2 * originally based on the dummy device.
4 * Copyright 1999, Thomas Davis, tadavis@lbl.gov.
5 * Licensed under the GPL. Based on dummy.c, and eql.c devices.
7 * bonding.c: an Ethernet Bonding driver
9 * This is useful to talk to a Cisco EtherChannel compatible equipment:
11 * Sun Trunking (Solaris)
12 * Alteon AceDirector Trunks
14 * and probably many L2 switches ...
17 * ifconfig bond0 ipaddress netmask up
18 * will setup a network device, with an ip address. No mac address
19 * will be assigned at this time. The hw mac address will come from
20 * the first slave bonded to the channel. All slaves will then use
21 * this hw mac address.
24 * will release all slaves, marking them as down.
26 * ifenslave bond0 eth0
27 * will attach eth0 to bond0 as a slave. eth0 hw mac address will either
28 * a: be used as initial mac address
29 * b: if a hw mac address already is there, eth0's hw mac address
30 * will then be set from bond0.
34 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
36 #include <linux/kernel.h>
37 #include <linux/module.h>
38 #include <linux/types.h>
39 #include <linux/fcntl.h>
40 #include <linux/interrupt.h>
41 #include <linux/ptrace.h>
42 #include <linux/ioport.h>
46 #include <linux/tcp.h>
47 #include <linux/udp.h>
48 #include <linux/slab.h>
49 #include <linux/string.h>
50 #include <linux/init.h>
51 #include <linux/timer.h>
52 #include <linux/socket.h>
53 #include <linux/ctype.h>
54 #include <linux/inet.h>
55 #include <linux/bitops.h>
57 #include <asm/system.h>
59 #include <linux/uaccess.h>
60 #include <linux/errno.h>
61 #include <linux/netdevice.h>
62 #include <linux/netpoll.h>
63 #include <linux/inetdevice.h>
64 #include <linux/igmp.h>
65 #include <linux/etherdevice.h>
66 #include <linux/skbuff.h>
68 #include <linux/rtnetlink.h>
69 #include <linux/proc_fs.h>
70 #include <linux/seq_file.h>
71 #include <linux/smp.h>
72 #include <linux/if_ether.h>
74 #include <linux/mii.h>
75 #include <linux/ethtool.h>
76 #include <linux/if_vlan.h>
77 #include <linux/if_bonding.h>
78 #include <linux/jiffies.h>
79 #include <net/route.h>
80 #include <net/net_namespace.h>
81 #include <net/netns/generic.h>
86 /*---------------------------- Module parameters ----------------------------*/
88 /* monitor all links that often (in milliseconds). <=0 disables monitoring */
89 #define BOND_LINK_MON_INTERV 0
90 #define BOND_LINK_ARP_INTERV 0
92 static int max_bonds = BOND_DEFAULT_MAX_BONDS;
93 static int tx_queues = BOND_DEFAULT_TX_QUEUES;
94 static int num_grat_arp = 1;
95 static int num_unsol_na = 1;
96 static int miimon = BOND_LINK_MON_INTERV;
99 static int use_carrier = 1;
101 static char *primary;
102 static char *primary_reselect;
103 static char *lacp_rate;
104 static char *ad_select;
105 static char *xmit_hash_policy;
106 static int arp_interval = BOND_LINK_ARP_INTERV;
107 static char *arp_ip_target[BOND_MAX_ARP_TARGETS];
108 static char *arp_validate;
109 static char *fail_over_mac;
110 static int all_slaves_active = 0;
111 static struct bond_params bonding_defaults;
112 static int resend_igmp = BOND_DEFAULT_RESEND_IGMP;
114 module_param(max_bonds, int, 0);
115 MODULE_PARM_DESC(max_bonds, "Max number of bonded devices");
116 module_param(tx_queues, int, 0);
117 MODULE_PARM_DESC(tx_queues, "Max number of transmit queues (default = 16)");
118 module_param(num_grat_arp, int, 0644);
119 MODULE_PARM_DESC(num_grat_arp, "Number of gratuitous ARP packets to send on failover event");
120 module_param(num_unsol_na, int, 0644);
121 MODULE_PARM_DESC(num_unsol_na, "Number of unsolicited IPv6 Neighbor Advertisements packets to send on failover event");
122 module_param(miimon, int, 0);
123 MODULE_PARM_DESC(miimon, "Link check interval in milliseconds");
124 module_param(updelay, int, 0);
125 MODULE_PARM_DESC(updelay, "Delay before considering link up, in milliseconds");
126 module_param(downdelay, int, 0);
127 MODULE_PARM_DESC(downdelay, "Delay before considering link down, "
129 module_param(use_carrier, int, 0);
130 MODULE_PARM_DESC(use_carrier, "Use netif_carrier_ok (vs MII ioctls) in miimon; "
131 "0 for off, 1 for on (default)");
132 module_param(mode, charp, 0);
133 MODULE_PARM_DESC(mode, "Mode of operation : 0 for balance-rr, "
134 "1 for active-backup, 2 for balance-xor, "
135 "3 for broadcast, 4 for 802.3ad, 5 for balance-tlb, "
136 "6 for balance-alb");
137 module_param(primary, charp, 0);
138 MODULE_PARM_DESC(primary, "Primary network device to use");
139 module_param(primary_reselect, charp, 0);
140 MODULE_PARM_DESC(primary_reselect, "Reselect primary slave "
142 "0 for always (default), "
143 "1 for only if speed of primary is "
145 "2 for only on active slave "
147 module_param(lacp_rate, charp, 0);
148 MODULE_PARM_DESC(lacp_rate, "LACPDU tx rate to request from 802.3ad partner "
150 module_param(ad_select, charp, 0);
151 MODULE_PARM_DESC(ad_select, "803.ad aggregation selection logic: stable (0, default), bandwidth (1), count (2)");
152 module_param(xmit_hash_policy, charp, 0);
153 MODULE_PARM_DESC(xmit_hash_policy, "XOR hashing method: 0 for layer 2 (default)"
154 ", 1 for layer 3+4");
155 module_param(arp_interval, int, 0);
156 MODULE_PARM_DESC(arp_interval, "arp interval in milliseconds");
157 module_param_array(arp_ip_target, charp, NULL, 0);
158 MODULE_PARM_DESC(arp_ip_target, "arp targets in n.n.n.n form");
159 module_param(arp_validate, charp, 0);
160 MODULE_PARM_DESC(arp_validate, "validate src/dst of ARP probes: none (default), active, backup or all");
161 module_param(fail_over_mac, charp, 0);
162 MODULE_PARM_DESC(fail_over_mac, "For active-backup, do not set all slaves to the same MAC. none (default), active or follow");
163 module_param(all_slaves_active, int, 0);
164 MODULE_PARM_DESC(all_slaves_active, "Keep all frames received on an interface"
165 "by setting active flag for all slaves. "
166 "0 for never (default), 1 for always.");
167 module_param(resend_igmp, int, 0);
168 MODULE_PARM_DESC(resend_igmp, "Number of IGMP membership reports to send on link failure");
170 /*----------------------------- Global variables ----------------------------*/
172 static const char * const version =
173 DRV_DESCRIPTION ": v" DRV_VERSION " (" DRV_RELDATE ")\n";
175 int bond_net_id __read_mostly;
177 static __be32 arp_target[BOND_MAX_ARP_TARGETS];
178 static int arp_ip_count;
179 static int bond_mode = BOND_MODE_ROUNDROBIN;
180 static int xmit_hashtype = BOND_XMIT_POLICY_LAYER2;
181 static int lacp_fast;
182 #ifdef CONFIG_NET_POLL_CONTROLLER
183 static int disable_netpoll = 1;
186 const struct bond_parm_tbl bond_lacp_tbl[] = {
187 { "slow", AD_LACP_SLOW},
188 { "fast", AD_LACP_FAST},
192 const struct bond_parm_tbl bond_mode_tbl[] = {
193 { "balance-rr", BOND_MODE_ROUNDROBIN},
194 { "active-backup", BOND_MODE_ACTIVEBACKUP},
195 { "balance-xor", BOND_MODE_XOR},
196 { "broadcast", BOND_MODE_BROADCAST},
197 { "802.3ad", BOND_MODE_8023AD},
198 { "balance-tlb", BOND_MODE_TLB},
199 { "balance-alb", BOND_MODE_ALB},
203 const struct bond_parm_tbl xmit_hashtype_tbl[] = {
204 { "layer2", BOND_XMIT_POLICY_LAYER2},
205 { "layer3+4", BOND_XMIT_POLICY_LAYER34},
206 { "layer2+3", BOND_XMIT_POLICY_LAYER23},
210 const struct bond_parm_tbl arp_validate_tbl[] = {
211 { "none", BOND_ARP_VALIDATE_NONE},
212 { "active", BOND_ARP_VALIDATE_ACTIVE},
213 { "backup", BOND_ARP_VALIDATE_BACKUP},
214 { "all", BOND_ARP_VALIDATE_ALL},
218 const struct bond_parm_tbl fail_over_mac_tbl[] = {
219 { "none", BOND_FOM_NONE},
220 { "active", BOND_FOM_ACTIVE},
221 { "follow", BOND_FOM_FOLLOW},
225 const struct bond_parm_tbl pri_reselect_tbl[] = {
226 { "always", BOND_PRI_RESELECT_ALWAYS},
227 { "better", BOND_PRI_RESELECT_BETTER},
228 { "failure", BOND_PRI_RESELECT_FAILURE},
232 struct bond_parm_tbl ad_select_tbl[] = {
233 { "stable", BOND_AD_STABLE},
234 { "bandwidth", BOND_AD_BANDWIDTH},
235 { "count", BOND_AD_COUNT},
239 /*-------------------------- Forward declarations ---------------------------*/
241 static void bond_send_gratuitous_arp(struct bonding *bond);
242 static int bond_init(struct net_device *bond_dev);
243 static void bond_uninit(struct net_device *bond_dev);
245 /*---------------------------- General routines -----------------------------*/
247 static const char *bond_mode_name(int mode)
249 static const char *names[] = {
250 [BOND_MODE_ROUNDROBIN] = "load balancing (round-robin)",
251 [BOND_MODE_ACTIVEBACKUP] = "fault-tolerance (active-backup)",
252 [BOND_MODE_XOR] = "load balancing (xor)",
253 [BOND_MODE_BROADCAST] = "fault-tolerance (broadcast)",
254 [BOND_MODE_8023AD] = "IEEE 802.3ad Dynamic link aggregation",
255 [BOND_MODE_TLB] = "transmit load balancing",
256 [BOND_MODE_ALB] = "adaptive load balancing",
259 if (mode < 0 || mode > BOND_MODE_ALB)
265 /*---------------------------------- VLAN -----------------------------------*/
268 * bond_add_vlan - add a new vlan id on bond
269 * @bond: bond that got the notification
270 * @vlan_id: the vlan id to add
272 * Returns -ENOMEM if allocation failed.
274 static int bond_add_vlan(struct bonding *bond, unsigned short vlan_id)
276 struct vlan_entry *vlan;
278 pr_debug("bond: %s, vlan id %d\n",
279 (bond ? bond->dev->name : "None"), vlan_id);
281 vlan = kzalloc(sizeof(struct vlan_entry), GFP_KERNEL);
285 INIT_LIST_HEAD(&vlan->vlan_list);
286 vlan->vlan_id = vlan_id;
288 write_lock_bh(&bond->lock);
290 list_add_tail(&vlan->vlan_list, &bond->vlan_list);
292 write_unlock_bh(&bond->lock);
294 pr_debug("added VLAN ID %d on bond %s\n", vlan_id, bond->dev->name);
300 * bond_del_vlan - delete a vlan id from bond
301 * @bond: bond that got the notification
302 * @vlan_id: the vlan id to delete
304 * returns -ENODEV if @vlan_id was not found in @bond.
306 static int bond_del_vlan(struct bonding *bond, unsigned short vlan_id)
308 struct vlan_entry *vlan;
311 pr_debug("bond: %s, vlan id %d\n", bond->dev->name, vlan_id);
313 write_lock_bh(&bond->lock);
315 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
316 if (vlan->vlan_id == vlan_id) {
317 list_del(&vlan->vlan_list);
319 if (bond_is_lb(bond))
320 bond_alb_clear_vlan(bond, vlan_id);
322 pr_debug("removed VLAN ID %d from bond %s\n",
323 vlan_id, bond->dev->name);
327 if (list_empty(&bond->vlan_list) &&
328 (bond->slave_cnt == 0)) {
329 /* Last VLAN removed and no slaves, so
330 * restore block on adding VLANs. This will
331 * be removed once new slaves that are not
332 * VLAN challenged will be added.
334 bond->dev->features |= NETIF_F_VLAN_CHALLENGED;
342 pr_debug("couldn't find VLAN ID %d in bond %s\n",
343 vlan_id, bond->dev->name);
346 write_unlock_bh(&bond->lock);
351 * bond_has_challenged_slaves
352 * @bond: the bond we're working on
354 * Searches the slave list. Returns 1 if a vlan challenged slave
355 * was found, 0 otherwise.
357 * Assumes bond->lock is held.
359 static int bond_has_challenged_slaves(struct bonding *bond)
364 bond_for_each_slave(bond, slave, i) {
365 if (slave->dev->features & NETIF_F_VLAN_CHALLENGED) {
366 pr_debug("found VLAN challenged slave - %s\n",
372 pr_debug("no VLAN challenged slaves found\n");
377 * bond_next_vlan - safely skip to the next item in the vlans list.
378 * @bond: the bond we're working on
379 * @curr: item we're advancing from
381 * Returns %NULL if list is empty, bond->next_vlan if @curr is %NULL,
382 * or @curr->next otherwise (even if it is @curr itself again).
384 * Caller must hold bond->lock
386 struct vlan_entry *bond_next_vlan(struct bonding *bond, struct vlan_entry *curr)
388 struct vlan_entry *next, *last;
390 if (list_empty(&bond->vlan_list))
394 next = list_entry(bond->vlan_list.next,
395 struct vlan_entry, vlan_list);
397 last = list_entry(bond->vlan_list.prev,
398 struct vlan_entry, vlan_list);
400 next = list_entry(bond->vlan_list.next,
401 struct vlan_entry, vlan_list);
403 next = list_entry(curr->vlan_list.next,
404 struct vlan_entry, vlan_list);
412 * bond_dev_queue_xmit - Prepare skb for xmit.
414 * @bond: bond device that got this skb for tx.
415 * @skb: hw accel VLAN tagged skb to transmit
416 * @slave_dev: slave that is supposed to xmit this skbuff
418 * When the bond gets an skb to transmit that is
419 * already hardware accelerated VLAN tagged, and it
420 * needs to relay this skb to a slave that is not
421 * hw accel capable, the skb needs to be "unaccelerated",
422 * i.e. strip the hwaccel tag and re-insert it as part
425 int bond_dev_queue_xmit(struct bonding *bond, struct sk_buff *skb,
426 struct net_device *slave_dev)
428 unsigned short uninitialized_var(vlan_id);
430 /* Test vlan_list not vlgrp to catch and handle 802.1p tags */
431 if (!list_empty(&bond->vlan_list) &&
432 !(slave_dev->features & NETIF_F_HW_VLAN_TX) &&
433 vlan_get_tag(skb, &vlan_id) == 0) {
434 skb->dev = slave_dev;
435 skb = vlan_put_tag(skb, vlan_id);
437 /* vlan_put_tag() frees the skb in case of error,
438 * so return success here so the calling functions
439 * won't attempt to free is again.
444 skb->dev = slave_dev;
448 #ifdef CONFIG_NET_POLL_CONTROLLER
449 if (unlikely(bond->dev->priv_flags & IFF_IN_NETPOLL)) {
450 struct netpoll *np = bond->dev->npinfo->netpoll;
451 slave_dev->npinfo = bond->dev->npinfo;
452 np->real_dev = np->dev = skb->dev;
453 slave_dev->priv_flags |= IFF_IN_NETPOLL;
454 netpoll_send_skb(np, skb);
455 slave_dev->priv_flags &= ~IFF_IN_NETPOLL;
465 * In the following 3 functions, bond_vlan_rx_register(), bond_vlan_rx_add_vid
466 * and bond_vlan_rx_kill_vid, We don't protect the slave list iteration with a
468 * a. This operation is performed in IOCTL context,
469 * b. The operation is protected by the RTNL semaphore in the 8021q code,
470 * c. Holding a lock with BH disabled while directly calling a base driver
471 * entry point is generally a BAD idea.
473 * The design of synchronization/protection for this operation in the 8021q
474 * module is good for one or more VLAN devices over a single physical device
475 * and cannot be extended for a teaming solution like bonding, so there is a
476 * potential race condition here where a net device from the vlan group might
477 * be referenced (either by a base driver or the 8021q code) while it is being
478 * removed from the system. However, it turns out we're not making matters
479 * worse, and if it works for regular VLAN usage it will work here too.
483 * bond_vlan_rx_register - Propagates registration to slaves
484 * @bond_dev: bonding net device that got called
485 * @grp: vlan group being registered
487 static void bond_vlan_rx_register(struct net_device *bond_dev,
488 struct vlan_group *grp)
490 struct bonding *bond = netdev_priv(bond_dev);
494 write_lock(&bond->lock);
496 write_unlock(&bond->lock);
498 bond_for_each_slave(bond, slave, i) {
499 struct net_device *slave_dev = slave->dev;
500 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
502 if ((slave_dev->features & NETIF_F_HW_VLAN_RX) &&
503 slave_ops->ndo_vlan_rx_register) {
504 slave_ops->ndo_vlan_rx_register(slave_dev, grp);
510 * bond_vlan_rx_add_vid - Propagates adding an id to slaves
511 * @bond_dev: bonding net device that got called
512 * @vid: vlan id being added
514 static void bond_vlan_rx_add_vid(struct net_device *bond_dev, uint16_t vid)
516 struct bonding *bond = netdev_priv(bond_dev);
520 bond_for_each_slave(bond, slave, i) {
521 struct net_device *slave_dev = slave->dev;
522 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
524 if ((slave_dev->features & NETIF_F_HW_VLAN_FILTER) &&
525 slave_ops->ndo_vlan_rx_add_vid) {
526 slave_ops->ndo_vlan_rx_add_vid(slave_dev, vid);
530 res = bond_add_vlan(bond, vid);
532 pr_err("%s: Error: Failed to add vlan id %d\n",
533 bond_dev->name, vid);
538 * bond_vlan_rx_kill_vid - Propagates deleting an id to slaves
539 * @bond_dev: bonding net device that got called
540 * @vid: vlan id being removed
542 static void bond_vlan_rx_kill_vid(struct net_device *bond_dev, uint16_t vid)
544 struct bonding *bond = netdev_priv(bond_dev);
546 struct net_device *vlan_dev;
549 bond_for_each_slave(bond, slave, i) {
550 struct net_device *slave_dev = slave->dev;
551 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
553 if ((slave_dev->features & NETIF_F_HW_VLAN_FILTER) &&
554 slave_ops->ndo_vlan_rx_kill_vid) {
555 /* Save and then restore vlan_dev in the grp array,
556 * since the slave's driver might clear it.
558 vlan_dev = vlan_group_get_device(bond->vlgrp, vid);
559 slave_ops->ndo_vlan_rx_kill_vid(slave_dev, vid);
560 vlan_group_set_device(bond->vlgrp, vid, vlan_dev);
564 res = bond_del_vlan(bond, vid);
566 pr_err("%s: Error: Failed to remove vlan id %d\n",
567 bond_dev->name, vid);
571 static void bond_add_vlans_on_slave(struct bonding *bond, struct net_device *slave_dev)
573 struct vlan_entry *vlan;
574 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
579 if ((slave_dev->features & NETIF_F_HW_VLAN_RX) &&
580 slave_ops->ndo_vlan_rx_register)
581 slave_ops->ndo_vlan_rx_register(slave_dev, bond->vlgrp);
583 if (!(slave_dev->features & NETIF_F_HW_VLAN_FILTER) ||
584 !(slave_ops->ndo_vlan_rx_add_vid))
587 list_for_each_entry(vlan, &bond->vlan_list, vlan_list)
588 slave_ops->ndo_vlan_rx_add_vid(slave_dev, vlan->vlan_id);
591 static void bond_del_vlans_from_slave(struct bonding *bond,
592 struct net_device *slave_dev)
594 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
595 struct vlan_entry *vlan;
596 struct net_device *vlan_dev;
601 if (!(slave_dev->features & NETIF_F_HW_VLAN_FILTER) ||
602 !(slave_ops->ndo_vlan_rx_kill_vid))
605 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
608 /* Save and then restore vlan_dev in the grp array,
609 * since the slave's driver might clear it.
611 vlan_dev = vlan_group_get_device(bond->vlgrp, vlan->vlan_id);
612 slave_ops->ndo_vlan_rx_kill_vid(slave_dev, vlan->vlan_id);
613 vlan_group_set_device(bond->vlgrp, vlan->vlan_id, vlan_dev);
617 if ((slave_dev->features & NETIF_F_HW_VLAN_RX) &&
618 slave_ops->ndo_vlan_rx_register)
619 slave_ops->ndo_vlan_rx_register(slave_dev, NULL);
622 /*------------------------------- Link status -------------------------------*/
625 * Set the carrier state for the master according to the state of its
626 * slaves. If any slaves are up, the master is up. In 802.3ad mode,
627 * do special 802.3ad magic.
629 * Returns zero if carrier state does not change, nonzero if it does.
631 static int bond_set_carrier(struct bonding *bond)
636 if (bond->slave_cnt == 0)
639 if (bond->params.mode == BOND_MODE_8023AD)
640 return bond_3ad_set_carrier(bond);
642 bond_for_each_slave(bond, slave, i) {
643 if (slave->link == BOND_LINK_UP) {
644 if (!netif_carrier_ok(bond->dev)) {
645 netif_carrier_on(bond->dev);
653 if (netif_carrier_ok(bond->dev)) {
654 netif_carrier_off(bond->dev);
661 * Get link speed and duplex from the slave's base driver
662 * using ethtool. If for some reason the call fails or the
663 * values are invalid, fake speed and duplex to 100/Full
666 static int bond_update_speed_duplex(struct slave *slave)
668 struct net_device *slave_dev = slave->dev;
669 struct ethtool_cmd etool;
672 /* Fake speed and duplex */
673 slave->speed = SPEED_100;
674 slave->duplex = DUPLEX_FULL;
676 if (!slave_dev->ethtool_ops || !slave_dev->ethtool_ops->get_settings)
679 res = slave_dev->ethtool_ops->get_settings(slave_dev, &etool);
683 switch (etool.speed) {
693 switch (etool.duplex) {
701 slave->speed = etool.speed;
702 slave->duplex = etool.duplex;
708 * if <dev> supports MII link status reporting, check its link status.
710 * We either do MII/ETHTOOL ioctls, or check netif_carrier_ok(),
711 * depending upon the setting of the use_carrier parameter.
713 * Return either BMSR_LSTATUS, meaning that the link is up (or we
714 * can't tell and just pretend it is), or 0, meaning that the link is
717 * If reporting is non-zero, instead of faking link up, return -1 if
718 * both ETHTOOL and MII ioctls fail (meaning the device does not
719 * support them). If use_carrier is set, return whatever it says.
720 * It'd be nice if there was a good way to tell if a driver supports
721 * netif_carrier, but there really isn't.
723 static int bond_check_dev_link(struct bonding *bond,
724 struct net_device *slave_dev, int reporting)
726 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
727 int (*ioctl)(struct net_device *, struct ifreq *, int);
729 struct mii_ioctl_data *mii;
731 if (!reporting && !netif_running(slave_dev))
734 if (bond->params.use_carrier)
735 return netif_carrier_ok(slave_dev) ? BMSR_LSTATUS : 0;
737 /* Try to get link status using Ethtool first. */
738 if (slave_dev->ethtool_ops) {
739 if (slave_dev->ethtool_ops->get_link) {
742 link = slave_dev->ethtool_ops->get_link(slave_dev);
744 return link ? BMSR_LSTATUS : 0;
748 /* Ethtool can't be used, fallback to MII ioctls. */
749 ioctl = slave_ops->ndo_do_ioctl;
751 /* TODO: set pointer to correct ioctl on a per team member */
752 /* bases to make this more efficient. that is, once */
753 /* we determine the correct ioctl, we will always */
754 /* call it and not the others for that team */
758 * We cannot assume that SIOCGMIIPHY will also read a
759 * register; not all network drivers (e.g., e100)
763 /* Yes, the mii is overlaid on the ifreq.ifr_ifru */
764 strncpy(ifr.ifr_name, slave_dev->name, IFNAMSIZ);
766 if (IOCTL(slave_dev, &ifr, SIOCGMIIPHY) == 0) {
767 mii->reg_num = MII_BMSR;
768 if (IOCTL(slave_dev, &ifr, SIOCGMIIREG) == 0)
769 return mii->val_out & BMSR_LSTATUS;
774 * If reporting, report that either there's no dev->do_ioctl,
775 * or both SIOCGMIIREG and get_link failed (meaning that we
776 * cannot report link status). If not reporting, pretend
779 return reporting ? -1 : BMSR_LSTATUS;
782 /*----------------------------- Multicast list ------------------------------*/
785 * Push the promiscuity flag down to appropriate slaves
787 static int bond_set_promiscuity(struct bonding *bond, int inc)
790 if (USES_PRIMARY(bond->params.mode)) {
791 /* write lock already acquired */
792 if (bond->curr_active_slave) {
793 err = dev_set_promiscuity(bond->curr_active_slave->dev,
799 bond_for_each_slave(bond, slave, i) {
800 err = dev_set_promiscuity(slave->dev, inc);
809 * Push the allmulti flag down to all slaves
811 static int bond_set_allmulti(struct bonding *bond, int inc)
814 if (USES_PRIMARY(bond->params.mode)) {
815 /* write lock already acquired */
816 if (bond->curr_active_slave) {
817 err = dev_set_allmulti(bond->curr_active_slave->dev,
823 bond_for_each_slave(bond, slave, i) {
824 err = dev_set_allmulti(slave->dev, inc);
833 * Add a Multicast address to slaves
836 static void bond_mc_add(struct bonding *bond, void *addr)
838 if (USES_PRIMARY(bond->params.mode)) {
839 /* write lock already acquired */
840 if (bond->curr_active_slave)
841 dev_mc_add(bond->curr_active_slave->dev, addr);
846 bond_for_each_slave(bond, slave, i)
847 dev_mc_add(slave->dev, addr);
852 * Remove a multicast address from slave
855 static void bond_mc_del(struct bonding *bond, void *addr)
857 if (USES_PRIMARY(bond->params.mode)) {
858 /* write lock already acquired */
859 if (bond->curr_active_slave)
860 dev_mc_del(bond->curr_active_slave->dev, addr);
864 bond_for_each_slave(bond, slave, i) {
865 dev_mc_del(slave->dev, addr);
871 static void __bond_resend_igmp_join_requests(struct net_device *dev)
873 struct in_device *in_dev;
874 struct ip_mc_list *im;
877 in_dev = __in_dev_get_rcu(dev);
879 for (im = in_dev->mc_list; im; im = im->next)
880 ip_mc_rejoin_group(im);
887 * Retrieve the list of registered multicast addresses for the bonding
888 * device and retransmit an IGMP JOIN request to the current active
891 static void bond_resend_igmp_join_requests(struct bonding *bond)
893 struct net_device *vlan_dev;
894 struct vlan_entry *vlan;
896 read_lock(&bond->lock);
898 /* rejoin all groups on bond device */
899 __bond_resend_igmp_join_requests(bond->dev);
901 /* rejoin all groups on vlan devices */
903 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
904 vlan_dev = vlan_group_get_device(bond->vlgrp,
907 __bond_resend_igmp_join_requests(vlan_dev);
911 if (--bond->igmp_retrans > 0)
912 queue_delayed_work(bond->wq, &bond->mcast_work, HZ/5);
914 read_unlock(&bond->lock);
917 void bond_resend_igmp_join_requests_delayed(struct work_struct *work)
919 struct bonding *bond = container_of(work, struct bonding,
921 bond_resend_igmp_join_requests(bond);
925 * flush all members of flush->mc_list from device dev->mc_list
927 static void bond_mc_list_flush(struct net_device *bond_dev,
928 struct net_device *slave_dev)
930 struct bonding *bond = netdev_priv(bond_dev);
931 struct netdev_hw_addr *ha;
933 netdev_for_each_mc_addr(ha, bond_dev)
934 dev_mc_del(slave_dev, ha->addr);
936 if (bond->params.mode == BOND_MODE_8023AD) {
937 /* del lacpdu mc addr from mc list */
938 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
940 dev_mc_del(slave_dev, lacpdu_multicast);
944 /*--------------------------- Active slave change ---------------------------*/
947 * Update the mc list and multicast-related flags for the new and
948 * old active slaves (if any) according to the multicast mode, and
949 * promiscuous flags unconditionally.
951 static void bond_mc_swap(struct bonding *bond, struct slave *new_active,
952 struct slave *old_active)
954 struct netdev_hw_addr *ha;
956 if (!USES_PRIMARY(bond->params.mode))
957 /* nothing to do - mc list is already up-to-date on
963 if (bond->dev->flags & IFF_PROMISC)
964 dev_set_promiscuity(old_active->dev, -1);
966 if (bond->dev->flags & IFF_ALLMULTI)
967 dev_set_allmulti(old_active->dev, -1);
969 netdev_for_each_mc_addr(ha, bond->dev)
970 dev_mc_del(old_active->dev, ha->addr);
974 /* FIXME: Signal errors upstream. */
975 if (bond->dev->flags & IFF_PROMISC)
976 dev_set_promiscuity(new_active->dev, 1);
978 if (bond->dev->flags & IFF_ALLMULTI)
979 dev_set_allmulti(new_active->dev, 1);
981 netdev_for_each_mc_addr(ha, bond->dev)
982 dev_mc_add(new_active->dev, ha->addr);
987 * bond_do_fail_over_mac
989 * Perform special MAC address swapping for fail_over_mac settings
991 * Called with RTNL, bond->lock for read, curr_slave_lock for write_bh.
993 static void bond_do_fail_over_mac(struct bonding *bond,
994 struct slave *new_active,
995 struct slave *old_active)
996 __releases(&bond->curr_slave_lock)
997 __releases(&bond->lock)
998 __acquires(&bond->lock)
999 __acquires(&bond->curr_slave_lock)
1001 u8 tmp_mac[ETH_ALEN];
1002 struct sockaddr saddr;
1005 switch (bond->params.fail_over_mac) {
1006 case BOND_FOM_ACTIVE:
1008 memcpy(bond->dev->dev_addr, new_active->dev->dev_addr,
1009 new_active->dev->addr_len);
1011 case BOND_FOM_FOLLOW:
1013 * if new_active && old_active, swap them
1014 * if just old_active, do nothing (going to no active slave)
1015 * if just new_active, set new_active to bond's MAC
1020 write_unlock_bh(&bond->curr_slave_lock);
1021 read_unlock(&bond->lock);
1024 memcpy(tmp_mac, new_active->dev->dev_addr, ETH_ALEN);
1025 memcpy(saddr.sa_data, old_active->dev->dev_addr,
1027 saddr.sa_family = new_active->dev->type;
1029 memcpy(saddr.sa_data, bond->dev->dev_addr, ETH_ALEN);
1030 saddr.sa_family = bond->dev->type;
1033 rv = dev_set_mac_address(new_active->dev, &saddr);
1035 pr_err("%s: Error %d setting MAC of slave %s\n",
1036 bond->dev->name, -rv, new_active->dev->name);
1043 memcpy(saddr.sa_data, tmp_mac, ETH_ALEN);
1044 saddr.sa_family = old_active->dev->type;
1046 rv = dev_set_mac_address(old_active->dev, &saddr);
1048 pr_err("%s: Error %d setting MAC of slave %s\n",
1049 bond->dev->name, -rv, new_active->dev->name);
1051 read_lock(&bond->lock);
1052 write_lock_bh(&bond->curr_slave_lock);
1055 pr_err("%s: bond_do_fail_over_mac impossible: bad policy %d\n",
1056 bond->dev->name, bond->params.fail_over_mac);
1062 static bool bond_should_change_active(struct bonding *bond)
1064 struct slave *prim = bond->primary_slave;
1065 struct slave *curr = bond->curr_active_slave;
1067 if (!prim || !curr || curr->link != BOND_LINK_UP)
1069 if (bond->force_primary) {
1070 bond->force_primary = false;
1073 if (bond->params.primary_reselect == BOND_PRI_RESELECT_BETTER &&
1074 (prim->speed < curr->speed ||
1075 (prim->speed == curr->speed && prim->duplex <= curr->duplex)))
1077 if (bond->params.primary_reselect == BOND_PRI_RESELECT_FAILURE)
1083 * find_best_interface - select the best available slave to be the active one
1084 * @bond: our bonding struct
1086 * Warning: Caller must hold curr_slave_lock for writing.
1088 static struct slave *bond_find_best_slave(struct bonding *bond)
1090 struct slave *new_active, *old_active;
1091 struct slave *bestslave = NULL;
1092 int mintime = bond->params.updelay;
1095 new_active = bond->curr_active_slave;
1097 if (!new_active) { /* there were no active slaves left */
1098 if (bond->slave_cnt > 0) /* found one slave */
1099 new_active = bond->first_slave;
1101 return NULL; /* still no slave, return NULL */
1104 if ((bond->primary_slave) &&
1105 bond->primary_slave->link == BOND_LINK_UP &&
1106 bond_should_change_active(bond)) {
1107 new_active = bond->primary_slave;
1110 /* remember where to stop iterating over the slaves */
1111 old_active = new_active;
1113 bond_for_each_slave_from(bond, new_active, i, old_active) {
1114 if (new_active->link == BOND_LINK_UP) {
1116 } else if (new_active->link == BOND_LINK_BACK &&
1117 IS_UP(new_active->dev)) {
1118 /* link up, but waiting for stabilization */
1119 if (new_active->delay < mintime) {
1120 mintime = new_active->delay;
1121 bestslave = new_active;
1130 * change_active_interface - change the active slave into the specified one
1131 * @bond: our bonding struct
1132 * @new: the new slave to make the active one
1134 * Set the new slave to the bond's settings and unset them on the old
1135 * curr_active_slave.
1136 * Setting include flags, mc-list, promiscuity, allmulti, etc.
1138 * If @new's link state is %BOND_LINK_BACK we'll set it to %BOND_LINK_UP,
1139 * because it is apparently the best available slave we have, even though its
1140 * updelay hasn't timed out yet.
1142 * If new_active is not NULL, caller must hold bond->lock for read and
1143 * curr_slave_lock for write_bh.
1145 void bond_change_active_slave(struct bonding *bond, struct slave *new_active)
1147 struct slave *old_active = bond->curr_active_slave;
1149 if (old_active == new_active)
1153 new_active->jiffies = jiffies;
1155 if (new_active->link == BOND_LINK_BACK) {
1156 if (USES_PRIMARY(bond->params.mode)) {
1157 pr_info("%s: making interface %s the new active one %d ms earlier.\n",
1158 bond->dev->name, new_active->dev->name,
1159 (bond->params.updelay - new_active->delay) * bond->params.miimon);
1162 new_active->delay = 0;
1163 new_active->link = BOND_LINK_UP;
1165 if (bond->params.mode == BOND_MODE_8023AD)
1166 bond_3ad_handle_link_change(new_active, BOND_LINK_UP);
1168 if (bond_is_lb(bond))
1169 bond_alb_handle_link_change(bond, new_active, BOND_LINK_UP);
1171 if (USES_PRIMARY(bond->params.mode)) {
1172 pr_info("%s: making interface %s the new active one.\n",
1173 bond->dev->name, new_active->dev->name);
1178 if (USES_PRIMARY(bond->params.mode))
1179 bond_mc_swap(bond, new_active, old_active);
1181 if (bond_is_lb(bond)) {
1182 bond_alb_handle_active_change(bond, new_active);
1184 bond_set_slave_inactive_flags(old_active);
1186 bond_set_slave_active_flags(new_active);
1188 bond->curr_active_slave = new_active;
1191 if (bond->params.mode == BOND_MODE_ACTIVEBACKUP) {
1193 bond_set_slave_inactive_flags(old_active);
1196 bond_set_slave_active_flags(new_active);
1198 if (bond->params.fail_over_mac)
1199 bond_do_fail_over_mac(bond, new_active,
1202 bond->send_grat_arp = bond->params.num_grat_arp;
1203 bond_send_gratuitous_arp(bond);
1205 bond->send_unsol_na = bond->params.num_unsol_na;
1206 bond_send_unsolicited_na(bond);
1208 write_unlock_bh(&bond->curr_slave_lock);
1209 read_unlock(&bond->lock);
1211 netdev_bonding_change(bond->dev, NETDEV_BONDING_FAILOVER);
1213 read_lock(&bond->lock);
1214 write_lock_bh(&bond->curr_slave_lock);
1218 /* resend IGMP joins since active slave has changed or
1219 * all were sent on curr_active_slave */
1220 if ((USES_PRIMARY(bond->params.mode) && new_active) ||
1221 bond->params.mode == BOND_MODE_ROUNDROBIN) {
1222 bond->igmp_retrans = bond->params.resend_igmp;
1223 queue_delayed_work(bond->wq, &bond->mcast_work, 0);
1228 * bond_select_active_slave - select a new active slave, if needed
1229 * @bond: our bonding struct
1231 * This functions should be called when one of the following occurs:
1232 * - The old curr_active_slave has been released or lost its link.
1233 * - The primary_slave has got its link back.
1234 * - A slave has got its link back and there's no old curr_active_slave.
1236 * Caller must hold bond->lock for read and curr_slave_lock for write_bh.
1238 void bond_select_active_slave(struct bonding *bond)
1240 struct slave *best_slave;
1243 best_slave = bond_find_best_slave(bond);
1244 if (best_slave != bond->curr_active_slave) {
1245 bond_change_active_slave(bond, best_slave);
1246 rv = bond_set_carrier(bond);
1250 if (netif_carrier_ok(bond->dev)) {
1251 pr_info("%s: first active interface up!\n",
1254 pr_info("%s: now running without any active interface !\n",
1260 /*--------------------------- slave list handling ---------------------------*/
1263 * This function attaches the slave to the end of list.
1265 * bond->lock held for writing by caller.
1267 static void bond_attach_slave(struct bonding *bond, struct slave *new_slave)
1269 if (bond->first_slave == NULL) { /* attaching the first slave */
1270 new_slave->next = new_slave;
1271 new_slave->prev = new_slave;
1272 bond->first_slave = new_slave;
1274 new_slave->next = bond->first_slave;
1275 new_slave->prev = bond->first_slave->prev;
1276 new_slave->next->prev = new_slave;
1277 new_slave->prev->next = new_slave;
1284 * This function detaches the slave from the list.
1285 * WARNING: no check is made to verify if the slave effectively
1286 * belongs to <bond>.
1287 * Nothing is freed on return, structures are just unchained.
1288 * If any slave pointer in bond was pointing to <slave>,
1289 * it should be changed by the calling function.
1291 * bond->lock held for writing by caller.
1293 static void bond_detach_slave(struct bonding *bond, struct slave *slave)
1296 slave->next->prev = slave->prev;
1299 slave->prev->next = slave->next;
1301 if (bond->first_slave == slave) { /* slave is the first slave */
1302 if (bond->slave_cnt > 1) { /* there are more slave */
1303 bond->first_slave = slave->next;
1305 bond->first_slave = NULL; /* slave was the last one */
1314 #ifdef CONFIG_NET_POLL_CONTROLLER
1316 * You must hold read lock on bond->lock before calling this.
1318 static bool slaves_support_netpoll(struct net_device *bond_dev)
1320 struct bonding *bond = netdev_priv(bond_dev);
1321 struct slave *slave;
1325 bond_for_each_slave(bond, slave, i) {
1326 if ((slave->dev->priv_flags & IFF_DISABLE_NETPOLL) ||
1327 !slave->dev->netdev_ops->ndo_poll_controller)
1330 return i != 0 && ret;
1333 static void bond_poll_controller(struct net_device *bond_dev)
1335 struct net_device *dev = bond_dev->npinfo->netpoll->real_dev;
1336 if (dev != bond_dev)
1337 netpoll_poll_dev(dev);
1340 static void bond_netpoll_cleanup(struct net_device *bond_dev)
1342 struct bonding *bond = netdev_priv(bond_dev);
1343 struct slave *slave;
1344 const struct net_device_ops *ops;
1347 read_lock(&bond->lock);
1348 bond_dev->npinfo = NULL;
1349 bond_for_each_slave(bond, slave, i) {
1351 ops = slave->dev->netdev_ops;
1352 if (ops->ndo_netpoll_cleanup)
1353 ops->ndo_netpoll_cleanup(slave->dev);
1355 slave->dev->npinfo = NULL;
1358 read_unlock(&bond->lock);
1363 static void bond_netpoll_cleanup(struct net_device *bond_dev)
1369 /*---------------------------------- IOCTL ----------------------------------*/
1371 static int bond_sethwaddr(struct net_device *bond_dev,
1372 struct net_device *slave_dev)
1374 pr_debug("bond_dev=%p\n", bond_dev);
1375 pr_debug("slave_dev=%p\n", slave_dev);
1376 pr_debug("slave_dev->addr_len=%d\n", slave_dev->addr_len);
1377 memcpy(bond_dev->dev_addr, slave_dev->dev_addr, slave_dev->addr_len);
1381 #define BOND_VLAN_FEATURES \
1382 (NETIF_F_VLAN_CHALLENGED | NETIF_F_HW_VLAN_RX | NETIF_F_HW_VLAN_TX | \
1383 NETIF_F_HW_VLAN_FILTER)
1386 * Compute the common dev->feature set available to all slaves. Some
1387 * feature bits are managed elsewhere, so preserve those feature bits
1388 * on the master device.
1390 static int bond_compute_features(struct bonding *bond)
1392 struct slave *slave;
1393 struct net_device *bond_dev = bond->dev;
1394 unsigned long features = bond_dev->features;
1395 unsigned long vlan_features = 0;
1396 unsigned short max_hard_header_len = max((u16)ETH_HLEN,
1397 bond_dev->hard_header_len);
1400 features &= ~(NETIF_F_ALL_CSUM | BOND_VLAN_FEATURES);
1401 features |= NETIF_F_GSO_MASK | NETIF_F_NO_CSUM;
1403 if (!bond->first_slave)
1406 features &= ~NETIF_F_ONE_FOR_ALL;
1408 vlan_features = bond->first_slave->dev->vlan_features;
1409 bond_for_each_slave(bond, slave, i) {
1410 features = netdev_increment_features(features,
1411 slave->dev->features,
1412 NETIF_F_ONE_FOR_ALL);
1413 vlan_features = netdev_increment_features(vlan_features,
1414 slave->dev->vlan_features,
1415 NETIF_F_ONE_FOR_ALL);
1416 if (slave->dev->hard_header_len > max_hard_header_len)
1417 max_hard_header_len = slave->dev->hard_header_len;
1421 features |= (bond_dev->features & BOND_VLAN_FEATURES);
1422 bond_dev->features = netdev_fix_features(features, NULL);
1423 bond_dev->vlan_features = netdev_fix_features(vlan_features, NULL);
1424 bond_dev->hard_header_len = max_hard_header_len;
1429 static void bond_setup_by_slave(struct net_device *bond_dev,
1430 struct net_device *slave_dev)
1432 struct bonding *bond = netdev_priv(bond_dev);
1434 bond_dev->header_ops = slave_dev->header_ops;
1436 bond_dev->type = slave_dev->type;
1437 bond_dev->hard_header_len = slave_dev->hard_header_len;
1438 bond_dev->addr_len = slave_dev->addr_len;
1440 memcpy(bond_dev->broadcast, slave_dev->broadcast,
1441 slave_dev->addr_len);
1442 bond->setup_by_slave = 1;
1445 /* enslave device <slave> to bond device <master> */
1446 int bond_enslave(struct net_device *bond_dev, struct net_device *slave_dev)
1448 struct bonding *bond = netdev_priv(bond_dev);
1449 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
1450 struct slave *new_slave = NULL;
1451 struct netdev_hw_addr *ha;
1452 struct sockaddr addr;
1454 int old_features = bond_dev->features;
1457 if (!bond->params.use_carrier && slave_dev->ethtool_ops == NULL &&
1458 slave_ops->ndo_do_ioctl == NULL) {
1459 pr_warning("%s: Warning: no link monitoring support for %s\n",
1460 bond_dev->name, slave_dev->name);
1463 /* bond must be initialized by bond_open() before enslaving */
1464 if (!(bond_dev->flags & IFF_UP)) {
1465 pr_warning("%s: master_dev is not up in bond_enslave\n",
1469 /* already enslaved */
1470 if (slave_dev->flags & IFF_SLAVE) {
1471 pr_debug("Error, Device was already enslaved\n");
1475 /* vlan challenged mutual exclusion */
1476 /* no need to lock since we're protected by rtnl_lock */
1477 if (slave_dev->features & NETIF_F_VLAN_CHALLENGED) {
1478 pr_debug("%s: NETIF_F_VLAN_CHALLENGED\n", slave_dev->name);
1480 pr_err("%s: Error: cannot enslave VLAN challenged slave %s on VLAN enabled bond %s\n",
1481 bond_dev->name, slave_dev->name, bond_dev->name);
1484 pr_warning("%s: Warning: enslaved VLAN challenged slave %s. Adding VLANs will be blocked as long as %s is part of bond %s\n",
1485 bond_dev->name, slave_dev->name,
1486 slave_dev->name, bond_dev->name);
1487 bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
1490 pr_debug("%s: ! NETIF_F_VLAN_CHALLENGED\n", slave_dev->name);
1491 if (bond->slave_cnt == 0) {
1492 /* First slave, and it is not VLAN challenged,
1493 * so remove the block of adding VLANs over the bond.
1495 bond_dev->features &= ~NETIF_F_VLAN_CHALLENGED;
1500 * Old ifenslave binaries are no longer supported. These can
1501 * be identified with moderate accuracy by the state of the slave:
1502 * the current ifenslave will set the interface down prior to
1503 * enslaving it; the old ifenslave will not.
1505 if ((slave_dev->flags & IFF_UP)) {
1506 pr_err("%s is up. This may be due to an out of date ifenslave.\n",
1509 goto err_undo_flags;
1512 /* set bonding device ether type by slave - bonding netdevices are
1513 * created with ether_setup, so when the slave type is not ARPHRD_ETHER
1514 * there is a need to override some of the type dependent attribs/funcs.
1516 * bond ether type mutual exclusion - don't allow slaves of dissimilar
1517 * ether type (eg ARPHRD_ETHER and ARPHRD_INFINIBAND) share the same bond
1519 if (bond->slave_cnt == 0) {
1520 if (bond_dev->type != slave_dev->type) {
1521 pr_debug("%s: change device type from %d to %d\n",
1523 bond_dev->type, slave_dev->type);
1525 res = netdev_bonding_change(bond_dev,
1526 NETDEV_PRE_TYPE_CHANGE);
1527 res = notifier_to_errno(res);
1529 pr_err("%s: refused to change device type\n",
1532 goto err_undo_flags;
1535 /* Flush unicast and multicast addresses */
1536 dev_uc_flush(bond_dev);
1537 dev_mc_flush(bond_dev);
1539 if (slave_dev->type != ARPHRD_ETHER)
1540 bond_setup_by_slave(bond_dev, slave_dev);
1542 ether_setup(bond_dev);
1544 netdev_bonding_change(bond_dev,
1545 NETDEV_POST_TYPE_CHANGE);
1547 } else if (bond_dev->type != slave_dev->type) {
1548 pr_err("%s ether type (%d) is different from other slaves (%d), can not enslave it.\n",
1550 slave_dev->type, bond_dev->type);
1552 goto err_undo_flags;
1555 if (slave_ops->ndo_set_mac_address == NULL) {
1556 if (bond->slave_cnt == 0) {
1557 pr_warning("%s: Warning: The first slave device specified does not support setting the MAC address. Setting fail_over_mac to active.",
1559 bond->params.fail_over_mac = BOND_FOM_ACTIVE;
1560 } else if (bond->params.fail_over_mac != BOND_FOM_ACTIVE) {
1561 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",
1564 goto err_undo_flags;
1568 /* If this is the first slave, then we need to set the master's hardware
1569 * address to be the same as the slave's. */
1570 if (bond->slave_cnt == 0)
1571 memcpy(bond->dev->dev_addr, slave_dev->dev_addr,
1572 slave_dev->addr_len);
1575 new_slave = kzalloc(sizeof(struct slave), GFP_KERNEL);
1578 goto err_undo_flags;
1582 * Set the new_slave's queue_id to be zero. Queue ID mapping
1583 * is set via sysfs or module option if desired.
1585 new_slave->queue_id = 0;
1587 /* Save slave's original mtu and then set it to match the bond */
1588 new_slave->original_mtu = slave_dev->mtu;
1589 res = dev_set_mtu(slave_dev, bond->dev->mtu);
1591 pr_debug("Error %d calling dev_set_mtu\n", res);
1596 * Save slave's original ("permanent") mac address for modes
1597 * that need it, and for restoring it upon release, and then
1598 * set it to the master's address
1600 memcpy(new_slave->perm_hwaddr, slave_dev->dev_addr, ETH_ALEN);
1602 if (!bond->params.fail_over_mac) {
1604 * Set slave to master's mac address. The application already
1605 * set the master's mac address to that of the first slave
1607 memcpy(addr.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
1608 addr.sa_family = slave_dev->type;
1609 res = dev_set_mac_address(slave_dev, &addr);
1611 pr_debug("Error %d calling set_mac_address\n", res);
1612 goto err_restore_mtu;
1616 res = netdev_set_master(slave_dev, bond_dev);
1618 pr_debug("Error %d calling netdev_set_master\n", res);
1619 goto err_restore_mac;
1621 /* open the slave since the application closed it */
1622 res = dev_open(slave_dev);
1624 pr_debug("Opening slave %s failed\n", slave_dev->name);
1625 goto err_unset_master;
1628 new_slave->dev = slave_dev;
1629 slave_dev->priv_flags |= IFF_BONDING;
1631 if (bond_is_lb(bond)) {
1632 /* bond_alb_init_slave() must be called before all other stages since
1633 * it might fail and we do not want to have to undo everything
1635 res = bond_alb_init_slave(bond, new_slave);
1640 /* If the mode USES_PRIMARY, then the new slave gets the
1641 * master's promisc (and mc) settings only if it becomes the
1642 * curr_active_slave, and that is taken care of later when calling
1643 * bond_change_active()
1645 if (!USES_PRIMARY(bond->params.mode)) {
1646 /* set promiscuity level to new slave */
1647 if (bond_dev->flags & IFF_PROMISC) {
1648 res = dev_set_promiscuity(slave_dev, 1);
1653 /* set allmulti level to new slave */
1654 if (bond_dev->flags & IFF_ALLMULTI) {
1655 res = dev_set_allmulti(slave_dev, 1);
1660 netif_addr_lock_bh(bond_dev);
1661 /* upload master's mc_list to new slave */
1662 netdev_for_each_mc_addr(ha, bond_dev)
1663 dev_mc_add(slave_dev, ha->addr);
1664 netif_addr_unlock_bh(bond_dev);
1667 if (bond->params.mode == BOND_MODE_8023AD) {
1668 /* add lacpdu mc addr to mc list */
1669 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
1671 dev_mc_add(slave_dev, lacpdu_multicast);
1674 bond_add_vlans_on_slave(bond, slave_dev);
1676 write_lock_bh(&bond->lock);
1678 bond_attach_slave(bond, new_slave);
1680 new_slave->delay = 0;
1681 new_slave->link_failure_count = 0;
1683 bond_compute_features(bond);
1685 write_unlock_bh(&bond->lock);
1687 read_lock(&bond->lock);
1689 new_slave->last_arp_rx = jiffies;
1691 if (bond->params.miimon && !bond->params.use_carrier) {
1692 link_reporting = bond_check_dev_link(bond, slave_dev, 1);
1694 if ((link_reporting == -1) && !bond->params.arp_interval) {
1696 * miimon is set but a bonded network driver
1697 * does not support ETHTOOL/MII and
1698 * arp_interval is not set. Note: if
1699 * use_carrier is enabled, we will never go
1700 * here (because netif_carrier is always
1701 * supported); thus, we don't need to change
1702 * the messages for netif_carrier.
1704 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",
1705 bond_dev->name, slave_dev->name);
1706 } else if (link_reporting == -1) {
1707 /* unable get link status using mii/ethtool */
1708 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",
1709 bond_dev->name, slave_dev->name);
1713 /* check for initial state */
1714 if (!bond->params.miimon ||
1715 (bond_check_dev_link(bond, slave_dev, 0) == BMSR_LSTATUS)) {
1716 if (bond->params.updelay) {
1717 pr_debug("Initial state of slave_dev is BOND_LINK_BACK\n");
1718 new_slave->link = BOND_LINK_BACK;
1719 new_slave->delay = bond->params.updelay;
1721 pr_debug("Initial state of slave_dev is BOND_LINK_UP\n");
1722 new_slave->link = BOND_LINK_UP;
1724 new_slave->jiffies = jiffies;
1726 pr_debug("Initial state of slave_dev is BOND_LINK_DOWN\n");
1727 new_slave->link = BOND_LINK_DOWN;
1730 if (bond_update_speed_duplex(new_slave) &&
1731 (new_slave->link != BOND_LINK_DOWN)) {
1732 pr_warning("%s: Warning: failed to get speed and duplex from %s, assumed to be 100Mb/sec and Full.\n",
1733 bond_dev->name, new_slave->dev->name);
1735 if (bond->params.mode == BOND_MODE_8023AD) {
1736 pr_warning("%s: Warning: Operation of 802.3ad mode requires ETHTOOL support in base driver for proper aggregator selection.\n",
1741 if (USES_PRIMARY(bond->params.mode) && bond->params.primary[0]) {
1742 /* if there is a primary slave, remember it */
1743 if (strcmp(bond->params.primary, new_slave->dev->name) == 0) {
1744 bond->primary_slave = new_slave;
1745 bond->force_primary = true;
1749 write_lock_bh(&bond->curr_slave_lock);
1751 switch (bond->params.mode) {
1752 case BOND_MODE_ACTIVEBACKUP:
1753 bond_set_slave_inactive_flags(new_slave);
1754 bond_select_active_slave(bond);
1756 case BOND_MODE_8023AD:
1757 /* in 802.3ad mode, the internal mechanism
1758 * will activate the slaves in the selected
1761 bond_set_slave_inactive_flags(new_slave);
1762 /* if this is the first slave */
1763 if (bond->slave_cnt == 1) {
1764 SLAVE_AD_INFO(new_slave).id = 1;
1765 /* Initialize AD with the number of times that the AD timer is called in 1 second
1766 * can be called only after the mac address of the bond is set
1768 bond_3ad_initialize(bond, 1000/AD_TIMER_INTERVAL,
1769 bond->params.lacp_fast);
1771 SLAVE_AD_INFO(new_slave).id =
1772 SLAVE_AD_INFO(new_slave->prev).id + 1;
1775 bond_3ad_bind_slave(new_slave);
1779 new_slave->state = BOND_STATE_ACTIVE;
1780 bond_set_slave_inactive_flags(new_slave);
1781 bond_select_active_slave(bond);
1784 pr_debug("This slave is always active in trunk mode\n");
1786 /* always active in trunk mode */
1787 new_slave->state = BOND_STATE_ACTIVE;
1789 /* In trunking mode there is little meaning to curr_active_slave
1790 * anyway (it holds no special properties of the bond device),
1791 * so we can change it without calling change_active_interface()
1793 if (!bond->curr_active_slave)
1794 bond->curr_active_slave = new_slave;
1797 } /* switch(bond_mode) */
1799 write_unlock_bh(&bond->curr_slave_lock);
1801 bond_set_carrier(bond);
1803 #ifdef CONFIG_NET_POLL_CONTROLLER
1805 * Netpoll and bonding is broken, make sure it is not initialized
1806 * until it is fixed.
1808 if (disable_netpoll) {
1809 bond_dev->priv_flags |= IFF_DISABLE_NETPOLL;
1811 if (slaves_support_netpoll(bond_dev)) {
1812 bond_dev->priv_flags &= ~IFF_DISABLE_NETPOLL;
1813 if (bond_dev->npinfo)
1814 slave_dev->npinfo = bond_dev->npinfo;
1815 } else if (!(bond_dev->priv_flags & IFF_DISABLE_NETPOLL)) {
1816 bond_dev->priv_flags |= IFF_DISABLE_NETPOLL;
1817 pr_info("New slave device %s does not support netpoll\n",
1819 pr_info("Disabling netpoll support for %s\n", bond_dev->name);
1823 read_unlock(&bond->lock);
1825 res = bond_create_slave_symlinks(bond_dev, slave_dev);
1829 pr_info("%s: enslaving %s as a%s interface with a%s link.\n",
1830 bond_dev->name, slave_dev->name,
1831 new_slave->state == BOND_STATE_ACTIVE ? "n active" : " backup",
1832 new_slave->link != BOND_LINK_DOWN ? "n up" : " down");
1834 /* enslave is successful */
1837 /* Undo stages on error */
1839 dev_close(slave_dev);
1842 netdev_set_master(slave_dev, NULL);
1845 if (!bond->params.fail_over_mac) {
1846 /* XXX TODO - fom follow mode needs to change master's
1847 * MAC if this slave's MAC is in use by the bond, or at
1848 * least print a warning.
1850 memcpy(addr.sa_data, new_slave->perm_hwaddr, ETH_ALEN);
1851 addr.sa_family = slave_dev->type;
1852 dev_set_mac_address(slave_dev, &addr);
1856 dev_set_mtu(slave_dev, new_slave->original_mtu);
1862 bond_dev->features = old_features;
1868 * Try to release the slave device <slave> from the bond device <master>
1869 * It is legal to access curr_active_slave without a lock because all the function
1872 * The rules for slave state should be:
1873 * for Active/Backup:
1874 * Active stays on all backups go down
1875 * for Bonded connections:
1876 * The first up interface should be left on and all others downed.
1878 int bond_release(struct net_device *bond_dev, struct net_device *slave_dev)
1880 struct bonding *bond = netdev_priv(bond_dev);
1881 struct slave *slave, *oldcurrent;
1882 struct sockaddr addr;
1884 /* slave is not a slave or master is not master of this slave */
1885 if (!(slave_dev->flags & IFF_SLAVE) ||
1886 (slave_dev->master != bond_dev)) {
1887 pr_err("%s: Error: cannot release %s.\n",
1888 bond_dev->name, slave_dev->name);
1892 netdev_bonding_change(bond_dev, NETDEV_BONDING_DESLAVE);
1893 write_lock_bh(&bond->lock);
1895 slave = bond_get_slave_by_dev(bond, slave_dev);
1897 /* not a slave of this bond */
1898 pr_info("%s: %s not enslaved\n",
1899 bond_dev->name, slave_dev->name);
1900 write_unlock_bh(&bond->lock);
1904 if (!bond->params.fail_over_mac) {
1905 if (!compare_ether_addr(bond_dev->dev_addr, slave->perm_hwaddr) &&
1906 bond->slave_cnt > 1)
1907 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",
1908 bond_dev->name, slave_dev->name,
1910 bond_dev->name, slave_dev->name);
1913 /* Inform AD package of unbinding of slave. */
1914 if (bond->params.mode == BOND_MODE_8023AD) {
1915 /* must be called before the slave is
1916 * detached from the list
1918 bond_3ad_unbind_slave(slave);
1921 pr_info("%s: releasing %s interface %s\n",
1923 (slave->state == BOND_STATE_ACTIVE) ? "active" : "backup",
1926 oldcurrent = bond->curr_active_slave;
1928 bond->current_arp_slave = NULL;
1930 /* release the slave from its bond */
1931 bond_detach_slave(bond, slave);
1933 bond_compute_features(bond);
1935 if (bond->primary_slave == slave)
1936 bond->primary_slave = NULL;
1938 if (oldcurrent == slave)
1939 bond_change_active_slave(bond, NULL);
1941 if (bond_is_lb(bond)) {
1942 /* Must be called only after the slave has been
1943 * detached from the list and the curr_active_slave
1944 * has been cleared (if our_slave == old_current),
1945 * but before a new active slave is selected.
1947 write_unlock_bh(&bond->lock);
1948 bond_alb_deinit_slave(bond, slave);
1949 write_lock_bh(&bond->lock);
1952 if (oldcurrent == slave) {
1954 * Note that we hold RTNL over this sequence, so there
1955 * is no concern that another slave add/remove event
1958 write_unlock_bh(&bond->lock);
1959 read_lock(&bond->lock);
1960 write_lock_bh(&bond->curr_slave_lock);
1962 bond_select_active_slave(bond);
1964 write_unlock_bh(&bond->curr_slave_lock);
1965 read_unlock(&bond->lock);
1966 write_lock_bh(&bond->lock);
1969 if (bond->slave_cnt == 0) {
1970 bond_set_carrier(bond);
1972 /* if the last slave was removed, zero the mac address
1973 * of the master so it will be set by the application
1974 * to the mac address of the first slave
1976 memset(bond_dev->dev_addr, 0, bond_dev->addr_len);
1979 bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
1981 pr_warning("%s: Warning: clearing HW address of %s while it still has VLANs.\n",
1982 bond_dev->name, bond_dev->name);
1983 pr_warning("%s: When re-adding slaves, make sure the bond's HW address matches its VLANs'.\n",
1986 } else if ((bond_dev->features & NETIF_F_VLAN_CHALLENGED) &&
1987 !bond_has_challenged_slaves(bond)) {
1988 pr_info("%s: last VLAN challenged slave %s left bond %s. VLAN blocking is removed\n",
1989 bond_dev->name, slave_dev->name, bond_dev->name);
1990 bond_dev->features &= ~NETIF_F_VLAN_CHALLENGED;
1993 write_unlock_bh(&bond->lock);
1995 /* must do this from outside any spinlocks */
1996 bond_destroy_slave_symlinks(bond_dev, slave_dev);
1998 bond_del_vlans_from_slave(bond, slave_dev);
2000 /* If the mode USES_PRIMARY, then we should only remove its
2001 * promisc and mc settings if it was the curr_active_slave, but that was
2002 * already taken care of above when we detached the slave
2004 if (!USES_PRIMARY(bond->params.mode)) {
2005 /* unset promiscuity level from slave */
2006 if (bond_dev->flags & IFF_PROMISC)
2007 dev_set_promiscuity(slave_dev, -1);
2009 /* unset allmulti level from slave */
2010 if (bond_dev->flags & IFF_ALLMULTI)
2011 dev_set_allmulti(slave_dev, -1);
2013 /* flush master's mc_list from slave */
2014 netif_addr_lock_bh(bond_dev);
2015 bond_mc_list_flush(bond_dev, slave_dev);
2016 netif_addr_unlock_bh(bond_dev);
2019 netdev_set_master(slave_dev, NULL);
2021 #ifdef CONFIG_NET_POLL_CONTROLLER
2022 read_lock_bh(&bond->lock);
2024 /* Make sure netpoll over stays disabled until fixed. */
2025 if (!disable_netpoll)
2026 if (slaves_support_netpoll(bond_dev))
2027 bond_dev->priv_flags &= ~IFF_DISABLE_NETPOLL;
2028 read_unlock_bh(&bond->lock);
2029 if (slave_dev->netdev_ops->ndo_netpoll_cleanup)
2030 slave_dev->netdev_ops->ndo_netpoll_cleanup(slave_dev);
2032 slave_dev->npinfo = NULL;
2035 /* close slave before restoring its mac address */
2036 dev_close(slave_dev);
2038 if (bond->params.fail_over_mac != BOND_FOM_ACTIVE) {
2039 /* restore original ("permanent") mac address */
2040 memcpy(addr.sa_data, slave->perm_hwaddr, ETH_ALEN);
2041 addr.sa_family = slave_dev->type;
2042 dev_set_mac_address(slave_dev, &addr);
2045 dev_set_mtu(slave_dev, slave->original_mtu);
2047 slave_dev->priv_flags &= ~(IFF_MASTER_8023AD | IFF_MASTER_ALB |
2048 IFF_SLAVE_INACTIVE | IFF_BONDING |
2053 return 0; /* deletion OK */
2057 * First release a slave and than destroy the bond if no more slaves are left.
2058 * Must be under rtnl_lock when this function is called.
2060 int bond_release_and_destroy(struct net_device *bond_dev,
2061 struct net_device *slave_dev)
2063 struct bonding *bond = netdev_priv(bond_dev);
2066 ret = bond_release(bond_dev, slave_dev);
2067 if ((ret == 0) && (bond->slave_cnt == 0)) {
2068 pr_info("%s: destroying bond %s.\n",
2069 bond_dev->name, bond_dev->name);
2070 unregister_netdevice(bond_dev);
2076 * This function releases all slaves.
2078 static int bond_release_all(struct net_device *bond_dev)
2080 struct bonding *bond = netdev_priv(bond_dev);
2081 struct slave *slave;
2082 struct net_device *slave_dev;
2083 struct sockaddr addr;
2085 write_lock_bh(&bond->lock);
2087 netif_carrier_off(bond_dev);
2089 if (bond->slave_cnt == 0)
2092 bond->current_arp_slave = NULL;
2093 bond->primary_slave = NULL;
2094 bond_change_active_slave(bond, NULL);
2096 while ((slave = bond->first_slave) != NULL) {
2097 /* Inform AD package of unbinding of slave
2098 * before slave is detached from the list.
2100 if (bond->params.mode == BOND_MODE_8023AD)
2101 bond_3ad_unbind_slave(slave);
2103 slave_dev = slave->dev;
2104 bond_detach_slave(bond, slave);
2106 /* now that the slave is detached, unlock and perform
2107 * all the undo steps that should not be called from
2110 write_unlock_bh(&bond->lock);
2112 if (bond_is_lb(bond)) {
2113 /* must be called only after the slave
2114 * has been detached from the list
2116 bond_alb_deinit_slave(bond, slave);
2119 bond_compute_features(bond);
2121 bond_destroy_slave_symlinks(bond_dev, slave_dev);
2122 bond_del_vlans_from_slave(bond, slave_dev);
2124 /* If the mode USES_PRIMARY, then we should only remove its
2125 * promisc and mc settings if it was the curr_active_slave, but that was
2126 * already taken care of above when we detached the slave
2128 if (!USES_PRIMARY(bond->params.mode)) {
2129 /* unset promiscuity level from slave */
2130 if (bond_dev->flags & IFF_PROMISC)
2131 dev_set_promiscuity(slave_dev, -1);
2133 /* unset allmulti level from slave */
2134 if (bond_dev->flags & IFF_ALLMULTI)
2135 dev_set_allmulti(slave_dev, -1);
2137 /* flush master's mc_list from slave */
2138 netif_addr_lock_bh(bond_dev);
2139 bond_mc_list_flush(bond_dev, slave_dev);
2140 netif_addr_unlock_bh(bond_dev);
2143 netdev_set_master(slave_dev, NULL);
2145 /* close slave before restoring its mac address */
2146 dev_close(slave_dev);
2148 if (!bond->params.fail_over_mac) {
2149 /* restore original ("permanent") mac address*/
2150 memcpy(addr.sa_data, slave->perm_hwaddr, ETH_ALEN);
2151 addr.sa_family = slave_dev->type;
2152 dev_set_mac_address(slave_dev, &addr);
2155 slave_dev->priv_flags &= ~(IFF_MASTER_8023AD | IFF_MASTER_ALB |
2156 IFF_SLAVE_INACTIVE);
2160 /* re-acquire the lock before getting the next slave */
2161 write_lock_bh(&bond->lock);
2164 /* zero the mac address of the master so it will be
2165 * set by the application to the mac address of the
2168 memset(bond_dev->dev_addr, 0, bond_dev->addr_len);
2171 bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
2173 pr_warning("%s: Warning: clearing HW address of %s while it still has VLANs.\n",
2174 bond_dev->name, bond_dev->name);
2175 pr_warning("%s: When re-adding slaves, make sure the bond's HW address matches its VLANs'.\n",
2179 pr_info("%s: released all slaves\n", bond_dev->name);
2182 write_unlock_bh(&bond->lock);
2188 * This function changes the active slave to slave <slave_dev>.
2189 * It returns -EINVAL in the following cases.
2190 * - <slave_dev> is not found in the list.
2191 * - There is not active slave now.
2192 * - <slave_dev> is already active.
2193 * - The link state of <slave_dev> is not BOND_LINK_UP.
2194 * - <slave_dev> is not running.
2195 * In these cases, this function does nothing.
2196 * In the other cases, current_slave pointer is changed and 0 is returned.
2198 static int bond_ioctl_change_active(struct net_device *bond_dev, struct net_device *slave_dev)
2200 struct bonding *bond = netdev_priv(bond_dev);
2201 struct slave *old_active = NULL;
2202 struct slave *new_active = NULL;
2205 if (!USES_PRIMARY(bond->params.mode))
2208 /* Verify that master_dev is indeed the master of slave_dev */
2209 if (!(slave_dev->flags & IFF_SLAVE) || (slave_dev->master != bond_dev))
2212 read_lock(&bond->lock);
2214 read_lock(&bond->curr_slave_lock);
2215 old_active = bond->curr_active_slave;
2216 read_unlock(&bond->curr_slave_lock);
2218 new_active = bond_get_slave_by_dev(bond, slave_dev);
2221 * Changing to the current active: do nothing; return success.
2223 if (new_active && (new_active == old_active)) {
2224 read_unlock(&bond->lock);
2230 (new_active->link == BOND_LINK_UP) &&
2231 IS_UP(new_active->dev)) {
2232 write_lock_bh(&bond->curr_slave_lock);
2233 bond_change_active_slave(bond, new_active);
2234 write_unlock_bh(&bond->curr_slave_lock);
2238 read_unlock(&bond->lock);
2243 static int bond_info_query(struct net_device *bond_dev, struct ifbond *info)
2245 struct bonding *bond = netdev_priv(bond_dev);
2247 info->bond_mode = bond->params.mode;
2248 info->miimon = bond->params.miimon;
2250 read_lock(&bond->lock);
2251 info->num_slaves = bond->slave_cnt;
2252 read_unlock(&bond->lock);
2257 static int bond_slave_info_query(struct net_device *bond_dev, struct ifslave *info)
2259 struct bonding *bond = netdev_priv(bond_dev);
2260 struct slave *slave;
2261 int i, res = -ENODEV;
2263 read_lock(&bond->lock);
2265 bond_for_each_slave(bond, slave, i) {
2266 if (i == (int)info->slave_id) {
2268 strcpy(info->slave_name, slave->dev->name);
2269 info->link = slave->link;
2270 info->state = slave->state;
2271 info->link_failure_count = slave->link_failure_count;
2276 read_unlock(&bond->lock);
2281 /*-------------------------------- Monitoring -------------------------------*/
2284 static int bond_miimon_inspect(struct bonding *bond)
2286 struct slave *slave;
2287 int i, link_state, commit = 0;
2288 bool ignore_updelay;
2290 ignore_updelay = !bond->curr_active_slave ? true : false;
2292 bond_for_each_slave(bond, slave, i) {
2293 slave->new_link = BOND_LINK_NOCHANGE;
2295 link_state = bond_check_dev_link(bond, slave->dev, 0);
2297 switch (slave->link) {
2302 slave->link = BOND_LINK_FAIL;
2303 slave->delay = bond->params.downdelay;
2305 pr_info("%s: link status down for %sinterface %s, disabling it in %d ms.\n",
2307 (bond->params.mode ==
2308 BOND_MODE_ACTIVEBACKUP) ?
2309 ((slave->state == BOND_STATE_ACTIVE) ?
2310 "active " : "backup ") : "",
2312 bond->params.downdelay * bond->params.miimon);
2315 case BOND_LINK_FAIL:
2318 * recovered before downdelay expired
2320 slave->link = BOND_LINK_UP;
2321 slave->jiffies = jiffies;
2322 pr_info("%s: link status up again after %d ms for interface %s.\n",
2324 (bond->params.downdelay - slave->delay) *
2325 bond->params.miimon,
2330 if (slave->delay <= 0) {
2331 slave->new_link = BOND_LINK_DOWN;
2339 case BOND_LINK_DOWN:
2343 slave->link = BOND_LINK_BACK;
2344 slave->delay = bond->params.updelay;
2347 pr_info("%s: link status up for interface %s, enabling it in %d ms.\n",
2348 bond->dev->name, slave->dev->name,
2349 ignore_updelay ? 0 :
2350 bond->params.updelay *
2351 bond->params.miimon);
2354 case BOND_LINK_BACK:
2356 slave->link = BOND_LINK_DOWN;
2357 pr_info("%s: link status down again after %d ms for interface %s.\n",
2359 (bond->params.updelay - slave->delay) *
2360 bond->params.miimon,
2369 if (slave->delay <= 0) {
2370 slave->new_link = BOND_LINK_UP;
2372 ignore_updelay = false;
2384 static void bond_miimon_commit(struct bonding *bond)
2386 struct slave *slave;
2389 bond_for_each_slave(bond, slave, i) {
2390 switch (slave->new_link) {
2391 case BOND_LINK_NOCHANGE:
2395 slave->link = BOND_LINK_UP;
2396 slave->jiffies = jiffies;
2398 if (bond->params.mode == BOND_MODE_8023AD) {
2399 /* prevent it from being the active one */
2400 slave->state = BOND_STATE_BACKUP;
2401 } else if (bond->params.mode != BOND_MODE_ACTIVEBACKUP) {
2402 /* make it immediately active */
2403 slave->state = BOND_STATE_ACTIVE;
2404 } else if (slave != bond->primary_slave) {
2405 /* prevent it from being the active one */
2406 slave->state = BOND_STATE_BACKUP;
2409 bond_update_speed_duplex(slave);
2411 pr_info("%s: link status definitely up for interface %s, %d Mbps %s duplex.\n",
2412 bond->dev->name, slave->dev->name,
2413 slave->speed, slave->duplex ? "full" : "half");
2415 /* notify ad that the link status has changed */
2416 if (bond->params.mode == BOND_MODE_8023AD)
2417 bond_3ad_handle_link_change(slave, BOND_LINK_UP);
2419 if (bond_is_lb(bond))
2420 bond_alb_handle_link_change(bond, slave,
2423 if (!bond->curr_active_slave ||
2424 (slave == bond->primary_slave))
2429 case BOND_LINK_DOWN:
2430 if (slave->link_failure_count < UINT_MAX)
2431 slave->link_failure_count++;
2433 slave->link = BOND_LINK_DOWN;
2435 if (bond->params.mode == BOND_MODE_ACTIVEBACKUP ||
2436 bond->params.mode == BOND_MODE_8023AD)
2437 bond_set_slave_inactive_flags(slave);
2439 pr_info("%s: link status definitely down for interface %s, disabling it\n",
2440 bond->dev->name, slave->dev->name);
2442 if (bond->params.mode == BOND_MODE_8023AD)
2443 bond_3ad_handle_link_change(slave,
2446 if (bond_is_lb(bond))
2447 bond_alb_handle_link_change(bond, slave,
2450 if (slave == bond->curr_active_slave)
2456 pr_err("%s: invalid new link %d on slave %s\n",
2457 bond->dev->name, slave->new_link,
2459 slave->new_link = BOND_LINK_NOCHANGE;
2466 write_lock_bh(&bond->curr_slave_lock);
2467 bond_select_active_slave(bond);
2468 write_unlock_bh(&bond->curr_slave_lock);
2471 bond_set_carrier(bond);
2477 * Really a wrapper that splits the mii monitor into two phases: an
2478 * inspection, then (if inspection indicates something needs to be done)
2479 * an acquisition of appropriate locks followed by a commit phase to
2480 * implement whatever link state changes are indicated.
2482 void bond_mii_monitor(struct work_struct *work)
2484 struct bonding *bond = container_of(work, struct bonding,
2487 read_lock(&bond->lock);
2488 if (bond->kill_timers)
2491 if (bond->slave_cnt == 0)
2494 if (bond->send_grat_arp) {
2495 read_lock(&bond->curr_slave_lock);
2496 bond_send_gratuitous_arp(bond);
2497 read_unlock(&bond->curr_slave_lock);
2500 if (bond->send_unsol_na) {
2501 read_lock(&bond->curr_slave_lock);
2502 bond_send_unsolicited_na(bond);
2503 read_unlock(&bond->curr_slave_lock);
2506 if (bond_miimon_inspect(bond)) {
2507 read_unlock(&bond->lock);
2509 read_lock(&bond->lock);
2511 bond_miimon_commit(bond);
2513 read_unlock(&bond->lock);
2514 rtnl_unlock(); /* might sleep, hold no other locks */
2515 read_lock(&bond->lock);
2519 if (bond->params.miimon)
2520 queue_delayed_work(bond->wq, &bond->mii_work,
2521 msecs_to_jiffies(bond->params.miimon));
2523 read_unlock(&bond->lock);
2526 static __be32 bond_glean_dev_ip(struct net_device *dev)
2528 struct in_device *idev;
2529 struct in_ifaddr *ifa;
2536 idev = __in_dev_get_rcu(dev);
2540 ifa = idev->ifa_list;
2544 addr = ifa->ifa_local;
2550 static int bond_has_this_ip(struct bonding *bond, __be32 ip)
2552 struct vlan_entry *vlan;
2554 if (ip == bond->master_ip)
2557 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
2558 if (ip == vlan->vlan_ip)
2566 * We go to the (large) trouble of VLAN tagging ARP frames because
2567 * switches in VLAN mode (especially if ports are configured as
2568 * "native" to a VLAN) might not pass non-tagged frames.
2570 static void bond_arp_send(struct net_device *slave_dev, int arp_op, __be32 dest_ip, __be32 src_ip, unsigned short vlan_id)
2572 struct sk_buff *skb;
2574 pr_debug("arp %d on slave %s: dst %x src %x vid %d\n", arp_op,
2575 slave_dev->name, dest_ip, src_ip, vlan_id);
2577 skb = arp_create(arp_op, ETH_P_ARP, dest_ip, slave_dev, src_ip,
2578 NULL, slave_dev->dev_addr, NULL);
2581 pr_err("ARP packet allocation failed\n");
2585 skb = vlan_put_tag(skb, vlan_id);
2587 pr_err("failed to insert VLAN tag\n");
2595 static void bond_arp_send_all(struct bonding *bond, struct slave *slave)
2598 __be32 *targets = bond->params.arp_targets;
2599 struct vlan_entry *vlan;
2600 struct net_device *vlan_dev;
2604 for (i = 0; (i < BOND_MAX_ARP_TARGETS); i++) {
2607 pr_debug("basa: target %x\n", targets[i]);
2609 pr_debug("basa: empty vlan: arp_send\n");
2610 bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2611 bond->master_ip, 0);
2616 * If VLANs are configured, we do a route lookup to
2617 * determine which VLAN interface would be used, so we
2618 * can tag the ARP with the proper VLAN tag.
2620 memset(&fl, 0, sizeof(fl));
2621 fl.fl4_dst = targets[i];
2622 fl.fl4_tos = RTO_ONLINK;
2624 rv = ip_route_output_key(dev_net(bond->dev), &rt, &fl);
2626 if (net_ratelimit()) {
2627 pr_warning("%s: no route to arp_ip_target %pI4\n",
2628 bond->dev->name, &fl.fl4_dst);
2634 * This target is not on a VLAN
2636 if (rt->dst.dev == bond->dev) {
2638 pr_debug("basa: rtdev == bond->dev: arp_send\n");
2639 bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2640 bond->master_ip, 0);
2645 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
2646 vlan_dev = vlan_group_get_device(bond->vlgrp, vlan->vlan_id);
2647 if (vlan_dev == rt->dst.dev) {
2648 vlan_id = vlan->vlan_id;
2649 pr_debug("basa: vlan match on %s %d\n",
2650 vlan_dev->name, vlan_id);
2657 bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2658 vlan->vlan_ip, vlan_id);
2662 if (net_ratelimit()) {
2663 pr_warning("%s: no path to arp_ip_target %pI4 via rt.dev %s\n",
2664 bond->dev->name, &fl.fl4_dst,
2665 rt->dst.dev ? rt->dst.dev->name : "NULL");
2672 * Kick out a gratuitous ARP for an IP on the bonding master plus one
2673 * for each VLAN above us.
2675 * Caller must hold curr_slave_lock for read or better
2677 static void bond_send_gratuitous_arp(struct bonding *bond)
2679 struct slave *slave = bond->curr_active_slave;
2680 struct vlan_entry *vlan;
2681 struct net_device *vlan_dev;
2683 pr_debug("bond_send_grat_arp: bond %s slave %s\n",
2684 bond->dev->name, slave ? slave->dev->name : "NULL");
2686 if (!slave || !bond->send_grat_arp ||
2687 test_bit(__LINK_STATE_LINKWATCH_PENDING, &slave->dev->state))
2690 bond->send_grat_arp--;
2692 if (bond->master_ip) {
2693 bond_arp_send(slave->dev, ARPOP_REPLY, bond->master_ip,
2694 bond->master_ip, 0);
2700 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
2701 vlan_dev = vlan_group_get_device(bond->vlgrp, vlan->vlan_id);
2702 if (vlan->vlan_ip) {
2703 bond_arp_send(slave->dev, ARPOP_REPLY, vlan->vlan_ip,
2704 vlan->vlan_ip, vlan->vlan_id);
2709 static void bond_validate_arp(struct bonding *bond, struct slave *slave, __be32 sip, __be32 tip)
2712 __be32 *targets = bond->params.arp_targets;
2714 for (i = 0; (i < BOND_MAX_ARP_TARGETS) && targets[i]; i++) {
2715 pr_debug("bva: sip %pI4 tip %pI4 t[%d] %pI4 bhti(tip) %d\n",
2716 &sip, &tip, i, &targets[i],
2717 bond_has_this_ip(bond, tip));
2718 if (sip == targets[i]) {
2719 if (bond_has_this_ip(bond, tip))
2720 slave->last_arp_rx = jiffies;
2726 static int bond_arp_rcv(struct sk_buff *skb, struct net_device *dev, struct packet_type *pt, struct net_device *orig_dev)
2729 struct slave *slave;
2730 struct bonding *bond;
2731 unsigned char *arp_ptr;
2734 if (dev->priv_flags & IFF_802_1Q_VLAN) {
2736 * When using VLANS and bonding, dev and oriv_dev may be
2737 * incorrect if the physical interface supports VLAN
2738 * acceleration. With this change ARP validation now
2739 * works for hosts only reachable on the VLAN interface.
2741 dev = vlan_dev_real_dev(dev);
2742 orig_dev = dev_get_by_index_rcu(dev_net(skb->dev),skb->skb_iif);
2745 if (!(dev->priv_flags & IFF_BONDING) || !(dev->flags & IFF_MASTER))
2748 bond = netdev_priv(dev);
2749 read_lock(&bond->lock);
2751 pr_debug("bond_arp_rcv: bond %s skb->dev %s orig_dev %s\n",
2752 bond->dev->name, skb->dev ? skb->dev->name : "NULL",
2753 orig_dev ? orig_dev->name : "NULL");
2755 slave = bond_get_slave_by_dev(bond, orig_dev);
2756 if (!slave || !slave_do_arp_validate(bond, slave))
2759 if (!pskb_may_pull(skb, arp_hdr_len(dev)))
2763 if (arp->ar_hln != dev->addr_len ||
2764 skb->pkt_type == PACKET_OTHERHOST ||
2765 skb->pkt_type == PACKET_LOOPBACK ||
2766 arp->ar_hrd != htons(ARPHRD_ETHER) ||
2767 arp->ar_pro != htons(ETH_P_IP) ||
2771 arp_ptr = (unsigned char *)(arp + 1);
2772 arp_ptr += dev->addr_len;
2773 memcpy(&sip, arp_ptr, 4);
2774 arp_ptr += 4 + dev->addr_len;
2775 memcpy(&tip, arp_ptr, 4);
2777 pr_debug("bond_arp_rcv: %s %s/%d av %d sv %d sip %pI4 tip %pI4\n",
2778 bond->dev->name, slave->dev->name, slave->state,
2779 bond->params.arp_validate, slave_do_arp_validate(bond, slave),
2783 * Backup slaves won't see the ARP reply, but do come through
2784 * here for each ARP probe (so we swap the sip/tip to validate
2785 * the probe). In a "redundant switch, common router" type of
2786 * configuration, the ARP probe will (hopefully) travel from
2787 * the active, through one switch, the router, then the other
2788 * switch before reaching the backup.
2790 if (slave->state == BOND_STATE_ACTIVE)
2791 bond_validate_arp(bond, slave, sip, tip);
2793 bond_validate_arp(bond, slave, tip, sip);
2796 read_unlock(&bond->lock);
2799 return NET_RX_SUCCESS;
2803 * this function is called regularly to monitor each slave's link
2804 * ensuring that traffic is being sent and received when arp monitoring
2805 * is used in load-balancing mode. if the adapter has been dormant, then an
2806 * arp is transmitted to generate traffic. see activebackup_arp_monitor for
2807 * arp monitoring in active backup mode.
2809 void bond_loadbalance_arp_mon(struct work_struct *work)
2811 struct bonding *bond = container_of(work, struct bonding,
2813 struct slave *slave, *oldcurrent;
2814 int do_failover = 0;
2818 read_lock(&bond->lock);
2820 delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
2822 if (bond->kill_timers)
2825 if (bond->slave_cnt == 0)
2828 read_lock(&bond->curr_slave_lock);
2829 oldcurrent = bond->curr_active_slave;
2830 read_unlock(&bond->curr_slave_lock);
2832 /* see if any of the previous devices are up now (i.e. they have
2833 * xmt and rcv traffic). the curr_active_slave does not come into
2834 * the picture unless it is null. also, slave->jiffies is not needed
2835 * here because we send an arp on each slave and give a slave as
2836 * long as it needs to get the tx/rx within the delta.
2837 * TODO: what about up/down delay in arp mode? it wasn't here before
2840 bond_for_each_slave(bond, slave, i) {
2841 unsigned long trans_start = dev_trans_start(slave->dev);
2843 if (slave->link != BOND_LINK_UP) {
2844 if (time_in_range(jiffies,
2845 trans_start - delta_in_ticks,
2846 trans_start + delta_in_ticks) &&
2847 time_in_range(jiffies,
2848 slave->dev->last_rx - delta_in_ticks,
2849 slave->dev->last_rx + delta_in_ticks)) {
2851 slave->link = BOND_LINK_UP;
2852 slave->state = BOND_STATE_ACTIVE;
2854 /* primary_slave has no meaning in round-robin
2855 * mode. the window of a slave being up and
2856 * curr_active_slave being null after enslaving
2860 pr_info("%s: link status definitely up for interface %s, ",
2865 pr_info("%s: interface %s is now up\n",
2871 /* slave->link == BOND_LINK_UP */
2873 /* not all switches will respond to an arp request
2874 * when the source ip is 0, so don't take the link down
2875 * if we don't know our ip yet
2877 if (!time_in_range(jiffies,
2878 trans_start - delta_in_ticks,
2879 trans_start + 2 * delta_in_ticks) ||
2880 !time_in_range(jiffies,
2881 slave->dev->last_rx - delta_in_ticks,
2882 slave->dev->last_rx + 2 * delta_in_ticks)) {
2884 slave->link = BOND_LINK_DOWN;
2885 slave->state = BOND_STATE_BACKUP;
2887 if (slave->link_failure_count < UINT_MAX)
2888 slave->link_failure_count++;
2890 pr_info("%s: interface %s is now down.\n",
2894 if (slave == oldcurrent)
2899 /* note: if switch is in round-robin mode, all links
2900 * must tx arp to ensure all links rx an arp - otherwise
2901 * links may oscillate or not come up at all; if switch is
2902 * in something like xor mode, there is nothing we can
2903 * do - all replies will be rx'ed on same link causing slaves
2904 * to be unstable during low/no traffic periods
2906 if (IS_UP(slave->dev))
2907 bond_arp_send_all(bond, slave);
2911 write_lock_bh(&bond->curr_slave_lock);
2913 bond_select_active_slave(bond);
2915 write_unlock_bh(&bond->curr_slave_lock);
2919 if (bond->params.arp_interval)
2920 queue_delayed_work(bond->wq, &bond->arp_work, delta_in_ticks);
2922 read_unlock(&bond->lock);
2926 * Called to inspect slaves for active-backup mode ARP monitor link state
2927 * changes. Sets new_link in slaves to specify what action should take
2928 * place for the slave. Returns 0 if no changes are found, >0 if changes
2929 * to link states must be committed.
2931 * Called with bond->lock held for read.
2933 static int bond_ab_arp_inspect(struct bonding *bond, int delta_in_ticks)
2935 struct slave *slave;
2937 unsigned long trans_start;
2939 bond_for_each_slave(bond, slave, i) {
2940 slave->new_link = BOND_LINK_NOCHANGE;
2942 if (slave->link != BOND_LINK_UP) {
2943 if (time_in_range(jiffies,
2944 slave_last_rx(bond, slave) - delta_in_ticks,
2945 slave_last_rx(bond, slave) + delta_in_ticks)) {
2947 slave->new_link = BOND_LINK_UP;
2955 * Give slaves 2*delta after being enslaved or made
2956 * active. This avoids bouncing, as the last receive
2957 * times need a full ARP monitor cycle to be updated.
2959 if (time_in_range(jiffies,
2960 slave->jiffies - delta_in_ticks,
2961 slave->jiffies + 2 * delta_in_ticks))
2965 * Backup slave is down if:
2966 * - No current_arp_slave AND
2967 * - more than 3*delta since last receive AND
2968 * - the bond has an IP address
2970 * Note: a non-null current_arp_slave indicates
2971 * the curr_active_slave went down and we are
2972 * searching for a new one; under this condition
2973 * we only take the curr_active_slave down - this
2974 * gives each slave a chance to tx/rx traffic
2975 * before being taken out
2977 if (slave->state == BOND_STATE_BACKUP &&
2978 !bond->current_arp_slave &&
2979 !time_in_range(jiffies,
2980 slave_last_rx(bond, slave) - delta_in_ticks,
2981 slave_last_rx(bond, slave) + 3 * delta_in_ticks)) {
2983 slave->new_link = BOND_LINK_DOWN;
2988 * Active slave is down if:
2989 * - more than 2*delta since transmitting OR
2990 * - (more than 2*delta since receive AND
2991 * the bond has an IP address)
2993 trans_start = dev_trans_start(slave->dev);
2994 if ((slave->state == BOND_STATE_ACTIVE) &&
2995 (!time_in_range(jiffies,
2996 trans_start - delta_in_ticks,
2997 trans_start + 2 * delta_in_ticks) ||
2998 !time_in_range(jiffies,
2999 slave_last_rx(bond, slave) - delta_in_ticks,
3000 slave_last_rx(bond, slave) + 2 * delta_in_ticks))) {
3002 slave->new_link = BOND_LINK_DOWN;
3011 * Called to commit link state changes noted by inspection step of
3012 * active-backup mode ARP monitor.
3014 * Called with RTNL and bond->lock for read.
3016 static void bond_ab_arp_commit(struct bonding *bond, int delta_in_ticks)
3018 struct slave *slave;
3020 unsigned long trans_start;
3022 bond_for_each_slave(bond, slave, i) {
3023 switch (slave->new_link) {
3024 case BOND_LINK_NOCHANGE:
3028 trans_start = dev_trans_start(slave->dev);
3029 if ((!bond->curr_active_slave &&
3030 time_in_range(jiffies,
3031 trans_start - delta_in_ticks,
3032 trans_start + delta_in_ticks)) ||
3033 bond->curr_active_slave != slave) {
3034 slave->link = BOND_LINK_UP;
3035 bond->current_arp_slave = NULL;
3037 pr_info("%s: link status definitely up for interface %s.\n",
3038 bond->dev->name, slave->dev->name);
3040 if (!bond->curr_active_slave ||
3041 (slave == bond->primary_slave))
3048 case BOND_LINK_DOWN:
3049 if (slave->link_failure_count < UINT_MAX)
3050 slave->link_failure_count++;
3052 slave->link = BOND_LINK_DOWN;
3053 bond_set_slave_inactive_flags(slave);
3055 pr_info("%s: link status definitely down for interface %s, disabling it\n",
3056 bond->dev->name, slave->dev->name);
3058 if (slave == bond->curr_active_slave) {
3059 bond->current_arp_slave = NULL;
3066 pr_err("%s: impossible: new_link %d on slave %s\n",
3067 bond->dev->name, slave->new_link,
3074 write_lock_bh(&bond->curr_slave_lock);
3075 bond_select_active_slave(bond);
3076 write_unlock_bh(&bond->curr_slave_lock);
3079 bond_set_carrier(bond);
3083 * Send ARP probes for active-backup mode ARP monitor.
3085 * Called with bond->lock held for read.
3087 static void bond_ab_arp_probe(struct bonding *bond)
3089 struct slave *slave;
3092 read_lock(&bond->curr_slave_lock);
3094 if (bond->current_arp_slave && bond->curr_active_slave)
3095 pr_info("PROBE: c_arp %s && cas %s BAD\n",
3096 bond->current_arp_slave->dev->name,
3097 bond->curr_active_slave->dev->name);
3099 if (bond->curr_active_slave) {
3100 bond_arp_send_all(bond, bond->curr_active_slave);
3101 read_unlock(&bond->curr_slave_lock);
3105 read_unlock(&bond->curr_slave_lock);
3107 /* if we don't have a curr_active_slave, search for the next available
3108 * backup slave from the current_arp_slave and make it the candidate
3109 * for becoming the curr_active_slave
3112 if (!bond->current_arp_slave) {
3113 bond->current_arp_slave = bond->first_slave;
3114 if (!bond->current_arp_slave)
3118 bond_set_slave_inactive_flags(bond->current_arp_slave);
3120 /* search for next candidate */
3121 bond_for_each_slave_from(bond, slave, i, bond->current_arp_slave->next) {
3122 if (IS_UP(slave->dev)) {
3123 slave->link = BOND_LINK_BACK;
3124 bond_set_slave_active_flags(slave);
3125 bond_arp_send_all(bond, slave);
3126 slave->jiffies = jiffies;
3127 bond->current_arp_slave = slave;
3131 /* if the link state is up at this point, we
3132 * mark it down - this can happen if we have
3133 * simultaneous link failures and
3134 * reselect_active_interface doesn't make this
3135 * one the current slave so it is still marked
3136 * up when it is actually down
3138 if (slave->link == BOND_LINK_UP) {
3139 slave->link = BOND_LINK_DOWN;
3140 if (slave->link_failure_count < UINT_MAX)
3141 slave->link_failure_count++;
3143 bond_set_slave_inactive_flags(slave);
3145 pr_info("%s: backup interface %s is now down.\n",
3146 bond->dev->name, slave->dev->name);
3151 void bond_activebackup_arp_mon(struct work_struct *work)
3153 struct bonding *bond = container_of(work, struct bonding,
3157 read_lock(&bond->lock);
3159 if (bond->kill_timers)
3162 delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
3164 if (bond->slave_cnt == 0)
3167 if (bond->send_grat_arp) {
3168 read_lock(&bond->curr_slave_lock);
3169 bond_send_gratuitous_arp(bond);
3170 read_unlock(&bond->curr_slave_lock);
3173 if (bond->send_unsol_na) {
3174 read_lock(&bond->curr_slave_lock);
3175 bond_send_unsolicited_na(bond);
3176 read_unlock(&bond->curr_slave_lock);
3179 if (bond_ab_arp_inspect(bond, delta_in_ticks)) {
3180 read_unlock(&bond->lock);
3182 read_lock(&bond->lock);
3184 bond_ab_arp_commit(bond, delta_in_ticks);
3186 read_unlock(&bond->lock);
3188 read_lock(&bond->lock);
3191 bond_ab_arp_probe(bond);
3194 if (bond->params.arp_interval)
3195 queue_delayed_work(bond->wq, &bond->arp_work, delta_in_ticks);
3197 read_unlock(&bond->lock);
3200 /*------------------------------ proc/seq_file-------------------------------*/
3202 #ifdef CONFIG_PROC_FS
3204 static void *bond_info_seq_start(struct seq_file *seq, loff_t *pos)
3205 __acquires(&dev_base_lock)
3206 __acquires(&bond->lock)
3208 struct bonding *bond = seq->private;
3210 struct slave *slave;
3213 /* make sure the bond won't be taken away */
3214 read_lock(&dev_base_lock);
3215 read_lock(&bond->lock);
3218 return SEQ_START_TOKEN;
3220 bond_for_each_slave(bond, slave, i) {
3228 static void *bond_info_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3230 struct bonding *bond = seq->private;
3231 struct slave *slave = v;
3234 if (v == SEQ_START_TOKEN)
3235 return bond->first_slave;
3237 slave = slave->next;
3239 return (slave == bond->first_slave) ? NULL : slave;
3242 static void bond_info_seq_stop(struct seq_file *seq, void *v)
3243 __releases(&bond->lock)
3244 __releases(&dev_base_lock)
3246 struct bonding *bond = seq->private;
3248 read_unlock(&bond->lock);
3249 read_unlock(&dev_base_lock);
3252 static void bond_info_show_master(struct seq_file *seq)
3254 struct bonding *bond = seq->private;
3258 read_lock(&bond->curr_slave_lock);
3259 curr = bond->curr_active_slave;
3260 read_unlock(&bond->curr_slave_lock);
3262 seq_printf(seq, "Bonding Mode: %s",
3263 bond_mode_name(bond->params.mode));
3265 if (bond->params.mode == BOND_MODE_ACTIVEBACKUP &&
3266 bond->params.fail_over_mac)
3267 seq_printf(seq, " (fail_over_mac %s)",
3268 fail_over_mac_tbl[bond->params.fail_over_mac].modename);
3270 seq_printf(seq, "\n");
3272 if (bond->params.mode == BOND_MODE_XOR ||
3273 bond->params.mode == BOND_MODE_8023AD) {
3274 seq_printf(seq, "Transmit Hash Policy: %s (%d)\n",
3275 xmit_hashtype_tbl[bond->params.xmit_policy].modename,
3276 bond->params.xmit_policy);
3279 if (USES_PRIMARY(bond->params.mode)) {
3280 seq_printf(seq, "Primary Slave: %s",
3281 (bond->primary_slave) ?
3282 bond->primary_slave->dev->name : "None");
3283 if (bond->primary_slave)
3284 seq_printf(seq, " (primary_reselect %s)",
3285 pri_reselect_tbl[bond->params.primary_reselect].modename);
3287 seq_printf(seq, "\nCurrently Active Slave: %s\n",
3288 (curr) ? curr->dev->name : "None");
3291 seq_printf(seq, "MII Status: %s\n", netif_carrier_ok(bond->dev) ?
3293 seq_printf(seq, "MII Polling Interval (ms): %d\n", bond->params.miimon);
3294 seq_printf(seq, "Up Delay (ms): %d\n",
3295 bond->params.updelay * bond->params.miimon);
3296 seq_printf(seq, "Down Delay (ms): %d\n",
3297 bond->params.downdelay * bond->params.miimon);
3300 /* ARP information */
3301 if (bond->params.arp_interval > 0) {
3303 seq_printf(seq, "ARP Polling Interval (ms): %d\n",
3304 bond->params.arp_interval);
3306 seq_printf(seq, "ARP IP target/s (n.n.n.n form):");
3308 for (i = 0; (i < BOND_MAX_ARP_TARGETS); i++) {
3309 if (!bond->params.arp_targets[i])
3312 seq_printf(seq, ",");
3313 seq_printf(seq, " %pI4", &bond->params.arp_targets[i]);
3316 seq_printf(seq, "\n");
3319 if (bond->params.mode == BOND_MODE_8023AD) {
3320 struct ad_info ad_info;
3322 seq_puts(seq, "\n802.3ad info\n");
3323 seq_printf(seq, "LACP rate: %s\n",
3324 (bond->params.lacp_fast) ? "fast" : "slow");
3325 seq_printf(seq, "Aggregator selection policy (ad_select): %s\n",
3326 ad_select_tbl[bond->params.ad_select].modename);
3328 if (bond_3ad_get_active_agg_info(bond, &ad_info)) {
3329 seq_printf(seq, "bond %s has no active aggregator\n",
3332 seq_printf(seq, "Active Aggregator Info:\n");
3334 seq_printf(seq, "\tAggregator ID: %d\n",
3335 ad_info.aggregator_id);
3336 seq_printf(seq, "\tNumber of ports: %d\n",
3338 seq_printf(seq, "\tActor Key: %d\n",
3340 seq_printf(seq, "\tPartner Key: %d\n",
3341 ad_info.partner_key);
3342 seq_printf(seq, "\tPartner Mac Address: %pM\n",
3343 ad_info.partner_system);
3348 static void bond_info_show_slave(struct seq_file *seq,
3349 const struct slave *slave)
3351 struct bonding *bond = seq->private;
3353 seq_printf(seq, "\nSlave Interface: %s\n", slave->dev->name);
3354 seq_printf(seq, "MII Status: %s\n",
3355 (slave->link == BOND_LINK_UP) ? "up" : "down");
3356 seq_printf(seq, "Speed: %d Mbps\n", slave->speed);
3357 seq_printf(seq, "Duplex: %s\n", slave->duplex ? "full" : "half");
3358 seq_printf(seq, "Link Failure Count: %u\n",
3359 slave->link_failure_count);
3361 seq_printf(seq, "Permanent HW addr: %pM\n", slave->perm_hwaddr);
3363 if (bond->params.mode == BOND_MODE_8023AD) {
3364 const struct aggregator *agg
3365 = SLAVE_AD_INFO(slave).port.aggregator;
3368 seq_printf(seq, "Aggregator ID: %d\n",
3369 agg->aggregator_identifier);
3371 seq_puts(seq, "Aggregator ID: N/A\n");
3373 seq_printf(seq, "Slave queue ID: %d\n", slave->queue_id);
3376 static int bond_info_seq_show(struct seq_file *seq, void *v)
3378 if (v == SEQ_START_TOKEN) {
3379 seq_printf(seq, "%s\n", version);
3380 bond_info_show_master(seq);
3382 bond_info_show_slave(seq, v);
3387 static const struct seq_operations bond_info_seq_ops = {
3388 .start = bond_info_seq_start,
3389 .next = bond_info_seq_next,
3390 .stop = bond_info_seq_stop,
3391 .show = bond_info_seq_show,
3394 static int bond_info_open(struct inode *inode, struct file *file)
3396 struct seq_file *seq;
3397 struct proc_dir_entry *proc;
3400 res = seq_open(file, &bond_info_seq_ops);
3402 /* recover the pointer buried in proc_dir_entry data */
3403 seq = file->private_data;
3405 seq->private = proc->data;
3411 static const struct file_operations bond_info_fops = {
3412 .owner = THIS_MODULE,
3413 .open = bond_info_open,
3415 .llseek = seq_lseek,
3416 .release = seq_release,
3419 static void bond_create_proc_entry(struct bonding *bond)
3421 struct net_device *bond_dev = bond->dev;
3422 struct bond_net *bn = net_generic(dev_net(bond_dev), bond_net_id);
3425 bond->proc_entry = proc_create_data(bond_dev->name,
3426 S_IRUGO, bn->proc_dir,
3427 &bond_info_fops, bond);
3428 if (bond->proc_entry == NULL)
3429 pr_warning("Warning: Cannot create /proc/net/%s/%s\n",
3430 DRV_NAME, bond_dev->name);
3432 memcpy(bond->proc_file_name, bond_dev->name, IFNAMSIZ);
3436 static void bond_remove_proc_entry(struct bonding *bond)
3438 struct net_device *bond_dev = bond->dev;
3439 struct bond_net *bn = net_generic(dev_net(bond_dev), bond_net_id);
3441 if (bn->proc_dir && bond->proc_entry) {
3442 remove_proc_entry(bond->proc_file_name, bn->proc_dir);
3443 memset(bond->proc_file_name, 0, IFNAMSIZ);
3444 bond->proc_entry = NULL;
3448 /* Create the bonding directory under /proc/net, if doesn't exist yet.
3449 * Caller must hold rtnl_lock.
3451 static void __net_init bond_create_proc_dir(struct bond_net *bn)
3453 if (!bn->proc_dir) {
3454 bn->proc_dir = proc_mkdir(DRV_NAME, bn->net->proc_net);
3456 pr_warning("Warning: cannot create /proc/net/%s\n",
3461 /* Destroy the bonding directory under /proc/net, if empty.
3462 * Caller must hold rtnl_lock.
3464 static void __net_exit bond_destroy_proc_dir(struct bond_net *bn)
3467 remove_proc_entry(DRV_NAME, bn->net->proc_net);
3468 bn->proc_dir = NULL;
3472 #else /* !CONFIG_PROC_FS */
3474 static void bond_create_proc_entry(struct bonding *bond)
3478 static void bond_remove_proc_entry(struct bonding *bond)
3482 static inline void bond_create_proc_dir(struct bond_net *bn)
3486 static inline void bond_destroy_proc_dir(struct bond_net *bn)
3490 #endif /* CONFIG_PROC_FS */
3493 /*-------------------------- netdev event handling --------------------------*/
3496 * Change device name
3498 static int bond_event_changename(struct bonding *bond)
3500 bond_remove_proc_entry(bond);
3501 bond_create_proc_entry(bond);
3506 static int bond_master_netdev_event(unsigned long event,
3507 struct net_device *bond_dev)
3509 struct bonding *event_bond = netdev_priv(bond_dev);
3512 case NETDEV_CHANGENAME:
3513 return bond_event_changename(event_bond);
3521 static int bond_slave_netdev_event(unsigned long event,
3522 struct net_device *slave_dev)
3524 struct net_device *bond_dev = slave_dev->master;
3525 struct bonding *bond = netdev_priv(bond_dev);
3528 case NETDEV_UNREGISTER:
3530 if (bond->setup_by_slave)
3531 bond_release_and_destroy(bond_dev, slave_dev);
3533 bond_release(bond_dev, slave_dev);
3537 if (bond->params.mode == BOND_MODE_8023AD || bond_is_lb(bond)) {
3538 struct slave *slave;
3540 slave = bond_get_slave_by_dev(bond, slave_dev);
3542 u16 old_speed = slave->speed;
3543 u16 old_duplex = slave->duplex;
3545 bond_update_speed_duplex(slave);
3547 if (bond_is_lb(bond))
3550 if (old_speed != slave->speed)
3551 bond_3ad_adapter_speed_changed(slave);
3552 if (old_duplex != slave->duplex)
3553 bond_3ad_adapter_duplex_changed(slave);
3560 * ... Or is it this?
3563 case NETDEV_CHANGEMTU:
3565 * TODO: Should slaves be allowed to
3566 * independently alter their MTU? For
3567 * an active-backup bond, slaves need
3568 * not be the same type of device, so
3569 * MTUs may vary. For other modes,
3570 * slaves arguably should have the
3571 * same MTUs. To do this, we'd need to
3572 * take over the slave's change_mtu
3573 * function for the duration of their
3577 case NETDEV_CHANGENAME:
3579 * TODO: handle changing the primary's name
3582 case NETDEV_FEAT_CHANGE:
3583 bond_compute_features(bond);
3593 * bond_netdev_event: handle netdev notifier chain events.
3595 * This function receives events for the netdev chain. The caller (an
3596 * ioctl handler calling blocking_notifier_call_chain) holds the necessary
3597 * locks for us to safely manipulate the slave devices (RTNL lock,
3600 static int bond_netdev_event(struct notifier_block *this,
3601 unsigned long event, void *ptr)
3603 struct net_device *event_dev = (struct net_device *)ptr;
3605 pr_debug("event_dev: %s, event: %lx\n",
3606 event_dev ? event_dev->name : "None",
3609 if (!(event_dev->priv_flags & IFF_BONDING))
3612 if (event_dev->flags & IFF_MASTER) {
3613 pr_debug("IFF_MASTER\n");
3614 return bond_master_netdev_event(event, event_dev);
3617 if (event_dev->flags & IFF_SLAVE) {
3618 pr_debug("IFF_SLAVE\n");
3619 return bond_slave_netdev_event(event, event_dev);
3626 * bond_inetaddr_event: handle inetaddr notifier chain events.
3628 * We keep track of device IPs primarily to use as source addresses in
3629 * ARP monitor probes (rather than spewing out broadcasts all the time).
3631 * We track one IP for the main device (if it has one), plus one per VLAN.
3633 static int bond_inetaddr_event(struct notifier_block *this, unsigned long event, void *ptr)
3635 struct in_ifaddr *ifa = ptr;
3636 struct net_device *vlan_dev, *event_dev = ifa->ifa_dev->dev;
3637 struct bond_net *bn = net_generic(dev_net(event_dev), bond_net_id);
3638 struct bonding *bond;
3639 struct vlan_entry *vlan;
3641 list_for_each_entry(bond, &bn->dev_list, bond_list) {
3642 if (bond->dev == event_dev) {
3645 bond->master_ip = ifa->ifa_local;
3648 bond->master_ip = bond_glean_dev_ip(bond->dev);
3655 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
3658 vlan_dev = vlan_group_get_device(bond->vlgrp, vlan->vlan_id);
3659 if (vlan_dev == event_dev) {
3662 vlan->vlan_ip = ifa->ifa_local;
3666 bond_glean_dev_ip(vlan_dev);
3677 static struct notifier_block bond_netdev_notifier = {
3678 .notifier_call = bond_netdev_event,
3681 static struct notifier_block bond_inetaddr_notifier = {
3682 .notifier_call = bond_inetaddr_event,
3685 /*-------------------------- Packet type handling ---------------------------*/
3687 /* register to receive lacpdus on a bond */
3688 static void bond_register_lacpdu(struct bonding *bond)
3690 struct packet_type *pk_type = &(BOND_AD_INFO(bond).ad_pkt_type);
3692 /* initialize packet type */
3693 pk_type->type = PKT_TYPE_LACPDU;
3694 pk_type->dev = bond->dev;
3695 pk_type->func = bond_3ad_lacpdu_recv;
3697 dev_add_pack(pk_type);
3700 /* unregister to receive lacpdus on a bond */
3701 static void bond_unregister_lacpdu(struct bonding *bond)
3703 dev_remove_pack(&(BOND_AD_INFO(bond).ad_pkt_type));
3706 void bond_register_arp(struct bonding *bond)
3708 struct packet_type *pt = &bond->arp_mon_pt;
3713 pt->type = htons(ETH_P_ARP);
3714 pt->dev = bond->dev;
3715 pt->func = bond_arp_rcv;
3719 void bond_unregister_arp(struct bonding *bond)
3721 struct packet_type *pt = &bond->arp_mon_pt;
3723 dev_remove_pack(pt);
3727 /*---------------------------- Hashing Policies -----------------------------*/
3730 * Hash for the output device based upon layer 2 and layer 3 data. If
3731 * the packet is not IP mimic bond_xmit_hash_policy_l2()
3733 static int bond_xmit_hash_policy_l23(struct sk_buff *skb, int count)
3735 struct ethhdr *data = (struct ethhdr *)skb->data;
3736 struct iphdr *iph = ip_hdr(skb);
3738 if (skb->protocol == htons(ETH_P_IP)) {
3739 return ((ntohl(iph->saddr ^ iph->daddr) & 0xffff) ^
3740 (data->h_dest[5] ^ data->h_source[5])) % count;
3743 return (data->h_dest[5] ^ data->h_source[5]) % count;
3747 * Hash for the output device based upon layer 3 and layer 4 data. If
3748 * the packet is a frag or not TCP or UDP, just use layer 3 data. If it is
3749 * altogether not IP, mimic bond_xmit_hash_policy_l2()
3751 static int bond_xmit_hash_policy_l34(struct sk_buff *skb, int count)
3753 struct ethhdr *data = (struct ethhdr *)skb->data;
3754 struct iphdr *iph = ip_hdr(skb);
3755 __be16 *layer4hdr = (__be16 *)((u32 *)iph + iph->ihl);
3758 if (skb->protocol == htons(ETH_P_IP)) {
3759 if (!(iph->frag_off & htons(IP_MF|IP_OFFSET)) &&
3760 (iph->protocol == IPPROTO_TCP ||
3761 iph->protocol == IPPROTO_UDP)) {
3762 layer4_xor = ntohs((*layer4hdr ^ *(layer4hdr + 1)));
3764 return (layer4_xor ^
3765 ((ntohl(iph->saddr ^ iph->daddr)) & 0xffff)) % count;
3769 return (data->h_dest[5] ^ data->h_source[5]) % count;
3773 * Hash for the output device based upon layer 2 data
3775 static int bond_xmit_hash_policy_l2(struct sk_buff *skb, int count)
3777 struct ethhdr *data = (struct ethhdr *)skb->data;
3779 return (data->h_dest[5] ^ data->h_source[5]) % count;
3782 /*-------------------------- Device entry points ----------------------------*/
3784 static int bond_open(struct net_device *bond_dev)
3786 struct bonding *bond = netdev_priv(bond_dev);
3788 bond->kill_timers = 0;
3790 INIT_DELAYED_WORK(&bond->mcast_work, bond_resend_igmp_join_requests_delayed);
3792 if (bond_is_lb(bond)) {
3793 /* bond_alb_initialize must be called before the timer
3796 if (bond_alb_initialize(bond, (bond->params.mode == BOND_MODE_ALB))) {
3797 /* something went wrong - fail the open operation */
3801 INIT_DELAYED_WORK(&bond->alb_work, bond_alb_monitor);
3802 queue_delayed_work(bond->wq, &bond->alb_work, 0);
3805 if (bond->params.miimon) { /* link check interval, in milliseconds. */
3806 INIT_DELAYED_WORK(&bond->mii_work, bond_mii_monitor);
3807 queue_delayed_work(bond->wq, &bond->mii_work, 0);
3810 if (bond->params.arp_interval) { /* arp interval, in milliseconds. */
3811 if (bond->params.mode == BOND_MODE_ACTIVEBACKUP)
3812 INIT_DELAYED_WORK(&bond->arp_work,
3813 bond_activebackup_arp_mon);
3815 INIT_DELAYED_WORK(&bond->arp_work,
3816 bond_loadbalance_arp_mon);
3818 queue_delayed_work(bond->wq, &bond->arp_work, 0);
3819 if (bond->params.arp_validate)
3820 bond_register_arp(bond);
3823 if (bond->params.mode == BOND_MODE_8023AD) {
3824 INIT_DELAYED_WORK(&bond->ad_work, bond_3ad_state_machine_handler);
3825 queue_delayed_work(bond->wq, &bond->ad_work, 0);
3826 /* register to receive LACPDUs */
3827 bond_register_lacpdu(bond);
3828 bond_3ad_initiate_agg_selection(bond, 1);
3834 static int bond_close(struct net_device *bond_dev)
3836 struct bonding *bond = netdev_priv(bond_dev);
3838 if (bond->params.mode == BOND_MODE_8023AD) {
3839 /* Unregister the receive of LACPDUs */
3840 bond_unregister_lacpdu(bond);
3843 if (bond->params.arp_validate)
3844 bond_unregister_arp(bond);
3846 write_lock_bh(&bond->lock);
3848 bond->send_grat_arp = 0;
3849 bond->send_unsol_na = 0;
3851 /* signal timers not to re-arm */
3852 bond->kill_timers = 1;
3854 write_unlock_bh(&bond->lock);
3856 if (bond->params.miimon) { /* link check interval, in milliseconds. */
3857 cancel_delayed_work(&bond->mii_work);
3860 if (bond->params.arp_interval) { /* arp interval, in milliseconds. */
3861 cancel_delayed_work(&bond->arp_work);
3864 switch (bond->params.mode) {
3865 case BOND_MODE_8023AD:
3866 cancel_delayed_work(&bond->ad_work);
3870 cancel_delayed_work(&bond->alb_work);
3876 if (delayed_work_pending(&bond->mcast_work))
3877 cancel_delayed_work(&bond->mcast_work);
3879 if (bond_is_lb(bond)) {
3880 /* Must be called only after all
3881 * slaves have been released
3883 bond_alb_deinitialize(bond);
3889 static struct rtnl_link_stats64 *bond_get_stats(struct net_device *bond_dev,
3890 struct rtnl_link_stats64 *stats)
3892 struct bonding *bond = netdev_priv(bond_dev);
3893 struct rtnl_link_stats64 temp;
3894 struct slave *slave;
3897 memset(stats, 0, sizeof(*stats));
3899 read_lock_bh(&bond->lock);
3901 bond_for_each_slave(bond, slave, i) {
3902 const struct rtnl_link_stats64 *sstats =
3903 dev_get_stats(slave->dev, &temp);
3905 stats->rx_packets += sstats->rx_packets;
3906 stats->rx_bytes += sstats->rx_bytes;
3907 stats->rx_errors += sstats->rx_errors;
3908 stats->rx_dropped += sstats->rx_dropped;
3910 stats->tx_packets += sstats->tx_packets;
3911 stats->tx_bytes += sstats->tx_bytes;
3912 stats->tx_errors += sstats->tx_errors;
3913 stats->tx_dropped += sstats->tx_dropped;
3915 stats->multicast += sstats->multicast;
3916 stats->collisions += sstats->collisions;
3918 stats->rx_length_errors += sstats->rx_length_errors;
3919 stats->rx_over_errors += sstats->rx_over_errors;
3920 stats->rx_crc_errors += sstats->rx_crc_errors;
3921 stats->rx_frame_errors += sstats->rx_frame_errors;
3922 stats->rx_fifo_errors += sstats->rx_fifo_errors;
3923 stats->rx_missed_errors += sstats->rx_missed_errors;
3925 stats->tx_aborted_errors += sstats->tx_aborted_errors;
3926 stats->tx_carrier_errors += sstats->tx_carrier_errors;
3927 stats->tx_fifo_errors += sstats->tx_fifo_errors;
3928 stats->tx_heartbeat_errors += sstats->tx_heartbeat_errors;
3929 stats->tx_window_errors += sstats->tx_window_errors;
3932 read_unlock_bh(&bond->lock);
3937 static int bond_do_ioctl(struct net_device *bond_dev, struct ifreq *ifr, int cmd)
3939 struct net_device *slave_dev = NULL;
3940 struct ifbond k_binfo;
3941 struct ifbond __user *u_binfo = NULL;
3942 struct ifslave k_sinfo;
3943 struct ifslave __user *u_sinfo = NULL;
3944 struct mii_ioctl_data *mii = NULL;
3947 pr_debug("bond_ioctl: master=%s, cmd=%d\n", bond_dev->name, cmd);
3959 * We do this again just in case we were called by SIOCGMIIREG
3960 * instead of SIOCGMIIPHY.
3967 if (mii->reg_num == 1) {
3968 struct bonding *bond = netdev_priv(bond_dev);
3970 read_lock(&bond->lock);
3971 read_lock(&bond->curr_slave_lock);
3972 if (netif_carrier_ok(bond->dev))
3973 mii->val_out = BMSR_LSTATUS;
3975 read_unlock(&bond->curr_slave_lock);
3976 read_unlock(&bond->lock);
3980 case BOND_INFO_QUERY_OLD:
3981 case SIOCBONDINFOQUERY:
3982 u_binfo = (struct ifbond __user *)ifr->ifr_data;
3984 if (copy_from_user(&k_binfo, u_binfo, sizeof(ifbond)))
3987 res = bond_info_query(bond_dev, &k_binfo);
3989 copy_to_user(u_binfo, &k_binfo, sizeof(ifbond)))
3993 case BOND_SLAVE_INFO_QUERY_OLD:
3994 case SIOCBONDSLAVEINFOQUERY:
3995 u_sinfo = (struct ifslave __user *)ifr->ifr_data;
3997 if (copy_from_user(&k_sinfo, u_sinfo, sizeof(ifslave)))
4000 res = bond_slave_info_query(bond_dev, &k_sinfo);
4002 copy_to_user(u_sinfo, &k_sinfo, sizeof(ifslave)))
4011 if (!capable(CAP_NET_ADMIN))
4014 slave_dev = dev_get_by_name(dev_net(bond_dev), ifr->ifr_slave);
4016 pr_debug("slave_dev=%p:\n", slave_dev);
4021 pr_debug("slave_dev->name=%s:\n", slave_dev->name);
4023 case BOND_ENSLAVE_OLD:
4024 case SIOCBONDENSLAVE:
4025 res = bond_enslave(bond_dev, slave_dev);
4027 case BOND_RELEASE_OLD:
4028 case SIOCBONDRELEASE:
4029 res = bond_release(bond_dev, slave_dev);
4031 case BOND_SETHWADDR_OLD:
4032 case SIOCBONDSETHWADDR:
4033 res = bond_sethwaddr(bond_dev, slave_dev);
4035 case BOND_CHANGE_ACTIVE_OLD:
4036 case SIOCBONDCHANGEACTIVE:
4037 res = bond_ioctl_change_active(bond_dev, slave_dev);
4049 static bool bond_addr_in_mc_list(unsigned char *addr,
4050 struct netdev_hw_addr_list *list,
4053 struct netdev_hw_addr *ha;
4055 netdev_hw_addr_list_for_each(ha, list)
4056 if (!memcmp(ha->addr, addr, addrlen))
4062 static void bond_set_multicast_list(struct net_device *bond_dev)
4064 struct bonding *bond = netdev_priv(bond_dev);
4065 struct netdev_hw_addr *ha;
4069 * Do promisc before checking multicast_mode
4071 if ((bond_dev->flags & IFF_PROMISC) && !(bond->flags & IFF_PROMISC))
4073 * FIXME: Need to handle the error when one of the multi-slaves
4076 bond_set_promiscuity(bond, 1);
4079 if (!(bond_dev->flags & IFF_PROMISC) && (bond->flags & IFF_PROMISC))
4080 bond_set_promiscuity(bond, -1);
4083 /* set allmulti flag to slaves */
4084 if ((bond_dev->flags & IFF_ALLMULTI) && !(bond->flags & IFF_ALLMULTI))
4086 * FIXME: Need to handle the error when one of the multi-slaves
4089 bond_set_allmulti(bond, 1);
4092 if (!(bond_dev->flags & IFF_ALLMULTI) && (bond->flags & IFF_ALLMULTI))
4093 bond_set_allmulti(bond, -1);
4096 read_lock(&bond->lock);
4098 bond->flags = bond_dev->flags;
4100 /* looking for addresses to add to slaves' mc list */
4101 netdev_for_each_mc_addr(ha, bond_dev) {
4102 found = bond_addr_in_mc_list(ha->addr, &bond->mc_list,
4103 bond_dev->addr_len);
4105 bond_mc_add(bond, ha->addr);
4108 /* looking for addresses to delete from slaves' list */
4109 netdev_hw_addr_list_for_each(ha, &bond->mc_list) {
4110 found = bond_addr_in_mc_list(ha->addr, &bond_dev->mc,
4111 bond_dev->addr_len);
4113 bond_mc_del(bond, ha->addr);
4116 /* save master's multicast list */
4117 __hw_addr_flush(&bond->mc_list);
4118 __hw_addr_add_multiple(&bond->mc_list, &bond_dev->mc,
4119 bond_dev->addr_len, NETDEV_HW_ADDR_T_MULTICAST);
4121 read_unlock(&bond->lock);
4124 static int bond_neigh_setup(struct net_device *dev, struct neigh_parms *parms)
4126 struct bonding *bond = netdev_priv(dev);
4127 struct slave *slave = bond->first_slave;
4130 const struct net_device_ops *slave_ops
4131 = slave->dev->netdev_ops;
4132 if (slave_ops->ndo_neigh_setup)
4133 return slave_ops->ndo_neigh_setup(slave->dev, parms);
4139 * Change the MTU of all of a master's slaves to match the master
4141 static int bond_change_mtu(struct net_device *bond_dev, int new_mtu)
4143 struct bonding *bond = netdev_priv(bond_dev);
4144 struct slave *slave, *stop_at;
4148 pr_debug("bond=%p, name=%s, new_mtu=%d\n", bond,
4149 (bond_dev ? bond_dev->name : "None"), new_mtu);
4151 /* Can't hold bond->lock with bh disabled here since
4152 * some base drivers panic. On the other hand we can't
4153 * hold bond->lock without bh disabled because we'll
4154 * deadlock. The only solution is to rely on the fact
4155 * that we're under rtnl_lock here, and the slaves
4156 * list won't change. This doesn't solve the problem
4157 * of setting the slave's MTU while it is
4158 * transmitting, but the assumption is that the base
4159 * driver can handle that.
4161 * TODO: figure out a way to safely iterate the slaves
4162 * list, but without holding a lock around the actual
4163 * call to the base driver.
4166 bond_for_each_slave(bond, slave, i) {
4167 pr_debug("s %p s->p %p c_m %p\n",
4170 slave->dev->netdev_ops->ndo_change_mtu);
4172 res = dev_set_mtu(slave->dev, new_mtu);
4175 /* If we failed to set the slave's mtu to the new value
4176 * we must abort the operation even in ACTIVE_BACKUP
4177 * mode, because if we allow the backup slaves to have
4178 * different mtu values than the active slave we'll
4179 * need to change their mtu when doing a failover. That
4180 * means changing their mtu from timer context, which
4181 * is probably not a good idea.
4183 pr_debug("err %d %s\n", res, slave->dev->name);
4188 bond_dev->mtu = new_mtu;
4193 /* unwind from head to the slave that failed */
4195 bond_for_each_slave_from_to(bond, slave, i, bond->first_slave, stop_at) {
4198 tmp_res = dev_set_mtu(slave->dev, bond_dev->mtu);
4200 pr_debug("unwind err %d dev %s\n",
4201 tmp_res, slave->dev->name);
4211 * Note that many devices must be down to change the HW address, and
4212 * downing the master releases all slaves. We can make bonds full of
4213 * bonding devices to test this, however.
4215 static int bond_set_mac_address(struct net_device *bond_dev, void *addr)
4217 struct bonding *bond = netdev_priv(bond_dev);
4218 struct sockaddr *sa = addr, tmp_sa;
4219 struct slave *slave, *stop_at;
4223 if (bond->params.mode == BOND_MODE_ALB)
4224 return bond_alb_set_mac_address(bond_dev, addr);
4227 pr_debug("bond=%p, name=%s\n",
4228 bond, bond_dev ? bond_dev->name : "None");
4231 * If fail_over_mac is set to active, do nothing and return
4232 * success. Returning an error causes ifenslave to fail.
4234 if (bond->params.fail_over_mac == BOND_FOM_ACTIVE)
4237 if (!is_valid_ether_addr(sa->sa_data))
4238 return -EADDRNOTAVAIL;
4240 /* Can't hold bond->lock with bh disabled here since
4241 * some base drivers panic. On the other hand we can't
4242 * hold bond->lock without bh disabled because we'll
4243 * deadlock. The only solution is to rely on the fact
4244 * that we're under rtnl_lock here, and the slaves
4245 * list won't change. This doesn't solve the problem
4246 * of setting the slave's hw address while it is
4247 * transmitting, but the assumption is that the base
4248 * driver can handle that.
4250 * TODO: figure out a way to safely iterate the slaves
4251 * list, but without holding a lock around the actual
4252 * call to the base driver.
4255 bond_for_each_slave(bond, slave, i) {
4256 const struct net_device_ops *slave_ops = slave->dev->netdev_ops;
4257 pr_debug("slave %p %s\n", slave, slave->dev->name);
4259 if (slave_ops->ndo_set_mac_address == NULL) {
4261 pr_debug("EOPNOTSUPP %s\n", slave->dev->name);
4265 res = dev_set_mac_address(slave->dev, addr);
4267 /* TODO: consider downing the slave
4269 * User should expect communications
4270 * breakage anyway until ARP finish
4273 pr_debug("err %d %s\n", res, slave->dev->name);
4279 memcpy(bond_dev->dev_addr, sa->sa_data, bond_dev->addr_len);
4283 memcpy(tmp_sa.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
4284 tmp_sa.sa_family = bond_dev->type;
4286 /* unwind from head to the slave that failed */
4288 bond_for_each_slave_from_to(bond, slave, i, bond->first_slave, stop_at) {
4291 tmp_res = dev_set_mac_address(slave->dev, &tmp_sa);
4293 pr_debug("unwind err %d dev %s\n",
4294 tmp_res, slave->dev->name);
4301 static int bond_xmit_roundrobin(struct sk_buff *skb, struct net_device *bond_dev)
4303 struct bonding *bond = netdev_priv(bond_dev);
4304 struct slave *slave, *start_at;
4305 int i, slave_no, res = 1;
4306 struct iphdr *iph = ip_hdr(skb);
4308 read_lock(&bond->lock);
4310 if (!BOND_IS_OK(bond))
4313 * Start with the curr_active_slave that joined the bond as the
4314 * default for sending IGMP traffic. For failover purposes one
4315 * needs to maintain some consistency for the interface that will
4316 * send the join/membership reports. The curr_active_slave found
4317 * will send all of this type of traffic.
4319 if ((iph->protocol == IPPROTO_IGMP) &&
4320 (skb->protocol == htons(ETH_P_IP))) {
4322 read_lock(&bond->curr_slave_lock);
4323 slave = bond->curr_active_slave;
4324 read_unlock(&bond->curr_slave_lock);
4330 * Concurrent TX may collide on rr_tx_counter; we accept
4331 * that as being rare enough not to justify using an
4334 slave_no = bond->rr_tx_counter++ % bond->slave_cnt;
4336 bond_for_each_slave(bond, slave, i) {
4344 bond_for_each_slave_from(bond, slave, i, start_at) {
4345 if (IS_UP(slave->dev) &&
4346 (slave->link == BOND_LINK_UP) &&
4347 (slave->state == BOND_STATE_ACTIVE)) {
4348 res = bond_dev_queue_xmit(bond, skb, slave->dev);
4355 /* no suitable interface, frame not sent */
4358 read_unlock(&bond->lock);
4359 return NETDEV_TX_OK;
4364 * in active-backup mode, we know that bond->curr_active_slave is always valid if
4365 * the bond has a usable interface.
4367 static int bond_xmit_activebackup(struct sk_buff *skb, struct net_device *bond_dev)
4369 struct bonding *bond = netdev_priv(bond_dev);
4372 read_lock(&bond->lock);
4373 read_lock(&bond->curr_slave_lock);
4375 if (!BOND_IS_OK(bond))
4378 if (!bond->curr_active_slave)
4381 res = bond_dev_queue_xmit(bond, skb, bond->curr_active_slave->dev);
4385 /* no suitable interface, frame not sent */
4388 read_unlock(&bond->curr_slave_lock);
4389 read_unlock(&bond->lock);
4390 return NETDEV_TX_OK;
4394 * In bond_xmit_xor() , we determine the output device by using a pre-
4395 * determined xmit_hash_policy(), If the selected device is not enabled,
4396 * find the next active slave.
4398 static int bond_xmit_xor(struct sk_buff *skb, struct net_device *bond_dev)
4400 struct bonding *bond = netdev_priv(bond_dev);
4401 struct slave *slave, *start_at;
4406 read_lock(&bond->lock);
4408 if (!BOND_IS_OK(bond))
4411 slave_no = bond->xmit_hash_policy(skb, bond->slave_cnt);
4413 bond_for_each_slave(bond, slave, i) {
4421 bond_for_each_slave_from(bond, slave, i, start_at) {
4422 if (IS_UP(slave->dev) &&
4423 (slave->link == BOND_LINK_UP) &&
4424 (slave->state == BOND_STATE_ACTIVE)) {
4425 res = bond_dev_queue_xmit(bond, skb, slave->dev);
4432 /* no suitable interface, frame not sent */
4435 read_unlock(&bond->lock);
4436 return NETDEV_TX_OK;
4440 * in broadcast mode, we send everything to all usable interfaces.
4442 static int bond_xmit_broadcast(struct sk_buff *skb, struct net_device *bond_dev)
4444 struct bonding *bond = netdev_priv(bond_dev);
4445 struct slave *slave, *start_at;
4446 struct net_device *tx_dev = NULL;
4450 read_lock(&bond->lock);
4452 if (!BOND_IS_OK(bond))
4455 read_lock(&bond->curr_slave_lock);
4456 start_at = bond->curr_active_slave;
4457 read_unlock(&bond->curr_slave_lock);
4462 bond_for_each_slave_from(bond, slave, i, start_at) {
4463 if (IS_UP(slave->dev) &&
4464 (slave->link == BOND_LINK_UP) &&
4465 (slave->state == BOND_STATE_ACTIVE)) {
4467 struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
4469 pr_err("%s: Error: bond_xmit_broadcast(): skb_clone() failed\n",
4474 res = bond_dev_queue_xmit(bond, skb2, tx_dev);
4476 dev_kfree_skb(skb2);
4480 tx_dev = slave->dev;
4485 res = bond_dev_queue_xmit(bond, skb, tx_dev);
4489 /* no suitable interface, frame not sent */
4492 /* frame sent to all suitable interfaces */
4493 read_unlock(&bond->lock);
4494 return NETDEV_TX_OK;
4497 /*------------------------- Device initialization ---------------------------*/
4499 static void bond_set_xmit_hash_policy(struct bonding *bond)
4501 switch (bond->params.xmit_policy) {
4502 case BOND_XMIT_POLICY_LAYER23:
4503 bond->xmit_hash_policy = bond_xmit_hash_policy_l23;
4505 case BOND_XMIT_POLICY_LAYER34:
4506 bond->xmit_hash_policy = bond_xmit_hash_policy_l34;
4508 case BOND_XMIT_POLICY_LAYER2:
4510 bond->xmit_hash_policy = bond_xmit_hash_policy_l2;
4516 * Lookup the slave that corresponds to a qid
4518 static inline int bond_slave_override(struct bonding *bond,
4519 struct sk_buff *skb)
4522 struct slave *slave = NULL;
4523 struct slave *check_slave;
4525 read_lock(&bond->lock);
4527 if (!BOND_IS_OK(bond) || !skb->queue_mapping)
4530 /* Find out if any slaves have the same mapping as this skb. */
4531 bond_for_each_slave(bond, check_slave, i) {
4532 if (check_slave->queue_id == skb->queue_mapping) {
4533 slave = check_slave;
4538 /* If the slave isn't UP, use default transmit policy. */
4539 if (slave && slave->queue_id && IS_UP(slave->dev) &&
4540 (slave->link == BOND_LINK_UP)) {
4541 res = bond_dev_queue_xmit(bond, skb, slave->dev);
4545 read_unlock(&bond->lock);
4549 static u16 bond_select_queue(struct net_device *dev, struct sk_buff *skb)
4552 * This helper function exists to help dev_pick_tx get the correct
4553 * destination queue. Using a helper function skips the a call to
4554 * skb_tx_hash and will put the skbs in the queue we expect on their
4555 * way down to the bonding driver.
4557 return skb->queue_mapping;
4560 static netdev_tx_t bond_start_xmit(struct sk_buff *skb, struct net_device *dev)
4562 struct bonding *bond = netdev_priv(dev);
4564 if (TX_QUEUE_OVERRIDE(bond->params.mode)) {
4565 if (!bond_slave_override(bond, skb))
4566 return NETDEV_TX_OK;
4569 switch (bond->params.mode) {
4570 case BOND_MODE_ROUNDROBIN:
4571 return bond_xmit_roundrobin(skb, dev);
4572 case BOND_MODE_ACTIVEBACKUP:
4573 return bond_xmit_activebackup(skb, dev);
4575 return bond_xmit_xor(skb, dev);
4576 case BOND_MODE_BROADCAST:
4577 return bond_xmit_broadcast(skb, dev);
4578 case BOND_MODE_8023AD:
4579 return bond_3ad_xmit_xor(skb, dev);
4582 return bond_alb_xmit(skb, dev);
4584 /* Should never happen, mode already checked */
4585 pr_err("%s: Error: Unknown bonding mode %d\n",
4586 dev->name, bond->params.mode);
4589 return NETDEV_TX_OK;
4595 * set bond mode specific net device operations
4597 void bond_set_mode_ops(struct bonding *bond, int mode)
4599 struct net_device *bond_dev = bond->dev;
4602 case BOND_MODE_ROUNDROBIN:
4604 case BOND_MODE_ACTIVEBACKUP:
4607 bond_set_xmit_hash_policy(bond);
4609 case BOND_MODE_BROADCAST:
4611 case BOND_MODE_8023AD:
4612 bond_set_master_3ad_flags(bond);
4613 bond_set_xmit_hash_policy(bond);
4616 bond_set_master_alb_flags(bond);
4621 /* Should never happen, mode already checked */
4622 pr_err("%s: Error: Unknown bonding mode %d\n",
4623 bond_dev->name, mode);
4628 static void bond_ethtool_get_drvinfo(struct net_device *bond_dev,
4629 struct ethtool_drvinfo *drvinfo)
4631 strncpy(drvinfo->driver, DRV_NAME, 32);
4632 strncpy(drvinfo->version, DRV_VERSION, 32);
4633 snprintf(drvinfo->fw_version, 32, "%d", BOND_ABI_VERSION);
4636 static const struct ethtool_ops bond_ethtool_ops = {
4637 .get_drvinfo = bond_ethtool_get_drvinfo,
4638 .get_link = ethtool_op_get_link,
4639 .get_tx_csum = ethtool_op_get_tx_csum,
4640 .get_sg = ethtool_op_get_sg,
4641 .get_tso = ethtool_op_get_tso,
4642 .get_ufo = ethtool_op_get_ufo,
4643 .get_flags = ethtool_op_get_flags,
4646 static const struct net_device_ops bond_netdev_ops = {
4647 .ndo_init = bond_init,
4648 .ndo_uninit = bond_uninit,
4649 .ndo_open = bond_open,
4650 .ndo_stop = bond_close,
4651 .ndo_start_xmit = bond_start_xmit,
4652 .ndo_select_queue = bond_select_queue,
4653 .ndo_get_stats64 = bond_get_stats,
4654 .ndo_do_ioctl = bond_do_ioctl,
4655 .ndo_set_multicast_list = bond_set_multicast_list,
4656 .ndo_change_mtu = bond_change_mtu,
4657 .ndo_set_mac_address = bond_set_mac_address,
4658 .ndo_neigh_setup = bond_neigh_setup,
4659 .ndo_vlan_rx_register = bond_vlan_rx_register,
4660 .ndo_vlan_rx_add_vid = bond_vlan_rx_add_vid,
4661 .ndo_vlan_rx_kill_vid = bond_vlan_rx_kill_vid,
4662 #ifdef CONFIG_NET_POLL_CONTROLLER
4663 .ndo_netpoll_cleanup = bond_netpoll_cleanup,
4664 .ndo_poll_controller = bond_poll_controller,
4668 static void bond_destructor(struct net_device *bond_dev)
4670 struct bonding *bond = netdev_priv(bond_dev);
4672 destroy_workqueue(bond->wq);
4673 free_netdev(bond_dev);
4676 static void bond_setup(struct net_device *bond_dev)
4678 struct bonding *bond = netdev_priv(bond_dev);
4680 /* initialize rwlocks */
4681 rwlock_init(&bond->lock);
4682 rwlock_init(&bond->curr_slave_lock);
4684 bond->params = bonding_defaults;
4686 /* Initialize pointers */
4687 bond->dev = bond_dev;
4688 INIT_LIST_HEAD(&bond->vlan_list);
4690 /* Initialize the device entry points */
4691 ether_setup(bond_dev);
4692 bond_dev->netdev_ops = &bond_netdev_ops;
4693 bond_dev->ethtool_ops = &bond_ethtool_ops;
4694 bond_set_mode_ops(bond, bond->params.mode);
4696 bond_dev->destructor = bond_destructor;
4698 /* Initialize the device options */
4699 bond_dev->tx_queue_len = 0;
4700 bond_dev->flags |= IFF_MASTER|IFF_MULTICAST;
4701 bond_dev->priv_flags |= IFF_BONDING;
4702 bond_dev->priv_flags &= ~IFF_XMIT_DST_RELEASE;
4704 if (bond->params.arp_interval)
4705 bond_dev->priv_flags |= IFF_MASTER_ARPMON;
4707 /* At first, we block adding VLANs. That's the only way to
4708 * prevent problems that occur when adding VLANs over an
4709 * empty bond. The block will be removed once non-challenged
4710 * slaves are enslaved.
4712 bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
4714 /* don't acquire bond device's netif_tx_lock when
4716 bond_dev->features |= NETIF_F_LLTX;
4718 /* By default, we declare the bond to be fully
4719 * VLAN hardware accelerated capable. Special
4720 * care is taken in the various xmit functions
4721 * when there are slaves that are not hw accel
4724 bond_dev->features |= (NETIF_F_HW_VLAN_TX |
4725 NETIF_F_HW_VLAN_RX |
4726 NETIF_F_HW_VLAN_FILTER);
4728 /* By default, we enable GRO on bonding devices.
4729 * Actual support requires lowlevel drivers are GRO ready.
4731 bond_dev->features |= NETIF_F_GRO;
4734 static void bond_work_cancel_all(struct bonding *bond)
4736 write_lock_bh(&bond->lock);
4737 bond->kill_timers = 1;
4738 write_unlock_bh(&bond->lock);
4740 if (bond->params.miimon && delayed_work_pending(&bond->mii_work))
4741 cancel_delayed_work(&bond->mii_work);
4743 if (bond->params.arp_interval && delayed_work_pending(&bond->arp_work))
4744 cancel_delayed_work(&bond->arp_work);
4746 if (bond->params.mode == BOND_MODE_ALB &&
4747 delayed_work_pending(&bond->alb_work))
4748 cancel_delayed_work(&bond->alb_work);
4750 if (bond->params.mode == BOND_MODE_8023AD &&
4751 delayed_work_pending(&bond->ad_work))
4752 cancel_delayed_work(&bond->ad_work);
4754 if (delayed_work_pending(&bond->mcast_work))
4755 cancel_delayed_work(&bond->mcast_work);
4759 * Destroy a bonding device.
4760 * Must be under rtnl_lock when this function is called.
4762 static void bond_uninit(struct net_device *bond_dev)
4764 struct bonding *bond = netdev_priv(bond_dev);
4765 struct vlan_entry *vlan, *tmp;
4767 bond_netpoll_cleanup(bond_dev);
4769 /* Release the bonded slaves */
4770 bond_release_all(bond_dev);
4772 list_del(&bond->bond_list);
4774 bond_work_cancel_all(bond);
4776 bond_remove_proc_entry(bond);
4778 __hw_addr_flush(&bond->mc_list);
4780 list_for_each_entry_safe(vlan, tmp, &bond->vlan_list, vlan_list) {
4781 list_del(&vlan->vlan_list);
4786 /*------------------------- Module initialization ---------------------------*/
4789 * Convert string input module parms. Accept either the
4790 * number of the mode or its string name. A bit complicated because
4791 * some mode names are substrings of other names, and calls from sysfs
4792 * may have whitespace in the name (trailing newlines, for example).
4794 int bond_parse_parm(const char *buf, const struct bond_parm_tbl *tbl)
4796 int modeint = -1, i, rv;
4797 char *p, modestr[BOND_MAX_MODENAME_LEN + 1] = { 0, };
4799 for (p = (char *)buf; *p; p++)
4800 if (!(isdigit(*p) || isspace(*p)))
4804 rv = sscanf(buf, "%20s", modestr);
4806 rv = sscanf(buf, "%d", &modeint);
4811 for (i = 0; tbl[i].modename; i++) {
4812 if (modeint == tbl[i].mode)
4814 if (strcmp(modestr, tbl[i].modename) == 0)
4821 static int bond_check_params(struct bond_params *params)
4823 int arp_validate_value, fail_over_mac_value, primary_reselect_value;
4826 * Convert string parameters.
4829 bond_mode = bond_parse_parm(mode, bond_mode_tbl);
4830 if (bond_mode == -1) {
4831 pr_err("Error: Invalid bonding mode \"%s\"\n",
4832 mode == NULL ? "NULL" : mode);
4837 if (xmit_hash_policy) {
4838 if ((bond_mode != BOND_MODE_XOR) &&
4839 (bond_mode != BOND_MODE_8023AD)) {
4840 pr_info("xmit_hash_policy param is irrelevant in mode %s\n",
4841 bond_mode_name(bond_mode));
4843 xmit_hashtype = bond_parse_parm(xmit_hash_policy,
4845 if (xmit_hashtype == -1) {
4846 pr_err("Error: Invalid xmit_hash_policy \"%s\"\n",
4847 xmit_hash_policy == NULL ? "NULL" :
4855 if (bond_mode != BOND_MODE_8023AD) {
4856 pr_info("lacp_rate param is irrelevant in mode %s\n",
4857 bond_mode_name(bond_mode));
4859 lacp_fast = bond_parse_parm(lacp_rate, bond_lacp_tbl);
4860 if (lacp_fast == -1) {
4861 pr_err("Error: Invalid lacp rate \"%s\"\n",
4862 lacp_rate == NULL ? "NULL" : lacp_rate);
4869 params->ad_select = bond_parse_parm(ad_select, ad_select_tbl);
4870 if (params->ad_select == -1) {
4871 pr_err("Error: Invalid ad_select \"%s\"\n",
4872 ad_select == NULL ? "NULL" : ad_select);
4876 if (bond_mode != BOND_MODE_8023AD) {
4877 pr_warning("ad_select param only affects 802.3ad mode\n");
4880 params->ad_select = BOND_AD_STABLE;
4883 if (max_bonds < 0) {
4884 pr_warning("Warning: max_bonds (%d) not in range %d-%d, so it was reset to BOND_DEFAULT_MAX_BONDS (%d)\n",
4885 max_bonds, 0, INT_MAX, BOND_DEFAULT_MAX_BONDS);
4886 max_bonds = BOND_DEFAULT_MAX_BONDS;
4890 pr_warning("Warning: miimon module parameter (%d), not in range 0-%d, so it was reset to %d\n",
4891 miimon, INT_MAX, BOND_LINK_MON_INTERV);
4892 miimon = BOND_LINK_MON_INTERV;
4896 pr_warning("Warning: updelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4901 if (downdelay < 0) {
4902 pr_warning("Warning: downdelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4903 downdelay, INT_MAX);
4907 if ((use_carrier != 0) && (use_carrier != 1)) {
4908 pr_warning("Warning: use_carrier module parameter (%d), not of valid value (0/1), so it was set to 1\n",
4913 if (num_grat_arp < 0 || num_grat_arp > 255) {
4914 pr_warning("Warning: num_grat_arp (%d) not in range 0-255 so it was reset to 1\n",
4919 if (num_unsol_na < 0 || num_unsol_na > 255) {
4920 pr_warning("Warning: num_unsol_na (%d) not in range 0-255 so it was reset to 1\n",
4925 /* reset values for 802.3ad */
4926 if (bond_mode == BOND_MODE_8023AD) {
4928 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");
4929 pr_warning("Forcing miimon to 100msec\n");
4934 if (tx_queues < 1 || tx_queues > 255) {
4935 pr_warning("Warning: tx_queues (%d) should be between "
4936 "1 and 255, resetting to %d\n",
4937 tx_queues, BOND_DEFAULT_TX_QUEUES);
4938 tx_queues = BOND_DEFAULT_TX_QUEUES;
4941 if ((all_slaves_active != 0) && (all_slaves_active != 1)) {
4942 pr_warning("Warning: all_slaves_active module parameter (%d), "
4943 "not of valid value (0/1), so it was set to "
4944 "0\n", all_slaves_active);
4945 all_slaves_active = 0;
4948 if (resend_igmp < 0 || resend_igmp > 255) {
4949 pr_warning("Warning: resend_igmp (%d) should be between "
4950 "0 and 255, resetting to %d\n",
4951 resend_igmp, BOND_DEFAULT_RESEND_IGMP);
4952 resend_igmp = BOND_DEFAULT_RESEND_IGMP;
4955 /* reset values for TLB/ALB */
4956 if ((bond_mode == BOND_MODE_TLB) ||
4957 (bond_mode == BOND_MODE_ALB)) {
4959 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");
4960 pr_warning("Forcing miimon to 100msec\n");
4965 if (bond_mode == BOND_MODE_ALB) {
4966 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",
4971 if (updelay || downdelay) {
4972 /* just warn the user the up/down delay will have
4973 * no effect since miimon is zero...
4975 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",
4976 updelay, downdelay);
4979 /* don't allow arp monitoring */
4981 pr_warning("Warning: miimon (%d) and arp_interval (%d) can't be used simultaneously, disabling ARP monitoring\n",
4982 miimon, arp_interval);
4986 if ((updelay % miimon) != 0) {
4987 pr_warning("Warning: updelay (%d) is not a multiple of miimon (%d), updelay rounded to %d ms\n",
4989 (updelay / miimon) * miimon);
4994 if ((downdelay % miimon) != 0) {
4995 pr_warning("Warning: downdelay (%d) is not a multiple of miimon (%d), downdelay rounded to %d ms\n",
4997 (downdelay / miimon) * miimon);
5000 downdelay /= miimon;
5003 if (arp_interval < 0) {
5004 pr_warning("Warning: arp_interval module parameter (%d) , not in range 0-%d, so it was reset to %d\n",
5005 arp_interval, INT_MAX, BOND_LINK_ARP_INTERV);
5006 arp_interval = BOND_LINK_ARP_INTERV;
5009 for (arp_ip_count = 0;
5010 (arp_ip_count < BOND_MAX_ARP_TARGETS) && arp_ip_target[arp_ip_count];
5012 /* not complete check, but should be good enough to
5014 if (!isdigit(arp_ip_target[arp_ip_count][0])) {
5015 pr_warning("Warning: bad arp_ip_target module parameter (%s), ARP monitoring will not be performed\n",
5016 arp_ip_target[arp_ip_count]);
5019 __be32 ip = in_aton(arp_ip_target[arp_ip_count]);
5020 arp_target[arp_ip_count] = ip;
5024 if (arp_interval && !arp_ip_count) {
5025 /* don't allow arping if no arp_ip_target given... */
5026 pr_warning("Warning: arp_interval module parameter (%d) specified without providing an arp_ip_target parameter, arp_interval was reset to 0\n",
5032 if (bond_mode != BOND_MODE_ACTIVEBACKUP) {
5033 pr_err("arp_validate only supported in active-backup mode\n");
5036 if (!arp_interval) {
5037 pr_err("arp_validate requires arp_interval\n");
5041 arp_validate_value = bond_parse_parm(arp_validate,
5043 if (arp_validate_value == -1) {
5044 pr_err("Error: invalid arp_validate \"%s\"\n",
5045 arp_validate == NULL ? "NULL" : arp_validate);
5049 arp_validate_value = 0;
5052 pr_info("MII link monitoring set to %d ms\n", miimon);
5053 } else if (arp_interval) {
5056 pr_info("ARP monitoring set to %d ms, validate %s, with %d target(s):",
5058 arp_validate_tbl[arp_validate_value].modename,
5061 for (i = 0; i < arp_ip_count; i++)
5062 pr_info(" %s", arp_ip_target[i]);
5066 } else if (max_bonds) {
5067 /* miimon and arp_interval not set, we need one so things
5068 * work as expected, see bonding.txt for details
5070 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");
5073 if (primary && !USES_PRIMARY(bond_mode)) {
5074 /* currently, using a primary only makes sense
5075 * in active backup, TLB or ALB modes
5077 pr_warning("Warning: %s primary device specified but has no effect in %s mode\n",
5078 primary, bond_mode_name(bond_mode));
5082 if (primary && primary_reselect) {
5083 primary_reselect_value = bond_parse_parm(primary_reselect,
5085 if (primary_reselect_value == -1) {
5086 pr_err("Error: Invalid primary_reselect \"%s\"\n",
5088 NULL ? "NULL" : primary_reselect);
5092 primary_reselect_value = BOND_PRI_RESELECT_ALWAYS;
5095 if (fail_over_mac) {
5096 fail_over_mac_value = bond_parse_parm(fail_over_mac,
5098 if (fail_over_mac_value == -1) {
5099 pr_err("Error: invalid fail_over_mac \"%s\"\n",
5100 arp_validate == NULL ? "NULL" : arp_validate);
5104 if (bond_mode != BOND_MODE_ACTIVEBACKUP)
5105 pr_warning("Warning: fail_over_mac only affects active-backup mode.\n");
5107 fail_over_mac_value = BOND_FOM_NONE;
5110 /* fill params struct with the proper values */
5111 params->mode = bond_mode;
5112 params->xmit_policy = xmit_hashtype;
5113 params->miimon = miimon;
5114 params->num_grat_arp = num_grat_arp;
5115 params->num_unsol_na = num_unsol_na;
5116 params->arp_interval = arp_interval;
5117 params->arp_validate = arp_validate_value;
5118 params->updelay = updelay;
5119 params->downdelay = downdelay;
5120 params->use_carrier = use_carrier;
5121 params->lacp_fast = lacp_fast;
5122 params->primary[0] = 0;
5123 params->primary_reselect = primary_reselect_value;
5124 params->fail_over_mac = fail_over_mac_value;
5125 params->tx_queues = tx_queues;
5126 params->all_slaves_active = all_slaves_active;
5127 params->resend_igmp = resend_igmp;
5130 strncpy(params->primary, primary, IFNAMSIZ);
5131 params->primary[IFNAMSIZ - 1] = 0;
5134 memcpy(params->arp_targets, arp_target, sizeof(arp_target));
5139 static struct lock_class_key bonding_netdev_xmit_lock_key;
5140 static struct lock_class_key bonding_netdev_addr_lock_key;
5142 static void bond_set_lockdep_class_one(struct net_device *dev,
5143 struct netdev_queue *txq,
5146 lockdep_set_class(&txq->_xmit_lock,
5147 &bonding_netdev_xmit_lock_key);
5150 static void bond_set_lockdep_class(struct net_device *dev)
5152 lockdep_set_class(&dev->addr_list_lock,
5153 &bonding_netdev_addr_lock_key);
5154 netdev_for_each_tx_queue(dev, bond_set_lockdep_class_one, NULL);
5158 * Called from registration process
5160 static int bond_init(struct net_device *bond_dev)
5162 struct bonding *bond = netdev_priv(bond_dev);
5163 struct bond_net *bn = net_generic(dev_net(bond_dev), bond_net_id);
5165 pr_debug("Begin bond_init for %s\n", bond_dev->name);
5167 bond->wq = create_singlethread_workqueue(bond_dev->name);
5171 bond_set_lockdep_class(bond_dev);
5173 netif_carrier_off(bond_dev);
5175 bond_create_proc_entry(bond);
5176 list_add_tail(&bond->bond_list, &bn->dev_list);
5178 bond_prepare_sysfs_group(bond);
5180 __hw_addr_init(&bond->mc_list);
5184 static int bond_validate(struct nlattr *tb[], struct nlattr *data[])
5186 if (tb[IFLA_ADDRESS]) {
5187 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN)
5189 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS])))
5190 return -EADDRNOTAVAIL;
5195 static struct rtnl_link_ops bond_link_ops __read_mostly = {
5197 .priv_size = sizeof(struct bonding),
5198 .setup = bond_setup,
5199 .validate = bond_validate,
5202 /* Create a new bond based on the specified name and bonding parameters.
5203 * If name is NULL, obtain a suitable "bond%d" name for us.
5204 * Caller must NOT hold rtnl_lock; we need to release it here before we
5205 * set up our sysfs entries.
5207 int bond_create(struct net *net, const char *name)
5209 struct net_device *bond_dev;
5214 bond_dev = alloc_netdev_mq(sizeof(struct bonding), name ? name : "",
5215 bond_setup, tx_queues);
5217 pr_err("%s: eek! can't alloc netdev!\n", name);
5222 dev_net_set(bond_dev, net);
5223 bond_dev->rtnl_link_ops = &bond_link_ops;
5226 res = dev_alloc_name(bond_dev, "bond%d");
5231 * If we're given a name to register
5232 * we need to ensure that its not already
5236 if (__dev_get_by_name(net, name) != NULL)
5240 res = register_netdevice(bond_dev);
5245 bond_destructor(bond_dev);
5249 static int __net_init bond_net_init(struct net *net)
5251 struct bond_net *bn = net_generic(net, bond_net_id);
5254 INIT_LIST_HEAD(&bn->dev_list);
5256 bond_create_proc_dir(bn);
5261 static void __net_exit bond_net_exit(struct net *net)
5263 struct bond_net *bn = net_generic(net, bond_net_id);
5265 bond_destroy_proc_dir(bn);
5268 static struct pernet_operations bond_net_ops = {
5269 .init = bond_net_init,
5270 .exit = bond_net_exit,
5272 .size = sizeof(struct bond_net),
5275 static int __init bonding_init(void)
5280 pr_info("%s", version);
5282 res = bond_check_params(&bonding_defaults);
5286 res = register_pernet_subsys(&bond_net_ops);
5290 res = rtnl_link_register(&bond_link_ops);
5294 for (i = 0; i < max_bonds; i++) {
5295 res = bond_create(&init_net, NULL);
5300 res = bond_create_sysfs();
5304 register_netdevice_notifier(&bond_netdev_notifier);
5305 register_inetaddr_notifier(&bond_inetaddr_notifier);
5306 bond_register_ipv6_notifier();
5310 rtnl_link_unregister(&bond_link_ops);
5312 unregister_pernet_subsys(&bond_net_ops);
5317 static void __exit bonding_exit(void)
5319 unregister_netdevice_notifier(&bond_netdev_notifier);
5320 unregister_inetaddr_notifier(&bond_inetaddr_notifier);
5321 bond_unregister_ipv6_notifier();
5323 bond_destroy_sysfs();
5325 rtnl_link_unregister(&bond_link_ops);
5326 unregister_pernet_subsys(&bond_net_ops);
5329 module_init(bonding_init);
5330 module_exit(bonding_exit);
5331 MODULE_LICENSE("GPL");
5332 MODULE_VERSION(DRV_VERSION);
5333 MODULE_DESCRIPTION(DRV_DESCRIPTION ", v" DRV_VERSION);
5334 MODULE_AUTHOR("Thomas Davis, tadavis@lbl.gov and many others");
5335 MODULE_ALIAS_RTNL_LINK("bond");