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>
58 #include <linux/uaccess.h>
59 #include <linux/errno.h>
60 #include <linux/netdevice.h>
61 #include <linux/inetdevice.h>
62 #include <linux/igmp.h>
63 #include <linux/etherdevice.h>
64 #include <linux/skbuff.h>
66 #include <linux/rtnetlink.h>
67 #include <linux/smp.h>
68 #include <linux/if_ether.h>
70 #include <linux/mii.h>
71 #include <linux/ethtool.h>
72 #include <linux/if_vlan.h>
73 #include <linux/if_bonding.h>
74 #include <linux/jiffies.h>
75 #include <linux/preempt.h>
76 #include <net/route.h>
77 #include <net/net_namespace.h>
78 #include <net/netns/generic.h>
79 #include <net/pkt_sched.h>
80 #include <linux/rculist.h>
81 #include <net/flow_keys.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_peer_notif = 1;
95 static int miimon = BOND_LINK_MON_INTERV;
98 static int use_carrier = 1;
100 static char *primary;
101 static char *primary_reselect;
102 static char *lacp_rate;
103 static int min_links;
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 *arp_all_targets;
110 static char *fail_over_mac;
111 static int all_slaves_active;
112 static struct bond_params bonding_defaults;
113 static int resend_igmp = BOND_DEFAULT_RESEND_IGMP;
114 static int packets_per_slave = 1;
115 static int lp_interval = BOND_ALB_DEFAULT_LP_INTERVAL;
117 module_param(max_bonds, int, 0);
118 MODULE_PARM_DESC(max_bonds, "Max number of bonded devices");
119 module_param(tx_queues, int, 0);
120 MODULE_PARM_DESC(tx_queues, "Max number of transmit queues (default = 16)");
121 module_param_named(num_grat_arp, num_peer_notif, int, 0644);
122 MODULE_PARM_DESC(num_grat_arp, "Number of peer notifications to send on "
123 "failover event (alias of num_unsol_na)");
124 module_param_named(num_unsol_na, num_peer_notif, int, 0644);
125 MODULE_PARM_DESC(num_unsol_na, "Number of peer notifications to send on "
126 "failover event (alias of num_grat_arp)");
127 module_param(miimon, int, 0);
128 MODULE_PARM_DESC(miimon, "Link check interval in milliseconds");
129 module_param(updelay, int, 0);
130 MODULE_PARM_DESC(updelay, "Delay before considering link up, in milliseconds");
131 module_param(downdelay, int, 0);
132 MODULE_PARM_DESC(downdelay, "Delay before considering link down, "
134 module_param(use_carrier, int, 0);
135 MODULE_PARM_DESC(use_carrier, "Use netif_carrier_ok (vs MII ioctls) in miimon; "
136 "0 for off, 1 for on (default)");
137 module_param(mode, charp, 0);
138 MODULE_PARM_DESC(mode, "Mode of operation; 0 for balance-rr, "
139 "1 for active-backup, 2 for balance-xor, "
140 "3 for broadcast, 4 for 802.3ad, 5 for balance-tlb, "
141 "6 for balance-alb");
142 module_param(primary, charp, 0);
143 MODULE_PARM_DESC(primary, "Primary network device to use");
144 module_param(primary_reselect, charp, 0);
145 MODULE_PARM_DESC(primary_reselect, "Reselect primary slave "
147 "0 for always (default), "
148 "1 for only if speed of primary is "
150 "2 for only on active slave "
152 module_param(lacp_rate, charp, 0);
153 MODULE_PARM_DESC(lacp_rate, "LACPDU tx rate to request from 802.3ad partner; "
154 "0 for slow, 1 for fast");
155 module_param(ad_select, charp, 0);
156 MODULE_PARM_DESC(ad_select, "803.ad aggregation selection logic; "
157 "0 for stable (default), 1 for bandwidth, "
159 module_param(min_links, int, 0);
160 MODULE_PARM_DESC(min_links, "Minimum number of available links before turning on carrier");
162 module_param(xmit_hash_policy, charp, 0);
163 MODULE_PARM_DESC(xmit_hash_policy, "balance-xor and 802.3ad hashing method; "
164 "0 for layer 2 (default), 1 for layer 3+4, "
165 "2 for layer 2+3, 3 for encap layer 2+3, "
166 "4 for encap layer 3+4");
167 module_param(arp_interval, int, 0);
168 MODULE_PARM_DESC(arp_interval, "arp interval in milliseconds");
169 module_param_array(arp_ip_target, charp, NULL, 0);
170 MODULE_PARM_DESC(arp_ip_target, "arp targets in n.n.n.n form");
171 module_param(arp_validate, charp, 0);
172 MODULE_PARM_DESC(arp_validate, "validate src/dst of ARP probes; "
173 "0 for none (default), 1 for active, "
174 "2 for backup, 3 for all");
175 module_param(arp_all_targets, charp, 0);
176 MODULE_PARM_DESC(arp_all_targets, "fail on any/all arp targets timeout; 0 for any (default), 1 for all");
177 module_param(fail_over_mac, charp, 0);
178 MODULE_PARM_DESC(fail_over_mac, "For active-backup, do not set all slaves to "
179 "the same MAC; 0 for none (default), "
180 "1 for active, 2 for follow");
181 module_param(all_slaves_active, int, 0);
182 MODULE_PARM_DESC(all_slaves_active, "Keep all frames received on an interface"
183 "by setting active flag for all slaves; "
184 "0 for never (default), 1 for always.");
185 module_param(resend_igmp, int, 0);
186 MODULE_PARM_DESC(resend_igmp, "Number of IGMP membership reports to send on "
188 module_param(packets_per_slave, int, 0);
189 MODULE_PARM_DESC(packets_per_slave, "Packets to send per slave in balance-rr "
190 "mode; 0 for a random slave, 1 packet per "
191 "slave (default), >1 packets per slave.");
192 module_param(lp_interval, uint, 0);
193 MODULE_PARM_DESC(lp_interval, "The number of seconds between instances where "
194 "the bonding driver sends learning packets to "
195 "each slaves peer switch. The default is 1.");
197 /*----------------------------- Global variables ----------------------------*/
199 #ifdef CONFIG_NET_POLL_CONTROLLER
200 atomic_t netpoll_block_tx = ATOMIC_INIT(0);
203 int bond_net_id __read_mostly;
205 static __be32 arp_target[BOND_MAX_ARP_TARGETS];
206 static int arp_ip_count;
207 static int bond_mode = BOND_MODE_ROUNDROBIN;
208 static int xmit_hashtype = BOND_XMIT_POLICY_LAYER2;
209 static int lacp_fast;
211 const struct bond_parm_tbl bond_lacp_tbl[] = {
212 { "slow", AD_LACP_SLOW},
213 { "fast", AD_LACP_FAST},
217 const struct bond_parm_tbl xmit_hashtype_tbl[] = {
218 { "layer2", BOND_XMIT_POLICY_LAYER2},
219 { "layer3+4", BOND_XMIT_POLICY_LAYER34},
220 { "layer2+3", BOND_XMIT_POLICY_LAYER23},
221 { "encap2+3", BOND_XMIT_POLICY_ENCAP23},
222 { "encap3+4", BOND_XMIT_POLICY_ENCAP34},
226 const struct bond_parm_tbl arp_all_targets_tbl[] = {
227 { "any", BOND_ARP_TARGETS_ANY},
228 { "all", BOND_ARP_TARGETS_ALL},
232 const struct bond_parm_tbl arp_validate_tbl[] = {
233 { "none", BOND_ARP_VALIDATE_NONE},
234 { "active", BOND_ARP_VALIDATE_ACTIVE},
235 { "backup", BOND_ARP_VALIDATE_BACKUP},
236 { "all", BOND_ARP_VALIDATE_ALL},
240 const struct bond_parm_tbl fail_over_mac_tbl[] = {
241 { "none", BOND_FOM_NONE},
242 { "active", BOND_FOM_ACTIVE},
243 { "follow", BOND_FOM_FOLLOW},
247 const struct bond_parm_tbl pri_reselect_tbl[] = {
248 { "always", BOND_PRI_RESELECT_ALWAYS},
249 { "better", BOND_PRI_RESELECT_BETTER},
250 { "failure", BOND_PRI_RESELECT_FAILURE},
254 struct bond_parm_tbl ad_select_tbl[] = {
255 { "stable", BOND_AD_STABLE},
256 { "bandwidth", BOND_AD_BANDWIDTH},
257 { "count", BOND_AD_COUNT},
261 /*-------------------------- Forward declarations ---------------------------*/
263 static int bond_init(struct net_device *bond_dev);
264 static void bond_uninit(struct net_device *bond_dev);
266 /*---------------------------- General routines -----------------------------*/
268 const char *bond_mode_name(int mode)
270 static const char *names[] = {
271 [BOND_MODE_ROUNDROBIN] = "load balancing (round-robin)",
272 [BOND_MODE_ACTIVEBACKUP] = "fault-tolerance (active-backup)",
273 [BOND_MODE_XOR] = "load balancing (xor)",
274 [BOND_MODE_BROADCAST] = "fault-tolerance (broadcast)",
275 [BOND_MODE_8023AD] = "IEEE 802.3ad Dynamic link aggregation",
276 [BOND_MODE_TLB] = "transmit load balancing",
277 [BOND_MODE_ALB] = "adaptive load balancing",
280 if (mode < BOND_MODE_ROUNDROBIN || mode > BOND_MODE_ALB)
286 /*---------------------------------- VLAN -----------------------------------*/
289 * bond_dev_queue_xmit - Prepare skb for xmit.
291 * @bond: bond device that got this skb for tx.
292 * @skb: hw accel VLAN tagged skb to transmit
293 * @slave_dev: slave that is supposed to xmit this skbuff
295 void bond_dev_queue_xmit(struct bonding *bond, struct sk_buff *skb,
296 struct net_device *slave_dev)
298 skb->dev = slave_dev;
300 BUILD_BUG_ON(sizeof(skb->queue_mapping) !=
301 sizeof(qdisc_skb_cb(skb)->slave_dev_queue_mapping));
302 skb->queue_mapping = qdisc_skb_cb(skb)->slave_dev_queue_mapping;
304 if (unlikely(netpoll_tx_running(bond->dev)))
305 bond_netpoll_send_skb(bond_get_slave_by_dev(bond, slave_dev), skb);
311 * In the following 2 functions, bond_vlan_rx_add_vid and bond_vlan_rx_kill_vid,
312 * We don't protect the slave list iteration with a lock because:
313 * a. This operation is performed in IOCTL context,
314 * b. The operation is protected by the RTNL semaphore in the 8021q code,
315 * c. Holding a lock with BH disabled while directly calling a base driver
316 * entry point is generally a BAD idea.
318 * The design of synchronization/protection for this operation in the 8021q
319 * module is good for one or more VLAN devices over a single physical device
320 * and cannot be extended for a teaming solution like bonding, so there is a
321 * potential race condition here where a net device from the vlan group might
322 * be referenced (either by a base driver or the 8021q code) while it is being
323 * removed from the system. However, it turns out we're not making matters
324 * worse, and if it works for regular VLAN usage it will work here too.
328 * bond_vlan_rx_add_vid - Propagates adding an id to slaves
329 * @bond_dev: bonding net device that got called
330 * @vid: vlan id being added
332 static int bond_vlan_rx_add_vid(struct net_device *bond_dev,
333 __be16 proto, u16 vid)
335 struct bonding *bond = netdev_priv(bond_dev);
336 struct slave *slave, *rollback_slave;
337 struct list_head *iter;
340 bond_for_each_slave(bond, slave, iter) {
341 res = vlan_vid_add(slave->dev, proto, vid);
349 /* unwind to the slave that failed */
350 bond_for_each_slave(bond, rollback_slave, iter) {
351 if (rollback_slave == slave)
354 vlan_vid_del(rollback_slave->dev, proto, vid);
361 * bond_vlan_rx_kill_vid - Propagates deleting an id to slaves
362 * @bond_dev: bonding net device that got called
363 * @vid: vlan id being removed
365 static int bond_vlan_rx_kill_vid(struct net_device *bond_dev,
366 __be16 proto, u16 vid)
368 struct bonding *bond = netdev_priv(bond_dev);
369 struct list_head *iter;
372 bond_for_each_slave(bond, slave, iter)
373 vlan_vid_del(slave->dev, proto, vid);
375 if (bond_is_lb(bond))
376 bond_alb_clear_vlan(bond, vid);
381 /*------------------------------- Link status -------------------------------*/
384 * Set the carrier state for the master according to the state of its
385 * slaves. If any slaves are up, the master is up. In 802.3ad mode,
386 * do special 802.3ad magic.
388 * Returns zero if carrier state does not change, nonzero if it does.
390 static int bond_set_carrier(struct bonding *bond)
392 struct list_head *iter;
395 if (!bond_has_slaves(bond))
398 if (bond->params.mode == BOND_MODE_8023AD)
399 return bond_3ad_set_carrier(bond);
401 bond_for_each_slave(bond, slave, iter) {
402 if (slave->link == BOND_LINK_UP) {
403 if (!netif_carrier_ok(bond->dev)) {
404 netif_carrier_on(bond->dev);
412 if (netif_carrier_ok(bond->dev)) {
413 netif_carrier_off(bond->dev);
420 * Get link speed and duplex from the slave's base driver
421 * using ethtool. If for some reason the call fails or the
422 * values are invalid, set speed and duplex to -1,
425 static void bond_update_speed_duplex(struct slave *slave)
427 struct net_device *slave_dev = slave->dev;
428 struct ethtool_cmd ecmd;
432 slave->speed = SPEED_UNKNOWN;
433 slave->duplex = DUPLEX_UNKNOWN;
435 res = __ethtool_get_settings(slave_dev, &ecmd);
439 slave_speed = ethtool_cmd_speed(&ecmd);
440 if (slave_speed == 0 || slave_speed == ((__u32) -1))
443 switch (ecmd.duplex) {
451 slave->speed = slave_speed;
452 slave->duplex = ecmd.duplex;
457 const char *bond_slave_link_status(s8 link)
474 * if <dev> supports MII link status reporting, check its link status.
476 * We either do MII/ETHTOOL ioctls, or check netif_carrier_ok(),
477 * depending upon the setting of the use_carrier parameter.
479 * Return either BMSR_LSTATUS, meaning that the link is up (or we
480 * can't tell and just pretend it is), or 0, meaning that the link is
483 * If reporting is non-zero, instead of faking link up, return -1 if
484 * both ETHTOOL and MII ioctls fail (meaning the device does not
485 * support them). If use_carrier is set, return whatever it says.
486 * It'd be nice if there was a good way to tell if a driver supports
487 * netif_carrier, but there really isn't.
489 static int bond_check_dev_link(struct bonding *bond,
490 struct net_device *slave_dev, int reporting)
492 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
493 int (*ioctl)(struct net_device *, struct ifreq *, int);
495 struct mii_ioctl_data *mii;
497 if (!reporting && !netif_running(slave_dev))
500 if (bond->params.use_carrier)
501 return netif_carrier_ok(slave_dev) ? BMSR_LSTATUS : 0;
503 /* Try to get link status using Ethtool first. */
504 if (slave_dev->ethtool_ops->get_link)
505 return slave_dev->ethtool_ops->get_link(slave_dev) ?
508 /* Ethtool can't be used, fallback to MII ioctls. */
509 ioctl = slave_ops->ndo_do_ioctl;
511 /* TODO: set pointer to correct ioctl on a per team member */
512 /* bases to make this more efficient. that is, once */
513 /* we determine the correct ioctl, we will always */
514 /* call it and not the others for that team */
518 * We cannot assume that SIOCGMIIPHY will also read a
519 * register; not all network drivers (e.g., e100)
523 /* Yes, the mii is overlaid on the ifreq.ifr_ifru */
524 strncpy(ifr.ifr_name, slave_dev->name, IFNAMSIZ);
526 if (IOCTL(slave_dev, &ifr, SIOCGMIIPHY) == 0) {
527 mii->reg_num = MII_BMSR;
528 if (IOCTL(slave_dev, &ifr, SIOCGMIIREG) == 0)
529 return mii->val_out & BMSR_LSTATUS;
534 * If reporting, report that either there's no dev->do_ioctl,
535 * or both SIOCGMIIREG and get_link failed (meaning that we
536 * cannot report link status). If not reporting, pretend
539 return reporting ? -1 : BMSR_LSTATUS;
542 /*----------------------------- Multicast list ------------------------------*/
545 * Push the promiscuity flag down to appropriate slaves
547 static int bond_set_promiscuity(struct bonding *bond, int inc)
549 struct list_head *iter;
552 if (USES_PRIMARY(bond->params.mode)) {
553 /* write lock already acquired */
554 if (bond->curr_active_slave) {
555 err = dev_set_promiscuity(bond->curr_active_slave->dev,
561 bond_for_each_slave(bond, slave, iter) {
562 err = dev_set_promiscuity(slave->dev, inc);
571 * Push the allmulti flag down to all slaves
573 static int bond_set_allmulti(struct bonding *bond, int inc)
575 struct list_head *iter;
578 if (USES_PRIMARY(bond->params.mode)) {
579 /* write lock already acquired */
580 if (bond->curr_active_slave) {
581 err = dev_set_allmulti(bond->curr_active_slave->dev,
587 bond_for_each_slave(bond, slave, iter) {
588 err = dev_set_allmulti(slave->dev, inc);
597 * Retrieve the list of registered multicast addresses for the bonding
598 * device and retransmit an IGMP JOIN request to the current active
601 static void bond_resend_igmp_join_requests_delayed(struct work_struct *work)
603 struct bonding *bond = container_of(work, struct bonding,
606 if (!rtnl_trylock()) {
607 queue_delayed_work(bond->wq, &bond->mcast_work, 1);
610 call_netdevice_notifiers(NETDEV_RESEND_IGMP, bond->dev);
612 if (bond->igmp_retrans > 1) {
613 bond->igmp_retrans--;
614 queue_delayed_work(bond->wq, &bond->mcast_work, HZ/5);
619 /* Flush bond's hardware addresses from slave
621 static void bond_hw_addr_flush(struct net_device *bond_dev,
622 struct net_device *slave_dev)
624 struct bonding *bond = netdev_priv(bond_dev);
626 dev_uc_unsync(slave_dev, bond_dev);
627 dev_mc_unsync(slave_dev, bond_dev);
629 if (bond->params.mode == BOND_MODE_8023AD) {
630 /* del lacpdu mc addr from mc list */
631 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
633 dev_mc_del(slave_dev, lacpdu_multicast);
637 /*--------------------------- Active slave change ---------------------------*/
639 /* Update the hardware address list and promisc/allmulti for the new and
640 * old active slaves (if any). Modes that are !USES_PRIMARY keep all
641 * slaves up date at all times; only the USES_PRIMARY modes need to call
642 * this function to swap these settings during a failover.
644 static void bond_hw_addr_swap(struct bonding *bond, struct slave *new_active,
645 struct slave *old_active)
650 if (bond->dev->flags & IFF_PROMISC)
651 dev_set_promiscuity(old_active->dev, -1);
653 if (bond->dev->flags & IFF_ALLMULTI)
654 dev_set_allmulti(old_active->dev, -1);
656 bond_hw_addr_flush(bond->dev, old_active->dev);
660 /* FIXME: Signal errors upstream. */
661 if (bond->dev->flags & IFF_PROMISC)
662 dev_set_promiscuity(new_active->dev, 1);
664 if (bond->dev->flags & IFF_ALLMULTI)
665 dev_set_allmulti(new_active->dev, 1);
667 netif_addr_lock_bh(bond->dev);
668 dev_uc_sync(new_active->dev, bond->dev);
669 dev_mc_sync(new_active->dev, bond->dev);
670 netif_addr_unlock_bh(bond->dev);
675 * bond_set_dev_addr - clone slave's address to bond
676 * @bond_dev: bond net device
677 * @slave_dev: slave net device
679 * Should be called with RTNL held.
681 static void bond_set_dev_addr(struct net_device *bond_dev,
682 struct net_device *slave_dev)
684 pr_debug("bond_dev=%p slave_dev=%p slave_dev->addr_len=%d\n",
685 bond_dev, slave_dev, slave_dev->addr_len);
686 memcpy(bond_dev->dev_addr, slave_dev->dev_addr, slave_dev->addr_len);
687 bond_dev->addr_assign_type = NET_ADDR_STOLEN;
688 call_netdevice_notifiers(NETDEV_CHANGEADDR, bond_dev);
692 * bond_do_fail_over_mac
694 * Perform special MAC address swapping for fail_over_mac settings
696 * Called with RTNL, curr_slave_lock for write_bh.
698 static void bond_do_fail_over_mac(struct bonding *bond,
699 struct slave *new_active,
700 struct slave *old_active)
701 __releases(&bond->curr_slave_lock)
702 __acquires(&bond->curr_slave_lock)
704 u8 tmp_mac[ETH_ALEN];
705 struct sockaddr saddr;
708 switch (bond->params.fail_over_mac) {
709 case BOND_FOM_ACTIVE:
711 write_unlock_bh(&bond->curr_slave_lock);
712 bond_set_dev_addr(bond->dev, new_active->dev);
713 write_lock_bh(&bond->curr_slave_lock);
716 case BOND_FOM_FOLLOW:
718 * if new_active && old_active, swap them
719 * if just old_active, do nothing (going to no active slave)
720 * if just new_active, set new_active to bond's MAC
725 write_unlock_bh(&bond->curr_slave_lock);
728 memcpy(tmp_mac, new_active->dev->dev_addr, ETH_ALEN);
729 memcpy(saddr.sa_data, old_active->dev->dev_addr,
731 saddr.sa_family = new_active->dev->type;
733 memcpy(saddr.sa_data, bond->dev->dev_addr, ETH_ALEN);
734 saddr.sa_family = bond->dev->type;
737 rv = dev_set_mac_address(new_active->dev, &saddr);
739 pr_err("%s: Error %d setting MAC of slave %s\n",
740 bond->dev->name, -rv, new_active->dev->name);
747 memcpy(saddr.sa_data, tmp_mac, ETH_ALEN);
748 saddr.sa_family = old_active->dev->type;
750 rv = dev_set_mac_address(old_active->dev, &saddr);
752 pr_err("%s: Error %d setting MAC of slave %s\n",
753 bond->dev->name, -rv, new_active->dev->name);
755 write_lock_bh(&bond->curr_slave_lock);
758 pr_err("%s: bond_do_fail_over_mac impossible: bad policy %d\n",
759 bond->dev->name, bond->params.fail_over_mac);
765 static bool bond_should_change_active(struct bonding *bond)
767 struct slave *prim = bond->primary_slave;
768 struct slave *curr = bond->curr_active_slave;
770 if (!prim || !curr || curr->link != BOND_LINK_UP)
772 if (bond->force_primary) {
773 bond->force_primary = false;
776 if (bond->params.primary_reselect == BOND_PRI_RESELECT_BETTER &&
777 (prim->speed < curr->speed ||
778 (prim->speed == curr->speed && prim->duplex <= curr->duplex)))
780 if (bond->params.primary_reselect == BOND_PRI_RESELECT_FAILURE)
786 * find_best_interface - select the best available slave to be the active one
787 * @bond: our bonding struct
789 static struct slave *bond_find_best_slave(struct bonding *bond)
791 struct slave *slave, *bestslave = NULL;
792 struct list_head *iter;
793 int mintime = bond->params.updelay;
795 if (bond->primary_slave && bond->primary_slave->link == BOND_LINK_UP &&
796 bond_should_change_active(bond))
797 return bond->primary_slave;
799 bond_for_each_slave(bond, slave, iter) {
800 if (slave->link == BOND_LINK_UP)
802 if (slave->link == BOND_LINK_BACK && IS_UP(slave->dev) &&
803 slave->delay < mintime) {
804 mintime = slave->delay;
812 static bool bond_should_notify_peers(struct bonding *bond)
817 slave = rcu_dereference(bond->curr_active_slave);
820 pr_debug("bond_should_notify_peers: bond %s slave %s\n",
821 bond->dev->name, slave ? slave->dev->name : "NULL");
823 if (!slave || !bond->send_peer_notif ||
824 test_bit(__LINK_STATE_LINKWATCH_PENDING, &slave->dev->state))
831 * change_active_interface - change the active slave into the specified one
832 * @bond: our bonding struct
833 * @new: the new slave to make the active one
835 * Set the new slave to the bond's settings and unset them on the old
837 * Setting include flags, mc-list, promiscuity, allmulti, etc.
839 * If @new's link state is %BOND_LINK_BACK we'll set it to %BOND_LINK_UP,
840 * because it is apparently the best available slave we have, even though its
841 * updelay hasn't timed out yet.
843 * If new_active is not NULL, caller must hold curr_slave_lock for write_bh.
845 void bond_change_active_slave(struct bonding *bond, struct slave *new_active)
847 struct slave *old_active = bond->curr_active_slave;
849 if (old_active == new_active)
853 new_active->jiffies = jiffies;
855 if (new_active->link == BOND_LINK_BACK) {
856 if (USES_PRIMARY(bond->params.mode)) {
857 pr_info("%s: making interface %s the new active one %d ms earlier.\n",
858 bond->dev->name, new_active->dev->name,
859 (bond->params.updelay - new_active->delay) * bond->params.miimon);
862 new_active->delay = 0;
863 new_active->link = BOND_LINK_UP;
865 if (bond->params.mode == BOND_MODE_8023AD)
866 bond_3ad_handle_link_change(new_active, BOND_LINK_UP);
868 if (bond_is_lb(bond))
869 bond_alb_handle_link_change(bond, new_active, BOND_LINK_UP);
871 if (USES_PRIMARY(bond->params.mode)) {
872 pr_info("%s: making interface %s the new active one.\n",
873 bond->dev->name, new_active->dev->name);
878 if (USES_PRIMARY(bond->params.mode))
879 bond_hw_addr_swap(bond, new_active, old_active);
881 if (bond_is_lb(bond)) {
882 bond_alb_handle_active_change(bond, new_active);
884 bond_set_slave_inactive_flags(old_active);
886 bond_set_slave_active_flags(new_active);
888 rcu_assign_pointer(bond->curr_active_slave, new_active);
891 if (bond->params.mode == BOND_MODE_ACTIVEBACKUP) {
893 bond_set_slave_inactive_flags(old_active);
896 bool should_notify_peers = false;
898 bond_set_slave_active_flags(new_active);
900 if (bond->params.fail_over_mac)
901 bond_do_fail_over_mac(bond, new_active,
904 if (netif_running(bond->dev)) {
905 bond->send_peer_notif =
906 bond->params.num_peer_notif;
907 should_notify_peers =
908 bond_should_notify_peers(bond);
911 write_unlock_bh(&bond->curr_slave_lock);
913 call_netdevice_notifiers(NETDEV_BONDING_FAILOVER, bond->dev);
914 if (should_notify_peers)
915 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS,
918 write_lock_bh(&bond->curr_slave_lock);
922 /* resend IGMP joins since active slave has changed or
923 * all were sent on curr_active_slave.
924 * resend only if bond is brought up with the affected
925 * bonding modes and the retransmission is enabled */
926 if (netif_running(bond->dev) && (bond->params.resend_igmp > 0) &&
927 ((USES_PRIMARY(bond->params.mode) && new_active) ||
928 bond->params.mode == BOND_MODE_ROUNDROBIN)) {
929 bond->igmp_retrans = bond->params.resend_igmp;
930 queue_delayed_work(bond->wq, &bond->mcast_work, 1);
935 * bond_select_active_slave - select a new active slave, if needed
936 * @bond: our bonding struct
938 * This functions should be called when one of the following occurs:
939 * - The old curr_active_slave has been released or lost its link.
940 * - The primary_slave has got its link back.
941 * - A slave has got its link back and there's no old curr_active_slave.
943 * Caller must hold curr_slave_lock for write_bh.
945 void bond_select_active_slave(struct bonding *bond)
947 struct slave *best_slave;
950 best_slave = bond_find_best_slave(bond);
951 if (best_slave != bond->curr_active_slave) {
952 bond_change_active_slave(bond, best_slave);
953 rv = bond_set_carrier(bond);
957 if (netif_carrier_ok(bond->dev)) {
958 pr_info("%s: first active interface up!\n",
961 pr_info("%s: now running without any active interface !\n",
967 #ifdef CONFIG_NET_POLL_CONTROLLER
968 static inline int slave_enable_netpoll(struct slave *slave)
973 np = kzalloc(sizeof(*np), GFP_ATOMIC);
978 err = __netpoll_setup(np, slave->dev, GFP_ATOMIC);
987 static inline void slave_disable_netpoll(struct slave *slave)
989 struct netpoll *np = slave->np;
995 __netpoll_free_async(np);
997 static inline bool slave_dev_support_netpoll(struct net_device *slave_dev)
999 if (slave_dev->priv_flags & IFF_DISABLE_NETPOLL)
1001 if (!slave_dev->netdev_ops->ndo_poll_controller)
1006 static void bond_poll_controller(struct net_device *bond_dev)
1010 static void bond_netpoll_cleanup(struct net_device *bond_dev)
1012 struct bonding *bond = netdev_priv(bond_dev);
1013 struct list_head *iter;
1014 struct slave *slave;
1016 bond_for_each_slave(bond, slave, iter)
1017 if (IS_UP(slave->dev))
1018 slave_disable_netpoll(slave);
1021 static int bond_netpoll_setup(struct net_device *dev, struct netpoll_info *ni, gfp_t gfp)
1023 struct bonding *bond = netdev_priv(dev);
1024 struct list_head *iter;
1025 struct slave *slave;
1028 bond_for_each_slave(bond, slave, iter) {
1029 err = slave_enable_netpoll(slave);
1031 bond_netpoll_cleanup(dev);
1038 static inline int slave_enable_netpoll(struct slave *slave)
1042 static inline void slave_disable_netpoll(struct slave *slave)
1045 static void bond_netpoll_cleanup(struct net_device *bond_dev)
1050 /*---------------------------------- IOCTL ----------------------------------*/
1052 static netdev_features_t bond_fix_features(struct net_device *dev,
1053 netdev_features_t features)
1055 struct bonding *bond = netdev_priv(dev);
1056 struct list_head *iter;
1057 netdev_features_t mask;
1058 struct slave *slave;
1060 if (!bond_has_slaves(bond)) {
1061 /* Disable adding VLANs to empty bond. But why? --mq */
1062 features |= NETIF_F_VLAN_CHALLENGED;
1067 features &= ~NETIF_F_ONE_FOR_ALL;
1068 features |= NETIF_F_ALL_FOR_ALL;
1070 bond_for_each_slave(bond, slave, iter) {
1071 features = netdev_increment_features(features,
1072 slave->dev->features,
1075 features = netdev_add_tso_features(features, mask);
1080 #define BOND_VLAN_FEATURES (NETIF_F_ALL_CSUM | NETIF_F_SG | \
1081 NETIF_F_FRAGLIST | NETIF_F_ALL_TSO | \
1082 NETIF_F_HIGHDMA | NETIF_F_LRO)
1084 static void bond_compute_features(struct bonding *bond)
1086 unsigned int flags, dst_release_flag = IFF_XMIT_DST_RELEASE;
1087 netdev_features_t vlan_features = BOND_VLAN_FEATURES;
1088 struct net_device *bond_dev = bond->dev;
1089 struct list_head *iter;
1090 struct slave *slave;
1091 unsigned short max_hard_header_len = ETH_HLEN;
1092 unsigned int gso_max_size = GSO_MAX_SIZE;
1093 u16 gso_max_segs = GSO_MAX_SEGS;
1095 if (!bond_has_slaves(bond))
1098 bond_for_each_slave(bond, slave, iter) {
1099 vlan_features = netdev_increment_features(vlan_features,
1100 slave->dev->vlan_features, BOND_VLAN_FEATURES);
1102 dst_release_flag &= slave->dev->priv_flags;
1103 if (slave->dev->hard_header_len > max_hard_header_len)
1104 max_hard_header_len = slave->dev->hard_header_len;
1106 gso_max_size = min(gso_max_size, slave->dev->gso_max_size);
1107 gso_max_segs = min(gso_max_segs, slave->dev->gso_max_segs);
1111 bond_dev->vlan_features = vlan_features;
1112 bond_dev->hard_header_len = max_hard_header_len;
1113 bond_dev->gso_max_segs = gso_max_segs;
1114 netif_set_gso_max_size(bond_dev, gso_max_size);
1116 flags = bond_dev->priv_flags & ~IFF_XMIT_DST_RELEASE;
1117 bond_dev->priv_flags = flags | dst_release_flag;
1119 netdev_change_features(bond_dev);
1122 static void bond_setup_by_slave(struct net_device *bond_dev,
1123 struct net_device *slave_dev)
1125 bond_dev->header_ops = slave_dev->header_ops;
1127 bond_dev->type = slave_dev->type;
1128 bond_dev->hard_header_len = slave_dev->hard_header_len;
1129 bond_dev->addr_len = slave_dev->addr_len;
1131 memcpy(bond_dev->broadcast, slave_dev->broadcast,
1132 slave_dev->addr_len);
1135 /* On bonding slaves other than the currently active slave, suppress
1136 * duplicates except for alb non-mcast/bcast.
1138 static bool bond_should_deliver_exact_match(struct sk_buff *skb,
1139 struct slave *slave,
1140 struct bonding *bond)
1142 if (bond_is_slave_inactive(slave)) {
1143 if (bond->params.mode == BOND_MODE_ALB &&
1144 skb->pkt_type != PACKET_BROADCAST &&
1145 skb->pkt_type != PACKET_MULTICAST)
1152 static rx_handler_result_t bond_handle_frame(struct sk_buff **pskb)
1154 struct sk_buff *skb = *pskb;
1155 struct slave *slave;
1156 struct bonding *bond;
1157 int (*recv_probe)(const struct sk_buff *, struct bonding *,
1159 int ret = RX_HANDLER_ANOTHER;
1161 skb = skb_share_check(skb, GFP_ATOMIC);
1163 return RX_HANDLER_CONSUMED;
1167 slave = bond_slave_get_rcu(skb->dev);
1170 if (bond->params.arp_interval)
1171 slave->dev->last_rx = jiffies;
1173 recv_probe = ACCESS_ONCE(bond->recv_probe);
1175 ret = recv_probe(skb, bond, slave);
1176 if (ret == RX_HANDLER_CONSUMED) {
1182 if (bond_should_deliver_exact_match(skb, slave, bond)) {
1183 return RX_HANDLER_EXACT;
1186 skb->dev = bond->dev;
1188 if (bond->params.mode == BOND_MODE_ALB &&
1189 bond->dev->priv_flags & IFF_BRIDGE_PORT &&
1190 skb->pkt_type == PACKET_HOST) {
1192 if (unlikely(skb_cow_head(skb,
1193 skb->data - skb_mac_header(skb)))) {
1195 return RX_HANDLER_CONSUMED;
1197 memcpy(eth_hdr(skb)->h_dest, bond->dev->dev_addr, ETH_ALEN);
1203 static int bond_master_upper_dev_link(struct net_device *bond_dev,
1204 struct net_device *slave_dev,
1205 struct slave *slave)
1209 err = netdev_master_upper_dev_link_private(slave_dev, bond_dev, slave);
1212 slave_dev->flags |= IFF_SLAVE;
1213 rtmsg_ifinfo(RTM_NEWLINK, slave_dev, IFF_SLAVE, GFP_KERNEL);
1217 static void bond_upper_dev_unlink(struct net_device *bond_dev,
1218 struct net_device *slave_dev)
1220 netdev_upper_dev_unlink(slave_dev, bond_dev);
1221 slave_dev->flags &= ~IFF_SLAVE;
1222 rtmsg_ifinfo(RTM_NEWLINK, slave_dev, IFF_SLAVE, GFP_KERNEL);
1225 /* enslave device <slave> to bond device <master> */
1226 int bond_enslave(struct net_device *bond_dev, struct net_device *slave_dev)
1228 struct bonding *bond = netdev_priv(bond_dev);
1229 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
1230 struct slave *new_slave = NULL, *prev_slave;
1231 struct sockaddr addr;
1235 if (!bond->params.use_carrier &&
1236 slave_dev->ethtool_ops->get_link == NULL &&
1237 slave_ops->ndo_do_ioctl == NULL) {
1238 pr_warning("%s: Warning: no link monitoring support for %s\n",
1239 bond_dev->name, slave_dev->name);
1242 /* already enslaved */
1243 if (slave_dev->flags & IFF_SLAVE) {
1244 pr_debug("Error, Device was already enslaved\n");
1248 /* vlan challenged mutual exclusion */
1249 /* no need to lock since we're protected by rtnl_lock */
1250 if (slave_dev->features & NETIF_F_VLAN_CHALLENGED) {
1251 pr_debug("%s: NETIF_F_VLAN_CHALLENGED\n", slave_dev->name);
1252 if (vlan_uses_dev(bond_dev)) {
1253 pr_err("%s: Error: cannot enslave VLAN challenged slave %s on VLAN enabled bond %s\n",
1254 bond_dev->name, slave_dev->name, bond_dev->name);
1257 pr_warning("%s: Warning: enslaved VLAN challenged slave %s. Adding VLANs will be blocked as long as %s is part of bond %s\n",
1258 bond_dev->name, slave_dev->name,
1259 slave_dev->name, bond_dev->name);
1262 pr_debug("%s: ! NETIF_F_VLAN_CHALLENGED\n", slave_dev->name);
1266 * Old ifenslave binaries are no longer supported. These can
1267 * be identified with moderate accuracy by the state of the slave:
1268 * the current ifenslave will set the interface down prior to
1269 * enslaving it; the old ifenslave will not.
1271 if ((slave_dev->flags & IFF_UP)) {
1272 pr_err("%s is up. This may be due to an out of date ifenslave.\n",
1275 goto err_undo_flags;
1278 /* set bonding device ether type by slave - bonding netdevices are
1279 * created with ether_setup, so when the slave type is not ARPHRD_ETHER
1280 * there is a need to override some of the type dependent attribs/funcs.
1282 * bond ether type mutual exclusion - don't allow slaves of dissimilar
1283 * ether type (eg ARPHRD_ETHER and ARPHRD_INFINIBAND) share the same bond
1285 if (!bond_has_slaves(bond)) {
1286 if (bond_dev->type != slave_dev->type) {
1287 pr_debug("%s: change device type from %d to %d\n",
1289 bond_dev->type, slave_dev->type);
1291 res = call_netdevice_notifiers(NETDEV_PRE_TYPE_CHANGE,
1293 res = notifier_to_errno(res);
1295 pr_err("%s: refused to change device type\n",
1298 goto err_undo_flags;
1301 /* Flush unicast and multicast addresses */
1302 dev_uc_flush(bond_dev);
1303 dev_mc_flush(bond_dev);
1305 if (slave_dev->type != ARPHRD_ETHER)
1306 bond_setup_by_slave(bond_dev, slave_dev);
1308 ether_setup(bond_dev);
1309 bond_dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1312 call_netdevice_notifiers(NETDEV_POST_TYPE_CHANGE,
1315 } else if (bond_dev->type != slave_dev->type) {
1316 pr_err("%s ether type (%d) is different from other slaves (%d), can not enslave it.\n",
1318 slave_dev->type, bond_dev->type);
1320 goto err_undo_flags;
1323 if (slave_ops->ndo_set_mac_address == NULL) {
1324 if (!bond_has_slaves(bond)) {
1325 pr_warning("%s: Warning: The first slave device specified does not support setting the MAC address. Setting fail_over_mac to active.",
1327 bond->params.fail_over_mac = BOND_FOM_ACTIVE;
1328 } else if (bond->params.fail_over_mac != BOND_FOM_ACTIVE) {
1329 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",
1332 goto err_undo_flags;
1336 call_netdevice_notifiers(NETDEV_JOIN, slave_dev);
1338 /* If this is the first slave, then we need to set the master's hardware
1339 * address to be the same as the slave's. */
1340 if (!bond_has_slaves(bond) &&
1341 bond->dev->addr_assign_type == NET_ADDR_RANDOM)
1342 bond_set_dev_addr(bond->dev, slave_dev);
1344 new_slave = kzalloc(sizeof(struct slave), GFP_KERNEL);
1347 goto err_undo_flags;
1350 * Set the new_slave's queue_id to be zero. Queue ID mapping
1351 * is set via sysfs or module option if desired.
1353 new_slave->queue_id = 0;
1355 /* Save slave's original mtu and then set it to match the bond */
1356 new_slave->original_mtu = slave_dev->mtu;
1357 res = dev_set_mtu(slave_dev, bond->dev->mtu);
1359 pr_debug("Error %d calling dev_set_mtu\n", res);
1364 * Save slave's original ("permanent") mac address for modes
1365 * that need it, and for restoring it upon release, and then
1366 * set it to the master's address
1368 memcpy(new_slave->perm_hwaddr, slave_dev->dev_addr, ETH_ALEN);
1370 if (!bond->params.fail_over_mac) {
1372 * Set slave to master's mac address. The application already
1373 * set the master's mac address to that of the first slave
1375 memcpy(addr.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
1376 addr.sa_family = slave_dev->type;
1377 res = dev_set_mac_address(slave_dev, &addr);
1379 pr_debug("Error %d calling set_mac_address\n", res);
1380 goto err_restore_mtu;
1384 /* open the slave since the application closed it */
1385 res = dev_open(slave_dev);
1387 pr_debug("Opening slave %s failed\n", slave_dev->name);
1388 goto err_restore_mac;
1391 new_slave->bond = bond;
1392 new_slave->dev = slave_dev;
1393 slave_dev->priv_flags |= IFF_BONDING;
1395 if (bond_is_lb(bond)) {
1396 /* bond_alb_init_slave() must be called before all other stages since
1397 * it might fail and we do not want to have to undo everything
1399 res = bond_alb_init_slave(bond, new_slave);
1404 /* If the mode USES_PRIMARY, then the following is handled by
1405 * bond_change_active_slave().
1407 if (!USES_PRIMARY(bond->params.mode)) {
1408 /* set promiscuity level to new slave */
1409 if (bond_dev->flags & IFF_PROMISC) {
1410 res = dev_set_promiscuity(slave_dev, 1);
1415 /* set allmulti level to new slave */
1416 if (bond_dev->flags & IFF_ALLMULTI) {
1417 res = dev_set_allmulti(slave_dev, 1);
1422 netif_addr_lock_bh(bond_dev);
1424 dev_mc_sync_multiple(slave_dev, bond_dev);
1425 dev_uc_sync_multiple(slave_dev, bond_dev);
1427 netif_addr_unlock_bh(bond_dev);
1430 if (bond->params.mode == BOND_MODE_8023AD) {
1431 /* add lacpdu mc addr to mc list */
1432 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
1434 dev_mc_add(slave_dev, lacpdu_multicast);
1437 res = vlan_vids_add_by_dev(slave_dev, bond_dev);
1439 pr_err("%s: Error: Couldn't add bond vlan ids to %s\n",
1440 bond_dev->name, slave_dev->name);
1444 prev_slave = bond_last_slave(bond);
1446 new_slave->delay = 0;
1447 new_slave->link_failure_count = 0;
1449 bond_update_speed_duplex(new_slave);
1451 new_slave->last_arp_rx = jiffies -
1452 (msecs_to_jiffies(bond->params.arp_interval) + 1);
1453 for (i = 0; i < BOND_MAX_ARP_TARGETS; i++)
1454 new_slave->target_last_arp_rx[i] = new_slave->last_arp_rx;
1456 if (bond->params.miimon && !bond->params.use_carrier) {
1457 link_reporting = bond_check_dev_link(bond, slave_dev, 1);
1459 if ((link_reporting == -1) && !bond->params.arp_interval) {
1461 * miimon is set but a bonded network driver
1462 * does not support ETHTOOL/MII and
1463 * arp_interval is not set. Note: if
1464 * use_carrier is enabled, we will never go
1465 * here (because netif_carrier is always
1466 * supported); thus, we don't need to change
1467 * the messages for netif_carrier.
1469 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",
1470 bond_dev->name, slave_dev->name);
1471 } else if (link_reporting == -1) {
1472 /* unable get link status using mii/ethtool */
1473 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",
1474 bond_dev->name, slave_dev->name);
1478 /* check for initial state */
1479 if (bond->params.miimon) {
1480 if (bond_check_dev_link(bond, slave_dev, 0) == BMSR_LSTATUS) {
1481 if (bond->params.updelay) {
1482 new_slave->link = BOND_LINK_BACK;
1483 new_slave->delay = bond->params.updelay;
1485 new_slave->link = BOND_LINK_UP;
1488 new_slave->link = BOND_LINK_DOWN;
1490 } else if (bond->params.arp_interval) {
1491 new_slave->link = (netif_carrier_ok(slave_dev) ?
1492 BOND_LINK_UP : BOND_LINK_DOWN);
1494 new_slave->link = BOND_LINK_UP;
1497 if (new_slave->link != BOND_LINK_DOWN)
1498 new_slave->jiffies = jiffies;
1499 pr_debug("Initial state of slave_dev is BOND_LINK_%s\n",
1500 new_slave->link == BOND_LINK_DOWN ? "DOWN" :
1501 (new_slave->link == BOND_LINK_UP ? "UP" : "BACK"));
1503 if (USES_PRIMARY(bond->params.mode) && bond->params.primary[0]) {
1504 /* if there is a primary slave, remember it */
1505 if (strcmp(bond->params.primary, new_slave->dev->name) == 0) {
1506 bond->primary_slave = new_slave;
1507 bond->force_primary = true;
1511 switch (bond->params.mode) {
1512 case BOND_MODE_ACTIVEBACKUP:
1513 bond_set_slave_inactive_flags(new_slave);
1515 case BOND_MODE_8023AD:
1516 /* in 802.3ad mode, the internal mechanism
1517 * will activate the slaves in the selected
1520 bond_set_slave_inactive_flags(new_slave);
1521 /* if this is the first slave */
1523 SLAVE_AD_INFO(new_slave).id = 1;
1524 /* Initialize AD with the number of times that the AD timer is called in 1 second
1525 * can be called only after the mac address of the bond is set
1527 bond_3ad_initialize(bond, 1000/AD_TIMER_INTERVAL);
1529 SLAVE_AD_INFO(new_slave).id =
1530 SLAVE_AD_INFO(prev_slave).id + 1;
1533 bond_3ad_bind_slave(new_slave);
1537 bond_set_active_slave(new_slave);
1538 bond_set_slave_inactive_flags(new_slave);
1541 pr_debug("This slave is always active in trunk mode\n");
1543 /* always active in trunk mode */
1544 bond_set_active_slave(new_slave);
1546 /* In trunking mode there is little meaning to curr_active_slave
1547 * anyway (it holds no special properties of the bond device),
1548 * so we can change it without calling change_active_interface()
1550 if (!bond->curr_active_slave && new_slave->link == BOND_LINK_UP)
1551 rcu_assign_pointer(bond->curr_active_slave, new_slave);
1554 } /* switch(bond_mode) */
1556 #ifdef CONFIG_NET_POLL_CONTROLLER
1557 slave_dev->npinfo = bond->dev->npinfo;
1558 if (slave_dev->npinfo) {
1559 if (slave_enable_netpoll(new_slave)) {
1560 read_unlock(&bond->lock);
1561 pr_info("Error, %s: master_dev is using netpoll, "
1562 "but new slave device does not support netpoll.\n",
1570 res = netdev_rx_handler_register(slave_dev, bond_handle_frame,
1573 pr_debug("Error %d calling netdev_rx_handler_register\n", res);
1577 res = bond_master_upper_dev_link(bond_dev, slave_dev, new_slave);
1579 pr_debug("Error %d calling bond_master_upper_dev_link\n", res);
1580 goto err_unregister;
1583 res = bond_sysfs_slave_add(new_slave);
1585 pr_debug("Error %d calling bond_sysfs_slave_add\n", res);
1586 goto err_upper_unlink;
1590 bond_compute_features(bond);
1591 bond_set_carrier(bond);
1593 if (USES_PRIMARY(bond->params.mode)) {
1594 write_lock_bh(&bond->curr_slave_lock);
1595 bond_select_active_slave(bond);
1596 write_unlock_bh(&bond->curr_slave_lock);
1599 pr_info("%s: enslaving %s as a%s interface with a%s link.\n",
1600 bond_dev->name, slave_dev->name,
1601 bond_is_active_slave(new_slave) ? "n active" : " backup",
1602 new_slave->link != BOND_LINK_DOWN ? "n up" : " down");
1604 /* enslave is successful */
1607 /* Undo stages on error */
1609 bond_upper_dev_unlink(bond_dev, slave_dev);
1612 netdev_rx_handler_unregister(slave_dev);
1615 if (!USES_PRIMARY(bond->params.mode))
1616 bond_hw_addr_flush(bond_dev, slave_dev);
1618 vlan_vids_del_by_dev(slave_dev, bond_dev);
1619 if (bond->primary_slave == new_slave)
1620 bond->primary_slave = NULL;
1621 if (bond->curr_active_slave == new_slave) {
1622 write_lock_bh(&bond->curr_slave_lock);
1623 bond_change_active_slave(bond, NULL);
1624 bond_select_active_slave(bond);
1625 write_unlock_bh(&bond->curr_slave_lock);
1627 slave_disable_netpoll(new_slave);
1630 slave_dev->priv_flags &= ~IFF_BONDING;
1631 dev_close(slave_dev);
1634 if (!bond->params.fail_over_mac) {
1635 /* XXX TODO - fom follow mode needs to change master's
1636 * MAC if this slave's MAC is in use by the bond, or at
1637 * least print a warning.
1639 memcpy(addr.sa_data, new_slave->perm_hwaddr, ETH_ALEN);
1640 addr.sa_family = slave_dev->type;
1641 dev_set_mac_address(slave_dev, &addr);
1645 dev_set_mtu(slave_dev, new_slave->original_mtu);
1651 /* Enslave of first slave has failed and we need to fix master's mac */
1652 if (!bond_has_slaves(bond) &&
1653 ether_addr_equal_64bits(bond_dev->dev_addr, slave_dev->dev_addr))
1654 eth_hw_addr_random(bond_dev);
1660 * Try to release the slave device <slave> from the bond device <master>
1661 * It is legal to access curr_active_slave without a lock because all the function
1662 * is write-locked. If "all" is true it means that the function is being called
1663 * while destroying a bond interface and all slaves are being released.
1665 * The rules for slave state should be:
1666 * for Active/Backup:
1667 * Active stays on all backups go down
1668 * for Bonded connections:
1669 * The first up interface should be left on and all others downed.
1671 static int __bond_release_one(struct net_device *bond_dev,
1672 struct net_device *slave_dev,
1675 struct bonding *bond = netdev_priv(bond_dev);
1676 struct slave *slave, *oldcurrent;
1677 struct sockaddr addr;
1678 int old_flags = bond_dev->flags;
1679 netdev_features_t old_features = bond_dev->features;
1681 /* slave is not a slave or master is not master of this slave */
1682 if (!(slave_dev->flags & IFF_SLAVE) ||
1683 !netdev_has_upper_dev(slave_dev, bond_dev)) {
1684 pr_err("%s: Error: cannot release %s.\n",
1685 bond_dev->name, slave_dev->name);
1691 slave = bond_get_slave_by_dev(bond, slave_dev);
1693 /* not a slave of this bond */
1694 pr_info("%s: %s not enslaved\n",
1695 bond_dev->name, slave_dev->name);
1696 unblock_netpoll_tx();
1700 /* release the slave from its bond */
1703 bond_sysfs_slave_del(slave);
1705 bond_upper_dev_unlink(bond_dev, slave_dev);
1706 /* unregister rx_handler early so bond_handle_frame wouldn't be called
1707 * for this slave anymore.
1709 netdev_rx_handler_unregister(slave_dev);
1710 write_lock_bh(&bond->lock);
1712 /* Inform AD package of unbinding of slave. */
1713 if (bond->params.mode == BOND_MODE_8023AD)
1714 bond_3ad_unbind_slave(slave);
1716 write_unlock_bh(&bond->lock);
1718 pr_info("%s: releasing %s interface %s\n",
1720 bond_is_active_slave(slave) ? "active" : "backup",
1723 oldcurrent = bond->curr_active_slave;
1725 bond->current_arp_slave = NULL;
1727 if (!all && !bond->params.fail_over_mac) {
1728 if (ether_addr_equal_64bits(bond_dev->dev_addr, slave->perm_hwaddr) &&
1729 bond_has_slaves(bond))
1730 pr_warn("%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",
1731 bond_dev->name, slave_dev->name,
1733 bond_dev->name, slave_dev->name);
1736 if (bond->primary_slave == slave)
1737 bond->primary_slave = NULL;
1739 if (oldcurrent == slave) {
1740 write_lock_bh(&bond->curr_slave_lock);
1741 bond_change_active_slave(bond, NULL);
1742 write_unlock_bh(&bond->curr_slave_lock);
1745 if (bond_is_lb(bond)) {
1746 /* Must be called only after the slave has been
1747 * detached from the list and the curr_active_slave
1748 * has been cleared (if our_slave == old_current),
1749 * but before a new active slave is selected.
1751 bond_alb_deinit_slave(bond, slave);
1755 rcu_assign_pointer(bond->curr_active_slave, NULL);
1756 } else if (oldcurrent == slave) {
1758 * Note that we hold RTNL over this sequence, so there
1759 * is no concern that another slave add/remove event
1762 write_lock_bh(&bond->curr_slave_lock);
1764 bond_select_active_slave(bond);
1766 write_unlock_bh(&bond->curr_slave_lock);
1769 if (!bond_has_slaves(bond)) {
1770 bond_set_carrier(bond);
1771 eth_hw_addr_random(bond_dev);
1773 if (vlan_uses_dev(bond_dev)) {
1774 pr_warning("%s: Warning: clearing HW address of %s while it still has VLANs.\n",
1775 bond_dev->name, bond_dev->name);
1776 pr_warning("%s: When re-adding slaves, make sure the bond's HW address matches its VLANs'.\n",
1781 unblock_netpoll_tx();
1784 if (!bond_has_slaves(bond)) {
1785 call_netdevice_notifiers(NETDEV_CHANGEADDR, bond->dev);
1786 call_netdevice_notifiers(NETDEV_RELEASE, bond->dev);
1789 bond_compute_features(bond);
1790 if (!(bond_dev->features & NETIF_F_VLAN_CHALLENGED) &&
1791 (old_features & NETIF_F_VLAN_CHALLENGED))
1792 pr_info("%s: last VLAN challenged slave %s left bond %s. VLAN blocking is removed\n",
1793 bond_dev->name, slave_dev->name, bond_dev->name);
1795 /* must do this from outside any spinlocks */
1796 vlan_vids_del_by_dev(slave_dev, bond_dev);
1798 /* If the mode USES_PRIMARY, then this cases was handled above by
1799 * bond_change_active_slave(..., NULL)
1801 if (!USES_PRIMARY(bond->params.mode)) {
1802 /* unset promiscuity level from slave
1803 * NOTE: The NETDEV_CHANGEADDR call above may change the value
1804 * of the IFF_PROMISC flag in the bond_dev, but we need the
1805 * value of that flag before that change, as that was the value
1806 * when this slave was attached, so we cache at the start of the
1807 * function and use it here. Same goes for ALLMULTI below
1809 if (old_flags & IFF_PROMISC)
1810 dev_set_promiscuity(slave_dev, -1);
1812 /* unset allmulti level from slave */
1813 if (old_flags & IFF_ALLMULTI)
1814 dev_set_allmulti(slave_dev, -1);
1816 bond_hw_addr_flush(bond_dev, slave_dev);
1819 slave_disable_netpoll(slave);
1821 /* close slave before restoring its mac address */
1822 dev_close(slave_dev);
1824 if (bond->params.fail_over_mac != BOND_FOM_ACTIVE) {
1825 /* restore original ("permanent") mac address */
1826 memcpy(addr.sa_data, slave->perm_hwaddr, ETH_ALEN);
1827 addr.sa_family = slave_dev->type;
1828 dev_set_mac_address(slave_dev, &addr);
1831 dev_set_mtu(slave_dev, slave->original_mtu);
1833 slave_dev->priv_flags &= ~IFF_BONDING;
1837 return 0; /* deletion OK */
1840 /* A wrapper used because of ndo_del_link */
1841 int bond_release(struct net_device *bond_dev, struct net_device *slave_dev)
1843 return __bond_release_one(bond_dev, slave_dev, false);
1847 * First release a slave and then destroy the bond if no more slaves are left.
1848 * Must be under rtnl_lock when this function is called.
1850 static int bond_release_and_destroy(struct net_device *bond_dev,
1851 struct net_device *slave_dev)
1853 struct bonding *bond = netdev_priv(bond_dev);
1856 ret = bond_release(bond_dev, slave_dev);
1857 if (ret == 0 && !bond_has_slaves(bond)) {
1858 bond_dev->priv_flags |= IFF_DISABLE_NETPOLL;
1859 pr_info("%s: destroying bond %s.\n",
1860 bond_dev->name, bond_dev->name);
1861 unregister_netdevice(bond_dev);
1866 static int bond_info_query(struct net_device *bond_dev, struct ifbond *info)
1868 struct bonding *bond = netdev_priv(bond_dev);
1870 info->bond_mode = bond->params.mode;
1871 info->miimon = bond->params.miimon;
1873 read_lock(&bond->lock);
1874 info->num_slaves = bond->slave_cnt;
1875 read_unlock(&bond->lock);
1880 static int bond_slave_info_query(struct net_device *bond_dev, struct ifslave *info)
1882 struct bonding *bond = netdev_priv(bond_dev);
1883 struct list_head *iter;
1884 int i = 0, res = -ENODEV;
1885 struct slave *slave;
1887 read_lock(&bond->lock);
1888 bond_for_each_slave(bond, slave, iter) {
1889 if (i++ == (int)info->slave_id) {
1891 strcpy(info->slave_name, slave->dev->name);
1892 info->link = slave->link;
1893 info->state = bond_slave_state(slave);
1894 info->link_failure_count = slave->link_failure_count;
1898 read_unlock(&bond->lock);
1903 /*-------------------------------- Monitoring -------------------------------*/
1906 static int bond_miimon_inspect(struct bonding *bond)
1908 int link_state, commit = 0;
1909 struct list_head *iter;
1910 struct slave *slave;
1911 bool ignore_updelay;
1913 ignore_updelay = !bond->curr_active_slave ? true : false;
1915 bond_for_each_slave_rcu(bond, slave, iter) {
1916 slave->new_link = BOND_LINK_NOCHANGE;
1918 link_state = bond_check_dev_link(bond, slave->dev, 0);
1920 switch (slave->link) {
1925 slave->link = BOND_LINK_FAIL;
1926 slave->delay = bond->params.downdelay;
1928 pr_info("%s: link status down for %sinterface %s, disabling it in %d ms.\n",
1930 (bond->params.mode ==
1931 BOND_MODE_ACTIVEBACKUP) ?
1932 (bond_is_active_slave(slave) ?
1933 "active " : "backup ") : "",
1935 bond->params.downdelay * bond->params.miimon);
1938 case BOND_LINK_FAIL:
1941 * recovered before downdelay expired
1943 slave->link = BOND_LINK_UP;
1944 slave->jiffies = jiffies;
1945 pr_info("%s: link status up again after %d ms for interface %s.\n",
1947 (bond->params.downdelay - slave->delay) *
1948 bond->params.miimon,
1953 if (slave->delay <= 0) {
1954 slave->new_link = BOND_LINK_DOWN;
1962 case BOND_LINK_DOWN:
1966 slave->link = BOND_LINK_BACK;
1967 slave->delay = bond->params.updelay;
1970 pr_info("%s: link status up for interface %s, enabling it in %d ms.\n",
1971 bond->dev->name, slave->dev->name,
1972 ignore_updelay ? 0 :
1973 bond->params.updelay *
1974 bond->params.miimon);
1977 case BOND_LINK_BACK:
1979 slave->link = BOND_LINK_DOWN;
1980 pr_info("%s: link status down again after %d ms for interface %s.\n",
1982 (bond->params.updelay - slave->delay) *
1983 bond->params.miimon,
1992 if (slave->delay <= 0) {
1993 slave->new_link = BOND_LINK_UP;
1995 ignore_updelay = false;
2007 static void bond_miimon_commit(struct bonding *bond)
2009 struct list_head *iter;
2010 struct slave *slave;
2012 bond_for_each_slave(bond, slave, iter) {
2013 switch (slave->new_link) {
2014 case BOND_LINK_NOCHANGE:
2018 slave->link = BOND_LINK_UP;
2019 slave->jiffies = jiffies;
2021 if (bond->params.mode == BOND_MODE_8023AD) {
2022 /* prevent it from being the active one */
2023 bond_set_backup_slave(slave);
2024 } else if (bond->params.mode != BOND_MODE_ACTIVEBACKUP) {
2025 /* make it immediately active */
2026 bond_set_active_slave(slave);
2027 } else if (slave != bond->primary_slave) {
2028 /* prevent it from being the active one */
2029 bond_set_backup_slave(slave);
2032 pr_info("%s: link status definitely up for interface %s, %u Mbps %s duplex.\n",
2033 bond->dev->name, slave->dev->name,
2034 slave->speed == SPEED_UNKNOWN ? 0 : slave->speed,
2035 slave->duplex ? "full" : "half");
2037 /* notify ad that the link status has changed */
2038 if (bond->params.mode == BOND_MODE_8023AD)
2039 bond_3ad_handle_link_change(slave, BOND_LINK_UP);
2041 if (bond_is_lb(bond))
2042 bond_alb_handle_link_change(bond, slave,
2045 if (!bond->curr_active_slave ||
2046 (slave == bond->primary_slave))
2051 case BOND_LINK_DOWN:
2052 if (slave->link_failure_count < UINT_MAX)
2053 slave->link_failure_count++;
2055 slave->link = BOND_LINK_DOWN;
2057 if (bond->params.mode == BOND_MODE_ACTIVEBACKUP ||
2058 bond->params.mode == BOND_MODE_8023AD)
2059 bond_set_slave_inactive_flags(slave);
2061 pr_info("%s: link status definitely down for interface %s, disabling it\n",
2062 bond->dev->name, slave->dev->name);
2064 if (bond->params.mode == BOND_MODE_8023AD)
2065 bond_3ad_handle_link_change(slave,
2068 if (bond_is_lb(bond))
2069 bond_alb_handle_link_change(bond, slave,
2072 if (slave == bond->curr_active_slave)
2078 pr_err("%s: invalid new link %d on slave %s\n",
2079 bond->dev->name, slave->new_link,
2081 slave->new_link = BOND_LINK_NOCHANGE;
2089 write_lock_bh(&bond->curr_slave_lock);
2090 bond_select_active_slave(bond);
2091 write_unlock_bh(&bond->curr_slave_lock);
2092 unblock_netpoll_tx();
2095 bond_set_carrier(bond);
2101 * Really a wrapper that splits the mii monitor into two phases: an
2102 * inspection, then (if inspection indicates something needs to be done)
2103 * an acquisition of appropriate locks followed by a commit phase to
2104 * implement whatever link state changes are indicated.
2106 static void bond_mii_monitor(struct work_struct *work)
2108 struct bonding *bond = container_of(work, struct bonding,
2110 bool should_notify_peers = false;
2111 unsigned long delay;
2113 delay = msecs_to_jiffies(bond->params.miimon);
2115 if (!bond_has_slaves(bond))
2120 should_notify_peers = bond_should_notify_peers(bond);
2122 if (bond_miimon_inspect(bond)) {
2125 /* Race avoidance with bond_close cancel of workqueue */
2126 if (!rtnl_trylock()) {
2128 should_notify_peers = false;
2132 bond_miimon_commit(bond);
2134 rtnl_unlock(); /* might sleep, hold no other locks */
2139 if (bond->params.miimon)
2140 queue_delayed_work(bond->wq, &bond->mii_work, delay);
2142 if (should_notify_peers) {
2143 if (!rtnl_trylock())
2145 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, bond->dev);
2150 static bool bond_has_this_ip(struct bonding *bond, __be32 ip)
2152 struct net_device *upper;
2153 struct list_head *iter;
2156 if (ip == bond_confirm_addr(bond->dev, 0, ip))
2160 netdev_for_each_all_upper_dev_rcu(bond->dev, upper, iter) {
2161 if (ip == bond_confirm_addr(upper, 0, ip)) {
2172 * We go to the (large) trouble of VLAN tagging ARP frames because
2173 * switches in VLAN mode (especially if ports are configured as
2174 * "native" to a VLAN) might not pass non-tagged frames.
2176 static void bond_arp_send(struct net_device *slave_dev, int arp_op, __be32 dest_ip, __be32 src_ip, unsigned short vlan_id)
2178 struct sk_buff *skb;
2180 pr_debug("arp %d on slave %s: dst %pI4 src %pI4 vid %d\n", arp_op,
2181 slave_dev->name, &dest_ip, &src_ip, vlan_id);
2183 skb = arp_create(arp_op, ETH_P_ARP, dest_ip, slave_dev, src_ip,
2184 NULL, slave_dev->dev_addr, NULL);
2187 pr_err("ARP packet allocation failed\n");
2191 skb = vlan_put_tag(skb, htons(ETH_P_8021Q), vlan_id);
2193 pr_err("failed to insert VLAN tag\n");
2201 static void bond_arp_send_all(struct bonding *bond, struct slave *slave)
2203 struct net_device *upper, *vlan_upper;
2204 struct list_head *iter, *vlan_iter;
2206 __be32 *targets = bond->params.arp_targets, addr;
2209 for (i = 0; i < BOND_MAX_ARP_TARGETS && targets[i]; i++) {
2210 pr_debug("basa: target %pI4\n", &targets[i]);
2212 /* Find out through which dev should the packet go */
2213 rt = ip_route_output(dev_net(bond->dev), targets[i], 0,
2216 pr_debug("%s: no route to arp_ip_target %pI4\n",
2217 bond->dev->name, &targets[i]);
2223 /* bond device itself */
2224 if (rt->dst.dev == bond->dev)
2228 /* first we search only for vlan devices. for every vlan
2229 * found we verify its upper dev list, searching for the
2230 * rt->dst.dev. If found we save the tag of the vlan and
2231 * proceed to send the packet.
2235 netdev_for_each_all_upper_dev_rcu(bond->dev, vlan_upper,
2237 if (!is_vlan_dev(vlan_upper))
2239 netdev_for_each_all_upper_dev_rcu(vlan_upper, upper,
2241 if (upper == rt->dst.dev) {
2242 vlan_id = vlan_dev_vlan_id(vlan_upper);
2249 /* if the device we're looking for is not on top of any of
2250 * our upper vlans, then just search for any dev that
2251 * matches, and in case it's a vlan - save the id
2253 netdev_for_each_all_upper_dev_rcu(bond->dev, upper, iter) {
2254 if (upper == rt->dst.dev) {
2255 /* if it's a vlan - get its VID */
2256 if (is_vlan_dev(upper))
2257 vlan_id = vlan_dev_vlan_id(upper);
2265 /* Not our device - skip */
2266 pr_debug("%s: no path to arp_ip_target %pI4 via rt.dev %s\n",
2267 bond->dev->name, &targets[i],
2268 rt->dst.dev ? rt->dst.dev->name : "NULL");
2274 addr = bond_confirm_addr(rt->dst.dev, targets[i], 0);
2276 bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2281 static void bond_validate_arp(struct bonding *bond, struct slave *slave, __be32 sip, __be32 tip)
2285 if (!sip || !bond_has_this_ip(bond, tip)) {
2286 pr_debug("bva: sip %pI4 tip %pI4 not found\n", &sip, &tip);
2290 i = bond_get_targets_ip(bond->params.arp_targets, sip);
2292 pr_debug("bva: sip %pI4 not found in targets\n", &sip);
2295 slave->last_arp_rx = jiffies;
2296 slave->target_last_arp_rx[i] = jiffies;
2299 int bond_arp_rcv(const struct sk_buff *skb, struct bonding *bond,
2300 struct slave *slave)
2302 struct arphdr *arp = (struct arphdr *)skb->data;
2303 unsigned char *arp_ptr;
2307 if (skb->protocol != __cpu_to_be16(ETH_P_ARP))
2308 return RX_HANDLER_ANOTHER;
2310 read_lock(&bond->lock);
2312 if (!slave_do_arp_validate(bond, slave))
2315 alen = arp_hdr_len(bond->dev);
2317 pr_debug("bond_arp_rcv: bond %s skb->dev %s\n",
2318 bond->dev->name, skb->dev->name);
2320 if (alen > skb_headlen(skb)) {
2321 arp = kmalloc(alen, GFP_ATOMIC);
2324 if (skb_copy_bits(skb, 0, arp, alen) < 0)
2328 if (arp->ar_hln != bond->dev->addr_len ||
2329 skb->pkt_type == PACKET_OTHERHOST ||
2330 skb->pkt_type == PACKET_LOOPBACK ||
2331 arp->ar_hrd != htons(ARPHRD_ETHER) ||
2332 arp->ar_pro != htons(ETH_P_IP) ||
2336 arp_ptr = (unsigned char *)(arp + 1);
2337 arp_ptr += bond->dev->addr_len;
2338 memcpy(&sip, arp_ptr, 4);
2339 arp_ptr += 4 + bond->dev->addr_len;
2340 memcpy(&tip, arp_ptr, 4);
2342 pr_debug("bond_arp_rcv: %s %s/%d av %d sv %d sip %pI4 tip %pI4\n",
2343 bond->dev->name, slave->dev->name, bond_slave_state(slave),
2344 bond->params.arp_validate, slave_do_arp_validate(bond, slave),
2348 * Backup slaves won't see the ARP reply, but do come through
2349 * here for each ARP probe (so we swap the sip/tip to validate
2350 * the probe). In a "redundant switch, common router" type of
2351 * configuration, the ARP probe will (hopefully) travel from
2352 * the active, through one switch, the router, then the other
2353 * switch before reaching the backup.
2355 * We 'trust' the arp requests if there is an active slave and
2356 * it received valid arp reply(s) after it became active. This
2357 * is done to avoid endless looping when we can't reach the
2358 * arp_ip_target and fool ourselves with our own arp requests.
2360 if (bond_is_active_slave(slave))
2361 bond_validate_arp(bond, slave, sip, tip);
2362 else if (bond->curr_active_slave &&
2363 time_after(slave_last_rx(bond, bond->curr_active_slave),
2364 bond->curr_active_slave->jiffies))
2365 bond_validate_arp(bond, slave, tip, sip);
2368 read_unlock(&bond->lock);
2369 if (arp != (struct arphdr *)skb->data)
2371 return RX_HANDLER_ANOTHER;
2374 /* function to verify if we're in the arp_interval timeslice, returns true if
2375 * (last_act - arp_interval) <= jiffies <= (last_act + mod * arp_interval +
2376 * arp_interval/2) . the arp_interval/2 is needed for really fast networks.
2378 static bool bond_time_in_interval(struct bonding *bond, unsigned long last_act,
2381 int delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
2383 return time_in_range(jiffies,
2384 last_act - delta_in_ticks,
2385 last_act + mod * delta_in_ticks + delta_in_ticks/2);
2389 * this function is called regularly to monitor each slave's link
2390 * ensuring that traffic is being sent and received when arp monitoring
2391 * is used in load-balancing mode. if the adapter has been dormant, then an
2392 * arp is transmitted to generate traffic. see activebackup_arp_monitor for
2393 * arp monitoring in active backup mode.
2395 static void bond_loadbalance_arp_mon(struct work_struct *work)
2397 struct bonding *bond = container_of(work, struct bonding,
2399 struct slave *slave, *oldcurrent;
2400 struct list_head *iter;
2401 int do_failover = 0;
2403 if (!bond_has_slaves(bond))
2408 oldcurrent = ACCESS_ONCE(bond->curr_active_slave);
2409 /* see if any of the previous devices are up now (i.e. they have
2410 * xmt and rcv traffic). the curr_active_slave does not come into
2411 * the picture unless it is null. also, slave->jiffies is not needed
2412 * here because we send an arp on each slave and give a slave as
2413 * long as it needs to get the tx/rx within the delta.
2414 * TODO: what about up/down delay in arp mode? it wasn't here before
2417 bond_for_each_slave_rcu(bond, slave, iter) {
2418 unsigned long trans_start = dev_trans_start(slave->dev);
2420 if (slave->link != BOND_LINK_UP) {
2421 if (bond_time_in_interval(bond, trans_start, 1) &&
2422 bond_time_in_interval(bond, slave->dev->last_rx, 1)) {
2424 slave->link = BOND_LINK_UP;
2425 bond_set_active_slave(slave);
2427 /* primary_slave has no meaning in round-robin
2428 * mode. the window of a slave being up and
2429 * curr_active_slave being null after enslaving
2433 pr_info("%s: link status definitely up for interface %s, ",
2438 pr_info("%s: interface %s is now up\n",
2444 /* slave->link == BOND_LINK_UP */
2446 /* not all switches will respond to an arp request
2447 * when the source ip is 0, so don't take the link down
2448 * if we don't know our ip yet
2450 if (!bond_time_in_interval(bond, trans_start, 2) ||
2451 !bond_time_in_interval(bond, slave->dev->last_rx, 2)) {
2453 slave->link = BOND_LINK_DOWN;
2454 bond_set_backup_slave(slave);
2456 if (slave->link_failure_count < UINT_MAX)
2457 slave->link_failure_count++;
2459 pr_info("%s: interface %s is now down.\n",
2463 if (slave == oldcurrent)
2468 /* note: if switch is in round-robin mode, all links
2469 * must tx arp to ensure all links rx an arp - otherwise
2470 * links may oscillate or not come up at all; if switch is
2471 * in something like xor mode, there is nothing we can
2472 * do - all replies will be rx'ed on same link causing slaves
2473 * to be unstable during low/no traffic periods
2475 if (IS_UP(slave->dev))
2476 bond_arp_send_all(bond, slave);
2482 /* the bond_select_active_slave must hold RTNL
2483 * and curr_slave_lock for write.
2485 if (!rtnl_trylock())
2488 write_lock_bh(&bond->curr_slave_lock);
2490 bond_select_active_slave(bond);
2492 write_unlock_bh(&bond->curr_slave_lock);
2493 unblock_netpoll_tx();
2498 if (bond->params.arp_interval)
2499 queue_delayed_work(bond->wq, &bond->arp_work,
2500 msecs_to_jiffies(bond->params.arp_interval));
2504 * Called to inspect slaves for active-backup mode ARP monitor link state
2505 * changes. Sets new_link in slaves to specify what action should take
2506 * place for the slave. Returns 0 if no changes are found, >0 if changes
2507 * to link states must be committed.
2509 * Called with rcu_read_lock hold.
2511 static int bond_ab_arp_inspect(struct bonding *bond)
2513 unsigned long trans_start, last_rx;
2514 struct list_head *iter;
2515 struct slave *slave;
2518 bond_for_each_slave_rcu(bond, slave, iter) {
2519 slave->new_link = BOND_LINK_NOCHANGE;
2520 last_rx = slave_last_rx(bond, slave);
2522 if (slave->link != BOND_LINK_UP) {
2523 if (bond_time_in_interval(bond, last_rx, 1)) {
2524 slave->new_link = BOND_LINK_UP;
2531 * Give slaves 2*delta after being enslaved or made
2532 * active. This avoids bouncing, as the last receive
2533 * times need a full ARP monitor cycle to be updated.
2535 if (bond_time_in_interval(bond, slave->jiffies, 2))
2539 * Backup slave is down if:
2540 * - No current_arp_slave AND
2541 * - more than 3*delta since last receive AND
2542 * - the bond has an IP address
2544 * Note: a non-null current_arp_slave indicates
2545 * the curr_active_slave went down and we are
2546 * searching for a new one; under this condition
2547 * we only take the curr_active_slave down - this
2548 * gives each slave a chance to tx/rx traffic
2549 * before being taken out
2551 if (!bond_is_active_slave(slave) &&
2552 !bond->current_arp_slave &&
2553 !bond_time_in_interval(bond, last_rx, 3)) {
2554 slave->new_link = BOND_LINK_DOWN;
2559 * Active slave is down if:
2560 * - more than 2*delta since transmitting OR
2561 * - (more than 2*delta since receive AND
2562 * the bond has an IP address)
2564 trans_start = dev_trans_start(slave->dev);
2565 if (bond_is_active_slave(slave) &&
2566 (!bond_time_in_interval(bond, trans_start, 2) ||
2567 !bond_time_in_interval(bond, last_rx, 2))) {
2568 slave->new_link = BOND_LINK_DOWN;
2577 * Called to commit link state changes noted by inspection step of
2578 * active-backup mode ARP monitor.
2580 * Called with RTNL hold.
2582 static void bond_ab_arp_commit(struct bonding *bond)
2584 unsigned long trans_start;
2585 struct list_head *iter;
2586 struct slave *slave;
2588 bond_for_each_slave(bond, slave, iter) {
2589 switch (slave->new_link) {
2590 case BOND_LINK_NOCHANGE:
2594 trans_start = dev_trans_start(slave->dev);
2595 if (bond->curr_active_slave != slave ||
2596 (!bond->curr_active_slave &&
2597 bond_time_in_interval(bond, trans_start, 1))) {
2598 slave->link = BOND_LINK_UP;
2599 if (bond->current_arp_slave) {
2600 bond_set_slave_inactive_flags(
2601 bond->current_arp_slave);
2602 bond->current_arp_slave = NULL;
2605 pr_info("%s: link status definitely up for interface %s.\n",
2606 bond->dev->name, slave->dev->name);
2608 if (!bond->curr_active_slave ||
2609 (slave == bond->primary_slave))
2616 case BOND_LINK_DOWN:
2617 if (slave->link_failure_count < UINT_MAX)
2618 slave->link_failure_count++;
2620 slave->link = BOND_LINK_DOWN;
2621 bond_set_slave_inactive_flags(slave);
2623 pr_info("%s: link status definitely down for interface %s, disabling it\n",
2624 bond->dev->name, slave->dev->name);
2626 if (slave == bond->curr_active_slave) {
2627 bond->current_arp_slave = NULL;
2634 pr_err("%s: impossible: new_link %d on slave %s\n",
2635 bond->dev->name, slave->new_link,
2643 write_lock_bh(&bond->curr_slave_lock);
2644 bond_select_active_slave(bond);
2645 write_unlock_bh(&bond->curr_slave_lock);
2646 unblock_netpoll_tx();
2649 bond_set_carrier(bond);
2653 * Send ARP probes for active-backup mode ARP monitor.
2655 * Called with rcu_read_lock hold.
2657 static void bond_ab_arp_probe(struct bonding *bond)
2659 struct slave *slave, *before = NULL, *new_slave = NULL,
2660 *curr_arp_slave = rcu_dereference(bond->current_arp_slave);
2661 struct list_head *iter;
2664 read_lock(&bond->curr_slave_lock);
2666 if (curr_arp_slave && bond->curr_active_slave)
2667 pr_info("PROBE: c_arp %s && cas %s BAD\n",
2668 curr_arp_slave->dev->name,
2669 bond->curr_active_slave->dev->name);
2671 if (bond->curr_active_slave) {
2672 bond_arp_send_all(bond, bond->curr_active_slave);
2673 read_unlock(&bond->curr_slave_lock);
2677 read_unlock(&bond->curr_slave_lock);
2679 /* if we don't have a curr_active_slave, search for the next available
2680 * backup slave from the current_arp_slave and make it the candidate
2681 * for becoming the curr_active_slave
2684 if (!curr_arp_slave) {
2685 curr_arp_slave = bond_first_slave_rcu(bond);
2686 if (!curr_arp_slave)
2690 bond_set_slave_inactive_flags(curr_arp_slave);
2692 bond_for_each_slave_rcu(bond, slave, iter) {
2693 if (!found && !before && IS_UP(slave->dev))
2696 if (found && !new_slave && IS_UP(slave->dev))
2698 /* if the link state is up at this point, we
2699 * mark it down - this can happen if we have
2700 * simultaneous link failures and
2701 * reselect_active_interface doesn't make this
2702 * one the current slave so it is still marked
2703 * up when it is actually down
2705 if (!IS_UP(slave->dev) && slave->link == BOND_LINK_UP) {
2706 slave->link = BOND_LINK_DOWN;
2707 if (slave->link_failure_count < UINT_MAX)
2708 slave->link_failure_count++;
2710 bond_set_slave_inactive_flags(slave);
2712 pr_info("%s: backup interface %s is now down.\n",
2713 bond->dev->name, slave->dev->name);
2715 if (slave == curr_arp_slave)
2719 if (!new_slave && before)
2725 new_slave->link = BOND_LINK_BACK;
2726 bond_set_slave_active_flags(new_slave);
2727 bond_arp_send_all(bond, new_slave);
2728 new_slave->jiffies = jiffies;
2729 rcu_assign_pointer(bond->current_arp_slave, new_slave);
2732 static void bond_activebackup_arp_mon(struct work_struct *work)
2734 struct bonding *bond = container_of(work, struct bonding,
2736 bool should_notify_peers = false;
2739 delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
2741 if (!bond_has_slaves(bond))
2746 should_notify_peers = bond_should_notify_peers(bond);
2748 if (bond_ab_arp_inspect(bond)) {
2751 /* Race avoidance with bond_close flush of workqueue */
2752 if (!rtnl_trylock()) {
2754 should_notify_peers = false;
2758 bond_ab_arp_commit(bond);
2764 bond_ab_arp_probe(bond);
2768 if (bond->params.arp_interval)
2769 queue_delayed_work(bond->wq, &bond->arp_work, delta_in_ticks);
2771 if (should_notify_peers) {
2772 if (!rtnl_trylock())
2774 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, bond->dev);
2779 /*-------------------------- netdev event handling --------------------------*/
2782 * Change device name
2784 static int bond_event_changename(struct bonding *bond)
2786 bond_remove_proc_entry(bond);
2787 bond_create_proc_entry(bond);
2789 bond_debug_reregister(bond);
2794 static int bond_master_netdev_event(unsigned long event,
2795 struct net_device *bond_dev)
2797 struct bonding *event_bond = netdev_priv(bond_dev);
2800 case NETDEV_CHANGENAME:
2801 return bond_event_changename(event_bond);
2802 case NETDEV_UNREGISTER:
2803 bond_remove_proc_entry(event_bond);
2805 case NETDEV_REGISTER:
2806 bond_create_proc_entry(event_bond);
2808 case NETDEV_NOTIFY_PEERS:
2809 if (event_bond->send_peer_notif)
2810 event_bond->send_peer_notif--;
2819 static int bond_slave_netdev_event(unsigned long event,
2820 struct net_device *slave_dev)
2822 struct slave *slave = bond_slave_get_rtnl(slave_dev);
2823 struct bonding *bond;
2824 struct net_device *bond_dev;
2828 /* A netdev event can be generated while enslaving a device
2829 * before netdev_rx_handler_register is called in which case
2830 * slave will be NULL
2834 bond_dev = slave->bond->dev;
2838 case NETDEV_UNREGISTER:
2839 if (bond_dev->type != ARPHRD_ETHER)
2840 bond_release_and_destroy(bond_dev, slave_dev);
2842 bond_release(bond_dev, slave_dev);
2846 old_speed = slave->speed;
2847 old_duplex = slave->duplex;
2849 bond_update_speed_duplex(slave);
2851 if (bond->params.mode == BOND_MODE_8023AD) {
2852 if (old_speed != slave->speed)
2853 bond_3ad_adapter_speed_changed(slave);
2854 if (old_duplex != slave->duplex)
2855 bond_3ad_adapter_duplex_changed(slave);
2860 * ... Or is it this?
2863 case NETDEV_CHANGEMTU:
2865 * TODO: Should slaves be allowed to
2866 * independently alter their MTU? For
2867 * an active-backup bond, slaves need
2868 * not be the same type of device, so
2869 * MTUs may vary. For other modes,
2870 * slaves arguably should have the
2871 * same MTUs. To do this, we'd need to
2872 * take over the slave's change_mtu
2873 * function for the duration of their
2877 case NETDEV_CHANGENAME:
2878 /* we don't care if we don't have primary set */
2879 if (!USES_PRIMARY(bond->params.mode) ||
2880 !bond->params.primary[0])
2883 if (slave == bond->primary_slave) {
2884 /* slave's name changed - he's no longer primary */
2885 bond->primary_slave = NULL;
2886 } else if (!strcmp(slave_dev->name, bond->params.primary)) {
2887 /* we have a new primary slave */
2888 bond->primary_slave = slave;
2889 } else { /* we didn't change primary - exit */
2893 pr_info("%s: Primary slave changed to %s, reselecting active slave.\n",
2894 bond->dev->name, bond->primary_slave ? slave_dev->name :
2896 write_lock_bh(&bond->curr_slave_lock);
2897 bond_select_active_slave(bond);
2898 write_unlock_bh(&bond->curr_slave_lock);
2900 case NETDEV_FEAT_CHANGE:
2901 bond_compute_features(bond);
2903 case NETDEV_RESEND_IGMP:
2904 /* Propagate to master device */
2905 call_netdevice_notifiers(event, slave->bond->dev);
2915 * bond_netdev_event: handle netdev notifier chain events.
2917 * This function receives events for the netdev chain. The caller (an
2918 * ioctl handler calling blocking_notifier_call_chain) holds the necessary
2919 * locks for us to safely manipulate the slave devices (RTNL lock,
2922 static int bond_netdev_event(struct notifier_block *this,
2923 unsigned long event, void *ptr)
2925 struct net_device *event_dev = netdev_notifier_info_to_dev(ptr);
2927 pr_debug("event_dev: %s, event: %lx\n",
2928 event_dev ? event_dev->name : "None",
2931 if (!(event_dev->priv_flags & IFF_BONDING))
2934 if (event_dev->flags & IFF_MASTER) {
2935 pr_debug("IFF_MASTER\n");
2936 return bond_master_netdev_event(event, event_dev);
2939 if (event_dev->flags & IFF_SLAVE) {
2940 pr_debug("IFF_SLAVE\n");
2941 return bond_slave_netdev_event(event, event_dev);
2947 static struct notifier_block bond_netdev_notifier = {
2948 .notifier_call = bond_netdev_event,
2951 /*---------------------------- Hashing Policies -----------------------------*/
2953 /* L2 hash helper */
2954 static inline u32 bond_eth_hash(struct sk_buff *skb)
2956 struct ethhdr *data = (struct ethhdr *)skb->data;
2958 if (skb_headlen(skb) >= offsetof(struct ethhdr, h_proto))
2959 return data->h_dest[5] ^ data->h_source[5];
2964 /* Extract the appropriate headers based on bond's xmit policy */
2965 static bool bond_flow_dissect(struct bonding *bond, struct sk_buff *skb,
2966 struct flow_keys *fk)
2968 const struct ipv6hdr *iph6;
2969 const struct iphdr *iph;
2970 int noff, proto = -1;
2972 if (bond->params.xmit_policy > BOND_XMIT_POLICY_LAYER23)
2973 return skb_flow_dissect(skb, fk);
2976 noff = skb_network_offset(skb);
2977 if (skb->protocol == htons(ETH_P_IP)) {
2978 if (!pskb_may_pull(skb, noff + sizeof(*iph)))
2981 fk->src = iph->saddr;
2982 fk->dst = iph->daddr;
2983 noff += iph->ihl << 2;
2984 if (!ip_is_fragment(iph))
2985 proto = iph->protocol;
2986 } else if (skb->protocol == htons(ETH_P_IPV6)) {
2987 if (!pskb_may_pull(skb, noff + sizeof(*iph6)))
2989 iph6 = ipv6_hdr(skb);
2990 fk->src = (__force __be32)ipv6_addr_hash(&iph6->saddr);
2991 fk->dst = (__force __be32)ipv6_addr_hash(&iph6->daddr);
2992 noff += sizeof(*iph6);
2993 proto = iph6->nexthdr;
2997 if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER34 && proto >= 0)
2998 fk->ports = skb_flow_get_ports(skb, noff, proto);
3004 * bond_xmit_hash - generate a hash value based on the xmit policy
3005 * @bond: bonding device
3006 * @skb: buffer to use for headers
3007 * @count: modulo value
3009 * This function will extract the necessary headers from the skb buffer and use
3010 * them to generate a hash based on the xmit_policy set in the bonding device
3011 * which will be reduced modulo count before returning.
3013 int bond_xmit_hash(struct bonding *bond, struct sk_buff *skb, int count)
3015 struct flow_keys flow;
3018 if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER2 ||
3019 !bond_flow_dissect(bond, skb, &flow))
3020 return bond_eth_hash(skb) % count;
3022 if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER23 ||
3023 bond->params.xmit_policy == BOND_XMIT_POLICY_ENCAP23)
3024 hash = bond_eth_hash(skb);
3026 hash = (__force u32)flow.ports;
3027 hash ^= (__force u32)flow.dst ^ (__force u32)flow.src;
3028 hash ^= (hash >> 16);
3029 hash ^= (hash >> 8);
3031 return hash % count;
3034 /*-------------------------- Device entry points ----------------------------*/
3036 static void bond_work_init_all(struct bonding *bond)
3038 INIT_DELAYED_WORK(&bond->mcast_work,
3039 bond_resend_igmp_join_requests_delayed);
3040 INIT_DELAYED_WORK(&bond->alb_work, bond_alb_monitor);
3041 INIT_DELAYED_WORK(&bond->mii_work, bond_mii_monitor);
3042 if (bond->params.mode == BOND_MODE_ACTIVEBACKUP)
3043 INIT_DELAYED_WORK(&bond->arp_work, bond_activebackup_arp_mon);
3045 INIT_DELAYED_WORK(&bond->arp_work, bond_loadbalance_arp_mon);
3046 INIT_DELAYED_WORK(&bond->ad_work, bond_3ad_state_machine_handler);
3049 static void bond_work_cancel_all(struct bonding *bond)
3051 cancel_delayed_work_sync(&bond->mii_work);
3052 cancel_delayed_work_sync(&bond->arp_work);
3053 cancel_delayed_work_sync(&bond->alb_work);
3054 cancel_delayed_work_sync(&bond->ad_work);
3055 cancel_delayed_work_sync(&bond->mcast_work);
3058 static int bond_open(struct net_device *bond_dev)
3060 struct bonding *bond = netdev_priv(bond_dev);
3061 struct list_head *iter;
3062 struct slave *slave;
3064 /* reset slave->backup and slave->inactive */
3065 read_lock(&bond->lock);
3066 if (bond_has_slaves(bond)) {
3067 read_lock(&bond->curr_slave_lock);
3068 bond_for_each_slave(bond, slave, iter) {
3069 if ((bond->params.mode == BOND_MODE_ACTIVEBACKUP)
3070 && (slave != bond->curr_active_slave)) {
3071 bond_set_slave_inactive_flags(slave);
3073 bond_set_slave_active_flags(slave);
3076 read_unlock(&bond->curr_slave_lock);
3078 read_unlock(&bond->lock);
3080 bond_work_init_all(bond);
3082 if (bond_is_lb(bond)) {
3083 /* bond_alb_initialize must be called before the timer
3086 if (bond_alb_initialize(bond, (bond->params.mode == BOND_MODE_ALB)))
3088 queue_delayed_work(bond->wq, &bond->alb_work, 0);
3091 if (bond->params.miimon) /* link check interval, in milliseconds. */
3092 queue_delayed_work(bond->wq, &bond->mii_work, 0);
3094 if (bond->params.arp_interval) { /* arp interval, in milliseconds. */
3095 queue_delayed_work(bond->wq, &bond->arp_work, 0);
3096 if (bond->params.arp_validate)
3097 bond->recv_probe = bond_arp_rcv;
3100 if (bond->params.mode == BOND_MODE_8023AD) {
3101 queue_delayed_work(bond->wq, &bond->ad_work, 0);
3102 /* register to receive LACPDUs */
3103 bond->recv_probe = bond_3ad_lacpdu_recv;
3104 bond_3ad_initiate_agg_selection(bond, 1);
3110 static int bond_close(struct net_device *bond_dev)
3112 struct bonding *bond = netdev_priv(bond_dev);
3114 bond_work_cancel_all(bond);
3115 bond->send_peer_notif = 0;
3116 if (bond_is_lb(bond))
3117 bond_alb_deinitialize(bond);
3118 bond->recv_probe = NULL;
3123 static struct rtnl_link_stats64 *bond_get_stats(struct net_device *bond_dev,
3124 struct rtnl_link_stats64 *stats)
3126 struct bonding *bond = netdev_priv(bond_dev);
3127 struct rtnl_link_stats64 temp;
3128 struct list_head *iter;
3129 struct slave *slave;
3131 memset(stats, 0, sizeof(*stats));
3133 read_lock_bh(&bond->lock);
3134 bond_for_each_slave(bond, slave, iter) {
3135 const struct rtnl_link_stats64 *sstats =
3136 dev_get_stats(slave->dev, &temp);
3138 stats->rx_packets += sstats->rx_packets;
3139 stats->rx_bytes += sstats->rx_bytes;
3140 stats->rx_errors += sstats->rx_errors;
3141 stats->rx_dropped += sstats->rx_dropped;
3143 stats->tx_packets += sstats->tx_packets;
3144 stats->tx_bytes += sstats->tx_bytes;
3145 stats->tx_errors += sstats->tx_errors;
3146 stats->tx_dropped += sstats->tx_dropped;
3148 stats->multicast += sstats->multicast;
3149 stats->collisions += sstats->collisions;
3151 stats->rx_length_errors += sstats->rx_length_errors;
3152 stats->rx_over_errors += sstats->rx_over_errors;
3153 stats->rx_crc_errors += sstats->rx_crc_errors;
3154 stats->rx_frame_errors += sstats->rx_frame_errors;
3155 stats->rx_fifo_errors += sstats->rx_fifo_errors;
3156 stats->rx_missed_errors += sstats->rx_missed_errors;
3158 stats->tx_aborted_errors += sstats->tx_aborted_errors;
3159 stats->tx_carrier_errors += sstats->tx_carrier_errors;
3160 stats->tx_fifo_errors += sstats->tx_fifo_errors;
3161 stats->tx_heartbeat_errors += sstats->tx_heartbeat_errors;
3162 stats->tx_window_errors += sstats->tx_window_errors;
3164 read_unlock_bh(&bond->lock);
3169 static int bond_do_ioctl(struct net_device *bond_dev, struct ifreq *ifr, int cmd)
3171 struct bonding *bond = netdev_priv(bond_dev);
3172 struct net_device *slave_dev = NULL;
3173 struct ifbond k_binfo;
3174 struct ifbond __user *u_binfo = NULL;
3175 struct ifslave k_sinfo;
3176 struct ifslave __user *u_sinfo = NULL;
3177 struct mii_ioctl_data *mii = NULL;
3181 pr_debug("bond_ioctl: master=%s, cmd=%d\n", bond_dev->name, cmd);
3193 * We do this again just in case we were called by SIOCGMIIREG
3194 * instead of SIOCGMIIPHY.
3201 if (mii->reg_num == 1) {
3203 read_lock(&bond->lock);
3204 read_lock(&bond->curr_slave_lock);
3205 if (netif_carrier_ok(bond->dev))
3206 mii->val_out = BMSR_LSTATUS;
3208 read_unlock(&bond->curr_slave_lock);
3209 read_unlock(&bond->lock);
3213 case BOND_INFO_QUERY_OLD:
3214 case SIOCBONDINFOQUERY:
3215 u_binfo = (struct ifbond __user *)ifr->ifr_data;
3217 if (copy_from_user(&k_binfo, u_binfo, sizeof(ifbond)))
3220 res = bond_info_query(bond_dev, &k_binfo);
3222 copy_to_user(u_binfo, &k_binfo, sizeof(ifbond)))
3226 case BOND_SLAVE_INFO_QUERY_OLD:
3227 case SIOCBONDSLAVEINFOQUERY:
3228 u_sinfo = (struct ifslave __user *)ifr->ifr_data;
3230 if (copy_from_user(&k_sinfo, u_sinfo, sizeof(ifslave)))
3233 res = bond_slave_info_query(bond_dev, &k_sinfo);
3235 copy_to_user(u_sinfo, &k_sinfo, sizeof(ifslave)))
3244 net = dev_net(bond_dev);
3246 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
3249 slave_dev = __dev_get_by_name(net, ifr->ifr_slave);
3251 pr_debug("slave_dev=%p:\n", slave_dev);
3256 pr_debug("slave_dev->name=%s:\n", slave_dev->name);
3258 case BOND_ENSLAVE_OLD:
3259 case SIOCBONDENSLAVE:
3260 res = bond_enslave(bond_dev, slave_dev);
3262 case BOND_RELEASE_OLD:
3263 case SIOCBONDRELEASE:
3264 res = bond_release(bond_dev, slave_dev);
3266 case BOND_SETHWADDR_OLD:
3267 case SIOCBONDSETHWADDR:
3268 bond_set_dev_addr(bond_dev, slave_dev);
3271 case BOND_CHANGE_ACTIVE_OLD:
3272 case SIOCBONDCHANGEACTIVE:
3273 res = bond_option_active_slave_set(bond, slave_dev);
3282 static void bond_change_rx_flags(struct net_device *bond_dev, int change)
3284 struct bonding *bond = netdev_priv(bond_dev);
3286 if (change & IFF_PROMISC)
3287 bond_set_promiscuity(bond,
3288 bond_dev->flags & IFF_PROMISC ? 1 : -1);
3290 if (change & IFF_ALLMULTI)
3291 bond_set_allmulti(bond,
3292 bond_dev->flags & IFF_ALLMULTI ? 1 : -1);
3295 static void bond_set_rx_mode(struct net_device *bond_dev)
3297 struct bonding *bond = netdev_priv(bond_dev);
3298 struct list_head *iter;
3299 struct slave *slave;
3303 if (USES_PRIMARY(bond->params.mode)) {
3304 slave = rcu_dereference(bond->curr_active_slave);
3306 dev_uc_sync(slave->dev, bond_dev);
3307 dev_mc_sync(slave->dev, bond_dev);
3310 bond_for_each_slave_rcu(bond, slave, iter) {
3311 dev_uc_sync_multiple(slave->dev, bond_dev);
3312 dev_mc_sync_multiple(slave->dev, bond_dev);
3318 static int bond_neigh_init(struct neighbour *n)
3320 struct bonding *bond = netdev_priv(n->dev);
3321 const struct net_device_ops *slave_ops;
3322 struct neigh_parms parms;
3323 struct slave *slave;
3326 slave = bond_first_slave(bond);
3329 slave_ops = slave->dev->netdev_ops;
3330 if (!slave_ops->ndo_neigh_setup)
3333 parms.neigh_setup = NULL;
3334 parms.neigh_cleanup = NULL;
3335 ret = slave_ops->ndo_neigh_setup(slave->dev, &parms);
3340 * Assign slave's neigh_cleanup to neighbour in case cleanup is called
3341 * after the last slave has been detached. Assumes that all slaves
3342 * utilize the same neigh_cleanup (true at this writing as only user
3345 n->parms->neigh_cleanup = parms.neigh_cleanup;
3347 if (!parms.neigh_setup)
3350 return parms.neigh_setup(n);
3354 * The bonding ndo_neigh_setup is called at init time beofre any
3355 * slave exists. So we must declare proxy setup function which will
3356 * be used at run time to resolve the actual slave neigh param setup.
3358 * It's also called by master devices (such as vlans) to setup their
3359 * underlying devices. In that case - do nothing, we're already set up from
3362 static int bond_neigh_setup(struct net_device *dev,
3363 struct neigh_parms *parms)
3365 /* modify only our neigh_parms */
3366 if (parms->dev == dev)
3367 parms->neigh_setup = bond_neigh_init;
3373 * Change the MTU of all of a master's slaves to match the master
3375 static int bond_change_mtu(struct net_device *bond_dev, int new_mtu)
3377 struct bonding *bond = netdev_priv(bond_dev);
3378 struct slave *slave, *rollback_slave;
3379 struct list_head *iter;
3382 pr_debug("bond=%p, name=%s, new_mtu=%d\n", bond,
3383 (bond_dev ? bond_dev->name : "None"), new_mtu);
3385 /* Can't hold bond->lock with bh disabled here since
3386 * some base drivers panic. On the other hand we can't
3387 * hold bond->lock without bh disabled because we'll
3388 * deadlock. The only solution is to rely on the fact
3389 * that we're under rtnl_lock here, and the slaves
3390 * list won't change. This doesn't solve the problem
3391 * of setting the slave's MTU while it is
3392 * transmitting, but the assumption is that the base
3393 * driver can handle that.
3395 * TODO: figure out a way to safely iterate the slaves
3396 * list, but without holding a lock around the actual
3397 * call to the base driver.
3400 bond_for_each_slave(bond, slave, iter) {
3401 pr_debug("s %p c_m %p\n",
3403 slave->dev->netdev_ops->ndo_change_mtu);
3405 res = dev_set_mtu(slave->dev, new_mtu);
3408 /* If we failed to set the slave's mtu to the new value
3409 * we must abort the operation even in ACTIVE_BACKUP
3410 * mode, because if we allow the backup slaves to have
3411 * different mtu values than the active slave we'll
3412 * need to change their mtu when doing a failover. That
3413 * means changing their mtu from timer context, which
3414 * is probably not a good idea.
3416 pr_debug("err %d %s\n", res, slave->dev->name);
3421 bond_dev->mtu = new_mtu;
3426 /* unwind from head to the slave that failed */
3427 bond_for_each_slave(bond, rollback_slave, iter) {
3430 if (rollback_slave == slave)
3433 tmp_res = dev_set_mtu(rollback_slave->dev, bond_dev->mtu);
3435 pr_debug("unwind err %d dev %s\n",
3436 tmp_res, rollback_slave->dev->name);
3446 * Note that many devices must be down to change the HW address, and
3447 * downing the master releases all slaves. We can make bonds full of
3448 * bonding devices to test this, however.
3450 static int bond_set_mac_address(struct net_device *bond_dev, void *addr)
3452 struct bonding *bond = netdev_priv(bond_dev);
3453 struct slave *slave, *rollback_slave;
3454 struct sockaddr *sa = addr, tmp_sa;
3455 struct list_head *iter;
3458 if (bond->params.mode == BOND_MODE_ALB)
3459 return bond_alb_set_mac_address(bond_dev, addr);
3462 pr_debug("bond=%p, name=%s\n",
3463 bond, bond_dev ? bond_dev->name : "None");
3465 /* If fail_over_mac is enabled, do nothing and return success.
3466 * Returning an error causes ifenslave to fail.
3468 if (bond->params.fail_over_mac)
3471 if (!is_valid_ether_addr(sa->sa_data))
3472 return -EADDRNOTAVAIL;
3474 /* Can't hold bond->lock with bh disabled here since
3475 * some base drivers panic. On the other hand we can't
3476 * hold bond->lock without bh disabled because we'll
3477 * deadlock. The only solution is to rely on the fact
3478 * that we're under rtnl_lock here, and the slaves
3479 * list won't change. This doesn't solve the problem
3480 * of setting the slave's hw address while it is
3481 * transmitting, but the assumption is that the base
3482 * driver can handle that.
3484 * TODO: figure out a way to safely iterate the slaves
3485 * list, but without holding a lock around the actual
3486 * call to the base driver.
3489 bond_for_each_slave(bond, slave, iter) {
3490 const struct net_device_ops *slave_ops = slave->dev->netdev_ops;
3491 pr_debug("slave %p %s\n", slave, slave->dev->name);
3493 if (slave_ops->ndo_set_mac_address == NULL) {
3495 pr_debug("EOPNOTSUPP %s\n", slave->dev->name);
3499 res = dev_set_mac_address(slave->dev, addr);
3501 /* TODO: consider downing the slave
3503 * User should expect communications
3504 * breakage anyway until ARP finish
3507 pr_debug("err %d %s\n", res, slave->dev->name);
3513 memcpy(bond_dev->dev_addr, sa->sa_data, bond_dev->addr_len);
3517 memcpy(tmp_sa.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
3518 tmp_sa.sa_family = bond_dev->type;
3520 /* unwind from head to the slave that failed */
3521 bond_for_each_slave(bond, rollback_slave, iter) {
3524 if (rollback_slave == slave)
3527 tmp_res = dev_set_mac_address(rollback_slave->dev, &tmp_sa);
3529 pr_debug("unwind err %d dev %s\n",
3530 tmp_res, rollback_slave->dev->name);
3538 * bond_xmit_slave_id - transmit skb through slave with slave_id
3539 * @bond: bonding device that is transmitting
3540 * @skb: buffer to transmit
3541 * @slave_id: slave id up to slave_cnt-1 through which to transmit
3543 * This function tries to transmit through slave with slave_id but in case
3544 * it fails, it tries to find the first available slave for transmission.
3545 * The skb is consumed in all cases, thus the function is void.
3547 static void bond_xmit_slave_id(struct bonding *bond, struct sk_buff *skb, int slave_id)
3549 struct list_head *iter;
3550 struct slave *slave;
3553 /* Here we start from the slave with slave_id */
3554 bond_for_each_slave_rcu(bond, slave, iter) {
3556 if (slave_can_tx(slave)) {
3557 bond_dev_queue_xmit(bond, skb, slave->dev);
3563 /* Here we start from the first slave up to slave_id */
3565 bond_for_each_slave_rcu(bond, slave, iter) {
3568 if (slave_can_tx(slave)) {
3569 bond_dev_queue_xmit(bond, skb, slave->dev);
3573 /* no slave that can tx has been found */
3578 * bond_rr_gen_slave_id - generate slave id based on packets_per_slave
3579 * @bond: bonding device to use
3581 * Based on the value of the bonding device's packets_per_slave parameter
3582 * this function generates a slave id, which is usually used as the next
3583 * slave to transmit through.
3585 static u32 bond_rr_gen_slave_id(struct bonding *bond)
3588 struct reciprocal_value reciprocal_packets_per_slave;
3589 int packets_per_slave = bond->params.packets_per_slave;
3591 switch (packets_per_slave) {
3593 slave_id = prandom_u32();
3596 slave_id = bond->rr_tx_counter;
3599 reciprocal_packets_per_slave =
3600 bond->params.reciprocal_packets_per_slave;
3601 slave_id = reciprocal_divide(bond->rr_tx_counter,
3602 reciprocal_packets_per_slave);
3605 bond->rr_tx_counter++;
3610 static int bond_xmit_roundrobin(struct sk_buff *skb, struct net_device *bond_dev)
3612 struct bonding *bond = netdev_priv(bond_dev);
3613 struct iphdr *iph = ip_hdr(skb);
3614 struct slave *slave;
3617 /* Start with the curr_active_slave that joined the bond as the
3618 * default for sending IGMP traffic. For failover purposes one
3619 * needs to maintain some consistency for the interface that will
3620 * send the join/membership reports. The curr_active_slave found
3621 * will send all of this type of traffic.
3623 if (iph->protocol == IPPROTO_IGMP && skb->protocol == htons(ETH_P_IP)) {
3624 slave = rcu_dereference(bond->curr_active_slave);
3625 if (slave && slave_can_tx(slave))
3626 bond_dev_queue_xmit(bond, skb, slave->dev);
3628 bond_xmit_slave_id(bond, skb, 0);
3630 slave_id = bond_rr_gen_slave_id(bond);
3631 bond_xmit_slave_id(bond, skb, slave_id % bond->slave_cnt);
3634 return NETDEV_TX_OK;
3638 * in active-backup mode, we know that bond->curr_active_slave is always valid if
3639 * the bond has a usable interface.
3641 static int bond_xmit_activebackup(struct sk_buff *skb, struct net_device *bond_dev)
3643 struct bonding *bond = netdev_priv(bond_dev);
3644 struct slave *slave;
3646 slave = rcu_dereference(bond->curr_active_slave);
3648 bond_dev_queue_xmit(bond, skb, slave->dev);
3652 return NETDEV_TX_OK;
3655 /* In bond_xmit_xor() , we determine the output device by using a pre-
3656 * determined xmit_hash_policy(), If the selected device is not enabled,
3657 * find the next active slave.
3659 static int bond_xmit_xor(struct sk_buff *skb, struct net_device *bond_dev)
3661 struct bonding *bond = netdev_priv(bond_dev);
3663 bond_xmit_slave_id(bond, skb, bond_xmit_hash(bond, skb, bond->slave_cnt));
3665 return NETDEV_TX_OK;
3668 /* in broadcast mode, we send everything to all usable interfaces. */
3669 static int bond_xmit_broadcast(struct sk_buff *skb, struct net_device *bond_dev)
3671 struct bonding *bond = netdev_priv(bond_dev);
3672 struct slave *slave = NULL;
3673 struct list_head *iter;
3675 bond_for_each_slave_rcu(bond, slave, iter) {
3676 if (bond_is_last_slave(bond, slave))
3678 if (IS_UP(slave->dev) && slave->link == BOND_LINK_UP) {
3679 struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
3682 pr_err("%s: Error: bond_xmit_broadcast(): skb_clone() failed\n",
3686 /* bond_dev_queue_xmit always returns 0 */
3687 bond_dev_queue_xmit(bond, skb2, slave->dev);
3690 if (slave && IS_UP(slave->dev) && slave->link == BOND_LINK_UP)
3691 bond_dev_queue_xmit(bond, skb, slave->dev);
3695 return NETDEV_TX_OK;
3698 /*------------------------- Device initialization ---------------------------*/
3701 * Lookup the slave that corresponds to a qid
3703 static inline int bond_slave_override(struct bonding *bond,
3704 struct sk_buff *skb)
3706 struct slave *slave = NULL;
3707 struct list_head *iter;
3709 if (!skb->queue_mapping)
3712 /* Find out if any slaves have the same mapping as this skb. */
3713 bond_for_each_slave_rcu(bond, slave, iter) {
3714 if (slave->queue_id == skb->queue_mapping) {
3715 if (slave_can_tx(slave)) {
3716 bond_dev_queue_xmit(bond, skb, slave->dev);
3719 /* If the slave isn't UP, use default transmit policy. */
3728 static u16 bond_select_queue(struct net_device *dev, struct sk_buff *skb,
3732 * This helper function exists to help dev_pick_tx get the correct
3733 * destination queue. Using a helper function skips a call to
3734 * skb_tx_hash and will put the skbs in the queue we expect on their
3735 * way down to the bonding driver.
3737 u16 txq = skb_rx_queue_recorded(skb) ? skb_get_rx_queue(skb) : 0;
3740 * Save the original txq to restore before passing to the driver
3742 qdisc_skb_cb(skb)->slave_dev_queue_mapping = skb->queue_mapping;
3744 if (unlikely(txq >= dev->real_num_tx_queues)) {
3746 txq -= dev->real_num_tx_queues;
3747 } while (txq >= dev->real_num_tx_queues);
3752 static netdev_tx_t __bond_start_xmit(struct sk_buff *skb, struct net_device *dev)
3754 struct bonding *bond = netdev_priv(dev);
3756 if (TX_QUEUE_OVERRIDE(bond->params.mode)) {
3757 if (!bond_slave_override(bond, skb))
3758 return NETDEV_TX_OK;
3761 switch (bond->params.mode) {
3762 case BOND_MODE_ROUNDROBIN:
3763 return bond_xmit_roundrobin(skb, dev);
3764 case BOND_MODE_ACTIVEBACKUP:
3765 return bond_xmit_activebackup(skb, dev);
3767 return bond_xmit_xor(skb, dev);
3768 case BOND_MODE_BROADCAST:
3769 return bond_xmit_broadcast(skb, dev);
3770 case BOND_MODE_8023AD:
3771 return bond_3ad_xmit_xor(skb, dev);
3774 return bond_alb_xmit(skb, dev);
3776 /* Should never happen, mode already checked */
3777 pr_err("%s: Error: Unknown bonding mode %d\n",
3778 dev->name, bond->params.mode);
3781 return NETDEV_TX_OK;
3785 static netdev_tx_t bond_start_xmit(struct sk_buff *skb, struct net_device *dev)
3787 struct bonding *bond = netdev_priv(dev);
3788 netdev_tx_t ret = NETDEV_TX_OK;
3791 * If we risk deadlock from transmitting this in the
3792 * netpoll path, tell netpoll to queue the frame for later tx
3794 if (is_netpoll_tx_blocked(dev))
3795 return NETDEV_TX_BUSY;
3798 if (bond_has_slaves(bond))
3799 ret = __bond_start_xmit(skb, dev);
3807 static int bond_ethtool_get_settings(struct net_device *bond_dev,
3808 struct ethtool_cmd *ecmd)
3810 struct bonding *bond = netdev_priv(bond_dev);
3811 unsigned long speed = 0;
3812 struct list_head *iter;
3813 struct slave *slave;
3815 ecmd->duplex = DUPLEX_UNKNOWN;
3816 ecmd->port = PORT_OTHER;
3818 /* Since SLAVE_IS_OK returns false for all inactive or down slaves, we
3819 * do not need to check mode. Though link speed might not represent
3820 * the true receive or transmit bandwidth (not all modes are symmetric)
3821 * this is an accurate maximum.
3823 read_lock(&bond->lock);
3824 bond_for_each_slave(bond, slave, iter) {
3825 if (SLAVE_IS_OK(slave)) {
3826 if (slave->speed != SPEED_UNKNOWN)
3827 speed += slave->speed;
3828 if (ecmd->duplex == DUPLEX_UNKNOWN &&
3829 slave->duplex != DUPLEX_UNKNOWN)
3830 ecmd->duplex = slave->duplex;
3833 ethtool_cmd_speed_set(ecmd, speed ? : SPEED_UNKNOWN);
3834 read_unlock(&bond->lock);
3839 static void bond_ethtool_get_drvinfo(struct net_device *bond_dev,
3840 struct ethtool_drvinfo *drvinfo)
3842 strlcpy(drvinfo->driver, DRV_NAME, sizeof(drvinfo->driver));
3843 strlcpy(drvinfo->version, DRV_VERSION, sizeof(drvinfo->version));
3844 snprintf(drvinfo->fw_version, sizeof(drvinfo->fw_version), "%d",
3848 static const struct ethtool_ops bond_ethtool_ops = {
3849 .get_drvinfo = bond_ethtool_get_drvinfo,
3850 .get_settings = bond_ethtool_get_settings,
3851 .get_link = ethtool_op_get_link,
3854 static const struct net_device_ops bond_netdev_ops = {
3855 .ndo_init = bond_init,
3856 .ndo_uninit = bond_uninit,
3857 .ndo_open = bond_open,
3858 .ndo_stop = bond_close,
3859 .ndo_start_xmit = bond_start_xmit,
3860 .ndo_select_queue = bond_select_queue,
3861 .ndo_get_stats64 = bond_get_stats,
3862 .ndo_do_ioctl = bond_do_ioctl,
3863 .ndo_change_rx_flags = bond_change_rx_flags,
3864 .ndo_set_rx_mode = bond_set_rx_mode,
3865 .ndo_change_mtu = bond_change_mtu,
3866 .ndo_set_mac_address = bond_set_mac_address,
3867 .ndo_neigh_setup = bond_neigh_setup,
3868 .ndo_vlan_rx_add_vid = bond_vlan_rx_add_vid,
3869 .ndo_vlan_rx_kill_vid = bond_vlan_rx_kill_vid,
3870 #ifdef CONFIG_NET_POLL_CONTROLLER
3871 .ndo_netpoll_setup = bond_netpoll_setup,
3872 .ndo_netpoll_cleanup = bond_netpoll_cleanup,
3873 .ndo_poll_controller = bond_poll_controller,
3875 .ndo_add_slave = bond_enslave,
3876 .ndo_del_slave = bond_release,
3877 .ndo_get_slave = bond_get_slave,
3878 .ndo_fix_features = bond_fix_features,
3881 static const struct device_type bond_type = {
3885 static void bond_destructor(struct net_device *bond_dev)
3887 struct bonding *bond = netdev_priv(bond_dev);
3889 destroy_workqueue(bond->wq);
3890 free_netdev(bond_dev);
3893 void bond_setup(struct net_device *bond_dev)
3895 struct bonding *bond = netdev_priv(bond_dev);
3897 /* initialize rwlocks */
3898 rwlock_init(&bond->lock);
3899 rwlock_init(&bond->curr_slave_lock);
3900 bond->params = bonding_defaults;
3902 /* Initialize pointers */
3903 bond->dev = bond_dev;
3905 /* Initialize the device entry points */
3906 ether_setup(bond_dev);
3907 bond_dev->netdev_ops = &bond_netdev_ops;
3908 bond_dev->ethtool_ops = &bond_ethtool_ops;
3910 bond_dev->destructor = bond_destructor;
3912 SET_NETDEV_DEVTYPE(bond_dev, &bond_type);
3914 /* Initialize the device options */
3915 bond_dev->tx_queue_len = 0;
3916 bond_dev->flags |= IFF_MASTER|IFF_MULTICAST;
3917 bond_dev->priv_flags |= IFF_BONDING;
3918 bond_dev->priv_flags &= ~(IFF_XMIT_DST_RELEASE | IFF_TX_SKB_SHARING);
3920 /* At first, we block adding VLANs. That's the only way to
3921 * prevent problems that occur when adding VLANs over an
3922 * empty bond. The block will be removed once non-challenged
3923 * slaves are enslaved.
3925 bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
3927 /* don't acquire bond device's netif_tx_lock when
3929 bond_dev->features |= NETIF_F_LLTX;
3931 /* By default, we declare the bond to be fully
3932 * VLAN hardware accelerated capable. Special
3933 * care is taken in the various xmit functions
3934 * when there are slaves that are not hw accel
3938 bond_dev->hw_features = BOND_VLAN_FEATURES |
3939 NETIF_F_HW_VLAN_CTAG_TX |
3940 NETIF_F_HW_VLAN_CTAG_RX |
3941 NETIF_F_HW_VLAN_CTAG_FILTER;
3943 bond_dev->hw_features &= ~(NETIF_F_ALL_CSUM & ~NETIF_F_HW_CSUM);
3944 bond_dev->features |= bond_dev->hw_features;
3948 * Destroy a bonding device.
3949 * Must be under rtnl_lock when this function is called.
3951 static void bond_uninit(struct net_device *bond_dev)
3953 struct bonding *bond = netdev_priv(bond_dev);
3954 struct list_head *iter;
3955 struct slave *slave;
3957 bond_netpoll_cleanup(bond_dev);
3959 /* Release the bonded slaves */
3960 bond_for_each_slave(bond, slave, iter)
3961 __bond_release_one(bond_dev, slave->dev, true);
3962 pr_info("%s: released all slaves\n", bond_dev->name);
3964 list_del(&bond->bond_list);
3966 bond_debug_unregister(bond);
3969 /*------------------------- Module initialization ---------------------------*/
3971 int bond_parm_tbl_lookup(int mode, const struct bond_parm_tbl *tbl)
3975 for (i = 0; tbl[i].modename; i++)
3976 if (mode == tbl[i].mode)
3982 static int bond_parm_tbl_lookup_name(const char *modename,
3983 const struct bond_parm_tbl *tbl)
3987 for (i = 0; tbl[i].modename; i++)
3988 if (strcmp(modename, tbl[i].modename) == 0)
3995 * Convert string input module parms. Accept either the
3996 * number of the mode or its string name. A bit complicated because
3997 * some mode names are substrings of other names, and calls from sysfs
3998 * may have whitespace in the name (trailing newlines, for example).
4000 int bond_parse_parm(const char *buf, const struct bond_parm_tbl *tbl)
4003 char *p, modestr[BOND_MAX_MODENAME_LEN + 1];
4005 for (p = (char *)buf; *p; p++)
4006 if (!(isdigit(*p) || isspace(*p)))
4009 if (*p && sscanf(buf, "%20s", modestr) != 0)
4010 return bond_parm_tbl_lookup_name(modestr, tbl);
4011 else if (sscanf(buf, "%d", &modeint) != 0)
4012 return bond_parm_tbl_lookup(modeint, tbl);
4017 static int bond_check_params(struct bond_params *params)
4019 int arp_validate_value, fail_over_mac_value, primary_reselect_value, i;
4020 struct bond_opt_value newval, *valptr;
4021 int arp_all_targets_value;
4024 * Convert string parameters.
4027 bond_opt_initstr(&newval, mode);
4028 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_MODE), &newval);
4030 pr_err("Error: Invalid bonding mode \"%s\"\n", mode);
4033 bond_mode = valptr->value;
4036 if (xmit_hash_policy) {
4037 if ((bond_mode != BOND_MODE_XOR) &&
4038 (bond_mode != BOND_MODE_8023AD)) {
4039 pr_info("xmit_hash_policy param is irrelevant in mode %s\n",
4040 bond_mode_name(bond_mode));
4042 xmit_hashtype = bond_parse_parm(xmit_hash_policy,
4044 if (xmit_hashtype == -1) {
4045 pr_err("Error: Invalid xmit_hash_policy \"%s\"\n",
4046 xmit_hash_policy == NULL ? "NULL" :
4054 if (bond_mode != BOND_MODE_8023AD) {
4055 pr_info("lacp_rate param is irrelevant in mode %s\n",
4056 bond_mode_name(bond_mode));
4058 lacp_fast = bond_parse_parm(lacp_rate, bond_lacp_tbl);
4059 if (lacp_fast == -1) {
4060 pr_err("Error: Invalid lacp rate \"%s\"\n",
4061 lacp_rate == NULL ? "NULL" : lacp_rate);
4068 params->ad_select = bond_parse_parm(ad_select, ad_select_tbl);
4069 if (params->ad_select == -1) {
4070 pr_err("Error: Invalid ad_select \"%s\"\n",
4071 ad_select == NULL ? "NULL" : ad_select);
4075 if (bond_mode != BOND_MODE_8023AD) {
4076 pr_warning("ad_select param only affects 802.3ad mode\n");
4079 params->ad_select = BOND_AD_STABLE;
4082 if (max_bonds < 0) {
4083 pr_warning("Warning: max_bonds (%d) not in range %d-%d, so it was reset to BOND_DEFAULT_MAX_BONDS (%d)\n",
4084 max_bonds, 0, INT_MAX, BOND_DEFAULT_MAX_BONDS);
4085 max_bonds = BOND_DEFAULT_MAX_BONDS;
4089 pr_warning("Warning: miimon module parameter (%d), not in range 0-%d, so it was reset to %d\n",
4090 miimon, INT_MAX, BOND_LINK_MON_INTERV);
4091 miimon = BOND_LINK_MON_INTERV;
4095 pr_warning("Warning: updelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4100 if (downdelay < 0) {
4101 pr_warning("Warning: downdelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4102 downdelay, INT_MAX);
4106 if ((use_carrier != 0) && (use_carrier != 1)) {
4107 pr_warning("Warning: use_carrier module parameter (%d), not of valid value (0/1), so it was set to 1\n",
4112 if (num_peer_notif < 0 || num_peer_notif > 255) {
4113 pr_warning("Warning: num_grat_arp/num_unsol_na (%d) not in range 0-255 so it was reset to 1\n",
4118 /* reset values for 802.3ad/TLB/ALB */
4119 if (BOND_NO_USES_ARP(bond_mode)) {
4121 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");
4122 pr_warning("Forcing miimon to 100msec\n");
4123 miimon = BOND_DEFAULT_MIIMON;
4127 if (tx_queues < 1 || tx_queues > 255) {
4128 pr_warning("Warning: tx_queues (%d) should be between "
4129 "1 and 255, resetting to %d\n",
4130 tx_queues, BOND_DEFAULT_TX_QUEUES);
4131 tx_queues = BOND_DEFAULT_TX_QUEUES;
4134 if ((all_slaves_active != 0) && (all_slaves_active != 1)) {
4135 pr_warning("Warning: all_slaves_active module parameter (%d), "
4136 "not of valid value (0/1), so it was set to "
4137 "0\n", all_slaves_active);
4138 all_slaves_active = 0;
4141 if (resend_igmp < 0 || resend_igmp > 255) {
4142 pr_warning("Warning: resend_igmp (%d) should be between "
4143 "0 and 255, resetting to %d\n",
4144 resend_igmp, BOND_DEFAULT_RESEND_IGMP);
4145 resend_igmp = BOND_DEFAULT_RESEND_IGMP;
4148 if (packets_per_slave < 0 || packets_per_slave > USHRT_MAX) {
4149 pr_warn("Warning: packets_per_slave (%d) should be between 0 and %u resetting to 1\n",
4150 packets_per_slave, USHRT_MAX);
4151 packets_per_slave = 1;
4154 if (bond_mode == BOND_MODE_ALB) {
4155 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",
4160 if (updelay || downdelay) {
4161 /* just warn the user the up/down delay will have
4162 * no effect since miimon is zero...
4164 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",
4165 updelay, downdelay);
4168 /* don't allow arp monitoring */
4170 pr_warning("Warning: miimon (%d) and arp_interval (%d) can't be used simultaneously, disabling ARP monitoring\n",
4171 miimon, arp_interval);
4175 if ((updelay % miimon) != 0) {
4176 pr_warning("Warning: updelay (%d) is not a multiple of miimon (%d), updelay rounded to %d ms\n",
4178 (updelay / miimon) * miimon);
4183 if ((downdelay % miimon) != 0) {
4184 pr_warning("Warning: downdelay (%d) is not a multiple of miimon (%d), downdelay rounded to %d ms\n",
4186 (downdelay / miimon) * miimon);
4189 downdelay /= miimon;
4192 if (arp_interval < 0) {
4193 pr_warning("Warning: arp_interval module parameter (%d) , not in range 0-%d, so it was reset to %d\n",
4194 arp_interval, INT_MAX, BOND_LINK_ARP_INTERV);
4195 arp_interval = BOND_LINK_ARP_INTERV;
4198 for (arp_ip_count = 0, i = 0;
4199 (arp_ip_count < BOND_MAX_ARP_TARGETS) && arp_ip_target[i]; i++) {
4200 /* not complete check, but should be good enough to
4203 if (!in4_pton(arp_ip_target[i], -1, (u8 *)&ip, -1, NULL) ||
4204 IS_IP_TARGET_UNUSABLE_ADDRESS(ip)) {
4205 pr_warning("Warning: bad arp_ip_target module parameter (%s), ARP monitoring will not be performed\n",
4209 if (bond_get_targets_ip(arp_target, ip) == -1)
4210 arp_target[arp_ip_count++] = ip;
4212 pr_warning("Warning: duplicate address %pI4 in arp_ip_target, skipping\n",
4217 if (arp_interval && !arp_ip_count) {
4218 /* don't allow arping if no arp_ip_target given... */
4219 pr_warning("Warning: arp_interval module parameter (%d) specified without providing an arp_ip_target parameter, arp_interval was reset to 0\n",
4225 if (bond_mode != BOND_MODE_ACTIVEBACKUP) {
4226 pr_err("arp_validate only supported in active-backup mode\n");
4229 if (!arp_interval) {
4230 pr_err("arp_validate requires arp_interval\n");
4234 arp_validate_value = bond_parse_parm(arp_validate,
4236 if (arp_validate_value == -1) {
4237 pr_err("Error: invalid arp_validate \"%s\"\n",
4238 arp_validate == NULL ? "NULL" : arp_validate);
4242 arp_validate_value = 0;
4244 arp_all_targets_value = 0;
4245 if (arp_all_targets) {
4246 arp_all_targets_value = bond_parse_parm(arp_all_targets,
4247 arp_all_targets_tbl);
4249 if (arp_all_targets_value == -1) {
4250 pr_err("Error: invalid arp_all_targets_value \"%s\"\n",
4252 arp_all_targets_value = 0;
4257 pr_info("MII link monitoring set to %d ms\n", miimon);
4258 } else if (arp_interval) {
4259 pr_info("ARP monitoring set to %d ms, validate %s, with %d target(s):",
4261 arp_validate_tbl[arp_validate_value].modename,
4264 for (i = 0; i < arp_ip_count; i++)
4265 pr_info(" %s", arp_ip_target[i]);
4269 } else if (max_bonds) {
4270 /* miimon and arp_interval not set, we need one so things
4271 * work as expected, see bonding.txt for details
4273 pr_debug("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");
4276 if (primary && !USES_PRIMARY(bond_mode)) {
4277 /* currently, using a primary only makes sense
4278 * in active backup, TLB or ALB modes
4280 pr_warning("Warning: %s primary device specified but has no effect in %s mode\n",
4281 primary, bond_mode_name(bond_mode));
4285 if (primary && primary_reselect) {
4286 primary_reselect_value = bond_parse_parm(primary_reselect,
4288 if (primary_reselect_value == -1) {
4289 pr_err("Error: Invalid primary_reselect \"%s\"\n",
4291 NULL ? "NULL" : primary_reselect);
4295 primary_reselect_value = BOND_PRI_RESELECT_ALWAYS;
4298 if (fail_over_mac) {
4299 fail_over_mac_value = bond_parse_parm(fail_over_mac,
4301 if (fail_over_mac_value == -1) {
4302 pr_err("Error: invalid fail_over_mac \"%s\"\n",
4303 arp_validate == NULL ? "NULL" : arp_validate);
4307 if (bond_mode != BOND_MODE_ACTIVEBACKUP)
4308 pr_warning("Warning: fail_over_mac only affects active-backup mode.\n");
4310 fail_over_mac_value = BOND_FOM_NONE;
4313 if (lp_interval == 0) {
4314 pr_warning("Warning: ip_interval must be between 1 and %d, so it was reset to %d\n",
4315 INT_MAX, BOND_ALB_DEFAULT_LP_INTERVAL);
4316 lp_interval = BOND_ALB_DEFAULT_LP_INTERVAL;
4319 /* fill params struct with the proper values */
4320 params->mode = bond_mode;
4321 params->xmit_policy = xmit_hashtype;
4322 params->miimon = miimon;
4323 params->num_peer_notif = num_peer_notif;
4324 params->arp_interval = arp_interval;
4325 params->arp_validate = arp_validate_value;
4326 params->arp_all_targets = arp_all_targets_value;
4327 params->updelay = updelay;
4328 params->downdelay = downdelay;
4329 params->use_carrier = use_carrier;
4330 params->lacp_fast = lacp_fast;
4331 params->primary[0] = 0;
4332 params->primary_reselect = primary_reselect_value;
4333 params->fail_over_mac = fail_over_mac_value;
4334 params->tx_queues = tx_queues;
4335 params->all_slaves_active = all_slaves_active;
4336 params->resend_igmp = resend_igmp;
4337 params->min_links = min_links;
4338 params->lp_interval = lp_interval;
4339 params->packets_per_slave = packets_per_slave;
4340 if (packets_per_slave > 0) {
4341 params->reciprocal_packets_per_slave =
4342 reciprocal_value(packets_per_slave);
4344 /* reciprocal_packets_per_slave is unused if
4345 * packets_per_slave is 0 or 1, just initialize it
4347 params->reciprocal_packets_per_slave =
4348 (struct reciprocal_value) { 0 };
4352 strncpy(params->primary, primary, IFNAMSIZ);
4353 params->primary[IFNAMSIZ - 1] = 0;
4356 memcpy(params->arp_targets, arp_target, sizeof(arp_target));
4361 static struct lock_class_key bonding_netdev_xmit_lock_key;
4362 static struct lock_class_key bonding_netdev_addr_lock_key;
4363 static struct lock_class_key bonding_tx_busylock_key;
4365 static void bond_set_lockdep_class_one(struct net_device *dev,
4366 struct netdev_queue *txq,
4369 lockdep_set_class(&txq->_xmit_lock,
4370 &bonding_netdev_xmit_lock_key);
4373 static void bond_set_lockdep_class(struct net_device *dev)
4375 lockdep_set_class(&dev->addr_list_lock,
4376 &bonding_netdev_addr_lock_key);
4377 netdev_for_each_tx_queue(dev, bond_set_lockdep_class_one, NULL);
4378 dev->qdisc_tx_busylock = &bonding_tx_busylock_key;
4382 * Called from registration process
4384 static int bond_init(struct net_device *bond_dev)
4386 struct bonding *bond = netdev_priv(bond_dev);
4387 struct bond_net *bn = net_generic(dev_net(bond_dev), bond_net_id);
4388 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
4390 pr_debug("Begin bond_init for %s\n", bond_dev->name);
4393 * Initialize locks that may be required during
4394 * en/deslave operations. All of the bond_open work
4395 * (of which this is part) should really be moved to
4396 * a phase prior to dev_open
4398 spin_lock_init(&(bond_info->tx_hashtbl_lock));
4399 spin_lock_init(&(bond_info->rx_hashtbl_lock));
4401 bond->wq = create_singlethread_workqueue(bond_dev->name);
4405 bond_set_lockdep_class(bond_dev);
4407 list_add_tail(&bond->bond_list, &bn->dev_list);
4409 bond_prepare_sysfs_group(bond);
4411 bond_debug_register(bond);
4413 /* Ensure valid dev_addr */
4414 if (is_zero_ether_addr(bond_dev->dev_addr) &&
4415 bond_dev->addr_assign_type == NET_ADDR_PERM)
4416 eth_hw_addr_random(bond_dev);
4421 unsigned int bond_get_num_tx_queues(void)
4426 /* Create a new bond based on the specified name and bonding parameters.
4427 * If name is NULL, obtain a suitable "bond%d" name for us.
4428 * Caller must NOT hold rtnl_lock; we need to release it here before we
4429 * set up our sysfs entries.
4431 int bond_create(struct net *net, const char *name)
4433 struct net_device *bond_dev;
4438 bond_dev = alloc_netdev_mq(sizeof(struct bonding),
4439 name ? name : "bond%d",
4440 bond_setup, tx_queues);
4442 pr_err("%s: eek! can't alloc netdev!\n", name);
4447 dev_net_set(bond_dev, net);
4448 bond_dev->rtnl_link_ops = &bond_link_ops;
4450 res = register_netdevice(bond_dev);
4452 netif_carrier_off(bond_dev);
4456 bond_destructor(bond_dev);
4460 static int __net_init bond_net_init(struct net *net)
4462 struct bond_net *bn = net_generic(net, bond_net_id);
4465 INIT_LIST_HEAD(&bn->dev_list);
4467 bond_create_proc_dir(bn);
4468 bond_create_sysfs(bn);
4473 static void __net_exit bond_net_exit(struct net *net)
4475 struct bond_net *bn = net_generic(net, bond_net_id);
4476 struct bonding *bond, *tmp_bond;
4479 bond_destroy_sysfs(bn);
4480 bond_destroy_proc_dir(bn);
4482 /* Kill off any bonds created after unregistering bond rtnl ops */
4484 list_for_each_entry_safe(bond, tmp_bond, &bn->dev_list, bond_list)
4485 unregister_netdevice_queue(bond->dev, &list);
4486 unregister_netdevice_many(&list);
4490 static struct pernet_operations bond_net_ops = {
4491 .init = bond_net_init,
4492 .exit = bond_net_exit,
4494 .size = sizeof(struct bond_net),
4497 static int __init bonding_init(void)
4502 pr_info("%s", bond_version);
4504 res = bond_check_params(&bonding_defaults);
4508 res = register_pernet_subsys(&bond_net_ops);
4512 res = bond_netlink_init();
4516 bond_create_debugfs();
4518 for (i = 0; i < max_bonds; i++) {
4519 res = bond_create(&init_net, NULL);
4524 register_netdevice_notifier(&bond_netdev_notifier);
4528 bond_netlink_fini();
4530 unregister_pernet_subsys(&bond_net_ops);
4535 static void __exit bonding_exit(void)
4537 unregister_netdevice_notifier(&bond_netdev_notifier);
4539 bond_destroy_debugfs();
4541 bond_netlink_fini();
4542 unregister_pernet_subsys(&bond_net_ops);
4544 #ifdef CONFIG_NET_POLL_CONTROLLER
4546 * Make sure we don't have an imbalance on our netpoll blocking
4548 WARN_ON(atomic_read(&netpoll_block_tx));
4552 module_init(bonding_init);
4553 module_exit(bonding_exit);
4554 MODULE_LICENSE("GPL");
4555 MODULE_VERSION(DRV_VERSION);
4556 MODULE_DESCRIPTION(DRV_DESCRIPTION ", v" DRV_VERSION);
4557 MODULE_AUTHOR("Thomas Davis, tadavis@lbl.gov and many others");