projects / deliverable / linux.git / blob
? search:


ea6b9043a5e72005b6e4c6251117cdb13bbba451
[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 #include <linux/kernel.h>
35 #include <linux/module.h>
36 #include <linux/types.h>
37 #include <linux/fcntl.h>
38 #include <linux/interrupt.h>
39 #include <linux/ptrace.h>
40 #include <linux/ioport.h>
41 #include <linux/in.h>
42 #include <net/ip.h>
43 #include <linux/ip.h>
44 #include <linux/tcp.h>
45 #include <linux/udp.h>
46 #include <linux/slab.h>
47 #include <linux/string.h>
48 #include <linux/init.h>
49 #include <linux/timer.h>
50 #include <linux/socket.h>
51 #include <linux/ctype.h>
52 #include <linux/inet.h>
53 #include <linux/bitops.h>
54 #include <linux/io.h>
55 #include <asm/system.h>
56 #include <asm/dma.h>
57 #include <linux/uaccess.h>
58 #include <linux/errno.h>
59 #include <linux/netdevice.h>
60 #include <linux/inetdevice.h>
61 #include <linux/igmp.h>
62 #include <linux/etherdevice.h>
63 #include <linux/skbuff.h>
64 #include <net/sock.h>
65 #include <linux/rtnetlink.h>
66 #include <linux/proc_fs.h>
67 #include <linux/seq_file.h>
68 #include <linux/smp.h>
69 #include <linux/if_ether.h>
70 #include <net/arp.h>
71 #include <linux/mii.h>
72 #include <linux/ethtool.h>
73 #include <linux/if_vlan.h>
74 #include <linux/if_bonding.h>
75 #include <linux/jiffies.h>
76 #include <net/route.h>
77 #include <net/net_namespace.h>
78 #include "bonding.h"
79 #include "bond_3ad.h"
80 #include "bond_alb.h"
81
82 /*---------------------------- Module parameters ----------------------------*/
83
84 /* monitor all links that often (in milliseconds). <=0 disables monitoring */
85 #define BOND_LINK_MON_INTERV 0
86 #define BOND_LINK_ARP_INTERV 0
87
88 static int max_bonds = BOND_DEFAULT_MAX_BONDS;
89 static int num_grat_arp = 1;
90 static int num_unsol_na = 1;
91 static int miimon = BOND_LINK_MON_INTERV;
92 static int updelay;
93 static int downdelay;
94 static int use_carrier = 1;
95 static char *mode;
96 static char *primary;
97 static char *lacp_rate;
98 static char *ad_select;
99 static char *xmit_hash_policy;
100 static int arp_interval = BOND_LINK_ARP_INTERV;
101 static char *arp_ip_target[BOND_MAX_ARP_TARGETS];
102 static char *arp_validate;
103 static char *fail_over_mac;
104 static struct bond_params bonding_defaults;
105
106 module_param(max_bonds, int, 0);
107 MODULE_PARM_DESC(max_bonds, "Max number of bonded devices");
108 module_param(num_grat_arp, int, 0644);
109 MODULE_PARM_DESC(num_grat_arp, "Number of gratuitous ARP packets to send on failover event");
110 module_param(num_unsol_na, int, 0644);
111 MODULE_PARM_DESC(num_unsol_na, "Number of unsolicited IPv6 Neighbor Advertisements packets to send on failover event");
112 module_param(miimon, int, 0);
113 MODULE_PARM_DESC(miimon, "Link check interval in milliseconds");
114 module_param(updelay, int, 0);
115 MODULE_PARM_DESC(updelay, "Delay before considering link up, in milliseconds");
116 module_param(downdelay, int, 0);
117 MODULE_PARM_DESC(downdelay, "Delay before considering link down, "
118 "in milliseconds");
119 module_param(use_carrier, int, 0);
120 MODULE_PARM_DESC(use_carrier, "Use netif_carrier_ok (vs MII ioctls) in miimon; "
121 "0 for off, 1 for on (default)");
122 module_param(mode, charp, 0);
123 MODULE_PARM_DESC(mode, "Mode of operation : 0 for balance-rr, "
124 "1 for active-backup, 2 for balance-xor, "
125 "3 for broadcast, 4 for 802.3ad, 5 for balance-tlb, "
126 "6 for balance-alb");
127 module_param(primary, charp, 0);
128 MODULE_PARM_DESC(primary, "Primary network device to use");
129 module_param(lacp_rate, charp, 0);
130 MODULE_PARM_DESC(lacp_rate, "LACPDU tx rate to request from 802.3ad partner "
131 "(slow/fast)");
132 module_param(ad_select, charp, 0);
133 MODULE_PARM_DESC(ad_select, "803.ad aggregation selection logic: stable (0, default), bandwidth (1), count (2)");
134 module_param(xmit_hash_policy, charp, 0);
135 MODULE_PARM_DESC(xmit_hash_policy, "XOR hashing method: 0 for layer 2 (default)"
136 ", 1 for layer 3+4");
137 module_param(arp_interval, int, 0);
138 MODULE_PARM_DESC(arp_interval, "arp interval in milliseconds");
139 module_param_array(arp_ip_target, charp, NULL, 0);
140 MODULE_PARM_DESC(arp_ip_target, "arp targets in n.n.n.n form");
141 module_param(arp_validate, charp, 0);
142 MODULE_PARM_DESC(arp_validate, "validate src/dst of ARP probes: none (default), active, backup or all");
143 module_param(fail_over_mac, charp, 0);
144 MODULE_PARM_DESC(fail_over_mac, "For active-backup, do not set all slaves to the same MAC. none (default), active or follow");
145
146 /*----------------------------- Global variables ----------------------------*/
147
148 static const char * const version =
149 DRV_DESCRIPTION ": v" DRV_VERSION " (" DRV_RELDATE ")\n";
150
151 LIST_HEAD(bond_dev_list);
152
153 #ifdef CONFIG_PROC_FS
154 static struct proc_dir_entry *bond_proc_dir;
155 #endif
156
157 static __be32 arp_target[BOND_MAX_ARP_TARGETS];
158 static int arp_ip_count;
159 static int bond_mode = BOND_MODE_ROUNDROBIN;
160 static int xmit_hashtype = BOND_XMIT_POLICY_LAYER2;
161 static int lacp_fast;
162
163
164 const struct bond_parm_tbl bond_lacp_tbl[] = {
165 { "slow", AD_LACP_SLOW},
166 { "fast", AD_LACP_FAST},
167 { NULL, -1},
168 };
169
170 const struct bond_parm_tbl bond_mode_tbl[] = {
171 { "balance-rr", BOND_MODE_ROUNDROBIN},
172 { "active-backup", BOND_MODE_ACTIVEBACKUP},
173 { "balance-xor", BOND_MODE_XOR},
174 { "broadcast", BOND_MODE_BROADCAST},
175 { "802.3ad", BOND_MODE_8023AD},
176 { "balance-tlb", BOND_MODE_TLB},
177 { "balance-alb", BOND_MODE_ALB},
178 { NULL, -1},
179 };
180
181 const struct bond_parm_tbl xmit_hashtype_tbl[] = {
182 { "layer2", BOND_XMIT_POLICY_LAYER2},
183 { "layer3+4", BOND_XMIT_POLICY_LAYER34},
184 { "layer2+3", BOND_XMIT_POLICY_LAYER23},
185 { NULL, -1},
186 };
187
188 const struct bond_parm_tbl arp_validate_tbl[] = {
189 { "none", BOND_ARP_VALIDATE_NONE},
190 { "active", BOND_ARP_VALIDATE_ACTIVE},
191 { "backup", BOND_ARP_VALIDATE_BACKUP},
192 { "all", BOND_ARP_VALIDATE_ALL},
193 { NULL, -1},
194 };
195
196 const struct bond_parm_tbl fail_over_mac_tbl[] = {
197 { "none", BOND_FOM_NONE},
198 { "active", BOND_FOM_ACTIVE},
199 { "follow", BOND_FOM_FOLLOW},
200 { NULL, -1},
201 };
202
203 struct bond_parm_tbl ad_select_tbl[] = {
204 { "stable", BOND_AD_STABLE},
205 { "bandwidth", BOND_AD_BANDWIDTH},
206 { "count", BOND_AD_COUNT},
207 { NULL, -1},
208 };
209
210 /*-------------------------- Forward declarations ---------------------------*/
211
212 static void bond_send_gratuitous_arp(struct bonding *bond);
213 static void bond_deinit(struct net_device *bond_dev);
214
215 /*---------------------------- General routines -----------------------------*/
216
217 static const char *bond_mode_name(int mode)
218 {
219 static const char *names[] = {
220 [BOND_MODE_ROUNDROBIN] = "load balancing (round-robin)",
221 [BOND_MODE_ACTIVEBACKUP] = "fault-tolerance (active-backup)",
222 [BOND_MODE_XOR] = "load balancing (xor)",
223 [BOND_MODE_BROADCAST] = "fault-tolerance (broadcast)",
224 [BOND_MODE_8023AD] = "IEEE 802.3ad Dynamic link aggregation",
225 [BOND_MODE_TLB] = "transmit load balancing",
226 [BOND_MODE_ALB] = "adaptive load balancing",
227 };
228
229 if (mode < 0 || mode > BOND_MODE_ALB)
230 return "unknown";
231
232 return names[mode];
233 }
234
235 /*---------------------------------- VLAN -----------------------------------*/
236
237 /**
238 * bond_add_vlan - add a new vlan id on bond
239 * @bond: bond that got the notification
240 * @vlan_id: the vlan id to add
241 *
242 * Returns -ENOMEM if allocation failed.
243 */
244 static int bond_add_vlan(struct bonding *bond, unsigned short vlan_id)
245 {
246 struct vlan_entry *vlan;
247
248 pr_debug("bond: %s, vlan id %d\n",
249 (bond ? bond->dev->name : "None"), vlan_id);
250
251 vlan = kzalloc(sizeof(struct vlan_entry), GFP_KERNEL);
252 if (!vlan)
253 return -ENOMEM;
254
255 INIT_LIST_HEAD(&vlan->vlan_list);
256 vlan->vlan_id = vlan_id;
257
258 write_lock_bh(&bond->lock);
259
260 list_add_tail(&vlan->vlan_list, &bond->vlan_list);
261
262 write_unlock_bh(&bond->lock);
263
264 pr_debug("added VLAN ID %d on bond %s\n", vlan_id, bond->dev->name);
265
266 return 0;
267 }
268
269 /**
270 * bond_del_vlan - delete a vlan id from bond
271 * @bond: bond that got the notification
272 * @vlan_id: the vlan id to delete
273 *
274 * returns -ENODEV if @vlan_id was not found in @bond.
275 */
276 static int bond_del_vlan(struct bonding *bond, unsigned short vlan_id)
277 {
278 struct vlan_entry *vlan;
279 int res = -ENODEV;
280
281 pr_debug("bond: %s, vlan id %d\n", bond->dev->name, vlan_id);
282
283 write_lock_bh(&bond->lock);
284
285 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
286 if (vlan->vlan_id == vlan_id) {
287 list_del(&vlan->vlan_list);
288
289 if (bond_is_lb(bond))
290 bond_alb_clear_vlan(bond, vlan_id);
291
292 pr_debug("removed VLAN ID %d from bond %s\n", vlan_id,
293 bond->dev->name);
294
295 kfree(vlan);
296
297 if (list_empty(&bond->vlan_list) &&
298 (bond->slave_cnt == 0)) {
299 /* Last VLAN removed and no slaves, so
300 * restore block on adding VLANs. This will
301 * be removed once new slaves that are not
302 * VLAN challenged will be added.
303 */
304 bond->dev->features |= NETIF_F_VLAN_CHALLENGED;
305 }
306
307 res = 0;
308 goto out;
309 }
310 }
311
312 pr_debug("couldn't find VLAN ID %d in bond %s\n", vlan_id,
313 bond->dev->name);
314
315 out:
316 write_unlock_bh(&bond->lock);
317 return res;
318 }
319
320 /**
321 * bond_has_challenged_slaves
322 * @bond: the bond we're working on
323 *
324 * Searches the slave list. Returns 1 if a vlan challenged slave
325 * was found, 0 otherwise.
326 *
327 * Assumes bond->lock is held.
328 */
329 static int bond_has_challenged_slaves(struct bonding *bond)
330 {
331 struct slave *slave;
332 int i;
333
334 bond_for_each_slave(bond, slave, i) {
335 if (slave->dev->features & NETIF_F_VLAN_CHALLENGED) {
336 pr_debug("found VLAN challenged slave - %s\n",
337 slave->dev->name);
338 return 1;
339 }
340 }
341
342 pr_debug("no VLAN challenged slaves found\n");
343 return 0;
344 }
345
346 /**
347 * bond_next_vlan - safely skip to the next item in the vlans list.
348 * @bond: the bond we're working on
349 * @curr: item we're advancing from
350 *
351 * Returns %NULL if list is empty, bond->next_vlan if @curr is %NULL,
352 * or @curr->next otherwise (even if it is @curr itself again).
353 *
354 * Caller must hold bond->lock
355 */
356 struct vlan_entry *bond_next_vlan(struct bonding *bond, struct vlan_entry *curr)
357 {
358 struct vlan_entry *next, *last;
359
360 if (list_empty(&bond->vlan_list))
361 return NULL;
362
363 if (!curr) {
364 next = list_entry(bond->vlan_list.next,
365 struct vlan_entry, vlan_list);
366 } else {
367 last = list_entry(bond->vlan_list.prev,
368 struct vlan_entry, vlan_list);
369 if (last == curr) {
370 next = list_entry(bond->vlan_list.next,
371 struct vlan_entry, vlan_list);
372 } else {
373 next = list_entry(curr->vlan_list.next,
374 struct vlan_entry, vlan_list);
375 }
376 }
377
378 return next;
379 }
380
381 /**
382 * bond_dev_queue_xmit - Prepare skb for xmit.
383 *
384 * @bond: bond device that got this skb for tx.
385 * @skb: hw accel VLAN tagged skb to transmit
386 * @slave_dev: slave that is supposed to xmit this skbuff
387 *
388 * When the bond gets an skb to transmit that is
389 * already hardware accelerated VLAN tagged, and it
390 * needs to relay this skb to a slave that is not
391 * hw accel capable, the skb needs to be "unaccelerated",
392 * i.e. strip the hwaccel tag and re-insert it as part
393 * of the payload.
394 */
395 int bond_dev_queue_xmit(struct bonding *bond, struct sk_buff *skb,
396 struct net_device *slave_dev)
397 {
398 unsigned short uninitialized_var(vlan_id);
399
400 if (!list_empty(&bond->vlan_list) &&
401 !(slave_dev->features & NETIF_F_HW_VLAN_TX) &&
402 vlan_get_tag(skb, &vlan_id) == 0) {
403 skb->dev = slave_dev;
404 skb = vlan_put_tag(skb, vlan_id);
405 if (!skb) {
406 /* vlan_put_tag() frees the skb in case of error,
407 * so return success here so the calling functions
408 * won't attempt to free is again.
409 */
410 return 0;
411 }
412 } else {
413 skb->dev = slave_dev;
414 }
415
416 skb->priority = 1;
417 dev_queue_xmit(skb);
418
419 return 0;
420 }
421
422 /*
423 * In the following 3 functions, bond_vlan_rx_register(), bond_vlan_rx_add_vid
424 * and bond_vlan_rx_kill_vid, We don't protect the slave list iteration with a
425 * lock because:
426 * a. This operation is performed in IOCTL context,
427 * b. The operation is protected by the RTNL semaphore in the 8021q code,
428 * c. Holding a lock with BH disabled while directly calling a base driver
429 * entry point is generally a BAD idea.
430 *
431 * The design of synchronization/protection for this operation in the 8021q
432 * module is good for one or more VLAN devices over a single physical device
433 * and cannot be extended for a teaming solution like bonding, so there is a
434 * potential race condition here where a net device from the vlan group might
435 * be referenced (either by a base driver or the 8021q code) while it is being
436 * removed from the system. However, it turns out we're not making matters
437 * worse, and if it works for regular VLAN usage it will work here too.
438 */
439
440 /**
441 * bond_vlan_rx_register - Propagates registration to slaves
442 * @bond_dev: bonding net device that got called
443 * @grp: vlan group being registered
444 */
445 static void bond_vlan_rx_register(struct net_device *bond_dev,
446 struct vlan_group *grp)
447 {
448 struct bonding *bond = netdev_priv(bond_dev);
449 struct slave *slave;
450 int i;
451
452 bond->vlgrp = grp;
453
454 bond_for_each_slave(bond, slave, i) {
455 struct net_device *slave_dev = slave->dev;
456 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
457
458 if ((slave_dev->features & NETIF_F_HW_VLAN_RX) &&
459 slave_ops->ndo_vlan_rx_register) {
460 slave_ops->ndo_vlan_rx_register(slave_dev, grp);
461 }
462 }
463 }
464
465 /**
466 * bond_vlan_rx_add_vid - Propagates adding an id to slaves
467 * @bond_dev: bonding net device that got called
468 * @vid: vlan id being added
469 */
470 static void bond_vlan_rx_add_vid(struct net_device *bond_dev, uint16_t vid)
471 {
472 struct bonding *bond = netdev_priv(bond_dev);
473 struct slave *slave;
474 int i, res;
475
476 bond_for_each_slave(bond, slave, i) {
477 struct net_device *slave_dev = slave->dev;
478 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
479
480 if ((slave_dev->features & NETIF_F_HW_VLAN_FILTER) &&
481 slave_ops->ndo_vlan_rx_add_vid) {
482 slave_ops->ndo_vlan_rx_add_vid(slave_dev, vid);
483 }
484 }
485
486 res = bond_add_vlan(bond, vid);
487 if (res) {
488 pr_err(DRV_NAME
489 ": %s: Error: Failed to add vlan id %d\n",
490 bond_dev->name, vid);
491 }
492 }
493
494 /**
495 * bond_vlan_rx_kill_vid - Propagates deleting an id to slaves
496 * @bond_dev: bonding net device that got called
497 * @vid: vlan id being removed
498 */
499 static void bond_vlan_rx_kill_vid(struct net_device *bond_dev, uint16_t vid)
500 {
501 struct bonding *bond = netdev_priv(bond_dev);
502 struct slave *slave;
503 struct net_device *vlan_dev;
504 int i, res;
505
506 bond_for_each_slave(bond, slave, i) {
507 struct net_device *slave_dev = slave->dev;
508 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
509
510 if ((slave_dev->features & NETIF_F_HW_VLAN_FILTER) &&
511 slave_ops->ndo_vlan_rx_kill_vid) {
512 /* Save and then restore vlan_dev in the grp array,
513 * since the slave's driver might clear it.
514 */
515 vlan_dev = vlan_group_get_device(bond->vlgrp, vid);
516 slave_ops->ndo_vlan_rx_kill_vid(slave_dev, vid);
517 vlan_group_set_device(bond->vlgrp, vid, vlan_dev);
518 }
519 }
520
521 res = bond_del_vlan(bond, vid);
522 if (res) {
523 pr_err(DRV_NAME
524 ": %s: Error: Failed to remove vlan id %d\n",
525 bond_dev->name, vid);
526 }
527 }
528
529 static void bond_add_vlans_on_slave(struct bonding *bond, struct net_device *slave_dev)
530 {
531 struct vlan_entry *vlan;
532 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
533
534 write_lock_bh(&bond->lock);
535
536 if (list_empty(&bond->vlan_list))
537 goto out;
538
539 if ((slave_dev->features & NETIF_F_HW_VLAN_RX) &&
540 slave_ops->ndo_vlan_rx_register)
541 slave_ops->ndo_vlan_rx_register(slave_dev, bond->vlgrp);
542
543 if (!(slave_dev->features & NETIF_F_HW_VLAN_FILTER) ||
544 !(slave_ops->ndo_vlan_rx_add_vid))
545 goto out;
546
547 list_for_each_entry(vlan, &bond->vlan_list, vlan_list)
548 slave_ops->ndo_vlan_rx_add_vid(slave_dev, vlan->vlan_id);
549
550 out:
551 write_unlock_bh(&bond->lock);
552 }
553
554 static void bond_del_vlans_from_slave(struct bonding *bond,
555 struct net_device *slave_dev)
556 {
557 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
558 struct vlan_entry *vlan;
559 struct net_device *vlan_dev;
560
561 write_lock_bh(&bond->lock);
562
563 if (list_empty(&bond->vlan_list))
564 goto out;
565
566 if (!(slave_dev->features & NETIF_F_HW_VLAN_FILTER) ||
567 !(slave_ops->ndo_vlan_rx_kill_vid))
568 goto unreg;
569
570 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
571 /* Save and then restore vlan_dev in the grp array,
572 * since the slave's driver might clear it.
573 */
574 vlan_dev = vlan_group_get_device(bond->vlgrp, vlan->vlan_id);
575 slave_ops->ndo_vlan_rx_kill_vid(slave_dev, vlan->vlan_id);
576 vlan_group_set_device(bond->vlgrp, vlan->vlan_id, vlan_dev);
577 }
578
579 unreg:
580 if ((slave_dev->features & NETIF_F_HW_VLAN_RX) &&
581 slave_ops->ndo_vlan_rx_register)
582 slave_ops->ndo_vlan_rx_register(slave_dev, NULL);
583
584 out:
585 write_unlock_bh(&bond->lock);
586 }
587
588 /*------------------------------- Link status -------------------------------*/
589
590 /*
591 * Set the carrier state for the master according to the state of its
592 * slaves. If any slaves are up, the master is up. In 802.3ad mode,
593 * do special 802.3ad magic.
594 *
595 * Returns zero if carrier state does not change, nonzero if it does.
596 */
597 static int bond_set_carrier(struct bonding *bond)
598 {
599 struct slave *slave;
600 int i;
601
602 if (bond->slave_cnt == 0)
603 goto down;
604
605 if (bond->params.mode == BOND_MODE_8023AD)
606 return bond_3ad_set_carrier(bond);
607
608 bond_for_each_slave(bond, slave, i) {
609 if (slave->link == BOND_LINK_UP) {
610 if (!netif_carrier_ok(bond->dev)) {
611 netif_carrier_on(bond->dev);
612 return 1;
613 }
614 return 0;
615 }
616 }
617
618 down:
619 if (netif_carrier_ok(bond->dev)) {
620 netif_carrier_off(bond->dev);
621 return 1;
622 }
623 return 0;
624 }
625
626 /*
627 * Get link speed and duplex from the slave's base driver
628 * using ethtool. If for some reason the call fails or the
629 * values are invalid, fake speed and duplex to 100/Full
630 * and return error.
631 */
632 static int bond_update_speed_duplex(struct slave *slave)
633 {
634 struct net_device *slave_dev = slave->dev;
635 struct ethtool_cmd etool;
636 int res;
637
638 /* Fake speed and duplex */
639 slave->speed = SPEED_100;
640 slave->duplex = DUPLEX_FULL;
641
642 if (!slave_dev->ethtool_ops || !slave_dev->ethtool_ops->get_settings)
643 return -1;
644
645 res = slave_dev->ethtool_ops->get_settings(slave_dev, &etool);
646 if (res < 0)
647 return -1;
648
649 switch (etool.speed) {
650 case SPEED_10:
651 case SPEED_100:
652 case SPEED_1000:
653 case SPEED_10000:
654 break;
655 default:
656 return -1;
657 }
658
659 switch (etool.duplex) {
660 case DUPLEX_FULL:
661 case DUPLEX_HALF:
662 break;
663 default:
664 return -1;
665 }
666
667 slave->speed = etool.speed;
668 slave->duplex = etool.duplex;
669
670 return 0;
671 }
672
673 /*
674 * if <dev> supports MII link status reporting, check its link status.
675 *
676 * We either do MII/ETHTOOL ioctls, or check netif_carrier_ok(),
677 * depending upon the setting of the use_carrier parameter.
678 *
679 * Return either BMSR_LSTATUS, meaning that the link is up (or we
680 * can't tell and just pretend it is), or 0, meaning that the link is
681 * down.
682 *
683 * If reporting is non-zero, instead of faking link up, return -1 if
684 * both ETHTOOL and MII ioctls fail (meaning the device does not
685 * support them). If use_carrier is set, return whatever it says.
686 * It'd be nice if there was a good way to tell if a driver supports
687 * netif_carrier, but there really isn't.
688 */
689 static int bond_check_dev_link(struct bonding *bond,
690 struct net_device *slave_dev, int reporting)
691 {
692 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
693 static int (*ioctl)(struct net_device *, struct ifreq *, int);
694 struct ifreq ifr;
695 struct mii_ioctl_data *mii;
696
697 if (bond->params.use_carrier)
698 return netif_carrier_ok(slave_dev) ? BMSR_LSTATUS : 0;
699
700 /* Try to get link status using Ethtool first. */
701 if (slave_dev->ethtool_ops) {
702 if (slave_dev->ethtool_ops->get_link) {
703 u32 link;
704
705 link = slave_dev->ethtool_ops->get_link(slave_dev);
706
707 return link ? BMSR_LSTATUS : 0;
708 }
709 }
710
711 /* Ethtool can't be used, fallback to MII ioctls. */
712 ioctl = slave_ops->ndo_do_ioctl;
713 if (ioctl) {
714 /* TODO: set pointer to correct ioctl on a per team member */
715 /* bases to make this more efficient. that is, once */
716 /* we determine the correct ioctl, we will always */
717 /* call it and not the others for that team */
718 /* member. */
719
720 /*
721 * We cannot assume that SIOCGMIIPHY will also read a
722 * register; not all network drivers (e.g., e100)
723 * support that.
724 */
725
726 /* Yes, the mii is overlaid on the ifreq.ifr_ifru */
727 strncpy(ifr.ifr_name, slave_dev->name, IFNAMSIZ);
728 mii = if_mii(&ifr);
729 if (IOCTL(slave_dev, &ifr, SIOCGMIIPHY) == 0) {
730 mii->reg_num = MII_BMSR;
731 if (IOCTL(slave_dev, &ifr, SIOCGMIIREG) == 0)
732 return mii->val_out & BMSR_LSTATUS;
733 }
734 }
735
736 /*
737 * If reporting, report that either there's no dev->do_ioctl,
738 * or both SIOCGMIIREG and get_link failed (meaning that we
739 * cannot report link status). If not reporting, pretend
740 * we're ok.
741 */
742 return reporting ? -1 : BMSR_LSTATUS;
743 }
744
745 /*----------------------------- Multicast list ------------------------------*/
746
747 /*
748 * Returns 0 if dmi1 and dmi2 are the same, non-0 otherwise
749 */
750 static inline int bond_is_dmi_same(const struct dev_mc_list *dmi1,
751 const struct dev_mc_list *dmi2)
752 {
753 return memcmp(dmi1->dmi_addr, dmi2->dmi_addr, dmi1->dmi_addrlen) == 0 &&
754 dmi1->dmi_addrlen == dmi2->dmi_addrlen;
755 }
756
757 /*
758 * returns dmi entry if found, NULL otherwise
759 */
760 static struct dev_mc_list *bond_mc_list_find_dmi(struct dev_mc_list *dmi,
761 struct dev_mc_list *mc_list)
762 {
763 struct dev_mc_list *idmi;
764
765 for (idmi = mc_list; idmi; idmi = idmi->next) {
766 if (bond_is_dmi_same(dmi, idmi))
767 return idmi;
768 }
769
770 return NULL;
771 }
772
773 /*
774 * Push the promiscuity flag down to appropriate slaves
775 */
776 static int bond_set_promiscuity(struct bonding *bond, int inc)
777 {
778 int err = 0;
779 if (USES_PRIMARY(bond->params.mode)) {
780 /* write lock already acquired */
781 if (bond->curr_active_slave) {
782 err = dev_set_promiscuity(bond->curr_active_slave->dev,
783 inc);
784 }
785 } else {
786 struct slave *slave;
787 int i;
788 bond_for_each_slave(bond, slave, i) {
789 err = dev_set_promiscuity(slave->dev, inc);
790 if (err)
791 return err;
792 }
793 }
794 return err;
795 }
796
797 /*
798 * Push the allmulti flag down to all slaves
799 */
800 static int bond_set_allmulti(struct bonding *bond, int inc)
801 {
802 int err = 0;
803 if (USES_PRIMARY(bond->params.mode)) {
804 /* write lock already acquired */
805 if (bond->curr_active_slave) {
806 err = dev_set_allmulti(bond->curr_active_slave->dev,
807 inc);
808 }
809 } else {
810 struct slave *slave;
811 int i;
812 bond_for_each_slave(bond, slave, i) {
813 err = dev_set_allmulti(slave->dev, inc);
814 if (err)
815 return err;
816 }
817 }
818 return err;
819 }
820
821 /*
822 * Add a Multicast address to slaves
823 * according to mode
824 */
825 static void bond_mc_add(struct bonding *bond, void *addr, int alen)
826 {
827 if (USES_PRIMARY(bond->params.mode)) {
828 /* write lock already acquired */
829 if (bond->curr_active_slave)
830 dev_mc_add(bond->curr_active_slave->dev, addr, alen, 0);
831 } else {
832 struct slave *slave;
833 int i;
834
835 bond_for_each_slave(bond, slave, i)
836 dev_mc_add(slave->dev, addr, alen, 0);
837 }
838 }
839
840 /*
841 * Remove a multicast address from slave
842 * according to mode
843 */
844 static void bond_mc_delete(struct bonding *bond, void *addr, int alen)
845 {
846 if (USES_PRIMARY(bond->params.mode)) {
847 /* write lock already acquired */
848 if (bond->curr_active_slave)
849 dev_mc_delete(bond->curr_active_slave->dev, addr,
850 alen, 0);
851 } else {
852 struct slave *slave;
853 int i;
854 bond_for_each_slave(bond, slave, i) {
855 dev_mc_delete(slave->dev, addr, alen, 0);
856 }
857 }
858 }
859
860
861 /*
862 * Retrieve the list of registered multicast addresses for the bonding
863 * device and retransmit an IGMP JOIN request to the current active
864 * slave.
865 */
866 static void bond_resend_igmp_join_requests(struct bonding *bond)
867 {
868 struct in_device *in_dev;
869 struct ip_mc_list *im;
870
871 rcu_read_lock();
872 in_dev = __in_dev_get_rcu(bond->dev);
873 if (in_dev) {
874 for (im = in_dev->mc_list; im; im = im->next)
875 ip_mc_rejoin_group(im);
876 }
877
878 rcu_read_unlock();
879 }
880
881 /*
882 * Totally destroys the mc_list in bond
883 */
884 static void bond_mc_list_destroy(struct bonding *bond)
885 {
886 struct dev_mc_list *dmi;
887
888 dmi = bond->mc_list;
889 while (dmi) {
890 bond->mc_list = dmi->next;
891 kfree(dmi);
892 dmi = bond->mc_list;
893 }
894
895 bond->mc_list = NULL;
896 }
897
898 /*
899 * Copy all the Multicast addresses from src to the bonding device dst
900 */
901 static int bond_mc_list_copy(struct dev_mc_list *mc_list, struct bonding *bond,
902 gfp_t gfp_flag)
903 {
904 struct dev_mc_list *dmi, *new_dmi;
905
906 for (dmi = mc_list; dmi; dmi = dmi->next) {
907 new_dmi = kmalloc(sizeof(struct dev_mc_list), gfp_flag);
908
909 if (!new_dmi) {
910 /* FIXME: Potential memory leak !!! */
911 return -ENOMEM;
912 }
913
914 new_dmi->next = bond->mc_list;
915 bond->mc_list = new_dmi;
916 new_dmi->dmi_addrlen = dmi->dmi_addrlen;
917 memcpy(new_dmi->dmi_addr, dmi->dmi_addr, dmi->dmi_addrlen);
918 new_dmi->dmi_users = dmi->dmi_users;
919 new_dmi->dmi_gusers = dmi->dmi_gusers;
920 }
921
922 return 0;
923 }
924
925 /*
926 * flush all members of flush->mc_list from device dev->mc_list
927 */
928 static void bond_mc_list_flush(struct net_device *bond_dev,
929 struct net_device *slave_dev)
930 {
931 struct bonding *bond = netdev_priv(bond_dev);
932 struct dev_mc_list *dmi;
933
934 for (dmi = bond_dev->mc_list; dmi; dmi = dmi->next)
935 dev_mc_delete(slave_dev, dmi->dmi_addr, dmi->dmi_addrlen, 0);
936
937 if (bond->params.mode == BOND_MODE_8023AD) {
938 /* del lacpdu mc addr from mc list */
939 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
940
941 dev_mc_delete(slave_dev, lacpdu_multicast, ETH_ALEN, 0);
942 }
943 }
944
945 /*--------------------------- Active slave change ---------------------------*/
946
947 /*
948 * Update the mc list and multicast-related flags for the new and
949 * old active slaves (if any) according to the multicast mode, and
950 * promiscuous flags unconditionally.
951 */
952 static void bond_mc_swap(struct bonding *bond, struct slave *new_active,
953 struct slave *old_active)
954 {
955 struct dev_mc_list *dmi;
956
957 if (!USES_PRIMARY(bond->params.mode))
958 /* nothing to do - mc list is already up-to-date on
959 * all slaves
960 */
961 return;
962
963 if (old_active) {
964 if (bond->dev->flags & IFF_PROMISC)
965 dev_set_promiscuity(old_active->dev, -1);
966
967 if (bond->dev->flags & IFF_ALLMULTI)
968 dev_set_allmulti(old_active->dev, -1);
969
970 for (dmi = bond->dev->mc_list; dmi; dmi = dmi->next)
971 dev_mc_delete(old_active->dev, dmi->dmi_addr,
972 dmi->dmi_addrlen, 0);
973 }
974
975 if (new_active) {
976 /* FIXME: Signal errors upstream. */
977 if (bond->dev->flags & IFF_PROMISC)
978 dev_set_promiscuity(new_active->dev, 1);
979
980 if (bond->dev->flags & IFF_ALLMULTI)
981 dev_set_allmulti(new_active->dev, 1);
982
983 for (dmi = bond->dev->mc_list; dmi; dmi = dmi->next)
984 dev_mc_add(new_active->dev, dmi->dmi_addr,
985 dmi->dmi_addrlen, 0);
986 bond_resend_igmp_join_requests(bond);
987 }
988 }
989
990 /*
991 * bond_do_fail_over_mac
992 *
993 * Perform special MAC address swapping for fail_over_mac settings
994 *
995 * Called with RTNL, bond->lock for read, curr_slave_lock for write_bh.
996 */
997 static void bond_do_fail_over_mac(struct bonding *bond,
998 struct slave *new_active,
999 struct slave *old_active)
1000 __releases(&bond->curr_slave_lock)
1001 __releases(&bond->lock)
1002 __acquires(&bond->lock)
1003 __acquires(&bond->curr_slave_lock)
1004 {
1005 u8 tmp_mac[ETH_ALEN];
1006 struct sockaddr saddr;
1007 int rv;
1008
1009 switch (bond->params.fail_over_mac) {
1010 case BOND_FOM_ACTIVE:
1011 if (new_active)
1012 memcpy(bond->dev->dev_addr, new_active->dev->dev_addr,
1013 new_active->dev->addr_len);
1014 break;
1015 case BOND_FOM_FOLLOW:
1016 /*
1017 * if new_active && old_active, swap them
1018 * if just old_active, do nothing (going to no active slave)
1019 * if just new_active, set new_active to bond's MAC
1020 */
1021 if (!new_active)
1022 return;
1023
1024 write_unlock_bh(&bond->curr_slave_lock);
1025 read_unlock(&bond->lock);
1026
1027 if (old_active) {
1028 memcpy(tmp_mac, new_active->dev->dev_addr, ETH_ALEN);
1029 memcpy(saddr.sa_data, old_active->dev->dev_addr,
1030 ETH_ALEN);
1031 saddr.sa_family = new_active->dev->type;
1032 } else {
1033 memcpy(saddr.sa_data, bond->dev->dev_addr, ETH_ALEN);
1034 saddr.sa_family = bond->dev->type;
1035 }
1036
1037 rv = dev_set_mac_address(new_active->dev, &saddr);
1038 if (rv) {
1039 pr_err(DRV_NAME
1040 ": %s: Error %d setting MAC of slave %s\n",
1041 bond->dev->name, -rv, new_active->dev->name);
1042 goto out;
1043 }
1044
1045 if (!old_active)
1046 goto out;
1047
1048 memcpy(saddr.sa_data, tmp_mac, ETH_ALEN);
1049 saddr.sa_family = old_active->dev->type;
1050
1051 rv = dev_set_mac_address(old_active->dev, &saddr);
1052 if (rv)
1053 pr_err(DRV_NAME
1054 ": %s: Error %d setting MAC of slave %s\n",
1055 bond->dev->name, -rv, new_active->dev->name);
1056 out:
1057 read_lock(&bond->lock);
1058 write_lock_bh(&bond->curr_slave_lock);
1059 break;
1060 default:
1061 pr_err(DRV_NAME
1062 ": %s: bond_do_fail_over_mac impossible: bad policy %d\n",
1063 bond->dev->name, bond->params.fail_over_mac);
1064 break;
1065 }
1066
1067 }
1068
1069
1070 /**
1071 * find_best_interface - select the best available slave to be the active one
1072 * @bond: our bonding struct
1073 *
1074 * Warning: Caller must hold curr_slave_lock for writing.
1075 */
1076 static struct slave *bond_find_best_slave(struct bonding *bond)
1077 {
1078 struct slave *new_active, *old_active;
1079 struct slave *bestslave = NULL;
1080 int mintime = bond->params.updelay;
1081 int i;
1082
1083 new_active = old_active = bond->curr_active_slave;
1084
1085 if (!new_active) { /* there were no active slaves left */
1086 if (bond->slave_cnt > 0) /* found one slave */
1087 new_active = bond->first_slave;
1088 else
1089 return NULL; /* still no slave, return NULL */
1090 }
1091
1092 /*
1093 * first try the primary link; if arping, a link must tx/rx
1094 * traffic before it can be considered the curr_active_slave.
1095 * also, we would skip slaves between the curr_active_slave
1096 * and primary_slave that may be up and able to arp
1097 */
1098 if ((bond->primary_slave) &&
1099 (!bond->params.arp_interval) &&
1100 (IS_UP(bond->primary_slave->dev))) {
1101 new_active = bond->primary_slave;
1102 }
1103
1104 /* remember where to stop iterating over the slaves */
1105 old_active = new_active;
1106
1107 bond_for_each_slave_from(bond, new_active, i, old_active) {
1108 if (IS_UP(new_active->dev)) {
1109 if (new_active->link == BOND_LINK_UP) {
1110 return new_active;
1111 } else if (new_active->link == BOND_LINK_BACK) {
1112 /* link up, but waiting for stabilization */
1113 if (new_active->delay < mintime) {
1114 mintime = new_active->delay;
1115 bestslave = new_active;
1116 }
1117 }
1118 }
1119 }
1120
1121 return bestslave;
1122 }
1123
1124 /**
1125 * change_active_interface - change the active slave into the specified one
1126 * @bond: our bonding struct
1127 * @new: the new slave to make the active one
1128 *
1129 * Set the new slave to the bond's settings and unset them on the old
1130 * curr_active_slave.
1131 * Setting include flags, mc-list, promiscuity, allmulti, etc.
1132 *
1133 * If @new's link state is %BOND_LINK_BACK we'll set it to %BOND_LINK_UP,
1134 * because it is apparently the best available slave we have, even though its
1135 * updelay hasn't timed out yet.
1136 *
1137 * If new_active is not NULL, caller must hold bond->lock for read and
1138 * curr_slave_lock for write_bh.
1139 */
1140 void bond_change_active_slave(struct bonding *bond, struct slave *new_active)
1141 {
1142 struct slave *old_active = bond->curr_active_slave;
1143
1144 if (old_active == new_active)
1145 return;
1146
1147 if (new_active) {
1148 new_active->jiffies = jiffies;
1149
1150 if (new_active->link == BOND_LINK_BACK) {
1151 if (USES_PRIMARY(bond->params.mode)) {
1152 pr_info(DRV_NAME
1153 ": %s: making interface %s the new "
1154 "active one %d ms earlier.\n",
1155 bond->dev->name, new_active->dev->name,
1156 (bond->params.updelay - new_active->delay) * bond->params.miimon);
1157 }
1158
1159 new_active->delay = 0;
1160 new_active->link = BOND_LINK_UP;
1161
1162 if (bond->params.mode == BOND_MODE_8023AD)
1163 bond_3ad_handle_link_change(new_active, BOND_LINK_UP);
1164
1165 if (bond_is_lb(bond))
1166 bond_alb_handle_link_change(bond, new_active, BOND_LINK_UP);
1167 } else {
1168 if (USES_PRIMARY(bond->params.mode)) {
1169 pr_info(DRV_NAME
1170 ": %s: making interface %s the new "
1171 "active one.\n",
1172 bond->dev->name, new_active->dev->name);
1173 }
1174 }
1175 }
1176
1177 if (USES_PRIMARY(bond->params.mode))
1178 bond_mc_swap(bond, new_active, old_active);
1179
1180 if (bond_is_lb(bond)) {
1181 bond_alb_handle_active_change(bond, new_active);
1182 if (old_active)
1183 bond_set_slave_inactive_flags(old_active);
1184 if (new_active)
1185 bond_set_slave_active_flags(new_active);
1186 } else {
1187 bond->curr_active_slave = new_active;
1188 }
1189
1190 if (bond->params.mode == BOND_MODE_ACTIVEBACKUP) {
1191 if (old_active)
1192 bond_set_slave_inactive_flags(old_active);
1193
1194 if (new_active) {
1195 bond_set_slave_active_flags(new_active);
1196
1197 if (bond->params.fail_over_mac)
1198 bond_do_fail_over_mac(bond, new_active,
1199 old_active);
1200
1201 bond->send_grat_arp = bond->params.num_grat_arp;
1202 bond_send_gratuitous_arp(bond);
1203
1204 bond->send_unsol_na = bond->params.num_unsol_na;
1205 bond_send_unsolicited_na(bond);
1206
1207 write_unlock_bh(&bond->curr_slave_lock);
1208 read_unlock(&bond->lock);
1209
1210 netdev_bonding_change(bond->dev);
1211
1212 read_lock(&bond->lock);
1213 write_lock_bh(&bond->curr_slave_lock);
1214 }
1215 }
1216 }
1217
1218 /**
1219 * bond_select_active_slave - select a new active slave, if needed
1220 * @bond: our bonding struct
1221 *
1222 * This functions should be called when one of the following occurs:
1223 * - The old curr_active_slave has been released or lost its link.
1224 * - The primary_slave has got its link back.
1225 * - A slave has got its link back and there's no old curr_active_slave.
1226 *
1227 * Caller must hold bond->lock for read and curr_slave_lock for write_bh.
1228 */
1229 void bond_select_active_slave(struct bonding *bond)
1230 {
1231 struct slave *best_slave;
1232 int rv;
1233
1234 best_slave = bond_find_best_slave(bond);
1235 if (best_slave != bond->curr_active_slave) {
1236 bond_change_active_slave(bond, best_slave);
1237 rv = bond_set_carrier(bond);
1238 if (!rv)
1239 return;
1240
1241 if (netif_carrier_ok(bond->dev)) {
1242 pr_info(DRV_NAME
1243 ": %s: first active interface up!\n",
1244 bond->dev->name);
1245 } else {
1246 pr_info(DRV_NAME ": %s: "
1247 "now running without any active interface !\n",
1248 bond->dev->name);
1249 }
1250 }
1251 }
1252
1253 /*--------------------------- slave list handling ---------------------------*/
1254
1255 /*
1256 * This function attaches the slave to the end of list.
1257 *
1258 * bond->lock held for writing by caller.
1259 */
1260 static void bond_attach_slave(struct bonding *bond, struct slave *new_slave)
1261 {
1262 if (bond->first_slave == NULL) { /* attaching the first slave */
1263 new_slave->next = new_slave;
1264 new_slave->prev = new_slave;
1265 bond->first_slave = new_slave;
1266 } else {
1267 new_slave->next = bond->first_slave;
1268 new_slave->prev = bond->first_slave->prev;
1269 new_slave->next->prev = new_slave;
1270 new_slave->prev->next = new_slave;
1271 }
1272
1273 bond->slave_cnt++;
1274 }
1275
1276 /*
1277 * This function detaches the slave from the list.
1278 * WARNING: no check is made to verify if the slave effectively
1279 * belongs to <bond>.
1280 * Nothing is freed on return, structures are just unchained.
1281 * If any slave pointer in bond was pointing to <slave>,
1282 * it should be changed by the calling function.
1283 *
1284 * bond->lock held for writing by caller.
1285 */
1286 static void bond_detach_slave(struct bonding *bond, struct slave *slave)
1287 {
1288 if (slave->next)
1289 slave->next->prev = slave->prev;
1290
1291 if (slave->prev)
1292 slave->prev->next = slave->next;
1293
1294 if (bond->first_slave == slave) { /* slave is the first slave */
1295 if (bond->slave_cnt > 1) { /* there are more slave */
1296 bond->first_slave = slave->next;
1297 } else {
1298 bond->first_slave = NULL; /* slave was the last one */
1299 }
1300 }
1301
1302 slave->next = NULL;
1303 slave->prev = NULL;
1304 bond->slave_cnt--;
1305 }
1306
1307 /*---------------------------------- IOCTL ----------------------------------*/
1308
1309 static int bond_sethwaddr(struct net_device *bond_dev,
1310 struct net_device *slave_dev)
1311 {
1312 pr_debug("bond_dev=%p\n", bond_dev);
1313 pr_debug("slave_dev=%p\n", slave_dev);
1314 pr_debug("slave_dev->addr_len=%d\n", slave_dev->addr_len);
1315 memcpy(bond_dev->dev_addr, slave_dev->dev_addr, slave_dev->addr_len);
1316 return 0;
1317 }
1318
1319 #define BOND_VLAN_FEATURES \
1320 (NETIF_F_VLAN_CHALLENGED | NETIF_F_HW_VLAN_RX | NETIF_F_HW_VLAN_TX | \
1321 NETIF_F_HW_VLAN_FILTER)
1322
1323 /*
1324 * Compute the common dev->feature set available to all slaves. Some
1325 * feature bits are managed elsewhere, so preserve those feature bits
1326 * on the master device.
1327 */
1328 static int bond_compute_features(struct bonding *bond)
1329 {
1330 struct slave *slave;
1331 struct net_device *bond_dev = bond->dev;
1332 unsigned long features = bond_dev->features;
1333 unsigned short max_hard_header_len = max((u16)ETH_HLEN,
1334 bond_dev->hard_header_len);
1335 int i;
1336
1337 features &= ~(NETIF_F_ALL_CSUM | BOND_VLAN_FEATURES);
1338 features |= NETIF_F_GSO_MASK | NETIF_F_NO_CSUM;
1339
1340 if (!bond->first_slave)
1341 goto done;
1342
1343 features &= ~NETIF_F_ONE_FOR_ALL;
1344
1345 bond_for_each_slave(bond, slave, i) {
1346 features = netdev_increment_features(features,
1347 slave->dev->features,
1348 NETIF_F_ONE_FOR_ALL);
1349 if (slave->dev->hard_header_len > max_hard_header_len)
1350 max_hard_header_len = slave->dev->hard_header_len;
1351 }
1352
1353 done:
1354 features |= (bond_dev->features & BOND_VLAN_FEATURES);
1355 bond_dev->features = netdev_fix_features(features, NULL);
1356 bond_dev->hard_header_len = max_hard_header_len;
1357
1358 return 0;
1359 }
1360
1361 static void bond_setup_by_slave(struct net_device *bond_dev,
1362 struct net_device *slave_dev)
1363 {
1364 struct bonding *bond = netdev_priv(bond_dev);
1365
1366 bond_dev->header_ops = slave_dev->header_ops;
1367
1368 bond_dev->type = slave_dev->type;
1369 bond_dev->hard_header_len = slave_dev->hard_header_len;
1370 bond_dev->addr_len = slave_dev->addr_len;
1371
1372 memcpy(bond_dev->broadcast, slave_dev->broadcast,
1373 slave_dev->addr_len);
1374 bond->setup_by_slave = 1;
1375 }
1376
1377 /* enslave device <slave> to bond device <master> */
1378 int bond_enslave(struct net_device *bond_dev, struct net_device *slave_dev)
1379 {
1380 struct bonding *bond = netdev_priv(bond_dev);
1381 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
1382 struct slave *new_slave = NULL;
1383 struct dev_mc_list *dmi;
1384 struct sockaddr addr;
1385 int link_reporting;
1386 int old_features = bond_dev->features;
1387 int res = 0;
1388
1389 if (!bond->params.use_carrier && slave_dev->ethtool_ops == NULL &&
1390 slave_ops->ndo_do_ioctl == NULL) {
1391 pr_warning(DRV_NAME
1392 ": %s: Warning: no link monitoring support for %s\n",
1393 bond_dev->name, slave_dev->name);
1394 }
1395
1396 /* bond must be initialized by bond_open() before enslaving */
1397 if (!(bond_dev->flags & IFF_UP)) {
1398 pr_warning(DRV_NAME
1399 " %s: master_dev is not up in bond_enslave\n",
1400 bond_dev->name);
1401 }
1402
1403 /* already enslaved */
1404 if (slave_dev->flags & IFF_SLAVE) {
1405 pr_debug("Error, Device was already enslaved\n");
1406 return -EBUSY;
1407 }
1408
1409 /* vlan challenged mutual exclusion */
1410 /* no need to lock since we're protected by rtnl_lock */
1411 if (slave_dev->features & NETIF_F_VLAN_CHALLENGED) {
1412 pr_debug("%s: NETIF_F_VLAN_CHALLENGED\n", slave_dev->name);
1413 if (!list_empty(&bond->vlan_list)) {
1414 pr_err(DRV_NAME
1415 ": %s: Error: cannot enslave VLAN "
1416 "challenged slave %s on VLAN enabled "
1417 "bond %s\n", bond_dev->name, slave_dev->name,
1418 bond_dev->name);
1419 return -EPERM;
1420 } else {
1421 pr_warning(DRV_NAME
1422 ": %s: Warning: enslaved VLAN challenged "
1423 "slave %s. Adding VLANs will be blocked as "
1424 "long as %s is part of bond %s\n",
1425 bond_dev->name, slave_dev->name, slave_dev->name,
1426 bond_dev->name);
1427 bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
1428 }
1429 } else {
1430 pr_debug("%s: ! NETIF_F_VLAN_CHALLENGED\n", slave_dev->name);
1431 if (bond->slave_cnt == 0) {
1432 /* First slave, and it is not VLAN challenged,
1433 * so remove the block of adding VLANs over the bond.
1434 */
1435 bond_dev->features &= ~NETIF_F_VLAN_CHALLENGED;
1436 }
1437 }
1438
1439 /*
1440 * Old ifenslave binaries are no longer supported. These can
1441 * be identified with moderate accuracy by the state of the slave:
1442 * the current ifenslave will set the interface down prior to
1443 * enslaving it; the old ifenslave will not.
1444 */
1445 if ((slave_dev->flags & IFF_UP)) {
1446 pr_err(DRV_NAME ": %s is up. "
1447 "This may be due to an out of date ifenslave.\n",
1448 slave_dev->name);
1449 res = -EPERM;
1450 goto err_undo_flags;
1451 }
1452
1453 /* set bonding device ether type by slave - bonding netdevices are
1454 * created with ether_setup, so when the slave type is not ARPHRD_ETHER
1455 * there is a need to override some of the type dependent attribs/funcs.
1456 *
1457 * bond ether type mutual exclusion - don't allow slaves of dissimilar
1458 * ether type (eg ARPHRD_ETHER and ARPHRD_INFINIBAND) share the same bond
1459 */
1460 if (bond->slave_cnt == 0) {
1461 if (slave_dev->type != ARPHRD_ETHER)
1462 bond_setup_by_slave(bond_dev, slave_dev);
1463 } else if (bond_dev->type != slave_dev->type) {
1464 pr_err(DRV_NAME ": %s ether type (%d) is different "
1465 "from other slaves (%d), can not enslave it.\n",
1466 slave_dev->name,
1467 slave_dev->type, bond_dev->type);
1468 res = -EINVAL;
1469 goto err_undo_flags;
1470 }
1471
1472 if (slave_ops->ndo_set_mac_address == NULL) {
1473 if (bond->slave_cnt == 0) {
1474 pr_warning(DRV_NAME
1475 ": %s: Warning: The first slave device "
1476 "specified does not support setting the MAC "
1477 "address. Setting fail_over_mac to active.",
1478 bond_dev->name);
1479 bond->params.fail_over_mac = BOND_FOM_ACTIVE;
1480 } else if (bond->params.fail_over_mac != BOND_FOM_ACTIVE) {
1481 pr_err(DRV_NAME
1482 ": %s: Error: The slave device specified "
1483 "does not support setting the MAC address, "
1484 "but fail_over_mac is not set to active.\n"
1485 , bond_dev->name);
1486 res = -EOPNOTSUPP;
1487 goto err_undo_flags;
1488 }
1489 }
1490
1491 new_slave = kzalloc(sizeof(struct slave), GFP_KERNEL);
1492 if (!new_slave) {
1493 res = -ENOMEM;
1494 goto err_undo_flags;
1495 }
1496
1497 /* save slave's original flags before calling
1498 * netdev_set_master and dev_open
1499 */
1500 new_slave->original_flags = slave_dev->flags;
1501
1502 /*
1503 * Save slave's original ("permanent") mac address for modes
1504 * that need it, and for restoring it upon release, and then
1505 * set it to the master's address
1506 */
1507 memcpy(new_slave->perm_hwaddr, slave_dev->dev_addr, ETH_ALEN);
1508
1509 if (!bond->params.fail_over_mac) {
1510 /*
1511 * Set slave to master's mac address. The application already
1512 * set the master's mac address to that of the first slave
1513 */
1514 memcpy(addr.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
1515 addr.sa_family = slave_dev->type;
1516 res = dev_set_mac_address(slave_dev, &addr);
1517 if (res) {
1518 pr_debug("Error %d calling set_mac_address\n", res);
1519 goto err_free;
1520 }
1521 }
1522
1523 res = netdev_set_master(slave_dev, bond_dev);
1524 if (res) {
1525 pr_debug("Error %d calling netdev_set_master\n", res);
1526 goto err_restore_mac;
1527 }
1528 /* open the slave since the application closed it */
1529 res = dev_open(slave_dev);
1530 if (res) {
1531 pr_debug("Opening slave %s failed\n", slave_dev->name);
1532 goto err_unset_master;
1533 }
1534
1535 new_slave->dev = slave_dev;
1536 slave_dev->priv_flags |= IFF_BONDING;
1537
1538 if (bond_is_lb(bond)) {
1539 /* bond_alb_init_slave() must be called before all other stages since
1540 * it might fail and we do not want to have to undo everything
1541 */
1542 res = bond_alb_init_slave(bond, new_slave);
1543 if (res)
1544 goto err_close;
1545 }
1546
1547 /* If the mode USES_PRIMARY, then the new slave gets the
1548 * master's promisc (and mc) settings only if it becomes the
1549 * curr_active_slave, and that is taken care of later when calling
1550 * bond_change_active()
1551 */
1552 if (!USES_PRIMARY(bond->params.mode)) {
1553 /* set promiscuity level to new slave */
1554 if (bond_dev->flags & IFF_PROMISC) {
1555 res = dev_set_promiscuity(slave_dev, 1);
1556 if (res)
1557 goto err_close;
1558 }
1559
1560 /* set allmulti level to new slave */
1561 if (bond_dev->flags & IFF_ALLMULTI) {
1562 res = dev_set_allmulti(slave_dev, 1);
1563 if (res)
1564 goto err_close;
1565 }
1566
1567 netif_addr_lock_bh(bond_dev);
1568 /* upload master's mc_list to new slave */
1569 for (dmi = bond_dev->mc_list; dmi; dmi = dmi->next)
1570 dev_mc_add(slave_dev, dmi->dmi_addr,
1571 dmi->dmi_addrlen, 0);
1572 netif_addr_unlock_bh(bond_dev);
1573 }
1574
1575 if (bond->params.mode == BOND_MODE_8023AD) {
1576 /* add lacpdu mc addr to mc list */
1577 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
1578
1579 dev_mc_add(slave_dev, lacpdu_multicast, ETH_ALEN, 0);
1580 }
1581
1582 bond_add_vlans_on_slave(bond, slave_dev);
1583
1584 write_lock_bh(&bond->lock);
1585
1586 bond_attach_slave(bond, new_slave);
1587
1588 new_slave->delay = 0;
1589 new_slave->link_failure_count = 0;
1590
1591 bond_compute_features(bond);
1592
1593 write_unlock_bh(&bond->lock);
1594
1595 read_lock(&bond->lock);
1596
1597 new_slave->last_arp_rx = jiffies;
1598
1599 if (bond->params.miimon && !bond->params.use_carrier) {
1600 link_reporting = bond_check_dev_link(bond, slave_dev, 1);
1601
1602 if ((link_reporting == -1) && !bond->params.arp_interval) {
1603 /*
1604 * miimon is set but a bonded network driver
1605 * does not support ETHTOOL/MII and
1606 * arp_interval is not set. Note: if
1607 * use_carrier is enabled, we will never go
1608 * here (because netif_carrier is always
1609 * supported); thus, we don't need to change
1610 * the messages for netif_carrier.
1611 */
1612 pr_warning(DRV_NAME
1613 ": %s: Warning: MII and ETHTOOL support not "
1614 "available for interface %s, and "
1615 "arp_interval/arp_ip_target module parameters "
1616 "not specified, thus bonding will not detect "
1617 "link failures! see bonding.txt for details.\n",
1618 bond_dev->name, slave_dev->name);
1619 } else if (link_reporting == -1) {
1620 /* unable get link status using mii/ethtool */
1621 pr_warning(DRV_NAME
1622 ": %s: Warning: can't get link status from "
1623 "interface %s; the network driver associated "
1624 "with this interface does not support MII or "
1625 "ETHTOOL link status reporting, thus miimon "
1626 "has no effect on this interface.\n",
1627 bond_dev->name, slave_dev->name);
1628 }
1629 }
1630
1631 /* check for initial state */
1632 if (!bond->params.miimon ||
1633 (bond_check_dev_link(bond, slave_dev, 0) == BMSR_LSTATUS)) {
1634 if (bond->params.updelay) {
1635 pr_debug("Initial state of slave_dev is "
1636 "BOND_LINK_BACK\n");
1637 new_slave->link = BOND_LINK_BACK;
1638 new_slave->delay = bond->params.updelay;
1639 } else {
1640 pr_debug("Initial state of slave_dev is "
1641 "BOND_LINK_UP\n");
1642 new_slave->link = BOND_LINK_UP;
1643 }
1644 new_slave->jiffies = jiffies;
1645 } else {
1646 pr_debug("Initial state of slave_dev is "
1647 "BOND_LINK_DOWN\n");
1648 new_slave->link = BOND_LINK_DOWN;
1649 }
1650
1651 if (bond_update_speed_duplex(new_slave) &&
1652 (new_slave->link != BOND_LINK_DOWN)) {
1653 pr_warning(DRV_NAME
1654 ": %s: Warning: failed to get speed and duplex from %s, "
1655 "assumed to be 100Mb/sec and Full.\n",
1656 bond_dev->name, new_slave->dev->name);
1657
1658 if (bond->params.mode == BOND_MODE_8023AD) {
1659 pr_warning(DRV_NAME
1660 ": %s: Warning: Operation of 802.3ad mode requires ETHTOOL "
1661 "support in base driver for proper aggregator "
1662 "selection.\n", bond_dev->name);
1663 }
1664 }
1665
1666 if (USES_PRIMARY(bond->params.mode) && bond->params.primary[0]) {
1667 /* if there is a primary slave, remember it */
1668 if (strcmp(bond->params.primary, new_slave->dev->name) == 0)
1669 bond->primary_slave = new_slave;
1670 }
1671
1672 write_lock_bh(&bond->curr_slave_lock);
1673
1674 switch (bond->params.mode) {
1675 case BOND_MODE_ACTIVEBACKUP:
1676 bond_set_slave_inactive_flags(new_slave);
1677 bond_select_active_slave(bond);
1678 break;
1679 case BOND_MODE_8023AD:
1680 /* in 802.3ad mode, the internal mechanism
1681 * will activate the slaves in the selected
1682 * aggregator
1683 */
1684 bond_set_slave_inactive_flags(new_slave);
1685 /* if this is the first slave */
1686 if (bond->slave_cnt == 1) {
1687 SLAVE_AD_INFO(new_slave).id = 1;
1688 /* Initialize AD with the number of times that the AD timer is called in 1 second
1689 * can be called only after the mac address of the bond is set
1690 */
1691 bond_3ad_initialize(bond, 1000/AD_TIMER_INTERVAL,
1692 bond->params.lacp_fast);
1693 } else {
1694 SLAVE_AD_INFO(new_slave).id =
1695 SLAVE_AD_INFO(new_slave->prev).id + 1;
1696 }
1697
1698 bond_3ad_bind_slave(new_slave);
1699 break;
1700 case BOND_MODE_TLB:
1701 case BOND_MODE_ALB:
1702 new_slave->state = BOND_STATE_ACTIVE;
1703 bond_set_slave_inactive_flags(new_slave);
1704 bond_select_active_slave(bond);
1705 break;
1706 default:
1707 pr_debug("This slave is always active in trunk mode\n");
1708
1709 /* always active in trunk mode */
1710 new_slave->state = BOND_STATE_ACTIVE;
1711
1712 /* In trunking mode there is little meaning to curr_active_slave
1713 * anyway (it holds no special properties of the bond device),
1714 * so we can change it without calling change_active_interface()
1715 */
1716 if (!bond->curr_active_slave)
1717 bond->curr_active_slave = new_slave;
1718
1719 break;
1720 } /* switch(bond_mode) */
1721
1722 write_unlock_bh(&bond->curr_slave_lock);
1723
1724 bond_set_carrier(bond);
1725
1726 read_unlock(&bond->lock);
1727
1728 res = bond_create_slave_symlinks(bond_dev, slave_dev);
1729 if (res)
1730 goto err_close;
1731
1732 pr_info(DRV_NAME
1733 ": %s: enslaving %s as a%s interface with a%s link.\n",
1734 bond_dev->name, slave_dev->name,
1735 new_slave->state == BOND_STATE_ACTIVE ? "n active" : " backup",
1736 new_slave->link != BOND_LINK_DOWN ? "n up" : " down");
1737
1738 /* enslave is successful */
1739 return 0;
1740
1741 /* Undo stages on error */
1742 err_close:
1743 dev_close(slave_dev);
1744
1745 err_unset_master:
1746 netdev_set_master(slave_dev, NULL);
1747
1748 err_restore_mac:
1749 if (!bond->params.fail_over_mac) {
1750 /* XXX TODO - fom follow mode needs to change master's
1751 * MAC if this slave's MAC is in use by the bond, or at
1752 * least print a warning.
1753 */
1754 memcpy(addr.sa_data, new_slave->perm_hwaddr, ETH_ALEN);
1755 addr.sa_family = slave_dev->type;
1756 dev_set_mac_address(slave_dev, &addr);
1757 }
1758
1759 err_free:
1760 kfree(new_slave);
1761
1762 err_undo_flags:
1763 bond_dev->features = old_features;
1764
1765 return res;
1766 }
1767
1768 /*
1769 * Try to release the slave device <slave> from the bond device <master>
1770 * It is legal to access curr_active_slave without a lock because all the function
1771 * is write-locked.
1772 *
1773 * The rules for slave state should be:
1774 * for Active/Backup:
1775 * Active stays on all backups go down
1776 * for Bonded connections:
1777 * The first up interface should be left on and all others downed.
1778 */
1779 int bond_release(struct net_device *bond_dev, struct net_device *slave_dev)
1780 {
1781 struct bonding *bond = netdev_priv(bond_dev);
1782 struct slave *slave, *oldcurrent;
1783 struct sockaddr addr;
1784 int mac_addr_differ;
1785
1786 /* slave is not a slave or master is not master of this slave */
1787 if (!(slave_dev->flags & IFF_SLAVE) ||
1788 (slave_dev->master != bond_dev)) {
1789 pr_err(DRV_NAME
1790 ": %s: Error: cannot release %s.\n",
1791 bond_dev->name, slave_dev->name);
1792 return -EINVAL;
1793 }
1794
1795 write_lock_bh(&bond->lock);
1796
1797 slave = bond_get_slave_by_dev(bond, slave_dev);
1798 if (!slave) {
1799 /* not a slave of this bond */
1800 pr_info(DRV_NAME
1801 ": %s: %s not enslaved\n",
1802 bond_dev->name, slave_dev->name);
1803 write_unlock_bh(&bond->lock);
1804 return -EINVAL;
1805 }
1806
1807 if (!bond->params.fail_over_mac) {
1808 mac_addr_differ = memcmp(bond_dev->dev_addr, slave->perm_hwaddr,
1809 ETH_ALEN);
1810 if (!mac_addr_differ && (bond->slave_cnt > 1))
1811 pr_warning(DRV_NAME
1812 ": %s: Warning: the permanent HWaddr of %s - "
1813 "%pM - is still in use by %s. "
1814 "Set the HWaddr of %s to a different address "
1815 "to avoid conflicts.\n",
1816 bond_dev->name, slave_dev->name,
1817 slave->perm_hwaddr,
1818 bond_dev->name, slave_dev->name);
1819 }
1820
1821 /* Inform AD package of unbinding of slave. */
1822 if (bond->params.mode == BOND_MODE_8023AD) {
1823 /* must be called before the slave is
1824 * detached from the list
1825 */
1826 bond_3ad_unbind_slave(slave);
1827 }
1828
1829 pr_info(DRV_NAME
1830 ": %s: releasing %s interface %s\n",
1831 bond_dev->name,
1832 (slave->state == BOND_STATE_ACTIVE)
1833 ? "active" : "backup",
1834 slave_dev->name);
1835
1836 oldcurrent = bond->curr_active_slave;
1837
1838 bond->current_arp_slave = NULL;
1839
1840 /* release the slave from its bond */
1841 bond_detach_slave(bond, slave);
1842
1843 bond_compute_features(bond);
1844
1845 if (bond->primary_slave == slave)
1846 bond->primary_slave = NULL;
1847
1848 if (oldcurrent == slave)
1849 bond_change_active_slave(bond, NULL);
1850
1851 if (bond_is_lb(bond)) {
1852 /* Must be called only after the slave has been
1853 * detached from the list and the curr_active_slave
1854 * has been cleared (if our_slave == old_current),
1855 * but before a new active slave is selected.
1856 */
1857 write_unlock_bh(&bond->lock);
1858 bond_alb_deinit_slave(bond, slave);
1859 write_lock_bh(&bond->lock);
1860 }
1861
1862 if (oldcurrent == slave) {
1863 /*
1864 * Note that we hold RTNL over this sequence, so there
1865 * is no concern that another slave add/remove event
1866 * will interfere.
1867 */
1868 write_unlock_bh(&bond->lock);
1869 read_lock(&bond->lock);
1870 write_lock_bh(&bond->curr_slave_lock);
1871
1872 bond_select_active_slave(bond);
1873
1874 write_unlock_bh(&bond->curr_slave_lock);
1875 read_unlock(&bond->lock);
1876 write_lock_bh(&bond->lock);
1877 }
1878
1879 if (bond->slave_cnt == 0) {
1880 bond_set_carrier(bond);
1881
1882 /* if the last slave was removed, zero the mac address
1883 * of the master so it will be set by the application
1884 * to the mac address of the first slave
1885 */
1886 memset(bond_dev->dev_addr, 0, bond_dev->addr_len);
1887
1888 if (list_empty(&bond->vlan_list)) {
1889 bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
1890 } else {
1891 pr_warning(DRV_NAME
1892 ": %s: Warning: clearing HW address of %s while it "
1893 "still has VLANs.\n",
1894 bond_dev->name, bond_dev->name);
1895 pr_warning(DRV_NAME
1896 ": %s: When re-adding slaves, make sure the bond's "
1897 "HW address matches its VLANs'.\n",
1898 bond_dev->name);
1899 }
1900 } else if ((bond_dev->features & NETIF_F_VLAN_CHALLENGED) &&
1901 !bond_has_challenged_slaves(bond)) {
1902 pr_info(DRV_NAME
1903 ": %s: last VLAN challenged slave %s "
1904 "left bond %s. VLAN blocking is removed\n",
1905 bond_dev->name, slave_dev->name, bond_dev->name);
1906 bond_dev->features &= ~NETIF_F_VLAN_CHALLENGED;
1907 }
1908
1909 write_unlock_bh(&bond->lock);
1910
1911 /* must do this from outside any spinlocks */
1912 bond_destroy_slave_symlinks(bond_dev, slave_dev);
1913
1914 bond_del_vlans_from_slave(bond, slave_dev);
1915
1916 /* If the mode USES_PRIMARY, then we should only remove its
1917 * promisc and mc settings if it was the curr_active_slave, but that was
1918 * already taken care of above when we detached the slave
1919 */
1920 if (!USES_PRIMARY(bond->params.mode)) {
1921 /* unset promiscuity level from slave */
1922 if (bond_dev->flags & IFF_PROMISC)
1923 dev_set_promiscuity(slave_dev, -1);
1924
1925 /* unset allmulti level from slave */
1926 if (bond_dev->flags & IFF_ALLMULTI)
1927 dev_set_allmulti(slave_dev, -1);
1928
1929 /* flush master's mc_list from slave */
1930 netif_addr_lock_bh(bond_dev);
1931 bond_mc_list_flush(bond_dev, slave_dev);
1932 netif_addr_unlock_bh(bond_dev);
1933 }
1934
1935 netdev_set_master(slave_dev, NULL);
1936
1937 /* close slave before restoring its mac address */
1938 dev_close(slave_dev);
1939
1940 if (bond->params.fail_over_mac != BOND_FOM_ACTIVE) {
1941 /* restore original ("permanent") mac address */
1942 memcpy(addr.sa_data, slave->perm_hwaddr, ETH_ALEN);
1943 addr.sa_family = slave_dev->type;
1944 dev_set_mac_address(slave_dev, &addr);
1945 }
1946
1947 slave_dev->priv_flags &= ~(IFF_MASTER_8023AD | IFF_MASTER_ALB |
1948 IFF_SLAVE_INACTIVE | IFF_BONDING |
1949 IFF_SLAVE_NEEDARP);
1950
1951 kfree(slave);
1952
1953 return 0; /* deletion OK */
1954 }
1955
1956 /*
1957 * Destroy a bonding device.
1958 * Must be under rtnl_lock when this function is called.
1959 */
1960 static void bond_uninit(struct net_device *bond_dev)
1961 {
1962 struct bonding *bond = netdev_priv(bond_dev);
1963
1964 bond_deinit(bond_dev);
1965 bond_destroy_sysfs_entry(bond);
1966
1967 if (bond->wq)
1968 destroy_workqueue(bond->wq);
1969
1970 netif_addr_lock_bh(bond_dev);
1971 bond_mc_list_destroy(bond);
1972 netif_addr_unlock_bh(bond_dev);
1973 }
1974
1975 /*
1976 * First release a slave and than destroy the bond if no more slaves are left.
1977 * Must be under rtnl_lock when this function is called.
1978 */
1979 int bond_release_and_destroy(struct net_device *bond_dev,
1980 struct net_device *slave_dev)
1981 {
1982 struct bonding *bond = netdev_priv(bond_dev);
1983 int ret;
1984
1985 ret = bond_release(bond_dev, slave_dev);
1986 if ((ret == 0) && (bond->slave_cnt == 0)) {
1987 pr_info(DRV_NAME ": %s: destroying bond %s.\n",
1988 bond_dev->name, bond_dev->name);
1989 unregister_netdevice(bond_dev);
1990 }
1991 return ret;
1992 }
1993
1994 /*
1995 * This function releases all slaves.
1996 */
1997 static int bond_release_all(struct net_device *bond_dev)
1998 {
1999 struct bonding *bond = netdev_priv(bond_dev);
2000 struct slave *slave;
2001 struct net_device *slave_dev;
2002 struct sockaddr addr;
2003
2004 write_lock_bh(&bond->lock);
2005
2006 netif_carrier_off(bond_dev);
2007
2008 if (bond->slave_cnt == 0)
2009 goto out;
2010
2011 bond->current_arp_slave = NULL;
2012 bond->primary_slave = NULL;
2013 bond_change_active_slave(bond, NULL);
2014
2015 while ((slave = bond->first_slave) != NULL) {
2016 /* Inform AD package of unbinding of slave
2017 * before slave is detached from the list.
2018 */
2019 if (bond->params.mode == BOND_MODE_8023AD)
2020 bond_3ad_unbind_slave(slave);
2021
2022 slave_dev = slave->dev;
2023 bond_detach_slave(bond, slave);
2024
2025 /* now that the slave is detached, unlock and perform
2026 * all the undo steps that should not be called from
2027 * within a lock.
2028 */
2029 write_unlock_bh(&bond->lock);
2030
2031 if (bond_is_lb(bond)) {
2032 /* must be called only after the slave
2033 * has been detached from the list
2034 */
2035 bond_alb_deinit_slave(bond, slave);
2036 }
2037
2038 bond_compute_features(bond);
2039
2040 bond_destroy_slave_symlinks(bond_dev, slave_dev);
2041 bond_del_vlans_from_slave(bond, slave_dev);
2042
2043 /* If the mode USES_PRIMARY, then we should only remove its
2044 * promisc and mc settings if it was the curr_active_slave, but that was
2045 * already taken care of above when we detached the slave
2046 */
2047 if (!USES_PRIMARY(bond->params.mode)) {
2048 /* unset promiscuity level from slave */
2049 if (bond_dev->flags & IFF_PROMISC)
2050 dev_set_promiscuity(slave_dev, -1);
2051
2052 /* unset allmulti level from slave */
2053 if (bond_dev->flags & IFF_ALLMULTI)
2054 dev_set_allmulti(slave_dev, -1);
2055
2056 /* flush master's mc_list from slave */
2057 netif_addr_lock_bh(bond_dev);
2058 bond_mc_list_flush(bond_dev, slave_dev);
2059 netif_addr_unlock_bh(bond_dev);
2060 }
2061
2062 netdev_set_master(slave_dev, NULL);
2063
2064 /* close slave before restoring its mac address */
2065 dev_close(slave_dev);
2066
2067 if (!bond->params.fail_over_mac) {
2068 /* restore original ("permanent") mac address*/
2069 memcpy(addr.sa_data, slave->perm_hwaddr, ETH_ALEN);
2070 addr.sa_family = slave_dev->type;
2071 dev_set_mac_address(slave_dev, &addr);
2072 }
2073
2074 slave_dev->priv_flags &= ~(IFF_MASTER_8023AD | IFF_MASTER_ALB |
2075 IFF_SLAVE_INACTIVE);
2076
2077 kfree(slave);
2078
2079 /* re-acquire the lock before getting the next slave */
2080 write_lock_bh(&bond->lock);
2081 }
2082
2083 /* zero the mac address of the master so it will be
2084 * set by the application to the mac address of the
2085 * first slave
2086 */
2087 memset(bond_dev->dev_addr, 0, bond_dev->addr_len);
2088
2089 if (list_empty(&bond->vlan_list))
2090 bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
2091 else {
2092 pr_warning(DRV_NAME
2093 ": %s: Warning: clearing HW address of %s while it "
2094 "still has VLANs.\n",
2095 bond_dev->name, bond_dev->name);
2096 pr_warning(DRV_NAME
2097 ": %s: When re-adding slaves, make sure the bond's "
2098 "HW address matches its VLANs'.\n",
2099 bond_dev->name);
2100 }
2101
2102 pr_info(DRV_NAME
2103 ": %s: released all slaves\n",
2104 bond_dev->name);
2105
2106 out:
2107 write_unlock_bh(&bond->lock);
2108
2109 return 0;
2110 }
2111
2112 /*
2113 * This function changes the active slave to slave <slave_dev>.
2114 * It returns -EINVAL in the following cases.
2115 * - <slave_dev> is not found in the list.
2116 * - There is not active slave now.
2117 * - <slave_dev> is already active.
2118 * - The link state of <slave_dev> is not BOND_LINK_UP.
2119 * - <slave_dev> is not running.
2120 * In these cases, this function does nothing.
2121 * In the other cases, current_slave pointer is changed and 0 is returned.
2122 */
2123 static int bond_ioctl_change_active(struct net_device *bond_dev, struct net_device *slave_dev)
2124 {
2125 struct bonding *bond = netdev_priv(bond_dev);
2126 struct slave *old_active = NULL;
2127 struct slave *new_active = NULL;
2128 int res = 0;
2129
2130 if (!USES_PRIMARY(bond->params.mode))
2131 return -EINVAL;
2132
2133 /* Verify that master_dev is indeed the master of slave_dev */
2134 if (!(slave_dev->flags & IFF_SLAVE) || (slave_dev->master != bond_dev))
2135 return -EINVAL;
2136
2137 read_lock(&bond->lock);
2138
2139 read_lock(&bond->curr_slave_lock);
2140 old_active = bond->curr_active_slave;
2141 read_unlock(&bond->curr_slave_lock);
2142
2143 new_active = bond_get_slave_by_dev(bond, slave_dev);
2144
2145 /*
2146 * Changing to the current active: do nothing; return success.
2147 */
2148 if (new_active && (new_active == old_active)) {
2149 read_unlock(&bond->lock);
2150 return 0;
2151 }
2152
2153 if ((new_active) &&
2154 (old_active) &&
2155 (new_active->link == BOND_LINK_UP) &&
2156 IS_UP(new_active->dev)) {
2157 write_lock_bh(&bond->curr_slave_lock);
2158 bond_change_active_slave(bond, new_active);
2159 write_unlock_bh(&bond->curr_slave_lock);
2160 } else
2161 res = -EINVAL;
2162
2163 read_unlock(&bond->lock);
2164
2165 return res;
2166 }
2167
2168 static int bond_info_query(struct net_device *bond_dev, struct ifbond *info)
2169 {
2170 struct bonding *bond = netdev_priv(bond_dev);
2171
2172 info->bond_mode = bond->params.mode;
2173 info->miimon = bond->params.miimon;
2174
2175 read_lock(&bond->lock);
2176 info->num_slaves = bond->slave_cnt;
2177 read_unlock(&bond->lock);
2178
2179 return 0;
2180 }
2181
2182 static int bond_slave_info_query(struct net_device *bond_dev, struct ifslave *info)
2183 {
2184 struct bonding *bond = netdev_priv(bond_dev);
2185 struct slave *slave;
2186 int i, res = -ENODEV;
2187
2188 read_lock(&bond->lock);
2189
2190 bond_for_each_slave(bond, slave, i) {
2191 if (i == (int)info->slave_id) {
2192 res = 0;
2193 strcpy(info->slave_name, slave->dev->name);
2194 info->link = slave->link;
2195 info->state = slave->state;
2196 info->link_failure_count = slave->link_failure_count;
2197 break;
2198 }
2199 }
2200
2201 read_unlock(&bond->lock);
2202
2203 return res;
2204 }
2205
2206 /*-------------------------------- Monitoring -------------------------------*/
2207
2208
2209 static int bond_miimon_inspect(struct bonding *bond)
2210 {
2211 struct slave *slave;
2212 int i, link_state, commit = 0;
2213 bool ignore_updelay;
2214
2215 ignore_updelay = !bond->curr_active_slave ? true : false;
2216
2217 bond_for_each_slave(bond, slave, i) {
2218 slave->new_link = BOND_LINK_NOCHANGE;
2219
2220 link_state = bond_check_dev_link(bond, slave->dev, 0);
2221
2222 switch (slave->link) {
2223 case BOND_LINK_UP:
2224 if (link_state)
2225 continue;
2226
2227 slave->link = BOND_LINK_FAIL;
2228 slave->delay = bond->params.downdelay;
2229 if (slave->delay) {
2230 pr_info(DRV_NAME
2231 ": %s: link status down for %s"
2232 "interface %s, disabling it in %d ms.\n",
2233 bond->dev->name,
2234 (bond->params.mode ==
2235 BOND_MODE_ACTIVEBACKUP) ?
2236 ((slave->state == BOND_STATE_ACTIVE) ?
2237 "active " : "backup ") : "",
2238 slave->dev->name,
2239 bond->params.downdelay * bond->params.miimon);
2240 }
2241 /*FALLTHRU*/
2242 case BOND_LINK_FAIL:
2243 if (link_state) {
2244 /*
2245 * recovered before downdelay expired
2246 */
2247 slave->link = BOND_LINK_UP;
2248 slave->jiffies = jiffies;
2249 pr_info(DRV_NAME
2250 ": %s: link status up again after %d "
2251 "ms for interface %s.\n",
2252 bond->dev->name,
2253 (bond->params.downdelay - slave->delay) *
2254 bond->params.miimon,
2255 slave->dev->name);
2256 continue;
2257 }
2258
2259 if (slave->delay <= 0) {
2260 slave->new_link = BOND_LINK_DOWN;
2261 commit++;
2262 continue;
2263 }
2264
2265 slave->delay--;
2266 break;
2267
2268 case BOND_LINK_DOWN:
2269 if (!link_state)
2270 continue;
2271
2272 slave->link = BOND_LINK_BACK;
2273 slave->delay = bond->params.updelay;
2274
2275 if (slave->delay) {
2276 pr_info(DRV_NAME
2277 ": %s: link status up for "
2278 "interface %s, enabling it in %d ms.\n",
2279 bond->dev->name, slave->dev->name,
2280 ignore_updelay ? 0 :
2281 bond->params.updelay *
2282 bond->params.miimon);
2283 }
2284 /*FALLTHRU*/
2285 case BOND_LINK_BACK:
2286 if (!link_state) {
2287 slave->link = BOND_LINK_DOWN;
2288 pr_info(DRV_NAME
2289 ": %s: link status down again after %d "
2290 "ms for interface %s.\n",
2291 bond->dev->name,
2292 (bond->params.updelay - slave->delay) *
2293 bond->params.miimon,
2294 slave->dev->name);
2295
2296 continue;
2297 }
2298
2299 if (ignore_updelay)
2300 slave->delay = 0;
2301
2302 if (slave->delay <= 0) {
2303 slave->new_link = BOND_LINK_UP;
2304 commit++;
2305 ignore_updelay = false;
2306 continue;
2307 }
2308
2309 slave->delay--;
2310 break;
2311 }
2312 }
2313
2314 return commit;
2315 }
2316
2317 static void bond_miimon_commit(struct bonding *bond)
2318 {
2319 struct slave *slave;
2320 int i;
2321
2322 bond_for_each_slave(bond, slave, i) {
2323 switch (slave->new_link) {
2324 case BOND_LINK_NOCHANGE:
2325 continue;
2326
2327 case BOND_LINK_UP:
2328 slave->link = BOND_LINK_UP;
2329 slave->jiffies = jiffies;
2330
2331 if (bond->params.mode == BOND_MODE_8023AD) {
2332 /* prevent it from being the active one */
2333 slave->state = BOND_STATE_BACKUP;
2334 } else if (bond->params.mode != BOND_MODE_ACTIVEBACKUP) {
2335 /* make it immediately active */
2336 slave->state = BOND_STATE_ACTIVE;
2337 } else if (slave != bond->primary_slave) {
2338 /* prevent it from being the active one */
2339 slave->state = BOND_STATE_BACKUP;
2340 }
2341
2342 pr_info(DRV_NAME
2343 ": %s: link status definitely "
2344 "up for interface %s.\n",
2345 bond->dev->name, slave->dev->name);
2346
2347 /* notify ad that the link status has changed */
2348 if (bond->params.mode == BOND_MODE_8023AD)
2349 bond_3ad_handle_link_change(slave, BOND_LINK_UP);
2350
2351 if (bond_is_lb(bond))
2352 bond_alb_handle_link_change(bond, slave,
2353 BOND_LINK_UP);
2354
2355 if (!bond->curr_active_slave ||
2356 (slave == bond->primary_slave))
2357 goto do_failover;
2358
2359 continue;
2360
2361 case BOND_LINK_DOWN:
2362 if (slave->link_failure_count < UINT_MAX)
2363 slave->link_failure_count++;
2364
2365 slave->link = BOND_LINK_DOWN;
2366
2367 if (bond->params.mode == BOND_MODE_ACTIVEBACKUP ||
2368 bond->params.mode == BOND_MODE_8023AD)
2369 bond_set_slave_inactive_flags(slave);
2370
2371 pr_info(DRV_NAME
2372 ": %s: link status definitely down for "
2373 "interface %s, disabling it\n",
2374 bond->dev->name, slave->dev->name);
2375
2376 if (bond->params.mode == BOND_MODE_8023AD)
2377 bond_3ad_handle_link_change(slave,
2378 BOND_LINK_DOWN);
2379
2380 if (bond_is_lb(bond))
2381 bond_alb_handle_link_change(bond, slave,
2382 BOND_LINK_DOWN);
2383
2384 if (slave == bond->curr_active_slave)
2385 goto do_failover;
2386
2387 continue;
2388
2389 default:
2390 pr_err(DRV_NAME
2391 ": %s: invalid new link %d on slave %s\n",
2392 bond->dev->name, slave->new_link,
2393 slave->dev->name);
2394 slave->new_link = BOND_LINK_NOCHANGE;
2395
2396 continue;
2397 }
2398
2399 do_failover:
2400 ASSERT_RTNL();
2401 write_lock_bh(&bond->curr_slave_lock);
2402 bond_select_active_slave(bond);
2403 write_unlock_bh(&bond->curr_slave_lock);
2404 }
2405
2406 bond_set_carrier(bond);
2407 }
2408
2409 /*
2410 * bond_mii_monitor
2411 *
2412 * Really a wrapper that splits the mii monitor into two phases: an
2413 * inspection, then (if inspection indicates something needs to be done)
2414 * an acquisition of appropriate locks followed by a commit phase to
2415 * implement whatever link state changes are indicated.
2416 */
2417 void bond_mii_monitor(struct work_struct *work)
2418 {
2419 struct bonding *bond = container_of(work, struct bonding,
2420 mii_work.work);
2421
2422 read_lock(&bond->lock);
2423 if (bond->kill_timers)
2424 goto out;
2425
2426 if (bond->slave_cnt == 0)
2427 goto re_arm;
2428
2429 if (bond->send_grat_arp) {
2430 read_lock(&bond->curr_slave_lock);
2431 bond_send_gratuitous_arp(bond);
2432 read_unlock(&bond->curr_slave_lock);
2433 }
2434
2435 if (bond->send_unsol_na) {
2436 read_lock(&bond->curr_slave_lock);
2437 bond_send_unsolicited_na(bond);
2438 read_unlock(&bond->curr_slave_lock);
2439 }
2440
2441 if (bond_miimon_inspect(bond)) {
2442 read_unlock(&bond->lock);
2443 rtnl_lock();
2444 read_lock(&bond->lock);
2445
2446 bond_miimon_commit(bond);
2447
2448 read_unlock(&bond->lock);
2449 rtnl_unlock(); /* might sleep, hold no other locks */
2450 read_lock(&bond->lock);
2451 }
2452
2453 re_arm:
2454 if (bond->params.miimon)
2455 queue_delayed_work(bond->wq, &bond->mii_work,
2456 msecs_to_jiffies(bond->params.miimon));
2457 out:
2458 read_unlock(&bond->lock);
2459 }
2460
2461 static __be32 bond_glean_dev_ip(struct net_device *dev)
2462 {
2463 struct in_device *idev;
2464 struct in_ifaddr *ifa;
2465 __be32 addr = 0;
2466
2467 if (!dev)
2468 return 0;
2469
2470 rcu_read_lock();
2471 idev = __in_dev_get_rcu(dev);
2472 if (!idev)
2473 goto out;
2474
2475 ifa = idev->ifa_list;
2476 if (!ifa)
2477 goto out;
2478
2479 addr = ifa->ifa_local;
2480 out:
2481 rcu_read_unlock();
2482 return addr;
2483 }
2484
2485 static int bond_has_this_ip(struct bonding *bond, __be32 ip)
2486 {
2487 struct vlan_entry *vlan;
2488
2489 if (ip == bond->master_ip)
2490 return 1;
2491
2492 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
2493 if (ip == vlan->vlan_ip)
2494 return 1;
2495 }
2496
2497 return 0;
2498 }
2499
2500 /*
2501 * We go to the (large) trouble of VLAN tagging ARP frames because
2502 * switches in VLAN mode (especially if ports are configured as
2503 * "native" to a VLAN) might not pass non-tagged frames.
2504 */
2505 static void bond_arp_send(struct net_device *slave_dev, int arp_op, __be32 dest_ip, __be32 src_ip, unsigned short vlan_id)
2506 {
2507 struct sk_buff *skb;
2508
2509 pr_debug("arp %d on slave %s: dst %x src %x vid %d\n", arp_op,
2510 slave_dev->name, dest_ip, src_ip, vlan_id);
2511
2512 skb = arp_create(arp_op, ETH_P_ARP, dest_ip, slave_dev, src_ip,
2513 NULL, slave_dev->dev_addr, NULL);
2514
2515 if (!skb) {
2516 pr_err(DRV_NAME ": ARP packet allocation failed\n");
2517 return;
2518 }
2519 if (vlan_id) {
2520 skb = vlan_put_tag(skb, vlan_id);
2521 if (!skb) {
2522 pr_err(DRV_NAME ": failed to insert VLAN tag\n");
2523 return;
2524 }
2525 }
2526 arp_xmit(skb);
2527 }
2528
2529
2530 static void bond_arp_send_all(struct bonding *bond, struct slave *slave)
2531 {
2532 int i, vlan_id, rv;
2533 __be32 *targets = bond->params.arp_targets;
2534 struct vlan_entry *vlan;
2535 struct net_device *vlan_dev;
2536 struct flowi fl;
2537 struct rtable *rt;
2538
2539 for (i = 0; (i < BOND_MAX_ARP_TARGETS); i++) {
2540 if (!targets[i])
2541 break;
2542 pr_debug("basa: target %x\n", targets[i]);
2543 if (list_empty(&bond->vlan_list)) {
2544 pr_debug("basa: empty vlan: arp_send\n");
2545 bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2546 bond->master_ip, 0);
2547 continue;
2548 }
2549
2550 /*
2551 * If VLANs are configured, we do a route lookup to
2552 * determine which VLAN interface would be used, so we
2553 * can tag the ARP with the proper VLAN tag.
2554 */
2555 memset(&fl, 0, sizeof(fl));
2556 fl.fl4_dst = targets[i];
2557 fl.fl4_tos = RTO_ONLINK;
2558
2559 rv = ip_route_output_key(&init_net, &rt, &fl);
2560 if (rv) {
2561 if (net_ratelimit()) {
2562 pr_warning(DRV_NAME
2563 ": %s: no route to arp_ip_target %pI4\n",
2564 bond->dev->name, &fl.fl4_dst);
2565 }
2566 continue;
2567 }
2568
2569 /*
2570 * This target is not on a VLAN
2571 */
2572 if (rt->u.dst.dev == bond->dev) {
2573 ip_rt_put(rt);
2574 pr_debug("basa: rtdev == bond->dev: arp_send\n");
2575 bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2576 bond->master_ip, 0);
2577 continue;
2578 }
2579
2580 vlan_id = 0;
2581 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
2582 vlan_dev = vlan_group_get_device(bond->vlgrp, vlan->vlan_id);
2583 if (vlan_dev == rt->u.dst.dev) {
2584 vlan_id = vlan->vlan_id;
2585 pr_debug("basa: vlan match on %s %d\n",
2586 vlan_dev->name, vlan_id);
2587 break;
2588 }
2589 }
2590
2591 if (vlan_id) {
2592 ip_rt_put(rt);
2593 bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2594 vlan->vlan_ip, vlan_id);
2595 continue;
2596 }
2597
2598 if (net_ratelimit()) {
2599 pr_warning(DRV_NAME
2600 ": %s: no path to arp_ip_target %pI4 via rt.dev %s\n",
2601 bond->dev->name, &fl.fl4_dst,
2602 rt->u.dst.dev ? rt->u.dst.dev->name : "NULL");
2603 }
2604 ip_rt_put(rt);
2605 }
2606 }
2607
2608 /*
2609 * Kick out a gratuitous ARP for an IP on the bonding master plus one
2610 * for each VLAN above us.
2611 *
2612 * Caller must hold curr_slave_lock for read or better
2613 */
2614 static void bond_send_gratuitous_arp(struct bonding *bond)
2615 {
2616 struct slave *slave = bond->curr_active_slave;
2617 struct vlan_entry *vlan;
2618 struct net_device *vlan_dev;
2619
2620 pr_debug("bond_send_grat_arp: bond %s slave %s\n", bond->dev->name,
2621 slave ? slave->dev->name : "NULL");
2622
2623 if (!slave || !bond->send_grat_arp ||
2624 test_bit(__LINK_STATE_LINKWATCH_PENDING, &slave->dev->state))
2625 return;
2626
2627 bond->send_grat_arp--;
2628
2629 if (bond->master_ip) {
2630 bond_arp_send(slave->dev, ARPOP_REPLY, bond->master_ip,
2631 bond->master_ip, 0);
2632 }
2633
2634 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
2635 vlan_dev = vlan_group_get_device(bond->vlgrp, vlan->vlan_id);
2636 if (vlan->vlan_ip) {
2637 bond_arp_send(slave->dev, ARPOP_REPLY, vlan->vlan_ip,
2638 vlan->vlan_ip, vlan->vlan_id);
2639 }
2640 }
2641 }
2642
2643 static void bond_validate_arp(struct bonding *bond, struct slave *slave, __be32 sip, __be32 tip)
2644 {
2645 int i;
2646 __be32 *targets = bond->params.arp_targets;
2647
2648 for (i = 0; (i < BOND_MAX_ARP_TARGETS) && targets[i]; i++) {
2649 pr_debug("bva: sip %pI4 tip %pI4 t[%d] %pI4 bhti(tip) %d\n",
2650 &sip, &tip, i, &targets[i], bond_has_this_ip(bond, tip));
2651 if (sip == targets[i]) {
2652 if (bond_has_this_ip(bond, tip))
2653 slave->last_arp_rx = jiffies;
2654 return;
2655 }
2656 }
2657 }
2658
2659 static int bond_arp_rcv(struct sk_buff *skb, struct net_device *dev, struct packet_type *pt, struct net_device *orig_dev)
2660 {
2661 struct arphdr *arp;
2662 struct slave *slave;
2663 struct bonding *bond;
2664 unsigned char *arp_ptr;
2665 __be32 sip, tip;
2666
2667 if (dev_net(dev) != &init_net)
2668 goto out;
2669
2670 if (!(dev->priv_flags & IFF_BONDING) || !(dev->flags & IFF_MASTER))
2671 goto out;
2672
2673 bond = netdev_priv(dev);
2674 read_lock(&bond->lock);
2675
2676 pr_debug("bond_arp_rcv: bond %s skb->dev %s orig_dev %s\n",
2677 bond->dev->name, skb->dev ? skb->dev->name : "NULL",
2678 orig_dev ? orig_dev->name : "NULL");
2679
2680 slave = bond_get_slave_by_dev(bond, orig_dev);
2681 if (!slave || !slave_do_arp_validate(bond, slave))
2682 goto out_unlock;
2683
2684 if (!pskb_may_pull(skb, arp_hdr_len(dev)))
2685 goto out_unlock;
2686
2687 arp = arp_hdr(skb);
2688 if (arp->ar_hln != dev->addr_len ||
2689 skb->pkt_type == PACKET_OTHERHOST ||
2690 skb->pkt_type == PACKET_LOOPBACK ||
2691 arp->ar_hrd != htons(ARPHRD_ETHER) ||
2692 arp->ar_pro != htons(ETH_P_IP) ||
2693 arp->ar_pln != 4)
2694 goto out_unlock;
2695
2696 arp_ptr = (unsigned char *)(arp + 1);
2697 arp_ptr += dev->addr_len;
2698 memcpy(&sip, arp_ptr, 4);
2699 arp_ptr += 4 + dev->addr_len;
2700 memcpy(&tip, arp_ptr, 4);
2701
2702 pr_debug("bond_arp_rcv: %s %s/%d av %d sv %d sip %pI4 tip %pI4\n",
2703 bond->dev->name, slave->dev->name, slave->state,
2704 bond->params.arp_validate, slave_do_arp_validate(bond, slave),
2705 &sip, &tip);
2706
2707 /*
2708 * Backup slaves won't see the ARP reply, but do come through
2709 * here for each ARP probe (so we swap the sip/tip to validate
2710 * the probe). In a "redundant switch, common router" type of
2711 * configuration, the ARP probe will (hopefully) travel from
2712 * the active, through one switch, the router, then the other
2713 * switch before reaching the backup.
2714 */
2715 if (slave->state == BOND_STATE_ACTIVE)
2716 bond_validate_arp(bond, slave, sip, tip);
2717 else
2718 bond_validate_arp(bond, slave, tip, sip);
2719
2720 out_unlock:
2721 read_unlock(&bond->lock);
2722 out:
2723 dev_kfree_skb(skb);
2724 return NET_RX_SUCCESS;
2725 }
2726
2727 /*
2728 * this function is called regularly to monitor each slave's link
2729 * ensuring that traffic is being sent and received when arp monitoring
2730 * is used in load-balancing mode. if the adapter has been dormant, then an
2731 * arp is transmitted to generate traffic. see activebackup_arp_monitor for
2732 * arp monitoring in active backup mode.
2733 */
2734 void bond_loadbalance_arp_mon(struct work_struct *work)
2735 {
2736 struct bonding *bond = container_of(work, struct bonding,
2737 arp_work.work);
2738 struct slave *slave, *oldcurrent;
2739 int do_failover = 0;
2740 int delta_in_ticks;
2741 int i;
2742
2743 read_lock(&bond->lock);
2744
2745 delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
2746
2747 if (bond->kill_timers)
2748 goto out;
2749
2750 if (bond->slave_cnt == 0)
2751 goto re_arm;
2752
2753 read_lock(&bond->curr_slave_lock);
2754 oldcurrent = bond->curr_active_slave;
2755 read_unlock(&bond->curr_slave_lock);
2756
2757 /* see if any of the previous devices are up now (i.e. they have
2758 * xmt and rcv traffic). the curr_active_slave does not come into
2759 * the picture unless it is null. also, slave->jiffies is not needed
2760 * here because we send an arp on each slave and give a slave as
2761 * long as it needs to get the tx/rx within the delta.
2762 * TODO: what about up/down delay in arp mode? it wasn't here before
2763 * so it can wait
2764 */
2765 bond_for_each_slave(bond, slave, i) {
2766 if (slave->link != BOND_LINK_UP) {
2767 if (time_before_eq(jiffies, dev_trans_start(slave->dev) + delta_in_ticks) &&
2768 time_before_eq(jiffies, slave->dev->last_rx + delta_in_ticks)) {
2769
2770 slave->link = BOND_LINK_UP;
2771 slave->state = BOND_STATE_ACTIVE;
2772
2773 /* primary_slave has no meaning in round-robin
2774 * mode. the window of a slave being up and
2775 * curr_active_slave being null after enslaving
2776 * is closed.
2777 */
2778 if (!oldcurrent) {
2779 pr_info(DRV_NAME
2780 ": %s: link status definitely "
2781 "up for interface %s, ",
2782 bond->dev->name,
2783 slave->dev->name);
2784 do_failover = 1;
2785 } else {
2786 pr_info(DRV_NAME
2787 ": %s: interface %s is now up\n",
2788 bond->dev->name,
2789 slave->dev->name);
2790 }
2791 }
2792 } else {
2793 /* slave->link == BOND_LINK_UP */
2794
2795 /* not all switches will respond to an arp request
2796 * when the source ip is 0, so don't take the link down
2797 * if we don't know our ip yet
2798 */
2799 if (time_after_eq(jiffies, dev_trans_start(slave->dev) + 2*delta_in_ticks) ||
2800 (time_after_eq(jiffies, slave->dev->last_rx + 2*delta_in_ticks))) {
2801
2802 slave->link = BOND_LINK_DOWN;
2803 slave->state = BOND_STATE_BACKUP;
2804
2805 if (slave->link_failure_count < UINT_MAX)
2806 slave->link_failure_count++;
2807
2808 pr_info(DRV_NAME
2809 ": %s: interface %s is now down.\n",
2810 bond->dev->name,
2811 slave->dev->name);
2812
2813 if (slave == oldcurrent)
2814 do_failover = 1;
2815 }
2816 }
2817
2818 /* note: if switch is in round-robin mode, all links
2819 * must tx arp to ensure all links rx an arp - otherwise
2820 * links may oscillate or not come up at all; if switch is
2821 * in something like xor mode, there is nothing we can
2822 * do - all replies will be rx'ed on same link causing slaves
2823 * to be unstable during low/no traffic periods
2824 */
2825 if (IS_UP(slave->dev))
2826 bond_arp_send_all(bond, slave);
2827 }
2828
2829 if (do_failover) {
2830 write_lock_bh(&bond->curr_slave_lock);
2831
2832 bond_select_active_slave(bond);
2833
2834 write_unlock_bh(&bond->curr_slave_lock);
2835 }
2836
2837 re_arm:
2838 if (bond->params.arp_interval)
2839 queue_delayed_work(bond->wq, &bond->arp_work, delta_in_ticks);
2840 out:
2841 read_unlock(&bond->lock);
2842 }
2843
2844 /*
2845 * Called to inspect slaves for active-backup mode ARP monitor link state
2846 * changes. Sets new_link in slaves to specify what action should take
2847 * place for the slave. Returns 0 if no changes are found, >0 if changes
2848 * to link states must be committed.
2849 *
2850 * Called with bond->lock held for read.
2851 */
2852 static int bond_ab_arp_inspect(struct bonding *bond, int delta_in_ticks)
2853 {
2854 struct slave *slave;
2855 int i, commit = 0;
2856
2857 bond_for_each_slave(bond, slave, i) {
2858 slave->new_link = BOND_LINK_NOCHANGE;
2859
2860 if (slave->link != BOND_LINK_UP) {
2861 if (time_before_eq(jiffies, slave_last_rx(bond, slave) +
2862 delta_in_ticks)) {
2863 slave->new_link = BOND_LINK_UP;
2864 commit++;
2865 }
2866
2867 continue;
2868 }
2869
2870 /*
2871 * Give slaves 2*delta after being enslaved or made
2872 * active. This avoids bouncing, as the last receive
2873 * times need a full ARP monitor cycle to be updated.
2874 */
2875 if (!time_after_eq(jiffies, slave->jiffies +
2876 2 * delta_in_ticks))
2877 continue;
2878
2879 /*
2880 * Backup slave is down if:
2881 * - No current_arp_slave AND
2882 * - more than 3*delta since last receive AND
2883 * - the bond has an IP address
2884 *
2885 * Note: a non-null current_arp_slave indicates
2886 * the curr_active_slave went down and we are
2887 * searching for a new one; under this condition
2888 * we only take the curr_active_slave down - this
2889 * gives each slave a chance to tx/rx traffic
2890 * before being taken out
2891 */
2892 if (slave->state == BOND_STATE_BACKUP &&
2893 !bond->current_arp_slave &&
2894 time_after(jiffies, slave_last_rx(bond, slave) +
2895 3 * delta_in_ticks)) {
2896 slave->new_link = BOND_LINK_DOWN;
2897 commit++;
2898 }
2899
2900 /*
2901 * Active slave is down if:
2902 * - more than 2*delta since transmitting OR
2903 * - (more than 2*delta since receive AND
2904 * the bond has an IP address)
2905 */
2906 if ((slave->state == BOND_STATE_ACTIVE) &&
2907 (time_after_eq(jiffies, dev_trans_start(slave->dev) +
2908 2 * delta_in_ticks) ||
2909 (time_after_eq(jiffies, slave_last_rx(bond, slave)
2910 + 2 * delta_in_ticks)))) {
2911 slave->new_link = BOND_LINK_DOWN;
2912 commit++;
2913 }
2914 }
2915
2916 read_lock(&bond->curr_slave_lock);
2917
2918 /*
2919 * Trigger a commit if the primary option setting has changed.
2920 */
2921 if (bond->primary_slave &&
2922 (bond->primary_slave != bond->curr_active_slave) &&
2923 (bond->primary_slave->link == BOND_LINK_UP))
2924 commit++;
2925
2926 read_unlock(&bond->curr_slave_lock);
2927
2928 return commit;
2929 }
2930
2931 /*
2932 * Called to commit link state changes noted by inspection step of
2933 * active-backup mode ARP monitor.
2934 *
2935 * Called with RTNL and bond->lock for read.
2936 */
2937 static void bond_ab_arp_commit(struct bonding *bond, int delta_in_ticks)
2938 {
2939 struct slave *slave;
2940 int i;
2941
2942 bond_for_each_slave(bond, slave, i) {
2943 switch (slave->new_link) {
2944 case BOND_LINK_NOCHANGE:
2945 continue;
2946
2947 case BOND_LINK_UP:
2948 write_lock_bh(&bond->curr_slave_lock);
2949
2950 if (!bond->curr_active_slave &&
2951 time_before_eq(jiffies, dev_trans_start(slave->dev) +
2952 delta_in_ticks)) {
2953 slave->link = BOND_LINK_UP;
2954 bond_change_active_slave(bond, slave);
2955 bond->current_arp_slave = NULL;
2956
2957 pr_info(DRV_NAME
2958 ": %s: %s is up and now the "
2959 "active interface\n",
2960 bond->dev->name, slave->dev->name);
2961
2962 } else if (bond->curr_active_slave != slave) {
2963 /* this slave has just come up but we
2964 * already have a current slave; this can
2965 * also happen if bond_enslave adds a new
2966 * slave that is up while we are searching
2967 * for a new slave
2968 */
2969 slave->link = BOND_LINK_UP;
2970 bond_set_slave_inactive_flags(slave);
2971 bond->current_arp_slave = NULL;
2972
2973 pr_info(DRV_NAME
2974 ": %s: backup interface %s is now up\n",
2975 bond->dev->name, slave->dev->name);
2976 }
2977
2978 write_unlock_bh(&bond->curr_slave_lock);
2979
2980 break;
2981
2982 case BOND_LINK_DOWN:
2983 if (slave->link_failure_count < UINT_MAX)
2984 slave->link_failure_count++;
2985
2986 slave->link = BOND_LINK_DOWN;
2987
2988 if (slave == bond->curr_active_slave) {
2989 pr_info(DRV_NAME
2990 ": %s: link status down for active "
2991 "interface %s, disabling it\n",
2992 bond->dev->name, slave->dev->name);
2993
2994 bond_set_slave_inactive_flags(slave);
2995
2996 write_lock_bh(&bond->curr_slave_lock);
2997
2998 bond_select_active_slave(bond);
2999 if (bond->curr_active_slave)
3000 bond->curr_active_slave->jiffies =
3001 jiffies;
3002
3003 write_unlock_bh(&bond->curr_slave_lock);
3004
3005 bond->current_arp_slave = NULL;
3006
3007 } else if (slave->state == BOND_STATE_BACKUP) {
3008 pr_info(DRV_NAME
3009 ": %s: backup interface %s is now down\n",
3010 bond->dev->name, slave->dev->name);
3011
3012 bond_set_slave_inactive_flags(slave);
3013 }
3014 break;
3015
3016 default:
3017 pr_err(DRV_NAME
3018 ": %s: impossible: new_link %d on slave %s\n",
3019 bond->dev->name, slave->new_link,
3020 slave->dev->name);
3021 }
3022 }
3023
3024 /*
3025 * No race with changes to primary via sysfs, as we hold rtnl.
3026 */
3027 if (bond->primary_slave &&
3028 (bond->primary_slave != bond->curr_active_slave) &&
3029 (bond->primary_slave->link == BOND_LINK_UP)) {
3030 write_lock_bh(&bond->curr_slave_lock);
3031 bond_change_active_slave(bond, bond->primary_slave);
3032 write_unlock_bh(&bond->curr_slave_lock);
3033 }
3034
3035 bond_set_carrier(bond);
3036 }
3037
3038 /*
3039 * Send ARP probes for active-backup mode ARP monitor.
3040 *
3041 * Called with bond->lock held for read.
3042 */
3043 static void bond_ab_arp_probe(struct bonding *bond)
3044 {
3045 struct slave *slave;
3046 int i;
3047
3048 read_lock(&bond->curr_slave_lock);
3049
3050 if (bond->current_arp_slave && bond->curr_active_slave)
3051 pr_info(DRV_NAME "PROBE: c_arp %s && cas %s BAD\n",
3052 bond->current_arp_slave->dev->name,
3053 bond->curr_active_slave->dev->name);
3054
3055 if (bond->curr_active_slave) {
3056 bond_arp_send_all(bond, bond->curr_active_slave);
3057 read_unlock(&bond->curr_slave_lock);
3058 return;
3059 }
3060
3061 read_unlock(&bond->curr_slave_lock);
3062
3063 /* if we don't have a curr_active_slave, search for the next available
3064 * backup slave from the current_arp_slave and make it the candidate
3065 * for becoming the curr_active_slave
3066 */
3067
3068 if (!bond->current_arp_slave) {
3069 bond->current_arp_slave = bond->first_slave;
3070 if (!bond->current_arp_slave)
3071 return;
3072 }
3073
3074 bond_set_slave_inactive_flags(bond->current_arp_slave);
3075
3076 /* search for next candidate */
3077 bond_for_each_slave_from(bond, slave, i, bond->current_arp_slave->next) {
3078 if (IS_UP(slave->dev)) {
3079 slave->link = BOND_LINK_BACK;
3080 bond_set_slave_active_flags(slave);
3081 bond_arp_send_all(bond, slave);
3082 slave->jiffies = jiffies;
3083 bond->current_arp_slave = slave;
3084 break;
3085 }
3086
3087 /* if the link state is up at this point, we
3088 * mark it down - this can happen if we have
3089 * simultaneous link failures and
3090 * reselect_active_interface doesn't make this
3091 * one the current slave so it is still marked
3092 * up when it is actually down
3093 */
3094 if (slave->link == BOND_LINK_UP) {
3095 slave->link = BOND_LINK_DOWN;
3096 if (slave->link_failure_count < UINT_MAX)
3097 slave->link_failure_count++;
3098
3099 bond_set_slave_inactive_flags(slave);
3100
3101 pr_info(DRV_NAME
3102 ": %s: backup interface %s is now down.\n",
3103 bond->dev->name, slave->dev->name);
3104 }
3105 }
3106 }
3107
3108 void bond_activebackup_arp_mon(struct work_struct *work)
3109 {
3110 struct bonding *bond = container_of(work, struct bonding,
3111 arp_work.work);
3112 int delta_in_ticks;
3113
3114 read_lock(&bond->lock);
3115
3116 if (bond->kill_timers)
3117 goto out;
3118
3119 delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
3120
3121 if (bond->slave_cnt == 0)
3122 goto re_arm;
3123
3124 if (bond->send_grat_arp) {
3125 read_lock(&bond->curr_slave_lock);
3126 bond_send_gratuitous_arp(bond);
3127 read_unlock(&bond->curr_slave_lock);
3128 }
3129
3130 if (bond->send_unsol_na) {
3131 read_lock(&bond->curr_slave_lock);
3132 bond_send_unsolicited_na(bond);
3133 read_unlock(&bond->curr_slave_lock);
3134 }
3135
3136 if (bond_ab_arp_inspect(bond, delta_in_ticks)) {
3137 read_unlock(&bond->lock);
3138 rtnl_lock();
3139 read_lock(&bond->lock);
3140
3141 bond_ab_arp_commit(bond, delta_in_ticks);
3142
3143 read_unlock(&bond->lock);
3144 rtnl_unlock();
3145 read_lock(&bond->lock);
3146 }
3147
3148 bond_ab_arp_probe(bond);
3149
3150 re_arm:
3151 if (bond->params.arp_interval)
3152 queue_delayed_work(bond->wq, &bond->arp_work, delta_in_ticks);
3153 out:
3154 read_unlock(&bond->lock);
3155 }
3156
3157 /*------------------------------ proc/seq_file-------------------------------*/
3158
3159 #ifdef CONFIG_PROC_FS
3160
3161 static void *bond_info_seq_start(struct seq_file *seq, loff_t *pos)
3162 __acquires(&dev_base_lock)
3163 __acquires(&bond->lock)
3164 {
3165 struct bonding *bond = seq->private;
3166 loff_t off = 0;
3167 struct slave *slave;
3168 int i;
3169
3170 /* make sure the bond won't be taken away */
3171 read_lock(&dev_base_lock);
3172 read_lock(&bond->lock);
3173
3174 if (*pos == 0)
3175 return SEQ_START_TOKEN;
3176
3177 bond_for_each_slave(bond, slave, i) {
3178 if (++off == *pos)
3179 return slave;
3180 }
3181
3182 return NULL;
3183 }
3184
3185 static void *bond_info_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3186 {
3187 struct bonding *bond = seq->private;
3188 struct slave *slave = v;
3189
3190 ++*pos;
3191 if (v == SEQ_START_TOKEN)
3192 return bond->first_slave;
3193
3194 slave = slave->next;
3195
3196 return (slave == bond->first_slave) ? NULL : slave;
3197 }
3198
3199 static void bond_info_seq_stop(struct seq_file *seq, void *v)
3200 __releases(&bond->lock)
3201 __releases(&dev_base_lock)
3202 {
3203 struct bonding *bond = seq->private;
3204
3205 read_unlock(&bond->lock);
3206 read_unlock(&dev_base_lock);
3207 }
3208
3209 static void bond_info_show_master(struct seq_file *seq)
3210 {
3211 struct bonding *bond = seq->private;
3212 struct slave *curr;
3213 int i;
3214
3215 read_lock(&bond->curr_slave_lock);
3216 curr = bond->curr_active_slave;
3217 read_unlock(&bond->curr_slave_lock);
3218
3219 seq_printf(seq, "Bonding Mode: %s",
3220 bond_mode_name(bond->params.mode));
3221
3222 if (bond->params.mode == BOND_MODE_ACTIVEBACKUP &&
3223 bond->params.fail_over_mac)
3224 seq_printf(seq, " (fail_over_mac %s)",
3225 fail_over_mac_tbl[bond->params.fail_over_mac].modename);
3226
3227 seq_printf(seq, "\n");
3228
3229 if (bond->params.mode == BOND_MODE_XOR ||
3230 bond->params.mode == BOND_MODE_8023AD) {
3231 seq_printf(seq, "Transmit Hash Policy: %s (%d)\n",
3232 xmit_hashtype_tbl[bond->params.xmit_policy].modename,
3233 bond->params.xmit_policy);
3234 }
3235
3236 if (USES_PRIMARY(bond->params.mode)) {
3237 seq_printf(seq, "Primary Slave: %s\n",
3238 (bond->primary_slave) ?
3239 bond->primary_slave->dev->name : "None");
3240
3241 seq_printf(seq, "Currently Active Slave: %s\n",
3242 (curr) ? curr->dev->name : "None");
3243 }
3244
3245 seq_printf(seq, "MII Status: %s\n", netif_carrier_ok(bond->dev) ?
3246 "up" : "down");
3247 seq_printf(seq, "MII Polling Interval (ms): %d\n", bond->params.miimon);
3248 seq_printf(seq, "Up Delay (ms): %d\n",
3249 bond->params.updelay * bond->params.miimon);
3250 seq_printf(seq, "Down Delay (ms): %d\n",
3251 bond->params.downdelay * bond->params.miimon);
3252
3253
3254 /* ARP information */
3255 if (bond->params.arp_interval > 0) {
3256 int printed = 0;
3257 seq_printf(seq, "ARP Polling Interval (ms): %d\n",
3258 bond->params.arp_interval);
3259
3260 seq_printf(seq, "ARP IP target/s (n.n.n.n form):");
3261
3262 for (i = 0; (i < BOND_MAX_ARP_TARGETS); i++) {
3263 if (!bond->params.arp_targets[i])
3264 break;
3265 if (printed)
3266 seq_printf(seq, ",");
3267 seq_printf(seq, " %pI4", &bond->params.arp_targets[i]);
3268 printed = 1;
3269 }
3270 seq_printf(seq, "\n");
3271 }
3272
3273 if (bond->params.mode == BOND_MODE_8023AD) {
3274 struct ad_info ad_info;
3275
3276 seq_puts(seq, "\n802.3ad info\n");
3277 seq_printf(seq, "LACP rate: %s\n",
3278 (bond->params.lacp_fast) ? "fast" : "slow");
3279 seq_printf(seq, "Aggregator selection policy (ad_select): %s\n",
3280 ad_select_tbl[bond->params.ad_select].modename);
3281
3282 if (bond_3ad_get_active_agg_info(bond, &ad_info)) {
3283 seq_printf(seq, "bond %s has no active aggregator\n",
3284 bond->dev->name);
3285 } else {
3286 seq_printf(seq, "Active Aggregator Info:\n");
3287
3288 seq_printf(seq, "\tAggregator ID: %d\n",
3289 ad_info.aggregator_id);
3290 seq_printf(seq, "\tNumber of ports: %d\n",
3291 ad_info.ports);
3292 seq_printf(seq, "\tActor Key: %d\n",
3293 ad_info.actor_key);
3294 seq_printf(seq, "\tPartner Key: %d\n",
3295 ad_info.partner_key);
3296 seq_printf(seq, "\tPartner Mac Address: %pM\n",
3297 ad_info.partner_system);
3298 }
3299 }
3300 }
3301
3302 static void bond_info_show_slave(struct seq_file *seq,
3303 const struct slave *slave)
3304 {
3305 struct bonding *bond = seq->private;
3306
3307 seq_printf(seq, "\nSlave Interface: %s\n", slave->dev->name);
3308 seq_printf(seq, "MII Status: %s\n",
3309 (slave->link == BOND_LINK_UP) ? "up" : "down");
3310 seq_printf(seq, "Link Failure Count: %u\n",
3311 slave->link_failure_count);
3312
3313 seq_printf(seq, "Permanent HW addr: %pM\n", slave->perm_hwaddr);
3314
3315 if (bond->params.mode == BOND_MODE_8023AD) {
3316 const struct aggregator *agg
3317 = SLAVE_AD_INFO(slave).port.aggregator;
3318
3319 if (agg)
3320 seq_printf(seq, "Aggregator ID: %d\n",
3321 agg->aggregator_identifier);
3322 else
3323 seq_puts(seq, "Aggregator ID: N/A\n");
3324 }
3325 }
3326
3327 static int bond_info_seq_show(struct seq_file *seq, void *v)
3328 {
3329 if (v == SEQ_START_TOKEN) {
3330 seq_printf(seq, "%s\n", version);
3331 bond_info_show_master(seq);
3332 } else
3333 bond_info_show_slave(seq, v);
3334
3335 return 0;
3336 }
3337
3338 static const struct seq_operations bond_info_seq_ops = {
3339 .start = bond_info_seq_start,
3340 .next = bond_info_seq_next,
3341 .stop = bond_info_seq_stop,
3342 .show = bond_info_seq_show,
3343 };
3344
3345 static int bond_info_open(struct inode *inode, struct file *file)
3346 {
3347 struct seq_file *seq;
3348 struct proc_dir_entry *proc;
3349 int res;
3350
3351 res = seq_open(file, &bond_info_seq_ops);
3352 if (!res) {
3353 /* recover the pointer buried in proc_dir_entry data */
3354 seq = file->private_data;
3355 proc = PDE(inode);
3356 seq->private = proc->data;
3357 }
3358
3359 return res;
3360 }
3361
3362 static const struct file_operations bond_info_fops = {
3363 .owner = THIS_MODULE,
3364 .open = bond_info_open,
3365 .read = seq_read,
3366 .llseek = seq_lseek,
3367 .release = seq_release,
3368 };
3369
3370 static int bond_create_proc_entry(struct bonding *bond)
3371 {
3372 struct net_device *bond_dev = bond->dev;
3373
3374 if (bond_proc_dir) {
3375 bond->proc_entry = proc_create_data(bond_dev->name,
3376 S_IRUGO, bond_proc_dir,
3377 &bond_info_fops, bond);
3378 if (bond->proc_entry == NULL)
3379 pr_warning(DRV_NAME
3380 ": Warning: Cannot create /proc/net/%s/%s\n",
3381 DRV_NAME, bond_dev->name);
3382 else
3383 memcpy(bond->proc_file_name, bond_dev->name, IFNAMSIZ);
3384 }
3385
3386 return 0;
3387 }
3388
3389 static void bond_remove_proc_entry(struct bonding *bond)
3390 {
3391 if (bond_proc_dir && bond->proc_entry) {
3392 remove_proc_entry(bond->proc_file_name, bond_proc_dir);
3393 memset(bond->proc_file_name, 0, IFNAMSIZ);
3394 bond->proc_entry = NULL;
3395 }
3396 }
3397
3398 /* Create the bonding directory under /proc/net, if doesn't exist yet.
3399 * Caller must hold rtnl_lock.
3400 */
3401 static void bond_create_proc_dir(void)
3402 {
3403 if (!bond_proc_dir) {
3404 bond_proc_dir = proc_mkdir(DRV_NAME, init_net.proc_net);
3405 if (!bond_proc_dir)
3406 pr_warning(DRV_NAME
3407 ": Warning: cannot create /proc/net/%s\n",
3408 DRV_NAME);
3409 }
3410 }
3411
3412 /* Destroy the bonding directory under /proc/net, if empty.
3413 * Caller must hold rtnl_lock.
3414 */
3415 static void bond_destroy_proc_dir(void)
3416 {
3417 if (bond_proc_dir) {
3418 remove_proc_entry(DRV_NAME, init_net.proc_net);
3419 bond_proc_dir = NULL;
3420 }
3421 }
3422
3423 #else /* !CONFIG_PROC_FS */
3424
3425 static int bond_create_proc_entry(struct bonding *bond)
3426 {
3427 }
3428
3429 static void bond_remove_proc_entry(struct bonding *bond)
3430 {
3431 }
3432
3433 static void bond_create_proc_dir(void)
3434 {
3435 }
3436
3437 static void bond_destroy_proc_dir(void)
3438 {
3439 }
3440
3441 #endif /* CONFIG_PROC_FS */
3442
3443
3444 /*-------------------------- netdev event handling --------------------------*/
3445
3446 /*
3447 * Change device name
3448 */
3449 static int bond_event_changename(struct bonding *bond)
3450 {
3451 bond_remove_proc_entry(bond);
3452 bond_create_proc_entry(bond);
3453
3454 bond_destroy_sysfs_entry(bond);
3455 bond_create_sysfs_entry(bond);
3456
3457 return NOTIFY_DONE;
3458 }
3459
3460 static int bond_master_netdev_event(unsigned long event,
3461 struct net_device *bond_dev)
3462 {
3463 struct bonding *event_bond = netdev_priv(bond_dev);
3464
3465 switch (event) {
3466 case NETDEV_CHANGENAME:
3467 return bond_event_changename(event_bond);
3468 case NETDEV_UNREGISTER:
3469 bond_release_all(event_bond->dev);
3470 break;
3471 default:
3472 break;
3473 }
3474
3475 return NOTIFY_DONE;
3476 }
3477
3478 static int bond_slave_netdev_event(unsigned long event,
3479 struct net_device *slave_dev)
3480 {
3481 struct net_device *bond_dev = slave_dev->master;
3482 struct bonding *bond = netdev_priv(bond_dev);
3483
3484 switch (event) {
3485 case NETDEV_UNREGISTER:
3486 if (bond_dev) {
3487 if (bond->setup_by_slave)
3488 bond_release_and_destroy(bond_dev, slave_dev);
3489 else
3490 bond_release(bond_dev, slave_dev);
3491 }
3492 break;
3493 case NETDEV_CHANGE:
3494 if (bond->params.mode == BOND_MODE_8023AD || bond_is_lb(bond)) {
3495 struct slave *slave;
3496
3497 slave = bond_get_slave_by_dev(bond, slave_dev);
3498 if (slave) {
3499 u16 old_speed = slave->speed;
3500 u16 old_duplex = slave->duplex;
3501
3502 bond_update_speed_duplex(slave);
3503
3504 if (bond_is_lb(bond))
3505 break;
3506
3507 if (old_speed != slave->speed)
3508 bond_3ad_adapter_speed_changed(slave);
3509 if (old_duplex != slave->duplex)
3510 bond_3ad_adapter_duplex_changed(slave);
3511 }
3512 }
3513
3514 break;
3515 case NETDEV_DOWN:
3516 /*
3517 * ... Or is it this?
3518 */
3519 break;
3520 case NETDEV_CHANGEMTU:
3521 /*
3522 * TODO: Should slaves be allowed to
3523 * independently alter their MTU? For
3524 * an active-backup bond, slaves need
3525 * not be the same type of device, so
3526 * MTUs may vary. For other modes,
3527 * slaves arguably should have the
3528 * same MTUs. To do this, we'd need to
3529 * take over the slave's change_mtu
3530 * function for the duration of their
3531 * servitude.
3532 */
3533 break;
3534 case NETDEV_CHANGENAME:
3535 /*
3536 * TODO: handle changing the primary's name
3537 */
3538 break;
3539 case NETDEV_FEAT_CHANGE:
3540 bond_compute_features(bond);
3541 break;
3542 default:
3543 break;
3544 }
3545
3546 return NOTIFY_DONE;
3547 }
3548
3549 /*
3550 * bond_netdev_event: handle netdev notifier chain events.
3551 *
3552 * This function receives events for the netdev chain. The caller (an
3553 * ioctl handler calling blocking_notifier_call_chain) holds the necessary
3554 * locks for us to safely manipulate the slave devices (RTNL lock,
3555 * dev_probe_lock).
3556 */
3557 static int bond_netdev_event(struct notifier_block *this,
3558 unsigned long event, void *ptr)
3559 {
3560 struct net_device *event_dev = (struct net_device *)ptr;
3561
3562 if (dev_net(event_dev) != &init_net)
3563 return NOTIFY_DONE;
3564
3565 pr_debug("event_dev: %s, event: %lx\n",
3566 (event_dev ? event_dev->name : "None"),
3567 event);
3568
3569 if (!(event_dev->priv_flags & IFF_BONDING))
3570 return NOTIFY_DONE;
3571
3572 if (event_dev->flags & IFF_MASTER) {
3573 pr_debug("IFF_MASTER\n");
3574 return bond_master_netdev_event(event, event_dev);
3575 }
3576
3577 if (event_dev->flags & IFF_SLAVE) {
3578 pr_debug("IFF_SLAVE\n");
3579 return bond_slave_netdev_event(event, event_dev);
3580 }
3581
3582 return NOTIFY_DONE;
3583 }
3584
3585 /*
3586 * bond_inetaddr_event: handle inetaddr notifier chain events.
3587 *
3588 * We keep track of device IPs primarily to use as source addresses in
3589 * ARP monitor probes (rather than spewing out broadcasts all the time).
3590 *
3591 * We track one IP for the main device (if it has one), plus one per VLAN.
3592 */
3593 static int bond_inetaddr_event(struct notifier_block *this, unsigned long event, void *ptr)
3594 {
3595 struct in_ifaddr *ifa = ptr;
3596 struct net_device *vlan_dev, *event_dev = ifa->ifa_dev->dev;
3597 struct bonding *bond;
3598 struct vlan_entry *vlan;
3599
3600 if (dev_net(ifa->ifa_dev->dev) != &init_net)
3601 return NOTIFY_DONE;
3602
3603 list_for_each_entry(bond, &bond_dev_list, bond_list) {
3604 if (bond->dev == event_dev) {
3605 switch (event) {
3606 case NETDEV_UP:
3607 bond->master_ip = ifa->ifa_local;
3608 return NOTIFY_OK;
3609 case NETDEV_DOWN:
3610 bond->master_ip = bond_glean_dev_ip(bond->dev);
3611 return NOTIFY_OK;
3612 default:
3613 return NOTIFY_DONE;
3614 }
3615 }
3616
3617 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
3618 vlan_dev = vlan_group_get_device(bond->vlgrp, vlan->vlan_id);
3619 if (vlan_dev == event_dev) {
3620 switch (event) {
3621 case NETDEV_UP:
3622 vlan->vlan_ip = ifa->ifa_local;
3623 return NOTIFY_OK;
3624 case NETDEV_DOWN:
3625 vlan->vlan_ip =
3626 bond_glean_dev_ip(vlan_dev);
3627 return NOTIFY_OK;
3628 default:
3629 return NOTIFY_DONE;
3630 }
3631 }
3632 }
3633 }
3634 return NOTIFY_DONE;
3635 }
3636
3637 static struct notifier_block bond_netdev_notifier = {
3638 .notifier_call = bond_netdev_event,
3639 };
3640
3641 static struct notifier_block bond_inetaddr_notifier = {
3642 .notifier_call = bond_inetaddr_event,
3643 };
3644
3645 /*-------------------------- Packet type handling ---------------------------*/
3646
3647 /* register to receive lacpdus on a bond */
3648 static void bond_register_lacpdu(struct bonding *bond)
3649 {
3650 struct packet_type *pk_type = &(BOND_AD_INFO(bond).ad_pkt_type);
3651
3652 /* initialize packet type */
3653 pk_type->type = PKT_TYPE_LACPDU;
3654 pk_type->dev = bond->dev;
3655 pk_type->func = bond_3ad_lacpdu_recv;
3656
3657 dev_add_pack(pk_type);
3658 }
3659
3660 /* unregister to receive lacpdus on a bond */
3661 static void bond_unregister_lacpdu(struct bonding *bond)
3662 {
3663 dev_remove_pack(&(BOND_AD_INFO(bond).ad_pkt_type));
3664 }
3665
3666 void bond_register_arp(struct bonding *bond)
3667 {
3668 struct packet_type *pt = &bond->arp_mon_pt;
3669
3670 if (pt->type)
3671 return;
3672
3673 pt->type = htons(ETH_P_ARP);
3674 pt->dev = bond->dev;
3675 pt->func = bond_arp_rcv;
3676 dev_add_pack(pt);
3677 }
3678
3679 void bond_unregister_arp(struct bonding *bond)
3680 {
3681 struct packet_type *pt = &bond->arp_mon_pt;
3682
3683 dev_remove_pack(pt);
3684 pt->type = 0;
3685 }
3686
3687 /*---------------------------- Hashing Policies -----------------------------*/
3688
3689 /*
3690 * Hash for the output device based upon layer 2 and layer 3 data. If
3691 * the packet is not IP mimic bond_xmit_hash_policy_l2()
3692 */
3693 static int bond_xmit_hash_policy_l23(struct sk_buff *skb,
3694 struct net_device *bond_dev, int count)
3695 {
3696 struct ethhdr *data = (struct ethhdr *)skb->data;
3697 struct iphdr *iph = ip_hdr(skb);
3698
3699 if (skb->protocol == htons(ETH_P_IP)) {
3700 return ((ntohl(iph->saddr ^ iph->daddr) & 0xffff) ^
3701 (data->h_dest[5] ^ bond_dev->dev_addr[5])) % count;
3702 }
3703
3704 return (data->h_dest[5] ^ bond_dev->dev_addr[5]) % count;
3705 }
3706
3707 /*
3708 * Hash for the output device based upon layer 3 and layer 4 data. If
3709 * the packet is a frag or not TCP or UDP, just use layer 3 data. If it is
3710 * altogether not IP, mimic bond_xmit_hash_policy_l2()
3711 */
3712 static int bond_xmit_hash_policy_l34(struct sk_buff *skb,
3713 struct net_device *bond_dev, int count)
3714 {
3715 struct ethhdr *data = (struct ethhdr *)skb->data;
3716 struct iphdr *iph = ip_hdr(skb);
3717 __be16 *layer4hdr = (__be16 *)((u32 *)iph + iph->ihl);
3718 int layer4_xor = 0;
3719
3720 if (skb->protocol == htons(ETH_P_IP)) {
3721 if (!(iph->frag_off & htons(IP_MF|IP_OFFSET)) &&
3722 (iph->protocol == IPPROTO_TCP ||
3723 iph->protocol == IPPROTO_UDP)) {
3724 layer4_xor = ntohs((*layer4hdr ^ *(layer4hdr + 1)));
3725 }
3726 return (layer4_xor ^
3727 ((ntohl(iph->saddr ^ iph->daddr)) & 0xffff)) % count;
3728
3729 }
3730
3731 return (data->h_dest[5] ^ bond_dev->dev_addr[5]) % count;
3732 }
3733
3734 /*
3735 * Hash for the output device based upon layer 2 data
3736 */
3737 static int bond_xmit_hash_policy_l2(struct sk_buff *skb,
3738 struct net_device *bond_dev, int count)
3739 {
3740 struct ethhdr *data = (struct ethhdr *)skb->data;
3741
3742 return (data->h_dest[5] ^ bond_dev->dev_addr[5]) % count;
3743 }
3744
3745 /*-------------------------- Device entry points ----------------------------*/
3746
3747 static int bond_open(struct net_device *bond_dev)
3748 {
3749 struct bonding *bond = netdev_priv(bond_dev);
3750
3751 bond->kill_timers = 0;
3752
3753 if (bond_is_lb(bond)) {
3754 /* bond_alb_initialize must be called before the timer
3755 * is started.
3756 */
3757 if (bond_alb_initialize(bond, (bond->params.mode == BOND_MODE_ALB))) {
3758 /* something went wrong - fail the open operation */
3759 return -1;
3760 }
3761
3762 INIT_DELAYED_WORK(&bond->alb_work, bond_alb_monitor);
3763 queue_delayed_work(bond->wq, &bond->alb_work, 0);
3764 }
3765
3766 if (bond->params.miimon) { /* link check interval, in milliseconds. */
3767 INIT_DELAYED_WORK(&bond->mii_work, bond_mii_monitor);
3768 queue_delayed_work(bond->wq, &bond->mii_work, 0);
3769 }
3770
3771 if (bond->params.arp_interval) { /* arp interval, in milliseconds. */
3772 if (bond->params.mode == BOND_MODE_ACTIVEBACKUP)
3773 INIT_DELAYED_WORK(&bond->arp_work,
3774 bond_activebackup_arp_mon);
3775 else
3776 INIT_DELAYED_WORK(&bond->arp_work,
3777 bond_loadbalance_arp_mon);
3778
3779 queue_delayed_work(bond->wq, &bond->arp_work, 0);
3780 if (bond->params.arp_validate)
3781 bond_register_arp(bond);
3782 }
3783
3784 if (bond->params.mode == BOND_MODE_8023AD) {
3785 INIT_DELAYED_WORK(&bond->ad_work, bond_3ad_state_machine_handler);
3786 queue_delayed_work(bond->wq, &bond->ad_work, 0);
3787 /* register to receive LACPDUs */
3788 bond_register_lacpdu(bond);
3789 bond_3ad_initiate_agg_selection(bond, 1);
3790 }
3791
3792 return 0;
3793 }
3794
3795 static int bond_close(struct net_device *bond_dev)
3796 {
3797 struct bonding *bond = netdev_priv(bond_dev);
3798
3799 if (bond->params.mode == BOND_MODE_8023AD) {
3800 /* Unregister the receive of LACPDUs */
3801 bond_unregister_lacpdu(bond);
3802 }
3803
3804 if (bond->params.arp_validate)
3805 bond_unregister_arp(bond);
3806
3807 write_lock_bh(&bond->lock);
3808
3809 bond->send_grat_arp = 0;
3810 bond->send_unsol_na = 0;
3811
3812 /* signal timers not to re-arm */
3813 bond->kill_timers = 1;
3814
3815 write_unlock_bh(&bond->lock);
3816
3817 if (bond->params.miimon) { /* link check interval, in milliseconds. */
3818 cancel_delayed_work(&bond->mii_work);
3819 }
3820
3821 if (bond->params.arp_interval) { /* arp interval, in milliseconds. */
3822 cancel_delayed_work(&bond->arp_work);
3823 }
3824
3825 switch (bond->params.mode) {
3826 case BOND_MODE_8023AD:
3827 cancel_delayed_work(&bond->ad_work);
3828 break;
3829 case BOND_MODE_TLB:
3830 case BOND_MODE_ALB:
3831 cancel_delayed_work(&bond->alb_work);
3832 break;
3833 default:
3834 break;
3835 }
3836
3837
3838 if (bond_is_lb(bond)) {
3839 /* Must be called only after all
3840 * slaves have been released
3841 */
3842 bond_alb_deinitialize(bond);
3843 }
3844
3845 return 0;
3846 }
3847
3848 static struct net_device_stats *bond_get_stats(struct net_device *bond_dev)
3849 {
3850 struct bonding *bond = netdev_priv(bond_dev);
3851 struct net_device_stats *stats = &bond->stats;
3852 struct net_device_stats local_stats;
3853 struct slave *slave;
3854 int i;
3855
3856 memset(&local_stats, 0, sizeof(struct net_device_stats));
3857
3858 read_lock_bh(&bond->lock);
3859
3860 bond_for_each_slave(bond, slave, i) {
3861 const struct net_device_stats *sstats = dev_get_stats(slave->dev);
3862
3863 local_stats.rx_packets += sstats->rx_packets;
3864 local_stats.rx_bytes += sstats->rx_bytes;
3865 local_stats.rx_errors += sstats->rx_errors;
3866 local_stats.rx_dropped += sstats->rx_dropped;
3867
3868 local_stats.tx_packets += sstats->tx_packets;
3869 local_stats.tx_bytes += sstats->tx_bytes;
3870 local_stats.tx_errors += sstats->tx_errors;
3871 local_stats.tx_dropped += sstats->tx_dropped;
3872
3873 local_stats.multicast += sstats->multicast;
3874 local_stats.collisions += sstats->collisions;
3875
3876 local_stats.rx_length_errors += sstats->rx_length_errors;
3877 local_stats.rx_over_errors += sstats->rx_over_errors;
3878 local_stats.rx_crc_errors += sstats->rx_crc_errors;
3879 local_stats.rx_frame_errors += sstats->rx_frame_errors;
3880 local_stats.rx_fifo_errors += sstats->rx_fifo_errors;
3881 local_stats.rx_missed_errors += sstats->rx_missed_errors;
3882
3883 local_stats.tx_aborted_errors += sstats->tx_aborted_errors;
3884 local_stats.tx_carrier_errors += sstats->tx_carrier_errors;
3885 local_stats.tx_fifo_errors += sstats->tx_fifo_errors;
3886 local_stats.tx_heartbeat_errors += sstats->tx_heartbeat_errors;
3887 local_stats.tx_window_errors += sstats->tx_window_errors;
3888 }
3889
3890 memcpy(stats, &local_stats, sizeof(struct net_device_stats));
3891
3892 read_unlock_bh(&bond->lock);
3893
3894 return stats;
3895 }
3896
3897 static int bond_do_ioctl(struct net_device *bond_dev, struct ifreq *ifr, int cmd)
3898 {
3899 struct net_device *slave_dev = NULL;
3900 struct ifbond k_binfo;
3901 struct ifbond __user *u_binfo = NULL;
3902 struct ifslave k_sinfo;
3903 struct ifslave __user *u_sinfo = NULL;
3904 struct mii_ioctl_data *mii = NULL;
3905 int res = 0;
3906
3907 pr_debug("bond_ioctl: master=%s, cmd=%d\n",
3908 bond_dev->name, cmd);
3909
3910 switch (cmd) {
3911 case SIOCGMIIPHY:
3912 mii = if_mii(ifr);
3913 if (!mii)
3914 return -EINVAL;
3915
3916 mii->phy_id = 0;
3917 /* Fall Through */
3918 case SIOCGMIIREG:
3919 /*
3920 * We do this again just in case we were called by SIOCGMIIREG
3921 * instead of SIOCGMIIPHY.
3922 */
3923 mii = if_mii(ifr);
3924 if (!mii)
3925 return -EINVAL;
3926
3927
3928 if (mii->reg_num == 1) {
3929 struct bonding *bond = netdev_priv(bond_dev);
3930 mii->val_out = 0;
3931 read_lock(&bond->lock);
3932 read_lock(&bond->curr_slave_lock);
3933 if (netif_carrier_ok(bond->dev))
3934 mii->val_out = BMSR_LSTATUS;
3935
3936 read_unlock(&bond->curr_slave_lock);
3937 read_unlock(&bond->lock);
3938 }
3939
3940 return 0;
3941 case BOND_INFO_QUERY_OLD:
3942 case SIOCBONDINFOQUERY:
3943 u_binfo = (struct ifbond __user *)ifr->ifr_data;
3944
3945 if (copy_from_user(&k_binfo, u_binfo, sizeof(ifbond)))
3946 return -EFAULT;
3947
3948 res = bond_info_query(bond_dev, &k_binfo);
3949 if (res == 0 &&
3950 copy_to_user(u_binfo, &k_binfo, sizeof(ifbond)))
3951 return -EFAULT;
3952
3953 return res;
3954 case BOND_SLAVE_INFO_QUERY_OLD:
3955 case SIOCBONDSLAVEINFOQUERY:
3956 u_sinfo = (struct ifslave __user *)ifr->ifr_data;
3957
3958 if (copy_from_user(&k_sinfo, u_sinfo, sizeof(ifslave)))
3959 return -EFAULT;
3960
3961 res = bond_slave_info_query(bond_dev, &k_sinfo);
3962 if (res == 0 &&
3963 copy_to_user(u_sinfo, &k_sinfo, sizeof(ifslave)))
3964 return -EFAULT;
3965
3966 return res;
3967 default:
3968 /* Go on */
3969 break;
3970 }
3971
3972 if (!capable(CAP_NET_ADMIN))
3973 return -EPERM;
3974
3975 slave_dev = dev_get_by_name(&init_net, ifr->ifr_slave);
3976
3977 pr_debug("slave_dev=%p: \n", slave_dev);
3978
3979 if (!slave_dev)
3980 res = -ENODEV;
3981 else {
3982 pr_debug("slave_dev->name=%s: \n", slave_dev->name);
3983 switch (cmd) {
3984 case BOND_ENSLAVE_OLD:
3985 case SIOCBONDENSLAVE:
3986 res = bond_enslave(bond_dev, slave_dev);
3987 break;
3988 case BOND_RELEASE_OLD:
3989 case SIOCBONDRELEASE:
3990 res = bond_release(bond_dev, slave_dev);
3991 break;
3992 case BOND_SETHWADDR_OLD:
3993 case SIOCBONDSETHWADDR:
3994 res = bond_sethwaddr(bond_dev, slave_dev);
3995 break;
3996 case BOND_CHANGE_ACTIVE_OLD:
3997 case SIOCBONDCHANGEACTIVE:
3998 res = bond_ioctl_change_active(bond_dev, slave_dev);
3999 break;
4000 default:
4001 res = -EOPNOTSUPP;
4002 }
4003
4004 dev_put(slave_dev);
4005 }
4006
4007 return res;
4008 }
4009
4010 static void bond_set_multicast_list(struct net_device *bond_dev)
4011 {
4012 struct bonding *bond = netdev_priv(bond_dev);
4013 struct dev_mc_list *dmi;
4014
4015 /*
4016 * Do promisc before checking multicast_mode
4017 */
4018 if ((bond_dev->flags & IFF_PROMISC) && !(bond->flags & IFF_PROMISC))
4019 /*
4020 * FIXME: Need to handle the error when one of the multi-slaves
4021 * encounters error.
4022 */
4023 bond_set_promiscuity(bond, 1);
4024
4025
4026 if (!(bond_dev->flags & IFF_PROMISC) && (bond->flags & IFF_PROMISC))
4027 bond_set_promiscuity(bond, -1);
4028
4029
4030 /* set allmulti flag to slaves */
4031 if ((bond_dev->flags & IFF_ALLMULTI) && !(bond->flags & IFF_ALLMULTI))
4032 /*
4033 * FIXME: Need to handle the error when one of the multi-slaves
4034 * encounters error.
4035 */
4036 bond_set_allmulti(bond, 1);
4037
4038
4039 if (!(bond_dev->flags & IFF_ALLMULTI) && (bond->flags & IFF_ALLMULTI))
4040 bond_set_allmulti(bond, -1);
4041
4042
4043 read_lock(&bond->lock);
4044
4045 bond->flags = bond_dev->flags;
4046
4047 /* looking for addresses to add to slaves' mc list */
4048 for (dmi = bond_dev->mc_list; dmi; dmi = dmi->next) {
4049 if (!bond_mc_list_find_dmi(dmi, bond->mc_list))
4050 bond_mc_add(bond, dmi->dmi_addr, dmi->dmi_addrlen);
4051 }
4052
4053 /* looking for addresses to delete from slaves' list */
4054 for (dmi = bond->mc_list; dmi; dmi = dmi->next) {
4055 if (!bond_mc_list_find_dmi(dmi, bond_dev->mc_list))
4056 bond_mc_delete(bond, dmi->dmi_addr, dmi->dmi_addrlen);
4057 }
4058
4059 /* save master's multicast list */
4060 bond_mc_list_destroy(bond);
4061 bond_mc_list_copy(bond_dev->mc_list, bond, GFP_ATOMIC);
4062
4063 read_unlock(&bond->lock);
4064 }
4065
4066 static int bond_neigh_setup(struct net_device *dev, struct neigh_parms *parms)
4067 {
4068 struct bonding *bond = netdev_priv(dev);
4069 struct slave *slave = bond->first_slave;
4070
4071 if (slave) {
4072 const struct net_device_ops *slave_ops
4073 = slave->dev->netdev_ops;
4074 if (slave_ops->ndo_neigh_setup)
4075 return slave_ops->ndo_neigh_setup(slave->dev, parms);
4076 }
4077 return 0;
4078 }
4079
4080 /*
4081 * Change the MTU of all of a master's slaves to match the master
4082 */
4083 static int bond_change_mtu(struct net_device *bond_dev, int new_mtu)
4084 {
4085 struct bonding *bond = netdev_priv(bond_dev);
4086 struct slave *slave, *stop_at;
4087 int res = 0;
4088 int i;
4089
4090 pr_debug("bond=%p, name=%s, new_mtu=%d\n", bond,
4091 (bond_dev ? bond_dev->name : "None"), new_mtu);
4092
4093 /* Can't hold bond->lock with bh disabled here since
4094 * some base drivers panic. On the other hand we can't
4095 * hold bond->lock without bh disabled because we'll
4096 * deadlock. The only solution is to rely on the fact
4097 * that we're under rtnl_lock here, and the slaves
4098 * list won't change. This doesn't solve the problem
4099 * of setting the slave's MTU while it is
4100 * transmitting, but the assumption is that the base
4101 * driver can handle that.
4102 *
4103 * TODO: figure out a way to safely iterate the slaves
4104 * list, but without holding a lock around the actual
4105 * call to the base driver.
4106 */
4107
4108 bond_for_each_slave(bond, slave, i) {
4109 pr_debug("s %p s->p %p c_m %p\n", slave,
4110 slave->prev, slave->dev->netdev_ops->ndo_change_mtu);
4111
4112 res = dev_set_mtu(slave->dev, new_mtu);
4113
4114 if (res) {
4115 /* If we failed to set the slave's mtu to the new value
4116 * we must abort the operation even in ACTIVE_BACKUP
4117 * mode, because if we allow the backup slaves to have
4118 * different mtu values than the active slave we'll
4119 * need to change their mtu when doing a failover. That
4120 * means changing their mtu from timer context, which
4121 * is probably not a good idea.
4122 */
4123 pr_debug("err %d %s\n", res, slave->dev->name);
4124 goto unwind;
4125 }
4126 }
4127
4128 bond_dev->mtu = new_mtu;
4129
4130 return 0;
4131
4132 unwind:
4133 /* unwind from head to the slave that failed */
4134 stop_at = slave;
4135 bond_for_each_slave_from_to(bond, slave, i, bond->first_slave, stop_at) {
4136 int tmp_res;
4137
4138 tmp_res = dev_set_mtu(slave->dev, bond_dev->mtu);
4139 if (tmp_res) {
4140 pr_debug("unwind err %d dev %s\n", tmp_res,
4141 slave->dev->name);
4142 }
4143 }
4144
4145 return res;
4146 }
4147
4148 /*
4149 * Change HW address
4150 *
4151 * Note that many devices must be down to change the HW address, and
4152 * downing the master releases all slaves. We can make bonds full of
4153 * bonding devices to test this, however.
4154 */
4155 static int bond_set_mac_address(struct net_device *bond_dev, void *addr)
4156 {
4157 struct bonding *bond = netdev_priv(bond_dev);
4158 struct sockaddr *sa = addr, tmp_sa;
4159 struct slave *slave, *stop_at;
4160 int res = 0;
4161 int i;
4162
4163 if (bond->params.mode == BOND_MODE_ALB)
4164 return bond_alb_set_mac_address(bond_dev, addr);
4165
4166
4167 pr_debug("bond=%p, name=%s\n", bond, (bond_dev ? bond_dev->name : "None"));
4168
4169 /*
4170 * If fail_over_mac is set to active, do nothing and return
4171 * success. Returning an error causes ifenslave to fail.
4172 */
4173 if (bond->params.fail_over_mac == BOND_FOM_ACTIVE)
4174 return 0;
4175
4176 if (!is_valid_ether_addr(sa->sa_data))
4177 return -EADDRNOTAVAIL;
4178
4179 /* Can't hold bond->lock with bh disabled here since
4180 * some base drivers panic. On the other hand we can't
4181 * hold bond->lock without bh disabled because we'll
4182 * deadlock. The only solution is to rely on the fact
4183 * that we're under rtnl_lock here, and the slaves
4184 * list won't change. This doesn't solve the problem
4185 * of setting the slave's hw address while it is
4186 * transmitting, but the assumption is that the base
4187 * driver can handle that.
4188 *
4189 * TODO: figure out a way to safely iterate the slaves
4190 * list, but without holding a lock around the actual
4191 * call to the base driver.
4192 */
4193
4194 bond_for_each_slave(bond, slave, i) {
4195 const struct net_device_ops *slave_ops = slave->dev->netdev_ops;
4196 pr_debug("slave %p %s\n", slave, slave->dev->name);
4197
4198 if (slave_ops->ndo_set_mac_address == NULL) {
4199 res = -EOPNOTSUPP;
4200 pr_debug("EOPNOTSUPP %s\n", slave->dev->name);
4201 goto unwind;
4202 }
4203
4204 res = dev_set_mac_address(slave->dev, addr);
4205 if (res) {
4206 /* TODO: consider downing the slave
4207 * and retry ?
4208 * User should expect communications
4209 * breakage anyway until ARP finish
4210 * updating, so...
4211 */
4212 pr_debug("err %d %s\n", res, slave->dev->name);
4213 goto unwind;
4214 }
4215 }
4216
4217 /* success */
4218 memcpy(bond_dev->dev_addr, sa->sa_data, bond_dev->addr_len);
4219 return 0;
4220
4221 unwind:
4222 memcpy(tmp_sa.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
4223 tmp_sa.sa_family = bond_dev->type;
4224
4225 /* unwind from head to the slave that failed */
4226 stop_at = slave;
4227 bond_for_each_slave_from_to(bond, slave, i, bond->first_slave, stop_at) {
4228 int tmp_res;
4229
4230 tmp_res = dev_set_mac_address(slave->dev, &tmp_sa);
4231 if (tmp_res) {
4232 pr_debug("unwind err %d dev %s\n", tmp_res,
4233 slave->dev->name);
4234 }
4235 }
4236
4237 return res;
4238 }
4239
4240 static int bond_xmit_roundrobin(struct sk_buff *skb, struct net_device *bond_dev)
4241 {
4242 struct bonding *bond = netdev_priv(bond_dev);
4243 struct slave *slave, *start_at;
4244 int i, slave_no, res = 1;
4245
4246 read_lock(&bond->lock);
4247
4248 if (!BOND_IS_OK(bond))
4249 goto out;
4250
4251 /*
4252 * Concurrent TX may collide on rr_tx_counter; we accept that
4253 * as being rare enough not to justify using an atomic op here
4254 */
4255 slave_no = bond->rr_tx_counter++ % bond->slave_cnt;
4256
4257 bond_for_each_slave(bond, slave, i) {
4258 slave_no--;
4259 if (slave_no < 0)
4260 break;
4261 }
4262
4263 start_at = slave;
4264 bond_for_each_slave_from(bond, slave, i, start_at) {
4265 if (IS_UP(slave->dev) &&
4266 (slave->link == BOND_LINK_UP) &&
4267 (slave->state == BOND_STATE_ACTIVE)) {
4268 res = bond_dev_queue_xmit(bond, skb, slave->dev);
4269 break;
4270 }
4271 }
4272
4273 out:
4274 if (res) {
4275 /* no suitable interface, frame not sent */
4276 dev_kfree_skb(skb);
4277 }
4278 read_unlock(&bond->lock);
4279 return 0;
4280 }
4281
4282
4283 /*
4284 * in active-backup mode, we know that bond->curr_active_slave is always valid if
4285 * the bond has a usable interface.
4286 */
4287 static int bond_xmit_activebackup(struct sk_buff *skb, struct net_device *bond_dev)
4288 {
4289 struct bonding *bond = netdev_priv(bond_dev);
4290 int res = 1;
4291
4292 read_lock(&bond->lock);
4293 read_lock(&bond->curr_slave_lock);
4294
4295 if (!BOND_IS_OK(bond))
4296 goto out;
4297
4298 if (!bond->curr_active_slave)
4299 goto out;
4300
4301 res = bond_dev_queue_xmit(bond, skb, bond->curr_active_slave->dev);
4302
4303 out:
4304 if (res)
4305 /* no suitable interface, frame not sent */
4306 dev_kfree_skb(skb);
4307
4308 read_unlock(&bond->curr_slave_lock);
4309 read_unlock(&bond->lock);
4310 return 0;
4311 }
4312
4313 /*
4314 * In bond_xmit_xor() , we determine the output device by using a pre-
4315 * determined xmit_hash_policy(), If the selected device is not enabled,
4316 * find the next active slave.
4317 */
4318 static int bond_xmit_xor(struct sk_buff *skb, struct net_device *bond_dev)
4319 {
4320 struct bonding *bond = netdev_priv(bond_dev);
4321 struct slave *slave, *start_at;
4322 int slave_no;
4323 int i;
4324 int res = 1;
4325
4326 read_lock(&bond->lock);
4327
4328 if (!BOND_IS_OK(bond))
4329 goto out;
4330
4331 slave_no = bond->xmit_hash_policy(skb, bond_dev, bond->slave_cnt);
4332
4333 bond_for_each_slave(bond, slave, i) {
4334 slave_no--;
4335 if (slave_no < 0)
4336 break;
4337 }
4338
4339 start_at = slave;
4340
4341 bond_for_each_slave_from(bond, slave, i, start_at) {
4342 if (IS_UP(slave->dev) &&
4343 (slave->link == BOND_LINK_UP) &&
4344 (slave->state == BOND_STATE_ACTIVE)) {
4345 res = bond_dev_queue_xmit(bond, skb, slave->dev);
4346 break;
4347 }
4348 }
4349
4350 out:
4351 if (res) {
4352 /* no suitable interface, frame not sent */
4353 dev_kfree_skb(skb);
4354 }
4355 read_unlock(&bond->lock);
4356 return 0;
4357 }
4358
4359 /*
4360 * in broadcast mode, we send everything to all usable interfaces.
4361 */
4362 static int bond_xmit_broadcast(struct sk_buff *skb, struct net_device *bond_dev)
4363 {
4364 struct bonding *bond = netdev_priv(bond_dev);
4365 struct slave *slave, *start_at;
4366 struct net_device *tx_dev = NULL;
4367 int i;
4368 int res = 1;
4369
4370 read_lock(&bond->lock);
4371
4372 if (!BOND_IS_OK(bond))
4373 goto out;
4374
4375 read_lock(&bond->curr_slave_lock);
4376 start_at = bond->curr_active_slave;
4377 read_unlock(&bond->curr_slave_lock);
4378
4379 if (!start_at)
4380 goto out;
4381
4382 bond_for_each_slave_from(bond, slave, i, start_at) {
4383 if (IS_UP(slave->dev) &&
4384 (slave->link == BOND_LINK_UP) &&
4385 (slave->state == BOND_STATE_ACTIVE)) {
4386 if (tx_dev) {
4387 struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
4388 if (!skb2) {
4389 pr_err(DRV_NAME
4390 ": %s: Error: bond_xmit_broadcast(): "
4391 "skb_clone() failed\n",
4392 bond_dev->name);
4393 continue;
4394 }
4395
4396 res = bond_dev_queue_xmit(bond, skb2, tx_dev);
4397 if (res) {
4398 dev_kfree_skb(skb2);
4399 continue;
4400 }
4401 }
4402 tx_dev = slave->dev;
4403 }
4404 }
4405
4406 if (tx_dev)
4407 res = bond_dev_queue_xmit(bond, skb, tx_dev);
4408
4409 out:
4410 if (res)
4411 /* no suitable interface, frame not sent */
4412 dev_kfree_skb(skb);
4413
4414 /* frame sent to all suitable interfaces */
4415 read_unlock(&bond->lock);
4416 return 0;
4417 }
4418
4419 /*------------------------- Device initialization ---------------------------*/
4420
4421 static void bond_set_xmit_hash_policy(struct bonding *bond)
4422 {
4423 switch (bond->params.xmit_policy) {
4424 case BOND_XMIT_POLICY_LAYER23:
4425 bond->xmit_hash_policy = bond_xmit_hash_policy_l23;
4426 break;
4427 case BOND_XMIT_POLICY_LAYER34:
4428 bond->xmit_hash_policy = bond_xmit_hash_policy_l34;
4429 break;
4430 case BOND_XMIT_POLICY_LAYER2:
4431 default:
4432 bond->xmit_hash_policy = bond_xmit_hash_policy_l2;
4433 break;
4434 }
4435 }
4436
4437 static int bond_start_xmit(struct sk_buff *skb, struct net_device *dev)
4438 {
4439 const struct bonding *bond = netdev_priv(dev);
4440
4441 switch (bond->params.mode) {
4442 case BOND_MODE_ROUNDROBIN:
4443 return bond_xmit_roundrobin(skb, dev);
4444 case BOND_MODE_ACTIVEBACKUP:
4445 return bond_xmit_activebackup(skb, dev);
4446 case BOND_MODE_XOR:
4447 return bond_xmit_xor(skb, dev);
4448 case BOND_MODE_BROADCAST:
4449 return bond_xmit_broadcast(skb, dev);
4450 case BOND_MODE_8023AD:
4451 return bond_3ad_xmit_xor(skb, dev);
4452 case BOND_MODE_ALB:
4453 case BOND_MODE_TLB:
4454 return bond_alb_xmit(skb, dev);
4455 default:
4456 /* Should never happen, mode already checked */
4457 pr_err(DRV_NAME ": %s: Error: Unknown bonding mode %d\n",
4458 dev->name, bond->params.mode);
4459 WARN_ON_ONCE(1);
4460 dev_kfree_skb(skb);
4461 return NETDEV_TX_OK;
4462 }
4463 }
4464
4465
4466 /*
4467 * set bond mode specific net device operations
4468 */
4469 void bond_set_mode_ops(struct bonding *bond, int mode)
4470 {
4471 struct net_device *bond_dev = bond->dev;
4472
4473 switch (mode) {
4474 case BOND_MODE_ROUNDROBIN:
4475 break;
4476 case BOND_MODE_ACTIVEBACKUP:
4477 break;
4478 case BOND_MODE_XOR:
4479 bond_set_xmit_hash_policy(bond);
4480 break;
4481 case BOND_MODE_BROADCAST:
4482 break;
4483 case BOND_MODE_8023AD:
4484 bond_set_master_3ad_flags(bond);
4485 bond_set_xmit_hash_policy(bond);
4486 break;
4487 case BOND_MODE_ALB:
4488 bond_set_master_alb_flags(bond);
4489 /* FALLTHRU */
4490 case BOND_MODE_TLB:
4491 break;
4492 default:
4493 /* Should never happen, mode already checked */
4494 pr_err(DRV_NAME
4495 ": %s: Error: Unknown bonding mode %d\n",
4496 bond_dev->name,
4497 mode);
4498 break;
4499 }
4500 }
4501
4502 static void bond_ethtool_get_drvinfo(struct net_device *bond_dev,
4503 struct ethtool_drvinfo *drvinfo)
4504 {
4505 strncpy(drvinfo->driver, DRV_NAME, 32);
4506 strncpy(drvinfo->version, DRV_VERSION, 32);
4507 snprintf(drvinfo->fw_version, 32, "%d", BOND_ABI_VERSION);
4508 }
4509
4510 static const struct ethtool_ops bond_ethtool_ops = {
4511 .get_drvinfo = bond_ethtool_get_drvinfo,
4512 .get_link = ethtool_op_get_link,
4513 .get_tx_csum = ethtool_op_get_tx_csum,
4514 .get_sg = ethtool_op_get_sg,
4515 .get_tso = ethtool_op_get_tso,
4516 .get_ufo = ethtool_op_get_ufo,
4517 .get_flags = ethtool_op_get_flags,
4518 };
4519
4520 static const struct net_device_ops bond_netdev_ops = {
4521 .ndo_uninit = bond_uninit,
4522 .ndo_open = bond_open,
4523 .ndo_stop = bond_close,
4524 .ndo_start_xmit = bond_start_xmit,
4525 .ndo_get_stats = bond_get_stats,
4526 .ndo_do_ioctl = bond_do_ioctl,
4527 .ndo_set_multicast_list = bond_set_multicast_list,
4528 .ndo_change_mtu = bond_change_mtu,
4529 .ndo_set_mac_address = bond_set_mac_address,
4530 .ndo_neigh_setup = bond_neigh_setup,
4531 .ndo_vlan_rx_register = bond_vlan_rx_register,
4532 .ndo_vlan_rx_add_vid = bond_vlan_rx_add_vid,
4533 .ndo_vlan_rx_kill_vid = bond_vlan_rx_kill_vid,
4534 };
4535
4536 /*
4537 * Does not allocate but creates a /proc entry.
4538 * Allowed to fail.
4539 */
4540 static int bond_init(struct net_device *bond_dev)
4541 {
4542 struct bonding *bond = netdev_priv(bond_dev);
4543
4544 pr_debug("Begin bond_init for %s\n", bond_dev->name);
4545
4546 /* initialize rwlocks */
4547 rwlock_init(&bond->lock);
4548 rwlock_init(&bond->curr_slave_lock);
4549
4550 bond->params = bonding_defaults;
4551
4552 bond->wq = create_singlethread_workqueue(bond_dev->name);
4553 if (!bond->wq)
4554 return -ENOMEM;
4555
4556 /* Initialize pointers */
4557 bond->first_slave = NULL;
4558 bond->curr_active_slave = NULL;
4559 bond->current_arp_slave = NULL;
4560 bond->primary_slave = NULL;
4561 bond->dev = bond_dev;
4562 bond->send_grat_arp = 0;
4563 bond->send_unsol_na = 0;
4564 bond->setup_by_slave = 0;
4565 INIT_LIST_HEAD(&bond->vlan_list);
4566
4567 /* Initialize the device entry points */
4568 bond_dev->netdev_ops = &bond_netdev_ops;
4569 bond_dev->ethtool_ops = &bond_ethtool_ops;
4570 bond_set_mode_ops(bond, bond->params.mode);
4571
4572 bond_dev->destructor = free_netdev;
4573
4574 /* Initialize the device options */
4575 bond_dev->tx_queue_len = 0;
4576 bond_dev->flags |= IFF_MASTER|IFF_MULTICAST;
4577 bond_dev->priv_flags |= IFF_BONDING;
4578 if (bond->params.arp_interval)
4579 bond_dev->priv_flags |= IFF_MASTER_ARPMON;
4580
4581 /* At first, we block adding VLANs. That's the only way to
4582 * prevent problems that occur when adding VLANs over an
4583 * empty bond. The block will be removed once non-challenged
4584 * slaves are enslaved.
4585 */
4586 bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
4587
4588 /* don't acquire bond device's netif_tx_lock when
4589 * transmitting */
4590 bond_dev->features |= NETIF_F_LLTX;
4591
4592 /* By default, we declare the bond to be fully
4593 * VLAN hardware accelerated capable. Special
4594 * care is taken in the various xmit functions
4595 * when there are slaves that are not hw accel
4596 * capable
4597 */
4598 bond_dev->features |= (NETIF_F_HW_VLAN_TX |
4599 NETIF_F_HW_VLAN_RX |
4600 NETIF_F_HW_VLAN_FILTER);
4601
4602 bond_create_proc_entry(bond);
4603 list_add_tail(&bond->bond_list, &bond_dev_list);
4604
4605 return 0;
4606 }
4607
4608 static void bond_work_cancel_all(struct bonding *bond)
4609 {
4610 write_lock_bh(&bond->lock);
4611 bond->kill_timers = 1;
4612 write_unlock_bh(&bond->lock);
4613
4614 if (bond->params.miimon && delayed_work_pending(&bond->mii_work))
4615 cancel_delayed_work(&bond->mii_work);
4616
4617 if (bond->params.arp_interval && delayed_work_pending(&bond->arp_work))
4618 cancel_delayed_work(&bond->arp_work);
4619
4620 if (bond->params.mode == BOND_MODE_ALB &&
4621 delayed_work_pending(&bond->alb_work))
4622 cancel_delayed_work(&bond->alb_work);
4623
4624 if (bond->params.mode == BOND_MODE_8023AD &&
4625 delayed_work_pending(&bond->ad_work))
4626 cancel_delayed_work(&bond->ad_work);
4627 }
4628
4629 /* De-initialize device specific data.
4630 * Caller must hold rtnl_lock.
4631 */
4632 static void bond_deinit(struct net_device *bond_dev)
4633 {
4634 struct bonding *bond = netdev_priv(bond_dev);
4635
4636 list_del(&bond->bond_list);
4637
4638 bond_work_cancel_all(bond);
4639
4640 bond_remove_proc_entry(bond);
4641 }
4642
4643 /* Unregister and free all bond devices.
4644 * Caller must hold rtnl_lock.
4645 */
4646 static void bond_free_all(void)
4647 {
4648 struct bonding *bond, *nxt;
4649
4650 list_for_each_entry_safe(bond, nxt, &bond_dev_list, bond_list) {
4651 struct net_device *bond_dev = bond->dev;
4652
4653 bond_work_cancel_all(bond);
4654 /* Release the bonded slaves */
4655 bond_release_all(bond_dev);
4656 unregister_netdevice(bond_dev);
4657 }
4658
4659 bond_destroy_proc_dir();
4660 }
4661
4662 /*------------------------- Module initialization ---------------------------*/
4663
4664 /*
4665 * Convert string input module parms. Accept either the
4666 * number of the mode or its string name. A bit complicated because
4667 * some mode names are substrings of other names, and calls from sysfs
4668 * may have whitespace in the name (trailing newlines, for example).
4669 */
4670 int bond_parse_parm(const char *buf, const struct bond_parm_tbl *tbl)
4671 {
4672 int modeint = -1, i, rv;
4673 char *p, modestr[BOND_MAX_MODENAME_LEN + 1] = { 0, };
4674
4675 for (p = (char *)buf; *p; p++)
4676 if (!(isdigit(*p) || isspace(*p)))
4677 break;
4678
4679 if (*p)
4680 rv = sscanf(buf, "%20s", modestr);
4681 else
4682 rv = sscanf(buf, "%d", &modeint);
4683
4684 if (!rv)
4685 return -1;
4686
4687 for (i = 0; tbl[i].modename; i++) {
4688 if (modeint == tbl[i].mode)
4689 return tbl[i].mode;
4690 if (strcmp(modestr, tbl[i].modename) == 0)
4691 return tbl[i].mode;
4692 }
4693
4694 return -1;
4695 }
4696
4697 static int bond_check_params(struct bond_params *params)
4698 {
4699 int arp_validate_value, fail_over_mac_value;
4700
4701 /*
4702 * Convert string parameters.
4703 */
4704 if (mode) {
4705 bond_mode = bond_parse_parm(mode, bond_mode_tbl);
4706 if (bond_mode == -1) {
4707 pr_err(DRV_NAME
4708 ": Error: Invalid bonding mode \"%s\"\n",
4709 mode == NULL ? "NULL" : mode);
4710 return -EINVAL;
4711 }
4712 }
4713
4714 if (xmit_hash_policy) {
4715 if ((bond_mode != BOND_MODE_XOR) &&
4716 (bond_mode != BOND_MODE_8023AD)) {
4717 pr_info(DRV_NAME
4718 ": xor_mode param is irrelevant in mode %s\n",
4719 bond_mode_name(bond_mode));
4720 } else {
4721 xmit_hashtype = bond_parse_parm(xmit_hash_policy,
4722 xmit_hashtype_tbl);
4723 if (xmit_hashtype == -1) {
4724 pr_err(DRV_NAME
4725 ": Error: Invalid xmit_hash_policy \"%s\"\n",
4726 xmit_hash_policy == NULL ? "NULL" :
4727 xmit_hash_policy);
4728 return -EINVAL;
4729 }
4730 }
4731 }
4732
4733 if (lacp_rate) {
4734 if (bond_mode != BOND_MODE_8023AD) {
4735 pr_info(DRV_NAME
4736 ": lacp_rate param is irrelevant in mode %s\n",
4737 bond_mode_name(bond_mode));
4738 } else {
4739 lacp_fast = bond_parse_parm(lacp_rate, bond_lacp_tbl);
4740 if (lacp_fast == -1) {
4741 pr_err(DRV_NAME
4742 ": Error: Invalid lacp rate \"%s\"\n",
4743 lacp_rate == NULL ? "NULL" : lacp_rate);
4744 return -EINVAL;
4745 }
4746 }
4747 }
4748
4749 if (ad_select) {
4750 params->ad_select = bond_parse_parm(ad_select, ad_select_tbl);
4751 if (params->ad_select == -1) {
4752 pr_err(DRV_NAME
4753 ": Error: Invalid ad_select \"%s\"\n",
4754 ad_select == NULL ? "NULL" : ad_select);
4755 return -EINVAL;
4756 }
4757
4758 if (bond_mode != BOND_MODE_8023AD) {
4759 pr_warning(DRV_NAME
4760 ": ad_select param only affects 802.3ad mode\n");
4761 }
4762 } else {
4763 params->ad_select = BOND_AD_STABLE;
4764 }
4765
4766 if (max_bonds < 0 || max_bonds > INT_MAX) {
4767 pr_warning(DRV_NAME
4768 ": Warning: max_bonds (%d) not in range %d-%d, so it "
4769 "was reset to BOND_DEFAULT_MAX_BONDS (%d)\n",
4770 max_bonds, 0, INT_MAX, BOND_DEFAULT_MAX_BONDS);
4771 max_bonds = BOND_DEFAULT_MAX_BONDS;
4772 }
4773
4774 if (miimon < 0) {
4775 pr_warning(DRV_NAME
4776 ": Warning: miimon module parameter (%d), "
4777 "not in range 0-%d, so it was reset to %d\n",
4778 miimon, INT_MAX, BOND_LINK_MON_INTERV);
4779 miimon = BOND_LINK_MON_INTERV;
4780 }
4781
4782 if (updelay < 0) {
4783 pr_warning(DRV_NAME
4784 ": Warning: updelay module parameter (%d), "
4785 "not in range 0-%d, so it was reset to 0\n",
4786 updelay, INT_MAX);
4787 updelay = 0;
4788 }
4789
4790 if (downdelay < 0) {
4791 pr_warning(DRV_NAME
4792 ": Warning: downdelay module parameter (%d), "
4793 "not in range 0-%d, so it was reset to 0\n",
4794 downdelay, INT_MAX);
4795 downdelay = 0;
4796 }
4797
4798 if ((use_carrier != 0) && (use_carrier != 1)) {
4799 pr_warning(DRV_NAME
4800 ": Warning: use_carrier module parameter (%d), "
4801 "not of valid value (0/1), so it was set to 1\n",
4802 use_carrier);
4803 use_carrier = 1;
4804 }
4805
4806 if (num_grat_arp < 0 || num_grat_arp > 255) {
4807 pr_warning(DRV_NAME
4808 ": Warning: num_grat_arp (%d) not in range 0-255 so it "
4809 "was reset to 1 \n", num_grat_arp);
4810 num_grat_arp = 1;
4811 }
4812
4813 if (num_unsol_na < 0 || num_unsol_na > 255) {
4814 pr_warning(DRV_NAME
4815 ": Warning: num_unsol_na (%d) not in range 0-255 so it "
4816 "was reset to 1 \n", num_unsol_na);
4817 num_unsol_na = 1;
4818 }
4819
4820 /* reset values for 802.3ad */
4821 if (bond_mode == BOND_MODE_8023AD) {
4822 if (!miimon) {
4823 pr_warning(DRV_NAME
4824 ": Warning: miimon must be specified, "
4825 "otherwise bonding will not detect link "
4826 "failure, speed and duplex which are "
4827 "essential for 802.3ad operation\n");
4828 pr_warning("Forcing miimon to 100msec\n");
4829 miimon = 100;
4830 }
4831 }
4832
4833 /* reset values for TLB/ALB */
4834 if ((bond_mode == BOND_MODE_TLB) ||
4835 (bond_mode == BOND_MODE_ALB)) {
4836 if (!miimon) {
4837 pr_warning(DRV_NAME
4838 ": Warning: miimon must be specified, "
4839 "otherwise bonding will not detect link "
4840 "failure and link speed which are essential "
4841 "for TLB/ALB load balancing\n");
4842 pr_warning("Forcing miimon to 100msec\n");
4843 miimon = 100;
4844 }
4845 }
4846
4847 if (bond_mode == BOND_MODE_ALB) {
4848 printk(KERN_NOTICE DRV_NAME
4849 ": In ALB mode you might experience client "
4850 "disconnections upon reconnection of a link if the "
4851 "bonding module updelay parameter (%d msec) is "
4852 "incompatible with the forwarding delay time of the "
4853 "switch\n",
4854 updelay);
4855 }
4856
4857 if (!miimon) {
4858 if (updelay || downdelay) {
4859 /* just warn the user the up/down delay will have
4860 * no effect since miimon is zero...
4861 */
4862 pr_warning(DRV_NAME
4863 ": Warning: miimon module parameter not set "
4864 "and updelay (%d) or downdelay (%d) module "
4865 "parameter is set; updelay and downdelay have "
4866 "no effect unless miimon is set\n",
4867 updelay, downdelay);
4868 }
4869 } else {
4870 /* don't allow arp monitoring */
4871 if (arp_interval) {
4872 pr_warning(DRV_NAME
4873 ": Warning: miimon (%d) and arp_interval (%d) "
4874 "can't be used simultaneously, disabling ARP "
4875 "monitoring\n",
4876 miimon, arp_interval);
4877 arp_interval = 0;
4878 }
4879
4880 if ((updelay % miimon) != 0) {
4881 pr_warning(DRV_NAME
4882 ": Warning: updelay (%d) is not a multiple "
4883 "of miimon (%d), updelay rounded to %d ms\n",
4884 updelay, miimon, (updelay / miimon) * miimon);
4885 }
4886
4887 updelay /= miimon;
4888
4889 if ((downdelay % miimon) != 0) {
4890 pr_warning(DRV_NAME
4891 ": Warning: downdelay (%d) is not a multiple "
4892 "of miimon (%d), downdelay rounded to %d ms\n",
4893 downdelay, miimon,
4894 (downdelay / miimon) * miimon);
4895 }
4896
4897 downdelay /= miimon;
4898 }
4899
4900 if (arp_interval < 0) {
4901 pr_warning(DRV_NAME
4902 ": Warning: arp_interval module parameter (%d) "
4903 ", not in range 0-%d, so it was reset to %d\n",
4904 arp_interval, INT_MAX, BOND_LINK_ARP_INTERV);
4905 arp_interval = BOND_LINK_ARP_INTERV;
4906 }
4907
4908 for (arp_ip_count = 0;
4909 (arp_ip_count < BOND_MAX_ARP_TARGETS) && arp_ip_target[arp_ip_count];
4910 arp_ip_count++) {
4911 /* not complete check, but should be good enough to
4912 catch mistakes */
4913 if (!isdigit(arp_ip_target[arp_ip_count][0])) {
4914 pr_warning(DRV_NAME
4915 ": Warning: bad arp_ip_target module parameter "
4916 "(%s), ARP monitoring will not be performed\n",
4917 arp_ip_target[arp_ip_count]);
4918 arp_interval = 0;
4919 } else {
4920 __be32 ip = in_aton(arp_ip_target[arp_ip_count]);
4921 arp_target[arp_ip_count] = ip;
4922 }
4923 }
4924
4925 if (arp_interval && !arp_ip_count) {
4926 /* don't allow arping if no arp_ip_target given... */
4927 pr_warning(DRV_NAME
4928 ": Warning: arp_interval module parameter (%d) "
4929 "specified without providing an arp_ip_target "
4930 "parameter, arp_interval was reset to 0\n",
4931 arp_interval);
4932 arp_interval = 0;
4933 }
4934
4935 if (arp_validate) {
4936 if (bond_mode != BOND_MODE_ACTIVEBACKUP) {
4937 pr_err(DRV_NAME
4938 ": arp_validate only supported in active-backup mode\n");
4939 return -EINVAL;
4940 }
4941 if (!arp_interval) {
4942 pr_err(DRV_NAME
4943 ": arp_validate requires arp_interval\n");
4944 return -EINVAL;
4945 }
4946
4947 arp_validate_value = bond_parse_parm(arp_validate,
4948 arp_validate_tbl);
4949 if (arp_validate_value == -1) {
4950 pr_err(DRV_NAME
4951 ": Error: invalid arp_validate \"%s\"\n",
4952 arp_validate == NULL ? "NULL" : arp_validate);
4953 return -EINVAL;
4954 }
4955 } else
4956 arp_validate_value = 0;
4957
4958 if (miimon) {
4959 pr_info(DRV_NAME
4960 ": MII link monitoring set to %d ms\n",
4961 miimon);
4962 } else if (arp_interval) {
4963 int i;
4964
4965 pr_info(DRV_NAME ": ARP monitoring set to %d ms,"
4966 " validate %s, with %d target(s):",
4967 arp_interval,
4968 arp_validate_tbl[arp_validate_value].modename,
4969 arp_ip_count);
4970
4971 for (i = 0; i < arp_ip_count; i++)
4972 printk(" %s", arp_ip_target[i]);
4973
4974 printk("\n");
4975
4976 } else if (max_bonds) {
4977 /* miimon and arp_interval not set, we need one so things
4978 * work as expected, see bonding.txt for details
4979 */
4980 pr_warning(DRV_NAME
4981 ": Warning: either miimon or arp_interval and "
4982 "arp_ip_target module parameters must be specified, "
4983 "otherwise bonding will not detect link failures! see "
4984 "bonding.txt for details.\n");
4985 }
4986
4987 if (primary && !USES_PRIMARY(bond_mode)) {
4988 /* currently, using a primary only makes sense
4989 * in active backup, TLB or ALB modes
4990 */
4991 pr_warning(DRV_NAME
4992 ": Warning: %s primary device specified but has no "
4993 "effect in %s mode\n",
4994 primary, bond_mode_name(bond_mode));
4995 primary = NULL;
4996 }
4997
4998 if (fail_over_mac) {
4999 fail_over_mac_value = bond_parse_parm(fail_over_mac,
5000 fail_over_mac_tbl);
5001 if (fail_over_mac_value == -1) {
5002 pr_err(DRV_NAME
5003 ": Error: invalid fail_over_mac \"%s\"\n",
5004 arp_validate == NULL ? "NULL" : arp_validate);
5005 return -EINVAL;
5006 }
5007
5008 if (bond_mode != BOND_MODE_ACTIVEBACKUP)
5009 pr_warning(DRV_NAME
5010 ": Warning: fail_over_mac only affects "
5011 "active-backup mode.\n");
5012 } else {
5013 fail_over_mac_value = BOND_FOM_NONE;
5014 }
5015
5016 /* fill params struct with the proper values */
5017 params->mode = bond_mode;
5018 params->xmit_policy = xmit_hashtype;
5019 params->miimon = miimon;
5020 params->num_grat_arp = num_grat_arp;
5021 params->num_unsol_na = num_unsol_na;
5022 params->arp_interval = arp_interval;
5023 params->arp_validate = arp_validate_value;
5024 params->updelay = updelay;
5025 params->downdelay = downdelay;
5026 params->use_carrier = use_carrier;
5027 params->lacp_fast = lacp_fast;
5028 params->primary[0] = 0;
5029 params->fail_over_mac = fail_over_mac_value;
5030
5031 if (primary) {
5032 strncpy(params->primary, primary, IFNAMSIZ);
5033 params->primary[IFNAMSIZ - 1] = 0;
5034 }
5035
5036 memcpy(params->arp_targets, arp_target, sizeof(arp_target));
5037
5038 return 0;
5039 }
5040
5041 static struct lock_class_key bonding_netdev_xmit_lock_key;
5042 static struct lock_class_key bonding_netdev_addr_lock_key;
5043
5044 static void bond_set_lockdep_class_one(struct net_device *dev,
5045 struct netdev_queue *txq,
5046 void *_unused)
5047 {
5048 lockdep_set_class(&txq->_xmit_lock,
5049 &bonding_netdev_xmit_lock_key);
5050 }
5051
5052 static void bond_set_lockdep_class(struct net_device *dev)
5053 {
5054 lockdep_set_class(&dev->addr_list_lock,
5055 &bonding_netdev_addr_lock_key);
5056 netdev_for_each_tx_queue(dev, bond_set_lockdep_class_one, NULL);
5057 }
5058
5059 /* Create a new bond based on the specified name and bonding parameters.
5060 * If name is NULL, obtain a suitable "bond%d" name for us.
5061 * Caller must NOT hold rtnl_lock; we need to release it here before we
5062 * set up our sysfs entries.
5063 */
5064 int bond_create(const char *name)
5065 {
5066 struct net_device *bond_dev;
5067 struct bonding *bond;
5068 int res;
5069
5070 rtnl_lock();
5071 /* Check to see if the bond already exists. */
5072 if (name) {
5073 list_for_each_entry(bond, &bond_dev_list, bond_list)
5074 if (strnicmp(bond->dev->name, name, IFNAMSIZ) == 0) {
5075 pr_err(DRV_NAME ": cannot add bond %s;"
5076 " it already exists\n", name);
5077 res = -EPERM;
5078 goto out_rtnl;
5079 }
5080 }
5081
5082 bond_dev = alloc_netdev(sizeof(struct bonding), name ? name : "",
5083 ether_setup);
5084 if (!bond_dev) {
5085 pr_err(DRV_NAME ": %s: eek! can't alloc netdev!\n",
5086 name);
5087 res = -ENOMEM;
5088 goto out_rtnl;
5089 }
5090
5091 bond_dev->priv_flags &= ~IFF_XMIT_DST_RELEASE;
5092 if (!name) {
5093 res = dev_alloc_name(bond_dev, "bond%d");
5094 if (res < 0)
5095 goto out_netdev;
5096 }
5097
5098 /* bond_init() must be called after dev_alloc_name() (for the
5099 * /proc files), but before register_netdevice(), because we
5100 * need to set function pointers.
5101 */
5102
5103 res = bond_init(bond_dev);
5104 if (res < 0)
5105 goto out_netdev;
5106
5107 bond_set_lockdep_class(bond_dev);
5108
5109 netif_carrier_off(bond_dev);
5110
5111 res = register_netdevice(bond_dev);
5112 if (res < 0)
5113 goto out_bond;
5114
5115 res = bond_create_sysfs_entry(netdev_priv(bond_dev));
5116 if (res < 0)
5117 goto out_unreg;
5118
5119 rtnl_unlock();
5120 return 0;
5121
5122 out_unreg:
5123 unregister_netdevice(bond_dev);
5124 out_bond:
5125 bond_deinit(bond_dev);
5126 out_netdev:
5127 free_netdev(bond_dev);
5128 out_rtnl:
5129 rtnl_unlock();
5130 return res;
5131 }
5132
5133 static int __init bonding_init(void)
5134 {
5135 int i;
5136 int res;
5137
5138 pr_info("%s", version);
5139
5140 res = bond_check_params(&bonding_defaults);
5141 if (res)
5142 goto out;
5143
5144 bond_create_proc_dir();
5145
5146 for (i = 0; i < max_bonds; i++) {
5147 res = bond_create(NULL);
5148 if (res)
5149 goto err;
5150 }
5151
5152 res = bond_create_sysfs();
5153 if (res)
5154 goto err;
5155
5156 register_netdevice_notifier(&bond_netdev_notifier);
5157 register_inetaddr_notifier(&bond_inetaddr_notifier);
5158 bond_register_ipv6_notifier();
5159
5160 goto out;
5161 err:
5162 rtnl_lock();
5163 bond_free_all();
5164 rtnl_unlock();
5165 out:
5166 return res;
5167
5168 }
5169
5170 static void __exit bonding_exit(void)
5171 {
5172 unregister_netdevice_notifier(&bond_netdev_notifier);
5173 unregister_inetaddr_notifier(&bond_inetaddr_notifier);
5174 bond_unregister_ipv6_notifier();
5175
5176 bond_destroy_sysfs();
5177
5178 rtnl_lock();
5179 bond_free_all();
5180 rtnl_unlock();
5181 }
5182
5183 module_init(bonding_init);
5184 module_exit(bonding_exit);
5185 MODULE_LICENSE("GPL");
5186 MODULE_VERSION(DRV_VERSION);
5187 MODULE_DESCRIPTION(DRV_DESCRIPTION ", v" DRV_VERSION);
5188 MODULE_AUTHOR("Thomas Davis, tadavis@lbl.gov and many others");
Linux kernel - Deliverables
This page took 0.153219 seconds and 4 git commands to generate.