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 <linux/preempt.h>
80 #include <net/route.h>
81 #include <net/net_namespace.h>
82 #include <net/netns/generic.h>
87 /*---------------------------- Module parameters ----------------------------*/
89 /* monitor all links that often (in milliseconds). <=0 disables monitoring */
90 #define BOND_LINK_MON_INTERV 0
91 #define BOND_LINK_ARP_INTERV 0
93 static int max_bonds = BOND_DEFAULT_MAX_BONDS;
94 static int tx_queues = BOND_DEFAULT_TX_QUEUES;
95 static int num_grat_arp = 1;
96 static int num_unsol_na = 1;
97 static int miimon = BOND_LINK_MON_INTERV;
100 static int use_carrier = 1;
102 static char *primary;
103 static char *primary_reselect;
104 static char *lacp_rate;
105 static char *ad_select;
106 static char *xmit_hash_policy;
107 static int arp_interval = BOND_LINK_ARP_INTERV;
108 static char *arp_ip_target[BOND_MAX_ARP_TARGETS];
109 static char *arp_validate;
110 static char *fail_over_mac;
111 static int all_slaves_active = 0;
112 static struct bond_params bonding_defaults;
113 static int resend_igmp = BOND_DEFAULT_RESEND_IGMP;
115 module_param(max_bonds, int, 0);
116 MODULE_PARM_DESC(max_bonds, "Max number of bonded devices");
117 module_param(tx_queues, int, 0);
118 MODULE_PARM_DESC(tx_queues, "Max number of transmit queues (default = 16)");
119 module_param(num_grat_arp, int, 0644);
120 MODULE_PARM_DESC(num_grat_arp, "Number of gratuitous ARP packets to send on failover event");
121 module_param(num_unsol_na, int, 0644);
122 MODULE_PARM_DESC(num_unsol_na, "Number of unsolicited IPv6 Neighbor Advertisements packets to send on failover event");
123 module_param(miimon, int, 0);
124 MODULE_PARM_DESC(miimon, "Link check interval in milliseconds");
125 module_param(updelay, int, 0);
126 MODULE_PARM_DESC(updelay, "Delay before considering link up, in milliseconds");
127 module_param(downdelay, int, 0);
128 MODULE_PARM_DESC(downdelay, "Delay before considering link down, "
130 module_param(use_carrier, int, 0);
131 MODULE_PARM_DESC(use_carrier, "Use netif_carrier_ok (vs MII ioctls) in miimon; "
132 "0 for off, 1 for on (default)");
133 module_param(mode, charp, 0);
134 MODULE_PARM_DESC(mode, "Mode of operation : 0 for balance-rr, "
135 "1 for active-backup, 2 for balance-xor, "
136 "3 for broadcast, 4 for 802.3ad, 5 for balance-tlb, "
137 "6 for balance-alb");
138 module_param(primary, charp, 0);
139 MODULE_PARM_DESC(primary, "Primary network device to use");
140 module_param(primary_reselect, charp, 0);
141 MODULE_PARM_DESC(primary_reselect, "Reselect primary slave "
143 "0 for always (default), "
144 "1 for only if speed of primary is "
146 "2 for only on active slave "
148 module_param(lacp_rate, charp, 0);
149 MODULE_PARM_DESC(lacp_rate, "LACPDU tx rate to request from 802.3ad partner "
151 module_param(ad_select, charp, 0);
152 MODULE_PARM_DESC(ad_select, "803.ad aggregation selection logic: stable (0, default), bandwidth (1), count (2)");
153 module_param(xmit_hash_policy, charp, 0);
154 MODULE_PARM_DESC(xmit_hash_policy, "XOR hashing method: 0 for layer 2 (default)"
155 ", 1 for layer 3+4");
156 module_param(arp_interval, int, 0);
157 MODULE_PARM_DESC(arp_interval, "arp interval in milliseconds");
158 module_param_array(arp_ip_target, charp, NULL, 0);
159 MODULE_PARM_DESC(arp_ip_target, "arp targets in n.n.n.n form");
160 module_param(arp_validate, charp, 0);
161 MODULE_PARM_DESC(arp_validate, "validate src/dst of ARP probes: none (default), active, backup or all");
162 module_param(fail_over_mac, charp, 0);
163 MODULE_PARM_DESC(fail_over_mac, "For active-backup, do not set all slaves to the same MAC. none (default), active or follow");
164 module_param(all_slaves_active, int, 0);
165 MODULE_PARM_DESC(all_slaves_active, "Keep all frames received on an interface"
166 "by setting active flag for all slaves. "
167 "0 for never (default), 1 for always.");
168 module_param(resend_igmp, int, 0);
169 MODULE_PARM_DESC(resend_igmp, "Number of IGMP membership reports to send on link failure");
171 /*----------------------------- Global variables ----------------------------*/
173 #ifdef CONFIG_NET_POLL_CONTROLLER
174 atomic_t netpoll_block_tx = ATOMIC_INIT(0);
177 static const char * const version =
178 DRV_DESCRIPTION ": v" DRV_VERSION " (" DRV_RELDATE ")\n";
180 int bond_net_id __read_mostly;
182 static __be32 arp_target[BOND_MAX_ARP_TARGETS];
183 static int arp_ip_count;
184 static int bond_mode = BOND_MODE_ROUNDROBIN;
185 static int xmit_hashtype = BOND_XMIT_POLICY_LAYER2;
186 static int lacp_fast;
188 const struct bond_parm_tbl bond_lacp_tbl[] = {
189 { "slow", AD_LACP_SLOW},
190 { "fast", AD_LACP_FAST},
194 const struct bond_parm_tbl bond_mode_tbl[] = {
195 { "balance-rr", BOND_MODE_ROUNDROBIN},
196 { "active-backup", BOND_MODE_ACTIVEBACKUP},
197 { "balance-xor", BOND_MODE_XOR},
198 { "broadcast", BOND_MODE_BROADCAST},
199 { "802.3ad", BOND_MODE_8023AD},
200 { "balance-tlb", BOND_MODE_TLB},
201 { "balance-alb", BOND_MODE_ALB},
205 const struct bond_parm_tbl xmit_hashtype_tbl[] = {
206 { "layer2", BOND_XMIT_POLICY_LAYER2},
207 { "layer3+4", BOND_XMIT_POLICY_LAYER34},
208 { "layer2+3", BOND_XMIT_POLICY_LAYER23},
212 const struct bond_parm_tbl arp_validate_tbl[] = {
213 { "none", BOND_ARP_VALIDATE_NONE},
214 { "active", BOND_ARP_VALIDATE_ACTIVE},
215 { "backup", BOND_ARP_VALIDATE_BACKUP},
216 { "all", BOND_ARP_VALIDATE_ALL},
220 const struct bond_parm_tbl fail_over_mac_tbl[] = {
221 { "none", BOND_FOM_NONE},
222 { "active", BOND_FOM_ACTIVE},
223 { "follow", BOND_FOM_FOLLOW},
227 const struct bond_parm_tbl pri_reselect_tbl[] = {
228 { "always", BOND_PRI_RESELECT_ALWAYS},
229 { "better", BOND_PRI_RESELECT_BETTER},
230 { "failure", BOND_PRI_RESELECT_FAILURE},
234 struct bond_parm_tbl ad_select_tbl[] = {
235 { "stable", BOND_AD_STABLE},
236 { "bandwidth", BOND_AD_BANDWIDTH},
237 { "count", BOND_AD_COUNT},
241 /*-------------------------- Forward declarations ---------------------------*/
243 static void bond_send_gratuitous_arp(struct bonding *bond);
244 static int bond_init(struct net_device *bond_dev);
245 static void bond_uninit(struct net_device *bond_dev);
247 /*---------------------------- General routines -----------------------------*/
249 static const char *bond_mode_name(int mode)
251 static const char *names[] = {
252 [BOND_MODE_ROUNDROBIN] = "load balancing (round-robin)",
253 [BOND_MODE_ACTIVEBACKUP] = "fault-tolerance (active-backup)",
254 [BOND_MODE_XOR] = "load balancing (xor)",
255 [BOND_MODE_BROADCAST] = "fault-tolerance (broadcast)",
256 [BOND_MODE_8023AD] = "IEEE 802.3ad Dynamic link aggregation",
257 [BOND_MODE_TLB] = "transmit load balancing",
258 [BOND_MODE_ALB] = "adaptive load balancing",
261 if (mode < 0 || mode > BOND_MODE_ALB)
267 /*---------------------------------- VLAN -----------------------------------*/
270 * bond_add_vlan - add a new vlan id on bond
271 * @bond: bond that got the notification
272 * @vlan_id: the vlan id to add
274 * Returns -ENOMEM if allocation failed.
276 static int bond_add_vlan(struct bonding *bond, unsigned short vlan_id)
278 struct vlan_entry *vlan;
280 pr_debug("bond: %s, vlan id %d\n",
281 (bond ? bond->dev->name : "None"), vlan_id);
283 vlan = kzalloc(sizeof(struct vlan_entry), GFP_KERNEL);
287 INIT_LIST_HEAD(&vlan->vlan_list);
288 vlan->vlan_id = vlan_id;
290 write_lock_bh(&bond->lock);
292 list_add_tail(&vlan->vlan_list, &bond->vlan_list);
294 write_unlock_bh(&bond->lock);
296 pr_debug("added VLAN ID %d on bond %s\n", vlan_id, bond->dev->name);
302 * bond_del_vlan - delete a vlan id from bond
303 * @bond: bond that got the notification
304 * @vlan_id: the vlan id to delete
306 * returns -ENODEV if @vlan_id was not found in @bond.
308 static int bond_del_vlan(struct bonding *bond, unsigned short vlan_id)
310 struct vlan_entry *vlan;
313 pr_debug("bond: %s, vlan id %d\n", bond->dev->name, vlan_id);
316 write_lock_bh(&bond->lock);
318 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
319 if (vlan->vlan_id == vlan_id) {
320 list_del(&vlan->vlan_list);
322 if (bond_is_lb(bond))
323 bond_alb_clear_vlan(bond, vlan_id);
325 pr_debug("removed VLAN ID %d from bond %s\n",
326 vlan_id, bond->dev->name);
330 if (list_empty(&bond->vlan_list) &&
331 (bond->slave_cnt == 0)) {
332 /* Last VLAN removed and no slaves, so
333 * restore block on adding VLANs. This will
334 * be removed once new slaves that are not
335 * VLAN challenged will be added.
337 bond->dev->features |= NETIF_F_VLAN_CHALLENGED;
345 pr_debug("couldn't find VLAN ID %d in bond %s\n",
346 vlan_id, bond->dev->name);
349 write_unlock_bh(&bond->lock);
350 unblock_netpoll_tx();
355 * bond_has_challenged_slaves
356 * @bond: the bond we're working on
358 * Searches the slave list. Returns 1 if a vlan challenged slave
359 * was found, 0 otherwise.
361 * Assumes bond->lock is held.
363 static int bond_has_challenged_slaves(struct bonding *bond)
368 bond_for_each_slave(bond, slave, i) {
369 if (slave->dev->features & NETIF_F_VLAN_CHALLENGED) {
370 pr_debug("found VLAN challenged slave - %s\n",
376 pr_debug("no VLAN challenged slaves found\n");
381 * bond_next_vlan - safely skip to the next item in the vlans list.
382 * @bond: the bond we're working on
383 * @curr: item we're advancing from
385 * Returns %NULL if list is empty, bond->next_vlan if @curr is %NULL,
386 * or @curr->next otherwise (even if it is @curr itself again).
388 * Caller must hold bond->lock
390 struct vlan_entry *bond_next_vlan(struct bonding *bond, struct vlan_entry *curr)
392 struct vlan_entry *next, *last;
394 if (list_empty(&bond->vlan_list))
398 next = list_entry(bond->vlan_list.next,
399 struct vlan_entry, vlan_list);
401 last = list_entry(bond->vlan_list.prev,
402 struct vlan_entry, vlan_list);
404 next = list_entry(bond->vlan_list.next,
405 struct vlan_entry, vlan_list);
407 next = list_entry(curr->vlan_list.next,
408 struct vlan_entry, vlan_list);
416 * bond_dev_queue_xmit - Prepare skb for xmit.
418 * @bond: bond device that got this skb for tx.
419 * @skb: hw accel VLAN tagged skb to transmit
420 * @slave_dev: slave that is supposed to xmit this skbuff
422 * When the bond gets an skb to transmit that is
423 * already hardware accelerated VLAN tagged, and it
424 * needs to relay this skb to a slave that is not
425 * hw accel capable, the skb needs to be "unaccelerated",
426 * i.e. strip the hwaccel tag and re-insert it as part
429 int bond_dev_queue_xmit(struct bonding *bond, struct sk_buff *skb,
430 struct net_device *slave_dev)
432 unsigned short uninitialized_var(vlan_id);
434 /* Test vlan_list not vlgrp to catch and handle 802.1p tags */
435 if (!list_empty(&bond->vlan_list) &&
436 !(slave_dev->features & NETIF_F_HW_VLAN_TX) &&
437 vlan_get_tag(skb, &vlan_id) == 0) {
438 skb->dev = slave_dev;
439 skb = vlan_put_tag(skb, vlan_id);
441 /* vlan_put_tag() frees the skb in case of error,
442 * so return success here so the calling functions
443 * won't attempt to free is again.
448 skb->dev = slave_dev;
452 #ifdef CONFIG_NET_POLL_CONTROLLER
453 if (unlikely(bond->dev->priv_flags & IFF_IN_NETPOLL)) {
454 struct netpoll *np = bond->dev->npinfo->netpoll;
455 slave_dev->npinfo = bond->dev->npinfo;
456 slave_dev->priv_flags |= IFF_IN_NETPOLL;
457 netpoll_send_skb_on_dev(np, skb, slave_dev);
458 slave_dev->priv_flags &= ~IFF_IN_NETPOLL;
467 * In the following 3 functions, bond_vlan_rx_register(), bond_vlan_rx_add_vid
468 * and bond_vlan_rx_kill_vid, We don't protect the slave list iteration with a
470 * a. This operation is performed in IOCTL context,
471 * b. The operation is protected by the RTNL semaphore in the 8021q code,
472 * c. Holding a lock with BH disabled while directly calling a base driver
473 * entry point is generally a BAD idea.
475 * The design of synchronization/protection for this operation in the 8021q
476 * module is good for one or more VLAN devices over a single physical device
477 * and cannot be extended for a teaming solution like bonding, so there is a
478 * potential race condition here where a net device from the vlan group might
479 * be referenced (either by a base driver or the 8021q code) while it is being
480 * removed from the system. However, it turns out we're not making matters
481 * worse, and if it works for regular VLAN usage it will work here too.
485 * bond_vlan_rx_register - Propagates registration to slaves
486 * @bond_dev: bonding net device that got called
487 * @grp: vlan group being registered
489 static void bond_vlan_rx_register(struct net_device *bond_dev,
490 struct vlan_group *grp)
492 struct bonding *bond = netdev_priv(bond_dev);
496 write_lock_bh(&bond->lock);
498 write_unlock_bh(&bond->lock);
500 bond_for_each_slave(bond, slave, i) {
501 struct net_device *slave_dev = slave->dev;
502 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
504 if ((slave_dev->features & NETIF_F_HW_VLAN_RX) &&
505 slave_ops->ndo_vlan_rx_register) {
506 slave_ops->ndo_vlan_rx_register(slave_dev, grp);
512 * bond_vlan_rx_add_vid - Propagates adding an id to slaves
513 * @bond_dev: bonding net device that got called
514 * @vid: vlan id being added
516 static void bond_vlan_rx_add_vid(struct net_device *bond_dev, uint16_t vid)
518 struct bonding *bond = netdev_priv(bond_dev);
522 bond_for_each_slave(bond, slave, i) {
523 struct net_device *slave_dev = slave->dev;
524 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
526 if ((slave_dev->features & NETIF_F_HW_VLAN_FILTER) &&
527 slave_ops->ndo_vlan_rx_add_vid) {
528 slave_ops->ndo_vlan_rx_add_vid(slave_dev, vid);
532 res = bond_add_vlan(bond, vid);
534 pr_err("%s: Error: Failed to add vlan id %d\n",
535 bond_dev->name, vid);
540 * bond_vlan_rx_kill_vid - Propagates deleting an id to slaves
541 * @bond_dev: bonding net device that got called
542 * @vid: vlan id being removed
544 static void bond_vlan_rx_kill_vid(struct net_device *bond_dev, uint16_t vid)
546 struct bonding *bond = netdev_priv(bond_dev);
548 struct net_device *vlan_dev;
551 bond_for_each_slave(bond, slave, i) {
552 struct net_device *slave_dev = slave->dev;
553 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
555 if ((slave_dev->features & NETIF_F_HW_VLAN_FILTER) &&
556 slave_ops->ndo_vlan_rx_kill_vid) {
557 /* Save and then restore vlan_dev in the grp array,
558 * since the slave's driver might clear it.
560 vlan_dev = vlan_group_get_device(bond->vlgrp, vid);
561 slave_ops->ndo_vlan_rx_kill_vid(slave_dev, vid);
562 vlan_group_set_device(bond->vlgrp, vid, vlan_dev);
566 res = bond_del_vlan(bond, vid);
568 pr_err("%s: Error: Failed to remove vlan id %d\n",
569 bond_dev->name, vid);
573 static void bond_add_vlans_on_slave(struct bonding *bond, struct net_device *slave_dev)
575 struct vlan_entry *vlan;
576 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
581 if ((slave_dev->features & NETIF_F_HW_VLAN_RX) &&
582 slave_ops->ndo_vlan_rx_register)
583 slave_ops->ndo_vlan_rx_register(slave_dev, bond->vlgrp);
585 if (!(slave_dev->features & NETIF_F_HW_VLAN_FILTER) ||
586 !(slave_ops->ndo_vlan_rx_add_vid))
589 list_for_each_entry(vlan, &bond->vlan_list, vlan_list)
590 slave_ops->ndo_vlan_rx_add_vid(slave_dev, vlan->vlan_id);
593 static void bond_del_vlans_from_slave(struct bonding *bond,
594 struct net_device *slave_dev)
596 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
597 struct vlan_entry *vlan;
598 struct net_device *vlan_dev;
603 if (!(slave_dev->features & NETIF_F_HW_VLAN_FILTER) ||
604 !(slave_ops->ndo_vlan_rx_kill_vid))
607 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
610 /* Save and then restore vlan_dev in the grp array,
611 * since the slave's driver might clear it.
613 vlan_dev = vlan_group_get_device(bond->vlgrp, vlan->vlan_id);
614 slave_ops->ndo_vlan_rx_kill_vid(slave_dev, vlan->vlan_id);
615 vlan_group_set_device(bond->vlgrp, vlan->vlan_id, vlan_dev);
619 if ((slave_dev->features & NETIF_F_HW_VLAN_RX) &&
620 slave_ops->ndo_vlan_rx_register)
621 slave_ops->ndo_vlan_rx_register(slave_dev, NULL);
624 /*------------------------------- Link status -------------------------------*/
627 * Set the carrier state for the master according to the state of its
628 * slaves. If any slaves are up, the master is up. In 802.3ad mode,
629 * do special 802.3ad magic.
631 * Returns zero if carrier state does not change, nonzero if it does.
633 static int bond_set_carrier(struct bonding *bond)
638 if (bond->slave_cnt == 0)
641 if (bond->params.mode == BOND_MODE_8023AD)
642 return bond_3ad_set_carrier(bond);
644 bond_for_each_slave(bond, slave, i) {
645 if (slave->link == BOND_LINK_UP) {
646 if (!netif_carrier_ok(bond->dev)) {
647 netif_carrier_on(bond->dev);
655 if (netif_carrier_ok(bond->dev)) {
656 netif_carrier_off(bond->dev);
663 * Get link speed and duplex from the slave's base driver
664 * using ethtool. If for some reason the call fails or the
665 * values are invalid, fake speed and duplex to 100/Full
668 static int bond_update_speed_duplex(struct slave *slave)
670 struct net_device *slave_dev = slave->dev;
671 struct ethtool_cmd etool;
674 /* Fake speed and duplex */
675 slave->speed = SPEED_100;
676 slave->duplex = DUPLEX_FULL;
678 if (!slave_dev->ethtool_ops || !slave_dev->ethtool_ops->get_settings)
681 res = slave_dev->ethtool_ops->get_settings(slave_dev, &etool);
685 switch (etool.speed) {
695 switch (etool.duplex) {
703 slave->speed = etool.speed;
704 slave->duplex = etool.duplex;
710 * if <dev> supports MII link status reporting, check its link status.
712 * We either do MII/ETHTOOL ioctls, or check netif_carrier_ok(),
713 * depending upon the setting of the use_carrier parameter.
715 * Return either BMSR_LSTATUS, meaning that the link is up (or we
716 * can't tell and just pretend it is), or 0, meaning that the link is
719 * If reporting is non-zero, instead of faking link up, return -1 if
720 * both ETHTOOL and MII ioctls fail (meaning the device does not
721 * support them). If use_carrier is set, return whatever it says.
722 * It'd be nice if there was a good way to tell if a driver supports
723 * netif_carrier, but there really isn't.
725 static int bond_check_dev_link(struct bonding *bond,
726 struct net_device *slave_dev, int reporting)
728 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
729 int (*ioctl)(struct net_device *, struct ifreq *, int);
731 struct mii_ioctl_data *mii;
733 if (!reporting && !netif_running(slave_dev))
736 if (bond->params.use_carrier)
737 return netif_carrier_ok(slave_dev) ? BMSR_LSTATUS : 0;
739 /* Try to get link status using Ethtool first. */
740 if (slave_dev->ethtool_ops) {
741 if (slave_dev->ethtool_ops->get_link) {
744 link = slave_dev->ethtool_ops->get_link(slave_dev);
746 return link ? BMSR_LSTATUS : 0;
750 /* Ethtool can't be used, fallback to MII ioctls. */
751 ioctl = slave_ops->ndo_do_ioctl;
753 /* TODO: set pointer to correct ioctl on a per team member */
754 /* bases to make this more efficient. that is, once */
755 /* we determine the correct ioctl, we will always */
756 /* call it and not the others for that team */
760 * We cannot assume that SIOCGMIIPHY will also read a
761 * register; not all network drivers (e.g., e100)
765 /* Yes, the mii is overlaid on the ifreq.ifr_ifru */
766 strncpy(ifr.ifr_name, slave_dev->name, IFNAMSIZ);
768 if (IOCTL(slave_dev, &ifr, SIOCGMIIPHY) == 0) {
769 mii->reg_num = MII_BMSR;
770 if (IOCTL(slave_dev, &ifr, SIOCGMIIREG) == 0)
771 return mii->val_out & BMSR_LSTATUS;
776 * If reporting, report that either there's no dev->do_ioctl,
777 * or both SIOCGMIIREG and get_link failed (meaning that we
778 * cannot report link status). If not reporting, pretend
781 return reporting ? -1 : BMSR_LSTATUS;
784 /*----------------------------- Multicast list ------------------------------*/
787 * Push the promiscuity flag down to appropriate slaves
789 static int bond_set_promiscuity(struct bonding *bond, int inc)
792 if (USES_PRIMARY(bond->params.mode)) {
793 /* write lock already acquired */
794 if (bond->curr_active_slave) {
795 err = dev_set_promiscuity(bond->curr_active_slave->dev,
801 bond_for_each_slave(bond, slave, i) {
802 err = dev_set_promiscuity(slave->dev, inc);
811 * Push the allmulti flag down to all slaves
813 static int bond_set_allmulti(struct bonding *bond, int inc)
816 if (USES_PRIMARY(bond->params.mode)) {
817 /* write lock already acquired */
818 if (bond->curr_active_slave) {
819 err = dev_set_allmulti(bond->curr_active_slave->dev,
825 bond_for_each_slave(bond, slave, i) {
826 err = dev_set_allmulti(slave->dev, inc);
835 * Add a Multicast address to slaves
838 static void bond_mc_add(struct bonding *bond, void *addr)
840 if (USES_PRIMARY(bond->params.mode)) {
841 /* write lock already acquired */
842 if (bond->curr_active_slave)
843 dev_mc_add(bond->curr_active_slave->dev, addr);
848 bond_for_each_slave(bond, slave, i)
849 dev_mc_add(slave->dev, addr);
854 * Remove a multicast address from slave
857 static void bond_mc_del(struct bonding *bond, void *addr)
859 if (USES_PRIMARY(bond->params.mode)) {
860 /* write lock already acquired */
861 if (bond->curr_active_slave)
862 dev_mc_del(bond->curr_active_slave->dev, addr);
866 bond_for_each_slave(bond, slave, i) {
867 dev_mc_del(slave->dev, addr);
873 static void __bond_resend_igmp_join_requests(struct net_device *dev)
875 struct in_device *in_dev;
876 struct ip_mc_list *im;
879 in_dev = __in_dev_get_rcu(dev);
881 read_lock(&in_dev->mc_list_lock);
882 for (im = in_dev->mc_list; im; im = im->next)
883 ip_mc_rejoin_group(im);
884 read_unlock(&in_dev->mc_list_lock);
891 * Retrieve the list of registered multicast addresses for the bonding
892 * device and retransmit an IGMP JOIN request to the current active
895 static void bond_resend_igmp_join_requests(struct bonding *bond)
897 struct net_device *vlan_dev;
898 struct vlan_entry *vlan;
900 read_lock(&bond->lock);
902 /* rejoin all groups on bond device */
903 __bond_resend_igmp_join_requests(bond->dev);
905 /* rejoin all groups on vlan devices */
907 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
908 vlan_dev = vlan_group_get_device(bond->vlgrp,
911 __bond_resend_igmp_join_requests(vlan_dev);
915 if (--bond->igmp_retrans > 0)
916 queue_delayed_work(bond->wq, &bond->mcast_work, HZ/5);
918 read_unlock(&bond->lock);
921 static void bond_resend_igmp_join_requests_delayed(struct work_struct *work)
923 struct bonding *bond = container_of(work, struct bonding,
925 bond_resend_igmp_join_requests(bond);
929 * flush all members of flush->mc_list from device dev->mc_list
931 static void bond_mc_list_flush(struct net_device *bond_dev,
932 struct net_device *slave_dev)
934 struct bonding *bond = netdev_priv(bond_dev);
935 struct netdev_hw_addr *ha;
937 netdev_for_each_mc_addr(ha, bond_dev)
938 dev_mc_del(slave_dev, ha->addr);
940 if (bond->params.mode == BOND_MODE_8023AD) {
941 /* del lacpdu mc addr from mc list */
942 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
944 dev_mc_del(slave_dev, lacpdu_multicast);
948 /*--------------------------- Active slave change ---------------------------*/
951 * Update the mc list and multicast-related flags for the new and
952 * old active slaves (if any) according to the multicast mode, and
953 * promiscuous flags unconditionally.
955 static void bond_mc_swap(struct bonding *bond, struct slave *new_active,
956 struct slave *old_active)
958 struct netdev_hw_addr *ha;
960 if (!USES_PRIMARY(bond->params.mode))
961 /* nothing to do - mc list is already up-to-date on
967 if (bond->dev->flags & IFF_PROMISC)
968 dev_set_promiscuity(old_active->dev, -1);
970 if (bond->dev->flags & IFF_ALLMULTI)
971 dev_set_allmulti(old_active->dev, -1);
973 netdev_for_each_mc_addr(ha, bond->dev)
974 dev_mc_del(old_active->dev, ha->addr);
978 /* FIXME: Signal errors upstream. */
979 if (bond->dev->flags & IFF_PROMISC)
980 dev_set_promiscuity(new_active->dev, 1);
982 if (bond->dev->flags & IFF_ALLMULTI)
983 dev_set_allmulti(new_active->dev, 1);
985 netdev_for_each_mc_addr(ha, bond->dev)
986 dev_mc_add(new_active->dev, ha->addr);
991 * bond_do_fail_over_mac
993 * Perform special MAC address swapping for fail_over_mac settings
995 * Called with RTNL, bond->lock for read, curr_slave_lock for write_bh.
997 static void bond_do_fail_over_mac(struct bonding *bond,
998 struct slave *new_active,
999 struct slave *old_active)
1000 __releases(&bond->curr_slave_lock)
1001 __releases(&bond->lock)
1002 __acquires(&bond->lock)
1003 __acquires(&bond->curr_slave_lock)
1005 u8 tmp_mac[ETH_ALEN];
1006 struct sockaddr saddr;
1009 switch (bond->params.fail_over_mac) {
1010 case BOND_FOM_ACTIVE:
1012 memcpy(bond->dev->dev_addr, new_active->dev->dev_addr,
1013 new_active->dev->addr_len);
1015 case BOND_FOM_FOLLOW:
1017 * if new_active && old_active, swap them
1018 * if just old_active, do nothing (going to no active slave)
1019 * if just new_active, set new_active to bond's MAC
1024 write_unlock_bh(&bond->curr_slave_lock);
1025 read_unlock(&bond->lock);
1028 memcpy(tmp_mac, new_active->dev->dev_addr, ETH_ALEN);
1029 memcpy(saddr.sa_data, old_active->dev->dev_addr,
1031 saddr.sa_family = new_active->dev->type;
1033 memcpy(saddr.sa_data, bond->dev->dev_addr, ETH_ALEN);
1034 saddr.sa_family = bond->dev->type;
1037 rv = dev_set_mac_address(new_active->dev, &saddr);
1039 pr_err("%s: Error %d setting MAC of slave %s\n",
1040 bond->dev->name, -rv, new_active->dev->name);
1047 memcpy(saddr.sa_data, tmp_mac, ETH_ALEN);
1048 saddr.sa_family = old_active->dev->type;
1050 rv = dev_set_mac_address(old_active->dev, &saddr);
1052 pr_err("%s: Error %d setting MAC of slave %s\n",
1053 bond->dev->name, -rv, new_active->dev->name);
1055 read_lock(&bond->lock);
1056 write_lock_bh(&bond->curr_slave_lock);
1059 pr_err("%s: bond_do_fail_over_mac impossible: bad policy %d\n",
1060 bond->dev->name, bond->params.fail_over_mac);
1066 static bool bond_should_change_active(struct bonding *bond)
1068 struct slave *prim = bond->primary_slave;
1069 struct slave *curr = bond->curr_active_slave;
1071 if (!prim || !curr || curr->link != BOND_LINK_UP)
1073 if (bond->force_primary) {
1074 bond->force_primary = false;
1077 if (bond->params.primary_reselect == BOND_PRI_RESELECT_BETTER &&
1078 (prim->speed < curr->speed ||
1079 (prim->speed == curr->speed && prim->duplex <= curr->duplex)))
1081 if (bond->params.primary_reselect == BOND_PRI_RESELECT_FAILURE)
1087 * find_best_interface - select the best available slave to be the active one
1088 * @bond: our bonding struct
1090 * Warning: Caller must hold curr_slave_lock for writing.
1092 static struct slave *bond_find_best_slave(struct bonding *bond)
1094 struct slave *new_active, *old_active;
1095 struct slave *bestslave = NULL;
1096 int mintime = bond->params.updelay;
1099 new_active = bond->curr_active_slave;
1101 if (!new_active) { /* there were no active slaves left */
1102 if (bond->slave_cnt > 0) /* found one slave */
1103 new_active = bond->first_slave;
1105 return NULL; /* still no slave, return NULL */
1108 if ((bond->primary_slave) &&
1109 bond->primary_slave->link == BOND_LINK_UP &&
1110 bond_should_change_active(bond)) {
1111 new_active = bond->primary_slave;
1114 /* remember where to stop iterating over the slaves */
1115 old_active = new_active;
1117 bond_for_each_slave_from(bond, new_active, i, old_active) {
1118 if (new_active->link == BOND_LINK_UP) {
1120 } else if (new_active->link == BOND_LINK_BACK &&
1121 IS_UP(new_active->dev)) {
1122 /* link up, but waiting for stabilization */
1123 if (new_active->delay < mintime) {
1124 mintime = new_active->delay;
1125 bestslave = new_active;
1134 * change_active_interface - change the active slave into the specified one
1135 * @bond: our bonding struct
1136 * @new: the new slave to make the active one
1138 * Set the new slave to the bond's settings and unset them on the old
1139 * curr_active_slave.
1140 * Setting include flags, mc-list, promiscuity, allmulti, etc.
1142 * If @new's link state is %BOND_LINK_BACK we'll set it to %BOND_LINK_UP,
1143 * because it is apparently the best available slave we have, even though its
1144 * updelay hasn't timed out yet.
1146 * If new_active is not NULL, caller must hold bond->lock for read and
1147 * curr_slave_lock for write_bh.
1149 void bond_change_active_slave(struct bonding *bond, struct slave *new_active)
1151 struct slave *old_active = bond->curr_active_slave;
1153 if (old_active == new_active)
1157 new_active->jiffies = jiffies;
1159 if (new_active->link == BOND_LINK_BACK) {
1160 if (USES_PRIMARY(bond->params.mode)) {
1161 pr_info("%s: making interface %s the new active one %d ms earlier.\n",
1162 bond->dev->name, new_active->dev->name,
1163 (bond->params.updelay - new_active->delay) * bond->params.miimon);
1166 new_active->delay = 0;
1167 new_active->link = BOND_LINK_UP;
1169 if (bond->params.mode == BOND_MODE_8023AD)
1170 bond_3ad_handle_link_change(new_active, BOND_LINK_UP);
1172 if (bond_is_lb(bond))
1173 bond_alb_handle_link_change(bond, new_active, BOND_LINK_UP);
1175 if (USES_PRIMARY(bond->params.mode)) {
1176 pr_info("%s: making interface %s the new active one.\n",
1177 bond->dev->name, new_active->dev->name);
1182 if (USES_PRIMARY(bond->params.mode))
1183 bond_mc_swap(bond, new_active, old_active);
1185 if (bond_is_lb(bond)) {
1186 bond_alb_handle_active_change(bond, new_active);
1188 bond_set_slave_inactive_flags(old_active);
1190 bond_set_slave_active_flags(new_active);
1192 bond->curr_active_slave = new_active;
1195 if (bond->params.mode == BOND_MODE_ACTIVEBACKUP) {
1197 bond_set_slave_inactive_flags(old_active);
1200 bond_set_slave_active_flags(new_active);
1202 if (bond->params.fail_over_mac)
1203 bond_do_fail_over_mac(bond, new_active,
1206 bond->send_grat_arp = bond->params.num_grat_arp;
1207 bond_send_gratuitous_arp(bond);
1209 bond->send_unsol_na = bond->params.num_unsol_na;
1210 bond_send_unsolicited_na(bond);
1212 write_unlock_bh(&bond->curr_slave_lock);
1213 read_unlock(&bond->lock);
1215 netdev_bonding_change(bond->dev, NETDEV_BONDING_FAILOVER);
1217 read_lock(&bond->lock);
1218 write_lock_bh(&bond->curr_slave_lock);
1222 /* resend IGMP joins since active slave has changed or
1223 * all were sent on curr_active_slave */
1224 if ((USES_PRIMARY(bond->params.mode) && new_active) ||
1225 bond->params.mode == BOND_MODE_ROUNDROBIN) {
1226 bond->igmp_retrans = bond->params.resend_igmp;
1227 queue_delayed_work(bond->wq, &bond->mcast_work, 0);
1232 * bond_select_active_slave - select a new active slave, if needed
1233 * @bond: our bonding struct
1235 * This functions should be called when one of the following occurs:
1236 * - The old curr_active_slave has been released or lost its link.
1237 * - The primary_slave has got its link back.
1238 * - A slave has got its link back and there's no old curr_active_slave.
1240 * Caller must hold bond->lock for read and curr_slave_lock for write_bh.
1242 void bond_select_active_slave(struct bonding *bond)
1244 struct slave *best_slave;
1247 best_slave = bond_find_best_slave(bond);
1248 if (best_slave != bond->curr_active_slave) {
1249 bond_change_active_slave(bond, best_slave);
1250 rv = bond_set_carrier(bond);
1254 if (netif_carrier_ok(bond->dev)) {
1255 pr_info("%s: first active interface up!\n",
1258 pr_info("%s: now running without any active interface !\n",
1264 /*--------------------------- slave list handling ---------------------------*/
1267 * This function attaches the slave to the end of list.
1269 * bond->lock held for writing by caller.
1271 static void bond_attach_slave(struct bonding *bond, struct slave *new_slave)
1273 if (bond->first_slave == NULL) { /* attaching the first slave */
1274 new_slave->next = new_slave;
1275 new_slave->prev = new_slave;
1276 bond->first_slave = new_slave;
1278 new_slave->next = bond->first_slave;
1279 new_slave->prev = bond->first_slave->prev;
1280 new_slave->next->prev = new_slave;
1281 new_slave->prev->next = new_slave;
1288 * This function detaches the slave from the list.
1289 * WARNING: no check is made to verify if the slave effectively
1290 * belongs to <bond>.
1291 * Nothing is freed on return, structures are just unchained.
1292 * If any slave pointer in bond was pointing to <slave>,
1293 * it should be changed by the calling function.
1295 * bond->lock held for writing by caller.
1297 static void bond_detach_slave(struct bonding *bond, struct slave *slave)
1300 slave->next->prev = slave->prev;
1303 slave->prev->next = slave->next;
1305 if (bond->first_slave == slave) { /* slave is the first slave */
1306 if (bond->slave_cnt > 1) { /* there are more slave */
1307 bond->first_slave = slave->next;
1309 bond->first_slave = NULL; /* slave was the last one */
1318 #ifdef CONFIG_NET_POLL_CONTROLLER
1320 * You must hold read lock on bond->lock before calling this.
1322 static bool slaves_support_netpoll(struct net_device *bond_dev)
1324 struct bonding *bond = netdev_priv(bond_dev);
1325 struct slave *slave;
1329 bond_for_each_slave(bond, slave, i) {
1330 if ((slave->dev->priv_flags & IFF_DISABLE_NETPOLL) ||
1331 !slave->dev->netdev_ops->ndo_poll_controller)
1334 return i != 0 && ret;
1337 static void bond_poll_controller(struct net_device *bond_dev)
1339 struct bonding *bond = netdev_priv(bond_dev);
1340 struct slave *slave;
1343 bond_for_each_slave(bond, slave, i) {
1344 if (slave->dev && IS_UP(slave->dev))
1345 netpoll_poll_dev(slave->dev);
1349 static void bond_netpoll_cleanup(struct net_device *bond_dev)
1351 struct bonding *bond = netdev_priv(bond_dev);
1352 struct slave *slave;
1353 const struct net_device_ops *ops;
1356 read_lock(&bond->lock);
1357 bond_dev->npinfo = NULL;
1358 bond_for_each_slave(bond, slave, i) {
1360 ops = slave->dev->netdev_ops;
1361 if (ops->ndo_netpoll_cleanup)
1362 ops->ndo_netpoll_cleanup(slave->dev);
1364 slave->dev->npinfo = NULL;
1367 read_unlock(&bond->lock);
1372 static void bond_netpoll_cleanup(struct net_device *bond_dev)
1378 /*---------------------------------- IOCTL ----------------------------------*/
1380 static int bond_sethwaddr(struct net_device *bond_dev,
1381 struct net_device *slave_dev)
1383 pr_debug("bond_dev=%p\n", bond_dev);
1384 pr_debug("slave_dev=%p\n", slave_dev);
1385 pr_debug("slave_dev->addr_len=%d\n", slave_dev->addr_len);
1386 memcpy(bond_dev->dev_addr, slave_dev->dev_addr, slave_dev->addr_len);
1390 #define BOND_VLAN_FEATURES \
1391 (NETIF_F_VLAN_CHALLENGED | NETIF_F_HW_VLAN_RX | NETIF_F_HW_VLAN_TX | \
1392 NETIF_F_HW_VLAN_FILTER)
1395 * Compute the common dev->feature set available to all slaves. Some
1396 * feature bits are managed elsewhere, so preserve those feature bits
1397 * on the master device.
1399 static int bond_compute_features(struct bonding *bond)
1401 struct slave *slave;
1402 struct net_device *bond_dev = bond->dev;
1403 unsigned long features = bond_dev->features;
1404 unsigned long vlan_features = 0;
1405 unsigned short max_hard_header_len = max((u16)ETH_HLEN,
1406 bond_dev->hard_header_len);
1409 features &= ~(NETIF_F_ALL_CSUM | BOND_VLAN_FEATURES);
1410 features |= NETIF_F_GSO_MASK | NETIF_F_NO_CSUM;
1412 if (!bond->first_slave)
1415 features &= ~NETIF_F_ONE_FOR_ALL;
1417 vlan_features = bond->first_slave->dev->vlan_features;
1418 bond_for_each_slave(bond, slave, i) {
1419 features = netdev_increment_features(features,
1420 slave->dev->features,
1421 NETIF_F_ONE_FOR_ALL);
1422 vlan_features = netdev_increment_features(vlan_features,
1423 slave->dev->vlan_features,
1424 NETIF_F_ONE_FOR_ALL);
1425 if (slave->dev->hard_header_len > max_hard_header_len)
1426 max_hard_header_len = slave->dev->hard_header_len;
1430 features |= (bond_dev->features & BOND_VLAN_FEATURES);
1431 bond_dev->features = netdev_fix_features(features, NULL);
1432 bond_dev->vlan_features = netdev_fix_features(vlan_features, NULL);
1433 bond_dev->hard_header_len = max_hard_header_len;
1438 static void bond_setup_by_slave(struct net_device *bond_dev,
1439 struct net_device *slave_dev)
1441 struct bonding *bond = netdev_priv(bond_dev);
1443 bond_dev->header_ops = slave_dev->header_ops;
1445 bond_dev->type = slave_dev->type;
1446 bond_dev->hard_header_len = slave_dev->hard_header_len;
1447 bond_dev->addr_len = slave_dev->addr_len;
1449 memcpy(bond_dev->broadcast, slave_dev->broadcast,
1450 slave_dev->addr_len);
1451 bond->setup_by_slave = 1;
1454 /* enslave device <slave> to bond device <master> */
1455 int bond_enslave(struct net_device *bond_dev, struct net_device *slave_dev)
1457 struct bonding *bond = netdev_priv(bond_dev);
1458 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
1459 struct slave *new_slave = NULL;
1460 struct netdev_hw_addr *ha;
1461 struct sockaddr addr;
1463 int old_features = bond_dev->features;
1466 if (!bond->params.use_carrier && slave_dev->ethtool_ops == NULL &&
1467 slave_ops->ndo_do_ioctl == NULL) {
1468 pr_warning("%s: Warning: no link monitoring support for %s\n",
1469 bond_dev->name, slave_dev->name);
1472 /* bond must be initialized by bond_open() before enslaving */
1473 if (!(bond_dev->flags & IFF_UP)) {
1474 pr_warning("%s: master_dev is not up in bond_enslave\n",
1478 /* already enslaved */
1479 if (slave_dev->flags & IFF_SLAVE) {
1480 pr_debug("Error, Device was already enslaved\n");
1484 /* vlan challenged mutual exclusion */
1485 /* no need to lock since we're protected by rtnl_lock */
1486 if (slave_dev->features & NETIF_F_VLAN_CHALLENGED) {
1487 pr_debug("%s: NETIF_F_VLAN_CHALLENGED\n", slave_dev->name);
1489 pr_err("%s: Error: cannot enslave VLAN challenged slave %s on VLAN enabled bond %s\n",
1490 bond_dev->name, slave_dev->name, bond_dev->name);
1493 pr_warning("%s: Warning: enslaved VLAN challenged slave %s. Adding VLANs will be blocked as long as %s is part of bond %s\n",
1494 bond_dev->name, slave_dev->name,
1495 slave_dev->name, bond_dev->name);
1496 bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
1499 pr_debug("%s: ! NETIF_F_VLAN_CHALLENGED\n", slave_dev->name);
1500 if (bond->slave_cnt == 0) {
1501 /* First slave, and it is not VLAN challenged,
1502 * so remove the block of adding VLANs over the bond.
1504 bond_dev->features &= ~NETIF_F_VLAN_CHALLENGED;
1509 * Old ifenslave binaries are no longer supported. These can
1510 * be identified with moderate accuracy by the state of the slave:
1511 * the current ifenslave will set the interface down prior to
1512 * enslaving it; the old ifenslave will not.
1514 if ((slave_dev->flags & IFF_UP)) {
1515 pr_err("%s is up. This may be due to an out of date ifenslave.\n",
1518 goto err_undo_flags;
1521 /* set bonding device ether type by slave - bonding netdevices are
1522 * created with ether_setup, so when the slave type is not ARPHRD_ETHER
1523 * there is a need to override some of the type dependent attribs/funcs.
1525 * bond ether type mutual exclusion - don't allow slaves of dissimilar
1526 * ether type (eg ARPHRD_ETHER and ARPHRD_INFINIBAND) share the same bond
1528 if (bond->slave_cnt == 0) {
1529 if (bond_dev->type != slave_dev->type) {
1530 pr_debug("%s: change device type from %d to %d\n",
1532 bond_dev->type, slave_dev->type);
1534 res = netdev_bonding_change(bond_dev,
1535 NETDEV_PRE_TYPE_CHANGE);
1536 res = notifier_to_errno(res);
1538 pr_err("%s: refused to change device type\n",
1541 goto err_undo_flags;
1544 /* Flush unicast and multicast addresses */
1545 dev_uc_flush(bond_dev);
1546 dev_mc_flush(bond_dev);
1548 if (slave_dev->type != ARPHRD_ETHER)
1549 bond_setup_by_slave(bond_dev, slave_dev);
1551 ether_setup(bond_dev);
1553 netdev_bonding_change(bond_dev,
1554 NETDEV_POST_TYPE_CHANGE);
1556 } else if (bond_dev->type != slave_dev->type) {
1557 pr_err("%s ether type (%d) is different from other slaves (%d), can not enslave it.\n",
1559 slave_dev->type, bond_dev->type);
1561 goto err_undo_flags;
1564 if (slave_ops->ndo_set_mac_address == NULL) {
1565 if (bond->slave_cnt == 0) {
1566 pr_warning("%s: Warning: The first slave device specified does not support setting the MAC address. Setting fail_over_mac to active.",
1568 bond->params.fail_over_mac = BOND_FOM_ACTIVE;
1569 } else if (bond->params.fail_over_mac != BOND_FOM_ACTIVE) {
1570 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",
1573 goto err_undo_flags;
1577 /* If this is the first slave, then we need to set the master's hardware
1578 * address to be the same as the slave's. */
1579 if (is_zero_ether_addr(bond->dev->dev_addr))
1580 memcpy(bond->dev->dev_addr, slave_dev->dev_addr,
1581 slave_dev->addr_len);
1584 new_slave = kzalloc(sizeof(struct slave), GFP_KERNEL);
1587 goto err_undo_flags;
1591 * Set the new_slave's queue_id to be zero. Queue ID mapping
1592 * is set via sysfs or module option if desired.
1594 new_slave->queue_id = 0;
1596 /* Save slave's original mtu and then set it to match the bond */
1597 new_slave->original_mtu = slave_dev->mtu;
1598 res = dev_set_mtu(slave_dev, bond->dev->mtu);
1600 pr_debug("Error %d calling dev_set_mtu\n", res);
1605 * Save slave's original ("permanent") mac address for modes
1606 * that need it, and for restoring it upon release, and then
1607 * set it to the master's address
1609 memcpy(new_slave->perm_hwaddr, slave_dev->dev_addr, ETH_ALEN);
1611 if (!bond->params.fail_over_mac) {
1613 * Set slave to master's mac address. The application already
1614 * set the master's mac address to that of the first slave
1616 memcpy(addr.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
1617 addr.sa_family = slave_dev->type;
1618 res = dev_set_mac_address(slave_dev, &addr);
1620 pr_debug("Error %d calling set_mac_address\n", res);
1621 goto err_restore_mtu;
1625 res = netdev_set_master(slave_dev, bond_dev);
1627 pr_debug("Error %d calling netdev_set_master\n", res);
1628 goto err_restore_mac;
1630 /* open the slave since the application closed it */
1631 res = dev_open(slave_dev);
1633 pr_debug("Opening slave %s failed\n", slave_dev->name);
1634 goto err_unset_master;
1637 new_slave->dev = slave_dev;
1638 slave_dev->priv_flags |= IFF_BONDING;
1640 if (bond_is_lb(bond)) {
1641 /* bond_alb_init_slave() must be called before all other stages since
1642 * it might fail and we do not want to have to undo everything
1644 res = bond_alb_init_slave(bond, new_slave);
1649 /* If the mode USES_PRIMARY, then the new slave gets the
1650 * master's promisc (and mc) settings only if it becomes the
1651 * curr_active_slave, and that is taken care of later when calling
1652 * bond_change_active()
1654 if (!USES_PRIMARY(bond->params.mode)) {
1655 /* set promiscuity level to new slave */
1656 if (bond_dev->flags & IFF_PROMISC) {
1657 res = dev_set_promiscuity(slave_dev, 1);
1662 /* set allmulti level to new slave */
1663 if (bond_dev->flags & IFF_ALLMULTI) {
1664 res = dev_set_allmulti(slave_dev, 1);
1669 netif_addr_lock_bh(bond_dev);
1670 /* upload master's mc_list to new slave */
1671 netdev_for_each_mc_addr(ha, bond_dev)
1672 dev_mc_add(slave_dev, ha->addr);
1673 netif_addr_unlock_bh(bond_dev);
1676 if (bond->params.mode == BOND_MODE_8023AD) {
1677 /* add lacpdu mc addr to mc list */
1678 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
1680 dev_mc_add(slave_dev, lacpdu_multicast);
1683 bond_add_vlans_on_slave(bond, slave_dev);
1685 write_lock_bh(&bond->lock);
1687 bond_attach_slave(bond, new_slave);
1689 new_slave->delay = 0;
1690 new_slave->link_failure_count = 0;
1692 bond_compute_features(bond);
1694 write_unlock_bh(&bond->lock);
1696 read_lock(&bond->lock);
1698 new_slave->last_arp_rx = jiffies;
1700 if (bond->params.miimon && !bond->params.use_carrier) {
1701 link_reporting = bond_check_dev_link(bond, slave_dev, 1);
1703 if ((link_reporting == -1) && !bond->params.arp_interval) {
1705 * miimon is set but a bonded network driver
1706 * does not support ETHTOOL/MII and
1707 * arp_interval is not set. Note: if
1708 * use_carrier is enabled, we will never go
1709 * here (because netif_carrier is always
1710 * supported); thus, we don't need to change
1711 * the messages for netif_carrier.
1713 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",
1714 bond_dev->name, slave_dev->name);
1715 } else if (link_reporting == -1) {
1716 /* unable get link status using mii/ethtool */
1717 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",
1718 bond_dev->name, slave_dev->name);
1722 /* check for initial state */
1723 if (!bond->params.miimon ||
1724 (bond_check_dev_link(bond, slave_dev, 0) == BMSR_LSTATUS)) {
1725 if (bond->params.updelay) {
1726 pr_debug("Initial state of slave_dev is BOND_LINK_BACK\n");
1727 new_slave->link = BOND_LINK_BACK;
1728 new_slave->delay = bond->params.updelay;
1730 pr_debug("Initial state of slave_dev is BOND_LINK_UP\n");
1731 new_slave->link = BOND_LINK_UP;
1733 new_slave->jiffies = jiffies;
1735 pr_debug("Initial state of slave_dev is BOND_LINK_DOWN\n");
1736 new_slave->link = BOND_LINK_DOWN;
1739 if (bond_update_speed_duplex(new_slave) &&
1740 (new_slave->link != BOND_LINK_DOWN)) {
1741 pr_warning("%s: Warning: failed to get speed and duplex from %s, assumed to be 100Mb/sec and Full.\n",
1742 bond_dev->name, new_slave->dev->name);
1744 if (bond->params.mode == BOND_MODE_8023AD) {
1745 pr_warning("%s: Warning: Operation of 802.3ad mode requires ETHTOOL support in base driver for proper aggregator selection.\n",
1750 if (USES_PRIMARY(bond->params.mode) && bond->params.primary[0]) {
1751 /* if there is a primary slave, remember it */
1752 if (strcmp(bond->params.primary, new_slave->dev->name) == 0) {
1753 bond->primary_slave = new_slave;
1754 bond->force_primary = true;
1758 write_lock_bh(&bond->curr_slave_lock);
1760 switch (bond->params.mode) {
1761 case BOND_MODE_ACTIVEBACKUP:
1762 bond_set_slave_inactive_flags(new_slave);
1763 bond_select_active_slave(bond);
1765 case BOND_MODE_8023AD:
1766 /* in 802.3ad mode, the internal mechanism
1767 * will activate the slaves in the selected
1770 bond_set_slave_inactive_flags(new_slave);
1771 /* if this is the first slave */
1772 if (bond->slave_cnt == 1) {
1773 SLAVE_AD_INFO(new_slave).id = 1;
1774 /* Initialize AD with the number of times that the AD timer is called in 1 second
1775 * can be called only after the mac address of the bond is set
1777 bond_3ad_initialize(bond, 1000/AD_TIMER_INTERVAL,
1778 bond->params.lacp_fast);
1780 SLAVE_AD_INFO(new_slave).id =
1781 SLAVE_AD_INFO(new_slave->prev).id + 1;
1784 bond_3ad_bind_slave(new_slave);
1788 new_slave->state = BOND_STATE_ACTIVE;
1789 bond_set_slave_inactive_flags(new_slave);
1790 bond_select_active_slave(bond);
1793 pr_debug("This slave is always active in trunk mode\n");
1795 /* always active in trunk mode */
1796 new_slave->state = BOND_STATE_ACTIVE;
1798 /* In trunking mode there is little meaning to curr_active_slave
1799 * anyway (it holds no special properties of the bond device),
1800 * so we can change it without calling change_active_interface()
1802 if (!bond->curr_active_slave)
1803 bond->curr_active_slave = new_slave;
1806 } /* switch(bond_mode) */
1808 write_unlock_bh(&bond->curr_slave_lock);
1810 bond_set_carrier(bond);
1812 #ifdef CONFIG_NET_POLL_CONTROLLER
1813 if (slaves_support_netpoll(bond_dev)) {
1814 bond_dev->priv_flags &= ~IFF_DISABLE_NETPOLL;
1815 if (bond_dev->npinfo)
1816 slave_dev->npinfo = bond_dev->npinfo;
1817 } else if (!(bond_dev->priv_flags & IFF_DISABLE_NETPOLL)) {
1818 bond_dev->priv_flags |= IFF_DISABLE_NETPOLL;
1819 pr_info("New slave device %s does not support netpoll\n",
1821 pr_info("Disabling netpoll support for %s\n", bond_dev->name);
1824 read_unlock(&bond->lock);
1826 res = bond_create_slave_symlinks(bond_dev, slave_dev);
1830 pr_info("%s: enslaving %s as a%s interface with a%s link.\n",
1831 bond_dev->name, slave_dev->name,
1832 new_slave->state == BOND_STATE_ACTIVE ? "n active" : " backup",
1833 new_slave->link != BOND_LINK_DOWN ? "n up" : " down");
1835 /* enslave is successful */
1838 /* Undo stages on error */
1840 dev_close(slave_dev);
1843 netdev_set_master(slave_dev, NULL);
1846 if (!bond->params.fail_over_mac) {
1847 /* XXX TODO - fom follow mode needs to change master's
1848 * MAC if this slave's MAC is in use by the bond, or at
1849 * least print a warning.
1851 memcpy(addr.sa_data, new_slave->perm_hwaddr, ETH_ALEN);
1852 addr.sa_family = slave_dev->type;
1853 dev_set_mac_address(slave_dev, &addr);
1857 dev_set_mtu(slave_dev, new_slave->original_mtu);
1863 bond_dev->features = old_features;
1869 * Try to release the slave device <slave> from the bond device <master>
1870 * It is legal to access curr_active_slave without a lock because all the function
1873 * The rules for slave state should be:
1874 * for Active/Backup:
1875 * Active stays on all backups go down
1876 * for Bonded connections:
1877 * The first up interface should be left on and all others downed.
1879 int bond_release(struct net_device *bond_dev, struct net_device *slave_dev)
1881 struct bonding *bond = netdev_priv(bond_dev);
1882 struct slave *slave, *oldcurrent;
1883 struct sockaddr addr;
1885 /* slave is not a slave or master is not master of this slave */
1886 if (!(slave_dev->flags & IFF_SLAVE) ||
1887 (slave_dev->master != bond_dev)) {
1888 pr_err("%s: Error: cannot release %s.\n",
1889 bond_dev->name, slave_dev->name);
1894 netdev_bonding_change(bond_dev, NETDEV_BONDING_DESLAVE);
1895 write_lock_bh(&bond->lock);
1897 slave = bond_get_slave_by_dev(bond, slave_dev);
1899 /* not a slave of this bond */
1900 pr_info("%s: %s not enslaved\n",
1901 bond_dev->name, slave_dev->name);
1902 write_unlock_bh(&bond->lock);
1903 unblock_netpoll_tx();
1907 if (!bond->params.fail_over_mac) {
1908 if (!compare_ether_addr(bond_dev->dev_addr, slave->perm_hwaddr) &&
1909 bond->slave_cnt > 1)
1910 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",
1911 bond_dev->name, slave_dev->name,
1913 bond_dev->name, slave_dev->name);
1916 /* Inform AD package of unbinding of slave. */
1917 if (bond->params.mode == BOND_MODE_8023AD) {
1918 /* must be called before the slave is
1919 * detached from the list
1921 bond_3ad_unbind_slave(slave);
1924 pr_info("%s: releasing %s interface %s\n",
1926 (slave->state == BOND_STATE_ACTIVE) ? "active" : "backup",
1929 oldcurrent = bond->curr_active_slave;
1931 bond->current_arp_slave = NULL;
1933 /* release the slave from its bond */
1934 bond_detach_slave(bond, slave);
1936 bond_compute_features(bond);
1938 if (bond->primary_slave == slave)
1939 bond->primary_slave = NULL;
1941 if (oldcurrent == slave)
1942 bond_change_active_slave(bond, NULL);
1944 if (bond_is_lb(bond)) {
1945 /* Must be called only after the slave has been
1946 * detached from the list and the curr_active_slave
1947 * has been cleared (if our_slave == old_current),
1948 * but before a new active slave is selected.
1950 write_unlock_bh(&bond->lock);
1951 bond_alb_deinit_slave(bond, slave);
1952 write_lock_bh(&bond->lock);
1955 if (oldcurrent == slave) {
1957 * Note that we hold RTNL over this sequence, so there
1958 * is no concern that another slave add/remove event
1961 write_unlock_bh(&bond->lock);
1962 read_lock(&bond->lock);
1963 write_lock_bh(&bond->curr_slave_lock);
1965 bond_select_active_slave(bond);
1967 write_unlock_bh(&bond->curr_slave_lock);
1968 read_unlock(&bond->lock);
1969 write_lock_bh(&bond->lock);
1972 if (bond->slave_cnt == 0) {
1973 bond_set_carrier(bond);
1975 /* if the last slave was removed, zero the mac address
1976 * of the master so it will be set by the application
1977 * to the mac address of the first slave
1979 memset(bond_dev->dev_addr, 0, bond_dev->addr_len);
1982 bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
1984 pr_warning("%s: Warning: clearing HW address of %s while it still has VLANs.\n",
1985 bond_dev->name, bond_dev->name);
1986 pr_warning("%s: When re-adding slaves, make sure the bond's HW address matches its VLANs'.\n",
1989 } else if ((bond_dev->features & NETIF_F_VLAN_CHALLENGED) &&
1990 !bond_has_challenged_slaves(bond)) {
1991 pr_info("%s: last VLAN challenged slave %s left bond %s. VLAN blocking is removed\n",
1992 bond_dev->name, slave_dev->name, bond_dev->name);
1993 bond_dev->features &= ~NETIF_F_VLAN_CHALLENGED;
1996 write_unlock_bh(&bond->lock);
1997 unblock_netpoll_tx();
1999 /* must do this from outside any spinlocks */
2000 bond_destroy_slave_symlinks(bond_dev, slave_dev);
2002 bond_del_vlans_from_slave(bond, slave_dev);
2004 /* If the mode USES_PRIMARY, then we should only remove its
2005 * promisc and mc settings if it was the curr_active_slave, but that was
2006 * already taken care of above when we detached the slave
2008 if (!USES_PRIMARY(bond->params.mode)) {
2009 /* unset promiscuity level from slave */
2010 if (bond_dev->flags & IFF_PROMISC)
2011 dev_set_promiscuity(slave_dev, -1);
2013 /* unset allmulti level from slave */
2014 if (bond_dev->flags & IFF_ALLMULTI)
2015 dev_set_allmulti(slave_dev, -1);
2017 /* flush master's mc_list from slave */
2018 netif_addr_lock_bh(bond_dev);
2019 bond_mc_list_flush(bond_dev, slave_dev);
2020 netif_addr_unlock_bh(bond_dev);
2023 netdev_set_master(slave_dev, NULL);
2025 #ifdef CONFIG_NET_POLL_CONTROLLER
2026 read_lock_bh(&bond->lock);
2028 if (slaves_support_netpoll(bond_dev))
2029 bond_dev->priv_flags &= ~IFF_DISABLE_NETPOLL;
2030 read_unlock_bh(&bond->lock);
2031 if (slave_dev->netdev_ops->ndo_netpoll_cleanup)
2032 slave_dev->netdev_ops->ndo_netpoll_cleanup(slave_dev);
2034 slave_dev->npinfo = NULL;
2037 /* close slave before restoring its mac address */
2038 dev_close(slave_dev);
2040 if (bond->params.fail_over_mac != BOND_FOM_ACTIVE) {
2041 /* restore original ("permanent") mac address */
2042 memcpy(addr.sa_data, slave->perm_hwaddr, ETH_ALEN);
2043 addr.sa_family = slave_dev->type;
2044 dev_set_mac_address(slave_dev, &addr);
2047 dev_set_mtu(slave_dev, slave->original_mtu);
2049 slave_dev->priv_flags &= ~(IFF_MASTER_8023AD | IFF_MASTER_ALB |
2050 IFF_SLAVE_INACTIVE | IFF_BONDING |
2055 return 0; /* deletion OK */
2059 * First release a slave and than destroy the bond if no more slaves are left.
2060 * Must be under rtnl_lock when this function is called.
2062 static int bond_release_and_destroy(struct net_device *bond_dev,
2063 struct net_device *slave_dev)
2065 struct bonding *bond = netdev_priv(bond_dev);
2068 ret = bond_release(bond_dev, slave_dev);
2069 if ((ret == 0) && (bond->slave_cnt == 0)) {
2070 pr_info("%s: destroying bond %s.\n",
2071 bond_dev->name, bond_dev->name);
2072 unregister_netdevice(bond_dev);
2078 * This function releases all slaves.
2080 static int bond_release_all(struct net_device *bond_dev)
2082 struct bonding *bond = netdev_priv(bond_dev);
2083 struct slave *slave;
2084 struct net_device *slave_dev;
2085 struct sockaddr addr;
2087 write_lock_bh(&bond->lock);
2089 netif_carrier_off(bond_dev);
2091 if (bond->slave_cnt == 0)
2094 bond->current_arp_slave = NULL;
2095 bond->primary_slave = NULL;
2096 bond_change_active_slave(bond, NULL);
2098 while ((slave = bond->first_slave) != NULL) {
2099 /* Inform AD package of unbinding of slave
2100 * before slave is detached from the list.
2102 if (bond->params.mode == BOND_MODE_8023AD)
2103 bond_3ad_unbind_slave(slave);
2105 slave_dev = slave->dev;
2106 bond_detach_slave(bond, slave);
2108 /* now that the slave is detached, unlock and perform
2109 * all the undo steps that should not be called from
2112 write_unlock_bh(&bond->lock);
2114 if (bond_is_lb(bond)) {
2115 /* must be called only after the slave
2116 * has been detached from the list
2118 bond_alb_deinit_slave(bond, slave);
2121 bond_compute_features(bond);
2123 bond_destroy_slave_symlinks(bond_dev, slave_dev);
2124 bond_del_vlans_from_slave(bond, slave_dev);
2126 /* If the mode USES_PRIMARY, then we should only remove its
2127 * promisc and mc settings if it was the curr_active_slave, but that was
2128 * already taken care of above when we detached the slave
2130 if (!USES_PRIMARY(bond->params.mode)) {
2131 /* unset promiscuity level from slave */
2132 if (bond_dev->flags & IFF_PROMISC)
2133 dev_set_promiscuity(slave_dev, -1);
2135 /* unset allmulti level from slave */
2136 if (bond_dev->flags & IFF_ALLMULTI)
2137 dev_set_allmulti(slave_dev, -1);
2139 /* flush master's mc_list from slave */
2140 netif_addr_lock_bh(bond_dev);
2141 bond_mc_list_flush(bond_dev, slave_dev);
2142 netif_addr_unlock_bh(bond_dev);
2145 netdev_set_master(slave_dev, NULL);
2147 /* close slave before restoring its mac address */
2148 dev_close(slave_dev);
2150 if (!bond->params.fail_over_mac) {
2151 /* restore original ("permanent") mac address*/
2152 memcpy(addr.sa_data, slave->perm_hwaddr, ETH_ALEN);
2153 addr.sa_family = slave_dev->type;
2154 dev_set_mac_address(slave_dev, &addr);
2157 slave_dev->priv_flags &= ~(IFF_MASTER_8023AD | IFF_MASTER_ALB |
2158 IFF_SLAVE_INACTIVE);
2162 /* re-acquire the lock before getting the next slave */
2163 write_lock_bh(&bond->lock);
2166 /* zero the mac address of the master so it will be
2167 * set by the application to the mac address of the
2170 memset(bond_dev->dev_addr, 0, bond_dev->addr_len);
2173 bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
2175 pr_warning("%s: Warning: clearing HW address of %s while it still has VLANs.\n",
2176 bond_dev->name, bond_dev->name);
2177 pr_warning("%s: When re-adding slaves, make sure the bond's HW address matches its VLANs'.\n",
2181 pr_info("%s: released all slaves\n", bond_dev->name);
2184 write_unlock_bh(&bond->lock);
2189 * This function changes the active slave to slave <slave_dev>.
2190 * It returns -EINVAL in the following cases.
2191 * - <slave_dev> is not found in the list.
2192 * - There is not active slave now.
2193 * - <slave_dev> is already active.
2194 * - The link state of <slave_dev> is not BOND_LINK_UP.
2195 * - <slave_dev> is not running.
2196 * In these cases, this function does nothing.
2197 * In the other cases, current_slave pointer is changed and 0 is returned.
2199 static int bond_ioctl_change_active(struct net_device *bond_dev, struct net_device *slave_dev)
2201 struct bonding *bond = netdev_priv(bond_dev);
2202 struct slave *old_active = NULL;
2203 struct slave *new_active = NULL;
2206 if (!USES_PRIMARY(bond->params.mode))
2209 /* Verify that master_dev is indeed the master of slave_dev */
2210 if (!(slave_dev->flags & IFF_SLAVE) || (slave_dev->master != bond_dev))
2213 read_lock(&bond->lock);
2215 read_lock(&bond->curr_slave_lock);
2216 old_active = bond->curr_active_slave;
2217 read_unlock(&bond->curr_slave_lock);
2219 new_active = bond_get_slave_by_dev(bond, slave_dev);
2222 * Changing to the current active: do nothing; return success.
2224 if (new_active && (new_active == old_active)) {
2225 read_unlock(&bond->lock);
2231 (new_active->link == BOND_LINK_UP) &&
2232 IS_UP(new_active->dev)) {
2234 write_lock_bh(&bond->curr_slave_lock);
2235 bond_change_active_slave(bond, new_active);
2236 write_unlock_bh(&bond->curr_slave_lock);
2237 unblock_netpoll_tx();
2241 read_unlock(&bond->lock);
2246 static int bond_info_query(struct net_device *bond_dev, struct ifbond *info)
2248 struct bonding *bond = netdev_priv(bond_dev);
2250 info->bond_mode = bond->params.mode;
2251 info->miimon = bond->params.miimon;
2253 read_lock(&bond->lock);
2254 info->num_slaves = bond->slave_cnt;
2255 read_unlock(&bond->lock);
2260 static int bond_slave_info_query(struct net_device *bond_dev, struct ifslave *info)
2262 struct bonding *bond = netdev_priv(bond_dev);
2263 struct slave *slave;
2264 int i, res = -ENODEV;
2266 read_lock(&bond->lock);
2268 bond_for_each_slave(bond, slave, i) {
2269 if (i == (int)info->slave_id) {
2271 strcpy(info->slave_name, slave->dev->name);
2272 info->link = slave->link;
2273 info->state = slave->state;
2274 info->link_failure_count = slave->link_failure_count;
2279 read_unlock(&bond->lock);
2284 /*-------------------------------- Monitoring -------------------------------*/
2287 static int bond_miimon_inspect(struct bonding *bond)
2289 struct slave *slave;
2290 int i, link_state, commit = 0;
2291 bool ignore_updelay;
2293 ignore_updelay = !bond->curr_active_slave ? true : false;
2295 bond_for_each_slave(bond, slave, i) {
2296 slave->new_link = BOND_LINK_NOCHANGE;
2298 link_state = bond_check_dev_link(bond, slave->dev, 0);
2300 switch (slave->link) {
2305 slave->link = BOND_LINK_FAIL;
2306 slave->delay = bond->params.downdelay;
2308 pr_info("%s: link status down for %sinterface %s, disabling it in %d ms.\n",
2310 (bond->params.mode ==
2311 BOND_MODE_ACTIVEBACKUP) ?
2312 ((slave->state == BOND_STATE_ACTIVE) ?
2313 "active " : "backup ") : "",
2315 bond->params.downdelay * bond->params.miimon);
2318 case BOND_LINK_FAIL:
2321 * recovered before downdelay expired
2323 slave->link = BOND_LINK_UP;
2324 slave->jiffies = jiffies;
2325 pr_info("%s: link status up again after %d ms for interface %s.\n",
2327 (bond->params.downdelay - slave->delay) *
2328 bond->params.miimon,
2333 if (slave->delay <= 0) {
2334 slave->new_link = BOND_LINK_DOWN;
2342 case BOND_LINK_DOWN:
2346 slave->link = BOND_LINK_BACK;
2347 slave->delay = bond->params.updelay;
2350 pr_info("%s: link status up for interface %s, enabling it in %d ms.\n",
2351 bond->dev->name, slave->dev->name,
2352 ignore_updelay ? 0 :
2353 bond->params.updelay *
2354 bond->params.miimon);
2357 case BOND_LINK_BACK:
2359 slave->link = BOND_LINK_DOWN;
2360 pr_info("%s: link status down again after %d ms for interface %s.\n",
2362 (bond->params.updelay - slave->delay) *
2363 bond->params.miimon,
2372 if (slave->delay <= 0) {
2373 slave->new_link = BOND_LINK_UP;
2375 ignore_updelay = false;
2387 static void bond_miimon_commit(struct bonding *bond)
2389 struct slave *slave;
2392 bond_for_each_slave(bond, slave, i) {
2393 switch (slave->new_link) {
2394 case BOND_LINK_NOCHANGE:
2398 slave->link = BOND_LINK_UP;
2399 slave->jiffies = jiffies;
2401 if (bond->params.mode == BOND_MODE_8023AD) {
2402 /* prevent it from being the active one */
2403 slave->state = BOND_STATE_BACKUP;
2404 } else if (bond->params.mode != BOND_MODE_ACTIVEBACKUP) {
2405 /* make it immediately active */
2406 slave->state = BOND_STATE_ACTIVE;
2407 } else if (slave != bond->primary_slave) {
2408 /* prevent it from being the active one */
2409 slave->state = BOND_STATE_BACKUP;
2412 bond_update_speed_duplex(slave);
2414 pr_info("%s: link status definitely up for interface %s, %d Mbps %s duplex.\n",
2415 bond->dev->name, slave->dev->name,
2416 slave->speed, slave->duplex ? "full" : "half");
2418 /* notify ad that the link status has changed */
2419 if (bond->params.mode == BOND_MODE_8023AD)
2420 bond_3ad_handle_link_change(slave, BOND_LINK_UP);
2422 if (bond_is_lb(bond))
2423 bond_alb_handle_link_change(bond, slave,
2426 if (!bond->curr_active_slave ||
2427 (slave == bond->primary_slave))
2432 case BOND_LINK_DOWN:
2433 if (slave->link_failure_count < UINT_MAX)
2434 slave->link_failure_count++;
2436 slave->link = BOND_LINK_DOWN;
2438 if (bond->params.mode == BOND_MODE_ACTIVEBACKUP ||
2439 bond->params.mode == BOND_MODE_8023AD)
2440 bond_set_slave_inactive_flags(slave);
2442 pr_info("%s: link status definitely down for interface %s, disabling it\n",
2443 bond->dev->name, slave->dev->name);
2445 if (bond->params.mode == BOND_MODE_8023AD)
2446 bond_3ad_handle_link_change(slave,
2449 if (bond_is_lb(bond))
2450 bond_alb_handle_link_change(bond, slave,
2453 if (slave == bond->curr_active_slave)
2459 pr_err("%s: invalid new link %d on slave %s\n",
2460 bond->dev->name, slave->new_link,
2462 slave->new_link = BOND_LINK_NOCHANGE;
2470 write_lock_bh(&bond->curr_slave_lock);
2471 bond_select_active_slave(bond);
2472 write_unlock_bh(&bond->curr_slave_lock);
2473 unblock_netpoll_tx();
2476 bond_set_carrier(bond);
2482 * Really a wrapper that splits the mii monitor into two phases: an
2483 * inspection, then (if inspection indicates something needs to be done)
2484 * an acquisition of appropriate locks followed by a commit phase to
2485 * implement whatever link state changes are indicated.
2487 void bond_mii_monitor(struct work_struct *work)
2489 struct bonding *bond = container_of(work, struct bonding,
2492 read_lock(&bond->lock);
2493 if (bond->kill_timers)
2496 if (bond->slave_cnt == 0)
2499 if (bond->send_grat_arp) {
2500 read_lock(&bond->curr_slave_lock);
2501 bond_send_gratuitous_arp(bond);
2502 read_unlock(&bond->curr_slave_lock);
2505 if (bond->send_unsol_na) {
2506 read_lock(&bond->curr_slave_lock);
2507 bond_send_unsolicited_na(bond);
2508 read_unlock(&bond->curr_slave_lock);
2511 if (bond_miimon_inspect(bond)) {
2512 read_unlock(&bond->lock);
2514 read_lock(&bond->lock);
2516 bond_miimon_commit(bond);
2518 read_unlock(&bond->lock);
2519 rtnl_unlock(); /* might sleep, hold no other locks */
2520 read_lock(&bond->lock);
2524 if (bond->params.miimon)
2525 queue_delayed_work(bond->wq, &bond->mii_work,
2526 msecs_to_jiffies(bond->params.miimon));
2528 read_unlock(&bond->lock);
2531 static __be32 bond_glean_dev_ip(struct net_device *dev)
2533 struct in_device *idev;
2534 struct in_ifaddr *ifa;
2541 idev = __in_dev_get_rcu(dev);
2545 ifa = idev->ifa_list;
2549 addr = ifa->ifa_local;
2555 static int bond_has_this_ip(struct bonding *bond, __be32 ip)
2557 struct vlan_entry *vlan;
2559 if (ip == bond->master_ip)
2562 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
2563 if (ip == vlan->vlan_ip)
2571 * We go to the (large) trouble of VLAN tagging ARP frames because
2572 * switches in VLAN mode (especially if ports are configured as
2573 * "native" to a VLAN) might not pass non-tagged frames.
2575 static void bond_arp_send(struct net_device *slave_dev, int arp_op, __be32 dest_ip, __be32 src_ip, unsigned short vlan_id)
2577 struct sk_buff *skb;
2579 pr_debug("arp %d on slave %s: dst %x src %x vid %d\n", arp_op,
2580 slave_dev->name, dest_ip, src_ip, vlan_id);
2582 skb = arp_create(arp_op, ETH_P_ARP, dest_ip, slave_dev, src_ip,
2583 NULL, slave_dev->dev_addr, NULL);
2586 pr_err("ARP packet allocation failed\n");
2590 skb = vlan_put_tag(skb, vlan_id);
2592 pr_err("failed to insert VLAN tag\n");
2600 static void bond_arp_send_all(struct bonding *bond, struct slave *slave)
2603 __be32 *targets = bond->params.arp_targets;
2604 struct vlan_entry *vlan;
2605 struct net_device *vlan_dev;
2609 for (i = 0; (i < BOND_MAX_ARP_TARGETS); i++) {
2612 pr_debug("basa: target %x\n", targets[i]);
2614 pr_debug("basa: empty vlan: arp_send\n");
2615 bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2616 bond->master_ip, 0);
2621 * If VLANs are configured, we do a route lookup to
2622 * determine which VLAN interface would be used, so we
2623 * can tag the ARP with the proper VLAN tag.
2625 memset(&fl, 0, sizeof(fl));
2626 fl.fl4_dst = targets[i];
2627 fl.fl4_tos = RTO_ONLINK;
2629 rv = ip_route_output_key(dev_net(bond->dev), &rt, &fl);
2631 if (net_ratelimit()) {
2632 pr_warning("%s: no route to arp_ip_target %pI4\n",
2633 bond->dev->name, &fl.fl4_dst);
2639 * This target is not on a VLAN
2641 if (rt->dst.dev == bond->dev) {
2643 pr_debug("basa: rtdev == bond->dev: arp_send\n");
2644 bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2645 bond->master_ip, 0);
2650 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
2651 vlan_dev = vlan_group_get_device(bond->vlgrp, vlan->vlan_id);
2652 if (vlan_dev == rt->dst.dev) {
2653 vlan_id = vlan->vlan_id;
2654 pr_debug("basa: vlan match on %s %d\n",
2655 vlan_dev->name, vlan_id);
2662 bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2663 vlan->vlan_ip, vlan_id);
2667 if (net_ratelimit()) {
2668 pr_warning("%s: no path to arp_ip_target %pI4 via rt.dev %s\n",
2669 bond->dev->name, &fl.fl4_dst,
2670 rt->dst.dev ? rt->dst.dev->name : "NULL");
2677 * Kick out a gratuitous ARP for an IP on the bonding master plus one
2678 * for each VLAN above us.
2680 * Caller must hold curr_slave_lock for read or better
2682 static void bond_send_gratuitous_arp(struct bonding *bond)
2684 struct slave *slave = bond->curr_active_slave;
2685 struct vlan_entry *vlan;
2686 struct net_device *vlan_dev;
2688 pr_debug("bond_send_grat_arp: bond %s slave %s\n",
2689 bond->dev->name, slave ? slave->dev->name : "NULL");
2691 if (!slave || !bond->send_grat_arp ||
2692 test_bit(__LINK_STATE_LINKWATCH_PENDING, &slave->dev->state))
2695 bond->send_grat_arp--;
2697 if (bond->master_ip) {
2698 bond_arp_send(slave->dev, ARPOP_REPLY, bond->master_ip,
2699 bond->master_ip, 0);
2705 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
2706 vlan_dev = vlan_group_get_device(bond->vlgrp, vlan->vlan_id);
2707 if (vlan->vlan_ip) {
2708 bond_arp_send(slave->dev, ARPOP_REPLY, vlan->vlan_ip,
2709 vlan->vlan_ip, vlan->vlan_id);
2714 static void bond_validate_arp(struct bonding *bond, struct slave *slave, __be32 sip, __be32 tip)
2717 __be32 *targets = bond->params.arp_targets;
2719 for (i = 0; (i < BOND_MAX_ARP_TARGETS) && targets[i]; i++) {
2720 pr_debug("bva: sip %pI4 tip %pI4 t[%d] %pI4 bhti(tip) %d\n",
2721 &sip, &tip, i, &targets[i],
2722 bond_has_this_ip(bond, tip));
2723 if (sip == targets[i]) {
2724 if (bond_has_this_ip(bond, tip))
2725 slave->last_arp_rx = jiffies;
2731 static int bond_arp_rcv(struct sk_buff *skb, struct net_device *dev, struct packet_type *pt, struct net_device *orig_dev)
2734 struct slave *slave;
2735 struct bonding *bond;
2736 unsigned char *arp_ptr;
2739 if (dev->priv_flags & IFF_802_1Q_VLAN) {
2741 * When using VLANS and bonding, dev and oriv_dev may be
2742 * incorrect if the physical interface supports VLAN
2743 * acceleration. With this change ARP validation now
2744 * works for hosts only reachable on the VLAN interface.
2746 dev = vlan_dev_real_dev(dev);
2747 orig_dev = dev_get_by_index_rcu(dev_net(skb->dev),skb->skb_iif);
2750 if (!(dev->priv_flags & IFF_BONDING) || !(dev->flags & IFF_MASTER))
2753 bond = netdev_priv(dev);
2754 read_lock(&bond->lock);
2756 pr_debug("bond_arp_rcv: bond %s skb->dev %s orig_dev %s\n",
2757 bond->dev->name, skb->dev ? skb->dev->name : "NULL",
2758 orig_dev ? orig_dev->name : "NULL");
2760 slave = bond_get_slave_by_dev(bond, orig_dev);
2761 if (!slave || !slave_do_arp_validate(bond, slave))
2764 if (!pskb_may_pull(skb, arp_hdr_len(dev)))
2768 if (arp->ar_hln != dev->addr_len ||
2769 skb->pkt_type == PACKET_OTHERHOST ||
2770 skb->pkt_type == PACKET_LOOPBACK ||
2771 arp->ar_hrd != htons(ARPHRD_ETHER) ||
2772 arp->ar_pro != htons(ETH_P_IP) ||
2776 arp_ptr = (unsigned char *)(arp + 1);
2777 arp_ptr += dev->addr_len;
2778 memcpy(&sip, arp_ptr, 4);
2779 arp_ptr += 4 + dev->addr_len;
2780 memcpy(&tip, arp_ptr, 4);
2782 pr_debug("bond_arp_rcv: %s %s/%d av %d sv %d sip %pI4 tip %pI4\n",
2783 bond->dev->name, slave->dev->name, slave->state,
2784 bond->params.arp_validate, slave_do_arp_validate(bond, slave),
2788 * Backup slaves won't see the ARP reply, but do come through
2789 * here for each ARP probe (so we swap the sip/tip to validate
2790 * the probe). In a "redundant switch, common router" type of
2791 * configuration, the ARP probe will (hopefully) travel from
2792 * the active, through one switch, the router, then the other
2793 * switch before reaching the backup.
2795 if (slave->state == BOND_STATE_ACTIVE)
2796 bond_validate_arp(bond, slave, sip, tip);
2798 bond_validate_arp(bond, slave, tip, sip);
2801 read_unlock(&bond->lock);
2804 return NET_RX_SUCCESS;
2808 * this function is called regularly to monitor each slave's link
2809 * ensuring that traffic is being sent and received when arp monitoring
2810 * is used in load-balancing mode. if the adapter has been dormant, then an
2811 * arp is transmitted to generate traffic. see activebackup_arp_monitor for
2812 * arp monitoring in active backup mode.
2814 void bond_loadbalance_arp_mon(struct work_struct *work)
2816 struct bonding *bond = container_of(work, struct bonding,
2818 struct slave *slave, *oldcurrent;
2819 int do_failover = 0;
2823 read_lock(&bond->lock);
2825 delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
2827 if (bond->kill_timers)
2830 if (bond->slave_cnt == 0)
2833 read_lock(&bond->curr_slave_lock);
2834 oldcurrent = bond->curr_active_slave;
2835 read_unlock(&bond->curr_slave_lock);
2837 /* see if any of the previous devices are up now (i.e. they have
2838 * xmt and rcv traffic). the curr_active_slave does not come into
2839 * the picture unless it is null. also, slave->jiffies is not needed
2840 * here because we send an arp on each slave and give a slave as
2841 * long as it needs to get the tx/rx within the delta.
2842 * TODO: what about up/down delay in arp mode? it wasn't here before
2845 bond_for_each_slave(bond, slave, i) {
2846 unsigned long trans_start = dev_trans_start(slave->dev);
2848 if (slave->link != BOND_LINK_UP) {
2849 if (time_in_range(jiffies,
2850 trans_start - delta_in_ticks,
2851 trans_start + delta_in_ticks) &&
2852 time_in_range(jiffies,
2853 slave->dev->last_rx - delta_in_ticks,
2854 slave->dev->last_rx + delta_in_ticks)) {
2856 slave->link = BOND_LINK_UP;
2857 slave->state = BOND_STATE_ACTIVE;
2859 /* primary_slave has no meaning in round-robin
2860 * mode. the window of a slave being up and
2861 * curr_active_slave being null after enslaving
2865 pr_info("%s: link status definitely up for interface %s, ",
2870 pr_info("%s: interface %s is now up\n",
2876 /* slave->link == BOND_LINK_UP */
2878 /* not all switches will respond to an arp request
2879 * when the source ip is 0, so don't take the link down
2880 * if we don't know our ip yet
2882 if (!time_in_range(jiffies,
2883 trans_start - delta_in_ticks,
2884 trans_start + 2 * delta_in_ticks) ||
2885 !time_in_range(jiffies,
2886 slave->dev->last_rx - delta_in_ticks,
2887 slave->dev->last_rx + 2 * delta_in_ticks)) {
2889 slave->link = BOND_LINK_DOWN;
2890 slave->state = BOND_STATE_BACKUP;
2892 if (slave->link_failure_count < UINT_MAX)
2893 slave->link_failure_count++;
2895 pr_info("%s: interface %s is now down.\n",
2899 if (slave == oldcurrent)
2904 /* note: if switch is in round-robin mode, all links
2905 * must tx arp to ensure all links rx an arp - otherwise
2906 * links may oscillate or not come up at all; if switch is
2907 * in something like xor mode, there is nothing we can
2908 * do - all replies will be rx'ed on same link causing slaves
2909 * to be unstable during low/no traffic periods
2911 if (IS_UP(slave->dev))
2912 bond_arp_send_all(bond, slave);
2917 write_lock_bh(&bond->curr_slave_lock);
2919 bond_select_active_slave(bond);
2921 write_unlock_bh(&bond->curr_slave_lock);
2922 unblock_netpoll_tx();
2926 if (bond->params.arp_interval)
2927 queue_delayed_work(bond->wq, &bond->arp_work, delta_in_ticks);
2929 read_unlock(&bond->lock);
2933 * Called to inspect slaves for active-backup mode ARP monitor link state
2934 * changes. Sets new_link in slaves to specify what action should take
2935 * place for the slave. Returns 0 if no changes are found, >0 if changes
2936 * to link states must be committed.
2938 * Called with bond->lock held for read.
2940 static int bond_ab_arp_inspect(struct bonding *bond, int delta_in_ticks)
2942 struct slave *slave;
2944 unsigned long trans_start;
2946 bond_for_each_slave(bond, slave, i) {
2947 slave->new_link = BOND_LINK_NOCHANGE;
2949 if (slave->link != BOND_LINK_UP) {
2950 if (time_in_range(jiffies,
2951 slave_last_rx(bond, slave) - delta_in_ticks,
2952 slave_last_rx(bond, slave) + delta_in_ticks)) {
2954 slave->new_link = BOND_LINK_UP;
2962 * Give slaves 2*delta after being enslaved or made
2963 * active. This avoids bouncing, as the last receive
2964 * times need a full ARP monitor cycle to be updated.
2966 if (time_in_range(jiffies,
2967 slave->jiffies - delta_in_ticks,
2968 slave->jiffies + 2 * delta_in_ticks))
2972 * Backup slave is down if:
2973 * - No current_arp_slave AND
2974 * - more than 3*delta since last receive AND
2975 * - the bond has an IP address
2977 * Note: a non-null current_arp_slave indicates
2978 * the curr_active_slave went down and we are
2979 * searching for a new one; under this condition
2980 * we only take the curr_active_slave down - this
2981 * gives each slave a chance to tx/rx traffic
2982 * before being taken out
2984 if (slave->state == BOND_STATE_BACKUP &&
2985 !bond->current_arp_slave &&
2986 !time_in_range(jiffies,
2987 slave_last_rx(bond, slave) - delta_in_ticks,
2988 slave_last_rx(bond, slave) + 3 * delta_in_ticks)) {
2990 slave->new_link = BOND_LINK_DOWN;
2995 * Active slave is down if:
2996 * - more than 2*delta since transmitting OR
2997 * - (more than 2*delta since receive AND
2998 * the bond has an IP address)
3000 trans_start = dev_trans_start(slave->dev);
3001 if ((slave->state == BOND_STATE_ACTIVE) &&
3002 (!time_in_range(jiffies,
3003 trans_start - delta_in_ticks,
3004 trans_start + 2 * delta_in_ticks) ||
3005 !time_in_range(jiffies,
3006 slave_last_rx(bond, slave) - delta_in_ticks,
3007 slave_last_rx(bond, slave) + 2 * delta_in_ticks))) {
3009 slave->new_link = BOND_LINK_DOWN;
3018 * Called to commit link state changes noted by inspection step of
3019 * active-backup mode ARP monitor.
3021 * Called with RTNL and bond->lock for read.
3023 static void bond_ab_arp_commit(struct bonding *bond, int delta_in_ticks)
3025 struct slave *slave;
3027 unsigned long trans_start;
3029 bond_for_each_slave(bond, slave, i) {
3030 switch (slave->new_link) {
3031 case BOND_LINK_NOCHANGE:
3035 trans_start = dev_trans_start(slave->dev);
3036 if ((!bond->curr_active_slave &&
3037 time_in_range(jiffies,
3038 trans_start - delta_in_ticks,
3039 trans_start + delta_in_ticks)) ||
3040 bond->curr_active_slave != slave) {
3041 slave->link = BOND_LINK_UP;
3042 bond->current_arp_slave = NULL;
3044 pr_info("%s: link status definitely up for interface %s.\n",
3045 bond->dev->name, slave->dev->name);
3047 if (!bond->curr_active_slave ||
3048 (slave == bond->primary_slave))
3055 case BOND_LINK_DOWN:
3056 if (slave->link_failure_count < UINT_MAX)
3057 slave->link_failure_count++;
3059 slave->link = BOND_LINK_DOWN;
3060 bond_set_slave_inactive_flags(slave);
3062 pr_info("%s: link status definitely down for interface %s, disabling it\n",
3063 bond->dev->name, slave->dev->name);
3065 if (slave == bond->curr_active_slave) {
3066 bond->current_arp_slave = NULL;
3073 pr_err("%s: impossible: new_link %d on slave %s\n",
3074 bond->dev->name, slave->new_link,
3082 write_lock_bh(&bond->curr_slave_lock);
3083 bond_select_active_slave(bond);
3084 write_unlock_bh(&bond->curr_slave_lock);
3085 unblock_netpoll_tx();
3088 bond_set_carrier(bond);
3092 * Send ARP probes for active-backup mode ARP monitor.
3094 * Called with bond->lock held for read.
3096 static void bond_ab_arp_probe(struct bonding *bond)
3098 struct slave *slave;
3101 read_lock(&bond->curr_slave_lock);
3103 if (bond->current_arp_slave && bond->curr_active_slave)
3104 pr_info("PROBE: c_arp %s && cas %s BAD\n",
3105 bond->current_arp_slave->dev->name,
3106 bond->curr_active_slave->dev->name);
3108 if (bond->curr_active_slave) {
3109 bond_arp_send_all(bond, bond->curr_active_slave);
3110 read_unlock(&bond->curr_slave_lock);
3114 read_unlock(&bond->curr_slave_lock);
3116 /* if we don't have a curr_active_slave, search for the next available
3117 * backup slave from the current_arp_slave and make it the candidate
3118 * for becoming the curr_active_slave
3121 if (!bond->current_arp_slave) {
3122 bond->current_arp_slave = bond->first_slave;
3123 if (!bond->current_arp_slave)
3127 bond_set_slave_inactive_flags(bond->current_arp_slave);
3129 /* search for next candidate */
3130 bond_for_each_slave_from(bond, slave, i, bond->current_arp_slave->next) {
3131 if (IS_UP(slave->dev)) {
3132 slave->link = BOND_LINK_BACK;
3133 bond_set_slave_active_flags(slave);
3134 bond_arp_send_all(bond, slave);
3135 slave->jiffies = jiffies;
3136 bond->current_arp_slave = slave;
3140 /* if the link state is up at this point, we
3141 * mark it down - this can happen if we have
3142 * simultaneous link failures and
3143 * reselect_active_interface doesn't make this
3144 * one the current slave so it is still marked
3145 * up when it is actually down
3147 if (slave->link == BOND_LINK_UP) {
3148 slave->link = BOND_LINK_DOWN;
3149 if (slave->link_failure_count < UINT_MAX)
3150 slave->link_failure_count++;
3152 bond_set_slave_inactive_flags(slave);
3154 pr_info("%s: backup interface %s is now down.\n",
3155 bond->dev->name, slave->dev->name);
3160 void bond_activebackup_arp_mon(struct work_struct *work)
3162 struct bonding *bond = container_of(work, struct bonding,
3166 read_lock(&bond->lock);
3168 if (bond->kill_timers)
3171 delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
3173 if (bond->slave_cnt == 0)
3176 if (bond->send_grat_arp) {
3177 read_lock(&bond->curr_slave_lock);
3178 bond_send_gratuitous_arp(bond);
3179 read_unlock(&bond->curr_slave_lock);
3182 if (bond->send_unsol_na) {
3183 read_lock(&bond->curr_slave_lock);
3184 bond_send_unsolicited_na(bond);
3185 read_unlock(&bond->curr_slave_lock);
3188 if (bond_ab_arp_inspect(bond, delta_in_ticks)) {
3189 read_unlock(&bond->lock);
3191 read_lock(&bond->lock);
3193 bond_ab_arp_commit(bond, delta_in_ticks);
3195 read_unlock(&bond->lock);
3197 read_lock(&bond->lock);
3200 bond_ab_arp_probe(bond);
3203 if (bond->params.arp_interval)
3204 queue_delayed_work(bond->wq, &bond->arp_work, delta_in_ticks);
3206 read_unlock(&bond->lock);
3209 /*------------------------------ proc/seq_file-------------------------------*/
3211 #ifdef CONFIG_PROC_FS
3213 static void *bond_info_seq_start(struct seq_file *seq, loff_t *pos)
3214 __acquires(&dev_base_lock)
3215 __acquires(&bond->lock)
3217 struct bonding *bond = seq->private;
3219 struct slave *slave;
3222 /* make sure the bond won't be taken away */
3223 read_lock(&dev_base_lock);
3224 read_lock(&bond->lock);
3227 return SEQ_START_TOKEN;
3229 bond_for_each_slave(bond, slave, i) {
3237 static void *bond_info_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3239 struct bonding *bond = seq->private;
3240 struct slave *slave = v;
3243 if (v == SEQ_START_TOKEN)
3244 return bond->first_slave;
3246 slave = slave->next;
3248 return (slave == bond->first_slave) ? NULL : slave;
3251 static void bond_info_seq_stop(struct seq_file *seq, void *v)
3252 __releases(&bond->lock)
3253 __releases(&dev_base_lock)
3255 struct bonding *bond = seq->private;
3257 read_unlock(&bond->lock);
3258 read_unlock(&dev_base_lock);
3261 static void bond_info_show_master(struct seq_file *seq)
3263 struct bonding *bond = seq->private;
3267 read_lock(&bond->curr_slave_lock);
3268 curr = bond->curr_active_slave;
3269 read_unlock(&bond->curr_slave_lock);
3271 seq_printf(seq, "Bonding Mode: %s",
3272 bond_mode_name(bond->params.mode));
3274 if (bond->params.mode == BOND_MODE_ACTIVEBACKUP &&
3275 bond->params.fail_over_mac)
3276 seq_printf(seq, " (fail_over_mac %s)",
3277 fail_over_mac_tbl[bond->params.fail_over_mac].modename);
3279 seq_printf(seq, "\n");
3281 if (bond->params.mode == BOND_MODE_XOR ||
3282 bond->params.mode == BOND_MODE_8023AD) {
3283 seq_printf(seq, "Transmit Hash Policy: %s (%d)\n",
3284 xmit_hashtype_tbl[bond->params.xmit_policy].modename,
3285 bond->params.xmit_policy);
3288 if (USES_PRIMARY(bond->params.mode)) {
3289 seq_printf(seq, "Primary Slave: %s",
3290 (bond->primary_slave) ?
3291 bond->primary_slave->dev->name : "None");
3292 if (bond->primary_slave)
3293 seq_printf(seq, " (primary_reselect %s)",
3294 pri_reselect_tbl[bond->params.primary_reselect].modename);
3296 seq_printf(seq, "\nCurrently Active Slave: %s\n",
3297 (curr) ? curr->dev->name : "None");
3300 seq_printf(seq, "MII Status: %s\n", netif_carrier_ok(bond->dev) ?
3302 seq_printf(seq, "MII Polling Interval (ms): %d\n", bond->params.miimon);
3303 seq_printf(seq, "Up Delay (ms): %d\n",
3304 bond->params.updelay * bond->params.miimon);
3305 seq_printf(seq, "Down Delay (ms): %d\n",
3306 bond->params.downdelay * bond->params.miimon);
3309 /* ARP information */
3310 if (bond->params.arp_interval > 0) {
3312 seq_printf(seq, "ARP Polling Interval (ms): %d\n",
3313 bond->params.arp_interval);
3315 seq_printf(seq, "ARP IP target/s (n.n.n.n form):");
3317 for (i = 0; (i < BOND_MAX_ARP_TARGETS); i++) {
3318 if (!bond->params.arp_targets[i])
3321 seq_printf(seq, ",");
3322 seq_printf(seq, " %pI4", &bond->params.arp_targets[i]);
3325 seq_printf(seq, "\n");
3328 if (bond->params.mode == BOND_MODE_8023AD) {
3329 struct ad_info ad_info;
3331 seq_puts(seq, "\n802.3ad info\n");
3332 seq_printf(seq, "LACP rate: %s\n",
3333 (bond->params.lacp_fast) ? "fast" : "slow");
3334 seq_printf(seq, "Aggregator selection policy (ad_select): %s\n",
3335 ad_select_tbl[bond->params.ad_select].modename);
3337 if (bond_3ad_get_active_agg_info(bond, &ad_info)) {
3338 seq_printf(seq, "bond %s has no active aggregator\n",
3341 seq_printf(seq, "Active Aggregator Info:\n");
3343 seq_printf(seq, "\tAggregator ID: %d\n",
3344 ad_info.aggregator_id);
3345 seq_printf(seq, "\tNumber of ports: %d\n",
3347 seq_printf(seq, "\tActor Key: %d\n",
3349 seq_printf(seq, "\tPartner Key: %d\n",
3350 ad_info.partner_key);
3351 seq_printf(seq, "\tPartner Mac Address: %pM\n",
3352 ad_info.partner_system);
3357 static void bond_info_show_slave(struct seq_file *seq,
3358 const struct slave *slave)
3360 struct bonding *bond = seq->private;
3362 seq_printf(seq, "\nSlave Interface: %s\n", slave->dev->name);
3363 seq_printf(seq, "MII Status: %s\n",
3364 (slave->link == BOND_LINK_UP) ? "up" : "down");
3365 seq_printf(seq, "Speed: %d Mbps\n", slave->speed);
3366 seq_printf(seq, "Duplex: %s\n", slave->duplex ? "full" : "half");
3367 seq_printf(seq, "Link Failure Count: %u\n",
3368 slave->link_failure_count);
3370 seq_printf(seq, "Permanent HW addr: %pM\n", slave->perm_hwaddr);
3372 if (bond->params.mode == BOND_MODE_8023AD) {
3373 const struct aggregator *agg
3374 = SLAVE_AD_INFO(slave).port.aggregator;
3377 seq_printf(seq, "Aggregator ID: %d\n",
3378 agg->aggregator_identifier);
3380 seq_puts(seq, "Aggregator ID: N/A\n");
3382 seq_printf(seq, "Slave queue ID: %d\n", slave->queue_id);
3385 static int bond_info_seq_show(struct seq_file *seq, void *v)
3387 if (v == SEQ_START_TOKEN) {
3388 seq_printf(seq, "%s\n", version);
3389 bond_info_show_master(seq);
3391 bond_info_show_slave(seq, v);
3396 static const struct seq_operations bond_info_seq_ops = {
3397 .start = bond_info_seq_start,
3398 .next = bond_info_seq_next,
3399 .stop = bond_info_seq_stop,
3400 .show = bond_info_seq_show,
3403 static int bond_info_open(struct inode *inode, struct file *file)
3405 struct seq_file *seq;
3406 struct proc_dir_entry *proc;
3409 res = seq_open(file, &bond_info_seq_ops);
3411 /* recover the pointer buried in proc_dir_entry data */
3412 seq = file->private_data;
3414 seq->private = proc->data;
3420 static const struct file_operations bond_info_fops = {
3421 .owner = THIS_MODULE,
3422 .open = bond_info_open,
3424 .llseek = seq_lseek,
3425 .release = seq_release,
3428 static void bond_create_proc_entry(struct bonding *bond)
3430 struct net_device *bond_dev = bond->dev;
3431 struct bond_net *bn = net_generic(dev_net(bond_dev), bond_net_id);
3434 bond->proc_entry = proc_create_data(bond_dev->name,
3435 S_IRUGO, bn->proc_dir,
3436 &bond_info_fops, bond);
3437 if (bond->proc_entry == NULL)
3438 pr_warning("Warning: Cannot create /proc/net/%s/%s\n",
3439 DRV_NAME, bond_dev->name);
3441 memcpy(bond->proc_file_name, bond_dev->name, IFNAMSIZ);
3445 static void bond_remove_proc_entry(struct bonding *bond)
3447 struct net_device *bond_dev = bond->dev;
3448 struct bond_net *bn = net_generic(dev_net(bond_dev), bond_net_id);
3450 if (bn->proc_dir && bond->proc_entry) {
3451 remove_proc_entry(bond->proc_file_name, bn->proc_dir);
3452 memset(bond->proc_file_name, 0, IFNAMSIZ);
3453 bond->proc_entry = NULL;
3457 /* Create the bonding directory under /proc/net, if doesn't exist yet.
3458 * Caller must hold rtnl_lock.
3460 static void __net_init bond_create_proc_dir(struct bond_net *bn)
3462 if (!bn->proc_dir) {
3463 bn->proc_dir = proc_mkdir(DRV_NAME, bn->net->proc_net);
3465 pr_warning("Warning: cannot create /proc/net/%s\n",
3470 /* Destroy the bonding directory under /proc/net, if empty.
3471 * Caller must hold rtnl_lock.
3473 static void __net_exit bond_destroy_proc_dir(struct bond_net *bn)
3476 remove_proc_entry(DRV_NAME, bn->net->proc_net);
3477 bn->proc_dir = NULL;
3481 #else /* !CONFIG_PROC_FS */
3483 static void bond_create_proc_entry(struct bonding *bond)
3487 static void bond_remove_proc_entry(struct bonding *bond)
3491 static inline void bond_create_proc_dir(struct bond_net *bn)
3495 static inline void bond_destroy_proc_dir(struct bond_net *bn)
3499 #endif /* CONFIG_PROC_FS */
3502 /*-------------------------- netdev event handling --------------------------*/
3505 * Change device name
3507 static int bond_event_changename(struct bonding *bond)
3509 bond_remove_proc_entry(bond);
3510 bond_create_proc_entry(bond);
3515 static int bond_master_netdev_event(unsigned long event,
3516 struct net_device *bond_dev)
3518 struct bonding *event_bond = netdev_priv(bond_dev);
3521 case NETDEV_CHANGENAME:
3522 return bond_event_changename(event_bond);
3530 static int bond_slave_netdev_event(unsigned long event,
3531 struct net_device *slave_dev)
3533 struct net_device *bond_dev = slave_dev->master;
3534 struct bonding *bond = netdev_priv(bond_dev);
3537 case NETDEV_UNREGISTER:
3539 if (bond->setup_by_slave)
3540 bond_release_and_destroy(bond_dev, slave_dev);
3542 bond_release(bond_dev, slave_dev);
3546 if (bond->params.mode == BOND_MODE_8023AD || bond_is_lb(bond)) {
3547 struct slave *slave;
3549 slave = bond_get_slave_by_dev(bond, slave_dev);
3551 u16 old_speed = slave->speed;
3552 u16 old_duplex = slave->duplex;
3554 bond_update_speed_duplex(slave);
3556 if (bond_is_lb(bond))
3559 if (old_speed != slave->speed)
3560 bond_3ad_adapter_speed_changed(slave);
3561 if (old_duplex != slave->duplex)
3562 bond_3ad_adapter_duplex_changed(slave);
3569 * ... Or is it this?
3572 case NETDEV_CHANGEMTU:
3574 * TODO: Should slaves be allowed to
3575 * independently alter their MTU? For
3576 * an active-backup bond, slaves need
3577 * not be the same type of device, so
3578 * MTUs may vary. For other modes,
3579 * slaves arguably should have the
3580 * same MTUs. To do this, we'd need to
3581 * take over the slave's change_mtu
3582 * function for the duration of their
3586 case NETDEV_CHANGENAME:
3588 * TODO: handle changing the primary's name
3591 case NETDEV_FEAT_CHANGE:
3592 bond_compute_features(bond);
3602 * bond_netdev_event: handle netdev notifier chain events.
3604 * This function receives events for the netdev chain. The caller (an
3605 * ioctl handler calling blocking_notifier_call_chain) holds the necessary
3606 * locks for us to safely manipulate the slave devices (RTNL lock,
3609 static int bond_netdev_event(struct notifier_block *this,
3610 unsigned long event, void *ptr)
3612 struct net_device *event_dev = (struct net_device *)ptr;
3614 pr_debug("event_dev: %s, event: %lx\n",
3615 event_dev ? event_dev->name : "None",
3618 if (!(event_dev->priv_flags & IFF_BONDING))
3621 if (event_dev->flags & IFF_MASTER) {
3622 pr_debug("IFF_MASTER\n");
3623 return bond_master_netdev_event(event, event_dev);
3626 if (event_dev->flags & IFF_SLAVE) {
3627 pr_debug("IFF_SLAVE\n");
3628 return bond_slave_netdev_event(event, event_dev);
3635 * bond_inetaddr_event: handle inetaddr notifier chain events.
3637 * We keep track of device IPs primarily to use as source addresses in
3638 * ARP monitor probes (rather than spewing out broadcasts all the time).
3640 * We track one IP for the main device (if it has one), plus one per VLAN.
3642 static int bond_inetaddr_event(struct notifier_block *this, unsigned long event, void *ptr)
3644 struct in_ifaddr *ifa = ptr;
3645 struct net_device *vlan_dev, *event_dev = ifa->ifa_dev->dev;
3646 struct bond_net *bn = net_generic(dev_net(event_dev), bond_net_id);
3647 struct bonding *bond;
3648 struct vlan_entry *vlan;
3650 list_for_each_entry(bond, &bn->dev_list, bond_list) {
3651 if (bond->dev == event_dev) {
3654 bond->master_ip = ifa->ifa_local;
3657 bond->master_ip = bond_glean_dev_ip(bond->dev);
3664 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
3667 vlan_dev = vlan_group_get_device(bond->vlgrp, vlan->vlan_id);
3668 if (vlan_dev == event_dev) {
3671 vlan->vlan_ip = ifa->ifa_local;
3675 bond_glean_dev_ip(vlan_dev);
3686 static struct notifier_block bond_netdev_notifier = {
3687 .notifier_call = bond_netdev_event,
3690 static struct notifier_block bond_inetaddr_notifier = {
3691 .notifier_call = bond_inetaddr_event,
3694 /*-------------------------- Packet type handling ---------------------------*/
3696 /* register to receive lacpdus on a bond */
3697 static void bond_register_lacpdu(struct bonding *bond)
3699 struct packet_type *pk_type = &(BOND_AD_INFO(bond).ad_pkt_type);
3701 /* initialize packet type */
3702 pk_type->type = PKT_TYPE_LACPDU;
3703 pk_type->dev = bond->dev;
3704 pk_type->func = bond_3ad_lacpdu_recv;
3706 dev_add_pack(pk_type);
3709 /* unregister to receive lacpdus on a bond */
3710 static void bond_unregister_lacpdu(struct bonding *bond)
3712 dev_remove_pack(&(BOND_AD_INFO(bond).ad_pkt_type));
3715 void bond_register_arp(struct bonding *bond)
3717 struct packet_type *pt = &bond->arp_mon_pt;
3722 pt->type = htons(ETH_P_ARP);
3723 pt->dev = bond->dev;
3724 pt->func = bond_arp_rcv;
3728 void bond_unregister_arp(struct bonding *bond)
3730 struct packet_type *pt = &bond->arp_mon_pt;
3732 dev_remove_pack(pt);
3736 /*---------------------------- Hashing Policies -----------------------------*/
3739 * Hash for the output device based upon layer 2 and layer 3 data. If
3740 * the packet is not IP mimic bond_xmit_hash_policy_l2()
3742 static int bond_xmit_hash_policy_l23(struct sk_buff *skb, int count)
3744 struct ethhdr *data = (struct ethhdr *)skb->data;
3745 struct iphdr *iph = ip_hdr(skb);
3747 if (skb->protocol == htons(ETH_P_IP)) {
3748 return ((ntohl(iph->saddr ^ iph->daddr) & 0xffff) ^
3749 (data->h_dest[5] ^ data->h_source[5])) % count;
3752 return (data->h_dest[5] ^ data->h_source[5]) % count;
3756 * Hash for the output device based upon layer 3 and layer 4 data. If
3757 * the packet is a frag or not TCP or UDP, just use layer 3 data. If it is
3758 * altogether not IP, mimic bond_xmit_hash_policy_l2()
3760 static int bond_xmit_hash_policy_l34(struct sk_buff *skb, int count)
3762 struct ethhdr *data = (struct ethhdr *)skb->data;
3763 struct iphdr *iph = ip_hdr(skb);
3764 __be16 *layer4hdr = (__be16 *)((u32 *)iph + iph->ihl);
3767 if (skb->protocol == htons(ETH_P_IP)) {
3768 if (!(iph->frag_off & htons(IP_MF|IP_OFFSET)) &&
3769 (iph->protocol == IPPROTO_TCP ||
3770 iph->protocol == IPPROTO_UDP)) {
3771 layer4_xor = ntohs((*layer4hdr ^ *(layer4hdr + 1)));
3773 return (layer4_xor ^
3774 ((ntohl(iph->saddr ^ iph->daddr)) & 0xffff)) % count;
3778 return (data->h_dest[5] ^ data->h_source[5]) % count;
3782 * Hash for the output device based upon layer 2 data
3784 static int bond_xmit_hash_policy_l2(struct sk_buff *skb, int count)
3786 struct ethhdr *data = (struct ethhdr *)skb->data;
3788 return (data->h_dest[5] ^ data->h_source[5]) % count;
3791 /*-------------------------- Device entry points ----------------------------*/
3793 static int bond_open(struct net_device *bond_dev)
3795 struct bonding *bond = netdev_priv(bond_dev);
3797 bond->kill_timers = 0;
3799 INIT_DELAYED_WORK(&bond->mcast_work, bond_resend_igmp_join_requests_delayed);
3801 if (bond_is_lb(bond)) {
3802 /* bond_alb_initialize must be called before the timer
3805 if (bond_alb_initialize(bond, (bond->params.mode == BOND_MODE_ALB))) {
3806 /* something went wrong - fail the open operation */
3810 INIT_DELAYED_WORK(&bond->alb_work, bond_alb_monitor);
3811 queue_delayed_work(bond->wq, &bond->alb_work, 0);
3814 if (bond->params.miimon) { /* link check interval, in milliseconds. */
3815 INIT_DELAYED_WORK(&bond->mii_work, bond_mii_monitor);
3816 queue_delayed_work(bond->wq, &bond->mii_work, 0);
3819 if (bond->params.arp_interval) { /* arp interval, in milliseconds. */
3820 if (bond->params.mode == BOND_MODE_ACTIVEBACKUP)
3821 INIT_DELAYED_WORK(&bond->arp_work,
3822 bond_activebackup_arp_mon);
3824 INIT_DELAYED_WORK(&bond->arp_work,
3825 bond_loadbalance_arp_mon);
3827 queue_delayed_work(bond->wq, &bond->arp_work, 0);
3828 if (bond->params.arp_validate)
3829 bond_register_arp(bond);
3832 if (bond->params.mode == BOND_MODE_8023AD) {
3833 INIT_DELAYED_WORK(&bond->ad_work, bond_3ad_state_machine_handler);
3834 queue_delayed_work(bond->wq, &bond->ad_work, 0);
3835 /* register to receive LACPDUs */
3836 bond_register_lacpdu(bond);
3837 bond_3ad_initiate_agg_selection(bond, 1);
3843 static int bond_close(struct net_device *bond_dev)
3845 struct bonding *bond = netdev_priv(bond_dev);
3847 if (bond->params.mode == BOND_MODE_8023AD) {
3848 /* Unregister the receive of LACPDUs */
3849 bond_unregister_lacpdu(bond);
3852 if (bond->params.arp_validate)
3853 bond_unregister_arp(bond);
3855 write_lock_bh(&bond->lock);
3857 bond->send_grat_arp = 0;
3858 bond->send_unsol_na = 0;
3860 /* signal timers not to re-arm */
3861 bond->kill_timers = 1;
3863 write_unlock_bh(&bond->lock);
3865 if (bond->params.miimon) { /* link check interval, in milliseconds. */
3866 cancel_delayed_work(&bond->mii_work);
3869 if (bond->params.arp_interval) { /* arp interval, in milliseconds. */
3870 cancel_delayed_work(&bond->arp_work);
3873 switch (bond->params.mode) {
3874 case BOND_MODE_8023AD:
3875 cancel_delayed_work(&bond->ad_work);
3879 cancel_delayed_work(&bond->alb_work);
3885 if (delayed_work_pending(&bond->mcast_work))
3886 cancel_delayed_work(&bond->mcast_work);
3888 if (bond_is_lb(bond)) {
3889 /* Must be called only after all
3890 * slaves have been released
3892 bond_alb_deinitialize(bond);
3898 static struct rtnl_link_stats64 *bond_get_stats(struct net_device *bond_dev,
3899 struct rtnl_link_stats64 *stats)
3901 struct bonding *bond = netdev_priv(bond_dev);
3902 struct rtnl_link_stats64 temp;
3903 struct slave *slave;
3906 memset(stats, 0, sizeof(*stats));
3908 read_lock_bh(&bond->lock);
3910 bond_for_each_slave(bond, slave, i) {
3911 const struct rtnl_link_stats64 *sstats =
3912 dev_get_stats(slave->dev, &temp);
3914 stats->rx_packets += sstats->rx_packets;
3915 stats->rx_bytes += sstats->rx_bytes;
3916 stats->rx_errors += sstats->rx_errors;
3917 stats->rx_dropped += sstats->rx_dropped;
3919 stats->tx_packets += sstats->tx_packets;
3920 stats->tx_bytes += sstats->tx_bytes;
3921 stats->tx_errors += sstats->tx_errors;
3922 stats->tx_dropped += sstats->tx_dropped;
3924 stats->multicast += sstats->multicast;
3925 stats->collisions += sstats->collisions;
3927 stats->rx_length_errors += sstats->rx_length_errors;
3928 stats->rx_over_errors += sstats->rx_over_errors;
3929 stats->rx_crc_errors += sstats->rx_crc_errors;
3930 stats->rx_frame_errors += sstats->rx_frame_errors;
3931 stats->rx_fifo_errors += sstats->rx_fifo_errors;
3932 stats->rx_missed_errors += sstats->rx_missed_errors;
3934 stats->tx_aborted_errors += sstats->tx_aborted_errors;
3935 stats->tx_carrier_errors += sstats->tx_carrier_errors;
3936 stats->tx_fifo_errors += sstats->tx_fifo_errors;
3937 stats->tx_heartbeat_errors += sstats->tx_heartbeat_errors;
3938 stats->tx_window_errors += sstats->tx_window_errors;
3941 read_unlock_bh(&bond->lock);
3946 static int bond_do_ioctl(struct net_device *bond_dev, struct ifreq *ifr, int cmd)
3948 struct net_device *slave_dev = NULL;
3949 struct ifbond k_binfo;
3950 struct ifbond __user *u_binfo = NULL;
3951 struct ifslave k_sinfo;
3952 struct ifslave __user *u_sinfo = NULL;
3953 struct mii_ioctl_data *mii = NULL;
3956 pr_debug("bond_ioctl: master=%s, cmd=%d\n", bond_dev->name, cmd);
3968 * We do this again just in case we were called by SIOCGMIIREG
3969 * instead of SIOCGMIIPHY.
3976 if (mii->reg_num == 1) {
3977 struct bonding *bond = netdev_priv(bond_dev);
3979 read_lock(&bond->lock);
3980 read_lock(&bond->curr_slave_lock);
3981 if (netif_carrier_ok(bond->dev))
3982 mii->val_out = BMSR_LSTATUS;
3984 read_unlock(&bond->curr_slave_lock);
3985 read_unlock(&bond->lock);
3989 case BOND_INFO_QUERY_OLD:
3990 case SIOCBONDINFOQUERY:
3991 u_binfo = (struct ifbond __user *)ifr->ifr_data;
3993 if (copy_from_user(&k_binfo, u_binfo, sizeof(ifbond)))
3996 res = bond_info_query(bond_dev, &k_binfo);
3998 copy_to_user(u_binfo, &k_binfo, sizeof(ifbond)))
4002 case BOND_SLAVE_INFO_QUERY_OLD:
4003 case SIOCBONDSLAVEINFOQUERY:
4004 u_sinfo = (struct ifslave __user *)ifr->ifr_data;
4006 if (copy_from_user(&k_sinfo, u_sinfo, sizeof(ifslave)))
4009 res = bond_slave_info_query(bond_dev, &k_sinfo);
4011 copy_to_user(u_sinfo, &k_sinfo, sizeof(ifslave)))
4020 if (!capable(CAP_NET_ADMIN))
4023 slave_dev = dev_get_by_name(dev_net(bond_dev), ifr->ifr_slave);
4025 pr_debug("slave_dev=%p:\n", slave_dev);
4030 pr_debug("slave_dev->name=%s:\n", slave_dev->name);
4032 case BOND_ENSLAVE_OLD:
4033 case SIOCBONDENSLAVE:
4034 res = bond_enslave(bond_dev, slave_dev);
4036 case BOND_RELEASE_OLD:
4037 case SIOCBONDRELEASE:
4038 res = bond_release(bond_dev, slave_dev);
4040 case BOND_SETHWADDR_OLD:
4041 case SIOCBONDSETHWADDR:
4042 res = bond_sethwaddr(bond_dev, slave_dev);
4044 case BOND_CHANGE_ACTIVE_OLD:
4045 case SIOCBONDCHANGEACTIVE:
4046 res = bond_ioctl_change_active(bond_dev, slave_dev);
4058 static bool bond_addr_in_mc_list(unsigned char *addr,
4059 struct netdev_hw_addr_list *list,
4062 struct netdev_hw_addr *ha;
4064 netdev_hw_addr_list_for_each(ha, list)
4065 if (!memcmp(ha->addr, addr, addrlen))
4071 static void bond_set_multicast_list(struct net_device *bond_dev)
4073 struct bonding *bond = netdev_priv(bond_dev);
4074 struct netdev_hw_addr *ha;
4078 * Do promisc before checking multicast_mode
4080 if ((bond_dev->flags & IFF_PROMISC) && !(bond->flags & IFF_PROMISC))
4082 * FIXME: Need to handle the error when one of the multi-slaves
4085 bond_set_promiscuity(bond, 1);
4088 if (!(bond_dev->flags & IFF_PROMISC) && (bond->flags & IFF_PROMISC))
4089 bond_set_promiscuity(bond, -1);
4092 /* set allmulti flag to slaves */
4093 if ((bond_dev->flags & IFF_ALLMULTI) && !(bond->flags & IFF_ALLMULTI))
4095 * FIXME: Need to handle the error when one of the multi-slaves
4098 bond_set_allmulti(bond, 1);
4101 if (!(bond_dev->flags & IFF_ALLMULTI) && (bond->flags & IFF_ALLMULTI))
4102 bond_set_allmulti(bond, -1);
4105 read_lock(&bond->lock);
4107 bond->flags = bond_dev->flags;
4109 /* looking for addresses to add to slaves' mc list */
4110 netdev_for_each_mc_addr(ha, bond_dev) {
4111 found = bond_addr_in_mc_list(ha->addr, &bond->mc_list,
4112 bond_dev->addr_len);
4114 bond_mc_add(bond, ha->addr);
4117 /* looking for addresses to delete from slaves' list */
4118 netdev_hw_addr_list_for_each(ha, &bond->mc_list) {
4119 found = bond_addr_in_mc_list(ha->addr, &bond_dev->mc,
4120 bond_dev->addr_len);
4122 bond_mc_del(bond, ha->addr);
4125 /* save master's multicast list */
4126 __hw_addr_flush(&bond->mc_list);
4127 __hw_addr_add_multiple(&bond->mc_list, &bond_dev->mc,
4128 bond_dev->addr_len, NETDEV_HW_ADDR_T_MULTICAST);
4130 read_unlock(&bond->lock);
4133 static int bond_neigh_setup(struct net_device *dev, struct neigh_parms *parms)
4135 struct bonding *bond = netdev_priv(dev);
4136 struct slave *slave = bond->first_slave;
4139 const struct net_device_ops *slave_ops
4140 = slave->dev->netdev_ops;
4141 if (slave_ops->ndo_neigh_setup)
4142 return slave_ops->ndo_neigh_setup(slave->dev, parms);
4148 * Change the MTU of all of a master's slaves to match the master
4150 static int bond_change_mtu(struct net_device *bond_dev, int new_mtu)
4152 struct bonding *bond = netdev_priv(bond_dev);
4153 struct slave *slave, *stop_at;
4157 pr_debug("bond=%p, name=%s, new_mtu=%d\n", bond,
4158 (bond_dev ? bond_dev->name : "None"), new_mtu);
4160 /* Can't hold bond->lock with bh disabled here since
4161 * some base drivers panic. On the other hand we can't
4162 * hold bond->lock without bh disabled because we'll
4163 * deadlock. The only solution is to rely on the fact
4164 * that we're under rtnl_lock here, and the slaves
4165 * list won't change. This doesn't solve the problem
4166 * of setting the slave's MTU while it is
4167 * transmitting, but the assumption is that the base
4168 * driver can handle that.
4170 * TODO: figure out a way to safely iterate the slaves
4171 * list, but without holding a lock around the actual
4172 * call to the base driver.
4175 bond_for_each_slave(bond, slave, i) {
4176 pr_debug("s %p s->p %p c_m %p\n",
4179 slave->dev->netdev_ops->ndo_change_mtu);
4181 res = dev_set_mtu(slave->dev, new_mtu);
4184 /* If we failed to set the slave's mtu to the new value
4185 * we must abort the operation even in ACTIVE_BACKUP
4186 * mode, because if we allow the backup slaves to have
4187 * different mtu values than the active slave we'll
4188 * need to change their mtu when doing a failover. That
4189 * means changing their mtu from timer context, which
4190 * is probably not a good idea.
4192 pr_debug("err %d %s\n", res, slave->dev->name);
4197 bond_dev->mtu = new_mtu;
4202 /* unwind from head to the slave that failed */
4204 bond_for_each_slave_from_to(bond, slave, i, bond->first_slave, stop_at) {
4207 tmp_res = dev_set_mtu(slave->dev, bond_dev->mtu);
4209 pr_debug("unwind err %d dev %s\n",
4210 tmp_res, slave->dev->name);
4220 * Note that many devices must be down to change the HW address, and
4221 * downing the master releases all slaves. We can make bonds full of
4222 * bonding devices to test this, however.
4224 static int bond_set_mac_address(struct net_device *bond_dev, void *addr)
4226 struct bonding *bond = netdev_priv(bond_dev);
4227 struct sockaddr *sa = addr, tmp_sa;
4228 struct slave *slave, *stop_at;
4232 if (bond->params.mode == BOND_MODE_ALB)
4233 return bond_alb_set_mac_address(bond_dev, addr);
4236 pr_debug("bond=%p, name=%s\n",
4237 bond, bond_dev ? bond_dev->name : "None");
4240 * If fail_over_mac is set to active, do nothing and return
4241 * success. Returning an error causes ifenslave to fail.
4243 if (bond->params.fail_over_mac == BOND_FOM_ACTIVE)
4246 if (!is_valid_ether_addr(sa->sa_data))
4247 return -EADDRNOTAVAIL;
4249 /* Can't hold bond->lock with bh disabled here since
4250 * some base drivers panic. On the other hand we can't
4251 * hold bond->lock without bh disabled because we'll
4252 * deadlock. The only solution is to rely on the fact
4253 * that we're under rtnl_lock here, and the slaves
4254 * list won't change. This doesn't solve the problem
4255 * of setting the slave's hw address while it is
4256 * transmitting, but the assumption is that the base
4257 * driver can handle that.
4259 * TODO: figure out a way to safely iterate the slaves
4260 * list, but without holding a lock around the actual
4261 * call to the base driver.
4264 bond_for_each_slave(bond, slave, i) {
4265 const struct net_device_ops *slave_ops = slave->dev->netdev_ops;
4266 pr_debug("slave %p %s\n", slave, slave->dev->name);
4268 if (slave_ops->ndo_set_mac_address == NULL) {
4270 pr_debug("EOPNOTSUPP %s\n", slave->dev->name);
4274 res = dev_set_mac_address(slave->dev, addr);
4276 /* TODO: consider downing the slave
4278 * User should expect communications
4279 * breakage anyway until ARP finish
4282 pr_debug("err %d %s\n", res, slave->dev->name);
4288 memcpy(bond_dev->dev_addr, sa->sa_data, bond_dev->addr_len);
4292 memcpy(tmp_sa.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
4293 tmp_sa.sa_family = bond_dev->type;
4295 /* unwind from head to the slave that failed */
4297 bond_for_each_slave_from_to(bond, slave, i, bond->first_slave, stop_at) {
4300 tmp_res = dev_set_mac_address(slave->dev, &tmp_sa);
4302 pr_debug("unwind err %d dev %s\n",
4303 tmp_res, slave->dev->name);
4310 static int bond_xmit_roundrobin(struct sk_buff *skb, struct net_device *bond_dev)
4312 struct bonding *bond = netdev_priv(bond_dev);
4313 struct slave *slave, *start_at;
4314 int i, slave_no, res = 1;
4315 struct iphdr *iph = ip_hdr(skb);
4317 read_lock(&bond->lock);
4319 if (!BOND_IS_OK(bond))
4322 * Start with the curr_active_slave that joined the bond as the
4323 * default for sending IGMP traffic. For failover purposes one
4324 * needs to maintain some consistency for the interface that will
4325 * send the join/membership reports. The curr_active_slave found
4326 * will send all of this type of traffic.
4328 if ((iph->protocol == IPPROTO_IGMP) &&
4329 (skb->protocol == htons(ETH_P_IP))) {
4331 read_lock(&bond->curr_slave_lock);
4332 slave = bond->curr_active_slave;
4333 read_unlock(&bond->curr_slave_lock);
4339 * Concurrent TX may collide on rr_tx_counter; we accept
4340 * that as being rare enough not to justify using an
4343 slave_no = bond->rr_tx_counter++ % bond->slave_cnt;
4345 bond_for_each_slave(bond, slave, i) {
4353 bond_for_each_slave_from(bond, slave, i, start_at) {
4354 if (IS_UP(slave->dev) &&
4355 (slave->link == BOND_LINK_UP) &&
4356 (slave->state == BOND_STATE_ACTIVE)) {
4357 res = bond_dev_queue_xmit(bond, skb, slave->dev);
4364 /* no suitable interface, frame not sent */
4367 read_unlock(&bond->lock);
4368 return NETDEV_TX_OK;
4373 * in active-backup mode, we know that bond->curr_active_slave is always valid if
4374 * the bond has a usable interface.
4376 static int bond_xmit_activebackup(struct sk_buff *skb, struct net_device *bond_dev)
4378 struct bonding *bond = netdev_priv(bond_dev);
4381 read_lock(&bond->lock);
4382 read_lock(&bond->curr_slave_lock);
4384 if (!BOND_IS_OK(bond))
4387 if (!bond->curr_active_slave)
4390 res = bond_dev_queue_xmit(bond, skb, bond->curr_active_slave->dev);
4394 /* no suitable interface, frame not sent */
4397 read_unlock(&bond->curr_slave_lock);
4398 read_unlock(&bond->lock);
4399 return NETDEV_TX_OK;
4403 * In bond_xmit_xor() , we determine the output device by using a pre-
4404 * determined xmit_hash_policy(), If the selected device is not enabled,
4405 * find the next active slave.
4407 static int bond_xmit_xor(struct sk_buff *skb, struct net_device *bond_dev)
4409 struct bonding *bond = netdev_priv(bond_dev);
4410 struct slave *slave, *start_at;
4415 read_lock(&bond->lock);
4417 if (!BOND_IS_OK(bond))
4420 slave_no = bond->xmit_hash_policy(skb, bond->slave_cnt);
4422 bond_for_each_slave(bond, slave, i) {
4430 bond_for_each_slave_from(bond, slave, i, start_at) {
4431 if (IS_UP(slave->dev) &&
4432 (slave->link == BOND_LINK_UP) &&
4433 (slave->state == BOND_STATE_ACTIVE)) {
4434 res = bond_dev_queue_xmit(bond, skb, slave->dev);
4441 /* no suitable interface, frame not sent */
4444 read_unlock(&bond->lock);
4445 return NETDEV_TX_OK;
4449 * in broadcast mode, we send everything to all usable interfaces.
4451 static int bond_xmit_broadcast(struct sk_buff *skb, struct net_device *bond_dev)
4453 struct bonding *bond = netdev_priv(bond_dev);
4454 struct slave *slave, *start_at;
4455 struct net_device *tx_dev = NULL;
4459 read_lock(&bond->lock);
4461 if (!BOND_IS_OK(bond))
4464 read_lock(&bond->curr_slave_lock);
4465 start_at = bond->curr_active_slave;
4466 read_unlock(&bond->curr_slave_lock);
4471 bond_for_each_slave_from(bond, slave, i, start_at) {
4472 if (IS_UP(slave->dev) &&
4473 (slave->link == BOND_LINK_UP) &&
4474 (slave->state == BOND_STATE_ACTIVE)) {
4476 struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
4478 pr_err("%s: Error: bond_xmit_broadcast(): skb_clone() failed\n",
4483 res = bond_dev_queue_xmit(bond, skb2, tx_dev);
4485 dev_kfree_skb(skb2);
4489 tx_dev = slave->dev;
4494 res = bond_dev_queue_xmit(bond, skb, tx_dev);
4498 /* no suitable interface, frame not sent */
4501 /* frame sent to all suitable interfaces */
4502 read_unlock(&bond->lock);
4503 return NETDEV_TX_OK;
4506 /*------------------------- Device initialization ---------------------------*/
4508 static void bond_set_xmit_hash_policy(struct bonding *bond)
4510 switch (bond->params.xmit_policy) {
4511 case BOND_XMIT_POLICY_LAYER23:
4512 bond->xmit_hash_policy = bond_xmit_hash_policy_l23;
4514 case BOND_XMIT_POLICY_LAYER34:
4515 bond->xmit_hash_policy = bond_xmit_hash_policy_l34;
4517 case BOND_XMIT_POLICY_LAYER2:
4519 bond->xmit_hash_policy = bond_xmit_hash_policy_l2;
4525 * Lookup the slave that corresponds to a qid
4527 static inline int bond_slave_override(struct bonding *bond,
4528 struct sk_buff *skb)
4531 struct slave *slave = NULL;
4532 struct slave *check_slave;
4534 read_lock(&bond->lock);
4536 if (!BOND_IS_OK(bond) || !skb->queue_mapping)
4539 /* Find out if any slaves have the same mapping as this skb. */
4540 bond_for_each_slave(bond, check_slave, i) {
4541 if (check_slave->queue_id == skb->queue_mapping) {
4542 slave = check_slave;
4547 /* If the slave isn't UP, use default transmit policy. */
4548 if (slave && slave->queue_id && IS_UP(slave->dev) &&
4549 (slave->link == BOND_LINK_UP)) {
4550 res = bond_dev_queue_xmit(bond, skb, slave->dev);
4554 read_unlock(&bond->lock);
4558 static u16 bond_select_queue(struct net_device *dev, struct sk_buff *skb)
4561 * This helper function exists to help dev_pick_tx get the correct
4562 * destination queue. Using a helper function skips the a call to
4563 * skb_tx_hash and will put the skbs in the queue we expect on their
4564 * way down to the bonding driver.
4566 return skb->queue_mapping;
4569 static netdev_tx_t bond_start_xmit(struct sk_buff *skb, struct net_device *dev)
4571 struct bonding *bond = netdev_priv(dev);
4574 * If we risk deadlock from transmitting this in the
4575 * netpoll path, tell netpoll to queue the frame for later tx
4577 if (is_netpoll_tx_blocked(dev))
4578 return NETDEV_TX_BUSY;
4580 if (TX_QUEUE_OVERRIDE(bond->params.mode)) {
4581 if (!bond_slave_override(bond, skb))
4582 return NETDEV_TX_OK;
4585 switch (bond->params.mode) {
4586 case BOND_MODE_ROUNDROBIN:
4587 return bond_xmit_roundrobin(skb, dev);
4588 case BOND_MODE_ACTIVEBACKUP:
4589 return bond_xmit_activebackup(skb, dev);
4591 return bond_xmit_xor(skb, dev);
4592 case BOND_MODE_BROADCAST:
4593 return bond_xmit_broadcast(skb, dev);
4594 case BOND_MODE_8023AD:
4595 return bond_3ad_xmit_xor(skb, dev);
4598 return bond_alb_xmit(skb, dev);
4600 /* Should never happen, mode already checked */
4601 pr_err("%s: Error: Unknown bonding mode %d\n",
4602 dev->name, bond->params.mode);
4605 return NETDEV_TX_OK;
4611 * set bond mode specific net device operations
4613 void bond_set_mode_ops(struct bonding *bond, int mode)
4615 struct net_device *bond_dev = bond->dev;
4618 case BOND_MODE_ROUNDROBIN:
4620 case BOND_MODE_ACTIVEBACKUP:
4623 bond_set_xmit_hash_policy(bond);
4625 case BOND_MODE_BROADCAST:
4627 case BOND_MODE_8023AD:
4628 bond_set_master_3ad_flags(bond);
4629 bond_set_xmit_hash_policy(bond);
4632 bond_set_master_alb_flags(bond);
4637 /* Should never happen, mode already checked */
4638 pr_err("%s: Error: Unknown bonding mode %d\n",
4639 bond_dev->name, mode);
4644 static void bond_ethtool_get_drvinfo(struct net_device *bond_dev,
4645 struct ethtool_drvinfo *drvinfo)
4647 strncpy(drvinfo->driver, DRV_NAME, 32);
4648 strncpy(drvinfo->version, DRV_VERSION, 32);
4649 snprintf(drvinfo->fw_version, 32, "%d", BOND_ABI_VERSION);
4652 static const struct ethtool_ops bond_ethtool_ops = {
4653 .get_drvinfo = bond_ethtool_get_drvinfo,
4654 .get_link = ethtool_op_get_link,
4655 .get_tx_csum = ethtool_op_get_tx_csum,
4656 .get_sg = ethtool_op_get_sg,
4657 .get_tso = ethtool_op_get_tso,
4658 .get_ufo = ethtool_op_get_ufo,
4659 .get_flags = ethtool_op_get_flags,
4662 static const struct net_device_ops bond_netdev_ops = {
4663 .ndo_init = bond_init,
4664 .ndo_uninit = bond_uninit,
4665 .ndo_open = bond_open,
4666 .ndo_stop = bond_close,
4667 .ndo_start_xmit = bond_start_xmit,
4668 .ndo_select_queue = bond_select_queue,
4669 .ndo_get_stats64 = bond_get_stats,
4670 .ndo_do_ioctl = bond_do_ioctl,
4671 .ndo_set_multicast_list = bond_set_multicast_list,
4672 .ndo_change_mtu = bond_change_mtu,
4673 .ndo_set_mac_address = bond_set_mac_address,
4674 .ndo_neigh_setup = bond_neigh_setup,
4675 .ndo_vlan_rx_register = bond_vlan_rx_register,
4676 .ndo_vlan_rx_add_vid = bond_vlan_rx_add_vid,
4677 .ndo_vlan_rx_kill_vid = bond_vlan_rx_kill_vid,
4678 #ifdef CONFIG_NET_POLL_CONTROLLER
4679 .ndo_netpoll_cleanup = bond_netpoll_cleanup,
4680 .ndo_poll_controller = bond_poll_controller,
4684 static void bond_destructor(struct net_device *bond_dev)
4686 struct bonding *bond = netdev_priv(bond_dev);
4688 destroy_workqueue(bond->wq);
4689 free_netdev(bond_dev);
4692 static void bond_setup(struct net_device *bond_dev)
4694 struct bonding *bond = netdev_priv(bond_dev);
4696 /* initialize rwlocks */
4697 rwlock_init(&bond->lock);
4698 rwlock_init(&bond->curr_slave_lock);
4700 bond->params = bonding_defaults;
4702 /* Initialize pointers */
4703 bond->dev = bond_dev;
4704 INIT_LIST_HEAD(&bond->vlan_list);
4706 /* Initialize the device entry points */
4707 ether_setup(bond_dev);
4708 bond_dev->netdev_ops = &bond_netdev_ops;
4709 bond_dev->ethtool_ops = &bond_ethtool_ops;
4710 bond_set_mode_ops(bond, bond->params.mode);
4712 bond_dev->destructor = bond_destructor;
4714 /* Initialize the device options */
4715 bond_dev->tx_queue_len = 0;
4716 bond_dev->flags |= IFF_MASTER|IFF_MULTICAST;
4717 bond_dev->priv_flags |= IFF_BONDING;
4718 bond_dev->priv_flags &= ~IFF_XMIT_DST_RELEASE;
4720 if (bond->params.arp_interval)
4721 bond_dev->priv_flags |= IFF_MASTER_ARPMON;
4723 /* At first, we block adding VLANs. That's the only way to
4724 * prevent problems that occur when adding VLANs over an
4725 * empty bond. The block will be removed once non-challenged
4726 * slaves are enslaved.
4728 bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
4730 /* don't acquire bond device's netif_tx_lock when
4732 bond_dev->features |= NETIF_F_LLTX;
4734 /* By default, we declare the bond to be fully
4735 * VLAN hardware accelerated capable. Special
4736 * care is taken in the various xmit functions
4737 * when there are slaves that are not hw accel
4740 bond_dev->features |= (NETIF_F_HW_VLAN_TX |
4741 NETIF_F_HW_VLAN_RX |
4742 NETIF_F_HW_VLAN_FILTER);
4744 /* By default, we enable GRO on bonding devices.
4745 * Actual support requires lowlevel drivers are GRO ready.
4747 bond_dev->features |= NETIF_F_GRO;
4750 static void bond_work_cancel_all(struct bonding *bond)
4752 write_lock_bh(&bond->lock);
4753 bond->kill_timers = 1;
4754 write_unlock_bh(&bond->lock);
4756 if (bond->params.miimon && delayed_work_pending(&bond->mii_work))
4757 cancel_delayed_work(&bond->mii_work);
4759 if (bond->params.arp_interval && delayed_work_pending(&bond->arp_work))
4760 cancel_delayed_work(&bond->arp_work);
4762 if (bond->params.mode == BOND_MODE_ALB &&
4763 delayed_work_pending(&bond->alb_work))
4764 cancel_delayed_work(&bond->alb_work);
4766 if (bond->params.mode == BOND_MODE_8023AD &&
4767 delayed_work_pending(&bond->ad_work))
4768 cancel_delayed_work(&bond->ad_work);
4770 if (delayed_work_pending(&bond->mcast_work))
4771 cancel_delayed_work(&bond->mcast_work);
4775 * Destroy a bonding device.
4776 * Must be under rtnl_lock when this function is called.
4778 static void bond_uninit(struct net_device *bond_dev)
4780 struct bonding *bond = netdev_priv(bond_dev);
4781 struct vlan_entry *vlan, *tmp;
4783 bond_netpoll_cleanup(bond_dev);
4785 /* Release the bonded slaves */
4786 bond_release_all(bond_dev);
4788 list_del(&bond->bond_list);
4790 bond_work_cancel_all(bond);
4792 bond_remove_proc_entry(bond);
4794 __hw_addr_flush(&bond->mc_list);
4796 list_for_each_entry_safe(vlan, tmp, &bond->vlan_list, vlan_list) {
4797 list_del(&vlan->vlan_list);
4802 /*------------------------- Module initialization ---------------------------*/
4805 * Convert string input module parms. Accept either the
4806 * number of the mode or its string name. A bit complicated because
4807 * some mode names are substrings of other names, and calls from sysfs
4808 * may have whitespace in the name (trailing newlines, for example).
4810 int bond_parse_parm(const char *buf, const struct bond_parm_tbl *tbl)
4812 int modeint = -1, i, rv;
4813 char *p, modestr[BOND_MAX_MODENAME_LEN + 1] = { 0, };
4815 for (p = (char *)buf; *p; p++)
4816 if (!(isdigit(*p) || isspace(*p)))
4820 rv = sscanf(buf, "%20s", modestr);
4822 rv = sscanf(buf, "%d", &modeint);
4827 for (i = 0; tbl[i].modename; i++) {
4828 if (modeint == tbl[i].mode)
4830 if (strcmp(modestr, tbl[i].modename) == 0)
4837 static int bond_check_params(struct bond_params *params)
4839 int arp_validate_value, fail_over_mac_value, primary_reselect_value;
4842 * Convert string parameters.
4845 bond_mode = bond_parse_parm(mode, bond_mode_tbl);
4846 if (bond_mode == -1) {
4847 pr_err("Error: Invalid bonding mode \"%s\"\n",
4848 mode == NULL ? "NULL" : mode);
4853 if (xmit_hash_policy) {
4854 if ((bond_mode != BOND_MODE_XOR) &&
4855 (bond_mode != BOND_MODE_8023AD)) {
4856 pr_info("xmit_hash_policy param is irrelevant in mode %s\n",
4857 bond_mode_name(bond_mode));
4859 xmit_hashtype = bond_parse_parm(xmit_hash_policy,
4861 if (xmit_hashtype == -1) {
4862 pr_err("Error: Invalid xmit_hash_policy \"%s\"\n",
4863 xmit_hash_policy == NULL ? "NULL" :
4871 if (bond_mode != BOND_MODE_8023AD) {
4872 pr_info("lacp_rate param is irrelevant in mode %s\n",
4873 bond_mode_name(bond_mode));
4875 lacp_fast = bond_parse_parm(lacp_rate, bond_lacp_tbl);
4876 if (lacp_fast == -1) {
4877 pr_err("Error: Invalid lacp rate \"%s\"\n",
4878 lacp_rate == NULL ? "NULL" : lacp_rate);
4885 params->ad_select = bond_parse_parm(ad_select, ad_select_tbl);
4886 if (params->ad_select == -1) {
4887 pr_err("Error: Invalid ad_select \"%s\"\n",
4888 ad_select == NULL ? "NULL" : ad_select);
4892 if (bond_mode != BOND_MODE_8023AD) {
4893 pr_warning("ad_select param only affects 802.3ad mode\n");
4896 params->ad_select = BOND_AD_STABLE;
4899 if (max_bonds < 0) {
4900 pr_warning("Warning: max_bonds (%d) not in range %d-%d, so it was reset to BOND_DEFAULT_MAX_BONDS (%d)\n",
4901 max_bonds, 0, INT_MAX, BOND_DEFAULT_MAX_BONDS);
4902 max_bonds = BOND_DEFAULT_MAX_BONDS;
4906 pr_warning("Warning: miimon module parameter (%d), not in range 0-%d, so it was reset to %d\n",
4907 miimon, INT_MAX, BOND_LINK_MON_INTERV);
4908 miimon = BOND_LINK_MON_INTERV;
4912 pr_warning("Warning: updelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4917 if (downdelay < 0) {
4918 pr_warning("Warning: downdelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4919 downdelay, INT_MAX);
4923 if ((use_carrier != 0) && (use_carrier != 1)) {
4924 pr_warning("Warning: use_carrier module parameter (%d), not of valid value (0/1), so it was set to 1\n",
4929 if (num_grat_arp < 0 || num_grat_arp > 255) {
4930 pr_warning("Warning: num_grat_arp (%d) not in range 0-255 so it was reset to 1\n",
4935 if (num_unsol_na < 0 || num_unsol_na > 255) {
4936 pr_warning("Warning: num_unsol_na (%d) not in range 0-255 so it was reset to 1\n",
4941 /* reset values for 802.3ad */
4942 if (bond_mode == BOND_MODE_8023AD) {
4944 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");
4945 pr_warning("Forcing miimon to 100msec\n");
4950 if (tx_queues < 1 || tx_queues > 255) {
4951 pr_warning("Warning: tx_queues (%d) should be between "
4952 "1 and 255, resetting to %d\n",
4953 tx_queues, BOND_DEFAULT_TX_QUEUES);
4954 tx_queues = BOND_DEFAULT_TX_QUEUES;
4957 if ((all_slaves_active != 0) && (all_slaves_active != 1)) {
4958 pr_warning("Warning: all_slaves_active module parameter (%d), "
4959 "not of valid value (0/1), so it was set to "
4960 "0\n", all_slaves_active);
4961 all_slaves_active = 0;
4964 if (resend_igmp < 0 || resend_igmp > 255) {
4965 pr_warning("Warning: resend_igmp (%d) should be between "
4966 "0 and 255, resetting to %d\n",
4967 resend_igmp, BOND_DEFAULT_RESEND_IGMP);
4968 resend_igmp = BOND_DEFAULT_RESEND_IGMP;
4971 /* reset values for TLB/ALB */
4972 if ((bond_mode == BOND_MODE_TLB) ||
4973 (bond_mode == BOND_MODE_ALB)) {
4975 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");
4976 pr_warning("Forcing miimon to 100msec\n");
4981 if (bond_mode == BOND_MODE_ALB) {
4982 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",
4987 if (updelay || downdelay) {
4988 /* just warn the user the up/down delay will have
4989 * no effect since miimon is zero...
4991 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",
4992 updelay, downdelay);
4995 /* don't allow arp monitoring */
4997 pr_warning("Warning: miimon (%d) and arp_interval (%d) can't be used simultaneously, disabling ARP monitoring\n",
4998 miimon, arp_interval);
5002 if ((updelay % miimon) != 0) {
5003 pr_warning("Warning: updelay (%d) is not a multiple of miimon (%d), updelay rounded to %d ms\n",
5005 (updelay / miimon) * miimon);
5010 if ((downdelay % miimon) != 0) {
5011 pr_warning("Warning: downdelay (%d) is not a multiple of miimon (%d), downdelay rounded to %d ms\n",
5013 (downdelay / miimon) * miimon);
5016 downdelay /= miimon;
5019 if (arp_interval < 0) {
5020 pr_warning("Warning: arp_interval module parameter (%d) , not in range 0-%d, so it was reset to %d\n",
5021 arp_interval, INT_MAX, BOND_LINK_ARP_INTERV);
5022 arp_interval = BOND_LINK_ARP_INTERV;
5025 for (arp_ip_count = 0;
5026 (arp_ip_count < BOND_MAX_ARP_TARGETS) && arp_ip_target[arp_ip_count];
5028 /* not complete check, but should be good enough to
5030 if (!isdigit(arp_ip_target[arp_ip_count][0])) {
5031 pr_warning("Warning: bad arp_ip_target module parameter (%s), ARP monitoring will not be performed\n",
5032 arp_ip_target[arp_ip_count]);
5035 __be32 ip = in_aton(arp_ip_target[arp_ip_count]);
5036 arp_target[arp_ip_count] = ip;
5040 if (arp_interval && !arp_ip_count) {
5041 /* don't allow arping if no arp_ip_target given... */
5042 pr_warning("Warning: arp_interval module parameter (%d) specified without providing an arp_ip_target parameter, arp_interval was reset to 0\n",
5048 if (bond_mode != BOND_MODE_ACTIVEBACKUP) {
5049 pr_err("arp_validate only supported in active-backup mode\n");
5052 if (!arp_interval) {
5053 pr_err("arp_validate requires arp_interval\n");
5057 arp_validate_value = bond_parse_parm(arp_validate,
5059 if (arp_validate_value == -1) {
5060 pr_err("Error: invalid arp_validate \"%s\"\n",
5061 arp_validate == NULL ? "NULL" : arp_validate);
5065 arp_validate_value = 0;
5068 pr_info("MII link monitoring set to %d ms\n", miimon);
5069 } else if (arp_interval) {
5072 pr_info("ARP monitoring set to %d ms, validate %s, with %d target(s):",
5074 arp_validate_tbl[arp_validate_value].modename,
5077 for (i = 0; i < arp_ip_count; i++)
5078 pr_info(" %s", arp_ip_target[i]);
5082 } else if (max_bonds) {
5083 /* miimon and arp_interval not set, we need one so things
5084 * work as expected, see bonding.txt for details
5086 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");
5089 if (primary && !USES_PRIMARY(bond_mode)) {
5090 /* currently, using a primary only makes sense
5091 * in active backup, TLB or ALB modes
5093 pr_warning("Warning: %s primary device specified but has no effect in %s mode\n",
5094 primary, bond_mode_name(bond_mode));
5098 if (primary && primary_reselect) {
5099 primary_reselect_value = bond_parse_parm(primary_reselect,
5101 if (primary_reselect_value == -1) {
5102 pr_err("Error: Invalid primary_reselect \"%s\"\n",
5104 NULL ? "NULL" : primary_reselect);
5108 primary_reselect_value = BOND_PRI_RESELECT_ALWAYS;
5111 if (fail_over_mac) {
5112 fail_over_mac_value = bond_parse_parm(fail_over_mac,
5114 if (fail_over_mac_value == -1) {
5115 pr_err("Error: invalid fail_over_mac \"%s\"\n",
5116 arp_validate == NULL ? "NULL" : arp_validate);
5120 if (bond_mode != BOND_MODE_ACTIVEBACKUP)
5121 pr_warning("Warning: fail_over_mac only affects active-backup mode.\n");
5123 fail_over_mac_value = BOND_FOM_NONE;
5126 /* fill params struct with the proper values */
5127 params->mode = bond_mode;
5128 params->xmit_policy = xmit_hashtype;
5129 params->miimon = miimon;
5130 params->num_grat_arp = num_grat_arp;
5131 params->num_unsol_na = num_unsol_na;
5132 params->arp_interval = arp_interval;
5133 params->arp_validate = arp_validate_value;
5134 params->updelay = updelay;
5135 params->downdelay = downdelay;
5136 params->use_carrier = use_carrier;
5137 params->lacp_fast = lacp_fast;
5138 params->primary[0] = 0;
5139 params->primary_reselect = primary_reselect_value;
5140 params->fail_over_mac = fail_over_mac_value;
5141 params->tx_queues = tx_queues;
5142 params->all_slaves_active = all_slaves_active;
5143 params->resend_igmp = resend_igmp;
5146 strncpy(params->primary, primary, IFNAMSIZ);
5147 params->primary[IFNAMSIZ - 1] = 0;
5150 memcpy(params->arp_targets, arp_target, sizeof(arp_target));
5155 static struct lock_class_key bonding_netdev_xmit_lock_key;
5156 static struct lock_class_key bonding_netdev_addr_lock_key;
5158 static void bond_set_lockdep_class_one(struct net_device *dev,
5159 struct netdev_queue *txq,
5162 lockdep_set_class(&txq->_xmit_lock,
5163 &bonding_netdev_xmit_lock_key);
5166 static void bond_set_lockdep_class(struct net_device *dev)
5168 lockdep_set_class(&dev->addr_list_lock,
5169 &bonding_netdev_addr_lock_key);
5170 netdev_for_each_tx_queue(dev, bond_set_lockdep_class_one, NULL);
5174 * Called from registration process
5176 static int bond_init(struct net_device *bond_dev)
5178 struct bonding *bond = netdev_priv(bond_dev);
5179 struct bond_net *bn = net_generic(dev_net(bond_dev), bond_net_id);
5181 pr_debug("Begin bond_init for %s\n", bond_dev->name);
5183 bond->wq = create_singlethread_workqueue(bond_dev->name);
5187 bond_set_lockdep_class(bond_dev);
5189 netif_carrier_off(bond_dev);
5191 bond_create_proc_entry(bond);
5192 list_add_tail(&bond->bond_list, &bn->dev_list);
5194 bond_prepare_sysfs_group(bond);
5196 __hw_addr_init(&bond->mc_list);
5200 static int bond_validate(struct nlattr *tb[], struct nlattr *data[])
5202 if (tb[IFLA_ADDRESS]) {
5203 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN)
5205 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS])))
5206 return -EADDRNOTAVAIL;
5211 static struct rtnl_link_ops bond_link_ops __read_mostly = {
5213 .priv_size = sizeof(struct bonding),
5214 .setup = bond_setup,
5215 .validate = bond_validate,
5218 /* Create a new bond based on the specified name and bonding parameters.
5219 * If name is NULL, obtain a suitable "bond%d" name for us.
5220 * Caller must NOT hold rtnl_lock; we need to release it here before we
5221 * set up our sysfs entries.
5223 int bond_create(struct net *net, const char *name)
5225 struct net_device *bond_dev;
5230 bond_dev = alloc_netdev_mq(sizeof(struct bonding), name ? name : "",
5231 bond_setup, tx_queues);
5233 pr_err("%s: eek! can't alloc netdev!\n", name);
5238 dev_net_set(bond_dev, net);
5239 bond_dev->rtnl_link_ops = &bond_link_ops;
5242 res = dev_alloc_name(bond_dev, "bond%d");
5247 * If we're given a name to register
5248 * we need to ensure that its not already
5252 if (__dev_get_by_name(net, name) != NULL)
5256 res = register_netdevice(bond_dev);
5261 bond_destructor(bond_dev);
5265 static int __net_init bond_net_init(struct net *net)
5267 struct bond_net *bn = net_generic(net, bond_net_id);
5270 INIT_LIST_HEAD(&bn->dev_list);
5272 bond_create_proc_dir(bn);
5277 static void __net_exit bond_net_exit(struct net *net)
5279 struct bond_net *bn = net_generic(net, bond_net_id);
5281 bond_destroy_proc_dir(bn);
5284 static struct pernet_operations bond_net_ops = {
5285 .init = bond_net_init,
5286 .exit = bond_net_exit,
5288 .size = sizeof(struct bond_net),
5291 static int __init bonding_init(void)
5296 pr_info("%s", version);
5298 res = bond_check_params(&bonding_defaults);
5302 res = register_pernet_subsys(&bond_net_ops);
5306 res = rtnl_link_register(&bond_link_ops);
5310 for (i = 0; i < max_bonds; i++) {
5311 res = bond_create(&init_net, NULL);
5316 res = bond_create_sysfs();
5321 register_netdevice_notifier(&bond_netdev_notifier);
5322 register_inetaddr_notifier(&bond_inetaddr_notifier);
5323 bond_register_ipv6_notifier();
5327 rtnl_link_unregister(&bond_link_ops);
5329 unregister_pernet_subsys(&bond_net_ops);
5334 static void __exit bonding_exit(void)
5336 unregister_netdevice_notifier(&bond_netdev_notifier);
5337 unregister_inetaddr_notifier(&bond_inetaddr_notifier);
5338 bond_unregister_ipv6_notifier();
5340 bond_destroy_sysfs();
5342 rtnl_link_unregister(&bond_link_ops);
5343 unregister_pernet_subsys(&bond_net_ops);
5345 #ifdef CONFIG_NET_POLL_CONTROLLER
5347 * Make sure we don't have an imbalance on our netpoll blocking
5349 WARN_ON(atomic_read(&netpoll_block_tx));
5353 module_init(bonding_init);
5354 module_exit(bonding_exit);
5355 MODULE_LICENSE("GPL");
5356 MODULE_VERSION(DRV_VERSION);
5357 MODULE_DESCRIPTION(DRV_DESCRIPTION ", v" DRV_VERSION);
5358 MODULE_AUTHOR("Thomas Davis, tadavis@lbl.gov and many others");
5359 MODULE_ALIAS_RTNL_LINK("bond");