);
2789 return NET_RX_SUCCESS;
2793 * this function is called regularly to monitor each slave's link
2794 * ensuring that traffic is being sent and received when arp monitoring
2795 * is used in load-balancing mode. if the adapter has been dormant, then an
2796 * arp is transmitted to generate traffic. see activebackup_arp_monitor for
2797 * arp monitoring in active backup mode.
2799 void bond_loadbalance_arp_mon(struct work_struct *work)
2801 struct bonding *bond = container_of(work, struct bonding,
2803 struct slave *slave, *oldcurrent;
2804 int do_failover = 0;
2808 read_lock(&bond->lock);
2810 delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
2812 if (bond->kill_timers)
2815 if (bond->slave_cnt == 0)
2818 read_lock(&bond->curr_slave_lock);
2819 oldcurrent = bond->curr_active_slave;
2820 read_unlock(&bond->curr_slave_lock);
2822 /* see if any of the previous devices are up now (i.e. they have
2823 * xmt and rcv traffic). the curr_active_slave does not come into
2824 * the picture unless it is null. also, slave->jiffies is not needed
2825 * here because we send an arp on each slave and give a slave as
2826 * long as it needs to get the tx/rx within the delta.
2827 * TODO: what about up/down delay in arp mode? it wasn't here before
2830 bond_for_each_slave(bond, slave, i) {
2831 unsigned long trans_start = dev_trans_start(slave->dev);
2833 if (slave->link != BOND_LINK_UP) {
2834 if (time_in_range(jiffies,
2835 trans_start - delta_in_ticks,
2836 trans_start + delta_in_ticks) &&
2837 time_in_range(jiffies,
2838 slave->dev->last_rx - delta_in_ticks,
2839 slave->dev->last_rx + delta_in_ticks)) {
2841 slave->link = BOND_LINK_UP;
2842 slave->state = BOND_STATE_ACTIVE;
2844 /* primary_slave has no meaning in round-robin
2845 * mode. the window of a slave being up and
2846 * curr_active_slave being null after enslaving
2850 pr_info("%s: link status definitely up for interface %s, ",
2855 pr_info("%s: interface %s is now up\n",
2861 /* slave->link == BOND_LINK_UP */
2863 /* not all switches will respond to an arp request
2864 * when the source ip is 0, so don't take the link down
2865 * if we don't know our ip yet
2867 if (!time_in_range(jiffies,
2868 trans_start - delta_in_ticks,
2869 trans_start + 2 * delta_in_ticks) ||
2870 !time_in_range(jiffies,
2871 slave->dev->last_rx - delta_in_ticks,
2872 slave->dev->last_rx + 2 * delta_in_ticks)) {
2874 slave->link = BOND_LINK_DOWN;
2875 slave->state = BOND_STATE_BACKUP;
2877 if (slave->link_failure_count < UINT_MAX)
2878 slave->link_failure_count++;
2880 pr_info("%s: interface %s is now down.\n",
2884 if (slave == oldcurrent)
2889 /* note: if switch is in round-robin mode, all links
2890 * must tx arp to ensure all links rx an arp - otherwise
2891 * links may oscillate or not come up at all; if switch is
2892 * in something like xor mode, there is nothing we can
2893 * do - all replies will be rx'ed on same link causing slaves
2894 * to be unstable during low/no traffic periods
2896 if (IS_UP(slave->dev))
2897 bond_arp_send_all(bond, slave);
2901 write_lock_bh(&bond->curr_slave_lock);
2903 bond_select_active_slave(bond);
2905 write_unlock_bh(&bond->curr_slave_lock);
2909 if (bond->params.arp_interval)
2910 queue_delayed_work(bond->wq, &bond->arp_work, delta_in_ticks);
2912 read_unlock(&bond->lock);
2916 * Called to inspect slaves for active-backup mode ARP monitor link state
2917 * changes. Sets new_link in slaves to specify what action should take
2918 * place for the slave. Returns 0 if no changes are found, >0 if changes
2919 * to link states must be committed.
2921 * Called with bond->lock held for read.
2923 static int bond_ab_arp_inspect(struct bonding *bond, int delta_in_ticks)
2925 struct slave *slave;
2927 unsigned long trans_start;
2929 bond_for_each_slave(bond, slave, i) {
2930 slave->new_link = BOND_LINK_NOCHANGE;
2932 if (slave->link != BOND_LINK_UP) {
2933 if (time_in_range(jiffies,
2934 slave_last_rx(bond, slave) - delta_in_ticks,
2935 slave_last_rx(bond, slave) + delta_in_ticks)) {
2937 slave->new_link = BOND_LINK_UP;
2945 * Give slaves 2*delta after being enslaved or made
2946 * active. This avoids bouncing, as the last receive
2947 * times need a full ARP monitor cycle to be updated.
2949 if (time_in_range(jiffies,
2950 slave->jiffies - delta_in_ticks,
2951 slave->jiffies + 2 * delta_in_ticks))
2955 * Backup slave is down if:
2956 * - No current_arp_slave AND
2957 * - more than 3*delta since last receive AND
2958 * - the bond has an IP address
2960 * Note: a non-null current_arp_slave indicates
2961 * the curr_active_slave went down and we are
2962 * searching for a new one; under this condition
2963 * we only take the curr_active_slave down - this
2964 * gives each slave a chance to tx/rx traffic
2965 * before being taken out
2967 if (slave->state == BOND_STATE_BACKUP &&
2968 !bond->current_arp_slave &&
2969 !time_in_range(jiffies,
2970 slave_last_rx(bond, slave) - delta_in_ticks,
2971 slave_last_rx(bond, slave) + 3 * delta_in_ticks)) {
2973 slave->new_link = BOND_LINK_DOWN;
2978 * Active slave is down if:
2979 * - more than 2*delta since transmitting OR
2980 * - (more than 2*delta since receive AND
2981 * the bond has an IP address)
2983 trans_start = dev_trans_start(slave->dev);
2984 if ((slave->state == BOND_STATE_ACTIVE) &&
2985 (!time_in_range(jiffies,
2986 trans_start - delta_in_ticks,
2987 trans_start + 2 * delta_in_ticks) ||
2988 !time_in_range(jiffies,
2989 slave_last_rx(bond, slave) - delta_in_ticks,
2990 slave_last_rx(bond, slave) + 2 * delta_in_ticks))) {
2992 slave->new_link = BOND_LINK_DOWN;
3001 * Called to commit link state changes noted by inspection step of
3002 * active-backup mode ARP monitor.
3004 * Called with RTNL and bond->lock for read.
3006 static void bond_ab_arp_commit(struct bonding *bond, int delta_in_ticks)
3008 struct slave *slave;
3010 unsigned long trans_start;
3012 bond_for_each_slave(bond, slave, i) {
3013 switch (slave->new_link) {
3014 case BOND_LINK_NOCHANGE:
3018 trans_start = dev_trans_start(slave->dev);
3019 if ((!bond->curr_active_slave &&
3020 time_in_range(jiffies,
3021 trans_start - delta_in_ticks,
3022 trans_start + delta_in_ticks)) ||
3023 bond->curr_active_slave != slave) {
3024 slave->link = BOND_LINK_UP;
3025 bond->current_arp_slave = NULL;
3027 pr_info("%s: link status definitely up for interface %s.\n",
3028 bond->dev->name, slave->dev->name);
3030 if (!bond->curr_active_slave ||
3031 (slave == bond->primary_slave))
3038 case BOND_LINK_DOWN:
3039 if (slave->link_failure_count < UINT_MAX)
3040 slave->link_failure_count++;
3042 slave->link = BOND_LINK_DOWN;
3043 bond_set_slave_inactive_flags(slave);
3045 pr_info("%s: link status definitely down for interface %s, disabling it\n",
3046 bond->dev->name, slave->dev->name);
3048 if (slave == bond->curr_active_slave) {
3049 bond->current_arp_slave = NULL;
3056 pr_err("%s: impossible: new_link %d on slave %s\n",
3057 bond->dev->name, slave->new_link,
3064 write_lock_bh(&bond->curr_slave_lock);
3065 bond_select_active_slave(bond);
3066 write_unlock_bh(&bond->curr_slave_lock);
3069 bond_set_carrier(bond);
3073 * Send ARP probes for active-backup mode ARP monitor.
3075 * Called with bond->lock held for read.
3077 static void bond_ab_arp_probe(struct bonding *bond)
3079 struct slave *slave;
3082 read_lock(&bond->curr_slave_lock);
3084 if (bond->current_arp_slave && bond->curr_active_slave)
3085 pr_info("PROBE: c_arp %s && cas %s BAD\n",
3086 bond->current_arp_slave->dev->name,
3087 bond->curr_active_slave->dev->name);
3089 if (bond->curr_active_slave) {
3090 bond_arp_send_all(bond, bond->curr_active_slave);
3091 read_unlock(&bond->curr_slave_lock);
3095 read_unlock(&bond->curr_slave_lock);
3097 /* if we don't have a curr_active_slave, search for the next available
3098 * backup slave from the current_arp_slave and make it the candidate
3099 * for becoming the curr_active_slave
3102 if (!bond->current_arp_slave) {
3103 bond->current_arp_slave = bond->first_slave;
3104 if (!bond->current_arp_slave)
3108 bond_set_slave_inactive_flags(bond->current_arp_slave);
3110 /* search for next candidate */
3111 bond_for_each_slave_from(bond, slave, i, bond->current_arp_slave->next) {
3112 if (IS_UP(slave->dev)) {
3113 slave->link = BOND_LINK_BACK;
3114 bond_set_slave_active_flags(slave);
3115 bond_arp_send_all(bond, slave);
3116 slave->jiffies = jiffies;
3117 bond->current_arp_slave = slave;
3121 /* if the link state is up at this point, we
3122 * mark it down - this can happen if we have
3123 * simultaneous link failures and
3124 * reselect_active_interface doesn't make this
3125 * one the current slave so it is still marked
3126 * up when it is actually down
3128 if (slave->link == BOND_LINK_UP) {
3129 slave->link = BOND_LINK_DOWN;
3130 if (slave->link_failure_count < UINT_MAX)
3131 slave->link_failure_count++;
3133 bond_set_slave_inactive_flags(slave);
3135 pr_info("%s: backup interface %s is now down.\n",
3136 bond->dev->name, slave->dev->name);
3141 void bond_activebackup_arp_mon(struct work_struct *work)
3143 struct bonding *bond = container_of(work, struct bonding,
3147 read_lock(&bond->lock);
3149 if (bond->kill_timers)
3152 delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
3154 if (bond->slave_cnt == 0)
3157 if (bond->send_grat_arp) {
3158 read_lock(&bond->curr_slave_lock);
3159 bond_send_gratuitous_arp(bond);
3160 read_unlock(&bond->curr_slave_lock);
3163 if (bond->send_unsol_na) {
3164 read_lock(&bond->curr_slave_lock);
3165 bond_send_unsolicited_na(bond);
3166 read_unlock(&bond->curr_slave_lock);
3169 if (bond_ab_arp_inspect(bond, delta_in_ticks)) {
3170 read_unlock(&bond->lock);
3172 read_lock(&bond->lock);
3174 bond_ab_arp_commit(bond, delta_in_ticks);
3176 read_unlock(&bond->lock);
3178 read_lock(&bond->lock);
3181 bond_ab_arp_probe(bond);
3184 if (bond->params.arp_interval)
3185 queue_delayed_work(bond->wq, &bond->arp_work, delta_in_ticks);
3187 read_unlock(&bond->lock);
3190 /*------------------------------ proc/seq_file-------------------------------*/
3192 #ifdef CONFIG_PROC_FS
3194 static void *bond_info_seq_start(struct seq_file *seq, loff_t *pos)
3195 __acquires(&dev_base_lock)
3196 __acquires(&bond->lock)
3198 struct bonding *bond = seq->private;
3200 struct slave *slave;
3203 /* make sure the bond won't be taken away */
3204 read_lock(&dev_base_lock);
3205 read_lock(&bond->lock);
3208 return SEQ_START_TOKEN;
3210 bond_for_each_slave(bond, slave, i) {
3218 static void *bond_info_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3220 struct bonding *bond = seq->private;
3221 struct slave *slave = v;
3224 if (v == SEQ_START_TOKEN)
3225 return bond->first_slave;
3227 slave = slave->next;
3229 return (slave == bond->first_slave) ? NULL : slave;
3232 static void bond_info_seq_stop(struct seq_file *seq, void *v)
3233 __releases(&bond->lock)
3234 __releases(&dev_base_lock)
3236 struct bonding *bond = seq->private;
3238 read_unlock(&bond->lock);
3239 read_unlock(&dev_base_lock);
3242 static void bond_info_show_master(struct seq_file *seq)
3244 struct bonding *bond = seq->private;
3248 read_lock(&bond->curr_slave_lock);
3249 curr = bond->curr_active_slave;
3250 read_unlock(&bond->curr_slave_lock);
3252 seq_printf(seq, "Bonding Mode: %s",
3253 bond_mode_name(bond->params.mode));
3255 if (bond->params.mode == BOND_MODE_ACTIVEBACKUP &&
3256 bond->params.fail_over_mac)
3257 seq_printf(seq, " (fail_over_mac %s)",
3258 fail_over_mac_tbl[bond->params.fail_over_mac].modename);
3260 seq_printf(seq, "\n");
3262 if (bond->params.mode == BOND_MODE_XOR ||
3263 bond->params.mode == BOND_MODE_8023AD) {
3264 seq_printf(seq, "Transmit Hash Policy: %s (%d)\n",
3265 xmit_hashtype_tbl[bond->params.xmit_policy].modename,
3266 bond->params.xmit_policy);
3269 if (USES_PRIMARY(bond->params.mode)) {
3270 seq_printf(seq, "Primary Slave: %s",
3271 (bond->primary_slave) ?
3272 bond->primary_slave->dev->name : "None");
3273 if (bond->primary_slave)
3274 seq_printf(seq, " (primary_reselect %s)",
3275 pri_reselect_tbl[bond->params.primary_reselect].modename);
3277 seq_printf(seq, "\nCurrently Active Slave: %s\n",
3278 (curr) ? curr->dev->name : "None");
3281 seq_printf(seq, "MII Status: %s\n", netif_carrier_ok(bond->dev) ?
3283 seq_printf(seq, "MII Polling Interval (ms): %d\n", bond->params.miimon);
3284 seq_printf(seq, "Up Delay (ms): %d\n",
3285 bond->params.updelay * bond->params.miimon);
3286 seq_printf(seq, "Down Delay (ms): %d\n",
3287 bond->params.downdelay * bond->params.miimon);
3290 /* ARP information */
3291 if (bond->params.arp_interval > 0) {
3293 seq_printf(seq, "ARP Polling Interval (ms): %d\n",
3294 bond->params.arp_interval);
3296 seq_printf(seq, "ARP IP target/s (n.n.n.n form):");
3298 for (i = 0; (i < BOND_MAX_ARP_TARGETS); i++) {
3299 if (!bond->params.arp_targets[i])
3302 seq_printf(seq, ",");
3303 seq_printf(seq, " %pI4", &bond->params.arp_targets[i]);
3306 seq_printf(seq, "\n");
3309 if (bond->params.mode == BOND_MODE_8023AD) {
3310 struct ad_info ad_info;
3312 seq_puts(seq, "\n802.3ad info\n");
3313 seq_printf(seq, "LACP rate: %s\n",
3314 (bond->params.lacp_fast) ? "fast" : "slow");
3315 seq_printf(seq, "Aggregator selection policy (ad_select): %s\n",
3316 ad_select_tbl[bond->params.ad_select].modename);
3318 if (bond_3ad_get_active_agg_info(bond, &ad_info)) {
3319 seq_printf(seq, "bond %s has no active aggregator\n",
3322 seq_printf(seq, "Active Aggregator Info:\n");
3324 seq_printf(seq, "\tAggregator ID: %d\n",
3325 ad_info.aggregator_id);
3326 seq_printf(seq, "\tNumber of ports: %d\n",
3328 seq_printf(seq, "\tActor Key: %d\n",
3330 seq_printf(seq, "\tPartner Key: %d\n",
3331 ad_info.partner_key);
3332 seq_printf(seq, "\tPartner Mac Address: %pM\n",
3333 ad_info.partner_system);
3338 static void bond_info_show_slave(struct seq_file *seq,
3339 const struct slave *slave)
3341 struct bonding *bond = seq->private;
3343 seq_printf(seq, "\nSlave Interface: %s\n", slave->dev->name);
3344 seq_printf(seq, "MII Status: %s\n",
3345 (slave->link == BOND_LINK_UP) ? "up" : "down");
3346 seq_printf(seq, "Link Failure Count: %u\n",
3347 slave->link_failure_count);
3349 seq_printf(seq, "Permanent HW addr: %pM\n", slave->perm_hwaddr);
3351 if (bond->params.mode == BOND_MODE_8023AD) {
3352 const struct aggregator *agg
3353 = SLAVE_AD_INFO(slave).port.aggregator;
3356 seq_printf(seq, "Aggregator ID: %d\n",
3357 agg->aggregator_identifier);
3359 seq_puts(seq, "Aggregator ID: N/A\n");
3361 seq_printf(seq, "Slave queue ID: %d\n", slave->queue_id);
3364 static int bond_info_seq_show(struct seq_file *seq, void *v)
3366 if (v == SEQ_START_TOKEN) {
3367 seq_printf(seq, "%s\n", version);
3368 bond_info_show_master(seq);
3370 bond_info_show_slave(seq, v);
3375 static const struct seq_operations bond_info_seq_ops = {
3376 .start = bond_info_seq_start,
3377 .next = bond_info_seq_next,
3378 .stop = bond_info_seq_stop,
3379 .show = bond_info_seq_show,
3382 static int bond_info_open(struct inode *inode, struct file *file)
3384 struct seq_file *seq;
3385 struct proc_dir_entry *proc;
3388 res = seq_open(file, &bond_info_seq_ops);
3390 /* recover the pointer buried in proc_dir_entry data */
3391 seq = file->private_data;
3393 seq->private = proc->data;
3399 static const struct file_operations bond_info_fops = {
3400 .owner = THIS_MODULE,
3401 .open = bond_info_open,
3403 .llseek = seq_lseek,
3404 .release = seq_release,
3407 static void bond_create_proc_entry(struct bonding *bond)
3409 struct net_device *bond_dev = bond->dev;
3410 struct bond_net *bn = net_generic(dev_net(bond_dev), bond_net_id);
3413 bond->proc_entry = proc_create_data(bond_dev->name,
3414 S_IRUGO, bn->proc_dir,
3415 &bond_info_fops, bond);
3416 if (bond->proc_entry == NULL)
3417 pr_warning("Warning: Cannot create /proc/net/%s/%s\n",
3418 DRV_NAME, bond_dev->name);
3420 memcpy(bond->proc_file_name, bond_dev->name, IFNAMSIZ);
3424 static void bond_remove_proc_entry(struct bonding *bond)
3426 struct net_device *bond_dev = bond->dev;
3427 struct bond_net *bn = net_generic(dev_net(bond_dev), bond_net_id);
3429 if (bn->proc_dir && bond->proc_entry) {
3430 remove_proc_entry(bond->proc_file_name, bn->proc_dir);
3431 memset(bond->proc_file_name, 0, IFNAMSIZ);
3432 bond->proc_entry = NULL;
3436 /* Create the bonding directory under /proc/net, if doesn't exist yet.
3437 * Caller must hold rtnl_lock.
3439 static void __net_init bond_create_proc_dir(struct bond_net *bn)
3441 if (!bn->proc_dir) {
3442 bn->proc_dir = proc_mkdir(DRV_NAME, bn->net->proc_net);
3444 pr_warning("Warning: cannot create /proc/net/%s\n",
3449 /* Destroy the bonding directory under /proc/net, if empty.
3450 * Caller must hold rtnl_lock.
3452 static void __net_exit bond_destroy_proc_dir(struct bond_net *bn)
3455 remove_proc_entry(DRV_NAME, bn->net->proc_net);
3456 bn->proc_dir = NULL;
3460 #else /* !CONFIG_PROC_FS */
3462 static void bond_create_proc_entry(struct bonding *bond)
3466 static void bond_remove_proc_entry(struct bonding *bond)
3470 static inline void bond_create_proc_dir(struct bond_net *bn)
3474 static inline void bond_destroy_proc_dir(struct bond_net *bn)
3478 #endif /* CONFIG_PROC_FS */
3481 /*-------------------------- netdev event handling --------------------------*/
3484 * Change device name
3486 static int bond_event_changename(struct bonding *bond)
3488 bond_remove_proc_entry(bond);
3489 bond_create_proc_entry(bond);
3494 static int bond_master_netdev_event(unsigned long event,
3495 struct net_device *bond_dev)
3497 struct bonding *event_bond = netdev_priv(bond_dev);
3500 case NETDEV_CHANGENAME:
3501 return bond_event_changename(event_bond);
3509 static int bond_slave_netdev_event(unsigned long event,
3510 struct net_device *slave_dev)
3512 struct net_device *bond_dev = slave_dev->master;
3513 struct bonding *bond = netdev_priv(bond_dev);
3516 case NETDEV_UNREGISTER:
3518 if (bond->setup_by_slave)
3519 bond_release_and_destroy(bond_dev, slave_dev);
3521 bond_release(bond_dev, slave_dev);
3525 if (bond->params.mode == BOND_MODE_8023AD || bond_is_lb(bond)) {
3526 struct slave *slave;
3528 slave = bond_get_slave_by_dev(bond, slave_dev);
3530 u16 old_speed = slave->speed;
3531 u16 old_duplex = slave->duplex;
3533 bond_update_speed_duplex(slave);
3535 if (bond_is_lb(bond))
3538 if (old_speed != slave->speed)
3539 bond_3ad_adapter_speed_changed(slave);
3540 if (old_duplex != slave->duplex)
3541 bond_3ad_adapter_duplex_changed(slave);
3548 * ... Or is it this?
3551 case NETDEV_CHANGEMTU:
3553 * TODO: Should slaves be allowed to
3554 * independently alter their MTU? For
3555 * an active-backup bond, slaves need
3556 * not be the same type of device, so
3557 * MTUs may vary. For other modes,
3558 * slaves arguably should have the
3559 * same MTUs. To do this, we'd need to
3560 * take over the slave's change_mtu
3561 * function for the duration of their
3565 case NETDEV_CHANGENAME:
3567 * TODO: handle changing the primary's name
3570 case NETDEV_FEAT_CHANGE:
3571 bond_compute_features(bond);
3581 * bond_netdev_event: handle netdev notifier chain events.
3583 * This function receives events for the netdev chain. The caller (an
3584 * ioctl handler calling blocking_notifier_call_chain) holds the necessary
3585 * locks for us to safely manipulate the slave devices (RTNL lock,
3588 static int bond_netdev_event(struct notifier_block *this,
3589 unsigned long event, void *ptr)
3591 struct net_device *event_dev = (struct net_device *)ptr;
3593 pr_debug("event_dev: %s, event: %lx\n",
3594 event_dev ? event_dev->name : "None",
3597 if (!(event_dev->priv_flags & IFF_BONDING))
3600 if (event_dev->flags & IFF_MASTER) {
3601 pr_debug("IFF_MASTER\n");
3602 return bond_master_netdev_event(event, event_dev);
3605 if (event_dev->flags & IFF_SLAVE) {
3606 pr_debug("IFF_SLAVE\n");
3607 return bond_slave_netdev_event(event, event_dev);
3614 * bond_inetaddr_event: handle inetaddr notifier chain events.
3616 * We keep track of device IPs primarily to use as source addresses in
3617 * ARP monitor probes (rather than spewing out broadcasts all the time).
3619 * We track one IP for the main device (if it has one), plus one per VLAN.
3621 static int bond_inetaddr_event(struct notifier_block *this, unsigned long event, void *ptr)
3623 struct in_ifaddr *ifa = ptr;
3624 struct net_device *vlan_dev, *event_dev = ifa->ifa_dev->dev;
3625 struct bond_net *bn = net_generic(dev_net(event_dev), bond_net_id);
3626 struct bonding *bond;
3627 struct vlan_entry *vlan;
3629 list_for_each_entry(bond, &bn->dev_list, bond_list) {
3630 if (bond->dev == event_dev) {
3633 bond->master_ip = ifa->ifa_local;
3636 bond->master_ip = bond_glean_dev_ip(bond->dev);
3643 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
3646 vlan_dev = vlan_group_get_device(bond->vlgrp, vlan->vlan_id);
3647 if (vlan_dev == event_dev) {
3650 vlan->vlan_ip = ifa->ifa_local;
3654 bond_glean_dev_ip(vlan_dev);
3665 static struct notifier_block bond_netdev_notifier = {
3666 .notifier_call = bond_netdev_event,
3669 static struct notifier_block bond_inetaddr_notifier = {
3670 .notifier_call = bond_inetaddr_event,
3673 /*-------------------------- Packet type handling ---------------------------*/
3675 /* register to receive lacpdus on a bond */
3676 static void bond_register_lacpdu(struct bonding *bond)
3678 struct packet_type *pk_type = &(BOND_AD_INFO(bond).ad_pkt_type);
3680 /* initialize packet type */
3681 pk_type->type = PKT_TYPE_LACPDU;
3682 pk_type->dev = bond->dev;
3683 pk_type->func = bond_3ad_lacpdu_recv;
3685 dev_add_pack(pk_type);
3688 /* unregister to receive lacpdus on a bond */
3689 static void bond_unregister_lacpdu(struct bonding *bond)
3691 dev_remove_pack(&(BOND_AD_INFO(bond).ad_pkt_type));
3694 void bond_register_arp(struct bonding *bond)
3696 struct packet_type *pt = &bond->arp_mon_pt;
3701 pt->type = htons(ETH_P_ARP);
3702 pt->dev = bond->dev;
3703 pt->func = bond_arp_rcv;
3707 void bond_unregister_arp(struct bonding *bond)
3709 struct packet_type *pt = &bond->arp_mon_pt;
3711 dev_remove_pack(pt);
3715 /*---------------------------- Hashing Policies -----------------------------*/
3718 * Hash for the output device based upon layer 2 and layer 3 data. If
3719 * the packet is not IP mimic bond_xmit_hash_policy_l2()
3721 static int bond_xmit_hash_policy_l23(struct sk_buff *skb, int count)
3723 struct ethhdr *data = (struct ethhdr *)skb->data;
3724 struct iphdr *iph = ip_hdr(skb);
3726 if (skb->protocol == htons(ETH_P_IP)) {
3727 return ((ntohl(iph->saddr ^ iph->daddr) & 0xffff) ^
3728 (data->h_dest[5] ^ data->h_source[5])) % count;
3731 return (data->h_dest[5] ^ data->h_source[5]) % count;
3735 * Hash for the output device based upon layer 3 and layer 4 data. If
3736 * the packet is a frag or not TCP or UDP, just use layer 3 data. If it is
3737 * altogether not IP, mimic bond_xmit_hash_policy_l2()
3739 static int bond_xmit_hash_policy_l34(struct sk_buff *skb, int count)
3741 struct ethhdr *data = (struct ethhdr *)skb->data;
3742 struct iphdr *iph = ip_hdr(skb);
3743 __be16 *layer4hdr = (__be16 *)((u32 *)iph + iph->ihl);
3746 if (skb->protocol == htons(ETH_P_IP)) {
3747 if (!(iph->frag_off & htons(IP_MF|IP_OFFSET)) &&
3748 (iph->protocol == IPPROTO_TCP ||
3749 iph->protocol == IPPROTO_UDP)) {
3750 layer4_xor = ntohs((*layer4hdr ^ *(layer4hdr + 1)));
3752 return (layer4_xor ^
3753 ((ntohl(iph->saddr ^ iph->daddr)) & 0xffff)) % count;
3757 return (data->h_dest[5] ^ data->h_source[5]) % count;
3761 * Hash for the output device based upon layer 2 data
3763 static int bond_xmit_hash_policy_l2(struct sk_buff *skb, int count)
3765 struct ethhdr *data = (struct ethhdr *)skb->data;
3767 return (data->h_dest[5] ^ data->h_source[5]) % count;
3770 /*-------------------------- Device entry points ----------------------------*/
3772 static int bond_open(struct net_device *bond_dev)
3774 struct bonding *bond = netdev_priv(bond_dev);
3776 bond->kill_timers = 0;
3778 INIT_DELAYED_WORK(&bond->mcast_work, bond_resend_igmp_join_requests_delayed);
3780 if (bond_is_lb(bond)) {
3781 /* bond_alb_initialize must be called before the timer
3784 if (bond_alb_initialize(bond, (bond->params.mode == BOND_MODE_ALB))) {
3785 /* something went wrong - fail the open operation */
3789 INIT_DELAYED_WORK(&bond->alb_work, bond_alb_monitor);
3790 queue_delayed_work(bond->wq, &bond->alb_work, 0);
3793 if (bond->params.miimon) { /* link check interval, in milliseconds. */
3794 INIT_DELAYED_WORK(&bond->mii_work, bond_mii_monitor);
3795 queue_delayed_work(bond->wq, &bond->mii_work, 0);
3798 if (bond->params.arp_interval) { /* arp interval, in milliseconds. */
3799 if (bond->params.mode == BOND_MODE_ACTIVEBACKUP)
3800 INIT_DELAYED_WORK(&bond->arp_work,
3801 bond_activebackup_arp_mon);
3803 INIT_DELAYED_WORK(&bond->arp_work,
3804 bond_loadbalance_arp_mon);
3806 queue_delayed_work(bond->wq, &bond->arp_work, 0);
3807 if (bond->params.arp_validate)
3808 bond_register_arp(bond);
3811 if (bond->params.mode == BOND_MODE_8023AD) {
3812 INIT_DELAYED_WORK(&bond->ad_work, bond_3ad_state_machine_handler);
3813 queue_delayed_work(bond->wq, &bond->ad_work, 0);
3814 /* register to receive LACPDUs */
3815 bond_register_lacpdu(bond);
3816 bond_3ad_initiate_agg_selection(bond, 1);
3822 static int bond_close(struct net_device *bond_dev)
3824 struct bonding *bond = netdev_priv(bond_dev);
3826 if (bond->params.mode == BOND_MODE_8023AD) {
3827 /* Unregister the receive of LACPDUs */
3828 bond_unregister_lacpdu(bond);
3831 if (bond->params.arp_validate)
3832 bond_unregister_arp(bond);
3834 write_lock_bh(&bond->lock);
3836 bond->send_grat_arp = 0;
3837 bond->send_unsol_na = 0;
3839 /* signal timers not to re-arm */
3840 bond->kill_timers = 1;
3842 write_unlock_bh(&bond->lock);
3844 if (bond->params.miimon) { /* link check interval, in milliseconds. */
3845 cancel_delayed_work(&bond->mii_work);
3848 if (bond->params.arp_interval) { /* arp interval, in milliseconds. */
3849 cancel_delayed_work(&bond->arp_work);
3852 switch (bond->params.mode) {
3853 case BOND_MODE_8023AD:
3854 cancel_delayed_work(&bond->ad_work);
3858 cancel_delayed_work(&bond->alb_work);
3864 if (delayed_work_pending(&bond->mcast_work))
3865 cancel_delayed_work(&bond->mcast_work);
3867 if (bond_is_lb(bond)) {
3868 /* Must be called only after all
3869 * slaves have been released
3871 bond_alb_deinitialize(bond);
3877 static struct rtnl_link_stats64 *bond_get_stats(struct net_device *bond_dev,
3878 struct rtnl_link_stats64 *stats)
3880 struct bonding *bond = netdev_priv(bond_dev);
3881 struct rtnl_link_stats64 temp;
3882 struct slave *slave;
3885 memset(stats, 0, sizeof(*stats));
3887 read_lock_bh(&bond->lock);
3889 bond_for_each_slave(bond, slave, i) {
3890 const struct rtnl_link_stats64 *sstats =
3891 dev_get_stats(slave->dev, &temp);
3893 stats->rx_packets += sstats->rx_packets;
3894 stats->rx_bytes += sstats->rx_bytes;
3895 stats->rx_errors += sstats->rx_errors;
3896 stats->rx_dropped += sstats->rx_dropped;
3898 stats->tx_packets += sstats->tx_packets;
3899 stats->tx_bytes += sstats->tx_bytes;
3900 stats->tx_errors += sstats->tx_errors;
3901 stats->tx_dropped += sstats->tx_dropped;
3903 stats->multicast += sstats->multicast;
3904 stats->collisions += sstats->collisions;
3906 stats->rx_length_errors += sstats->rx_length_errors;
3907 stats->rx_over_errors += sstats->rx_over_errors;
3908 stats->rx_crc_errors += sstats->rx_crc_errors;
3909 stats->rx_frame_errors += sstats->rx_frame_errors;
3910 stats->rx_fifo_errors += sstats->rx_fifo_errors;
3911 stats->rx_missed_errors += sstats->rx_missed_errors;
3913 stats->tx_aborted_errors += sstats->tx_aborted_errors;
3914 stats->tx_carrier_errors += sstats->tx_carrier_errors;
3915 stats->tx_fifo_errors += sstats->tx_fifo_errors;
3916 stats->tx_heartbeat_errors += sstats->tx_heartbeat_errors;
3917 stats->tx_window_errors += sstats->tx_window_errors;
3920 read_unlock_bh(&bond->lock);
3925 static int bond_do_ioctl(struct net_device *bond_dev, struct ifreq *ifr, int cmd)
3927 struct net_device *slave_dev = NULL;
3928 struct ifbond k_binfo;
3929 struct ifbond __user *u_binfo = NULL;
3930 struct ifslave k_sinfo;
3931 struct ifslave __user *u_sinfo = NULL;
3932 struct mii_ioctl_data *mii = NULL;
3935 pr_debug("bond_ioctl: master=%s, cmd=%d\n", bond_dev->name, cmd);
3947 * We do this again just in case we were called by SIOCGMIIREG
3948 * instead of SIOCGMIIPHY.
3955 if (mii->reg_num == 1) {
3956 struct bonding *bond = netdev_priv(bond_dev);
3958 read_lock(&bond->lock);
3959 read_lock(&bond->curr_slave_lock);
3960 if (netif_carrier_ok(bond->dev))
3961 mii->val_out = BMSR_LSTATUS;
3963 read_unlock(&bond->curr_slave_lock);
3964 read_unlock(&bond->lock);
3968 case BOND_INFO_QUERY_OLD:
3969 case SIOCBONDINFOQUERY:
3970 u_binfo = (struct ifbond __user *)ifr->ifr_data;
3972 if (copy_from_user(&k_binfo, u_binfo, sizeof(ifbond)))
3975 res = bond_info_query(bond_dev, &k_binfo);
3977 copy_to_user(u_binfo, &k_binfo, sizeof(ifbond)))
3981 case BOND_SLAVE_INFO_QUERY_OLD:
3982 case SIOCBONDSLAVEINFOQUERY:
3983 u_sinfo = (struct ifslave __user *)ifr->ifr_data;
3985 if (copy_from_user(&k_sinfo, u_sinfo, sizeof(ifslave)))
3988 res = bond_slave_info_query(bond_dev, &k_sinfo);
3990 copy_to_user(u_sinfo, &k_sinfo, sizeof(ifslave)))
3999 if (!capable(CAP_NET_ADMIN))
4002 slave_dev = dev_get_by_name(dev_net(bond_dev), ifr->ifr_slave);
4004 pr_debug("slave_dev=%p:\n", slave_dev);
4009 pr_debug("slave_dev->name=%s:\n", slave_dev->name);
4011 case BOND_ENSLAVE_OLD:
4012 case SIOCBONDENSLAVE:
4013 res = bond_enslave(bond_dev, slave_dev);
4015 case BOND_RELEASE_OLD:
4016 case SIOCBONDRELEASE:
4017 res = bond_release(bond_dev, slave_dev);
4019 case BOND_SETHWADDR_OLD:
4020 case SIOCBONDSETHWADDR:
4021 res = bond_sethwaddr(bond_dev, slave_dev);
4023 case BOND_CHANGE_ACTIVE_OLD:
4024 case SIOCBONDCHANGEACTIVE:
4025 res = bond_ioctl_change_active(bond_dev, slave_dev);
4037 static bool bond_addr_in_mc_list(unsigned char *addr,
4038 struct netdev_hw_addr_list *list,
4041 struct netdev_hw_addr *ha;
4043 netdev_hw_addr_list_for_each(ha, list)
4044 if (!memcmp(ha->addr, addr, addrlen))
4050 static void bond_set_multicast_list(struct net_device *bond_dev)
4052 struct bonding *bond = netdev_priv(bond_dev);
4053 struct netdev_hw_addr *ha;
4057 * Do promisc before checking multicast_mode
4059 if ((bond_dev->flags & IFF_PROMISC) && !(bond->flags & IFF_PROMISC))
4061 * FIXME: Need to handle the error when one of the multi-slaves
4064 bond_set_promiscuity(bond, 1);
4067 if (!(bond_dev->flags & IFF_PROMISC) && (bond->flags & IFF_PROMISC))
4068 bond_set_promiscuity(bond, -1);
4071 /* set allmulti flag to slaves */
4072 if ((bond_dev->flags & IFF_ALLMULTI) && !(bond->flags & IFF_ALLMULTI))
4074 * FIXME: Need to handle the error when one of the multi-slaves
4077 bond_set_allmulti(bond, 1);
4080 if (!(bond_dev->flags & IFF_ALLMULTI) && (bond->flags & IFF_ALLMULTI))
4081 bond_set_allmulti(bond, -1);
4084 read_lock(&bond->lock);
4086 bond->flags = bond_dev->flags;
4088 /* looking for addresses to add to slaves' mc list */
4089 netdev_for_each_mc_addr(ha, bond_dev) {
4090 found = bond_addr_in_mc_list(ha->addr, &bond->mc_list,
4091 bond_dev->addr_len);
4093 bond_mc_add(bond, ha->addr);
4096 /* looking for addresses to delete from slaves' list */
4097 netdev_hw_addr_list_for_each(ha, &bond->mc_list) {
4098 found = bond_addr_in_mc_list(ha->addr, &bond_dev->mc,
4099 bond_dev->addr_len);
4101 bond_mc_del(bond, ha->addr);
4104 /* save master's multicast list */
4105 __hw_addr_flush(&bond->mc_list);
4106 __hw_addr_add_multiple(&bond->mc_list, &bond_dev->mc,
4107 bond_dev->addr_len, NETDEV_HW_ADDR_T_MULTICAST);
4109 read_unlock(&bond->lock);
4112 static int bond_neigh_setup(struct net_device *dev, struct neigh_parms *parms)
4114 struct bonding *bond = netdev_priv(dev);
4115 struct slave *slave = bond->first_slave;
4118 const struct net_device_ops *slave_ops
4119 = slave->dev->netdev_ops;
4120 if (slave_ops->ndo_neigh_setup)
4121 return slave_ops->ndo_neigh_setup(slave->dev, parms);
4127 * Change the MTU of all of a master's slaves to match the master
4129 static int bond_change_mtu(struct net_device *bond_dev, int new_mtu)
4131 struct bonding *bond = netdev_priv(bond_dev);
4132 struct slave *slave, *stop_at;
4136 pr_debug("bond=%p, name=%s, new_mtu=%d\n", bond,
4137 (bond_dev ? bond_dev->name : "None"), new_mtu);
4139 /* Can't hold bond->lock with bh disabled here since
4140 * some base drivers panic. On the other hand we can't
4141 * hold bond->lock without bh disabled because we'll
4142 * deadlock. The only solution is to rely on the fact
4143 * that we're under rtnl_lock here, and the slaves
4144 * list won't change. This doesn't solve the problem
4145 * of setting the slave's MTU while it is
4146 * transmitting, but the assumption is that the base
4147 * driver can handle that.
4149 * TODO: figure out a way to safely iterate the slaves
4150 * list, but without holding a lock around the actual
4151 * call to the base driver.
4154 bond_for_each_slave(bond, slave, i) {
4155 pr_debug("s %p s->p %p c_m %p\n",
4158 slave->dev->netdev_ops->ndo_change_mtu);
4160 res = dev_set_mtu(slave->dev, new_mtu);
4163 /* If we failed to set the slave's mtu to the new value
4164 * we must abort the operation even in ACTIVE_BACKUP
4165 * mode, because if we allow the backup slaves to have
4166 * different mtu values than the active slave we'll
4167 * need to change their mtu when doing a failover. That
4168 * means changing their mtu from timer context, which
4169 * is probably not a good idea.
4171 pr_debug("err %d %s\n", res, slave->dev->name);
4176 bond_dev->mtu = new_mtu;
4181 /* unwind from head to the slave that failed */
4183 bond_for_each_slave_from_to(bond, slave, i, bond->first_slave, stop_at) {
4186 tmp_res = dev_set_mtu(slave->dev, bond_dev->mtu);
4188 pr_debug("unwind err %d dev %s\n",
4189 tmp_res, slave->dev->name);
4199 * Note that many devices must be down to change the HW address, and
4200 * downing the master releases all slaves. We can make bonds full of
4201 * bonding devices to test this, however.
4203 static int bond_set_mac_address(struct net_device *bond_dev, void *addr)
4205 struct bonding *bond = netdev_priv(bond_dev);
4206 struct sockaddr *sa = addr, tmp_sa;
4207 struct slave *slave, *stop_at;
4211 if (bond->params.mode == BOND_MODE_ALB)
4212 return bond_alb_set_mac_address(bond_dev, addr);
4215 pr_debug("bond=%p, name=%s\n",
4216 bond, bond_dev ? bond_dev->name : "None");
4219 * If fail_over_mac is set to active, do nothing and return
4220 * success. Returning an error causes ifenslave to fail.
4222 if (bond->params.fail_over_mac == BOND_FOM_ACTIVE)
4225 if (!is_valid_ether_addr(sa->sa_data))
4226 return -EADDRNOTAVAIL;
4228 /* Can't hold bond->lock with bh disabled here since
4229 * some base drivers panic. On the other hand we can't
4230 * hold bond->lock without bh disabled because we'll
4231 * deadlock. The only solution is to rely on the fact
4232 * that we're under rtnl_lock here, and the slaves
4233 * list won't change. This doesn't solve the problem
4234 * of setting the slave's hw address while it is
4235 * transmitting, but the assumption is that the base
4236 * driver can handle that.
4238 * TODO: figure out a way to safely iterate the slaves
4239 * list, but without holding a lock around the actual
4240 * call to the base driver.
4243 bond_for_each_slave(bond, slave, i) {
4244 const struct net_device_ops *slave_ops = slave->dev->netdev_ops;
4245 pr_debug("slave %p %s\n", slave, slave->dev->name);
4247 if (slave_ops->ndo_set_mac_address == NULL) {
4249 pr_debug("EOPNOTSUPP %s\n", slave->dev->name);
4253 res = dev_set_mac_address(slave->dev, addr);
4255 /* TODO: consider downing the slave
4257 * User should expect communications
4258 * breakage anyway until ARP finish
4261 pr_debug("err %d %s\n", res, slave->dev->name);
4267 memcpy(bond_dev->dev_addr, sa->sa_data, bond_dev->addr_len);
4271 memcpy(tmp_sa.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
4272 tmp_sa.sa_family = bond_dev->type;
4274 /* unwind from head to the slave that failed */
4276 bond_for_each_slave_from_to(bond, slave, i, bond->first_slave, stop_at) {
4279 tmp_res = dev_set_mac_address(slave->dev, &tmp_sa);
4281 pr_debug("unwind err %d dev %s\n",
4282 tmp_res, slave->dev->name);
4289 static int bond_xmit_roundrobin(struct sk_buff *skb, struct net_device *bond_dev)
4291 struct bonding *bond = netdev_priv(bond_dev);
4292 struct slave *slave, *start_at;
4293 int i, slave_no, res = 1;
4294 struct iphdr *iph = ip_hdr(skb);
4296 read_lock(&bond->lock);
4298 if (!BOND_IS_OK(bond))
4301 * Start with the curr_active_slave that joined the bond as the
4302 * default for sending IGMP traffic. For failover purposes one
4303 * needs to maintain some consistency for the interface that will
4304 * send the join/membership reports. The curr_active_slave found
4305 * will send all of this type of traffic.
4307 if ((iph->protocol == IPPROTO_IGMP) &&
4308 (skb->protocol == htons(ETH_P_IP))) {
4310 read_lock(&bond->curr_slave_lock);
4311 slave = bond->curr_active_slave;
4312 read_unlock(&bond->curr_slave_lock);
4318 * Concurrent TX may collide on rr_tx_counter; we accept
4319 * that as being rare enough not to justify using an
4322 slave_no = bond->rr_tx_counter++ % bond->slave_cnt;
4324 bond_for_each_slave(bond, slave, i) {
4332 bond_for_each_slave_from(bond, slave, i, start_at) {
4333 if (IS_UP(slave->dev) &&
4334 (slave->link == BOND_LINK_UP) &&
4335 (slave->state == BOND_STATE_ACTIVE)) {
4336 res = bond_dev_queue_xmit(bond, skb, slave->dev);
4343 /* no suitable interface, frame not sent */
4346 read_unlock(&bond->lock);
4347 return NETDEV_TX_OK;
4352 * in active-backup mode, we know that bond->curr_active_slave is always valid if
4353 * the bond has a usable interface.
4355 static int bond_xmit_activebackup(struct sk_buff *skb, struct net_device *bond_dev)
4357 struct bonding *bond = netdev_priv(bond_dev);
4360 read_lock(&bond->lock);
4361 read_lock(&bond->curr_slave_lock);
4363 if (!BOND_IS_OK(bond))
4366 if (!bond->curr_active_slave)
4369 res = bond_dev_queue_xmit(bond, skb, bond->curr_active_slave->dev);
4373 /* no suitable interface, frame not sent */
4376 read_unlock(&bond->curr_slave_lock);
4377 read_unlock(&bond->lock);
4378 return NETDEV_TX_OK;
4382 * In bond_xmit_xor() , we determine the output device by using a pre-
4383 * determined xmit_hash_policy(), If the selected device is not enabled,
4384 * find the next active slave.
4386 static int bond_xmit_xor(struct sk_buff *skb, struct net_device *bond_dev)
4388 struct bonding *bond = netdev_priv(bond_dev);
4389 struct slave *slave, *start_at;
4394 read_lock(&bond->lock);
4396 if (!BOND_IS_OK(bond))
4399 slave_no = bond->xmit_hash_policy(skb, bond->slave_cnt);
4401 bond_for_each_slave(bond, slave, i) {
4409 bond_for_each_slave_from(bond, slave, i, start_at) {
4410 if (IS_UP(slave->dev) &&
4411 (slave->link == BOND_LINK_UP) &&
4412 (slave->state == BOND_STATE_ACTIVE)) {
4413 res = bond_dev_queue_xmit(bond, skb, slave->dev);
4420 /* no suitable interface, frame not sent */
4423 read_unlock(&bond->lock);
4424 return NETDEV_TX_OK;
4428 * in broadcast mode, we send everything to all usable interfaces.
4430 static int bond_xmit_broadcast(struct sk_buff *skb, struct net_device *bond_dev)
4432 struct bonding *bond = netdev_priv(bond_dev);
4433 struct slave *slave, *start_at;
4434 struct net_device *tx_dev = NULL;
4438 read_lock(&bond->lock);
4440 if (!BOND_IS_OK(bond))
4443 read_lock(&bond->curr_slave_lock);
4444 start_at = bond->curr_active_slave;
4445 read_unlock(&bond->curr_slave_lock);
4450 bond_for_each_slave_from(bond, slave, i, start_at) {
4451 if (IS_UP(slave->dev) &&
4452 (slave->link == BOND_LINK_UP) &&
4453 (slave->state == BOND_STATE_ACTIVE)) {
4455 struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
4457 pr_err("%s: Error: bond_xmit_broadcast(): skb_clone() failed\n",
4462 res = bond_dev_queue_xmit(bond, skb2, tx_dev);
4464 dev_kfree_skb(skb2);
4468 tx_dev = slave->dev;
4473 res = bond_dev_queue_xmit(bond, skb, tx_dev);
4477 /* no suitable interface, frame not sent */
4480 /* frame sent to all suitable interfaces */
4481 read_unlock(&bond->lock);
4482 return NETDEV_TX_OK;
4485 /*------------------------- Device initialization ---------------------------*/
4487 static void bond_set_xmit_hash_policy(struct bonding *bond)
4489 switch (bond->params.xmit_policy) {
4490 case BOND_XMIT_POLICY_LAYER23:
4491 bond->xmit_hash_policy = bond_xmit_hash_policy_l23;
4493 case BOND_XMIT_POLICY_LAYER34:
4494 bond->xmit_hash_policy = bond_xmit_hash_policy_l34;
4496 case BOND_XMIT_POLICY_LAYER2:
4498 bond->xmit_hash_policy = bond_xmit_hash_policy_l2;
4504 * Lookup the slave that corresponds to a qid
4506 static inline int bond_slave_override(struct bonding *bond,
4507 struct sk_buff *skb)
4510 struct slave *slave = NULL;
4511 struct slave *check_slave;
4513 read_lock(&bond->lock);
4515 if (!BOND_IS_OK(bond) || !skb->queue_mapping)
4518 /* Find out if any slaves have the same mapping as this skb. */
4519 bond_for_each_slave(bond, check_slave, i) {
4520 if (check_slave->queue_id == skb->queue_mapping) {
4521 slave = check_slave;
4526 /* If the slave isn't UP, use default transmit policy. */
4527 if (slave && slave->queue_id && IS_UP(slave->dev) &&
4528 (slave->link == BOND_LINK_UP)) {
4529 res = bond_dev_queue_xmit(bond, skb, slave->dev);
4533 read_unlock(&bond->lock);
4537 static u16 bond_select_queue(struct net_device *dev, struct sk_buff *skb)
4540 * This helper function exists to help dev_pick_tx get the correct
4541 * destination queue. Using a helper function skips the a call to
4542 * skb_tx_hash and will put the skbs in the queue we expect on their
4543 * way down to the bonding driver.
4545 return skb->queue_mapping;
4548 static netdev_tx_t bond_start_xmit(struct sk_buff *skb, struct net_device *dev)
4550 struct bonding *bond = netdev_priv(dev);
4552 if (TX_QUEUE_OVERRIDE(bond->params.mode)) {
4553 if (!bond_slave_override(bond, skb))
4554 return NETDEV_TX_OK;
4557 switch (bond->params.mode) {
4558 case BOND_MODE_ROUNDROBIN:
4559 return bond_xmit_roundrobin(skb, dev);
4560 case BOND_MODE_ACTIVEBACKUP:
4561 return bond_xmit_activebackup(skb, dev);
4563 return bond_xmit_xor(skb, dev);
4564 case BOND_MODE_BROADCAST:
4565 return bond_xmit_broadcast(skb, dev);
4566 case BOND_MODE_8023AD:
4567 return bond_3ad_xmit_xor(skb, dev);
4570 return bond_alb_xmit(skb, dev);
4572 /* Should never happen, mode already checked */
4573 pr_err("%s: Error: Unknown bonding mode %d\n",
4574 dev->name, bond->params.mode);
4577 return NETDEV_TX_OK;
4583 * set bond mode specific net device operations
4585 void bond_set_mode_ops(struct bonding *bond, int mode)
4587 struct net_device *bond_dev = bond->dev;
4590 case BOND_MODE_ROUNDROBIN:
4592 case BOND_MODE_ACTIVEBACKUP:
4595 bond_set_xmit_hash_policy(bond);
4597 case BOND_MODE_BROADCAST:
4599 case BOND_MODE_8023AD:
4600 bond_set_master_3ad_flags(bond);
4601 bond_set_xmit_hash_policy(bond);
4604 bond_set_master_alb_flags(bond);
4609 /* Should never happen, mode already checked */
4610 pr_err("%s: Error: Unknown bonding mode %d\n",
4611 bond_dev->name, mode);
4616 static void bond_ethtool_get_drvinfo(struct net_device *bond_dev,
4617 struct ethtool_drvinfo *drvinfo)
4619 strncpy(drvinfo->driver, DRV_NAME, 32);
4620 strncpy(drvinfo->version, DRV_VERSION, 32);
4621 snprintf(drvinfo->fw_version, 32, "%d", BOND_ABI_VERSION);
4624 static const struct ethtool_ops bond_ethtool_ops = {
4625 .get_drvinfo = bond_ethtool_get_drvinfo,
4626 .get_link = ethtool_op_get_link,
4627 .get_tx_csum = ethtool_op_get_tx_csum,
4628 .get_sg = ethtool_op_get_sg,
4629 .get_tso = ethtool_op_get_tso,
4630 .get_ufo = ethtool_op_get_ufo,
4631 .get_flags = ethtool_op_get_flags,
4634 static const struct net_device_ops bond_netdev_ops = {
4635 .ndo_init = bond_init,
4636 .ndo_uninit = bond_uninit,
4637 .ndo_open = bond_open,
4638 .ndo_stop = bond_close,
4639 .ndo_start_xmit = bond_start_xmit,
4640 .ndo_select_queue = bond_select_queue,
4641 .ndo_get_stats64 = bond_get_stats,
4642 .ndo_do_ioctl = bond_do_ioctl,
4643 .ndo_set_multicast_list = bond_set_multicast_list,
4644 .ndo_change_mtu = bond_change_mtu,
4645 .ndo_set_mac_address = bond_set_mac_address,
4646 .ndo_neigh_setup = bond_neigh_setup,
4647 .ndo_vlan_rx_register = bond_vlan_rx_register,
4648 .ndo_vlan_rx_add_vid = bond_vlan_rx_add_vid,
4649 .ndo_vlan_rx_kill_vid = bond_vlan_rx_kill_vid,
4650 #ifdef CONFIG_NET_POLL_CONTROLLER
4651 .ndo_netpoll_cleanup = bond_netpoll_cleanup,
4652 .ndo_poll_controller = bond_poll_controller,
4656 static void bond_destructor(struct net_device *bond_dev)
4658 struct bonding *bond = netdev_priv(bond_dev);
4660 destroy_workqueue(bond->wq);
4661 free_netdev(bond_dev);
4664 static void bond_setup(struct net_device *bond_dev)
4666 struct bonding *bond = netdev_priv(bond_dev);
4668 /* initialize rwlocks */
4669 rwlock_init(&bond->lock);
4670 rwlock_init(&bond->curr_slave_lock);
4672 bond->params = bonding_defaults;
4674 /* Initialize pointers */
4675 bond->dev = bond_dev;
4676 INIT_LIST_HEAD(&bond->vlan_list);
4678 /* Initialize the device entry points */
4679 ether_setup(bond_dev);
4680 bond_dev->netdev_ops = &bond_netdev_ops;
4681 bond_dev->ethtool_ops = &bond_ethtool_ops;
4682 bond_set_mode_ops(bond, bond->params.mode);
4684 bond_dev->destructor = bond_destructor;
4686 /* Initialize the device options */
4687 bond_dev->tx_queue_len = 0;
4688 bond_dev->flags |= IFF_MASTER|IFF_MULTICAST;
4689 bond_dev->priv_flags |= IFF_BONDING;
4690 bond_dev->priv_flags &= ~IFF_XMIT_DST_RELEASE;
4692 if (bond->params.arp_interval)
4693 bond_dev->priv_flags |= IFF_MASTER_ARPMON;
4695 /* At first, we block adding VLANs. That's the only way to
4696 * prevent problems that occur when adding VLANs over an
4697 * empty bond. The block will be removed once non-challenged
4698 * slaves are enslaved.
4700 bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
4702 /* don't acquire bond device's netif_tx_lock when
4704 bond_dev->features |= NETIF_F_LLTX;
4706 /* By default, we declare the bond to be fully
4707 * VLAN hardware accelerated capable. Special
4708 * care is taken in the various xmit functions
4709 * when there are slaves that are not hw accel
4712 bond_dev->features |= (NETIF_F_HW_VLAN_TX |
4713 NETIF_F_HW_VLAN_RX |
4714 NETIF_F_HW_VLAN_FILTER);
4716 /* By default, we enable GRO on bonding devices.
4717 * Actual support requires lowlevel drivers are GRO ready.
4719 bond_dev->features |= NETIF_F_GRO;
4722 static void bond_work_cancel_all(struct bonding *bond)
4724 write_lock_bh(&bond->lock);
4725 bond->kill_timers = 1;
4726 write_unlock_bh(&bond->lock);
4728 if (bond->params.miimon && delayed_work_pending(&bond->mii_work))
4729 cancel_delayed_work(&bond->mii_work);
4731 if (bond->params.arp_interval && delayed_work_pending(&bond->arp_work))
4732 cancel_delayed_work(&bond->arp_work);
4734 if (bond->params.mode == BOND_MODE_ALB &&
4735 delayed_work_pending(&bond->alb_work))
4736 cancel_delayed_work(&bond->alb_work);
4738 if (bond->params.mode == BOND_MODE_8023AD &&
4739 delayed_work_pending(&bond->ad_work))
4740 cancel_delayed_work(&bond->ad_work);
4742 if (delayed_work_pending(&bond->mcast_work))
4743 cancel_delayed_work(&bond->mcast_work);
4747 * Destroy a bonding device.
4748 * Must be under rtnl_lock when this function is called.
4750 static void bond_uninit(struct net_device *bond_dev)
4752 struct bonding *bond = netdev_priv(bond_dev);
4753 struct vlan_entry *vlan, *tmp;
4755 bond_netpoll_cleanup(bond_dev);
4757 /* Release the bonded slaves */
4758 bond_release_all(bond_dev);
4760 list_del(&bond->bond_list);
4762 bond_work_cancel_all(bond);
4764 bond_remove_proc_entry(bond);
4766 __hw_addr_flush(&bond->mc_list);
4768 list_for_each_entry_safe(vlan, tmp, &bond->vlan_list, vlan_list) {
4769 list_del(&vlan->vlan_list);
4774 /*------------------------- Module initialization ---------------------------*/
4777 * Convert string input module parms. Accept either the
4778 * number of the mode or its string name. A bit complicated because
4779 * some mode names are substrings of other names, and calls from sysfs
4780 * may have whitespace in the name (trailing newlines, for example).
4782 int bond_parse_parm(const char *buf, const struct bond_parm_tbl *tbl)
4784 int modeint = -1, i, rv;
4785 char *p, modestr[BOND_MAX_MODENAME_LEN + 1] = { 0, };
4787 for (p = (char *)buf; *p; p++)
4788 if (!(isdigit(*p) || isspace(*p)))
4792 rv = sscanf(buf, "%20s", modestr);
4794 rv = sscanf(buf, "%d", &modeint);
4799 for (i = 0; tbl[i].modename; i++) {
4800 if (modeint == tbl[i].mode)
4802 if (strcmp(modestr, tbl[i].modename) == 0)
4809 static int bond_check_params(struct bond_params *params)
4811 int arp_validate_value, fail_over_mac_value, primary_reselect_value;
4814 * Convert string parameters.
4817 bond_mode = bond_parse_parm(mode, bond_mode_tbl);
4818 if (bond_mode == -1) {
4819 pr_err("Error: Invalid bonding mode \"%s\"\n",
4820 mode == NULL ? "NULL" : mode);
4825 if (xmit_hash_policy) {
4826 if ((bond_mode != BOND_MODE_XOR) &&
4827 (bond_mode != BOND_MODE_8023AD)) {
4828 pr_info("xmit_hash_policy param is irrelevant in mode %s\n",
4829 bond_mode_name(bond_mode));
4831 xmit_hashtype = bond_parse_parm(xmit_hash_policy,
4833 if (xmit_hashtype == -1) {
4834 pr_err("Error: Invalid xmit_hash_policy \"%s\"\n",
4835 xmit_hash_policy == NULL ? "NULL" :
4843 if (bond_mode != BOND_MODE_8023AD) {
4844 pr_info("lacp_rate param is irrelevant in mode %s\n",
4845 bond_mode_name(bond_mode));
4847 lacp_fast = bond_parse_parm(lacp_rate, bond_lacp_tbl);
4848 if (lacp_fast == -1) {
4849 pr_err("Error: Invalid lacp rate \"%s\"\n",
4850 lacp_rate == NULL ? "NULL" : lacp_rate);
4857 params->ad_select = bond_parse_parm(ad_select, ad_select_tbl);
4858 if (params->ad_select == -1) {
4859 pr_err("Error: Invalid ad_select \"%s\"\n",
4860 ad_select == NULL ? "NULL" : ad_select);
4864 if (bond_mode != BOND_MODE_8023AD) {
4865 pr_warning("ad_select param only affects 802.3ad mode\n");
4868 params->ad_select = BOND_AD_STABLE;
4871 if (max_bonds < 0) {
4872 pr_warning("Warning: max_bonds (%d) not in range %d-%d, so it was reset to BOND_DEFAULT_MAX_BONDS (%d)\n",
4873 max_bonds, 0, INT_MAX, BOND_DEFAULT_MAX_BONDS);
4874 max_bonds = BOND_DEFAULT_MAX_BONDS;
4878 pr_warning("Warning: miimon module parameter (%d), not in range 0-%d, so it was reset to %d\n",
4879 miimon, INT_MAX, BOND_LINK_MON_INTERV);
4880 miimon = BOND_LINK_MON_INTERV;
4884 pr_warning("Warning: updelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4889 if (downdelay < 0) {
4890 pr_warning("Warning: downdelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4891 downdelay, INT_MAX);
4895 if ((use_carrier != 0) && (use_carrier != 1)) {
4896 pr_warning("Warning: use_carrier module parameter (%d), not of valid value (0/1), so it was set to 1\n",
4901 if (num_grat_arp < 0 || num_grat_arp > 255) {
4902 pr_warning("Warning: num_grat_arp (%d) not in range 0-255 so it was reset to 1\n",
4907 if (num_unsol_na < 0 || num_unsol_na > 255) {
4908 pr_warning("Warning: num_unsol_na (%d) not in range 0-255 so it was reset to 1\n",
4913 /* reset values for 802.3ad */
4914 if (bond_mode == BOND_MODE_8023AD) {
4916 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");
4917 pr_warning("Forcing miimon to 100msec\n");
4922 if (tx_queues < 1 || tx_queues > 255) {
4923 pr_warning("Warning: tx_queues (%d) should be between "
4924 "1 and 255, resetting to %d\n",
4925 tx_queues, BOND_DEFAULT_TX_QUEUES);
4926 tx_queues = BOND_DEFAULT_TX_QUEUES;
4929 if ((all_slaves_active != 0) && (all_slaves_active != 1)) {
4930 pr_warning("Warning: all_slaves_active module parameter (%d), "
4931 "not of valid value (0/1), so it was set to "
4932 "0\n", all_slaves_active);
4933 all_slaves_active = 0;
4936 /* reset values for TLB/ALB */
4937 if ((bond_mode == BOND_MODE_TLB) ||
4938 (bond_mode == BOND_MODE_ALB)) {
4940 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");
4941 pr_warning("Forcing miimon to 100msec\n");
4946 if (bond_mode == BOND_MODE_ALB) {
4947 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",
4952 if (updelay || downdelay) {
4953 /* just warn the user the up/down delay will have
4954 * no effect since miimon is zero...
4956 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",
4957 updelay, downdelay);
4960 /* don't allow arp monitoring */
4962 pr_warning("Warning: miimon (%d) and arp_interval (%d) can't be used simultaneously, disabling ARP monitoring\n",
4963 miimon, arp_interval);
4967 if ((updelay % miimon) != 0) {
4968 pr_warning("Warning: updelay (%d) is not a multiple of miimon (%d), updelay rounded to %d ms\n",
4970 (updelay / miimon) * miimon);
4975 if ((downdelay % miimon) != 0) {
4976 pr_warning("Warning: downdelay (%d) is not a multiple of miimon (%d), downdelay rounded to %d ms\n",
4978 (downdelay / miimon) * miimon);
4981 downdelay /= miimon;
4984 if (arp_interval < 0) {
4985 pr_warning("Warning: arp_interval module parameter (%d) , not in range 0-%d, so it was reset to %d\n",
4986 arp_interval, INT_MAX, BOND_LINK_ARP_INTERV);
4987 arp_interval = BOND_LINK_ARP_INTERV;
4990 for (arp_ip_count = 0;
4991 (arp_ip_count < BOND_MAX_ARP_TARGETS) && arp_ip_target[arp_ip_count];
4993 /* not complete check, but should be good enough to
4995 if (!isdigit(arp_ip_target[arp_ip_count][0])) {
4996 pr_warning("Warning: bad arp_ip_target module parameter (%s), ARP monitoring will not be performed\n",
4997 arp_ip_target[arp_ip_count]);
5000 __be32 ip = in_aton(arp_ip_target[arp_ip_count]);
5001 arp_target[arp_ip_count] = ip;
5005 if (arp_interval && !arp_ip_count) {
5006 /* don't allow arping if no arp_ip_target given... */
5007 pr_warning("Warning: arp_interval module parameter (%d) specified without providing an arp_ip_target parameter, arp_interval was reset to 0\n",
5013 if (bond_mode != BOND_MODE_ACTIVEBACKUP) {
5014 pr_err("arp_validate only supported in active-backup mode\n");
5017 if (!arp_interval) {
5018 pr_err("arp_validate requires arp_interval\n");
5022 arp_validate_value = bond_parse_parm(arp_validate,
5024 if (arp_validate_value == -1) {
5025 pr_err("Error: invalid arp_validate \"%s\"\n",
5026 arp_validate == NULL ? "NULL" : arp_validate);
5030 arp_validate_value = 0;
5033 pr_info("MII link monitoring set to %d ms\n", miimon);
5034 } else if (arp_interval) {
5037 pr_info("ARP monitoring set to %d ms, validate %s, with %d target(s):",
5039 arp_validate_tbl[arp_validate_value].modename,
5042 for (i = 0; i < arp_ip_count; i++)
5043 pr_info(" %s", arp_ip_target[i]);
5047 } else if (max_bonds) {
5048 /* miimon and arp_interval not set, we need one so things
5049 * work as expected, see bonding.txt for details
5051 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");
5054 if (primary && !USES_PRIMARY(bond_mode)) {
5055 /* currently, using a primary only makes sense
5056 * in active backup, TLB or ALB modes
5058 pr_warning("Warning: %s primary device specified but has no effect in %s mode\n",
5059 primary, bond_mode_name(bond_mode));
5063 if (primary && primary_reselect) {
5064 primary_reselect_value = bond_parse_parm(primary_reselect,
5066 if (primary_reselect_value == -1) {
5067 pr_err("Error: Invalid primary_reselect \"%s\"\n",
5069 NULL ? "NULL" : primary_reselect);
5073 primary_reselect_value = BOND_PRI_RESELECT_ALWAYS;
5076 if (fail_over_mac) {
5077 fail_over_mac_value = bond_parse_parm(fail_over_mac,
5079 if (fail_over_mac_value == -1) {
5080 pr_err("Error: invalid fail_over_mac \"%s\"\n",
5081 arp_validate == NULL ? "NULL" : arp_validate);
5085 if (bond_mode != BOND_MODE_ACTIVEBACKUP)
5086 pr_warning("Warning: fail_over_mac only affects active-backup mode.\n");
5088 fail_over_mac_value = BOND_FOM_NONE;
5091 /* fill params struct with the proper values */
5092 params->mode = bond_mode;
5093 params->xmit_policy = xmit_hashtype;
5094 params->miimon = miimon;
5095 params->num_grat_arp = num_grat_arp;
5096 params->num_unsol_na = num_unsol_na;
5097 params->arp_interval = arp_interval;
5098 params->arp_validate = arp_validate_value;
5099 params->updelay = updelay;
5100 params->downdelay = downdelay;
5101 params->use_carrier = use_carrier;
5102 params->lacp_fast = lacp_fast;
5103 params->primary[0] = 0;
5104 params->primary_reselect = primary_reselect_value;
5105 params->fail_over_mac = fail_over_mac_value;
5106 params->tx_queues = tx_queues;
5107 params->all_slaves_active = all_slaves_active;
5110 strncpy(params->primary, primary, IFNAMSIZ);
5111 params->primary[IFNAMSIZ - 1] = 0;
5114 memcpy(params->arp_targets, arp_target, sizeof(arp_target));
5119 static struct lock_class_key bonding_netdev_xmit_lock_key;
5120 static struct lock_class_key bonding_netdev_addr_lock_key;
5122 static void bond_set_lockdep_class_one(struct net_device *dev,
5123 struct netdev_queue *txq,
5126 lockdep_set_class(&txq->_xmit_lock,
5127 &bonding_netdev_xmit_lock_key);
5130 static void bond_set_lockdep_class(struct net_device *dev)
5132 lockdep_set_class(&dev->addr_list_lock,
5133 &bonding_netdev_addr_lock_key);
5134 netdev_for_each_tx_queue(dev, bond_set_lockdep_class_one, NULL);
5138 * Called from registration process
5140 static int bond_init(struct net_device *bond_dev)
5142 struct bonding *bond = netdev_priv(bond_dev);
5143 struct bond_net *bn = net_generic(dev_net(bond_dev), bond_net_id);
5145 pr_debug("Begin bond_init for %s\n", bond_dev->name);
5147 bond->wq = create_singlethread_workqueue(bond_dev->name);
5151 bond_set_lockdep_class(bond_dev);
5153 netif_carrier_off(bond_dev);
5155 bond_create_proc_entry(bond);
5156 list_add_tail(&bond->bond_list, &bn->dev_list);
5158 bond_prepare_sysfs_group(bond);
5160 __hw_addr_init(&bond->mc_list);
5164 static int bond_validate(struct nlattr *tb[], struct nlattr *data[])
5166 if (tb[IFLA_ADDRESS]) {
5167 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN)
5169 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS])))
5170 return -EADDRNOTAVAIL;
5175 static struct rtnl_link_ops bond_link_ops __read_mostly = {
5177 .priv_size = sizeof(struct bonding),
5178 .setup = bond_setup,
5179 .validate = bond_validate,
5182 /* Create a new bond based on the specified name and bonding parameters.
5183 * If name is NULL, obtain a suitable "bond%d" name for us.
5184 * Caller must NOT hold rtnl_lock; we need to release it here before we
5185 * set up our sysfs entries.
5187 int bond_create(struct net *net, const char *name)
5189 struct net_device *bond_dev;
5194 bond_dev = alloc_netdev_mq(sizeof(struct bonding), name ? name : "",
5195 bond_setup, tx_queues);
5197 pr_err("%s: eek! can't alloc netdev!\n", name);
5202 dev_net_set(bond_dev, net);
5203 bond_dev->rtnl_link_ops = &bond_link_ops;
5206 res = dev_alloc_name(bond_dev, "bond%d");
5211 res = register_netdevice(bond_dev);
5216 bond_destructor(bond_dev);
5220 static int __net_init bond_net_init(struct net *net)
5222 struct bond_net *bn = net_generic(net, bond_net_id);
5225 INIT_LIST_HEAD(&bn->dev_list);
5227 bond_create_proc_dir(bn);
5232 static void __net_exit bond_net_exit(struct net *net)
5234 struct bond_net *bn = net_generic(net, bond_net_id);
5236 bond_destroy_proc_dir(bn);
5239 static struct pernet_operations bond_net_ops = {
5240 .init = bond_net_init,
5241 .exit = bond_net_exit,
5243 .size = sizeof(struct bond_net),
5246 static int __init bonding_init(void)
5251 pr_info("%s", version);
5253 res = bond_check_params(&bonding_defaults);
5257 res = register_pernet_subsys(&bond_net_ops);
5261 res = rtnl_link_register(&bond_link_ops);
5265 for (i = 0; i < max_bonds; i++) {
5266 res = bond_create(&init_net, NULL);
5271 res = bond_create_sysfs();
5275 register_netdevice_notifier(&bond_netdev_notifier);
5276 register_inetaddr_notifier(&bond_inetaddr_notifier);
5277 bond_register_ipv6_notifier();
5281 rtnl_link_unregister(&bond_link_ops);
5283 unregister_pernet_subsys(&bond_net_ops);
5288 static void __exit bonding_exit(void)
5290 unregister_netdevice_notifier(&bond_netdev_notifier);
5291 unregister_inetaddr_notifier(&bond_inetaddr_notifier);
5292 bond_unregister_ipv6_notifier();
5294 bond_destroy_sysfs();
5296 rtnl_link_unregister(&bond_link_ops);
5297 unregister_pernet_subsys(&bond_net_ops);
5300 module_init(bonding_init);
5301 module_exit(bonding_exit);
5302 MODULE_LICENSE("GPL");
5303 MODULE_VERSION(DRV_VERSION);
5304 MODULE_DESCRIPTION(DRV_DESCRIPTION ", v" DRV_VERSION);
5305 MODULE_AUTHOR("Thomas Davis, tadavis@lbl.gov and many others");
5306 MODULE_ALIAS_RTNL_LINK("bond");