1 /*
2 * originally based on the dummy device.
3 *
4 * Copyright 1999, Thomas Davis, tadavis@lbl.gov.
5 * Licensed under the GPL. Based on dummy.c, and eql.c devices.
6 *
7 * bonding.c: an Ethernet Bonding driver
8 *
9 * This is useful to talk to a Cisco EtherChannel compatible equipment:
10 * Cisco 5500
11 * Sun Trunking (Solaris)
12 * Alteon AceDirector Trunks
13 * Linux Bonding
14 * and probably many L2 switches ...
15 *
16 * How it works:
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.
22 *
23 * ifconfig bond0 down
24 * will release all slaves, marking them as down.
25 *
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.
31 *
32 */
34 #include <linux/kernel.h>
35 #include <linux/module.h>
36 #include <linux/types.h>
37 #include <linux/fcntl.h>
38 #include <linux/interrupt.h>
39 #include <linux/ptrace.h>
40 #include <linux/ioport.h>
41 #include <linux/in.h>
42 #include <net/ip.h>
43 #include <linux/ip.h>
44 #include <linux/tcp.h>
45 #include <linux/udp.h>
46 #include <linux/slab.h>
47 #include <linux/string.h>
48 #include <linux/init.h>
49 #include <linux/timer.h>
50 #include <linux/socket.h>
51 #include <linux/ctype.h>
52 #include <linux/inet.h>
53 #include <linux/bitops.h>
54 #include <linux/io.h>
55 #include <asm/dma.h>
56 #include <linux/uaccess.h>
57 #include <linux/errno.h>
58 #include <linux/netdevice.h>
59 #include <linux/inetdevice.h>
60 #include <linux/igmp.h>
61 #include <linux/etherdevice.h>
62 #include <linux/skbuff.h>
63 #include <net/sock.h>
64 #include <linux/rtnetlink.h>
65 #include <linux/smp.h>
66 #include <linux/if_ether.h>
67 #include <net/arp.h>
68 #include <linux/mii.h>
69 #include <linux/ethtool.h>
70 #include <linux/if_vlan.h>
71 #include <linux/if_bonding.h>
72 #include <linux/jiffies.h>
73 #include <linux/preempt.h>
74 #include <net/route.h>
75 #include <net/net_namespace.h>
76 #include <net/netns/generic.h>
77 #include <net/pkt_sched.h>
78 #include <linux/rculist.h>
79 #include <net/flow_dissector.h>
80 #include <net/switchdev.h>
81 #include <net/bonding.h>
82 #include <net/bond_3ad.h>
83 #include <net/bond_alb.h>
85 #include "bonding_priv.h"
87 /*---------------------------- Module parameters ----------------------------*/
89 /* monitor all links that often (in milliseconds). <=0 disables monitoring */
91 static int max_bonds = BOND_DEFAULT_MAX_BONDS;
92 static int tx_queues = BOND_DEFAULT_TX_QUEUES;
93 static int num_peer_notif = 1;
94 static int miimon;
95 static int updelay;
96 static int downdelay;
97 static int use_carrier = 1;
98 static char *mode;
99 static char *primary;
100 static char *primary_reselect;
101 static char *lacp_rate;
102 static int min_links;
103 static char *ad_select;
104 static char *xmit_hash_policy;
105 static int arp_interval;
106 static char *arp_ip_target[BOND_MAX_ARP_TARGETS];
107 static char *arp_validate;
108 static char *arp_all_targets;
109 static char *fail_over_mac;
110 static int all_slaves_active;
111 static struct bond_params bonding_defaults;
112 static int resend_igmp = BOND_DEFAULT_RESEND_IGMP;
113 static int packets_per_slave = 1;
114 static int lp_interval = BOND_ALB_DEFAULT_LP_INTERVAL;
116 module_param(max_bonds, int, 0);
117 MODULE_PARM_DESC(max_bonds, "Max number of bonded devices");
118 module_param(tx_queues, int, 0);
119 MODULE_PARM_DESC(tx_queues, "Max number of transmit queues (default = 16)");
120 module_param_named(num_grat_arp, num_peer_notif, int, 0644);
121 MODULE_PARM_DESC(num_grat_arp, "Number of peer notifications to send on "
122 "failover event (alias of num_unsol_na)");
123 module_param_named(num_unsol_na, num_peer_notif, int, 0644);
124 MODULE_PARM_DESC(num_unsol_na, "Number of peer notifications to send on "
125 "failover event (alias of num_grat_arp)");
126 module_param(miimon, int, 0);
127 MODULE_PARM_DESC(miimon, "Link check interval in milliseconds");
128 module_param(updelay, int, 0);
129 MODULE_PARM_DESC(updelay, "Delay before considering link up, in milliseconds");
130 module_param(downdelay, int, 0);
131 MODULE_PARM_DESC(downdelay, "Delay before considering link down, "
132 "in milliseconds");
133 module_param(use_carrier, int, 0);
134 MODULE_PARM_DESC(use_carrier, "Use netif_carrier_ok (vs MII ioctls) in miimon; "
135 "0 for off, 1 for on (default)");
136 module_param(mode, charp, 0);
137 MODULE_PARM_DESC(mode, "Mode of operation; 0 for balance-rr, "
138 "1 for active-backup, 2 for balance-xor, "
139 "3 for broadcast, 4 for 802.3ad, 5 for balance-tlb, "
140 "6 for balance-alb");
141 module_param(primary, charp, 0);
142 MODULE_PARM_DESC(primary, "Primary network device to use");
143 module_param(primary_reselect, charp, 0);
144 MODULE_PARM_DESC(primary_reselect, "Reselect primary slave "
145 "once it comes up; "
146 "0 for always (default), "
147 "1 for only if speed of primary is "
148 "better, "
149 "2 for only on active slave "
150 "failure");
151 module_param(lacp_rate, charp, 0);
152 MODULE_PARM_DESC(lacp_rate, "LACPDU tx rate to request from 802.3ad partner; "
153 "0 for slow, 1 for fast");
154 module_param(ad_select, charp, 0);
155 MODULE_PARM_DESC(ad_select, "802.3ad aggregation selection logic; "
156 "0 for stable (default), 1 for bandwidth, "
157 "2 for count");
158 module_param(min_links, int, 0);
159 MODULE_PARM_DESC(min_links, "Minimum number of available links before turning on carrier");
161 module_param(xmit_hash_policy, charp, 0);
162 MODULE_PARM_DESC(xmit_hash_policy, "balance-alb, balance-tlb, balance-xor, 802.3ad hashing method; "
163 "0 for layer 2 (default), 1 for layer 3+4, "
164 "2 for layer 2+3, 3 for encap layer 2+3, "
165 "4 for encap layer 3+4");
166 module_param(arp_interval, int, 0);
167 MODULE_PARM_DESC(arp_interval, "arp interval in milliseconds");
168 module_param_array(arp_ip_target, charp, NULL, 0);
169 MODULE_PARM_DESC(arp_ip_target, "arp targets in n.n.n.n form");
170 module_param(arp_validate, charp, 0);
171 MODULE_PARM_DESC(arp_validate, "validate src/dst of ARP probes; "
172 "0 for none (default), 1 for active, "
173 "2 for backup, 3 for all");
174 module_param(arp_all_targets, charp, 0);
175 MODULE_PARM_DESC(arp_all_targets, "fail on any/all arp targets timeout; 0 for any (default), 1 for all");
176 module_param(fail_over_mac, charp, 0);
177 MODULE_PARM_DESC(fail_over_mac, "For active-backup, do not set all slaves to "
178 "the same MAC; 0 for none (default), "
179 "1 for active, 2 for follow");
180 module_param(all_slaves_active, int, 0);
181 MODULE_PARM_DESC(all_slaves_active, "Keep all frames received on an interface "
182 "by setting active flag for all slaves; "
183 "0 for never (default), 1 for always.");
184 module_param(resend_igmp, int, 0);
185 MODULE_PARM_DESC(resend_igmp, "Number of IGMP membership reports to send on "
186 "link failure");
187 module_param(packets_per_slave, int, 0);
188 MODULE_PARM_DESC(packets_per_slave, "Packets to send per slave in balance-rr "
189 "mode; 0 for a random slave, 1 packet per "
190 "slave (default), >1 packets per slave.");
191 module_param(lp_interval, uint, 0);
192 MODULE_PARM_DESC(lp_interval, "The number of seconds between instances where "
193 "the bonding driver sends learning packets to "
194 "each slaves peer switch. The default is 1.");
196 /*----------------------------- Global variables ----------------------------*/
198 #ifdef CONFIG_NET_POLL_CONTROLLER
199 atomic_t netpoll_block_tx = ATOMIC_INIT(0);
200 #endif
202 unsigned int bond_net_id __read_mostly;
204 /*-------------------------- Forward declarations ---------------------------*/
206 static int bond_init(struct net_device *bond_dev);
207 static void bond_uninit(struct net_device *bond_dev);
208 static void bond_get_stats(struct net_device *bond_dev,
209 struct rtnl_link_stats64 *stats);
210 static void bond_slave_arr_handler(struct work_struct *work);
211 static bool bond_time_in_interval(struct bonding *bond, unsigned long last_act,
212 int mod);
213 static void bond_netdev_notify_work(struct work_struct *work);
215 /*---------------------------- General routines -----------------------------*/
217 const char *bond_mode_name(int mode)
218 {
219 static const char *names[] = {
220 [BOND_MODE_ROUNDROBIN] = "load balancing (round-robin)",
221 [BOND_MODE_ACTIVEBACKUP] = "fault-tolerance (active-backup)",
222 [BOND_MODE_XOR] = "load balancing (xor)",
223 [BOND_MODE_BROADCAST] = "fault-tolerance (broadcast)",
224 [BOND_MODE_8023AD] = "IEEE 802.3ad Dynamic link aggregation",
225 [BOND_MODE_TLB] = "transmit load balancing",
226 [BOND_MODE_ALB] = "adaptive load balancing",
227 };
229 if (mode < BOND_MODE_ROUNDROBIN || mode > BOND_MODE_ALB)
230 return "unknown";
232 return names[mode];
233 }
235 /*---------------------------------- VLAN -----------------------------------*/
237 /**
238 * bond_dev_queue_xmit - Prepare skb for xmit.
239 *
240 * @bond: bond device that got this skb for tx.
241 * @skb: hw accel VLAN tagged skb to transmit
242 * @slave_dev: slave that is supposed to xmit this skbuff
243 */
244 void bond_dev_queue_xmit(struct bonding *bond, struct sk_buff *skb,
245 struct net_device *slave_dev)
246 {
247 skb->dev = slave_dev;
249 BUILD_BUG_ON(sizeof(skb->queue_mapping) !=
250 sizeof(qdisc_skb_cb(skb)->slave_dev_queue_mapping));
251 skb_set_queue_mapping(skb, qdisc_skb_cb(skb)->slave_dev_queue_mapping);
253 if (unlikely(netpoll_tx_running(bond->dev)))
254 bond_netpoll_send_skb(bond_get_slave_by_dev(bond, slave_dev), skb);
255 else
256 dev_queue_xmit(skb);
257 }
259 /* In the following 2 functions, bond_vlan_rx_add_vid and bond_vlan_rx_kill_vid,
260 * We don't protect the slave list iteration with a lock because:
261 * a. This operation is performed in IOCTL context,
262 * b. The operation is protected by the RTNL semaphore in the 8021q code,
263 * c. Holding a lock with BH disabled while directly calling a base driver
264 * entry point is generally a BAD idea.
265 *
266 * The design of synchronization/protection for this operation in the 8021q
267 * module is good for one or more VLAN devices over a single physical device
268 * and cannot be extended for a teaming solution like bonding, so there is a
269 * potential race condition here where a net device from the vlan group might
270 * be referenced (either by a base driver or the 8021q code) while it is being
271 * removed from the system. However, it turns out we're not making matters
272 * worse, and if it works for regular VLAN usage it will work here too.
273 */
275 /**
276 * bond_vlan_rx_add_vid - Propagates adding an id to slaves
277 * @bond_dev: bonding net device that got called
278 * @vid: vlan id being added
279 */
280 static int bond_vlan_rx_add_vid(struct net_device *bond_dev,
281 __be16 proto, u16 vid)
282 {
283 struct bonding *bond = netdev_priv(bond_dev);
284 struct slave *slave, *rollback_slave;
285 struct list_head *iter;
286 int res;
288 bond_for_each_slave(bond, slave, iter) {
289 res = vlan_vid_add(slave->dev, proto, vid);
290 if (res)
291 goto unwind;
292 }
294 return 0;
296 unwind:
297 /* unwind to the slave that failed */
298 bond_for_each_slave(bond, rollback_slave, iter) {
299 if (rollback_slave == slave)
300 break;
302 vlan_vid_del(rollback_slave->dev, proto, vid);
303 }
305 return res;
306 }
308 /**
309 * bond_vlan_rx_kill_vid - Propagates deleting an id to slaves
310 * @bond_dev: bonding net device that got called
311 * @vid: vlan id being removed
312 */
313 static int bond_vlan_rx_kill_vid(struct net_device *bond_dev,
314 __be16 proto, u16 vid)
315 {
316 struct bonding *bond = netdev_priv(bond_dev);
317 struct list_head *iter;
318 struct slave *slave;
320 bond_for_each_slave(bond, slave, iter)
321 vlan_vid_del(slave->dev, proto, vid);
323 if (bond_is_lb(bond))
324 bond_alb_clear_vlan(bond, vid);
326 return 0;
327 }
329 /*------------------------------- Link status -------------------------------*/
331 /* Set the carrier state for the master according to the state of its
332 * slaves. If any slaves are up, the master is up. In 802.3ad mode,
333 * do special 802.3ad magic.
334 *
335 * Returns zero if carrier state does not change, nonzero if it does.
336 */
337 int bond_set_carrier(struct bonding *bond)
338 {
339 struct list_head *iter;
340 struct slave *slave;
342 if (!bond_has_slaves(bond))
343 goto down;
345 if (BOND_MODE(bond) == BOND_MODE_8023AD)
346 return bond_3ad_set_carrier(bond);
348 bond_for_each_slave(bond, slave, iter) {
349 if (slave->link == BOND_LINK_UP) {
350 if (!netif_carrier_ok(bond->dev)) {
351 netif_carrier_on(bond->dev);
352 return 1;
353 }
354 return 0;
355 }
356 }
358 down:
359 if (netif_carrier_ok(bond->dev)) {
360 netif_carrier_off(bond->dev);
361 return 1;
362 }
363 return 0;
364 }
366 /* Get link speed and duplex from the slave's base driver
367 * using ethtool. If for some reason the call fails or the
368 * values are invalid, set speed and duplex to -1,
369 * and return. Return 1 if speed or duplex settings are
370 * UNKNOWN; 0 otherwise.
371 */
372 static int bond_update_speed_duplex(struct slave *slave)
373 {
374 struct net_device *slave_dev = slave->dev;
375 struct ethtool_link_ksettings ecmd;
376 int res;
378 slave->speed = SPEED_UNKNOWN;
379 slave->duplex = DUPLEX_UNKNOWN;
381 res = __ethtool_get_link_ksettings(slave_dev, &ecmd);
382 if (res < 0)
383 return 1;
384 if (ecmd.base.speed == 0 || ecmd.base.speed == ((__u32)-1))
385 return 1;
386 switch (ecmd.base.duplex) {
387 case DUPLEX_FULL:
388 case DUPLEX_HALF:
389 break;
390 default:
391 return 1;
392 }
394 slave->speed = ecmd.base.speed;
395 slave->duplex = ecmd.base.duplex;
397 return 0;
398 }
400 const char *bond_slave_link_status(s8 link)
401 {
402 switch (link) {
403 case BOND_LINK_UP:
404 return "up";
405 case BOND_LINK_FAIL:
406 return "going down";
407 case BOND_LINK_DOWN:
408 return "down";
409 case BOND_LINK_BACK:
410 return "going back";
411 default:
412 return "unknown";
413 }
414 }
416 /* if <dev> supports MII link status reporting, check its link status.
417 *
418 * We either do MII/ETHTOOL ioctls, or check netif_carrier_ok(),
419 * depending upon the setting of the use_carrier parameter.
420 *
421 * Return either BMSR_LSTATUS, meaning that the link is up (or we
422 * can't tell and just pretend it is), or 0, meaning that the link is
423 * down.
424 *
425 * If reporting is non-zero, instead of faking link up, return -1 if
426 * both ETHTOOL and MII ioctls fail (meaning the device does not
427 * support them). If use_carrier is set, return whatever it says.
428 * It'd be nice if there was a good way to tell if a driver supports
429 * netif_carrier, but there really isn't.
430 */
431 static int bond_check_dev_link(struct bonding *bond,
432 struct net_device *slave_dev, int reporting)
433 {
434 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
435 int (*ioctl)(struct net_device *, struct ifreq *, int);
436 struct ifreq ifr;
437 struct mii_ioctl_data *mii;
439 if (!reporting && !netif_running(slave_dev))
440 return 0;
442 if (bond->params.use_carrier)
443 return netif_carrier_ok(slave_dev) ? BMSR_LSTATUS : 0;
445 /* Try to get link status using Ethtool first. */
446 if (slave_dev->ethtool_ops->get_link)
447 return slave_dev->ethtool_ops->get_link(slave_dev) ?
448 BMSR_LSTATUS : 0;
450 /* Ethtool can't be used, fallback to MII ioctls. */
451 ioctl = slave_ops->ndo_do_ioctl;
452 if (ioctl) {
453 /* TODO: set pointer to correct ioctl on a per team member
454 * bases to make this more efficient. that is, once
455 * we determine the correct ioctl, we will always
456 * call it and not the others for that team
457 * member.
458 */
460 /* We cannot assume that SIOCGMIIPHY will also read a
461 * register; not all network drivers (e.g., e100)
462 * support that.
463 */
465 /* Yes, the mii is overlaid on the ifreq.ifr_ifru */
466 strncpy(ifr.ifr_name, slave_dev->name, IFNAMSIZ);
467 mii = if_mii(&ifr);
468 if (ioctl(slave_dev, &ifr, SIOCGMIIPHY) == 0) {
469 mii->reg_num = MII_BMSR;
470 if (ioctl(slave_dev, &ifr, SIOCGMIIREG) == 0)
471 return mii->val_out & BMSR_LSTATUS;
472 }
473 }
475 /* If reporting, report that either there's no dev->do_ioctl,
476 * or both SIOCGMIIREG and get_link failed (meaning that we
477 * cannot report link status). If not reporting, pretend
478 * we're ok.
479 */
480 return reporting ? -1 : BMSR_LSTATUS;
481 }
483 /*----------------------------- Multicast list ------------------------------*/
485 /* Push the promiscuity flag down to appropriate slaves */
486 static int bond_set_promiscuity(struct bonding *bond, int inc)
487 {
488 struct list_head *iter;
489 int err = 0;
491 if (bond_uses_primary(bond)) {
492 struct slave *curr_active = rtnl_dereference(bond->curr_active_slave);
494 if (curr_active)
495 err = dev_set_promiscuity(curr_active->dev, inc);
496 } else {
497 struct slave *slave;
499 bond_for_each_slave(bond, slave, iter) {
500 err = dev_set_promiscuity(slave->dev, inc);
501 if (err)
502 return err;
503 }
504 }
505 return err;
506 }
508 /* Push the allmulti flag down to all slaves */
509 static int bond_set_allmulti(struct bonding *bond, int inc)
510 {
511 struct list_head *iter;
512 int err = 0;
514 if (bond_uses_primary(bond)) {
515 struct slave *curr_active = rtnl_dereference(bond->curr_active_slave);
517 if (curr_active)
518 err = dev_set_allmulti(curr_active->dev, inc);
519 } else {
520 struct slave *slave;
522 bond_for_each_slave(bond, slave, iter) {
523 err = dev_set_allmulti(slave->dev, inc);
524 if (err)
525 return err;
526 }
527 }
528 return err;
529 }
531 /* Retrieve the list of registered multicast addresses for the bonding
532 * device and retransmit an IGMP JOIN request to the current active
533 * slave.
534 */
535 static void bond_resend_igmp_join_requests_delayed(struct work_struct *work)
536 {
537 struct bonding *bond = container_of(work, struct bonding,
538 mcast_work.work);
540 if (!rtnl_trylock()) {
541 queue_delayed_work(bond->wq, &bond->mcast_work, 1);
542 return;
543 }
544 call_netdevice_notifiers(NETDEV_RESEND_IGMP, bond->dev);
546 if (bond->igmp_retrans > 1) {
547 bond->igmp_retrans--;
548 queue_delayed_work(bond->wq, &bond->mcast_work, HZ/5);
549 }
550 rtnl_unlock();
551 }
553 /* Flush bond's hardware addresses from slave */
554 static void bond_hw_addr_flush(struct net_device *bond_dev,
555 struct net_device *slave_dev)
556 {
557 struct bonding *bond = netdev_priv(bond_dev);
559 dev_uc_unsync(slave_dev, bond_dev);
560 dev_mc_unsync(slave_dev, bond_dev);
562 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
563 /* del lacpdu mc addr from mc list */
564 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
566 dev_mc_del(slave_dev, lacpdu_multicast);
567 }
568 }
570 /*--------------------------- Active slave change ---------------------------*/
572 /* Update the hardware address list and promisc/allmulti for the new and
573 * old active slaves (if any). Modes that are not using primary keep all
574 * slaves up date at all times; only the modes that use primary need to call
575 * this function to swap these settings during a failover.
576 */
577 static void bond_hw_addr_swap(struct bonding *bond, struct slave *new_active,
578 struct slave *old_active)
579 {
580 if (old_active) {
581 if (bond->dev->flags & IFF_PROMISC)
582 dev_set_promiscuity(old_active->dev, -1);
584 if (bond->dev->flags & IFF_ALLMULTI)
585 dev_set_allmulti(old_active->dev, -1);
587 bond_hw_addr_flush(bond->dev, old_active->dev);
588 }
590 if (new_active) {
591 /* FIXME: Signal errors upstream. */
592 if (bond->dev->flags & IFF_PROMISC)
593 dev_set_promiscuity(new_active->dev, 1);
595 if (bond->dev->flags & IFF_ALLMULTI)
596 dev_set_allmulti(new_active->dev, 1);
598 netif_addr_lock_bh(bond->dev);
599 dev_uc_sync(new_active->dev, bond->dev);
600 dev_mc_sync(new_active->dev, bond->dev);
601 netif_addr_unlock_bh(bond->dev);
602 }
603 }
605 /**
606 * bond_set_dev_addr - clone slave's address to bond
607 * @bond_dev: bond net device
608 * @slave_dev: slave net device
609 *
610 * Should be called with RTNL held.
611 */
612 static void bond_set_dev_addr(struct net_device *bond_dev,
613 struct net_device *slave_dev)
614 {
615 netdev_dbg(bond_dev, "bond_dev=%p slave_dev=%p slave_dev->name=%s slave_dev->addr_len=%d\n",
616 bond_dev, slave_dev, slave_dev->name, slave_dev->addr_len);
617 memcpy(bond_dev->dev_addr, slave_dev->dev_addr, slave_dev->addr_len);
618 bond_dev->addr_assign_type = NET_ADDR_STOLEN;
619 call_netdevice_notifiers(NETDEV_CHANGEADDR, bond_dev);
620 }
622 static struct slave *bond_get_old_active(struct bonding *bond,
623 struct slave *new_active)
624 {
625 struct slave *slave;
626 struct list_head *iter;
628 bond_for_each_slave(bond, slave, iter) {
629 if (slave == new_active)
630 continue;
632 if (ether_addr_equal(bond->dev->dev_addr, slave->dev->dev_addr))
633 return slave;
634 }
636 return NULL;
637 }
639 /* bond_do_fail_over_mac
640 *
641 * Perform special MAC address swapping for fail_over_mac settings
642 *
643 * Called with RTNL
644 */
645 static void bond_do_fail_over_mac(struct bonding *bond,
646 struct slave *new_active,
647 struct slave *old_active)
648 {
649 u8 tmp_mac[MAX_ADDR_LEN];
650 struct sockaddr_storage ss;
651 int rv;
653 switch (bond->params.fail_over_mac) {
654 case BOND_FOM_ACTIVE:
655 if (new_active)
656 bond_set_dev_addr(bond->dev, new_active->dev);
657 break;
658 case BOND_FOM_FOLLOW:
659 /* if new_active && old_active, swap them
660 * if just old_active, do nothing (going to no active slave)
661 * if just new_active, set new_active to bond's MAC
662 */
663 if (!new_active)
664 return;
666 if (!old_active)
667 old_active = bond_get_old_active(bond, new_active);
669 if (old_active) {
670 bond_hw_addr_copy(tmp_mac, new_active->dev->dev_addr,
671 new_active->dev->addr_len);
672 bond_hw_addr_copy(ss.__data,
673 old_active->dev->dev_addr,
674 old_active->dev->addr_len);
675 ss.ss_family = new_active->dev->type;
676 } else {
677 bond_hw_addr_copy(ss.__data, bond->dev->dev_addr,
678 bond->dev->addr_len);
679 ss.ss_family = bond->dev->type;
680 }
682 rv = dev_set_mac_address(new_active->dev,
683 (struct sockaddr *)&ss);
684 if (rv) {
685 netdev_err(bond->dev, "Error %d setting MAC of slave %s\n",
686 -rv, new_active->dev->name);
687 goto out;
688 }
690 if (!old_active)
691 goto out;
693 bond_hw_addr_copy(ss.__data, tmp_mac,
694 new_active->dev->addr_len);
695 ss.ss_family = old_active->dev->type;
697 rv = dev_set_mac_address(old_active->dev,
698 (struct sockaddr *)&ss);
699 if (rv)
700 netdev_err(bond->dev, "Error %d setting MAC of slave %s\n",
701 -rv, new_active->dev->name);
702 out:
703 break;
704 default:
705 netdev_err(bond->dev, "bond_do_fail_over_mac impossible: bad policy %d\n",
706 bond->params.fail_over_mac);
707 break;
708 }
710 }
712 static struct slave *bond_choose_primary_or_current(struct bonding *bond)
713 {
714 struct slave *prim = rtnl_dereference(bond->primary_slave);
715 struct slave *curr = rtnl_dereference(bond->curr_active_slave);
717 if (!prim || prim->link != BOND_LINK_UP) {
718 if (!curr || curr->link != BOND_LINK_UP)
719 return NULL;
720 return curr;
721 }
723 if (bond->force_primary) {
724 bond->force_primary = false;
725 return prim;
726 }
728 if (!curr || curr->link != BOND_LINK_UP)
729 return prim;
731 /* At this point, prim and curr are both up */
732 switch (bond->params.primary_reselect) {
733 case BOND_PRI_RESELECT_ALWAYS:
734 return prim;
735 case BOND_PRI_RESELECT_BETTER:
736 if (prim->speed < curr->speed)
737 return curr;
738 if (prim->speed == curr->speed && prim->duplex <= curr->duplex)
739 return curr;
740 return prim;
741 case BOND_PRI_RESELECT_FAILURE:
742 return curr;
743 default:
744 netdev_err(bond->dev, "impossible primary_reselect %d\n",
745 bond->params.primary_reselect);
746 return curr;
747 }
748 }
750 /**
751 * bond_find_best_slave - select the best available slave to be the active one
752 * @bond: our bonding struct
753 */
754 static struct slave *bond_find_best_slave(struct bonding *bond)
755 {
756 struct slave *slave, *bestslave = NULL;
757 struct list_head *iter;
758 int mintime = bond->params.updelay;
760 slave = bond_choose_primary_or_current(bond);
761 if (slave)
762 return slave;
764 bond_for_each_slave(bond, slave, iter) {
765 if (slave->link == BOND_LINK_UP)
766 return slave;
767 if (slave->link == BOND_LINK_BACK && bond_slave_is_up(slave) &&
768 slave->delay < mintime) {
769 mintime = slave->delay;
770 bestslave = slave;
771 }
772 }
774 return bestslave;
775 }
777 static bool bond_should_notify_peers(struct bonding *bond)
778 {
779 struct slave *slave;
781 rcu_read_lock();
782 slave = rcu_dereference(bond->curr_active_slave);
783 rcu_read_unlock();
785 netdev_dbg(bond->dev, "bond_should_notify_peers: slave %s\n",
786 slave ? slave->dev->name : "NULL");
788 if (!slave || !bond->send_peer_notif ||
789 !netif_carrier_ok(bond->dev) ||
790 test_bit(__LINK_STATE_LINKWATCH_PENDING, &slave->dev->state))
791 return false;
793 return true;
794 }
796 /**
797 * change_active_interface - change the active slave into the specified one
798 * @bond: our bonding struct
799 * @new: the new slave to make the active one
800 *
801 * Set the new slave to the bond's settings and unset them on the old
802 * curr_active_slave.
803 * Setting include flags, mc-list, promiscuity, allmulti, etc.
804 *
805 * If @new's link state is %BOND_LINK_BACK we'll set it to %BOND_LINK_UP,
806 * because it is apparently the best available slave we have, even though its
807 * updelay hasn't timed out yet.
808 *
809 * Caller must hold RTNL.
810 */
811 void bond_change_active_slave(struct bonding *bond, struct slave *new_active)
812 {
813 struct slave *old_active;
815 ASSERT_RTNL();
817 old_active = rtnl_dereference(bond->curr_active_slave);
819 if (old_active == new_active)
820 return;
822 if (new_active) {
823 new_active->last_link_up = jiffies;
825 if (new_active->link == BOND_LINK_BACK) {
826 if (bond_uses_primary(bond)) {
827 netdev_info(bond->dev, "making interface %s the new active one %d ms earlier\n",
828 new_active->dev->name,
829 (bond->params.updelay - new_active->delay) * bond->params.miimon);
830 }
832 new_active->delay = 0;
833 bond_set_slave_link_state(new_active, BOND_LINK_UP,
834 BOND_SLAVE_NOTIFY_NOW);
836 if (BOND_MODE(bond) == BOND_MODE_8023AD)
837 bond_3ad_handle_link_change(new_active, BOND_LINK_UP);
839 if (bond_is_lb(bond))
840 bond_alb_handle_link_change(bond, new_active, BOND_LINK_UP);
841 } else {
842 if (bond_uses_primary(bond)) {
843 netdev_info(bond->dev, "making interface %s the new active one\n",
844 new_active->dev->name);
845 }
846 }
847 }
849 if (bond_uses_primary(bond))
850 bond_hw_addr_swap(bond, new_active, old_active);
852 if (bond_is_lb(bond)) {
853 bond_alb_handle_active_change(bond, new_active);
854 if (old_active)
855 bond_set_slave_inactive_flags(old_active,
856 BOND_SLAVE_NOTIFY_NOW);
857 if (new_active)
858 bond_set_slave_active_flags(new_active,
859 BOND_SLAVE_NOTIFY_NOW);
860 } else {
861 rcu_assign_pointer(bond->curr_active_slave, new_active);
862 }
864 if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP) {
865 if (old_active)
866 bond_set_slave_inactive_flags(old_active,
867 BOND_SLAVE_NOTIFY_NOW);
869 if (new_active) {
870 bool should_notify_peers = false;
872 bond_set_slave_active_flags(new_active,
873 BOND_SLAVE_NOTIFY_NOW);
875 if (bond->params.fail_over_mac)
876 bond_do_fail_over_mac(bond, new_active,
877 old_active);
879 if (netif_running(bond->dev)) {
880 bond->send_peer_notif =
881 bond->params.num_peer_notif;
882 should_notify_peers =
883 bond_should_notify_peers(bond);
884 }
886 call_netdevice_notifiers(NETDEV_BONDING_FAILOVER, bond->dev);
887 if (should_notify_peers)
888 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS,
889 bond->dev);
890 }
891 }
893 /* resend IGMP joins since active slave has changed or
894 * all were sent on curr_active_slave.
895 * resend only if bond is brought up with the affected
896 * bonding modes and the retransmission is enabled
897 */
898 if (netif_running(bond->dev) && (bond->params.resend_igmp > 0) &&
899 ((bond_uses_primary(bond) && new_active) ||
900 BOND_MODE(bond) == BOND_MODE_ROUNDROBIN)) {
901 bond->igmp_retrans = bond->params.resend_igmp;
902 queue_delayed_work(bond->wq, &bond->mcast_work, 1);
903 }
904 }
906 /**
907 * bond_select_active_slave - select a new active slave, if needed
908 * @bond: our bonding struct
909 *
910 * This functions should be called when one of the following occurs:
911 * - The old curr_active_slave has been released or lost its link.
912 * - The primary_slave has got its link back.
913 * - A slave has got its link back and there's no old curr_active_slave.
914 *
915 * Caller must hold RTNL.
916 */
917 void bond_select_active_slave(struct bonding *bond)
918 {
919 struct slave *best_slave;
920 int rv;
922 ASSERT_RTNL();
924 best_slave = bond_find_best_slave(bond);
925 if (best_slave != rtnl_dereference(bond->curr_active_slave)) {
926 bond_change_active_slave(bond, best_slave);
927 rv = bond_set_carrier(bond);
928 if (!rv)
929 return;
931 if (netif_carrier_ok(bond->dev))
932 netdev_info(bond->dev, "first active interface up!\n");
933 else
934 netdev_info(bond->dev, "now running without any active interface!\n");
935 }
936 }
938 #ifdef CONFIG_NET_POLL_CONTROLLER
939 static inline int slave_enable_netpoll(struct slave *slave)
940 {
941 struct netpoll *np;
942 int err = 0;
944 np = kzalloc(sizeof(*np), GFP_KERNEL);
945 err = -ENOMEM;
946 if (!np)
947 goto out;
949 err = __netpoll_setup(np, slave->dev);
950 if (err) {
951 kfree(np);
952 goto out;
953 }
954 slave->np = np;
955 out:
956 return err;
957 }
958 static inline void slave_disable_netpoll(struct slave *slave)
959 {
960 struct netpoll *np = slave->np;
962 if (!np)
963 return;
965 slave->np = NULL;
966 __netpoll_free_async(np);
967 }
969 static void bond_poll_controller(struct net_device *bond_dev)
970 {
971 struct bonding *bond = netdev_priv(bond_dev);
972 struct slave *slave = NULL;
973 struct list_head *iter;
974 struct ad_info ad_info;
976 if (BOND_MODE(bond) == BOND_MODE_8023AD)
977 if (bond_3ad_get_active_agg_info(bond, &ad_info))
978 return;
980 bond_for_each_slave_rcu(bond, slave, iter) {
981 if (!bond_slave_is_up(slave))
982 continue;
984 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
985 struct aggregator *agg =
986 SLAVE_AD_INFO(slave)->port.aggregator;
988 if (agg &&
989 agg->aggregator_identifier != ad_info.aggregator_id)
990 continue;
991 }
993 netpoll_poll_dev(slave->dev);
994 }
995 }
997 static void bond_netpoll_cleanup(struct net_device *bond_dev)
998 {
999 struct bonding *bond = netdev_priv(bond_dev);
1000 struct list_head *iter;
1001 struct slave *slave;
1003 bond_for_each_slave(bond, slave, iter)
1004 if (bond_slave_is_up(slave))
1005 slave_disable_netpoll(slave);
1006 }
1008 static int bond_netpoll_setup(struct net_device *dev, struct netpoll_info *ni)
1009 {
1010 struct bonding *bond = netdev_priv(dev);
1011 struct list_head *iter;
1012 struct slave *slave;
1013 int err = 0;
1015 bond_for_each_slave(bond, slave, iter) {
1016 err = slave_enable_netpoll(slave);
1017 if (err) {
1018 bond_netpoll_cleanup(dev);
1019 break;
1020 }
1021 }
1022 return err;
1023 }
1024 #else
1025 static inline int slave_enable_netpoll(struct slave *slave)
1026 {
1027 return 0;
1028 }
1029 static inline void slave_disable_netpoll(struct slave *slave)
1030 {
1031 }
1032 static void bond_netpoll_cleanup(struct net_device *bond_dev)
1033 {
1034 }
1035 #endif
1037 /*---------------------------------- IOCTL ----------------------------------*/
1039 static netdev_features_t bond_fix_features(struct net_device *dev,
1040 netdev_features_t features)
1041 {
1042 struct bonding *bond = netdev_priv(dev);
1043 struct list_head *iter;
1044 netdev_features_t mask;
1045 struct slave *slave;
1047 mask = features;
1049 features &= ~NETIF_F_ONE_FOR_ALL;
1050 features |= NETIF_F_ALL_FOR_ALL;
1052 bond_for_each_slave(bond, slave, iter) {
1053 features = netdev_increment_features(features,
1054 slave->dev->features,
1055 mask);
1056 }
1057 features = netdev_add_tso_features(features, mask);
1059 return features;
1060 }
1062 #define BOND_VLAN_FEATURES (NETIF_F_HW_CSUM | NETIF_F_SG | \
1063 NETIF_F_FRAGLIST | NETIF_F_ALL_TSO | \
1064 NETIF_F_HIGHDMA | NETIF_F_LRO)
1066 #define BOND_ENC_FEATURES (NETIF_F_HW_CSUM | NETIF_F_SG | \
1067 NETIF_F_RXCSUM | NETIF_F_ALL_TSO)
1069 static void bond_compute_features(struct bonding *bond)
1070 {
1071 unsigned int dst_release_flag = IFF_XMIT_DST_RELEASE |
1072 IFF_XMIT_DST_RELEASE_PERM;
1073 netdev_features_t vlan_features = BOND_VLAN_FEATURES;
1074 netdev_features_t enc_features = BOND_ENC_FEATURES;
1075 struct net_device *bond_dev = bond->dev;
1076 struct list_head *iter;
1077 struct slave *slave;
1078 unsigned short max_hard_header_len = ETH_HLEN;
1079 unsigned int gso_max_size = GSO_MAX_SIZE;
1080 u16 gso_max_segs = GSO_MAX_SEGS;
1082 if (!bond_has_slaves(bond))
1083 goto done;
1084 vlan_features &= NETIF_F_ALL_FOR_ALL;
1086 bond_for_each_slave(bond, slave, iter) {
1087 vlan_features = netdev_increment_features(vlan_features,
1088 slave->dev->vlan_features, BOND_VLAN_FEATURES);
1090 enc_features = netdev_increment_features(enc_features,
1091 slave->dev->hw_enc_features,
1092 BOND_ENC_FEATURES);
1093 dst_release_flag &= slave->dev->priv_flags;
1094 if (slave->dev->hard_header_len > max_hard_header_len)
1095 max_hard_header_len = slave->dev->hard_header_len;
1097 gso_max_size = min(gso_max_size, slave->dev->gso_max_size);
1098 gso_max_segs = min(gso_max_segs, slave->dev->gso_max_segs);
1099 }
1100 bond_dev->hard_header_len = max_hard_header_len;
1102 done:
1103 bond_dev->vlan_features = vlan_features;
1104 bond_dev->hw_enc_features = enc_features | NETIF_F_GSO_ENCAP_ALL |
1105 NETIF_F_GSO_UDP_L4;
1106 bond_dev->gso_max_segs = gso_max_segs;
1107 netif_set_gso_max_size(bond_dev, gso_max_size);
1109 bond_dev->priv_flags &= ~IFF_XMIT_DST_RELEASE;
1110 if ((bond_dev->priv_flags & IFF_XMIT_DST_RELEASE_PERM) &&
1111 dst_release_flag == (IFF_XMIT_DST_RELEASE | IFF_XMIT_DST_RELEASE_PERM))
1112 bond_dev->priv_flags |= IFF_XMIT_DST_RELEASE;
1114 netdev_change_features(bond_dev);
1115 }
1117 static void bond_setup_by_slave(struct net_device *bond_dev,
1118 struct net_device *slave_dev)
1119 {
1120 bond_dev->header_ops = slave_dev->header_ops;
1122 bond_dev->type = slave_dev->type;
1123 bond_dev->hard_header_len = slave_dev->hard_header_len;
1124 bond_dev->addr_len = slave_dev->addr_len;
1126 memcpy(bond_dev->broadcast, slave_dev->broadcast,
1127 slave_dev->addr_len);
1128 }
1130 /* On bonding slaves other than the currently active slave, suppress
1131 * duplicates except for alb non-mcast/bcast.
1132 */
1133 static bool bond_should_deliver_exact_match(struct sk_buff *skb,
1134 struct slave *slave,
1135 struct bonding *bond)
1136 {
1137 if (bond_is_slave_inactive(slave)) {
1138 if (BOND_MODE(bond) == BOND_MODE_ALB &&
1139 skb->pkt_type != PACKET_BROADCAST &&
1140 skb->pkt_type != PACKET_MULTICAST)
1141 return false;
1142 return true;
1143 }
1144 return false;
1145 }
1147 static rx_handler_result_t bond_handle_frame(struct sk_buff **pskb)
1148 {
1149 struct sk_buff *skb = *pskb;
1150 struct slave *slave;
1151 struct bonding *bond;
1152 int (*recv_probe)(const struct sk_buff *, struct bonding *,
1153 struct slave *);
1154 int ret = RX_HANDLER_ANOTHER;
1156 skb = skb_share_check(skb, GFP_ATOMIC);
1157 if (unlikely(!skb))
1158 return RX_HANDLER_CONSUMED;
1160 *pskb = skb;
1162 slave = bond_slave_get_rcu(skb->dev);
1163 bond = slave->bond;
1165 recv_probe = READ_ONCE(bond->recv_probe);
1166 if (recv_probe) {
1167 ret = recv_probe(skb, bond, slave);
1168 if (ret == RX_HANDLER_CONSUMED) {
1169 consume_skb(skb);
1170 return ret;
1171 }
1172 }
1174 /* Link-local multicast packets should be passed to the
1175 * stack on the link they arrive as well as pass them to the
1176 * bond-master device. These packets are mostly usable when
1177 * stack receives it with the link on which they arrive
1178 * (e.g. LLDP) they also must be available on master. Some of
1179 * the use cases include (but are not limited to): LLDP agents
1180 * that must be able to operate both on enslaved interfaces as
1181 * well as on bonds themselves; linux bridges that must be able
1182 * to process/pass BPDUs from attached bonds when any kind of
1183 * STP version is enabled on the network.
1184 */
1185 if (is_link_local_ether_addr(eth_hdr(skb)->h_dest)) {
1186 struct sk_buff *nskb = skb_clone(skb, GFP_ATOMIC);
1188 if (nskb) {
1189 nskb->dev = bond->dev;
1190 nskb->queue_mapping = 0;
1191 netif_rx(nskb);
1192 }
1193 return RX_HANDLER_PASS;
1194 }
1195 if (bond_should_deliver_exact_match(skb, slave, bond))
1196 return RX_HANDLER_EXACT;
1198 skb->dev = bond->dev;
1200 if (BOND_MODE(bond) == BOND_MODE_ALB &&
1201 bond->dev->priv_flags & IFF_BRIDGE_PORT &&
1202 skb->pkt_type == PACKET_HOST) {
1204 if (unlikely(skb_cow_head(skb,
1205 skb->data - skb_mac_header(skb)))) {
1206 kfree_skb(skb);
1207 return RX_HANDLER_CONSUMED;
1208 }
1209 bond_hw_addr_copy(eth_hdr(skb)->h_dest, bond->dev->dev_addr,
1210 bond->dev->addr_len);
1211 }
1213 return ret;
1214 }
1216 static enum netdev_lag_tx_type bond_lag_tx_type(struct bonding *bond)
1217 {
1218 switch (BOND_MODE(bond)) {
1219 case BOND_MODE_ROUNDROBIN:
1220 return NETDEV_LAG_TX_TYPE_ROUNDROBIN;
1221 case BOND_MODE_ACTIVEBACKUP:
1222 return NETDEV_LAG_TX_TYPE_ACTIVEBACKUP;
1223 case BOND_MODE_BROADCAST:
1224 return NETDEV_LAG_TX_TYPE_BROADCAST;
1225 case BOND_MODE_XOR:
1226 case BOND_MODE_8023AD:
1227 return NETDEV_LAG_TX_TYPE_HASH;
1228 default:
1229 return NETDEV_LAG_TX_TYPE_UNKNOWN;
1230 }
1231 }
1233 static enum netdev_lag_hash bond_lag_hash_type(struct bonding *bond,
1234 enum netdev_lag_tx_type type)
1235 {
1236 if (type != NETDEV_LAG_TX_TYPE_HASH)
1237 return NETDEV_LAG_HASH_NONE;
1239 switch (bond->params.xmit_policy) {
1240 case BOND_XMIT_POLICY_LAYER2:
1241 return NETDEV_LAG_HASH_L2;
1242 case BOND_XMIT_POLICY_LAYER34:
1243 return NETDEV_LAG_HASH_L34;
1244 case BOND_XMIT_POLICY_LAYER23:
1245 return NETDEV_LAG_HASH_L23;
1246 case BOND_XMIT_POLICY_ENCAP23:
1247 return NETDEV_LAG_HASH_E23;
1248 case BOND_XMIT_POLICY_ENCAP34:
1249 return NETDEV_LAG_HASH_E34;
1250 default:
1251 return NETDEV_LAG_HASH_UNKNOWN;
1252 }
1253 }
1255 static int bond_master_upper_dev_link(struct bonding *bond, struct slave *slave,
1256 struct netlink_ext_ack *extack)
1257 {
1258 struct netdev_lag_upper_info lag_upper_info;
1259 enum netdev_lag_tx_type type;
1261 type = bond_lag_tx_type(bond);
1262 lag_upper_info.tx_type = type;
1263 lag_upper_info.hash_type = bond_lag_hash_type(bond, type);
1265 return netdev_master_upper_dev_link(slave->dev, bond->dev, slave,
1266 &lag_upper_info, extack);
1267 }
1269 static void bond_upper_dev_unlink(struct bonding *bond, struct slave *slave)
1270 {
1271 netdev_upper_dev_unlink(slave->dev, bond->dev);
1272 slave->dev->flags &= ~IFF_SLAVE;
1273 }
1275 static struct slave *bond_alloc_slave(struct bonding *bond)
1276 {
1277 struct slave *slave = NULL;
1279 slave = kzalloc(sizeof(*slave), GFP_KERNEL);
1280 if (!slave)
1281 return NULL;
1283 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
1284 SLAVE_AD_INFO(slave) = kzalloc(sizeof(struct ad_slave_info),
1285 GFP_KERNEL);
1286 if (!SLAVE_AD_INFO(slave)) {
1287 kfree(slave);
1288 return NULL;
1289 }
1290 }
1291 INIT_DELAYED_WORK(&slave->notify_work, bond_netdev_notify_work);
1293 return slave;
1294 }
1296 static void bond_free_slave(struct slave *slave)
1297 {
1298 struct bonding *bond = bond_get_bond_by_slave(slave);
1300 cancel_delayed_work_sync(&slave->notify_work);
1301 if (BOND_MODE(bond) == BOND_MODE_8023AD)
1302 kfree(SLAVE_AD_INFO(slave));
1304 kfree(slave);
1305 }
1307 static void bond_fill_ifbond(struct bonding *bond, struct ifbond *info)
1308 {
1309 info->bond_mode = BOND_MODE(bond);
1310 info->miimon = bond->params.miimon;
1311 info->num_slaves = bond->slave_cnt;
1312 }
1314 static void bond_fill_ifslave(struct slave *slave, struct ifslave *info)
1315 {
1316 strcpy(info->slave_name, slave->dev->name);
1317 info->link = slave->link;
1318 info->state = bond_slave_state(slave);
1319 info->link_failure_count = slave->link_failure_count;
1320 }
1322 static void bond_netdev_notify_work(struct work_struct *_work)
1323 {
1324 struct slave *slave = container_of(_work, struct slave,
1325 notify_work.work);
1327 if (rtnl_trylock()) {
1328 struct netdev_bonding_info binfo;
1330 bond_fill_ifslave(slave, &binfo.slave);
1331 bond_fill_ifbond(slave->bond, &binfo.master);
1332 netdev_bonding_info_change(slave->dev, &binfo);
1333 rtnl_unlock();
1334 } else {
1335 queue_delayed_work(slave->bond->wq, &slave->notify_work, 1);
1336 }
1337 }
1339 void bond_queue_slave_event(struct slave *slave)
1340 {
1341 queue_delayed_work(slave->bond->wq, &slave->notify_work, 0);
1342 }
1344 void bond_lower_state_changed(struct slave *slave)
1345 {
1346 struct netdev_lag_lower_state_info info;
1348 info.link_up = slave->link == BOND_LINK_UP ||
1349 slave->link == BOND_LINK_FAIL;
1350 info.tx_enabled = bond_is_active_slave(slave);
1351 netdev_lower_state_changed(slave->dev, &info);
1352 }
1354 /* enslave device <slave> to bond device <master> */
1355 int bond_enslave(struct net_device *bond_dev, struct net_device *slave_dev,
1356 struct netlink_ext_ack *extack)
1357 {
1358 struct bonding *bond = netdev_priv(bond_dev);
1359 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
1360 struct slave *new_slave = NULL, *prev_slave;
1361 struct sockaddr_storage ss;
1362 int link_reporting;
1363 int res = 0, i;
1365 if (!bond->params.use_carrier &&
1366 slave_dev->ethtool_ops->get_link == NULL &&
1367 slave_ops->ndo_do_ioctl == NULL) {
1368 netdev_warn(bond_dev, "no link monitoring support for %s\n",
1369 slave_dev->name);
1370 }
1372 /* already in-use? */
1373 if (netdev_is_rx_handler_busy(slave_dev)) {
1374 NL_SET_ERR_MSG(extack, "Device is in use and cannot be enslaved");
1375 netdev_err(bond_dev,
1376 "Error: Device is in use and cannot be enslaved\n");
1377 return -EBUSY;
1378 }
1380 if (bond_dev == slave_dev) {
1381 NL_SET_ERR_MSG(extack, "Cannot enslave bond to itself.");
1382 netdev_err(bond_dev, "cannot enslave bond to itself.\n");
1383 return -EPERM;
1384 }
1386 /* vlan challenged mutual exclusion */
1387 /* no need to lock since we're protected by rtnl_lock */
1388 if (slave_dev->features & NETIF_F_VLAN_CHALLENGED) {
1389 netdev_dbg(bond_dev, "%s is NETIF_F_VLAN_CHALLENGED\n",
1390 slave_dev->name);
1391 if (vlan_uses_dev(bond_dev)) {
1392 NL_SET_ERR_MSG(extack, "Can not enslave VLAN challenged device to VLAN enabled bond");
1393 netdev_err(bond_dev, "Error: cannot enslave VLAN challenged slave %s on VLAN enabled bond %s\n",
1394 slave_dev->name, bond_dev->name);
1395 return -EPERM;
1396 } else {
1397 netdev_warn(bond_dev, "enslaved VLAN challenged slave %s. Adding VLANs will be blocked as long as %s is part of bond %s\n",
1398 slave_dev->name, slave_dev->name,
1399 bond_dev->name);
1400 }
1401 } else {
1402 netdev_dbg(bond_dev, "%s is !NETIF_F_VLAN_CHALLENGED\n",
1403 slave_dev->name);
1404 }
1406 /* Old ifenslave binaries are no longer supported. These can
1407 * be identified with moderate accuracy by the state of the slave:
1408 * the current ifenslave will set the interface down prior to
1409 * enslaving it; the old ifenslave will not.
1410 */
1411 if (slave_dev->flags & IFF_UP) {
1412 NL_SET_ERR_MSG(extack, "Device can not be enslaved while up");
1413 netdev_err(bond_dev, "%s is up - this may be due to an out of date ifenslave\n",
1414 slave_dev->name);
1415 return -EPERM;
1416 }
1418 /* set bonding device ether type by slave - bonding netdevices are
1419 * created with ether_setup, so when the slave type is not ARPHRD_ETHER
1420 * there is a need to override some of the type dependent attribs/funcs.
1421 *
1422 * bond ether type mutual exclusion - don't allow slaves of dissimilar
1423 * ether type (eg ARPHRD_ETHER and ARPHRD_INFINIBAND) share the same bond
1424 */
1425 if (!bond_has_slaves(bond)) {
1426 if (bond_dev->type != slave_dev->type) {
1427 netdev_dbg(bond_dev, "change device type from %d to %d\n",
1428 bond_dev->type, slave_dev->type);
1430 res = call_netdevice_notifiers(NETDEV_PRE_TYPE_CHANGE,
1431 bond_dev);
1432 res = notifier_to_errno(res);
1433 if (res) {
1434 netdev_err(bond_dev, "refused to change device type\n");
1435 return -EBUSY;
1436 }
1438 /* Flush unicast and multicast addresses */
1439 dev_uc_flush(bond_dev);
1440 dev_mc_flush(bond_dev);
1442 if (slave_dev->type != ARPHRD_ETHER)
1443 bond_setup_by_slave(bond_dev, slave_dev);
1444 else {
1445 ether_setup(bond_dev);
1446 bond_dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1447 }
1449 call_netdevice_notifiers(NETDEV_POST_TYPE_CHANGE,
1450 bond_dev);
1451 }
1452 } else if (bond_dev->type != slave_dev->type) {
1453 NL_SET_ERR_MSG(extack, "Device type is different from other slaves");
1454 netdev_err(bond_dev, "%s ether type (%d) is different from other slaves (%d), can not enslave it\n",
1455 slave_dev->name, slave_dev->type, bond_dev->type);
1456 return -EINVAL;
1457 }
1459 if (slave_dev->type == ARPHRD_INFINIBAND &&
1460 BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
1461 NL_SET_ERR_MSG(extack, "Only active-backup mode is supported for infiniband slaves");
1462 netdev_warn(bond_dev, "Type (%d) supports only active-backup mode\n",
1463 slave_dev->type);
1464 res = -EOPNOTSUPP;
1465 goto err_undo_flags;
1466 }
1468 if (!slave_ops->ndo_set_mac_address ||
1469 slave_dev->type == ARPHRD_INFINIBAND) {
1470 netdev_warn(bond_dev, "The slave device specified does not support setting the MAC address\n");
1471 if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP &&
1472 bond->params.fail_over_mac != BOND_FOM_ACTIVE) {
1473 if (!bond_has_slaves(bond)) {
1474 bond->params.fail_over_mac = BOND_FOM_ACTIVE;
1475 netdev_warn(bond_dev, "Setting fail_over_mac to active for active-backup mode\n");
1476 } else {
1477 NL_SET_ERR_MSG(extack, "Slave device does not support setting the MAC address, but fail_over_mac is not set to active");
1478 netdev_err(bond_dev, "The slave device specified does not support setting the MAC address, but fail_over_mac is not set to active\n");
1479 res = -EOPNOTSUPP;
1480 goto err_undo_flags;
1481 }
1482 }
1483 }
1485 call_netdevice_notifiers(NETDEV_JOIN, slave_dev);
1487 /* If this is the first slave, then we need to set the master's hardware
1488 * address to be the same as the slave's.
1489 */
1490 if (!bond_has_slaves(bond) &&
1491 bond->dev->addr_assign_type == NET_ADDR_RANDOM)
1492 bond_set_dev_addr(bond->dev, slave_dev);
1494 new_slave = bond_alloc_slave(bond);
1495 if (!new_slave) {
1496 res = -ENOMEM;
1497 goto err_undo_flags;
1498 }
1500 new_slave->bond = bond;
1501 new_slave->dev = slave_dev;
1502 /* Set the new_slave's queue_id to be zero. Queue ID mapping
1503 * is set via sysfs or module option if desired.
1504 */
1505 new_slave->queue_id = 0;
1507 /* Save slave's original mtu and then set it to match the bond */
1508 new_slave->original_mtu = slave_dev->mtu;
1509 res = dev_set_mtu(slave_dev, bond->dev->mtu);
1510 if (res) {
1511 netdev_dbg(bond_dev, "Error %d calling dev_set_mtu\n", res);
1512 goto err_free;
1513 }
1515 /* Save slave's original ("permanent") mac address for modes
1516 * that need it, and for restoring it upon release, and then
1517 * set it to the master's address
1518 */
1519 bond_hw_addr_copy(new_slave->perm_hwaddr, slave_dev->dev_addr,
1520 slave_dev->addr_len);
1522 if (!bond->params.fail_over_mac ||
1523 BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
1524 /* Set slave to master's mac address. The application already
1525 * set the master's mac address to that of the first slave
1526 */
1527 memcpy(ss.__data, bond_dev->dev_addr, bond_dev->addr_len);
1528 ss.ss_family = slave_dev->type;
1529 res = dev_set_mac_address(slave_dev, (struct sockaddr *)&ss);
1530 if (res) {
1531 netdev_dbg(bond_dev, "Error %d calling set_mac_address\n", res);
1532 goto err_restore_mtu;
1533 }
1534 }
1536 /* set slave flag before open to prevent IPv6 addrconf */
1537 slave_dev->flags |= IFF_SLAVE;
1539 /* open the slave since the application closed it */
1540 res = dev_open(slave_dev);
1541 if (res) {
1542 netdev_dbg(bond_dev, "Opening slave %s failed\n", slave_dev->name);
1543 goto err_restore_mac;
1544 }
1546 slave_dev->priv_flags |= IFF_BONDING;
1547 /* initialize slave stats */
1548 dev_get_stats(new_slave->dev, &new_slave->slave_stats);
1550 if (bond_is_lb(bond)) {
1551 /* bond_alb_init_slave() must be called before all other stages since
1552 * it might fail and we do not want to have to undo everything
1553 */
1554 res = bond_alb_init_slave(bond, new_slave);
1555 if (res)
1556 goto err_close;
1557 }
1559 res = vlan_vids_add_by_dev(slave_dev, bond_dev);
1560 if (res) {
1561 netdev_err(bond_dev, "Couldn't add bond vlan ids to %s\n",
1562 slave_dev->name);
1563 goto err_close;
1564 }
1566 prev_slave = bond_last_slave(bond);
1568 new_slave->delay = 0;
1569 new_slave->link_failure_count = 0;
1571 if (bond_update_speed_duplex(new_slave) &&
1572 bond_needs_speed_duplex(bond))
1573 new_slave->link = BOND_LINK_DOWN;
1575 new_slave->last_rx = jiffies -
1576 (msecs_to_jiffies(bond->params.arp_interval) + 1);
1577 for (i = 0; i < BOND_MAX_ARP_TARGETS; i++)
1578 new_slave->target_last_arp_rx[i] = new_slave->last_rx;
1580 if (bond->params.miimon && !bond->params.use_carrier) {
1581 link_reporting = bond_check_dev_link(bond, slave_dev, 1);
1583 if ((link_reporting == -1) && !bond->params.arp_interval) {
1584 /* miimon is set but a bonded network driver
1585 * does not support ETHTOOL/MII and
1586 * arp_interval is not set. Note: if
1587 * use_carrier is enabled, we will never go
1588 * here (because netif_carrier is always
1589 * supported); thus, we don't need to change
1590 * the messages for netif_carrier.
1591 */
1592 netdev_warn(bond_dev, "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",
1593 slave_dev->name);
1594 } else if (link_reporting == -1) {
1595 /* unable get link status using mii/ethtool */
1596 netdev_warn(bond_dev, "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",
1597 slave_dev->name);
1598 }
1599 }
1601 /* check for initial state */
1602 new_slave->link = BOND_LINK_NOCHANGE;
1603 if (bond->params.miimon) {
1604 if (bond_check_dev_link(bond, slave_dev, 0) == BMSR_LSTATUS) {
1605 if (bond->params.updelay) {
1606 bond_set_slave_link_state(new_slave,
1607 BOND_LINK_BACK,
1608 BOND_SLAVE_NOTIFY_NOW);
1609 new_slave->delay = bond->params.updelay;
1610 } else {
1611 bond_set_slave_link_state(new_slave,
1612 BOND_LINK_UP,
1613 BOND_SLAVE_NOTIFY_NOW);
1614 }
1615 } else {
1616 bond_set_slave_link_state(new_slave, BOND_LINK_DOWN,
1617 BOND_SLAVE_NOTIFY_NOW);
1618 }
1619 } else if (bond->params.arp_interval) {
1620 bond_set_slave_link_state(new_slave,
1621 (netif_carrier_ok(slave_dev) ?
1622 BOND_LINK_UP : BOND_LINK_DOWN),
1623 BOND_SLAVE_NOTIFY_NOW);
1624 } else {
1625 bond_set_slave_link_state(new_slave, BOND_LINK_UP,
1626 BOND_SLAVE_NOTIFY_NOW);
1627 }
1629 if (new_slave->link != BOND_LINK_DOWN)
1630 new_slave->last_link_up = jiffies;
1631 netdev_dbg(bond_dev, "Initial state of slave_dev is BOND_LINK_%s\n",
1632 new_slave->link == BOND_LINK_DOWN ? "DOWN" :
1633 (new_slave->link == BOND_LINK_UP ? "UP" : "BACK"));
1635 if (bond_uses_primary(bond) && bond->params.primary[0]) {
1636 /* if there is a primary slave, remember it */
1637 if (strcmp(bond->params.primary, new_slave->dev->name) == 0) {
1638 rcu_assign_pointer(bond->primary_slave, new_slave);
1639 bond->force_primary = true;
1640 }
1641 }
1643 switch (BOND_MODE(bond)) {
1644 case BOND_MODE_ACTIVEBACKUP:
1645 bond_set_slave_inactive_flags(new_slave,
1646 BOND_SLAVE_NOTIFY_NOW);
1647 break;
1648 case BOND_MODE_8023AD:
1649 /* in 802.3ad mode, the internal mechanism
1650 * will activate the slaves in the selected
1651 * aggregator
1652 */
1653 bond_set_slave_inactive_flags(new_slave, BOND_SLAVE_NOTIFY_NOW);
1654 /* if this is the first slave */
1655 if (!prev_slave) {
1656 SLAVE_AD_INFO(new_slave)->id = 1;
1657 /* Initialize AD with the number of times that the AD timer is called in 1 second
1658 * can be called only after the mac address of the bond is set
1659 */
1660 bond_3ad_initialize(bond, 1000/AD_TIMER_INTERVAL);
1661 } else {
1662 SLAVE_AD_INFO(new_slave)->id =
1663 SLAVE_AD_INFO(prev_slave)->id + 1;
1664 }
1666 bond_3ad_bind_slave(new_slave);
1667 break;
1668 case BOND_MODE_TLB:
1669 case BOND_MODE_ALB:
1670 bond_set_active_slave(new_slave);
1671 bond_set_slave_inactive_flags(new_slave, BOND_SLAVE_NOTIFY_NOW);
1672 break;
1673 default:
1674 netdev_dbg(bond_dev, "This slave is always active in trunk mode\n");
1676 /* always active in trunk mode */
1677 bond_set_active_slave(new_slave);
1679 /* In trunking mode there is little meaning to curr_active_slave
1680 * anyway (it holds no special properties of the bond device),
1681 * so we can change it without calling change_active_interface()
1682 */
1683 if (!rcu_access_pointer(bond->curr_active_slave) &&
1684 new_slave->link == BOND_LINK_UP)
1685 rcu_assign_pointer(bond->curr_active_slave, new_slave);
1687 break;
1688 } /* switch(bond_mode) */
1690 #ifdef CONFIG_NET_POLL_CONTROLLER
1691 if (bond->dev->npinfo) {
1692 if (slave_enable_netpoll(new_slave)) {
1693 netdev_info(bond_dev, "master_dev is using netpoll, but new slave device does not support netpoll\n");
1694 res = -EBUSY;
1695 goto err_detach;
1696 }
1697 }
1698 #endif
1700 if (!(bond_dev->features & NETIF_F_LRO))
1701 dev_disable_lro(slave_dev);
1703 res = netdev_rx_handler_register(slave_dev, bond_handle_frame,
1704 new_slave);
1705 if (res) {
1706 netdev_dbg(bond_dev, "Error %d calling netdev_rx_handler_register\n", res);
1707 goto err_detach;
1708 }
1710 res = bond_master_upper_dev_link(bond, new_slave, extack);
1711 if (res) {
1712 netdev_dbg(bond_dev, "Error %d calling bond_master_upper_dev_link\n", res);
1713 goto err_unregister;
1714 }
1716 res = bond_sysfs_slave_add(new_slave);
1717 if (res) {
1718 netdev_dbg(bond_dev, "Error %d calling bond_sysfs_slave_add\n", res);
1719 goto err_upper_unlink;
1720 }
1722 bond->nest_level = dev_get_nest_level(bond_dev) + 1;
1724 /* If the mode uses primary, then the following is handled by
1725 * bond_change_active_slave().
1726 */
1727 if (!bond_uses_primary(bond)) {
1728 /* set promiscuity level to new slave */
1729 if (bond_dev->flags & IFF_PROMISC) {
1730 res = dev_set_promiscuity(slave_dev, 1);
1731 if (res)
1732 goto err_sysfs_del;
1733 }
1735 /* set allmulti level to new slave */
1736 if (bond_dev->flags & IFF_ALLMULTI) {
1737 res = dev_set_allmulti(slave_dev, 1);
1738 if (res) {
1739 if (bond_dev->flags & IFF_PROMISC)
1740 dev_set_promiscuity(slave_dev, -1);
1741 goto err_sysfs_del;
1742 }
1743 }
1745 netif_addr_lock_bh(bond_dev);
1746 dev_mc_sync_multiple(slave_dev, bond_dev);
1747 dev_uc_sync_multiple(slave_dev, bond_dev);
1748 netif_addr_unlock_bh(bond_dev);
1750 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
1751 /* add lacpdu mc addr to mc list */
1752 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
1754 dev_mc_add(slave_dev, lacpdu_multicast);
1755 }
1756 }
1758 bond->slave_cnt++;
1759 bond_compute_features(bond);
1760 bond_set_carrier(bond);
1762 if (bond_uses_primary(bond)) {
1763 block_netpoll_tx();
1764 bond_select_active_slave(bond);
1765 unblock_netpoll_tx();
1766 }
1768 if (bond_mode_can_use_xmit_hash(bond))
1769 bond_update_slave_arr(bond, NULL);
1772 netdev_info(bond_dev, "Enslaving %s as %s interface with %s link\n",
1773 slave_dev->name,
1774 bond_is_active_slave(new_slave) ? "an active" : "a backup",
1775 new_slave->link != BOND_LINK_DOWN ? "an up" : "a down");
1777 /* enslave is successful */
1778 bond_queue_slave_event(new_slave);
1779 return 0;
1781 /* Undo stages on error */
1782 err_sysfs_del:
1783 bond_sysfs_slave_del(new_slave);
1785 err_upper_unlink:
1786 bond_upper_dev_unlink(bond, new_slave);
1788 err_unregister:
1789 netdev_rx_handler_unregister(slave_dev);
1791 err_detach:
1792 vlan_vids_del_by_dev(slave_dev, bond_dev);
1793 if (rcu_access_pointer(bond->primary_slave) == new_slave)
1794 RCU_INIT_POINTER(bond->primary_slave, NULL);
1795 if (rcu_access_pointer(bond->curr_active_slave) == new_slave) {
1796 block_netpoll_tx();
1797 bond_change_active_slave(bond, NULL);
1798 bond_select_active_slave(bond);
1799 unblock_netpoll_tx();
1800 }
1801 /* either primary_slave or curr_active_slave might've changed */
1802 synchronize_rcu();
1803 slave_disable_netpoll(new_slave);
1805 err_close:
1806 slave_dev->priv_flags &= ~IFF_BONDING;
1807 dev_close(slave_dev);
1809 err_restore_mac:
1810 slave_dev->flags &= ~IFF_SLAVE;
1811 if (!bond->params.fail_over_mac ||
1812 BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
1813 /* XXX TODO - fom follow mode needs to change master's
1814 * MAC if this slave's MAC is in use by the bond, or at
1815 * least print a warning.
1816 */
1817 bond_hw_addr_copy(ss.__data, new_slave->perm_hwaddr,
1818 new_slave->dev->addr_len);
1819 ss.ss_family = slave_dev->type;
1820 dev_set_mac_address(slave_dev, (struct sockaddr *)&ss);
1821 }
1823 err_restore_mtu:
1824 dev_set_mtu(slave_dev, new_slave->original_mtu);
1826 err_free:
1827 bond_free_slave(new_slave);
1829 err_undo_flags:
1830 /* Enslave of first slave has failed and we need to fix master's mac */
1831 if (!bond_has_slaves(bond)) {
1832 if (ether_addr_equal_64bits(bond_dev->dev_addr,
1833 slave_dev->dev_addr))
1834 eth_hw_addr_random(bond_dev);
1835 if (bond_dev->type != ARPHRD_ETHER) {
1836 dev_close(bond_dev);
1837 ether_setup(bond_dev);
1838 bond_dev->flags |= IFF_MASTER;
1839 bond_dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1840 }
1841 }
1843 return res;
1844 }
1846 /* Try to release the slave device <slave> from the bond device <master>
1847 * It is legal to access curr_active_slave without a lock because all the function
1848 * is RTNL-locked. If "all" is true it means that the function is being called
1849 * while destroying a bond interface and all slaves are being released.
1850 *
1851 * The rules for slave state should be:
1852 * for Active/Backup:
1853 * Active stays on all backups go down
1854 * for Bonded connections:
1855 * The first up interface should be left on and all others downed.
1856 */
1857 static int __bond_release_one(struct net_device *bond_dev,
1858 struct net_device *slave_dev,
1859 bool all, bool unregister)
1860 {
1861 struct bonding *bond = netdev_priv(bond_dev);
1862 struct slave *slave, *oldcurrent;
1863 struct sockaddr_storage ss;
1864 int old_flags = bond_dev->flags;
1865 netdev_features_t old_features = bond_dev->features;
1867 /* slave is not a slave or master is not master of this slave */
1868 if (!(slave_dev->flags & IFF_SLAVE) ||
1869 !netdev_has_upper_dev(slave_dev, bond_dev)) {
1870 netdev_dbg(bond_dev, "cannot release %s\n",
1871 slave_dev->name);
1872 return -EINVAL;
1873 }
1875 block_netpoll_tx();
1877 slave = bond_get_slave_by_dev(bond, slave_dev);
1878 if (!slave) {
1879 /* not a slave of this bond */
1880 netdev_info(bond_dev, "%s not enslaved\n",
1881 slave_dev->name);
1882 unblock_netpoll_tx();
1883 return -EINVAL;
1884 }
1886 bond_set_slave_inactive_flags(slave, BOND_SLAVE_NOTIFY_NOW);
1888 bond_sysfs_slave_del(slave);
1890 /* recompute stats just before removing the slave */
1891 bond_get_stats(bond->dev, &bond->bond_stats);
1893 bond_upper_dev_unlink(bond, slave);
1894 /* unregister rx_handler early so bond_handle_frame wouldn't be called
1895 * for this slave anymore.
1896 */
1897 netdev_rx_handler_unregister(slave_dev);
1899 if (BOND_MODE(bond) == BOND_MODE_8023AD)
1900 bond_3ad_unbind_slave(slave);
1902 if (bond_mode_can_use_xmit_hash(bond))
1903 bond_update_slave_arr(bond, slave);
1905 netdev_info(bond_dev, "Releasing %s interface %s\n",
1906 bond_is_active_slave(slave) ? "active" : "backup",
1907 slave_dev->name);
1909 oldcurrent = rcu_access_pointer(bond->curr_active_slave);
1911 RCU_INIT_POINTER(bond->current_arp_slave, NULL);
1913 if (!all && (!bond->params.fail_over_mac ||
1914 BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP)) {
1915 if (ether_addr_equal_64bits(bond_dev->dev_addr, slave->perm_hwaddr) &&
1916 bond_has_slaves(bond))
1917 netdev_warn(bond_dev, "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",
1918 slave_dev->name, slave->perm_hwaddr,
1919 bond_dev->name, slave_dev->name);
1920 }
1922 if (rtnl_dereference(bond->primary_slave) == slave)
1923 RCU_INIT_POINTER(bond->primary_slave, NULL);
1925 if (oldcurrent == slave)
1926 bond_change_active_slave(bond, NULL);
1928 if (bond_is_lb(bond)) {
1929 /* Must be called only after the slave has been
1930 * detached from the list and the curr_active_slave
1931 * has been cleared (if our_slave == old_current),
1932 * but before a new active slave is selected.
1933 */
1934 bond_alb_deinit_slave(bond, slave);
1935 }
1937 if (all) {
1938 RCU_INIT_POINTER(bond->curr_active_slave, NULL);
1939 } else if (oldcurrent == slave) {
1940 /* Note that we hold RTNL over this sequence, so there
1941 * is no concern that another slave add/remove event
1942 * will interfere.
1943 */
1944 bond_select_active_slave(bond);
1945 }
1947 if (!bond_has_slaves(bond)) {
1948 bond_set_carrier(bond);
1949 eth_hw_addr_random(bond_dev);
1950 }
1952 unblock_netpoll_tx();
1953 synchronize_rcu();
1954 bond->slave_cnt--;
1956 if (!bond_has_slaves(bond)) {
1957 call_netdevice_notifiers(NETDEV_CHANGEADDR, bond->dev);
1958 call_netdevice_notifiers(NETDEV_RELEASE, bond->dev);
1959 }
1961 bond_compute_features(bond);
1962 if (!(bond_dev->features & NETIF_F_VLAN_CHALLENGED) &&
1963 (old_features & NETIF_F_VLAN_CHALLENGED))
1964 netdev_info(bond_dev, "last VLAN challenged slave %s left bond %s - VLAN blocking is removed\n",
1965 slave_dev->name, bond_dev->name);
1967 vlan_vids_del_by_dev(slave_dev, bond_dev);
1969 /* If the mode uses primary, then this case was handled above by
1970 * bond_change_active_slave(..., NULL)
1971 */
1972 if (!bond_uses_primary(bond)) {
1973 /* unset promiscuity level from slave
1974 * NOTE: The NETDEV_CHANGEADDR call above may change the value
1975 * of the IFF_PROMISC flag in the bond_dev, but we need the
1976 * value of that flag before that change, as that was the value
1977 * when this slave was attached, so we cache at the start of the
1978 * function and use it here. Same goes for ALLMULTI below
1979 */
1980 if (old_flags & IFF_PROMISC)
1981 dev_set_promiscuity(slave_dev, -1);
1983 /* unset allmulti level from slave */
1984 if (old_flags & IFF_ALLMULTI)
1985 dev_set_allmulti(slave_dev, -1);
1987 bond_hw_addr_flush(bond_dev, slave_dev);
1988 }
1990 slave_disable_netpoll(slave);
1992 /* close slave before restoring its mac address */
1993 dev_close(slave_dev);
1995 if (bond->params.fail_over_mac != BOND_FOM_ACTIVE ||
1996 BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
1997 /* restore original ("permanent") mac address */
1998 bond_hw_addr_copy(ss.__data, slave->perm_hwaddr,
1999 slave->dev->addr_len);
2000 ss.ss_family = slave_dev->type;
2001 dev_set_mac_address(slave_dev, (struct sockaddr *)&ss);
2002 }
2004 if (unregister)
2005 __dev_set_mtu(slave_dev, slave->original_mtu);
2006 else
2007 dev_set_mtu(slave_dev, slave->original_mtu);
2009 slave_dev->priv_flags &= ~IFF_BONDING;
2011 bond_free_slave(slave);
2013 return 0;
2014 }
2016 /* A wrapper used because of ndo_del_link */
2017 int bond_release(struct net_device *bond_dev, struct net_device *slave_dev)
2018 {
2019 return __bond_release_one(bond_dev, slave_dev, false, false);
2020 }
2022 /* First release a slave and then destroy the bond if no more slaves are left.
2023 * Must be under rtnl_lock when this function is called.
2024 */
2025 static int bond_release_and_destroy(struct net_device *bond_dev,
2026 struct net_device *slave_dev)
2027 {
2028 struct bonding *bond = netdev_priv(bond_dev);
2029 int ret;
2031 ret = __bond_release_one(bond_dev, slave_dev, false, true);
2032 if (ret == 0 && !bond_has_slaves(bond)) {
2033 bond_dev->priv_flags |= IFF_DISABLE_NETPOLL;
2034 netdev_info(bond_dev, "Destroying bond %s\n",
2035 bond_dev->name);
2036 bond_remove_proc_entry(bond);
2037 unregister_netdevice(bond_dev);
2038 }
2039 return ret;
2040 }
2042 static void bond_info_query(struct net_device *bond_dev, struct ifbond *info)
2043 {
2044 struct bonding *bond = netdev_priv(bond_dev);
2045 bond_fill_ifbond(bond, info);
2046 }
2048 static int bond_slave_info_query(struct net_device *bond_dev, struct ifslave *info)
2049 {
2050 struct bonding *bond = netdev_priv(bond_dev);
2051 struct list_head *iter;
2052 int i = 0, res = -ENODEV;
2053 struct slave *slave;
2055 bond_for_each_slave(bond, slave, iter) {
2056 if (i++ == (int)info->slave_id) {
2057 res = 0;
2058 bond_fill_ifslave(slave, info);
2059 break;
2060 }
2061 }
2063 return res;
2064 }
2066 /*-------------------------------- Monitoring -------------------------------*/
2068 /* called with rcu_read_lock() */
2069 static int bond_miimon_inspect(struct bonding *bond)
2070 {
2071 int link_state, commit = 0;
2072 struct list_head *iter;
2073 struct slave *slave;
2074 bool ignore_updelay;
2076 ignore_updelay = !rcu_dereference(bond->curr_active_slave);
2078 bond_for_each_slave_rcu(bond, slave, iter) {
2079 slave->new_link = BOND_LINK_NOCHANGE;
2080 slave->link_new_state = slave->link;
2082 link_state = bond_check_dev_link(bond, slave->dev, 0);
2084 switch (slave->link) {
2085 case BOND_LINK_UP:
2086 if (link_state)
2087 continue;
2089 bond_propose_link_state(slave, BOND_LINK_FAIL);
2090 commit++;
2091 slave->delay = bond->params.downdelay;
2092 if (slave->delay) {
2093 netdev_info(bond->dev, "link status down for %sinterface %s, disabling it in %d ms\n",
2094 (BOND_MODE(bond) ==
2095 BOND_MODE_ACTIVEBACKUP) ?
2096 (bond_is_active_slave(slave) ?
2097 "active " : "backup ") : "",
2098 slave->dev->name,
2099 bond->params.downdelay * bond->params.miimon);
2100 }
2101 /*FALLTHRU*/
2102 case BOND_LINK_FAIL:
2103 if (link_state) {
2104 /* recovered before downdelay expired */
2105 bond_propose_link_state(slave, BOND_LINK_UP);
2106 slave->last_link_up = jiffies;
2107 netdev_info(bond->dev, "link status up again after %d ms for interface %s\n",
2108 (bond->params.downdelay - slave->delay) *
2109 bond->params.miimon,
2110 slave->dev->name);
2111 commit++;
2112 continue;
2113 }
2115 if (slave->delay <= 0) {
2116 slave->new_link = BOND_LINK_DOWN;
2117 commit++;
2118 continue;
2119 }
2121 slave->delay--;
2122 break;
2124 case BOND_LINK_DOWN:
2125 if (!link_state)
2126 continue;
2128 bond_propose_link_state(slave, BOND_LINK_BACK);
2129 commit++;
2130 slave->delay = bond->params.updelay;
2132 if (slave->delay) {
2133 netdev_info(bond->dev, "link status up for interface %s, enabling it in %d ms\n",
2134 slave->dev->name,
2135 ignore_updelay ? 0 :
2136 bond->params.updelay *
2137 bond->params.miimon);
2138 }
2139 /*FALLTHRU*/
2140 case BOND_LINK_BACK:
2141 if (!link_state) {
2142 bond_propose_link_state(slave, BOND_LINK_DOWN);
2143 netdev_info(bond->dev, "link status down again after %d ms for interface %s\n",
2144 (bond->params.updelay - slave->delay) *
2145 bond->params.miimon,
2146 slave->dev->name);
2147 commit++;
2148 continue;
2149 }
2151 if (ignore_updelay)
2152 slave->delay = 0;
2154 if (slave->delay <= 0) {
2155 slave->new_link = BOND_LINK_UP;
2156 commit++;
2157 ignore_updelay = false;
2158 continue;
2159 }
2161 slave->delay--;
2162 break;
2163 }
2164 }
2166 return commit;
2167 }
2169 static void bond_miimon_link_change(struct bonding *bond,
2170 struct slave *slave,
2171 char link)
2172 {
2173 switch (BOND_MODE(bond)) {
2174 case BOND_MODE_8023AD:
2175 bond_3ad_handle_link_change(slave, link);
2176 break;
2177 case BOND_MODE_TLB:
2178 case BOND_MODE_ALB:
2179 bond_alb_handle_link_change(bond, slave, link);
2180 break;
2181 case BOND_MODE_XOR:
2182 bond_update_slave_arr(bond, NULL);
2183 break;
2184 }
2185 }
2187 static void bond_miimon_commit(struct bonding *bond)
2188 {
2189 struct list_head *iter;
2190 struct slave *slave, *primary;
2192 bond_for_each_slave(bond, slave, iter) {
2193 switch (slave->new_link) {
2194 case BOND_LINK_NOCHANGE:
2195 continue;
2197 case BOND_LINK_UP:
2198 if (bond_update_speed_duplex(slave) &&
2199 bond_needs_speed_duplex(bond)) {
2200 slave->link = BOND_LINK_DOWN;
2201 if (net_ratelimit())
2202 netdev_warn(bond->dev,
2203 "failed to get link speed/duplex for %s\n",
2204 slave->dev->name);
2205 continue;
2206 }
2207 bond_set_slave_link_state(slave, BOND_LINK_UP,
2208 BOND_SLAVE_NOTIFY_NOW);
2209 slave->last_link_up = jiffies;
2211 primary = rtnl_dereference(bond->primary_slave);
2212 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
2213 /* prevent it from being the active one */
2214 bond_set_backup_slave(slave);
2215 } else if (BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
2216 /* make it immediately active */
2217 bond_set_active_slave(slave);
2218 } else if (slave != primary) {
2219 /* prevent it from being the active one */
2220 bond_set_backup_slave(slave);
2221 }
2223 netdev_info(bond->dev, "link status definitely up for interface %s, %u Mbps %s duplex\n",
2224 slave->dev->name,
2225 slave->speed == SPEED_UNKNOWN ? 0 : slave->speed,
2226 slave->duplex ? "full" : "half");
2228 bond_miimon_link_change(bond, slave, BOND_LINK_UP);
2230 if (!bond->curr_active_slave || slave == primary)
2231 goto do_failover;
2233 continue;
2235 case BOND_LINK_DOWN:
2236 if (slave->link_failure_count < UINT_MAX)
2237 slave->link_failure_count++;
2239 bond_set_slave_link_state(slave, BOND_LINK_DOWN,
2240 BOND_SLAVE_NOTIFY_NOW);
2242 if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP ||
2243 BOND_MODE(bond) == BOND_MODE_8023AD)
2244 bond_set_slave_inactive_flags(slave,
2245 BOND_SLAVE_NOTIFY_NOW);
2247 netdev_info(bond->dev, "link status definitely down for interface %s, disabling it\n",
2248 slave->dev->name);
2250 bond_miimon_link_change(bond, slave, BOND_LINK_DOWN);
2252 if (slave == rcu_access_pointer(bond->curr_active_slave))
2253 goto do_failover;
2255 continue;
2257 default:
2258 netdev_err(bond->dev, "invalid new link %d on slave %s\n",
2259 slave->new_link, slave->dev->name);
2260 slave->new_link = BOND_LINK_NOCHANGE;
2262 continue;
2263 }
2265 do_failover:
2266 block_netpoll_tx();
2267 bond_select_active_slave(bond);
2268 unblock_netpoll_tx();
2269 }
2271 bond_set_carrier(bond);
2272 }
2274 /* bond_mii_monitor
2275 *
2276 * Really a wrapper that splits the mii monitor into two phases: an
2277 * inspection, then (if inspection indicates something needs to be done)
2278 * an acquisition of appropriate locks followed by a commit phase to
2279 * implement whatever link state changes are indicated.
2280 */
2281 static void bond_mii_monitor(struct work_struct *work)
2282 {
2283 struct bonding *bond = container_of(work, struct bonding,
2284 mii_work.work);
2285 bool should_notify_peers = false;
2286 unsigned long delay;
2287 struct slave *slave;
2288 struct list_head *iter;
2290 delay = msecs_to_jiffies(bond->params.miimon);
2292 if (!bond_has_slaves(bond))
2293 goto re_arm;
2295 rcu_read_lock();
2297 should_notify_peers = bond_should_notify_peers(bond);
2299 if (bond_miimon_inspect(bond)) {
2300 rcu_read_unlock();
2302 /* Race avoidance with bond_close cancel of workqueue */
2303 if (!rtnl_trylock()) {
2304 delay = 1;
2305 should_notify_peers = false;
2306 goto re_arm;
2307 }
2309 bond_for_each_slave(bond, slave, iter) {
2310 bond_commit_link_state(slave, BOND_SLAVE_NOTIFY_LATER);
2311 }
2312 bond_miimon_commit(bond);
2314 rtnl_unlock(); /* might sleep, hold no other locks */
2315 } else
2316 rcu_read_unlock();
2318 re_arm:
2319 if (bond->params.miimon)
2320 queue_delayed_work(bond->wq, &bond->mii_work, delay);
2322 if (should_notify_peers) {
2323 if (!rtnl_trylock())
2324 return;
2325 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, bond->dev);
2326 rtnl_unlock();
2327 }
2328 }
2330 static int bond_upper_dev_walk(struct net_device *upper, void *data)
2331 {
2332 __be32 ip = *((__be32 *)data);
2334 return ip == bond_confirm_addr(upper, 0, ip);
2335 }
2337 static bool bond_has_this_ip(struct bonding *bond, __be32 ip)
2338 {
2339 bool ret = false;
2341 if (ip == bond_confirm_addr(bond->dev, 0, ip))
2342 return true;
2344 rcu_read_lock();
2345 if (netdev_walk_all_upper_dev_rcu(bond->dev, bond_upper_dev_walk, &ip))
2346 ret = true;
2347 rcu_read_unlock();
2349 return ret;
2350 }
2352 /* We go to the (large) trouble of VLAN tagging ARP frames because
2353 * switches in VLAN mode (especially if ports are configured as
2354 * "native" to a VLAN) might not pass non-tagged frames.
2355 */
2356 static void bond_arp_send(struct net_device *slave_dev, int arp_op,
2357 __be32 dest_ip, __be32 src_ip,
2358 struct bond_vlan_tag *tags)
2359 {
2360 struct sk_buff *skb;
2361 struct bond_vlan_tag *outer_tag = tags;
2363 netdev_dbg(slave_dev, "arp %d on slave %s: dst %pI4 src %pI4\n",
2364 arp_op, slave_dev->name, &dest_ip, &src_ip);
2366 skb = arp_create(arp_op, ETH_P_ARP, dest_ip, slave_dev, src_ip,
2367 NULL, slave_dev->dev_addr, NULL);
2369 if (!skb) {
2370 net_err_ratelimited("ARP packet allocation failed\n");
2371 return;
2372 }
2374 if (!tags || tags->vlan_proto == VLAN_N_VID)
2375 goto xmit;
2377 tags++;
2379 /* Go through all the tags backwards and add them to the packet */
2380 while (tags->vlan_proto != VLAN_N_VID) {
2381 if (!tags->vlan_id) {
2382 tags++;
2383 continue;
2384 }
2386 netdev_dbg(slave_dev, "inner tag: proto %X vid %X\n",
2387 ntohs(outer_tag->vlan_proto), tags->vlan_id);
2388 skb = vlan_insert_tag_set_proto(skb, tags->vlan_proto,
2389 tags->vlan_id);
2390 if (!skb) {
2391 net_err_ratelimited("failed to insert inner VLAN tag\n");
2392 return;
2393 }
2395 tags++;
2396 }
2397 /* Set the outer tag */
2398 if (outer_tag->vlan_id) {
2399 netdev_dbg(slave_dev, "outer tag: proto %X vid %X\n",
2400 ntohs(outer_tag->vlan_proto), outer_tag->vlan_id);
2401 __vlan_hwaccel_put_tag(skb, outer_tag->vlan_proto,
2402 outer_tag->vlan_id);
2403 }
2405 xmit:
2406 arp_xmit(skb);
2407 }
2409 /* Validate the device path between the @start_dev and the @end_dev.
2410 * The path is valid if the @end_dev is reachable through device
2411 * stacking.
2412 * When the path is validated, collect any vlan information in the
2413 * path.
2414 */
2415 struct bond_vlan_tag *bond_verify_device_path(struct net_device *start_dev,
2416 struct net_device *end_dev,
2417 int level)
2418 {
2419 struct bond_vlan_tag *tags;
2420 struct net_device *upper;
2421 struct list_head *iter;
2423 if (start_dev == end_dev) {
2424 tags = kcalloc(level + 1, sizeof(*tags), GFP_ATOMIC);
2425 if (!tags)
2426 return ERR_PTR(-ENOMEM);
2427 tags[level].vlan_proto = VLAN_N_VID;
2428 return tags;
2429 }
2431 netdev_for_each_upper_dev_rcu(start_dev, upper, iter) {
2432 tags = bond_verify_device_path(upper, end_dev, level + 1);
2433 if (IS_ERR_OR_NULL(tags)) {
2434 if (IS_ERR(tags))
2435 return tags;
2436 continue;
2437 }
2438 if (is_vlan_dev(upper)) {
2439 tags[level].vlan_proto = vlan_dev_vlan_proto(upper);
2440 tags[level].vlan_id = vlan_dev_vlan_id(upper);
2441 }
2443 return tags;
2444 }
2446 return NULL;
2447 }
2449 static void bond_arp_send_all(struct bonding *bond, struct slave *slave)
2450 {
2451 struct rtable *rt;
2452 struct bond_vlan_tag *tags;
2453 __be32 *targets = bond->params.arp_targets, addr;
2454 int i;
2456 for (i = 0; i < BOND_MAX_ARP_TARGETS && targets[i]; i++) {
2457 netdev_dbg(bond->dev, "basa: target %pI4\n", &targets[i]);
2458 tags = NULL;
2460 /* Find out through which dev should the packet go */
2461 rt = ip_route_output(dev_net(bond->dev), targets[i], 0,
2462 RTO_ONLINK, 0);
2463 if (IS_ERR(rt)) {
2464 /* there's no route to target - try to send arp
2465 * probe to generate any traffic (arp_validate=0)
2466 */
2467 if (bond->params.arp_validate)
2468 net_warn_ratelimited("%s: no route to arp_ip_target %pI4 and arp_validate is set\n",
2469 bond->dev->name,
2470 &targets[i]);
2471 bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2472 0, tags);
2473 continue;
2474 }
2476 /* bond device itself */
2477 if (rt->dst.dev == bond->dev)
2478 goto found;
2480 rcu_read_lock();
2481 tags = bond_verify_device_path(bond->dev, rt->dst.dev, 0);
2482 rcu_read_unlock();
2484 if (!IS_ERR_OR_NULL(tags))
2485 goto found;
2487 /* Not our device - skip */
2488 netdev_dbg(bond->dev, "no path to arp_ip_target %pI4 via rt.dev %s\n",
2489 &targets[i], rt->dst.dev ? rt->dst.dev->name : "NULL");
2491 ip_rt_put(rt);
2492 continue;
2494 found:
2495 addr = bond_confirm_addr(rt->dst.dev, targets[i], 0);
2496 ip_rt_put(rt);
2497 bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2498 addr, tags);
2499 kfree(tags);
2500 }
2501 }
2503 static void bond_validate_arp(struct bonding *bond, struct slave *slave, __be32 sip, __be32 tip)
2504 {
2505 int i;
2507 if (!sip || !bond_has_this_ip(bond, tip)) {
2508 netdev_dbg(bond->dev, "bva: sip %pI4 tip %pI4 not found\n",
2509 &sip, &tip);
2510 return;
2511 }
2513 i = bond_get_targets_ip(bond->params.arp_targets, sip);
2514 if (i == -1) {
2515 netdev_dbg(bond->dev, "bva: sip %pI4 not found in targets\n",
2516 &sip);
2517 return;
2518 }
2519 slave->last_rx = jiffies;
2520 slave->target_last_arp_rx[i] = jiffies;
2521 }
2523 int bond_arp_rcv(const struct sk_buff *skb, struct bonding *bond,
2524 struct slave *slave)
2525 {
2526 struct arphdr *arp = (struct arphdr *)skb->data;
2527 struct slave *curr_active_slave, *curr_arp_slave;
2528 unsigned char *arp_ptr;
2529 __be32 sip, tip;
2530 int is_arp = skb->protocol == __cpu_to_be16(ETH_P_ARP);
2531 unsigned int alen;
2533 if (!slave_do_arp_validate(bond, slave)) {
2534 if ((slave_do_arp_validate_only(bond) && is_arp) ||
2535 !slave_do_arp_validate_only(bond))
2536 slave->last_rx = jiffies;
2537 return RX_HANDLER_ANOTHER;
2538 } else if (!is_arp) {
2539 return RX_HANDLER_ANOTHER;
2540 }
2542 alen = arp_hdr_len(bond->dev);
2544 netdev_dbg(bond->dev, "bond_arp_rcv: skb->dev %s\n",
2545 skb->dev->name);
2547 if (alen > skb_headlen(skb)) {
2548 arp = kmalloc(alen, GFP_ATOMIC);
2549 if (!arp)
2550 goto out_unlock;
2551 if (skb_copy_bits(skb, 0, arp, alen) < 0)
2552 goto out_unlock;
2553 }
2555 if (arp->ar_hln != bond->dev->addr_len ||
2556 skb->pkt_type == PACKET_OTHERHOST ||
2557 skb->pkt_type == PACKET_LOOPBACK ||
2558 arp->ar_hrd != htons(ARPHRD_ETHER) ||
2559 arp->ar_pro != htons(ETH_P_IP) ||
2560 arp->ar_pln != 4)
2561 goto out_unlock;
2563 arp_ptr = (unsigned char *)(arp + 1);
2564 arp_ptr += bond->dev->addr_len;
2565 memcpy(&sip, arp_ptr, 4);
2566 arp_ptr += 4 + bond->dev->addr_len;
2567 memcpy(&tip, arp_ptr, 4);
2569 netdev_dbg(bond->dev, "bond_arp_rcv: %s/%d av %d sv %d sip %pI4 tip %pI4\n",
2570 slave->dev->name, bond_slave_state(slave),
2571 bond->params.arp_validate, slave_do_arp_validate(bond, slave),
2572 &sip, &tip);
2574 curr_active_slave = rcu_dereference(bond->curr_active_slave);
2575 curr_arp_slave = rcu_dereference(bond->current_arp_slave);
2577 /* We 'trust' the received ARP enough to validate it if:
2578 *
2579 * (a) the slave receiving the ARP is active (which includes the
2580 * current ARP slave, if any), or
2581 *
2582 * (b) the receiving slave isn't active, but there is a currently
2583 * active slave and it received valid arp reply(s) after it became
2584 * the currently active slave, or
2585 *
2586 * (c) there is an ARP slave that sent an ARP during the prior ARP
2587 * interval, and we receive an ARP reply on any slave. We accept
2588 * these because switch FDB update delays may deliver the ARP
2589 * reply to a slave other than the sender of the ARP request.
2590 *
2591 * Note: for (b), backup slaves are receiving the broadcast ARP
2592 * request, not a reply. This request passes from the sending
2593 * slave through the L2 switch(es) to the receiving slave. Since
2594 * this is checking the request, sip/tip are swapped for
2595 * validation.
2596 *
2597 * This is done to avoid endless looping when we can't reach the
2598 * arp_ip_target and fool ourselves with our own arp requests.
2599 */
2600 if (bond_is_active_slave(slave))
2601 bond_validate_arp(bond, slave, sip, tip);
2602 else if (curr_active_slave &&
2603 time_after(slave_last_rx(bond, curr_active_slave),
2604 curr_active_slave->last_link_up))
2605 bond_validate_arp(bond, slave, tip, sip);
2606 else if (curr_arp_slave && (arp->ar_op == htons(ARPOP_REPLY)) &&
2607 bond_time_in_interval(bond,
2608 dev_trans_start(curr_arp_slave->dev), 1))
2609 bond_validate_arp(bond, slave, sip, tip);
2611 out_unlock:
2612 if (arp != (struct arphdr *)skb->data)
2613 kfree(arp);
2614 return RX_HANDLER_ANOTHER;
2615 }
2617 /* function to verify if we're in the arp_interval timeslice, returns true if
2618 * (last_act - arp_interval) <= jiffies <= (last_act + mod * arp_interval +
2619 * arp_interval/2) . the arp_interval/2 is needed for really fast networks.
2620 */
2621 static bool bond_time_in_interval(struct bonding *bond, unsigned long last_act,
2622 int mod)
2623 {
2624 int delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
2626 return time_in_range(jiffies,
2627 last_act - delta_in_ticks,
2628 last_act + mod * delta_in_ticks + delta_in_ticks/2);
2629 }
2631 /* This function is called regularly to monitor each slave's link
2632 * ensuring that traffic is being sent and received when arp monitoring
2633 * is used in load-balancing mode. if the adapter has been dormant, then an
2634 * arp is transmitted to generate traffic. see activebackup_arp_monitor for
2635 * arp monitoring in active backup mode.
2636 */
2637 static void bond_loadbalance_arp_mon(struct bonding *bond)
2638 {
2639 struct slave *slave, *oldcurrent;
2640 struct list_head *iter;
2641 int do_failover = 0, slave_state_changed = 0;
2643 if (!bond_has_slaves(bond))
2644 goto re_arm;
2646 rcu_read_lock();
2648 oldcurrent = rcu_dereference(bond->curr_active_slave);
2649 /* see if any of the previous devices are up now (i.e. they have
2650 * xmt and rcv traffic). the curr_active_slave does not come into
2651 * the picture unless it is null. also, slave->last_link_up is not
2652 * needed here because we send an arp on each slave and give a slave
2653 * as long as it needs to get the tx/rx within the delta.
2654 * TODO: what about up/down delay in arp mode? it wasn't here before
2655 * so it can wait
2656 */
2657 bond_for_each_slave_rcu(bond, slave, iter) {
2658 unsigned long trans_start = dev_trans_start(slave->dev);
2660 slave->new_link = BOND_LINK_NOCHANGE;
2662 if (slave->link != BOND_LINK_UP) {
2663 if (bond_time_in_interval(bond, trans_start, 1) &&
2664 bond_time_in_interval(bond, slave->last_rx, 1)) {
2666 slave->new_link = BOND_LINK_UP;
2667 slave_state_changed = 1;
2669 /* primary_slave has no meaning in round-robin
2670 * mode. the window of a slave being up and
2671 * curr_active_slave being null after enslaving
2672 * is closed.
2673 */
2674 if (!oldcurrent) {
2675 netdev_info(bond->dev, "link status definitely up for interface %s\n",
2676 slave->dev->name);
2677 do_failover = 1;
2678 } else {
2679 netdev_info(bond->dev, "interface %s is now up\n",
2680 slave->dev->name);
2681 }
2682 }
2683 } else {
2684 /* slave->link == BOND_LINK_UP */
2686 /* not all switches will respond to an arp request
2687 * when the source ip is 0, so don't take the link down
2688 * if we don't know our ip yet
2689 */
2690 if (!bond_time_in_interval(bond, trans_start, 2) ||
2691 !bond_time_in_interval(bond, slave->last_rx, 2)) {
2693 slave->new_link = BOND_LINK_DOWN;
2694 slave_state_changed = 1;
2696 if (slave->link_failure_count < UINT_MAX)
2697 slave->link_failure_count++;
2699 netdev_info(bond->dev, "interface %s is now down\n",
2700 slave->dev->name);
2702 if (slave == oldcurrent)
2703 do_failover = 1;
2704 }
2705 }
2707 /* note: if switch is in round-robin mode, all links
2708 * must tx arp to ensure all links rx an arp - otherwise
2709 * links may oscillate or not come up at all; if switch is
2710 * in something like xor mode, there is nothing we can
2711 * do - all replies will be rx'ed on same link causing slaves
2712 * to be unstable during low/no traffic periods
2713 */
2714 if (bond_slave_is_up(slave))
2715 bond_arp_send_all(bond, slave);
2716 }
2718 rcu_read_unlock();
2720 if (do_failover || slave_state_changed) {
2721 if (!rtnl_trylock())
2722 goto re_arm;
2724 bond_for_each_slave(bond, slave, iter) {
2725 if (slave->new_link != BOND_LINK_NOCHANGE)
2726 slave->link = slave->new_link;
2727 }
2729 if (slave_state_changed) {
2730 bond_slave_state_change(bond);
2731 if (BOND_MODE(bond) == BOND_MODE_XOR)
2732 bond_update_slave_arr(bond, NULL);
2733 }
2734 if (do_failover) {
2735 block_netpoll_tx();
2736 bond_select_active_slave(bond);
2737 unblock_netpoll_tx();
2738 }
2739 rtnl_unlock();
2740 }
2742 re_arm:
2743 if (bond->params.arp_interval)
2744 queue_delayed_work(bond->wq, &bond->arp_work,
2745 msecs_to_jiffies(bond->params.arp_interval));
2746 }
2748 /* Called to inspect slaves for active-backup mode ARP monitor link state
2749 * changes. Sets new_link in slaves to specify what action should take
2750 * place for the slave. Returns 0 if no changes are found, >0 if changes
2751 * to link states must be committed.
2752 *
2753 * Called with rcu_read_lock held.
2754 */
2755 static int bond_ab_arp_inspect(struct bonding *bond)
2756 {
2757 unsigned long trans_start, last_rx;
2758 struct list_head *iter;
2759 struct slave *slave;
2760 int commit = 0;
2762 bond_for_each_slave_rcu(bond, slave, iter) {
2763 slave->new_link = BOND_LINK_NOCHANGE;
2764 last_rx = slave_last_rx(bond, slave);
2766 if (slave->link != BOND_LINK_UP) {
2767 if (bond_time_in_interval(bond, last_rx, 1)) {
2768 slave->new_link = BOND_LINK_UP;
2769 commit++;
2770 }
2771 continue;
2772 }
2774 /* Give slaves 2*delta after being enslaved or made
2775 * active. This avoids bouncing, as the last receive
2776 * times need a full ARP monitor cycle to be updated.
2777 */
2778 if (bond_time_in_interval(bond, slave->last_link_up, 2))
2779 continue;
2781 /* Backup slave is down if:
2782 * - No current_arp_slave AND
2783 * - more than 3*delta since last receive AND
2784 * - the bond has an IP address
2785 *
2786 * Note: a non-null current_arp_slave indicates
2787 * the curr_active_slave went down and we are
2788 * searching for a new one; under this condition
2789 * we only take the curr_active_slave down - this
2790 * gives each slave a chance to tx/rx traffic
2791 * before being taken out
2792 */
2793 if (!bond_is_active_slave(slave) &&
2794 !rcu_access_pointer(bond->current_arp_slave) &&
2795 !bond_time_in_interval(bond, last_rx, 3)) {
2796 slave->new_link = BOND_LINK_DOWN;
2797 commit++;
2798 }
2800 /* Active slave is down if:
2801 * - more than 2*delta since transmitting OR
2802 * - (more than 2*delta since receive AND
2803 * the bond has an IP address)
2804 */
2805 trans_start = dev_trans_start(slave->dev);
2806 if (bond_is_active_slave(slave) &&
2807 (!bond_time_in_interval(bond, trans_start, 2) ||
2808 !bond_time_in_interval(bond, last_rx, 2))) {
2809 slave->new_link = BOND_LINK_DOWN;
2810 commit++;
2811 }
2812 }
2814 return commit;
2815 }
2817 /* Called to commit link state changes noted by inspection step of
2818 * active-backup mode ARP monitor.
2819 *
2820 * Called with RTNL hold.
2821 */
2822 static void bond_ab_arp_commit(struct bonding *bond)
2823 {
2824 unsigned long trans_start;
2825 struct list_head *iter;
2826 struct slave *slave;
2828 bond_for_each_slave(bond, slave, iter) {
2829 switch (slave->new_link) {
2830 case BOND_LINK_NOCHANGE:
2831 continue;
2833 case BOND_LINK_UP:
2834 trans_start = dev_trans_start(slave->dev);
2835 if (rtnl_dereference(bond->curr_active_slave) != slave ||
2836 (!rtnl_dereference(bond->curr_active_slave) &&
2837 bond_time_in_interval(bond, trans_start, 1))) {
2838 struct slave *current_arp_slave;
2840 current_arp_slave = rtnl_dereference(bond->current_arp_slave);
2841 bond_set_slave_link_state(slave, BOND_LINK_UP,
2842 BOND_SLAVE_NOTIFY_NOW);
2843 if (current_arp_slave) {
2844 bond_set_slave_inactive_flags(
2845 current_arp_slave,
2846 BOND_SLAVE_NOTIFY_NOW);
2847 RCU_INIT_POINTER(bond->current_arp_slave, NULL);
2848 }
2850 netdev_info(bond->dev, "link status definitely up for interface %s\n",
2851 slave->dev->name);
2853 if (!rtnl_dereference(bond->curr_active_slave) ||
2854 slave == rtnl_dereference(bond->primary_slave))
2855 goto do_failover;
2857 }
2859 continue;
2861 case BOND_LINK_DOWN:
2862 if (slave->link_failure_count < UINT_MAX)
2863 slave->link_failure_count++;
2865 bond_set_slave_link_state(slave, BOND_LINK_DOWN,
2866 BOND_SLAVE_NOTIFY_NOW);
2867 bond_set_slave_inactive_flags(slave,
2868 BOND_SLAVE_NOTIFY_NOW);
2870 netdev_info(bond->dev, "link status definitely down for interface %s, disabling it\n",
2871 slave->dev->name);
2873 if (slave == rtnl_dereference(bond->curr_active_slave)) {
2874 RCU_INIT_POINTER(bond->current_arp_slave, NULL);
2875 goto do_failover;
2876 }
2878 continue;
2880 default:
2881 netdev_err(bond->dev, "impossible: new_link %d on slave %s\n",
2882 slave->new_link, slave->dev->name);
2883 continue;
2884 }
2886 do_failover:
2887 block_netpoll_tx();
2888 bond_select_active_slave(bond);
2889 unblock_netpoll_tx();
2890 }
2892 bond_set_carrier(bond);
2893 }
2895 /* Send ARP probes for active-backup mode ARP monitor.
2896 *
2897 * Called with rcu_read_lock held.
2898 */
2899 static bool bond_ab_arp_probe(struct bonding *bond)
2900 {
2901 struct slave *slave, *before = NULL, *new_slave = NULL,
2902 *curr_arp_slave = rcu_dereference(bond->current_arp_slave),
2903 *curr_active_slave = rcu_dereference(bond->curr_active_slave);
2904 struct list_head *iter;
2905 bool found = false;
2906 bool should_notify_rtnl = BOND_SLAVE_NOTIFY_LATER;
2908 if (curr_arp_slave && curr_active_slave)
2909 netdev_info(bond->dev, "PROBE: c_arp %s && cas %s BAD\n",
2910 curr_arp_slave->dev->name,
2911 curr_active_slave->dev->name);
2913 if (curr_active_slave) {
2914 bond_arp_send_all(bond, curr_active_slave);
2915 return should_notify_rtnl;
2916 }
2918 /* if we don't have a curr_active_slave, search for the next available
2919 * backup slave from the current_arp_slave and make it the candidate
2920 * for becoming the curr_active_slave
2921 */
2923 if (!curr_arp_slave) {
2924 curr_arp_slave = bond_first_slave_rcu(bond);
2925 if (!curr_arp_slave)
2926 return should_notify_rtnl;
2927 }
2929 bond_set_slave_inactive_flags(curr_arp_slave, BOND_SLAVE_NOTIFY_LATER);
2931 bond_for_each_slave_rcu(bond, slave, iter) {
2932 if (!found && !before && bond_slave_is_up(slave))
2933 before = slave;
2935 if (found && !new_slave && bond_slave_is_up(slave))
2936 new_slave = slave;
2937 /* if the link state is up at this point, we
2938 * mark it down - this can happen if we have
2939 * simultaneous link failures and
2940 * reselect_active_interface doesn't make this
2941 * one the current slave so it is still marked
2942 * up when it is actually down
2943 */
2944 if (!bond_slave_is_up(slave) && slave->link == BOND_LINK_UP) {
2945 bond_set_slave_link_state(slave, BOND_LINK_DOWN,
2946 BOND_SLAVE_NOTIFY_LATER);
2947 if (slave->link_failure_count < UINT_MAX)
2948 slave->link_failure_count++;
2950 bond_set_slave_inactive_flags(slave,
2951 BOND_SLAVE_NOTIFY_LATER);
2953 netdev_info(bond->dev, "backup interface %s is now down\n",
2954 slave->dev->name);
2955 }
2956 if (slave == curr_arp_slave)
2957 found = true;
2958 }
2960 if (!new_slave && before)
2961 new_slave = before;
2963 if (!new_slave)
2964 goto check_state;
2966 bond_set_slave_link_state(new_slave, BOND_LINK_BACK,
2967 BOND_SLAVE_NOTIFY_LATER);
2968 bond_set_slave_active_flags(new_slave, BOND_SLAVE_NOTIFY_LATER);
2969 bond_arp_send_all(bond, new_slave);
2970 new_slave->last_link_up = jiffies;
2971 rcu_assign_pointer(bond->current_arp_slave, new_slave);
2973 check_state:
2974 bond_for_each_slave_rcu(bond, slave, iter) {
2975 if (slave->should_notify || slave->should_notify_link) {
2976 should_notify_rtnl = BOND_SLAVE_NOTIFY_NOW;
2977 break;
2978 }
2979 }
2980 return should_notify_rtnl;
2981 }
2983 static void bond_activebackup_arp_mon(struct bonding *bond)
2984 {
2985 bool should_notify_peers = false;
2986 bool should_notify_rtnl = false;
2987 int delta_in_ticks;
2989 delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
2991 if (!bond_has_slaves(bond))
2992 goto re_arm;
2994 rcu_read_lock();
2996 should_notify_peers = bond_should_notify_peers(bond);
2998 if (bond_ab_arp_inspect(bond)) {
2999 rcu_read_unlock();
3001 /* Race avoidance with bond_close flush of workqueue */
3002 if (!rtnl_trylock()) {
3003 delta_in_ticks = 1;
3004 should_notify_peers = false;
3005 goto re_arm;
3006 }
3008 bond_ab_arp_commit(bond);
3010 rtnl_unlock();
3011 rcu_read_lock();
3012 }
3014 should_notify_rtnl = bond_ab_arp_probe(bond);
3015 rcu_read_unlock();
3017 re_arm:
3018 if (bond->params.arp_interval)
3019 queue_delayed_work(bond->wq, &bond->arp_work, delta_in_ticks);
3021 if (should_notify_peers || should_notify_rtnl) {
3022 if (!rtnl_trylock())
3023 return;
3025 if (should_notify_peers)
3026 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS,
3027 bond->dev);
3028 if (should_notify_rtnl) {
3029 bond_slave_state_notify(bond);
3030 bond_slave_link_notify(bond);
3031 }
3033 rtnl_unlock();
3034 }
3035 }
3037 static void bond_arp_monitor(struct work_struct *work)
3038 {
3039 struct bonding *bond = container_of(work, struct bonding,
3040 arp_work.work);
3042 if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP)
3043 bond_activebackup_arp_mon(bond);
3044 else
3045 bond_loadbalance_arp_mon(bond);
3046 }
3048 /*-------------------------- netdev event handling --------------------------*/
3050 /* Change device name */
3051 static int bond_event_changename(struct bonding *bond)
3052 {
3053 bond_remove_proc_entry(bond);
3054 bond_create_proc_entry(bond);
3056 bond_debug_reregister(bond);
3058 return NOTIFY_DONE;
3059 }
3061 static int bond_master_netdev_event(unsigned long event,
3062 struct net_device *bond_dev)
3063 {
3064 struct bonding *event_bond = netdev_priv(bond_dev);
3066 switch (event) {
3067 case NETDEV_CHANGENAME:
3068 return bond_event_changename(event_bond);
3069 case NETDEV_UNREGISTER:
3070 bond_remove_proc_entry(event_bond);
3071 break;
3072 case NETDEV_REGISTER:
3073 bond_create_proc_entry(event_bond);
3074 break;
3075 case NETDEV_NOTIFY_PEERS:
3076 if (event_bond->send_peer_notif)
3077 event_bond->send_peer_notif--;
3078 break;
3079 default:
3080 break;
3081 }
3083 return NOTIFY_DONE;
3084 }
3086 static int bond_slave_netdev_event(unsigned long event,
3087 struct net_device *slave_dev)
3088 {
3089 struct slave *slave = bond_slave_get_rtnl(slave_dev), *primary;
3090 struct bonding *bond;
3091 struct net_device *bond_dev;
3093 /* A netdev event can be generated while enslaving a device
3094 * before netdev_rx_handler_register is called in which case
3095 * slave will be NULL
3096 */
3097 if (!slave)
3098 return NOTIFY_DONE;
3099 bond_dev = slave->bond->dev;
3100 bond = slave->bond;
3101 primary = rtnl_dereference(bond->primary_slave);
3103 switch (event) {
3104 case NETDEV_UNREGISTER:
3105 if (bond_dev->type != ARPHRD_ETHER)
3106 bond_release_and_destroy(bond_dev, slave_dev);
3107 else
3108 __bond_release_one(bond_dev, slave_dev, false, true);
3109 break;
3110 case NETDEV_UP:
3111 case NETDEV_CHANGE:
3112 /* For 802.3ad mode only:
3113 * Getting invalid Speed/Duplex values here will put slave
3114 * in weird state. So mark it as link-down for the time
3115 * being and let link-monitoring (miimon) set it right when
3116 * correct speeds/duplex are available.
3117 */
3118 if (bond_update_speed_duplex(slave) &&
3119 BOND_MODE(bond) == BOND_MODE_8023AD)
3120 slave->link = BOND_LINK_DOWN;
3122 if (BOND_MODE(bond) == BOND_MODE_8023AD)
3123 bond_3ad_adapter_speed_duplex_changed(slave);
3124 /* Fallthrough */
3125 case NETDEV_DOWN:
3126 /* Refresh slave-array if applicable!
3127 * If the setup does not use miimon or arpmon (mode-specific!),
3128 * then these events will not cause the slave-array to be
3129 * refreshed. This will cause xmit to use a slave that is not
3130 * usable. Avoid such situation by refeshing the array at these
3131 * events. If these (miimon/arpmon) parameters are configured
3132 * then array gets refreshed twice and that should be fine!
3133 */
3134 if (bond_mode_can_use_xmit_hash(bond))
3135 bond_update_slave_arr(bond, NULL);
3136 break;
3137 case NETDEV_CHANGEMTU:
3138 /* TODO: Should slaves be allowed to
3139 * independently alter their MTU? For
3140 * an active-backup bond, slaves need
3141 * not be the same type of device, so
3142 * MTUs may vary. For other modes,
3143 * slaves arguably should have the
3144 * same MTUs. To do this, we'd need to
3145 * take over the slave's change_mtu
3146 * function for the duration of their
3147 * servitude.
3148 */
3149 break;
3150 case NETDEV_CHANGENAME:
3151 /* we don't care if we don't have primary set */
3152 if (!bond_uses_primary(bond) ||
3153 !bond->params.primary[0])
3154 break;
3156 if (slave == primary) {
3157 /* slave's name changed - he's no longer primary */
3158 RCU_INIT_POINTER(bond->primary_slave, NULL);
3159 } else if (!strcmp(slave_dev->name, bond->params.primary)) {
3160 /* we have a new primary slave */
3161 rcu_assign_pointer(bond->primary_slave, slave);
3162 } else { /* we didn't change primary - exit */
3163 break;
3164 }
3166 netdev_info(bond->dev, "Primary slave changed to %s, reselecting active slave\n",
3167 primary ? slave_dev->name : "none");
3169 block_netpoll_tx();
3170 bond_select_active_slave(bond);
3171 unblock_netpoll_tx();
3172 break;
3173 case NETDEV_FEAT_CHANGE:
3174 bond_compute_features(bond);
3175 break;
3176 case NETDEV_RESEND_IGMP:
3177 /* Propagate to master device */
3178 call_netdevice_notifiers(event, slave->bond->dev);
3179 break;
3180 default:
3181 break;
3182 }
3184 return NOTIFY_DONE;
3185 }
3187 /* bond_netdev_event: handle netdev notifier chain events.
3188 *
3189 * This function receives events for the netdev chain. The caller (an
3190 * ioctl handler calling blocking_notifier_call_chain) holds the necessary
3191 * locks for us to safely manipulate the slave devices (RTNL lock,
3192 * dev_probe_lock).
3193 */
3194 static int bond_netdev_event(struct notifier_block *this,
3195 unsigned long event, void *ptr)
3196 {
3197 struct net_device *event_dev = netdev_notifier_info_to_dev(ptr);
3199 netdev_dbg(event_dev, "event: %lx\n", event);
3201 if (!(event_dev->priv_flags & IFF_BONDING))
3202 return NOTIFY_DONE;
3204 if (event_dev->flags & IFF_MASTER) {
3205 netdev_dbg(event_dev, "IFF_MASTER\n");
3206 return bond_master_netdev_event(event, event_dev);
3207 }
3209 if (event_dev->flags & IFF_SLAVE) {
3210 netdev_dbg(event_dev, "IFF_SLAVE\n");
3211 return bond_slave_netdev_event(event, event_dev);
3212 }
3214 return NOTIFY_DONE;
3215 }
3217 static struct notifier_block bond_netdev_notifier = {
3218 .notifier_call = bond_netdev_event,
3219 };
3221 /*---------------------------- Hashing Policies -----------------------------*/
3223 /* L2 hash helper */
3224 static inline u32 bond_eth_hash(struct sk_buff *skb)
3225 {
3226 struct ethhdr *ep, hdr_tmp;
3228 ep = skb_header_pointer(skb, 0, sizeof(hdr_tmp), &hdr_tmp);
3229 if (ep)
3230 return ep->h_dest[5] ^ ep->h_source[5] ^ ep->h_proto;
3231 return 0;
3232 }
3234 /* Extract the appropriate headers based on bond's xmit policy */
3235 static bool bond_flow_dissect(struct bonding *bond, struct sk_buff *skb,
3236 struct flow_keys *fk)
3237 {
3238 const struct ipv6hdr *iph6;
3239 const struct iphdr *iph;
3240 int noff, proto = -1;
3242 if (bond->params.xmit_policy > BOND_XMIT_POLICY_LAYER23)
3243 return skb_flow_dissect_flow_keys(skb, fk, 0);
3245 fk->ports.ports = 0;
3246 noff = skb_network_offset(skb);
3247 if (skb->protocol == htons(ETH_P_IP)) {
3248 if (unlikely(!pskb_may_pull(skb, noff + sizeof(*iph))))
3249 return false;
3250 iph = ip_hdr(skb);
3251 iph_to_flow_copy_v4addrs(fk, iph);
3252 noff += iph->ihl << 2;
3253 if (!ip_is_fragment(iph))
3254 proto = iph->protocol;
3255 } else if (skb->protocol == htons(ETH_P_IPV6)) {
3256 if (unlikely(!pskb_may_pull(skb, noff + sizeof(*iph6))))
3257 return false;
3258 iph6 = ipv6_hdr(skb);
3259 iph_to_flow_copy_v6addrs(fk, iph6);
3260 noff += sizeof(*iph6);
3261 proto = iph6->nexthdr;
3262 } else {
3263 return false;
3264 }
3265 if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER34 && proto >= 0)
3266 fk->ports.ports = skb_flow_get_ports(skb, noff, proto);
3268 return true;
3269 }
3271 /**
3272 * bond_xmit_hash - generate a hash value based on the xmit policy
3273 * @bond: bonding device
3274 * @skb: buffer to use for headers
3275 *
3276 * This function will extract the necessary headers from the skb buffer and use
3277 * them to generate a hash based on the xmit_policy set in the bonding device
3278 */
3279 u32 bond_xmit_hash(struct bonding *bond, struct sk_buff *skb)
3280 {
3281 struct flow_keys flow;
3282 u32 hash;
3284 if (bond->params.xmit_policy == BOND_XMIT_POLICY_ENCAP34 &&
3285 skb->l4_hash)
3286 return skb->hash;
3288 if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER2 ||
3289 !bond_flow_dissect(bond, skb, &flow))
3290 return bond_eth_hash(skb);
3292 if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER23 ||
3293 bond->params.xmit_policy == BOND_XMIT_POLICY_ENCAP23)
3294 hash = bond_eth_hash(skb);
3295 else
3296 hash = (__force u32)flow.ports.ports;
3297 hash ^= (__force u32)flow_get_u32_dst(&flow) ^
3298 (__force u32)flow_get_u32_src(&flow);
3299 hash ^= (hash >> 16);
3300 hash ^= (hash >> 8);
3302 return hash >> 1;
3303 }
3305 /*-------------------------- Device entry points ----------------------------*/
3307 void bond_work_init_all(struct bonding *bond)
3308 {
3309 INIT_DELAYED_WORK(&bond->mcast_work,
3310 bond_resend_igmp_join_requests_delayed);
3311 INIT_DELAYED_WORK(&bond->alb_work, bond_alb_monitor);
3312 INIT_DELAYED_WORK(&bond->mii_work, bond_mii_monitor);
3313 INIT_DELAYED_WORK(&bond->arp_work, bond_arp_monitor);
3314 INIT_DELAYED_WORK(&bond->ad_work, bond_3ad_state_machine_handler);
3315 INIT_DELAYED_WORK(&bond->slave_arr_work, bond_slave_arr_handler);
3316 }
3318 static void bond_work_cancel_all(struct bonding *bond)
3319 {
3320 cancel_delayed_work_sync(&bond->mii_work);
3321 cancel_delayed_work_sync(&bond->arp_work);
3322 cancel_delayed_work_sync(&bond->alb_work);
3323 cancel_delayed_work_sync(&bond->ad_work);
3324 cancel_delayed_work_sync(&bond->mcast_work);
3325 cancel_delayed_work_sync(&bond->slave_arr_work);
3326 }
3328 static int bond_open(struct net_device *bond_dev)
3329 {
3330 struct bonding *bond = netdev_priv(bond_dev);
3331 struct list_head *iter;
3332 struct slave *slave;
3334 /* reset slave->backup and slave->inactive */
3335 if (bond_has_slaves(bond)) {
3336 bond_for_each_slave(bond, slave, iter) {
3337 if (bond_uses_primary(bond) &&
3338 slave != rcu_access_pointer(bond->curr_active_slave)) {
3339 bond_set_slave_inactive_flags(slave,
3340 BOND_SLAVE_NOTIFY_NOW);
3341 } else if (BOND_MODE(bond) != BOND_MODE_8023AD) {
3342 bond_set_slave_active_flags(slave,
3343 BOND_SLAVE_NOTIFY_NOW);
3344 }
3345 }
3346 }
3348 if (bond_is_lb(bond)) {
3349 /* bond_alb_initialize must be called before the timer
3350 * is started.
3351 */
3352 if (bond_alb_initialize(bond, (BOND_MODE(bond) == BOND_MODE_ALB)))
3353 return -ENOMEM;
3354 if (bond->params.tlb_dynamic_lb || BOND_MODE(bond) == BOND_MODE_ALB)
3355 queue_delayed_work(bond->wq, &bond->alb_work, 0);
3356 }
3358 if (bond->params.miimon) /* link check interval, in milliseconds. */
3359 queue_delayed_work(bond->wq, &bond->mii_work, 0);
3361 if (bond->params.arp_interval) { /* arp interval, in milliseconds. */
3362 queue_delayed_work(bond->wq, &bond->arp_work, 0);
3363 bond->recv_probe = bond_arp_rcv;
3364 }
3366 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
3367 queue_delayed_work(bond->wq, &bond->ad_work, 0);
3368 /* register to receive LACPDUs */
3369 bond->recv_probe = bond_3ad_lacpdu_recv;
3370 bond_3ad_initiate_agg_selection(bond, 1);
3371 }
3373 if (bond_mode_can_use_xmit_hash(bond))
3374 bond_update_slave_arr(bond, NULL);
3376 return 0;
3377 }
3379 static int bond_close(struct net_device *bond_dev)
3380 {
3381 struct bonding *bond = netdev_priv(bond_dev);
3383 bond_work_cancel_all(bond);
3384 bond->send_peer_notif = 0;
3385 if (bond_is_lb(bond))
3386 bond_alb_deinitialize(bond);
3387 bond->recv_probe = NULL;
3389 return 0;
3390 }
3392 /* fold stats, assuming all rtnl_link_stats64 fields are u64, but
3393 * that some drivers can provide 32bit values only.
3394 */
3395 static void bond_fold_stats(struct rtnl_link_stats64 *_res,
3396 const struct rtnl_link_stats64 *_new,
3397 const struct rtnl_link_stats64 *_old)
3398 {
3399 const u64 *new = (const u64 *)_new;
3400 const u64 *old = (const u64 *)_old;
3401 u64 *res = (u64 *)_res;
3402 int i;
3404 for (i = 0; i < sizeof(*_res) / sizeof(u64); i++) {
3405 u64 nv = new[i];
3406 u64 ov = old[i];
3407 s64 delta = nv - ov;
3409 /* detects if this particular field is 32bit only */
3410 if (((nv | ov) >> 32) == 0)
3411 delta = (s64)(s32)((u32)nv - (u32)ov);
3413 /* filter anomalies, some drivers reset their stats
3414 * at down/up events.
3415 */
3416 if (delta > 0)
3417 res[i] += delta;
3418 }
3419 }
3421 static int bond_get_nest_level(struct net_device *bond_dev)
3422 {
3423 struct bonding *bond = netdev_priv(bond_dev);
3425 return bond->nest_level;
3426 }
3428 static void bond_get_stats(struct net_device *bond_dev,
3429 struct rtnl_link_stats64 *stats)
3430 {
3431 struct bonding *bond = netdev_priv(bond_dev);
3432 struct rtnl_link_stats64 temp;
3433 struct list_head *iter;
3434 struct slave *slave;
3436 spin_lock_nested(&bond->stats_lock, bond_get_nest_level(bond_dev));
3437 memcpy(stats, &bond->bond_stats, sizeof(*stats));
3439 rcu_read_lock();
3440 bond_for_each_slave_rcu(bond, slave, iter) {
3441 const struct rtnl_link_stats64 *new =
3442 dev_get_stats(slave->dev, &temp);
3444 bond_fold_stats(stats, new, &slave->slave_stats);
3446 /* save off the slave stats for the next run */
3447 memcpy(&slave->slave_stats, new, sizeof(*new));
3448 }
3449 rcu_read_unlock();
3451 memcpy(&bond->bond_stats, stats, sizeof(*stats));
3452 spin_unlock(&bond->stats_lock);
3453 }
3455 static int bond_do_ioctl(struct net_device *bond_dev, struct ifreq *ifr, int cmd)
3456 {
3457 struct bonding *bond = netdev_priv(bond_dev);
3458 struct net_device *slave_dev = NULL;
3459 struct ifbond k_binfo;
3460 struct ifbond __user *u_binfo = NULL;
3461 struct ifslave k_sinfo;
3462 struct ifslave __user *u_sinfo = NULL;
3463 struct mii_ioctl_data *mii = NULL;
3464 struct bond_opt_value newval;
3465 struct net *net;
3466 int res = 0;
3468 netdev_dbg(bond_dev, "bond_ioctl: cmd=%d\n", cmd);
3470 switch (cmd) {
3471 case SIOCGMIIPHY:
3472 mii = if_mii(ifr);
3473 if (!mii)
3474 return -EINVAL;
3476 mii->phy_id = 0;
3477 /* Fall Through */
3478 case SIOCGMIIREG:
3479 /* We do this again just in case we were called by SIOCGMIIREG
3480 * instead of SIOCGMIIPHY.
3481 */
3482 mii = if_mii(ifr);
3483 if (!mii)
3484 return -EINVAL;
3486 if (mii->reg_num == 1) {
3487 mii->val_out = 0;
3488 if (netif_carrier_ok(bond->dev))
3489 mii->val_out = BMSR_LSTATUS;
3490 }
3492 return 0;
3493 case BOND_INFO_QUERY_OLD:
3494 case SIOCBONDINFOQUERY:
3495 u_binfo = (struct ifbond __user *)ifr->ifr_data;
3497 if (copy_from_user(&k_binfo, u_binfo, sizeof(ifbond)))
3498 return -EFAULT;
3500 bond_info_query(bond_dev, &k_binfo);
3501 if (copy_to_user(u_binfo, &k_binfo, sizeof(ifbond)))
3502 return -EFAULT;
3504 return 0;
3505 case BOND_SLAVE_INFO_QUERY_OLD:
3506 case SIOCBONDSLAVEINFOQUERY:
3507 u_sinfo = (struct ifslave __user *)ifr->ifr_data;
3509 if (copy_from_user(&k_sinfo, u_sinfo, sizeof(ifslave)))
3510 return -EFAULT;
3512 res = bond_slave_info_query(bond_dev, &k_sinfo);
3513 if (res == 0 &&
3514 copy_to_user(u_sinfo, &k_sinfo, sizeof(ifslave)))
3515 return -EFAULT;
3517 return res;
3518 default:
3519 break;
3520 }
3522 net = dev_net(bond_dev);
3524 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
3525 return -EPERM;
3527 slave_dev = __dev_get_by_name(net, ifr->ifr_slave);
3529 netdev_dbg(bond_dev, "slave_dev=%p:\n", slave_dev);
3531 if (!slave_dev)
3532 return -ENODEV;
3534 netdev_dbg(bond_dev, "slave_dev->name=%s:\n", slave_dev->name);
3535 switch (cmd) {
3536 case BOND_ENSLAVE_OLD:
3537 case SIOCBONDENSLAVE:
3538 res = bond_enslave(bond_dev, slave_dev, NULL);
3539 break;
3540 case BOND_RELEASE_OLD:
3541 case SIOCBONDRELEASE:
3542 res = bond_release(bond_dev, slave_dev);
3543 break;
3544 case BOND_SETHWADDR_OLD:
3545 case SIOCBONDSETHWADDR:
3546 bond_set_dev_addr(bond_dev, slave_dev);
3547 res = 0;
3548 break;
3549 case BOND_CHANGE_ACTIVE_OLD:
3550 case SIOCBONDCHANGEACTIVE:
3551 bond_opt_initstr(&newval, slave_dev->name);
3552 res = __bond_opt_set_notify(bond, BOND_OPT_ACTIVE_SLAVE,
3553 &newval);
3554 break;
3555 default:
3556 res = -EOPNOTSUPP;
3557 }
3559 return res;
3560 }
3562 static void bond_change_rx_flags(struct net_device *bond_dev, int change)
3563 {
3564 struct bonding *bond = netdev_priv(bond_dev);
3566 if (change & IFF_PROMISC)
3567 bond_set_promiscuity(bond,
3568 bond_dev->flags & IFF_PROMISC ? 1 : -1);
3570 if (change & IFF_ALLMULTI)
3571 bond_set_allmulti(bond,
3572 bond_dev->flags & IFF_ALLMULTI ? 1 : -1);
3573 }
3575 static void bond_set_rx_mode(struct net_device *bond_dev)
3576 {
3577 struct bonding *bond = netdev_priv(bond_dev);
3578 struct list_head *iter;
3579 struct slave *slave;
3581 rcu_read_lock();
3582 if (bond_uses_primary(bond)) {
3583 slave = rcu_dereference(bond->curr_active_slave);
3584 if (slave) {
3585 dev_uc_sync(slave->dev, bond_dev);
3586 dev_mc_sync(slave->dev, bond_dev);
3587 }
3588 } else {
3589 bond_for_each_slave_rcu(bond, slave, iter) {
3590 dev_uc_sync_multiple(slave->dev, bond_dev);
3591 dev_mc_sync_multiple(slave->dev, bond_dev);
3592 }
3593 }
3594 rcu_read_unlock();
3595 }
3597 static int bond_neigh_init(struct neighbour *n)
3598 {
3599 struct bonding *bond = netdev_priv(n->dev);
3600 const struct net_device_ops *slave_ops;
3601 struct neigh_parms parms;
3602 struct slave *slave;
3603 int ret;
3605 slave = bond_first_slave(bond);
3606 if (!slave)
3607 return 0;
3608 slave_ops = slave->dev->netdev_ops;
3609 if (!slave_ops->ndo_neigh_setup)
3610 return 0;
3612 parms.neigh_setup = NULL;
3613 parms.neigh_cleanup = NULL;
3614 ret = slave_ops->ndo_neigh_setup(slave->dev, &parms);
3615 if (ret)
3616 return ret;
3618 /* Assign slave's neigh_cleanup to neighbour in case cleanup is called
3619 * after the last slave has been detached. Assumes that all slaves
3620 * utilize the same neigh_cleanup (true at this writing as only user
3621 * is ipoib).
3622 */
3623 n->parms->neigh_cleanup = parms.neigh_cleanup;
3625 if (!parms.neigh_setup)
3626 return 0;
3628 return parms.neigh_setup(n);
3629 }
3631 /* The bonding ndo_neigh_setup is called at init time beofre any
3632 * slave exists. So we must declare proxy setup function which will
3633 * be used at run time to resolve the actual slave neigh param setup.
3634 *
3635 * It's also called by master devices (such as vlans) to setup their
3636 * underlying devices. In that case - do nothing, we're already set up from
3637 * our init.
3638 */
3639 static int bond_neigh_setup(struct net_device *dev,
3640 struct neigh_parms *parms)
3641 {
3642 /* modify only our neigh_parms */
3643 if (parms->dev == dev)
3644 parms->neigh_setup = bond_neigh_init;
3646 return 0;
3647 }
3649 /* Change the MTU of all of a master's slaves to match the master */
3650 static int bond_change_mtu(struct net_device *bond_dev, int new_mtu)
3651 {
3652 struct bonding *bond = netdev_priv(bond_dev);
3653 struct slave *slave, *rollback_slave;
3654 struct list_head *iter;
3655 int res = 0;
3657 netdev_dbg(bond_dev, "bond=%p, new_mtu=%d\n", bond, new_mtu);
3659 bond_for_each_slave(bond, slave, iter) {
3660 netdev_dbg(bond_dev, "s %p c_m %p\n",
3661 slave, slave->dev->netdev_ops->ndo_change_mtu);
3663 res = dev_set_mtu(slave->dev, new_mtu);
3665 if (res) {
3666 /* If we failed to set the slave's mtu to the new value
3667 * we must abort the operation even in ACTIVE_BACKUP
3668 * mode, because if we allow the backup slaves to have
3669 * different mtu values than the active slave we'll
3670 * need to change their mtu when doing a failover. That
3671 * means changing their mtu from timer context, which
3672 * is probably not a good idea.
3673 */
3674 netdev_dbg(bond_dev, "err %d %s\n", res,
3675 slave->dev->name);
3676 goto unwind;
3677 }
3678 }
3680 bond_dev->mtu = new_mtu;
3682 return 0;
3684 unwind:
3685 /* unwind from head to the slave that failed */
3686 bond_for_each_slave(bond, rollback_slave, iter) {
3687 int tmp_res;
3689 if (rollback_slave == slave)
3690 break;
3692 tmp_res = dev_set_mtu(rollback_slave->dev, bond_dev->mtu);
3693 if (tmp_res) {
3694 netdev_dbg(bond_dev, "unwind err %d dev %s\n",
3695 tmp_res, rollback_slave->dev->name);
3696 }
3697 }
3699 return res;
3700 }
3702 /* Change HW address
3703 *
3704 * Note that many devices must be down to change the HW address, and
3705 * downing the master releases all slaves. We can make bonds full of
3706 * bonding devices to test this, however.
3707 */
3708 static int bond_set_mac_address(struct net_device *bond_dev, void *addr)
3709 {
3710 struct bonding *bond = netdev_priv(bond_dev);
3711 struct slave *slave, *rollback_slave;
3712 struct sockaddr_storage *ss = addr, tmp_ss;
3713 struct list_head *iter;
3714 int res = 0;
3716 if (BOND_MODE(bond) == BOND_MODE_ALB)
3717 return bond_alb_set_mac_address(bond_dev, addr);
3720 netdev_dbg(bond_dev, "bond=%p\n", bond);
3722 /* If fail_over_mac is enabled, do nothing and return success.
3723 * Returning an error causes ifenslave to fail.
3724 */
3725 if (bond->params.fail_over_mac &&
3726 BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP)
3727 return 0;
3729 if (!is_valid_ether_addr(ss->__data))
3730 return -EADDRNOTAVAIL;
3732 bond_for_each_slave(bond, slave, iter) {
3733 netdev_dbg(bond_dev, "slave %p %s\n", slave, slave->dev->name);
3734 res = dev_set_mac_address(slave->dev, addr);
3735 if (res) {
3736 /* TODO: consider downing the slave
3737 * and retry ?
3738 * User should expect communications
3739 * breakage anyway until ARP finish
3740 * updating, so...
3741 */
3742 netdev_dbg(bond_dev, "err %d %s\n", res, slave->dev->name);
3743 goto unwind;
3744 }
3745 }
3747 /* success */
3748 memcpy(bond_dev->dev_addr, ss->__data, bond_dev->addr_len);
3749 return 0;
3751 unwind:
3752 memcpy(tmp_ss.__data, bond_dev->dev_addr, bond_dev->addr_len);
3753 tmp_ss.ss_family = bond_dev->type;
3755 /* unwind from head to the slave that failed */
3756 bond_for_each_slave(bond, rollback_slave, iter) {
3757 int tmp_res;
3759 if (rollback_slave == slave)
3760 break;
3762 tmp_res = dev_set_mac_address(rollback_slave->dev,
3763 (struct sockaddr *)&tmp_ss);
3764 if (tmp_res) {
3765 netdev_dbg(bond_dev, "unwind err %d dev %s\n",
3766 tmp_res, rollback_slave->dev->name);
3767 }
3768 }
3770 return res;
3771 }
3773 /**
3774 * bond_xmit_slave_id - transmit skb through slave with slave_id
3775 * @bond: bonding device that is transmitting
3776 * @skb: buffer to transmit
3777 * @slave_id: slave id up to slave_cnt-1 through which to transmit
3778 *
3779 * This function tries to transmit through slave with slave_id but in case
3780 * it fails, it tries to find the first available slave for transmission.
3781 * The skb is consumed in all cases, thus the function is void.
3782 */
3783 static void bond_xmit_slave_id(struct bonding *bond, struct sk_buff *skb, int slave_id)
3784 {
3785 struct list_head *iter;
3786 struct slave *slave;
3787 int i = slave_id;
3789 /* Here we start from the slave with slave_id */
3790 bond_for_each_slave_rcu(bond, slave, iter) {
3791 if (--i < 0) {
3792 if (bond_slave_can_tx(slave)) {
3793 bond_dev_queue_xmit(bond, skb, slave->dev);
3794 return;
3795 }
3796 }
3797 }
3799 /* Here we start from the first slave up to slave_id */
3800 i = slave_id;
3801 bond_for_each_slave_rcu(bond, slave, iter) {
3802 if (--i < 0)
3803 break;
3804 if (bond_slave_can_tx(slave)) {
3805 bond_dev_queue_xmit(bond, skb, slave->dev);
3806 return;
3807 }
3808 }
3809 /* no slave that can tx has been found */
3810 bond_tx_drop(bond->dev, skb);
3811 }
3813 /**
3814 * bond_rr_gen_slave_id - generate slave id based on packets_per_slave
3815 * @bond: bonding device to use
3816 *
3817 * Based on the value of the bonding device's packets_per_slave parameter
3818 * this function generates a slave id, which is usually used as the next
3819 * slave to transmit through.
3820 */
3821 static u32 bond_rr_gen_slave_id(struct bonding *bond)
3822 {
3823 u32 slave_id;
3824 struct reciprocal_value reciprocal_packets_per_slave;
3825 int packets_per_slave = bond->params.packets_per_slave;
3827 switch (packets_per_slave) {
3828 case 0:
3829 slave_id = prandom_u32();
3830 break;
3831 case 1:
3832 slave_id = bond->rr_tx_counter;
3833 break;
3834 default:
3835 reciprocal_packets_per_slave =
3836 bond->params.reciprocal_packets_per_slave;
3837 slave_id = reciprocal_divide(bond->rr_tx_counter,
3838 reciprocal_packets_per_slave);
3839 break;
3840 }
3841 bond->rr_tx_counter++;
3843 return slave_id;
3844 }
3846 static netdev_tx_t bond_xmit_roundrobin(struct sk_buff *skb,
3847 struct net_device *bond_dev)
3848 {
3849 struct bonding *bond = netdev_priv(bond_dev);
3850 struct iphdr *iph = ip_hdr(skb);
3851 struct slave *slave;
3852 u32 slave_id;
3854 /* Start with the curr_active_slave that joined the bond as the
3855 * default for sending IGMP traffic. For failover purposes one
3856 * needs to maintain some consistency for the interface that will
3857 * send the join/membership reports. The curr_active_slave found
3858 * will send all of this type of traffic.
3859 */
3860 if (iph->protocol == IPPROTO_IGMP && skb->protocol == htons(ETH_P_IP)) {
3861 slave = rcu_dereference(bond->curr_active_slave);
3862 if (slave)
3863 bond_dev_queue_xmit(bond, skb, slave->dev);
3864 else
3865 bond_xmit_slave_id(bond, skb, 0);
3866 } else {
3867 int slave_cnt = READ_ONCE(bond->slave_cnt);
3869 if (likely(slave_cnt)) {
3870 slave_id = bond_rr_gen_slave_id(bond);
3871 bond_xmit_slave_id(bond, skb, slave_id % slave_cnt);
3872 } else {
3873 bond_tx_drop(bond_dev, skb);
3874 }
3875 }
3877 return NETDEV_TX_OK;
3878 }
3880 /* In active-backup mode, we know that bond->curr_active_slave is always valid if
3881 * the bond has a usable interface.
3882 */
3883 static netdev_tx_t bond_xmit_activebackup(struct sk_buff *skb,
3884 struct net_device *bond_dev)
3885 {
3886 struct bonding *bond = netdev_priv(bond_dev);
3887 struct slave *slave;
3889 slave = rcu_dereference(bond->curr_active_slave);
3890 if (slave)
3891 bond_dev_queue_xmit(bond, skb, slave->dev);
3892 else
3893 bond_tx_drop(bond_dev, skb);
3895 return NETDEV_TX_OK;
3896 }
3898 /* Use this to update slave_array when (a) it's not appropriate to update
3899 * slave_array right away (note that update_slave_array() may sleep)
3900 * and / or (b) RTNL is not held.
3901 */
3902 void bond_slave_arr_work_rearm(struct bonding *bond, unsigned long delay)
3903 {
3904 queue_delayed_work(bond->wq, &bond->slave_arr_work, delay);
3905 }
3907 /* Slave array work handler. Holds only RTNL */
3908 static void bond_slave_arr_handler(struct work_struct *work)
3909 {
3910 struct bonding *bond = container_of(work, struct bonding,
3911 slave_arr_work.work);
3912 int ret;
3914 if (!rtnl_trylock())
3915 goto err;
3917 ret = bond_update_slave_arr(bond, NULL);
3918 rtnl_unlock();
3919 if (ret) {
3920 pr_warn_ratelimited("Failed to update slave array from WT\n");
3921 goto err;
3922 }
3923 return;
3925 err:
3926 bond_slave_arr_work_rearm(bond, 1);
3927 }
3929 /* Build the usable slaves array in control path for modes that use xmit-hash
3930 * to determine the slave interface -
3931 * (a) BOND_MODE_8023AD
3932 * (b) BOND_MODE_XOR
3933 * (c) (BOND_MODE_TLB || BOND_MODE_ALB) && tlb_dynamic_lb == 0
3934 *
3935 * The caller is expected to hold RTNL only and NO other lock!
3936 */
3937 int bond_update_slave_arr(struct bonding *bond, struct slave *skipslave)
3938 {
3939 struct slave *slave;
3940 struct list_head *iter;
3941 struct bond_up_slave *new_arr, *old_arr;
3942 int agg_id = 0;
3943 int ret = 0;
3945 #ifdef CONFIG_LOCKDEP
3946 WARN_ON(lockdep_is_held(&bond->mode_lock));
3947 #endif
3949 new_arr = kzalloc(offsetof(struct bond_up_slave, arr[bond->slave_cnt]),
3950 GFP_KERNEL);
3951 if (!new_arr) {
3952 ret = -ENOMEM;
3953 pr_err("Failed to build slave-array.\n");
3954 goto out;
3955 }
3956 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
3957 struct ad_info ad_info;
3959 if (bond_3ad_get_active_agg_info(bond, &ad_info)) {
3960 pr_debug("bond_3ad_get_active_agg_info failed\n");
3961 kfree_rcu(new_arr, rcu);
3962 /* No active aggragator means it's not safe to use
3963 * the previous array.
3964 */
3965 old_arr = rtnl_dereference(bond->slave_arr);
3966 if (old_arr) {
3967 RCU_INIT_POINTER(bond->slave_arr, NULL);
3968 kfree_rcu(old_arr, rcu);
3969 }
3970 goto out;
3971 }
3972 agg_id = ad_info.aggregator_id;
3973 }
3974 bond_for_each_slave(bond, slave, iter) {
3975 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
3976 struct aggregator *agg;
3978 agg = SLAVE_AD_INFO(slave)->port.aggregator;
3979 if (!agg || agg->aggregator_identifier != agg_id)
3980 continue;
3981 }
3982 if (!bond_slave_can_tx(slave))
3983 continue;
3984 if (skipslave == slave)
3985 continue;
3987 netdev_dbg(bond->dev,
3988 "Adding slave dev %s to tx hash array[%d]\n",
3989 slave->dev->name, new_arr->count);
3991 new_arr->arr[new_arr->count++] = slave;
3992 }
3994 old_arr = rtnl_dereference(bond->slave_arr);
3995 rcu_assign_pointer(bond->slave_arr, new_arr);
3996 if (old_arr)
3997 kfree_rcu(old_arr, rcu);
3998 out:
3999 if (ret != 0 && skipslave) {
4000 int idx;
4002 /* Rare situation where caller has asked to skip a specific
4003 * slave but allocation failed (most likely!). BTW this is
4004 * only possible when the call is initiated from
4005 * __bond_release_one(). In this situation; overwrite the
4006 * skipslave entry in the array with the last entry from the
4007 * array to avoid a situation where the xmit path may choose
4008 * this to-be-skipped slave to send a packet out.
4009 */
4010 old_arr = rtnl_dereference(bond->slave_arr);
4011 for (idx = 0; idx < old_arr->count; idx++) {
4012 if (skipslave == old_arr->arr[idx]) {
4013 old_arr->arr[idx] =
4014 old_arr->arr[old_arr->count-1];
4015 old_arr->count--;
4016 break;
4017 }
4018 }
4019 }
4020 return ret;
4021 }
4023 /* Use this Xmit function for 3AD as well as XOR modes. The current
4024 * usable slave array is formed in the control path. The xmit function
4025 * just calculates hash and sends the packet out.
4026 */
4027 static netdev_tx_t bond_3ad_xor_xmit(struct sk_buff *skb,
4028 struct net_device *dev)
4029 {
4030 struct bonding *bond = netdev_priv(dev);
4031 struct slave *slave;
4032 struct bond_up_slave *slaves;
4033 unsigned int count;
4035 slaves = rcu_dereference(bond->slave_arr);
4036 count = slaves ? READ_ONCE(slaves->count) : 0;
4037 if (likely(count)) {
4038 slave = slaves->arr[bond_xmit_hash(bond, skb) % count];
4039 bond_dev_queue_xmit(bond, skb, slave->dev);
4040 } else {
4041 bond_tx_drop(dev, skb);
4042 }
4044 return NETDEV_TX_OK;
4045 }
4047 /* in broadcast mode, we send everything to all usable interfaces. */
4048 static netdev_tx_t bond_xmit_broadcast(struct sk_buff *skb,
4049 struct net_device *bond_dev)
4050 {
4051 struct bonding *bond = netdev_priv(bond_dev);
4052 struct slave *slave = NULL;
4053 struct list_head *iter;
4055 bond_for_each_slave_rcu(bond, slave, iter) {
4056 if (bond_is_last_slave(bond, slave))
4057 break;
4058 if (bond_slave_is_up(slave) && slave->link == BOND_LINK_UP) {
4059 struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
4061 if (!skb2) {
4062 net_err_ratelimited("%s: Error: %s: skb_clone() failed\n",
4063 bond_dev->name, __func__);
4064 continue;
4065 }
4066 bond_dev_queue_xmit(bond, skb2, slave->dev);
4067 }
4068 }
4069 if (slave && bond_slave_is_up(slave) && slave->link == BOND_LINK_UP)
4070 bond_dev_queue_xmit(bond, skb, slave->dev);
4071 else
4072 bond_tx_drop(bond_dev, skb);
4074 return NETDEV_TX_OK;
4075 }
4077 /*------------------------- Device initialization ---------------------------*/
4079 /* Lookup the slave that corresponds to a qid */
4080 static inline int bond_slave_override(struct bonding *bond,
4081 struct sk_buff *skb)
4082 {
4083 struct slave *slave = NULL;
4084 struct list_head *iter;
4086 if (!skb_rx_queue_recorded(skb))
4087 return 1;
4089 /* Find out if any slaves have the same mapping as this skb. */
4090 bond_for_each_slave_rcu(bond, slave, iter) {
4091 if (slave->queue_id == skb_get_queue_mapping(skb)) {
4092 if (bond_slave_is_up(slave) &&
4093 slave->link == BOND_LINK_UP) {
4094 bond_dev_queue_xmit(bond, skb, slave->dev);
4095 return 0;
4096 }
4097 /* If the slave isn't UP, use default transmit policy. */
4098 break;
4099 }
4100 }
4102 return 1;
4103 }
4106 static u16 bond_select_queue(struct net_device *dev, struct sk_buff *skb,
4107 struct net_device *sb_dev,
4108 select_queue_fallback_t fallback)
4109 {
4110 /* This helper function exists to help dev_pick_tx get the correct
4111 * destination queue. Using a helper function skips a call to
4112 * skb_tx_hash and will put the skbs in the queue we expect on their
4113 * way down to the bonding driver.
4114 */
4115 u16 txq = skb_rx_queue_recorded(skb) ? skb_get_rx_queue(skb) : 0;
4117 /* Save the original txq to restore before passing to the driver */
4118 qdisc_skb_cb(skb)->slave_dev_queue_mapping = skb_get_queue_mapping(skb);
4120 if (unlikely(txq >= dev->real_num_tx_queues)) {
4121 do {
4122 txq -= dev->real_num_tx_queues;
4123 } while (txq >= dev->real_num_tx_queues);
4124 }
4125 return txq;
4126 }
4128 static netdev_tx_t __bond_start_xmit(struct sk_buff *skb, struct net_device *dev)
4129 {
4130 struct bonding *bond = netdev_priv(dev);
4132 if (bond_should_override_tx_queue(bond) &&
4133 !bond_slave_override(bond, skb))
4134 return NETDEV_TX_OK;
4136 switch (BOND_MODE(bond)) {
4137 case BOND_MODE_ROUNDROBIN:
4138 return bond_xmit_roundrobin(skb, dev);
4139 case BOND_MODE_ACTIVEBACKUP:
4140 return bond_xmit_activebackup(skb, dev);
4141 case BOND_MODE_8023AD:
4142 case BOND_MODE_XOR:
4143 return bond_3ad_xor_xmit(skb, dev);
4144 case BOND_MODE_BROADCAST:
4145 return bond_xmit_broadcast(skb, dev);
4146 case BOND_MODE_ALB:
4147 return bond_alb_xmit(skb, dev);
4148 case BOND_MODE_TLB:
4149 return bond_tlb_xmit(skb, dev);
4150 default:
4151 /* Should never happen, mode already checked */
4152 netdev_err(dev, "Unknown bonding mode %d\n", BOND_MODE(bond));
4153 WARN_ON_ONCE(1);
4154 bond_tx_drop(dev, skb);
4155 return NETDEV_TX_OK;
4156 }
4157 }
4159 static netdev_tx_t bond_start_xmit(struct sk_buff *skb, struct net_device *dev)
4160 {
4161 struct bonding *bond = netdev_priv(dev);
4162 netdev_tx_t ret = NETDEV_TX_OK;
4164 /* If we risk deadlock from transmitting this in the
4165 * netpoll path, tell netpoll to queue the frame for later tx
4166 */
4167 if (unlikely(is_netpoll_tx_blocked(dev)))
4168 return NETDEV_TX_BUSY;
4170 rcu_read_lock();
4171 if (bond_has_slaves(bond))
4172 ret = __bond_start_xmit(skb, dev);
4173 else
4174 bond_tx_drop(dev, skb);
4175 rcu_read_unlock();
4177 return ret;
4178 }
4180 static int bond_ethtool_get_link_ksettings(struct net_device *bond_dev,
4181 struct ethtool_link_ksettings *cmd)
4182 {
4183 struct bonding *bond = netdev_priv(bond_dev);
4184 unsigned long speed = 0;
4185 struct list_head *iter;
4186 struct slave *slave;
4188 cmd->base.duplex = DUPLEX_UNKNOWN;
4189 cmd->base.port = PORT_OTHER;
4191 /* Since bond_slave_can_tx returns false for all inactive or down slaves, we
4192 * do not need to check mode. Though link speed might not represent
4193 * the true receive or transmit bandwidth (not all modes are symmetric)
4194 * this is an accurate maximum.
4195 */
4196 bond_for_each_slave(bond, slave, iter) {
4197 if (bond_slave_can_tx(slave)) {
4198 if (slave->speed != SPEED_UNKNOWN)
4199 speed += slave->speed;
4200 if (cmd->base.duplex == DUPLEX_UNKNOWN &&
4201 slave->duplex != DUPLEX_UNKNOWN)
4202 cmd->base.duplex = slave->duplex;
4203 }
4204 }
4205 cmd->base.speed = speed ? : SPEED_UNKNOWN;
4207 return 0;
4208 }
4210 static void bond_ethtool_get_drvinfo(struct net_device *bond_dev,
4211 struct ethtool_drvinfo *drvinfo)
4212 {
4213 strlcpy(drvinfo->driver, DRV_NAME, sizeof(drvinfo->driver));
4214 strlcpy(drvinfo->version, DRV_VERSION, sizeof(drvinfo->version));
4215 snprintf(drvinfo->fw_version, sizeof(drvinfo->fw_version), "%d",
4216 BOND_ABI_VERSION);
4217 }
4219 static const struct ethtool_ops bond_ethtool_ops = {
4220 .get_drvinfo = bond_ethtool_get_drvinfo,
4221 .get_link = ethtool_op_get_link,
4222 .get_link_ksettings = bond_ethtool_get_link_ksettings,
4223 };
4225 static const struct net_device_ops bond_netdev_ops = {
4226 .ndo_init = bond_init,
4227 .ndo_uninit = bond_uninit,
4228 .ndo_open = bond_open,
4229 .ndo_stop = bond_close,
4230 .ndo_start_xmit = bond_start_xmit,
4231 .ndo_select_queue = bond_select_queue,
4232 .ndo_get_stats64 = bond_get_stats,
4233 .ndo_do_ioctl = bond_do_ioctl,
4234 .ndo_change_rx_flags = bond_change_rx_flags,
4235 .ndo_set_rx_mode = bond_set_rx_mode,
4236 .ndo_change_mtu = bond_change_mtu,
4237 .ndo_set_mac_address = bond_set_mac_address,
4238 .ndo_neigh_setup = bond_neigh_setup,
4239 .ndo_vlan_rx_add_vid = bond_vlan_rx_add_vid,
4240 .ndo_vlan_rx_kill_vid = bond_vlan_rx_kill_vid,
4241 .ndo_get_lock_subclass = bond_get_nest_level,
4242 #ifdef CONFIG_NET_POLL_CONTROLLER
4243 .ndo_netpoll_setup = bond_netpoll_setup,
4244 .ndo_netpoll_cleanup = bond_netpoll_cleanup,
4245 .ndo_poll_controller = bond_poll_controller,
4246 #endif
4247 .ndo_add_slave = bond_enslave,
4248 .ndo_del_slave = bond_release,
4249 .ndo_fix_features = bond_fix_features,
4250 .ndo_features_check = passthru_features_check,
4251 };
4253 static const struct device_type bond_type = {
4254 .name = "bond",
4255 };
4257 static void bond_destructor(struct net_device *bond_dev)
4258 {
4259 struct bonding *bond = netdev_priv(bond_dev);
4260 if (bond->wq)
4261 destroy_workqueue(bond->wq);
4262 }
4264 void bond_setup(struct net_device *bond_dev)
4265 {
4266 struct bonding *bond = netdev_priv(bond_dev);
4268 spin_lock_init(&bond->mode_lock);
4269 spin_lock_init(&bond->stats_lock);
4270 bond->params = bonding_defaults;
4272 /* Initialize pointers */
4273 bond->dev = bond_dev;
4275 /* Initialize the device entry points */
4276 ether_setup(bond_dev);
4277 bond_dev->max_mtu = ETH_MAX_MTU;
4278 bond_dev->netdev_ops = &bond_netdev_ops;
4279 bond_dev->ethtool_ops = &bond_ethtool_ops;
4281 bond_dev->needs_free_netdev = true;
4282 bond_dev->priv_destructor = bond_destructor;
4284 SET_NETDEV_DEVTYPE(bond_dev, &bond_type);
4286 /* Initialize the device options */
4287 bond_dev->flags |= IFF_MASTER;
4288 bond_dev->priv_flags |= IFF_BONDING | IFF_UNICAST_FLT | IFF_NO_QUEUE;
4289 bond_dev->priv_flags &= ~(IFF_XMIT_DST_RELEASE | IFF_TX_SKB_SHARING);
4291 /* don't acquire bond device's netif_tx_lock when transmitting */
4292 bond_dev->features |= NETIF_F_LLTX;
4294 /* By default, we declare the bond to be fully
4295 * VLAN hardware accelerated capable. Special
4296 * care is taken in the various xmit functions
4297 * when there are slaves that are not hw accel
4298 * capable
4299 */
4301 /* Don't allow bond devices to change network namespaces. */
4302 bond_dev->features |= NETIF_F_NETNS_LOCAL;
4304 bond_dev->hw_features = BOND_VLAN_FEATURES |
4305 NETIF_F_HW_VLAN_CTAG_TX |
4306 NETIF_F_HW_VLAN_CTAG_RX |
4307 NETIF_F_HW_VLAN_CTAG_FILTER;
4309 bond_dev->hw_features |= NETIF_F_GSO_ENCAP_ALL | NETIF_F_GSO_UDP_L4;
4310 bond_dev->features |= bond_dev->hw_features;
4311 }
4313 /* Destroy a bonding device.
4314 * Must be under rtnl_lock when this function is called.
4315 */
4316 static void bond_uninit(struct net_device *bond_dev)
4317 {
4318 struct bonding *bond = netdev_priv(bond_dev);
4319 struct list_head *iter;
4320 struct slave *slave;
4321 struct bond_up_slave *arr;
4323 bond_netpoll_cleanup(bond_dev);
4325 /* Release the bonded slaves */
4326 bond_for_each_slave(bond, slave, iter)
4327 __bond_release_one(bond_dev, slave->dev, true, true);
4328 netdev_info(bond_dev, "Released all slaves\n");
4330 arr = rtnl_dereference(bond->slave_arr);
4331 if (arr) {
4332 RCU_INIT_POINTER(bond->slave_arr, NULL);
4333 kfree_rcu(arr, rcu);
4334 }
4336 list_del(&bond->bond_list);
4338 bond_debug_unregister(bond);
4339 }
4341 /*------------------------- Module initialization ---------------------------*/
4343 static int bond_check_params(struct bond_params *params)
4344 {
4345 int arp_validate_value, fail_over_mac_value, primary_reselect_value, i;
4346 struct bond_opt_value newval;
4347 const struct bond_opt_value *valptr;
4348 int arp_all_targets_value = 0;
4349 u16 ad_actor_sys_prio = 0;
4350 u16 ad_user_port_key = 0;
4351 __be32 arp_target[BOND_MAX_ARP_TARGETS] = { 0 };
4352 int arp_ip_count;
4353 int bond_mode = BOND_MODE_ROUNDROBIN;
4354 int xmit_hashtype = BOND_XMIT_POLICY_LAYER2;
4355 int lacp_fast = 0;
4356 int tlb_dynamic_lb;
4358 /* Convert string parameters. */
4359 if (mode) {
4360 bond_opt_initstr(&newval, mode);
4361 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_MODE), &newval);
4362 if (!valptr) {
4363 pr_err("Error: Invalid bonding mode \"%s\"\n", mode);
4364 return -EINVAL;
4365 }
4366 bond_mode = valptr->value;
4367 }
4369 if (xmit_hash_policy) {
4370 if (bond_mode == BOND_MODE_ROUNDROBIN ||
4371 bond_mode == BOND_MODE_ACTIVEBACKUP ||
4372 bond_mode == BOND_MODE_BROADCAST) {
4373 pr_info("xmit_hash_policy param is irrelevant in mode %s\n",
4374 bond_mode_name(bond_mode));
4375 } else {
4376 bond_opt_initstr(&newval, xmit_hash_policy);
4377 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_XMIT_HASH),
4378 &newval);
4379 if (!valptr) {
4380 pr_err("Error: Invalid xmit_hash_policy \"%s\"\n",
4381 xmit_hash_policy);
4382 return -EINVAL;
4383 }
4384 xmit_hashtype = valptr->value;
4385 }
4386 }
4388 if (lacp_rate) {
4389 if (bond_mode != BOND_MODE_8023AD) {
4390 pr_info("lacp_rate param is irrelevant in mode %s\n",
4391 bond_mode_name(bond_mode));
4392 } else {
4393 bond_opt_initstr(&newval, lacp_rate);
4394 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_LACP_RATE),
4395 &newval);
4396 if (!valptr) {
4397 pr_err("Error: Invalid lacp rate \"%s\"\n",
4398 lacp_rate);
4399 return -EINVAL;
4400 }
4401 lacp_fast = valptr->value;
4402 }
4403 }
4405 if (ad_select) {
4406 bond_opt_initstr(&newval, ad_select);
4407 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_AD_SELECT),
4408 &newval);
4409 if (!valptr) {
4410 pr_err("Error: Invalid ad_select \"%s\"\n", ad_select);
4411 return -EINVAL;
4412 }
4413 params->ad_select = valptr->value;
4414 if (bond_mode != BOND_MODE_8023AD)
4415 pr_warn("ad_select param only affects 802.3ad mode\n");
4416 } else {
4417 params->ad_select = BOND_AD_STABLE;
4418 }
4420 if (max_bonds < 0) {
4421 pr_warn("Warning: max_bonds (%d) not in range %d-%d, so it was reset to BOND_DEFAULT_MAX_BONDS (%d)\n",
4422 max_bonds, 0, INT_MAX, BOND_DEFAULT_MAX_BONDS);
4423 max_bonds = BOND_DEFAULT_MAX_BONDS;
4424 }
4426 if (miimon < 0) {
4427 pr_warn("Warning: miimon module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4428 miimon, INT_MAX);
4429 miimon = 0;
4430 }
4432 if (updelay < 0) {
4433 pr_warn("Warning: updelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4434 updelay, INT_MAX);
4435 updelay = 0;
4436 }
4438 if (downdelay < 0) {
4439 pr_warn("Warning: downdelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4440 downdelay, INT_MAX);
4441 downdelay = 0;
4442 }
4444 if ((use_carrier != 0) && (use_carrier != 1)) {
4445 pr_warn("Warning: use_carrier module parameter (%d), not of valid value (0/1), so it was set to 1\n",
4446 use_carrier);
4447 use_carrier = 1;
4448 }
4450 if (num_peer_notif < 0 || num_peer_notif > 255) {
4451 pr_warn("Warning: num_grat_arp/num_unsol_na (%d) not in range 0-255 so it was reset to 1\n",
4452 num_peer_notif);
4453 num_peer_notif = 1;
4454 }
4456 /* reset values for 802.3ad/TLB/ALB */
4457 if (!bond_mode_uses_arp(bond_mode)) {
4458 if (!miimon) {
4459 pr_warn("Warning: miimon must be specified, otherwise bonding will not detect link failure, speed and duplex which are essential for 802.3ad operation\n");
4460 pr_warn("Forcing miimon to 100msec\n");
4461 miimon = BOND_DEFAULT_MIIMON;
4462 }
4463 }
4465 if (tx_queues < 1 || tx_queues > 255) {
4466 pr_warn("Warning: tx_queues (%d) should be between 1 and 255, resetting to %d\n",
4467 tx_queues, BOND_DEFAULT_TX_QUEUES);
4468 tx_queues = BOND_DEFAULT_TX_QUEUES;
4469 }
4471 if ((all_slaves_active != 0) && (all_slaves_active != 1)) {
4472 pr_warn("Warning: all_slaves_active module parameter (%d), not of valid value (0/1), so it was set to 0\n",
4473 all_slaves_active);
4474 all_slaves_active = 0;
4475 }
4477 if (resend_igmp < 0 || resend_igmp > 255) {
4478 pr_warn("Warning: resend_igmp (%d) should be between 0 and 255, resetting to %d\n",
4479 resend_igmp, BOND_DEFAULT_RESEND_IGMP);
4480 resend_igmp = BOND_DEFAULT_RESEND_IGMP;
4481 }
4483 bond_opt_initval(&newval, packets_per_slave);
4484 if (!bond_opt_parse(bond_opt_get(BOND_OPT_PACKETS_PER_SLAVE), &newval)) {
4485 pr_warn("Warning: packets_per_slave (%d) should be between 0 and %u resetting to 1\n",
4486 packets_per_slave, USHRT_MAX);
4487 packets_per_slave = 1;
4488 }
4490 if (bond_mode == BOND_MODE_ALB) {
4491 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",
4492 updelay);
4493 }
4495 if (!miimon) {
4496 if (updelay || downdelay) {
4497 /* just warn the user the up/down delay will have
4498 * no effect since miimon is zero...
4499 */
4500 pr_warn("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",
4501 updelay, downdelay);
4502 }
4503 } else {
4504 /* don't allow arp monitoring */
4505 if (arp_interval) {
4506 pr_warn("Warning: miimon (%d) and arp_interval (%d) can't be used simultaneously, disabling ARP monitoring\n",
4507 miimon, arp_interval);
4508 arp_interval = 0;
4509 }
4511 if ((updelay % miimon) != 0) {
4512 pr_warn("Warning: updelay (%d) is not a multiple of miimon (%d), updelay rounded to %d ms\n",
4513 updelay, miimon, (updelay / miimon) * miimon);
4514 }
4516 updelay /= miimon;
4518 if ((downdelay % miimon) != 0) {
4519 pr_warn("Warning: downdelay (%d) is not a multiple of miimon (%d), downdelay rounded to %d ms\n",
4520 downdelay, miimon,
4521 (downdelay / miimon) * miimon);
4522 }
4524 downdelay /= miimon;
4525 }
4527 if (arp_interval < 0) {
4528 pr_warn("Warning: arp_interval module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4529 arp_interval, INT_MAX);
4530 arp_interval = 0;
4531 }
4533 for (arp_ip_count = 0, i = 0;
4534 (arp_ip_count < BOND_MAX_ARP_TARGETS) && arp_ip_target[i]; i++) {
4535 __be32 ip;
4537 /* not a complete check, but good enough to catch mistakes */
4538 if (!in4_pton(arp_ip_target[i], -1, (u8 *)&ip, -1, NULL) ||
4539 !bond_is_ip_target_ok(ip)) {
4540 pr_warn("Warning: bad arp_ip_target module parameter (%s), ARP monitoring will not be performed\n",
4541 arp_ip_target[i]);
4542 arp_interval = 0;
4543 } else {
4544 if (bond_get_targets_ip(arp_target, ip) == -1)
4545 arp_target[arp_ip_count++] = ip;
4546 else
4547 pr_warn("Warning: duplicate address %pI4 in arp_ip_target, skipping\n",
4548 &ip);
4549 }
4550 }
4552 if (arp_interval && !arp_ip_count) {
4553 /* don't allow arping if no arp_ip_target given... */
4554 pr_warn("Warning: arp_interval module parameter (%d) specified without providing an arp_ip_target parameter, arp_interval was reset to 0\n",
4555 arp_interval);
4556 arp_interval = 0;
4557 }
4559 if (arp_validate) {
4560 if (!arp_interval) {
4561 pr_err("arp_validate requires arp_interval\n");
4562 return -EINVAL;
4563 }
4565 bond_opt_initstr(&newval, arp_validate);
4566 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_ARP_VALIDATE),
4567 &newval);
4568 if (!valptr) {
4569 pr_err("Error: invalid arp_validate \"%s\"\n",
4570 arp_validate);
4571 return -EINVAL;
4572 }
4573 arp_validate_value = valptr->value;
4574 } else {
4575 arp_validate_value = 0;
4576 }
4578 if (arp_all_targets) {
4579 bond_opt_initstr(&newval, arp_all_targets);
4580 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_ARP_ALL_TARGETS),
4581 &newval);
4582 if (!valptr) {
4583 pr_err("Error: invalid arp_all_targets_value \"%s\"\n",
4584 arp_all_targets);
4585 arp_all_targets_value = 0;
4586 } else {
4587 arp_all_targets_value = valptr->value;
4588 }
4589 }
4591 if (miimon) {
4592 pr_info("MII link monitoring set to %d ms\n", miimon);
4593 } else if (arp_interval) {
4594 valptr = bond_opt_get_val(BOND_OPT_ARP_VALIDATE,
4595 arp_validate_value);
4596 pr_info("ARP monitoring set to %d ms, validate %s, with %d target(s):",
4597 arp_interval, valptr->string, arp_ip_count);
4599 for (i = 0; i < arp_ip_count; i++)
4600 pr_cont(" %s", arp_ip_target[i]);
4602 pr_cont("\n");
4604 } else if (max_bonds) {
4605 /* miimon and arp_interval not set, we need one so things
4606 * work as expected, see bonding.txt for details
4607 */
4608 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");
4609 }
4611 if (primary && !bond_mode_uses_primary(bond_mode)) {
4612 /* currently, using a primary only makes sense
4613 * in active backup, TLB or ALB modes
4614 */
4615 pr_warn("Warning: %s primary device specified but has no effect in %s mode\n",
4616 primary, bond_mode_name(bond_mode));
4617 primary = NULL;
4618 }
4620 if (primary && primary_reselect) {
4621 bond_opt_initstr(&newval, primary_reselect);
4622 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_PRIMARY_RESELECT),
4623 &newval);
4624 if (!valptr) {
4625 pr_err("Error: Invalid primary_reselect \"%s\"\n",
4626 primary_reselect);
4627 return -EINVAL;
4628 }
4629 primary_reselect_value = valptr->value;
4630 } else {
4631 primary_reselect_value = BOND_PRI_RESELECT_ALWAYS;
4632 }
4634 if (fail_over_mac) {
4635 bond_opt_initstr(&newval, fail_over_mac);
4636 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_FAIL_OVER_MAC),
4637 &newval);
4638 if (!valptr) {
4639 pr_err("Error: invalid fail_over_mac \"%s\"\n",
4640 fail_over_mac);
4641 return -EINVAL;
4642 }
4643 fail_over_mac_value = valptr->value;
4644 if (bond_mode != BOND_MODE_ACTIVEBACKUP)
4645 pr_warn("Warning: fail_over_mac only affects active-backup mode\n");
4646 } else {
4647 fail_over_mac_value = BOND_FOM_NONE;
4648 }
4650 bond_opt_initstr(&newval, "default");
4651 valptr = bond_opt_parse(
4652 bond_opt_get(BOND_OPT_AD_ACTOR_SYS_PRIO),
4653 &newval);
4654 if (!valptr) {
4655 pr_err("Error: No ad_actor_sys_prio default value");
4656 return -EINVAL;
4657 }
4658 ad_actor_sys_prio = valptr->value;
4660 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_AD_USER_PORT_KEY),
4661 &newval);
4662 if (!valptr) {
4663 pr_err("Error: No ad_user_port_key default value");
4664 return -EINVAL;
4665 }
4666 ad_user_port_key = valptr->value;
4668 bond_opt_initstr(&newval, "default");
4669 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_TLB_DYNAMIC_LB), &newval);
4670 if (!valptr) {
4671 pr_err("Error: No tlb_dynamic_lb default value");
4672 return -EINVAL;
4673 }
4674 tlb_dynamic_lb = valptr->value;
4676 if (lp_interval == 0) {
4677 pr_warn("Warning: ip_interval must be between 1 and %d, so it was reset to %d\n",
4678 INT_MAX, BOND_ALB_DEFAULT_LP_INTERVAL);
4679 lp_interval = BOND_ALB_DEFAULT_LP_INTERVAL;
4680 }
4682 /* fill params struct with the proper values */
4683 params->mode = bond_mode;
4684 params->xmit_policy = xmit_hashtype;
4685 params->miimon = miimon;
4686 params->num_peer_notif = num_peer_notif;
4687 params->arp_interval = arp_interval;
4688 params->arp_validate = arp_validate_value;
4689 params->arp_all_targets = arp_all_targets_value;
4690 params->updelay = updelay;
4691 params->downdelay = downdelay;
4692 params->use_carrier = use_carrier;
4693 params->lacp_fast = lacp_fast;
4694 params->primary[0] = 0;
4695 params->primary_reselect = primary_reselect_value;
4696 params->fail_over_mac = fail_over_mac_value;
4697 params->tx_queues = tx_queues;
4698 params->all_slaves_active = all_slaves_active;
4699 params->resend_igmp = resend_igmp;
4700 params->min_links = min_links;
4701 params->lp_interval = lp_interval;
4702 params->packets_per_slave = packets_per_slave;
4703 params->tlb_dynamic_lb = tlb_dynamic_lb;
4704 params->ad_actor_sys_prio = ad_actor_sys_prio;
4705 eth_zero_addr(params->ad_actor_system);
4706 params->ad_user_port_key = ad_user_port_key;
4707 if (packets_per_slave > 0) {
4708 params->reciprocal_packets_per_slave =
4709 reciprocal_value(packets_per_slave);
4710 } else {
4711 /* reciprocal_packets_per_slave is unused if
4712 * packets_per_slave is 0 or 1, just initialize it
4713 */
4714 params->reciprocal_packets_per_slave =
4715 (struct reciprocal_value) { 0 };
4716 }
4718 if (primary) {
4719 strncpy(params->primary, primary, IFNAMSIZ);
4720 params->primary[IFNAMSIZ - 1] = 0;
4721 }
4723 memcpy(params->arp_targets, arp_target, sizeof(arp_target));
4725 return 0;
4726 }
4728 /* Called from registration process */
4729 static int bond_init(struct net_device *bond_dev)
4730 {
4731 struct bonding *bond = netdev_priv(bond_dev);
4732 struct bond_net *bn = net_generic(dev_net(bond_dev), bond_net_id);
4734 netdev_dbg(bond_dev, "Begin bond_init\n");
4736 bond->wq = alloc_ordered_workqueue(bond_dev->name, WQ_MEM_RECLAIM);
4737 if (!bond->wq)
4738 return -ENOMEM;
4740 bond->nest_level = SINGLE_DEPTH_NESTING;
4741 netdev_lockdep_set_classes(bond_dev);
4743 list_add_tail(&bond->bond_list, &bn->dev_list);
4745 bond_prepare_sysfs_group(bond);
4747 bond_debug_register(bond);
4749 /* Ensure valid dev_addr */
4750 if (is_zero_ether_addr(bond_dev->dev_addr) &&
4751 bond_dev->addr_assign_type == NET_ADDR_PERM)
4752 eth_hw_addr_random(bond_dev);
4754 return 0;
4755 }
4757 unsigned int bond_get_num_tx_queues(void)
4758 {
4759 return tx_queues;
4760 }
4762 /* Create a new bond based on the specified name and bonding parameters.
4763 * If name is NULL, obtain a suitable "bond%d" name for us.
4764 * Caller must NOT hold rtnl_lock; we need to release it here before we
4765 * set up our sysfs entries.
4766 */
4767 int bond_create(struct net *net, const char *name)
4768 {
4769 struct net_device *bond_dev;
4770 struct bonding *bond;
4771 struct alb_bond_info *bond_info;
4772 int res;
4774 rtnl_lock();
4776 bond_dev = alloc_netdev_mq(sizeof(struct bonding),
4777 name ? name : "bond%d", NET_NAME_UNKNOWN,
4778 bond_setup, tx_queues);
4779 if (!bond_dev) {
4780 pr_err("%s: eek! can't alloc netdev!\n", name);
4781 rtnl_unlock();
4782 return -ENOMEM;
4783 }
4785 /*
4786 * Initialize rx_hashtbl_used_head to RLB_NULL_INDEX.
4787 * It is set to 0 by default which is wrong.
4788 */
4789 bond = netdev_priv(bond_dev);
4790 bond_info = &(BOND_ALB_INFO(bond));
4791 bond_info->rx_hashtbl_used_head = RLB_NULL_INDEX;
4793 dev_net_set(bond_dev, net);
4794 bond_dev->rtnl_link_ops = &bond_link_ops;
4796 res = register_netdevice(bond_dev);
4798 netif_carrier_off(bond_dev);
4800 bond_work_init_all(bond);
4802 rtnl_unlock();
4803 if (res < 0)
4804 free_netdev(bond_dev);
4805 return res;
4806 }
4808 static int __net_init bond_net_init(struct net *net)
4809 {
4810 struct bond_net *bn = net_generic(net, bond_net_id);
4812 bn->net = net;
4813 INIT_LIST_HEAD(&bn->dev_list);
4815 bond_create_proc_dir(bn);
4816 bond_create_sysfs(bn);
4818 return 0;
4819 }
4821 static void __net_exit bond_net_exit(struct net *net)
4822 {
4823 struct bond_net *bn = net_generic(net, bond_net_id);
4824 struct bonding *bond, *tmp_bond;
4825 LIST_HEAD(list);
4827 bond_destroy_sysfs(bn);
4829 /* Kill off any bonds created after unregistering bond rtnl ops */
4830 rtnl_lock();
4831 list_for_each_entry_safe(bond, tmp_bond, &bn->dev_list, bond_list)
4832 unregister_netdevice_queue(bond->dev, &list);
4833 unregister_netdevice_many(&list);
4834 rtnl_unlock();
4836 bond_destroy_proc_dir(bn);
4837 }
4839 static struct pernet_operations bond_net_ops = {
4840 .init = bond_net_init,
4841 .exit = bond_net_exit,
4842 .id = &bond_net_id,
4843 .size = sizeof(struct bond_net),
4844 };
4846 static int __init bonding_init(void)
4847 {
4848 int i;
4849 int res;
4851 pr_info("%s", bond_version);
4853 res = bond_check_params(&bonding_defaults);
4854 if (res)
4855 goto out;
4857 res = register_pernet_subsys(&bond_net_ops);
4858 if (res)
4859 goto out;
4861 res = bond_netlink_init();
4862 if (res)
4863 goto err_link;
4865 bond_create_debugfs();
4867 for (i = 0; i < max_bonds; i++) {
4868 res = bond_create(&init_net, NULL);
4869 if (res)
4870 goto err;
4871 }
4873 register_netdevice_notifier(&bond_netdev_notifier);
4874 out:
4875 return res;
4876 err:
4877 bond_destroy_debugfs();
4878 bond_netlink_fini();
4879 err_link:
4880 unregister_pernet_subsys(&bond_net_ops);
4881 goto out;
4883 }
4885 static void __exit bonding_exit(void)
4886 {
4887 unregister_netdevice_notifier(&bond_netdev_notifier);
4889 bond_destroy_debugfs();
4891 bond_netlink_fini();
4892 unregister_pernet_subsys(&bond_net_ops);
4894 #ifdef CONFIG_NET_POLL_CONTROLLER
4895 /* Make sure we don't have an imbalance on our netpoll blocking */
4896 WARN_ON(atomic_read(&netpoll_block_tx));
4897 #endif
4898 }
4900 module_init(bonding_init);
4901 module_exit(bonding_exit);
4902 MODULE_LICENSE("GPL");
4903 MODULE_VERSION(DRV_VERSION);
4904 MODULE_DESCRIPTION(DRV_DESCRIPTION ", v" DRV_VERSION);
4905 MODULE_AUTHOR("Thomas Davis, tadavis@lbl.gov and many others");