?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
Merge branch 'x86-urgent-for-linus' of git://git.kernel.org/pub/scm/linux/kernel...
[cris-mirror.git] / drivers / net / bonding / bond_main.c
blobc669554d70bb7c7ba2fe3091ed1c58bd3026229f
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 */
33
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>
84
85 #include "bonding_priv.h"
86
87 /*---------------------------- Module parameters ----------------------------*/
88
89 /* monitor all links that often (in milliseconds). <=0 disables monitoring */
90
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;
115
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");
160
161 module_param(xmit_hash_policy, charp, 0);
162 MODULE_PARM_DESC(xmit_hash_policy, "balance-xor and 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.");
195
196 /*----------------------------- Global variables ----------------------------*/
197
198 #ifdef CONFIG_NET_POLL_CONTROLLER
199 atomic_t netpoll_block_tx = ATOMIC_INIT(0);
200 #endif
201
202 unsigned int bond_net_id __read_mostly;
203
204 /*-------------------------- Forward declarations ---------------------------*/
205
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
214 /*---------------------------- General routines -----------------------------*/
215
216 const char *bond_mode_name(int mode)
217 {
218 static const char *names[] = {
219 [BOND_MODE_ROUNDROBIN] = "load balancing (round-robin)",
220 [BOND_MODE_ACTIVEBACKUP] = "fault-tolerance (active-backup)",
221 [BOND_MODE_XOR] = "load balancing (xor)",
222 [BOND_MODE_BROADCAST] = "fault-tolerance (broadcast)",
223 [BOND_MODE_8023AD] = "IEEE 802.3ad Dynamic link aggregation",
224 [BOND_MODE_TLB] = "transmit load balancing",
225 [BOND_MODE_ALB] = "adaptive load balancing",
226 };
227
228 if (mode < BOND_MODE_ROUNDROBIN || mode > BOND_MODE_ALB)
229 return "unknown";
230
231 return names[mode];
232 }
233
234 /*---------------------------------- VLAN -----------------------------------*/
235
236 /**
237 * bond_dev_queue_xmit - Prepare skb for xmit.
238 *
239 * @bond: bond device that got this skb for tx.
240 * @skb: hw accel VLAN tagged skb to transmit
241 * @slave_dev: slave that is supposed to xmit this skbuff
242 */
243 void bond_dev_queue_xmit(struct bonding *bond, struct sk_buff *skb,
244 struct net_device *slave_dev)
245 {
246 skb->dev = slave_dev;
247
248 BUILD_BUG_ON(sizeof(skb->queue_mapping) !=
249 sizeof(qdisc_skb_cb(skb)->slave_dev_queue_mapping));
250 skb->queue_mapping = qdisc_skb_cb(skb)->slave_dev_queue_mapping;
251
252 if (unlikely(netpoll_tx_running(bond->dev)))
253 bond_netpoll_send_skb(bond_get_slave_by_dev(bond, slave_dev), skb);
254 else
255 dev_queue_xmit(skb);
256 }
257
258 /* In the following 2 functions, bond_vlan_rx_add_vid and bond_vlan_rx_kill_vid,
259 * We don't protect the slave list iteration with a lock because:
260 * a. This operation is performed in IOCTL context,
261 * b. The operation is protected by the RTNL semaphore in the 8021q code,
262 * c. Holding a lock with BH disabled while directly calling a base driver
263 * entry point is generally a BAD idea.
264 *
265 * The design of synchronization/protection for this operation in the 8021q
266 * module is good for one or more VLAN devices over a single physical device
267 * and cannot be extended for a teaming solution like bonding, so there is a
268 * potential race condition here where a net device from the vlan group might
269 * be referenced (either by a base driver or the 8021q code) while it is being
270 * removed from the system. However, it turns out we're not making matters
271 * worse, and if it works for regular VLAN usage it will work here too.
272 */
273
274 /**
275 * bond_vlan_rx_add_vid - Propagates adding an id to slaves
276 * @bond_dev: bonding net device that got called
277 * @vid: vlan id being added
278 */
279 static int bond_vlan_rx_add_vid(struct net_device *bond_dev,
280 __be16 proto, u16 vid)
281 {
282 struct bonding *bond = netdev_priv(bond_dev);
283 struct slave *slave, *rollback_slave;
284 struct list_head *iter;
285 int res;
286
287 bond_for_each_slave(bond, slave, iter) {
288 res = vlan_vid_add(slave->dev, proto, vid);
289 if (res)
290 goto unwind;
291 }
292
293 return 0;
294
295 unwind:
296 /* unwind to the slave that failed */
297 bond_for_each_slave(bond, rollback_slave, iter) {
298 if (rollback_slave == slave)
299 break;
300
301 vlan_vid_del(rollback_slave->dev, proto, vid);
302 }
303
304 return res;
305 }
306
307 /**
308 * bond_vlan_rx_kill_vid - Propagates deleting an id to slaves
309 * @bond_dev: bonding net device that got called
310 * @vid: vlan id being removed
311 */
312 static int bond_vlan_rx_kill_vid(struct net_device *bond_dev,
313 __be16 proto, u16 vid)
314 {
315 struct bonding *bond = netdev_priv(bond_dev);
316 struct list_head *iter;
317 struct slave *slave;
318
319 bond_for_each_slave(bond, slave, iter)
320 vlan_vid_del(slave->dev, proto, vid);
321
322 if (bond_is_lb(bond))
323 bond_alb_clear_vlan(bond, vid);
324
325 return 0;
326 }
327
328 /*------------------------------- Link status -------------------------------*/
329
330 /* Set the carrier state for the master according to the state of its
331 * slaves. If any slaves are up, the master is up. In 802.3ad mode,
332 * do special 802.3ad magic.
333 *
334 * Returns zero if carrier state does not change, nonzero if it does.
335 */
336 int bond_set_carrier(struct bonding *bond)
337 {
338 struct list_head *iter;
339 struct slave *slave;
340
341 if (!bond_has_slaves(bond))
342 goto down;
343
344 if (BOND_MODE(bond) == BOND_MODE_8023AD)
345 return bond_3ad_set_carrier(bond);
346
347 bond_for_each_slave(bond, slave, iter) {
348 if (slave->link == BOND_LINK_UP) {
349 if (!netif_carrier_ok(bond->dev)) {
350 netif_carrier_on(bond->dev);
351 return 1;
352 }
353 return 0;
354 }
355 }
356
357 down:
358 if (netif_carrier_ok(bond->dev)) {
359 netif_carrier_off(bond->dev);
360 return 1;
361 }
362 return 0;
363 }
364
365 /* Get link speed and duplex from the slave's base driver
366 * using ethtool. If for some reason the call fails or the
367 * values are invalid, set speed and duplex to -1,
368 * and return. Return 1 if speed or duplex settings are
369 * UNKNOWN; 0 otherwise.
370 */
371 static int bond_update_speed_duplex(struct slave *slave)
372 {
373 struct net_device *slave_dev = slave->dev;
374 struct ethtool_link_ksettings ecmd;
375 int res;
376
377 slave->speed = SPEED_UNKNOWN;
378 slave->duplex = DUPLEX_UNKNOWN;
379
380 res = __ethtool_get_link_ksettings(slave_dev, &ecmd);
381 if (res < 0)
382 return 1;
383 if (ecmd.base.speed == 0 || ecmd.base.speed == ((__u32)-1))
384 return 1;
385 switch (ecmd.base.duplex) {
386 case DUPLEX_FULL:
387 case DUPLEX_HALF:
388 break;
389 default:
390 return 1;
391 }
392
393 slave->speed = ecmd.base.speed;
394 slave->duplex = ecmd.base.duplex;
395
396 return 0;
397 }
398
399 const char *bond_slave_link_status(s8 link)
400 {
401 switch (link) {
402 case BOND_LINK_UP:
403 return "up";
404 case BOND_LINK_FAIL:
405 return "going down";
406 case BOND_LINK_DOWN:
407 return "down";
408 case BOND_LINK_BACK:
409 return "going back";
410 default:
411 return "unknown";
412 }
413 }
414
415 /* if <dev> supports MII link status reporting, check its link status.
416 *
417 * We either do MII/ETHTOOL ioctls, or check netif_carrier_ok(),
418 * depending upon the setting of the use_carrier parameter.
419 *
420 * Return either BMSR_LSTATUS, meaning that the link is up (or we
421 * can't tell and just pretend it is), or 0, meaning that the link is
422 * down.
423 *
424 * If reporting is non-zero, instead of faking link up, return -1 if
425 * both ETHTOOL and MII ioctls fail (meaning the device does not
426 * support them). If use_carrier is set, return whatever it says.
427 * It'd be nice if there was a good way to tell if a driver supports
428 * netif_carrier, but there really isn't.
429 */
430 static int bond_check_dev_link(struct bonding *bond,
431 struct net_device *slave_dev, int reporting)
432 {
433 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
434 int (*ioctl)(struct net_device *, struct ifreq *, int);
435 struct ifreq ifr;
436 struct mii_ioctl_data *mii;
437
438 if (!reporting && !netif_running(slave_dev))
439 return 0;
440
441 if (bond->params.use_carrier)
442 return netif_carrier_ok(slave_dev) ? BMSR_LSTATUS : 0;
443
444 /* Try to get link status using Ethtool first. */
445 if (slave_dev->ethtool_ops->get_link)
446 return slave_dev->ethtool_ops->get_link(slave_dev) ?
447 BMSR_LSTATUS : 0;
448
449 /* Ethtool can't be used, fallback to MII ioctls. */
450 ioctl = slave_ops->ndo_do_ioctl;
451 if (ioctl) {
452 /* TODO: set pointer to correct ioctl on a per team member
453 * bases to make this more efficient. that is, once
454 * we determine the correct ioctl, we will always
455 * call it and not the others for that team
456 * member.
457 */
458
459 /* We cannot assume that SIOCGMIIPHY will also read a
460 * register; not all network drivers (e.g., e100)
461 * support that.
462 */
463
464 /* Yes, the mii is overlaid on the ifreq.ifr_ifru */
465 strncpy(ifr.ifr_name, slave_dev->name, IFNAMSIZ);
466 mii = if_mii(&ifr);
467 if (ioctl(slave_dev, &ifr, SIOCGMIIPHY) == 0) {
468 mii->reg_num = MII_BMSR;
469 if (ioctl(slave_dev, &ifr, SIOCGMIIREG) == 0)
470 return mii->val_out & BMSR_LSTATUS;
471 }
472 }
473
474 /* If reporting, report that either there's no dev->do_ioctl,
475 * or both SIOCGMIIREG and get_link failed (meaning that we
476 * cannot report link status). If not reporting, pretend
477 * we're ok.
478 */
479 return reporting ? -1 : BMSR_LSTATUS;
480 }
481
482 /*----------------------------- Multicast list ------------------------------*/
483
484 /* Push the promiscuity flag down to appropriate slaves */
485 static int bond_set_promiscuity(struct bonding *bond, int inc)
486 {
487 struct list_head *iter;
488 int err = 0;
489
490 if (bond_uses_primary(bond)) {
491 struct slave *curr_active = rtnl_dereference(bond->curr_active_slave);
492
493 if (curr_active)
494 err = dev_set_promiscuity(curr_active->dev, inc);
495 } else {
496 struct slave *slave;
497
498 bond_for_each_slave(bond, slave, iter) {
499 err = dev_set_promiscuity(slave->dev, inc);
500 if (err)
501 return err;
502 }
503 }
504 return err;
505 }
506
507 /* Push the allmulti flag down to all slaves */
508 static int bond_set_allmulti(struct bonding *bond, int inc)
509 {
510 struct list_head *iter;
511 int err = 0;
512
513 if (bond_uses_primary(bond)) {
514 struct slave *curr_active = rtnl_dereference(bond->curr_active_slave);
515
516 if (curr_active)
517 err = dev_set_allmulti(curr_active->dev, inc);
518 } else {
519 struct slave *slave;
520
521 bond_for_each_slave(bond, slave, iter) {
522 err = dev_set_allmulti(slave->dev, inc);
523 if (err)
524 return err;
525 }
526 }
527 return err;
528 }
529
530 /* Retrieve the list of registered multicast addresses for the bonding
531 * device and retransmit an IGMP JOIN request to the current active
532 * slave.
533 */
534 static void bond_resend_igmp_join_requests_delayed(struct work_struct *work)
535 {
536 struct bonding *bond = container_of(work, struct bonding,
537 mcast_work.work);
538
539 if (!rtnl_trylock()) {
540 queue_delayed_work(bond->wq, &bond->mcast_work, 1);
541 return;
542 }
543 call_netdevice_notifiers(NETDEV_RESEND_IGMP, bond->dev);
544
545 if (bond->igmp_retrans > 1) {
546 bond->igmp_retrans--;
547 queue_delayed_work(bond->wq, &bond->mcast_work, HZ/5);
548 }
549 rtnl_unlock();
550 }
551
552 /* Flush bond's hardware addresses from slave */
553 static void bond_hw_addr_flush(struct net_device *bond_dev,
554 struct net_device *slave_dev)
555 {
556 struct bonding *bond = netdev_priv(bond_dev);
557
558 dev_uc_unsync(slave_dev, bond_dev);
559 dev_mc_unsync(slave_dev, bond_dev);
560
561 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
562 /* del lacpdu mc addr from mc list */
563 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
564
565 dev_mc_del(slave_dev, lacpdu_multicast);
566 }
567 }
568
569 /*--------------------------- Active slave change ---------------------------*/
570
571 /* Update the hardware address list and promisc/allmulti for the new and
572 * old active slaves (if any). Modes that are not using primary keep all
573 * slaves up date at all times; only the modes that use primary need to call
574 * this function to swap these settings during a failover.
575 */
576 static void bond_hw_addr_swap(struct bonding *bond, struct slave *new_active,
577 struct slave *old_active)
578 {
579 if (old_active) {
580 if (bond->dev->flags & IFF_PROMISC)
581 dev_set_promiscuity(old_active->dev, -1);
582
583 if (bond->dev->flags & IFF_ALLMULTI)
584 dev_set_allmulti(old_active->dev, -1);
585
586 bond_hw_addr_flush(bond->dev, old_active->dev);
587 }
588
589 if (new_active) {
590 /* FIXME: Signal errors upstream. */
591 if (bond->dev->flags & IFF_PROMISC)
592 dev_set_promiscuity(new_active->dev, 1);
593
594 if (bond->dev->flags & IFF_ALLMULTI)
595 dev_set_allmulti(new_active->dev, 1);
596
597 netif_addr_lock_bh(bond->dev);
598 dev_uc_sync(new_active->dev, bond->dev);
599 dev_mc_sync(new_active->dev, bond->dev);
600 netif_addr_unlock_bh(bond->dev);
601 }
602 }
603
604 /**
605 * bond_set_dev_addr - clone slave's address to bond
606 * @bond_dev: bond net device
607 * @slave_dev: slave net device
608 *
609 * Should be called with RTNL held.
610 */
611 static void bond_set_dev_addr(struct net_device *bond_dev,
612 struct net_device *slave_dev)
613 {
614 netdev_dbg(bond_dev, "bond_dev=%p slave_dev=%p slave_dev->name=%s slave_dev->addr_len=%d\n",
615 bond_dev, slave_dev, slave_dev->name, slave_dev->addr_len);
616 memcpy(bond_dev->dev_addr, slave_dev->dev_addr, slave_dev->addr_len);
617 bond_dev->addr_assign_type = NET_ADDR_STOLEN;
618 call_netdevice_notifiers(NETDEV_CHANGEADDR, bond_dev);
619 }
620
621 static struct slave *bond_get_old_active(struct bonding *bond,
622 struct slave *new_active)
623 {
624 struct slave *slave;
625 struct list_head *iter;
626
627 bond_for_each_slave(bond, slave, iter) {
628 if (slave == new_active)
629 continue;
630
631 if (ether_addr_equal(bond->dev->dev_addr, slave->dev->dev_addr))
632 return slave;
633 }
634
635 return NULL;
636 }
637
638 /* bond_do_fail_over_mac
639 *
640 * Perform special MAC address swapping for fail_over_mac settings
641 *
642 * Called with RTNL
643 */
644 static void bond_do_fail_over_mac(struct bonding *bond,
645 struct slave *new_active,
646 struct slave *old_active)
647 {
648 u8 tmp_mac[MAX_ADDR_LEN];
649 struct sockaddr_storage ss;
650 int rv;
651
652 switch (bond->params.fail_over_mac) {
653 case BOND_FOM_ACTIVE:
654 if (new_active)
655 bond_set_dev_addr(bond->dev, new_active->dev);
656 break;
657 case BOND_FOM_FOLLOW:
658 /* if new_active && old_active, swap them
659 * if just old_active, do nothing (going to no active slave)
660 * if just new_active, set new_active to bond's MAC
661 */
662 if (!new_active)
663 return;
664
665 if (!old_active)
666 old_active = bond_get_old_active(bond, new_active);
667
668 if (old_active) {
669 bond_hw_addr_copy(tmp_mac, new_active->dev->dev_addr,
670 new_active->dev->addr_len);
671 bond_hw_addr_copy(ss.__data,
672 old_active->dev->dev_addr,
673 old_active->dev->addr_len);
674 ss.ss_family = new_active->dev->type;
675 } else {
676 bond_hw_addr_copy(ss.__data, bond->dev->dev_addr,
677 bond->dev->addr_len);
678 ss.ss_family = bond->dev->type;
679 }
680
681 rv = dev_set_mac_address(new_active->dev,
682 (struct sockaddr *)&ss);
683 if (rv) {
684 netdev_err(bond->dev, "Error %d setting MAC of slave %s\n",
685 -rv, new_active->dev->name);
686 goto out;
687 }
688
689 if (!old_active)
690 goto out;
691
692 bond_hw_addr_copy(ss.__data, tmp_mac,
693 new_active->dev->addr_len);
694 ss.ss_family = old_active->dev->type;
695
696 rv = dev_set_mac_address(old_active->dev,
697 (struct sockaddr *)&ss);
698 if (rv)
699 netdev_err(bond->dev, "Error %d setting MAC of slave %s\n",
700 -rv, new_active->dev->name);
701 out:
702 break;
703 default:
704 netdev_err(bond->dev, "bond_do_fail_over_mac impossible: bad policy %d\n",
705 bond->params.fail_over_mac);
706 break;
707 }
708
709 }
710
711 static struct slave *bond_choose_primary_or_current(struct bonding *bond)
712 {
713 struct slave *prim = rtnl_dereference(bond->primary_slave);
714 struct slave *curr = rtnl_dereference(bond->curr_active_slave);
715
716 if (!prim || prim->link != BOND_LINK_UP) {
717 if (!curr || curr->link != BOND_LINK_UP)
718 return NULL;
719 return curr;
720 }
721
722 if (bond->force_primary) {
723 bond->force_primary = false;
724 return prim;
725 }
726
727 if (!curr || curr->link != BOND_LINK_UP)
728 return prim;
729
730 /* At this point, prim and curr are both up */
731 switch (bond->params.primary_reselect) {
732 case BOND_PRI_RESELECT_ALWAYS:
733 return prim;
734 case BOND_PRI_RESELECT_BETTER:
735 if (prim->speed < curr->speed)
736 return curr;
737 if (prim->speed == curr->speed && prim->duplex <= curr->duplex)
738 return curr;
739 return prim;
740 case BOND_PRI_RESELECT_FAILURE:
741 return curr;
742 default:
743 netdev_err(bond->dev, "impossible primary_reselect %d\n",
744 bond->params.primary_reselect);
745 return curr;
746 }
747 }
748
749 /**
750 * bond_find_best_slave - select the best available slave to be the active one
751 * @bond: our bonding struct
752 */
753 static struct slave *bond_find_best_slave(struct bonding *bond)
754 {
755 struct slave *slave, *bestslave = NULL;
756 struct list_head *iter;
757 int mintime = bond->params.updelay;
758
759 slave = bond_choose_primary_or_current(bond);
760 if (slave)
761 return slave;
762
763 bond_for_each_slave(bond, slave, iter) {
764 if (slave->link == BOND_LINK_UP)
765 return slave;
766 if (slave->link == BOND_LINK_BACK && bond_slave_is_up(slave) &&
767 slave->delay < mintime) {
768 mintime = slave->delay;
769 bestslave = slave;
770 }
771 }
772
773 return bestslave;
774 }
775
776 static bool bond_should_notify_peers(struct bonding *bond)
777 {
778 struct slave *slave;
779
780 rcu_read_lock();
781 slave = rcu_dereference(bond->curr_active_slave);
782 rcu_read_unlock();
783
784 netdev_dbg(bond->dev, "bond_should_notify_peers: slave %s\n",
785 slave ? slave->dev->name : "NULL");
786
787 if (!slave || !bond->send_peer_notif ||
788 !netif_carrier_ok(bond->dev) ||
789 test_bit(__LINK_STATE_LINKWATCH_PENDING, &slave->dev->state))
790 return false;
791
792 return true;
793 }
794
795 /**
796 * change_active_interface - change the active slave into the specified one
797 * @bond: our bonding struct
798 * @new: the new slave to make the active one
799 *
800 * Set the new slave to the bond's settings and unset them on the old
801 * curr_active_slave.
802 * Setting include flags, mc-list, promiscuity, allmulti, etc.
803 *
804 * If @new's link state is %BOND_LINK_BACK we'll set it to %BOND_LINK_UP,
805 * because it is apparently the best available slave we have, even though its
806 * updelay hasn't timed out yet.
807 *
808 * Caller must hold RTNL.
809 */
810 void bond_change_active_slave(struct bonding *bond, struct slave *new_active)
811 {
812 struct slave *old_active;
813
814 ASSERT_RTNL();
815
816 old_active = rtnl_dereference(bond->curr_active_slave);
817
818 if (old_active == new_active)
819 return;
820
821 if (new_active) {
822 new_active->last_link_up = jiffies;
823
824 if (new_active->link == BOND_LINK_BACK) {
825 if (bond_uses_primary(bond)) {
826 netdev_info(bond->dev, "making interface %s the new active one %d ms earlier\n",
827 new_active->dev->name,
828 (bond->params.updelay - new_active->delay) * bond->params.miimon);
829 }
830
831 new_active->delay = 0;
832 bond_set_slave_link_state(new_active, BOND_LINK_UP,
833 BOND_SLAVE_NOTIFY_NOW);
834
835 if (BOND_MODE(bond) == BOND_MODE_8023AD)
836 bond_3ad_handle_link_change(new_active, BOND_LINK_UP);
837
838 if (bond_is_lb(bond))
839 bond_alb_handle_link_change(bond, new_active, BOND_LINK_UP);
840 } else {
841 if (bond_uses_primary(bond)) {
842 netdev_info(bond->dev, "making interface %s the new active one\n",
843 new_active->dev->name);
844 }
845 }
846 }
847
848 if (bond_uses_primary(bond))
849 bond_hw_addr_swap(bond, new_active, old_active);
850
851 if (bond_is_lb(bond)) {
852 bond_alb_handle_active_change(bond, new_active);
853 if (old_active)
854 bond_set_slave_inactive_flags(old_active,
855 BOND_SLAVE_NOTIFY_NOW);
856 if (new_active)
857 bond_set_slave_active_flags(new_active,
858 BOND_SLAVE_NOTIFY_NOW);
859 } else {
860 rcu_assign_pointer(bond->curr_active_slave, new_active);
861 }
862
863 if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP) {
864 if (old_active)
865 bond_set_slave_inactive_flags(old_active,
866 BOND_SLAVE_NOTIFY_NOW);
867
868 if (new_active) {
869 bool should_notify_peers = false;
870
871 bond_set_slave_active_flags(new_active,
872 BOND_SLAVE_NOTIFY_NOW);
873
874 if (bond->params.fail_over_mac)
875 bond_do_fail_over_mac(bond, new_active,
876 old_active);
877
878 if (netif_running(bond->dev)) {
879 bond->send_peer_notif =
880 bond->params.num_peer_notif;
881 should_notify_peers =
882 bond_should_notify_peers(bond);
883 }
884
885 call_netdevice_notifiers(NETDEV_BONDING_FAILOVER, bond->dev);
886 if (should_notify_peers)
887 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS,
888 bond->dev);
889 }
890 }
891
892 /* resend IGMP joins since active slave has changed or
893 * all were sent on curr_active_slave.
894 * resend only if bond is brought up with the affected
895 * bonding modes and the retransmission is enabled
896 */
897 if (netif_running(bond->dev) && (bond->params.resend_igmp > 0) &&
898 ((bond_uses_primary(bond) && new_active) ||
899 BOND_MODE(bond) == BOND_MODE_ROUNDROBIN)) {
900 bond->igmp_retrans = bond->params.resend_igmp;
901 queue_delayed_work(bond->wq, &bond->mcast_work, 1);
902 }
903 }
904
905 /**
906 * bond_select_active_slave - select a new active slave, if needed
907 * @bond: our bonding struct
908 *
909 * This functions should be called when one of the following occurs:
910 * - The old curr_active_slave has been released or lost its link.
911 * - The primary_slave has got its link back.
912 * - A slave has got its link back and there's no old curr_active_slave.
913 *
914 * Caller must hold RTNL.
915 */
916 void bond_select_active_slave(struct bonding *bond)
917 {
918 struct slave *best_slave;
919 int rv;
920
921 ASSERT_RTNL();
922
923 best_slave = bond_find_best_slave(bond);
924 if (best_slave != rtnl_dereference(bond->curr_active_slave)) {
925 bond_change_active_slave(bond, best_slave);
926 rv = bond_set_carrier(bond);
927 if (!rv)
928 return;
929
930 if (netif_carrier_ok(bond->dev))
931 netdev_info(bond->dev, "first active interface up!\n");
932 else
933 netdev_info(bond->dev, "now running without any active interface!\n");
934 }
935 }
936
937 #ifdef CONFIG_NET_POLL_CONTROLLER
938 static inline int slave_enable_netpoll(struct slave *slave)
939 {
940 struct netpoll *np;
941 int err = 0;
942
943 np = kzalloc(sizeof(*np), GFP_KERNEL);
944 err = -ENOMEM;
945 if (!np)
946 goto out;
947
948 err = __netpoll_setup(np, slave->dev);
949 if (err) {
950 kfree(np);
951 goto out;
952 }
953 slave->np = np;
954 out:
955 return err;
956 }
957 static inline void slave_disable_netpoll(struct slave *slave)
958 {
959 struct netpoll *np = slave->np;
960
961 if (!np)
962 return;
963
964 slave->np = NULL;
965 __netpoll_free_async(np);
966 }
967
968 static void bond_poll_controller(struct net_device *bond_dev)
969 {
970 struct bonding *bond = netdev_priv(bond_dev);
971 struct slave *slave = NULL;
972 struct list_head *iter;
973 struct ad_info ad_info;
974 struct netpoll_info *ni;
975 const struct net_device_ops *ops;
976
977 if (BOND_MODE(bond) == BOND_MODE_8023AD)
978 if (bond_3ad_get_active_agg_info(bond, &ad_info))
979 return;
980
981 bond_for_each_slave_rcu(bond, slave, iter) {
982 ops = slave->dev->netdev_ops;
983 if (!bond_slave_is_up(slave) || !ops->ndo_poll_controller)
984 continue;
985
986 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
987 struct aggregator *agg =
988 SLAVE_AD_INFO(slave)->port.aggregator;
989
990 if (agg &&
991 agg->aggregator_identifier != ad_info.aggregator_id)
992 continue;
993 }
994
995 ni = rcu_dereference_bh(slave->dev->npinfo);
996 if (down_trylock(&ni->dev_lock))
997 continue;
998 ops->ndo_poll_controller(slave->dev);
999 up(&ni->dev_lock);
1000 }
1001 }
1002
1003 static void bond_netpoll_cleanup(struct net_device *bond_dev)
1004 {
1005 struct bonding *bond = netdev_priv(bond_dev);
1006 struct list_head *iter;
1007 struct slave *slave;
1008
1009 bond_for_each_slave(bond, slave, iter)
1010 if (bond_slave_is_up(slave))
1011 slave_disable_netpoll(slave);
1012 }
1013
1014 static int bond_netpoll_setup(struct net_device *dev, struct netpoll_info *ni)
1015 {
1016 struct bonding *bond = netdev_priv(dev);
1017 struct list_head *iter;
1018 struct slave *slave;
1019 int err = 0;
1020
1021 bond_for_each_slave(bond, slave, iter) {
1022 err = slave_enable_netpoll(slave);
1023 if (err) {
1024 bond_netpoll_cleanup(dev);
1025 break;
1026 }
1027 }
1028 return err;
1029 }
1030 #else
1031 static inline int slave_enable_netpoll(struct slave *slave)
1032 {
1033 return 0;
1034 }
1035 static inline void slave_disable_netpoll(struct slave *slave)
1036 {
1037 }
1038 static void bond_netpoll_cleanup(struct net_device *bond_dev)
1039 {
1040 }
1041 #endif
1042
1043 /*---------------------------------- IOCTL ----------------------------------*/
1044
1045 static netdev_features_t bond_fix_features(struct net_device *dev,
1046 netdev_features_t features)
1047 {
1048 struct bonding *bond = netdev_priv(dev);
1049 struct list_head *iter;
1050 netdev_features_t mask;
1051 struct slave *slave;
1052
1053 mask = features;
1054
1055 features &= ~NETIF_F_ONE_FOR_ALL;
1056 features |= NETIF_F_ALL_FOR_ALL;
1057
1058 bond_for_each_slave(bond, slave, iter) {
1059 features = netdev_increment_features(features,
1060 slave->dev->features,
1061 mask);
1062 }
1063 features = netdev_add_tso_features(features, mask);
1064
1065 return features;
1066 }
1067
1068 #define BOND_VLAN_FEATURES (NETIF_F_HW_CSUM | NETIF_F_SG | \
1069 NETIF_F_FRAGLIST | NETIF_F_ALL_TSO | \
1070 NETIF_F_HIGHDMA | NETIF_F_LRO)
1071
1072 #define BOND_ENC_FEATURES (NETIF_F_HW_CSUM | NETIF_F_SG | \
1073 NETIF_F_RXCSUM | NETIF_F_ALL_TSO)
1074
1075 static void bond_compute_features(struct bonding *bond)
1076 {
1077 unsigned int dst_release_flag = IFF_XMIT_DST_RELEASE |
1078 IFF_XMIT_DST_RELEASE_PERM;
1079 netdev_features_t vlan_features = BOND_VLAN_FEATURES;
1080 netdev_features_t enc_features = BOND_ENC_FEATURES;
1081 struct net_device *bond_dev = bond->dev;
1082 struct list_head *iter;
1083 struct slave *slave;
1084 unsigned short max_hard_header_len = ETH_HLEN;
1085 unsigned int gso_max_size = GSO_MAX_SIZE;
1086 u16 gso_max_segs = GSO_MAX_SEGS;
1087
1088 if (!bond_has_slaves(bond))
1089 goto done;
1090 vlan_features &= NETIF_F_ALL_FOR_ALL;
1091
1092 bond_for_each_slave(bond, slave, iter) {
1093 vlan_features = netdev_increment_features(vlan_features,
1094 slave->dev->vlan_features, BOND_VLAN_FEATURES);
1095
1096 enc_features = netdev_increment_features(enc_features,
1097 slave->dev->hw_enc_features,
1098 BOND_ENC_FEATURES);
1099 dst_release_flag &= slave->dev->priv_flags;
1100 if (slave->dev->hard_header_len > max_hard_header_len)
1101 max_hard_header_len = slave->dev->hard_header_len;
1102
1103 gso_max_size = min(gso_max_size, slave->dev->gso_max_size);
1104 gso_max_segs = min(gso_max_segs, slave->dev->gso_max_segs);
1105 }
1106 bond_dev->hard_header_len = max_hard_header_len;
1107
1108 done:
1109 bond_dev->vlan_features = vlan_features;
1110 bond_dev->hw_enc_features = enc_features | NETIF_F_GSO_ENCAP_ALL;
1111 bond_dev->gso_max_segs = gso_max_segs;
1112 netif_set_gso_max_size(bond_dev, gso_max_size);
1113
1114 bond_dev->priv_flags &= ~IFF_XMIT_DST_RELEASE;
1115 if ((bond_dev->priv_flags & IFF_XMIT_DST_RELEASE_PERM) &&
1116 dst_release_flag == (IFF_XMIT_DST_RELEASE | IFF_XMIT_DST_RELEASE_PERM))
1117 bond_dev->priv_flags |= IFF_XMIT_DST_RELEASE;
1118
1119 netdev_change_features(bond_dev);
1120 }
1121
1122 static void bond_setup_by_slave(struct net_device *bond_dev,
1123 struct net_device *slave_dev)
1124 {
1125 bond_dev->header_ops = slave_dev->header_ops;
1126
1127 bond_dev->type = slave_dev->type;
1128 bond_dev->hard_header_len = slave_dev->hard_header_len;
1129 bond_dev->addr_len = slave_dev->addr_len;
1130
1131 memcpy(bond_dev->broadcast, slave_dev->broadcast,
1132 slave_dev->addr_len);
1133 }
1134
1135 /* On bonding slaves other than the currently active slave, suppress
1136 * duplicates except for alb non-mcast/bcast.
1137 */
1138 static bool bond_should_deliver_exact_match(struct sk_buff *skb,
1139 struct slave *slave,
1140 struct bonding *bond)
1141 {
1142 if (bond_is_slave_inactive(slave)) {
1143 if (BOND_MODE(bond) == BOND_MODE_ALB &&
1144 skb->pkt_type != PACKET_BROADCAST &&
1145 skb->pkt_type != PACKET_MULTICAST)
1146 return false;
1147 return true;
1148 }
1149 return false;
1150 }
1151
1152 static rx_handler_result_t bond_handle_frame(struct sk_buff **pskb)
1153 {
1154 struct sk_buff *skb = *pskb;
1155 struct slave *slave;
1156 struct bonding *bond;
1157 int (*recv_probe)(const struct sk_buff *, struct bonding *,
1158 struct slave *);
1159 int ret = RX_HANDLER_ANOTHER;
1160
1161 skb = skb_share_check(skb, GFP_ATOMIC);
1162 if (unlikely(!skb))
1163 return RX_HANDLER_CONSUMED;
1164
1165 *pskb = skb;
1166
1167 slave = bond_slave_get_rcu(skb->dev);
1168 bond = slave->bond;
1169
1170 recv_probe = READ_ONCE(bond->recv_probe);
1171 if (recv_probe) {
1172 ret = recv_probe(skb, bond, slave);
1173 if (ret == RX_HANDLER_CONSUMED) {
1174 consume_skb(skb);
1175 return ret;
1176 }
1177 }
1178
1179 /* don't change skb->dev for link-local packets */
1180 if (is_link_local_ether_addr(eth_hdr(skb)->h_dest))
1181 return RX_HANDLER_PASS;
1182 if (bond_should_deliver_exact_match(skb, slave, bond))
1183 return RX_HANDLER_EXACT;
1184
1185 skb->dev = bond->dev;
1186
1187 if (BOND_MODE(bond) == BOND_MODE_ALB &&
1188 bond->dev->priv_flags & IFF_BRIDGE_PORT &&
1189 skb->pkt_type == PACKET_HOST) {
1190
1191 if (unlikely(skb_cow_head(skb,
1192 skb->data - skb_mac_header(skb)))) {
1193 kfree_skb(skb);
1194 return RX_HANDLER_CONSUMED;
1195 }
1196 bond_hw_addr_copy(eth_hdr(skb)->h_dest, bond->dev->dev_addr,
1197 bond->dev->addr_len);
1198 }
1199
1200 return ret;
1201 }
1202
1203 static enum netdev_lag_tx_type bond_lag_tx_type(struct bonding *bond)
1204 {
1205 switch (BOND_MODE(bond)) {
1206 case BOND_MODE_ROUNDROBIN:
1207 return NETDEV_LAG_TX_TYPE_ROUNDROBIN;
1208 case BOND_MODE_ACTIVEBACKUP:
1209 return NETDEV_LAG_TX_TYPE_ACTIVEBACKUP;
1210 case BOND_MODE_BROADCAST:
1211 return NETDEV_LAG_TX_TYPE_BROADCAST;
1212 case BOND_MODE_XOR:
1213 case BOND_MODE_8023AD:
1214 return NETDEV_LAG_TX_TYPE_HASH;
1215 default:
1216 return NETDEV_LAG_TX_TYPE_UNKNOWN;
1217 }
1218 }
1219
1220 static int bond_master_upper_dev_link(struct bonding *bond, struct slave *slave,
1221 struct netlink_ext_ack *extack)
1222 {
1223 struct netdev_lag_upper_info lag_upper_info;
1224
1225 lag_upper_info.tx_type = bond_lag_tx_type(bond);
1226
1227 return netdev_master_upper_dev_link(slave->dev, bond->dev, slave,
1228 &lag_upper_info, extack);
1229 }
1230
1231 static void bond_upper_dev_unlink(struct bonding *bond, struct slave *slave)
1232 {
1233 netdev_upper_dev_unlink(slave->dev, bond->dev);
1234 slave->dev->flags &= ~IFF_SLAVE;
1235 }
1236
1237 static struct slave *bond_alloc_slave(struct bonding *bond)
1238 {
1239 struct slave *slave = NULL;
1240
1241 slave = kzalloc(sizeof(*slave), GFP_KERNEL);
1242 if (!slave)
1243 return NULL;
1244
1245 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
1246 SLAVE_AD_INFO(slave) = kzalloc(sizeof(struct ad_slave_info),
1247 GFP_KERNEL);
1248 if (!SLAVE_AD_INFO(slave)) {
1249 kfree(slave);
1250 return NULL;
1251 }
1252 }
1253 return slave;
1254 }
1255
1256 static void bond_free_slave(struct slave *slave)
1257 {
1258 struct bonding *bond = bond_get_bond_by_slave(slave);
1259
1260 if (BOND_MODE(bond) == BOND_MODE_8023AD)
1261 kfree(SLAVE_AD_INFO(slave));
1262
1263 kfree(slave);
1264 }
1265
1266 static void bond_fill_ifbond(struct bonding *bond, struct ifbond *info)
1267 {
1268 info->bond_mode = BOND_MODE(bond);
1269 info->miimon = bond->params.miimon;
1270 info->num_slaves = bond->slave_cnt;
1271 }
1272
1273 static void bond_fill_ifslave(struct slave *slave, struct ifslave *info)
1274 {
1275 strcpy(info->slave_name, slave->dev->name);
1276 info->link = slave->link;
1277 info->state = bond_slave_state(slave);
1278 info->link_failure_count = slave->link_failure_count;
1279 }
1280
1281 static void bond_netdev_notify(struct net_device *dev,
1282 struct netdev_bonding_info *info)
1283 {
1284 rtnl_lock();
1285 netdev_bonding_info_change(dev, info);
1286 rtnl_unlock();
1287 }
1288
1289 static void bond_netdev_notify_work(struct work_struct *_work)
1290 {
1291 struct netdev_notify_work *w =
1292 container_of(_work, struct netdev_notify_work, work.work);
1293
1294 bond_netdev_notify(w->dev, &w->bonding_info);
1295 dev_put(w->dev);
1296 kfree(w);
1297 }
1298
1299 void bond_queue_slave_event(struct slave *slave)
1300 {
1301 struct bonding *bond = slave->bond;
1302 struct netdev_notify_work *nnw = kzalloc(sizeof(*nnw), GFP_ATOMIC);
1303
1304 if (!nnw)
1305 return;
1306
1307 dev_hold(slave->dev);
1308 nnw->dev = slave->dev;
1309 bond_fill_ifslave(slave, &nnw->bonding_info.slave);
1310 bond_fill_ifbond(bond, &nnw->bonding_info.master);
1311 INIT_DELAYED_WORK(&nnw->work, bond_netdev_notify_work);
1312
1313 queue_delayed_work(slave->bond->wq, &nnw->work, 0);
1314 }
1315
1316 void bond_lower_state_changed(struct slave *slave)
1317 {
1318 struct netdev_lag_lower_state_info info;
1319
1320 info.link_up = slave->link == BOND_LINK_UP ||
1321 slave->link == BOND_LINK_FAIL;
1322 info.tx_enabled = bond_is_active_slave(slave);
1323 netdev_lower_state_changed(slave->dev, &info);
1324 }
1325
1326 /* enslave device <slave> to bond device <master> */
1327 int bond_enslave(struct net_device *bond_dev, struct net_device *slave_dev,
1328 struct netlink_ext_ack *extack)
1329 {
1330 struct bonding *bond = netdev_priv(bond_dev);
1331 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
1332 struct slave *new_slave = NULL, *prev_slave;
1333 struct sockaddr_storage ss;
1334 int link_reporting;
1335 int res = 0, i;
1336
1337 if (!bond->params.use_carrier &&
1338 slave_dev->ethtool_ops->get_link == NULL &&
1339 slave_ops->ndo_do_ioctl == NULL) {
1340 netdev_warn(bond_dev, "no link monitoring support for %s\n",
1341 slave_dev->name);
1342 }
1343
1344 /* already in-use? */
1345 if (netdev_is_rx_handler_busy(slave_dev)) {
1346 NL_SET_ERR_MSG(extack, "Device is in use and cannot be enslaved");
1347 netdev_err(bond_dev,
1348 "Error: Device is in use and cannot be enslaved\n");
1349 return -EBUSY;
1350 }
1351
1352 if (bond_dev == slave_dev) {
1353 NL_SET_ERR_MSG(extack, "Cannot enslave bond to itself.");
1354 netdev_err(bond_dev, "cannot enslave bond to itself.\n");
1355 return -EPERM;
1356 }
1357
1358 /* vlan challenged mutual exclusion */
1359 /* no need to lock since we're protected by rtnl_lock */
1360 if (slave_dev->features & NETIF_F_VLAN_CHALLENGED) {
1361 netdev_dbg(bond_dev, "%s is NETIF_F_VLAN_CHALLENGED\n",
1362 slave_dev->name);
1363 if (vlan_uses_dev(bond_dev)) {
1364 NL_SET_ERR_MSG(extack, "Can not enslave VLAN challenged device to VLAN enabled bond");
1365 netdev_err(bond_dev, "Error: cannot enslave VLAN challenged slave %s on VLAN enabled bond %s\n",
1366 slave_dev->name, bond_dev->name);
1367 return -EPERM;
1368 } else {
1369 netdev_warn(bond_dev, "enslaved VLAN challenged slave %s. Adding VLANs will be blocked as long as %s is part of bond %s\n",
1370 slave_dev->name, slave_dev->name,
1371 bond_dev->name);
1372 }
1373 } else {
1374 netdev_dbg(bond_dev, "%s is !NETIF_F_VLAN_CHALLENGED\n",
1375 slave_dev->name);
1376 }
1377
1378 /* Old ifenslave binaries are no longer supported. These can
1379 * be identified with moderate accuracy by the state of the slave:
1380 * the current ifenslave will set the interface down prior to
1381 * enslaving it; the old ifenslave will not.
1382 */
1383 if (slave_dev->flags & IFF_UP) {
1384 NL_SET_ERR_MSG(extack, "Device can not be enslaved while up");
1385 netdev_err(bond_dev, "%s is up - this may be due to an out of date ifenslave\n",
1386 slave_dev->name);
1387 return -EPERM;
1388 }
1389
1390 /* set bonding device ether type by slave - bonding netdevices are
1391 * created with ether_setup, so when the slave type is not ARPHRD_ETHER
1392 * there is a need to override some of the type dependent attribs/funcs.
1393 *
1394 * bond ether type mutual exclusion - don't allow slaves of dissimilar
1395 * ether type (eg ARPHRD_ETHER and ARPHRD_INFINIBAND) share the same bond
1396 */
1397 if (!bond_has_slaves(bond)) {
1398 if (bond_dev->type != slave_dev->type) {
1399 netdev_dbg(bond_dev, "change device type from %d to %d\n",
1400 bond_dev->type, slave_dev->type);
1401
1402 res = call_netdevice_notifiers(NETDEV_PRE_TYPE_CHANGE,
1403 bond_dev);
1404 res = notifier_to_errno(res);
1405 if (res) {
1406 netdev_err(bond_dev, "refused to change device type\n");
1407 return -EBUSY;
1408 }
1409
1410 /* Flush unicast and multicast addresses */
1411 dev_uc_flush(bond_dev);
1412 dev_mc_flush(bond_dev);
1413
1414 if (slave_dev->type != ARPHRD_ETHER)
1415 bond_setup_by_slave(bond_dev, slave_dev);
1416 else {
1417 ether_setup(bond_dev);
1418 bond_dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1419 }
1420
1421 call_netdevice_notifiers(NETDEV_POST_TYPE_CHANGE,
1422 bond_dev);
1423 }
1424 } else if (bond_dev->type != slave_dev->type) {
1425 NL_SET_ERR_MSG(extack, "Device type is different from other slaves");
1426 netdev_err(bond_dev, "%s ether type (%d) is different from other slaves (%d), can not enslave it\n",
1427 slave_dev->name, slave_dev->type, bond_dev->type);
1428 return -EINVAL;
1429 }
1430
1431 if (slave_dev->type == ARPHRD_INFINIBAND &&
1432 BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
1433 NL_SET_ERR_MSG(extack, "Only active-backup mode is supported for infiniband slaves");
1434 netdev_warn(bond_dev, "Type (%d) supports only active-backup mode\n",
1435 slave_dev->type);
1436 res = -EOPNOTSUPP;
1437 goto err_undo_flags;
1438 }
1439
1440 if (!slave_ops->ndo_set_mac_address ||
1441 slave_dev->type == ARPHRD_INFINIBAND) {
1442 netdev_warn(bond_dev, "The slave device specified does not support setting the MAC address\n");
1443 if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP &&
1444 bond->params.fail_over_mac != BOND_FOM_ACTIVE) {
1445 if (!bond_has_slaves(bond)) {
1446 bond->params.fail_over_mac = BOND_FOM_ACTIVE;
1447 netdev_warn(bond_dev, "Setting fail_over_mac to active for active-backup mode\n");
1448 } else {
1449 NL_SET_ERR_MSG(extack, "Slave device does not support setting the MAC address, but fail_over_mac is not set to active");
1450 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");
1451 res = -EOPNOTSUPP;
1452 goto err_undo_flags;
1453 }
1454 }
1455 }
1456
1457 call_netdevice_notifiers(NETDEV_JOIN, slave_dev);
1458
1459 /* If this is the first slave, then we need to set the master's hardware
1460 * address to be the same as the slave's.
1461 */
1462 if (!bond_has_slaves(bond) &&
1463 bond->dev->addr_assign_type == NET_ADDR_RANDOM)
1464 bond_set_dev_addr(bond->dev, slave_dev);
1465
1466 new_slave = bond_alloc_slave(bond);
1467 if (!new_slave) {
1468 res = -ENOMEM;
1469 goto err_undo_flags;
1470 }
1471
1472 new_slave->bond = bond;
1473 new_slave->dev = slave_dev;
1474 /* Set the new_slave's queue_id to be zero. Queue ID mapping
1475 * is set via sysfs or module option if desired.
1476 */
1477 new_slave->queue_id = 0;
1478
1479 /* Save slave's original mtu and then set it to match the bond */
1480 new_slave->original_mtu = slave_dev->mtu;
1481 res = dev_set_mtu(slave_dev, bond->dev->mtu);
1482 if (res) {
1483 netdev_dbg(bond_dev, "Error %d calling dev_set_mtu\n", res);
1484 goto err_free;
1485 }
1486
1487 /* Save slave's original ("permanent") mac address for modes
1488 * that need it, and for restoring it upon release, and then
1489 * set it to the master's address
1490 */
1491 bond_hw_addr_copy(new_slave->perm_hwaddr, slave_dev->dev_addr,
1492 slave_dev->addr_len);
1493
1494 if (!bond->params.fail_over_mac ||
1495 BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
1496 /* Set slave to master's mac address. The application already
1497 * set the master's mac address to that of the first slave
1498 */
1499 memcpy(ss.__data, bond_dev->dev_addr, bond_dev->addr_len);
1500 ss.ss_family = slave_dev->type;
1501 res = dev_set_mac_address(slave_dev, (struct sockaddr *)&ss);
1502 if (res) {
1503 netdev_dbg(bond_dev, "Error %d calling set_mac_address\n", res);
1504 goto err_restore_mtu;
1505 }
1506 }
1507
1508 /* set slave flag before open to prevent IPv6 addrconf */
1509 slave_dev->flags |= IFF_SLAVE;
1510
1511 /* open the slave since the application closed it */
1512 res = dev_open(slave_dev);
1513 if (res) {
1514 netdev_dbg(bond_dev, "Opening slave %s failed\n", slave_dev->name);
1515 goto err_restore_mac;
1516 }
1517
1518 slave_dev->priv_flags |= IFF_BONDING;
1519 /* initialize slave stats */
1520 dev_get_stats(new_slave->dev, &new_slave->slave_stats);
1521
1522 if (bond_is_lb(bond)) {
1523 /* bond_alb_init_slave() must be called before all other stages since
1524 * it might fail and we do not want to have to undo everything
1525 */
1526 res = bond_alb_init_slave(bond, new_slave);
1527 if (res)
1528 goto err_close;
1529 }
1530
1531 /* If the mode uses primary, then the following is handled by
1532 * bond_change_active_slave().
1533 */
1534 if (!bond_uses_primary(bond)) {
1535 /* set promiscuity level to new slave */
1536 if (bond_dev->flags & IFF_PROMISC) {
1537 res = dev_set_promiscuity(slave_dev, 1);
1538 if (res)
1539 goto err_close;
1540 }
1541
1542 /* set allmulti level to new slave */
1543 if (bond_dev->flags & IFF_ALLMULTI) {
1544 res = dev_set_allmulti(slave_dev, 1);
1545 if (res)
1546 goto err_close;
1547 }
1548
1549 netif_addr_lock_bh(bond_dev);
1550
1551 dev_mc_sync_multiple(slave_dev, bond_dev);
1552 dev_uc_sync_multiple(slave_dev, bond_dev);
1553
1554 netif_addr_unlock_bh(bond_dev);
1555 }
1556
1557 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
1558 /* add lacpdu mc addr to mc list */
1559 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
1560
1561 dev_mc_add(slave_dev, lacpdu_multicast);
1562 }
1563
1564 res = vlan_vids_add_by_dev(slave_dev, bond_dev);
1565 if (res) {
1566 netdev_err(bond_dev, "Couldn't add bond vlan ids to %s\n",
1567 slave_dev->name);
1568 goto err_close;
1569 }
1570
1571 prev_slave = bond_last_slave(bond);
1572
1573 new_slave->delay = 0;
1574 new_slave->link_failure_count = 0;
1575
1576 if (bond_update_speed_duplex(new_slave) &&
1577 bond_needs_speed_duplex(bond))
1578 new_slave->link = BOND_LINK_DOWN;
1579
1580 new_slave->last_rx = jiffies -
1581 (msecs_to_jiffies(bond->params.arp_interval) + 1);
1582 for (i = 0; i < BOND_MAX_ARP_TARGETS; i++)
1583 new_slave->target_last_arp_rx[i] = new_slave->last_rx;
1584
1585 if (bond->params.miimon && !bond->params.use_carrier) {
1586 link_reporting = bond_check_dev_link(bond, slave_dev, 1);
1587
1588 if ((link_reporting == -1) && !bond->params.arp_interval) {
1589 /* miimon is set but a bonded network driver
1590 * does not support ETHTOOL/MII and
1591 * arp_interval is not set. Note: if
1592 * use_carrier is enabled, we will never go
1593 * here (because netif_carrier is always
1594 * supported); thus, we don't need to change
1595 * the messages for netif_carrier.
1596 */
1597 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",
1598 slave_dev->name);
1599 } else if (link_reporting == -1) {
1600 /* unable get link status using mii/ethtool */
1601 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",
1602 slave_dev->name);
1603 }
1604 }
1605
1606 /* check for initial state */
1607 new_slave->link = BOND_LINK_NOCHANGE;
1608 if (bond->params.miimon) {
1609 if (bond_check_dev_link(bond, slave_dev, 0) == BMSR_LSTATUS) {
1610 if (bond->params.updelay) {
1611 bond_set_slave_link_state(new_slave,
1612 BOND_LINK_BACK,
1613 BOND_SLAVE_NOTIFY_NOW);
1614 new_slave->delay = bond->params.updelay;
1615 } else {
1616 bond_set_slave_link_state(new_slave,
1617 BOND_LINK_UP,
1618 BOND_SLAVE_NOTIFY_NOW);
1619 }
1620 } else {
1621 bond_set_slave_link_state(new_slave, BOND_LINK_DOWN,
1622 BOND_SLAVE_NOTIFY_NOW);
1623 }
1624 } else if (bond->params.arp_interval) {
1625 bond_set_slave_link_state(new_slave,
1626 (netif_carrier_ok(slave_dev) ?
1627 BOND_LINK_UP : BOND_LINK_DOWN),
1628 BOND_SLAVE_NOTIFY_NOW);
1629 } else {
1630 bond_set_slave_link_state(new_slave, BOND_LINK_UP,
1631 BOND_SLAVE_NOTIFY_NOW);
1632 }
1633
1634 if (new_slave->link != BOND_LINK_DOWN)
1635 new_slave->last_link_up = jiffies;
1636 netdev_dbg(bond_dev, "Initial state of slave_dev is BOND_LINK_%s\n",
1637 new_slave->link == BOND_LINK_DOWN ? "DOWN" :
1638 (new_slave->link == BOND_LINK_UP ? "UP" : "BACK"));
1639
1640 if (bond_uses_primary(bond) && bond->params.primary[0]) {
1641 /* if there is a primary slave, remember it */
1642 if (strcmp(bond->params.primary, new_slave->dev->name) == 0) {
1643 rcu_assign_pointer(bond->primary_slave, new_slave);
1644 bond->force_primary = true;
1645 }
1646 }
1647
1648 switch (BOND_MODE(bond)) {
1649 case BOND_MODE_ACTIVEBACKUP:
1650 bond_set_slave_inactive_flags(new_slave,
1651 BOND_SLAVE_NOTIFY_NOW);
1652 break;
1653 case BOND_MODE_8023AD:
1654 /* in 802.3ad mode, the internal mechanism
1655 * will activate the slaves in the selected
1656 * aggregator
1657 */
1658 bond_set_slave_inactive_flags(new_slave, BOND_SLAVE_NOTIFY_NOW);
1659 /* if this is the first slave */
1660 if (!prev_slave) {
1661 SLAVE_AD_INFO(new_slave)->id = 1;
1662 /* Initialize AD with the number of times that the AD timer is called in 1 second
1663 * can be called only after the mac address of the bond is set
1664 */
1665 bond_3ad_initialize(bond, 1000/AD_TIMER_INTERVAL);
1666 } else {
1667 SLAVE_AD_INFO(new_slave)->id =
1668 SLAVE_AD_INFO(prev_slave)->id + 1;
1669 }
1670
1671 bond_3ad_bind_slave(new_slave);
1672 break;
1673 case BOND_MODE_TLB:
1674 case BOND_MODE_ALB:
1675 bond_set_active_slave(new_slave);
1676 bond_set_slave_inactive_flags(new_slave, BOND_SLAVE_NOTIFY_NOW);
1677 break;
1678 default:
1679 netdev_dbg(bond_dev, "This slave is always active in trunk mode\n");
1680
1681 /* always active in trunk mode */
1682 bond_set_active_slave(new_slave);
1683
1684 /* In trunking mode there is little meaning to curr_active_slave
1685 * anyway (it holds no special properties of the bond device),
1686 * so we can change it without calling change_active_interface()
1687 */
1688 if (!rcu_access_pointer(bond->curr_active_slave) &&
1689 new_slave->link == BOND_LINK_UP)
1690 rcu_assign_pointer(bond->curr_active_slave, new_slave);
1691
1692 break;
1693 } /* switch(bond_mode) */
1694
1695 #ifdef CONFIG_NET_POLL_CONTROLLER
1696 slave_dev->npinfo = bond->dev->npinfo;
1697 if (slave_dev->npinfo) {
1698 if (slave_enable_netpoll(new_slave)) {
1699 netdev_info(bond_dev, "master_dev is using netpoll, but new slave device does not support netpoll\n");
1700 res = -EBUSY;
1701 goto err_detach;
1702 }
1703 }
1704 #endif
1705
1706 if (!(bond_dev->features & NETIF_F_LRO))
1707 dev_disable_lro(slave_dev);
1708
1709 res = netdev_rx_handler_register(slave_dev, bond_handle_frame,
1710 new_slave);
1711 if (res) {
1712 netdev_dbg(bond_dev, "Error %d calling netdev_rx_handler_register\n", res);
1713 goto err_detach;
1714 }
1715
1716 res = bond_master_upper_dev_link(bond, new_slave, extack);
1717 if (res) {
1718 netdev_dbg(bond_dev, "Error %d calling bond_master_upper_dev_link\n", res);
1719 goto err_unregister;
1720 }
1721
1722 res = bond_sysfs_slave_add(new_slave);
1723 if (res) {
1724 netdev_dbg(bond_dev, "Error %d calling bond_sysfs_slave_add\n", res);
1725 goto err_upper_unlink;
1726 }
1727
1728 bond->slave_cnt++;
1729 bond_compute_features(bond);
1730 bond_set_carrier(bond);
1731
1732 if (bond_uses_primary(bond)) {
1733 block_netpoll_tx();
1734 bond_select_active_slave(bond);
1735 unblock_netpoll_tx();
1736 }
1737
1738 if (bond_mode_uses_xmit_hash(bond))
1739 bond_update_slave_arr(bond, NULL);
1740
1741 netdev_info(bond_dev, "Enslaving %s as %s interface with %s link\n",
1742 slave_dev->name,
1743 bond_is_active_slave(new_slave) ? "an active" : "a backup",
1744 new_slave->link != BOND_LINK_DOWN ? "an up" : "a down");
1745
1746 /* enslave is successful */
1747 bond_queue_slave_event(new_slave);
1748 return 0;
1749
1750 /* Undo stages on error */
1751 err_upper_unlink:
1752 bond_upper_dev_unlink(bond, new_slave);
1753
1754 err_unregister:
1755 netdev_rx_handler_unregister(slave_dev);
1756
1757 err_detach:
1758 if (!bond_uses_primary(bond))
1759 bond_hw_addr_flush(bond_dev, slave_dev);
1760
1761 vlan_vids_del_by_dev(slave_dev, bond_dev);
1762 if (rcu_access_pointer(bond->primary_slave) == new_slave)
1763 RCU_INIT_POINTER(bond->primary_slave, NULL);
1764 if (rcu_access_pointer(bond->curr_active_slave) == new_slave) {
1765 block_netpoll_tx();
1766 bond_change_active_slave(bond, NULL);
1767 bond_select_active_slave(bond);
1768 unblock_netpoll_tx();
1769 }
1770 /* either primary_slave or curr_active_slave might've changed */
1771 synchronize_rcu();
1772 slave_disable_netpoll(new_slave);
1773
1774 err_close:
1775 slave_dev->priv_flags &= ~IFF_BONDING;
1776 dev_close(slave_dev);
1777
1778 err_restore_mac:
1779 slave_dev->flags &= ~IFF_SLAVE;
1780 if (!bond->params.fail_over_mac ||
1781 BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
1782 /* XXX TODO - fom follow mode needs to change master's
1783 * MAC if this slave's MAC is in use by the bond, or at
1784 * least print a warning.
1785 */
1786 bond_hw_addr_copy(ss.__data, new_slave->perm_hwaddr,
1787 new_slave->dev->addr_len);
1788 ss.ss_family = slave_dev->type;
1789 dev_set_mac_address(slave_dev, (struct sockaddr *)&ss);
1790 }
1791
1792 err_restore_mtu:
1793 dev_set_mtu(slave_dev, new_slave->original_mtu);
1794
1795 err_free:
1796 bond_free_slave(new_slave);
1797
1798 err_undo_flags:
1799 /* Enslave of first slave has failed and we need to fix master's mac */
1800 if (!bond_has_slaves(bond)) {
1801 if (ether_addr_equal_64bits(bond_dev->dev_addr,
1802 slave_dev->dev_addr))
1803 eth_hw_addr_random(bond_dev);
1804 if (bond_dev->type != ARPHRD_ETHER) {
1805 dev_close(bond_dev);
1806 ether_setup(bond_dev);
1807 bond_dev->flags |= IFF_MASTER;
1808 bond_dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1809 }
1810 }
1811
1812 return res;
1813 }
1814
1815 /* Try to release the slave device <slave> from the bond device <master>
1816 * It is legal to access curr_active_slave without a lock because all the function
1817 * is RTNL-locked. If "all" is true it means that the function is being called
1818 * while destroying a bond interface and all slaves are being released.
1819 *
1820 * The rules for slave state should be:
1821 * for Active/Backup:
1822 * Active stays on all backups go down
1823 * for Bonded connections:
1824 * The first up interface should be left on and all others downed.
1825 */
1826 static int __bond_release_one(struct net_device *bond_dev,
1827 struct net_device *slave_dev,
1828 bool all, bool unregister)
1829 {
1830 struct bonding *bond = netdev_priv(bond_dev);
1831 struct slave *slave, *oldcurrent;
1832 struct sockaddr_storage ss;
1833 int old_flags = bond_dev->flags;
1834 netdev_features_t old_features = bond_dev->features;
1835
1836 /* slave is not a slave or master is not master of this slave */
1837 if (!(slave_dev->flags & IFF_SLAVE) ||
1838 !netdev_has_upper_dev(slave_dev, bond_dev)) {
1839 netdev_dbg(bond_dev, "cannot release %s\n",
1840 slave_dev->name);
1841 return -EINVAL;
1842 }
1843
1844 block_netpoll_tx();
1845
1846 slave = bond_get_slave_by_dev(bond, slave_dev);
1847 if (!slave) {
1848 /* not a slave of this bond */
1849 netdev_info(bond_dev, "%s not enslaved\n",
1850 slave_dev->name);
1851 unblock_netpoll_tx();
1852 return -EINVAL;
1853 }
1854
1855 bond_set_slave_inactive_flags(slave, BOND_SLAVE_NOTIFY_NOW);
1856
1857 bond_sysfs_slave_del(slave);
1858
1859 /* recompute stats just before removing the slave */
1860 bond_get_stats(bond->dev, &bond->bond_stats);
1861
1862 bond_upper_dev_unlink(bond, slave);
1863 /* unregister rx_handler early so bond_handle_frame wouldn't be called
1864 * for this slave anymore.
1865 */
1866 netdev_rx_handler_unregister(slave_dev);
1867
1868 if (BOND_MODE(bond) == BOND_MODE_8023AD)
1869 bond_3ad_unbind_slave(slave);
1870
1871 if (bond_mode_uses_xmit_hash(bond))
1872 bond_update_slave_arr(bond, slave);
1873
1874 netdev_info(bond_dev, "Releasing %s interface %s\n",
1875 bond_is_active_slave(slave) ? "active" : "backup",
1876 slave_dev->name);
1877
1878 oldcurrent = rcu_access_pointer(bond->curr_active_slave);
1879
1880 RCU_INIT_POINTER(bond->current_arp_slave, NULL);
1881
1882 if (!all && (!bond->params.fail_over_mac ||
1883 BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP)) {
1884 if (ether_addr_equal_64bits(bond_dev->dev_addr, slave->perm_hwaddr) &&
1885 bond_has_slaves(bond))
1886 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",
1887 slave_dev->name, slave->perm_hwaddr,
1888 bond_dev->name, slave_dev->name);
1889 }
1890
1891 if (rtnl_dereference(bond->primary_slave) == slave)
1892 RCU_INIT_POINTER(bond->primary_slave, NULL);
1893
1894 if (oldcurrent == slave)
1895 bond_change_active_slave(bond, NULL);
1896
1897 if (bond_is_lb(bond)) {
1898 /* Must be called only after the slave has been
1899 * detached from the list and the curr_active_slave
1900 * has been cleared (if our_slave == old_current),
1901 * but before a new active slave is selected.
1902 */
1903 bond_alb_deinit_slave(bond, slave);
1904 }
1905
1906 if (all) {
1907 RCU_INIT_POINTER(bond->curr_active_slave, NULL);
1908 } else if (oldcurrent == slave) {
1909 /* Note that we hold RTNL over this sequence, so there
1910 * is no concern that another slave add/remove event
1911 * will interfere.
1912 */
1913 bond_select_active_slave(bond);
1914 }
1915
1916 if (!bond_has_slaves(bond)) {
1917 bond_set_carrier(bond);
1918 eth_hw_addr_random(bond_dev);
1919 }
1920
1921 unblock_netpoll_tx();
1922 synchronize_rcu();
1923 bond->slave_cnt--;
1924
1925 if (!bond_has_slaves(bond)) {
1926 call_netdevice_notifiers(NETDEV_CHANGEADDR, bond->dev);
1927 call_netdevice_notifiers(NETDEV_RELEASE, bond->dev);
1928 }
1929
1930 bond_compute_features(bond);
1931 if (!(bond_dev->features & NETIF_F_VLAN_CHALLENGED) &&
1932 (old_features & NETIF_F_VLAN_CHALLENGED))
1933 netdev_info(bond_dev, "last VLAN challenged slave %s left bond %s - VLAN blocking is removed\n",
1934 slave_dev->name, bond_dev->name);
1935
1936 vlan_vids_del_by_dev(slave_dev, bond_dev);
1937
1938 /* If the mode uses primary, then this case was handled above by
1939 * bond_change_active_slave(..., NULL)
1940 */
1941 if (!bond_uses_primary(bond)) {
1942 /* unset promiscuity level from slave
1943 * NOTE: The NETDEV_CHANGEADDR call above may change the value
1944 * of the IFF_PROMISC flag in the bond_dev, but we need the
1945 * value of that flag before that change, as that was the value
1946 * when this slave was attached, so we cache at the start of the
1947 * function and use it here. Same goes for ALLMULTI below
1948 */
1949 if (old_flags & IFF_PROMISC)
1950 dev_set_promiscuity(slave_dev, -1);
1951
1952 /* unset allmulti level from slave */
1953 if (old_flags & IFF_ALLMULTI)
1954 dev_set_allmulti(slave_dev, -1);
1955
1956 bond_hw_addr_flush(bond_dev, slave_dev);
1957 }
1958
1959 slave_disable_netpoll(slave);
1960
1961 /* close slave before restoring its mac address */
1962 dev_close(slave_dev);
1963
1964 if (bond->params.fail_over_mac != BOND_FOM_ACTIVE ||
1965 BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
1966 /* restore original ("permanent") mac address */
1967 bond_hw_addr_copy(ss.__data, slave->perm_hwaddr,
1968 slave->dev->addr_len);
1969 ss.ss_family = slave_dev->type;
1970 dev_set_mac_address(slave_dev, (struct sockaddr *)&ss);
1971 }
1972
1973 if (unregister)
1974 __dev_set_mtu(slave_dev, slave->original_mtu);
1975 else
1976 dev_set_mtu(slave_dev, slave->original_mtu);
1977
1978 slave_dev->priv_flags &= ~IFF_BONDING;
1979
1980 bond_free_slave(slave);
1981
1982 return 0;
1983 }
1984
1985 /* A wrapper used because of ndo_del_link */
1986 int bond_release(struct net_device *bond_dev, struct net_device *slave_dev)
1987 {
1988 return __bond_release_one(bond_dev, slave_dev, false, false);
1989 }
1990
1991 /* First release a slave and then destroy the bond if no more slaves are left.
1992 * Must be under rtnl_lock when this function is called.
1993 */
1994 static int bond_release_and_destroy(struct net_device *bond_dev,
1995 struct net_device *slave_dev)
1996 {
1997 struct bonding *bond = netdev_priv(bond_dev);
1998 int ret;
1999
2000 ret = __bond_release_one(bond_dev, slave_dev, false, true);
2001 if (ret == 0 && !bond_has_slaves(bond)) {
2002 bond_dev->priv_flags |= IFF_DISABLE_NETPOLL;
2003 netdev_info(bond_dev, "Destroying bond %s\n",
2004 bond_dev->name);
2005 bond_remove_proc_entry(bond);
2006 unregister_netdevice(bond_dev);
2007 }
2008 return ret;
2009 }
2010
2011 static void bond_info_query(struct net_device *bond_dev, struct ifbond *info)
2012 {
2013 struct bonding *bond = netdev_priv(bond_dev);
2014 bond_fill_ifbond(bond, info);
2015 }
2016
2017 static int bond_slave_info_query(struct net_device *bond_dev, struct ifslave *info)
2018 {
2019 struct bonding *bond = netdev_priv(bond_dev);
2020 struct list_head *iter;
2021 int i = 0, res = -ENODEV;
2022 struct slave *slave;
2023
2024 bond_for_each_slave(bond, slave, iter) {
2025 if (i++ == (int)info->slave_id) {
2026 res = 0;
2027 bond_fill_ifslave(slave, info);
2028 break;
2029 }
2030 }
2031
2032 return res;
2033 }
2034
2035 /*-------------------------------- Monitoring -------------------------------*/
2036
2037 /* called with rcu_read_lock() */
2038 static int bond_miimon_inspect(struct bonding *bond)
2039 {
2040 int link_state, commit = 0;
2041 struct list_head *iter;
2042 struct slave *slave;
2043 bool ignore_updelay;
2044
2045 ignore_updelay = !rcu_dereference(bond->curr_active_slave);
2046
2047 bond_for_each_slave_rcu(bond, slave, iter) {
2048 slave->new_link = BOND_LINK_NOCHANGE;
2049 slave->link_new_state = slave->link;
2050
2051 link_state = bond_check_dev_link(bond, slave->dev, 0);
2052
2053 switch (slave->link) {
2054 case BOND_LINK_UP:
2055 if (link_state)
2056 continue;
2057
2058 bond_propose_link_state(slave, BOND_LINK_FAIL);
2059 commit++;
2060 slave->delay = bond->params.downdelay;
2061 if (slave->delay) {
2062 netdev_info(bond->dev, "link status down for %sinterface %s, disabling it in %d ms\n",
2063 (BOND_MODE(bond) ==
2064 BOND_MODE_ACTIVEBACKUP) ?
2065 (bond_is_active_slave(slave) ?
2066 "active " : "backup ") : "",
2067 slave->dev->name,
2068 bond->params.downdelay * bond->params.miimon);
2069 }
2070 /*FALLTHRU*/
2071 case BOND_LINK_FAIL:
2072 if (link_state) {
2073 /* recovered before downdelay expired */
2074 bond_propose_link_state(slave, BOND_LINK_UP);
2075 slave->last_link_up = jiffies;
2076 netdev_info(bond->dev, "link status up again after %d ms for interface %s\n",
2077 (bond->params.downdelay - slave->delay) *
2078 bond->params.miimon,
2079 slave->dev->name);
2080 commit++;
2081 continue;
2082 }
2083
2084 if (slave->delay <= 0) {
2085 slave->new_link = BOND_LINK_DOWN;
2086 commit++;
2087 continue;
2088 }
2089
2090 slave->delay--;
2091 break;
2092
2093 case BOND_LINK_DOWN:
2094 if (!link_state)
2095 continue;
2096
2097 bond_propose_link_state(slave, BOND_LINK_BACK);
2098 commit++;
2099 slave->delay = bond->params.updelay;
2100
2101 if (slave->delay) {
2102 netdev_info(bond->dev, "link status up for interface %s, enabling it in %d ms\n",
2103 slave->dev->name,
2104 ignore_updelay ? 0 :
2105 bond->params.updelay *
2106 bond->params.miimon);
2107 }
2108 /*FALLTHRU*/
2109 case BOND_LINK_BACK:
2110 if (!link_state) {
2111 bond_propose_link_state(slave, BOND_LINK_DOWN);
2112 netdev_info(bond->dev, "link status down again after %d ms for interface %s\n",
2113 (bond->params.updelay - slave->delay) *
2114 bond->params.miimon,
2115 slave->dev->name);
2116 commit++;
2117 continue;
2118 }
2119
2120 if (ignore_updelay)
2121 slave->delay = 0;
2122
2123 if (slave->delay <= 0) {
2124 slave->new_link = BOND_LINK_UP;
2125 commit++;
2126 ignore_updelay = false;
2127 continue;
2128 }
2129
2130 slave->delay--;
2131 break;
2132 }
2133 }
2134
2135 return commit;
2136 }
2137
2138 static void bond_miimon_commit(struct bonding *bond)
2139 {
2140 struct list_head *iter;
2141 struct slave *slave, *primary;
2142
2143 bond_for_each_slave(bond, slave, iter) {
2144 switch (slave->new_link) {
2145 case BOND_LINK_NOCHANGE:
2146 continue;
2147
2148 case BOND_LINK_UP:
2149 if (bond_update_speed_duplex(slave) &&
2150 bond_needs_speed_duplex(bond)) {
2151 slave->link = BOND_LINK_DOWN;
2152 if (net_ratelimit())
2153 netdev_warn(bond->dev,
2154 "failed to get link speed/duplex for %s\n",
2155 slave->dev->name);
2156 continue;
2157 }
2158 bond_set_slave_link_state(slave, BOND_LINK_UP,
2159 BOND_SLAVE_NOTIFY_NOW);
2160 slave->last_link_up = jiffies;
2161
2162 primary = rtnl_dereference(bond->primary_slave);
2163 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
2164 /* prevent it from being the active one */
2165 bond_set_backup_slave(slave);
2166 } else if (BOND_MODE(bond) != BOND_MODE_ACTIVEBACKUP) {
2167 /* make it immediately active */
2168 bond_set_active_slave(slave);
2169 } else if (slave != primary) {
2170 /* prevent it from being the active one */
2171 bond_set_backup_slave(slave);
2172 }
2173
2174 netdev_info(bond->dev, "link status definitely up for interface %s, %u Mbps %s duplex\n",
2175 slave->dev->name,
2176 slave->speed == SPEED_UNKNOWN ? 0 : slave->speed,
2177 slave->duplex ? "full" : "half");
2178
2179 /* notify ad that the link status has changed */
2180 if (BOND_MODE(bond) == BOND_MODE_8023AD)
2181 bond_3ad_handle_link_change(slave, BOND_LINK_UP);
2182
2183 if (bond_is_lb(bond))
2184 bond_alb_handle_link_change(bond, slave,
2185 BOND_LINK_UP);
2186
2187 if (BOND_MODE(bond) == BOND_MODE_XOR)
2188 bond_update_slave_arr(bond, NULL);
2189
2190 if (!bond->curr_active_slave || slave == primary)
2191 goto do_failover;
2192
2193 continue;
2194
2195 case BOND_LINK_DOWN:
2196 if (slave->link_failure_count < UINT_MAX)
2197 slave->link_failure_count++;
2198
2199 bond_set_slave_link_state(slave, BOND_LINK_DOWN,
2200 BOND_SLAVE_NOTIFY_NOW);
2201
2202 if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP ||
2203 BOND_MODE(bond) == BOND_MODE_8023AD)
2204 bond_set_slave_inactive_flags(slave,
2205 BOND_SLAVE_NOTIFY_NOW);
2206
2207 netdev_info(bond->dev, "link status definitely down for interface %s, disabling it\n",
2208 slave->dev->name);
2209
2210 if (BOND_MODE(bond) == BOND_MODE_8023AD)
2211 bond_3ad_handle_link_change(slave,
2212 BOND_LINK_DOWN);
2213
2214 if (bond_is_lb(bond))
2215 bond_alb_handle_link_change(bond, slave,
2216 BOND_LINK_DOWN);
2217
2218 if (BOND_MODE(bond) == BOND_MODE_XOR)
2219 bond_update_slave_arr(bond, NULL);
2220
2221 if (slave == rcu_access_pointer(bond->curr_active_slave))
2222 goto do_failover;
2223
2224 continue;
2225
2226 default:
2227 netdev_err(bond->dev, "invalid new link %d on slave %s\n",
2228 slave->new_link, slave->dev->name);
2229 slave->new_link = BOND_LINK_NOCHANGE;
2230
2231 continue;
2232 }
2233
2234 do_failover:
2235 block_netpoll_tx();
2236 bond_select_active_slave(bond);
2237 unblock_netpoll_tx();
2238 }
2239
2240 bond_set_carrier(bond);
2241 }
2242
2243 /* bond_mii_monitor
2244 *
2245 * Really a wrapper that splits the mii monitor into two phases: an
2246 * inspection, then (if inspection indicates something needs to be done)
2247 * an acquisition of appropriate locks followed by a commit phase to
2248 * implement whatever link state changes are indicated.
2249 */
2250 static void bond_mii_monitor(struct work_struct *work)
2251 {
2252 struct bonding *bond = container_of(work, struct bonding,
2253 mii_work.work);
2254 bool should_notify_peers = false;
2255 unsigned long delay;
2256 struct slave *slave;
2257 struct list_head *iter;
2258
2259 delay = msecs_to_jiffies(bond->params.miimon);
2260
2261 if (!bond_has_slaves(bond))
2262 goto re_arm;
2263
2264 rcu_read_lock();
2265
2266 should_notify_peers = bond_should_notify_peers(bond);
2267
2268 if (bond_miimon_inspect(bond)) {
2269 rcu_read_unlock();
2270
2271 /* Race avoidance with bond_close cancel of workqueue */
2272 if (!rtnl_trylock()) {
2273 delay = 1;
2274 should_notify_peers = false;
2275 goto re_arm;
2276 }
2277
2278 bond_for_each_slave(bond, slave, iter) {
2279 bond_commit_link_state(slave, BOND_SLAVE_NOTIFY_LATER);
2280 }
2281 bond_miimon_commit(bond);
2282
2283 rtnl_unlock(); /* might sleep, hold no other locks */
2284 } else
2285 rcu_read_unlock();
2286
2287 re_arm:
2288 if (bond->params.miimon)
2289 queue_delayed_work(bond->wq, &bond->mii_work, delay);
2290
2291 if (should_notify_peers) {
2292 if (!rtnl_trylock())
2293 return;
2294 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, bond->dev);
2295 rtnl_unlock();
2296 }
2297 }
2298
2299 static int bond_upper_dev_walk(struct net_device *upper, void *data)
2300 {
2301 __be32 ip = *((__be32 *)data);
2302
2303 return ip == bond_confirm_addr(upper, 0, ip);
2304 }
2305
2306 static bool bond_has_this_ip(struct bonding *bond, __be32 ip)
2307 {
2308 bool ret = false;
2309
2310 if (ip == bond_confirm_addr(bond->dev, 0, ip))
2311 return true;
2312
2313 rcu_read_lock();
2314 if (netdev_walk_all_upper_dev_rcu(bond->dev, bond_upper_dev_walk, &ip))
2315 ret = true;
2316 rcu_read_unlock();
2317
2318 return ret;
2319 }
2320
2321 /* We go to the (large) trouble of VLAN tagging ARP frames because
2322 * switches in VLAN mode (especially if ports are configured as
2323 * "native" to a VLAN) might not pass non-tagged frames.
2324 */
2325 static void bond_arp_send(struct net_device *slave_dev, int arp_op,
2326 __be32 dest_ip, __be32 src_ip,
2327 struct bond_vlan_tag *tags)
2328 {
2329 struct sk_buff *skb;
2330 struct bond_vlan_tag *outer_tag = tags;
2331
2332 netdev_dbg(slave_dev, "arp %d on slave %s: dst %pI4 src %pI4\n",
2333 arp_op, slave_dev->name, &dest_ip, &src_ip);
2334
2335 skb = arp_create(arp_op, ETH_P_ARP, dest_ip, slave_dev, src_ip,
2336 NULL, slave_dev->dev_addr, NULL);
2337
2338 if (!skb) {
2339 net_err_ratelimited("ARP packet allocation failed\n");
2340 return;
2341 }
2342
2343 if (!tags || tags->vlan_proto == VLAN_N_VID)
2344 goto xmit;
2345
2346 tags++;
2347
2348 /* Go through all the tags backwards and add them to the packet */
2349 while (tags->vlan_proto != VLAN_N_VID) {
2350 if (!tags->vlan_id) {
2351 tags++;
2352 continue;
2353 }
2354
2355 netdev_dbg(slave_dev, "inner tag: proto %X vid %X\n",
2356 ntohs(outer_tag->vlan_proto), tags->vlan_id);
2357 skb = vlan_insert_tag_set_proto(skb, tags->vlan_proto,
2358 tags->vlan_id);
2359 if (!skb) {
2360 net_err_ratelimited("failed to insert inner VLAN tag\n");
2361 return;
2362 }
2363
2364 tags++;
2365 }
2366 /* Set the outer tag */
2367 if (outer_tag->vlan_id) {
2368 netdev_dbg(slave_dev, "outer tag: proto %X vid %X\n",
2369 ntohs(outer_tag->vlan_proto), outer_tag->vlan_id);
2370 __vlan_hwaccel_put_tag(skb, outer_tag->vlan_proto,
2371 outer_tag->vlan_id);
2372 }
2373
2374 xmit:
2375 arp_xmit(skb);
2376 }
2377
2378 /* Validate the device path between the @start_dev and the @end_dev.
2379 * The path is valid if the @end_dev is reachable through device
2380 * stacking.
2381 * When the path is validated, collect any vlan information in the
2382 * path.
2383 */
2384 struct bond_vlan_tag *bond_verify_device_path(struct net_device *start_dev,
2385 struct net_device *end_dev,
2386 int level)
2387 {
2388 struct bond_vlan_tag *tags;
2389 struct net_device *upper;
2390 struct list_head *iter;
2391
2392 if (start_dev == end_dev) {
2393 tags = kzalloc(sizeof(*tags) * (level + 1), GFP_ATOMIC);
2394 if (!tags)
2395 return ERR_PTR(-ENOMEM);
2396 tags[level].vlan_proto = VLAN_N_VID;
2397 return tags;
2398 }
2399
2400 netdev_for_each_upper_dev_rcu(start_dev, upper, iter) {
2401 tags = bond_verify_device_path(upper, end_dev, level + 1);
2402 if (IS_ERR_OR_NULL(tags)) {
2403 if (IS_ERR(tags))
2404 return tags;
2405 continue;
2406 }
2407 if (is_vlan_dev(upper)) {
2408 tags[level].vlan_proto = vlan_dev_vlan_proto(upper);
2409 tags[level].vlan_id = vlan_dev_vlan_id(upper);
2410 }
2411
2412 return tags;
2413 }
2414
2415 return NULL;
2416 }
2417
2418 static void bond_arp_send_all(struct bonding *bond, struct slave *slave)
2419 {
2420 struct rtable *rt;
2421 struct bond_vlan_tag *tags;
2422 __be32 *targets = bond->params.arp_targets, addr;
2423 int i;
2424
2425 for (i = 0; i < BOND_MAX_ARP_TARGETS && targets[i]; i++) {
2426 netdev_dbg(bond->dev, "basa: target %pI4\n", &targets[i]);
2427 tags = NULL;
2428
2429 /* Find out through which dev should the packet go */
2430 rt = ip_route_output(dev_net(bond->dev), targets[i], 0,
2431 RTO_ONLINK, 0);
2432 if (IS_ERR(rt)) {
2433 /* there's no route to target - try to send arp
2434 * probe to generate any traffic (arp_validate=0)
2435 */
2436 if (bond->params.arp_validate)
2437 net_warn_ratelimited("%s: no route to arp_ip_target %pI4 and arp_validate is set\n",
2438 bond->dev->name,
2439 &targets[i]);
2440 bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2441 0, tags);
2442 continue;
2443 }
2444
2445 /* bond device itself */
2446 if (rt->dst.dev == bond->dev)
2447 goto found;
2448
2449 rcu_read_lock();
2450 tags = bond_verify_device_path(bond->dev, rt->dst.dev, 0);
2451 rcu_read_unlock();
2452
2453 if (!IS_ERR_OR_NULL(tags))
2454 goto found;
2455
2456 /* Not our device - skip */
2457 netdev_dbg(bond->dev, "no path to arp_ip_target %pI4 via rt.dev %s\n",
2458 &targets[i], rt->dst.dev ? rt->dst.dev->name : "NULL");
2459
2460 ip_rt_put(rt);
2461 continue;
2462
2463 found:
2464 addr = bond_confirm_addr(rt->dst.dev, targets[i], 0);
2465 ip_rt_put(rt);
2466 bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2467 addr, tags);
2468 kfree(tags);
2469 }
2470 }
2471
2472 static void bond_validate_arp(struct bonding *bond, struct slave *slave, __be32 sip, __be32 tip)
2473 {
2474 int i;
2475
2476 if (!sip || !bond_has_this_ip(bond, tip)) {
2477 netdev_dbg(bond->dev, "bva: sip %pI4 tip %pI4 not found\n",
2478 &sip, &tip);
2479 return;
2480 }
2481
2482 i = bond_get_targets_ip(bond->params.arp_targets, sip);
2483 if (i == -1) {
2484 netdev_dbg(bond->dev, "bva: sip %pI4 not found in targets\n",
2485 &sip);
2486 return;
2487 }
2488 slave->last_rx = jiffies;
2489 slave->target_last_arp_rx[i] = jiffies;
2490 }
2491
2492 int bond_arp_rcv(const struct sk_buff *skb, struct bonding *bond,
2493 struct slave *slave)
2494 {
2495 struct arphdr *arp = (struct arphdr *)skb->data;
2496 struct slave *curr_active_slave, *curr_arp_slave;
2497 unsigned char *arp_ptr;
2498 __be32 sip, tip;
2499 int is_arp = skb->protocol == __cpu_to_be16(ETH_P_ARP);
2500 unsigned int alen;
2501
2502 if (!slave_do_arp_validate(bond, slave)) {
2503 if ((slave_do_arp_validate_only(bond) && is_arp) ||
2504 !slave_do_arp_validate_only(bond))
2505 slave->last_rx = jiffies;
2506 return RX_HANDLER_ANOTHER;
2507 } else if (!is_arp) {
2508 return RX_HANDLER_ANOTHER;
2509 }
2510
2511 alen = arp_hdr_len(bond->dev);
2512
2513 netdev_dbg(bond->dev, "bond_arp_rcv: skb->dev %s\n",
2514 skb->dev->name);
2515
2516 if (alen > skb_headlen(skb)) {
2517 arp = kmalloc(alen, GFP_ATOMIC);
2518 if (!arp)
2519 goto out_unlock;
2520 if (skb_copy_bits(skb, 0, arp, alen) < 0)
2521 goto out_unlock;
2522 }
2523
2524 if (arp->ar_hln != bond->dev->addr_len ||
2525 skb->pkt_type == PACKET_OTHERHOST ||
2526 skb->pkt_type == PACKET_LOOPBACK ||
2527 arp->ar_hrd != htons(ARPHRD_ETHER) ||
2528 arp->ar_pro != htons(ETH_P_IP) ||
2529 arp->ar_pln != 4)
2530 goto out_unlock;
2531
2532 arp_ptr = (unsigned char *)(arp + 1);
2533 arp_ptr += bond->dev->addr_len;
2534 memcpy(&sip, arp_ptr, 4);
2535 arp_ptr += 4 + bond->dev->addr_len;
2536 memcpy(&tip, arp_ptr, 4);
2537
2538 netdev_dbg(bond->dev, "bond_arp_rcv: %s/%d av %d sv %d sip %pI4 tip %pI4\n",
2539 slave->dev->name, bond_slave_state(slave),
2540 bond->params.arp_validate, slave_do_arp_validate(bond, slave),
2541 &sip, &tip);
2542
2543 curr_active_slave = rcu_dereference(bond->curr_active_slave);
2544 curr_arp_slave = rcu_dereference(bond->current_arp_slave);
2545
2546 /* We 'trust' the received ARP enough to validate it if:
2547 *
2548 * (a) the slave receiving the ARP is active (which includes the
2549 * current ARP slave, if any), or
2550 *
2551 * (b) the receiving slave isn't active, but there is a currently
2552 * active slave and it received valid arp reply(s) after it became
2553 * the currently active slave, or
2554 *
2555 * (c) there is an ARP slave that sent an ARP during the prior ARP
2556 * interval, and we receive an ARP reply on any slave. We accept
2557 * these because switch FDB update delays may deliver the ARP
2558 * reply to a slave other than the sender of the ARP request.
2559 *
2560 * Note: for (b), backup slaves are receiving the broadcast ARP
2561 * request, not a reply. This request passes from the sending
2562 * slave through the L2 switch(es) to the receiving slave. Since
2563 * this is checking the request, sip/tip are swapped for
2564 * validation.
2565 *
2566 * This is done to avoid endless looping when we can't reach the
2567 * arp_ip_target and fool ourselves with our own arp requests.
2568 */
2569 if (bond_is_active_slave(slave))
2570 bond_validate_arp(bond, slave, sip, tip);
2571 else if (curr_active_slave &&
2572 time_after(slave_last_rx(bond, curr_active_slave),
2573 curr_active_slave->last_link_up))
2574 bond_validate_arp(bond, slave, tip, sip);
2575 else if (curr_arp_slave && (arp->ar_op == htons(ARPOP_REPLY)) &&
2576 bond_time_in_interval(bond,
2577 dev_trans_start(curr_arp_slave->dev), 1))
2578 bond_validate_arp(bond, slave, sip, tip);
2579
2580 out_unlock:
2581 if (arp != (struct arphdr *)skb->data)
2582 kfree(arp);
2583 return RX_HANDLER_ANOTHER;
2584 }
2585
2586 /* function to verify if we're in the arp_interval timeslice, returns true if
2587 * (last_act - arp_interval) <= jiffies <= (last_act + mod * arp_interval +
2588 * arp_interval/2) . the arp_interval/2 is needed for really fast networks.
2589 */
2590 static bool bond_time_in_interval(struct bonding *bond, unsigned long last_act,
2591 int mod)
2592 {
2593 int delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
2594
2595 return time_in_range(jiffies,
2596 last_act - delta_in_ticks,
2597 last_act + mod * delta_in_ticks + delta_in_ticks/2);
2598 }
2599
2600 /* This function is called regularly to monitor each slave's link
2601 * ensuring that traffic is being sent and received when arp monitoring
2602 * is used in load-balancing mode. if the adapter has been dormant, then an
2603 * arp is transmitted to generate traffic. see activebackup_arp_monitor for
2604 * arp monitoring in active backup mode.
2605 */
2606 static void bond_loadbalance_arp_mon(struct bonding *bond)
2607 {
2608 struct slave *slave, *oldcurrent;
2609 struct list_head *iter;
2610 int do_failover = 0, slave_state_changed = 0;
2611
2612 if (!bond_has_slaves(bond))
2613 goto re_arm;
2614
2615 rcu_read_lock();
2616
2617 oldcurrent = rcu_dereference(bond->curr_active_slave);
2618 /* see if any of the previous devices are up now (i.e. they have
2619 * xmt and rcv traffic). the curr_active_slave does not come into
2620 * the picture unless it is null. also, slave->last_link_up is not
2621 * needed here because we send an arp on each slave and give a slave
2622 * as long as it needs to get the tx/rx within the delta.
2623 * TODO: what about up/down delay in arp mode? it wasn't here before
2624 * so it can wait
2625 */
2626 bond_for_each_slave_rcu(bond, slave, iter) {
2627 unsigned long trans_start = dev_trans_start(slave->dev);
2628
2629 slave->new_link = BOND_LINK_NOCHANGE;
2630
2631 if (slave->link != BOND_LINK_UP) {
2632 if (bond_time_in_interval(bond, trans_start, 1) &&
2633 bond_time_in_interval(bond, slave->last_rx, 1)) {
2634
2635 slave->new_link = BOND_LINK_UP;
2636 slave_state_changed = 1;
2637
2638 /* primary_slave has no meaning in round-robin
2639 * mode. the window of a slave being up and
2640 * curr_active_slave being null after enslaving
2641 * is closed.
2642 */
2643 if (!oldcurrent) {
2644 netdev_info(bond->dev, "link status definitely up for interface %s\n",
2645 slave->dev->name);
2646 do_failover = 1;
2647 } else {
2648 netdev_info(bond->dev, "interface %s is now up\n",
2649 slave->dev->name);
2650 }
2651 }
2652 } else {
2653 /* slave->link == BOND_LINK_UP */
2654
2655 /* not all switches will respond to an arp request
2656 * when the source ip is 0, so don't take the link down
2657 * if we don't know our ip yet
2658 */
2659 if (!bond_time_in_interval(bond, trans_start, 2) ||
2660 !bond_time_in_interval(bond, slave->last_rx, 2)) {
2661
2662 slave->new_link = BOND_LINK_DOWN;
2663 slave_state_changed = 1;
2664
2665 if (slave->link_failure_count < UINT_MAX)
2666 slave->link_failure_count++;
2667
2668 netdev_info(bond->dev, "interface %s is now down\n",
2669 slave->dev->name);
2670
2671 if (slave == oldcurrent)
2672 do_failover = 1;
2673 }
2674 }
2675
2676 /* note: if switch is in round-robin mode, all links
2677 * must tx arp to ensure all links rx an arp - otherwise
2678 * links may oscillate or not come up at all; if switch is
2679 * in something like xor mode, there is nothing we can
2680 * do - all replies will be rx'ed on same link causing slaves
2681 * to be unstable during low/no traffic periods
2682 */
2683 if (bond_slave_is_up(slave))
2684 bond_arp_send_all(bond, slave);
2685 }
2686
2687 rcu_read_unlock();
2688
2689 if (do_failover || slave_state_changed) {
2690 if (!rtnl_trylock())
2691 goto re_arm;
2692
2693 bond_for_each_slave(bond, slave, iter) {
2694 if (slave->new_link != BOND_LINK_NOCHANGE)
2695 slave->link = slave->new_link;
2696 }
2697
2698 if (slave_state_changed) {
2699 bond_slave_state_change(bond);
2700 if (BOND_MODE(bond) == BOND_MODE_XOR)
2701 bond_update_slave_arr(bond, NULL);
2702 }
2703 if (do_failover) {
2704 block_netpoll_tx();
2705 bond_select_active_slave(bond);
2706 unblock_netpoll_tx();
2707 }
2708 rtnl_unlock();
2709 }
2710
2711 re_arm:
2712 if (bond->params.arp_interval)
2713 queue_delayed_work(bond->wq, &bond->arp_work,
2714 msecs_to_jiffies(bond->params.arp_interval));
2715 }
2716
2717 /* Called to inspect slaves for active-backup mode ARP monitor link state
2718 * changes. Sets new_link in slaves to specify what action should take
2719 * place for the slave. Returns 0 if no changes are found, >0 if changes
2720 * to link states must be committed.
2721 *
2722 * Called with rcu_read_lock held.
2723 */
2724 static int bond_ab_arp_inspect(struct bonding *bond)
2725 {
2726 unsigned long trans_start, last_rx;
2727 struct list_head *iter;
2728 struct slave *slave;
2729 int commit = 0;
2730
2731 bond_for_each_slave_rcu(bond, slave, iter) {
2732 slave->new_link = BOND_LINK_NOCHANGE;
2733 last_rx = slave_last_rx(bond, slave);
2734
2735 if (slave->link != BOND_LINK_UP) {
2736 if (bond_time_in_interval(bond, last_rx, 1)) {
2737 slave->new_link = BOND_LINK_UP;
2738 commit++;
2739 }
2740 continue;
2741 }
2742
2743 /* Give slaves 2*delta after being enslaved or made
2744 * active. This avoids bouncing, as the last receive
2745 * times need a full ARP monitor cycle to be updated.
2746 */
2747 if (bond_time_in_interval(bond, slave->last_link_up, 2))
2748 continue;
2749
2750 /* Backup slave is down if:
2751 * - No current_arp_slave AND
2752 * - more than 3*delta since last receive AND
2753 * - the bond has an IP address
2754 *
2755 * Note: a non-null current_arp_slave indicates
2756 * the curr_active_slave went down and we are
2757 * searching for a new one; under this condition
2758 * we only take the curr_active_slave down - this
2759 * gives each slave a chance to tx/rx traffic
2760 * before being taken out
2761 */
2762 if (!bond_is_active_slave(slave) &&
2763 !rcu_access_pointer(bond->current_arp_slave) &&
2764 !bond_time_in_interval(bond, last_rx, 3)) {
2765 slave->new_link = BOND_LINK_DOWN;
2766 commit++;
2767 }
2768
2769 /* Active slave is down if:
2770 * - more than 2*delta since transmitting OR
2771 * - (more than 2*delta since receive AND
2772 * the bond has an IP address)
2773 */
2774 trans_start = dev_trans_start(slave->dev);
2775 if (bond_is_active_slave(slave) &&
2776 (!bond_time_in_interval(bond, trans_start, 2) ||
2777 !bond_time_in_interval(bond, last_rx, 2))) {
2778 slave->new_link = BOND_LINK_DOWN;
2779 commit++;
2780 }
2781 }
2782
2783 return commit;
2784 }
2785
2786 /* Called to commit link state changes noted by inspection step of
2787 * active-backup mode ARP monitor.
2788 *
2789 * Called with RTNL hold.
2790 */
2791 static void bond_ab_arp_commit(struct bonding *bond)
2792 {
2793 unsigned long trans_start;
2794 struct list_head *iter;
2795 struct slave *slave;
2796
2797 bond_for_each_slave(bond, slave, iter) {
2798 switch (slave->new_link) {
2799 case BOND_LINK_NOCHANGE:
2800 continue;
2801
2802 case BOND_LINK_UP:
2803 trans_start = dev_trans_start(slave->dev);
2804 if (rtnl_dereference(bond->curr_active_slave) != slave ||
2805 (!rtnl_dereference(bond->curr_active_slave) &&
2806 bond_time_in_interval(bond, trans_start, 1))) {
2807 struct slave *current_arp_slave;
2808
2809 current_arp_slave = rtnl_dereference(bond->current_arp_slave);
2810 bond_set_slave_link_state(slave, BOND_LINK_UP,
2811 BOND_SLAVE_NOTIFY_NOW);
2812 if (current_arp_slave) {
2813 bond_set_slave_inactive_flags(
2814 current_arp_slave,
2815 BOND_SLAVE_NOTIFY_NOW);
2816 RCU_INIT_POINTER(bond->current_arp_slave, NULL);
2817 }
2818
2819 netdev_info(bond->dev, "link status definitely up for interface %s\n",
2820 slave->dev->name);
2821
2822 if (!rtnl_dereference(bond->curr_active_slave) ||
2823 slave == rtnl_dereference(bond->primary_slave))
2824 goto do_failover;
2825
2826 }
2827
2828 continue;
2829
2830 case BOND_LINK_DOWN:
2831 if (slave->link_failure_count < UINT_MAX)
2832 slave->link_failure_count++;
2833
2834 bond_set_slave_link_state(slave, BOND_LINK_DOWN,
2835 BOND_SLAVE_NOTIFY_NOW);
2836 bond_set_slave_inactive_flags(slave,
2837 BOND_SLAVE_NOTIFY_NOW);
2838
2839 netdev_info(bond->dev, "link status definitely down for interface %s, disabling it\n",
2840 slave->dev->name);
2841
2842 if (slave == rtnl_dereference(bond->curr_active_slave)) {
2843 RCU_INIT_POINTER(bond->current_arp_slave, NULL);
2844 goto do_failover;
2845 }
2846
2847 continue;
2848
2849 default:
2850 netdev_err(bond->dev, "impossible: new_link %d on slave %s\n",
2851 slave->new_link, slave->dev->name);
2852 continue;
2853 }
2854
2855 do_failover:
2856 block_netpoll_tx();
2857 bond_select_active_slave(bond);
2858 unblock_netpoll_tx();
2859 }
2860
2861 bond_set_carrier(bond);
2862 }
2863
2864 /* Send ARP probes for active-backup mode ARP monitor.
2865 *
2866 * Called with rcu_read_lock held.
2867 */
2868 static bool bond_ab_arp_probe(struct bonding *bond)
2869 {
2870 struct slave *slave, *before = NULL, *new_slave = NULL,
2871 *curr_arp_slave = rcu_dereference(bond->current_arp_slave),
2872 *curr_active_slave = rcu_dereference(bond->curr_active_slave);
2873 struct list_head *iter;
2874 bool found = false;
2875 bool should_notify_rtnl = BOND_SLAVE_NOTIFY_LATER;
2876
2877 if (curr_arp_slave && curr_active_slave)
2878 netdev_info(bond->dev, "PROBE: c_arp %s && cas %s BAD\n",
2879 curr_arp_slave->dev->name,
2880 curr_active_slave->dev->name);
2881
2882 if (curr_active_slave) {
2883 bond_arp_send_all(bond, curr_active_slave);
2884 return should_notify_rtnl;
2885 }
2886
2887 /* if we don't have a curr_active_slave, search for the next available
2888 * backup slave from the current_arp_slave and make it the candidate
2889 * for becoming the curr_active_slave
2890 */
2891
2892 if (!curr_arp_slave) {
2893 curr_arp_slave = bond_first_slave_rcu(bond);
2894 if (!curr_arp_slave)
2895 return should_notify_rtnl;
2896 }
2897
2898 bond_set_slave_inactive_flags(curr_arp_slave, BOND_SLAVE_NOTIFY_LATER);
2899
2900 bond_for_each_slave_rcu(bond, slave, iter) {
2901 if (!found && !before && bond_slave_is_up(slave))
2902 before = slave;
2903
2904 if (found && !new_slave && bond_slave_is_up(slave))
2905 new_slave = slave;
2906 /* if the link state is up at this point, we
2907 * mark it down - this can happen if we have
2908 * simultaneous link failures and
2909 * reselect_active_interface doesn't make this
2910 * one the current slave so it is still marked
2911 * up when it is actually down
2912 */
2913 if (!bond_slave_is_up(slave) && slave->link == BOND_LINK_UP) {
2914 bond_set_slave_link_state(slave, BOND_LINK_DOWN,
2915 BOND_SLAVE_NOTIFY_LATER);
2916 if (slave->link_failure_count < UINT_MAX)
2917 slave->link_failure_count++;
2918
2919 bond_set_slave_inactive_flags(slave,
2920 BOND_SLAVE_NOTIFY_LATER);
2921
2922 netdev_info(bond->dev, "backup interface %s is now down\n",
2923 slave->dev->name);
2924 }
2925 if (slave == curr_arp_slave)
2926 found = true;
2927 }
2928
2929 if (!new_slave && before)
2930 new_slave = before;
2931
2932 if (!new_slave)
2933 goto check_state;
2934
2935 bond_set_slave_link_state(new_slave, BOND_LINK_BACK,
2936 BOND_SLAVE_NOTIFY_LATER);
2937 bond_set_slave_active_flags(new_slave, BOND_SLAVE_NOTIFY_LATER);
2938 bond_arp_send_all(bond, new_slave);
2939 new_slave->last_link_up = jiffies;
2940 rcu_assign_pointer(bond->current_arp_slave, new_slave);
2941
2942 check_state:
2943 bond_for_each_slave_rcu(bond, slave, iter) {
2944 if (slave->should_notify || slave->should_notify_link) {
2945 should_notify_rtnl = BOND_SLAVE_NOTIFY_NOW;
2946 break;
2947 }
2948 }
2949 return should_notify_rtnl;
2950 }
2951
2952 static void bond_activebackup_arp_mon(struct bonding *bond)
2953 {
2954 bool should_notify_peers = false;
2955 bool should_notify_rtnl = false;
2956 int delta_in_ticks;
2957
2958 delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
2959
2960 if (!bond_has_slaves(bond))
2961 goto re_arm;
2962
2963 rcu_read_lock();
2964
2965 should_notify_peers = bond_should_notify_peers(bond);
2966
2967 if (bond_ab_arp_inspect(bond)) {
2968 rcu_read_unlock();
2969
2970 /* Race avoidance with bond_close flush of workqueue */
2971 if (!rtnl_trylock()) {
2972 delta_in_ticks = 1;
2973 should_notify_peers = false;
2974 goto re_arm;
2975 }
2976
2977 bond_ab_arp_commit(bond);
2978
2979 rtnl_unlock();
2980 rcu_read_lock();
2981 }
2982
2983 should_notify_rtnl = bond_ab_arp_probe(bond);
2984 rcu_read_unlock();
2985
2986 re_arm:
2987 if (bond->params.arp_interval)
2988 queue_delayed_work(bond->wq, &bond->arp_work, delta_in_ticks);
2989
2990 if (should_notify_peers || should_notify_rtnl) {
2991 if (!rtnl_trylock())
2992 return;
2993
2994 if (should_notify_peers)
2995 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS,
2996 bond->dev);
2997 if (should_notify_rtnl) {
2998 bond_slave_state_notify(bond);
2999 bond_slave_link_notify(bond);
3000 }
3001
3002 rtnl_unlock();
3003 }
3004 }
3005
3006 static void bond_arp_monitor(struct work_struct *work)
3007 {
3008 struct bonding *bond = container_of(work, struct bonding,
3009 arp_work.work);
3010
3011 if (BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP)
3012 bond_activebackup_arp_mon(bond);
3013 else
3014 bond_loadbalance_arp_mon(bond);
3015 }
3016
3017 /*-------------------------- netdev event handling --------------------------*/
3018
3019 /* Change device name */
3020 static int bond_event_changename(struct bonding *bond)
3021 {
3022 bond_remove_proc_entry(bond);
3023 bond_create_proc_entry(bond);
3024
3025 bond_debug_reregister(bond);
3026
3027 return NOTIFY_DONE;
3028 }
3029
3030 static int bond_master_netdev_event(unsigned long event,
3031 struct net_device *bond_dev)
3032 {
3033 struct bonding *event_bond = netdev_priv(bond_dev);
3034
3035 switch (event) {
3036 case NETDEV_CHANGENAME:
3037 return bond_event_changename(event_bond);
3038 case NETDEV_UNREGISTER:
3039 bond_remove_proc_entry(event_bond);
3040 break;
3041 case NETDEV_REGISTER:
3042 bond_create_proc_entry(event_bond);
3043 break;
3044 case NETDEV_NOTIFY_PEERS:
3045 if (event_bond->send_peer_notif)
3046 event_bond->send_peer_notif--;
3047 break;
3048 default:
3049 break;
3050 }
3051
3052 return NOTIFY_DONE;
3053 }
3054
3055 static int bond_slave_netdev_event(unsigned long event,
3056 struct net_device *slave_dev)
3057 {
3058 struct slave *slave = bond_slave_get_rtnl(slave_dev), *primary;
3059 struct bonding *bond;
3060 struct net_device *bond_dev;
3061
3062 /* A netdev event can be generated while enslaving a device
3063 * before netdev_rx_handler_register is called in which case
3064 * slave will be NULL
3065 */
3066 if (!slave)
3067 return NOTIFY_DONE;
3068 bond_dev = slave->bond->dev;
3069 bond = slave->bond;
3070 primary = rtnl_dereference(bond->primary_slave);
3071
3072 switch (event) {
3073 case NETDEV_UNREGISTER:
3074 if (bond_dev->type != ARPHRD_ETHER)
3075 bond_release_and_destroy(bond_dev, slave_dev);
3076 else
3077 __bond_release_one(bond_dev, slave_dev, false, true);
3078 break;
3079 case NETDEV_UP:
3080 case NETDEV_CHANGE:
3081 /* For 802.3ad mode only:
3082 * Getting invalid Speed/Duplex values here will put slave
3083 * in weird state. So mark it as link-down for the time
3084 * being and let link-monitoring (miimon) set it right when
3085 * correct speeds/duplex are available.
3086 */
3087 if (bond_update_speed_duplex(slave) &&
3088 BOND_MODE(bond) == BOND_MODE_8023AD)
3089 slave->link = BOND_LINK_DOWN;
3090
3091 if (BOND_MODE(bond) == BOND_MODE_8023AD)
3092 bond_3ad_adapter_speed_duplex_changed(slave);
3093 /* Fallthrough */
3094 case NETDEV_DOWN:
3095 /* Refresh slave-array if applicable!
3096 * If the setup does not use miimon or arpmon (mode-specific!),
3097 * then these events will not cause the slave-array to be
3098 * refreshed. This will cause xmit to use a slave that is not
3099 * usable. Avoid such situation by refeshing the array at these
3100 * events. If these (miimon/arpmon) parameters are configured
3101 * then array gets refreshed twice and that should be fine!
3102 */
3103 if (bond_mode_uses_xmit_hash(bond))
3104 bond_update_slave_arr(bond, NULL);
3105 break;
3106 case NETDEV_CHANGEMTU:
3107 /* TODO: Should slaves be allowed to
3108 * independently alter their MTU? For
3109 * an active-backup bond, slaves need
3110 * not be the same type of device, so
3111 * MTUs may vary. For other modes,
3112 * slaves arguably should have the
3113 * same MTUs. To do this, we'd need to
3114 * take over the slave's change_mtu
3115 * function for the duration of their
3116 * servitude.
3117 */
3118 break;
3119 case NETDEV_CHANGENAME:
3120 /* we don't care if we don't have primary set */
3121 if (!bond_uses_primary(bond) ||
3122 !bond->params.primary[0])
3123 break;
3124
3125 if (slave == primary) {
3126 /* slave's name changed - he's no longer primary */
3127 RCU_INIT_POINTER(bond->primary_slave, NULL);
3128 } else if (!strcmp(slave_dev->name, bond->params.primary)) {
3129 /* we have a new primary slave */
3130 rcu_assign_pointer(bond->primary_slave, slave);
3131 } else { /* we didn't change primary - exit */
3132 break;
3133 }
3134
3135 netdev_info(bond->dev, "Primary slave changed to %s, reselecting active slave\n",
3136 primary ? slave_dev->name : "none");
3137
3138 block_netpoll_tx();
3139 bond_select_active_slave(bond);
3140 unblock_netpoll_tx();
3141 break;
3142 case NETDEV_FEAT_CHANGE:
3143 bond_compute_features(bond);
3144 break;
3145 case NETDEV_RESEND_IGMP:
3146 /* Propagate to master device */
3147 call_netdevice_notifiers(event, slave->bond->dev);
3148 break;
3149 default:
3150 break;
3151 }
3152
3153 return NOTIFY_DONE;
3154 }
3155
3156 /* bond_netdev_event: handle netdev notifier chain events.
3157 *
3158 * This function receives events for the netdev chain. The caller (an
3159 * ioctl handler calling blocking_notifier_call_chain) holds the necessary
3160 * locks for us to safely manipulate the slave devices (RTNL lock,
3161 * dev_probe_lock).
3162 */
3163 static int bond_netdev_event(struct notifier_block *this,
3164 unsigned long event, void *ptr)
3165 {
3166 struct net_device *event_dev = netdev_notifier_info_to_dev(ptr);
3167
3168 netdev_dbg(event_dev, "event: %lx\n", event);
3169
3170 if (!(event_dev->priv_flags & IFF_BONDING))
3171 return NOTIFY_DONE;
3172
3173 if (event_dev->flags & IFF_MASTER) {
3174 netdev_dbg(event_dev, "IFF_MASTER\n");
3175 return bond_master_netdev_event(event, event_dev);
3176 }
3177
3178 if (event_dev->flags & IFF_SLAVE) {
3179 netdev_dbg(event_dev, "IFF_SLAVE\n");
3180 return bond_slave_netdev_event(event, event_dev);
3181 }
3182
3183 return NOTIFY_DONE;
3184 }
3185
3186 static struct notifier_block bond_netdev_notifier = {
3187 .notifier_call = bond_netdev_event,
3188 };
3189
3190 /*---------------------------- Hashing Policies -----------------------------*/
3191
3192 /* L2 hash helper */
3193 static inline u32 bond_eth_hash(struct sk_buff *skb)
3194 {
3195 struct ethhdr *ep, hdr_tmp;
3196
3197 ep = skb_header_pointer(skb, 0, sizeof(hdr_tmp), &hdr_tmp);
3198 if (ep)
3199 return ep->h_dest[5] ^ ep->h_source[5] ^ ep->h_proto;
3200 return 0;
3201 }
3202
3203 /* Extract the appropriate headers based on bond's xmit policy */
3204 static bool bond_flow_dissect(struct bonding *bond, struct sk_buff *skb,
3205 struct flow_keys *fk)
3206 {
3207 const struct ipv6hdr *iph6;
3208 const struct iphdr *iph;
3209 int noff, proto = -1;
3210
3211 if (bond->params.xmit_policy > BOND_XMIT_POLICY_LAYER23)
3212 return skb_flow_dissect_flow_keys(skb, fk, 0);
3213
3214 fk->ports.ports = 0;
3215 noff = skb_network_offset(skb);
3216 if (skb->protocol == htons(ETH_P_IP)) {
3217 if (unlikely(!pskb_may_pull(skb, noff + sizeof(*iph))))
3218 return false;
3219 iph = ip_hdr(skb);
3220 iph_to_flow_copy_v4addrs(fk, iph);
3221 noff += iph->ihl << 2;
3222 if (!ip_is_fragment(iph))
3223 proto = iph->protocol;
3224 } else if (skb->protocol == htons(ETH_P_IPV6)) {
3225 if (unlikely(!pskb_may_pull(skb, noff + sizeof(*iph6))))
3226 return false;
3227 iph6 = ipv6_hdr(skb);
3228 iph_to_flow_copy_v6addrs(fk, iph6);
3229 noff += sizeof(*iph6);
3230 proto = iph6->nexthdr;
3231 } else {
3232 return false;
3233 }
3234 if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER34 && proto >= 0)
3235 fk->ports.ports = skb_flow_get_ports(skb, noff, proto);
3236
3237 return true;
3238 }
3239
3240 /**
3241 * bond_xmit_hash - generate a hash value based on the xmit policy
3242 * @bond: bonding device
3243 * @skb: buffer to use for headers
3244 *
3245 * This function will extract the necessary headers from the skb buffer and use
3246 * them to generate a hash based on the xmit_policy set in the bonding device
3247 */
3248 u32 bond_xmit_hash(struct bonding *bond, struct sk_buff *skb)
3249 {
3250 struct flow_keys flow;
3251 u32 hash;
3252
3253 if (bond->params.xmit_policy == BOND_XMIT_POLICY_ENCAP34 &&
3254 skb->l4_hash)
3255 return skb->hash;
3256
3257 if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER2 ||
3258 !bond_flow_dissect(bond, skb, &flow))
3259 return bond_eth_hash(skb);
3260
3261 if (bond->params.xmit_policy == BOND_XMIT_POLICY_LAYER23 ||
3262 bond->params.xmit_policy == BOND_XMIT_POLICY_ENCAP23)
3263 hash = bond_eth_hash(skb);
3264 else
3265 hash = (__force u32)flow.ports.ports;
3266 hash ^= (__force u32)flow_get_u32_dst(&flow) ^
3267 (__force u32)flow_get_u32_src(&flow);
3268 hash ^= (hash >> 16);
3269 hash ^= (hash >> 8);
3270
3271 return hash >> 1;
3272 }
3273
3274 /*-------------------------- Device entry points ----------------------------*/
3275
3276 void bond_work_init_all(struct bonding *bond)
3277 {
3278 INIT_DELAYED_WORK(&bond->mcast_work,
3279 bond_resend_igmp_join_requests_delayed);
3280 INIT_DELAYED_WORK(&bond->alb_work, bond_alb_monitor);
3281 INIT_DELAYED_WORK(&bond->mii_work, bond_mii_monitor);
3282 INIT_DELAYED_WORK(&bond->arp_work, bond_arp_monitor);
3283 INIT_DELAYED_WORK(&bond->ad_work, bond_3ad_state_machine_handler);
3284 INIT_DELAYED_WORK(&bond->slave_arr_work, bond_slave_arr_handler);
3285 }
3286
3287 static void bond_work_cancel_all(struct bonding *bond)
3288 {
3289 cancel_delayed_work_sync(&bond->mii_work);
3290 cancel_delayed_work_sync(&bond->arp_work);
3291 cancel_delayed_work_sync(&bond->alb_work);
3292 cancel_delayed_work_sync(&bond->ad_work);
3293 cancel_delayed_work_sync(&bond->mcast_work);
3294 cancel_delayed_work_sync(&bond->slave_arr_work);
3295 }
3296
3297 static int bond_open(struct net_device *bond_dev)
3298 {
3299 struct bonding *bond = netdev_priv(bond_dev);
3300 struct list_head *iter;
3301 struct slave *slave;
3302
3303 /* reset slave->backup and slave->inactive */
3304 if (bond_has_slaves(bond)) {
3305 bond_for_each_slave(bond, slave, iter) {
3306 if (bond_uses_primary(bond) &&
3307 slave != rcu_access_pointer(bond->curr_active_slave)) {
3308 bond_set_slave_inactive_flags(slave,
3309 BOND_SLAVE_NOTIFY_NOW);
3310 } else if (BOND_MODE(bond) != BOND_MODE_8023AD) {
3311 bond_set_slave_active_flags(slave,
3312 BOND_SLAVE_NOTIFY_NOW);
3313 }
3314 }
3315 }
3316
3317 if (bond_is_lb(bond)) {
3318 /* bond_alb_initialize must be called before the timer
3319 * is started.
3320 */
3321 if (bond_alb_initialize(bond, (BOND_MODE(bond) == BOND_MODE_ALB)))
3322 return -ENOMEM;
3323 if (bond->params.tlb_dynamic_lb)
3324 queue_delayed_work(bond->wq, &bond->alb_work, 0);
3325 }
3326
3327 if (bond->params.miimon) /* link check interval, in milliseconds. */
3328 queue_delayed_work(bond->wq, &bond->mii_work, 0);
3329
3330 if (bond->params.arp_interval) { /* arp interval, in milliseconds. */
3331 queue_delayed_work(bond->wq, &bond->arp_work, 0);
3332 bond->recv_probe = bond_arp_rcv;
3333 }
3334
3335 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
3336 queue_delayed_work(bond->wq, &bond->ad_work, 0);
3337 /* register to receive LACPDUs */
3338 bond->recv_probe = bond_3ad_lacpdu_recv;
3339 bond_3ad_initiate_agg_selection(bond, 1);
3340 }
3341
3342 if (bond_mode_uses_xmit_hash(bond))
3343 bond_update_slave_arr(bond, NULL);
3344
3345 return 0;
3346 }
3347
3348 static int bond_close(struct net_device *bond_dev)
3349 {
3350 struct bonding *bond = netdev_priv(bond_dev);
3351
3352 bond_work_cancel_all(bond);
3353 bond->send_peer_notif = 0;
3354 if (bond_is_lb(bond))
3355 bond_alb_deinitialize(bond);
3356 bond->recv_probe = NULL;
3357
3358 return 0;
3359 }
3360
3361 /* fold stats, assuming all rtnl_link_stats64 fields are u64, but
3362 * that some drivers can provide 32bit values only.
3363 */
3364 static void bond_fold_stats(struct rtnl_link_stats64 *_res,
3365 const struct rtnl_link_stats64 *_new,
3366 const struct rtnl_link_stats64 *_old)
3367 {
3368 const u64 *new = (const u64 *)_new;
3369 const u64 *old = (const u64 *)_old;
3370 u64 *res = (u64 *)_res;
3371 int i;
3372
3373 for (i = 0; i < sizeof(*_res) / sizeof(u64); i++) {
3374 u64 nv = new[i];
3375 u64 ov = old[i];
3376 s64 delta = nv - ov;
3377
3378 /* detects if this particular field is 32bit only */
3379 if (((nv | ov) >> 32) == 0)
3380 delta = (s64)(s32)((u32)nv - (u32)ov);
3381
3382 /* filter anomalies, some drivers reset their stats
3383 * at down/up events.
3384 */
3385 if (delta > 0)
3386 res[i] += delta;
3387 }
3388 }
3389
3390 static void bond_get_stats(struct net_device *bond_dev,
3391 struct rtnl_link_stats64 *stats)
3392 {
3393 struct bonding *bond = netdev_priv(bond_dev);
3394 struct rtnl_link_stats64 temp;
3395 struct list_head *iter;
3396 struct slave *slave;
3397
3398 spin_lock(&bond->stats_lock);
3399 memcpy(stats, &bond->bond_stats, sizeof(*stats));
3400
3401 rcu_read_lock();
3402 bond_for_each_slave_rcu(bond, slave, iter) {
3403 const struct rtnl_link_stats64 *new =
3404 dev_get_stats(slave->dev, &temp);
3405
3406 bond_fold_stats(stats, new, &slave->slave_stats);
3407
3408 /* save off the slave stats for the next run */
3409 memcpy(&slave->slave_stats, new, sizeof(*new));
3410 }
3411 rcu_read_unlock();
3412
3413 memcpy(&bond->bond_stats, stats, sizeof(*stats));
3414 spin_unlock(&bond->stats_lock);
3415 }
3416
3417 static int bond_do_ioctl(struct net_device *bond_dev, struct ifreq *ifr, int cmd)
3418 {
3419 struct bonding *bond = netdev_priv(bond_dev);
3420 struct net_device *slave_dev = NULL;
3421 struct ifbond k_binfo;
3422 struct ifbond __user *u_binfo = NULL;
3423 struct ifslave k_sinfo;
3424 struct ifslave __user *u_sinfo = NULL;
3425 struct mii_ioctl_data *mii = NULL;
3426 struct bond_opt_value newval;
3427 struct net *net;
3428 int res = 0;
3429
3430 netdev_dbg(bond_dev, "bond_ioctl: cmd=%d\n", cmd);
3431
3432 switch (cmd) {
3433 case SIOCGMIIPHY:
3434 mii = if_mii(ifr);
3435 if (!mii)
3436 return -EINVAL;
3437
3438 mii->phy_id = 0;
3439 /* Fall Through */
3440 case SIOCGMIIREG:
3441 /* We do this again just in case we were called by SIOCGMIIREG
3442 * instead of SIOCGMIIPHY.
3443 */
3444 mii = if_mii(ifr);
3445 if (!mii)
3446 return -EINVAL;
3447
3448 if (mii->reg_num == 1) {
3449 mii->val_out = 0;
3450 if (netif_carrier_ok(bond->dev))
3451 mii->val_out = BMSR_LSTATUS;
3452 }
3453
3454 return 0;
3455 case BOND_INFO_QUERY_OLD:
3456 case SIOCBONDINFOQUERY:
3457 u_binfo = (struct ifbond __user *)ifr->ifr_data;
3458
3459 if (copy_from_user(&k_binfo, u_binfo, sizeof(ifbond)))
3460 return -EFAULT;
3461
3462 bond_info_query(bond_dev, &k_binfo);
3463 if (copy_to_user(u_binfo, &k_binfo, sizeof(ifbond)))
3464 return -EFAULT;
3465
3466 return 0;
3467 case BOND_SLAVE_INFO_QUERY_OLD:
3468 case SIOCBONDSLAVEINFOQUERY:
3469 u_sinfo = (struct ifslave __user *)ifr->ifr_data;
3470
3471 if (copy_from_user(&k_sinfo, u_sinfo, sizeof(ifslave)))
3472 return -EFAULT;
3473
3474 res = bond_slave_info_query(bond_dev, &k_sinfo);
3475 if (res == 0 &&
3476 copy_to_user(u_sinfo, &k_sinfo, sizeof(ifslave)))
3477 return -EFAULT;
3478
3479 return res;
3480 default:
3481 break;
3482 }
3483
3484 net = dev_net(bond_dev);
3485
3486 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
3487 return -EPERM;
3488
3489 slave_dev = __dev_get_by_name(net, ifr->ifr_slave);
3490
3491 netdev_dbg(bond_dev, "slave_dev=%p:\n", slave_dev);
3492
3493 if (!slave_dev)
3494 return -ENODEV;
3495
3496 netdev_dbg(bond_dev, "slave_dev->name=%s:\n", slave_dev->name);
3497 switch (cmd) {
3498 case BOND_ENSLAVE_OLD:
3499 case SIOCBONDENSLAVE:
3500 res = bond_enslave(bond_dev, slave_dev, NULL);
3501 break;
3502 case BOND_RELEASE_OLD:
3503 case SIOCBONDRELEASE:
3504 res = bond_release(bond_dev, slave_dev);
3505 break;
3506 case BOND_SETHWADDR_OLD:
3507 case SIOCBONDSETHWADDR:
3508 bond_set_dev_addr(bond_dev, slave_dev);
3509 res = 0;
3510 break;
3511 case BOND_CHANGE_ACTIVE_OLD:
3512 case SIOCBONDCHANGEACTIVE:
3513 bond_opt_initstr(&newval, slave_dev->name);
3514 res = __bond_opt_set_notify(bond, BOND_OPT_ACTIVE_SLAVE,
3515 &newval);
3516 break;
3517 default:
3518 res = -EOPNOTSUPP;
3519 }
3520
3521 return res;
3522 }
3523
3524 static void bond_change_rx_flags(struct net_device *bond_dev, int change)
3525 {
3526 struct bonding *bond = netdev_priv(bond_dev);
3527
3528 if (change & IFF_PROMISC)
3529 bond_set_promiscuity(bond,
3530 bond_dev->flags & IFF_PROMISC ? 1 : -1);
3531
3532 if (change & IFF_ALLMULTI)
3533 bond_set_allmulti(bond,
3534 bond_dev->flags & IFF_ALLMULTI ? 1 : -1);
3535 }
3536
3537 static void bond_set_rx_mode(struct net_device *bond_dev)
3538 {
3539 struct bonding *bond = netdev_priv(bond_dev);
3540 struct list_head *iter;
3541 struct slave *slave;
3542
3543 rcu_read_lock();
3544 if (bond_uses_primary(bond)) {
3545 slave = rcu_dereference(bond->curr_active_slave);
3546 if (slave) {
3547 dev_uc_sync(slave->dev, bond_dev);
3548 dev_mc_sync(slave->dev, bond_dev);
3549 }
3550 } else {
3551 bond_for_each_slave_rcu(bond, slave, iter) {
3552 dev_uc_sync_multiple(slave->dev, bond_dev);
3553 dev_mc_sync_multiple(slave->dev, bond_dev);
3554 }
3555 }
3556 rcu_read_unlock();
3557 }
3558
3559 static int bond_neigh_init(struct neighbour *n)
3560 {
3561 struct bonding *bond = netdev_priv(n->dev);
3562 const struct net_device_ops *slave_ops;
3563 struct neigh_parms parms;
3564 struct slave *slave;
3565 int ret;
3566
3567 slave = bond_first_slave(bond);
3568 if (!slave)
3569 return 0;
3570 slave_ops = slave->dev->netdev_ops;
3571 if (!slave_ops->ndo_neigh_setup)
3572 return 0;
3573
3574 parms.neigh_setup = NULL;
3575 parms.neigh_cleanup = NULL;
3576 ret = slave_ops->ndo_neigh_setup(slave->dev, &parms);
3577 if (ret)
3578 return ret;
3579
3580 /* Assign slave's neigh_cleanup to neighbour in case cleanup is called
3581 * after the last slave has been detached. Assumes that all slaves
3582 * utilize the same neigh_cleanup (true at this writing as only user
3583 * is ipoib).
3584 */
3585 n->parms->neigh_cleanup = parms.neigh_cleanup;
3586
3587 if (!parms.neigh_setup)
3588 return 0;
3589
3590 return parms.neigh_setup(n);
3591 }
3592
3593 /* The bonding ndo_neigh_setup is called at init time beofre any
3594 * slave exists. So we must declare proxy setup function which will
3595 * be used at run time to resolve the actual slave neigh param setup.
3596 *
3597 * It's also called by master devices (such as vlans) to setup their
3598 * underlying devices. In that case - do nothing, we're already set up from
3599 * our init.
3600 */
3601 static int bond_neigh_setup(struct net_device *dev,
3602 struct neigh_parms *parms)
3603 {
3604 /* modify only our neigh_parms */
3605 if (parms->dev == dev)
3606 parms->neigh_setup = bond_neigh_init;
3607
3608 return 0;
3609 }
3610
3611 /* Change the MTU of all of a master's slaves to match the master */
3612 static int bond_change_mtu(struct net_device *bond_dev, int new_mtu)
3613 {
3614 struct bonding *bond = netdev_priv(bond_dev);
3615 struct slave *slave, *rollback_slave;
3616 struct list_head *iter;
3617 int res = 0;
3618
3619 netdev_dbg(bond_dev, "bond=%p, new_mtu=%d\n", bond, new_mtu);
3620
3621 bond_for_each_slave(bond, slave, iter) {
3622 netdev_dbg(bond_dev, "s %p c_m %p\n",
3623 slave, slave->dev->netdev_ops->ndo_change_mtu);
3624
3625 res = dev_set_mtu(slave->dev, new_mtu);
3626
3627 if (res) {
3628 /* If we failed to set the slave's mtu to the new value
3629 * we must abort the operation even in ACTIVE_BACKUP
3630 * mode, because if we allow the backup slaves to have
3631 * different mtu values than the active slave we'll
3632 * need to change their mtu when doing a failover. That
3633 * means changing their mtu from timer context, which
3634 * is probably not a good idea.
3635 */
3636 netdev_dbg(bond_dev, "err %d %s\n", res,
3637 slave->dev->name);
3638 goto unwind;
3639 }
3640 }
3641
3642 bond_dev->mtu = new_mtu;
3643
3644 return 0;
3645
3646 unwind:
3647 /* unwind from head to the slave that failed */
3648 bond_for_each_slave(bond, rollback_slave, iter) {
3649 int tmp_res;
3650
3651 if (rollback_slave == slave)
3652 break;
3653
3654 tmp_res = dev_set_mtu(rollback_slave->dev, bond_dev->mtu);
3655 if (tmp_res) {
3656 netdev_dbg(bond_dev, "unwind err %d dev %s\n",
3657 tmp_res, rollback_slave->dev->name);
3658 }
3659 }
3660
3661 return res;
3662 }
3663
3664 /* Change HW address
3665 *
3666 * Note that many devices must be down to change the HW address, and
3667 * downing the master releases all slaves. We can make bonds full of
3668 * bonding devices to test this, however.
3669 */
3670 static int bond_set_mac_address(struct net_device *bond_dev, void *addr)
3671 {
3672 struct bonding *bond = netdev_priv(bond_dev);
3673 struct slave *slave, *rollback_slave;
3674 struct sockaddr_storage *ss = addr, tmp_ss;
3675 struct list_head *iter;
3676 int res = 0;
3677
3678 if (BOND_MODE(bond) == BOND_MODE_ALB)
3679 return bond_alb_set_mac_address(bond_dev, addr);
3680
3681
3682 netdev_dbg(bond_dev, "bond=%p\n", bond);
3683
3684 /* If fail_over_mac is enabled, do nothing and return success.
3685 * Returning an error causes ifenslave to fail.
3686 */
3687 if (bond->params.fail_over_mac &&
3688 BOND_MODE(bond) == BOND_MODE_ACTIVEBACKUP)
3689 return 0;
3690
3691 if (!is_valid_ether_addr(ss->__data))
3692 return -EADDRNOTAVAIL;
3693
3694 bond_for_each_slave(bond, slave, iter) {
3695 netdev_dbg(bond_dev, "slave %p %s\n", slave, slave->dev->name);
3696 res = dev_set_mac_address(slave->dev, addr);
3697 if (res) {
3698 /* TODO: consider downing the slave
3699 * and retry ?
3700 * User should expect communications
3701 * breakage anyway until ARP finish
3702 * updating, so...
3703 */
3704 netdev_dbg(bond_dev, "err %d %s\n", res, slave->dev->name);
3705 goto unwind;
3706 }
3707 }
3708
3709 /* success */
3710 memcpy(bond_dev->dev_addr, ss->__data, bond_dev->addr_len);
3711 return 0;
3712
3713 unwind:
3714 memcpy(tmp_ss.__data, bond_dev->dev_addr, bond_dev->addr_len);
3715 tmp_ss.ss_family = bond_dev->type;
3716
3717 /* unwind from head to the slave that failed */
3718 bond_for_each_slave(bond, rollback_slave, iter) {
3719 int tmp_res;
3720
3721 if (rollback_slave == slave)
3722 break;
3723
3724 tmp_res = dev_set_mac_address(rollback_slave->dev,
3725 (struct sockaddr *)&tmp_ss);
3726 if (tmp_res) {
3727 netdev_dbg(bond_dev, "unwind err %d dev %s\n",
3728 tmp_res, rollback_slave->dev->name);
3729 }
3730 }
3731
3732 return res;
3733 }
3734
3735 /**
3736 * bond_xmit_slave_id - transmit skb through slave with slave_id
3737 * @bond: bonding device that is transmitting
3738 * @skb: buffer to transmit
3739 * @slave_id: slave id up to slave_cnt-1 through which to transmit
3740 *
3741 * This function tries to transmit through slave with slave_id but in case
3742 * it fails, it tries to find the first available slave for transmission.
3743 * The skb is consumed in all cases, thus the function is void.
3744 */
3745 static void bond_xmit_slave_id(struct bonding *bond, struct sk_buff *skb, int slave_id)
3746 {
3747 struct list_head *iter;
3748 struct slave *slave;
3749 int i = slave_id;
3750
3751 /* Here we start from the slave with slave_id */
3752 bond_for_each_slave_rcu(bond, slave, iter) {
3753 if (--i < 0) {
3754 if (bond_slave_can_tx(slave)) {
3755 bond_dev_queue_xmit(bond, skb, slave->dev);
3756 return;
3757 }
3758 }
3759 }
3760
3761 /* Here we start from the first slave up to slave_id */
3762 i = slave_id;
3763 bond_for_each_slave_rcu(bond, slave, iter) {
3764 if (--i < 0)
3765 break;
3766 if (bond_slave_can_tx(slave)) {
3767 bond_dev_queue_xmit(bond, skb, slave->dev);
3768 return;
3769 }
3770 }
3771 /* no slave that can tx has been found */
3772 bond_tx_drop(bond->dev, skb);
3773 }
3774
3775 /**
3776 * bond_rr_gen_slave_id - generate slave id based on packets_per_slave
3777 * @bond: bonding device to use
3778 *
3779 * Based on the value of the bonding device's packets_per_slave parameter
3780 * this function generates a slave id, which is usually used as the next
3781 * slave to transmit through.
3782 */
3783 static u32 bond_rr_gen_slave_id(struct bonding *bond)
3784 {
3785 u32 slave_id;
3786 struct reciprocal_value reciprocal_packets_per_slave;
3787 int packets_per_slave = bond->params.packets_per_slave;
3788
3789 switch (packets_per_slave) {
3790 case 0:
3791 slave_id = prandom_u32();
3792 break;
3793 case 1:
3794 slave_id = bond->rr_tx_counter;
3795 break;
3796 default:
3797 reciprocal_packets_per_slave =
3798 bond->params.reciprocal_packets_per_slave;
3799 slave_id = reciprocal_divide(bond->rr_tx_counter,
3800 reciprocal_packets_per_slave);
3801 break;
3802 }
3803 bond->rr_tx_counter++;
3804
3805 return slave_id;
3806 }
3807
3808 static int bond_xmit_roundrobin(struct sk_buff *skb, struct net_device *bond_dev)
3809 {
3810 struct bonding *bond = netdev_priv(bond_dev);
3811 struct iphdr *iph = ip_hdr(skb);
3812 struct slave *slave;
3813 u32 slave_id;
3814
3815 /* Start with the curr_active_slave that joined the bond as the
3816 * default for sending IGMP traffic. For failover purposes one
3817 * needs to maintain some consistency for the interface that will
3818 * send the join/membership reports. The curr_active_slave found
3819 * will send all of this type of traffic.
3820 */
3821 if (iph->protocol == IPPROTO_IGMP && skb->protocol == htons(ETH_P_IP)) {
3822 slave = rcu_dereference(bond->curr_active_slave);
3823 if (slave)
3824 bond_dev_queue_xmit(bond, skb, slave->dev);
3825 else
3826 bond_xmit_slave_id(bond, skb, 0);
3827 } else {
3828 int slave_cnt = READ_ONCE(bond->slave_cnt);
3829
3830 if (likely(slave_cnt)) {
3831 slave_id = bond_rr_gen_slave_id(bond);
3832 bond_xmit_slave_id(bond, skb, slave_id % slave_cnt);
3833 } else {
3834 bond_tx_drop(bond_dev, skb);
3835 }
3836 }
3837
3838 return NETDEV_TX_OK;
3839 }
3840
3841 /* In active-backup mode, we know that bond->curr_active_slave is always valid if
3842 * the bond has a usable interface.
3843 */
3844 static int bond_xmit_activebackup(struct sk_buff *skb, struct net_device *bond_dev)
3845 {
3846 struct bonding *bond = netdev_priv(bond_dev);
3847 struct slave *slave;
3848
3849 slave = rcu_dereference(bond->curr_active_slave);
3850 if (slave)
3851 bond_dev_queue_xmit(bond, skb, slave->dev);
3852 else
3853 bond_tx_drop(bond_dev, skb);
3854
3855 return NETDEV_TX_OK;
3856 }
3857
3858 /* Use this to update slave_array when (a) it's not appropriate to update
3859 * slave_array right away (note that update_slave_array() may sleep)
3860 * and / or (b) RTNL is not held.
3861 */
3862 void bond_slave_arr_work_rearm(struct bonding *bond, unsigned long delay)
3863 {
3864 queue_delayed_work(bond->wq, &bond->slave_arr_work, delay);
3865 }
3866
3867 /* Slave array work handler. Holds only RTNL */
3868 static void bond_slave_arr_handler(struct work_struct *work)
3869 {
3870 struct bonding *bond = container_of(work, struct bonding,
3871 slave_arr_work.work);
3872 int ret;
3873
3874 if (!rtnl_trylock())
3875 goto err;
3876
3877 ret = bond_update_slave_arr(bond, NULL);
3878 rtnl_unlock();
3879 if (ret) {
3880 pr_warn_ratelimited("Failed to update slave array from WT\n");
3881 goto err;
3882 }
3883 return;
3884
3885 err:
3886 bond_slave_arr_work_rearm(bond, 1);
3887 }
3888
3889 /* Build the usable slaves array in control path for modes that use xmit-hash
3890 * to determine the slave interface -
3891 * (a) BOND_MODE_8023AD
3892 * (b) BOND_MODE_XOR
3893 * (c) BOND_MODE_TLB && tlb_dynamic_lb == 0
3894 *
3895 * The caller is expected to hold RTNL only and NO other lock!
3896 */
3897 int bond_update_slave_arr(struct bonding *bond, struct slave *skipslave)
3898 {
3899 struct slave *slave;
3900 struct list_head *iter;
3901 struct bond_up_slave *new_arr, *old_arr;
3902 int agg_id = 0;
3903 int ret = 0;
3904
3905 #ifdef CONFIG_LOCKDEP
3906 WARN_ON(lockdep_is_held(&bond->mode_lock));
3907 #endif
3908
3909 new_arr = kzalloc(offsetof(struct bond_up_slave, arr[bond->slave_cnt]),
3910 GFP_KERNEL);
3911 if (!new_arr) {
3912 ret = -ENOMEM;
3913 pr_err("Failed to build slave-array.\n");
3914 goto out;
3915 }
3916 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
3917 struct ad_info ad_info;
3918
3919 if (bond_3ad_get_active_agg_info(bond, &ad_info)) {
3920 pr_debug("bond_3ad_get_active_agg_info failed\n");
3921 kfree_rcu(new_arr, rcu);
3922 /* No active aggragator means it's not safe to use
3923 * the previous array.
3924 */
3925 old_arr = rtnl_dereference(bond->slave_arr);
3926 if (old_arr) {
3927 RCU_INIT_POINTER(bond->slave_arr, NULL);
3928 kfree_rcu(old_arr, rcu);
3929 }
3930 goto out;
3931 }
3932 agg_id = ad_info.aggregator_id;
3933 }
3934 bond_for_each_slave(bond, slave, iter) {
3935 if (BOND_MODE(bond) == BOND_MODE_8023AD) {
3936 struct aggregator *agg;
3937
3938 agg = SLAVE_AD_INFO(slave)->port.aggregator;
3939 if (!agg || agg->aggregator_identifier != agg_id)
3940 continue;
3941 }
3942 if (!bond_slave_can_tx(slave))
3943 continue;
3944 if (skipslave == slave)
3945 continue;
3946 new_arr->arr[new_arr->count++] = slave;
3947 }
3948
3949 old_arr = rtnl_dereference(bond->slave_arr);
3950 rcu_assign_pointer(bond->slave_arr, new_arr);
3951 if (old_arr)
3952 kfree_rcu(old_arr, rcu);
3953 out:
3954 if (ret != 0 && skipslave) {
3955 int idx;
3956
3957 /* Rare situation where caller has asked to skip a specific
3958 * slave but allocation failed (most likely!). BTW this is
3959 * only possible when the call is initiated from
3960 * __bond_release_one(). In this situation; overwrite the
3961 * skipslave entry in the array with the last entry from the
3962 * array to avoid a situation where the xmit path may choose
3963 * this to-be-skipped slave to send a packet out.
3964 */
3965 old_arr = rtnl_dereference(bond->slave_arr);
3966 for (idx = 0; idx < old_arr->count; idx++) {
3967 if (skipslave == old_arr->arr[idx]) {
3968 old_arr->arr[idx] =
3969 old_arr->arr[old_arr->count-1];
3970 old_arr->count--;
3971 break;
3972 }
3973 }
3974 }
3975 return ret;
3976 }
3977
3978 /* Use this Xmit function for 3AD as well as XOR modes. The current
3979 * usable slave array is formed in the control path. The xmit function
3980 * just calculates hash and sends the packet out.
3981 */
3982 static int bond_3ad_xor_xmit(struct sk_buff *skb, struct net_device *dev)
3983 {
3984 struct bonding *bond = netdev_priv(dev);
3985 struct slave *slave;
3986 struct bond_up_slave *slaves;
3987 unsigned int count;
3988
3989 slaves = rcu_dereference(bond->slave_arr);
3990 count = slaves ? READ_ONCE(slaves->count) : 0;
3991 if (likely(count)) {
3992 slave = slaves->arr[bond_xmit_hash(bond, skb) % count];
3993 bond_dev_queue_xmit(bond, skb, slave->dev);
3994 } else {
3995 bond_tx_drop(dev, skb);
3996 }
3997
3998 return NETDEV_TX_OK;
3999 }
4000
4001 /* in broadcast mode, we send everything to all usable interfaces. */
4002 static int bond_xmit_broadcast(struct sk_buff *skb, struct net_device *bond_dev)
4003 {
4004 struct bonding *bond = netdev_priv(bond_dev);
4005 struct slave *slave = NULL;
4006 struct list_head *iter;
4007
4008 bond_for_each_slave_rcu(bond, slave, iter) {
4009 if (bond_is_last_slave(bond, slave))
4010 break;
4011 if (bond_slave_is_up(slave) && slave->link == BOND_LINK_UP) {
4012 struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
4013
4014 if (!skb2) {
4015 net_err_ratelimited("%s: Error: %s: skb_clone() failed\n",
4016 bond_dev->name, __func__);
4017 continue;
4018 }
4019 bond_dev_queue_xmit(bond, skb2, slave->dev);
4020 }
4021 }
4022 if (slave && bond_slave_is_up(slave) && slave->link == BOND_LINK_UP)
4023 bond_dev_queue_xmit(bond, skb, slave->dev);
4024 else
4025 bond_tx_drop(bond_dev, skb);
4026
4027 return NETDEV_TX_OK;
4028 }
4029
4030 /*------------------------- Device initialization ---------------------------*/
4031
4032 /* Lookup the slave that corresponds to a qid */
4033 static inline int bond_slave_override(struct bonding *bond,
4034 struct sk_buff *skb)
4035 {
4036 struct slave *slave = NULL;
4037 struct list_head *iter;
4038
4039 if (!skb->queue_mapping)
4040 return 1;
4041
4042 /* Find out if any slaves have the same mapping as this skb. */
4043 bond_for_each_slave_rcu(bond, slave, iter) {
4044 if (slave->queue_id == skb->queue_mapping) {
4045 if (bond_slave_is_up(slave) &&
4046 slave->link == BOND_LINK_UP) {
4047 bond_dev_queue_xmit(bond, skb, slave->dev);
4048 return 0;
4049 }
4050 /* If the slave isn't UP, use default transmit policy. */
4051 break;
4052 }
4053 }
4054
4055 return 1;
4056 }
4057
4058
4059 static u16 bond_select_queue(struct net_device *dev, struct sk_buff *skb,
4060 void *accel_priv, select_queue_fallback_t fallback)
4061 {
4062 /* This helper function exists to help dev_pick_tx get the correct
4063 * destination queue. Using a helper function skips a call to
4064 * skb_tx_hash and will put the skbs in the queue we expect on their
4065 * way down to the bonding driver.
4066 */
4067 u16 txq = skb_rx_queue_recorded(skb) ? skb_get_rx_queue(skb) : 0;
4068
4069 /* Save the original txq to restore before passing to the driver */
4070 qdisc_skb_cb(skb)->slave_dev_queue_mapping = skb->queue_mapping;
4071
4072 if (unlikely(txq >= dev->real_num_tx_queues)) {
4073 do {
4074 txq -= dev->real_num_tx_queues;
4075 } while (txq >= dev->real_num_tx_queues);
4076 }
4077 return txq;
4078 }
4079
4080 static netdev_tx_t __bond_start_xmit(struct sk_buff *skb, struct net_device *dev)
4081 {
4082 struct bonding *bond = netdev_priv(dev);
4083
4084 if (bond_should_override_tx_queue(bond) &&
4085 !bond_slave_override(bond, skb))
4086 return NETDEV_TX_OK;
4087
4088 switch (BOND_MODE(bond)) {
4089 case BOND_MODE_ROUNDROBIN:
4090 return bond_xmit_roundrobin(skb, dev);
4091 case BOND_MODE_ACTIVEBACKUP:
4092 return bond_xmit_activebackup(skb, dev);
4093 case BOND_MODE_8023AD:
4094 case BOND_MODE_XOR:
4095 return bond_3ad_xor_xmit(skb, dev);
4096 case BOND_MODE_BROADCAST:
4097 return bond_xmit_broadcast(skb, dev);
4098 case BOND_MODE_ALB:
4099 return bond_alb_xmit(skb, dev);
4100 case BOND_MODE_TLB:
4101 return bond_tlb_xmit(skb, dev);
4102 default:
4103 /* Should never happen, mode already checked */
4104 netdev_err(dev, "Unknown bonding mode %d\n", BOND_MODE(bond));
4105 WARN_ON_ONCE(1);
4106 bond_tx_drop(dev, skb);
4107 return NETDEV_TX_OK;
4108 }
4109 }
4110
4111 static netdev_tx_t bond_start_xmit(struct sk_buff *skb, struct net_device *dev)
4112 {
4113 struct bonding *bond = netdev_priv(dev);
4114 netdev_tx_t ret = NETDEV_TX_OK;
4115
4116 /* If we risk deadlock from transmitting this in the
4117 * netpoll path, tell netpoll to queue the frame for later tx
4118 */
4119 if (unlikely(is_netpoll_tx_blocked(dev)))
4120 return NETDEV_TX_BUSY;
4121
4122 rcu_read_lock();
4123 if (bond_has_slaves(bond))
4124 ret = __bond_start_xmit(skb, dev);
4125 else
4126 bond_tx_drop(dev, skb);
4127 rcu_read_unlock();
4128
4129 return ret;
4130 }
4131
4132 static int bond_ethtool_get_link_ksettings(struct net_device *bond_dev,
4133 struct ethtool_link_ksettings *cmd)
4134 {
4135 struct bonding *bond = netdev_priv(bond_dev);
4136 unsigned long speed = 0;
4137 struct list_head *iter;
4138 struct slave *slave;
4139
4140 cmd->base.duplex = DUPLEX_UNKNOWN;
4141 cmd->base.port = PORT_OTHER;
4142
4143 /* Since bond_slave_can_tx returns false for all inactive or down slaves, we
4144 * do not need to check mode. Though link speed might not represent
4145 * the true receive or transmit bandwidth (not all modes are symmetric)
4146 * this is an accurate maximum.
4147 */
4148 bond_for_each_slave(bond, slave, iter) {
4149 if (bond_slave_can_tx(slave)) {
4150 if (slave->speed != SPEED_UNKNOWN)
4151 speed += slave->speed;
4152 if (cmd->base.duplex == DUPLEX_UNKNOWN &&
4153 slave->duplex != DUPLEX_UNKNOWN)
4154 cmd->base.duplex = slave->duplex;
4155 }
4156 }
4157 cmd->base.speed = speed ? : SPEED_UNKNOWN;
4158
4159 return 0;
4160 }
4161
4162 static void bond_ethtool_get_drvinfo(struct net_device *bond_dev,
4163 struct ethtool_drvinfo *drvinfo)
4164 {
4165 strlcpy(drvinfo->driver, DRV_NAME, sizeof(drvinfo->driver));
4166 strlcpy(drvinfo->version, DRV_VERSION, sizeof(drvinfo->version));
4167 snprintf(drvinfo->fw_version, sizeof(drvinfo->fw_version), "%d",
4168 BOND_ABI_VERSION);
4169 }
4170
4171 static const struct ethtool_ops bond_ethtool_ops = {
4172 .get_drvinfo = bond_ethtool_get_drvinfo,
4173 .get_link = ethtool_op_get_link,
4174 .get_link_ksettings = bond_ethtool_get_link_ksettings,
4175 };
4176
4177 static const struct net_device_ops bond_netdev_ops = {
4178 .ndo_init = bond_init,
4179 .ndo_uninit = bond_uninit,
4180 .ndo_open = bond_open,
4181 .ndo_stop = bond_close,
4182 .ndo_start_xmit = bond_start_xmit,
4183 .ndo_select_queue = bond_select_queue,
4184 .ndo_get_stats64 = bond_get_stats,
4185 .ndo_do_ioctl = bond_do_ioctl,
4186 .ndo_change_rx_flags = bond_change_rx_flags,
4187 .ndo_set_rx_mode = bond_set_rx_mode,
4188 .ndo_change_mtu = bond_change_mtu,
4189 .ndo_set_mac_address = bond_set_mac_address,
4190 .ndo_neigh_setup = bond_neigh_setup,
4191 .ndo_vlan_rx_add_vid = bond_vlan_rx_add_vid,
4192 .ndo_vlan_rx_kill_vid = bond_vlan_rx_kill_vid,
4193 #ifdef CONFIG_NET_POLL_CONTROLLER
4194 .ndo_netpoll_setup = bond_netpoll_setup,
4195 .ndo_netpoll_cleanup = bond_netpoll_cleanup,
4196 .ndo_poll_controller = bond_poll_controller,
4197 #endif
4198 .ndo_add_slave = bond_enslave,
4199 .ndo_del_slave = bond_release,
4200 .ndo_fix_features = bond_fix_features,
4201 .ndo_features_check = passthru_features_check,
4202 };
4203
4204 static const struct device_type bond_type = {
4205 .name = "bond",
4206 };
4207
4208 static void bond_destructor(struct net_device *bond_dev)
4209 {
4210 struct bonding *bond = netdev_priv(bond_dev);
4211 if (bond->wq)
4212 destroy_workqueue(bond->wq);
4213 }
4214
4215 void bond_setup(struct net_device *bond_dev)
4216 {
4217 struct bonding *bond = netdev_priv(bond_dev);
4218
4219 spin_lock_init(&bond->mode_lock);
4220 spin_lock_init(&bond->stats_lock);
4221 bond->params = bonding_defaults;
4222
4223 /* Initialize pointers */
4224 bond->dev = bond_dev;
4225
4226 /* Initialize the device entry points */
4227 ether_setup(bond_dev);
4228 bond_dev->max_mtu = ETH_MAX_MTU;
4229 bond_dev->netdev_ops = &bond_netdev_ops;
4230 bond_dev->ethtool_ops = &bond_ethtool_ops;
4231
4232 bond_dev->needs_free_netdev = true;
4233 bond_dev->priv_destructor = bond_destructor;
4234
4235 SET_NETDEV_DEVTYPE(bond_dev, &bond_type);
4236
4237 /* Initialize the device options */
4238 bond_dev->flags |= IFF_MASTER;
4239 bond_dev->priv_flags |= IFF_BONDING | IFF_UNICAST_FLT | IFF_NO_QUEUE;
4240 bond_dev->priv_flags &= ~(IFF_XMIT_DST_RELEASE | IFF_TX_SKB_SHARING);
4241
4242 /* don't acquire bond device's netif_tx_lock when transmitting */
4243 bond_dev->features |= NETIF_F_LLTX;
4244
4245 /* By default, we declare the bond to be fully
4246 * VLAN hardware accelerated capable. Special
4247 * care is taken in the various xmit functions
4248 * when there are slaves that are not hw accel
4249 * capable
4250 */
4251
4252 /* Don't allow bond devices to change network namespaces. */
4253 bond_dev->features |= NETIF_F_NETNS_LOCAL;
4254
4255 bond_dev->hw_features = BOND_VLAN_FEATURES |
4256 NETIF_F_HW_VLAN_CTAG_TX |
4257 NETIF_F_HW_VLAN_CTAG_RX |
4258 NETIF_F_HW_VLAN_CTAG_FILTER;
4259
4260 bond_dev->hw_features |= NETIF_F_GSO_ENCAP_ALL;
4261 bond_dev->features |= bond_dev->hw_features;
4262 }
4263
4264 /* Destroy a bonding device.
4265 * Must be under rtnl_lock when this function is called.
4266 */
4267 static void bond_uninit(struct net_device *bond_dev)
4268 {
4269 struct bonding *bond = netdev_priv(bond_dev);
4270 struct list_head *iter;
4271 struct slave *slave;
4272 struct bond_up_slave *arr;
4273
4274 bond_netpoll_cleanup(bond_dev);
4275
4276 /* Release the bonded slaves */
4277 bond_for_each_slave(bond, slave, iter)
4278 __bond_release_one(bond_dev, slave->dev, true, true);
4279 netdev_info(bond_dev, "Released all slaves\n");
4280
4281 arr = rtnl_dereference(bond->slave_arr);
4282 if (arr) {
4283 RCU_INIT_POINTER(bond->slave_arr, NULL);
4284 kfree_rcu(arr, rcu);
4285 }
4286
4287 list_del(&bond->bond_list);
4288
4289 bond_debug_unregister(bond);
4290 }
4291
4292 /*------------------------- Module initialization ---------------------------*/
4293
4294 static int bond_check_params(struct bond_params *params)
4295 {
4296 int arp_validate_value, fail_over_mac_value, primary_reselect_value, i;
4297 struct bond_opt_value newval;
4298 const struct bond_opt_value *valptr;
4299 int arp_all_targets_value = 0;
4300 u16 ad_actor_sys_prio = 0;
4301 u16 ad_user_port_key = 0;
4302 __be32 arp_target[BOND_MAX_ARP_TARGETS] = { 0 };
4303 int arp_ip_count;
4304 int bond_mode = BOND_MODE_ROUNDROBIN;
4305 int xmit_hashtype = BOND_XMIT_POLICY_LAYER2;
4306 int lacp_fast = 0;
4307 int tlb_dynamic_lb;
4308
4309 /* Convert string parameters. */
4310 if (mode) {
4311 bond_opt_initstr(&newval, mode);
4312 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_MODE), &newval);
4313 if (!valptr) {
4314 pr_err("Error: Invalid bonding mode \"%s\"\n", mode);
4315 return -EINVAL;
4316 }
4317 bond_mode = valptr->value;
4318 }
4319
4320 if (xmit_hash_policy) {
4321 if ((bond_mode != BOND_MODE_XOR) &&
4322 (bond_mode != BOND_MODE_8023AD) &&
4323 (bond_mode != BOND_MODE_TLB)) {
4324 pr_info("xmit_hash_policy param is irrelevant in mode %s\n",
4325 bond_mode_name(bond_mode));
4326 } else {
4327 bond_opt_initstr(&newval, xmit_hash_policy);
4328 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_XMIT_HASH),
4329 &newval);
4330 if (!valptr) {
4331 pr_err("Error: Invalid xmit_hash_policy \"%s\"\n",
4332 xmit_hash_policy);
4333 return -EINVAL;
4334 }
4335 xmit_hashtype = valptr->value;
4336 }
4337 }
4338
4339 if (lacp_rate) {
4340 if (bond_mode != BOND_MODE_8023AD) {
4341 pr_info("lacp_rate param is irrelevant in mode %s\n",
4342 bond_mode_name(bond_mode));
4343 } else {
4344 bond_opt_initstr(&newval, lacp_rate);
4345 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_LACP_RATE),
4346 &newval);
4347 if (!valptr) {
4348 pr_err("Error: Invalid lacp rate \"%s\"\n",
4349 lacp_rate);
4350 return -EINVAL;
4351 }
4352 lacp_fast = valptr->value;
4353 }
4354 }
4355
4356 if (ad_select) {
4357 bond_opt_initstr(&newval, ad_select);
4358 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_AD_SELECT),
4359 &newval);
4360 if (!valptr) {
4361 pr_err("Error: Invalid ad_select \"%s\"\n", ad_select);
4362 return -EINVAL;
4363 }
4364 params->ad_select = valptr->value;
4365 if (bond_mode != BOND_MODE_8023AD)
4366 pr_warn("ad_select param only affects 802.3ad mode\n");
4367 } else {
4368 params->ad_select = BOND_AD_STABLE;
4369 }
4370
4371 if (max_bonds < 0) {
4372 pr_warn("Warning: max_bonds (%d) not in range %d-%d, so it was reset to BOND_DEFAULT_MAX_BONDS (%d)\n",
4373 max_bonds, 0, INT_MAX, BOND_DEFAULT_MAX_BONDS);
4374 max_bonds = BOND_DEFAULT_MAX_BONDS;
4375 }
4376
4377 if (miimon < 0) {
4378 pr_warn("Warning: miimon module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4379 miimon, INT_MAX);
4380 miimon = 0;
4381 }
4382
4383 if (updelay < 0) {
4384 pr_warn("Warning: updelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4385 updelay, INT_MAX);
4386 updelay = 0;
4387 }
4388
4389 if (downdelay < 0) {
4390 pr_warn("Warning: downdelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4391 downdelay, INT_MAX);
4392 downdelay = 0;
4393 }
4394
4395 if ((use_carrier != 0) && (use_carrier != 1)) {
4396 pr_warn("Warning: use_carrier module parameter (%d), not of valid value (0/1), so it was set to 1\n",
4397 use_carrier);
4398 use_carrier = 1;
4399 }
4400
4401 if (num_peer_notif < 0 || num_peer_notif > 255) {
4402 pr_warn("Warning: num_grat_arp/num_unsol_na (%d) not in range 0-255 so it was reset to 1\n",
4403 num_peer_notif);
4404 num_peer_notif = 1;
4405 }
4406
4407 /* reset values for 802.3ad/TLB/ALB */
4408 if (!bond_mode_uses_arp(bond_mode)) {
4409 if (!miimon) {
4410 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");
4411 pr_warn("Forcing miimon to 100msec\n");
4412 miimon = BOND_DEFAULT_MIIMON;
4413 }
4414 }
4415
4416 if (tx_queues < 1 || tx_queues > 255) {
4417 pr_warn("Warning: tx_queues (%d) should be between 1 and 255, resetting to %d\n",
4418 tx_queues, BOND_DEFAULT_TX_QUEUES);
4419 tx_queues = BOND_DEFAULT_TX_QUEUES;
4420 }
4421
4422 if ((all_slaves_active != 0) && (all_slaves_active != 1)) {
4423 pr_warn("Warning: all_slaves_active module parameter (%d), not of valid value (0/1), so it was set to 0\n",
4424 all_slaves_active);
4425 all_slaves_active = 0;
4426 }
4427
4428 if (resend_igmp < 0 || resend_igmp > 255) {
4429 pr_warn("Warning: resend_igmp (%d) should be between 0 and 255, resetting to %d\n",
4430 resend_igmp, BOND_DEFAULT_RESEND_IGMP);
4431 resend_igmp = BOND_DEFAULT_RESEND_IGMP;
4432 }
4433
4434 bond_opt_initval(&newval, packets_per_slave);
4435 if (!bond_opt_parse(bond_opt_get(BOND_OPT_PACKETS_PER_SLAVE), &newval)) {
4436 pr_warn("Warning: packets_per_slave (%d) should be between 0 and %u resetting to 1\n",
4437 packets_per_slave, USHRT_MAX);
4438 packets_per_slave = 1;
4439 }
4440
4441 if (bond_mode == BOND_MODE_ALB) {
4442 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",
4443 updelay);
4444 }
4445
4446 if (!miimon) {
4447 if (updelay || downdelay) {
4448 /* just warn the user the up/down delay will have
4449 * no effect since miimon is zero...
4450 */
4451 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",
4452 updelay, downdelay);
4453 }
4454 } else {
4455 /* don't allow arp monitoring */
4456 if (arp_interval) {
4457 pr_warn("Warning: miimon (%d) and arp_interval (%d) can't be used simultaneously, disabling ARP monitoring\n",
4458 miimon, arp_interval);
4459 arp_interval = 0;
4460 }
4461
4462 if ((updelay % miimon) != 0) {
4463 pr_warn("Warning: updelay (%d) is not a multiple of miimon (%d), updelay rounded to %d ms\n",
4464 updelay, miimon, (updelay / miimon) * miimon);
4465 }
4466
4467 updelay /= miimon;
4468
4469 if ((downdelay % miimon) != 0) {
4470 pr_warn("Warning: downdelay (%d) is not a multiple of miimon (%d), downdelay rounded to %d ms\n",
4471 downdelay, miimon,
4472 (downdelay / miimon) * miimon);
4473 }
4474
4475 downdelay /= miimon;
4476 }
4477
4478 if (arp_interval < 0) {
4479 pr_warn("Warning: arp_interval module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4480 arp_interval, INT_MAX);
4481 arp_interval = 0;
4482 }
4483
4484 for (arp_ip_count = 0, i = 0;
4485 (arp_ip_count < BOND_MAX_ARP_TARGETS) && arp_ip_target[i]; i++) {
4486 __be32 ip;
4487
4488 /* not a complete check, but good enough to catch mistakes */
4489 if (!in4_pton(arp_ip_target[i], -1, (u8 *)&ip, -1, NULL) ||
4490 !bond_is_ip_target_ok(ip)) {
4491 pr_warn("Warning: bad arp_ip_target module parameter (%s), ARP monitoring will not be performed\n",
4492 arp_ip_target[i]);
4493 arp_interval = 0;
4494 } else {
4495 if (bond_get_targets_ip(arp_target, ip) == -1)
4496 arp_target[arp_ip_count++] = ip;
4497 else
4498 pr_warn("Warning: duplicate address %pI4 in arp_ip_target, skipping\n",
4499 &ip);
4500 }
4501 }
4502
4503 if (arp_interval && !arp_ip_count) {
4504 /* don't allow arping if no arp_ip_target given... */
4505 pr_warn("Warning: arp_interval module parameter (%d) specified without providing an arp_ip_target parameter, arp_interval was reset to 0\n",
4506 arp_interval);
4507 arp_interval = 0;
4508 }
4509
4510 if (arp_validate) {
4511 if (!arp_interval) {
4512 pr_err("arp_validate requires arp_interval\n");
4513 return -EINVAL;
4514 }
4515
4516 bond_opt_initstr(&newval, arp_validate);
4517 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_ARP_VALIDATE),
4518 &newval);
4519 if (!valptr) {
4520 pr_err("Error: invalid arp_validate \"%s\"\n",
4521 arp_validate);
4522 return -EINVAL;
4523 }
4524 arp_validate_value = valptr->value;
4525 } else {
4526 arp_validate_value = 0;
4527 }
4528
4529 if (arp_all_targets) {
4530 bond_opt_initstr(&newval, arp_all_targets);
4531 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_ARP_ALL_TARGETS),
4532 &newval);
4533 if (!valptr) {
4534 pr_err("Error: invalid arp_all_targets_value \"%s\"\n",
4535 arp_all_targets);
4536 arp_all_targets_value = 0;
4537 } else {
4538 arp_all_targets_value = valptr->value;
4539 }
4540 }
4541
4542 if (miimon) {
4543 pr_info("MII link monitoring set to %d ms\n", miimon);
4544 } else if (arp_interval) {
4545 valptr = bond_opt_get_val(BOND_OPT_ARP_VALIDATE,
4546 arp_validate_value);
4547 pr_info("ARP monitoring set to %d ms, validate %s, with %d target(s):",
4548 arp_interval, valptr->string, arp_ip_count);
4549
4550 for (i = 0; i < arp_ip_count; i++)
4551 pr_cont(" %s", arp_ip_target[i]);
4552
4553 pr_cont("\n");
4554
4555 } else if (max_bonds) {
4556 /* miimon and arp_interval not set, we need one so things
4557 * work as expected, see bonding.txt for details
4558 */
4559 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");
4560 }
4561
4562 if (primary && !bond_mode_uses_primary(bond_mode)) {
4563 /* currently, using a primary only makes sense
4564 * in active backup, TLB or ALB modes
4565 */
4566 pr_warn("Warning: %s primary device specified but has no effect in %s mode\n",
4567 primary, bond_mode_name(bond_mode));
4568 primary = NULL;
4569 }
4570
4571 if (primary && primary_reselect) {
4572 bond_opt_initstr(&newval, primary_reselect);
4573 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_PRIMARY_RESELECT),
4574 &newval);
4575 if (!valptr) {
4576 pr_err("Error: Invalid primary_reselect \"%s\"\n",
4577 primary_reselect);
4578 return -EINVAL;
4579 }
4580 primary_reselect_value = valptr->value;
4581 } else {
4582 primary_reselect_value = BOND_PRI_RESELECT_ALWAYS;
4583 }
4584
4585 if (fail_over_mac) {
4586 bond_opt_initstr(&newval, fail_over_mac);
4587 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_FAIL_OVER_MAC),
4588 &newval);
4589 if (!valptr) {
4590 pr_err("Error: invalid fail_over_mac \"%s\"\n",
4591 fail_over_mac);
4592 return -EINVAL;
4593 }
4594 fail_over_mac_value = valptr->value;
4595 if (bond_mode != BOND_MODE_ACTIVEBACKUP)
4596 pr_warn("Warning: fail_over_mac only affects active-backup mode\n");
4597 } else {
4598 fail_over_mac_value = BOND_FOM_NONE;
4599 }
4600
4601 bond_opt_initstr(&newval, "default");
4602 valptr = bond_opt_parse(
4603 bond_opt_get(BOND_OPT_AD_ACTOR_SYS_PRIO),
4604 &newval);
4605 if (!valptr) {
4606 pr_err("Error: No ad_actor_sys_prio default value");
4607 return -EINVAL;
4608 }
4609 ad_actor_sys_prio = valptr->value;
4610
4611 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_AD_USER_PORT_KEY),
4612 &newval);
4613 if (!valptr) {
4614 pr_err("Error: No ad_user_port_key default value");
4615 return -EINVAL;
4616 }
4617 ad_user_port_key = valptr->value;
4618
4619 bond_opt_initstr(&newval, "default");
4620 valptr = bond_opt_parse(bond_opt_get(BOND_OPT_TLB_DYNAMIC_LB), &newval);
4621 if (!valptr) {
4622 pr_err("Error: No tlb_dynamic_lb default value");
4623 return -EINVAL;
4624 }
4625 tlb_dynamic_lb = valptr->value;
4626
4627 if (lp_interval == 0) {
4628 pr_warn("Warning: ip_interval must be between 1 and %d, so it was reset to %d\n",
4629 INT_MAX, BOND_ALB_DEFAULT_LP_INTERVAL);
4630 lp_interval = BOND_ALB_DEFAULT_LP_INTERVAL;
4631 }
4632
4633 /* fill params struct with the proper values */
4634 params->mode = bond_mode;
4635 params->xmit_policy = xmit_hashtype;
4636 params->miimon = miimon;
4637 params->num_peer_notif = num_peer_notif;
4638 params->arp_interval = arp_interval;
4639 params->arp_validate = arp_validate_value;
4640 params->arp_all_targets = arp_all_targets_value;
4641 params->updelay = updelay;
4642 params->downdelay = downdelay;
4643 params->use_carrier = use_carrier;
4644 params->lacp_fast = lacp_fast;
4645 params->primary[0] = 0;
4646 params->primary_reselect = primary_reselect_value;
4647 params->fail_over_mac = fail_over_mac_value;
4648 params->tx_queues = tx_queues;
4649 params->all_slaves_active = all_slaves_active;
4650 params->resend_igmp = resend_igmp;
4651 params->min_links = min_links;
4652 params->lp_interval = lp_interval;
4653 params->packets_per_slave = packets_per_slave;
4654 params->tlb_dynamic_lb = tlb_dynamic_lb;
4655 params->ad_actor_sys_prio = ad_actor_sys_prio;
4656 eth_zero_addr(params->ad_actor_system);
4657 params->ad_user_port_key = ad_user_port_key;
4658 if (packets_per_slave > 0) {
4659 params->reciprocal_packets_per_slave =
4660 reciprocal_value(packets_per_slave);
4661 } else {
4662 /* reciprocal_packets_per_slave is unused if
4663 * packets_per_slave is 0 or 1, just initialize it
4664 */
4665 params->reciprocal_packets_per_slave =
4666 (struct reciprocal_value) { 0 };
4667 }
4668
4669 if (primary) {
4670 strncpy(params->primary, primary, IFNAMSIZ);
4671 params->primary[IFNAMSIZ - 1] = 0;
4672 }
4673
4674 memcpy(params->arp_targets, arp_target, sizeof(arp_target));
4675
4676 return 0;
4677 }
4678
4679 /* Called from registration process */
4680 static int bond_init(struct net_device *bond_dev)
4681 {
4682 struct bonding *bond = netdev_priv(bond_dev);
4683 struct bond_net *bn = net_generic(dev_net(bond_dev), bond_net_id);
4684
4685 netdev_dbg(bond_dev, "Begin bond_init\n");
4686
4687 bond->wq = alloc_ordered_workqueue(bond_dev->name, WQ_MEM_RECLAIM);
4688 if (!bond->wq)
4689 return -ENOMEM;
4690
4691 netdev_lockdep_set_classes(bond_dev);
4692
4693 list_add_tail(&bond->bond_list, &bn->dev_list);
4694
4695 bond_prepare_sysfs_group(bond);
4696
4697 bond_debug_register(bond);
4698
4699 /* Ensure valid dev_addr */
4700 if (is_zero_ether_addr(bond_dev->dev_addr) &&
4701 bond_dev->addr_assign_type == NET_ADDR_PERM)
4702 eth_hw_addr_random(bond_dev);
4703
4704 return 0;
4705 }
4706
4707 unsigned int bond_get_num_tx_queues(void)
4708 {
4709 return tx_queues;
4710 }
4711
4712 /* Create a new bond based on the specified name and bonding parameters.
4713 * If name is NULL, obtain a suitable "bond%d" name for us.
4714 * Caller must NOT hold rtnl_lock; we need to release it here before we
4715 * set up our sysfs entries.
4716 */
4717 int bond_create(struct net *net, const char *name)
4718 {
4719 struct net_device *bond_dev;
4720 struct bonding *bond;
4721 struct alb_bond_info *bond_info;
4722 int res;
4723
4724 rtnl_lock();
4725
4726 bond_dev = alloc_netdev_mq(sizeof(struct bonding),
4727 name ? name : "bond%d", NET_NAME_UNKNOWN,
4728 bond_setup, tx_queues);
4729 if (!bond_dev) {
4730 pr_err("%s: eek! can't alloc netdev!\n", name);
4731 rtnl_unlock();
4732 return -ENOMEM;
4733 }
4734
4735 /*
4736 * Initialize rx_hashtbl_used_head to RLB_NULL_INDEX.
4737 * It is set to 0 by default which is wrong.
4738 */
4739 bond = netdev_priv(bond_dev);
4740 bond_info = &(BOND_ALB_INFO(bond));
4741 bond_info->rx_hashtbl_used_head = RLB_NULL_INDEX;
4742
4743 dev_net_set(bond_dev, net);
4744 bond_dev->rtnl_link_ops = &bond_link_ops;
4745
4746 res = register_netdevice(bond_dev);
4747
4748 netif_carrier_off(bond_dev);
4749
4750 bond_work_init_all(bond);
4751
4752 rtnl_unlock();
4753 if (res < 0)
4754 free_netdev(bond_dev);
4755 return res;
4756 }
4757
4758 static int __net_init bond_net_init(struct net *net)
4759 {
4760 struct bond_net *bn = net_generic(net, bond_net_id);
4761
4762 bn->net = net;
4763 INIT_LIST_HEAD(&bn->dev_list);
4764
4765 bond_create_proc_dir(bn);
4766 bond_create_sysfs(bn);
4767
4768 return 0;
4769 }
4770
4771 static void __net_exit bond_net_exit(struct net *net)
4772 {
4773 struct bond_net *bn = net_generic(net, bond_net_id);
4774 struct bonding *bond, *tmp_bond;
4775 LIST_HEAD(list);
4776
4777 bond_destroy_sysfs(bn);
4778
4779 /* Kill off any bonds created after unregistering bond rtnl ops */
4780 rtnl_lock();
4781 list_for_each_entry_safe(bond, tmp_bond, &bn->dev_list, bond_list)
4782 unregister_netdevice_queue(bond->dev, &list);
4783 unregister_netdevice_many(&list);
4784 rtnl_unlock();
4785
4786 bond_destroy_proc_dir(bn);
4787 }
4788
4789 static struct pernet_operations bond_net_ops = {
4790 .init = bond_net_init,
4791 .exit = bond_net_exit,
4792 .id = &bond_net_id,
4793 .size = sizeof(struct bond_net),
4794 };
4795
4796 static int __init bonding_init(void)
4797 {
4798 int i;
4799 int res;
4800
4801 pr_info("%s", bond_version);
4802
4803 res = bond_check_params(&bonding_defaults);
4804 if (res)
4805 goto out;
4806
4807 res = register_pernet_subsys(&bond_net_ops);
4808 if (res)
4809 goto out;
4810
4811 res = bond_netlink_init();
4812 if (res)
4813 goto err_link;
4814
4815 bond_create_debugfs();
4816
4817 for (i = 0; i < max_bonds; i++) {
4818 res = bond_create(&init_net, NULL);
4819 if (res)
4820 goto err;
4821 }
4822
4823 register_netdevice_notifier(&bond_netdev_notifier);
4824 out:
4825 return res;
4826 err:
4827 bond_destroy_debugfs();
4828 bond_netlink_fini();
4829 err_link:
4830 unregister_pernet_subsys(&bond_net_ops);
4831 goto out;
4832
4833 }
4834
4835 static void __exit bonding_exit(void)
4836 {
4837 unregister_netdevice_notifier(&bond_netdev_notifier);
4838
4839 bond_destroy_debugfs();
4840
4841 bond_netlink_fini();
4842 unregister_pernet_subsys(&bond_net_ops);
4843
4844 #ifdef CONFIG_NET_POLL_CONTROLLER
4845 /* Make sure we don't have an imbalance on our netpoll blocking */
4846 WARN_ON(atomic_read(&netpoll_block_tx));
4847 #endif
4848 }
4849
4850 module_init(bonding_init);
4851 module_exit(bonding_exit);
4852 MODULE_LICENSE("GPL");
4853 MODULE_VERSION(DRV_VERSION);
4854 MODULE_DESCRIPTION(DRV_DESCRIPTION ", v" DRV_VERSION);
4855 MODULE_AUTHOR("Thomas Davis, tadavis@lbl.gov and many others");
CRIS linux kernel tree