bonding: print information about speed and duplex seen by the driver
[deliverable/linux.git] / drivers / net / bonding / bond_main.c
1 /*
2 * originally based on the dummy device.
3 *
4 * Copyright 1999, Thomas Davis, tadavis@lbl.gov.
5 * Licensed under the GPL. Based on dummy.c, and eql.c devices.
6 *
7 * bonding.c: an Ethernet Bonding driver
8 *
9 * This is useful to talk to a Cisco EtherChannel compatible equipment:
10 * Cisco 5500
11 * Sun Trunking (Solaris)
12 * Alteon AceDirector Trunks
13 * Linux Bonding
14 * and probably many L2 switches ...
15 *
16 * How it works:
17 * ifconfig bond0 ipaddress netmask up
18 * will setup a network device, with an ip address. No mac address
19 * will be assigned at this time. The hw mac address will come from
20 * the first slave bonded to the channel. All slaves will then use
21 * this hw mac address.
22 *
23 * ifconfig bond0 down
24 * will release all slaves, marking them as down.
25 *
26 * ifenslave bond0 eth0
27 * will attach eth0 to bond0 as a slave. eth0 hw mac address will either
28 * a: be used as initial mac address
29 * b: if a hw mac address already is there, eth0's hw mac address
30 * will then be set from bond0.
31 *
32 */
33
34 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
35
36 #include <linux/kernel.h>
37 #include <linux/module.h>
38 #include <linux/types.h>
39 #include <linux/fcntl.h>
40 #include <linux/interrupt.h>
41 #include <linux/ptrace.h>
42 #include <linux/ioport.h>
43 #include <linux/in.h>
44 #include <net/ip.h>
45 #include <linux/ip.h>
46 #include <linux/tcp.h>
47 #include <linux/udp.h>
48 #include <linux/slab.h>
49 #include <linux/string.h>
50 #include <linux/init.h>
51 #include <linux/timer.h>
52 #include <linux/socket.h>
53 #include <linux/ctype.h>
54 #include <linux/inet.h>
55 #include <linux/bitops.h>
56 #include <linux/io.h>
57 #include <asm/system.h>
58 #include <asm/dma.h>
59 #include <linux/uaccess.h>
60 #include <linux/errno.h>
61 #include <linux/netdevice.h>
62 #include <linux/netpoll.h>
63 #include <linux/inetdevice.h>
64 #include <linux/igmp.h>
65 #include <linux/etherdevice.h>
66 #include <linux/skbuff.h>
67 #include <net/sock.h>
68 #include <linux/rtnetlink.h>
69 #include <linux/proc_fs.h>
70 #include <linux/seq_file.h>
71 #include <linux/smp.h>
72 #include <linux/if_ether.h>
73 #include <net/arp.h>
74 #include <linux/mii.h>
75 #include <linux/ethtool.h>
76 #include <linux/if_vlan.h>
77 #include <linux/if_bonding.h>
78 #include <linux/jiffies.h>
79 #include <net/route.h>
80 #include <net/net_namespace.h>
81 #include <net/netns/generic.h>
82 #include "bonding.h"
83 #include "bond_3ad.h"
84 #include "bond_alb.h"
85
86 /*---------------------------- Module parameters ----------------------------*/
87
88 /* monitor all links that often (in milliseconds). <=0 disables monitoring */
89 #define BOND_LINK_MON_INTERV 0
90 #define BOND_LINK_ARP_INTERV 0
91
92 static int max_bonds = BOND_DEFAULT_MAX_BONDS;
93 static int tx_queues = BOND_DEFAULT_TX_QUEUES;
94 static int num_grat_arp = 1;
95 static int num_unsol_na = 1;
96 static int miimon = BOND_LINK_MON_INTERV;
97 static int updelay;
98 static int downdelay;
99 static int use_carrier = 1;
100 static char *mode;
101 static char *primary;
102 static char *primary_reselect;
103 static char *lacp_rate;
104 static char *ad_select;
105 static char *xmit_hash_policy;
106 static int arp_interval = BOND_LINK_ARP_INTERV;
107 static char *arp_ip_target[BOND_MAX_ARP_TARGETS];
108 static char *arp_validate;
109 static char *fail_over_mac;
110 static int all_slaves_active = 0;
111 static struct bond_params bonding_defaults;
112 static int resend_igmp = BOND_DEFAULT_RESEND_IGMP;
113
114 module_param(max_bonds, int, 0);
115 MODULE_PARM_DESC(max_bonds, "Max number of bonded devices");
116 module_param(tx_queues, int, 0);
117 MODULE_PARM_DESC(tx_queues, "Max number of transmit queues (default = 16)");
118 module_param(num_grat_arp, int, 0644);
119 MODULE_PARM_DESC(num_grat_arp, "Number of gratuitous ARP packets to send on failover event");
120 module_param(num_unsol_na, int, 0644);
121 MODULE_PARM_DESC(num_unsol_na, "Number of unsolicited IPv6 Neighbor Advertisements packets to send on failover event");
122 module_param(miimon, int, 0);
123 MODULE_PARM_DESC(miimon, "Link check interval in milliseconds");
124 module_param(updelay, int, 0);
125 MODULE_PARM_DESC(updelay, "Delay before considering link up, in milliseconds");
126 module_param(downdelay, int, 0);
127 MODULE_PARM_DESC(downdelay, "Delay before considering link down, "
128 "in milliseconds");
129 module_param(use_carrier, int, 0);
130 MODULE_PARM_DESC(use_carrier, "Use netif_carrier_ok (vs MII ioctls) in miimon; "
131 "0 for off, 1 for on (default)");
132 module_param(mode, charp, 0);
133 MODULE_PARM_DESC(mode, "Mode of operation : 0 for balance-rr, "
134 "1 for active-backup, 2 for balance-xor, "
135 "3 for broadcast, 4 for 802.3ad, 5 for balance-tlb, "
136 "6 for balance-alb");
137 module_param(primary, charp, 0);
138 MODULE_PARM_DESC(primary, "Primary network device to use");
139 module_param(primary_reselect, charp, 0);
140 MODULE_PARM_DESC(primary_reselect, "Reselect primary slave "
141 "once it comes up; "
142 "0 for always (default), "
143 "1 for only if speed of primary is "
144 "better, "
145 "2 for only on active slave "
146 "failure");
147 module_param(lacp_rate, charp, 0);
148 MODULE_PARM_DESC(lacp_rate, "LACPDU tx rate to request from 802.3ad partner "
149 "(slow/fast)");
150 module_param(ad_select, charp, 0);
151 MODULE_PARM_DESC(ad_select, "803.ad aggregation selection logic: stable (0, default), bandwidth (1), count (2)");
152 module_param(xmit_hash_policy, charp, 0);
153 MODULE_PARM_DESC(xmit_hash_policy, "XOR hashing method: 0 for layer 2 (default)"
154 ", 1 for layer 3+4");
155 module_param(arp_interval, int, 0);
156 MODULE_PARM_DESC(arp_interval, "arp interval in milliseconds");
157 module_param_array(arp_ip_target, charp, NULL, 0);
158 MODULE_PARM_DESC(arp_ip_target, "arp targets in n.n.n.n form");
159 module_param(arp_validate, charp, 0);
160 MODULE_PARM_DESC(arp_validate, "validate src/dst of ARP probes: none (default), active, backup or all");
161 module_param(fail_over_mac, charp, 0);
162 MODULE_PARM_DESC(fail_over_mac, "For active-backup, do not set all slaves to the same MAC. none (default), active or follow");
163 module_param(all_slaves_active, int, 0);
164 MODULE_PARM_DESC(all_slaves_active, "Keep all frames received on an interface"
165 "by setting active flag for all slaves. "
166 "0 for never (default), 1 for always.");
167 module_param(resend_igmp, int, 0);
168 MODULE_PARM_DESC(resend_igmp, "Number of IGMP membership reports to send on link failure");
169
170 /*----------------------------- Global variables ----------------------------*/
171
172 static const char * const version =
173 DRV_DESCRIPTION ": v" DRV_VERSION " (" DRV_RELDATE ")\n";
174
175 int bond_net_id __read_mostly;
176
177 static __be32 arp_target[BOND_MAX_ARP_TARGETS];
178 static int arp_ip_count;
179 static int bond_mode = BOND_MODE_ROUNDROBIN;
180 static int xmit_hashtype = BOND_XMIT_POLICY_LAYER2;
181 static int lacp_fast;
182 #ifdef CONFIG_NET_POLL_CONTROLLER
183 static int disable_netpoll = 1;
184 #endif
185
186 const struct bond_parm_tbl bond_lacp_tbl[] = {
187 { "slow", AD_LACP_SLOW},
188 { "fast", AD_LACP_FAST},
189 { NULL, -1},
190 };
191
192 const struct bond_parm_tbl bond_mode_tbl[] = {
193 { "balance-rr", BOND_MODE_ROUNDROBIN},
194 { "active-backup", BOND_MODE_ACTIVEBACKUP},
195 { "balance-xor", BOND_MODE_XOR},
196 { "broadcast", BOND_MODE_BROADCAST},
197 { "802.3ad", BOND_MODE_8023AD},
198 { "balance-tlb", BOND_MODE_TLB},
199 { "balance-alb", BOND_MODE_ALB},
200 { NULL, -1},
201 };
202
203 const struct bond_parm_tbl xmit_hashtype_tbl[] = {
204 { "layer2", BOND_XMIT_POLICY_LAYER2},
205 { "layer3+4", BOND_XMIT_POLICY_LAYER34},
206 { "layer2+3", BOND_XMIT_POLICY_LAYER23},
207 { NULL, -1},
208 };
209
210 const struct bond_parm_tbl arp_validate_tbl[] = {
211 { "none", BOND_ARP_VALIDATE_NONE},
212 { "active", BOND_ARP_VALIDATE_ACTIVE},
213 { "backup", BOND_ARP_VALIDATE_BACKUP},
214 { "all", BOND_ARP_VALIDATE_ALL},
215 { NULL, -1},
216 };
217
218 const struct bond_parm_tbl fail_over_mac_tbl[] = {
219 { "none", BOND_FOM_NONE},
220 { "active", BOND_FOM_ACTIVE},
221 { "follow", BOND_FOM_FOLLOW},
222 { NULL, -1},
223 };
224
225 const struct bond_parm_tbl pri_reselect_tbl[] = {
226 { "always", BOND_PRI_RESELECT_ALWAYS},
227 { "better", BOND_PRI_RESELECT_BETTER},
228 { "failure", BOND_PRI_RESELECT_FAILURE},
229 { NULL, -1},
230 };
231
232 struct bond_parm_tbl ad_select_tbl[] = {
233 { "stable", BOND_AD_STABLE},
234 { "bandwidth", BOND_AD_BANDWIDTH},
235 { "count", BOND_AD_COUNT},
236 { NULL, -1},
237 };
238
239 /*-------------------------- Forward declarations ---------------------------*/
240
241 static void bond_send_gratuitous_arp(struct bonding *bond);
242 static int bond_init(struct net_device *bond_dev);
243 static void bond_uninit(struct net_device *bond_dev);
244
245 /*---------------------------- General routines -----------------------------*/
246
247 static const char *bond_mode_name(int mode)
248 {
249 static const char *names[] = {
250 [BOND_MODE_ROUNDROBIN] = "load balancing (round-robin)",
251 [BOND_MODE_ACTIVEBACKUP] = "fault-tolerance (active-backup)",
252 [BOND_MODE_XOR] = "load balancing (xor)",
253 [BOND_MODE_BROADCAST] = "fault-tolerance (broadcast)",
254 [BOND_MODE_8023AD] = "IEEE 802.3ad Dynamic link aggregation",
255 [BOND_MODE_TLB] = "transmit load balancing",
256 [BOND_MODE_ALB] = "adaptive load balancing",
257 };
258
259 if (mode < 0 || mode > BOND_MODE_ALB)
260 return "unknown";
261
262 return names[mode];
263 }
264
265 /*---------------------------------- VLAN -----------------------------------*/
266
267 /**
268 * bond_add_vlan - add a new vlan id on bond
269 * @bond: bond that got the notification
270 * @vlan_id: the vlan id to add
271 *
272 * Returns -ENOMEM if allocation failed.
273 */
274 static int bond_add_vlan(struct bonding *bond, unsigned short vlan_id)
275 {
276 struct vlan_entry *vlan;
277
278 pr_debug("bond: %s, vlan id %d\n",
279 (bond ? bond->dev->name : "None"), vlan_id);
280
281 vlan = kzalloc(sizeof(struct vlan_entry), GFP_KERNEL);
282 if (!vlan)
283 return -ENOMEM;
284
285 INIT_LIST_HEAD(&vlan->vlan_list);
286 vlan->vlan_id = vlan_id;
287
288 write_lock_bh(&bond->lock);
289
290 list_add_tail(&vlan->vlan_list, &bond->vlan_list);
291
292 write_unlock_bh(&bond->lock);
293
294 pr_debug("added VLAN ID %d on bond %s\n", vlan_id, bond->dev->name);
295
296 return 0;
297 }
298
299 /**
300 * bond_del_vlan - delete a vlan id from bond
301 * @bond: bond that got the notification
302 * @vlan_id: the vlan id to delete
303 *
304 * returns -ENODEV if @vlan_id was not found in @bond.
305 */
306 static int bond_del_vlan(struct bonding *bond, unsigned short vlan_id)
307 {
308 struct vlan_entry *vlan;
309 int res = -ENODEV;
310
311 pr_debug("bond: %s, vlan id %d\n", bond->dev->name, vlan_id);
312
313 write_lock_bh(&bond->lock);
314
315 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
316 if (vlan->vlan_id == vlan_id) {
317 list_del(&vlan->vlan_list);
318
319 if (bond_is_lb(bond))
320 bond_alb_clear_vlan(bond, vlan_id);
321
322 pr_debug("removed VLAN ID %d from bond %s\n",
323 vlan_id, bond->dev->name);
324
325 kfree(vlan);
326
327 if (list_empty(&bond->vlan_list) &&
328 (bond->slave_cnt == 0)) {
329 /* Last VLAN removed and no slaves, so
330 * restore block on adding VLANs. This will
331 * be removed once new slaves that are not
332 * VLAN challenged will be added.
333 */
334 bond->dev->features |= NETIF_F_VLAN_CHALLENGED;
335 }
336
337 res = 0;
338 goto out;
339 }
340 }
341
342 pr_debug("couldn't find VLAN ID %d in bond %s\n",
343 vlan_id, bond->dev->name);
344
345 out:
346 write_unlock_bh(&bond->lock);
347 return res;
348 }
349
350 /**
351 * bond_has_challenged_slaves
352 * @bond: the bond we're working on
353 *
354 * Searches the slave list. Returns 1 if a vlan challenged slave
355 * was found, 0 otherwise.
356 *
357 * Assumes bond->lock is held.
358 */
359 static int bond_has_challenged_slaves(struct bonding *bond)
360 {
361 struct slave *slave;
362 int i;
363
364 bond_for_each_slave(bond, slave, i) {
365 if (slave->dev->features & NETIF_F_VLAN_CHALLENGED) {
366 pr_debug("found VLAN challenged slave - %s\n",
367 slave->dev->name);
368 return 1;
369 }
370 }
371
372 pr_debug("no VLAN challenged slaves found\n");
373 return 0;
374 }
375
376 /**
377 * bond_next_vlan - safely skip to the next item in the vlans list.
378 * @bond: the bond we're working on
379 * @curr: item we're advancing from
380 *
381 * Returns %NULL if list is empty, bond->next_vlan if @curr is %NULL,
382 * or @curr->next otherwise (even if it is @curr itself again).
383 *
384 * Caller must hold bond->lock
385 */
386 struct vlan_entry *bond_next_vlan(struct bonding *bond, struct vlan_entry *curr)
387 {
388 struct vlan_entry *next, *last;
389
390 if (list_empty(&bond->vlan_list))
391 return NULL;
392
393 if (!curr) {
394 next = list_entry(bond->vlan_list.next,
395 struct vlan_entry, vlan_list);
396 } else {
397 last = list_entry(bond->vlan_list.prev,
398 struct vlan_entry, vlan_list);
399 if (last == curr) {
400 next = list_entry(bond->vlan_list.next,
401 struct vlan_entry, vlan_list);
402 } else {
403 next = list_entry(curr->vlan_list.next,
404 struct vlan_entry, vlan_list);
405 }
406 }
407
408 return next;
409 }
410
411 /**
412 * bond_dev_queue_xmit - Prepare skb for xmit.
413 *
414 * @bond: bond device that got this skb for tx.
415 * @skb: hw accel VLAN tagged skb to transmit
416 * @slave_dev: slave that is supposed to xmit this skbuff
417 *
418 * When the bond gets an skb to transmit that is
419 * already hardware accelerated VLAN tagged, and it
420 * needs to relay this skb to a slave that is not
421 * hw accel capable, the skb needs to be "unaccelerated",
422 * i.e. strip the hwaccel tag and re-insert it as part
423 * of the payload.
424 */
425 int bond_dev_queue_xmit(struct bonding *bond, struct sk_buff *skb,
426 struct net_device *slave_dev)
427 {
428 unsigned short uninitialized_var(vlan_id);
429
430 /* Test vlan_list not vlgrp to catch and handle 802.1p tags */
431 if (!list_empty(&bond->vlan_list) &&
432 !(slave_dev->features & NETIF_F_HW_VLAN_TX) &&
433 vlan_get_tag(skb, &vlan_id) == 0) {
434 skb->dev = slave_dev;
435 skb = vlan_put_tag(skb, vlan_id);
436 if (!skb) {
437 /* vlan_put_tag() frees the skb in case of error,
438 * so return success here so the calling functions
439 * won't attempt to free is again.
440 */
441 return 0;
442 }
443 } else {
444 skb->dev = slave_dev;
445 }
446
447 skb->priority = 1;
448 #ifdef CONFIG_NET_POLL_CONTROLLER
449 if (unlikely(bond->dev->priv_flags & IFF_IN_NETPOLL)) {
450 struct netpoll *np = bond->dev->npinfo->netpoll;
451 slave_dev->npinfo = bond->dev->npinfo;
452 np->real_dev = np->dev = skb->dev;
453 slave_dev->priv_flags |= IFF_IN_NETPOLL;
454 netpoll_send_skb(np, skb);
455 slave_dev->priv_flags &= ~IFF_IN_NETPOLL;
456 np->dev = bond->dev;
457 } else
458 #endif
459 dev_queue_xmit(skb);
460
461 return 0;
462 }
463
464 /*
465 * In the following 3 functions, bond_vlan_rx_register(), bond_vlan_rx_add_vid
466 * and bond_vlan_rx_kill_vid, We don't protect the slave list iteration with a
467 * lock because:
468 * a. This operation is performed in IOCTL context,
469 * b. The operation is protected by the RTNL semaphore in the 8021q code,
470 * c. Holding a lock with BH disabled while directly calling a base driver
471 * entry point is generally a BAD idea.
472 *
473 * The design of synchronization/protection for this operation in the 8021q
474 * module is good for one or more VLAN devices over a single physical device
475 * and cannot be extended for a teaming solution like bonding, so there is a
476 * potential race condition here where a net device from the vlan group might
477 * be referenced (either by a base driver or the 8021q code) while it is being
478 * removed from the system. However, it turns out we're not making matters
479 * worse, and if it works for regular VLAN usage it will work here too.
480 */
481
482 /**
483 * bond_vlan_rx_register - Propagates registration to slaves
484 * @bond_dev: bonding net device that got called
485 * @grp: vlan group being registered
486 */
487 static void bond_vlan_rx_register(struct net_device *bond_dev,
488 struct vlan_group *grp)
489 {
490 struct bonding *bond = netdev_priv(bond_dev);
491 struct slave *slave;
492 int i;
493
494 write_lock(&bond->lock);
495 bond->vlgrp = grp;
496 write_unlock(&bond->lock);
497
498 bond_for_each_slave(bond, slave, i) {
499 struct net_device *slave_dev = slave->dev;
500 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
501
502 if ((slave_dev->features & NETIF_F_HW_VLAN_RX) &&
503 slave_ops->ndo_vlan_rx_register) {
504 slave_ops->ndo_vlan_rx_register(slave_dev, grp);
505 }
506 }
507 }
508
509 /**
510 * bond_vlan_rx_add_vid - Propagates adding an id to slaves
511 * @bond_dev: bonding net device that got called
512 * @vid: vlan id being added
513 */
514 static void bond_vlan_rx_add_vid(struct net_device *bond_dev, uint16_t vid)
515 {
516 struct bonding *bond = netdev_priv(bond_dev);
517 struct slave *slave;
518 int i, res;
519
520 bond_for_each_slave(bond, slave, i) {
521 struct net_device *slave_dev = slave->dev;
522 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
523
524 if ((slave_dev->features & NETIF_F_HW_VLAN_FILTER) &&
525 slave_ops->ndo_vlan_rx_add_vid) {
526 slave_ops->ndo_vlan_rx_add_vid(slave_dev, vid);
527 }
528 }
529
530 res = bond_add_vlan(bond, vid);
531 if (res) {
532 pr_err("%s: Error: Failed to add vlan id %d\n",
533 bond_dev->name, vid);
534 }
535 }
536
537 /**
538 * bond_vlan_rx_kill_vid - Propagates deleting an id to slaves
539 * @bond_dev: bonding net device that got called
540 * @vid: vlan id being removed
541 */
542 static void bond_vlan_rx_kill_vid(struct net_device *bond_dev, uint16_t vid)
543 {
544 struct bonding *bond = netdev_priv(bond_dev);
545 struct slave *slave;
546 struct net_device *vlan_dev;
547 int i, res;
548
549 bond_for_each_slave(bond, slave, i) {
550 struct net_device *slave_dev = slave->dev;
551 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
552
553 if ((slave_dev->features & NETIF_F_HW_VLAN_FILTER) &&
554 slave_ops->ndo_vlan_rx_kill_vid) {
555 /* Save and then restore vlan_dev in the grp array,
556 * since the slave's driver might clear it.
557 */
558 vlan_dev = vlan_group_get_device(bond->vlgrp, vid);
559 slave_ops->ndo_vlan_rx_kill_vid(slave_dev, vid);
560 vlan_group_set_device(bond->vlgrp, vid, vlan_dev);
561 }
562 }
563
564 res = bond_del_vlan(bond, vid);
565 if (res) {
566 pr_err("%s: Error: Failed to remove vlan id %d\n",
567 bond_dev->name, vid);
568 }
569 }
570
571 static void bond_add_vlans_on_slave(struct bonding *bond, struct net_device *slave_dev)
572 {
573 struct vlan_entry *vlan;
574 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
575
576 if (!bond->vlgrp)
577 return;
578
579 if ((slave_dev->features & NETIF_F_HW_VLAN_RX) &&
580 slave_ops->ndo_vlan_rx_register)
581 slave_ops->ndo_vlan_rx_register(slave_dev, bond->vlgrp);
582
583 if (!(slave_dev->features & NETIF_F_HW_VLAN_FILTER) ||
584 !(slave_ops->ndo_vlan_rx_add_vid))
585 return;
586
587 list_for_each_entry(vlan, &bond->vlan_list, vlan_list)
588 slave_ops->ndo_vlan_rx_add_vid(slave_dev, vlan->vlan_id);
589 }
590
591 static void bond_del_vlans_from_slave(struct bonding *bond,
592 struct net_device *slave_dev)
593 {
594 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
595 struct vlan_entry *vlan;
596 struct net_device *vlan_dev;
597
598 if (!bond->vlgrp)
599 return;
600
601 if (!(slave_dev->features & NETIF_F_HW_VLAN_FILTER) ||
602 !(slave_ops->ndo_vlan_rx_kill_vid))
603 goto unreg;
604
605 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
606 if (!vlan->vlan_id)
607 continue;
608 /* Save and then restore vlan_dev in the grp array,
609 * since the slave's driver might clear it.
610 */
611 vlan_dev = vlan_group_get_device(bond->vlgrp, vlan->vlan_id);
612 slave_ops->ndo_vlan_rx_kill_vid(slave_dev, vlan->vlan_id);
613 vlan_group_set_device(bond->vlgrp, vlan->vlan_id, vlan_dev);
614 }
615
616 unreg:
617 if ((slave_dev->features & NETIF_F_HW_VLAN_RX) &&
618 slave_ops->ndo_vlan_rx_register)
619 slave_ops->ndo_vlan_rx_register(slave_dev, NULL);
620 }
621
622 /*------------------------------- Link status -------------------------------*/
623
624 /*
625 * Set the carrier state for the master according to the state of its
626 * slaves. If any slaves are up, the master is up. In 802.3ad mode,
627 * do special 802.3ad magic.
628 *
629 * Returns zero if carrier state does not change, nonzero if it does.
630 */
631 static int bond_set_carrier(struct bonding *bond)
632 {
633 struct slave *slave;
634 int i;
635
636 if (bond->slave_cnt == 0)
637 goto down;
638
639 if (bond->params.mode == BOND_MODE_8023AD)
640 return bond_3ad_set_carrier(bond);
641
642 bond_for_each_slave(bond, slave, i) {
643 if (slave->link == BOND_LINK_UP) {
644 if (!netif_carrier_ok(bond->dev)) {
645 netif_carrier_on(bond->dev);
646 return 1;
647 }
648 return 0;
649 }
650 }
651
652 down:
653 if (netif_carrier_ok(bond->dev)) {
654 netif_carrier_off(bond->dev);
655 return 1;
656 }
657 return 0;
658 }
659
660 /*
661 * Get link speed and duplex from the slave's base driver
662 * using ethtool. If for some reason the call fails or the
663 * values are invalid, fake speed and duplex to 100/Full
664 * and return error.
665 */
666 static int bond_update_speed_duplex(struct slave *slave)
667 {
668 struct net_device *slave_dev = slave->dev;
669 struct ethtool_cmd etool;
670 int res;
671
672 /* Fake speed and duplex */
673 slave->speed = SPEED_100;
674 slave->duplex = DUPLEX_FULL;
675
676 if (!slave_dev->ethtool_ops || !slave_dev->ethtool_ops->get_settings)
677 return -1;
678
679 res = slave_dev->ethtool_ops->get_settings(slave_dev, &etool);
680 if (res < 0)
681 return -1;
682
683 switch (etool.speed) {
684 case SPEED_10:
685 case SPEED_100:
686 case SPEED_1000:
687 case SPEED_10000:
688 break;
689 default:
690 return -1;
691 }
692
693 switch (etool.duplex) {
694 case DUPLEX_FULL:
695 case DUPLEX_HALF:
696 break;
697 default:
698 return -1;
699 }
700
701 slave->speed = etool.speed;
702 slave->duplex = etool.duplex;
703
704 return 0;
705 }
706
707 /*
708 * if <dev> supports MII link status reporting, check its link status.
709 *
710 * We either do MII/ETHTOOL ioctls, or check netif_carrier_ok(),
711 * depending upon the setting of the use_carrier parameter.
712 *
713 * Return either BMSR_LSTATUS, meaning that the link is up (or we
714 * can't tell and just pretend it is), or 0, meaning that the link is
715 * down.
716 *
717 * If reporting is non-zero, instead of faking link up, return -1 if
718 * both ETHTOOL and MII ioctls fail (meaning the device does not
719 * support them). If use_carrier is set, return whatever it says.
720 * It'd be nice if there was a good way to tell if a driver supports
721 * netif_carrier, but there really isn't.
722 */
723 static int bond_check_dev_link(struct bonding *bond,
724 struct net_device *slave_dev, int reporting)
725 {
726 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
727 int (*ioctl)(struct net_device *, struct ifreq *, int);
728 struct ifreq ifr;
729 struct mii_ioctl_data *mii;
730
731 if (!reporting && !netif_running(slave_dev))
732 return 0;
733
734 if (bond->params.use_carrier)
735 return netif_carrier_ok(slave_dev) ? BMSR_LSTATUS : 0;
736
737 /* Try to get link status using Ethtool first. */
738 if (slave_dev->ethtool_ops) {
739 if (slave_dev->ethtool_ops->get_link) {
740 u32 link;
741
742 link = slave_dev->ethtool_ops->get_link(slave_dev);
743
744 return link ? BMSR_LSTATUS : 0;
745 }
746 }
747
748 /* Ethtool can't be used, fallback to MII ioctls. */
749 ioctl = slave_ops->ndo_do_ioctl;
750 if (ioctl) {
751 /* TODO: set pointer to correct ioctl on a per team member */
752 /* bases to make this more efficient. that is, once */
753 /* we determine the correct ioctl, we will always */
754 /* call it and not the others for that team */
755 /* member. */
756
757 /*
758 * We cannot assume that SIOCGMIIPHY will also read a
759 * register; not all network drivers (e.g., e100)
760 * support that.
761 */
762
763 /* Yes, the mii is overlaid on the ifreq.ifr_ifru */
764 strncpy(ifr.ifr_name, slave_dev->name, IFNAMSIZ);
765 mii = if_mii(&ifr);
766 if (IOCTL(slave_dev, &ifr, SIOCGMIIPHY) == 0) {
767 mii->reg_num = MII_BMSR;
768 if (IOCTL(slave_dev, &ifr, SIOCGMIIREG) == 0)
769 return mii->val_out & BMSR_LSTATUS;
770 }
771 }
772
773 /*
774 * If reporting, report that either there's no dev->do_ioctl,
775 * or both SIOCGMIIREG and get_link failed (meaning that we
776 * cannot report link status). If not reporting, pretend
777 * we're ok.
778 */
779 return reporting ? -1 : BMSR_LSTATUS;
780 }
781
782 /*----------------------------- Multicast list ------------------------------*/
783
784 /*
785 * Push the promiscuity flag down to appropriate slaves
786 */
787 static int bond_set_promiscuity(struct bonding *bond, int inc)
788 {
789 int err = 0;
790 if (USES_PRIMARY(bond->params.mode)) {
791 /* write lock already acquired */
792 if (bond->curr_active_slave) {
793 err = dev_set_promiscuity(bond->curr_active_slave->dev,
794 inc);
795 }
796 } else {
797 struct slave *slave;
798 int i;
799 bond_for_each_slave(bond, slave, i) {
800 err = dev_set_promiscuity(slave->dev, inc);
801 if (err)
802 return err;
803 }
804 }
805 return err;
806 }
807
808 /*
809 * Push the allmulti flag down to all slaves
810 */
811 static int bond_set_allmulti(struct bonding *bond, int inc)
812 {
813 int err = 0;
814 if (USES_PRIMARY(bond->params.mode)) {
815 /* write lock already acquired */
816 if (bond->curr_active_slave) {
817 err = dev_set_allmulti(bond->curr_active_slave->dev,
818 inc);
819 }
820 } else {
821 struct slave *slave;
822 int i;
823 bond_for_each_slave(bond, slave, i) {
824 err = dev_set_allmulti(slave->dev, inc);
825 if (err)
826 return err;
827 }
828 }
829 return err;
830 }
831
832 /*
833 * Add a Multicast address to slaves
834 * according to mode
835 */
836 static void bond_mc_add(struct bonding *bond, void *addr)
837 {
838 if (USES_PRIMARY(bond->params.mode)) {
839 /* write lock already acquired */
840 if (bond->curr_active_slave)
841 dev_mc_add(bond->curr_active_slave->dev, addr);
842 } else {
843 struct slave *slave;
844 int i;
845
846 bond_for_each_slave(bond, slave, i)
847 dev_mc_add(slave->dev, addr);
848 }
849 }
850
851 /*
852 * Remove a multicast address from slave
853 * according to mode
854 */
855 static void bond_mc_del(struct bonding *bond, void *addr)
856 {
857 if (USES_PRIMARY(bond->params.mode)) {
858 /* write lock already acquired */
859 if (bond->curr_active_slave)
860 dev_mc_del(bond->curr_active_slave->dev, addr);
861 } else {
862 struct slave *slave;
863 int i;
864 bond_for_each_slave(bond, slave, i) {
865 dev_mc_del(slave->dev, addr);
866 }
867 }
868 }
869
870
871 static void __bond_resend_igmp_join_requests(struct net_device *dev)
872 {
873 struct in_device *in_dev;
874 struct ip_mc_list *im;
875
876 rcu_read_lock();
877 in_dev = __in_dev_get_rcu(dev);
878 if (in_dev) {
879 for (im = in_dev->mc_list; im; im = im->next)
880 ip_mc_rejoin_group(im);
881 }
882
883 rcu_read_unlock();
884 }
885
886 /*
887 * Retrieve the list of registered multicast addresses for the bonding
888 * device and retransmit an IGMP JOIN request to the current active
889 * slave.
890 */
891 static void bond_resend_igmp_join_requests(struct bonding *bond)
892 {
893 struct net_device *vlan_dev;
894 struct vlan_entry *vlan;
895
896 read_lock(&bond->lock);
897
898 /* rejoin all groups on bond device */
899 __bond_resend_igmp_join_requests(bond->dev);
900
901 /* rejoin all groups on vlan devices */
902 if (bond->vlgrp) {
903 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
904 vlan_dev = vlan_group_get_device(bond->vlgrp,
905 vlan->vlan_id);
906 if (vlan_dev)
907 __bond_resend_igmp_join_requests(vlan_dev);
908 }
909 }
910
911 if (--bond->igmp_retrans > 0)
912 queue_delayed_work(bond->wq, &bond->mcast_work, HZ/5);
913
914 read_unlock(&bond->lock);
915 }
916
917 void bond_resend_igmp_join_requests_delayed(struct work_struct *work)
918 {
919 struct bonding *bond = container_of(work, struct bonding,
920 mcast_work.work);
921 bond_resend_igmp_join_requests(bond);
922 }
923
924 /*
925 * flush all members of flush->mc_list from device dev->mc_list
926 */
927 static void bond_mc_list_flush(struct net_device *bond_dev,
928 struct net_device *slave_dev)
929 {
930 struct bonding *bond = netdev_priv(bond_dev);
931 struct netdev_hw_addr *ha;
932
933 netdev_for_each_mc_addr(ha, bond_dev)
934 dev_mc_del(slave_dev, ha->addr);
935
936 if (bond->params.mode == BOND_MODE_8023AD) {
937 /* del lacpdu mc addr from mc list */
938 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
939
940 dev_mc_del(slave_dev, lacpdu_multicast);
941 }
942 }
943
944 /*--------------------------- Active slave change ---------------------------*/
945
946 /*
947 * Update the mc list and multicast-related flags for the new and
948 * old active slaves (if any) according to the multicast mode, and
949 * promiscuous flags unconditionally.
950 */
951 static void bond_mc_swap(struct bonding *bond, struct slave *new_active,
952 struct slave *old_active)
953 {
954 struct netdev_hw_addr *ha;
955
956 if (!USES_PRIMARY(bond->params.mode))
957 /* nothing to do - mc list is already up-to-date on
958 * all slaves
959 */
960 return;
961
962 if (old_active) {
963 if (bond->dev->flags & IFF_PROMISC)
964 dev_set_promiscuity(old_active->dev, -1);
965
966 if (bond->dev->flags & IFF_ALLMULTI)
967 dev_set_allmulti(old_active->dev, -1);
968
969 netdev_for_each_mc_addr(ha, bond->dev)
970 dev_mc_del(old_active->dev, ha->addr);
971 }
972
973 if (new_active) {
974 /* FIXME: Signal errors upstream. */
975 if (bond->dev->flags & IFF_PROMISC)
976 dev_set_promiscuity(new_active->dev, 1);
977
978 if (bond->dev->flags & IFF_ALLMULTI)
979 dev_set_allmulti(new_active->dev, 1);
980
981 netdev_for_each_mc_addr(ha, bond->dev)
982 dev_mc_add(new_active->dev, ha->addr);
983 }
984 }
985
986 /*
987 * bond_do_fail_over_mac
988 *
989 * Perform special MAC address swapping for fail_over_mac settings
990 *
991 * Called with RTNL, bond->lock for read, curr_slave_lock for write_bh.
992 */
993 static void bond_do_fail_over_mac(struct bonding *bond,
994 struct slave *new_active,
995 struct slave *old_active)
996 __releases(&bond->curr_slave_lock)
997 __releases(&bond->lock)
998 __acquires(&bond->lock)
999 __acquires(&bond->curr_slave_lock)
1000 {
1001 u8 tmp_mac[ETH_ALEN];
1002 struct sockaddr saddr;
1003 int rv;
1004
1005 switch (bond->params.fail_over_mac) {
1006 case BOND_FOM_ACTIVE:
1007 if (new_active)
1008 memcpy(bond->dev->dev_addr, new_active->dev->dev_addr,
1009 new_active->dev->addr_len);
1010 break;
1011 case BOND_FOM_FOLLOW:
1012 /*
1013 * if new_active && old_active, swap them
1014 * if just old_active, do nothing (going to no active slave)
1015 * if just new_active, set new_active to bond's MAC
1016 */
1017 if (!new_active)
1018 return;
1019
1020 write_unlock_bh(&bond->curr_slave_lock);
1021 read_unlock(&bond->lock);
1022
1023 if (old_active) {
1024 memcpy(tmp_mac, new_active->dev->dev_addr, ETH_ALEN);
1025 memcpy(saddr.sa_data, old_active->dev->dev_addr,
1026 ETH_ALEN);
1027 saddr.sa_family = new_active->dev->type;
1028 } else {
1029 memcpy(saddr.sa_data, bond->dev->dev_addr, ETH_ALEN);
1030 saddr.sa_family = bond->dev->type;
1031 }
1032
1033 rv = dev_set_mac_address(new_active->dev, &saddr);
1034 if (rv) {
1035 pr_err("%s: Error %d setting MAC of slave %s\n",
1036 bond->dev->name, -rv, new_active->dev->name);
1037 goto out;
1038 }
1039
1040 if (!old_active)
1041 goto out;
1042
1043 memcpy(saddr.sa_data, tmp_mac, ETH_ALEN);
1044 saddr.sa_family = old_active->dev->type;
1045
1046 rv = dev_set_mac_address(old_active->dev, &saddr);
1047 if (rv)
1048 pr_err("%s: Error %d setting MAC of slave %s\n",
1049 bond->dev->name, -rv, new_active->dev->name);
1050 out:
1051 read_lock(&bond->lock);
1052 write_lock_bh(&bond->curr_slave_lock);
1053 break;
1054 default:
1055 pr_err("%s: bond_do_fail_over_mac impossible: bad policy %d\n",
1056 bond->dev->name, bond->params.fail_over_mac);
1057 break;
1058 }
1059
1060 }
1061
1062 static bool bond_should_change_active(struct bonding *bond)
1063 {
1064 struct slave *prim = bond->primary_slave;
1065 struct slave *curr = bond->curr_active_slave;
1066
1067 if (!prim || !curr || curr->link != BOND_LINK_UP)
1068 return true;
1069 if (bond->force_primary) {
1070 bond->force_primary = false;
1071 return true;
1072 }
1073 if (bond->params.primary_reselect == BOND_PRI_RESELECT_BETTER &&
1074 (prim->speed < curr->speed ||
1075 (prim->speed == curr->speed && prim->duplex <= curr->duplex)))
1076 return false;
1077 if (bond->params.primary_reselect == BOND_PRI_RESELECT_FAILURE)
1078 return false;
1079 return true;
1080 }
1081
1082 /**
1083 * find_best_interface - select the best available slave to be the active one
1084 * @bond: our bonding struct
1085 *
1086 * Warning: Caller must hold curr_slave_lock for writing.
1087 */
1088 static struct slave *bond_find_best_slave(struct bonding *bond)
1089 {
1090 struct slave *new_active, *old_active;
1091 struct slave *bestslave = NULL;
1092 int mintime = bond->params.updelay;
1093 int i;
1094
1095 new_active = bond->curr_active_slave;
1096
1097 if (!new_active) { /* there were no active slaves left */
1098 if (bond->slave_cnt > 0) /* found one slave */
1099 new_active = bond->first_slave;
1100 else
1101 return NULL; /* still no slave, return NULL */
1102 }
1103
1104 if ((bond->primary_slave) &&
1105 bond->primary_slave->link == BOND_LINK_UP &&
1106 bond_should_change_active(bond)) {
1107 new_active = bond->primary_slave;
1108 }
1109
1110 /* remember where to stop iterating over the slaves */
1111 old_active = new_active;
1112
1113 bond_for_each_slave_from(bond, new_active, i, old_active) {
1114 if (new_active->link == BOND_LINK_UP) {
1115 return new_active;
1116 } else if (new_active->link == BOND_LINK_BACK &&
1117 IS_UP(new_active->dev)) {
1118 /* link up, but waiting for stabilization */
1119 if (new_active->delay < mintime) {
1120 mintime = new_active->delay;
1121 bestslave = new_active;
1122 }
1123 }
1124 }
1125
1126 return bestslave;
1127 }
1128
1129 /**
1130 * change_active_interface - change the active slave into the specified one
1131 * @bond: our bonding struct
1132 * @new: the new slave to make the active one
1133 *
1134 * Set the new slave to the bond's settings and unset them on the old
1135 * curr_active_slave.
1136 * Setting include flags, mc-list, promiscuity, allmulti, etc.
1137 *
1138 * If @new's link state is %BOND_LINK_BACK we'll set it to %BOND_LINK_UP,
1139 * because it is apparently the best available slave we have, even though its
1140 * updelay hasn't timed out yet.
1141 *
1142 * If new_active is not NULL, caller must hold bond->lock for read and
1143 * curr_slave_lock for write_bh.
1144 */
1145 void bond_change_active_slave(struct bonding *bond, struct slave *new_active)
1146 {
1147 struct slave *old_active = bond->curr_active_slave;
1148
1149 if (old_active == new_active)
1150 return;
1151
1152 if (new_active) {
1153 new_active->jiffies = jiffies;
1154
1155 if (new_active->link == BOND_LINK_BACK) {
1156 if (USES_PRIMARY(bond->params.mode)) {
1157 pr_info("%s: making interface %s the new active one %d ms earlier.\n",
1158 bond->dev->name, new_active->dev->name,
1159 (bond->params.updelay - new_active->delay) * bond->params.miimon);
1160 }
1161
1162 new_active->delay = 0;
1163 new_active->link = BOND_LINK_UP;
1164
1165 if (bond->params.mode == BOND_MODE_8023AD)
1166 bond_3ad_handle_link_change(new_active, BOND_LINK_UP);
1167
1168 if (bond_is_lb(bond))
1169 bond_alb_handle_link_change(bond, new_active, BOND_LINK_UP);
1170 } else {
1171 if (USES_PRIMARY(bond->params.mode)) {
1172 pr_info("%s: making interface %s the new active one.\n",
1173 bond->dev->name, new_active->dev->name);
1174 }
1175 }
1176 }
1177
1178 if (USES_PRIMARY(bond->params.mode))
1179 bond_mc_swap(bond, new_active, old_active);
1180
1181 if (bond_is_lb(bond)) {
1182 bond_alb_handle_active_change(bond, new_active);
1183 if (old_active)
1184 bond_set_slave_inactive_flags(old_active);
1185 if (new_active)
1186 bond_set_slave_active_flags(new_active);
1187 } else {
1188 bond->curr_active_slave = new_active;
1189 }
1190
1191 if (bond->params.mode == BOND_MODE_ACTIVEBACKUP) {
1192 if (old_active)
1193 bond_set_slave_inactive_flags(old_active);
1194
1195 if (new_active) {
1196 bond_set_slave_active_flags(new_active);
1197
1198 if (bond->params.fail_over_mac)
1199 bond_do_fail_over_mac(bond, new_active,
1200 old_active);
1201
1202 bond->send_grat_arp = bond->params.num_grat_arp;
1203 bond_send_gratuitous_arp(bond);
1204
1205 bond->send_unsol_na = bond->params.num_unsol_na;
1206 bond_send_unsolicited_na(bond);
1207
1208 write_unlock_bh(&bond->curr_slave_lock);
1209 read_unlock(&bond->lock);
1210
1211 netdev_bonding_change(bond->dev, NETDEV_BONDING_FAILOVER);
1212
1213 read_lock(&bond->lock);
1214 write_lock_bh(&bond->curr_slave_lock);
1215 }
1216 }
1217
1218 /* resend IGMP joins since active slave has changed or
1219 * all were sent on curr_active_slave */
1220 if ((USES_PRIMARY(bond->params.mode) && new_active) ||
1221 bond->params.mode == BOND_MODE_ROUNDROBIN) {
1222 bond->igmp_retrans = bond->params.resend_igmp;
1223 queue_delayed_work(bond->wq, &bond->mcast_work, 0);
1224 }
1225 }
1226
1227 /**
1228 * bond_select_active_slave - select a new active slave, if needed
1229 * @bond: our bonding struct
1230 *
1231 * This functions should be called when one of the following occurs:
1232 * - The old curr_active_slave has been released or lost its link.
1233 * - The primary_slave has got its link back.
1234 * - A slave has got its link back and there's no old curr_active_slave.
1235 *
1236 * Caller must hold bond->lock for read and curr_slave_lock for write_bh.
1237 */
1238 void bond_select_active_slave(struct bonding *bond)
1239 {
1240 struct slave *best_slave;
1241 int rv;
1242
1243 best_slave = bond_find_best_slave(bond);
1244 if (best_slave != bond->curr_active_slave) {
1245 bond_change_active_slave(bond, best_slave);
1246 rv = bond_set_carrier(bond);
1247 if (!rv)
1248 return;
1249
1250 if (netif_carrier_ok(bond->dev)) {
1251 pr_info("%s: first active interface up!\n",
1252 bond->dev->name);
1253 } else {
1254 pr_info("%s: now running without any active interface !\n",
1255 bond->dev->name);
1256 }
1257 }
1258 }
1259
1260 /*--------------------------- slave list handling ---------------------------*/
1261
1262 /*
1263 * This function attaches the slave to the end of list.
1264 *
1265 * bond->lock held for writing by caller.
1266 */
1267 static void bond_attach_slave(struct bonding *bond, struct slave *new_slave)
1268 {
1269 if (bond->first_slave == NULL) { /* attaching the first slave */
1270 new_slave->next = new_slave;
1271 new_slave->prev = new_slave;
1272 bond->first_slave = new_slave;
1273 } else {
1274 new_slave->next = bond->first_slave;
1275 new_slave->prev = bond->first_slave->prev;
1276 new_slave->next->prev = new_slave;
1277 new_slave->prev->next = new_slave;
1278 }
1279
1280 bond->slave_cnt++;
1281 }
1282
1283 /*
1284 * This function detaches the slave from the list.
1285 * WARNING: no check is made to verify if the slave effectively
1286 * belongs to <bond>.
1287 * Nothing is freed on return, structures are just unchained.
1288 * If any slave pointer in bond was pointing to <slave>,
1289 * it should be changed by the calling function.
1290 *
1291 * bond->lock held for writing by caller.
1292 */
1293 static void bond_detach_slave(struct bonding *bond, struct slave *slave)
1294 {
1295 if (slave->next)
1296 slave->next->prev = slave->prev;
1297
1298 if (slave->prev)
1299 slave->prev->next = slave->next;
1300
1301 if (bond->first_slave == slave) { /* slave is the first slave */
1302 if (bond->slave_cnt > 1) { /* there are more slave */
1303 bond->first_slave = slave->next;
1304 } else {
1305 bond->first_slave = NULL; /* slave was the last one */
1306 }
1307 }
1308
1309 slave->next = NULL;
1310 slave->prev = NULL;
1311 bond->slave_cnt--;
1312 }
1313
1314 #ifdef CONFIG_NET_POLL_CONTROLLER
1315 /*
1316 * You must hold read lock on bond->lock before calling this.
1317 */
1318 static bool slaves_support_netpoll(struct net_device *bond_dev)
1319 {
1320 struct bonding *bond = netdev_priv(bond_dev);
1321 struct slave *slave;
1322 int i = 0;
1323 bool ret = true;
1324
1325 bond_for_each_slave(bond, slave, i) {
1326 if ((slave->dev->priv_flags & IFF_DISABLE_NETPOLL) ||
1327 !slave->dev->netdev_ops->ndo_poll_controller)
1328 ret = false;
1329 }
1330 return i != 0 && ret;
1331 }
1332
1333 static void bond_poll_controller(struct net_device *bond_dev)
1334 {
1335 struct net_device *dev = bond_dev->npinfo->netpoll->real_dev;
1336 if (dev != bond_dev)
1337 netpoll_poll_dev(dev);
1338 }
1339
1340 static void bond_netpoll_cleanup(struct net_device *bond_dev)
1341 {
1342 struct bonding *bond = netdev_priv(bond_dev);
1343 struct slave *slave;
1344 const struct net_device_ops *ops;
1345 int i;
1346
1347 read_lock(&bond->lock);
1348 bond_dev->npinfo = NULL;
1349 bond_for_each_slave(bond, slave, i) {
1350 if (slave->dev) {
1351 ops = slave->dev->netdev_ops;
1352 if (ops->ndo_netpoll_cleanup)
1353 ops->ndo_netpoll_cleanup(slave->dev);
1354 else
1355 slave->dev->npinfo = NULL;
1356 }
1357 }
1358 read_unlock(&bond->lock);
1359 }
1360
1361 #else
1362
1363 static void bond_netpoll_cleanup(struct net_device *bond_dev)
1364 {
1365 }
1366
1367 #endif
1368
1369 /*---------------------------------- IOCTL ----------------------------------*/
1370
1371 static int bond_sethwaddr(struct net_device *bond_dev,
1372 struct net_device *slave_dev)
1373 {
1374 pr_debug("bond_dev=%p\n", bond_dev);
1375 pr_debug("slave_dev=%p\n", slave_dev);
1376 pr_debug("slave_dev->addr_len=%d\n", slave_dev->addr_len);
1377 memcpy(bond_dev->dev_addr, slave_dev->dev_addr, slave_dev->addr_len);
1378 return 0;
1379 }
1380
1381 #define BOND_VLAN_FEATURES \
1382 (NETIF_F_VLAN_CHALLENGED | NETIF_F_HW_VLAN_RX | NETIF_F_HW_VLAN_TX | \
1383 NETIF_F_HW_VLAN_FILTER)
1384
1385 /*
1386 * Compute the common dev->feature set available to all slaves. Some
1387 * feature bits are managed elsewhere, so preserve those feature bits
1388 * on the master device.
1389 */
1390 static int bond_compute_features(struct bonding *bond)
1391 {
1392 struct slave *slave;
1393 struct net_device *bond_dev = bond->dev;
1394 unsigned long features = bond_dev->features;
1395 unsigned long vlan_features = 0;
1396 unsigned short max_hard_header_len = max((u16)ETH_HLEN,
1397 bond_dev->hard_header_len);
1398 int i;
1399
1400 features &= ~(NETIF_F_ALL_CSUM | BOND_VLAN_FEATURES);
1401 features |= NETIF_F_GSO_MASK | NETIF_F_NO_CSUM;
1402
1403 if (!bond->first_slave)
1404 goto done;
1405
1406 features &= ~NETIF_F_ONE_FOR_ALL;
1407
1408 vlan_features = bond->first_slave->dev->vlan_features;
1409 bond_for_each_slave(bond, slave, i) {
1410 features = netdev_increment_features(features,
1411 slave->dev->features,
1412 NETIF_F_ONE_FOR_ALL);
1413 vlan_features = netdev_increment_features(vlan_features,
1414 slave->dev->vlan_features,
1415 NETIF_F_ONE_FOR_ALL);
1416 if (slave->dev->hard_header_len > max_hard_header_len)
1417 max_hard_header_len = slave->dev->hard_header_len;
1418 }
1419
1420 done:
1421 features |= (bond_dev->features & BOND_VLAN_FEATURES);
1422 bond_dev->features = netdev_fix_features(features, NULL);
1423 bond_dev->vlan_features = netdev_fix_features(vlan_features, NULL);
1424 bond_dev->hard_header_len = max_hard_header_len;
1425
1426 return 0;
1427 }
1428
1429 static void bond_setup_by_slave(struct net_device *bond_dev,
1430 struct net_device *slave_dev)
1431 {
1432 struct bonding *bond = netdev_priv(bond_dev);
1433
1434 bond_dev->header_ops = slave_dev->header_ops;
1435
1436 bond_dev->type = slave_dev->type;
1437 bond_dev->hard_header_len = slave_dev->hard_header_len;
1438 bond_dev->addr_len = slave_dev->addr_len;
1439
1440 memcpy(bond_dev->broadcast, slave_dev->broadcast,
1441 slave_dev->addr_len);
1442 bond->setup_by_slave = 1;
1443 }
1444
1445 /* enslave device <slave> to bond device <master> */
1446 int bond_enslave(struct net_device *bond_dev, struct net_device *slave_dev)
1447 {
1448 struct bonding *bond = netdev_priv(bond_dev);
1449 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
1450 struct slave *new_slave = NULL;
1451 struct netdev_hw_addr *ha;
1452 struct sockaddr addr;
1453 int link_reporting;
1454 int old_features = bond_dev->features;
1455 int res = 0;
1456
1457 if (!bond->params.use_carrier && slave_dev->ethtool_ops == NULL &&
1458 slave_ops->ndo_do_ioctl == NULL) {
1459 pr_warning("%s: Warning: no link monitoring support for %s\n",
1460 bond_dev->name, slave_dev->name);
1461 }
1462
1463 /* bond must be initialized by bond_open() before enslaving */
1464 if (!(bond_dev->flags & IFF_UP)) {
1465 pr_warning("%s: master_dev is not up in bond_enslave\n",
1466 bond_dev->name);
1467 }
1468
1469 /* already enslaved */
1470 if (slave_dev->flags & IFF_SLAVE) {
1471 pr_debug("Error, Device was already enslaved\n");
1472 return -EBUSY;
1473 }
1474
1475 /* vlan challenged mutual exclusion */
1476 /* no need to lock since we're protected by rtnl_lock */
1477 if (slave_dev->features & NETIF_F_VLAN_CHALLENGED) {
1478 pr_debug("%s: NETIF_F_VLAN_CHALLENGED\n", slave_dev->name);
1479 if (bond->vlgrp) {
1480 pr_err("%s: Error: cannot enslave VLAN challenged slave %s on VLAN enabled bond %s\n",
1481 bond_dev->name, slave_dev->name, bond_dev->name);
1482 return -EPERM;
1483 } else {
1484 pr_warning("%s: Warning: enslaved VLAN challenged slave %s. Adding VLANs will be blocked as long as %s is part of bond %s\n",
1485 bond_dev->name, slave_dev->name,
1486 slave_dev->name, bond_dev->name);
1487 bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
1488 }
1489 } else {
1490 pr_debug("%s: ! NETIF_F_VLAN_CHALLENGED\n", slave_dev->name);
1491 if (bond->slave_cnt == 0) {
1492 /* First slave, and it is not VLAN challenged,
1493 * so remove the block of adding VLANs over the bond.
1494 */
1495 bond_dev->features &= ~NETIF_F_VLAN_CHALLENGED;
1496 }
1497 }
1498
1499 /*
1500 * Old ifenslave binaries are no longer supported. These can
1501 * be identified with moderate accuracy by the state of the slave:
1502 * the current ifenslave will set the interface down prior to
1503 * enslaving it; the old ifenslave will not.
1504 */
1505 if ((slave_dev->flags & IFF_UP)) {
1506 pr_err("%s is up. This may be due to an out of date ifenslave.\n",
1507 slave_dev->name);
1508 res = -EPERM;
1509 goto err_undo_flags;
1510 }
1511
1512 /* set bonding device ether type by slave - bonding netdevices are
1513 * created with ether_setup, so when the slave type is not ARPHRD_ETHER
1514 * there is a need to override some of the type dependent attribs/funcs.
1515 *
1516 * bond ether type mutual exclusion - don't allow slaves of dissimilar
1517 * ether type (eg ARPHRD_ETHER and ARPHRD_INFINIBAND) share the same bond
1518 */
1519 if (bond->slave_cnt == 0) {
1520 if (bond_dev->type != slave_dev->type) {
1521 pr_debug("%s: change device type from %d to %d\n",
1522 bond_dev->name,
1523 bond_dev->type, slave_dev->type);
1524
1525 res = netdev_bonding_change(bond_dev,
1526 NETDEV_PRE_TYPE_CHANGE);
1527 res = notifier_to_errno(res);
1528 if (res) {
1529 pr_err("%s: refused to change device type\n",
1530 bond_dev->name);
1531 res = -EBUSY;
1532 goto err_undo_flags;
1533 }
1534
1535 /* Flush unicast and multicast addresses */
1536 dev_uc_flush(bond_dev);
1537 dev_mc_flush(bond_dev);
1538
1539 if (slave_dev->type != ARPHRD_ETHER)
1540 bond_setup_by_slave(bond_dev, slave_dev);
1541 else
1542 ether_setup(bond_dev);
1543
1544 netdev_bonding_change(bond_dev,
1545 NETDEV_POST_TYPE_CHANGE);
1546 }
1547 } else if (bond_dev->type != slave_dev->type) {
1548 pr_err("%s ether type (%d) is different from other slaves (%d), can not enslave it.\n",
1549 slave_dev->name,
1550 slave_dev->type, bond_dev->type);
1551 res = -EINVAL;
1552 goto err_undo_flags;
1553 }
1554
1555 if (slave_ops->ndo_set_mac_address == NULL) {
1556 if (bond->slave_cnt == 0) {
1557 pr_warning("%s: Warning: The first slave device specified does not support setting the MAC address. Setting fail_over_mac to active.",
1558 bond_dev->name);
1559 bond->params.fail_over_mac = BOND_FOM_ACTIVE;
1560 } else if (bond->params.fail_over_mac != BOND_FOM_ACTIVE) {
1561 pr_err("%s: Error: The slave device specified does not support setting the MAC address, but fail_over_mac is not set to active.\n",
1562 bond_dev->name);
1563 res = -EOPNOTSUPP;
1564 goto err_undo_flags;
1565 }
1566 }
1567
1568 /* If this is the first slave, then we need to set the master's hardware
1569 * address to be the same as the slave's. */
1570 if (bond->slave_cnt == 0)
1571 memcpy(bond->dev->dev_addr, slave_dev->dev_addr,
1572 slave_dev->addr_len);
1573
1574
1575 new_slave = kzalloc(sizeof(struct slave), GFP_KERNEL);
1576 if (!new_slave) {
1577 res = -ENOMEM;
1578 goto err_undo_flags;
1579 }
1580
1581 /*
1582 * Set the new_slave's queue_id to be zero. Queue ID mapping
1583 * is set via sysfs or module option if desired.
1584 */
1585 new_slave->queue_id = 0;
1586
1587 /* Save slave's original mtu and then set it to match the bond */
1588 new_slave->original_mtu = slave_dev->mtu;
1589 res = dev_set_mtu(slave_dev, bond->dev->mtu);
1590 if (res) {
1591 pr_debug("Error %d calling dev_set_mtu\n", res);
1592 goto err_free;
1593 }
1594
1595 /*
1596 * Save slave's original ("permanent") mac address for modes
1597 * that need it, and for restoring it upon release, and then
1598 * set it to the master's address
1599 */
1600 memcpy(new_slave->perm_hwaddr, slave_dev->dev_addr, ETH_ALEN);
1601
1602 if (!bond->params.fail_over_mac) {
1603 /*
1604 * Set slave to master's mac address. The application already
1605 * set the master's mac address to that of the first slave
1606 */
1607 memcpy(addr.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
1608 addr.sa_family = slave_dev->type;
1609 res = dev_set_mac_address(slave_dev, &addr);
1610 if (res) {
1611 pr_debug("Error %d calling set_mac_address\n", res);
1612 goto err_restore_mtu;
1613 }
1614 }
1615
1616 res = netdev_set_master(slave_dev, bond_dev);
1617 if (res) {
1618 pr_debug("Error %d calling netdev_set_master\n", res);
1619 goto err_restore_mac;
1620 }
1621 /* open the slave since the application closed it */
1622 res = dev_open(slave_dev);
1623 if (res) {
1624 pr_debug("Opening slave %s failed\n", slave_dev->name);
1625 goto err_unset_master;
1626 }
1627
1628 new_slave->dev = slave_dev;
1629 slave_dev->priv_flags |= IFF_BONDING;
1630
1631 if (bond_is_lb(bond)) {
1632 /* bond_alb_init_slave() must be called before all other stages since
1633 * it might fail and we do not want to have to undo everything
1634 */
1635 res = bond_alb_init_slave(bond, new_slave);
1636 if (res)
1637 goto err_close;
1638 }
1639
1640 /* If the mode USES_PRIMARY, then the new slave gets the
1641 * master's promisc (and mc) settings only if it becomes the
1642 * curr_active_slave, and that is taken care of later when calling
1643 * bond_change_active()
1644 */
1645 if (!USES_PRIMARY(bond->params.mode)) {
1646 /* set promiscuity level to new slave */
1647 if (bond_dev->flags & IFF_PROMISC) {
1648 res = dev_set_promiscuity(slave_dev, 1);
1649 if (res)
1650 goto err_close;
1651 }
1652
1653 /* set allmulti level to new slave */
1654 if (bond_dev->flags & IFF_ALLMULTI) {
1655 res = dev_set_allmulti(slave_dev, 1);
1656 if (res)
1657 goto err_close;
1658 }
1659
1660 netif_addr_lock_bh(bond_dev);
1661 /* upload master's mc_list to new slave */
1662 netdev_for_each_mc_addr(ha, bond_dev)
1663 dev_mc_add(slave_dev, ha->addr);
1664 netif_addr_unlock_bh(bond_dev);
1665 }
1666
1667 if (bond->params.mode == BOND_MODE_8023AD) {
1668 /* add lacpdu mc addr to mc list */
1669 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
1670
1671 dev_mc_add(slave_dev, lacpdu_multicast);
1672 }
1673
1674 bond_add_vlans_on_slave(bond, slave_dev);
1675
1676 write_lock_bh(&bond->lock);
1677
1678 bond_attach_slave(bond, new_slave);
1679
1680 new_slave->delay = 0;
1681 new_slave->link_failure_count = 0;
1682
1683 bond_compute_features(bond);
1684
1685 write_unlock_bh(&bond->lock);
1686
1687 read_lock(&bond->lock);
1688
1689 new_slave->last_arp_rx = jiffies;
1690
1691 if (bond->params.miimon && !bond->params.use_carrier) {
1692 link_reporting = bond_check_dev_link(bond, slave_dev, 1);
1693
1694 if ((link_reporting == -1) && !bond->params.arp_interval) {
1695 /*
1696 * miimon is set but a bonded network driver
1697 * does not support ETHTOOL/MII and
1698 * arp_interval is not set. Note: if
1699 * use_carrier is enabled, we will never go
1700 * here (because netif_carrier is always
1701 * supported); thus, we don't need to change
1702 * the messages for netif_carrier.
1703 */
1704 pr_warning("%s: Warning: MII and ETHTOOL support not available for interface %s, and arp_interval/arp_ip_target module parameters not specified, thus bonding will not detect link failures! see bonding.txt for details.\n",
1705 bond_dev->name, slave_dev->name);
1706 } else if (link_reporting == -1) {
1707 /* unable get link status using mii/ethtool */
1708 pr_warning("%s: Warning: can't get link status from interface %s; the network driver associated with this interface does not support MII or ETHTOOL link status reporting, thus miimon has no effect on this interface.\n",
1709 bond_dev->name, slave_dev->name);
1710 }
1711 }
1712
1713 /* check for initial state */
1714 if (!bond->params.miimon ||
1715 (bond_check_dev_link(bond, slave_dev, 0) == BMSR_LSTATUS)) {
1716 if (bond->params.updelay) {
1717 pr_debug("Initial state of slave_dev is BOND_LINK_BACK\n");
1718 new_slave->link = BOND_LINK_BACK;
1719 new_slave->delay = bond->params.updelay;
1720 } else {
1721 pr_debug("Initial state of slave_dev is BOND_LINK_UP\n");
1722 new_slave->link = BOND_LINK_UP;
1723 }
1724 new_slave->jiffies = jiffies;
1725 } else {
1726 pr_debug("Initial state of slave_dev is BOND_LINK_DOWN\n");
1727 new_slave->link = BOND_LINK_DOWN;
1728 }
1729
1730 if (bond_update_speed_duplex(new_slave) &&
1731 (new_slave->link != BOND_LINK_DOWN)) {
1732 pr_warning("%s: Warning: failed to get speed and duplex from %s, assumed to be 100Mb/sec and Full.\n",
1733 bond_dev->name, new_slave->dev->name);
1734
1735 if (bond->params.mode == BOND_MODE_8023AD) {
1736 pr_warning("%s: Warning: Operation of 802.3ad mode requires ETHTOOL support in base driver for proper aggregator selection.\n",
1737 bond_dev->name);
1738 }
1739 }
1740
1741 if (USES_PRIMARY(bond->params.mode) && bond->params.primary[0]) {
1742 /* if there is a primary slave, remember it */
1743 if (strcmp(bond->params.primary, new_slave->dev->name) == 0) {
1744 bond->primary_slave = new_slave;
1745 bond->force_primary = true;
1746 }
1747 }
1748
1749 write_lock_bh(&bond->curr_slave_lock);
1750
1751 switch (bond->params.mode) {
1752 case BOND_MODE_ACTIVEBACKUP:
1753 bond_set_slave_inactive_flags(new_slave);
1754 bond_select_active_slave(bond);
1755 break;
1756 case BOND_MODE_8023AD:
1757 /* in 802.3ad mode, the internal mechanism
1758 * will activate the slaves in the selected
1759 * aggregator
1760 */
1761 bond_set_slave_inactive_flags(new_slave);
1762 /* if this is the first slave */
1763 if (bond->slave_cnt == 1) {
1764 SLAVE_AD_INFO(new_slave).id = 1;
1765 /* Initialize AD with the number of times that the AD timer is called in 1 second
1766 * can be called only after the mac address of the bond is set
1767 */
1768 bond_3ad_initialize(bond, 1000/AD_TIMER_INTERVAL,
1769 bond->params.lacp_fast);
1770 } else {
1771 SLAVE_AD_INFO(new_slave).id =
1772 SLAVE_AD_INFO(new_slave->prev).id + 1;
1773 }
1774
1775 bond_3ad_bind_slave(new_slave);
1776 break;
1777 case BOND_MODE_TLB:
1778 case BOND_MODE_ALB:
1779 new_slave->state = BOND_STATE_ACTIVE;
1780 bond_set_slave_inactive_flags(new_slave);
1781 bond_select_active_slave(bond);
1782 break;
1783 default:
1784 pr_debug("This slave is always active in trunk mode\n");
1785
1786 /* always active in trunk mode */
1787 new_slave->state = BOND_STATE_ACTIVE;
1788
1789 /* In trunking mode there is little meaning to curr_active_slave
1790 * anyway (it holds no special properties of the bond device),
1791 * so we can change it without calling change_active_interface()
1792 */
1793 if (!bond->curr_active_slave)
1794 bond->curr_active_slave = new_slave;
1795
1796 break;
1797 } /* switch(bond_mode) */
1798
1799 write_unlock_bh(&bond->curr_slave_lock);
1800
1801 bond_set_carrier(bond);
1802
1803 #ifdef CONFIG_NET_POLL_CONTROLLER
1804 /*
1805 * Netpoll and bonding is broken, make sure it is not initialized
1806 * until it is fixed.
1807 */
1808 if (disable_netpoll) {
1809 bond_dev->priv_flags |= IFF_DISABLE_NETPOLL;
1810 } else {
1811 if (slaves_support_netpoll(bond_dev)) {
1812 bond_dev->priv_flags &= ~IFF_DISABLE_NETPOLL;
1813 if (bond_dev->npinfo)
1814 slave_dev->npinfo = bond_dev->npinfo;
1815 } else if (!(bond_dev->priv_flags & IFF_DISABLE_NETPOLL)) {
1816 bond_dev->priv_flags |= IFF_DISABLE_NETPOLL;
1817 pr_info("New slave device %s does not support netpoll\n",
1818 slave_dev->name);
1819 pr_info("Disabling netpoll support for %s\n", bond_dev->name);
1820 }
1821 }
1822 #endif
1823 read_unlock(&bond->lock);
1824
1825 res = bond_create_slave_symlinks(bond_dev, slave_dev);
1826 if (res)
1827 goto err_close;
1828
1829 pr_info("%s: enslaving %s as a%s interface with a%s link.\n",
1830 bond_dev->name, slave_dev->name,
1831 new_slave->state == BOND_STATE_ACTIVE ? "n active" : " backup",
1832 new_slave->link != BOND_LINK_DOWN ? "n up" : " down");
1833
1834 /* enslave is successful */
1835 return 0;
1836
1837 /* Undo stages on error */
1838 err_close:
1839 dev_close(slave_dev);
1840
1841 err_unset_master:
1842 netdev_set_master(slave_dev, NULL);
1843
1844 err_restore_mac:
1845 if (!bond->params.fail_over_mac) {
1846 /* XXX TODO - fom follow mode needs to change master's
1847 * MAC if this slave's MAC is in use by the bond, or at
1848 * least print a warning.
1849 */
1850 memcpy(addr.sa_data, new_slave->perm_hwaddr, ETH_ALEN);
1851 addr.sa_family = slave_dev->type;
1852 dev_set_mac_address(slave_dev, &addr);
1853 }
1854
1855 err_restore_mtu:
1856 dev_set_mtu(slave_dev, new_slave->original_mtu);
1857
1858 err_free:
1859 kfree(new_slave);
1860
1861 err_undo_flags:
1862 bond_dev->features = old_features;
1863
1864 return res;
1865 }
1866
1867 /*
1868 * Try to release the slave device <slave> from the bond device <master>
1869 * It is legal to access curr_active_slave without a lock because all the function
1870 * is write-locked.
1871 *
1872 * The rules for slave state should be:
1873 * for Active/Backup:
1874 * Active stays on all backups go down
1875 * for Bonded connections:
1876 * The first up interface should be left on and all others downed.
1877 */
1878 int bond_release(struct net_device *bond_dev, struct net_device *slave_dev)
1879 {
1880 struct bonding *bond = netdev_priv(bond_dev);
1881 struct slave *slave, *oldcurrent;
1882 struct sockaddr addr;
1883
1884 /* slave is not a slave or master is not master of this slave */
1885 if (!(slave_dev->flags & IFF_SLAVE) ||
1886 (slave_dev->master != bond_dev)) {
1887 pr_err("%s: Error: cannot release %s.\n",
1888 bond_dev->name, slave_dev->name);
1889 return -EINVAL;
1890 }
1891
1892 netdev_bonding_change(bond_dev, NETDEV_BONDING_DESLAVE);
1893 write_lock_bh(&bond->lock);
1894
1895 slave = bond_get_slave_by_dev(bond, slave_dev);
1896 if (!slave) {
1897 /* not a slave of this bond */
1898 pr_info("%s: %s not enslaved\n",
1899 bond_dev->name, slave_dev->name);
1900 write_unlock_bh(&bond->lock);
1901 return -EINVAL;
1902 }
1903
1904 if (!bond->params.fail_over_mac) {
1905 if (!compare_ether_addr(bond_dev->dev_addr, slave->perm_hwaddr) &&
1906 bond->slave_cnt > 1)
1907 pr_warning("%s: Warning: the permanent HWaddr of %s - %pM - is still in use by %s. Set the HWaddr of %s to a different address to avoid conflicts.\n",
1908 bond_dev->name, slave_dev->name,
1909 slave->perm_hwaddr,
1910 bond_dev->name, slave_dev->name);
1911 }
1912
1913 /* Inform AD package of unbinding of slave. */
1914 if (bond->params.mode == BOND_MODE_8023AD) {
1915 /* must be called before the slave is
1916 * detached from the list
1917 */
1918 bond_3ad_unbind_slave(slave);
1919 }
1920
1921 pr_info("%s: releasing %s interface %s\n",
1922 bond_dev->name,
1923 (slave->state == BOND_STATE_ACTIVE) ? "active" : "backup",
1924 slave_dev->name);
1925
1926 oldcurrent = bond->curr_active_slave;
1927
1928 bond->current_arp_slave = NULL;
1929
1930 /* release the slave from its bond */
1931 bond_detach_slave(bond, slave);
1932
1933 bond_compute_features(bond);
1934
1935 if (bond->primary_slave == slave)
1936 bond->primary_slave = NULL;
1937
1938 if (oldcurrent == slave)
1939 bond_change_active_slave(bond, NULL);
1940
1941 if (bond_is_lb(bond)) {
1942 /* Must be called only after the slave has been
1943 * detached from the list and the curr_active_slave
1944 * has been cleared (if our_slave == old_current),
1945 * but before a new active slave is selected.
1946 */
1947 write_unlock_bh(&bond->lock);
1948 bond_alb_deinit_slave(bond, slave);
1949 write_lock_bh(&bond->lock);
1950 }
1951
1952 if (oldcurrent == slave) {
1953 /*
1954 * Note that we hold RTNL over this sequence, so there
1955 * is no concern that another slave add/remove event
1956 * will interfere.
1957 */
1958 write_unlock_bh(&bond->lock);
1959 read_lock(&bond->lock);
1960 write_lock_bh(&bond->curr_slave_lock);
1961
1962 bond_select_active_slave(bond);
1963
1964 write_unlock_bh(&bond->curr_slave_lock);
1965 read_unlock(&bond->lock);
1966 write_lock_bh(&bond->lock);
1967 }
1968
1969 if (bond->slave_cnt == 0) {
1970 bond_set_carrier(bond);
1971
1972 /* if the last slave was removed, zero the mac address
1973 * of the master so it will be set by the application
1974 * to the mac address of the first slave
1975 */
1976 memset(bond_dev->dev_addr, 0, bond_dev->addr_len);
1977
1978 if (!bond->vlgrp) {
1979 bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
1980 } else {
1981 pr_warning("%s: Warning: clearing HW address of %s while it still has VLANs.\n",
1982 bond_dev->name, bond_dev->name);
1983 pr_warning("%s: When re-adding slaves, make sure the bond's HW address matches its VLANs'.\n",
1984 bond_dev->name);
1985 }
1986 } else if ((bond_dev->features & NETIF_F_VLAN_CHALLENGED) &&
1987 !bond_has_challenged_slaves(bond)) {
1988 pr_info("%s: last VLAN challenged slave %s left bond %s. VLAN blocking is removed\n",
1989 bond_dev->name, slave_dev->name, bond_dev->name);
1990 bond_dev->features &= ~NETIF_F_VLAN_CHALLENGED;
1991 }
1992
1993 write_unlock_bh(&bond->lock);
1994
1995 /* must do this from outside any spinlocks */
1996 bond_destroy_slave_symlinks(bond_dev, slave_dev);
1997
1998 bond_del_vlans_from_slave(bond, slave_dev);
1999
2000 /* If the mode USES_PRIMARY, then we should only remove its
2001 * promisc and mc settings if it was the curr_active_slave, but that was
2002 * already taken care of above when we detached the slave
2003 */
2004 if (!USES_PRIMARY(bond->params.mode)) {
2005 /* unset promiscuity level from slave */
2006 if (bond_dev->flags & IFF_PROMISC)
2007 dev_set_promiscuity(slave_dev, -1);
2008
2009 /* unset allmulti level from slave */
2010 if (bond_dev->flags & IFF_ALLMULTI)
2011 dev_set_allmulti(slave_dev, -1);
2012
2013 /* flush master's mc_list from slave */
2014 netif_addr_lock_bh(bond_dev);
2015 bond_mc_list_flush(bond_dev, slave_dev);
2016 netif_addr_unlock_bh(bond_dev);
2017 }
2018
2019 netdev_set_master(slave_dev, NULL);
2020
2021 #ifdef CONFIG_NET_POLL_CONTROLLER
2022 read_lock_bh(&bond->lock);
2023
2024 /* Make sure netpoll over stays disabled until fixed. */
2025 if (!disable_netpoll)
2026 if (slaves_support_netpoll(bond_dev))
2027 bond_dev->priv_flags &= ~IFF_DISABLE_NETPOLL;
2028 read_unlock_bh(&bond->lock);
2029 if (slave_dev->netdev_ops->ndo_netpoll_cleanup)
2030 slave_dev->netdev_ops->ndo_netpoll_cleanup(slave_dev);
2031 else
2032 slave_dev->npinfo = NULL;
2033 #endif
2034
2035 /* close slave before restoring its mac address */
2036 dev_close(slave_dev);
2037
2038 if (bond->params.fail_over_mac != BOND_FOM_ACTIVE) {
2039 /* restore original ("permanent") mac address */
2040 memcpy(addr.sa_data, slave->perm_hwaddr, ETH_ALEN);
2041 addr.sa_family = slave_dev->type;
2042 dev_set_mac_address(slave_dev, &addr);
2043 }
2044
2045 dev_set_mtu(slave_dev, slave->original_mtu);
2046
2047 slave_dev->priv_flags &= ~(IFF_MASTER_8023AD | IFF_MASTER_ALB |
2048 IFF_SLAVE_INACTIVE | IFF_BONDING |
2049 IFF_SLAVE_NEEDARP);
2050
2051 kfree(slave);
2052
2053 return 0; /* deletion OK */
2054 }
2055
2056 /*
2057 * First release a slave and than destroy the bond if no more slaves are left.
2058 * Must be under rtnl_lock when this function is called.
2059 */
2060 int bond_release_and_destroy(struct net_device *bond_dev,
2061 struct net_device *slave_dev)
2062 {
2063 struct bonding *bond = netdev_priv(bond_dev);
2064 int ret;
2065
2066 ret = bond_release(bond_dev, slave_dev);
2067 if ((ret == 0) && (bond->slave_cnt == 0)) {
2068 pr_info("%s: destroying bond %s.\n",
2069 bond_dev->name, bond_dev->name);
2070 unregister_netdevice(bond_dev);
2071 }
2072 return ret;
2073 }
2074
2075 /*
2076 * This function releases all slaves.
2077 */
2078 static int bond_release_all(struct net_device *bond_dev)
2079 {
2080 struct bonding *bond = netdev_priv(bond_dev);
2081 struct slave *slave;
2082 struct net_device *slave_dev;
2083 struct sockaddr addr;
2084
2085 write_lock_bh(&bond->lock);
2086
2087 netif_carrier_off(bond_dev);
2088
2089 if (bond->slave_cnt == 0)
2090 goto out;
2091
2092 bond->current_arp_slave = NULL;
2093 bond->primary_slave = NULL;
2094 bond_change_active_slave(bond, NULL);
2095
2096 while ((slave = bond->first_slave) != NULL) {
2097 /* Inform AD package of unbinding of slave
2098 * before slave is detached from the list.
2099 */
2100 if (bond->params.mode == BOND_MODE_8023AD)
2101 bond_3ad_unbind_slave(slave);
2102
2103 slave_dev = slave->dev;
2104 bond_detach_slave(bond, slave);
2105
2106 /* now that the slave is detached, unlock and perform
2107 * all the undo steps that should not be called from
2108 * within a lock.
2109 */
2110 write_unlock_bh(&bond->lock);
2111
2112 if (bond_is_lb(bond)) {
2113 /* must be called only after the slave
2114 * has been detached from the list
2115 */
2116 bond_alb_deinit_slave(bond, slave);
2117 }
2118
2119 bond_compute_features(bond);
2120
2121 bond_destroy_slave_symlinks(bond_dev, slave_dev);
2122 bond_del_vlans_from_slave(bond, slave_dev);
2123
2124 /* If the mode USES_PRIMARY, then we should only remove its
2125 * promisc and mc settings if it was the curr_active_slave, but that was
2126 * already taken care of above when we detached the slave
2127 */
2128 if (!USES_PRIMARY(bond->params.mode)) {
2129 /* unset promiscuity level from slave */
2130 if (bond_dev->flags & IFF_PROMISC)
2131 dev_set_promiscuity(slave_dev, -1);
2132
2133 /* unset allmulti level from slave */
2134 if (bond_dev->flags & IFF_ALLMULTI)
2135 dev_set_allmulti(slave_dev, -1);
2136
2137 /* flush master's mc_list from slave */
2138 netif_addr_lock_bh(bond_dev);
2139 bond_mc_list_flush(bond_dev, slave_dev);
2140 netif_addr_unlock_bh(bond_dev);
2141 }
2142
2143 netdev_set_master(slave_dev, NULL);
2144
2145 /* close slave before restoring its mac address */
2146 dev_close(slave_dev);
2147
2148 if (!bond->params.fail_over_mac) {
2149 /* restore original ("permanent") mac address*/
2150 memcpy(addr.sa_data, slave->perm_hwaddr, ETH_ALEN);
2151 addr.sa_family = slave_dev->type;
2152 dev_set_mac_address(slave_dev, &addr);
2153 }
2154
2155 slave_dev->priv_flags &= ~(IFF_MASTER_8023AD | IFF_MASTER_ALB |
2156 IFF_SLAVE_INACTIVE);
2157
2158 kfree(slave);
2159
2160 /* re-acquire the lock before getting the next slave */
2161 write_lock_bh(&bond->lock);
2162 }
2163
2164 /* zero the mac address of the master so it will be
2165 * set by the application to the mac address of the
2166 * first slave
2167 */
2168 memset(bond_dev->dev_addr, 0, bond_dev->addr_len);
2169
2170 if (!bond->vlgrp) {
2171 bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
2172 } else {
2173 pr_warning("%s: Warning: clearing HW address of %s while it still has VLANs.\n",
2174 bond_dev->name, bond_dev->name);
2175 pr_warning("%s: When re-adding slaves, make sure the bond's HW address matches its VLANs'.\n",
2176 bond_dev->name);
2177 }
2178
2179 pr_info("%s: released all slaves\n", bond_dev->name);
2180
2181 out:
2182 write_unlock_bh(&bond->lock);
2183
2184 return 0;
2185 }
2186
2187 /*
2188 * This function changes the active slave to slave <slave_dev>.
2189 * It returns -EINVAL in the following cases.
2190 * - <slave_dev> is not found in the list.
2191 * - There is not active slave now.
2192 * - <slave_dev> is already active.
2193 * - The link state of <slave_dev> is not BOND_LINK_UP.
2194 * - <slave_dev> is not running.
2195 * In these cases, this function does nothing.
2196 * In the other cases, current_slave pointer is changed and 0 is returned.
2197 */
2198 static int bond_ioctl_change_active(struct net_device *bond_dev, struct net_device *slave_dev)
2199 {
2200 struct bonding *bond = netdev_priv(bond_dev);
2201 struct slave *old_active = NULL;
2202 struct slave *new_active = NULL;
2203 int res = 0;
2204
2205 if (!USES_PRIMARY(bond->params.mode))
2206 return -EINVAL;
2207
2208 /* Verify that master_dev is indeed the master of slave_dev */
2209 if (!(slave_dev->flags & IFF_SLAVE) || (slave_dev->master != bond_dev))
2210 return -EINVAL;
2211
2212 read_lock(&bond->lock);
2213
2214 read_lock(&bond->curr_slave_lock);
2215 old_active = bond->curr_active_slave;
2216 read_unlock(&bond->curr_slave_lock);
2217
2218 new_active = bond_get_slave_by_dev(bond, slave_dev);
2219
2220 /*
2221 * Changing to the current active: do nothing; return success.
2222 */
2223 if (new_active && (new_active == old_active)) {
2224 read_unlock(&bond->lock);
2225 return 0;
2226 }
2227
2228 if ((new_active) &&
2229 (old_active) &&
2230 (new_active->link == BOND_LINK_UP) &&
2231 IS_UP(new_active->dev)) {
2232 write_lock_bh(&bond->curr_slave_lock);
2233 bond_change_active_slave(bond, new_active);
2234 write_unlock_bh(&bond->curr_slave_lock);
2235 } else
2236 res = -EINVAL;
2237
2238 read_unlock(&bond->lock);
2239
2240 return res;
2241 }
2242
2243 static int bond_info_query(struct net_device *bond_dev, struct ifbond *info)
2244 {
2245 struct bonding *bond = netdev_priv(bond_dev);
2246
2247 info->bond_mode = bond->params.mode;
2248 info->miimon = bond->params.miimon;
2249
2250 read_lock(&bond->lock);
2251 info->num_slaves = bond->slave_cnt;
2252 read_unlock(&bond->lock);
2253
2254 return 0;
2255 }
2256
2257 static int bond_slave_info_query(struct net_device *bond_dev, struct ifslave *info)
2258 {
2259 struct bonding *bond = netdev_priv(bond_dev);
2260 struct slave *slave;
2261 int i, res = -ENODEV;
2262
2263 read_lock(&bond->lock);
2264
2265 bond_for_each_slave(bond, slave, i) {
2266 if (i == (int)info->slave_id) {
2267 res = 0;
2268 strcpy(info->slave_name, slave->dev->name);
2269 info->link = slave->link;
2270 info->state = slave->state;
2271 info->link_failure_count = slave->link_failure_count;
2272 break;
2273 }
2274 }
2275
2276 read_unlock(&bond->lock);
2277
2278 return res;
2279 }
2280
2281 /*-------------------------------- Monitoring -------------------------------*/
2282
2283
2284 static int bond_miimon_inspect(struct bonding *bond)
2285 {
2286 struct slave *slave;
2287 int i, link_state, commit = 0;
2288 bool ignore_updelay;
2289
2290 ignore_updelay = !bond->curr_active_slave ? true : false;
2291
2292 bond_for_each_slave(bond, slave, i) {
2293 slave->new_link = BOND_LINK_NOCHANGE;
2294
2295 link_state = bond_check_dev_link(bond, slave->dev, 0);
2296
2297 switch (slave->link) {
2298 case BOND_LINK_UP:
2299 if (link_state)
2300 continue;
2301
2302 slave->link = BOND_LINK_FAIL;
2303 slave->delay = bond->params.downdelay;
2304 if (slave->delay) {
2305 pr_info("%s: link status down for %sinterface %s, disabling it in %d ms.\n",
2306 bond->dev->name,
2307 (bond->params.mode ==
2308 BOND_MODE_ACTIVEBACKUP) ?
2309 ((slave->state == BOND_STATE_ACTIVE) ?
2310 "active " : "backup ") : "",
2311 slave->dev->name,
2312 bond->params.downdelay * bond->params.miimon);
2313 }
2314 /*FALLTHRU*/
2315 case BOND_LINK_FAIL:
2316 if (link_state) {
2317 /*
2318 * recovered before downdelay expired
2319 */
2320 slave->link = BOND_LINK_UP;
2321 slave->jiffies = jiffies;
2322 pr_info("%s: link status up again after %d ms for interface %s.\n",
2323 bond->dev->name,
2324 (bond->params.downdelay - slave->delay) *
2325 bond->params.miimon,
2326 slave->dev->name);
2327 continue;
2328 }
2329
2330 if (slave->delay <= 0) {
2331 slave->new_link = BOND_LINK_DOWN;
2332 commit++;
2333 continue;
2334 }
2335
2336 slave->delay--;
2337 break;
2338
2339 case BOND_LINK_DOWN:
2340 if (!link_state)
2341 continue;
2342
2343 slave->link = BOND_LINK_BACK;
2344 slave->delay = bond->params.updelay;
2345
2346 if (slave->delay) {
2347 pr_info("%s: link status up for interface %s, enabling it in %d ms.\n",
2348 bond->dev->name, slave->dev->name,
2349 ignore_updelay ? 0 :
2350 bond->params.updelay *
2351 bond->params.miimon);
2352 }
2353 /*FALLTHRU*/
2354 case BOND_LINK_BACK:
2355 if (!link_state) {
2356 slave->link = BOND_LINK_DOWN;
2357 pr_info("%s: link status down again after %d ms for interface %s.\n",
2358 bond->dev->name,
2359 (bond->params.updelay - slave->delay) *
2360 bond->params.miimon,
2361 slave->dev->name);
2362
2363 continue;
2364 }
2365
2366 if (ignore_updelay)
2367 slave->delay = 0;
2368
2369 if (slave->delay <= 0) {
2370 slave->new_link = BOND_LINK_UP;
2371 commit++;
2372 ignore_updelay = false;
2373 continue;
2374 }
2375
2376 slave->delay--;
2377 break;
2378 }
2379 }
2380
2381 return commit;
2382 }
2383
2384 static void bond_miimon_commit(struct bonding *bond)
2385 {
2386 struct slave *slave;
2387 int i;
2388
2389 bond_for_each_slave(bond, slave, i) {
2390 switch (slave->new_link) {
2391 case BOND_LINK_NOCHANGE:
2392 continue;
2393
2394 case BOND_LINK_UP:
2395 slave->link = BOND_LINK_UP;
2396 slave->jiffies = jiffies;
2397
2398 if (bond->params.mode == BOND_MODE_8023AD) {
2399 /* prevent it from being the active one */
2400 slave->state = BOND_STATE_BACKUP;
2401 } else if (bond->params.mode != BOND_MODE_ACTIVEBACKUP) {
2402 /* make it immediately active */
2403 slave->state = BOND_STATE_ACTIVE;
2404 } else if (slave != bond->primary_slave) {
2405 /* prevent it from being the active one */
2406 slave->state = BOND_STATE_BACKUP;
2407 }
2408
2409 pr_info("%s: link status definitely up for interface %s, %d Mbps %s duplex.\n",
2410 bond->dev->name, slave->dev->name,
2411 slave->speed, slave->duplex ? "full" : "half");
2412
2413 /* notify ad that the link status has changed */
2414 if (bond->params.mode == BOND_MODE_8023AD)
2415 bond_3ad_handle_link_change(slave, BOND_LINK_UP);
2416
2417 if (bond_is_lb(bond))
2418 bond_alb_handle_link_change(bond, slave,
2419 BOND_LINK_UP);
2420
2421 if (!bond->curr_active_slave ||
2422 (slave == bond->primary_slave))
2423 goto do_failover;
2424
2425 continue;
2426
2427 case BOND_LINK_DOWN:
2428 if (slave->link_failure_count < UINT_MAX)
2429 slave->link_failure_count++;
2430
2431 slave->link = BOND_LINK_DOWN;
2432
2433 if (bond->params.mode == BOND_MODE_ACTIVEBACKUP ||
2434 bond->params.mode == BOND_MODE_8023AD)
2435 bond_set_slave_inactive_flags(slave);
2436
2437 pr_info("%s: link status definitely down for interface %s, disabling it\n",
2438 bond->dev->name, slave->dev->name);
2439
2440 if (bond->params.mode == BOND_MODE_8023AD)
2441 bond_3ad_handle_link_change(slave,
2442 BOND_LINK_DOWN);
2443
2444 if (bond_is_lb(bond))
2445 bond_alb_handle_link_change(bond, slave,
2446 BOND_LINK_DOWN);
2447
2448 if (slave == bond->curr_active_slave)
2449 goto do_failover;
2450
2451 continue;
2452
2453 default:
2454 pr_err("%s: invalid new link %d on slave %s\n",
2455 bond->dev->name, slave->new_link,
2456 slave->dev->name);
2457 slave->new_link = BOND_LINK_NOCHANGE;
2458
2459 continue;
2460 }
2461
2462 do_failover:
2463 ASSERT_RTNL();
2464 write_lock_bh(&bond->curr_slave_lock);
2465 bond_select_active_slave(bond);
2466 write_unlock_bh(&bond->curr_slave_lock);
2467 }
2468
2469 bond_set_carrier(bond);
2470 }
2471
2472 /*
2473 * bond_mii_monitor
2474 *
2475 * Really a wrapper that splits the mii monitor into two phases: an
2476 * inspection, then (if inspection indicates something needs to be done)
2477 * an acquisition of appropriate locks followed by a commit phase to
2478 * implement whatever link state changes are indicated.
2479 */
2480 void bond_mii_monitor(struct work_struct *work)
2481 {
2482 struct bonding *bond = container_of(work, struct bonding,
2483 mii_work.work);
2484
2485 read_lock(&bond->lock);
2486 if (bond->kill_timers)
2487 goto out;
2488
2489 if (bond->slave_cnt == 0)
2490 goto re_arm;
2491
2492 if (bond->send_grat_arp) {
2493 read_lock(&bond->curr_slave_lock);
2494 bond_send_gratuitous_arp(bond);
2495 read_unlock(&bond->curr_slave_lock);
2496 }
2497
2498 if (bond->send_unsol_na) {
2499 read_lock(&bond->curr_slave_lock);
2500 bond_send_unsolicited_na(bond);
2501 read_unlock(&bond->curr_slave_lock);
2502 }
2503
2504 if (bond_miimon_inspect(bond)) {
2505 read_unlock(&bond->lock);
2506 rtnl_lock();
2507 read_lock(&bond->lock);
2508
2509 bond_miimon_commit(bond);
2510
2511 read_unlock(&bond->lock);
2512 rtnl_unlock(); /* might sleep, hold no other locks */
2513 read_lock(&bond->lock);
2514 }
2515
2516 re_arm:
2517 if (bond->params.miimon)
2518 queue_delayed_work(bond->wq, &bond->mii_work,
2519 msecs_to_jiffies(bond->params.miimon));
2520 out:
2521 read_unlock(&bond->lock);
2522 }
2523
2524 static __be32 bond_glean_dev_ip(struct net_device *dev)
2525 {
2526 struct in_device *idev;
2527 struct in_ifaddr *ifa;
2528 __be32 addr = 0;
2529
2530 if (!dev)
2531 return 0;
2532
2533 rcu_read_lock();
2534 idev = __in_dev_get_rcu(dev);
2535 if (!idev)
2536 goto out;
2537
2538 ifa = idev->ifa_list;
2539 if (!ifa)
2540 goto out;
2541
2542 addr = ifa->ifa_local;
2543 out:
2544 rcu_read_unlock();
2545 return addr;
2546 }
2547
2548 static int bond_has_this_ip(struct bonding *bond, __be32 ip)
2549 {
2550 struct vlan_entry *vlan;
2551
2552 if (ip == bond->master_ip)
2553 return 1;
2554
2555 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
2556 if (ip == vlan->vlan_ip)
2557 return 1;
2558 }
2559
2560 return 0;
2561 }
2562
2563 /*
2564 * We go to the (large) trouble of VLAN tagging ARP frames because
2565 * switches in VLAN mode (especially if ports are configured as
2566 * "native" to a VLAN) might not pass non-tagged frames.
2567 */
2568 static void bond_arp_send(struct net_device *slave_dev, int arp_op, __be32 dest_ip, __be32 src_ip, unsigned short vlan_id)
2569 {
2570 struct sk_buff *skb;
2571
2572 pr_debug("arp %d on slave %s: dst %x src %x vid %d\n", arp_op,
2573 slave_dev->name, dest_ip, src_ip, vlan_id);
2574
2575 skb = arp_create(arp_op, ETH_P_ARP, dest_ip, slave_dev, src_ip,
2576 NULL, slave_dev->dev_addr, NULL);
2577
2578 if (!skb) {
2579 pr_err("ARP packet allocation failed\n");
2580 return;
2581 }
2582 if (vlan_id) {
2583 skb = vlan_put_tag(skb, vlan_id);
2584 if (!skb) {
2585 pr_err("failed to insert VLAN tag\n");
2586 return;
2587 }
2588 }
2589 arp_xmit(skb);
2590 }
2591
2592
2593 static void bond_arp_send_all(struct bonding *bond, struct slave *slave)
2594 {
2595 int i, vlan_id, rv;
2596 __be32 *targets = bond->params.arp_targets;
2597 struct vlan_entry *vlan;
2598 struct net_device *vlan_dev;
2599 struct flowi fl;
2600 struct rtable *rt;
2601
2602 for (i = 0; (i < BOND_MAX_ARP_TARGETS); i++) {
2603 if (!targets[i])
2604 break;
2605 pr_debug("basa: target %x\n", targets[i]);
2606 if (!bond->vlgrp) {
2607 pr_debug("basa: empty vlan: arp_send\n");
2608 bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2609 bond->master_ip, 0);
2610 continue;
2611 }
2612
2613 /*
2614 * If VLANs are configured, we do a route lookup to
2615 * determine which VLAN interface would be used, so we
2616 * can tag the ARP with the proper VLAN tag.
2617 */
2618 memset(&fl, 0, sizeof(fl));
2619 fl.fl4_dst = targets[i];
2620 fl.fl4_tos = RTO_ONLINK;
2621
2622 rv = ip_route_output_key(dev_net(bond->dev), &rt, &fl);
2623 if (rv) {
2624 if (net_ratelimit()) {
2625 pr_warning("%s: no route to arp_ip_target %pI4\n",
2626 bond->dev->name, &fl.fl4_dst);
2627 }
2628 continue;
2629 }
2630
2631 /*
2632 * This target is not on a VLAN
2633 */
2634 if (rt->dst.dev == bond->dev) {
2635 ip_rt_put(rt);
2636 pr_debug("basa: rtdev == bond->dev: arp_send\n");
2637 bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2638 bond->master_ip, 0);
2639 continue;
2640 }
2641
2642 vlan_id = 0;
2643 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
2644 vlan_dev = vlan_group_get_device(bond->vlgrp, vlan->vlan_id);
2645 if (vlan_dev == rt->dst.dev) {
2646 vlan_id = vlan->vlan_id;
2647 pr_debug("basa: vlan match on %s %d\n",
2648 vlan_dev->name, vlan_id);
2649 break;
2650 }
2651 }
2652
2653 if (vlan_id) {
2654 ip_rt_put(rt);
2655 bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2656 vlan->vlan_ip, vlan_id);
2657 continue;
2658 }
2659
2660 if (net_ratelimit()) {
2661 pr_warning("%s: no path to arp_ip_target %pI4 via rt.dev %s\n",
2662 bond->dev->name, &fl.fl4_dst,
2663 rt->dst.dev ? rt->dst.dev->name : "NULL");
2664 }
2665 ip_rt_put(rt);
2666 }
2667 }
2668
2669 /*
2670 * Kick out a gratuitous ARP for an IP on the bonding master plus one
2671 * for each VLAN above us.
2672 *
2673 * Caller must hold curr_slave_lock for read or better
2674 */
2675 static void bond_send_gratuitous_arp(struct bonding *bond)
2676 {
2677 struct slave *slave = bond->curr_active_slave;
2678 struct vlan_entry *vlan;
2679 struct net_device *vlan_dev;
2680
2681 pr_debug("bond_send_grat_arp: bond %s slave %s\n",
2682 bond->dev->name, slave ? slave->dev->name : "NULL");
2683
2684 if (!slave || !bond->send_grat_arp ||
2685 test_bit(__LINK_STATE_LINKWATCH_PENDING, &slave->dev->state))
2686 return;
2687
2688 bond->send_grat_arp--;
2689
2690 if (bond->master_ip) {
2691 bond_arp_send(slave->dev, ARPOP_REPLY, bond->master_ip,
2692 bond->master_ip, 0);
2693 }
2694
2695 if (!bond->vlgrp)
2696 return;
2697
2698 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
2699 vlan_dev = vlan_group_get_device(bond->vlgrp, vlan->vlan_id);
2700 if (vlan->vlan_ip) {
2701 bond_arp_send(slave->dev, ARPOP_REPLY, vlan->vlan_ip,
2702 vlan->vlan_ip, vlan->vlan_id);
2703 }
2704 }
2705 }
2706
2707 static void bond_validate_arp(struct bonding *bond, struct slave *slave, __be32 sip, __be32 tip)
2708 {
2709 int i;
2710 __be32 *targets = bond->params.arp_targets;
2711
2712 for (i = 0; (i < BOND_MAX_ARP_TARGETS) && targets[i]; i++) {
2713 pr_debug("bva: sip %pI4 tip %pI4 t[%d] %pI4 bhti(tip) %d\n",
2714 &sip, &tip, i, &targets[i],
2715 bond_has_this_ip(bond, tip));
2716 if (sip == targets[i]) {
2717 if (bond_has_this_ip(bond, tip))
2718 slave->last_arp_rx = jiffies;
2719 return;
2720 }
2721 }
2722 }
2723
2724 static int bond_arp_rcv(struct sk_buff *skb, struct net_device *dev, struct packet_type *pt, struct net_device *orig_dev)
2725 {
2726 struct arphdr *arp;
2727 struct slave *slave;
2728 struct bonding *bond;
2729 unsigned char *arp_ptr;
2730 __be32 sip, tip;
2731
2732 if (dev->priv_flags & IFF_802_1Q_VLAN) {
2733 /*
2734 * When using VLANS and bonding, dev and oriv_dev may be
2735 * incorrect if the physical interface supports VLAN
2736 * acceleration. With this change ARP validation now
2737 * works for hosts only reachable on the VLAN interface.
2738 */
2739 dev = vlan_dev_real_dev(dev);
2740 orig_dev = dev_get_by_index_rcu(dev_net(skb->dev),skb->skb_iif);
2741 }
2742
2743 if (!(dev->priv_flags & IFF_BONDING) || !(dev->flags & IFF_MASTER))
2744 goto out;
2745
2746 bond = netdev_priv(dev);
2747 read_lock(&bond->lock);
2748
2749 pr_debug("bond_arp_rcv: bond %s skb->dev %s orig_dev %s\n",
2750 bond->dev->name, skb->dev ? skb->dev->name : "NULL",
2751 orig_dev ? orig_dev->name : "NULL");
2752
2753 slave = bond_get_slave_by_dev(bond, orig_dev);
2754 if (!slave || !slave_do_arp_validate(bond, slave))
2755 goto out_unlock;
2756
2757 if (!pskb_may_pull(skb, arp_hdr_len(dev)))
2758 goto out_unlock;
2759
2760 arp = arp_hdr(skb);
2761 if (arp->ar_hln != dev->addr_len ||
2762 skb->pkt_type == PACKET_OTHERHOST ||
2763 skb->pkt_type == PACKET_LOOPBACK ||
2764 arp->ar_hrd != htons(ARPHRD_ETHER) ||
2765 arp->ar_pro != htons(ETH_P_IP) ||
2766 arp->ar_pln != 4)
2767 goto out_unlock;
2768
2769 arp_ptr = (unsigned char *)(arp + 1);
2770 arp_ptr += dev->addr_len;
2771 memcpy(&sip, arp_ptr, 4);
2772 arp_ptr += 4 + dev->addr_len;
2773 memcpy(&tip, arp_ptr, 4);
2774
2775 pr_debug("bond_arp_rcv: %s %s/%d av %d sv %d sip %pI4 tip %pI4\n",
2776 bond->dev->name, slave->dev->name, slave->state,
2777 bond->params.arp_validate, slave_do_arp_validate(bond, slave),
2778 &sip, &tip);
2779
2780 /*
2781 * Backup slaves won't see the ARP reply, but do come through
2782 * here for each ARP probe (so we swap the sip/tip to validate
2783 * the probe). In a "redundant switch, common router" type of
2784 * configuration, the ARP probe will (hopefully) travel from
2785 * the active, through one switch, the router, then the other
2786 * switch before reaching the backup.
2787 */
2788 if (slave->state == BOND_STATE_ACTIVE)
2789 bond_validate_arp(bond, slave, sip, tip);
2790 else
2791 bond_validate_arp(bond, slave, tip, sip);
2792
2793 out_unlock:
2794 read_unlock(&bond->lock);
2795 out:
2796 dev_kfree_skb(skb);
2797 return NET_RX_SUCCESS;
2798 }
2799
2800 /*
2801 * this function is called regularly to monitor each slave's link
2802 * ensuring that traffic is being sent and received when arp monitoring
2803 * is used in load-balancing mode. if the adapter has been dormant, then an
2804 * arp is transmitted to generate traffic. see activebackup_arp_monitor for
2805 * arp monitoring in active backup mode.
2806 */
2807 void bond_loadbalance_arp_mon(struct work_struct *work)
2808 {
2809 struct bonding *bond = container_of(work, struct bonding,
2810 arp_work.work);
2811 struct slave *slave, *oldcurrent;
2812 int do_failover = 0;
2813 int delta_in_ticks;
2814 int i;
2815
2816 read_lock(&bond->lock);
2817
2818 delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
2819
2820 if (bond->kill_timers)
2821 goto out;
2822
2823 if (bond->slave_cnt == 0)
2824 goto re_arm;
2825
2826 read_lock(&bond->curr_slave_lock);
2827 oldcurrent = bond->curr_active_slave;
2828 read_unlock(&bond->curr_slave_lock);
2829
2830 /* see if any of the previous devices are up now (i.e. they have
2831 * xmt and rcv traffic). the curr_active_slave does not come into
2832 * the picture unless it is null. also, slave->jiffies is not needed
2833 * here because we send an arp on each slave and give a slave as
2834 * long as it needs to get the tx/rx within the delta.
2835 * TODO: what about up/down delay in arp mode? it wasn't here before
2836 * so it can wait
2837 */
2838 bond_for_each_slave(bond, slave, i) {
2839 unsigned long trans_start = dev_trans_start(slave->dev);
2840
2841 if (slave->link != BOND_LINK_UP) {
2842 if (time_in_range(jiffies,
2843 trans_start - delta_in_ticks,
2844 trans_start + delta_in_ticks) &&
2845 time_in_range(jiffies,
2846 slave->dev->last_rx - delta_in_ticks,
2847 slave->dev->last_rx + delta_in_ticks)) {
2848
2849 slave->link = BOND_LINK_UP;
2850 slave->state = BOND_STATE_ACTIVE;
2851
2852 /* primary_slave has no meaning in round-robin
2853 * mode. the window of a slave being up and
2854 * curr_active_slave being null after enslaving
2855 * is closed.
2856 */
2857 if (!oldcurrent) {
2858 pr_info("%s: link status definitely up for interface %s, ",
2859 bond->dev->name,
2860 slave->dev->name);
2861 do_failover = 1;
2862 } else {
2863 pr_info("%s: interface %s is now up\n",
2864 bond->dev->name,
2865 slave->dev->name);
2866 }
2867 }
2868 } else {
2869 /* slave->link == BOND_LINK_UP */
2870
2871 /* not all switches will respond to an arp request
2872 * when the source ip is 0, so don't take the link down
2873 * if we don't know our ip yet
2874 */
2875 if (!time_in_range(jiffies,
2876 trans_start - delta_in_ticks,
2877 trans_start + 2 * delta_in_ticks) ||
2878 !time_in_range(jiffies,
2879 slave->dev->last_rx - delta_in_ticks,
2880 slave->dev->last_rx + 2 * delta_in_ticks)) {
2881
2882 slave->link = BOND_LINK_DOWN;
2883 slave->state = BOND_STATE_BACKUP;
2884
2885 if (slave->link_failure_count < UINT_MAX)
2886 slave->link_failure_count++;
2887
2888 pr_info("%s: interface %s is now down.\n",
2889 bond->dev->name,
2890 slave->dev->name);
2891
2892 if (slave == oldcurrent)
2893 do_failover = 1;
2894 }
2895 }
2896
2897 /* note: if switch is in round-robin mode, all links
2898 * must tx arp to ensure all links rx an arp - otherwise
2899 * links may oscillate or not come up at all; if switch is
2900 * in something like xor mode, there is nothing we can
2901 * do - all replies will be rx'ed on same link causing slaves
2902 * to be unstable during low/no traffic periods
2903 */
2904 if (IS_UP(slave->dev))
2905 bond_arp_send_all(bond, slave);
2906 }
2907
2908 if (do_failover) {
2909 write_lock_bh(&bond->curr_slave_lock);
2910
2911 bond_select_active_slave(bond);
2912
2913 write_unlock_bh(&bond->curr_slave_lock);
2914 }
2915
2916 re_arm:
2917 if (bond->params.arp_interval)
2918 queue_delayed_work(bond->wq, &bond->arp_work, delta_in_ticks);
2919 out:
2920 read_unlock(&bond->lock);
2921 }
2922
2923 /*
2924 * Called to inspect slaves for active-backup mode ARP monitor link state
2925 * changes. Sets new_link in slaves to specify what action should take
2926 * place for the slave. Returns 0 if no changes are found, >0 if changes
2927 * to link states must be committed.
2928 *
2929 * Called with bond->lock held for read.
2930 */
2931 static int bond_ab_arp_inspect(struct bonding *bond, int delta_in_ticks)
2932 {
2933 struct slave *slave;
2934 int i, commit = 0;
2935 unsigned long trans_start;
2936
2937 bond_for_each_slave(bond, slave, i) {
2938 slave->new_link = BOND_LINK_NOCHANGE;
2939
2940 if (slave->link != BOND_LINK_UP) {
2941 if (time_in_range(jiffies,
2942 slave_last_rx(bond, slave) - delta_in_ticks,
2943 slave_last_rx(bond, slave) + delta_in_ticks)) {
2944
2945 slave->new_link = BOND_LINK_UP;
2946 commit++;
2947 }
2948
2949 continue;
2950 }
2951
2952 /*
2953 * Give slaves 2*delta after being enslaved or made
2954 * active. This avoids bouncing, as the last receive
2955 * times need a full ARP monitor cycle to be updated.
2956 */
2957 if (time_in_range(jiffies,
2958 slave->jiffies - delta_in_ticks,
2959 slave->jiffies + 2 * delta_in_ticks))
2960 continue;
2961
2962 /*
2963 * Backup slave is down if:
2964 * - No current_arp_slave AND
2965 * - more than 3*delta since last receive AND
2966 * - the bond has an IP address
2967 *
2968 * Note: a non-null current_arp_slave indicates
2969 * the curr_active_slave went down and we are
2970 * searching for a new one; under this condition
2971 * we only take the curr_active_slave down - this
2972 * gives each slave a chance to tx/rx traffic
2973 * before being taken out
2974 */
2975 if (slave->state == BOND_STATE_BACKUP &&
2976 !bond->current_arp_slave &&
2977 !time_in_range(jiffies,
2978 slave_last_rx(bond, slave) - delta_in_ticks,
2979 slave_last_rx(bond, slave) + 3 * delta_in_ticks)) {
2980
2981 slave->new_link = BOND_LINK_DOWN;
2982 commit++;
2983 }
2984
2985 /*
2986 * Active slave is down if:
2987 * - more than 2*delta since transmitting OR
2988 * - (more than 2*delta since receive AND
2989 * the bond has an IP address)
2990 */
2991 trans_start = dev_trans_start(slave->dev);
2992 if ((slave->state == BOND_STATE_ACTIVE) &&
2993 (!time_in_range(jiffies,
2994 trans_start - delta_in_ticks,
2995 trans_start + 2 * delta_in_ticks) ||
2996 !time_in_range(jiffies,
2997 slave_last_rx(bond, slave) - delta_in_ticks,
2998 slave_last_rx(bond, slave) + 2 * delta_in_ticks))) {
2999
3000 slave->new_link = BOND_LINK_DOWN;
3001 commit++;
3002 }
3003 }
3004
3005 return commit;
3006 }
3007
3008 /*
3009 * Called to commit link state changes noted by inspection step of
3010 * active-backup mode ARP monitor.
3011 *
3012 * Called with RTNL and bond->lock for read.
3013 */
3014 static void bond_ab_arp_commit(struct bonding *bond, int delta_in_ticks)
3015 {
3016 struct slave *slave;
3017 int i;
3018 unsigned long trans_start;
3019
3020 bond_for_each_slave(bond, slave, i) {
3021 switch (slave->new_link) {
3022 case BOND_LINK_NOCHANGE:
3023 continue;
3024
3025 case BOND_LINK_UP:
3026 trans_start = dev_trans_start(slave->dev);
3027 if ((!bond->curr_active_slave &&
3028 time_in_range(jiffies,
3029 trans_start - delta_in_ticks,
3030 trans_start + delta_in_ticks)) ||
3031 bond->curr_active_slave != slave) {
3032 slave->link = BOND_LINK_UP;
3033 bond->current_arp_slave = NULL;
3034
3035 pr_info("%s: link status definitely up for interface %s.\n",
3036 bond->dev->name, slave->dev->name);
3037
3038 if (!bond->curr_active_slave ||
3039 (slave == bond->primary_slave))
3040 goto do_failover;
3041
3042 }
3043
3044 continue;
3045
3046 case BOND_LINK_DOWN:
3047 if (slave->link_failure_count < UINT_MAX)
3048 slave->link_failure_count++;
3049
3050 slave->link = BOND_LINK_DOWN;
3051 bond_set_slave_inactive_flags(slave);
3052
3053 pr_info("%s: link status definitely down for interface %s, disabling it\n",
3054 bond->dev->name, slave->dev->name);
3055
3056 if (slave == bond->curr_active_slave) {
3057 bond->current_arp_slave = NULL;
3058 goto do_failover;
3059 }
3060
3061 continue;
3062
3063 default:
3064 pr_err("%s: impossible: new_link %d on slave %s\n",
3065 bond->dev->name, slave->new_link,
3066 slave->dev->name);
3067 continue;
3068 }
3069
3070 do_failover:
3071 ASSERT_RTNL();
3072 write_lock_bh(&bond->curr_slave_lock);
3073 bond_select_active_slave(bond);
3074 write_unlock_bh(&bond->curr_slave_lock);
3075 }
3076
3077 bond_set_carrier(bond);
3078 }
3079
3080 /*
3081 * Send ARP probes for active-backup mode ARP monitor.
3082 *
3083 * Called with bond->lock held for read.
3084 */
3085 static void bond_ab_arp_probe(struct bonding *bond)
3086 {
3087 struct slave *slave;
3088 int i;
3089
3090 read_lock(&bond->curr_slave_lock);
3091
3092 if (bond->current_arp_slave && bond->curr_active_slave)
3093 pr_info("PROBE: c_arp %s && cas %s BAD\n",
3094 bond->current_arp_slave->dev->name,
3095 bond->curr_active_slave->dev->name);
3096
3097 if (bond->curr_active_slave) {
3098 bond_arp_send_all(bond, bond->curr_active_slave);
3099 read_unlock(&bond->curr_slave_lock);
3100 return;
3101 }
3102
3103 read_unlock(&bond->curr_slave_lock);
3104
3105 /* if we don't have a curr_active_slave, search for the next available
3106 * backup slave from the current_arp_slave and make it the candidate
3107 * for becoming the curr_active_slave
3108 */
3109
3110 if (!bond->current_arp_slave) {
3111 bond->current_arp_slave = bond->first_slave;
3112 if (!bond->current_arp_slave)
3113 return;
3114 }
3115
3116 bond_set_slave_inactive_flags(bond->current_arp_slave);
3117
3118 /* search for next candidate */
3119 bond_for_each_slave_from(bond, slave, i, bond->current_arp_slave->next) {
3120 if (IS_UP(slave->dev)) {
3121 slave->link = BOND_LINK_BACK;
3122 bond_set_slave_active_flags(slave);
3123 bond_arp_send_all(bond, slave);
3124 slave->jiffies = jiffies;
3125 bond->current_arp_slave = slave;
3126 break;
3127 }
3128
3129 /* if the link state is up at this point, we
3130 * mark it down - this can happen if we have
3131 * simultaneous link failures and
3132 * reselect_active_interface doesn't make this
3133 * one the current slave so it is still marked
3134 * up when it is actually down
3135 */
3136 if (slave->link == BOND_LINK_UP) {
3137 slave->link = BOND_LINK_DOWN;
3138 if (slave->link_failure_count < UINT_MAX)
3139 slave->link_failure_count++;
3140
3141 bond_set_slave_inactive_flags(slave);
3142
3143 pr_info("%s: backup interface %s is now down.\n",
3144 bond->dev->name, slave->dev->name);
3145 }
3146 }
3147 }
3148
3149 void bond_activebackup_arp_mon(struct work_struct *work)
3150 {
3151 struct bonding *bond = container_of(work, struct bonding,
3152 arp_work.work);
3153 int delta_in_ticks;
3154
3155 read_lock(&bond->lock);
3156
3157 if (bond->kill_timers)
3158 goto out;
3159
3160 delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
3161
3162 if (bond->slave_cnt == 0)
3163 goto re_arm;
3164
3165 if (bond->send_grat_arp) {
3166 read_lock(&bond->curr_slave_lock);
3167 bond_send_gratuitous_arp(bond);
3168 read_unlock(&bond->curr_slave_lock);
3169 }
3170
3171 if (bond->send_unsol_na) {
3172 read_lock(&bond->curr_slave_lock);
3173 bond_send_unsolicited_na(bond);
3174 read_unlock(&bond->curr_slave_lock);
3175 }
3176
3177 if (bond_ab_arp_inspect(bond, delta_in_ticks)) {
3178 read_unlock(&bond->lock);
3179 rtnl_lock();
3180 read_lock(&bond->lock);
3181
3182 bond_ab_arp_commit(bond, delta_in_ticks);
3183
3184 read_unlock(&bond->lock);
3185 rtnl_unlock();
3186 read_lock(&bond->lock);
3187 }
3188
3189 bond_ab_arp_probe(bond);
3190
3191 re_arm:
3192 if (bond->params.arp_interval)
3193 queue_delayed_work(bond->wq, &bond->arp_work, delta_in_ticks);
3194 out:
3195 read_unlock(&bond->lock);
3196 }
3197
3198 /*------------------------------ proc/seq_file-------------------------------*/
3199
3200 #ifdef CONFIG_PROC_FS
3201
3202 static void *bond_info_seq_start(struct seq_file *seq, loff_t *pos)
3203 __acquires(&dev_base_lock)
3204 __acquires(&bond->lock)
3205 {
3206 struct bonding *bond = seq->private;
3207 loff_t off = 0;
3208 struct slave *slave;
3209 int i;
3210
3211 /* make sure the bond won't be taken away */
3212 read_lock(&dev_base_lock);
3213 read_lock(&bond->lock);
3214
3215 if (*pos == 0)
3216 return SEQ_START_TOKEN;
3217
3218 bond_for_each_slave(bond, slave, i) {
3219 if (++off == *pos)
3220 return slave;
3221 }
3222
3223 return NULL;
3224 }
3225
3226 static void *bond_info_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3227 {
3228 struct bonding *bond = seq->private;
3229 struct slave *slave = v;
3230
3231 ++*pos;
3232 if (v == SEQ_START_TOKEN)
3233 return bond->first_slave;
3234
3235 slave = slave->next;
3236
3237 return (slave == bond->first_slave) ? NULL : slave;
3238 }
3239
3240 static void bond_info_seq_stop(struct seq_file *seq, void *v)
3241 __releases(&bond->lock)
3242 __releases(&dev_base_lock)
3243 {
3244 struct bonding *bond = seq->private;
3245
3246 read_unlock(&bond->lock);
3247 read_unlock(&dev_base_lock);
3248 }
3249
3250 static void bond_info_show_master(struct seq_file *seq)
3251 {
3252 struct bonding *bond = seq->private;
3253 struct slave *curr;
3254 int i;
3255
3256 read_lock(&bond->curr_slave_lock);
3257 curr = bond->curr_active_slave;
3258 read_unlock(&bond->curr_slave_lock);
3259
3260 seq_printf(seq, "Bonding Mode: %s",
3261 bond_mode_name(bond->params.mode));
3262
3263 if (bond->params.mode == BOND_MODE_ACTIVEBACKUP &&
3264 bond->params.fail_over_mac)
3265 seq_printf(seq, " (fail_over_mac %s)",
3266 fail_over_mac_tbl[bond->params.fail_over_mac].modename);
3267
3268 seq_printf(seq, "\n");
3269
3270 if (bond->params.mode == BOND_MODE_XOR ||
3271 bond->params.mode == BOND_MODE_8023AD) {
3272 seq_printf(seq, "Transmit Hash Policy: %s (%d)\n",
3273 xmit_hashtype_tbl[bond->params.xmit_policy].modename,
3274 bond->params.xmit_policy);
3275 }
3276
3277 if (USES_PRIMARY(bond->params.mode)) {
3278 seq_printf(seq, "Primary Slave: %s",
3279 (bond->primary_slave) ?
3280 bond->primary_slave->dev->name : "None");
3281 if (bond->primary_slave)
3282 seq_printf(seq, " (primary_reselect %s)",
3283 pri_reselect_tbl[bond->params.primary_reselect].modename);
3284
3285 seq_printf(seq, "\nCurrently Active Slave: %s\n",
3286 (curr) ? curr->dev->name : "None");
3287 }
3288
3289 seq_printf(seq, "MII Status: %s\n", netif_carrier_ok(bond->dev) ?
3290 "up" : "down");
3291 seq_printf(seq, "MII Polling Interval (ms): %d\n", bond->params.miimon);
3292 seq_printf(seq, "Up Delay (ms): %d\n",
3293 bond->params.updelay * bond->params.miimon);
3294 seq_printf(seq, "Down Delay (ms): %d\n",
3295 bond->params.downdelay * bond->params.miimon);
3296
3297
3298 /* ARP information */
3299 if (bond->params.arp_interval > 0) {
3300 int printed = 0;
3301 seq_printf(seq, "ARP Polling Interval (ms): %d\n",
3302 bond->params.arp_interval);
3303
3304 seq_printf(seq, "ARP IP target/s (n.n.n.n form):");
3305
3306 for (i = 0; (i < BOND_MAX_ARP_TARGETS); i++) {
3307 if (!bond->params.arp_targets[i])
3308 break;
3309 if (printed)
3310 seq_printf(seq, ",");
3311 seq_printf(seq, " %pI4", &bond->params.arp_targets[i]);
3312 printed = 1;
3313 }
3314 seq_printf(seq, "\n");
3315 }
3316
3317 if (bond->params.mode == BOND_MODE_8023AD) {
3318 struct ad_info ad_info;
3319
3320 seq_puts(seq, "\n802.3ad info\n");
3321 seq_printf(seq, "LACP rate: %s\n",
3322 (bond->params.lacp_fast) ? "fast" : "slow");
3323 seq_printf(seq, "Aggregator selection policy (ad_select): %s\n",
3324 ad_select_tbl[bond->params.ad_select].modename);
3325
3326 if (bond_3ad_get_active_agg_info(bond, &ad_info)) {
3327 seq_printf(seq, "bond %s has no active aggregator\n",
3328 bond->dev->name);
3329 } else {
3330 seq_printf(seq, "Active Aggregator Info:\n");
3331
3332 seq_printf(seq, "\tAggregator ID: %d\n",
3333 ad_info.aggregator_id);
3334 seq_printf(seq, "\tNumber of ports: %d\n",
3335 ad_info.ports);
3336 seq_printf(seq, "\tActor Key: %d\n",
3337 ad_info.actor_key);
3338 seq_printf(seq, "\tPartner Key: %d\n",
3339 ad_info.partner_key);
3340 seq_printf(seq, "\tPartner Mac Address: %pM\n",
3341 ad_info.partner_system);
3342 }
3343 }
3344 }
3345
3346 static void bond_info_show_slave(struct seq_file *seq,
3347 const struct slave *slave)
3348 {
3349 struct bonding *bond = seq->private;
3350
3351 seq_printf(seq, "\nSlave Interface: %s\n", slave->dev->name);
3352 seq_printf(seq, "MII Status: %s\n",
3353 (slave->link == BOND_LINK_UP) ? "up" : "down");
3354 seq_printf(seq, "Link Failure Count: %u\n",
3355 slave->link_failure_count);
3356
3357 seq_printf(seq, "Permanent HW addr: %pM\n", slave->perm_hwaddr);
3358
3359 if (bond->params.mode == BOND_MODE_8023AD) {
3360 const struct aggregator *agg
3361 = SLAVE_AD_INFO(slave).port.aggregator;
3362
3363 if (agg)
3364 seq_printf(seq, "Aggregator ID: %d\n",
3365 agg->aggregator_identifier);
3366 else
3367 seq_puts(seq, "Aggregator ID: N/A\n");
3368 }
3369 seq_printf(seq, "Slave queue ID: %d\n", slave->queue_id);
3370 }
3371
3372 static int bond_info_seq_show(struct seq_file *seq, void *v)
3373 {
3374 if (v == SEQ_START_TOKEN) {
3375 seq_printf(seq, "%s\n", version);
3376 bond_info_show_master(seq);
3377 } else
3378 bond_info_show_slave(seq, v);
3379
3380 return 0;
3381 }
3382
3383 static const struct seq_operations bond_info_seq_ops = {
3384 .start = bond_info_seq_start,
3385 .next = bond_info_seq_next,
3386 .stop = bond_info_seq_stop,
3387 .show = bond_info_seq_show,
3388 };
3389
3390 static int bond_info_open(struct inode *inode, struct file *file)
3391 {
3392 struct seq_file *seq;
3393 struct proc_dir_entry *proc;
3394 int res;
3395
3396 res = seq_open(file, &bond_info_seq_ops);
3397 if (!res) {
3398 /* recover the pointer buried in proc_dir_entry data */
3399 seq = file->private_data;
3400 proc = PDE(inode);
3401 seq->private = proc->data;
3402 }
3403
3404 return res;
3405 }
3406
3407 static const struct file_operations bond_info_fops = {
3408 .owner = THIS_MODULE,
3409 .open = bond_info_open,
3410 .read = seq_read,
3411 .llseek = seq_lseek,
3412 .release = seq_release,
3413 };
3414
3415 static void bond_create_proc_entry(struct bonding *bond)
3416 {
3417 struct net_device *bond_dev = bond->dev;
3418 struct bond_net *bn = net_generic(dev_net(bond_dev), bond_net_id);
3419
3420 if (bn->proc_dir) {
3421 bond->proc_entry = proc_create_data(bond_dev->name,
3422 S_IRUGO, bn->proc_dir,
3423 &bond_info_fops, bond);
3424 if (bond->proc_entry == NULL)
3425 pr_warning("Warning: Cannot create /proc/net/%s/%s\n",
3426 DRV_NAME, bond_dev->name);
3427 else
3428 memcpy(bond->proc_file_name, bond_dev->name, IFNAMSIZ);
3429 }
3430 }
3431
3432 static void bond_remove_proc_entry(struct bonding *bond)
3433 {
3434 struct net_device *bond_dev = bond->dev;
3435 struct bond_net *bn = net_generic(dev_net(bond_dev), bond_net_id);
3436
3437 if (bn->proc_dir && bond->proc_entry) {
3438 remove_proc_entry(bond->proc_file_name, bn->proc_dir);
3439 memset(bond->proc_file_name, 0, IFNAMSIZ);
3440 bond->proc_entry = NULL;
3441 }
3442 }
3443
3444 /* Create the bonding directory under /proc/net, if doesn't exist yet.
3445 * Caller must hold rtnl_lock.
3446 */
3447 static void __net_init bond_create_proc_dir(struct bond_net *bn)
3448 {
3449 if (!bn->proc_dir) {
3450 bn->proc_dir = proc_mkdir(DRV_NAME, bn->net->proc_net);
3451 if (!bn->proc_dir)
3452 pr_warning("Warning: cannot create /proc/net/%s\n",
3453 DRV_NAME);
3454 }
3455 }
3456
3457 /* Destroy the bonding directory under /proc/net, if empty.
3458 * Caller must hold rtnl_lock.
3459 */
3460 static void __net_exit bond_destroy_proc_dir(struct bond_net *bn)
3461 {
3462 if (bn->proc_dir) {
3463 remove_proc_entry(DRV_NAME, bn->net->proc_net);
3464 bn->proc_dir = NULL;
3465 }
3466 }
3467
3468 #else /* !CONFIG_PROC_FS */
3469
3470 static void bond_create_proc_entry(struct bonding *bond)
3471 {
3472 }
3473
3474 static void bond_remove_proc_entry(struct bonding *bond)
3475 {
3476 }
3477
3478 static inline void bond_create_proc_dir(struct bond_net *bn)
3479 {
3480 }
3481
3482 static inline void bond_destroy_proc_dir(struct bond_net *bn)
3483 {
3484 }
3485
3486 #endif /* CONFIG_PROC_FS */
3487
3488
3489 /*-------------------------- netdev event handling --------------------------*/
3490
3491 /*
3492 * Change device name
3493 */
3494 static int bond_event_changename(struct bonding *bond)
3495 {
3496 bond_remove_proc_entry(bond);
3497 bond_create_proc_entry(bond);
3498
3499 return NOTIFY_DONE;
3500 }
3501
3502 static int bond_master_netdev_event(unsigned long event,
3503 struct net_device *bond_dev)
3504 {
3505 struct bonding *event_bond = netdev_priv(bond_dev);
3506
3507 switch (event) {
3508 case NETDEV_CHANGENAME:
3509 return bond_event_changename(event_bond);
3510 default:
3511 break;
3512 }
3513
3514 return NOTIFY_DONE;
3515 }
3516
3517 static int bond_slave_netdev_event(unsigned long event,
3518 struct net_device *slave_dev)
3519 {
3520 struct net_device *bond_dev = slave_dev->master;
3521 struct bonding *bond = netdev_priv(bond_dev);
3522
3523 switch (event) {
3524 case NETDEV_UNREGISTER:
3525 if (bond_dev) {
3526 if (bond->setup_by_slave)
3527 bond_release_and_destroy(bond_dev, slave_dev);
3528 else
3529 bond_release(bond_dev, slave_dev);
3530 }
3531 break;
3532 case NETDEV_CHANGE:
3533 if (bond->params.mode == BOND_MODE_8023AD || bond_is_lb(bond)) {
3534 struct slave *slave;
3535
3536 slave = bond_get_slave_by_dev(bond, slave_dev);
3537 if (slave) {
3538 u16 old_speed = slave->speed;
3539 u16 old_duplex = slave->duplex;
3540
3541 bond_update_speed_duplex(slave);
3542
3543 if (bond_is_lb(bond))
3544 break;
3545
3546 if (old_speed != slave->speed)
3547 bond_3ad_adapter_speed_changed(slave);
3548 if (old_duplex != slave->duplex)
3549 bond_3ad_adapter_duplex_changed(slave);
3550 }
3551 }
3552
3553 break;
3554 case NETDEV_DOWN:
3555 /*
3556 * ... Or is it this?
3557 */
3558 break;
3559 case NETDEV_CHANGEMTU:
3560 /*
3561 * TODO: Should slaves be allowed to
3562 * independently alter their MTU? For
3563 * an active-backup bond, slaves need
3564 * not be the same type of device, so
3565 * MTUs may vary. For other modes,
3566 * slaves arguably should have the
3567 * same MTUs. To do this, we'd need to
3568 * take over the slave's change_mtu
3569 * function for the duration of their
3570 * servitude.
3571 */
3572 break;
3573 case NETDEV_CHANGENAME:
3574 /*
3575 * TODO: handle changing the primary's name
3576 */
3577 break;
3578 case NETDEV_FEAT_CHANGE:
3579 bond_compute_features(bond);
3580 break;
3581 default:
3582 break;
3583 }
3584
3585 return NOTIFY_DONE;
3586 }
3587
3588 /*
3589 * bond_netdev_event: handle netdev notifier chain events.
3590 *
3591 * This function receives events for the netdev chain. The caller (an
3592 * ioctl handler calling blocking_notifier_call_chain) holds the necessary
3593 * locks for us to safely manipulate the slave devices (RTNL lock,
3594 * dev_probe_lock).
3595 */
3596 static int bond_netdev_event(struct notifier_block *this,
3597 unsigned long event, void *ptr)
3598 {
3599 struct net_device *event_dev = (struct net_device *)ptr;
3600
3601 pr_debug("event_dev: %s, event: %lx\n",
3602 event_dev ? event_dev->name : "None",
3603 event);
3604
3605 if (!(event_dev->priv_flags & IFF_BONDING))
3606 return NOTIFY_DONE;
3607
3608 if (event_dev->flags & IFF_MASTER) {
3609 pr_debug("IFF_MASTER\n");
3610 return bond_master_netdev_event(event, event_dev);
3611 }
3612
3613 if (event_dev->flags & IFF_SLAVE) {
3614 pr_debug("IFF_SLAVE\n");
3615 return bond_slave_netdev_event(event, event_dev);
3616 }
3617
3618 return NOTIFY_DONE;
3619 }
3620
3621 /*
3622 * bond_inetaddr_event: handle inetaddr notifier chain events.
3623 *
3624 * We keep track of device IPs primarily to use as source addresses in
3625 * ARP monitor probes (rather than spewing out broadcasts all the time).
3626 *
3627 * We track one IP for the main device (if it has one), plus one per VLAN.
3628 */
3629 static int bond_inetaddr_event(struct notifier_block *this, unsigned long event, void *ptr)
3630 {
3631 struct in_ifaddr *ifa = ptr;
3632 struct net_device *vlan_dev, *event_dev = ifa->ifa_dev->dev;
3633 struct bond_net *bn = net_generic(dev_net(event_dev), bond_net_id);
3634 struct bonding *bond;
3635 struct vlan_entry *vlan;
3636
3637 list_for_each_entry(bond, &bn->dev_list, bond_list) {
3638 if (bond->dev == event_dev) {
3639 switch (event) {
3640 case NETDEV_UP:
3641 bond->master_ip = ifa->ifa_local;
3642 return NOTIFY_OK;
3643 case NETDEV_DOWN:
3644 bond->master_ip = bond_glean_dev_ip(bond->dev);
3645 return NOTIFY_OK;
3646 default:
3647 return NOTIFY_DONE;
3648 }
3649 }
3650
3651 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
3652 if (!bond->vlgrp)
3653 continue;
3654 vlan_dev = vlan_group_get_device(bond->vlgrp, vlan->vlan_id);
3655 if (vlan_dev == event_dev) {
3656 switch (event) {
3657 case NETDEV_UP:
3658 vlan->vlan_ip = ifa->ifa_local;
3659 return NOTIFY_OK;
3660 case NETDEV_DOWN:
3661 vlan->vlan_ip =
3662 bond_glean_dev_ip(vlan_dev);
3663 return NOTIFY_OK;
3664 default:
3665 return NOTIFY_DONE;
3666 }
3667 }
3668 }
3669 }
3670 return NOTIFY_DONE;
3671 }
3672
3673 static struct notifier_block bond_netdev_notifier = {
3674 .notifier_call = bond_netdev_event,
3675 };
3676
3677 static struct notifier_block bond_inetaddr_notifier = {
3678 .notifier_call = bond_inetaddr_event,
3679 };
3680
3681 /*-------------------------- Packet type handling ---------------------------*/
3682
3683 /* register to receive lacpdus on a bond */
3684 static void bond_register_lacpdu(struct bonding *bond)
3685 {
3686 struct packet_type *pk_type = &(BOND_AD_INFO(bond).ad_pkt_type);
3687
3688 /* initialize packet type */
3689 pk_type->type = PKT_TYPE_LACPDU;
3690 pk_type->dev = bond->dev;
3691 pk_type->func = bond_3ad_lacpdu_recv;
3692
3693 dev_add_pack(pk_type);
3694 }
3695
3696 /* unregister to receive lacpdus on a bond */
3697 static void bond_unregister_lacpdu(struct bonding *bond)
3698 {
3699 dev_remove_pack(&(BOND_AD_INFO(bond).ad_pkt_type));
3700 }
3701
3702 void bond_register_arp(struct bonding *bond)
3703 {
3704 struct packet_type *pt = &bond->arp_mon_pt;
3705
3706 if (pt->type)
3707 return;
3708
3709 pt->type = htons(ETH_P_ARP);
3710 pt->dev = bond->dev;
3711 pt->func = bond_arp_rcv;
3712 dev_add_pack(pt);
3713 }
3714
3715 void bond_unregister_arp(struct bonding *bond)
3716 {
3717 struct packet_type *pt = &bond->arp_mon_pt;
3718
3719 dev_remove_pack(pt);
3720 pt->type = 0;
3721 }
3722
3723 /*---------------------------- Hashing Policies -----------------------------*/
3724
3725 /*
3726 * Hash for the output device based upon layer 2 and layer 3 data. If
3727 * the packet is not IP mimic bond_xmit_hash_policy_l2()
3728 */
3729 static int bond_xmit_hash_policy_l23(struct sk_buff *skb, int count)
3730 {
3731 struct ethhdr *data = (struct ethhdr *)skb->data;
3732 struct iphdr *iph = ip_hdr(skb);
3733
3734 if (skb->protocol == htons(ETH_P_IP)) {
3735 return ((ntohl(iph->saddr ^ iph->daddr) & 0xffff) ^
3736 (data->h_dest[5] ^ data->h_source[5])) % count;
3737 }
3738
3739 return (data->h_dest[5] ^ data->h_source[5]) % count;
3740 }
3741
3742 /*
3743 * Hash for the output device based upon layer 3 and layer 4 data. If
3744 * the packet is a frag or not TCP or UDP, just use layer 3 data. If it is
3745 * altogether not IP, mimic bond_xmit_hash_policy_l2()
3746 */
3747 static int bond_xmit_hash_policy_l34(struct sk_buff *skb, int count)
3748 {
3749 struct ethhdr *data = (struct ethhdr *)skb->data;
3750 struct iphdr *iph = ip_hdr(skb);
3751 __be16 *layer4hdr = (__be16 *)((u32 *)iph + iph->ihl);
3752 int layer4_xor = 0;
3753
3754 if (skb->protocol == htons(ETH_P_IP)) {
3755 if (!(iph->frag_off & htons(IP_MF|IP_OFFSET)) &&
3756 (iph->protocol == IPPROTO_TCP ||
3757 iph->protocol == IPPROTO_UDP)) {
3758 layer4_xor = ntohs((*layer4hdr ^ *(layer4hdr + 1)));
3759 }
3760 return (layer4_xor ^
3761 ((ntohl(iph->saddr ^ iph->daddr)) & 0xffff)) % count;
3762
3763 }
3764
3765 return (data->h_dest[5] ^ data->h_source[5]) % count;
3766 }
3767
3768 /*
3769 * Hash for the output device based upon layer 2 data
3770 */
3771 static int bond_xmit_hash_policy_l2(struct sk_buff *skb, int count)
3772 {
3773 struct ethhdr *data = (struct ethhdr *)skb->data;
3774
3775 return (data->h_dest[5] ^ data->h_source[5]) % count;
3776 }
3777
3778 /*-------------------------- Device entry points ----------------------------*/
3779
3780 static int bond_open(struct net_device *bond_dev)
3781 {
3782 struct bonding *bond = netdev_priv(bond_dev);
3783
3784 bond->kill_timers = 0;
3785
3786 INIT_DELAYED_WORK(&bond->mcast_work, bond_resend_igmp_join_requests_delayed);
3787
3788 if (bond_is_lb(bond)) {
3789 /* bond_alb_initialize must be called before the timer
3790 * is started.
3791 */
3792 if (bond_alb_initialize(bond, (bond->params.mode == BOND_MODE_ALB))) {
3793 /* something went wrong - fail the open operation */
3794 return -ENOMEM;
3795 }
3796
3797 INIT_DELAYED_WORK(&bond->alb_work, bond_alb_monitor);
3798 queue_delayed_work(bond->wq, &bond->alb_work, 0);
3799 }
3800
3801 if (bond->params.miimon) { /* link check interval, in milliseconds. */
3802 INIT_DELAYED_WORK(&bond->mii_work, bond_mii_monitor);
3803 queue_delayed_work(bond->wq, &bond->mii_work, 0);
3804 }
3805
3806 if (bond->params.arp_interval) { /* arp interval, in milliseconds. */
3807 if (bond->params.mode == BOND_MODE_ACTIVEBACKUP)
3808 INIT_DELAYED_WORK(&bond->arp_work,
3809 bond_activebackup_arp_mon);
3810 else
3811 INIT_DELAYED_WORK(&bond->arp_work,
3812 bond_loadbalance_arp_mon);
3813
3814 queue_delayed_work(bond->wq, &bond->arp_work, 0);
3815 if (bond->params.arp_validate)
3816 bond_register_arp(bond);
3817 }
3818
3819 if (bond->params.mode == BOND_MODE_8023AD) {
3820 INIT_DELAYED_WORK(&bond->ad_work, bond_3ad_state_machine_handler);
3821 queue_delayed_work(bond->wq, &bond->ad_work, 0);
3822 /* register to receive LACPDUs */
3823 bond_register_lacpdu(bond);
3824 bond_3ad_initiate_agg_selection(bond, 1);
3825 }
3826
3827 return 0;
3828 }
3829
3830 static int bond_close(struct net_device *bond_dev)
3831 {
3832 struct bonding *bond = netdev_priv(bond_dev);
3833
3834 if (bond->params.mode == BOND_MODE_8023AD) {
3835 /* Unregister the receive of LACPDUs */
3836 bond_unregister_lacpdu(bond);
3837 }
3838
3839 if (bond->params.arp_validate)
3840 bond_unregister_arp(bond);
3841
3842 write_lock_bh(&bond->lock);
3843
3844 bond->send_grat_arp = 0;
3845 bond->send_unsol_na = 0;
3846
3847 /* signal timers not to re-arm */
3848 bond->kill_timers = 1;
3849
3850 write_unlock_bh(&bond->lock);
3851
3852 if (bond->params.miimon) { /* link check interval, in milliseconds. */
3853 cancel_delayed_work(&bond->mii_work);
3854 }
3855
3856 if (bond->params.arp_interval) { /* arp interval, in milliseconds. */
3857 cancel_delayed_work(&bond->arp_work);
3858 }
3859
3860 switch (bond->params.mode) {
3861 case BOND_MODE_8023AD:
3862 cancel_delayed_work(&bond->ad_work);
3863 break;
3864 case BOND_MODE_TLB:
3865 case BOND_MODE_ALB:
3866 cancel_delayed_work(&bond->alb_work);
3867 break;
3868 default:
3869 break;
3870 }
3871
3872 if (delayed_work_pending(&bond->mcast_work))
3873 cancel_delayed_work(&bond->mcast_work);
3874
3875 if (bond_is_lb(bond)) {
3876 /* Must be called only after all
3877 * slaves have been released
3878 */
3879 bond_alb_deinitialize(bond);
3880 }
3881
3882 return 0;
3883 }
3884
3885 static struct rtnl_link_stats64 *bond_get_stats(struct net_device *bond_dev,
3886 struct rtnl_link_stats64 *stats)
3887 {
3888 struct bonding *bond = netdev_priv(bond_dev);
3889 struct rtnl_link_stats64 temp;
3890 struct slave *slave;
3891 int i;
3892
3893 memset(stats, 0, sizeof(*stats));
3894
3895 read_lock_bh(&bond->lock);
3896
3897 bond_for_each_slave(bond, slave, i) {
3898 const struct rtnl_link_stats64 *sstats =
3899 dev_get_stats(slave->dev, &temp);
3900
3901 stats->rx_packets += sstats->rx_packets;
3902 stats->rx_bytes += sstats->rx_bytes;
3903 stats->rx_errors += sstats->rx_errors;
3904 stats->rx_dropped += sstats->rx_dropped;
3905
3906 stats->tx_packets += sstats->tx_packets;
3907 stats->tx_bytes += sstats->tx_bytes;
3908 stats->tx_errors += sstats->tx_errors;
3909 stats->tx_dropped += sstats->tx_dropped;
3910
3911 stats->multicast += sstats->multicast;
3912 stats->collisions += sstats->collisions;
3913
3914 stats->rx_length_errors += sstats->rx_length_errors;
3915 stats->rx_over_errors += sstats->rx_over_errors;
3916 stats->rx_crc_errors += sstats->rx_crc_errors;
3917 stats->rx_frame_errors += sstats->rx_frame_errors;
3918 stats->rx_fifo_errors += sstats->rx_fifo_errors;
3919 stats->rx_missed_errors += sstats->rx_missed_errors;
3920
3921 stats->tx_aborted_errors += sstats->tx_aborted_errors;
3922 stats->tx_carrier_errors += sstats->tx_carrier_errors;
3923 stats->tx_fifo_errors += sstats->tx_fifo_errors;
3924 stats->tx_heartbeat_errors += sstats->tx_heartbeat_errors;
3925 stats->tx_window_errors += sstats->tx_window_errors;
3926 }
3927
3928 read_unlock_bh(&bond->lock);
3929
3930 return stats;
3931 }
3932
3933 static int bond_do_ioctl(struct net_device *bond_dev, struct ifreq *ifr, int cmd)
3934 {
3935 struct net_device *slave_dev = NULL;
3936 struct ifbond k_binfo;
3937 struct ifbond __user *u_binfo = NULL;
3938 struct ifslave k_sinfo;
3939 struct ifslave __user *u_sinfo = NULL;
3940 struct mii_ioctl_data *mii = NULL;
3941 int res = 0;
3942
3943 pr_debug("bond_ioctl: master=%s, cmd=%d\n", bond_dev->name, cmd);
3944
3945 switch (cmd) {
3946 case SIOCGMIIPHY:
3947 mii = if_mii(ifr);
3948 if (!mii)
3949 return -EINVAL;
3950
3951 mii->phy_id = 0;
3952 /* Fall Through */
3953 case SIOCGMIIREG:
3954 /*
3955 * We do this again just in case we were called by SIOCGMIIREG
3956 * instead of SIOCGMIIPHY.
3957 */
3958 mii = if_mii(ifr);
3959 if (!mii)
3960 return -EINVAL;
3961
3962
3963 if (mii->reg_num == 1) {
3964 struct bonding *bond = netdev_priv(bond_dev);
3965 mii->val_out = 0;
3966 read_lock(&bond->lock);
3967 read_lock(&bond->curr_slave_lock);
3968 if (netif_carrier_ok(bond->dev))
3969 mii->val_out = BMSR_LSTATUS;
3970
3971 read_unlock(&bond->curr_slave_lock);
3972 read_unlock(&bond->lock);
3973 }
3974
3975 return 0;
3976 case BOND_INFO_QUERY_OLD:
3977 case SIOCBONDINFOQUERY:
3978 u_binfo = (struct ifbond __user *)ifr->ifr_data;
3979
3980 if (copy_from_user(&k_binfo, u_binfo, sizeof(ifbond)))
3981 return -EFAULT;
3982
3983 res = bond_info_query(bond_dev, &k_binfo);
3984 if (res == 0 &&
3985 copy_to_user(u_binfo, &k_binfo, sizeof(ifbond)))
3986 return -EFAULT;
3987
3988 return res;
3989 case BOND_SLAVE_INFO_QUERY_OLD:
3990 case SIOCBONDSLAVEINFOQUERY:
3991 u_sinfo = (struct ifslave __user *)ifr->ifr_data;
3992
3993 if (copy_from_user(&k_sinfo, u_sinfo, sizeof(ifslave)))
3994 return -EFAULT;
3995
3996 res = bond_slave_info_query(bond_dev, &k_sinfo);
3997 if (res == 0 &&
3998 copy_to_user(u_sinfo, &k_sinfo, sizeof(ifslave)))
3999 return -EFAULT;
4000
4001 return res;
4002 default:
4003 /* Go on */
4004 break;
4005 }
4006
4007 if (!capable(CAP_NET_ADMIN))
4008 return -EPERM;
4009
4010 slave_dev = dev_get_by_name(dev_net(bond_dev), ifr->ifr_slave);
4011
4012 pr_debug("slave_dev=%p:\n", slave_dev);
4013
4014 if (!slave_dev)
4015 res = -ENODEV;
4016 else {
4017 pr_debug("slave_dev->name=%s:\n", slave_dev->name);
4018 switch (cmd) {
4019 case BOND_ENSLAVE_OLD:
4020 case SIOCBONDENSLAVE:
4021 res = bond_enslave(bond_dev, slave_dev);
4022 break;
4023 case BOND_RELEASE_OLD:
4024 case SIOCBONDRELEASE:
4025 res = bond_release(bond_dev, slave_dev);
4026 break;
4027 case BOND_SETHWADDR_OLD:
4028 case SIOCBONDSETHWADDR:
4029 res = bond_sethwaddr(bond_dev, slave_dev);
4030 break;
4031 case BOND_CHANGE_ACTIVE_OLD:
4032 case SIOCBONDCHANGEACTIVE:
4033 res = bond_ioctl_change_active(bond_dev, slave_dev);
4034 break;
4035 default:
4036 res = -EOPNOTSUPP;
4037 }
4038
4039 dev_put(slave_dev);
4040 }
4041
4042 return res;
4043 }
4044
4045 static bool bond_addr_in_mc_list(unsigned char *addr,
4046 struct netdev_hw_addr_list *list,
4047 int addrlen)
4048 {
4049 struct netdev_hw_addr *ha;
4050
4051 netdev_hw_addr_list_for_each(ha, list)
4052 if (!memcmp(ha->addr, addr, addrlen))
4053 return true;
4054
4055 return false;
4056 }
4057
4058 static void bond_set_multicast_list(struct net_device *bond_dev)
4059 {
4060 struct bonding *bond = netdev_priv(bond_dev);
4061 struct netdev_hw_addr *ha;
4062 bool found;
4063
4064 /*
4065 * Do promisc before checking multicast_mode
4066 */
4067 if ((bond_dev->flags & IFF_PROMISC) && !(bond->flags & IFF_PROMISC))
4068 /*
4069 * FIXME: Need to handle the error when one of the multi-slaves
4070 * encounters error.
4071 */
4072 bond_set_promiscuity(bond, 1);
4073
4074
4075 if (!(bond_dev->flags & IFF_PROMISC) && (bond->flags & IFF_PROMISC))
4076 bond_set_promiscuity(bond, -1);
4077
4078
4079 /* set allmulti flag to slaves */
4080 if ((bond_dev->flags & IFF_ALLMULTI) && !(bond->flags & IFF_ALLMULTI))
4081 /*
4082 * FIXME: Need to handle the error when one of the multi-slaves
4083 * encounters error.
4084 */
4085 bond_set_allmulti(bond, 1);
4086
4087
4088 if (!(bond_dev->flags & IFF_ALLMULTI) && (bond->flags & IFF_ALLMULTI))
4089 bond_set_allmulti(bond, -1);
4090
4091
4092 read_lock(&bond->lock);
4093
4094 bond->flags = bond_dev->flags;
4095
4096 /* looking for addresses to add to slaves' mc list */
4097 netdev_for_each_mc_addr(ha, bond_dev) {
4098 found = bond_addr_in_mc_list(ha->addr, &bond->mc_list,
4099 bond_dev->addr_len);
4100 if (!found)
4101 bond_mc_add(bond, ha->addr);
4102 }
4103
4104 /* looking for addresses to delete from slaves' list */
4105 netdev_hw_addr_list_for_each(ha, &bond->mc_list) {
4106 found = bond_addr_in_mc_list(ha->addr, &bond_dev->mc,
4107 bond_dev->addr_len);
4108 if (!found)
4109 bond_mc_del(bond, ha->addr);
4110 }
4111
4112 /* save master's multicast list */
4113 __hw_addr_flush(&bond->mc_list);
4114 __hw_addr_add_multiple(&bond->mc_list, &bond_dev->mc,
4115 bond_dev->addr_len, NETDEV_HW_ADDR_T_MULTICAST);
4116
4117 read_unlock(&bond->lock);
4118 }
4119
4120 static int bond_neigh_setup(struct net_device *dev, struct neigh_parms *parms)
4121 {
4122 struct bonding *bond = netdev_priv(dev);
4123 struct slave *slave = bond->first_slave;
4124
4125 if (slave) {
4126 const struct net_device_ops *slave_ops
4127 = slave->dev->netdev_ops;
4128 if (slave_ops->ndo_neigh_setup)
4129 return slave_ops->ndo_neigh_setup(slave->dev, parms);
4130 }
4131 return 0;
4132 }
4133
4134 /*
4135 * Change the MTU of all of a master's slaves to match the master
4136 */
4137 static int bond_change_mtu(struct net_device *bond_dev, int new_mtu)
4138 {
4139 struct bonding *bond = netdev_priv(bond_dev);
4140 struct slave *slave, *stop_at;
4141 int res = 0;
4142 int i;
4143
4144 pr_debug("bond=%p, name=%s, new_mtu=%d\n", bond,
4145 (bond_dev ? bond_dev->name : "None"), new_mtu);
4146
4147 /* Can't hold bond->lock with bh disabled here since
4148 * some base drivers panic. On the other hand we can't
4149 * hold bond->lock without bh disabled because we'll
4150 * deadlock. The only solution is to rely on the fact
4151 * that we're under rtnl_lock here, and the slaves
4152 * list won't change. This doesn't solve the problem
4153 * of setting the slave's MTU while it is
4154 * transmitting, but the assumption is that the base
4155 * driver can handle that.
4156 *
4157 * TODO: figure out a way to safely iterate the slaves
4158 * list, but without holding a lock around the actual
4159 * call to the base driver.
4160 */
4161
4162 bond_for_each_slave(bond, slave, i) {
4163 pr_debug("s %p s->p %p c_m %p\n",
4164 slave,
4165 slave->prev,
4166 slave->dev->netdev_ops->ndo_change_mtu);
4167
4168 res = dev_set_mtu(slave->dev, new_mtu);
4169
4170 if (res) {
4171 /* If we failed to set the slave's mtu to the new value
4172 * we must abort the operation even in ACTIVE_BACKUP
4173 * mode, because if we allow the backup slaves to have
4174 * different mtu values than the active slave we'll
4175 * need to change their mtu when doing a failover. That
4176 * means changing their mtu from timer context, which
4177 * is probably not a good idea.
4178 */
4179 pr_debug("err %d %s\n", res, slave->dev->name);
4180 goto unwind;
4181 }
4182 }
4183
4184 bond_dev->mtu = new_mtu;
4185
4186 return 0;
4187
4188 unwind:
4189 /* unwind from head to the slave that failed */
4190 stop_at = slave;
4191 bond_for_each_slave_from_to(bond, slave, i, bond->first_slave, stop_at) {
4192 int tmp_res;
4193
4194 tmp_res = dev_set_mtu(slave->dev, bond_dev->mtu);
4195 if (tmp_res) {
4196 pr_debug("unwind err %d dev %s\n",
4197 tmp_res, slave->dev->name);
4198 }
4199 }
4200
4201 return res;
4202 }
4203
4204 /*
4205 * Change HW address
4206 *
4207 * Note that many devices must be down to change the HW address, and
4208 * downing the master releases all slaves. We can make bonds full of
4209 * bonding devices to test this, however.
4210 */
4211 static int bond_set_mac_address(struct net_device *bond_dev, void *addr)
4212 {
4213 struct bonding *bond = netdev_priv(bond_dev);
4214 struct sockaddr *sa = addr, tmp_sa;
4215 struct slave *slave, *stop_at;
4216 int res = 0;
4217 int i;
4218
4219 if (bond->params.mode == BOND_MODE_ALB)
4220 return bond_alb_set_mac_address(bond_dev, addr);
4221
4222
4223 pr_debug("bond=%p, name=%s\n",
4224 bond, bond_dev ? bond_dev->name : "None");
4225
4226 /*
4227 * If fail_over_mac is set to active, do nothing and return
4228 * success. Returning an error causes ifenslave to fail.
4229 */
4230 if (bond->params.fail_over_mac == BOND_FOM_ACTIVE)
4231 return 0;
4232
4233 if (!is_valid_ether_addr(sa->sa_data))
4234 return -EADDRNOTAVAIL;
4235
4236 /* Can't hold bond->lock with bh disabled here since
4237 * some base drivers panic. On the other hand we can't
4238 * hold bond->lock without bh disabled because we'll
4239 * deadlock. The only solution is to rely on the fact
4240 * that we're under rtnl_lock here, and the slaves
4241 * list won't change. This doesn't solve the problem
4242 * of setting the slave's hw address while it is
4243 * transmitting, but the assumption is that the base
4244 * driver can handle that.
4245 *
4246 * TODO: figure out a way to safely iterate the slaves
4247 * list, but without holding a lock around the actual
4248 * call to the base driver.
4249 */
4250
4251 bond_for_each_slave(bond, slave, i) {
4252 const struct net_device_ops *slave_ops = slave->dev->netdev_ops;
4253 pr_debug("slave %p %s\n", slave, slave->dev->name);
4254
4255 if (slave_ops->ndo_set_mac_address == NULL) {
4256 res = -EOPNOTSUPP;
4257 pr_debug("EOPNOTSUPP %s\n", slave->dev->name);
4258 goto unwind;
4259 }
4260
4261 res = dev_set_mac_address(slave->dev, addr);
4262 if (res) {
4263 /* TODO: consider downing the slave
4264 * and retry ?
4265 * User should expect communications
4266 * breakage anyway until ARP finish
4267 * updating, so...
4268 */
4269 pr_debug("err %d %s\n", res, slave->dev->name);
4270 goto unwind;
4271 }
4272 }
4273
4274 /* success */
4275 memcpy(bond_dev->dev_addr, sa->sa_data, bond_dev->addr_len);
4276 return 0;
4277
4278 unwind:
4279 memcpy(tmp_sa.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
4280 tmp_sa.sa_family = bond_dev->type;
4281
4282 /* unwind from head to the slave that failed */
4283 stop_at = slave;
4284 bond_for_each_slave_from_to(bond, slave, i, bond->first_slave, stop_at) {
4285 int tmp_res;
4286
4287 tmp_res = dev_set_mac_address(slave->dev, &tmp_sa);
4288 if (tmp_res) {
4289 pr_debug("unwind err %d dev %s\n",
4290 tmp_res, slave->dev->name);
4291 }
4292 }
4293
4294 return res;
4295 }
4296
4297 static int bond_xmit_roundrobin(struct sk_buff *skb, struct net_device *bond_dev)
4298 {
4299 struct bonding *bond = netdev_priv(bond_dev);
4300 struct slave *slave, *start_at;
4301 int i, slave_no, res = 1;
4302 struct iphdr *iph = ip_hdr(skb);
4303
4304 read_lock(&bond->lock);
4305
4306 if (!BOND_IS_OK(bond))
4307 goto out;
4308 /*
4309 * Start with the curr_active_slave that joined the bond as the
4310 * default for sending IGMP traffic. For failover purposes one
4311 * needs to maintain some consistency for the interface that will
4312 * send the join/membership reports. The curr_active_slave found
4313 * will send all of this type of traffic.
4314 */
4315 if ((iph->protocol == IPPROTO_IGMP) &&
4316 (skb->protocol == htons(ETH_P_IP))) {
4317
4318 read_lock(&bond->curr_slave_lock);
4319 slave = bond->curr_active_slave;
4320 read_unlock(&bond->curr_slave_lock);
4321
4322 if (!slave)
4323 goto out;
4324 } else {
4325 /*
4326 * Concurrent TX may collide on rr_tx_counter; we accept
4327 * that as being rare enough not to justify using an
4328 * atomic op here.
4329 */
4330 slave_no = bond->rr_tx_counter++ % bond->slave_cnt;
4331
4332 bond_for_each_slave(bond, slave, i) {
4333 slave_no--;
4334 if (slave_no < 0)
4335 break;
4336 }
4337 }
4338
4339 start_at = slave;
4340 bond_for_each_slave_from(bond, slave, i, start_at) {
4341 if (IS_UP(slave->dev) &&
4342 (slave->link == BOND_LINK_UP) &&
4343 (slave->state == BOND_STATE_ACTIVE)) {
4344 res = bond_dev_queue_xmit(bond, skb, slave->dev);
4345 break;
4346 }
4347 }
4348
4349 out:
4350 if (res) {
4351 /* no suitable interface, frame not sent */
4352 dev_kfree_skb(skb);
4353 }
4354 read_unlock(&bond->lock);
4355 return NETDEV_TX_OK;
4356 }
4357
4358
4359 /*
4360 * in active-backup mode, we know that bond->curr_active_slave is always valid if
4361 * the bond has a usable interface.
4362 */
4363 static int bond_xmit_activebackup(struct sk_buff *skb, struct net_device *bond_dev)
4364 {
4365 struct bonding *bond = netdev_priv(bond_dev);
4366 int res = 1;
4367
4368 read_lock(&bond->lock);
4369 read_lock(&bond->curr_slave_lock);
4370
4371 if (!BOND_IS_OK(bond))
4372 goto out;
4373
4374 if (!bond->curr_active_slave)
4375 goto out;
4376
4377 res = bond_dev_queue_xmit(bond, skb, bond->curr_active_slave->dev);
4378
4379 out:
4380 if (res)
4381 /* no suitable interface, frame not sent */
4382 dev_kfree_skb(skb);
4383
4384 read_unlock(&bond->curr_slave_lock);
4385 read_unlock(&bond->lock);
4386 return NETDEV_TX_OK;
4387 }
4388
4389 /*
4390 * In bond_xmit_xor() , we determine the output device by using a pre-
4391 * determined xmit_hash_policy(), If the selected device is not enabled,
4392 * find the next active slave.
4393 */
4394 static int bond_xmit_xor(struct sk_buff *skb, struct net_device *bond_dev)
4395 {
4396 struct bonding *bond = netdev_priv(bond_dev);
4397 struct slave *slave, *start_at;
4398 int slave_no;
4399 int i;
4400 int res = 1;
4401
4402 read_lock(&bond->lock);
4403
4404 if (!BOND_IS_OK(bond))
4405 goto out;
4406
4407 slave_no = bond->xmit_hash_policy(skb, bond->slave_cnt);
4408
4409 bond_for_each_slave(bond, slave, i) {
4410 slave_no--;
4411 if (slave_no < 0)
4412 break;
4413 }
4414
4415 start_at = slave;
4416
4417 bond_for_each_slave_from(bond, slave, i, start_at) {
4418 if (IS_UP(slave->dev) &&
4419 (slave->link == BOND_LINK_UP) &&
4420 (slave->state == BOND_STATE_ACTIVE)) {
4421 res = bond_dev_queue_xmit(bond, skb, slave->dev);
4422 break;
4423 }
4424 }
4425
4426 out:
4427 if (res) {
4428 /* no suitable interface, frame not sent */
4429 dev_kfree_skb(skb);
4430 }
4431 read_unlock(&bond->lock);
4432 return NETDEV_TX_OK;
4433 }
4434
4435 /*
4436 * in broadcast mode, we send everything to all usable interfaces.
4437 */
4438 static int bond_xmit_broadcast(struct sk_buff *skb, struct net_device *bond_dev)
4439 {
4440 struct bonding *bond = netdev_priv(bond_dev);
4441 struct slave *slave, *start_at;
4442 struct net_device *tx_dev = NULL;
4443 int i;
4444 int res = 1;
4445
4446 read_lock(&bond->lock);
4447
4448 if (!BOND_IS_OK(bond))
4449 goto out;
4450
4451 read_lock(&bond->curr_slave_lock);
4452 start_at = bond->curr_active_slave;
4453 read_unlock(&bond->curr_slave_lock);
4454
4455 if (!start_at)
4456 goto out;
4457
4458 bond_for_each_slave_from(bond, slave, i, start_at) {
4459 if (IS_UP(slave->dev) &&
4460 (slave->link == BOND_LINK_UP) &&
4461 (slave->state == BOND_STATE_ACTIVE)) {
4462 if (tx_dev) {
4463 struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
4464 if (!skb2) {
4465 pr_err("%s: Error: bond_xmit_broadcast(): skb_clone() failed\n",
4466 bond_dev->name);
4467 continue;
4468 }
4469
4470 res = bond_dev_queue_xmit(bond, skb2, tx_dev);
4471 if (res) {
4472 dev_kfree_skb(skb2);
4473 continue;
4474 }
4475 }
4476 tx_dev = slave->dev;
4477 }
4478 }
4479
4480 if (tx_dev)
4481 res = bond_dev_queue_xmit(bond, skb, tx_dev);
4482
4483 out:
4484 if (res)
4485 /* no suitable interface, frame not sent */
4486 dev_kfree_skb(skb);
4487
4488 /* frame sent to all suitable interfaces */
4489 read_unlock(&bond->lock);
4490 return NETDEV_TX_OK;
4491 }
4492
4493 /*------------------------- Device initialization ---------------------------*/
4494
4495 static void bond_set_xmit_hash_policy(struct bonding *bond)
4496 {
4497 switch (bond->params.xmit_policy) {
4498 case BOND_XMIT_POLICY_LAYER23:
4499 bond->xmit_hash_policy = bond_xmit_hash_policy_l23;
4500 break;
4501 case BOND_XMIT_POLICY_LAYER34:
4502 bond->xmit_hash_policy = bond_xmit_hash_policy_l34;
4503 break;
4504 case BOND_XMIT_POLICY_LAYER2:
4505 default:
4506 bond->xmit_hash_policy = bond_xmit_hash_policy_l2;
4507 break;
4508 }
4509 }
4510
4511 /*
4512 * Lookup the slave that corresponds to a qid
4513 */
4514 static inline int bond_slave_override(struct bonding *bond,
4515 struct sk_buff *skb)
4516 {
4517 int i, res = 1;
4518 struct slave *slave = NULL;
4519 struct slave *check_slave;
4520
4521 read_lock(&bond->lock);
4522
4523 if (!BOND_IS_OK(bond) || !skb->queue_mapping)
4524 goto out;
4525
4526 /* Find out if any slaves have the same mapping as this skb. */
4527 bond_for_each_slave(bond, check_slave, i) {
4528 if (check_slave->queue_id == skb->queue_mapping) {
4529 slave = check_slave;
4530 break;
4531 }
4532 }
4533
4534 /* If the slave isn't UP, use default transmit policy. */
4535 if (slave && slave->queue_id && IS_UP(slave->dev) &&
4536 (slave->link == BOND_LINK_UP)) {
4537 res = bond_dev_queue_xmit(bond, skb, slave->dev);
4538 }
4539
4540 out:
4541 read_unlock(&bond->lock);
4542 return res;
4543 }
4544
4545 static u16 bond_select_queue(struct net_device *dev, struct sk_buff *skb)
4546 {
4547 /*
4548 * This helper function exists to help dev_pick_tx get the correct
4549 * destination queue. Using a helper function skips the a call to
4550 * skb_tx_hash and will put the skbs in the queue we expect on their
4551 * way down to the bonding driver.
4552 */
4553 return skb->queue_mapping;
4554 }
4555
4556 static netdev_tx_t bond_start_xmit(struct sk_buff *skb, struct net_device *dev)
4557 {
4558 struct bonding *bond = netdev_priv(dev);
4559
4560 if (TX_QUEUE_OVERRIDE(bond->params.mode)) {
4561 if (!bond_slave_override(bond, skb))
4562 return NETDEV_TX_OK;
4563 }
4564
4565 switch (bond->params.mode) {
4566 case BOND_MODE_ROUNDROBIN:
4567 return bond_xmit_roundrobin(skb, dev);
4568 case BOND_MODE_ACTIVEBACKUP:
4569 return bond_xmit_activebackup(skb, dev);
4570 case BOND_MODE_XOR:
4571 return bond_xmit_xor(skb, dev);
4572 case BOND_MODE_BROADCAST:
4573 return bond_xmit_broadcast(skb, dev);
4574 case BOND_MODE_8023AD:
4575 return bond_3ad_xmit_xor(skb, dev);
4576 case BOND_MODE_ALB:
4577 case BOND_MODE_TLB:
4578 return bond_alb_xmit(skb, dev);
4579 default:
4580 /* Should never happen, mode already checked */
4581 pr_err("%s: Error: Unknown bonding mode %d\n",
4582 dev->name, bond->params.mode);
4583 WARN_ON_ONCE(1);
4584 dev_kfree_skb(skb);
4585 return NETDEV_TX_OK;
4586 }
4587 }
4588
4589
4590 /*
4591 * set bond mode specific net device operations
4592 */
4593 void bond_set_mode_ops(struct bonding *bond, int mode)
4594 {
4595 struct net_device *bond_dev = bond->dev;
4596
4597 switch (mode) {
4598 case BOND_MODE_ROUNDROBIN:
4599 break;
4600 case BOND_MODE_ACTIVEBACKUP:
4601 break;
4602 case BOND_MODE_XOR:
4603 bond_set_xmit_hash_policy(bond);
4604 break;
4605 case BOND_MODE_BROADCAST:
4606 break;
4607 case BOND_MODE_8023AD:
4608 bond_set_master_3ad_flags(bond);
4609 bond_set_xmit_hash_policy(bond);
4610 break;
4611 case BOND_MODE_ALB:
4612 bond_set_master_alb_flags(bond);
4613 /* FALLTHRU */
4614 case BOND_MODE_TLB:
4615 break;
4616 default:
4617 /* Should never happen, mode already checked */
4618 pr_err("%s: Error: Unknown bonding mode %d\n",
4619 bond_dev->name, mode);
4620 break;
4621 }
4622 }
4623
4624 static void bond_ethtool_get_drvinfo(struct net_device *bond_dev,
4625 struct ethtool_drvinfo *drvinfo)
4626 {
4627 strncpy(drvinfo->driver, DRV_NAME, 32);
4628 strncpy(drvinfo->version, DRV_VERSION, 32);
4629 snprintf(drvinfo->fw_version, 32, "%d", BOND_ABI_VERSION);
4630 }
4631
4632 static const struct ethtool_ops bond_ethtool_ops = {
4633 .get_drvinfo = bond_ethtool_get_drvinfo,
4634 .get_link = ethtool_op_get_link,
4635 .get_tx_csum = ethtool_op_get_tx_csum,
4636 .get_sg = ethtool_op_get_sg,
4637 .get_tso = ethtool_op_get_tso,
4638 .get_ufo = ethtool_op_get_ufo,
4639 .get_flags = ethtool_op_get_flags,
4640 };
4641
4642 static const struct net_device_ops bond_netdev_ops = {
4643 .ndo_init = bond_init,
4644 .ndo_uninit = bond_uninit,
4645 .ndo_open = bond_open,
4646 .ndo_stop = bond_close,
4647 .ndo_start_xmit = bond_start_xmit,
4648 .ndo_select_queue = bond_select_queue,
4649 .ndo_get_stats64 = bond_get_stats,
4650 .ndo_do_ioctl = bond_do_ioctl,
4651 .ndo_set_multicast_list = bond_set_multicast_list,
4652 .ndo_change_mtu = bond_change_mtu,
4653 .ndo_set_mac_address = bond_set_mac_address,
4654 .ndo_neigh_setup = bond_neigh_setup,
4655 .ndo_vlan_rx_register = bond_vlan_rx_register,
4656 .ndo_vlan_rx_add_vid = bond_vlan_rx_add_vid,
4657 .ndo_vlan_rx_kill_vid = bond_vlan_rx_kill_vid,
4658 #ifdef CONFIG_NET_POLL_CONTROLLER
4659 .ndo_netpoll_cleanup = bond_netpoll_cleanup,
4660 .ndo_poll_controller = bond_poll_controller,
4661 #endif
4662 };
4663
4664 static void bond_destructor(struct net_device *bond_dev)
4665 {
4666 struct bonding *bond = netdev_priv(bond_dev);
4667 if (bond->wq)
4668 destroy_workqueue(bond->wq);
4669 free_netdev(bond_dev);
4670 }
4671
4672 static void bond_setup(struct net_device *bond_dev)
4673 {
4674 struct bonding *bond = netdev_priv(bond_dev);
4675
4676 /* initialize rwlocks */
4677 rwlock_init(&bond->lock);
4678 rwlock_init(&bond->curr_slave_lock);
4679
4680 bond->params = bonding_defaults;
4681
4682 /* Initialize pointers */
4683 bond->dev = bond_dev;
4684 INIT_LIST_HEAD(&bond->vlan_list);
4685
4686 /* Initialize the device entry points */
4687 ether_setup(bond_dev);
4688 bond_dev->netdev_ops = &bond_netdev_ops;
4689 bond_dev->ethtool_ops = &bond_ethtool_ops;
4690 bond_set_mode_ops(bond, bond->params.mode);
4691
4692 bond_dev->destructor = bond_destructor;
4693
4694 /* Initialize the device options */
4695 bond_dev->tx_queue_len = 0;
4696 bond_dev->flags |= IFF_MASTER|IFF_MULTICAST;
4697 bond_dev->priv_flags |= IFF_BONDING;
4698 bond_dev->priv_flags &= ~IFF_XMIT_DST_RELEASE;
4699
4700 if (bond->params.arp_interval)
4701 bond_dev->priv_flags |= IFF_MASTER_ARPMON;
4702
4703 /* At first, we block adding VLANs. That's the only way to
4704 * prevent problems that occur when adding VLANs over an
4705 * empty bond. The block will be removed once non-challenged
4706 * slaves are enslaved.
4707 */
4708 bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
4709
4710 /* don't acquire bond device's netif_tx_lock when
4711 * transmitting */
4712 bond_dev->features |= NETIF_F_LLTX;
4713
4714 /* By default, we declare the bond to be fully
4715 * VLAN hardware accelerated capable. Special
4716 * care is taken in the various xmit functions
4717 * when there are slaves that are not hw accel
4718 * capable
4719 */
4720 bond_dev->features |= (NETIF_F_HW_VLAN_TX |
4721 NETIF_F_HW_VLAN_RX |
4722 NETIF_F_HW_VLAN_FILTER);
4723
4724 /* By default, we enable GRO on bonding devices.
4725 * Actual support requires lowlevel drivers are GRO ready.
4726 */
4727 bond_dev->features |= NETIF_F_GRO;
4728 }
4729
4730 static void bond_work_cancel_all(struct bonding *bond)
4731 {
4732 write_lock_bh(&bond->lock);
4733 bond->kill_timers = 1;
4734 write_unlock_bh(&bond->lock);
4735
4736 if (bond->params.miimon && delayed_work_pending(&bond->mii_work))
4737 cancel_delayed_work(&bond->mii_work);
4738
4739 if (bond->params.arp_interval && delayed_work_pending(&bond->arp_work))
4740 cancel_delayed_work(&bond->arp_work);
4741
4742 if (bond->params.mode == BOND_MODE_ALB &&
4743 delayed_work_pending(&bond->alb_work))
4744 cancel_delayed_work(&bond->alb_work);
4745
4746 if (bond->params.mode == BOND_MODE_8023AD &&
4747 delayed_work_pending(&bond->ad_work))
4748 cancel_delayed_work(&bond->ad_work);
4749
4750 if (delayed_work_pending(&bond->mcast_work))
4751 cancel_delayed_work(&bond->mcast_work);
4752 }
4753
4754 /*
4755 * Destroy a bonding device.
4756 * Must be under rtnl_lock when this function is called.
4757 */
4758 static void bond_uninit(struct net_device *bond_dev)
4759 {
4760 struct bonding *bond = netdev_priv(bond_dev);
4761 struct vlan_entry *vlan, *tmp;
4762
4763 bond_netpoll_cleanup(bond_dev);
4764
4765 /* Release the bonded slaves */
4766 bond_release_all(bond_dev);
4767
4768 list_del(&bond->bond_list);
4769
4770 bond_work_cancel_all(bond);
4771
4772 bond_remove_proc_entry(bond);
4773
4774 __hw_addr_flush(&bond->mc_list);
4775
4776 list_for_each_entry_safe(vlan, tmp, &bond->vlan_list, vlan_list) {
4777 list_del(&vlan->vlan_list);
4778 kfree(vlan);
4779 }
4780 }
4781
4782 /*------------------------- Module initialization ---------------------------*/
4783
4784 /*
4785 * Convert string input module parms. Accept either the
4786 * number of the mode or its string name. A bit complicated because
4787 * some mode names are substrings of other names, and calls from sysfs
4788 * may have whitespace in the name (trailing newlines, for example).
4789 */
4790 int bond_parse_parm(const char *buf, const struct bond_parm_tbl *tbl)
4791 {
4792 int modeint = -1, i, rv;
4793 char *p, modestr[BOND_MAX_MODENAME_LEN + 1] = { 0, };
4794
4795 for (p = (char *)buf; *p; p++)
4796 if (!(isdigit(*p) || isspace(*p)))
4797 break;
4798
4799 if (*p)
4800 rv = sscanf(buf, "%20s", modestr);
4801 else
4802 rv = sscanf(buf, "%d", &modeint);
4803
4804 if (!rv)
4805 return -1;
4806
4807 for (i = 0; tbl[i].modename; i++) {
4808 if (modeint == tbl[i].mode)
4809 return tbl[i].mode;
4810 if (strcmp(modestr, tbl[i].modename) == 0)
4811 return tbl[i].mode;
4812 }
4813
4814 return -1;
4815 }
4816
4817 static int bond_check_params(struct bond_params *params)
4818 {
4819 int arp_validate_value, fail_over_mac_value, primary_reselect_value;
4820
4821 /*
4822 * Convert string parameters.
4823 */
4824 if (mode) {
4825 bond_mode = bond_parse_parm(mode, bond_mode_tbl);
4826 if (bond_mode == -1) {
4827 pr_err("Error: Invalid bonding mode \"%s\"\n",
4828 mode == NULL ? "NULL" : mode);
4829 return -EINVAL;
4830 }
4831 }
4832
4833 if (xmit_hash_policy) {
4834 if ((bond_mode != BOND_MODE_XOR) &&
4835 (bond_mode != BOND_MODE_8023AD)) {
4836 pr_info("xmit_hash_policy param is irrelevant in mode %s\n",
4837 bond_mode_name(bond_mode));
4838 } else {
4839 xmit_hashtype = bond_parse_parm(xmit_hash_policy,
4840 xmit_hashtype_tbl);
4841 if (xmit_hashtype == -1) {
4842 pr_err("Error: Invalid xmit_hash_policy \"%s\"\n",
4843 xmit_hash_policy == NULL ? "NULL" :
4844 xmit_hash_policy);
4845 return -EINVAL;
4846 }
4847 }
4848 }
4849
4850 if (lacp_rate) {
4851 if (bond_mode != BOND_MODE_8023AD) {
4852 pr_info("lacp_rate param is irrelevant in mode %s\n",
4853 bond_mode_name(bond_mode));
4854 } else {
4855 lacp_fast = bond_parse_parm(lacp_rate, bond_lacp_tbl);
4856 if (lacp_fast == -1) {
4857 pr_err("Error: Invalid lacp rate \"%s\"\n",
4858 lacp_rate == NULL ? "NULL" : lacp_rate);
4859 return -EINVAL;
4860 }
4861 }
4862 }
4863
4864 if (ad_select) {
4865 params->ad_select = bond_parse_parm(ad_select, ad_select_tbl);
4866 if (params->ad_select == -1) {
4867 pr_err("Error: Invalid ad_select \"%s\"\n",
4868 ad_select == NULL ? "NULL" : ad_select);
4869 return -EINVAL;
4870 }
4871
4872 if (bond_mode != BOND_MODE_8023AD) {
4873 pr_warning("ad_select param only affects 802.3ad mode\n");
4874 }
4875 } else {
4876 params->ad_select = BOND_AD_STABLE;
4877 }
4878
4879 if (max_bonds < 0) {
4880 pr_warning("Warning: max_bonds (%d) not in range %d-%d, so it was reset to BOND_DEFAULT_MAX_BONDS (%d)\n",
4881 max_bonds, 0, INT_MAX, BOND_DEFAULT_MAX_BONDS);
4882 max_bonds = BOND_DEFAULT_MAX_BONDS;
4883 }
4884
4885 if (miimon < 0) {
4886 pr_warning("Warning: miimon module parameter (%d), not in range 0-%d, so it was reset to %d\n",
4887 miimon, INT_MAX, BOND_LINK_MON_INTERV);
4888 miimon = BOND_LINK_MON_INTERV;
4889 }
4890
4891 if (updelay < 0) {
4892 pr_warning("Warning: updelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4893 updelay, INT_MAX);
4894 updelay = 0;
4895 }
4896
4897 if (downdelay < 0) {
4898 pr_warning("Warning: downdelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4899 downdelay, INT_MAX);
4900 downdelay = 0;
4901 }
4902
4903 if ((use_carrier != 0) && (use_carrier != 1)) {
4904 pr_warning("Warning: use_carrier module parameter (%d), not of valid value (0/1), so it was set to 1\n",
4905 use_carrier);
4906 use_carrier = 1;
4907 }
4908
4909 if (num_grat_arp < 0 || num_grat_arp > 255) {
4910 pr_warning("Warning: num_grat_arp (%d) not in range 0-255 so it was reset to 1\n",
4911 num_grat_arp);
4912 num_grat_arp = 1;
4913 }
4914
4915 if (num_unsol_na < 0 || num_unsol_na > 255) {
4916 pr_warning("Warning: num_unsol_na (%d) not in range 0-255 so it was reset to 1\n",
4917 num_unsol_na);
4918 num_unsol_na = 1;
4919 }
4920
4921 /* reset values for 802.3ad */
4922 if (bond_mode == BOND_MODE_8023AD) {
4923 if (!miimon) {
4924 pr_warning("Warning: miimon must be specified, otherwise bonding will not detect link failure, speed and duplex which are essential for 802.3ad operation\n");
4925 pr_warning("Forcing miimon to 100msec\n");
4926 miimon = 100;
4927 }
4928 }
4929
4930 if (tx_queues < 1 || tx_queues > 255) {
4931 pr_warning("Warning: tx_queues (%d) should be between "
4932 "1 and 255, resetting to %d\n",
4933 tx_queues, BOND_DEFAULT_TX_QUEUES);
4934 tx_queues = BOND_DEFAULT_TX_QUEUES;
4935 }
4936
4937 if ((all_slaves_active != 0) && (all_slaves_active != 1)) {
4938 pr_warning("Warning: all_slaves_active module parameter (%d), "
4939 "not of valid value (0/1), so it was set to "
4940 "0\n", all_slaves_active);
4941 all_slaves_active = 0;
4942 }
4943
4944 if (resend_igmp < 0 || resend_igmp > 255) {
4945 pr_warning("Warning: resend_igmp (%d) should be between "
4946 "0 and 255, resetting to %d\n",
4947 resend_igmp, BOND_DEFAULT_RESEND_IGMP);
4948 resend_igmp = BOND_DEFAULT_RESEND_IGMP;
4949 }
4950
4951 /* reset values for TLB/ALB */
4952 if ((bond_mode == BOND_MODE_TLB) ||
4953 (bond_mode == BOND_MODE_ALB)) {
4954 if (!miimon) {
4955 pr_warning("Warning: miimon must be specified, otherwise bonding will not detect link failure and link speed which are essential for TLB/ALB load balancing\n");
4956 pr_warning("Forcing miimon to 100msec\n");
4957 miimon = 100;
4958 }
4959 }
4960
4961 if (bond_mode == BOND_MODE_ALB) {
4962 pr_notice("In ALB mode you might experience client disconnections upon reconnection of a link if the bonding module updelay parameter (%d msec) is incompatible with the forwarding delay time of the switch\n",
4963 updelay);
4964 }
4965
4966 if (!miimon) {
4967 if (updelay || downdelay) {
4968 /* just warn the user the up/down delay will have
4969 * no effect since miimon is zero...
4970 */
4971 pr_warning("Warning: miimon module parameter not set and updelay (%d) or downdelay (%d) module parameter is set; updelay and downdelay have no effect unless miimon is set\n",
4972 updelay, downdelay);
4973 }
4974 } else {
4975 /* don't allow arp monitoring */
4976 if (arp_interval) {
4977 pr_warning("Warning: miimon (%d) and arp_interval (%d) can't be used simultaneously, disabling ARP monitoring\n",
4978 miimon, arp_interval);
4979 arp_interval = 0;
4980 }
4981
4982 if ((updelay % miimon) != 0) {
4983 pr_warning("Warning: updelay (%d) is not a multiple of miimon (%d), updelay rounded to %d ms\n",
4984 updelay, miimon,
4985 (updelay / miimon) * miimon);
4986 }
4987
4988 updelay /= miimon;
4989
4990 if ((downdelay % miimon) != 0) {
4991 pr_warning("Warning: downdelay (%d) is not a multiple of miimon (%d), downdelay rounded to %d ms\n",
4992 downdelay, miimon,
4993 (downdelay / miimon) * miimon);
4994 }
4995
4996 downdelay /= miimon;
4997 }
4998
4999 if (arp_interval < 0) {
5000 pr_warning("Warning: arp_interval module parameter (%d) , not in range 0-%d, so it was reset to %d\n",
5001 arp_interval, INT_MAX, BOND_LINK_ARP_INTERV);
5002 arp_interval = BOND_LINK_ARP_INTERV;
5003 }
5004
5005 for (arp_ip_count = 0;
5006 (arp_ip_count < BOND_MAX_ARP_TARGETS) && arp_ip_target[arp_ip_count];
5007 arp_ip_count++) {
5008 /* not complete check, but should be good enough to
5009 catch mistakes */
5010 if (!isdigit(arp_ip_target[arp_ip_count][0])) {
5011 pr_warning("Warning: bad arp_ip_target module parameter (%s), ARP monitoring will not be performed\n",
5012 arp_ip_target[arp_ip_count]);
5013 arp_interval = 0;
5014 } else {
5015 __be32 ip = in_aton(arp_ip_target[arp_ip_count]);
5016 arp_target[arp_ip_count] = ip;
5017 }
5018 }
5019
5020 if (arp_interval && !arp_ip_count) {
5021 /* don't allow arping if no arp_ip_target given... */
5022 pr_warning("Warning: arp_interval module parameter (%d) specified without providing an arp_ip_target parameter, arp_interval was reset to 0\n",
5023 arp_interval);
5024 arp_interval = 0;
5025 }
5026
5027 if (arp_validate) {
5028 if (bond_mode != BOND_MODE_ACTIVEBACKUP) {
5029 pr_err("arp_validate only supported in active-backup mode\n");
5030 return -EINVAL;
5031 }
5032 if (!arp_interval) {
5033 pr_err("arp_validate requires arp_interval\n");
5034 return -EINVAL;
5035 }
5036
5037 arp_validate_value = bond_parse_parm(arp_validate,
5038 arp_validate_tbl);
5039 if (arp_validate_value == -1) {
5040 pr_err("Error: invalid arp_validate \"%s\"\n",
5041 arp_validate == NULL ? "NULL" : arp_validate);
5042 return -EINVAL;
5043 }
5044 } else
5045 arp_validate_value = 0;
5046
5047 if (miimon) {
5048 pr_info("MII link monitoring set to %d ms\n", miimon);
5049 } else if (arp_interval) {
5050 int i;
5051
5052 pr_info("ARP monitoring set to %d ms, validate %s, with %d target(s):",
5053 arp_interval,
5054 arp_validate_tbl[arp_validate_value].modename,
5055 arp_ip_count);
5056
5057 for (i = 0; i < arp_ip_count; i++)
5058 pr_info(" %s", arp_ip_target[i]);
5059
5060 pr_info("\n");
5061
5062 } else if (max_bonds) {
5063 /* miimon and arp_interval not set, we need one so things
5064 * work as expected, see bonding.txt for details
5065 */
5066 pr_warning("Warning: either miimon or arp_interval and arp_ip_target module parameters must be specified, otherwise bonding will not detect link failures! see bonding.txt for details.\n");
5067 }
5068
5069 if (primary && !USES_PRIMARY(bond_mode)) {
5070 /* currently, using a primary only makes sense
5071 * in active backup, TLB or ALB modes
5072 */
5073 pr_warning("Warning: %s primary device specified but has no effect in %s mode\n",
5074 primary, bond_mode_name(bond_mode));
5075 primary = NULL;
5076 }
5077
5078 if (primary && primary_reselect) {
5079 primary_reselect_value = bond_parse_parm(primary_reselect,
5080 pri_reselect_tbl);
5081 if (primary_reselect_value == -1) {
5082 pr_err("Error: Invalid primary_reselect \"%s\"\n",
5083 primary_reselect ==
5084 NULL ? "NULL" : primary_reselect);
5085 return -EINVAL;
5086 }
5087 } else {
5088 primary_reselect_value = BOND_PRI_RESELECT_ALWAYS;
5089 }
5090
5091 if (fail_over_mac) {
5092 fail_over_mac_value = bond_parse_parm(fail_over_mac,
5093 fail_over_mac_tbl);
5094 if (fail_over_mac_value == -1) {
5095 pr_err("Error: invalid fail_over_mac \"%s\"\n",
5096 arp_validate == NULL ? "NULL" : arp_validate);
5097 return -EINVAL;
5098 }
5099
5100 if (bond_mode != BOND_MODE_ACTIVEBACKUP)
5101 pr_warning("Warning: fail_over_mac only affects active-backup mode.\n");
5102 } else {
5103 fail_over_mac_value = BOND_FOM_NONE;
5104 }
5105
5106 /* fill params struct with the proper values */
5107 params->mode = bond_mode;
5108 params->xmit_policy = xmit_hashtype;
5109 params->miimon = miimon;
5110 params->num_grat_arp = num_grat_arp;
5111 params->num_unsol_na = num_unsol_na;
5112 params->arp_interval = arp_interval;
5113 params->arp_validate = arp_validate_value;
5114 params->updelay = updelay;
5115 params->downdelay = downdelay;
5116 params->use_carrier = use_carrier;
5117 params->lacp_fast = lacp_fast;
5118 params->primary[0] = 0;
5119 params->primary_reselect = primary_reselect_value;
5120 params->fail_over_mac = fail_over_mac_value;
5121 params->tx_queues = tx_queues;
5122 params->all_slaves_active = all_slaves_active;
5123 params->resend_igmp = resend_igmp;
5124
5125 if (primary) {
5126 strncpy(params->primary, primary, IFNAMSIZ);
5127 params->primary[IFNAMSIZ - 1] = 0;
5128 }
5129
5130 memcpy(params->arp_targets, arp_target, sizeof(arp_target));
5131
5132 return 0;
5133 }
5134
5135 static struct lock_class_key bonding_netdev_xmit_lock_key;
5136 static struct lock_class_key bonding_netdev_addr_lock_key;
5137
5138 static void bond_set_lockdep_class_one(struct net_device *dev,
5139 struct netdev_queue *txq,
5140 void *_unused)
5141 {
5142 lockdep_set_class(&txq->_xmit_lock,
5143 &bonding_netdev_xmit_lock_key);
5144 }
5145
5146 static void bond_set_lockdep_class(struct net_device *dev)
5147 {
5148 lockdep_set_class(&dev->addr_list_lock,
5149 &bonding_netdev_addr_lock_key);
5150 netdev_for_each_tx_queue(dev, bond_set_lockdep_class_one, NULL);
5151 }
5152
5153 /*
5154 * Called from registration process
5155 */
5156 static int bond_init(struct net_device *bond_dev)
5157 {
5158 struct bonding *bond = netdev_priv(bond_dev);
5159 struct bond_net *bn = net_generic(dev_net(bond_dev), bond_net_id);
5160
5161 pr_debug("Begin bond_init for %s\n", bond_dev->name);
5162
5163 bond->wq = create_singlethread_workqueue(bond_dev->name);
5164 if (!bond->wq)
5165 return -ENOMEM;
5166
5167 bond_set_lockdep_class(bond_dev);
5168
5169 netif_carrier_off(bond_dev);
5170
5171 bond_create_proc_entry(bond);
5172 list_add_tail(&bond->bond_list, &bn->dev_list);
5173
5174 bond_prepare_sysfs_group(bond);
5175
5176 __hw_addr_init(&bond->mc_list);
5177 return 0;
5178 }
5179
5180 static int bond_validate(struct nlattr *tb[], struct nlattr *data[])
5181 {
5182 if (tb[IFLA_ADDRESS]) {
5183 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN)
5184 return -EINVAL;
5185 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS])))
5186 return -EADDRNOTAVAIL;
5187 }
5188 return 0;
5189 }
5190
5191 static struct rtnl_link_ops bond_link_ops __read_mostly = {
5192 .kind = "bond",
5193 .priv_size = sizeof(struct bonding),
5194 .setup = bond_setup,
5195 .validate = bond_validate,
5196 };
5197
5198 /* Create a new bond based on the specified name and bonding parameters.
5199 * If name is NULL, obtain a suitable "bond%d" name for us.
5200 * Caller must NOT hold rtnl_lock; we need to release it here before we
5201 * set up our sysfs entries.
5202 */
5203 int bond_create(struct net *net, const char *name)
5204 {
5205 struct net_device *bond_dev;
5206 int res;
5207
5208 rtnl_lock();
5209
5210 bond_dev = alloc_netdev_mq(sizeof(struct bonding), name ? name : "",
5211 bond_setup, tx_queues);
5212 if (!bond_dev) {
5213 pr_err("%s: eek! can't alloc netdev!\n", name);
5214 rtnl_unlock();
5215 return -ENOMEM;
5216 }
5217
5218 dev_net_set(bond_dev, net);
5219 bond_dev->rtnl_link_ops = &bond_link_ops;
5220
5221 if (!name) {
5222 res = dev_alloc_name(bond_dev, "bond%d");
5223 if (res < 0)
5224 goto out;
5225 }
5226
5227 res = register_netdevice(bond_dev);
5228
5229 out:
5230 rtnl_unlock();
5231 if (res < 0)
5232 bond_destructor(bond_dev);
5233 return res;
5234 }
5235
5236 static int __net_init bond_net_init(struct net *net)
5237 {
5238 struct bond_net *bn = net_generic(net, bond_net_id);
5239
5240 bn->net = net;
5241 INIT_LIST_HEAD(&bn->dev_list);
5242
5243 bond_create_proc_dir(bn);
5244
5245 return 0;
5246 }
5247
5248 static void __net_exit bond_net_exit(struct net *net)
5249 {
5250 struct bond_net *bn = net_generic(net, bond_net_id);
5251
5252 bond_destroy_proc_dir(bn);
5253 }
5254
5255 static struct pernet_operations bond_net_ops = {
5256 .init = bond_net_init,
5257 .exit = bond_net_exit,
5258 .id = &bond_net_id,
5259 .size = sizeof(struct bond_net),
5260 };
5261
5262 static int __init bonding_init(void)
5263 {
5264 int i;
5265 int res;
5266
5267 pr_info("%s", version);
5268
5269 res = bond_check_params(&bonding_defaults);
5270 if (res)
5271 goto out;
5272
5273 res = register_pernet_subsys(&bond_net_ops);
5274 if (res)
5275 goto out;
5276
5277 res = rtnl_link_register(&bond_link_ops);
5278 if (res)
5279 goto err_link;
5280
5281 for (i = 0; i < max_bonds; i++) {
5282 res = bond_create(&init_net, NULL);
5283 if (res)
5284 goto err;
5285 }
5286
5287 res = bond_create_sysfs();
5288 if (res)
5289 goto err;
5290
5291 register_netdevice_notifier(&bond_netdev_notifier);
5292 register_inetaddr_notifier(&bond_inetaddr_notifier);
5293 bond_register_ipv6_notifier();
5294 out:
5295 return res;
5296 err:
5297 rtnl_link_unregister(&bond_link_ops);
5298 err_link:
5299 unregister_pernet_subsys(&bond_net_ops);
5300 goto out;
5301
5302 }
5303
5304 static void __exit bonding_exit(void)
5305 {
5306 unregister_netdevice_notifier(&bond_netdev_notifier);
5307 unregister_inetaddr_notifier(&bond_inetaddr_notifier);
5308 bond_unregister_ipv6_notifier();
5309
5310 bond_destroy_sysfs();
5311
5312 rtnl_link_unregister(&bond_link_ops);
5313 unregister_pernet_subsys(&bond_net_ops);
5314 }
5315
5316 module_init(bonding_init);
5317 module_exit(bonding_exit);
5318 MODULE_LICENSE("GPL");
5319 MODULE_VERSION(DRV_VERSION);
5320 MODULE_DESCRIPTION(DRV_DESCRIPTION ", v" DRV_VERSION);
5321 MODULE_AUTHOR("Thomas Davis, tadavis@lbl.gov and many others");
5322 MODULE_ALIAS_RTNL_LINK("bond");
This page took 0.144516 seconds and 5 git commands to generate.