projects / deliverable / linux.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
Merge branch 'acpica-gpe' into release
[deliverable/linux.git] / drivers / net / bonding / bond_main.c
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 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
35
36 #include <linux/kernel.h>
37 #include <linux/module.h>
38 #include <linux/types.h>
39 #include <linux/fcntl.h>
40 #include <linux/interrupt.h>
41 #include <linux/ptrace.h>
42 #include <linux/ioport.h>
43 #include <linux/in.h>
44 #include <net/ip.h>
45 #include <linux/ip.h>
46 #include <linux/tcp.h>
47 #include <linux/udp.h>
48 #include <linux/slab.h>
49 #include <linux/string.h>
50 #include <linux/init.h>
51 #include <linux/timer.h>
52 #include <linux/socket.h>
53 #include <linux/ctype.h>
54 #include <linux/inet.h>
55 #include <linux/bitops.h>
56 #include <linux/io.h>
57 #include <asm/system.h>
58 #include <asm/dma.h>
59 #include <linux/uaccess.h>
60 #include <linux/errno.h>
61 #include <linux/netdevice.h>
62 #include <linux/netpoll.h>
63 #include <linux/inetdevice.h>
64 #include <linux/igmp.h>
65 #include <linux/etherdevice.h>
66 #include <linux/skbuff.h>
67 #include <net/sock.h>
68 #include <linux/rtnetlink.h>
69 #include <linux/proc_fs.h>
70 #include <linux/seq_file.h>
71 #include <linux/smp.h>
72 #include <linux/if_ether.h>
73 #include <net/arp.h>
74 #include <linux/mii.h>
75 #include <linux/ethtool.h>
76 #include <linux/if_vlan.h>
77 #include <linux/if_bonding.h>
78 #include <linux/jiffies.h>
79 #include <net/route.h>
80 #include <net/net_namespace.h>
81 #include <net/netns/generic.h>
82 #include "bonding.h"
83 #include "bond_3ad.h"
84 #include "bond_alb.h"
85
86 /*---------------------------- Module parameters ----------------------------*/
87
88 /* monitor all links that often (in milliseconds). <=0 disables monitoring */
89 #define BOND_LINK_MON_INTERV 0
90 #define BOND_LINK_ARP_INTERV 0
91
92 static int max_bonds = BOND_DEFAULT_MAX_BONDS;
93 static int num_grat_arp = 1;
94 static int num_unsol_na = 1;
95 static int miimon = BOND_LINK_MON_INTERV;
96 static int updelay;
97 static int downdelay;
98 static int use_carrier = 1;
99 static char *mode;
100 static char *primary;
101 static char *primary_reselect;
102 static char *lacp_rate;
103 static char *ad_select;
104 static char *xmit_hash_policy;
105 static int arp_interval = BOND_LINK_ARP_INTERV;
106 static char *arp_ip_target[BOND_MAX_ARP_TARGETS];
107 static char *arp_validate;
108 static char *fail_over_mac;
109 static struct bond_params bonding_defaults;
110
111 module_param(max_bonds, int, 0);
112 MODULE_PARM_DESC(max_bonds, "Max number of bonded devices");
113 module_param(num_grat_arp, int, 0644);
114 MODULE_PARM_DESC(num_grat_arp, "Number of gratuitous ARP packets to send on failover event");
115 module_param(num_unsol_na, int, 0644);
116 MODULE_PARM_DESC(num_unsol_na, "Number of unsolicited IPv6 Neighbor Advertisements packets to send on failover event");
117 module_param(miimon, int, 0);
118 MODULE_PARM_DESC(miimon, "Link check interval in milliseconds");
119 module_param(updelay, int, 0);
120 MODULE_PARM_DESC(updelay, "Delay before considering link up, in milliseconds");
121 module_param(downdelay, int, 0);
122 MODULE_PARM_DESC(downdelay, "Delay before considering link down, "
123 "in milliseconds");
124 module_param(use_carrier, int, 0);
125 MODULE_PARM_DESC(use_carrier, "Use netif_carrier_ok (vs MII ioctls) in miimon; "
126 "0 for off, 1 for on (default)");
127 module_param(mode, charp, 0);
128 MODULE_PARM_DESC(mode, "Mode of operation : 0 for balance-rr, "
129 "1 for active-backup, 2 for balance-xor, "
130 "3 for broadcast, 4 for 802.3ad, 5 for balance-tlb, "
131 "6 for balance-alb");
132 module_param(primary, charp, 0);
133 MODULE_PARM_DESC(primary, "Primary network device to use");
134 module_param(primary_reselect, charp, 0);
135 MODULE_PARM_DESC(primary_reselect, "Reselect primary slave "
136 "once it comes up; "
137 "0 for always (default), "
138 "1 for only if speed of primary is "
139 "better, "
140 "2 for only on active slave "
141 "failure");
142 module_param(lacp_rate, charp, 0);
143 MODULE_PARM_DESC(lacp_rate, "LACPDU tx rate to request from 802.3ad partner "
144 "(slow/fast)");
145 module_param(ad_select, charp, 0);
146 MODULE_PARM_DESC(ad_select, "803.ad aggregation selection logic: stable (0, default), bandwidth (1), count (2)");
147 module_param(xmit_hash_policy, charp, 0);
148 MODULE_PARM_DESC(xmit_hash_policy, "XOR hashing method: 0 for layer 2 (default)"
149 ", 1 for layer 3+4");
150 module_param(arp_interval, int, 0);
151 MODULE_PARM_DESC(arp_interval, "arp interval in milliseconds");
152 module_param_array(arp_ip_target, charp, NULL, 0);
153 MODULE_PARM_DESC(arp_ip_target, "arp targets in n.n.n.n form");
154 module_param(arp_validate, charp, 0);
155 MODULE_PARM_DESC(arp_validate, "validate src/dst of ARP probes: none (default), active, backup or all");
156 module_param(fail_over_mac, charp, 0);
157 MODULE_PARM_DESC(fail_over_mac, "For active-backup, do not set all slaves to the same MAC. none (default), active or follow");
158
159 /*----------------------------- Global variables ----------------------------*/
160
161 static const char * const version =
162 DRV_DESCRIPTION ": v" DRV_VERSION " (" DRV_RELDATE ")\n";
163
164 int bond_net_id __read_mostly;
165
166 static __be32 arp_target[BOND_MAX_ARP_TARGETS];
167 static int arp_ip_count;
168 static int bond_mode = BOND_MODE_ROUNDROBIN;
169 static int xmit_hashtype = BOND_XMIT_POLICY_LAYER2;
170 static int lacp_fast;
171 static int disable_netpoll = 1;
172
173 const struct bond_parm_tbl bond_lacp_tbl[] = {
174 { "slow", AD_LACP_SLOW},
175 { "fast", AD_LACP_FAST},
176 { NULL, -1},
177 };
178
179 const struct bond_parm_tbl bond_mode_tbl[] = {
180 { "balance-rr", BOND_MODE_ROUNDROBIN},
181 { "active-backup", BOND_MODE_ACTIVEBACKUP},
182 { "balance-xor", BOND_MODE_XOR},
183 { "broadcast", BOND_MODE_BROADCAST},
184 { "802.3ad", BOND_MODE_8023AD},
185 { "balance-tlb", BOND_MODE_TLB},
186 { "balance-alb", BOND_MODE_ALB},
187 { NULL, -1},
188 };
189
190 const struct bond_parm_tbl xmit_hashtype_tbl[] = {
191 { "layer2", BOND_XMIT_POLICY_LAYER2},
192 { "layer3+4", BOND_XMIT_POLICY_LAYER34},
193 { "layer2+3", BOND_XMIT_POLICY_LAYER23},
194 { NULL, -1},
195 };
196
197 const struct bond_parm_tbl arp_validate_tbl[] = {
198 { "none", BOND_ARP_VALIDATE_NONE},
199 { "active", BOND_ARP_VALIDATE_ACTIVE},
200 { "backup", BOND_ARP_VALIDATE_BACKUP},
201 { "all", BOND_ARP_VALIDATE_ALL},
202 { NULL, -1},
203 };
204
205 const struct bond_parm_tbl fail_over_mac_tbl[] = {
206 { "none", BOND_FOM_NONE},
207 { "active", BOND_FOM_ACTIVE},
208 { "follow", BOND_FOM_FOLLOW},
209 { NULL, -1},
210 };
211
212 const struct bond_parm_tbl pri_reselect_tbl[] = {
213 { "always", BOND_PRI_RESELECT_ALWAYS},
214 { "better", BOND_PRI_RESELECT_BETTER},
215 { "failure", BOND_PRI_RESELECT_FAILURE},
216 { NULL, -1},
217 };
218
219 struct bond_parm_tbl ad_select_tbl[] = {
220 { "stable", BOND_AD_STABLE},
221 { "bandwidth", BOND_AD_BANDWIDTH},
222 { "count", BOND_AD_COUNT},
223 { NULL, -1},
224 };
225
226 /*-------------------------- Forward declarations ---------------------------*/
227
228 static void bond_send_gratuitous_arp(struct bonding *bond);
229 static int bond_init(struct net_device *bond_dev);
230 static void bond_uninit(struct net_device *bond_dev);
231
232 /*---------------------------- General routines -----------------------------*/
233
234 static const char *bond_mode_name(int mode)
235 {
236 static const char *names[] = {
237 [BOND_MODE_ROUNDROBIN] = "load balancing (round-robin)",
238 [BOND_MODE_ACTIVEBACKUP] = "fault-tolerance (active-backup)",
239 [BOND_MODE_XOR] = "load balancing (xor)",
240 [BOND_MODE_BROADCAST] = "fault-tolerance (broadcast)",
241 [BOND_MODE_8023AD] = "IEEE 802.3ad Dynamic link aggregation",
242 [BOND_MODE_TLB] = "transmit load balancing",
243 [BOND_MODE_ALB] = "adaptive load balancing",
244 };
245
246 if (mode < 0 || mode > BOND_MODE_ALB)
247 return "unknown";
248
249 return names[mode];
250 }
251
252 /*---------------------------------- VLAN -----------------------------------*/
253
254 /**
255 * bond_add_vlan - add a new vlan id on bond
256 * @bond: bond that got the notification
257 * @vlan_id: the vlan id to add
258 *
259 * Returns -ENOMEM if allocation failed.
260 */
261 static int bond_add_vlan(struct bonding *bond, unsigned short vlan_id)
262 {
263 struct vlan_entry *vlan;
264
265 pr_debug("bond: %s, vlan id %d\n",
266 (bond ? bond->dev->name : "None"), vlan_id);
267
268 vlan = kzalloc(sizeof(struct vlan_entry), GFP_KERNEL);
269 if (!vlan)
270 return -ENOMEM;
271
272 INIT_LIST_HEAD(&vlan->vlan_list);
273 vlan->vlan_id = vlan_id;
274
275 write_lock_bh(&bond->lock);
276
277 list_add_tail(&vlan->vlan_list, &bond->vlan_list);
278
279 write_unlock_bh(&bond->lock);
280
281 pr_debug("added VLAN ID %d on bond %s\n", vlan_id, bond->dev->name);
282
283 return 0;
284 }
285
286 /**
287 * bond_del_vlan - delete a vlan id from bond
288 * @bond: bond that got the notification
289 * @vlan_id: the vlan id to delete
290 *
291 * returns -ENODEV if @vlan_id was not found in @bond.
292 */
293 static int bond_del_vlan(struct bonding *bond, unsigned short vlan_id)
294 {
295 struct vlan_entry *vlan;
296 int res = -ENODEV;
297
298 pr_debug("bond: %s, vlan id %d\n", bond->dev->name, vlan_id);
299
300 write_lock_bh(&bond->lock);
301
302 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
303 if (vlan->vlan_id == vlan_id) {
304 list_del(&vlan->vlan_list);
305
306 if (bond_is_lb(bond))
307 bond_alb_clear_vlan(bond, vlan_id);
308
309 pr_debug("removed VLAN ID %d from bond %s\n",
310 vlan_id, bond->dev->name);
311
312 kfree(vlan);
313
314 if (list_empty(&bond->vlan_list) &&
315 (bond->slave_cnt == 0)) {
316 /* Last VLAN removed and no slaves, so
317 * restore block on adding VLANs. This will
318 * be removed once new slaves that are not
319 * VLAN challenged will be added.
320 */
321 bond->dev->features |= NETIF_F_VLAN_CHALLENGED;
322 }
323
324 res = 0;
325 goto out;
326 }
327 }
328
329 pr_debug("couldn't find VLAN ID %d in bond %s\n",
330 vlan_id, bond->dev->name);
331
332 out:
333 write_unlock_bh(&bond->lock);
334 return res;
335 }
336
337 /**
338 * bond_has_challenged_slaves
339 * @bond: the bond we're working on
340 *
341 * Searches the slave list. Returns 1 if a vlan challenged slave
342 * was found, 0 otherwise.
343 *
344 * Assumes bond->lock is held.
345 */
346 static int bond_has_challenged_slaves(struct bonding *bond)
347 {
348 struct slave *slave;
349 int i;
350
351 bond_for_each_slave(bond, slave, i) {
352 if (slave->dev->features & NETIF_F_VLAN_CHALLENGED) {
353 pr_debug("found VLAN challenged slave - %s\n",
354 slave->dev->name);
355 return 1;
356 }
357 }
358
359 pr_debug("no VLAN challenged slaves found\n");
360 return 0;
361 }
362
363 /**
364 * bond_next_vlan - safely skip to the next item in the vlans list.
365 * @bond: the bond we're working on
366 * @curr: item we're advancing from
367 *
368 * Returns %NULL if list is empty, bond->next_vlan if @curr is %NULL,
369 * or @curr->next otherwise (even if it is @curr itself again).
370 *
371 * Caller must hold bond->lock
372 */
373 struct vlan_entry *bond_next_vlan(struct bonding *bond, struct vlan_entry *curr)
374 {
375 struct vlan_entry *next, *last;
376
377 if (list_empty(&bond->vlan_list))
378 return NULL;
379
380 if (!curr) {
381 next = list_entry(bond->vlan_list.next,
382 struct vlan_entry, vlan_list);
383 } else {
384 last = list_entry(bond->vlan_list.prev,
385 struct vlan_entry, vlan_list);
386 if (last == curr) {
387 next = list_entry(bond->vlan_list.next,
388 struct vlan_entry, vlan_list);
389 } else {
390 next = list_entry(curr->vlan_list.next,
391 struct vlan_entry, vlan_list);
392 }
393 }
394
395 return next;
396 }
397
398 /**
399 * bond_dev_queue_xmit - Prepare skb for xmit.
400 *
401 * @bond: bond device that got this skb for tx.
402 * @skb: hw accel VLAN tagged skb to transmit
403 * @slave_dev: slave that is supposed to xmit this skbuff
404 *
405 * When the bond gets an skb to transmit that is
406 * already hardware accelerated VLAN tagged, and it
407 * needs to relay this skb to a slave that is not
408 * hw accel capable, the skb needs to be "unaccelerated",
409 * i.e. strip the hwaccel tag and re-insert it as part
410 * of the payload.
411 */
412 int bond_dev_queue_xmit(struct bonding *bond, struct sk_buff *skb,
413 struct net_device *slave_dev)
414 {
415 unsigned short uninitialized_var(vlan_id);
416
417 if (!list_empty(&bond->vlan_list) &&
418 !(slave_dev->features & NETIF_F_HW_VLAN_TX) &&
419 vlan_get_tag(skb, &vlan_id) == 0) {
420 skb->dev = slave_dev;
421 skb = vlan_put_tag(skb, vlan_id);
422 if (!skb) {
423 /* vlan_put_tag() frees the skb in case of error,
424 * so return success here so the calling functions
425 * won't attempt to free is again.
426 */
427 return 0;
428 }
429 } else {
430 skb->dev = slave_dev;
431 }
432
433 skb->priority = 1;
434 #ifdef CONFIG_NET_POLL_CONTROLLER
435 if (unlikely(bond->dev->priv_flags & IFF_IN_NETPOLL)) {
436 struct netpoll *np = bond->dev->npinfo->netpoll;
437 slave_dev->npinfo = bond->dev->npinfo;
438 np->real_dev = np->dev = skb->dev;
439 slave_dev->priv_flags |= IFF_IN_NETPOLL;
440 netpoll_send_skb(np, skb);
441 slave_dev->priv_flags &= ~IFF_IN_NETPOLL;
442 np->dev = bond->dev;
443 } else
444 #endif
445 dev_queue_xmit(skb);
446
447 return 0;
448 }
449
450 /*
451 * In the following 3 functions, bond_vlan_rx_register(), bond_vlan_rx_add_vid
452 * and bond_vlan_rx_kill_vid, We don't protect the slave list iteration with a
453 * lock because:
454 * a. This operation is performed in IOCTL context,
455 * b. The operation is protected by the RTNL semaphore in the 8021q code,
456 * c. Holding a lock with BH disabled while directly calling a base driver
457 * entry point is generally a BAD idea.
458 *
459 * The design of synchronization/protection for this operation in the 8021q
460 * module is good for one or more VLAN devices over a single physical device
461 * and cannot be extended for a teaming solution like bonding, so there is a
462 * potential race condition here where a net device from the vlan group might
463 * be referenced (either by a base driver or the 8021q code) while it is being
464 * removed from the system. However, it turns out we're not making matters
465 * worse, and if it works for regular VLAN usage it will work here too.
466 */
467
468 /**
469 * bond_vlan_rx_register - Propagates registration to slaves
470 * @bond_dev: bonding net device that got called
471 * @grp: vlan group being registered
472 */
473 static void bond_vlan_rx_register(struct net_device *bond_dev,
474 struct vlan_group *grp)
475 {
476 struct bonding *bond = netdev_priv(bond_dev);
477 struct slave *slave;
478 int i;
479
480 bond->vlgrp = grp;
481
482 bond_for_each_slave(bond, slave, i) {
483 struct net_device *slave_dev = slave->dev;
484 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
485
486 if ((slave_dev->features & NETIF_F_HW_VLAN_RX) &&
487 slave_ops->ndo_vlan_rx_register) {
488 slave_ops->ndo_vlan_rx_register(slave_dev, grp);
489 }
490 }
491 }
492
493 /**
494 * bond_vlan_rx_add_vid - Propagates adding an id to slaves
495 * @bond_dev: bonding net device that got called
496 * @vid: vlan id being added
497 */
498 static void bond_vlan_rx_add_vid(struct net_device *bond_dev, uint16_t vid)
499 {
500 struct bonding *bond = netdev_priv(bond_dev);
501 struct slave *slave;
502 int i, res;
503
504 bond_for_each_slave(bond, slave, i) {
505 struct net_device *slave_dev = slave->dev;
506 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
507
508 if ((slave_dev->features & NETIF_F_HW_VLAN_FILTER) &&
509 slave_ops->ndo_vlan_rx_add_vid) {
510 slave_ops->ndo_vlan_rx_add_vid(slave_dev, vid);
511 }
512 }
513
514 res = bond_add_vlan(bond, vid);
515 if (res) {
516 pr_err("%s: Error: Failed to add vlan id %d\n",
517 bond_dev->name, vid);
518 }
519 }
520
521 /**
522 * bond_vlan_rx_kill_vid - Propagates deleting an id to slaves
523 * @bond_dev: bonding net device that got called
524 * @vid: vlan id being removed
525 */
526 static void bond_vlan_rx_kill_vid(struct net_device *bond_dev, uint16_t vid)
527 {
528 struct bonding *bond = netdev_priv(bond_dev);
529 struct slave *slave;
530 struct net_device *vlan_dev;
531 int i, res;
532
533 bond_for_each_slave(bond, slave, i) {
534 struct net_device *slave_dev = slave->dev;
535 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
536
537 if ((slave_dev->features & NETIF_F_HW_VLAN_FILTER) &&
538 slave_ops->ndo_vlan_rx_kill_vid) {
539 /* Save and then restore vlan_dev in the grp array,
540 * since the slave's driver might clear it.
541 */
542 vlan_dev = vlan_group_get_device(bond->vlgrp, vid);
543 slave_ops->ndo_vlan_rx_kill_vid(slave_dev, vid);
544 vlan_group_set_device(bond->vlgrp, vid, vlan_dev);
545 }
546 }
547
548 res = bond_del_vlan(bond, vid);
549 if (res) {
550 pr_err("%s: Error: Failed to remove vlan id %d\n",
551 bond_dev->name, vid);
552 }
553 }
554
555 static void bond_add_vlans_on_slave(struct bonding *bond, struct net_device *slave_dev)
556 {
557 struct vlan_entry *vlan;
558 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
559
560 write_lock_bh(&bond->lock);
561
562 if (list_empty(&bond->vlan_list))
563 goto out;
564
565 if ((slave_dev->features & NETIF_F_HW_VLAN_RX) &&
566 slave_ops->ndo_vlan_rx_register)
567 slave_ops->ndo_vlan_rx_register(slave_dev, bond->vlgrp);
568
569 if (!(slave_dev->features & NETIF_F_HW_VLAN_FILTER) ||
570 !(slave_ops->ndo_vlan_rx_add_vid))
571 goto out;
572
573 list_for_each_entry(vlan, &bond->vlan_list, vlan_list)
574 slave_ops->ndo_vlan_rx_add_vid(slave_dev, vlan->vlan_id);
575
576 out:
577 write_unlock_bh(&bond->lock);
578 }
579
580 static void bond_del_vlans_from_slave(struct bonding *bond,
581 struct net_device *slave_dev)
582 {
583 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
584 struct vlan_entry *vlan;
585 struct net_device *vlan_dev;
586
587 write_lock_bh(&bond->lock);
588
589 if (list_empty(&bond->vlan_list))
590 goto out;
591
592 if (!(slave_dev->features & NETIF_F_HW_VLAN_FILTER) ||
593 !(slave_ops->ndo_vlan_rx_kill_vid))
594 goto unreg;
595
596 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
597 /* Save and then restore vlan_dev in the grp array,
598 * since the slave's driver might clear it.
599 */
600 vlan_dev = vlan_group_get_device(bond->vlgrp, vlan->vlan_id);
601 slave_ops->ndo_vlan_rx_kill_vid(slave_dev, vlan->vlan_id);
602 vlan_group_set_device(bond->vlgrp, vlan->vlan_id, vlan_dev);
603 }
604
605 unreg:
606 if ((slave_dev->features & NETIF_F_HW_VLAN_RX) &&
607 slave_ops->ndo_vlan_rx_register)
608 slave_ops->ndo_vlan_rx_register(slave_dev, NULL);
609
610 out:
611 write_unlock_bh(&bond->lock);
612 }
613
614 /*------------------------------- Link status -------------------------------*/
615
616 /*
617 * Set the carrier state for the master according to the state of its
618 * slaves. If any slaves are up, the master is up. In 802.3ad mode,
619 * do special 802.3ad magic.
620 *
621 * Returns zero if carrier state does not change, nonzero if it does.
622 */
623 static int bond_set_carrier(struct bonding *bond)
624 {
625 struct slave *slave;
626 int i;
627
628 if (bond->slave_cnt == 0)
629 goto down;
630
631 if (bond->params.mode == BOND_MODE_8023AD)
632 return bond_3ad_set_carrier(bond);
633
634 bond_for_each_slave(bond, slave, i) {
635 if (slave->link == BOND_LINK_UP) {
636 if (!netif_carrier_ok(bond->dev)) {
637 netif_carrier_on(bond->dev);
638 return 1;
639 }
640 return 0;
641 }
642 }
643
644 down:
645 if (netif_carrier_ok(bond->dev)) {
646 netif_carrier_off(bond->dev);
647 return 1;
648 }
649 return 0;
650 }
651
652 /*
653 * Get link speed and duplex from the slave's base driver
654 * using ethtool. If for some reason the call fails or the
655 * values are invalid, fake speed and duplex to 100/Full
656 * and return error.
657 */
658 static int bond_update_speed_duplex(struct slave *slave)
659 {
660 struct net_device *slave_dev = slave->dev;
661 struct ethtool_cmd etool;
662 int res;
663
664 /* Fake speed and duplex */
665 slave->speed = SPEED_100;
666 slave->duplex = DUPLEX_FULL;
667
668 if (!slave_dev->ethtool_ops || !slave_dev->ethtool_ops->get_settings)
669 return -1;
670
671 res = slave_dev->ethtool_ops->get_settings(slave_dev, &etool);
672 if (res < 0)
673 return -1;
674
675 switch (etool.speed) {
676 case SPEED_10:
677 case SPEED_100:
678 case SPEED_1000:
679 case SPEED_10000:
680 break;
681 default:
682 return -1;
683 }
684
685 switch (etool.duplex) {
686 case DUPLEX_FULL:
687 case DUPLEX_HALF:
688 break;
689 default:
690 return -1;
691 }
692
693 slave->speed = etool.speed;
694 slave->duplex = etool.duplex;
695
696 return 0;
697 }
698
699 /*
700 * if <dev> supports MII link status reporting, check its link status.
701 *
702 * We either do MII/ETHTOOL ioctls, or check netif_carrier_ok(),
703 * depending upon the setting of the use_carrier parameter.
704 *
705 * Return either BMSR_LSTATUS, meaning that the link is up (or we
706 * can't tell and just pretend it is), or 0, meaning that the link is
707 * down.
708 *
709 * If reporting is non-zero, instead of faking link up, return -1 if
710 * both ETHTOOL and MII ioctls fail (meaning the device does not
711 * support them). If use_carrier is set, return whatever it says.
712 * It'd be nice if there was a good way to tell if a driver supports
713 * netif_carrier, but there really isn't.
714 */
715 static int bond_check_dev_link(struct bonding *bond,
716 struct net_device *slave_dev, int reporting)
717 {
718 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
719 int (*ioctl)(struct net_device *, struct ifreq *, int);
720 struct ifreq ifr;
721 struct mii_ioctl_data *mii;
722
723 if (!reporting && !netif_running(slave_dev))
724 return 0;
725
726 if (bond->params.use_carrier)
727 return netif_carrier_ok(slave_dev) ? BMSR_LSTATUS : 0;
728
729 /* Try to get link status using Ethtool first. */
730 if (slave_dev->ethtool_ops) {
731 if (slave_dev->ethtool_ops->get_link) {
732 u32 link;
733
734 link = slave_dev->ethtool_ops->get_link(slave_dev);
735
736 return link ? BMSR_LSTATUS : 0;
737 }
738 }
739
740 /* Ethtool can't be used, fallback to MII ioctls. */
741 ioctl = slave_ops->ndo_do_ioctl;
742 if (ioctl) {
743 /* TODO: set pointer to correct ioctl on a per team member */
744 /* bases to make this more efficient. that is, once */
745 /* we determine the correct ioctl, we will always */
746 /* call it and not the others for that team */
747 /* member. */
748
749 /*
750 * We cannot assume that SIOCGMIIPHY will also read a
751 * register; not all network drivers (e.g., e100)
752 * support that.
753 */
754
755 /* Yes, the mii is overlaid on the ifreq.ifr_ifru */
756 strncpy(ifr.ifr_name, slave_dev->name, IFNAMSIZ);
757 mii = if_mii(&ifr);
758 if (IOCTL(slave_dev, &ifr, SIOCGMIIPHY) == 0) {
759 mii->reg_num = MII_BMSR;
760 if (IOCTL(slave_dev, &ifr, SIOCGMIIREG) == 0)
761 return mii->val_out & BMSR_LSTATUS;
762 }
763 }
764
765 /*
766 * If reporting, report that either there's no dev->do_ioctl,
767 * or both SIOCGMIIREG and get_link failed (meaning that we
768 * cannot report link status). If not reporting, pretend
769 * we're ok.
770 */
771 return reporting ? -1 : BMSR_LSTATUS;
772 }
773
774 /*----------------------------- Multicast list ------------------------------*/
775
776 /*
777 * Push the promiscuity flag down to appropriate slaves
778 */
779 static int bond_set_promiscuity(struct bonding *bond, int inc)
780 {
781 int err = 0;
782 if (USES_PRIMARY(bond->params.mode)) {
783 /* write lock already acquired */
784 if (bond->curr_active_slave) {
785 err = dev_set_promiscuity(bond->curr_active_slave->dev,
786 inc);
787 }
788 } else {
789 struct slave *slave;
790 int i;
791 bond_for_each_slave(bond, slave, i) {
792 err = dev_set_promiscuity(slave->dev, inc);
793 if (err)
794 return err;
795 }
796 }
797 return err;
798 }
799
800 /*
801 * Push the allmulti flag down to all slaves
802 */
803 static int bond_set_allmulti(struct bonding *bond, int inc)
804 {
805 int err = 0;
806 if (USES_PRIMARY(bond->params.mode)) {
807 /* write lock already acquired */
808 if (bond->curr_active_slave) {
809 err = dev_set_allmulti(bond->curr_active_slave->dev,
810 inc);
811 }
812 } else {
813 struct slave *slave;
814 int i;
815 bond_for_each_slave(bond, slave, i) {
816 err = dev_set_allmulti(slave->dev, inc);
817 if (err)
818 return err;
819 }
820 }
821 return err;
822 }
823
824 /*
825 * Add a Multicast address to slaves
826 * according to mode
827 */
828 static void bond_mc_add(struct bonding *bond, void *addr)
829 {
830 if (USES_PRIMARY(bond->params.mode)) {
831 /* write lock already acquired */
832 if (bond->curr_active_slave)
833 dev_mc_add(bond->curr_active_slave->dev, addr);
834 } else {
835 struct slave *slave;
836 int i;
837
838 bond_for_each_slave(bond, slave, i)
839 dev_mc_add(slave->dev, addr);
840 }
841 }
842
843 /*
844 * Remove a multicast address from slave
845 * according to mode
846 */
847 static void bond_mc_del(struct bonding *bond, void *addr)
848 {
849 if (USES_PRIMARY(bond->params.mode)) {
850 /* write lock already acquired */
851 if (bond->curr_active_slave)
852 dev_mc_del(bond->curr_active_slave->dev, addr);
853 } else {
854 struct slave *slave;
855 int i;
856 bond_for_each_slave(bond, slave, i) {
857 dev_mc_del(slave->dev, addr);
858 }
859 }
860 }
861
862
863 /*
864 * Retrieve the list of registered multicast addresses for the bonding
865 * device and retransmit an IGMP JOIN request to the current active
866 * slave.
867 */
868 static void bond_resend_igmp_join_requests(struct bonding *bond)
869 {
870 struct in_device *in_dev;
871 struct ip_mc_list *im;
872
873 rcu_read_lock();
874 in_dev = __in_dev_get_rcu(bond->dev);
875 if (in_dev) {
876 for (im = in_dev->mc_list; im; im = im->next)
877 ip_mc_rejoin_group(im);
878 }
879
880 rcu_read_unlock();
881 }
882
883 /*
884 * flush all members of flush->mc_list from device dev->mc_list
885 */
886 static void bond_mc_list_flush(struct net_device *bond_dev,
887 struct net_device *slave_dev)
888 {
889 struct bonding *bond = netdev_priv(bond_dev);
890 struct netdev_hw_addr *ha;
891
892 netdev_for_each_mc_addr(ha, bond_dev)
893 dev_mc_del(slave_dev, ha->addr);
894
895 if (bond->params.mode == BOND_MODE_8023AD) {
896 /* del lacpdu mc addr from mc list */
897 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
898
899 dev_mc_del(slave_dev, lacpdu_multicast);
900 }
901 }
902
903 /*--------------------------- Active slave change ---------------------------*/
904
905 /*
906 * Update the mc list and multicast-related flags for the new and
907 * old active slaves (if any) according to the multicast mode, and
908 * promiscuous flags unconditionally.
909 */
910 static void bond_mc_swap(struct bonding *bond, struct slave *new_active,
911 struct slave *old_active)
912 {
913 struct netdev_hw_addr *ha;
914
915 if (!USES_PRIMARY(bond->params.mode))
916 /* nothing to do - mc list is already up-to-date on
917 * all slaves
918 */
919 return;
920
921 if (old_active) {
922 if (bond->dev->flags & IFF_PROMISC)
923 dev_set_promiscuity(old_active->dev, -1);
924
925 if (bond->dev->flags & IFF_ALLMULTI)
926 dev_set_allmulti(old_active->dev, -1);
927
928 netdev_for_each_mc_addr(ha, bond->dev)
929 dev_mc_del(old_active->dev, ha->addr);
930 }
931
932 if (new_active) {
933 /* FIXME: Signal errors upstream. */
934 if (bond->dev->flags & IFF_PROMISC)
935 dev_set_promiscuity(new_active->dev, 1);
936
937 if (bond->dev->flags & IFF_ALLMULTI)
938 dev_set_allmulti(new_active->dev, 1);
939
940 netdev_for_each_mc_addr(ha, bond->dev)
941 dev_mc_add(new_active->dev, ha->addr);
942 bond_resend_igmp_join_requests(bond);
943 }
944 }
945
946 /*
947 * bond_do_fail_over_mac
948 *
949 * Perform special MAC address swapping for fail_over_mac settings
950 *
951 * Called with RTNL, bond->lock for read, curr_slave_lock for write_bh.
952 */
953 static void bond_do_fail_over_mac(struct bonding *bond,
954 struct slave *new_active,
955 struct slave *old_active)
956 __releases(&bond->curr_slave_lock)
957 __releases(&bond->lock)
958 __acquires(&bond->lock)
959 __acquires(&bond->curr_slave_lock)
960 {
961 u8 tmp_mac[ETH_ALEN];
962 struct sockaddr saddr;
963 int rv;
964
965 switch (bond->params.fail_over_mac) {
966 case BOND_FOM_ACTIVE:
967 if (new_active)
968 memcpy(bond->dev->dev_addr, new_active->dev->dev_addr,
969 new_active->dev->addr_len);
970 break;
971 case BOND_FOM_FOLLOW:
972 /*
973 * if new_active && old_active, swap them
974 * if just old_active, do nothing (going to no active slave)
975 * if just new_active, set new_active to bond's MAC
976 */
977 if (!new_active)
978 return;
979
980 write_unlock_bh(&bond->curr_slave_lock);
981 read_unlock(&bond->lock);
982
983 if (old_active) {
984 memcpy(tmp_mac, new_active->dev->dev_addr, ETH_ALEN);
985 memcpy(saddr.sa_data, old_active->dev->dev_addr,
986 ETH_ALEN);
987 saddr.sa_family = new_active->dev->type;
988 } else {
989 memcpy(saddr.sa_data, bond->dev->dev_addr, ETH_ALEN);
990 saddr.sa_family = bond->dev->type;
991 }
992
993 rv = dev_set_mac_address(new_active->dev, &saddr);
994 if (rv) {
995 pr_err("%s: Error %d setting MAC of slave %s\n",
996 bond->dev->name, -rv, new_active->dev->name);
997 goto out;
998 }
999
1000 if (!old_active)
1001 goto out;
1002
1003 memcpy(saddr.sa_data, tmp_mac, ETH_ALEN);
1004 saddr.sa_family = old_active->dev->type;
1005
1006 rv = dev_set_mac_address(old_active->dev, &saddr);
1007 if (rv)
1008 pr_err("%s: Error %d setting MAC of slave %s\n",
1009 bond->dev->name, -rv, new_active->dev->name);
1010 out:
1011 read_lock(&bond->lock);
1012 write_lock_bh(&bond->curr_slave_lock);
1013 break;
1014 default:
1015 pr_err("%s: bond_do_fail_over_mac impossible: bad policy %d\n",
1016 bond->dev->name, bond->params.fail_over_mac);
1017 break;
1018 }
1019
1020 }
1021
1022 static bool bond_should_change_active(struct bonding *bond)
1023 {
1024 struct slave *prim = bond->primary_slave;
1025 struct slave *curr = bond->curr_active_slave;
1026
1027 if (!prim || !curr || curr->link != BOND_LINK_UP)
1028 return true;
1029 if (bond->force_primary) {
1030 bond->force_primary = false;
1031 return true;
1032 }
1033 if (bond->params.primary_reselect == BOND_PRI_RESELECT_BETTER &&
1034 (prim->speed < curr->speed ||
1035 (prim->speed == curr->speed && prim->duplex <= curr->duplex)))
1036 return false;
1037 if (bond->params.primary_reselect == BOND_PRI_RESELECT_FAILURE)
1038 return false;
1039 return true;
1040 }
1041
1042 /**
1043 * find_best_interface - select the best available slave to be the active one
1044 * @bond: our bonding struct
1045 *
1046 * Warning: Caller must hold curr_slave_lock for writing.
1047 */
1048 static struct slave *bond_find_best_slave(struct bonding *bond)
1049 {
1050 struct slave *new_active, *old_active;
1051 struct slave *bestslave = NULL;
1052 int mintime = bond->params.updelay;
1053 int i;
1054
1055 new_active = bond->curr_active_slave;
1056
1057 if (!new_active) { /* there were no active slaves left */
1058 if (bond->slave_cnt > 0) /* found one slave */
1059 new_active = bond->first_slave;
1060 else
1061 return NULL; /* still no slave, return NULL */
1062 }
1063
1064 if ((bond->primary_slave) &&
1065 bond->primary_slave->link == BOND_LINK_UP &&
1066 bond_should_change_active(bond)) {
1067 new_active = bond->primary_slave;
1068 }
1069
1070 /* remember where to stop iterating over the slaves */
1071 old_active = new_active;
1072
1073 bond_for_each_slave_from(bond, new_active, i, old_active) {
1074 if (new_active->link == BOND_LINK_UP) {
1075 return new_active;
1076 } else if (new_active->link == BOND_LINK_BACK &&
1077 IS_UP(new_active->dev)) {
1078 /* link up, but waiting for stabilization */
1079 if (new_active->delay < mintime) {
1080 mintime = new_active->delay;
1081 bestslave = new_active;
1082 }
1083 }
1084 }
1085
1086 return bestslave;
1087 }
1088
1089 /**
1090 * change_active_interface - change the active slave into the specified one
1091 * @bond: our bonding struct
1092 * @new: the new slave to make the active one
1093 *
1094 * Set the new slave to the bond's settings and unset them on the old
1095 * curr_active_slave.
1096 * Setting include flags, mc-list, promiscuity, allmulti, etc.
1097 *
1098 * If @new's link state is %BOND_LINK_BACK we'll set it to %BOND_LINK_UP,
1099 * because it is apparently the best available slave we have, even though its
1100 * updelay hasn't timed out yet.
1101 *
1102 * If new_active is not NULL, caller must hold bond->lock for read and
1103 * curr_slave_lock for write_bh.
1104 */
1105 void bond_change_active_slave(struct bonding *bond, struct slave *new_active)
1106 {
1107 struct slave *old_active = bond->curr_active_slave;
1108
1109 if (old_active == new_active)
1110 return;
1111
1112 if (new_active) {
1113 new_active->jiffies = jiffies;
1114
1115 if (new_active->link == BOND_LINK_BACK) {
1116 if (USES_PRIMARY(bond->params.mode)) {
1117 pr_info("%s: making interface %s the new active one %d ms earlier.\n",
1118 bond->dev->name, new_active->dev->name,
1119 (bond->params.updelay - new_active->delay) * bond->params.miimon);
1120 }
1121
1122 new_active->delay = 0;
1123 new_active->link = BOND_LINK_UP;
1124
1125 if (bond->params.mode == BOND_MODE_8023AD)
1126 bond_3ad_handle_link_change(new_active, BOND_LINK_UP);
1127
1128 if (bond_is_lb(bond))
1129 bond_alb_handle_link_change(bond, new_active, BOND_LINK_UP);
1130 } else {
1131 if (USES_PRIMARY(bond->params.mode)) {
1132 pr_info("%s: making interface %s the new active one.\n",
1133 bond->dev->name, new_active->dev->name);
1134 }
1135 }
1136 }
1137
1138 if (USES_PRIMARY(bond->params.mode))
1139 bond_mc_swap(bond, new_active, old_active);
1140
1141 if (bond_is_lb(bond)) {
1142 bond_alb_handle_active_change(bond, new_active);
1143 if (old_active)
1144 bond_set_slave_inactive_flags(old_active);
1145 if (new_active)
1146 bond_set_slave_active_flags(new_active);
1147 } else {
1148 bond->curr_active_slave = new_active;
1149 }
1150
1151 if (bond->params.mode == BOND_MODE_ACTIVEBACKUP) {
1152 if (old_active)
1153 bond_set_slave_inactive_flags(old_active);
1154
1155 if (new_active) {
1156 bond_set_slave_active_flags(new_active);
1157
1158 if (bond->params.fail_over_mac)
1159 bond_do_fail_over_mac(bond, new_active,
1160 old_active);
1161
1162 bond->send_grat_arp = bond->params.num_grat_arp;
1163 bond_send_gratuitous_arp(bond);
1164
1165 bond->send_unsol_na = bond->params.num_unsol_na;
1166 bond_send_unsolicited_na(bond);
1167
1168 write_unlock_bh(&bond->curr_slave_lock);
1169 read_unlock(&bond->lock);
1170
1171 netdev_bonding_change(bond->dev, NETDEV_BONDING_FAILOVER);
1172
1173 read_lock(&bond->lock);
1174 write_lock_bh(&bond->curr_slave_lock);
1175 }
1176 }
1177
1178 /* resend IGMP joins since all were sent on curr_active_slave */
1179 if (bond->params.mode == BOND_MODE_ROUNDROBIN) {
1180 bond_resend_igmp_join_requests(bond);
1181 }
1182 }
1183
1184 /**
1185 * bond_select_active_slave - select a new active slave, if needed
1186 * @bond: our bonding struct
1187 *
1188 * This functions should be called when one of the following occurs:
1189 * - The old curr_active_slave has been released or lost its link.
1190 * - The primary_slave has got its link back.
1191 * - A slave has got its link back and there's no old curr_active_slave.
1192 *
1193 * Caller must hold bond->lock for read and curr_slave_lock for write_bh.
1194 */
1195 void bond_select_active_slave(struct bonding *bond)
1196 {
1197 struct slave *best_slave;
1198 int rv;
1199
1200 best_slave = bond_find_best_slave(bond);
1201 if (best_slave != bond->curr_active_slave) {
1202 bond_change_active_slave(bond, best_slave);
1203 rv = bond_set_carrier(bond);
1204 if (!rv)
1205 return;
1206
1207 if (netif_carrier_ok(bond->dev)) {
1208 pr_info("%s: first active interface up!\n",
1209 bond->dev->name);
1210 } else {
1211 pr_info("%s: now running without any active interface !\n",
1212 bond->dev->name);
1213 }
1214 }
1215 }
1216
1217 /*--------------------------- slave list handling ---------------------------*/
1218
1219 /*
1220 * This function attaches the slave to the end of list.
1221 *
1222 * bond->lock held for writing by caller.
1223 */
1224 static void bond_attach_slave(struct bonding *bond, struct slave *new_slave)
1225 {
1226 if (bond->first_slave == NULL) { /* attaching the first slave */
1227 new_slave->next = new_slave;
1228 new_slave->prev = new_slave;
1229 bond->first_slave = new_slave;
1230 } else {
1231 new_slave->next = bond->first_slave;
1232 new_slave->prev = bond->first_slave->prev;
1233 new_slave->next->prev = new_slave;
1234 new_slave->prev->next = new_slave;
1235 }
1236
1237 bond->slave_cnt++;
1238 }
1239
1240 /*
1241 * This function detaches the slave from the list.
1242 * WARNING: no check is made to verify if the slave effectively
1243 * belongs to <bond>.
1244 * Nothing is freed on return, structures are just unchained.
1245 * If any slave pointer in bond was pointing to <slave>,
1246 * it should be changed by the calling function.
1247 *
1248 * bond->lock held for writing by caller.
1249 */
1250 static void bond_detach_slave(struct bonding *bond, struct slave *slave)
1251 {
1252 if (slave->next)
1253 slave->next->prev = slave->prev;
1254
1255 if (slave->prev)
1256 slave->prev->next = slave->next;
1257
1258 if (bond->first_slave == slave) { /* slave is the first slave */
1259 if (bond->slave_cnt > 1) { /* there are more slave */
1260 bond->first_slave = slave->next;
1261 } else {
1262 bond->first_slave = NULL; /* slave was the last one */
1263 }
1264 }
1265
1266 slave->next = NULL;
1267 slave->prev = NULL;
1268 bond->slave_cnt--;
1269 }
1270
1271 #ifdef CONFIG_NET_POLL_CONTROLLER
1272 /*
1273 * You must hold read lock on bond->lock before calling this.
1274 */
1275 static bool slaves_support_netpoll(struct net_device *bond_dev)
1276 {
1277 struct bonding *bond = netdev_priv(bond_dev);
1278 struct slave *slave;
1279 int i = 0;
1280 bool ret = true;
1281
1282 bond_for_each_slave(bond, slave, i) {
1283 if ((slave->dev->priv_flags & IFF_DISABLE_NETPOLL) ||
1284 !slave->dev->netdev_ops->ndo_poll_controller)
1285 ret = false;
1286 }
1287 return i != 0 && ret;
1288 }
1289
1290 static void bond_poll_controller(struct net_device *bond_dev)
1291 {
1292 struct net_device *dev = bond_dev->npinfo->netpoll->real_dev;
1293 if (dev != bond_dev)
1294 netpoll_poll_dev(dev);
1295 }
1296
1297 static void bond_netpoll_cleanup(struct net_device *bond_dev)
1298 {
1299 struct bonding *bond = netdev_priv(bond_dev);
1300 struct slave *slave;
1301 const struct net_device_ops *ops;
1302 int i;
1303
1304 read_lock(&bond->lock);
1305 bond_dev->npinfo = NULL;
1306 bond_for_each_slave(bond, slave, i) {
1307 if (slave->dev) {
1308 ops = slave->dev->netdev_ops;
1309 if (ops->ndo_netpoll_cleanup)
1310 ops->ndo_netpoll_cleanup(slave->dev);
1311 else
1312 slave->dev->npinfo = NULL;
1313 }
1314 }
1315 read_unlock(&bond->lock);
1316 }
1317
1318 #else
1319
1320 static void bond_netpoll_cleanup(struct net_device *bond_dev)
1321 {
1322 }
1323
1324 #endif
1325
1326 /*---------------------------------- IOCTL ----------------------------------*/
1327
1328 static int bond_sethwaddr(struct net_device *bond_dev,
1329 struct net_device *slave_dev)
1330 {
1331 pr_debug("bond_dev=%p\n", bond_dev);
1332 pr_debug("slave_dev=%p\n", slave_dev);
1333 pr_debug("slave_dev->addr_len=%d\n", slave_dev->addr_len);
1334 memcpy(bond_dev->dev_addr, slave_dev->dev_addr, slave_dev->addr_len);
1335 return 0;
1336 }
1337
1338 #define BOND_VLAN_FEATURES \
1339 (NETIF_F_VLAN_CHALLENGED | NETIF_F_HW_VLAN_RX | NETIF_F_HW_VLAN_TX | \
1340 NETIF_F_HW_VLAN_FILTER)
1341
1342 /*
1343 * Compute the common dev->feature set available to all slaves. Some
1344 * feature bits are managed elsewhere, so preserve those feature bits
1345 * on the master device.
1346 */
1347 static int bond_compute_features(struct bonding *bond)
1348 {
1349 struct slave *slave;
1350 struct net_device *bond_dev = bond->dev;
1351 unsigned long features = bond_dev->features;
1352 unsigned long vlan_features = 0;
1353 unsigned short max_hard_header_len = max((u16)ETH_HLEN,
1354 bond_dev->hard_header_len);
1355 int i;
1356
1357 features &= ~(NETIF_F_ALL_CSUM | BOND_VLAN_FEATURES);
1358 features |= NETIF_F_GSO_MASK | NETIF_F_NO_CSUM;
1359
1360 if (!bond->first_slave)
1361 goto done;
1362
1363 features &= ~NETIF_F_ONE_FOR_ALL;
1364
1365 vlan_features = bond->first_slave->dev->vlan_features;
1366 bond_for_each_slave(bond, slave, i) {
1367 features = netdev_increment_features(features,
1368 slave->dev->features,
1369 NETIF_F_ONE_FOR_ALL);
1370 vlan_features = netdev_increment_features(vlan_features,
1371 slave->dev->vlan_features,
1372 NETIF_F_ONE_FOR_ALL);
1373 if (slave->dev->hard_header_len > max_hard_header_len)
1374 max_hard_header_len = slave->dev->hard_header_len;
1375 }
1376
1377 done:
1378 features |= (bond_dev->features & BOND_VLAN_FEATURES);
1379 bond_dev->features = netdev_fix_features(features, NULL);
1380 bond_dev->vlan_features = netdev_fix_features(vlan_features, NULL);
1381 bond_dev->hard_header_len = max_hard_header_len;
1382
1383 return 0;
1384 }
1385
1386 static void bond_setup_by_slave(struct net_device *bond_dev,
1387 struct net_device *slave_dev)
1388 {
1389 struct bonding *bond = netdev_priv(bond_dev);
1390
1391 bond_dev->header_ops = slave_dev->header_ops;
1392
1393 bond_dev->type = slave_dev->type;
1394 bond_dev->hard_header_len = slave_dev->hard_header_len;
1395 bond_dev->addr_len = slave_dev->addr_len;
1396
1397 memcpy(bond_dev->broadcast, slave_dev->broadcast,
1398 slave_dev->addr_len);
1399 bond->setup_by_slave = 1;
1400 }
1401
1402 /* enslave device <slave> to bond device <master> */
1403 int bond_enslave(struct net_device *bond_dev, struct net_device *slave_dev)
1404 {
1405 struct bonding *bond = netdev_priv(bond_dev);
1406 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
1407 struct slave *new_slave = NULL;
1408 struct netdev_hw_addr *ha;
1409 struct sockaddr addr;
1410 int link_reporting;
1411 int old_features = bond_dev->features;
1412 int res = 0;
1413
1414 if (!bond->params.use_carrier && slave_dev->ethtool_ops == NULL &&
1415 slave_ops->ndo_do_ioctl == NULL) {
1416 pr_warning("%s: Warning: no link monitoring support for %s\n",
1417 bond_dev->name, slave_dev->name);
1418 }
1419
1420 /* bond must be initialized by bond_open() before enslaving */
1421 if (!(bond_dev->flags & IFF_UP)) {
1422 pr_warning("%s: master_dev is not up in bond_enslave\n",
1423 bond_dev->name);
1424 }
1425
1426 /* already enslaved */
1427 if (slave_dev->flags & IFF_SLAVE) {
1428 pr_debug("Error, Device was already enslaved\n");
1429 return -EBUSY;
1430 }
1431
1432 /* vlan challenged mutual exclusion */
1433 /* no need to lock since we're protected by rtnl_lock */
1434 if (slave_dev->features & NETIF_F_VLAN_CHALLENGED) {
1435 pr_debug("%s: NETIF_F_VLAN_CHALLENGED\n", slave_dev->name);
1436 if (!list_empty(&bond->vlan_list)) {
1437 pr_err("%s: Error: cannot enslave VLAN challenged slave %s on VLAN enabled bond %s\n",
1438 bond_dev->name, slave_dev->name, bond_dev->name);
1439 return -EPERM;
1440 } else {
1441 pr_warning("%s: Warning: enslaved VLAN challenged slave %s. Adding VLANs will be blocked as long as %s is part of bond %s\n",
1442 bond_dev->name, slave_dev->name,
1443 slave_dev->name, bond_dev->name);
1444 bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
1445 }
1446 } else {
1447 pr_debug("%s: ! NETIF_F_VLAN_CHALLENGED\n", slave_dev->name);
1448 if (bond->slave_cnt == 0) {
1449 /* First slave, and it is not VLAN challenged,
1450 * so remove the block of adding VLANs over the bond.
1451 */
1452 bond_dev->features &= ~NETIF_F_VLAN_CHALLENGED;
1453 }
1454 }
1455
1456 /*
1457 * Old ifenslave binaries are no longer supported. These can
1458 * be identified with moderate accuracy by the state of the slave:
1459 * the current ifenslave will set the interface down prior to
1460 * enslaving it; the old ifenslave will not.
1461 */
1462 if ((slave_dev->flags & IFF_UP)) {
1463 pr_err("%s is up. This may be due to an out of date ifenslave.\n",
1464 slave_dev->name);
1465 res = -EPERM;
1466 goto err_undo_flags;
1467 }
1468
1469 /* set bonding device ether type by slave - bonding netdevices are
1470 * created with ether_setup, so when the slave type is not ARPHRD_ETHER
1471 * there is a need to override some of the type dependent attribs/funcs.
1472 *
1473 * bond ether type mutual exclusion - don't allow slaves of dissimilar
1474 * ether type (eg ARPHRD_ETHER and ARPHRD_INFINIBAND) share the same bond
1475 */
1476 if (bond->slave_cnt == 0) {
1477 if (bond_dev->type != slave_dev->type) {
1478 pr_debug("%s: change device type from %d to %d\n",
1479 bond_dev->name,
1480 bond_dev->type, slave_dev->type);
1481
1482 res = netdev_bonding_change(bond_dev,
1483 NETDEV_PRE_TYPE_CHANGE);
1484 res = notifier_to_errno(res);
1485 if (res) {
1486 pr_err("%s: refused to change device type\n",
1487 bond_dev->name);
1488 res = -EBUSY;
1489 goto err_undo_flags;
1490 }
1491
1492 /* Flush unicast and multicast addresses */
1493 dev_uc_flush(bond_dev);
1494 dev_mc_flush(bond_dev);
1495
1496 if (slave_dev->type != ARPHRD_ETHER)
1497 bond_setup_by_slave(bond_dev, slave_dev);
1498 else
1499 ether_setup(bond_dev);
1500
1501 netdev_bonding_change(bond_dev,
1502 NETDEV_POST_TYPE_CHANGE);
1503 }
1504 } else if (bond_dev->type != slave_dev->type) {
1505 pr_err("%s ether type (%d) is different from other slaves (%d), can not enslave it.\n",
1506 slave_dev->name,
1507 slave_dev->type, bond_dev->type);
1508 res = -EINVAL;
1509 goto err_undo_flags;
1510 }
1511
1512 if (slave_ops->ndo_set_mac_address == NULL) {
1513 if (bond->slave_cnt == 0) {
1514 pr_warning("%s: Warning: The first slave device specified does not support setting the MAC address. Setting fail_over_mac to active.",
1515 bond_dev->name);
1516 bond->params.fail_over_mac = BOND_FOM_ACTIVE;
1517 } else if (bond->params.fail_over_mac != BOND_FOM_ACTIVE) {
1518 pr_err("%s: Error: The slave device specified does not support setting the MAC address, but fail_over_mac is not set to active.\n",
1519 bond_dev->name);
1520 res = -EOPNOTSUPP;
1521 goto err_undo_flags;
1522 }
1523 }
1524
1525 new_slave = kzalloc(sizeof(struct slave), GFP_KERNEL);
1526 if (!new_slave) {
1527 res = -ENOMEM;
1528 goto err_undo_flags;
1529 }
1530
1531 /* save slave's original flags before calling
1532 * netdev_set_master and dev_open
1533 */
1534 new_slave->original_flags = slave_dev->flags;
1535
1536 /*
1537 * Save slave's original ("permanent") mac address for modes
1538 * that need it, and for restoring it upon release, and then
1539 * set it to the master's address
1540 */
1541 memcpy(new_slave->perm_hwaddr, slave_dev->dev_addr, ETH_ALEN);
1542
1543 if (!bond->params.fail_over_mac) {
1544 /*
1545 * Set slave to master's mac address. The application already
1546 * set the master's mac address to that of the first slave
1547 */
1548 memcpy(addr.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
1549 addr.sa_family = slave_dev->type;
1550 res = dev_set_mac_address(slave_dev, &addr);
1551 if (res) {
1552 pr_debug("Error %d calling set_mac_address\n", res);
1553 goto err_free;
1554 }
1555 }
1556
1557 res = netdev_set_master(slave_dev, bond_dev);
1558 if (res) {
1559 pr_debug("Error %d calling netdev_set_master\n", res);
1560 goto err_restore_mac;
1561 }
1562 /* open the slave since the application closed it */
1563 res = dev_open(slave_dev);
1564 if (res) {
1565 pr_debug("Opening slave %s failed\n", slave_dev->name);
1566 goto err_unset_master;
1567 }
1568
1569 new_slave->dev = slave_dev;
1570 slave_dev->priv_flags |= IFF_BONDING;
1571
1572 if (bond_is_lb(bond)) {
1573 /* bond_alb_init_slave() must be called before all other stages since
1574 * it might fail and we do not want to have to undo everything
1575 */
1576 res = bond_alb_init_slave(bond, new_slave);
1577 if (res)
1578 goto err_close;
1579 }
1580
1581 /* If the mode USES_PRIMARY, then the new slave gets the
1582 * master's promisc (and mc) settings only if it becomes the
1583 * curr_active_slave, and that is taken care of later when calling
1584 * bond_change_active()
1585 */
1586 if (!USES_PRIMARY(bond->params.mode)) {
1587 /* set promiscuity level to new slave */
1588 if (bond_dev->flags & IFF_PROMISC) {
1589 res = dev_set_promiscuity(slave_dev, 1);
1590 if (res)
1591 goto err_close;
1592 }
1593
1594 /* set allmulti level to new slave */
1595 if (bond_dev->flags & IFF_ALLMULTI) {
1596 res = dev_set_allmulti(slave_dev, 1);
1597 if (res)
1598 goto err_close;
1599 }
1600
1601 netif_addr_lock_bh(bond_dev);
1602 /* upload master's mc_list to new slave */
1603 netdev_for_each_mc_addr(ha, bond_dev)
1604 dev_mc_add(slave_dev, ha->addr);
1605 netif_addr_unlock_bh(bond_dev);
1606 }
1607
1608 if (bond->params.mode == BOND_MODE_8023AD) {
1609 /* add lacpdu mc addr to mc list */
1610 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
1611
1612 dev_mc_add(slave_dev, lacpdu_multicast);
1613 }
1614
1615 bond_add_vlans_on_slave(bond, slave_dev);
1616
1617 write_lock_bh(&bond->lock);
1618
1619 bond_attach_slave(bond, new_slave);
1620
1621 new_slave->delay = 0;
1622 new_slave->link_failure_count = 0;
1623
1624 bond_compute_features(bond);
1625
1626 write_unlock_bh(&bond->lock);
1627
1628 read_lock(&bond->lock);
1629
1630 new_slave->last_arp_rx = jiffies;
1631
1632 if (bond->params.miimon && !bond->params.use_carrier) {
1633 link_reporting = bond_check_dev_link(bond, slave_dev, 1);
1634
1635 if ((link_reporting == -1) && !bond->params.arp_interval) {
1636 /*
1637 * miimon is set but a bonded network driver
1638 * does not support ETHTOOL/MII and
1639 * arp_interval is not set. Note: if
1640 * use_carrier is enabled, we will never go
1641 * here (because netif_carrier is always
1642 * supported); thus, we don't need to change
1643 * the messages for netif_carrier.
1644 */
1645 pr_warning("%s: Warning: MII and ETHTOOL support not available for interface %s, and arp_interval/arp_ip_target module parameters not specified, thus bonding will not detect link failures! see bonding.txt for details.\n",
1646 bond_dev->name, slave_dev->name);
1647 } else if (link_reporting == -1) {
1648 /* unable get link status using mii/ethtool */
1649 pr_warning("%s: Warning: can't get link status from interface %s; the network driver associated with this interface does not support MII or ETHTOOL link status reporting, thus miimon has no effect on this interface.\n",
1650 bond_dev->name, slave_dev->name);
1651 }
1652 }
1653
1654 /* check for initial state */
1655 if (!bond->params.miimon ||
1656 (bond_check_dev_link(bond, slave_dev, 0) == BMSR_LSTATUS)) {
1657 if (bond->params.updelay) {
1658 pr_debug("Initial state of slave_dev is BOND_LINK_BACK\n");
1659 new_slave->link = BOND_LINK_BACK;
1660 new_slave->delay = bond->params.updelay;
1661 } else {
1662 pr_debug("Initial state of slave_dev is BOND_LINK_UP\n");
1663 new_slave->link = BOND_LINK_UP;
1664 }
1665 new_slave->jiffies = jiffies;
1666 } else {
1667 pr_debug("Initial state of slave_dev is BOND_LINK_DOWN\n");
1668 new_slave->link = BOND_LINK_DOWN;
1669 }
1670
1671 if (bond_update_speed_duplex(new_slave) &&
1672 (new_slave->link != BOND_LINK_DOWN)) {
1673 pr_warning("%s: Warning: failed to get speed and duplex from %s, assumed to be 100Mb/sec and Full.\n",
1674 bond_dev->name, new_slave->dev->name);
1675
1676 if (bond->params.mode == BOND_MODE_8023AD) {
1677 pr_warning("%s: Warning: Operation of 802.3ad mode requires ETHTOOL support in base driver for proper aggregator selection.\n",
1678 bond_dev->name);
1679 }
1680 }
1681
1682 if (USES_PRIMARY(bond->params.mode) && bond->params.primary[0]) {
1683 /* if there is a primary slave, remember it */
1684 if (strcmp(bond->params.primary, new_slave->dev->name) == 0) {
1685 bond->primary_slave = new_slave;
1686 bond->force_primary = true;
1687 }
1688 }
1689
1690 write_lock_bh(&bond->curr_slave_lock);
1691
1692 switch (bond->params.mode) {
1693 case BOND_MODE_ACTIVEBACKUP:
1694 bond_set_slave_inactive_flags(new_slave);
1695 bond_select_active_slave(bond);
1696 break;
1697 case BOND_MODE_8023AD:
1698 /* in 802.3ad mode, the internal mechanism
1699 * will activate the slaves in the selected
1700 * aggregator
1701 */
1702 bond_set_slave_inactive_flags(new_slave);
1703 /* if this is the first slave */
1704 if (bond->slave_cnt == 1) {
1705 SLAVE_AD_INFO(new_slave).id = 1;
1706 /* Initialize AD with the number of times that the AD timer is called in 1 second
1707 * can be called only after the mac address of the bond is set
1708 */
1709 bond_3ad_initialize(bond, 1000/AD_TIMER_INTERVAL,
1710 bond->params.lacp_fast);
1711 } else {
1712 SLAVE_AD_INFO(new_slave).id =
1713 SLAVE_AD_INFO(new_slave->prev).id + 1;
1714 }
1715
1716 bond_3ad_bind_slave(new_slave);
1717 break;
1718 case BOND_MODE_TLB:
1719 case BOND_MODE_ALB:
1720 new_slave->state = BOND_STATE_ACTIVE;
1721 bond_set_slave_inactive_flags(new_slave);
1722 bond_select_active_slave(bond);
1723 break;
1724 default:
1725 pr_debug("This slave is always active in trunk mode\n");
1726
1727 /* always active in trunk mode */
1728 new_slave->state = BOND_STATE_ACTIVE;
1729
1730 /* In trunking mode there is little meaning to curr_active_slave
1731 * anyway (it holds no special properties of the bond device),
1732 * so we can change it without calling change_active_interface()
1733 */
1734 if (!bond->curr_active_slave)
1735 bond->curr_active_slave = new_slave;
1736
1737 break;
1738 } /* switch(bond_mode) */
1739
1740 write_unlock_bh(&bond->curr_slave_lock);
1741
1742 bond_set_carrier(bond);
1743
1744 #ifdef CONFIG_NET_POLL_CONTROLLER
1745 /*
1746 * Netpoll and bonding is broken, make sure it is not initialized
1747 * until it is fixed.
1748 */
1749 if (disable_netpoll) {
1750 bond_dev->priv_flags |= IFF_DISABLE_NETPOLL;
1751 } else {
1752 if (slaves_support_netpoll(bond_dev)) {
1753 bond_dev->priv_flags &= ~IFF_DISABLE_NETPOLL;
1754 if (bond_dev->npinfo)
1755 slave_dev->npinfo = bond_dev->npinfo;
1756 } else if (!(bond_dev->priv_flags & IFF_DISABLE_NETPOLL)) {
1757 bond_dev->priv_flags |= IFF_DISABLE_NETPOLL;
1758 pr_info("New slave device %s does not support netpoll\n",
1759 slave_dev->name);
1760 pr_info("Disabling netpoll support for %s\n", bond_dev->name);
1761 }
1762 }
1763 #endif
1764 read_unlock(&bond->lock);
1765
1766 res = bond_create_slave_symlinks(bond_dev, slave_dev);
1767 if (res)
1768 goto err_close;
1769
1770 pr_info("%s: enslaving %s as a%s interface with a%s link.\n",
1771 bond_dev->name, slave_dev->name,
1772 new_slave->state == BOND_STATE_ACTIVE ? "n active" : " backup",
1773 new_slave->link != BOND_LINK_DOWN ? "n up" : " down");
1774
1775 /* enslave is successful */
1776 return 0;
1777
1778 /* Undo stages on error */
1779 err_close:
1780 dev_close(slave_dev);
1781
1782 err_unset_master:
1783 netdev_set_master(slave_dev, NULL);
1784
1785 err_restore_mac:
1786 if (!bond->params.fail_over_mac) {
1787 /* XXX TODO - fom follow mode needs to change master's
1788 * MAC if this slave's MAC is in use by the bond, or at
1789 * least print a warning.
1790 */
1791 memcpy(addr.sa_data, new_slave->perm_hwaddr, ETH_ALEN);
1792 addr.sa_family = slave_dev->type;
1793 dev_set_mac_address(slave_dev, &addr);
1794 }
1795
1796 err_free:
1797 kfree(new_slave);
1798
1799 err_undo_flags:
1800 bond_dev->features = old_features;
1801
1802 return res;
1803 }
1804
1805 /*
1806 * Try to release the slave device <slave> from the bond device <master>
1807 * It is legal to access curr_active_slave without a lock because all the function
1808 * is write-locked.
1809 *
1810 * The rules for slave state should be:
1811 * for Active/Backup:
1812 * Active stays on all backups go down
1813 * for Bonded connections:
1814 * The first up interface should be left on and all others downed.
1815 */
1816 int bond_release(struct net_device *bond_dev, struct net_device *slave_dev)
1817 {
1818 struct bonding *bond = netdev_priv(bond_dev);
1819 struct slave *slave, *oldcurrent;
1820 struct sockaddr addr;
1821
1822 /* slave is not a slave or master is not master of this slave */
1823 if (!(slave_dev->flags & IFF_SLAVE) ||
1824 (slave_dev->master != bond_dev)) {
1825 pr_err("%s: Error: cannot release %s.\n",
1826 bond_dev->name, slave_dev->name);
1827 return -EINVAL;
1828 }
1829
1830 netdev_bonding_change(bond_dev, NETDEV_BONDING_DESLAVE);
1831 write_lock_bh(&bond->lock);
1832
1833 slave = bond_get_slave_by_dev(bond, slave_dev);
1834 if (!slave) {
1835 /* not a slave of this bond */
1836 pr_info("%s: %s not enslaved\n",
1837 bond_dev->name, slave_dev->name);
1838 write_unlock_bh(&bond->lock);
1839 return -EINVAL;
1840 }
1841
1842 if (!bond->params.fail_over_mac) {
1843 if (!compare_ether_addr(bond_dev->dev_addr, slave->perm_hwaddr) &&
1844 bond->slave_cnt > 1)
1845 pr_warning("%s: Warning: the permanent HWaddr of %s - %pM - is still in use by %s. Set the HWaddr of %s to a different address to avoid conflicts.\n",
1846 bond_dev->name, slave_dev->name,
1847 slave->perm_hwaddr,
1848 bond_dev->name, slave_dev->name);
1849 }
1850
1851 /* Inform AD package of unbinding of slave. */
1852 if (bond->params.mode == BOND_MODE_8023AD) {
1853 /* must be called before the slave is
1854 * detached from the list
1855 */
1856 bond_3ad_unbind_slave(slave);
1857 }
1858
1859 pr_info("%s: releasing %s interface %s\n",
1860 bond_dev->name,
1861 (slave->state == BOND_STATE_ACTIVE) ? "active" : "backup",
1862 slave_dev->name);
1863
1864 oldcurrent = bond->curr_active_slave;
1865
1866 bond->current_arp_slave = NULL;
1867
1868 /* release the slave from its bond */
1869 bond_detach_slave(bond, slave);
1870
1871 bond_compute_features(bond);
1872
1873 if (bond->primary_slave == slave)
1874 bond->primary_slave = NULL;
1875
1876 if (oldcurrent == slave)
1877 bond_change_active_slave(bond, NULL);
1878
1879 if (bond_is_lb(bond)) {
1880 /* Must be called only after the slave has been
1881 * detached from the list and the curr_active_slave
1882 * has been cleared (if our_slave == old_current),
1883 * but before a new active slave is selected.
1884 */
1885 write_unlock_bh(&bond->lock);
1886 bond_alb_deinit_slave(bond, slave);
1887 write_lock_bh(&bond->lock);
1888 }
1889
1890 if (oldcurrent == slave) {
1891 /*
1892 * Note that we hold RTNL over this sequence, so there
1893 * is no concern that another slave add/remove event
1894 * will interfere.
1895 */
1896 write_unlock_bh(&bond->lock);
1897 read_lock(&bond->lock);
1898 write_lock_bh(&bond->curr_slave_lock);
1899
1900 bond_select_active_slave(bond);
1901
1902 write_unlock_bh(&bond->curr_slave_lock);
1903 read_unlock(&bond->lock);
1904 write_lock_bh(&bond->lock);
1905 }
1906
1907 if (bond->slave_cnt == 0) {
1908 bond_set_carrier(bond);
1909
1910 /* if the last slave was removed, zero the mac address
1911 * of the master so it will be set by the application
1912 * to the mac address of the first slave
1913 */
1914 memset(bond_dev->dev_addr, 0, bond_dev->addr_len);
1915
1916 if (list_empty(&bond->vlan_list)) {
1917 bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
1918 } else {
1919 pr_warning("%s: Warning: clearing HW address of %s while it still has VLANs.\n",
1920 bond_dev->name, bond_dev->name);
1921 pr_warning("%s: When re-adding slaves, make sure the bond's HW address matches its VLANs'.\n",
1922 bond_dev->name);
1923 }
1924 } else if ((bond_dev->features & NETIF_F_VLAN_CHALLENGED) &&
1925 !bond_has_challenged_slaves(bond)) {
1926 pr_info("%s: last VLAN challenged slave %s left bond %s. VLAN blocking is removed\n",
1927 bond_dev->name, slave_dev->name, bond_dev->name);
1928 bond_dev->features &= ~NETIF_F_VLAN_CHALLENGED;
1929 }
1930
1931 write_unlock_bh(&bond->lock);
1932
1933 /* must do this from outside any spinlocks */
1934 bond_destroy_slave_symlinks(bond_dev, slave_dev);
1935
1936 bond_del_vlans_from_slave(bond, slave_dev);
1937
1938 /* If the mode USES_PRIMARY, then we should only remove its
1939 * promisc and mc settings if it was the curr_active_slave, but that was
1940 * already taken care of above when we detached the slave
1941 */
1942 if (!USES_PRIMARY(bond->params.mode)) {
1943 /* unset promiscuity level from slave */
1944 if (bond_dev->flags & IFF_PROMISC)
1945 dev_set_promiscuity(slave_dev, -1);
1946
1947 /* unset allmulti level from slave */
1948 if (bond_dev->flags & IFF_ALLMULTI)
1949 dev_set_allmulti(slave_dev, -1);
1950
1951 /* flush master's mc_list from slave */
1952 netif_addr_lock_bh(bond_dev);
1953 bond_mc_list_flush(bond_dev, slave_dev);
1954 netif_addr_unlock_bh(bond_dev);
1955 }
1956
1957 netdev_set_master(slave_dev, NULL);
1958
1959 #ifdef CONFIG_NET_POLL_CONTROLLER
1960 read_lock_bh(&bond->lock);
1961
1962 /* Make sure netpoll over stays disabled until fixed. */
1963 if (!disable_netpoll)
1964 if (slaves_support_netpoll(bond_dev))
1965 bond_dev->priv_flags &= ~IFF_DISABLE_NETPOLL;
1966 read_unlock_bh(&bond->lock);
1967 if (slave_dev->netdev_ops->ndo_netpoll_cleanup)
1968 slave_dev->netdev_ops->ndo_netpoll_cleanup(slave_dev);
1969 else
1970 slave_dev->npinfo = NULL;
1971 #endif
1972
1973 /* close slave before restoring its mac address */
1974 dev_close(slave_dev);
1975
1976 if (bond->params.fail_over_mac != BOND_FOM_ACTIVE) {
1977 /* restore original ("permanent") mac address */
1978 memcpy(addr.sa_data, slave->perm_hwaddr, ETH_ALEN);
1979 addr.sa_family = slave_dev->type;
1980 dev_set_mac_address(slave_dev, &addr);
1981 }
1982
1983 slave_dev->priv_flags &= ~(IFF_MASTER_8023AD | IFF_MASTER_ALB |
1984 IFF_SLAVE_INACTIVE | IFF_BONDING |
1985 IFF_SLAVE_NEEDARP);
1986
1987 kfree(slave);
1988
1989 return 0; /* deletion OK */
1990 }
1991
1992 /*
1993 * First release a slave and than destroy the bond if no more slaves are left.
1994 * Must be under rtnl_lock when this function is called.
1995 */
1996 int bond_release_and_destroy(struct net_device *bond_dev,
1997 struct net_device *slave_dev)
1998 {
1999 struct bonding *bond = netdev_priv(bond_dev);
2000 int ret;
2001
2002 ret = bond_release(bond_dev, slave_dev);
2003 if ((ret == 0) && (bond->slave_cnt == 0)) {
2004 pr_info("%s: destroying bond %s.\n",
2005 bond_dev->name, bond_dev->name);
2006 unregister_netdevice(bond_dev);
2007 }
2008 return ret;
2009 }
2010
2011 /*
2012 * This function releases all slaves.
2013 */
2014 static int bond_release_all(struct net_device *bond_dev)
2015 {
2016 struct bonding *bond = netdev_priv(bond_dev);
2017 struct slave *slave;
2018 struct net_device *slave_dev;
2019 struct sockaddr addr;
2020
2021 write_lock_bh(&bond->lock);
2022
2023 netif_carrier_off(bond_dev);
2024
2025 if (bond->slave_cnt == 0)
2026 goto out;
2027
2028 bond->current_arp_slave = NULL;
2029 bond->primary_slave = NULL;
2030 bond_change_active_slave(bond, NULL);
2031
2032 while ((slave = bond->first_slave) != NULL) {
2033 /* Inform AD package of unbinding of slave
2034 * before slave is detached from the list.
2035 */
2036 if (bond->params.mode == BOND_MODE_8023AD)
2037 bond_3ad_unbind_slave(slave);
2038
2039 slave_dev = slave->dev;
2040 bond_detach_slave(bond, slave);
2041
2042 /* now that the slave is detached, unlock and perform
2043 * all the undo steps that should not be called from
2044 * within a lock.
2045 */
2046 write_unlock_bh(&bond->lock);
2047
2048 if (bond_is_lb(bond)) {
2049 /* must be called only after the slave
2050 * has been detached from the list
2051 */
2052 bond_alb_deinit_slave(bond, slave);
2053 }
2054
2055 bond_compute_features(bond);
2056
2057 bond_destroy_slave_symlinks(bond_dev, slave_dev);
2058 bond_del_vlans_from_slave(bond, slave_dev);
2059
2060 /* If the mode USES_PRIMARY, then we should only remove its
2061 * promisc and mc settings if it was the curr_active_slave, but that was
2062 * already taken care of above when we detached the slave
2063 */
2064 if (!USES_PRIMARY(bond->params.mode)) {
2065 /* unset promiscuity level from slave */
2066 if (bond_dev->flags & IFF_PROMISC)
2067 dev_set_promiscuity(slave_dev, -1);
2068
2069 /* unset allmulti level from slave */
2070 if (bond_dev->flags & IFF_ALLMULTI)
2071 dev_set_allmulti(slave_dev, -1);
2072
2073 /* flush master's mc_list from slave */
2074 netif_addr_lock_bh(bond_dev);
2075 bond_mc_list_flush(bond_dev, slave_dev);
2076 netif_addr_unlock_bh(bond_dev);
2077 }
2078
2079 netdev_set_master(slave_dev, NULL);
2080
2081 /* close slave before restoring its mac address */
2082 dev_close(slave_dev);
2083
2084 if (!bond->params.fail_over_mac) {
2085 /* restore original ("permanent") mac address*/
2086 memcpy(addr.sa_data, slave->perm_hwaddr, ETH_ALEN);
2087 addr.sa_family = slave_dev->type;
2088 dev_set_mac_address(slave_dev, &addr);
2089 }
2090
2091 slave_dev->priv_flags &= ~(IFF_MASTER_8023AD | IFF_MASTER_ALB |
2092 IFF_SLAVE_INACTIVE);
2093
2094 kfree(slave);
2095
2096 /* re-acquire the lock before getting the next slave */
2097 write_lock_bh(&bond->lock);
2098 }
2099
2100 /* zero the mac address of the master so it will be
2101 * set by the application to the mac address of the
2102 * first slave
2103 */
2104 memset(bond_dev->dev_addr, 0, bond_dev->addr_len);
2105
2106 if (list_empty(&bond->vlan_list))
2107 bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
2108 else {
2109 pr_warning("%s: Warning: clearing HW address of %s while it still has VLANs.\n",
2110 bond_dev->name, bond_dev->name);
2111 pr_warning("%s: When re-adding slaves, make sure the bond's HW address matches its VLANs'.\n",
2112 bond_dev->name);
2113 }
2114
2115 pr_info("%s: released all slaves\n", bond_dev->name);
2116
2117 out:
2118 write_unlock_bh(&bond->lock);
2119
2120 return 0;
2121 }
2122
2123 /*
2124 * This function changes the active slave to slave <slave_dev>.
2125 * It returns -EINVAL in the following cases.
2126 * - <slave_dev> is not found in the list.
2127 * - There is not active slave now.
2128 * - <slave_dev> is already active.
2129 * - The link state of <slave_dev> is not BOND_LINK_UP.
2130 * - <slave_dev> is not running.
2131 * In these cases, this function does nothing.
2132 * In the other cases, current_slave pointer is changed and 0 is returned.
2133 */
2134 static int bond_ioctl_change_active(struct net_device *bond_dev, struct net_device *slave_dev)
2135 {
2136 struct bonding *bond = netdev_priv(bond_dev);
2137 struct slave *old_active = NULL;
2138 struct slave *new_active = NULL;
2139 int res = 0;
2140
2141 if (!USES_PRIMARY(bond->params.mode))
2142 return -EINVAL;
2143
2144 /* Verify that master_dev is indeed the master of slave_dev */
2145 if (!(slave_dev->flags & IFF_SLAVE) || (slave_dev->master != bond_dev))
2146 return -EINVAL;
2147
2148 read_lock(&bond->lock);
2149
2150 read_lock(&bond->curr_slave_lock);
2151 old_active = bond->curr_active_slave;
2152 read_unlock(&bond->curr_slave_lock);
2153
2154 new_active = bond_get_slave_by_dev(bond, slave_dev);
2155
2156 /*
2157 * Changing to the current active: do nothing; return success.
2158 */
2159 if (new_active && (new_active == old_active)) {
2160 read_unlock(&bond->lock);
2161 return 0;
2162 }
2163
2164 if ((new_active) &&
2165 (old_active) &&
2166 (new_active->link == BOND_LINK_UP) &&
2167 IS_UP(new_active->dev)) {
2168 write_lock_bh(&bond->curr_slave_lock);
2169 bond_change_active_slave(bond, new_active);
2170 write_unlock_bh(&bond->curr_slave_lock);
2171 } else
2172 res = -EINVAL;
2173
2174 read_unlock(&bond->lock);
2175
2176 return res;
2177 }
2178
2179 static int bond_info_query(struct net_device *bond_dev, struct ifbond *info)
2180 {
2181 struct bonding *bond = netdev_priv(bond_dev);
2182
2183 info->bond_mode = bond->params.mode;
2184 info->miimon = bond->params.miimon;
2185
2186 read_lock(&bond->lock);
2187 info->num_slaves = bond->slave_cnt;
2188 read_unlock(&bond->lock);
2189
2190 return 0;
2191 }
2192
2193 static int bond_slave_info_query(struct net_device *bond_dev, struct ifslave *info)
2194 {
2195 struct bonding *bond = netdev_priv(bond_dev);
2196 struct slave *slave;
2197 int i, res = -ENODEV;
2198
2199 read_lock(&bond->lock);
2200
2201 bond_for_each_slave(bond, slave, i) {
2202 if (i == (int)info->slave_id) {
2203 res = 0;
2204 strcpy(info->slave_name, slave->dev->name);
2205 info->link = slave->link;
2206 info->state = slave->state;
2207 info->link_failure_count = slave->link_failure_count;
2208 break;
2209 }
2210 }
2211
2212 read_unlock(&bond->lock);
2213
2214 return res;
2215 }
2216
2217 /*-------------------------------- Monitoring -------------------------------*/
2218
2219
2220 static int bond_miimon_inspect(struct bonding *bond)
2221 {
2222 struct slave *slave;
2223 int i, link_state, commit = 0;
2224 bool ignore_updelay;
2225
2226 ignore_updelay = !bond->curr_active_slave ? true : false;
2227
2228 bond_for_each_slave(bond, slave, i) {
2229 slave->new_link = BOND_LINK_NOCHANGE;
2230
2231 link_state = bond_check_dev_link(bond, slave->dev, 0);
2232
2233 switch (slave->link) {
2234 case BOND_LINK_UP:
2235 if (link_state)
2236 continue;
2237
2238 slave->link = BOND_LINK_FAIL;
2239 slave->delay = bond->params.downdelay;
2240 if (slave->delay) {
2241 pr_info("%s: link status down for %sinterface %s, disabling it in %d ms.\n",
2242 bond->dev->name,
2243 (bond->params.mode ==
2244 BOND_MODE_ACTIVEBACKUP) ?
2245 ((slave->state == BOND_STATE_ACTIVE) ?
2246 "active " : "backup ") : "",
2247 slave->dev->name,
2248 bond->params.downdelay * bond->params.miimon);
2249 }
2250 /*FALLTHRU*/
2251 case BOND_LINK_FAIL:
2252 if (link_state) {
2253 /*
2254 * recovered before downdelay expired
2255 */
2256 slave->link = BOND_LINK_UP;
2257 slave->jiffies = jiffies;
2258 pr_info("%s: link status up again after %d ms for interface %s.\n",
2259 bond->dev->name,
2260 (bond->params.downdelay - slave->delay) *
2261 bond->params.miimon,
2262 slave->dev->name);
2263 continue;
2264 }
2265
2266 if (slave->delay <= 0) {
2267 slave->new_link = BOND_LINK_DOWN;
2268 commit++;
2269 continue;
2270 }
2271
2272 slave->delay--;
2273 break;
2274
2275 case BOND_LINK_DOWN:
2276 if (!link_state)
2277 continue;
2278
2279 slave->link = BOND_LINK_BACK;
2280 slave->delay = bond->params.updelay;
2281
2282 if (slave->delay) {
2283 pr_info("%s: link status up for interface %s, enabling it in %d ms.\n",
2284 bond->dev->name, slave->dev->name,
2285 ignore_updelay ? 0 :
2286 bond->params.updelay *
2287 bond->params.miimon);
2288 }
2289 /*FALLTHRU*/
2290 case BOND_LINK_BACK:
2291 if (!link_state) {
2292 slave->link = BOND_LINK_DOWN;
2293 pr_info("%s: link status down again after %d ms for interface %s.\n",
2294 bond->dev->name,
2295 (bond->params.updelay - slave->delay) *
2296 bond->params.miimon,
2297 slave->dev->name);
2298
2299 continue;
2300 }
2301
2302 if (ignore_updelay)
2303 slave->delay = 0;
2304
2305 if (slave->delay <= 0) {
2306 slave->new_link = BOND_LINK_UP;
2307 commit++;
2308 ignore_updelay = false;
2309 continue;
2310 }
2311
2312 slave->delay--;
2313 break;
2314 }
2315 }
2316
2317 return commit;
2318 }
2319
2320 static void bond_miimon_commit(struct bonding *bond)
2321 {
2322 struct slave *slave;
2323 int i;
2324
2325 bond_for_each_slave(bond, slave, i) {
2326 switch (slave->new_link) {
2327 case BOND_LINK_NOCHANGE:
2328 continue;
2329
2330 case BOND_LINK_UP:
2331 slave->link = BOND_LINK_UP;
2332 slave->jiffies = jiffies;
2333
2334 if (bond->params.mode == BOND_MODE_8023AD) {
2335 /* prevent it from being the active one */
2336 slave->state = BOND_STATE_BACKUP;
2337 } else if (bond->params.mode != BOND_MODE_ACTIVEBACKUP) {
2338 /* make it immediately active */
2339 slave->state = BOND_STATE_ACTIVE;
2340 } else if (slave != bond->primary_slave) {
2341 /* prevent it from being the active one */
2342 slave->state = BOND_STATE_BACKUP;
2343 }
2344
2345 pr_info("%s: link status definitely up for interface %s.\n",
2346 bond->dev->name, slave->dev->name);
2347
2348 /* notify ad that the link status has changed */
2349 if (bond->params.mode == BOND_MODE_8023AD)
2350 bond_3ad_handle_link_change(slave, BOND_LINK_UP);
2351
2352 if (bond_is_lb(bond))
2353 bond_alb_handle_link_change(bond, slave,
2354 BOND_LINK_UP);
2355
2356 if (!bond->curr_active_slave ||
2357 (slave == bond->primary_slave))
2358 goto do_failover;
2359
2360 continue;
2361
2362 case BOND_LINK_DOWN:
2363 if (slave->link_failure_count < UINT_MAX)
2364 slave->link_failure_count++;
2365
2366 slave->link = BOND_LINK_DOWN;
2367
2368 if (bond->params.mode == BOND_MODE_ACTIVEBACKUP ||
2369 bond->params.mode == BOND_MODE_8023AD)
2370 bond_set_slave_inactive_flags(slave);
2371
2372 pr_info("%s: link status definitely down for interface %s, disabling it\n",
2373 bond->dev->name, slave->dev->name);
2374
2375 if (bond->params.mode == BOND_MODE_8023AD)
2376 bond_3ad_handle_link_change(slave,
2377 BOND_LINK_DOWN);
2378
2379 if (bond_is_lb(bond))
2380 bond_alb_handle_link_change(bond, slave,
2381 BOND_LINK_DOWN);
2382
2383 if (slave == bond->curr_active_slave)
2384 goto do_failover;
2385
2386 continue;
2387
2388 default:
2389 pr_err("%s: invalid new link %d on slave %s\n",
2390 bond->dev->name, slave->new_link,
2391 slave->dev->name);
2392 slave->new_link = BOND_LINK_NOCHANGE;
2393
2394 continue;
2395 }
2396
2397 do_failover:
2398 ASSERT_RTNL();
2399 write_lock_bh(&bond->curr_slave_lock);
2400 bond_select_active_slave(bond);
2401 write_unlock_bh(&bond->curr_slave_lock);
2402 }
2403
2404 bond_set_carrier(bond);
2405 }
2406
2407 /*
2408 * bond_mii_monitor
2409 *
2410 * Really a wrapper that splits the mii monitor into two phases: an
2411 * inspection, then (if inspection indicates something needs to be done)
2412 * an acquisition of appropriate locks followed by a commit phase to
2413 * implement whatever link state changes are indicated.
2414 */
2415 void bond_mii_monitor(struct work_struct *work)
2416 {
2417 struct bonding *bond = container_of(work, struct bonding,
2418 mii_work.work);
2419
2420 read_lock(&bond->lock);
2421 if (bond->kill_timers)
2422 goto out;
2423
2424 if (bond->slave_cnt == 0)
2425 goto re_arm;
2426
2427 if (bond->send_grat_arp) {
2428 read_lock(&bond->curr_slave_lock);
2429 bond_send_gratuitous_arp(bond);
2430 read_unlock(&bond->curr_slave_lock);
2431 }
2432
2433 if (bond->send_unsol_na) {
2434 read_lock(&bond->curr_slave_lock);
2435 bond_send_unsolicited_na(bond);
2436 read_unlock(&bond->curr_slave_lock);
2437 }
2438
2439 if (bond_miimon_inspect(bond)) {
2440 read_unlock(&bond->lock);
2441 rtnl_lock();
2442 read_lock(&bond->lock);
2443
2444 bond_miimon_commit(bond);
2445
2446 read_unlock(&bond->lock);
2447 rtnl_unlock(); /* might sleep, hold no other locks */
2448 read_lock(&bond->lock);
2449 }
2450
2451 re_arm:
2452 if (bond->params.miimon)
2453 queue_delayed_work(bond->wq, &bond->mii_work,
2454 msecs_to_jiffies(bond->params.miimon));
2455 out:
2456 read_unlock(&bond->lock);
2457 }
2458
2459 static __be32 bond_glean_dev_ip(struct net_device *dev)
2460 {
2461 struct in_device *idev;
2462 struct in_ifaddr *ifa;
2463 __be32 addr = 0;
2464
2465 if (!dev)
2466 return 0;
2467
2468 rcu_read_lock();
2469 idev = __in_dev_get_rcu(dev);
2470 if (!idev)
2471 goto out;
2472
2473 ifa = idev->ifa_list;
2474 if (!ifa)
2475 goto out;
2476
2477 addr = ifa->ifa_local;
2478 out:
2479 rcu_read_unlock();
2480 return addr;
2481 }
2482
2483 static int bond_has_this_ip(struct bonding *bond, __be32 ip)
2484 {
2485 struct vlan_entry *vlan;
2486
2487 if (ip == bond->master_ip)
2488 return 1;
2489
2490 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
2491 if (ip == vlan->vlan_ip)
2492 return 1;
2493 }
2494
2495 return 0;
2496 }
2497
2498 /*
2499 * We go to the (large) trouble of VLAN tagging ARP frames because
2500 * switches in VLAN mode (especially if ports are configured as
2501 * "native" to a VLAN) might not pass non-tagged frames.
2502 */
2503 static void bond_arp_send(struct net_device *slave_dev, int arp_op, __be32 dest_ip, __be32 src_ip, unsigned short vlan_id)
2504 {
2505 struct sk_buff *skb;
2506
2507 pr_debug("arp %d on slave %s: dst %x src %x vid %d\n", arp_op,
2508 slave_dev->name, dest_ip, src_ip, vlan_id);
2509
2510 skb = arp_create(arp_op, ETH_P_ARP, dest_ip, slave_dev, src_ip,
2511 NULL, slave_dev->dev_addr, NULL);
2512
2513 if (!skb) {
2514 pr_err("ARP packet allocation failed\n");
2515 return;
2516 }
2517 if (vlan_id) {
2518 skb = vlan_put_tag(skb, vlan_id);
2519 if (!skb) {
2520 pr_err("failed to insert VLAN tag\n");
2521 return;
2522 }
2523 }
2524 arp_xmit(skb);
2525 }
2526
2527
2528 static void bond_arp_send_all(struct bonding *bond, struct slave *slave)
2529 {
2530 int i, vlan_id, rv;
2531 __be32 *targets = bond->params.arp_targets;
2532 struct vlan_entry *vlan;
2533 struct net_device *vlan_dev;
2534 struct flowi fl;
2535 struct rtable *rt;
2536
2537 for (i = 0; (i < BOND_MAX_ARP_TARGETS); i++) {
2538 if (!targets[i])
2539 break;
2540 pr_debug("basa: target %x\n", targets[i]);
2541 if (list_empty(&bond->vlan_list)) {
2542 pr_debug("basa: empty vlan: arp_send\n");
2543 bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2544 bond->master_ip, 0);
2545 continue;
2546 }
2547
2548 /*
2549 * If VLANs are configured, we do a route lookup to
2550 * determine which VLAN interface would be used, so we
2551 * can tag the ARP with the proper VLAN tag.
2552 */
2553 memset(&fl, 0, sizeof(fl));
2554 fl.fl4_dst = targets[i];
2555 fl.fl4_tos = RTO_ONLINK;
2556
2557 rv = ip_route_output_key(dev_net(bond->dev), &rt, &fl);
2558 if (rv) {
2559 if (net_ratelimit()) {
2560 pr_warning("%s: no route to arp_ip_target %pI4\n",
2561 bond->dev->name, &fl.fl4_dst);
2562 }
2563 continue;
2564 }
2565
2566 /*
2567 * This target is not on a VLAN
2568 */
2569 if (rt->u.dst.dev == bond->dev) {
2570 ip_rt_put(rt);
2571 pr_debug("basa: rtdev == bond->dev: arp_send\n");
2572 bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2573 bond->master_ip, 0);
2574 continue;
2575 }
2576
2577 vlan_id = 0;
2578 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
2579 vlan_dev = vlan_group_get_device(bond->vlgrp, vlan->vlan_id);
2580 if (vlan_dev == rt->u.dst.dev) {
2581 vlan_id = vlan->vlan_id;
2582 pr_debug("basa: vlan match on %s %d\n",
2583 vlan_dev->name, vlan_id);
2584 break;
2585 }
2586 }
2587
2588 if (vlan_id) {
2589 ip_rt_put(rt);
2590 bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2591 vlan->vlan_ip, vlan_id);
2592 continue;
2593 }
2594
2595 if (net_ratelimit()) {
2596 pr_warning("%s: no path to arp_ip_target %pI4 via rt.dev %s\n",
2597 bond->dev->name, &fl.fl4_dst,
2598 rt->u.dst.dev ? rt->u.dst.dev->name : "NULL");
2599 }
2600 ip_rt_put(rt);
2601 }
2602 }
2603
2604 /*
2605 * Kick out a gratuitous ARP for an IP on the bonding master plus one
2606 * for each VLAN above us.
2607 *
2608 * Caller must hold curr_slave_lock for read or better
2609 */
2610 static void bond_send_gratuitous_arp(struct bonding *bond)
2611 {
2612 struct slave *slave = bond->curr_active_slave;
2613 struct vlan_entry *vlan;
2614 struct net_device *vlan_dev;
2615
2616 pr_debug("bond_send_grat_arp: bond %s slave %s\n",
2617 bond->dev->name, slave ? slave->dev->name : "NULL");
2618
2619 if (!slave || !bond->send_grat_arp ||
2620 test_bit(__LINK_STATE_LINKWATCH_PENDING, &slave->dev->state))
2621 return;
2622
2623 bond->send_grat_arp--;
2624
2625 if (bond->master_ip) {
2626 bond_arp_send(slave->dev, ARPOP_REPLY, bond->master_ip,
2627 bond->master_ip, 0);
2628 }
2629
2630 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
2631 vlan_dev = vlan_group_get_device(bond->vlgrp, vlan->vlan_id);
2632 if (vlan->vlan_ip) {
2633 bond_arp_send(slave->dev, ARPOP_REPLY, vlan->vlan_ip,
2634 vlan->vlan_ip, vlan->vlan_id);
2635 }
2636 }
2637 }
2638
2639 static void bond_validate_arp(struct bonding *bond, struct slave *slave, __be32 sip, __be32 tip)
2640 {
2641 int i;
2642 __be32 *targets = bond->params.arp_targets;
2643
2644 for (i = 0; (i < BOND_MAX_ARP_TARGETS) && targets[i]; i++) {
2645 pr_debug("bva: sip %pI4 tip %pI4 t[%d] %pI4 bhti(tip) %d\n",
2646 &sip, &tip, i, &targets[i],
2647 bond_has_this_ip(bond, tip));
2648 if (sip == targets[i]) {
2649 if (bond_has_this_ip(bond, tip))
2650 slave->last_arp_rx = jiffies;
2651 return;
2652 }
2653 }
2654 }
2655
2656 static int bond_arp_rcv(struct sk_buff *skb, struct net_device *dev, struct packet_type *pt, struct net_device *orig_dev)
2657 {
2658 struct arphdr *arp;
2659 struct slave *slave;
2660 struct bonding *bond;
2661 unsigned char *arp_ptr;
2662 __be32 sip, tip;
2663
2664 if (dev->priv_flags & IFF_802_1Q_VLAN) {
2665 /*
2666 * When using VLANS and bonding, dev and oriv_dev may be
2667 * incorrect if the physical interface supports VLAN
2668 * acceleration. With this change ARP validation now
2669 * works for hosts only reachable on the VLAN interface.
2670 */
2671 dev = vlan_dev_real_dev(dev);
2672 orig_dev = dev_get_by_index_rcu(dev_net(skb->dev),skb->skb_iif);
2673 }
2674
2675 if (!(dev->priv_flags & IFF_BONDING) || !(dev->flags & IFF_MASTER))
2676 goto out;
2677
2678 bond = netdev_priv(dev);
2679 read_lock(&bond->lock);
2680
2681 pr_debug("bond_arp_rcv: bond %s skb->dev %s orig_dev %s\n",
2682 bond->dev->name, skb->dev ? skb->dev->name : "NULL",
2683 orig_dev ? orig_dev->name : "NULL");
2684
2685 slave = bond_get_slave_by_dev(bond, orig_dev);
2686 if (!slave || !slave_do_arp_validate(bond, slave))
2687 goto out_unlock;
2688
2689 if (!pskb_may_pull(skb, arp_hdr_len(dev)))
2690 goto out_unlock;
2691
2692 arp = arp_hdr(skb);
2693 if (arp->ar_hln != dev->addr_len ||
2694 skb->pkt_type == PACKET_OTHERHOST ||
2695 skb->pkt_type == PACKET_LOOPBACK ||
2696 arp->ar_hrd != htons(ARPHRD_ETHER) ||
2697 arp->ar_pro != htons(ETH_P_IP) ||
2698 arp->ar_pln != 4)
2699 goto out_unlock;
2700
2701 arp_ptr = (unsigned char *)(arp + 1);
2702 arp_ptr += dev->addr_len;
2703 memcpy(&sip, arp_ptr, 4);
2704 arp_ptr += 4 + dev->addr_len;
2705 memcpy(&tip, arp_ptr, 4);
2706
2707 pr_debug("bond_arp_rcv: %s %s/%d av %d sv %d sip %pI4 tip %pI4\n",
2708 bond->dev->name, slave->dev->name, slave->state,
2709 bond->params.arp_validate, slave_do_arp_validate(bond, slave),
2710 &sip, &tip);
2711
2712 /*
2713 * Backup slaves won't see the ARP reply, but do come through
2714 * here for each ARP probe (so we swap the sip/tip to validate
2715 * the probe). In a "redundant switch, common router" type of
2716 * configuration, the ARP probe will (hopefully) travel from
2717 * the active, through one switch, the router, then the other
2718 * switch before reaching the backup.
2719 */
2720 if (slave->state == BOND_STATE_ACTIVE)
2721 bond_validate_arp(bond, slave, sip, tip);
2722 else
2723 bond_validate_arp(bond, slave, tip, sip);
2724
2725 out_unlock:
2726 read_unlock(&bond->lock);
2727 out:
2728 dev_kfree_skb(skb);
2729 return NET_RX_SUCCESS;
2730 }
2731
2732 /*
2733 * this function is called regularly to monitor each slave's link
2734 * ensuring that traffic is being sent and received when arp monitoring
2735 * is used in load-balancing mode. if the adapter has been dormant, then an
2736 * arp is transmitted to generate traffic. see activebackup_arp_monitor for
2737 * arp monitoring in active backup mode.
2738 */
2739 void bond_loadbalance_arp_mon(struct work_struct *work)
2740 {
2741 struct bonding *bond = container_of(work, struct bonding,
2742 arp_work.work);
2743 struct slave *slave, *oldcurrent;
2744 int do_failover = 0;
2745 int delta_in_ticks;
2746 int i;
2747
2748 read_lock(&bond->lock);
2749
2750 delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
2751
2752 if (bond->kill_timers)
2753 goto out;
2754
2755 if (bond->slave_cnt == 0)
2756 goto re_arm;
2757
2758 read_lock(&bond->curr_slave_lock);
2759 oldcurrent = bond->curr_active_slave;
2760 read_unlock(&bond->curr_slave_lock);
2761
2762 /* see if any of the previous devices are up now (i.e. they have
2763 * xmt and rcv traffic). the curr_active_slave does not come into
2764 * the picture unless it is null. also, slave->jiffies is not needed
2765 * here because we send an arp on each slave and give a slave as
2766 * long as it needs to get the tx/rx within the delta.
2767 * TODO: what about up/down delay in arp mode? it wasn't here before
2768 * so it can wait
2769 */
2770 bond_for_each_slave(bond, slave, i) {
2771 if (slave->link != BOND_LINK_UP) {
2772 if (time_before_eq(jiffies, dev_trans_start(slave->dev) + delta_in_ticks) &&
2773 time_before_eq(jiffies, slave->dev->last_rx + delta_in_ticks)) {
2774
2775 slave->link = BOND_LINK_UP;
2776 slave->state = BOND_STATE_ACTIVE;
2777
2778 /* primary_slave has no meaning in round-robin
2779 * mode. the window of a slave being up and
2780 * curr_active_slave being null after enslaving
2781 * is closed.
2782 */
2783 if (!oldcurrent) {
2784 pr_info("%s: link status definitely up for interface %s, ",
2785 bond->dev->name,
2786 slave->dev->name);
2787 do_failover = 1;
2788 } else {
2789 pr_info("%s: interface %s is now up\n",
2790 bond->dev->name,
2791 slave->dev->name);
2792 }
2793 }
2794 } else {
2795 /* slave->link == BOND_LINK_UP */
2796
2797 /* not all switches will respond to an arp request
2798 * when the source ip is 0, so don't take the link down
2799 * if we don't know our ip yet
2800 */
2801 if (time_after_eq(jiffies, dev_trans_start(slave->dev) + 2*delta_in_ticks) ||
2802 (time_after_eq(jiffies, slave->dev->last_rx + 2*delta_in_ticks))) {
2803
2804 slave->link = BOND_LINK_DOWN;
2805 slave->state = BOND_STATE_BACKUP;
2806
2807 if (slave->link_failure_count < UINT_MAX)
2808 slave->link_failure_count++;
2809
2810 pr_info("%s: interface %s is now down.\n",
2811 bond->dev->name,
2812 slave->dev->name);
2813
2814 if (slave == oldcurrent)
2815 do_failover = 1;
2816 }
2817 }
2818
2819 /* note: if switch is in round-robin mode, all links
2820 * must tx arp to ensure all links rx an arp - otherwise
2821 * links may oscillate or not come up at all; if switch is
2822 * in something like xor mode, there is nothing we can
2823 * do - all replies will be rx'ed on same link causing slaves
2824 * to be unstable during low/no traffic periods
2825 */
2826 if (IS_UP(slave->dev))
2827 bond_arp_send_all(bond, slave);
2828 }
2829
2830 if (do_failover) {
2831 write_lock_bh(&bond->curr_slave_lock);
2832
2833 bond_select_active_slave(bond);
2834
2835 write_unlock_bh(&bond->curr_slave_lock);
2836 }
2837
2838 re_arm:
2839 if (bond->params.arp_interval)
2840 queue_delayed_work(bond->wq, &bond->arp_work, delta_in_ticks);
2841 out:
2842 read_unlock(&bond->lock);
2843 }
2844
2845 /*
2846 * Called to inspect slaves for active-backup mode ARP monitor link state
2847 * changes. Sets new_link in slaves to specify what action should take
2848 * place for the slave. Returns 0 if no changes are found, >0 if changes
2849 * to link states must be committed.
2850 *
2851 * Called with bond->lock held for read.
2852 */
2853 static int bond_ab_arp_inspect(struct bonding *bond, int delta_in_ticks)
2854 {
2855 struct slave *slave;
2856 int i, commit = 0;
2857
2858 bond_for_each_slave(bond, slave, i) {
2859 slave->new_link = BOND_LINK_NOCHANGE;
2860
2861 if (slave->link != BOND_LINK_UP) {
2862 if (time_before_eq(jiffies, slave_last_rx(bond, slave) +
2863 delta_in_ticks)) {
2864 slave->new_link = BOND_LINK_UP;
2865 commit++;
2866 }
2867
2868 continue;
2869 }
2870
2871 /*
2872 * Give slaves 2*delta after being enslaved or made
2873 * active. This avoids bouncing, as the last receive
2874 * times need a full ARP monitor cycle to be updated.
2875 */
2876 if (!time_after_eq(jiffies, slave->jiffies +
2877 2 * delta_in_ticks))
2878 continue;
2879
2880 /*
2881 * Backup slave is down if:
2882 * - No current_arp_slave AND
2883 * - more than 3*delta since last receive AND
2884 * - the bond has an IP address
2885 *
2886 * Note: a non-null current_arp_slave indicates
2887 * the curr_active_slave went down and we are
2888 * searching for a new one; under this condition
2889 * we only take the curr_active_slave down - this
2890 * gives each slave a chance to tx/rx traffic
2891 * before being taken out
2892 */
2893 if (slave->state == BOND_STATE_BACKUP &&
2894 !bond->current_arp_slave &&
2895 time_after(jiffies, slave_last_rx(bond, slave) +
2896 3 * delta_in_ticks)) {
2897 slave->new_link = BOND_LINK_DOWN;
2898 commit++;
2899 }
2900
2901 /*
2902 * Active slave is down if:
2903 * - more than 2*delta since transmitting OR
2904 * - (more than 2*delta since receive AND
2905 * the bond has an IP address)
2906 */
2907 if ((slave->state == BOND_STATE_ACTIVE) &&
2908 (time_after_eq(jiffies, dev_trans_start(slave->dev) +
2909 2 * delta_in_ticks) ||
2910 (time_after_eq(jiffies, slave_last_rx(bond, slave)
2911 + 2 * delta_in_ticks)))) {
2912 slave->new_link = BOND_LINK_DOWN;
2913 commit++;
2914 }
2915 }
2916
2917 return commit;
2918 }
2919
2920 /*
2921 * Called to commit link state changes noted by inspection step of
2922 * active-backup mode ARP monitor.
2923 *
2924 * Called with RTNL and bond->lock for read.
2925 */
2926 static void bond_ab_arp_commit(struct bonding *bond, int delta_in_ticks)
2927 {
2928 struct slave *slave;
2929 int i;
2930
2931 bond_for_each_slave(bond, slave, i) {
2932 switch (slave->new_link) {
2933 case BOND_LINK_NOCHANGE:
2934 continue;
2935
2936 case BOND_LINK_UP:
2937 if ((!bond->curr_active_slave &&
2938 time_before_eq(jiffies,
2939 dev_trans_start(slave->dev) +
2940 delta_in_ticks)) ||
2941 bond->curr_active_slave != slave) {
2942 slave->link = BOND_LINK_UP;
2943 bond->current_arp_slave = NULL;
2944
2945 pr_info("%s: link status definitely up for interface %s.\n",
2946 bond->dev->name, slave->dev->name);
2947
2948 if (!bond->curr_active_slave ||
2949 (slave == bond->primary_slave))
2950 goto do_failover;
2951
2952 }
2953
2954 continue;
2955
2956 case BOND_LINK_DOWN:
2957 if (slave->link_failure_count < UINT_MAX)
2958 slave->link_failure_count++;
2959
2960 slave->link = BOND_LINK_DOWN;
2961 bond_set_slave_inactive_flags(slave);
2962
2963 pr_info("%s: link status definitely down for interface %s, disabling it\n",
2964 bond->dev->name, slave->dev->name);
2965
2966 if (slave == bond->curr_active_slave) {
2967 bond->current_arp_slave = NULL;
2968 goto do_failover;
2969 }
2970
2971 continue;
2972
2973 default:
2974 pr_err("%s: impossible: new_link %d on slave %s\n",
2975 bond->dev->name, slave->new_link,
2976 slave->dev->name);
2977 continue;
2978 }
2979
2980 do_failover:
2981 ASSERT_RTNL();
2982 write_lock_bh(&bond->curr_slave_lock);
2983 bond_select_active_slave(bond);
2984 write_unlock_bh(&bond->curr_slave_lock);
2985 }
2986
2987 bond_set_carrier(bond);
2988 }
2989
2990 /*
2991 * Send ARP probes for active-backup mode ARP monitor.
2992 *
2993 * Called with bond->lock held for read.
2994 */
2995 static void bond_ab_arp_probe(struct bonding *bond)
2996 {
2997 struct slave *slave;
2998 int i;
2999
3000 read_lock(&bond->curr_slave_lock);
3001
3002 if (bond->current_arp_slave && bond->curr_active_slave)
3003 pr_info("PROBE: c_arp %s && cas %s BAD\n",
3004 bond->current_arp_slave->dev->name,
3005 bond->curr_active_slave->dev->name);
3006
3007 if (bond->curr_active_slave) {
3008 bond_arp_send_all(bond, bond->curr_active_slave);
3009 read_unlock(&bond->curr_slave_lock);
3010 return;
3011 }
3012
3013 read_unlock(&bond->curr_slave_lock);
3014
3015 /* if we don't have a curr_active_slave, search for the next available
3016 * backup slave from the current_arp_slave and make it the candidate
3017 * for becoming the curr_active_slave
3018 */
3019
3020 if (!bond->current_arp_slave) {
3021 bond->current_arp_slave = bond->first_slave;
3022 if (!bond->current_arp_slave)
3023 return;
3024 }
3025
3026 bond_set_slave_inactive_flags(bond->current_arp_slave);
3027
3028 /* search for next candidate */
3029 bond_for_each_slave_from(bond, slave, i, bond->current_arp_slave->next) {
3030 if (IS_UP(slave->dev)) {
3031 slave->link = BOND_LINK_BACK;
3032 bond_set_slave_active_flags(slave);
3033 bond_arp_send_all(bond, slave);
3034 slave->jiffies = jiffies;
3035 bond->current_arp_slave = slave;
3036 break;
3037 }
3038
3039 /* if the link state is up at this point, we
3040 * mark it down - this can happen if we have
3041 * simultaneous link failures and
3042 * reselect_active_interface doesn't make this
3043 * one the current slave so it is still marked
3044 * up when it is actually down
3045 */
3046 if (slave->link == BOND_LINK_UP) {
3047 slave->link = BOND_LINK_DOWN;
3048 if (slave->link_failure_count < UINT_MAX)
3049 slave->link_failure_count++;
3050
3051 bond_set_slave_inactive_flags(slave);
3052
3053 pr_info("%s: backup interface %s is now down.\n",
3054 bond->dev->name, slave->dev->name);
3055 }
3056 }
3057 }
3058
3059 void bond_activebackup_arp_mon(struct work_struct *work)
3060 {
3061 struct bonding *bond = container_of(work, struct bonding,
3062 arp_work.work);
3063 int delta_in_ticks;
3064
3065 read_lock(&bond->lock);
3066
3067 if (bond->kill_timers)
3068 goto out;
3069
3070 delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
3071
3072 if (bond->slave_cnt == 0)
3073 goto re_arm;
3074
3075 if (bond->send_grat_arp) {
3076 read_lock(&bond->curr_slave_lock);
3077 bond_send_gratuitous_arp(bond);
3078 read_unlock(&bond->curr_slave_lock);
3079 }
3080
3081 if (bond->send_unsol_na) {
3082 read_lock(&bond->curr_slave_lock);
3083 bond_send_unsolicited_na(bond);
3084 read_unlock(&bond->curr_slave_lock);
3085 }
3086
3087 if (bond_ab_arp_inspect(bond, delta_in_ticks)) {
3088 read_unlock(&bond->lock);
3089 rtnl_lock();
3090 read_lock(&bond->lock);
3091
3092 bond_ab_arp_commit(bond, delta_in_ticks);
3093
3094 read_unlock(&bond->lock);
3095 rtnl_unlock();
3096 read_lock(&bond->lock);
3097 }
3098
3099 bond_ab_arp_probe(bond);
3100
3101 re_arm:
3102 if (bond->params.arp_interval)
3103 queue_delayed_work(bond->wq, &bond->arp_work, delta_in_ticks);
3104 out:
3105 read_unlock(&bond->lock);
3106 }
3107
3108 /*------------------------------ proc/seq_file-------------------------------*/
3109
3110 #ifdef CONFIG_PROC_FS
3111
3112 static void *bond_info_seq_start(struct seq_file *seq, loff_t *pos)
3113 __acquires(&dev_base_lock)
3114 __acquires(&bond->lock)
3115 {
3116 struct bonding *bond = seq->private;
3117 loff_t off = 0;
3118 struct slave *slave;
3119 int i;
3120
3121 /* make sure the bond won't be taken away */
3122 read_lock(&dev_base_lock);
3123 read_lock(&bond->lock);
3124
3125 if (*pos == 0)
3126 return SEQ_START_TOKEN;
3127
3128 bond_for_each_slave(bond, slave, i) {
3129 if (++off == *pos)
3130 return slave;
3131 }
3132
3133 return NULL;
3134 }
3135
3136 static void *bond_info_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3137 {
3138 struct bonding *bond = seq->private;
3139 struct slave *slave = v;
3140
3141 ++*pos;
3142 if (v == SEQ_START_TOKEN)
3143 return bond->first_slave;
3144
3145 slave = slave->next;
3146
3147 return (slave == bond->first_slave) ? NULL : slave;
3148 }
3149
3150 static void bond_info_seq_stop(struct seq_file *seq, void *v)
3151 __releases(&bond->lock)
3152 __releases(&dev_base_lock)
3153 {
3154 struct bonding *bond = seq->private;
3155
3156 read_unlock(&bond->lock);
3157 read_unlock(&dev_base_lock);
3158 }
3159
3160 static void bond_info_show_master(struct seq_file *seq)
3161 {
3162 struct bonding *bond = seq->private;
3163 struct slave *curr;
3164 int i;
3165
3166 read_lock(&bond->curr_slave_lock);
3167 curr = bond->curr_active_slave;
3168 read_unlock(&bond->curr_slave_lock);
3169
3170 seq_printf(seq, "Bonding Mode: %s",
3171 bond_mode_name(bond->params.mode));
3172
3173 if (bond->params.mode == BOND_MODE_ACTIVEBACKUP &&
3174 bond->params.fail_over_mac)
3175 seq_printf(seq, " (fail_over_mac %s)",
3176 fail_over_mac_tbl[bond->params.fail_over_mac].modename);
3177
3178 seq_printf(seq, "\n");
3179
3180 if (bond->params.mode == BOND_MODE_XOR ||
3181 bond->params.mode == BOND_MODE_8023AD) {
3182 seq_printf(seq, "Transmit Hash Policy: %s (%d)\n",
3183 xmit_hashtype_tbl[bond->params.xmit_policy].modename,
3184 bond->params.xmit_policy);
3185 }
3186
3187 if (USES_PRIMARY(bond->params.mode)) {
3188 seq_printf(seq, "Primary Slave: %s",
3189 (bond->primary_slave) ?
3190 bond->primary_slave->dev->name : "None");
3191 if (bond->primary_slave)
3192 seq_printf(seq, " (primary_reselect %s)",
3193 pri_reselect_tbl[bond->params.primary_reselect].modename);
3194
3195 seq_printf(seq, "\nCurrently Active Slave: %s\n",
3196 (curr) ? curr->dev->name : "None");
3197 }
3198
3199 seq_printf(seq, "MII Status: %s\n", netif_carrier_ok(bond->dev) ?
3200 "up" : "down");
3201 seq_printf(seq, "MII Polling Interval (ms): %d\n", bond->params.miimon);
3202 seq_printf(seq, "Up Delay (ms): %d\n",
3203 bond->params.updelay * bond->params.miimon);
3204 seq_printf(seq, "Down Delay (ms): %d\n",
3205 bond->params.downdelay * bond->params.miimon);
3206
3207
3208 /* ARP information */
3209 if (bond->params.arp_interval > 0) {
3210 int printed = 0;
3211 seq_printf(seq, "ARP Polling Interval (ms): %d\n",
3212 bond->params.arp_interval);
3213
3214 seq_printf(seq, "ARP IP target/s (n.n.n.n form):");
3215
3216 for (i = 0; (i < BOND_MAX_ARP_TARGETS); i++) {
3217 if (!bond->params.arp_targets[i])
3218 break;
3219 if (printed)
3220 seq_printf(seq, ",");
3221 seq_printf(seq, " %pI4", &bond->params.arp_targets[i]);
3222 printed = 1;
3223 }
3224 seq_printf(seq, "\n");
3225 }
3226
3227 if (bond->params.mode == BOND_MODE_8023AD) {
3228 struct ad_info ad_info;
3229
3230 seq_puts(seq, "\n802.3ad info\n");
3231 seq_printf(seq, "LACP rate: %s\n",
3232 (bond->params.lacp_fast) ? "fast" : "slow");
3233 seq_printf(seq, "Aggregator selection policy (ad_select): %s\n",
3234 ad_select_tbl[bond->params.ad_select].modename);
3235
3236 if (bond_3ad_get_active_agg_info(bond, &ad_info)) {
3237 seq_printf(seq, "bond %s has no active aggregator\n",
3238 bond->dev->name);
3239 } else {
3240 seq_printf(seq, "Active Aggregator Info:\n");
3241
3242 seq_printf(seq, "\tAggregator ID: %d\n",
3243 ad_info.aggregator_id);
3244 seq_printf(seq, "\tNumber of ports: %d\n",
3245 ad_info.ports);
3246 seq_printf(seq, "\tActor Key: %d\n",
3247 ad_info.actor_key);
3248 seq_printf(seq, "\tPartner Key: %d\n",
3249 ad_info.partner_key);
3250 seq_printf(seq, "\tPartner Mac Address: %pM\n",
3251 ad_info.partner_system);
3252 }
3253 }
3254 }
3255
3256 static void bond_info_show_slave(struct seq_file *seq,
3257 const struct slave *slave)
3258 {
3259 struct bonding *bond = seq->private;
3260
3261 seq_printf(seq, "\nSlave Interface: %s\n", slave->dev->name);
3262 seq_printf(seq, "MII Status: %s\n",
3263 (slave->link == BOND_LINK_UP) ? "up" : "down");
3264 seq_printf(seq, "Link Failure Count: %u\n",
3265 slave->link_failure_count);
3266
3267 seq_printf(seq, "Permanent HW addr: %pM\n", slave->perm_hwaddr);
3268
3269 if (bond->params.mode == BOND_MODE_8023AD) {
3270 const struct aggregator *agg
3271 = SLAVE_AD_INFO(slave).port.aggregator;
3272
3273 if (agg)
3274 seq_printf(seq, "Aggregator ID: %d\n",
3275 agg->aggregator_identifier);
3276 else
3277 seq_puts(seq, "Aggregator ID: N/A\n");
3278 }
3279 }
3280
3281 static int bond_info_seq_show(struct seq_file *seq, void *v)
3282 {
3283 if (v == SEQ_START_TOKEN) {
3284 seq_printf(seq, "%s\n", version);
3285 bond_info_show_master(seq);
3286 } else
3287 bond_info_show_slave(seq, v);
3288
3289 return 0;
3290 }
3291
3292 static const struct seq_operations bond_info_seq_ops = {
3293 .start = bond_info_seq_start,
3294 .next = bond_info_seq_next,
3295 .stop = bond_info_seq_stop,
3296 .show = bond_info_seq_show,
3297 };
3298
3299 static int bond_info_open(struct inode *inode, struct file *file)
3300 {
3301 struct seq_file *seq;
3302 struct proc_dir_entry *proc;
3303 int res;
3304
3305 res = seq_open(file, &bond_info_seq_ops);
3306 if (!res) {
3307 /* recover the pointer buried in proc_dir_entry data */
3308 seq = file->private_data;
3309 proc = PDE(inode);
3310 seq->private = proc->data;
3311 }
3312
3313 return res;
3314 }
3315
3316 static const struct file_operations bond_info_fops = {
3317 .owner = THIS_MODULE,
3318 .open = bond_info_open,
3319 .read = seq_read,
3320 .llseek = seq_lseek,
3321 .release = seq_release,
3322 };
3323
3324 static void bond_create_proc_entry(struct bonding *bond)
3325 {
3326 struct net_device *bond_dev = bond->dev;
3327 struct bond_net *bn = net_generic(dev_net(bond_dev), bond_net_id);
3328
3329 if (bn->proc_dir) {
3330 bond->proc_entry = proc_create_data(bond_dev->name,
3331 S_IRUGO, bn->proc_dir,
3332 &bond_info_fops, bond);
3333 if (bond->proc_entry == NULL)
3334 pr_warning("Warning: Cannot create /proc/net/%s/%s\n",
3335 DRV_NAME, bond_dev->name);
3336 else
3337 memcpy(bond->proc_file_name, bond_dev->name, IFNAMSIZ);
3338 }
3339 }
3340
3341 static void bond_remove_proc_entry(struct bonding *bond)
3342 {
3343 struct net_device *bond_dev = bond->dev;
3344 struct bond_net *bn = net_generic(dev_net(bond_dev), bond_net_id);
3345
3346 if (bn->proc_dir && bond->proc_entry) {
3347 remove_proc_entry(bond->proc_file_name, bn->proc_dir);
3348 memset(bond->proc_file_name, 0, IFNAMSIZ);
3349 bond->proc_entry = NULL;
3350 }
3351 }
3352
3353 /* Create the bonding directory under /proc/net, if doesn't exist yet.
3354 * Caller must hold rtnl_lock.
3355 */
3356 static void __net_init bond_create_proc_dir(struct bond_net *bn)
3357 {
3358 if (!bn->proc_dir) {
3359 bn->proc_dir = proc_mkdir(DRV_NAME, bn->net->proc_net);
3360 if (!bn->proc_dir)
3361 pr_warning("Warning: cannot create /proc/net/%s\n",
3362 DRV_NAME);
3363 }
3364 }
3365
3366 /* Destroy the bonding directory under /proc/net, if empty.
3367 * Caller must hold rtnl_lock.
3368 */
3369 static void __net_exit bond_destroy_proc_dir(struct bond_net *bn)
3370 {
3371 if (bn->proc_dir) {
3372 remove_proc_entry(DRV_NAME, bn->net->proc_net);
3373 bn->proc_dir = NULL;
3374 }
3375 }
3376
3377 #else /* !CONFIG_PROC_FS */
3378
3379 static void bond_create_proc_entry(struct bonding *bond)
3380 {
3381 }
3382
3383 static void bond_remove_proc_entry(struct bonding *bond)
3384 {
3385 }
3386
3387 static inline void bond_create_proc_dir(struct bond_net *bn)
3388 {
3389 }
3390
3391 static inline void bond_destroy_proc_dir(struct bond_net *bn)
3392 {
3393 }
3394
3395 #endif /* CONFIG_PROC_FS */
3396
3397
3398 /*-------------------------- netdev event handling --------------------------*/
3399
3400 /*
3401 * Change device name
3402 */
3403 static int bond_event_changename(struct bonding *bond)
3404 {
3405 bond_remove_proc_entry(bond);
3406 bond_create_proc_entry(bond);
3407
3408 return NOTIFY_DONE;
3409 }
3410
3411 static int bond_master_netdev_event(unsigned long event,
3412 struct net_device *bond_dev)
3413 {
3414 struct bonding *event_bond = netdev_priv(bond_dev);
3415
3416 switch (event) {
3417 case NETDEV_CHANGENAME:
3418 return bond_event_changename(event_bond);
3419 default:
3420 break;
3421 }
3422
3423 return NOTIFY_DONE;
3424 }
3425
3426 static int bond_slave_netdev_event(unsigned long event,
3427 struct net_device *slave_dev)
3428 {
3429 struct net_device *bond_dev = slave_dev->master;
3430 struct bonding *bond = netdev_priv(bond_dev);
3431
3432 switch (event) {
3433 case NETDEV_UNREGISTER:
3434 if (bond_dev) {
3435 if (bond->setup_by_slave)
3436 bond_release_and_destroy(bond_dev, slave_dev);
3437 else
3438 bond_release(bond_dev, slave_dev);
3439 }
3440 break;
3441 case NETDEV_CHANGE:
3442 if (bond->params.mode == BOND_MODE_8023AD || bond_is_lb(bond)) {
3443 struct slave *slave;
3444
3445 slave = bond_get_slave_by_dev(bond, slave_dev);
3446 if (slave) {
3447 u16 old_speed = slave->speed;
3448 u16 old_duplex = slave->duplex;
3449
3450 bond_update_speed_duplex(slave);
3451
3452 if (bond_is_lb(bond))
3453 break;
3454
3455 if (old_speed != slave->speed)
3456 bond_3ad_adapter_speed_changed(slave);
3457 if (old_duplex != slave->duplex)
3458 bond_3ad_adapter_duplex_changed(slave);
3459 }
3460 }
3461
3462 break;
3463 case NETDEV_DOWN:
3464 /*
3465 * ... Or is it this?
3466 */
3467 break;
3468 case NETDEV_CHANGEMTU:
3469 /*
3470 * TODO: Should slaves be allowed to
3471 * independently alter their MTU? For
3472 * an active-backup bond, slaves need
3473 * not be the same type of device, so
3474 * MTUs may vary. For other modes,
3475 * slaves arguably should have the
3476 * same MTUs. To do this, we'd need to
3477 * take over the slave's change_mtu
3478 * function for the duration of their
3479 * servitude.
3480 */
3481 break;
3482 case NETDEV_CHANGENAME:
3483 /*
3484 * TODO: handle changing the primary's name
3485 */
3486 break;
3487 case NETDEV_FEAT_CHANGE:
3488 bond_compute_features(bond);
3489 break;
3490 default:
3491 break;
3492 }
3493
3494 return NOTIFY_DONE;
3495 }
3496
3497 /*
3498 * bond_netdev_event: handle netdev notifier chain events.
3499 *
3500 * This function receives events for the netdev chain. The caller (an
3501 * ioctl handler calling blocking_notifier_call_chain) holds the necessary
3502 * locks for us to safely manipulate the slave devices (RTNL lock,
3503 * dev_probe_lock).
3504 */
3505 static int bond_netdev_event(struct notifier_block *this,
3506 unsigned long event, void *ptr)
3507 {
3508 struct net_device *event_dev = (struct net_device *)ptr;
3509
3510 pr_debug("event_dev: %s, event: %lx\n",
3511 event_dev ? event_dev->name : "None",
3512 event);
3513
3514 if (!(event_dev->priv_flags & IFF_BONDING))
3515 return NOTIFY_DONE;
3516
3517 if (event_dev->flags & IFF_MASTER) {
3518 pr_debug("IFF_MASTER\n");
3519 return bond_master_netdev_event(event, event_dev);
3520 }
3521
3522 if (event_dev->flags & IFF_SLAVE) {
3523 pr_debug("IFF_SLAVE\n");
3524 return bond_slave_netdev_event(event, event_dev);
3525 }
3526
3527 return NOTIFY_DONE;
3528 }
3529
3530 /*
3531 * bond_inetaddr_event: handle inetaddr notifier chain events.
3532 *
3533 * We keep track of device IPs primarily to use as source addresses in
3534 * ARP monitor probes (rather than spewing out broadcasts all the time).
3535 *
3536 * We track one IP for the main device (if it has one), plus one per VLAN.
3537 */
3538 static int bond_inetaddr_event(struct notifier_block *this, unsigned long event, void *ptr)
3539 {
3540 struct in_ifaddr *ifa = ptr;
3541 struct net_device *vlan_dev, *event_dev = ifa->ifa_dev->dev;
3542 struct bond_net *bn = net_generic(dev_net(event_dev), bond_net_id);
3543 struct bonding *bond;
3544 struct vlan_entry *vlan;
3545
3546 list_for_each_entry(bond, &bn->dev_list, bond_list) {
3547 if (bond->dev == event_dev) {
3548 switch (event) {
3549 case NETDEV_UP:
3550 bond->master_ip = ifa->ifa_local;
3551 return NOTIFY_OK;
3552 case NETDEV_DOWN:
3553 bond->master_ip = bond_glean_dev_ip(bond->dev);
3554 return NOTIFY_OK;
3555 default:
3556 return NOTIFY_DONE;
3557 }
3558 }
3559
3560 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
3561 vlan_dev = vlan_group_get_device(bond->vlgrp, vlan->vlan_id);
3562 if (vlan_dev == event_dev) {
3563 switch (event) {
3564 case NETDEV_UP:
3565 vlan->vlan_ip = ifa->ifa_local;
3566 return NOTIFY_OK;
3567 case NETDEV_DOWN:
3568 vlan->vlan_ip =
3569 bond_glean_dev_ip(vlan_dev);
3570 return NOTIFY_OK;
3571 default:
3572 return NOTIFY_DONE;
3573 }
3574 }
3575 }
3576 }
3577 return NOTIFY_DONE;
3578 }
3579
3580 static struct notifier_block bond_netdev_notifier = {
3581 .notifier_call = bond_netdev_event,
3582 };
3583
3584 static struct notifier_block bond_inetaddr_notifier = {
3585 .notifier_call = bond_inetaddr_event,
3586 };
3587
3588 /*-------------------------- Packet type handling ---------------------------*/
3589
3590 /* register to receive lacpdus on a bond */
3591 static void bond_register_lacpdu(struct bonding *bond)
3592 {
3593 struct packet_type *pk_type = &(BOND_AD_INFO(bond).ad_pkt_type);
3594
3595 /* initialize packet type */
3596 pk_type->type = PKT_TYPE_LACPDU;
3597 pk_type->dev = bond->dev;
3598 pk_type->func = bond_3ad_lacpdu_recv;
3599
3600 dev_add_pack(pk_type);
3601 }
3602
3603 /* unregister to receive lacpdus on a bond */
3604 static void bond_unregister_lacpdu(struct bonding *bond)
3605 {
3606 dev_remove_pack(&(BOND_AD_INFO(bond).ad_pkt_type));
3607 }
3608
3609 void bond_register_arp(struct bonding *bond)
3610 {
3611 struct packet_type *pt = &bond->arp_mon_pt;
3612
3613 if (pt->type)
3614 return;
3615
3616 pt->type = htons(ETH_P_ARP);
3617 pt->dev = bond->dev;
3618 pt->func = bond_arp_rcv;
3619 dev_add_pack(pt);
3620 }
3621
3622 void bond_unregister_arp(struct bonding *bond)
3623 {
3624 struct packet_type *pt = &bond->arp_mon_pt;
3625
3626 dev_remove_pack(pt);
3627 pt->type = 0;
3628 }
3629
3630 /*---------------------------- Hashing Policies -----------------------------*/
3631
3632 /*
3633 * Hash for the output device based upon layer 2 and layer 3 data. If
3634 * the packet is not IP mimic bond_xmit_hash_policy_l2()
3635 */
3636 static int bond_xmit_hash_policy_l23(struct sk_buff *skb, int count)
3637 {
3638 struct ethhdr *data = (struct ethhdr *)skb->data;
3639 struct iphdr *iph = ip_hdr(skb);
3640
3641 if (skb->protocol == htons(ETH_P_IP)) {
3642 return ((ntohl(iph->saddr ^ iph->daddr) & 0xffff) ^
3643 (data->h_dest[5] ^ data->h_source[5])) % count;
3644 }
3645
3646 return (data->h_dest[5] ^ data->h_source[5]) % count;
3647 }
3648
3649 /*
3650 * Hash for the output device based upon layer 3 and layer 4 data. If
3651 * the packet is a frag or not TCP or UDP, just use layer 3 data. If it is
3652 * altogether not IP, mimic bond_xmit_hash_policy_l2()
3653 */
3654 static int bond_xmit_hash_policy_l34(struct sk_buff *skb, int count)
3655 {
3656 struct ethhdr *data = (struct ethhdr *)skb->data;
3657 struct iphdr *iph = ip_hdr(skb);
3658 __be16 *layer4hdr = (__be16 *)((u32 *)iph + iph->ihl);
3659 int layer4_xor = 0;
3660
3661 if (skb->protocol == htons(ETH_P_IP)) {
3662 if (!(iph->frag_off & htons(IP_MF|IP_OFFSET)) &&
3663 (iph->protocol == IPPROTO_TCP ||
3664 iph->protocol == IPPROTO_UDP)) {
3665 layer4_xor = ntohs((*layer4hdr ^ *(layer4hdr + 1)));
3666 }
3667 return (layer4_xor ^
3668 ((ntohl(iph->saddr ^ iph->daddr)) & 0xffff)) % count;
3669
3670 }
3671
3672 return (data->h_dest[5] ^ data->h_source[5]) % count;
3673 }
3674
3675 /*
3676 * Hash for the output device based upon layer 2 data
3677 */
3678 static int bond_xmit_hash_policy_l2(struct sk_buff *skb, int count)
3679 {
3680 struct ethhdr *data = (struct ethhdr *)skb->data;
3681
3682 return (data->h_dest[5] ^ data->h_source[5]) % count;
3683 }
3684
3685 /*-------------------------- Device entry points ----------------------------*/
3686
3687 static int bond_open(struct net_device *bond_dev)
3688 {
3689 struct bonding *bond = netdev_priv(bond_dev);
3690
3691 bond->kill_timers = 0;
3692
3693 if (bond_is_lb(bond)) {
3694 /* bond_alb_initialize must be called before the timer
3695 * is started.
3696 */
3697 if (bond_alb_initialize(bond, (bond->params.mode == BOND_MODE_ALB))) {
3698 /* something went wrong - fail the open operation */
3699 return -ENOMEM;
3700 }
3701
3702 INIT_DELAYED_WORK(&bond->alb_work, bond_alb_monitor);
3703 queue_delayed_work(bond->wq, &bond->alb_work, 0);
3704 }
3705
3706 if (bond->params.miimon) { /* link check interval, in milliseconds. */
3707 INIT_DELAYED_WORK(&bond->mii_work, bond_mii_monitor);
3708 queue_delayed_work(bond->wq, &bond->mii_work, 0);
3709 }
3710
3711 if (bond->params.arp_interval) { /* arp interval, in milliseconds. */
3712 if (bond->params.mode == BOND_MODE_ACTIVEBACKUP)
3713 INIT_DELAYED_WORK(&bond->arp_work,
3714 bond_activebackup_arp_mon);
3715 else
3716 INIT_DELAYED_WORK(&bond->arp_work,
3717 bond_loadbalance_arp_mon);
3718
3719 queue_delayed_work(bond->wq, &bond->arp_work, 0);
3720 if (bond->params.arp_validate)
3721 bond_register_arp(bond);
3722 }
3723
3724 if (bond->params.mode == BOND_MODE_8023AD) {
3725 INIT_DELAYED_WORK(&bond->ad_work, bond_3ad_state_machine_handler);
3726 queue_delayed_work(bond->wq, &bond->ad_work, 0);
3727 /* register to receive LACPDUs */
3728 bond_register_lacpdu(bond);
3729 bond_3ad_initiate_agg_selection(bond, 1);
3730 }
3731
3732 return 0;
3733 }
3734
3735 static int bond_close(struct net_device *bond_dev)
3736 {
3737 struct bonding *bond = netdev_priv(bond_dev);
3738
3739 if (bond->params.mode == BOND_MODE_8023AD) {
3740 /* Unregister the receive of LACPDUs */
3741 bond_unregister_lacpdu(bond);
3742 }
3743
3744 if (bond->params.arp_validate)
3745 bond_unregister_arp(bond);
3746
3747 write_lock_bh(&bond->lock);
3748
3749 bond->send_grat_arp = 0;
3750 bond->send_unsol_na = 0;
3751
3752 /* signal timers not to re-arm */
3753 bond->kill_timers = 1;
3754
3755 write_unlock_bh(&bond->lock);
3756
3757 if (bond->params.miimon) { /* link check interval, in milliseconds. */
3758 cancel_delayed_work(&bond->mii_work);
3759 }
3760
3761 if (bond->params.arp_interval) { /* arp interval, in milliseconds. */
3762 cancel_delayed_work(&bond->arp_work);
3763 }
3764
3765 switch (bond->params.mode) {
3766 case BOND_MODE_8023AD:
3767 cancel_delayed_work(&bond->ad_work);
3768 break;
3769 case BOND_MODE_TLB:
3770 case BOND_MODE_ALB:
3771 cancel_delayed_work(&bond->alb_work);
3772 break;
3773 default:
3774 break;
3775 }
3776
3777
3778 if (bond_is_lb(bond)) {
3779 /* Must be called only after all
3780 * slaves have been released
3781 */
3782 bond_alb_deinitialize(bond);
3783 }
3784
3785 return 0;
3786 }
3787
3788 static struct net_device_stats *bond_get_stats(struct net_device *bond_dev)
3789 {
3790 struct bonding *bond = netdev_priv(bond_dev);
3791 struct net_device_stats *stats = &bond_dev->stats;
3792 struct net_device_stats local_stats;
3793 struct slave *slave;
3794 int i;
3795
3796 memset(&local_stats, 0, sizeof(struct net_device_stats));
3797
3798 read_lock_bh(&bond->lock);
3799
3800 bond_for_each_slave(bond, slave, i) {
3801 const struct net_device_stats *sstats = dev_get_stats(slave->dev);
3802
3803 local_stats.rx_packets += sstats->rx_packets;
3804 local_stats.rx_bytes += sstats->rx_bytes;
3805 local_stats.rx_errors += sstats->rx_errors;
3806 local_stats.rx_dropped += sstats->rx_dropped;
3807
3808 local_stats.tx_packets += sstats->tx_packets;
3809 local_stats.tx_bytes += sstats->tx_bytes;
3810 local_stats.tx_errors += sstats->tx_errors;
3811 local_stats.tx_dropped += sstats->tx_dropped;
3812
3813 local_stats.multicast += sstats->multicast;
3814 local_stats.collisions += sstats->collisions;
3815
3816 local_stats.rx_length_errors += sstats->rx_length_errors;
3817 local_stats.rx_over_errors += sstats->rx_over_errors;
3818 local_stats.rx_crc_errors += sstats->rx_crc_errors;
3819 local_stats.rx_frame_errors += sstats->rx_frame_errors;
3820 local_stats.rx_fifo_errors += sstats->rx_fifo_errors;
3821 local_stats.rx_missed_errors += sstats->rx_missed_errors;
3822
3823 local_stats.tx_aborted_errors += sstats->tx_aborted_errors;
3824 local_stats.tx_carrier_errors += sstats->tx_carrier_errors;
3825 local_stats.tx_fifo_errors += sstats->tx_fifo_errors;
3826 local_stats.tx_heartbeat_errors += sstats->tx_heartbeat_errors;
3827 local_stats.tx_window_errors += sstats->tx_window_errors;
3828 }
3829
3830 memcpy(stats, &local_stats, sizeof(struct net_device_stats));
3831
3832 read_unlock_bh(&bond->lock);
3833
3834 return stats;
3835 }
3836
3837 static int bond_do_ioctl(struct net_device *bond_dev, struct ifreq *ifr, int cmd)
3838 {
3839 struct net_device *slave_dev = NULL;
3840 struct ifbond k_binfo;
3841 struct ifbond __user *u_binfo = NULL;
3842 struct ifslave k_sinfo;
3843 struct ifslave __user *u_sinfo = NULL;
3844 struct mii_ioctl_data *mii = NULL;
3845 int res = 0;
3846
3847 pr_debug("bond_ioctl: master=%s, cmd=%d\n", bond_dev->name, cmd);
3848
3849 switch (cmd) {
3850 case SIOCGMIIPHY:
3851 mii = if_mii(ifr);
3852 if (!mii)
3853 return -EINVAL;
3854
3855 mii->phy_id = 0;
3856 /* Fall Through */
3857 case SIOCGMIIREG:
3858 /*
3859 * We do this again just in case we were called by SIOCGMIIREG
3860 * instead of SIOCGMIIPHY.
3861 */
3862 mii = if_mii(ifr);
3863 if (!mii)
3864 return -EINVAL;
3865
3866
3867 if (mii->reg_num == 1) {
3868 struct bonding *bond = netdev_priv(bond_dev);
3869 mii->val_out = 0;
3870 read_lock(&bond->lock);
3871 read_lock(&bond->curr_slave_lock);
3872 if (netif_carrier_ok(bond->dev))
3873 mii->val_out = BMSR_LSTATUS;
3874
3875 read_unlock(&bond->curr_slave_lock);
3876 read_unlock(&bond->lock);
3877 }
3878
3879 return 0;
3880 case BOND_INFO_QUERY_OLD:
3881 case SIOCBONDINFOQUERY:
3882 u_binfo = (struct ifbond __user *)ifr->ifr_data;
3883
3884 if (copy_from_user(&k_binfo, u_binfo, sizeof(ifbond)))
3885 return -EFAULT;
3886
3887 res = bond_info_query(bond_dev, &k_binfo);
3888 if (res == 0 &&
3889 copy_to_user(u_binfo, &k_binfo, sizeof(ifbond)))
3890 return -EFAULT;
3891
3892 return res;
3893 case BOND_SLAVE_INFO_QUERY_OLD:
3894 case SIOCBONDSLAVEINFOQUERY:
3895 u_sinfo = (struct ifslave __user *)ifr->ifr_data;
3896
3897 if (copy_from_user(&k_sinfo, u_sinfo, sizeof(ifslave)))
3898 return -EFAULT;
3899
3900 res = bond_slave_info_query(bond_dev, &k_sinfo);
3901 if (res == 0 &&
3902 copy_to_user(u_sinfo, &k_sinfo, sizeof(ifslave)))
3903 return -EFAULT;
3904
3905 return res;
3906 default:
3907 /* Go on */
3908 break;
3909 }
3910
3911 if (!capable(CAP_NET_ADMIN))
3912 return -EPERM;
3913
3914 slave_dev = dev_get_by_name(dev_net(bond_dev), ifr->ifr_slave);
3915
3916 pr_debug("slave_dev=%p:\n", slave_dev);
3917
3918 if (!slave_dev)
3919 res = -ENODEV;
3920 else {
3921 pr_debug("slave_dev->name=%s:\n", slave_dev->name);
3922 switch (cmd) {
3923 case BOND_ENSLAVE_OLD:
3924 case SIOCBONDENSLAVE:
3925 res = bond_enslave(bond_dev, slave_dev);
3926 break;
3927 case BOND_RELEASE_OLD:
3928 case SIOCBONDRELEASE:
3929 res = bond_release(bond_dev, slave_dev);
3930 break;
3931 case BOND_SETHWADDR_OLD:
3932 case SIOCBONDSETHWADDR:
3933 res = bond_sethwaddr(bond_dev, slave_dev);
3934 break;
3935 case BOND_CHANGE_ACTIVE_OLD:
3936 case SIOCBONDCHANGEACTIVE:
3937 res = bond_ioctl_change_active(bond_dev, slave_dev);
3938 break;
3939 default:
3940 res = -EOPNOTSUPP;
3941 }
3942
3943 dev_put(slave_dev);
3944 }
3945
3946 return res;
3947 }
3948
3949 static bool bond_addr_in_mc_list(unsigned char *addr,
3950 struct netdev_hw_addr_list *list,
3951 int addrlen)
3952 {
3953 struct netdev_hw_addr *ha;
3954
3955 netdev_hw_addr_list_for_each(ha, list)
3956 if (!memcmp(ha->addr, addr, addrlen))
3957 return true;
3958
3959 return false;
3960 }
3961
3962 static void bond_set_multicast_list(struct net_device *bond_dev)
3963 {
3964 struct bonding *bond = netdev_priv(bond_dev);
3965 struct netdev_hw_addr *ha;
3966 bool found;
3967
3968 /*
3969 * Do promisc before checking multicast_mode
3970 */
3971 if ((bond_dev->flags & IFF_PROMISC) && !(bond->flags & IFF_PROMISC))
3972 /*
3973 * FIXME: Need to handle the error when one of the multi-slaves
3974 * encounters error.
3975 */
3976 bond_set_promiscuity(bond, 1);
3977
3978
3979 if (!(bond_dev->flags & IFF_PROMISC) && (bond->flags & IFF_PROMISC))
3980 bond_set_promiscuity(bond, -1);
3981
3982
3983 /* set allmulti flag to slaves */
3984 if ((bond_dev->flags & IFF_ALLMULTI) && !(bond->flags & IFF_ALLMULTI))
3985 /*
3986 * FIXME: Need to handle the error when one of the multi-slaves
3987 * encounters error.
3988 */
3989 bond_set_allmulti(bond, 1);
3990
3991
3992 if (!(bond_dev->flags & IFF_ALLMULTI) && (bond->flags & IFF_ALLMULTI))
3993 bond_set_allmulti(bond, -1);
3994
3995
3996 read_lock(&bond->lock);
3997
3998 bond->flags = bond_dev->flags;
3999
4000 /* looking for addresses to add to slaves' mc list */
4001 netdev_for_each_mc_addr(ha, bond_dev) {
4002 found = bond_addr_in_mc_list(ha->addr, &bond->mc_list,
4003 bond_dev->addr_len);
4004 if (!found)
4005 bond_mc_add(bond, ha->addr);
4006 }
4007
4008 /* looking for addresses to delete from slaves' list */
4009 netdev_hw_addr_list_for_each(ha, &bond->mc_list) {
4010 found = bond_addr_in_mc_list(ha->addr, &bond_dev->mc,
4011 bond_dev->addr_len);
4012 if (!found)
4013 bond_mc_del(bond, ha->addr);
4014 }
4015
4016 /* save master's multicast list */
4017 __hw_addr_flush(&bond->mc_list);
4018 __hw_addr_add_multiple(&bond->mc_list, &bond_dev->mc,
4019 bond_dev->addr_len, NETDEV_HW_ADDR_T_MULTICAST);
4020
4021 read_unlock(&bond->lock);
4022 }
4023
4024 static int bond_neigh_setup(struct net_device *dev, struct neigh_parms *parms)
4025 {
4026 struct bonding *bond = netdev_priv(dev);
4027 struct slave *slave = bond->first_slave;
4028
4029 if (slave) {
4030 const struct net_device_ops *slave_ops
4031 = slave->dev->netdev_ops;
4032 if (slave_ops->ndo_neigh_setup)
4033 return slave_ops->ndo_neigh_setup(slave->dev, parms);
4034 }
4035 return 0;
4036 }
4037
4038 /*
4039 * Change the MTU of all of a master's slaves to match the master
4040 */
4041 static int bond_change_mtu(struct net_device *bond_dev, int new_mtu)
4042 {
4043 struct bonding *bond = netdev_priv(bond_dev);
4044 struct slave *slave, *stop_at;
4045 int res = 0;
4046 int i;
4047
4048 pr_debug("bond=%p, name=%s, new_mtu=%d\n", bond,
4049 (bond_dev ? bond_dev->name : "None"), new_mtu);
4050
4051 /* Can't hold bond->lock with bh disabled here since
4052 * some base drivers panic. On the other hand we can't
4053 * hold bond->lock without bh disabled because we'll
4054 * deadlock. The only solution is to rely on the fact
4055 * that we're under rtnl_lock here, and the slaves
4056 * list won't change. This doesn't solve the problem
4057 * of setting the slave's MTU while it is
4058 * transmitting, but the assumption is that the base
4059 * driver can handle that.
4060 *
4061 * TODO: figure out a way to safely iterate the slaves
4062 * list, but without holding a lock around the actual
4063 * call to the base driver.
4064 */
4065
4066 bond_for_each_slave(bond, slave, i) {
4067 pr_debug("s %p s->p %p c_m %p\n",
4068 slave,
4069 slave->prev,
4070 slave->dev->netdev_ops->ndo_change_mtu);
4071
4072 res = dev_set_mtu(slave->dev, new_mtu);
4073
4074 if (res) {
4075 /* If we failed to set the slave's mtu to the new value
4076 * we must abort the operation even in ACTIVE_BACKUP
4077 * mode, because if we allow the backup slaves to have
4078 * different mtu values than the active slave we'll
4079 * need to change their mtu when doing a failover. That
4080 * means changing their mtu from timer context, which
4081 * is probably not a good idea.
4082 */
4083 pr_debug("err %d %s\n", res, slave->dev->name);
4084 goto unwind;
4085 }
4086 }
4087
4088 bond_dev->mtu = new_mtu;
4089
4090 return 0;
4091
4092 unwind:
4093 /* unwind from head to the slave that failed */
4094 stop_at = slave;
4095 bond_for_each_slave_from_to(bond, slave, i, bond->first_slave, stop_at) {
4096 int tmp_res;
4097
4098 tmp_res = dev_set_mtu(slave->dev, bond_dev->mtu);
4099 if (tmp_res) {
4100 pr_debug("unwind err %d dev %s\n",
4101 tmp_res, slave->dev->name);
4102 }
4103 }
4104
4105 return res;
4106 }
4107
4108 /*
4109 * Change HW address
4110 *
4111 * Note that many devices must be down to change the HW address, and
4112 * downing the master releases all slaves. We can make bonds full of
4113 * bonding devices to test this, however.
4114 */
4115 static int bond_set_mac_address(struct net_device *bond_dev, void *addr)
4116 {
4117 struct bonding *bond = netdev_priv(bond_dev);
4118 struct sockaddr *sa = addr, tmp_sa;
4119 struct slave *slave, *stop_at;
4120 int res = 0;
4121 int i;
4122
4123 if (bond->params.mode == BOND_MODE_ALB)
4124 return bond_alb_set_mac_address(bond_dev, addr);
4125
4126
4127 pr_debug("bond=%p, name=%s\n",
4128 bond, bond_dev ? bond_dev->name : "None");
4129
4130 /*
4131 * If fail_over_mac is set to active, do nothing and return
4132 * success. Returning an error causes ifenslave to fail.
4133 */
4134 if (bond->params.fail_over_mac == BOND_FOM_ACTIVE)
4135 return 0;
4136
4137 if (!is_valid_ether_addr(sa->sa_data))
4138 return -EADDRNOTAVAIL;
4139
4140 /* Can't hold bond->lock with bh disabled here since
4141 * some base drivers panic. On the other hand we can't
4142 * hold bond->lock without bh disabled because we'll
4143 * deadlock. The only solution is to rely on the fact
4144 * that we're under rtnl_lock here, and the slaves
4145 * list won't change. This doesn't solve the problem
4146 * of setting the slave's hw address while it is
4147 * transmitting, but the assumption is that the base
4148 * driver can handle that.
4149 *
4150 * TODO: figure out a way to safely iterate the slaves
4151 * list, but without holding a lock around the actual
4152 * call to the base driver.
4153 */
4154
4155 bond_for_each_slave(bond, slave, i) {
4156 const struct net_device_ops *slave_ops = slave->dev->netdev_ops;
4157 pr_debug("slave %p %s\n", slave, slave->dev->name);
4158
4159 if (slave_ops->ndo_set_mac_address == NULL) {
4160 res = -EOPNOTSUPP;
4161 pr_debug("EOPNOTSUPP %s\n", slave->dev->name);
4162 goto unwind;
4163 }
4164
4165 res = dev_set_mac_address(slave->dev, addr);
4166 if (res) {
4167 /* TODO: consider downing the slave
4168 * and retry ?
4169 * User should expect communications
4170 * breakage anyway until ARP finish
4171 * updating, so...
4172 */
4173 pr_debug("err %d %s\n", res, slave->dev->name);
4174 goto unwind;
4175 }
4176 }
4177
4178 /* success */
4179 memcpy(bond_dev->dev_addr, sa->sa_data, bond_dev->addr_len);
4180 return 0;
4181
4182 unwind:
4183 memcpy(tmp_sa.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
4184 tmp_sa.sa_family = bond_dev->type;
4185
4186 /* unwind from head to the slave that failed */
4187 stop_at = slave;
4188 bond_for_each_slave_from_to(bond, slave, i, bond->first_slave, stop_at) {
4189 int tmp_res;
4190
4191 tmp_res = dev_set_mac_address(slave->dev, &tmp_sa);
4192 if (tmp_res) {
4193 pr_debug("unwind err %d dev %s\n",
4194 tmp_res, slave->dev->name);
4195 }
4196 }
4197
4198 return res;
4199 }
4200
4201 static int bond_xmit_roundrobin(struct sk_buff *skb, struct net_device *bond_dev)
4202 {
4203 struct bonding *bond = netdev_priv(bond_dev);
4204 struct slave *slave, *start_at;
4205 int i, slave_no, res = 1;
4206 struct iphdr *iph = ip_hdr(skb);
4207
4208 read_lock(&bond->lock);
4209
4210 if (!BOND_IS_OK(bond))
4211 goto out;
4212 /*
4213 * Start with the curr_active_slave that joined the bond as the
4214 * default for sending IGMP traffic. For failover purposes one
4215 * needs to maintain some consistency for the interface that will
4216 * send the join/membership reports. The curr_active_slave found
4217 * will send all of this type of traffic.
4218 */
4219 if ((iph->protocol == IPPROTO_IGMP) &&
4220 (skb->protocol == htons(ETH_P_IP))) {
4221
4222 read_lock(&bond->curr_slave_lock);
4223 slave = bond->curr_active_slave;
4224 read_unlock(&bond->curr_slave_lock);
4225
4226 if (!slave)
4227 goto out;
4228 } else {
4229 /*
4230 * Concurrent TX may collide on rr_tx_counter; we accept
4231 * that as being rare enough not to justify using an
4232 * atomic op here.
4233 */
4234 slave_no = bond->rr_tx_counter++ % bond->slave_cnt;
4235
4236 bond_for_each_slave(bond, slave, i) {
4237 slave_no--;
4238 if (slave_no < 0)
4239 break;
4240 }
4241 }
4242
4243 start_at = slave;
4244 bond_for_each_slave_from(bond, slave, i, start_at) {
4245 if (IS_UP(slave->dev) &&
4246 (slave->link == BOND_LINK_UP) &&
4247 (slave->state == BOND_STATE_ACTIVE)) {
4248 res = bond_dev_queue_xmit(bond, skb, slave->dev);
4249 break;
4250 }
4251 }
4252
4253 out:
4254 if (res) {
4255 /* no suitable interface, frame not sent */
4256 dev_kfree_skb(skb);
4257 }
4258 read_unlock(&bond->lock);
4259 return NETDEV_TX_OK;
4260 }
4261
4262
4263 /*
4264 * in active-backup mode, we know that bond->curr_active_slave is always valid if
4265 * the bond has a usable interface.
4266 */
4267 static int bond_xmit_activebackup(struct sk_buff *skb, struct net_device *bond_dev)
4268 {
4269 struct bonding *bond = netdev_priv(bond_dev);
4270 int res = 1;
4271
4272 read_lock(&bond->lock);
4273 read_lock(&bond->curr_slave_lock);
4274
4275 if (!BOND_IS_OK(bond))
4276 goto out;
4277
4278 if (!bond->curr_active_slave)
4279 goto out;
4280
4281 res = bond_dev_queue_xmit(bond, skb, bond->curr_active_slave->dev);
4282
4283 out:
4284 if (res)
4285 /* no suitable interface, frame not sent */
4286 dev_kfree_skb(skb);
4287
4288 read_unlock(&bond->curr_slave_lock);
4289 read_unlock(&bond->lock);
4290 return NETDEV_TX_OK;
4291 }
4292
4293 /*
4294 * In bond_xmit_xor() , we determine the output device by using a pre-
4295 * determined xmit_hash_policy(), If the selected device is not enabled,
4296 * find the next active slave.
4297 */
4298 static int bond_xmit_xor(struct sk_buff *skb, struct net_device *bond_dev)
4299 {
4300 struct bonding *bond = netdev_priv(bond_dev);
4301 struct slave *slave, *start_at;
4302 int slave_no;
4303 int i;
4304 int res = 1;
4305
4306 read_lock(&bond->lock);
4307
4308 if (!BOND_IS_OK(bond))
4309 goto out;
4310
4311 slave_no = bond->xmit_hash_policy(skb, bond->slave_cnt);
4312
4313 bond_for_each_slave(bond, slave, i) {
4314 slave_no--;
4315 if (slave_no < 0)
4316 break;
4317 }
4318
4319 start_at = slave;
4320
4321 bond_for_each_slave_from(bond, slave, i, start_at) {
4322 if (IS_UP(slave->dev) &&
4323 (slave->link == BOND_LINK_UP) &&
4324 (slave->state == BOND_STATE_ACTIVE)) {
4325 res = bond_dev_queue_xmit(bond, skb, slave->dev);
4326 break;
4327 }
4328 }
4329
4330 out:
4331 if (res) {
4332 /* no suitable interface, frame not sent */
4333 dev_kfree_skb(skb);
4334 }
4335 read_unlock(&bond->lock);
4336 return NETDEV_TX_OK;
4337 }
4338
4339 /*
4340 * in broadcast mode, we send everything to all usable interfaces.
4341 */
4342 static int bond_xmit_broadcast(struct sk_buff *skb, struct net_device *bond_dev)
4343 {
4344 struct bonding *bond = netdev_priv(bond_dev);
4345 struct slave *slave, *start_at;
4346 struct net_device *tx_dev = NULL;
4347 int i;
4348 int res = 1;
4349
4350 read_lock(&bond->lock);
4351
4352 if (!BOND_IS_OK(bond))
4353 goto out;
4354
4355 read_lock(&bond->curr_slave_lock);
4356 start_at = bond->curr_active_slave;
4357 read_unlock(&bond->curr_slave_lock);
4358
4359 if (!start_at)
4360 goto out;
4361
4362 bond_for_each_slave_from(bond, slave, i, start_at) {
4363 if (IS_UP(slave->dev) &&
4364 (slave->link == BOND_LINK_UP) &&
4365 (slave->state == BOND_STATE_ACTIVE)) {
4366 if (tx_dev) {
4367 struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
4368 if (!skb2) {
4369 pr_err("%s: Error: bond_xmit_broadcast(): skb_clone() failed\n",
4370 bond_dev->name);
4371 continue;
4372 }
4373
4374 res = bond_dev_queue_xmit(bond, skb2, tx_dev);
4375 if (res) {
4376 dev_kfree_skb(skb2);
4377 continue;
4378 }
4379 }
4380 tx_dev = slave->dev;
4381 }
4382 }
4383
4384 if (tx_dev)
4385 res = bond_dev_queue_xmit(bond, skb, tx_dev);
4386
4387 out:
4388 if (res)
4389 /* no suitable interface, frame not sent */
4390 dev_kfree_skb(skb);
4391
4392 /* frame sent to all suitable interfaces */
4393 read_unlock(&bond->lock);
4394 return NETDEV_TX_OK;
4395 }
4396
4397 /*------------------------- Device initialization ---------------------------*/
4398
4399 static void bond_set_xmit_hash_policy(struct bonding *bond)
4400 {
4401 switch (bond->params.xmit_policy) {
4402 case BOND_XMIT_POLICY_LAYER23:
4403 bond->xmit_hash_policy = bond_xmit_hash_policy_l23;
4404 break;
4405 case BOND_XMIT_POLICY_LAYER34:
4406 bond->xmit_hash_policy = bond_xmit_hash_policy_l34;
4407 break;
4408 case BOND_XMIT_POLICY_LAYER2:
4409 default:
4410 bond->xmit_hash_policy = bond_xmit_hash_policy_l2;
4411 break;
4412 }
4413 }
4414
4415 static netdev_tx_t bond_start_xmit(struct sk_buff *skb, struct net_device *dev)
4416 {
4417 const struct bonding *bond = netdev_priv(dev);
4418
4419 switch (bond->params.mode) {
4420 case BOND_MODE_ROUNDROBIN:
4421 return bond_xmit_roundrobin(skb, dev);
4422 case BOND_MODE_ACTIVEBACKUP:
4423 return bond_xmit_activebackup(skb, dev);
4424 case BOND_MODE_XOR:
4425 return bond_xmit_xor(skb, dev);
4426 case BOND_MODE_BROADCAST:
4427 return bond_xmit_broadcast(skb, dev);
4428 case BOND_MODE_8023AD:
4429 return bond_3ad_xmit_xor(skb, dev);
4430 case BOND_MODE_ALB:
4431 case BOND_MODE_TLB:
4432 return bond_alb_xmit(skb, dev);
4433 default:
4434 /* Should never happen, mode already checked */
4435 pr_err("%s: Error: Unknown bonding mode %d\n",
4436 dev->name, bond->params.mode);
4437 WARN_ON_ONCE(1);
4438 dev_kfree_skb(skb);
4439 return NETDEV_TX_OK;
4440 }
4441 }
4442
4443
4444 /*
4445 * set bond mode specific net device operations
4446 */
4447 void bond_set_mode_ops(struct bonding *bond, int mode)
4448 {
4449 struct net_device *bond_dev = bond->dev;
4450
4451 switch (mode) {
4452 case BOND_MODE_ROUNDROBIN:
4453 break;
4454 case BOND_MODE_ACTIVEBACKUP:
4455 break;
4456 case BOND_MODE_XOR:
4457 bond_set_xmit_hash_policy(bond);
4458 break;
4459 case BOND_MODE_BROADCAST:
4460 break;
4461 case BOND_MODE_8023AD:
4462 bond_set_master_3ad_flags(bond);
4463 bond_set_xmit_hash_policy(bond);
4464 break;
4465 case BOND_MODE_ALB:
4466 bond_set_master_alb_flags(bond);
4467 /* FALLTHRU */
4468 case BOND_MODE_TLB:
4469 break;
4470 default:
4471 /* Should never happen, mode already checked */
4472 pr_err("%s: Error: Unknown bonding mode %d\n",
4473 bond_dev->name, mode);
4474 break;
4475 }
4476 }
4477
4478 static void bond_ethtool_get_drvinfo(struct net_device *bond_dev,
4479 struct ethtool_drvinfo *drvinfo)
4480 {
4481 strncpy(drvinfo->driver, DRV_NAME, 32);
4482 strncpy(drvinfo->version, DRV_VERSION, 32);
4483 snprintf(drvinfo->fw_version, 32, "%d", BOND_ABI_VERSION);
4484 }
4485
4486 static const struct ethtool_ops bond_ethtool_ops = {
4487 .get_drvinfo = bond_ethtool_get_drvinfo,
4488 .get_link = ethtool_op_get_link,
4489 .get_tx_csum = ethtool_op_get_tx_csum,
4490 .get_sg = ethtool_op_get_sg,
4491 .get_tso = ethtool_op_get_tso,
4492 .get_ufo = ethtool_op_get_ufo,
4493 .get_flags = ethtool_op_get_flags,
4494 };
4495
4496 static const struct net_device_ops bond_netdev_ops = {
4497 .ndo_init = bond_init,
4498 .ndo_uninit = bond_uninit,
4499 .ndo_open = bond_open,
4500 .ndo_stop = bond_close,
4501 .ndo_start_xmit = bond_start_xmit,
4502 .ndo_get_stats = bond_get_stats,
4503 .ndo_do_ioctl = bond_do_ioctl,
4504 .ndo_set_multicast_list = bond_set_multicast_list,
4505 .ndo_change_mtu = bond_change_mtu,
4506 .ndo_set_mac_address = bond_set_mac_address,
4507 .ndo_neigh_setup = bond_neigh_setup,
4508 .ndo_vlan_rx_register = bond_vlan_rx_register,
4509 .ndo_vlan_rx_add_vid = bond_vlan_rx_add_vid,
4510 .ndo_vlan_rx_kill_vid = bond_vlan_rx_kill_vid,
4511 #ifdef CONFIG_NET_POLL_CONTROLLER
4512 .ndo_netpoll_cleanup = bond_netpoll_cleanup,
4513 .ndo_poll_controller = bond_poll_controller,
4514 #endif
4515 };
4516
4517 static void bond_destructor(struct net_device *bond_dev)
4518 {
4519 struct bonding *bond = netdev_priv(bond_dev);
4520 if (bond->wq)
4521 destroy_workqueue(bond->wq);
4522 free_netdev(bond_dev);
4523 }
4524
4525 static void bond_setup(struct net_device *bond_dev)
4526 {
4527 struct bonding *bond = netdev_priv(bond_dev);
4528
4529 /* initialize rwlocks */
4530 rwlock_init(&bond->lock);
4531 rwlock_init(&bond->curr_slave_lock);
4532
4533 bond->params = bonding_defaults;
4534
4535 /* Initialize pointers */
4536 bond->dev = bond_dev;
4537 INIT_LIST_HEAD(&bond->vlan_list);
4538
4539 /* Initialize the device entry points */
4540 ether_setup(bond_dev);
4541 bond_dev->netdev_ops = &bond_netdev_ops;
4542 bond_dev->ethtool_ops = &bond_ethtool_ops;
4543 bond_set_mode_ops(bond, bond->params.mode);
4544
4545 bond_dev->destructor = bond_destructor;
4546
4547 /* Initialize the device options */
4548 bond_dev->tx_queue_len = 0;
4549 bond_dev->flags |= IFF_MASTER|IFF_MULTICAST;
4550 bond_dev->priv_flags |= IFF_BONDING;
4551 bond_dev->priv_flags &= ~IFF_XMIT_DST_RELEASE;
4552
4553 if (bond->params.arp_interval)
4554 bond_dev->priv_flags |= IFF_MASTER_ARPMON;
4555
4556 /* At first, we block adding VLANs. That's the only way to
4557 * prevent problems that occur when adding VLANs over an
4558 * empty bond. The block will be removed once non-challenged
4559 * slaves are enslaved.
4560 */
4561 bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
4562
4563 /* don't acquire bond device's netif_tx_lock when
4564 * transmitting */
4565 bond_dev->features |= NETIF_F_LLTX;
4566
4567 /* By default, we declare the bond to be fully
4568 * VLAN hardware accelerated capable. Special
4569 * care is taken in the various xmit functions
4570 * when there are slaves that are not hw accel
4571 * capable
4572 */
4573 bond_dev->features |= (NETIF_F_HW_VLAN_TX |
4574 NETIF_F_HW_VLAN_RX |
4575 NETIF_F_HW_VLAN_FILTER);
4576
4577 }
4578
4579 static void bond_work_cancel_all(struct bonding *bond)
4580 {
4581 write_lock_bh(&bond->lock);
4582 bond->kill_timers = 1;
4583 write_unlock_bh(&bond->lock);
4584
4585 if (bond->params.miimon && delayed_work_pending(&bond->mii_work))
4586 cancel_delayed_work(&bond->mii_work);
4587
4588 if (bond->params.arp_interval && delayed_work_pending(&bond->arp_work))
4589 cancel_delayed_work(&bond->arp_work);
4590
4591 if (bond->params.mode == BOND_MODE_ALB &&
4592 delayed_work_pending(&bond->alb_work))
4593 cancel_delayed_work(&bond->alb_work);
4594
4595 if (bond->params.mode == BOND_MODE_8023AD &&
4596 delayed_work_pending(&bond->ad_work))
4597 cancel_delayed_work(&bond->ad_work);
4598 }
4599
4600 /*
4601 * Destroy a bonding device.
4602 * Must be under rtnl_lock when this function is called.
4603 */
4604 static void bond_uninit(struct net_device *bond_dev)
4605 {
4606 struct bonding *bond = netdev_priv(bond_dev);
4607
4608 bond_netpoll_cleanup(bond_dev);
4609
4610 /* Release the bonded slaves */
4611 bond_release_all(bond_dev);
4612
4613 list_del(&bond->bond_list);
4614
4615 bond_work_cancel_all(bond);
4616
4617 bond_remove_proc_entry(bond);
4618
4619 __hw_addr_flush(&bond->mc_list);
4620 }
4621
4622 /*------------------------- Module initialization ---------------------------*/
4623
4624 /*
4625 * Convert string input module parms. Accept either the
4626 * number of the mode or its string name. A bit complicated because
4627 * some mode names are substrings of other names, and calls from sysfs
4628 * may have whitespace in the name (trailing newlines, for example).
4629 */
4630 int bond_parse_parm(const char *buf, const struct bond_parm_tbl *tbl)
4631 {
4632 int modeint = -1, i, rv;
4633 char *p, modestr[BOND_MAX_MODENAME_LEN + 1] = { 0, };
4634
4635 for (p = (char *)buf; *p; p++)
4636 if (!(isdigit(*p) || isspace(*p)))
4637 break;
4638
4639 if (*p)
4640 rv = sscanf(buf, "%20s", modestr);
4641 else
4642 rv = sscanf(buf, "%d", &modeint);
4643
4644 if (!rv)
4645 return -1;
4646
4647 for (i = 0; tbl[i].modename; i++) {
4648 if (modeint == tbl[i].mode)
4649 return tbl[i].mode;
4650 if (strcmp(modestr, tbl[i].modename) == 0)
4651 return tbl[i].mode;
4652 }
4653
4654 return -1;
4655 }
4656
4657 static int bond_check_params(struct bond_params *params)
4658 {
4659 int arp_validate_value, fail_over_mac_value, primary_reselect_value;
4660
4661 /*
4662 * Convert string parameters.
4663 */
4664 if (mode) {
4665 bond_mode = bond_parse_parm(mode, bond_mode_tbl);
4666 if (bond_mode == -1) {
4667 pr_err("Error: Invalid bonding mode \"%s\"\n",
4668 mode == NULL ? "NULL" : mode);
4669 return -EINVAL;
4670 }
4671 }
4672
4673 if (xmit_hash_policy) {
4674 if ((bond_mode != BOND_MODE_XOR) &&
4675 (bond_mode != BOND_MODE_8023AD)) {
4676 pr_info("xmit_hash_policy param is irrelevant in mode %s\n",
4677 bond_mode_name(bond_mode));
4678 } else {
4679 xmit_hashtype = bond_parse_parm(xmit_hash_policy,
4680 xmit_hashtype_tbl);
4681 if (xmit_hashtype == -1) {
4682 pr_err("Error: Invalid xmit_hash_policy \"%s\"\n",
4683 xmit_hash_policy == NULL ? "NULL" :
4684 xmit_hash_policy);
4685 return -EINVAL;
4686 }
4687 }
4688 }
4689
4690 if (lacp_rate) {
4691 if (bond_mode != BOND_MODE_8023AD) {
4692 pr_info("lacp_rate param is irrelevant in mode %s\n",
4693 bond_mode_name(bond_mode));
4694 } else {
4695 lacp_fast = bond_parse_parm(lacp_rate, bond_lacp_tbl);
4696 if (lacp_fast == -1) {
4697 pr_err("Error: Invalid lacp rate \"%s\"\n",
4698 lacp_rate == NULL ? "NULL" : lacp_rate);
4699 return -EINVAL;
4700 }
4701 }
4702 }
4703
4704 if (ad_select) {
4705 params->ad_select = bond_parse_parm(ad_select, ad_select_tbl);
4706 if (params->ad_select == -1) {
4707 pr_err("Error: Invalid ad_select \"%s\"\n",
4708 ad_select == NULL ? "NULL" : ad_select);
4709 return -EINVAL;
4710 }
4711
4712 if (bond_mode != BOND_MODE_8023AD) {
4713 pr_warning("ad_select param only affects 802.3ad mode\n");
4714 }
4715 } else {
4716 params->ad_select = BOND_AD_STABLE;
4717 }
4718
4719 if (max_bonds < 0) {
4720 pr_warning("Warning: max_bonds (%d) not in range %d-%d, so it was reset to BOND_DEFAULT_MAX_BONDS (%d)\n",
4721 max_bonds, 0, INT_MAX, BOND_DEFAULT_MAX_BONDS);
4722 max_bonds = BOND_DEFAULT_MAX_BONDS;
4723 }
4724
4725 if (miimon < 0) {
4726 pr_warning("Warning: miimon module parameter (%d), not in range 0-%d, so it was reset to %d\n",
4727 miimon, INT_MAX, BOND_LINK_MON_INTERV);
4728 miimon = BOND_LINK_MON_INTERV;
4729 }
4730
4731 if (updelay < 0) {
4732 pr_warning("Warning: updelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4733 updelay, INT_MAX);
4734 updelay = 0;
4735 }
4736
4737 if (downdelay < 0) {
4738 pr_warning("Warning: downdelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4739 downdelay, INT_MAX);
4740 downdelay = 0;
4741 }
4742
4743 if ((use_carrier != 0) && (use_carrier != 1)) {
4744 pr_warning("Warning: use_carrier module parameter (%d), not of valid value (0/1), so it was set to 1\n",
4745 use_carrier);
4746 use_carrier = 1;
4747 }
4748
4749 if (num_grat_arp < 0 || num_grat_arp > 255) {
4750 pr_warning("Warning: num_grat_arp (%d) not in range 0-255 so it was reset to 1\n",
4751 num_grat_arp);
4752 num_grat_arp = 1;
4753 }
4754
4755 if (num_unsol_na < 0 || num_unsol_na > 255) {
4756 pr_warning("Warning: num_unsol_na (%d) not in range 0-255 so it was reset to 1\n",
4757 num_unsol_na);
4758 num_unsol_na = 1;
4759 }
4760
4761 /* reset values for 802.3ad */
4762 if (bond_mode == BOND_MODE_8023AD) {
4763 if (!miimon) {
4764 pr_warning("Warning: miimon must be specified, otherwise bonding will not detect link failure, speed and duplex which are essential for 802.3ad operation\n");
4765 pr_warning("Forcing miimon to 100msec\n");
4766 miimon = 100;
4767 }
4768 }
4769
4770 /* reset values for TLB/ALB */
4771 if ((bond_mode == BOND_MODE_TLB) ||
4772 (bond_mode == BOND_MODE_ALB)) {
4773 if (!miimon) {
4774 pr_warning("Warning: miimon must be specified, otherwise bonding will not detect link failure and link speed which are essential for TLB/ALB load balancing\n");
4775 pr_warning("Forcing miimon to 100msec\n");
4776 miimon = 100;
4777 }
4778 }
4779
4780 if (bond_mode == BOND_MODE_ALB) {
4781 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",
4782 updelay);
4783 }
4784
4785 if (!miimon) {
4786 if (updelay || downdelay) {
4787 /* just warn the user the up/down delay will have
4788 * no effect since miimon is zero...
4789 */
4790 pr_warning("Warning: miimon module parameter not set and updelay (%d) or downdelay (%d) module parameter is set; updelay and downdelay have no effect unless miimon is set\n",
4791 updelay, downdelay);
4792 }
4793 } else {
4794 /* don't allow arp monitoring */
4795 if (arp_interval) {
4796 pr_warning("Warning: miimon (%d) and arp_interval (%d) can't be used simultaneously, disabling ARP monitoring\n",
4797 miimon, arp_interval);
4798 arp_interval = 0;
4799 }
4800
4801 if ((updelay % miimon) != 0) {
4802 pr_warning("Warning: updelay (%d) is not a multiple of miimon (%d), updelay rounded to %d ms\n",
4803 updelay, miimon,
4804 (updelay / miimon) * miimon);
4805 }
4806
4807 updelay /= miimon;
4808
4809 if ((downdelay % miimon) != 0) {
4810 pr_warning("Warning: downdelay (%d) is not a multiple of miimon (%d), downdelay rounded to %d ms\n",
4811 downdelay, miimon,
4812 (downdelay / miimon) * miimon);
4813 }
4814
4815 downdelay /= miimon;
4816 }
4817
4818 if (arp_interval < 0) {
4819 pr_warning("Warning: arp_interval module parameter (%d) , not in range 0-%d, so it was reset to %d\n",
4820 arp_interval, INT_MAX, BOND_LINK_ARP_INTERV);
4821 arp_interval = BOND_LINK_ARP_INTERV;
4822 }
4823
4824 for (arp_ip_count = 0;
4825 (arp_ip_count < BOND_MAX_ARP_TARGETS) && arp_ip_target[arp_ip_count];
4826 arp_ip_count++) {
4827 /* not complete check, but should be good enough to
4828 catch mistakes */
4829 if (!isdigit(arp_ip_target[arp_ip_count][0])) {
4830 pr_warning("Warning: bad arp_ip_target module parameter (%s), ARP monitoring will not be performed\n",
4831 arp_ip_target[arp_ip_count]);
4832 arp_interval = 0;
4833 } else {
4834 __be32 ip = in_aton(arp_ip_target[arp_ip_count]);
4835 arp_target[arp_ip_count] = ip;
4836 }
4837 }
4838
4839 if (arp_interval && !arp_ip_count) {
4840 /* don't allow arping if no arp_ip_target given... */
4841 pr_warning("Warning: arp_interval module parameter (%d) specified without providing an arp_ip_target parameter, arp_interval was reset to 0\n",
4842 arp_interval);
4843 arp_interval = 0;
4844 }
4845
4846 if (arp_validate) {
4847 if (bond_mode != BOND_MODE_ACTIVEBACKUP) {
4848 pr_err("arp_validate only supported in active-backup mode\n");
4849 return -EINVAL;
4850 }
4851 if (!arp_interval) {
4852 pr_err("arp_validate requires arp_interval\n");
4853 return -EINVAL;
4854 }
4855
4856 arp_validate_value = bond_parse_parm(arp_validate,
4857 arp_validate_tbl);
4858 if (arp_validate_value == -1) {
4859 pr_err("Error: invalid arp_validate \"%s\"\n",
4860 arp_validate == NULL ? "NULL" : arp_validate);
4861 return -EINVAL;
4862 }
4863 } else
4864 arp_validate_value = 0;
4865
4866 if (miimon) {
4867 pr_info("MII link monitoring set to %d ms\n", miimon);
4868 } else if (arp_interval) {
4869 int i;
4870
4871 pr_info("ARP monitoring set to %d ms, validate %s, with %d target(s):",
4872 arp_interval,
4873 arp_validate_tbl[arp_validate_value].modename,
4874 arp_ip_count);
4875
4876 for (i = 0; i < arp_ip_count; i++)
4877 pr_info(" %s", arp_ip_target[i]);
4878
4879 pr_info("\n");
4880
4881 } else if (max_bonds) {
4882 /* miimon and arp_interval not set, we need one so things
4883 * work as expected, see bonding.txt for details
4884 */
4885 pr_warning("Warning: either miimon or arp_interval and arp_ip_target module parameters must be specified, otherwise bonding will not detect link failures! see bonding.txt for details.\n");
4886 }
4887
4888 if (primary && !USES_PRIMARY(bond_mode)) {
4889 /* currently, using a primary only makes sense
4890 * in active backup, TLB or ALB modes
4891 */
4892 pr_warning("Warning: %s primary device specified but has no effect in %s mode\n",
4893 primary, bond_mode_name(bond_mode));
4894 primary = NULL;
4895 }
4896
4897 if (primary && primary_reselect) {
4898 primary_reselect_value = bond_parse_parm(primary_reselect,
4899 pri_reselect_tbl);
4900 if (primary_reselect_value == -1) {
4901 pr_err("Error: Invalid primary_reselect \"%s\"\n",
4902 primary_reselect ==
4903 NULL ? "NULL" : primary_reselect);
4904 return -EINVAL;
4905 }
4906 } else {
4907 primary_reselect_value = BOND_PRI_RESELECT_ALWAYS;
4908 }
4909
4910 if (fail_over_mac) {
4911 fail_over_mac_value = bond_parse_parm(fail_over_mac,
4912 fail_over_mac_tbl);
4913 if (fail_over_mac_value == -1) {
4914 pr_err("Error: invalid fail_over_mac \"%s\"\n",
4915 arp_validate == NULL ? "NULL" : arp_validate);
4916 return -EINVAL;
4917 }
4918
4919 if (bond_mode != BOND_MODE_ACTIVEBACKUP)
4920 pr_warning("Warning: fail_over_mac only affects active-backup mode.\n");
4921 } else {
4922 fail_over_mac_value = BOND_FOM_NONE;
4923 }
4924
4925 /* fill params struct with the proper values */
4926 params->mode = bond_mode;
4927 params->xmit_policy = xmit_hashtype;
4928 params->miimon = miimon;
4929 params->num_grat_arp = num_grat_arp;
4930 params->num_unsol_na = num_unsol_na;
4931 params->arp_interval = arp_interval;
4932 params->arp_validate = arp_validate_value;
4933 params->updelay = updelay;
4934 params->downdelay = downdelay;
4935 params->use_carrier = use_carrier;
4936 params->lacp_fast = lacp_fast;
4937 params->primary[0] = 0;
4938 params->primary_reselect = primary_reselect_value;
4939 params->fail_over_mac = fail_over_mac_value;
4940
4941 if (primary) {
4942 strncpy(params->primary, primary, IFNAMSIZ);
4943 params->primary[IFNAMSIZ - 1] = 0;
4944 }
4945
4946 memcpy(params->arp_targets, arp_target, sizeof(arp_target));
4947
4948 return 0;
4949 }
4950
4951 static struct lock_class_key bonding_netdev_xmit_lock_key;
4952 static struct lock_class_key bonding_netdev_addr_lock_key;
4953
4954 static void bond_set_lockdep_class_one(struct net_device *dev,
4955 struct netdev_queue *txq,
4956 void *_unused)
4957 {
4958 lockdep_set_class(&txq->_xmit_lock,
4959 &bonding_netdev_xmit_lock_key);
4960 }
4961
4962 static void bond_set_lockdep_class(struct net_device *dev)
4963 {
4964 lockdep_set_class(&dev->addr_list_lock,
4965 &bonding_netdev_addr_lock_key);
4966 netdev_for_each_tx_queue(dev, bond_set_lockdep_class_one, NULL);
4967 }
4968
4969 /*
4970 * Called from registration process
4971 */
4972 static int bond_init(struct net_device *bond_dev)
4973 {
4974 struct bonding *bond = netdev_priv(bond_dev);
4975 struct bond_net *bn = net_generic(dev_net(bond_dev), bond_net_id);
4976
4977 pr_debug("Begin bond_init for %s\n", bond_dev->name);
4978
4979 bond->wq = create_singlethread_workqueue(bond_dev->name);
4980 if (!bond->wq)
4981 return -ENOMEM;
4982
4983 bond_set_lockdep_class(bond_dev);
4984
4985 netif_carrier_off(bond_dev);
4986
4987 bond_create_proc_entry(bond);
4988 list_add_tail(&bond->bond_list, &bn->dev_list);
4989
4990 bond_prepare_sysfs_group(bond);
4991
4992 __hw_addr_init(&bond->mc_list);
4993 return 0;
4994 }
4995
4996 static int bond_validate(struct nlattr *tb[], struct nlattr *data[])
4997 {
4998 if (tb[IFLA_ADDRESS]) {
4999 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN)
5000 return -EINVAL;
5001 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS])))
5002 return -EADDRNOTAVAIL;
5003 }
5004 return 0;
5005 }
5006
5007 static struct rtnl_link_ops bond_link_ops __read_mostly = {
5008 .kind = "bond",
5009 .priv_size = sizeof(struct bonding),
5010 .setup = bond_setup,
5011 .validate = bond_validate,
5012 };
5013
5014 /* Create a new bond based on the specified name and bonding parameters.
5015 * If name is NULL, obtain a suitable "bond%d" name for us.
5016 * Caller must NOT hold rtnl_lock; we need to release it here before we
5017 * set up our sysfs entries.
5018 */
5019 int bond_create(struct net *net, const char *name)
5020 {
5021 struct net_device *bond_dev;
5022 int res;
5023
5024 rtnl_lock();
5025
5026 bond_dev = alloc_netdev(sizeof(struct bonding), name ? name : "",
5027 bond_setup);
5028 if (!bond_dev) {
5029 pr_err("%s: eek! can't alloc netdev!\n", name);
5030 rtnl_unlock();
5031 return -ENOMEM;
5032 }
5033
5034 dev_net_set(bond_dev, net);
5035 bond_dev->rtnl_link_ops = &bond_link_ops;
5036
5037 if (!name) {
5038 res = dev_alloc_name(bond_dev, "bond%d");
5039 if (res < 0)
5040 goto out;
5041 }
5042
5043 res = register_netdevice(bond_dev);
5044
5045 out:
5046 rtnl_unlock();
5047 if (res < 0)
5048 bond_destructor(bond_dev);
5049 return res;
5050 }
5051
5052 static int __net_init bond_net_init(struct net *net)
5053 {
5054 struct bond_net *bn = net_generic(net, bond_net_id);
5055
5056 bn->net = net;
5057 INIT_LIST_HEAD(&bn->dev_list);
5058
5059 bond_create_proc_dir(bn);
5060
5061 return 0;
5062 }
5063
5064 static void __net_exit bond_net_exit(struct net *net)
5065 {
5066 struct bond_net *bn = net_generic(net, bond_net_id);
5067
5068 bond_destroy_proc_dir(bn);
5069 }
5070
5071 static struct pernet_operations bond_net_ops = {
5072 .init = bond_net_init,
5073 .exit = bond_net_exit,
5074 .id = &bond_net_id,
5075 .size = sizeof(struct bond_net),
5076 };
5077
5078 static int __init bonding_init(void)
5079 {
5080 int i;
5081 int res;
5082
5083 pr_info("%s", version);
5084
5085 res = bond_check_params(&bonding_defaults);
5086 if (res)
5087 goto out;
5088
5089 res = register_pernet_subsys(&bond_net_ops);
5090 if (res)
5091 goto out;
5092
5093 res = rtnl_link_register(&bond_link_ops);
5094 if (res)
5095 goto err_link;
5096
5097 for (i = 0; i < max_bonds; i++) {
5098 res = bond_create(&init_net, NULL);
5099 if (res)
5100 goto err;
5101 }
5102
5103 res = bond_create_sysfs();
5104 if (res)
5105 goto err;
5106
5107 register_netdevice_notifier(&bond_netdev_notifier);
5108 register_inetaddr_notifier(&bond_inetaddr_notifier);
5109 bond_register_ipv6_notifier();
5110 out:
5111 return res;
5112 err:
5113 rtnl_link_unregister(&bond_link_ops);
5114 err_link:
5115 unregister_pernet_subsys(&bond_net_ops);
5116 goto out;
5117
5118 }
5119
5120 static void __exit bonding_exit(void)
5121 {
5122 unregister_netdevice_notifier(&bond_netdev_notifier);
5123 unregister_inetaddr_notifier(&bond_inetaddr_notifier);
5124 bond_unregister_ipv6_notifier();
5125
5126 bond_destroy_sysfs();
5127
5128 rtnl_link_unregister(&bond_link_ops);
5129 unregister_pernet_subsys(&bond_net_ops);
5130 }
5131
5132 module_init(bonding_init);
5133 module_exit(bonding_exit);
5134 MODULE_LICENSE("GPL");
5135 MODULE_VERSION(DRV_VERSION);
5136 MODULE_DESCRIPTION(DRV_DESCRIPTION ", v" DRV_VERSION);
5137 MODULE_AUTHOR("Thomas Davis, tadavis@lbl.gov and many others");
5138 MODULE_ALIAS_RTNL_LINK("bond");
Linux kernel - Deliverables
This page took 0.148051 seconds and 6 git commands to generate.