]> bbs.cooldavid.org Git - net-next-2.6.git/blame_incremental - drivers/net/bonding/bond_main.c
git://bbs.cooldavid.org / net-next-2.6.git / blame_incremental
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (non-incremental) | history | HEAD
xfrm: make xfrm_bundle_ok local
[net-next-2.6.git] / drivers / net / bonding / bond_main.c
... / ...
CommitLineData
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 <linux/preempt.h>
80#include <net/route.h>
81#include <net/net_namespace.h>
82#include <net/netns/generic.h>
83#include "bonding.h"
84#include "bond_3ad.h"
85#include "bond_alb.h"
86
87/*---------------------------- Module parameters ----------------------------*/
88
89/* monitor all links that often (in milliseconds). <=0 disables monitoring */
90#define BOND_LINK_MON_INTERV 0
91#define BOND_LINK_ARP_INTERV 0
92
93static int max_bonds = BOND_DEFAULT_MAX_BONDS;
94static int tx_queues = BOND_DEFAULT_TX_QUEUES;
95static int num_grat_arp = 1;
96static int num_unsol_na = 1;
97static int miimon = BOND_LINK_MON_INTERV;
98static int updelay;
99static int downdelay;
100static int use_carrier = 1;
101static char *mode;
102static char *primary;
103static char *primary_reselect;
104static char *lacp_rate;
105static char *ad_select;
106static char *xmit_hash_policy;
107static int arp_interval = BOND_LINK_ARP_INTERV;
108static char *arp_ip_target[BOND_MAX_ARP_TARGETS];
109static char *arp_validate;
110static char *fail_over_mac;
111static int all_slaves_active = 0;
112static struct bond_params bonding_defaults;
113static int resend_igmp = BOND_DEFAULT_RESEND_IGMP;
114
115module_param(max_bonds, int, 0);
116MODULE_PARM_DESC(max_bonds, "Max number of bonded devices");
117module_param(tx_queues, int, 0);
118MODULE_PARM_DESC(tx_queues, "Max number of transmit queues (default = 16)");
119module_param(num_grat_arp, int, 0644);
120MODULE_PARM_DESC(num_grat_arp, "Number of gratuitous ARP packets to send on failover event");
121module_param(num_unsol_na, int, 0644);
122MODULE_PARM_DESC(num_unsol_na, "Number of unsolicited IPv6 Neighbor Advertisements packets to send on failover event");
123module_param(miimon, int, 0);
124MODULE_PARM_DESC(miimon, "Link check interval in milliseconds");
125module_param(updelay, int, 0);
126MODULE_PARM_DESC(updelay, "Delay before considering link up, in milliseconds");
127module_param(downdelay, int, 0);
128MODULE_PARM_DESC(downdelay, "Delay before considering link down, "
129 "in milliseconds");
130module_param(use_carrier, int, 0);
131MODULE_PARM_DESC(use_carrier, "Use netif_carrier_ok (vs MII ioctls) in miimon; "
132 "0 for off, 1 for on (default)");
133module_param(mode, charp, 0);
134MODULE_PARM_DESC(mode, "Mode of operation : 0 for balance-rr, "
135 "1 for active-backup, 2 for balance-xor, "
136 "3 for broadcast, 4 for 802.3ad, 5 for balance-tlb, "
137 "6 for balance-alb");
138module_param(primary, charp, 0);
139MODULE_PARM_DESC(primary, "Primary network device to use");
140module_param(primary_reselect, charp, 0);
141MODULE_PARM_DESC(primary_reselect, "Reselect primary slave "
142 "once it comes up; "
143 "0 for always (default), "
144 "1 for only if speed of primary is "
145 "better, "
146 "2 for only on active slave "
147 "failure");
148module_param(lacp_rate, charp, 0);
149MODULE_PARM_DESC(lacp_rate, "LACPDU tx rate to request from 802.3ad partner "
150 "(slow/fast)");
151module_param(ad_select, charp, 0);
152MODULE_PARM_DESC(ad_select, "803.ad aggregation selection logic: stable (0, default), bandwidth (1), count (2)");
153module_param(xmit_hash_policy, charp, 0);
154MODULE_PARM_DESC(xmit_hash_policy, "XOR hashing method: 0 for layer 2 (default)"
155 ", 1 for layer 3+4");
156module_param(arp_interval, int, 0);
157MODULE_PARM_DESC(arp_interval, "arp interval in milliseconds");
158module_param_array(arp_ip_target, charp, NULL, 0);
159MODULE_PARM_DESC(arp_ip_target, "arp targets in n.n.n.n form");
160module_param(arp_validate, charp, 0);
161MODULE_PARM_DESC(arp_validate, "validate src/dst of ARP probes: none (default), active, backup or all");
162module_param(fail_over_mac, charp, 0);
163MODULE_PARM_DESC(fail_over_mac, "For active-backup, do not set all slaves to the same MAC. none (default), active or follow");
164module_param(all_slaves_active, int, 0);
165MODULE_PARM_DESC(all_slaves_active, "Keep all frames received on an interface"
166 "by setting active flag for all slaves. "
167 "0 for never (default), 1 for always.");
168module_param(resend_igmp, int, 0);
169MODULE_PARM_DESC(resend_igmp, "Number of IGMP membership reports to send on link failure");
170
171/*----------------------------- Global variables ----------------------------*/
172
173#ifdef CONFIG_NET_POLL_CONTROLLER
174cpumask_var_t netpoll_block_tx;
175#endif
176
177static const char * const version =
178 DRV_DESCRIPTION ": v" DRV_VERSION " (" DRV_RELDATE ")\n";
179
180int bond_net_id __read_mostly;
181
182static __be32 arp_target[BOND_MAX_ARP_TARGETS];
183static int arp_ip_count;
184static int bond_mode = BOND_MODE_ROUNDROBIN;
185static int xmit_hashtype = BOND_XMIT_POLICY_LAYER2;
186static int lacp_fast;
187
188const struct bond_parm_tbl bond_lacp_tbl[] = {
189{ "slow", AD_LACP_SLOW},
190{ "fast", AD_LACP_FAST},
191{ NULL, -1},
192};
193
194const struct bond_parm_tbl bond_mode_tbl[] = {
195{ "balance-rr", BOND_MODE_ROUNDROBIN},
196{ "active-backup", BOND_MODE_ACTIVEBACKUP},
197{ "balance-xor", BOND_MODE_XOR},
198{ "broadcast", BOND_MODE_BROADCAST},
199{ "802.3ad", BOND_MODE_8023AD},
200{ "balance-tlb", BOND_MODE_TLB},
201{ "balance-alb", BOND_MODE_ALB},
202{ NULL, -1},
203};
204
205const struct bond_parm_tbl xmit_hashtype_tbl[] = {
206{ "layer2", BOND_XMIT_POLICY_LAYER2},
207{ "layer3+4", BOND_XMIT_POLICY_LAYER34},
208{ "layer2+3", BOND_XMIT_POLICY_LAYER23},
209{ NULL, -1},
210};
211
212const struct bond_parm_tbl arp_validate_tbl[] = {
213{ "none", BOND_ARP_VALIDATE_NONE},
214{ "active", BOND_ARP_VALIDATE_ACTIVE},
215{ "backup", BOND_ARP_VALIDATE_BACKUP},
216{ "all", BOND_ARP_VALIDATE_ALL},
217{ NULL, -1},
218};
219
220const struct bond_parm_tbl fail_over_mac_tbl[] = {
221{ "none", BOND_FOM_NONE},
222{ "active", BOND_FOM_ACTIVE},
223{ "follow", BOND_FOM_FOLLOW},
224{ NULL, -1},
225};
226
227const struct bond_parm_tbl pri_reselect_tbl[] = {
228{ "always", BOND_PRI_RESELECT_ALWAYS},
229{ "better", BOND_PRI_RESELECT_BETTER},
230{ "failure", BOND_PRI_RESELECT_FAILURE},
231{ NULL, -1},
232};
233
234struct bond_parm_tbl ad_select_tbl[] = {
235{ "stable", BOND_AD_STABLE},
236{ "bandwidth", BOND_AD_BANDWIDTH},
237{ "count", BOND_AD_COUNT},
238{ NULL, -1},
239};
240
241/*-------------------------- Forward declarations ---------------------------*/
242
243static void bond_send_gratuitous_arp(struct bonding *bond);
244static int bond_init(struct net_device *bond_dev);
245static void bond_uninit(struct net_device *bond_dev);
246
247/*---------------------------- General routines -----------------------------*/
248
249static const char *bond_mode_name(int mode)
250{
251 static const char *names[] = {
252 [BOND_MODE_ROUNDROBIN] = "load balancing (round-robin)",
253 [BOND_MODE_ACTIVEBACKUP] = "fault-tolerance (active-backup)",
254 [BOND_MODE_XOR] = "load balancing (xor)",
255 [BOND_MODE_BROADCAST] = "fault-tolerance (broadcast)",
256 [BOND_MODE_8023AD] = "IEEE 802.3ad Dynamic link aggregation",
257 [BOND_MODE_TLB] = "transmit load balancing",
258 [BOND_MODE_ALB] = "adaptive load balancing",
259 };
260
261 if (mode < 0 || mode > BOND_MODE_ALB)
262 return "unknown";
263
264 return names[mode];
265}
266
267/*---------------------------------- VLAN -----------------------------------*/
268
269/**
270 * bond_add_vlan - add a new vlan id on bond
271 * @bond: bond that got the notification
272 * @vlan_id: the vlan id to add
273 *
274 * Returns -ENOMEM if allocation failed.
275 */
276static int bond_add_vlan(struct bonding *bond, unsigned short vlan_id)
277{
278 struct vlan_entry *vlan;
279
280 pr_debug("bond: %s, vlan id %d\n",
281 (bond ? bond->dev->name : "None"), vlan_id);
282
283 vlan = kzalloc(sizeof(struct vlan_entry), GFP_KERNEL);
284 if (!vlan)
285 return -ENOMEM;
286
287 INIT_LIST_HEAD(&vlan->vlan_list);
288 vlan->vlan_id = vlan_id;
289
290 write_lock_bh(&bond->lock);
291
292 list_add_tail(&vlan->vlan_list, &bond->vlan_list);
293
294 write_unlock_bh(&bond->lock);
295
296 pr_debug("added VLAN ID %d on bond %s\n", vlan_id, bond->dev->name);
297
298 return 0;
299}
300
301/**
302 * bond_del_vlan - delete a vlan id from bond
303 * @bond: bond that got the notification
304 * @vlan_id: the vlan id to delete
305 *
306 * returns -ENODEV if @vlan_id was not found in @bond.
307 */
308static int bond_del_vlan(struct bonding *bond, unsigned short vlan_id)
309{
310 struct vlan_entry *vlan;
311 int res = -ENODEV;
312
313 pr_debug("bond: %s, vlan id %d\n", bond->dev->name, vlan_id);
314
315 block_netpoll_tx();
316 write_lock_bh(&bond->lock);
317
318 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
319 if (vlan->vlan_id == vlan_id) {
320 list_del(&vlan->vlan_list);
321
322 if (bond_is_lb(bond))
323 bond_alb_clear_vlan(bond, vlan_id);
324
325 pr_debug("removed VLAN ID %d from bond %s\n",
326 vlan_id, bond->dev->name);
327
328 kfree(vlan);
329
330 if (list_empty(&bond->vlan_list) &&
331 (bond->slave_cnt == 0)) {
332 /* Last VLAN removed and no slaves, so
333 * restore block on adding VLANs. This will
334 * be removed once new slaves that are not
335 * VLAN challenged will be added.
336 */
337 bond->dev->features |= NETIF_F_VLAN_CHALLENGED;
338 }
339
340 res = 0;
341 goto out;
342 }
343 }
344
345 pr_debug("couldn't find VLAN ID %d in bond %s\n",
346 vlan_id, bond->dev->name);
347
348out:
349 write_unlock_bh(&bond->lock);
350 unblock_netpoll_tx();
351 return res;
352}
353
354/**
355 * bond_has_challenged_slaves
356 * @bond: the bond we're working on
357 *
358 * Searches the slave list. Returns 1 if a vlan challenged slave
359 * was found, 0 otherwise.
360 *
361 * Assumes bond->lock is held.
362 */
363static int bond_has_challenged_slaves(struct bonding *bond)
364{
365 struct slave *slave;
366 int i;
367
368 bond_for_each_slave(bond, slave, i) {
369 if (slave->dev->features & NETIF_F_VLAN_CHALLENGED) {
370 pr_debug("found VLAN challenged slave - %s\n",
371 slave->dev->name);
372 return 1;
373 }
374 }
375
376 pr_debug("no VLAN challenged slaves found\n");
377 return 0;
378}
379
380/**
381 * bond_next_vlan - safely skip to the next item in the vlans list.
382 * @bond: the bond we're working on
383 * @curr: item we're advancing from
384 *
385 * Returns %NULL if list is empty, bond->next_vlan if @curr is %NULL,
386 * or @curr->next otherwise (even if it is @curr itself again).
387 *
388 * Caller must hold bond->lock
389 */
390struct vlan_entry *bond_next_vlan(struct bonding *bond, struct vlan_entry *curr)
391{
392 struct vlan_entry *next, *last;
393
394 if (list_empty(&bond->vlan_list))
395 return NULL;
396
397 if (!curr) {
398 next = list_entry(bond->vlan_list.next,
399 struct vlan_entry, vlan_list);
400 } else {
401 last = list_entry(bond->vlan_list.prev,
402 struct vlan_entry, vlan_list);
403 if (last == curr) {
404 next = list_entry(bond->vlan_list.next,
405 struct vlan_entry, vlan_list);
406 } else {
407 next = list_entry(curr->vlan_list.next,
408 struct vlan_entry, vlan_list);
409 }
410 }
411
412 return next;
413}
414
415/**
416 * bond_dev_queue_xmit - Prepare skb for xmit.
417 *
418 * @bond: bond device that got this skb for tx.
419 * @skb: hw accel VLAN tagged skb to transmit
420 * @slave_dev: slave that is supposed to xmit this skbuff
421 *
422 * When the bond gets an skb to transmit that is
423 * already hardware accelerated VLAN tagged, and it
424 * needs to relay this skb to a slave that is not
425 * hw accel capable, the skb needs to be "unaccelerated",
426 * i.e. strip the hwaccel tag and re-insert it as part
427 * of the payload.
428 */
429int bond_dev_queue_xmit(struct bonding *bond, struct sk_buff *skb,
430 struct net_device *slave_dev)
431{
432 unsigned short uninitialized_var(vlan_id);
433
434 /* Test vlan_list not vlgrp to catch and handle 802.1p tags */
435 if (!list_empty(&bond->vlan_list) &&
436 !(slave_dev->features & NETIF_F_HW_VLAN_TX) &&
437 vlan_get_tag(skb, &vlan_id) == 0) {
438 skb->dev = slave_dev;
439 skb = vlan_put_tag(skb, vlan_id);
440 if (!skb) {
441 /* vlan_put_tag() frees the skb in case of error,
442 * so return success here so the calling functions
443 * won't attempt to free is again.
444 */
445 return 0;
446 }
447 } else {
448 skb->dev = slave_dev;
449 }
450
451 skb->priority = 1;
452#ifdef CONFIG_NET_POLL_CONTROLLER
453 if (unlikely(bond->dev->priv_flags & IFF_IN_NETPOLL)) {
454 struct netpoll *np = bond->dev->npinfo->netpoll;
455 slave_dev->npinfo = bond->dev->npinfo;
456 slave_dev->priv_flags |= IFF_IN_NETPOLL;
457 netpoll_send_skb_on_dev(np, skb, slave_dev);
458 slave_dev->priv_flags &= ~IFF_IN_NETPOLL;
459 } else
460#endif
461 dev_queue_xmit(skb);
462
463 return 0;
464}
465
466/*
467 * In the following 3 functions, bond_vlan_rx_register(), bond_vlan_rx_add_vid
468 * and bond_vlan_rx_kill_vid, We don't protect the slave list iteration with a
469 * lock because:
470 * a. This operation is performed in IOCTL context,
471 * b. The operation is protected by the RTNL semaphore in the 8021q code,
472 * c. Holding a lock with BH disabled while directly calling a base driver
473 * entry point is generally a BAD idea.
474 *
475 * The design of synchronization/protection for this operation in the 8021q
476 * module is good for one or more VLAN devices over a single physical device
477 * and cannot be extended for a teaming solution like bonding, so there is a
478 * potential race condition here where a net device from the vlan group might
479 * be referenced (either by a base driver or the 8021q code) while it is being
480 * removed from the system. However, it turns out we're not making matters
481 * worse, and if it works for regular VLAN usage it will work here too.
482*/
483
484/**
485 * bond_vlan_rx_register - Propagates registration to slaves
486 * @bond_dev: bonding net device that got called
487 * @grp: vlan group being registered
488 */
489static void bond_vlan_rx_register(struct net_device *bond_dev,
490 struct vlan_group *grp)
491{
492 struct bonding *bond = netdev_priv(bond_dev);
493 struct slave *slave;
494 int i;
495
496 write_lock(&bond->lock);
497 bond->vlgrp = grp;
498 write_unlock(&bond->lock);
499
500 bond_for_each_slave(bond, slave, i) {
501 struct net_device *slave_dev = slave->dev;
502 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
503
504 if ((slave_dev->features & NETIF_F_HW_VLAN_RX) &&
505 slave_ops->ndo_vlan_rx_register) {
506 slave_ops->ndo_vlan_rx_register(slave_dev, grp);
507 }
508 }
509}
510
511/**
512 * bond_vlan_rx_add_vid - Propagates adding an id to slaves
513 * @bond_dev: bonding net device that got called
514 * @vid: vlan id being added
515 */
516static void bond_vlan_rx_add_vid(struct net_device *bond_dev, uint16_t vid)
517{
518 struct bonding *bond = netdev_priv(bond_dev);
519 struct slave *slave;
520 int i, res;
521
522 bond_for_each_slave(bond, slave, i) {
523 struct net_device *slave_dev = slave->dev;
524 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
525
526 if ((slave_dev->features & NETIF_F_HW_VLAN_FILTER) &&
527 slave_ops->ndo_vlan_rx_add_vid) {
528 slave_ops->ndo_vlan_rx_add_vid(slave_dev, vid);
529 }
530 }
531
532 res = bond_add_vlan(bond, vid);
533 if (res) {
534 pr_err("%s: Error: Failed to add vlan id %d\n",
535 bond_dev->name, vid);
536 }
537}
538
539/**
540 * bond_vlan_rx_kill_vid - Propagates deleting an id to slaves
541 * @bond_dev: bonding net device that got called
542 * @vid: vlan id being removed
543 */
544static void bond_vlan_rx_kill_vid(struct net_device *bond_dev, uint16_t vid)
545{
546 struct bonding *bond = netdev_priv(bond_dev);
547 struct slave *slave;
548 struct net_device *vlan_dev;
549 int i, res;
550
551 bond_for_each_slave(bond, slave, i) {
552 struct net_device *slave_dev = slave->dev;
553 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
554
555 if ((slave_dev->features & NETIF_F_HW_VLAN_FILTER) &&
556 slave_ops->ndo_vlan_rx_kill_vid) {
557 /* Save and then restore vlan_dev in the grp array,
558 * since the slave's driver might clear it.
559 */
560 vlan_dev = vlan_group_get_device(bond->vlgrp, vid);
561 slave_ops->ndo_vlan_rx_kill_vid(slave_dev, vid);
562 vlan_group_set_device(bond->vlgrp, vid, vlan_dev);
563 }
564 }
565
566 res = bond_del_vlan(bond, vid);
567 if (res) {
568 pr_err("%s: Error: Failed to remove vlan id %d\n",
569 bond_dev->name, vid);
570 }
571}
572
573static void bond_add_vlans_on_slave(struct bonding *bond, struct net_device *slave_dev)
574{
575 struct vlan_entry *vlan;
576 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
577
578 if (!bond->vlgrp)
579 return;
580
581 if ((slave_dev->features & NETIF_F_HW_VLAN_RX) &&
582 slave_ops->ndo_vlan_rx_register)
583 slave_ops->ndo_vlan_rx_register(slave_dev, bond->vlgrp);
584
585 if (!(slave_dev->features & NETIF_F_HW_VLAN_FILTER) ||
586 !(slave_ops->ndo_vlan_rx_add_vid))
587 return;
588
589 list_for_each_entry(vlan, &bond->vlan_list, vlan_list)
590 slave_ops->ndo_vlan_rx_add_vid(slave_dev, vlan->vlan_id);
591}
592
593static void bond_del_vlans_from_slave(struct bonding *bond,
594 struct net_device *slave_dev)
595{
596 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
597 struct vlan_entry *vlan;
598 struct net_device *vlan_dev;
599
600 if (!bond->vlgrp)
601 return;
602
603 if (!(slave_dev->features & NETIF_F_HW_VLAN_FILTER) ||
604 !(slave_ops->ndo_vlan_rx_kill_vid))
605 goto unreg;
606
607 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
608 if (!vlan->vlan_id)
609 continue;
610 /* Save and then restore vlan_dev in the grp array,
611 * since the slave's driver might clear it.
612 */
613 vlan_dev = vlan_group_get_device(bond->vlgrp, vlan->vlan_id);
614 slave_ops->ndo_vlan_rx_kill_vid(slave_dev, vlan->vlan_id);
615 vlan_group_set_device(bond->vlgrp, vlan->vlan_id, vlan_dev);
616 }
617
618unreg:
619 if ((slave_dev->features & NETIF_F_HW_VLAN_RX) &&
620 slave_ops->ndo_vlan_rx_register)
621 slave_ops->ndo_vlan_rx_register(slave_dev, NULL);
622}
623
624/*------------------------------- Link status -------------------------------*/
625
626/*
627 * Set the carrier state for the master according to the state of its
628 * slaves. If any slaves are up, the master is up. In 802.3ad mode,
629 * do special 802.3ad magic.
630 *
631 * Returns zero if carrier state does not change, nonzero if it does.
632 */
633static int bond_set_carrier(struct bonding *bond)
634{
635 struct slave *slave;
636 int i;
637
638 if (bond->slave_cnt == 0)
639 goto down;
640
641 if (bond->params.mode == BOND_MODE_8023AD)
642 return bond_3ad_set_carrier(bond);
643
644 bond_for_each_slave(bond, slave, i) {
645 if (slave->link == BOND_LINK_UP) {
646 if (!netif_carrier_ok(bond->dev)) {
647 netif_carrier_on(bond->dev);
648 return 1;
649 }
650 return 0;
651 }
652 }
653
654down:
655 if (netif_carrier_ok(bond->dev)) {
656 netif_carrier_off(bond->dev);
657 return 1;
658 }
659 return 0;
660}
661
662/*
663 * Get link speed and duplex from the slave's base driver
664 * using ethtool. If for some reason the call fails or the
665 * values are invalid, fake speed and duplex to 100/Full
666 * and return error.
667 */
668static int bond_update_speed_duplex(struct slave *slave)
669{
670 struct net_device *slave_dev = slave->dev;
671 struct ethtool_cmd etool;
672 int res;
673
674 /* Fake speed and duplex */
675 slave->speed = SPEED_100;
676 slave->duplex = DUPLEX_FULL;
677
678 if (!slave_dev->ethtool_ops || !slave_dev->ethtool_ops->get_settings)
679 return -1;
680
681 res = slave_dev->ethtool_ops->get_settings(slave_dev, &etool);
682 if (res < 0)
683 return -1;
684
685 switch (etool.speed) {
686 case SPEED_10:
687 case SPEED_100:
688 case SPEED_1000:
689 case SPEED_10000:
690 break;
691 default:
692 return -1;
693 }
694
695 switch (etool.duplex) {
696 case DUPLEX_FULL:
697 case DUPLEX_HALF:
698 break;
699 default:
700 return -1;
701 }
702
703 slave->speed = etool.speed;
704 slave->duplex = etool.duplex;
705
706 return 0;
707}
708
709/*
710 * if <dev> supports MII link status reporting, check its link status.
711 *
712 * We either do MII/ETHTOOL ioctls, or check netif_carrier_ok(),
713 * depending upon the setting of the use_carrier parameter.
714 *
715 * Return either BMSR_LSTATUS, meaning that the link is up (or we
716 * can't tell and just pretend it is), or 0, meaning that the link is
717 * down.
718 *
719 * If reporting is non-zero, instead of faking link up, return -1 if
720 * both ETHTOOL and MII ioctls fail (meaning the device does not
721 * support them). If use_carrier is set, return whatever it says.
722 * It'd be nice if there was a good way to tell if a driver supports
723 * netif_carrier, but there really isn't.
724 */
725static int bond_check_dev_link(struct bonding *bond,
726 struct net_device *slave_dev, int reporting)
727{
728 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
729 int (*ioctl)(struct net_device *, struct ifreq *, int);
730 struct ifreq ifr;
731 struct mii_ioctl_data *mii;
732
733 if (!reporting && !netif_running(slave_dev))
734 return 0;
735
736 if (bond->params.use_carrier)
737 return netif_carrier_ok(slave_dev) ? BMSR_LSTATUS : 0;
738
739 /* Try to get link status using Ethtool first. */
740 if (slave_dev->ethtool_ops) {
741 if (slave_dev->ethtool_ops->get_link) {
742 u32 link;
743
744 link = slave_dev->ethtool_ops->get_link(slave_dev);
745
746 return link ? BMSR_LSTATUS : 0;
747 }
748 }
749
750 /* Ethtool can't be used, fallback to MII ioctls. */
751 ioctl = slave_ops->ndo_do_ioctl;
752 if (ioctl) {
753 /* TODO: set pointer to correct ioctl on a per team member */
754 /* bases to make this more efficient. that is, once */
755 /* we determine the correct ioctl, we will always */
756 /* call it and not the others for that team */
757 /* member. */
758
759 /*
760 * We cannot assume that SIOCGMIIPHY will also read a
761 * register; not all network drivers (e.g., e100)
762 * support that.
763 */
764
765 /* Yes, the mii is overlaid on the ifreq.ifr_ifru */
766 strncpy(ifr.ifr_name, slave_dev->name, IFNAMSIZ);
767 mii = if_mii(&ifr);
768 if (IOCTL(slave_dev, &ifr, SIOCGMIIPHY) == 0) {
769 mii->reg_num = MII_BMSR;
770 if (IOCTL(slave_dev, &ifr, SIOCGMIIREG) == 0)
771 return mii->val_out & BMSR_LSTATUS;
772 }
773 }
774
775 /*
776 * If reporting, report that either there's no dev->do_ioctl,
777 * or both SIOCGMIIREG and get_link failed (meaning that we
778 * cannot report link status). If not reporting, pretend
779 * we're ok.
780 */
781 return reporting ? -1 : BMSR_LSTATUS;
782}
783
784/*----------------------------- Multicast list ------------------------------*/
785
786/*
787 * Push the promiscuity flag down to appropriate slaves
788 */
789static int bond_set_promiscuity(struct bonding *bond, int inc)
790{
791 int err = 0;
792 if (USES_PRIMARY(bond->params.mode)) {
793 /* write lock already acquired */
794 if (bond->curr_active_slave) {
795 err = dev_set_promiscuity(bond->curr_active_slave->dev,
796 inc);
797 }
798 } else {
799 struct slave *slave;
800 int i;
801 bond_for_each_slave(bond, slave, i) {
802 err = dev_set_promiscuity(slave->dev, inc);
803 if (err)
804 return err;
805 }
806 }
807 return err;
808}
809
810/*
811 * Push the allmulti flag down to all slaves
812 */
813static int bond_set_allmulti(struct bonding *bond, int inc)
814{
815 int err = 0;
816 if (USES_PRIMARY(bond->params.mode)) {
817 /* write lock already acquired */
818 if (bond->curr_active_slave) {
819 err = dev_set_allmulti(bond->curr_active_slave->dev,
820 inc);
821 }
822 } else {
823 struct slave *slave;
824 int i;
825 bond_for_each_slave(bond, slave, i) {
826 err = dev_set_allmulti(slave->dev, inc);
827 if (err)
828 return err;
829 }
830 }
831 return err;
832}
833
834/*
835 * Add a Multicast address to slaves
836 * according to mode
837 */
838static void bond_mc_add(struct bonding *bond, void *addr)
839{
840 if (USES_PRIMARY(bond->params.mode)) {
841 /* write lock already acquired */
842 if (bond->curr_active_slave)
843 dev_mc_add(bond->curr_active_slave->dev, addr);
844 } else {
845 struct slave *slave;
846 int i;
847
848 bond_for_each_slave(bond, slave, i)
849 dev_mc_add(slave->dev, addr);
850 }
851}
852
853/*
854 * Remove a multicast address from slave
855 * according to mode
856 */
857static void bond_mc_del(struct bonding *bond, void *addr)
858{
859 if (USES_PRIMARY(bond->params.mode)) {
860 /* write lock already acquired */
861 if (bond->curr_active_slave)
862 dev_mc_del(bond->curr_active_slave->dev, addr);
863 } else {
864 struct slave *slave;
865 int i;
866 bond_for_each_slave(bond, slave, i) {
867 dev_mc_del(slave->dev, addr);
868 }
869 }
870}
871
872
873static void __bond_resend_igmp_join_requests(struct net_device *dev)
874{
875 struct in_device *in_dev;
876 struct ip_mc_list *im;
877
878 rcu_read_lock();
879 in_dev = __in_dev_get_rcu(dev);
880 if (in_dev) {
881 for (im = in_dev->mc_list; im; im = im->next)
882 ip_mc_rejoin_group(im);
883 }
884
885 rcu_read_unlock();
886}
887
888/*
889 * Retrieve the list of registered multicast addresses for the bonding
890 * device and retransmit an IGMP JOIN request to the current active
891 * slave.
892 */
893static void bond_resend_igmp_join_requests(struct bonding *bond)
894{
895 struct net_device *vlan_dev;
896 struct vlan_entry *vlan;
897
898 read_lock(&bond->lock);
899
900 /* rejoin all groups on bond device */
901 __bond_resend_igmp_join_requests(bond->dev);
902
903 /* rejoin all groups on vlan devices */
904 if (bond->vlgrp) {
905 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
906 vlan_dev = vlan_group_get_device(bond->vlgrp,
907 vlan->vlan_id);
908 if (vlan_dev)
909 __bond_resend_igmp_join_requests(vlan_dev);
910 }
911 }
912
913 if (--bond->igmp_retrans > 0)
914 queue_delayed_work(bond->wq, &bond->mcast_work, HZ/5);
915
916 read_unlock(&bond->lock);
917}
918
919static void bond_resend_igmp_join_requests_delayed(struct work_struct *work)
920{
921 struct bonding *bond = container_of(work, struct bonding,
922 mcast_work.work);
923 bond_resend_igmp_join_requests(bond);
924}
925
926/*
927 * flush all members of flush->mc_list from device dev->mc_list
928 */
929static void bond_mc_list_flush(struct net_device *bond_dev,
930 struct net_device *slave_dev)
931{
932 struct bonding *bond = netdev_priv(bond_dev);
933 struct netdev_hw_addr *ha;
934
935 netdev_for_each_mc_addr(ha, bond_dev)
936 dev_mc_del(slave_dev, ha->addr);
937
938 if (bond->params.mode == BOND_MODE_8023AD) {
939 /* del lacpdu mc addr from mc list */
940 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
941
942 dev_mc_del(slave_dev, lacpdu_multicast);
943 }
944}
945
946/*--------------------------- Active slave change ---------------------------*/
947
948/*
949 * Update the mc list and multicast-related flags for the new and
950 * old active slaves (if any) according to the multicast mode, and
951 * promiscuous flags unconditionally.
952 */
953static void bond_mc_swap(struct bonding *bond, struct slave *new_active,
954 struct slave *old_active)
955{
956 struct netdev_hw_addr *ha;
957
958 if (!USES_PRIMARY(bond->params.mode))
959 /* nothing to do - mc list is already up-to-date on
960 * all slaves
961 */
962 return;
963
964 if (old_active) {
965 if (bond->dev->flags & IFF_PROMISC)
966 dev_set_promiscuity(old_active->dev, -1);
967
968 if (bond->dev->flags & IFF_ALLMULTI)
969 dev_set_allmulti(old_active->dev, -1);
970
971 netdev_for_each_mc_addr(ha, bond->dev)
972 dev_mc_del(old_active->dev, ha->addr);
973 }
974
975 if (new_active) {
976 /* FIXME: Signal errors upstream. */
977 if (bond->dev->flags & IFF_PROMISC)
978 dev_set_promiscuity(new_active->dev, 1);
979
980 if (bond->dev->flags & IFF_ALLMULTI)
981 dev_set_allmulti(new_active->dev, 1);
982
983 netdev_for_each_mc_addr(ha, bond->dev)
984 dev_mc_add(new_active->dev, ha->addr);
985 }
986}
987
988/*
989 * bond_do_fail_over_mac
990 *
991 * Perform special MAC address swapping for fail_over_mac settings
992 *
993 * Called with RTNL, bond->lock for read, curr_slave_lock for write_bh.
994 */
995static void bond_do_fail_over_mac(struct bonding *bond,
996 struct slave *new_active,
997 struct slave *old_active)
998 __releases(&bond->curr_slave_lock)
999 __releases(&bond->lock)
1000 __acquires(&bond->lock)
1001 __acquires(&bond->curr_slave_lock)
1002{
1003 u8 tmp_mac[ETH_ALEN];
1004 struct sockaddr saddr;
1005 int rv;
1006
1007 switch (bond->params.fail_over_mac) {
1008 case BOND_FOM_ACTIVE:
1009 if (new_active)
1010 memcpy(bond->dev->dev_addr, new_active->dev->dev_addr,
1011 new_active->dev->addr_len);
1012 break;
1013 case BOND_FOM_FOLLOW:
1014 /*
1015 * if new_active && old_active, swap them
1016 * if just old_active, do nothing (going to no active slave)
1017 * if just new_active, set new_active to bond's MAC
1018 */
1019 if (!new_active)
1020 return;
1021
1022 write_unlock_bh(&bond->curr_slave_lock);
1023 read_unlock(&bond->lock);
1024
1025 if (old_active) {
1026 memcpy(tmp_mac, new_active->dev->dev_addr, ETH_ALEN);
1027 memcpy(saddr.sa_data, old_active->dev->dev_addr,
1028 ETH_ALEN);
1029 saddr.sa_family = new_active->dev->type;
1030 } else {
1031 memcpy(saddr.sa_data, bond->dev->dev_addr, ETH_ALEN);
1032 saddr.sa_family = bond->dev->type;
1033 }
1034
1035 rv = dev_set_mac_address(new_active->dev, &saddr);
1036 if (rv) {
1037 pr_err("%s: Error %d setting MAC of slave %s\n",
1038 bond->dev->name, -rv, new_active->dev->name);
1039 goto out;
1040 }
1041
1042 if (!old_active)
1043 goto out;
1044
1045 memcpy(saddr.sa_data, tmp_mac, ETH_ALEN);
1046 saddr.sa_family = old_active->dev->type;
1047
1048 rv = dev_set_mac_address(old_active->dev, &saddr);
1049 if (rv)
1050 pr_err("%s: Error %d setting MAC of slave %s\n",
1051 bond->dev->name, -rv, new_active->dev->name);
1052out:
1053 read_lock(&bond->lock);
1054 write_lock_bh(&bond->curr_slave_lock);
1055 break;
1056 default:
1057 pr_err("%s: bond_do_fail_over_mac impossible: bad policy %d\n",
1058 bond->dev->name, bond->params.fail_over_mac);
1059 break;
1060 }
1061
1062}
1063
1064static bool bond_should_change_active(struct bonding *bond)
1065{
1066 struct slave *prim = bond->primary_slave;
1067 struct slave *curr = bond->curr_active_slave;
1068
1069 if (!prim || !curr || curr->link != BOND_LINK_UP)
1070 return true;
1071 if (bond->force_primary) {
1072 bond->force_primary = false;
1073 return true;
1074 }
1075 if (bond->params.primary_reselect == BOND_PRI_RESELECT_BETTER &&
1076 (prim->speed < curr->speed ||
1077 (prim->speed == curr->speed && prim->duplex <= curr->duplex)))
1078 return false;
1079 if (bond->params.primary_reselect == BOND_PRI_RESELECT_FAILURE)
1080 return false;
1081 return true;
1082}
1083
1084/**
1085 * find_best_interface - select the best available slave to be the active one
1086 * @bond: our bonding struct
1087 *
1088 * Warning: Caller must hold curr_slave_lock for writing.
1089 */
1090static struct slave *bond_find_best_slave(struct bonding *bond)
1091{
1092 struct slave *new_active, *old_active;
1093 struct slave *bestslave = NULL;
1094 int mintime = bond->params.updelay;
1095 int i;
1096
1097 new_active = bond->curr_active_slave;
1098
1099 if (!new_active) { /* there were no active slaves left */
1100 if (bond->slave_cnt > 0) /* found one slave */
1101 new_active = bond->first_slave;
1102 else
1103 return NULL; /* still no slave, return NULL */
1104 }
1105
1106 if ((bond->primary_slave) &&
1107 bond->primary_slave->link == BOND_LINK_UP &&
1108 bond_should_change_active(bond)) {
1109 new_active = bond->primary_slave;
1110 }
1111
1112 /* remember where to stop iterating over the slaves */
1113 old_active = new_active;
1114
1115 bond_for_each_slave_from(bond, new_active, i, old_active) {
1116 if (new_active->link == BOND_LINK_UP) {
1117 return new_active;
1118 } else if (new_active->link == BOND_LINK_BACK &&
1119 IS_UP(new_active->dev)) {
1120 /* link up, but waiting for stabilization */
1121 if (new_active->delay < mintime) {
1122 mintime = new_active->delay;
1123 bestslave = new_active;
1124 }
1125 }
1126 }
1127
1128 return bestslave;
1129}
1130
1131/**
1132 * change_active_interface - change the active slave into the specified one
1133 * @bond: our bonding struct
1134 * @new: the new slave to make the active one
1135 *
1136 * Set the new slave to the bond's settings and unset them on the old
1137 * curr_active_slave.
1138 * Setting include flags, mc-list, promiscuity, allmulti, etc.
1139 *
1140 * If @new's link state is %BOND_LINK_BACK we'll set it to %BOND_LINK_UP,
1141 * because it is apparently the best available slave we have, even though its
1142 * updelay hasn't timed out yet.
1143 *
1144 * If new_active is not NULL, caller must hold bond->lock for read and
1145 * curr_slave_lock for write_bh.
1146 */
1147void bond_change_active_slave(struct bonding *bond, struct slave *new_active)
1148{
1149 struct slave *old_active = bond->curr_active_slave;
1150
1151 if (old_active == new_active)
1152 return;
1153
1154 if (new_active) {
1155 new_active->jiffies = jiffies;
1156
1157 if (new_active->link == BOND_LINK_BACK) {
1158 if (USES_PRIMARY(bond->params.mode)) {
1159 pr_info("%s: making interface %s the new active one %d ms earlier.\n",
1160 bond->dev->name, new_active->dev->name,
1161 (bond->params.updelay - new_active->delay) * bond->params.miimon);
1162 }
1163
1164 new_active->delay = 0;
1165 new_active->link = BOND_LINK_UP;
1166
1167 if (bond->params.mode == BOND_MODE_8023AD)
1168 bond_3ad_handle_link_change(new_active, BOND_LINK_UP);
1169
1170 if (bond_is_lb(bond))
1171 bond_alb_handle_link_change(bond, new_active, BOND_LINK_UP);
1172 } else {
1173 if (USES_PRIMARY(bond->params.mode)) {
1174 pr_info("%s: making interface %s the new active one.\n",
1175 bond->dev->name, new_active->dev->name);
1176 }
1177 }
1178 }
1179
1180 if (USES_PRIMARY(bond->params.mode))
1181 bond_mc_swap(bond, new_active, old_active);
1182
1183 if (bond_is_lb(bond)) {
1184 bond_alb_handle_active_change(bond, new_active);
1185 if (old_active)
1186 bond_set_slave_inactive_flags(old_active);
1187 if (new_active)
1188 bond_set_slave_active_flags(new_active);
1189 } else {
1190 bond->curr_active_slave = new_active;
1191 }
1192
1193 if (bond->params.mode == BOND_MODE_ACTIVEBACKUP) {
1194 if (old_active)
1195 bond_set_slave_inactive_flags(old_active);
1196
1197 if (new_active) {
1198 bond_set_slave_active_flags(new_active);
1199
1200 if (bond->params.fail_over_mac)
1201 bond_do_fail_over_mac(bond, new_active,
1202 old_active);
1203
1204 bond->send_grat_arp = bond->params.num_grat_arp;
1205 bond_send_gratuitous_arp(bond);
1206
1207 bond->send_unsol_na = bond->params.num_unsol_na;
1208 bond_send_unsolicited_na(bond);
1209
1210 write_unlock_bh(&bond->curr_slave_lock);
1211 read_unlock(&bond->lock);
1212
1213 netdev_bonding_change(bond->dev, NETDEV_BONDING_FAILOVER);
1214
1215 read_lock(&bond->lock);
1216 write_lock_bh(&bond->curr_slave_lock);
1217 }
1218 }
1219
1220 /* resend IGMP joins since active slave has changed or
1221 * all were sent on curr_active_slave */
1222 if ((USES_PRIMARY(bond->params.mode) && new_active) ||
1223 bond->params.mode == BOND_MODE_ROUNDROBIN) {
1224 bond->igmp_retrans = bond->params.resend_igmp;
1225 queue_delayed_work(bond->wq, &bond->mcast_work, 0);
1226 }
1227}
1228
1229/**
1230 * bond_select_active_slave - select a new active slave, if needed
1231 * @bond: our bonding struct
1232 *
1233 * This functions should be called when one of the following occurs:
1234 * - The old curr_active_slave has been released or lost its link.
1235 * - The primary_slave has got its link back.
1236 * - A slave has got its link back and there's no old curr_active_slave.
1237 *
1238 * Caller must hold bond->lock for read and curr_slave_lock for write_bh.
1239 */
1240void bond_select_active_slave(struct bonding *bond)
1241{
1242 struct slave *best_slave;
1243 int rv;
1244
1245 best_slave = bond_find_best_slave(bond);
1246 if (best_slave != bond->curr_active_slave) {
1247 bond_change_active_slave(bond, best_slave);
1248 rv = bond_set_carrier(bond);
1249 if (!rv)
1250 return;
1251
1252 if (netif_carrier_ok(bond->dev)) {
1253 pr_info("%s: first active interface up!\n",
1254 bond->dev->name);
1255 } else {
1256 pr_info("%s: now running without any active interface !\n",
1257 bond->dev->name);
1258 }
1259 }
1260}
1261
1262/*--------------------------- slave list handling ---------------------------*/
1263
1264/*
1265 * This function attaches the slave to the end of list.
1266 *
1267 * bond->lock held for writing by caller.
1268 */
1269static void bond_attach_slave(struct bonding *bond, struct slave *new_slave)
1270{
1271 if (bond->first_slave == NULL) { /* attaching the first slave */
1272 new_slave->next = new_slave;
1273 new_slave->prev = new_slave;
1274 bond->first_slave = new_slave;
1275 } else {
1276 new_slave->next = bond->first_slave;
1277 new_slave->prev = bond->first_slave->prev;
1278 new_slave->next->prev = new_slave;
1279 new_slave->prev->next = new_slave;
1280 }
1281
1282 bond->slave_cnt++;
1283}
1284
1285/*
1286 * This function detaches the slave from the list.
1287 * WARNING: no check is made to verify if the slave effectively
1288 * belongs to <bond>.
1289 * Nothing is freed on return, structures are just unchained.
1290 * If any slave pointer in bond was pointing to <slave>,
1291 * it should be changed by the calling function.
1292 *
1293 * bond->lock held for writing by caller.
1294 */
1295static void bond_detach_slave(struct bonding *bond, struct slave *slave)
1296{
1297 if (slave->next)
1298 slave->next->prev = slave->prev;
1299
1300 if (slave->prev)
1301 slave->prev->next = slave->next;
1302
1303 if (bond->first_slave == slave) { /* slave is the first slave */
1304 if (bond->slave_cnt > 1) { /* there are more slave */
1305 bond->first_slave = slave->next;
1306 } else {
1307 bond->first_slave = NULL; /* slave was the last one */
1308 }
1309 }
1310
1311 slave->next = NULL;
1312 slave->prev = NULL;
1313 bond->slave_cnt--;
1314}
1315
1316#ifdef CONFIG_NET_POLL_CONTROLLER
1317/*
1318 * You must hold read lock on bond->lock before calling this.
1319 */
1320static bool slaves_support_netpoll(struct net_device *bond_dev)
1321{
1322 struct bonding *bond = netdev_priv(bond_dev);
1323 struct slave *slave;
1324 int i = 0;
1325 bool ret = true;
1326
1327 bond_for_each_slave(bond, slave, i) {
1328 if ((slave->dev->priv_flags & IFF_DISABLE_NETPOLL) ||
1329 !slave->dev->netdev_ops->ndo_poll_controller)
1330 ret = false;
1331 }
1332 return i != 0 && ret;
1333}
1334
1335static void bond_poll_controller(struct net_device *bond_dev)
1336{
1337 struct bonding *bond = netdev_priv(bond_dev);
1338 struct slave *slave;
1339 int i;
1340
1341 bond_for_each_slave(bond, slave, i) {
1342 if (slave->dev && IS_UP(slave->dev))
1343 netpoll_poll_dev(slave->dev);
1344 }
1345}
1346
1347static void bond_netpoll_cleanup(struct net_device *bond_dev)
1348{
1349 struct bonding *bond = netdev_priv(bond_dev);
1350 struct slave *slave;
1351 const struct net_device_ops *ops;
1352 int i;
1353
1354 read_lock(&bond->lock);
1355 bond_dev->npinfo = NULL;
1356 bond_for_each_slave(bond, slave, i) {
1357 if (slave->dev) {
1358 ops = slave->dev->netdev_ops;
1359 if (ops->ndo_netpoll_cleanup)
1360 ops->ndo_netpoll_cleanup(slave->dev);
1361 else
1362 slave->dev->npinfo = NULL;
1363 }
1364 }
1365 read_unlock(&bond->lock);
1366}
1367
1368#else
1369
1370static void bond_netpoll_cleanup(struct net_device *bond_dev)
1371{
1372}
1373
1374#endif
1375
1376/*---------------------------------- IOCTL ----------------------------------*/
1377
1378static int bond_sethwaddr(struct net_device *bond_dev,
1379 struct net_device *slave_dev)
1380{
1381 pr_debug("bond_dev=%p\n", bond_dev);
1382 pr_debug("slave_dev=%p\n", slave_dev);
1383 pr_debug("slave_dev->addr_len=%d\n", slave_dev->addr_len);
1384 memcpy(bond_dev->dev_addr, slave_dev->dev_addr, slave_dev->addr_len);
1385 return 0;
1386}
1387
1388#define BOND_VLAN_FEATURES \
1389 (NETIF_F_VLAN_CHALLENGED | NETIF_F_HW_VLAN_RX | NETIF_F_HW_VLAN_TX | \
1390 NETIF_F_HW_VLAN_FILTER)
1391
1392/*
1393 * Compute the common dev->feature set available to all slaves. Some
1394 * feature bits are managed elsewhere, so preserve those feature bits
1395 * on the master device.
1396 */
1397static int bond_compute_features(struct bonding *bond)
1398{
1399 struct slave *slave;
1400 struct net_device *bond_dev = bond->dev;
1401 unsigned long features = bond_dev->features;
1402 unsigned long vlan_features = 0;
1403 unsigned short max_hard_header_len = max((u16)ETH_HLEN,
1404 bond_dev->hard_header_len);
1405 int i;
1406
1407 features &= ~(NETIF_F_ALL_CSUM | BOND_VLAN_FEATURES);
1408 features |= NETIF_F_GSO_MASK | NETIF_F_NO_CSUM;
1409
1410 if (!bond->first_slave)
1411 goto done;
1412
1413 features &= ~NETIF_F_ONE_FOR_ALL;
1414
1415 vlan_features = bond->first_slave->dev->vlan_features;
1416 bond_for_each_slave(bond, slave, i) {
1417 features = netdev_increment_features(features,
1418 slave->dev->features,
1419 NETIF_F_ONE_FOR_ALL);
1420 vlan_features = netdev_increment_features(vlan_features,
1421 slave->dev->vlan_features,
1422 NETIF_F_ONE_FOR_ALL);
1423 if (slave->dev->hard_header_len > max_hard_header_len)
1424 max_hard_header_len = slave->dev->hard_header_len;
1425 }
1426
1427done:
1428 features |= (bond_dev->features & BOND_VLAN_FEATURES);
1429 bond_dev->features = netdev_fix_features(features, NULL);
1430 bond_dev->vlan_features = netdev_fix_features(vlan_features, NULL);
1431 bond_dev->hard_header_len = max_hard_header_len;
1432
1433 return 0;
1434}
1435
1436static void bond_setup_by_slave(struct net_device *bond_dev,
1437 struct net_device *slave_dev)
1438{
1439 struct bonding *bond = netdev_priv(bond_dev);
1440
1441 bond_dev->header_ops = slave_dev->header_ops;
1442
1443 bond_dev->type = slave_dev->type;
1444 bond_dev->hard_header_len = slave_dev->hard_header_len;
1445 bond_dev->addr_len = slave_dev->addr_len;
1446
1447 memcpy(bond_dev->broadcast, slave_dev->broadcast,
1448 slave_dev->addr_len);
1449 bond->setup_by_slave = 1;
1450}
1451
1452/* enslave device <slave> to bond device <master> */
1453int bond_enslave(struct net_device *bond_dev, struct net_device *slave_dev)
1454{
1455 struct bonding *bond = netdev_priv(bond_dev);
1456 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
1457 struct slave *new_slave = NULL;
1458 struct netdev_hw_addr *ha;
1459 struct sockaddr addr;
1460 int link_reporting;
1461 int old_features = bond_dev->features;
1462 int res = 0;
1463
1464 if (!bond->params.use_carrier && slave_dev->ethtool_ops == NULL &&
1465 slave_ops->ndo_do_ioctl == NULL) {
1466 pr_warning("%s: Warning: no link monitoring support for %s\n",
1467 bond_dev->name, slave_dev->name);
1468 }
1469
1470 /* bond must be initialized by bond_open() before enslaving */
1471 if (!(bond_dev->flags & IFF_UP)) {
1472 pr_warning("%s: master_dev is not up in bond_enslave\n",
1473 bond_dev->name);
1474 }
1475
1476 /* already enslaved */
1477 if (slave_dev->flags & IFF_SLAVE) {
1478 pr_debug("Error, Device was already enslaved\n");
1479 return -EBUSY;
1480 }
1481
1482 /* vlan challenged mutual exclusion */
1483 /* no need to lock since we're protected by rtnl_lock */
1484 if (slave_dev->features & NETIF_F_VLAN_CHALLENGED) {
1485 pr_debug("%s: NETIF_F_VLAN_CHALLENGED\n", slave_dev->name);
1486 if (bond->vlgrp) {
1487 pr_err("%s: Error: cannot enslave VLAN challenged slave %s on VLAN enabled bond %s\n",
1488 bond_dev->name, slave_dev->name, bond_dev->name);
1489 return -EPERM;
1490 } else {
1491 pr_warning("%s: Warning: enslaved VLAN challenged slave %s. Adding VLANs will be blocked as long as %s is part of bond %s\n",
1492 bond_dev->name, slave_dev->name,
1493 slave_dev->name, bond_dev->name);
1494 bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
1495 }
1496 } else {
1497 pr_debug("%s: ! NETIF_F_VLAN_CHALLENGED\n", slave_dev->name);
1498 if (bond->slave_cnt == 0) {
1499 /* First slave, and it is not VLAN challenged,
1500 * so remove the block of adding VLANs over the bond.
1501 */
1502 bond_dev->features &= ~NETIF_F_VLAN_CHALLENGED;
1503 }
1504 }
1505
1506 /*
1507 * Old ifenslave binaries are no longer supported. These can
1508 * be identified with moderate accuracy by the state of the slave:
1509 * the current ifenslave will set the interface down prior to
1510 * enslaving it; the old ifenslave will not.
1511 */
1512 if ((slave_dev->flags & IFF_UP)) {
1513 pr_err("%s is up. This may be due to an out of date ifenslave.\n",
1514 slave_dev->name);
1515 res = -EPERM;
1516 goto err_undo_flags;
1517 }
1518
1519 /* set bonding device ether type by slave - bonding netdevices are
1520 * created with ether_setup, so when the slave type is not ARPHRD_ETHER
1521 * there is a need to override some of the type dependent attribs/funcs.
1522 *
1523 * bond ether type mutual exclusion - don't allow slaves of dissimilar
1524 * ether type (eg ARPHRD_ETHER and ARPHRD_INFINIBAND) share the same bond
1525 */
1526 if (bond->slave_cnt == 0) {
1527 if (bond_dev->type != slave_dev->type) {
1528 pr_debug("%s: change device type from %d to %d\n",
1529 bond_dev->name,
1530 bond_dev->type, slave_dev->type);
1531
1532 res = netdev_bonding_change(bond_dev,
1533 NETDEV_PRE_TYPE_CHANGE);
1534 res = notifier_to_errno(res);
1535 if (res) {
1536 pr_err("%s: refused to change device type\n",
1537 bond_dev->name);
1538 res = -EBUSY;
1539 goto err_undo_flags;
1540 }
1541
1542 /* Flush unicast and multicast addresses */
1543 dev_uc_flush(bond_dev);
1544 dev_mc_flush(bond_dev);
1545
1546 if (slave_dev->type != ARPHRD_ETHER)
1547 bond_setup_by_slave(bond_dev, slave_dev);
1548 else
1549 ether_setup(bond_dev);
1550
1551 netdev_bonding_change(bond_dev,
1552 NETDEV_POST_TYPE_CHANGE);
1553 }
1554 } else if (bond_dev->type != slave_dev->type) {
1555 pr_err("%s ether type (%d) is different from other slaves (%d), can not enslave it.\n",
1556 slave_dev->name,
1557 slave_dev->type, bond_dev->type);
1558 res = -EINVAL;
1559 goto err_undo_flags;
1560 }
1561
1562 if (slave_ops->ndo_set_mac_address == NULL) {
1563 if (bond->slave_cnt == 0) {
1564 pr_warning("%s: Warning: The first slave device specified does not support setting the MAC address. Setting fail_over_mac to active.",
1565 bond_dev->name);
1566 bond->params.fail_over_mac = BOND_FOM_ACTIVE;
1567 } else if (bond->params.fail_over_mac != BOND_FOM_ACTIVE) {
1568 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",
1569 bond_dev->name);
1570 res = -EOPNOTSUPP;
1571 goto err_undo_flags;
1572 }
1573 }
1574
1575 /* If this is the first slave, then we need to set the master's hardware
1576 * address to be the same as the slave's. */
1577 if (bond->slave_cnt == 0)
1578 memcpy(bond->dev->dev_addr, slave_dev->dev_addr,
1579 slave_dev->addr_len);
1580
1581
1582 new_slave = kzalloc(sizeof(struct slave), GFP_KERNEL);
1583 if (!new_slave) {
1584 res = -ENOMEM;
1585 goto err_undo_flags;
1586 }
1587
1588 /*
1589 * Set the new_slave's queue_id to be zero. Queue ID mapping
1590 * is set via sysfs or module option if desired.
1591 */
1592 new_slave->queue_id = 0;
1593
1594 /* Save slave's original mtu and then set it to match the bond */
1595 new_slave->original_mtu = slave_dev->mtu;
1596 res = dev_set_mtu(slave_dev, bond->dev->mtu);
1597 if (res) {
1598 pr_debug("Error %d calling dev_set_mtu\n", res);
1599 goto err_free;
1600 }
1601
1602 /*
1603 * Save slave's original ("permanent") mac address for modes
1604 * that need it, and for restoring it upon release, and then
1605 * set it to the master's address
1606 */
1607 memcpy(new_slave->perm_hwaddr, slave_dev->dev_addr, ETH_ALEN);
1608
1609 if (!bond->params.fail_over_mac) {
1610 /*
1611 * Set slave to master's mac address. The application already
1612 * set the master's mac address to that of the first slave
1613 */
1614 memcpy(addr.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
1615 addr.sa_family = slave_dev->type;
1616 res = dev_set_mac_address(slave_dev, &addr);
1617 if (res) {
1618 pr_debug("Error %d calling set_mac_address\n", res);
1619 goto err_restore_mtu;
1620 }
1621 }
1622
1623 res = netdev_set_master(slave_dev, bond_dev);
1624 if (res) {
1625 pr_debug("Error %d calling netdev_set_master\n", res);
1626 goto err_restore_mac;
1627 }
1628 /* open the slave since the application closed it */
1629 res = dev_open(slave_dev);
1630 if (res) {
1631 pr_debug("Opening slave %s failed\n", slave_dev->name);
1632 goto err_unset_master;
1633 }
1634
1635 new_slave->dev = slave_dev;
1636 slave_dev->priv_flags |= IFF_BONDING;
1637
1638 if (bond_is_lb(bond)) {
1639 /* bond_alb_init_slave() must be called before all other stages since
1640 * it might fail and we do not want to have to undo everything
1641 */
1642 res = bond_alb_init_slave(bond, new_slave);
1643 if (res)
1644 goto err_close;
1645 }
1646
1647 /* If the mode USES_PRIMARY, then the new slave gets the
1648 * master's promisc (and mc) settings only if it becomes the
1649 * curr_active_slave, and that is taken care of later when calling
1650 * bond_change_active()
1651 */
1652 if (!USES_PRIMARY(bond->params.mode)) {
1653 /* set promiscuity level to new slave */
1654 if (bond_dev->flags & IFF_PROMISC) {
1655 res = dev_set_promiscuity(slave_dev, 1);
1656 if (res)
1657 goto err_close;
1658 }
1659
1660 /* set allmulti level to new slave */
1661 if (bond_dev->flags & IFF_ALLMULTI) {
1662 res = dev_set_allmulti(slave_dev, 1);
1663 if (res)
1664 goto err_close;
1665 }
1666
1667 netif_addr_lock_bh(bond_dev);
1668 /* upload master's mc_list to new slave */
1669 netdev_for_each_mc_addr(ha, bond_dev)
1670 dev_mc_add(slave_dev, ha->addr);
1671 netif_addr_unlock_bh(bond_dev);
1672 }
1673
1674 if (bond->params.mode == BOND_MODE_8023AD) {
1675 /* add lacpdu mc addr to mc list */
1676 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
1677
1678 dev_mc_add(slave_dev, lacpdu_multicast);
1679 }
1680
1681 bond_add_vlans_on_slave(bond, slave_dev);
1682
1683 write_lock_bh(&bond->lock);
1684
1685 bond_attach_slave(bond, new_slave);
1686
1687 new_slave->delay = 0;
1688 new_slave->link_failure_count = 0;
1689
1690 bond_compute_features(bond);
1691
1692 write_unlock_bh(&bond->lock);
1693
1694 read_lock(&bond->lock);
1695
1696 new_slave->last_arp_rx = jiffies;
1697
1698 if (bond->params.miimon && !bond->params.use_carrier) {
1699 link_reporting = bond_check_dev_link(bond, slave_dev, 1);
1700
1701 if ((link_reporting == -1) && !bond->params.arp_interval) {
1702 /*
1703 * miimon is set but a bonded network driver
1704 * does not support ETHTOOL/MII and
1705 * arp_interval is not set. Note: if
1706 * use_carrier is enabled, we will never go
1707 * here (because netif_carrier is always
1708 * supported); thus, we don't need to change
1709 * the messages for netif_carrier.
1710 */
1711 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",
1712 bond_dev->name, slave_dev->name);
1713 } else if (link_reporting == -1) {
1714 /* unable get link status using mii/ethtool */
1715 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",
1716 bond_dev->name, slave_dev->name);
1717 }
1718 }
1719
1720 /* check for initial state */
1721 if (!bond->params.miimon ||
1722 (bond_check_dev_link(bond, slave_dev, 0) == BMSR_LSTATUS)) {
1723 if (bond->params.updelay) {
1724 pr_debug("Initial state of slave_dev is BOND_LINK_BACK\n");
1725 new_slave->link = BOND_LINK_BACK;
1726 new_slave->delay = bond->params.updelay;
1727 } else {
1728 pr_debug("Initial state of slave_dev is BOND_LINK_UP\n");
1729 new_slave->link = BOND_LINK_UP;
1730 }
1731 new_slave->jiffies = jiffies;
1732 } else {
1733 pr_debug("Initial state of slave_dev is BOND_LINK_DOWN\n");
1734 new_slave->link = BOND_LINK_DOWN;
1735 }
1736
1737 if (bond_update_speed_duplex(new_slave) &&
1738 (new_slave->link != BOND_LINK_DOWN)) {
1739 pr_warning("%s: Warning: failed to get speed and duplex from %s, assumed to be 100Mb/sec and Full.\n",
1740 bond_dev->name, new_slave->dev->name);
1741
1742 if (bond->params.mode == BOND_MODE_8023AD) {
1743 pr_warning("%s: Warning: Operation of 802.3ad mode requires ETHTOOL support in base driver for proper aggregator selection.\n",
1744 bond_dev->name);
1745 }
1746 }
1747
1748 if (USES_PRIMARY(bond->params.mode) && bond->params.primary[0]) {
1749 /* if there is a primary slave, remember it */
1750 if (strcmp(bond->params.primary, new_slave->dev->name) == 0) {
1751 bond->primary_slave = new_slave;
1752 bond->force_primary = true;
1753 }
1754 }
1755
1756 write_lock_bh(&bond->curr_slave_lock);
1757
1758 switch (bond->params.mode) {
1759 case BOND_MODE_ACTIVEBACKUP:
1760 bond_set_slave_inactive_flags(new_slave);
1761 bond_select_active_slave(bond);
1762 break;
1763 case BOND_MODE_8023AD:
1764 /* in 802.3ad mode, the internal mechanism
1765 * will activate the slaves in the selected
1766 * aggregator
1767 */
1768 bond_set_slave_inactive_flags(new_slave);
1769 /* if this is the first slave */
1770 if (bond->slave_cnt == 1) {
1771 SLAVE_AD_INFO(new_slave).id = 1;
1772 /* Initialize AD with the number of times that the AD timer is called in 1 second
1773 * can be called only after the mac address of the bond is set
1774 */
1775 bond_3ad_initialize(bond, 1000/AD_TIMER_INTERVAL,
1776 bond->params.lacp_fast);
1777 } else {
1778 SLAVE_AD_INFO(new_slave).id =
1779 SLAVE_AD_INFO(new_slave->prev).id + 1;
1780 }
1781
1782 bond_3ad_bind_slave(new_slave);
1783 break;
1784 case BOND_MODE_TLB:
1785 case BOND_MODE_ALB:
1786 new_slave->state = BOND_STATE_ACTIVE;
1787 bond_set_slave_inactive_flags(new_slave);
1788 bond_select_active_slave(bond);
1789 break;
1790 default:
1791 pr_debug("This slave is always active in trunk mode\n");
1792
1793 /* always active in trunk mode */
1794 new_slave->state = BOND_STATE_ACTIVE;
1795
1796 /* In trunking mode there is little meaning to curr_active_slave
1797 * anyway (it holds no special properties of the bond device),
1798 * so we can change it without calling change_active_interface()
1799 */
1800 if (!bond->curr_active_slave)
1801 bond->curr_active_slave = new_slave;
1802
1803 break;
1804 } /* switch(bond_mode) */
1805
1806 write_unlock_bh(&bond->curr_slave_lock);
1807
1808 bond_set_carrier(bond);
1809
1810#ifdef CONFIG_NET_POLL_CONTROLLER
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#endif
1822 read_unlock(&bond->lock);
1823
1824 res = bond_create_slave_symlinks(bond_dev, slave_dev);
1825 if (res)
1826 goto err_close;
1827
1828 pr_info("%s: enslaving %s as a%s interface with a%s link.\n",
1829 bond_dev->name, slave_dev->name,
1830 new_slave->state == BOND_STATE_ACTIVE ? "n active" : " backup",
1831 new_slave->link != BOND_LINK_DOWN ? "n up" : " down");
1832
1833 /* enslave is successful */
1834 return 0;
1835
1836/* Undo stages on error */
1837err_close:
1838 dev_close(slave_dev);
1839
1840err_unset_master:
1841 netdev_set_master(slave_dev, NULL);
1842
1843err_restore_mac:
1844 if (!bond->params.fail_over_mac) {
1845 /* XXX TODO - fom follow mode needs to change master's
1846 * MAC if this slave's MAC is in use by the bond, or at
1847 * least print a warning.
1848 */
1849 memcpy(addr.sa_data, new_slave->perm_hwaddr, ETH_ALEN);
1850 addr.sa_family = slave_dev->type;
1851 dev_set_mac_address(slave_dev, &addr);
1852 }
1853
1854err_restore_mtu:
1855 dev_set_mtu(slave_dev, new_slave->original_mtu);
1856
1857err_free:
1858 kfree(new_slave);
1859
1860err_undo_flags:
1861 bond_dev->features = old_features;
1862
1863 return res;
1864}
1865
1866/*
1867 * Try to release the slave device <slave> from the bond device <master>
1868 * It is legal to access curr_active_slave without a lock because all the function
1869 * is write-locked.
1870 *
1871 * The rules for slave state should be:
1872 * for Active/Backup:
1873 * Active stays on all backups go down
1874 * for Bonded connections:
1875 * The first up interface should be left on and all others downed.
1876 */
1877int bond_release(struct net_device *bond_dev, struct net_device *slave_dev)
1878{
1879 struct bonding *bond = netdev_priv(bond_dev);
1880 struct slave *slave, *oldcurrent;
1881 struct sockaddr addr;
1882
1883 /* slave is not a slave or master is not master of this slave */
1884 if (!(slave_dev->flags & IFF_SLAVE) ||
1885 (slave_dev->master != bond_dev)) {
1886 pr_err("%s: Error: cannot release %s.\n",
1887 bond_dev->name, slave_dev->name);
1888 return -EINVAL;
1889 }
1890
1891 block_netpoll_tx();
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 unblock_netpoll_tx();
1902 return -EINVAL;
1903 }
1904
1905 if (!bond->params.fail_over_mac) {
1906 if (!compare_ether_addr(bond_dev->dev_addr, slave->perm_hwaddr) &&
1907 bond->slave_cnt > 1)
1908 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",
1909 bond_dev->name, slave_dev->name,
1910 slave->perm_hwaddr,
1911 bond_dev->name, slave_dev->name);
1912 }
1913
1914 /* Inform AD package of unbinding of slave. */
1915 if (bond->params.mode == BOND_MODE_8023AD) {
1916 /* must be called before the slave is
1917 * detached from the list
1918 */
1919 bond_3ad_unbind_slave(slave);
1920 }
1921
1922 pr_info("%s: releasing %s interface %s\n",
1923 bond_dev->name,
1924 (slave->state == BOND_STATE_ACTIVE) ? "active" : "backup",
1925 slave_dev->name);
1926
1927 oldcurrent = bond->curr_active_slave;
1928
1929 bond->current_arp_slave = NULL;
1930
1931 /* release the slave from its bond */
1932 bond_detach_slave(bond, slave);
1933
1934 bond_compute_features(bond);
1935
1936 if (bond->primary_slave == slave)
1937 bond->primary_slave = NULL;
1938
1939 if (oldcurrent == slave)
1940 bond_change_active_slave(bond, NULL);
1941
1942 if (bond_is_lb(bond)) {
1943 /* Must be called only after the slave has been
1944 * detached from the list and the curr_active_slave
1945 * has been cleared (if our_slave == old_current),
1946 * but before a new active slave is selected.
1947 */
1948 write_unlock_bh(&bond->lock);
1949 bond_alb_deinit_slave(bond, slave);
1950 write_lock_bh(&bond->lock);
1951 }
1952
1953 if (oldcurrent == slave) {
1954 /*
1955 * Note that we hold RTNL over this sequence, so there
1956 * is no concern that another slave add/remove event
1957 * will interfere.
1958 */
1959 write_unlock_bh(&bond->lock);
1960 read_lock(&bond->lock);
1961 write_lock_bh(&bond->curr_slave_lock);
1962
1963 bond_select_active_slave(bond);
1964
1965 write_unlock_bh(&bond->curr_slave_lock);
1966 read_unlock(&bond->lock);
1967 write_lock_bh(&bond->lock);
1968 }
1969
1970 if (bond->slave_cnt == 0) {
1971 bond_set_carrier(bond);
1972
1973 /* if the last slave was removed, zero the mac address
1974 * of the master so it will be set by the application
1975 * to the mac address of the first slave
1976 */
1977 memset(bond_dev->dev_addr, 0, bond_dev->addr_len);
1978
1979 if (!bond->vlgrp) {
1980 bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
1981 } else {
1982 pr_warning("%s: Warning: clearing HW address of %s while it still has VLANs.\n",
1983 bond_dev->name, bond_dev->name);
1984 pr_warning("%s: When re-adding slaves, make sure the bond's HW address matches its VLANs'.\n",
1985 bond_dev->name);
1986 }
1987 } else if ((bond_dev->features & NETIF_F_VLAN_CHALLENGED) &&
1988 !bond_has_challenged_slaves(bond)) {
1989 pr_info("%s: last VLAN challenged slave %s left bond %s. VLAN blocking is removed\n",
1990 bond_dev->name, slave_dev->name, bond_dev->name);
1991 bond_dev->features &= ~NETIF_F_VLAN_CHALLENGED;
1992 }
1993
1994 write_unlock_bh(&bond->lock);
1995 unblock_netpoll_tx();
1996
1997 /* must do this from outside any spinlocks */
1998 bond_destroy_slave_symlinks(bond_dev, slave_dev);
1999
2000 bond_del_vlans_from_slave(bond, slave_dev);
2001
2002 /* If the mode USES_PRIMARY, then we should only remove its
2003 * promisc and mc settings if it was the curr_active_slave, but that was
2004 * already taken care of above when we detached the slave
2005 */
2006 if (!USES_PRIMARY(bond->params.mode)) {
2007 /* unset promiscuity level from slave */
2008 if (bond_dev->flags & IFF_PROMISC)
2009 dev_set_promiscuity(slave_dev, -1);
2010
2011 /* unset allmulti level from slave */
2012 if (bond_dev->flags & IFF_ALLMULTI)
2013 dev_set_allmulti(slave_dev, -1);
2014
2015 /* flush master's mc_list from slave */
2016 netif_addr_lock_bh(bond_dev);
2017 bond_mc_list_flush(bond_dev, slave_dev);
2018 netif_addr_unlock_bh(bond_dev);
2019 }
2020
2021 netdev_set_master(slave_dev, NULL);
2022
2023#ifdef CONFIG_NET_POLL_CONTROLLER
2024 read_lock_bh(&bond->lock);
2025
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*/
2060int 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 */
2078static 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
2181out:
2182 write_unlock_bh(&bond->lock);
2183 return 0;
2184}
2185
2186/*
2187 * This function changes the active slave to slave <slave_dev>.
2188 * It returns -EINVAL in the following cases.
2189 * - <slave_dev> is not found in the list.
2190 * - There is not active slave now.
2191 * - <slave_dev> is already active.
2192 * - The link state of <slave_dev> is not BOND_LINK_UP.
2193 * - <slave_dev> is not running.
2194 * In these cases, this function does nothing.
2195 * In the other cases, current_slave pointer is changed and 0 is returned.
2196 */
2197static int bond_ioctl_change_active(struct net_device *bond_dev, struct net_device *slave_dev)
2198{
2199 struct bonding *bond = netdev_priv(bond_dev);
2200 struct slave *old_active = NULL;
2201 struct slave *new_active = NULL;
2202 int res = 0;
2203
2204 if (!USES_PRIMARY(bond->params.mode))
2205 return -EINVAL;
2206
2207 /* Verify that master_dev is indeed the master of slave_dev */
2208 if (!(slave_dev->flags & IFF_SLAVE) || (slave_dev->master != bond_dev))
2209 return -EINVAL;
2210
2211 read_lock(&bond->lock);
2212
2213 read_lock(&bond->curr_slave_lock);
2214 old_active = bond->curr_active_slave;
2215 read_unlock(&bond->curr_slave_lock);
2216
2217 new_active = bond_get_slave_by_dev(bond, slave_dev);
2218
2219 /*
2220 * Changing to the current active: do nothing; return success.
2221 */
2222 if (new_active && (new_active == old_active)) {
2223 read_unlock(&bond->lock);
2224 return 0;
2225 }
2226
2227 if ((new_active) &&
2228 (old_active) &&
2229 (new_active->link == BOND_LINK_UP) &&
2230 IS_UP(new_active->dev)) {
2231 block_netpoll_tx();
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 unblock_netpoll_tx();
2236 } else
2237 res = -EINVAL;
2238
2239 read_unlock(&bond->lock);
2240
2241 return res;
2242}
2243
2244static int bond_info_query(struct net_device *bond_dev, struct ifbond *info)
2245{
2246 struct bonding *bond = netdev_priv(bond_dev);
2247
2248 info->bond_mode = bond->params.mode;
2249 info->miimon = bond->params.miimon;
2250
2251 read_lock(&bond->lock);
2252 info->num_slaves = bond->slave_cnt;
2253 read_unlock(&bond->lock);
2254
2255 return 0;
2256}
2257
2258static int bond_slave_info_query(struct net_device *bond_dev, struct ifslave *info)
2259{
2260 struct bonding *bond = netdev_priv(bond_dev);
2261 struct slave *slave;
2262 int i, res = -ENODEV;
2263
2264 read_lock(&bond->lock);
2265
2266 bond_for_each_slave(bond, slave, i) {
2267 if (i == (int)info->slave_id) {
2268 res = 0;
2269 strcpy(info->slave_name, slave->dev->name);
2270 info->link = slave->link;
2271 info->state = slave->state;
2272 info->link_failure_count = slave->link_failure_count;
2273 break;
2274 }
2275 }
2276
2277 read_unlock(&bond->lock);
2278
2279 return res;
2280}
2281
2282/*-------------------------------- Monitoring -------------------------------*/
2283
2284
2285static int bond_miimon_inspect(struct bonding *bond)
2286{
2287 struct slave *slave;
2288 int i, link_state, commit = 0;
2289 bool ignore_updelay;
2290
2291 ignore_updelay = !bond->curr_active_slave ? true : false;
2292
2293 bond_for_each_slave(bond, slave, i) {
2294 slave->new_link = BOND_LINK_NOCHANGE;
2295
2296 link_state = bond_check_dev_link(bond, slave->dev, 0);
2297
2298 switch (slave->link) {
2299 case BOND_LINK_UP:
2300 if (link_state)
2301 continue;
2302
2303 slave->link = BOND_LINK_FAIL;
2304 slave->delay = bond->params.downdelay;
2305 if (slave->delay) {
2306 pr_info("%s: link status down for %sinterface %s, disabling it in %d ms.\n",
2307 bond->dev->name,
2308 (bond->params.mode ==
2309 BOND_MODE_ACTIVEBACKUP) ?
2310 ((slave->state == BOND_STATE_ACTIVE) ?
2311 "active " : "backup ") : "",
2312 slave->dev->name,
2313 bond->params.downdelay * bond->params.miimon);
2314 }
2315 /*FALLTHRU*/
2316 case BOND_LINK_FAIL:
2317 if (link_state) {
2318 /*
2319 * recovered before downdelay expired
2320 */
2321 slave->link = BOND_LINK_UP;
2322 slave->jiffies = jiffies;
2323 pr_info("%s: link status up again after %d ms for interface %s.\n",
2324 bond->dev->name,
2325 (bond->params.downdelay - slave->delay) *
2326 bond->params.miimon,
2327 slave->dev->name);
2328 continue;
2329 }
2330
2331 if (slave->delay <= 0) {
2332 slave->new_link = BOND_LINK_DOWN;
2333 commit++;
2334 continue;
2335 }
2336
2337 slave->delay--;
2338 break;
2339
2340 case BOND_LINK_DOWN:
2341 if (!link_state)
2342 continue;
2343
2344 slave->link = BOND_LINK_BACK;
2345 slave->delay = bond->params.updelay;
2346
2347 if (slave->delay) {
2348 pr_info("%s: link status up for interface %s, enabling it in %d ms.\n",
2349 bond->dev->name, slave->dev->name,
2350 ignore_updelay ? 0 :
2351 bond->params.updelay *
2352 bond->params.miimon);
2353 }
2354 /*FALLTHRU*/
2355 case BOND_LINK_BACK:
2356 if (!link_state) {
2357 slave->link = BOND_LINK_DOWN;
2358 pr_info("%s: link status down again after %d ms for interface %s.\n",
2359 bond->dev->name,
2360 (bond->params.updelay - slave->delay) *
2361 bond->params.miimon,
2362 slave->dev->name);
2363
2364 continue;
2365 }
2366
2367 if (ignore_updelay)
2368 slave->delay = 0;
2369
2370 if (slave->delay <= 0) {
2371 slave->new_link = BOND_LINK_UP;
2372 commit++;
2373 ignore_updelay = false;
2374 continue;
2375 }
2376
2377 slave->delay--;
2378 break;
2379 }
2380 }
2381
2382 return commit;
2383}
2384
2385static void bond_miimon_commit(struct bonding *bond)
2386{
2387 struct slave *slave;
2388 int i;
2389
2390 bond_for_each_slave(bond, slave, i) {
2391 switch (slave->new_link) {
2392 case BOND_LINK_NOCHANGE:
2393 continue;
2394
2395 case BOND_LINK_UP:
2396 slave->link = BOND_LINK_UP;
2397 slave->jiffies = jiffies;
2398
2399 if (bond->params.mode == BOND_MODE_8023AD) {
2400 /* prevent it from being the active one */
2401 slave->state = BOND_STATE_BACKUP;
2402 } else if (bond->params.mode != BOND_MODE_ACTIVEBACKUP) {
2403 /* make it immediately active */
2404 slave->state = BOND_STATE_ACTIVE;
2405 } else if (slave != bond->primary_slave) {
2406 /* prevent it from being the active one */
2407 slave->state = BOND_STATE_BACKUP;
2408 }
2409
2410 bond_update_speed_duplex(slave);
2411
2412 pr_info("%s: link status definitely up for interface %s, %d Mbps %s duplex.\n",
2413 bond->dev->name, slave->dev->name,
2414 slave->speed, slave->duplex ? "full" : "half");
2415
2416 /* notify ad that the link status has changed */
2417 if (bond->params.mode == BOND_MODE_8023AD)
2418 bond_3ad_handle_link_change(slave, BOND_LINK_UP);
2419
2420 if (bond_is_lb(bond))
2421 bond_alb_handle_link_change(bond, slave,
2422 BOND_LINK_UP);
2423
2424 if (!bond->curr_active_slave ||
2425 (slave == bond->primary_slave))
2426 goto do_failover;
2427
2428 continue;
2429
2430 case BOND_LINK_DOWN:
2431 if (slave->link_failure_count < UINT_MAX)
2432 slave->link_failure_count++;
2433
2434 slave->link = BOND_LINK_DOWN;
2435
2436 if (bond->params.mode == BOND_MODE_ACTIVEBACKUP ||
2437 bond->params.mode == BOND_MODE_8023AD)
2438 bond_set_slave_inactive_flags(slave);
2439
2440 pr_info("%s: link status definitely down for interface %s, disabling it\n",
2441 bond->dev->name, slave->dev->name);
2442
2443 if (bond->params.mode == BOND_MODE_8023AD)
2444 bond_3ad_handle_link_change(slave,
2445 BOND_LINK_DOWN);
2446
2447 if (bond_is_lb(bond))
2448 bond_alb_handle_link_change(bond, slave,
2449 BOND_LINK_DOWN);
2450
2451 if (slave == bond->curr_active_slave)
2452 goto do_failover;
2453
2454 continue;
2455
2456 default:
2457 pr_err("%s: invalid new link %d on slave %s\n",
2458 bond->dev->name, slave->new_link,
2459 slave->dev->name);
2460 slave->new_link = BOND_LINK_NOCHANGE;
2461
2462 continue;
2463 }
2464
2465do_failover:
2466 ASSERT_RTNL();
2467 block_netpoll_tx();
2468 write_lock_bh(&bond->curr_slave_lock);
2469 bond_select_active_slave(bond);
2470 write_unlock_bh(&bond->curr_slave_lock);
2471 unblock_netpoll_tx();
2472 }
2473
2474 bond_set_carrier(bond);
2475}
2476
2477/*
2478 * bond_mii_monitor
2479 *
2480 * Really a wrapper that splits the mii monitor into two phases: an
2481 * inspection, then (if inspection indicates something needs to be done)
2482 * an acquisition of appropriate locks followed by a commit phase to
2483 * implement whatever link state changes are indicated.
2484 */
2485void bond_mii_monitor(struct work_struct *work)
2486{
2487 struct bonding *bond = container_of(work, struct bonding,
2488 mii_work.work);
2489
2490 read_lock(&bond->lock);
2491 if (bond->kill_timers)
2492 goto out;
2493
2494 if (bond->slave_cnt == 0)
2495 goto re_arm;
2496
2497 if (bond->send_grat_arp) {
2498 read_lock(&bond->curr_slave_lock);
2499 bond_send_gratuitous_arp(bond);
2500 read_unlock(&bond->curr_slave_lock);
2501 }
2502
2503 if (bond->send_unsol_na) {
2504 read_lock(&bond->curr_slave_lock);
2505 bond_send_unsolicited_na(bond);
2506 read_unlock(&bond->curr_slave_lock);
2507 }
2508
2509 if (bond_miimon_inspect(bond)) {
2510 read_unlock(&bond->lock);
2511 rtnl_lock();
2512 read_lock(&bond->lock);
2513
2514 bond_miimon_commit(bond);
2515
2516 read_unlock(&bond->lock);
2517 rtnl_unlock(); /* might sleep, hold no other locks */
2518 read_lock(&bond->lock);
2519 }
2520
2521re_arm:
2522 if (bond->params.miimon)
2523 queue_delayed_work(bond->wq, &bond->mii_work,
2524 msecs_to_jiffies(bond->params.miimon));
2525out:
2526 read_unlock(&bond->lock);
2527}
2528
2529static __be32 bond_glean_dev_ip(struct net_device *dev)
2530{
2531 struct in_device *idev;
2532 struct in_ifaddr *ifa;
2533 __be32 addr = 0;
2534
2535 if (!dev)
2536 return 0;
2537
2538 rcu_read_lock();
2539 idev = __in_dev_get_rcu(dev);
2540 if (!idev)
2541 goto out;
2542
2543 ifa = idev->ifa_list;
2544 if (!ifa)
2545 goto out;
2546
2547 addr = ifa->ifa_local;
2548out:
2549 rcu_read_unlock();
2550 return addr;
2551}
2552
2553static int bond_has_this_ip(struct bonding *bond, __be32 ip)
2554{
2555 struct vlan_entry *vlan;
2556
2557 if (ip == bond->master_ip)
2558 return 1;
2559
2560 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
2561 if (ip == vlan->vlan_ip)
2562 return 1;
2563 }
2564
2565 return 0;
2566}
2567
2568/*
2569 * We go to the (large) trouble of VLAN tagging ARP frames because
2570 * switches in VLAN mode (especially if ports are configured as
2571 * "native" to a VLAN) might not pass non-tagged frames.
2572 */
2573static void bond_arp_send(struct net_device *slave_dev, int arp_op, __be32 dest_ip, __be32 src_ip, unsigned short vlan_id)
2574{
2575 struct sk_buff *skb;
2576
2577 pr_debug("arp %d on slave %s: dst %x src %x vid %d\n", arp_op,
2578 slave_dev->name, dest_ip, src_ip, vlan_id);
2579
2580 skb = arp_create(arp_op, ETH_P_ARP, dest_ip, slave_dev, src_ip,
2581 NULL, slave_dev->dev_addr, NULL);
2582
2583 if (!skb) {
2584 pr_err("ARP packet allocation failed\n");
2585 return;
2586 }
2587 if (vlan_id) {
2588 skb = vlan_put_tag(skb, vlan_id);
2589 if (!skb) {
2590 pr_err("failed to insert VLAN tag\n");
2591 return;
2592 }
2593 }
2594 arp_xmit(skb);
2595}
2596
2597
2598static void bond_arp_send_all(struct bonding *bond, struct slave *slave)
2599{
2600 int i, vlan_id, rv;
2601 __be32 *targets = bond->params.arp_targets;
2602 struct vlan_entry *vlan;
2603 struct net_device *vlan_dev;
2604 struct flowi fl;
2605 struct rtable *rt;
2606
2607 for (i = 0; (i < BOND_MAX_ARP_TARGETS); i++) {
2608 if (!targets[i])
2609 break;
2610 pr_debug("basa: target %x\n", targets[i]);
2611 if (!bond->vlgrp) {
2612 pr_debug("basa: empty vlan: arp_send\n");
2613 bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2614 bond->master_ip, 0);
2615 continue;
2616 }
2617
2618 /*
2619 * If VLANs are configured, we do a route lookup to
2620 * determine which VLAN interface would be used, so we
2621 * can tag the ARP with the proper VLAN tag.
2622 */
2623 memset(&fl, 0, sizeof(fl));
2624 fl.fl4_dst = targets[i];
2625 fl.fl4_tos = RTO_ONLINK;
2626
2627 rv = ip_route_output_key(dev_net(bond->dev), &rt, &fl);
2628 if (rv) {
2629 if (net_ratelimit()) {
2630 pr_warning("%s: no route to arp_ip_target %pI4\n",
2631 bond->dev->name, &fl.fl4_dst);
2632 }
2633 continue;
2634 }
2635
2636 /*
2637 * This target is not on a VLAN
2638 */
2639 if (rt->dst.dev == bond->dev) {
2640 ip_rt_put(rt);
2641 pr_debug("basa: rtdev == bond->dev: arp_send\n");
2642 bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2643 bond->master_ip, 0);
2644 continue;
2645 }
2646
2647 vlan_id = 0;
2648 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
2649 vlan_dev = vlan_group_get_device(bond->vlgrp, vlan->vlan_id);
2650 if (vlan_dev == rt->dst.dev) {
2651 vlan_id = vlan->vlan_id;
2652 pr_debug("basa: vlan match on %s %d\n",
2653 vlan_dev->name, vlan_id);
2654 break;
2655 }
2656 }
2657
2658 if (vlan_id) {
2659 ip_rt_put(rt);
2660 bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2661 vlan->vlan_ip, vlan_id);
2662 continue;
2663 }
2664
2665 if (net_ratelimit()) {
2666 pr_warning("%s: no path to arp_ip_target %pI4 via rt.dev %s\n",
2667 bond->dev->name, &fl.fl4_dst,
2668 rt->dst.dev ? rt->dst.dev->name : "NULL");
2669 }
2670 ip_rt_put(rt);
2671 }
2672}
2673
2674/*
2675 * Kick out a gratuitous ARP for an IP on the bonding master plus one
2676 * for each VLAN above us.
2677 *
2678 * Caller must hold curr_slave_lock for read or better
2679 */
2680static void bond_send_gratuitous_arp(struct bonding *bond)
2681{
2682 struct slave *slave = bond->curr_active_slave;
2683 struct vlan_entry *vlan;
2684 struct net_device *vlan_dev;
2685
2686 pr_debug("bond_send_grat_arp: bond %s slave %s\n",
2687 bond->dev->name, slave ? slave->dev->name : "NULL");
2688
2689 if (!slave || !bond->send_grat_arp ||
2690 test_bit(__LINK_STATE_LINKWATCH_PENDING, &slave->dev->state))
2691 return;
2692
2693 bond->send_grat_arp--;
2694
2695 if (bond->master_ip) {
2696 bond_arp_send(slave->dev, ARPOP_REPLY, bond->master_ip,
2697 bond->master_ip, 0);
2698 }
2699
2700 if (!bond->vlgrp)
2701 return;
2702
2703 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
2704 vlan_dev = vlan_group_get_device(bond->vlgrp, vlan->vlan_id);
2705 if (vlan->vlan_ip) {
2706 bond_arp_send(slave->dev, ARPOP_REPLY, vlan->vlan_ip,
2707 vlan->vlan_ip, vlan->vlan_id);
2708 }
2709 }
2710}
2711
2712static void bond_validate_arp(struct bonding *bond, struct slave *slave, __be32 sip, __be32 tip)
2713{
2714 int i;
2715 __be32 *targets = bond->params.arp_targets;
2716
2717 for (i = 0; (i < BOND_MAX_ARP_TARGETS) && targets[i]; i++) {
2718 pr_debug("bva: sip %pI4 tip %pI4 t[%d] %pI4 bhti(tip) %d\n",
2719 &sip, &tip, i, &targets[i],
2720 bond_has_this_ip(bond, tip));
2721 if (sip == targets[i]) {
2722 if (bond_has_this_ip(bond, tip))
2723 slave->last_arp_rx = jiffies;
2724 return;
2725 }
2726 }
2727}
2728
2729static int bond_arp_rcv(struct sk_buff *skb, struct net_device *dev, struct packet_type *pt, struct net_device *orig_dev)
2730{
2731 struct arphdr *arp;
2732 struct slave *slave;
2733 struct bonding *bond;
2734 unsigned char *arp_ptr;
2735 __be32 sip, tip;
2736
2737 if (dev->priv_flags & IFF_802_1Q_VLAN) {
2738 /*
2739 * When using VLANS and bonding, dev and oriv_dev may be
2740 * incorrect if the physical interface supports VLAN
2741 * acceleration. With this change ARP validation now
2742 * works for hosts only reachable on the VLAN interface.
2743 */
2744 dev = vlan_dev_real_dev(dev);
2745 orig_dev = dev_get_by_index_rcu(dev_net(skb->dev),skb->skb_iif);
2746 }
2747
2748 if (!(dev->priv_flags & IFF_BONDING) || !(dev->flags & IFF_MASTER))
2749 goto out;
2750
2751 bond = netdev_priv(dev);
2752 read_lock(&bond->lock);
2753
2754 pr_debug("bond_arp_rcv: bond %s skb->dev %s orig_dev %s\n",
2755 bond->dev->name, skb->dev ? skb->dev->name : "NULL",
2756 orig_dev ? orig_dev->name : "NULL");
2757
2758 slave = bond_get_slave_by_dev(bond, orig_dev);
2759 if (!slave || !slave_do_arp_validate(bond, slave))
2760 goto out_unlock;
2761
2762 if (!pskb_may_pull(skb, arp_hdr_len(dev)))
2763 goto out_unlock;
2764
2765 arp = arp_hdr(skb);
2766 if (arp->ar_hln != dev->addr_len ||
2767 skb->pkt_type == PACKET_OTHERHOST ||
2768 skb->pkt_type == PACKET_LOOPBACK ||
2769 arp->ar_hrd != htons(ARPHRD_ETHER) ||
2770 arp->ar_pro != htons(ETH_P_IP) ||
2771 arp->ar_pln != 4)
2772 goto out_unlock;
2773
2774 arp_ptr = (unsigned char *)(arp + 1);
2775 arp_ptr += dev->addr_len;
2776 memcpy(&sip, arp_ptr, 4);
2777 arp_ptr += 4 + dev->addr_len;
2778 memcpy(&tip, arp_ptr, 4);
2779
2780 pr_debug("bond_arp_rcv: %s %s/%d av %d sv %d sip %pI4 tip %pI4\n",
2781 bond->dev->name, slave->dev->name, slave->state,
2782 bond->params.arp_validate, slave_do_arp_validate(bond, slave),
2783 &sip, &tip);
2784
2785 /*
2786 * Backup slaves won't see the ARP reply, but do come through
2787 * here for each ARP probe (so we swap the sip/tip to validate
2788 * the probe). In a "redundant switch, common router" type of
2789 * configuration, the ARP probe will (hopefully) travel from
2790 * the active, through one switch, the router, then the other
2791 * switch before reaching the backup.
2792 */
2793 if (slave->state == BOND_STATE_ACTIVE)
2794 bond_validate_arp(bond, slave, sip, tip);
2795 else
2796 bond_validate_arp(bond, slave, tip, sip);
2797
2798out_unlock:
2799 read_unlock(&bond->lock);
2800out:
2801 dev_kfree_skb(skb);
2802 return NET_RX_SUCCESS;
2803}
2804
2805/*
2806 * this function is called regularly to monitor each slave's link
2807 * ensuring that traffic is being sent and received when arp monitoring
2808 * is used in load-balancing mode. if the adapter has been dormant, then an
2809 * arp is transmitted to generate traffic. see activebackup_arp_monitor for
2810 * arp monitoring in active backup mode.
2811 */
2812void bond_loadbalance_arp_mon(struct work_struct *work)
2813{
2814 struct bonding *bond = container_of(work, struct bonding,
2815 arp_work.work);
2816 struct slave *slave, *oldcurrent;
2817 int do_failover = 0;
2818 int delta_in_ticks;
2819 int i;
2820
2821 read_lock(&bond->lock);
2822
2823 delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
2824
2825 if (bond->kill_timers)
2826 goto out;
2827
2828 if (bond->slave_cnt == 0)
2829 goto re_arm;
2830
2831 read_lock(&bond->curr_slave_lock);
2832 oldcurrent = bond->curr_active_slave;
2833 read_unlock(&bond->curr_slave_lock);
2834
2835 /* see if any of the previous devices are up now (i.e. they have
2836 * xmt and rcv traffic). the curr_active_slave does not come into
2837 * the picture unless it is null. also, slave->jiffies is not needed
2838 * here because we send an arp on each slave and give a slave as
2839 * long as it needs to get the tx/rx within the delta.
2840 * TODO: what about up/down delay in arp mode? it wasn't here before
2841 * so it can wait
2842 */
2843 bond_for_each_slave(bond, slave, i) {
2844 unsigned long trans_start = dev_trans_start(slave->dev);
2845
2846 if (slave->link != BOND_LINK_UP) {
2847 if (time_in_range(jiffies,
2848 trans_start - delta_in_ticks,
2849 trans_start + delta_in_ticks) &&
2850 time_in_range(jiffies,
2851 slave->dev->last_rx - delta_in_ticks,
2852 slave->dev->last_rx + delta_in_ticks)) {
2853
2854 slave->link = BOND_LINK_UP;
2855 slave->state = BOND_STATE_ACTIVE;
2856
2857 /* primary_slave has no meaning in round-robin
2858 * mode. the window of a slave being up and
2859 * curr_active_slave being null after enslaving
2860 * is closed.
2861 */
2862 if (!oldcurrent) {
2863 pr_info("%s: link status definitely up for interface %s, ",
2864 bond->dev->name,
2865 slave->dev->name);
2866 do_failover = 1;
2867 } else {
2868 pr_info("%s: interface %s is now up\n",
2869 bond->dev->name,
2870 slave->dev->name);
2871 }
2872 }
2873 } else {
2874 /* slave->link == BOND_LINK_UP */
2875
2876 /* not all switches will respond to an arp request
2877 * when the source ip is 0, so don't take the link down
2878 * if we don't know our ip yet
2879 */
2880 if (!time_in_range(jiffies,
2881 trans_start - delta_in_ticks,
2882 trans_start + 2 * delta_in_ticks) ||
2883 !time_in_range(jiffies,
2884 slave->dev->last_rx - delta_in_ticks,
2885 slave->dev->last_rx + 2 * delta_in_ticks)) {
2886
2887 slave->link = BOND_LINK_DOWN;
2888 slave->state = BOND_STATE_BACKUP;
2889
2890 if (slave->link_failure_count < UINT_MAX)
2891 slave->link_failure_count++;
2892
2893 pr_info("%s: interface %s is now down.\n",
2894 bond->dev->name,
2895 slave->dev->name);
2896
2897 if (slave == oldcurrent)
2898 do_failover = 1;
2899 }
2900 }
2901
2902 /* note: if switch is in round-robin mode, all links
2903 * must tx arp to ensure all links rx an arp - otherwise
2904 * links may oscillate or not come up at all; if switch is
2905 * in something like xor mode, there is nothing we can
2906 * do - all replies will be rx'ed on same link causing slaves
2907 * to be unstable during low/no traffic periods
2908 */
2909 if (IS_UP(slave->dev))
2910 bond_arp_send_all(bond, slave);
2911 }
2912
2913 if (do_failover) {
2914 block_netpoll_tx();
2915 write_lock_bh(&bond->curr_slave_lock);
2916
2917 bond_select_active_slave(bond);
2918
2919 write_unlock_bh(&bond->curr_slave_lock);
2920 unblock_netpoll_tx();
2921 }
2922
2923re_arm:
2924 if (bond->params.arp_interval)
2925 queue_delayed_work(bond->wq, &bond->arp_work, delta_in_ticks);
2926out:
2927 read_unlock(&bond->lock);
2928}
2929
2930/*
2931 * Called to inspect slaves for active-backup mode ARP monitor link state
2932 * changes. Sets new_link in slaves to specify what action should take
2933 * place for the slave. Returns 0 if no changes are found, >0 if changes
2934 * to link states must be committed.
2935 *
2936 * Called with bond->lock held for read.
2937 */
2938static int bond_ab_arp_inspect(struct bonding *bond, int delta_in_ticks)
2939{
2940 struct slave *slave;
2941 int i, commit = 0;
2942 unsigned long trans_start;
2943
2944 bond_for_each_slave(bond, slave, i) {
2945 slave->new_link = BOND_LINK_NOCHANGE;
2946
2947 if (slave->link != BOND_LINK_UP) {
2948 if (time_in_range(jiffies,
2949 slave_last_rx(bond, slave) - delta_in_ticks,
2950 slave_last_rx(bond, slave) + delta_in_ticks)) {
2951
2952 slave->new_link = BOND_LINK_UP;
2953 commit++;
2954 }
2955
2956 continue;
2957 }
2958
2959 /*
2960 * Give slaves 2*delta after being enslaved or made
2961 * active. This avoids bouncing, as the last receive
2962 * times need a full ARP monitor cycle to be updated.
2963 */
2964 if (time_in_range(jiffies,
2965 slave->jiffies - delta_in_ticks,
2966 slave->jiffies + 2 * delta_in_ticks))
2967 continue;
2968
2969 /*
2970 * Backup slave is down if:
2971 * - No current_arp_slave AND
2972 * - more than 3*delta since last receive AND
2973 * - the bond has an IP address
2974 *
2975 * Note: a non-null current_arp_slave indicates
2976 * the curr_active_slave went down and we are
2977 * searching for a new one; under this condition
2978 * we only take the curr_active_slave down - this
2979 * gives each slave a chance to tx/rx traffic
2980 * before being taken out
2981 */
2982 if (slave->state == BOND_STATE_BACKUP &&
2983 !bond->current_arp_slave &&
2984 !time_in_range(jiffies,
2985 slave_last_rx(bond, slave) - delta_in_ticks,
2986 slave_last_rx(bond, slave) + 3 * delta_in_ticks)) {
2987
2988 slave->new_link = BOND_LINK_DOWN;
2989 commit++;
2990 }
2991
2992 /*
2993 * Active slave is down if:
2994 * - more than 2*delta since transmitting OR
2995 * - (more than 2*delta since receive AND
2996 * the bond has an IP address)
2997 */
2998 trans_start = dev_trans_start(slave->dev);
2999 if ((slave->state == BOND_STATE_ACTIVE) &&
3000 (!time_in_range(jiffies,
3001 trans_start - delta_in_ticks,
3002 trans_start + 2 * delta_in_ticks) ||
3003 !time_in_range(jiffies,
3004 slave_last_rx(bond, slave) - delta_in_ticks,
3005 slave_last_rx(bond, slave) + 2 * delta_in_ticks))) {
3006
3007 slave->new_link = BOND_LINK_DOWN;
3008 commit++;
3009 }
3010 }
3011
3012 return commit;
3013}
3014
3015/*
3016 * Called to commit link state changes noted by inspection step of
3017 * active-backup mode ARP monitor.
3018 *
3019 * Called with RTNL and bond->lock for read.
3020 */
3021static void bond_ab_arp_commit(struct bonding *bond, int delta_in_ticks)
3022{
3023 struct slave *slave;
3024 int i;
3025 unsigned long trans_start;
3026
3027 bond_for_each_slave(bond, slave, i) {
3028 switch (slave->new_link) {
3029 case BOND_LINK_NOCHANGE:
3030 continue;
3031
3032 case BOND_LINK_UP:
3033 trans_start = dev_trans_start(slave->dev);
3034 if ((!bond->curr_active_slave &&
3035 time_in_range(jiffies,
3036 trans_start - delta_in_ticks,
3037 trans_start + delta_in_ticks)) ||
3038 bond->curr_active_slave != slave) {
3039 slave->link = BOND_LINK_UP;
3040 bond->current_arp_slave = NULL;
3041
3042 pr_info("%s: link status definitely up for interface %s.\n",
3043 bond->dev->name, slave->dev->name);
3044
3045 if (!bond->curr_active_slave ||
3046 (slave == bond->primary_slave))
3047 goto do_failover;
3048
3049 }
3050
3051 continue;
3052
3053 case BOND_LINK_DOWN:
3054 if (slave->link_failure_count < UINT_MAX)
3055 slave->link_failure_count++;
3056
3057 slave->link = BOND_LINK_DOWN;
3058 bond_set_slave_inactive_flags(slave);
3059
3060 pr_info("%s: link status definitely down for interface %s, disabling it\n",
3061 bond->dev->name, slave->dev->name);
3062
3063 if (slave == bond->curr_active_slave) {
3064 bond->current_arp_slave = NULL;
3065 goto do_failover;
3066 }
3067
3068 continue;
3069
3070 default:
3071 pr_err("%s: impossible: new_link %d on slave %s\n",
3072 bond->dev->name, slave->new_link,
3073 slave->dev->name);
3074 continue;
3075 }
3076
3077do_failover:
3078 ASSERT_RTNL();
3079 block_netpoll_tx();
3080 write_lock_bh(&bond->curr_slave_lock);
3081 bond_select_active_slave(bond);
3082 write_unlock_bh(&bond->curr_slave_lock);
3083 unblock_netpoll_tx();
3084 }
3085
3086 bond_set_carrier(bond);
3087}
3088
3089/*
3090 * Send ARP probes for active-backup mode ARP monitor.
3091 *
3092 * Called with bond->lock held for read.
3093 */
3094static void bond_ab_arp_probe(struct bonding *bond)
3095{
3096 struct slave *slave;
3097 int i;
3098
3099 read_lock(&bond->curr_slave_lock);
3100
3101 if (bond->current_arp_slave && bond->curr_active_slave)
3102 pr_info("PROBE: c_arp %s && cas %s BAD\n",
3103 bond->current_arp_slave->dev->name,
3104 bond->curr_active_slave->dev->name);
3105
3106 if (bond->curr_active_slave) {
3107 bond_arp_send_all(bond, bond->curr_active_slave);
3108 read_unlock(&bond->curr_slave_lock);
3109 return;
3110 }
3111
3112 read_unlock(&bond->curr_slave_lock);
3113
3114 /* if we don't have a curr_active_slave, search for the next available
3115 * backup slave from the current_arp_slave and make it the candidate
3116 * for becoming the curr_active_slave
3117 */
3118
3119 if (!bond->current_arp_slave) {
3120 bond->current_arp_slave = bond->first_slave;
3121 if (!bond->current_arp_slave)
3122 return;
3123 }
3124
3125 bond_set_slave_inactive_flags(bond->current_arp_slave);
3126
3127 /* search for next candidate */
3128 bond_for_each_slave_from(bond, slave, i, bond->current_arp_slave->next) {
3129 if (IS_UP(slave->dev)) {
3130 slave->link = BOND_LINK_BACK;
3131 bond_set_slave_active_flags(slave);
3132 bond_arp_send_all(bond, slave);
3133 slave->jiffies = jiffies;
3134 bond->current_arp_slave = slave;
3135 break;
3136 }
3137
3138 /* if the link state is up at this point, we
3139 * mark it down - this can happen if we have
3140 * simultaneous link failures and
3141 * reselect_active_interface doesn't make this
3142 * one the current slave so it is still marked
3143 * up when it is actually down
3144 */
3145 if (slave->link == BOND_LINK_UP) {
3146 slave->link = BOND_LINK_DOWN;
3147 if (slave->link_failure_count < UINT_MAX)
3148 slave->link_failure_count++;
3149
3150 bond_set_slave_inactive_flags(slave);
3151
3152 pr_info("%s: backup interface %s is now down.\n",
3153 bond->dev->name, slave->dev->name);
3154 }
3155 }
3156}
3157
3158void bond_activebackup_arp_mon(struct work_struct *work)
3159{
3160 struct bonding *bond = container_of(work, struct bonding,
3161 arp_work.work);
3162 int delta_in_ticks;
3163
3164 read_lock(&bond->lock);
3165
3166 if (bond->kill_timers)
3167 goto out;
3168
3169 delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
3170
3171 if (bond->slave_cnt == 0)
3172 goto re_arm;
3173
3174 if (bond->send_grat_arp) {
3175 read_lock(&bond->curr_slave_lock);
3176 bond_send_gratuitous_arp(bond);
3177 read_unlock(&bond->curr_slave_lock);
3178 }
3179
3180 if (bond->send_unsol_na) {
3181 read_lock(&bond->curr_slave_lock);
3182 bond_send_unsolicited_na(bond);
3183 read_unlock(&bond->curr_slave_lock);
3184 }
3185
3186 if (bond_ab_arp_inspect(bond, delta_in_ticks)) {
3187 read_unlock(&bond->lock);
3188 rtnl_lock();
3189 read_lock(&bond->lock);
3190
3191 bond_ab_arp_commit(bond, delta_in_ticks);
3192
3193 read_unlock(&bond->lock);
3194 rtnl_unlock();
3195 read_lock(&bond->lock);
3196 }
3197
3198 bond_ab_arp_probe(bond);
3199
3200re_arm:
3201 if (bond->params.arp_interval)
3202 queue_delayed_work(bond->wq, &bond->arp_work, delta_in_ticks);
3203out:
3204 read_unlock(&bond->lock);
3205}
3206
3207/*------------------------------ proc/seq_file-------------------------------*/
3208
3209#ifdef CONFIG_PROC_FS
3210
3211static void *bond_info_seq_start(struct seq_file *seq, loff_t *pos)
3212 __acquires(&dev_base_lock)
3213 __acquires(&bond->lock)
3214{
3215 struct bonding *bond = seq->private;
3216 loff_t off = 0;
3217 struct slave *slave;
3218 int i;
3219
3220 /* make sure the bond won't be taken away */
3221 read_lock(&dev_base_lock);
3222 read_lock(&bond->lock);
3223
3224 if (*pos == 0)
3225 return SEQ_START_TOKEN;
3226
3227 bond_for_each_slave(bond, slave, i) {
3228 if (++off == *pos)
3229 return slave;
3230 }
3231
3232 return NULL;
3233}
3234
3235static void *bond_info_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3236{
3237 struct bonding *bond = seq->private;
3238 struct slave *slave = v;
3239
3240 ++*pos;
3241 if (v == SEQ_START_TOKEN)
3242 return bond->first_slave;
3243
3244 slave = slave->next;
3245
3246 return (slave == bond->first_slave) ? NULL : slave;
3247}
3248
3249static void bond_info_seq_stop(struct seq_file *seq, void *v)
3250 __releases(&bond->lock)
3251 __releases(&dev_base_lock)
3252{
3253 struct bonding *bond = seq->private;
3254
3255 read_unlock(&bond->lock);
3256 read_unlock(&dev_base_lock);
3257}
3258
3259static void bond_info_show_master(struct seq_file *seq)
3260{
3261 struct bonding *bond = seq->private;
3262 struct slave *curr;
3263 int i;
3264
3265 read_lock(&bond->curr_slave_lock);
3266 curr = bond->curr_active_slave;
3267 read_unlock(&bond->curr_slave_lock);
3268
3269 seq_printf(seq, "Bonding Mode: %s",
3270 bond_mode_name(bond->params.mode));
3271
3272 if (bond->params.mode == BOND_MODE_ACTIVEBACKUP &&
3273 bond->params.fail_over_mac)
3274 seq_printf(seq, " (fail_over_mac %s)",
3275 fail_over_mac_tbl[bond->params.fail_over_mac].modename);
3276
3277 seq_printf(seq, "\n");
3278
3279 if (bond->params.mode == BOND_MODE_XOR ||
3280 bond->params.mode == BOND_MODE_8023AD) {
3281 seq_printf(seq, "Transmit Hash Policy: %s (%d)\n",
3282 xmit_hashtype_tbl[bond->params.xmit_policy].modename,
3283 bond->params.xmit_policy);
3284 }
3285
3286 if (USES_PRIMARY(bond->params.mode)) {
3287 seq_printf(seq, "Primary Slave: %s",
3288 (bond->primary_slave) ?
3289 bond->primary_slave->dev->name : "None");
3290 if (bond->primary_slave)
3291 seq_printf(seq, " (primary_reselect %s)",
3292 pri_reselect_tbl[bond->params.primary_reselect].modename);
3293
3294 seq_printf(seq, "\nCurrently Active Slave: %s\n",
3295 (curr) ? curr->dev->name : "None");
3296 }
3297
3298 seq_printf(seq, "MII Status: %s\n", netif_carrier_ok(bond->dev) ?
3299 "up" : "down");
3300 seq_printf(seq, "MII Polling Interval (ms): %d\n", bond->params.miimon);
3301 seq_printf(seq, "Up Delay (ms): %d\n",
3302 bond->params.updelay * bond->params.miimon);
3303 seq_printf(seq, "Down Delay (ms): %d\n",
3304 bond->params.downdelay * bond->params.miimon);
3305
3306
3307 /* ARP information */
3308 if (bond->params.arp_interval > 0) {
3309 int printed = 0;
3310 seq_printf(seq, "ARP Polling Interval (ms): %d\n",
3311 bond->params.arp_interval);
3312
3313 seq_printf(seq, "ARP IP target/s (n.n.n.n form):");
3314
3315 for (i = 0; (i < BOND_MAX_ARP_TARGETS); i++) {
3316 if (!bond->params.arp_targets[i])
3317 break;
3318 if (printed)
3319 seq_printf(seq, ",");
3320 seq_printf(seq, " %pI4", &bond->params.arp_targets[i]);
3321 printed = 1;
3322 }
3323 seq_printf(seq, "\n");
3324 }
3325
3326 if (bond->params.mode == BOND_MODE_8023AD) {
3327 struct ad_info ad_info;
3328
3329 seq_puts(seq, "\n802.3ad info\n");
3330 seq_printf(seq, "LACP rate: %s\n",
3331 (bond->params.lacp_fast) ? "fast" : "slow");
3332 seq_printf(seq, "Aggregator selection policy (ad_select): %s\n",
3333 ad_select_tbl[bond->params.ad_select].modename);
3334
3335 if (bond_3ad_get_active_agg_info(bond, &ad_info)) {
3336 seq_printf(seq, "bond %s has no active aggregator\n",
3337 bond->dev->name);
3338 } else {
3339 seq_printf(seq, "Active Aggregator Info:\n");
3340
3341 seq_printf(seq, "\tAggregator ID: %d\n",
3342 ad_info.aggregator_id);
3343 seq_printf(seq, "\tNumber of ports: %d\n",
3344 ad_info.ports);
3345 seq_printf(seq, "\tActor Key: %d\n",
3346 ad_info.actor_key);
3347 seq_printf(seq, "\tPartner Key: %d\n",
3348 ad_info.partner_key);
3349 seq_printf(seq, "\tPartner Mac Address: %pM\n",
3350 ad_info.partner_system);
3351 }
3352 }
3353}
3354
3355static void bond_info_show_slave(struct seq_file *seq,
3356 const struct slave *slave)
3357{
3358 struct bonding *bond = seq->private;
3359
3360 seq_printf(seq, "\nSlave Interface: %s\n", slave->dev->name);
3361 seq_printf(seq, "MII Status: %s\n",
3362 (slave->link == BOND_LINK_UP) ? "up" : "down");
3363 seq_printf(seq, "Speed: %d Mbps\n", slave->speed);
3364 seq_printf(seq, "Duplex: %s\n", slave->duplex ? "full" : "half");
3365 seq_printf(seq, "Link Failure Count: %u\n",
3366 slave->link_failure_count);
3367
3368 seq_printf(seq, "Permanent HW addr: %pM\n", slave->perm_hwaddr);
3369
3370 if (bond->params.mode == BOND_MODE_8023AD) {
3371 const struct aggregator *agg
3372 = SLAVE_AD_INFO(slave).port.aggregator;
3373
3374 if (agg)
3375 seq_printf(seq, "Aggregator ID: %d\n",
3376 agg->aggregator_identifier);
3377 else
3378 seq_puts(seq, "Aggregator ID: N/A\n");
3379 }
3380 seq_printf(seq, "Slave queue ID: %d\n", slave->queue_id);
3381}
3382
3383static int bond_info_seq_show(struct seq_file *seq, void *v)
3384{
3385 if (v == SEQ_START_TOKEN) {
3386 seq_printf(seq, "%s\n", version);
3387 bond_info_show_master(seq);
3388 } else
3389 bond_info_show_slave(seq, v);
3390
3391 return 0;
3392}
3393
3394static const struct seq_operations bond_info_seq_ops = {
3395 .start = bond_info_seq_start,
3396 .next = bond_info_seq_next,
3397 .stop = bond_info_seq_stop,
3398 .show = bond_info_seq_show,
3399};
3400
3401static int bond_info_open(struct inode *inode, struct file *file)
3402{
3403 struct seq_file *seq;
3404 struct proc_dir_entry *proc;
3405 int res;
3406
3407 res = seq_open(file, &bond_info_seq_ops);
3408 if (!res) {
3409 /* recover the pointer buried in proc_dir_entry data */
3410 seq = file->private_data;
3411 proc = PDE(inode);
3412 seq->private = proc->data;
3413 }
3414
3415 return res;
3416}
3417
3418static const struct file_operations bond_info_fops = {
3419 .owner = THIS_MODULE,
3420 .open = bond_info_open,
3421 .read = seq_read,
3422 .llseek = seq_lseek,
3423 .release = seq_release,
3424};
3425
3426static void bond_create_proc_entry(struct bonding *bond)
3427{
3428 struct net_device *bond_dev = bond->dev;
3429 struct bond_net *bn = net_generic(dev_net(bond_dev), bond_net_id);
3430
3431 if (bn->proc_dir) {
3432 bond->proc_entry = proc_create_data(bond_dev->name,
3433 S_IRUGO, bn->proc_dir,
3434 &bond_info_fops, bond);
3435 if (bond->proc_entry == NULL)
3436 pr_warning("Warning: Cannot create /proc/net/%s/%s\n",
3437 DRV_NAME, bond_dev->name);
3438 else
3439 memcpy(bond->proc_file_name, bond_dev->name, IFNAMSIZ);
3440 }
3441}
3442
3443static void bond_remove_proc_entry(struct bonding *bond)
3444{
3445 struct net_device *bond_dev = bond->dev;
3446 struct bond_net *bn = net_generic(dev_net(bond_dev), bond_net_id);
3447
3448 if (bn->proc_dir && bond->proc_entry) {
3449 remove_proc_entry(bond->proc_file_name, bn->proc_dir);
3450 memset(bond->proc_file_name, 0, IFNAMSIZ);
3451 bond->proc_entry = NULL;
3452 }
3453}
3454
3455/* Create the bonding directory under /proc/net, if doesn't exist yet.
3456 * Caller must hold rtnl_lock.
3457 */
3458static void __net_init bond_create_proc_dir(struct bond_net *bn)
3459{
3460 if (!bn->proc_dir) {
3461 bn->proc_dir = proc_mkdir(DRV_NAME, bn->net->proc_net);
3462 if (!bn->proc_dir)
3463 pr_warning("Warning: cannot create /proc/net/%s\n",
3464 DRV_NAME);
3465 }
3466}
3467
3468/* Destroy the bonding directory under /proc/net, if empty.
3469 * Caller must hold rtnl_lock.
3470 */
3471static void __net_exit bond_destroy_proc_dir(struct bond_net *bn)
3472{
3473 if (bn->proc_dir) {
3474 remove_proc_entry(DRV_NAME, bn->net->proc_net);
3475 bn->proc_dir = NULL;
3476 }
3477}
3478
3479#else /* !CONFIG_PROC_FS */
3480
3481static void bond_create_proc_entry(struct bonding *bond)
3482{
3483}
3484
3485static void bond_remove_proc_entry(struct bonding *bond)
3486{
3487}
3488
3489static inline void bond_create_proc_dir(struct bond_net *bn)
3490{
3491}
3492
3493static inline void bond_destroy_proc_dir(struct bond_net *bn)
3494{
3495}
3496
3497#endif /* CONFIG_PROC_FS */
3498
3499
3500/*-------------------------- netdev event handling --------------------------*/
3501
3502/*
3503 * Change device name
3504 */
3505static int bond_event_changename(struct bonding *bond)
3506{
3507 bond_remove_proc_entry(bond);
3508 bond_create_proc_entry(bond);
3509
3510 return NOTIFY_DONE;
3511}
3512
3513static int bond_master_netdev_event(unsigned long event,
3514 struct net_device *bond_dev)
3515{
3516 struct bonding *event_bond = netdev_priv(bond_dev);
3517
3518 switch (event) {
3519 case NETDEV_CHANGENAME:
3520 return bond_event_changename(event_bond);
3521 default:
3522 break;
3523 }
3524
3525 return NOTIFY_DONE;
3526}
3527
3528static int bond_slave_netdev_event(unsigned long event,
3529 struct net_device *slave_dev)
3530{
3531 struct net_device *bond_dev = slave_dev->master;
3532 struct bonding *bond = netdev_priv(bond_dev);
3533
3534 switch (event) {
3535 case NETDEV_UNREGISTER:
3536 if (bond_dev) {
3537 if (bond->setup_by_slave)
3538 bond_release_and_destroy(bond_dev, slave_dev);
3539 else
3540 bond_release(bond_dev, slave_dev);
3541 }
3542 break;
3543 case NETDEV_CHANGE:
3544 if (bond->params.mode == BOND_MODE_8023AD || bond_is_lb(bond)) {
3545 struct slave *slave;
3546
3547 slave = bond_get_slave_by_dev(bond, slave_dev);
3548 if (slave) {
3549 u16 old_speed = slave->speed;
3550 u16 old_duplex = slave->duplex;
3551
3552 bond_update_speed_duplex(slave);
3553
3554 if (bond_is_lb(bond))
3555 break;
3556
3557 if (old_speed != slave->speed)
3558 bond_3ad_adapter_speed_changed(slave);
3559 if (old_duplex != slave->duplex)
3560 bond_3ad_adapter_duplex_changed(slave);
3561 }
3562 }
3563
3564 break;
3565 case NETDEV_DOWN:
3566 /*
3567 * ... Or is it this?
3568 */
3569 break;
3570 case NETDEV_CHANGEMTU:
3571 /*
3572 * TODO: Should slaves be allowed to
3573 * independently alter their MTU? For
3574 * an active-backup bond, slaves need
3575 * not be the same type of device, so
3576 * MTUs may vary. For other modes,
3577 * slaves arguably should have the
3578 * same MTUs. To do this, we'd need to
3579 * take over the slave's change_mtu
3580 * function for the duration of their
3581 * servitude.
3582 */
3583 break;
3584 case NETDEV_CHANGENAME:
3585 /*
3586 * TODO: handle changing the primary's name
3587 */
3588 break;
3589 case NETDEV_FEAT_CHANGE:
3590 bond_compute_features(bond);
3591 break;
3592 default:
3593 break;
3594 }
3595
3596 return NOTIFY_DONE;
3597}
3598
3599/*
3600 * bond_netdev_event: handle netdev notifier chain events.
3601 *
3602 * This function receives events for the netdev chain. The caller (an
3603 * ioctl handler calling blocking_notifier_call_chain) holds the necessary
3604 * locks for us to safely manipulate the slave devices (RTNL lock,
3605 * dev_probe_lock).
3606 */
3607static int bond_netdev_event(struct notifier_block *this,
3608 unsigned long event, void *ptr)
3609{
3610 struct net_device *event_dev = (struct net_device *)ptr;
3611
3612 pr_debug("event_dev: %s, event: %lx\n",
3613 event_dev ? event_dev->name : "None",
3614 event);
3615
3616 if (!(event_dev->priv_flags & IFF_BONDING))
3617 return NOTIFY_DONE;
3618
3619 if (event_dev->flags & IFF_MASTER) {
3620 pr_debug("IFF_MASTER\n");
3621 return bond_master_netdev_event(event, event_dev);
3622 }
3623
3624 if (event_dev->flags & IFF_SLAVE) {
3625 pr_debug("IFF_SLAVE\n");
3626 return bond_slave_netdev_event(event, event_dev);
3627 }
3628
3629 return NOTIFY_DONE;
3630}
3631
3632/*
3633 * bond_inetaddr_event: handle inetaddr notifier chain events.
3634 *
3635 * We keep track of device IPs primarily to use as source addresses in
3636 * ARP monitor probes (rather than spewing out broadcasts all the time).
3637 *
3638 * We track one IP for the main device (if it has one), plus one per VLAN.
3639 */
3640static int bond_inetaddr_event(struct notifier_block *this, unsigned long event, void *ptr)
3641{
3642 struct in_ifaddr *ifa = ptr;
3643 struct net_device *vlan_dev, *event_dev = ifa->ifa_dev->dev;
3644 struct bond_net *bn = net_generic(dev_net(event_dev), bond_net_id);
3645 struct bonding *bond;
3646 struct vlan_entry *vlan;
3647
3648 list_for_each_entry(bond, &bn->dev_list, bond_list) {
3649 if (bond->dev == event_dev) {
3650 switch (event) {
3651 case NETDEV_UP:
3652 bond->master_ip = ifa->ifa_local;
3653 return NOTIFY_OK;
3654 case NETDEV_DOWN:
3655 bond->master_ip = bond_glean_dev_ip(bond->dev);
3656 return NOTIFY_OK;
3657 default:
3658 return NOTIFY_DONE;
3659 }
3660 }
3661
3662 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
3663 if (!bond->vlgrp)
3664 continue;
3665 vlan_dev = vlan_group_get_device(bond->vlgrp, vlan->vlan_id);
3666 if (vlan_dev == event_dev) {
3667 switch (event) {
3668 case NETDEV_UP:
3669 vlan->vlan_ip = ifa->ifa_local;
3670 return NOTIFY_OK;
3671 case NETDEV_DOWN:
3672 vlan->vlan_ip =
3673 bond_glean_dev_ip(vlan_dev);
3674 return NOTIFY_OK;
3675 default:
3676 return NOTIFY_DONE;
3677 }
3678 }
3679 }
3680 }
3681 return NOTIFY_DONE;
3682}
3683
3684static struct notifier_block bond_netdev_notifier = {
3685 .notifier_call = bond_netdev_event,
3686};
3687
3688static struct notifier_block bond_inetaddr_notifier = {
3689 .notifier_call = bond_inetaddr_event,
3690};
3691
3692/*-------------------------- Packet type handling ---------------------------*/
3693
3694/* register to receive lacpdus on a bond */
3695static void bond_register_lacpdu(struct bonding *bond)
3696{
3697 struct packet_type *pk_type = &(BOND_AD_INFO(bond).ad_pkt_type);
3698
3699 /* initialize packet type */
3700 pk_type->type = PKT_TYPE_LACPDU;
3701 pk_type->dev = bond->dev;
3702 pk_type->func = bond_3ad_lacpdu_recv;
3703
3704 dev_add_pack(pk_type);
3705}
3706
3707/* unregister to receive lacpdus on a bond */
3708static void bond_unregister_lacpdu(struct bonding *bond)
3709{
3710 dev_remove_pack(&(BOND_AD_INFO(bond).ad_pkt_type));
3711}
3712
3713void bond_register_arp(struct bonding *bond)
3714{
3715 struct packet_type *pt = &bond->arp_mon_pt;
3716
3717 if (pt->type)
3718 return;
3719
3720 pt->type = htons(ETH_P_ARP);
3721 pt->dev = bond->dev;
3722 pt->func = bond_arp_rcv;
3723 dev_add_pack(pt);
3724}
3725
3726void bond_unregister_arp(struct bonding *bond)
3727{
3728 struct packet_type *pt = &bond->arp_mon_pt;
3729
3730 dev_remove_pack(pt);
3731 pt->type = 0;
3732}
3733
3734/*---------------------------- Hashing Policies -----------------------------*/
3735
3736/*
3737 * Hash for the output device based upon layer 2 and layer 3 data. If
3738 * the packet is not IP mimic bond_xmit_hash_policy_l2()
3739 */
3740static int bond_xmit_hash_policy_l23(struct sk_buff *skb, int count)
3741{
3742 struct ethhdr *data = (struct ethhdr *)skb->data;
3743 struct iphdr *iph = ip_hdr(skb);
3744
3745 if (skb->protocol == htons(ETH_P_IP)) {
3746 return ((ntohl(iph->saddr ^ iph->daddr) & 0xffff) ^
3747 (data->h_dest[5] ^ data->h_source[5])) % count;
3748 }
3749
3750 return (data->h_dest[5] ^ data->h_source[5]) % count;
3751}
3752
3753/*
3754 * Hash for the output device based upon layer 3 and layer 4 data. If
3755 * the packet is a frag or not TCP or UDP, just use layer 3 data. If it is
3756 * altogether not IP, mimic bond_xmit_hash_policy_l2()
3757 */
3758static int bond_xmit_hash_policy_l34(struct sk_buff *skb, int count)
3759{
3760 struct ethhdr *data = (struct ethhdr *)skb->data;
3761 struct iphdr *iph = ip_hdr(skb);
3762 __be16 *layer4hdr = (__be16 *)((u32 *)iph + iph->ihl);
3763 int layer4_xor = 0;
3764
3765 if (skb->protocol == htons(ETH_P_IP)) {
3766 if (!(iph->frag_off & htons(IP_MF|IP_OFFSET)) &&
3767 (iph->protocol == IPPROTO_TCP ||
3768 iph->protocol == IPPROTO_UDP)) {
3769 layer4_xor = ntohs((*layer4hdr ^ *(layer4hdr + 1)));
3770 }
3771 return (layer4_xor ^
3772 ((ntohl(iph->saddr ^ iph->daddr)) & 0xffff)) % count;
3773
3774 }
3775
3776 return (data->h_dest[5] ^ data->h_source[5]) % count;
3777}
3778
3779/*
3780 * Hash for the output device based upon layer 2 data
3781 */
3782static int bond_xmit_hash_policy_l2(struct sk_buff *skb, int count)
3783{
3784 struct ethhdr *data = (struct ethhdr *)skb->data;
3785
3786 return (data->h_dest[5] ^ data->h_source[5]) % count;
3787}
3788
3789/*-------------------------- Device entry points ----------------------------*/
3790
3791static int bond_open(struct net_device *bond_dev)
3792{
3793 struct bonding *bond = netdev_priv(bond_dev);
3794
3795 bond->kill_timers = 0;
3796
3797 INIT_DELAYED_WORK(&bond->mcast_work, bond_resend_igmp_join_requests_delayed);
3798
3799 if (bond_is_lb(bond)) {
3800 /* bond_alb_initialize must be called before the timer
3801 * is started.
3802 */
3803 if (bond_alb_initialize(bond, (bond->params.mode == BOND_MODE_ALB))) {
3804 /* something went wrong - fail the open operation */
3805 return -ENOMEM;
3806 }
3807
3808 INIT_DELAYED_WORK(&bond->alb_work, bond_alb_monitor);
3809 queue_delayed_work(bond->wq, &bond->alb_work, 0);
3810 }
3811
3812 if (bond->params.miimon) { /* link check interval, in milliseconds. */
3813 INIT_DELAYED_WORK(&bond->mii_work, bond_mii_monitor);
3814 queue_delayed_work(bond->wq, &bond->mii_work, 0);
3815 }
3816
3817 if (bond->params.arp_interval) { /* arp interval, in milliseconds. */
3818 if (bond->params.mode == BOND_MODE_ACTIVEBACKUP)
3819 INIT_DELAYED_WORK(&bond->arp_work,
3820 bond_activebackup_arp_mon);
3821 else
3822 INIT_DELAYED_WORK(&bond->arp_work,
3823 bond_loadbalance_arp_mon);
3824
3825 queue_delayed_work(bond->wq, &bond->arp_work, 0);
3826 if (bond->params.arp_validate)
3827 bond_register_arp(bond);
3828 }
3829
3830 if (bond->params.mode == BOND_MODE_8023AD) {
3831 INIT_DELAYED_WORK(&bond->ad_work, bond_3ad_state_machine_handler);
3832 queue_delayed_work(bond->wq, &bond->ad_work, 0);
3833 /* register to receive LACPDUs */
3834 bond_register_lacpdu(bond);
3835 bond_3ad_initiate_agg_selection(bond, 1);
3836 }
3837
3838 return 0;
3839}
3840
3841static int bond_close(struct net_device *bond_dev)
3842{
3843 struct bonding *bond = netdev_priv(bond_dev);
3844
3845 if (bond->params.mode == BOND_MODE_8023AD) {
3846 /* Unregister the receive of LACPDUs */
3847 bond_unregister_lacpdu(bond);
3848 }
3849
3850 if (bond->params.arp_validate)
3851 bond_unregister_arp(bond);
3852
3853 write_lock_bh(&bond->lock);
3854
3855 bond->send_grat_arp = 0;
3856 bond->send_unsol_na = 0;
3857
3858 /* signal timers not to re-arm */
3859 bond->kill_timers = 1;
3860
3861 write_unlock_bh(&bond->lock);
3862
3863 if (bond->params.miimon) { /* link check interval, in milliseconds. */
3864 cancel_delayed_work(&bond->mii_work);
3865 }
3866
3867 if (bond->params.arp_interval) { /* arp interval, in milliseconds. */
3868 cancel_delayed_work(&bond->arp_work);
3869 }
3870
3871 switch (bond->params.mode) {
3872 case BOND_MODE_8023AD:
3873 cancel_delayed_work(&bond->ad_work);
3874 break;
3875 case BOND_MODE_TLB:
3876 case BOND_MODE_ALB:
3877 cancel_delayed_work(&bond->alb_work);
3878 break;
3879 default:
3880 break;
3881 }
3882
3883 if (delayed_work_pending(&bond->mcast_work))
3884 cancel_delayed_work(&bond->mcast_work);
3885
3886 if (bond_is_lb(bond)) {
3887 /* Must be called only after all
3888 * slaves have been released
3889 */
3890 bond_alb_deinitialize(bond);
3891 }
3892
3893 return 0;
3894}
3895
3896static struct rtnl_link_stats64 *bond_get_stats(struct net_device *bond_dev,
3897 struct rtnl_link_stats64 *stats)
3898{
3899 struct bonding *bond = netdev_priv(bond_dev);
3900 struct rtnl_link_stats64 temp;
3901 struct slave *slave;
3902 int i;
3903
3904 memset(stats, 0, sizeof(*stats));
3905
3906 read_lock_bh(&bond->lock);
3907
3908 bond_for_each_slave(bond, slave, i) {
3909 const struct rtnl_link_stats64 *sstats =
3910 dev_get_stats(slave->dev, &temp);
3911
3912 stats->rx_packets += sstats->rx_packets;
3913 stats->rx_bytes += sstats->rx_bytes;
3914 stats->rx_errors += sstats->rx_errors;
3915 stats->rx_dropped += sstats->rx_dropped;
3916
3917 stats->tx_packets += sstats->tx_packets;
3918 stats->tx_bytes += sstats->tx_bytes;
3919 stats->tx_errors += sstats->tx_errors;
3920 stats->tx_dropped += sstats->tx_dropped;
3921
3922 stats->multicast += sstats->multicast;
3923 stats->collisions += sstats->collisions;
3924
3925 stats->rx_length_errors += sstats->rx_length_errors;
3926 stats->rx_over_errors += sstats->rx_over_errors;
3927 stats->rx_crc_errors += sstats->rx_crc_errors;
3928 stats->rx_frame_errors += sstats->rx_frame_errors;
3929 stats->rx_fifo_errors += sstats->rx_fifo_errors;
3930 stats->rx_missed_errors += sstats->rx_missed_errors;
3931
3932 stats->tx_aborted_errors += sstats->tx_aborted_errors;
3933 stats->tx_carrier_errors += sstats->tx_carrier_errors;
3934 stats->tx_fifo_errors += sstats->tx_fifo_errors;
3935 stats->tx_heartbeat_errors += sstats->tx_heartbeat_errors;
3936 stats->tx_window_errors += sstats->tx_window_errors;
3937 }
3938
3939 read_unlock_bh(&bond->lock);
3940
3941 return stats;
3942}
3943
3944static int bond_do_ioctl(struct net_device *bond_dev, struct ifreq *ifr, int cmd)
3945{
3946 struct net_device *slave_dev = NULL;
3947 struct ifbond k_binfo;
3948 struct ifbond __user *u_binfo = NULL;
3949 struct ifslave k_sinfo;
3950 struct ifslave __user *u_sinfo = NULL;
3951 struct mii_ioctl_data *mii = NULL;
3952 int res = 0;
3953
3954 pr_debug("bond_ioctl: master=%s, cmd=%d\n", bond_dev->name, cmd);
3955
3956 switch (cmd) {
3957 case SIOCGMIIPHY:
3958 mii = if_mii(ifr);
3959 if (!mii)
3960 return -EINVAL;
3961
3962 mii->phy_id = 0;
3963 /* Fall Through */
3964 case SIOCGMIIREG:
3965 /*
3966 * We do this again just in case we were called by SIOCGMIIREG
3967 * instead of SIOCGMIIPHY.
3968 */
3969 mii = if_mii(ifr);
3970 if (!mii)
3971 return -EINVAL;
3972
3973
3974 if (mii->reg_num == 1) {
3975 struct bonding *bond = netdev_priv(bond_dev);
3976 mii->val_out = 0;
3977 read_lock(&bond->lock);
3978 read_lock(&bond->curr_slave_lock);
3979 if (netif_carrier_ok(bond->dev))
3980 mii->val_out = BMSR_LSTATUS;
3981
3982 read_unlock(&bond->curr_slave_lock);
3983 read_unlock(&bond->lock);
3984 }
3985
3986 return 0;
3987 case BOND_INFO_QUERY_OLD:
3988 case SIOCBONDINFOQUERY:
3989 u_binfo = (struct ifbond __user *)ifr->ifr_data;
3990
3991 if (copy_from_user(&k_binfo, u_binfo, sizeof(ifbond)))
3992 return -EFAULT;
3993
3994 res = bond_info_query(bond_dev, &k_binfo);
3995 if (res == 0 &&
3996 copy_to_user(u_binfo, &k_binfo, sizeof(ifbond)))
3997 return -EFAULT;
3998
3999 return res;
4000 case BOND_SLAVE_INFO_QUERY_OLD:
4001 case SIOCBONDSLAVEINFOQUERY:
4002 u_sinfo = (struct ifslave __user *)ifr->ifr_data;
4003
4004 if (copy_from_user(&k_sinfo, u_sinfo, sizeof(ifslave)))
4005 return -EFAULT;
4006
4007 res = bond_slave_info_query(bond_dev, &k_sinfo);
4008 if (res == 0 &&
4009 copy_to_user(u_sinfo, &k_sinfo, sizeof(ifslave)))
4010 return -EFAULT;
4011
4012 return res;
4013 default:
4014 /* Go on */
4015 break;
4016 }
4017
4018 if (!capable(CAP_NET_ADMIN))
4019 return -EPERM;
4020
4021 slave_dev = dev_get_by_name(dev_net(bond_dev), ifr->ifr_slave);
4022
4023 pr_debug("slave_dev=%p:\n", slave_dev);
4024
4025 if (!slave_dev)
4026 res = -ENODEV;
4027 else {
4028 pr_debug("slave_dev->name=%s:\n", slave_dev->name);
4029 switch (cmd) {
4030 case BOND_ENSLAVE_OLD:
4031 case SIOCBONDENSLAVE:
4032 res = bond_enslave(bond_dev, slave_dev);
4033 break;
4034 case BOND_RELEASE_OLD:
4035 case SIOCBONDRELEASE:
4036 res = bond_release(bond_dev, slave_dev);
4037 break;
4038 case BOND_SETHWADDR_OLD:
4039 case SIOCBONDSETHWADDR:
4040 res = bond_sethwaddr(bond_dev, slave_dev);
4041 break;
4042 case BOND_CHANGE_ACTIVE_OLD:
4043 case SIOCBONDCHANGEACTIVE:
4044 res = bond_ioctl_change_active(bond_dev, slave_dev);
4045 break;
4046 default:
4047 res = -EOPNOTSUPP;
4048 }
4049
4050 dev_put(slave_dev);
4051 }
4052
4053 return res;
4054}
4055
4056static bool bond_addr_in_mc_list(unsigned char *addr,
4057 struct netdev_hw_addr_list *list,
4058 int addrlen)
4059{
4060 struct netdev_hw_addr *ha;
4061
4062 netdev_hw_addr_list_for_each(ha, list)
4063 if (!memcmp(ha->addr, addr, addrlen))
4064 return true;
4065
4066 return false;
4067}
4068
4069static void bond_set_multicast_list(struct net_device *bond_dev)
4070{
4071 struct bonding *bond = netdev_priv(bond_dev);
4072 struct netdev_hw_addr *ha;
4073 bool found;
4074
4075 /*
4076 * Do promisc before checking multicast_mode
4077 */
4078 if ((bond_dev->flags & IFF_PROMISC) && !(bond->flags & IFF_PROMISC))
4079 /*
4080 * FIXME: Need to handle the error when one of the multi-slaves
4081 * encounters error.
4082 */
4083 bond_set_promiscuity(bond, 1);
4084
4085
4086 if (!(bond_dev->flags & IFF_PROMISC) && (bond->flags & IFF_PROMISC))
4087 bond_set_promiscuity(bond, -1);
4088
4089
4090 /* set allmulti flag to slaves */
4091 if ((bond_dev->flags & IFF_ALLMULTI) && !(bond->flags & IFF_ALLMULTI))
4092 /*
4093 * FIXME: Need to handle the error when one of the multi-slaves
4094 * encounters error.
4095 */
4096 bond_set_allmulti(bond, 1);
4097
4098
4099 if (!(bond_dev->flags & IFF_ALLMULTI) && (bond->flags & IFF_ALLMULTI))
4100 bond_set_allmulti(bond, -1);
4101
4102
4103 read_lock(&bond->lock);
4104
4105 bond->flags = bond_dev->flags;
4106
4107 /* looking for addresses to add to slaves' mc list */
4108 netdev_for_each_mc_addr(ha, bond_dev) {
4109 found = bond_addr_in_mc_list(ha->addr, &bond->mc_list,
4110 bond_dev->addr_len);
4111 if (!found)
4112 bond_mc_add(bond, ha->addr);
4113 }
4114
4115 /* looking for addresses to delete from slaves' list */
4116 netdev_hw_addr_list_for_each(ha, &bond->mc_list) {
4117 found = bond_addr_in_mc_list(ha->addr, &bond_dev->mc,
4118 bond_dev->addr_len);
4119 if (!found)
4120 bond_mc_del(bond, ha->addr);
4121 }
4122
4123 /* save master's multicast list */
4124 __hw_addr_flush(&bond->mc_list);
4125 __hw_addr_add_multiple(&bond->mc_list, &bond_dev->mc,
4126 bond_dev->addr_len, NETDEV_HW_ADDR_T_MULTICAST);
4127
4128 read_unlock(&bond->lock);
4129}
4130
4131static int bond_neigh_setup(struct net_device *dev, struct neigh_parms *parms)
4132{
4133 struct bonding *bond = netdev_priv(dev);
4134 struct slave *slave = bond->first_slave;
4135
4136 if (slave) {
4137 const struct net_device_ops *slave_ops
4138 = slave->dev->netdev_ops;
4139 if (slave_ops->ndo_neigh_setup)
4140 return slave_ops->ndo_neigh_setup(slave->dev, parms);
4141 }
4142 return 0;
4143}
4144
4145/*
4146 * Change the MTU of all of a master's slaves to match the master
4147 */
4148static int bond_change_mtu(struct net_device *bond_dev, int new_mtu)
4149{
4150 struct bonding *bond = netdev_priv(bond_dev);
4151 struct slave *slave, *stop_at;
4152 int res = 0;
4153 int i;
4154
4155 pr_debug("bond=%p, name=%s, new_mtu=%d\n", bond,
4156 (bond_dev ? bond_dev->name : "None"), new_mtu);
4157
4158 /* Can't hold bond->lock with bh disabled here since
4159 * some base drivers panic. On the other hand we can't
4160 * hold bond->lock without bh disabled because we'll
4161 * deadlock. The only solution is to rely on the fact
4162 * that we're under rtnl_lock here, and the slaves
4163 * list won't change. This doesn't solve the problem
4164 * of setting the slave's MTU while it is
4165 * transmitting, but the assumption is that the base
4166 * driver can handle that.
4167 *
4168 * TODO: figure out a way to safely iterate the slaves
4169 * list, but without holding a lock around the actual
4170 * call to the base driver.
4171 */
4172
4173 bond_for_each_slave(bond, slave, i) {
4174 pr_debug("s %p s->p %p c_m %p\n",
4175 slave,
4176 slave->prev,
4177 slave->dev->netdev_ops->ndo_change_mtu);
4178
4179 res = dev_set_mtu(slave->dev, new_mtu);
4180
4181 if (res) {
4182 /* If we failed to set the slave's mtu to the new value
4183 * we must abort the operation even in ACTIVE_BACKUP
4184 * mode, because if we allow the backup slaves to have
4185 * different mtu values than the active slave we'll
4186 * need to change their mtu when doing a failover. That
4187 * means changing their mtu from timer context, which
4188 * is probably not a good idea.
4189 */
4190 pr_debug("err %d %s\n", res, slave->dev->name);
4191 goto unwind;
4192 }
4193 }
4194
4195 bond_dev->mtu = new_mtu;
4196
4197 return 0;
4198
4199unwind:
4200 /* unwind from head to the slave that failed */
4201 stop_at = slave;
4202 bond_for_each_slave_from_to(bond, slave, i, bond->first_slave, stop_at) {
4203 int tmp_res;
4204
4205 tmp_res = dev_set_mtu(slave->dev, bond_dev->mtu);
4206 if (tmp_res) {
4207 pr_debug("unwind err %d dev %s\n",
4208 tmp_res, slave->dev->name);
4209 }
4210 }
4211
4212 return res;
4213}
4214
4215/*
4216 * Change HW address
4217 *
4218 * Note that many devices must be down to change the HW address, and
4219 * downing the master releases all slaves. We can make bonds full of
4220 * bonding devices to test this, however.
4221 */
4222static int bond_set_mac_address(struct net_device *bond_dev, void *addr)
4223{
4224 struct bonding *bond = netdev_priv(bond_dev);
4225 struct sockaddr *sa = addr, tmp_sa;
4226 struct slave *slave, *stop_at;
4227 int res = 0;
4228 int i;
4229
4230 if (bond->params.mode == BOND_MODE_ALB)
4231 return bond_alb_set_mac_address(bond_dev, addr);
4232
4233
4234 pr_debug("bond=%p, name=%s\n",
4235 bond, bond_dev ? bond_dev->name : "None");
4236
4237 /*
4238 * If fail_over_mac is set to active, do nothing and return
4239 * success. Returning an error causes ifenslave to fail.
4240 */
4241 if (bond->params.fail_over_mac == BOND_FOM_ACTIVE)
4242 return 0;
4243
4244 if (!is_valid_ether_addr(sa->sa_data))
4245 return -EADDRNOTAVAIL;
4246
4247 /* Can't hold bond->lock with bh disabled here since
4248 * some base drivers panic. On the other hand we can't
4249 * hold bond->lock without bh disabled because we'll
4250 * deadlock. The only solution is to rely on the fact
4251 * that we're under rtnl_lock here, and the slaves
4252 * list won't change. This doesn't solve the problem
4253 * of setting the slave's hw address while it is
4254 * transmitting, but the assumption is that the base
4255 * driver can handle that.
4256 *
4257 * TODO: figure out a way to safely iterate the slaves
4258 * list, but without holding a lock around the actual
4259 * call to the base driver.
4260 */
4261
4262 bond_for_each_slave(bond, slave, i) {
4263 const struct net_device_ops *slave_ops = slave->dev->netdev_ops;
4264 pr_debug("slave %p %s\n", slave, slave->dev->name);
4265
4266 if (slave_ops->ndo_set_mac_address == NULL) {
4267 res = -EOPNOTSUPP;
4268 pr_debug("EOPNOTSUPP %s\n", slave->dev->name);
4269 goto unwind;
4270 }
4271
4272 res = dev_set_mac_address(slave->dev, addr);
4273 if (res) {
4274 /* TODO: consider downing the slave
4275 * and retry ?
4276 * User should expect communications
4277 * breakage anyway until ARP finish
4278 * updating, so...
4279 */
4280 pr_debug("err %d %s\n", res, slave->dev->name);
4281 goto unwind;
4282 }
4283 }
4284
4285 /* success */
4286 memcpy(bond_dev->dev_addr, sa->sa_data, bond_dev->addr_len);
4287 return 0;
4288
4289unwind:
4290 memcpy(tmp_sa.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
4291 tmp_sa.sa_family = bond_dev->type;
4292
4293 /* unwind from head to the slave that failed */
4294 stop_at = slave;
4295 bond_for_each_slave_from_to(bond, slave, i, bond->first_slave, stop_at) {
4296 int tmp_res;
4297
4298 tmp_res = dev_set_mac_address(slave->dev, &tmp_sa);
4299 if (tmp_res) {
4300 pr_debug("unwind err %d dev %s\n",
4301 tmp_res, slave->dev->name);
4302 }
4303 }
4304
4305 return res;
4306}
4307
4308static int bond_xmit_roundrobin(struct sk_buff *skb, struct net_device *bond_dev)
4309{
4310 struct bonding *bond = netdev_priv(bond_dev);
4311 struct slave *slave, *start_at;
4312 int i, slave_no, res = 1;
4313 struct iphdr *iph = ip_hdr(skb);
4314
4315 read_lock(&bond->lock);
4316
4317 if (!BOND_IS_OK(bond))
4318 goto out;
4319 /*
4320 * Start with the curr_active_slave that joined the bond as the
4321 * default for sending IGMP traffic. For failover purposes one
4322 * needs to maintain some consistency for the interface that will
4323 * send the join/membership reports. The curr_active_slave found
4324 * will send all of this type of traffic.
4325 */
4326 if ((iph->protocol == IPPROTO_IGMP) &&
4327 (skb->protocol == htons(ETH_P_IP))) {
4328
4329 read_lock(&bond->curr_slave_lock);
4330 slave = bond->curr_active_slave;
4331 read_unlock(&bond->curr_slave_lock);
4332
4333 if (!slave)
4334 goto out;
4335 } else {
4336 /*
4337 * Concurrent TX may collide on rr_tx_counter; we accept
4338 * that as being rare enough not to justify using an
4339 * atomic op here.
4340 */
4341 slave_no = bond->rr_tx_counter++ % bond->slave_cnt;
4342
4343 bond_for_each_slave(bond, slave, i) {
4344 slave_no--;
4345 if (slave_no < 0)
4346 break;
4347 }
4348 }
4349
4350 start_at = slave;
4351 bond_for_each_slave_from(bond, slave, i, start_at) {
4352 if (IS_UP(slave->dev) &&
4353 (slave->link == BOND_LINK_UP) &&
4354 (slave->state == BOND_STATE_ACTIVE)) {
4355 res = bond_dev_queue_xmit(bond, skb, slave->dev);
4356 break;
4357 }
4358 }
4359
4360out:
4361 if (res) {
4362 /* no suitable interface, frame not sent */
4363 dev_kfree_skb(skb);
4364 }
4365 read_unlock(&bond->lock);
4366 return NETDEV_TX_OK;
4367}
4368
4369
4370/*
4371 * in active-backup mode, we know that bond->curr_active_slave is always valid if
4372 * the bond has a usable interface.
4373 */
4374static int bond_xmit_activebackup(struct sk_buff *skb, struct net_device *bond_dev)
4375{
4376 struct bonding *bond = netdev_priv(bond_dev);
4377 int res = 1;
4378
4379 read_lock(&bond->lock);
4380 read_lock(&bond->curr_slave_lock);
4381
4382 if (!BOND_IS_OK(bond))
4383 goto out;
4384
4385 if (!bond->curr_active_slave)
4386 goto out;
4387
4388 res = bond_dev_queue_xmit(bond, skb, bond->curr_active_slave->dev);
4389
4390out:
4391 if (res)
4392 /* no suitable interface, frame not sent */
4393 dev_kfree_skb(skb);
4394
4395 read_unlock(&bond->curr_slave_lock);
4396 read_unlock(&bond->lock);
4397 return NETDEV_TX_OK;
4398}
4399
4400/*
4401 * In bond_xmit_xor() , we determine the output device by using a pre-
4402 * determined xmit_hash_policy(), If the selected device is not enabled,
4403 * find the next active slave.
4404 */
4405static int bond_xmit_xor(struct sk_buff *skb, struct net_device *bond_dev)
4406{
4407 struct bonding *bond = netdev_priv(bond_dev);
4408 struct slave *slave, *start_at;
4409 int slave_no;
4410 int i;
4411 int res = 1;
4412
4413 read_lock(&bond->lock);
4414
4415 if (!BOND_IS_OK(bond))
4416 goto out;
4417
4418 slave_no = bond->xmit_hash_policy(skb, bond->slave_cnt);
4419
4420 bond_for_each_slave(bond, slave, i) {
4421 slave_no--;
4422 if (slave_no < 0)
4423 break;
4424 }
4425
4426 start_at = slave;
4427
4428 bond_for_each_slave_from(bond, slave, i, start_at) {
4429 if (IS_UP(slave->dev) &&
4430 (slave->link == BOND_LINK_UP) &&
4431 (slave->state == BOND_STATE_ACTIVE)) {
4432 res = bond_dev_queue_xmit(bond, skb, slave->dev);
4433 break;
4434 }
4435 }
4436
4437out:
4438 if (res) {
4439 /* no suitable interface, frame not sent */
4440 dev_kfree_skb(skb);
4441 }
4442 read_unlock(&bond->lock);
4443 return NETDEV_TX_OK;
4444}
4445
4446/*
4447 * in broadcast mode, we send everything to all usable interfaces.
4448 */
4449static int bond_xmit_broadcast(struct sk_buff *skb, struct net_device *bond_dev)
4450{
4451 struct bonding *bond = netdev_priv(bond_dev);
4452 struct slave *slave, *start_at;
4453 struct net_device *tx_dev = NULL;
4454 int i;
4455 int res = 1;
4456
4457 read_lock(&bond->lock);
4458
4459 if (!BOND_IS_OK(bond))
4460 goto out;
4461
4462 read_lock(&bond->curr_slave_lock);
4463 start_at = bond->curr_active_slave;
4464 read_unlock(&bond->curr_slave_lock);
4465
4466 if (!start_at)
4467 goto out;
4468
4469 bond_for_each_slave_from(bond, slave, i, start_at) {
4470 if (IS_UP(slave->dev) &&
4471 (slave->link == BOND_LINK_UP) &&
4472 (slave->state == BOND_STATE_ACTIVE)) {
4473 if (tx_dev) {
4474 struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
4475 if (!skb2) {
4476 pr_err("%s: Error: bond_xmit_broadcast(): skb_clone() failed\n",
4477 bond_dev->name);
4478 continue;
4479 }
4480
4481 res = bond_dev_queue_xmit(bond, skb2, tx_dev);
4482 if (res) {
4483 dev_kfree_skb(skb2);
4484 continue;
4485 }
4486 }
4487 tx_dev = slave->dev;
4488 }
4489 }
4490
4491 if (tx_dev)
4492 res = bond_dev_queue_xmit(bond, skb, tx_dev);
4493
4494out:
4495 if (res)
4496 /* no suitable interface, frame not sent */
4497 dev_kfree_skb(skb);
4498
4499 /* frame sent to all suitable interfaces */
4500 read_unlock(&bond->lock);
4501 return NETDEV_TX_OK;
4502}
4503
4504/*------------------------- Device initialization ---------------------------*/
4505
4506static void bond_set_xmit_hash_policy(struct bonding *bond)
4507{
4508 switch (bond->params.xmit_policy) {
4509 case BOND_XMIT_POLICY_LAYER23:
4510 bond->xmit_hash_policy = bond_xmit_hash_policy_l23;
4511 break;
4512 case BOND_XMIT_POLICY_LAYER34:
4513 bond->xmit_hash_policy = bond_xmit_hash_policy_l34;
4514 break;
4515 case BOND_XMIT_POLICY_LAYER2:
4516 default:
4517 bond->xmit_hash_policy = bond_xmit_hash_policy_l2;
4518 break;
4519 }
4520}
4521
4522/*
4523 * Lookup the slave that corresponds to a qid
4524 */
4525static inline int bond_slave_override(struct bonding *bond,
4526 struct sk_buff *skb)
4527{
4528 int i, res = 1;
4529 struct slave *slave = NULL;
4530 struct slave *check_slave;
4531
4532 read_lock(&bond->lock);
4533
4534 if (!BOND_IS_OK(bond) || !skb->queue_mapping)
4535 goto out;
4536
4537 /* Find out if any slaves have the same mapping as this skb. */
4538 bond_for_each_slave(bond, check_slave, i) {
4539 if (check_slave->queue_id == skb->queue_mapping) {
4540 slave = check_slave;
4541 break;
4542 }
4543 }
4544
4545 /* If the slave isn't UP, use default transmit policy. */
4546 if (slave && slave->queue_id && IS_UP(slave->dev) &&
4547 (slave->link == BOND_LINK_UP)) {
4548 res = bond_dev_queue_xmit(bond, skb, slave->dev);
4549 }
4550
4551out:
4552 read_unlock(&bond->lock);
4553 return res;
4554}
4555
4556static u16 bond_select_queue(struct net_device *dev, struct sk_buff *skb)
4557{
4558 /*
4559 * This helper function exists to help dev_pick_tx get the correct
4560 * destination queue. Using a helper function skips the a call to
4561 * skb_tx_hash and will put the skbs in the queue we expect on their
4562 * way down to the bonding driver.
4563 */
4564 return skb->queue_mapping;
4565}
4566
4567static netdev_tx_t bond_start_xmit(struct sk_buff *skb, struct net_device *dev)
4568{
4569 struct bonding *bond = netdev_priv(dev);
4570
4571 /*
4572 * If we risk deadlock from transmitting this in the
4573 * netpoll path, tell netpoll to queue the frame for later tx
4574 */
4575 if (is_netpoll_tx_blocked(dev))
4576 return NETDEV_TX_BUSY;
4577
4578 if (TX_QUEUE_OVERRIDE(bond->params.mode)) {
4579 if (!bond_slave_override(bond, skb))
4580 return NETDEV_TX_OK;
4581 }
4582
4583 switch (bond->params.mode) {
4584 case BOND_MODE_ROUNDROBIN:
4585 return bond_xmit_roundrobin(skb, dev);
4586 case BOND_MODE_ACTIVEBACKUP:
4587 return bond_xmit_activebackup(skb, dev);
4588 case BOND_MODE_XOR:
4589 return bond_xmit_xor(skb, dev);
4590 case BOND_MODE_BROADCAST:
4591 return bond_xmit_broadcast(skb, dev);
4592 case BOND_MODE_8023AD:
4593 return bond_3ad_xmit_xor(skb, dev);
4594 case BOND_MODE_ALB:
4595 case BOND_MODE_TLB:
4596 return bond_alb_xmit(skb, dev);
4597 default:
4598 /* Should never happen, mode already checked */
4599 pr_err("%s: Error: Unknown bonding mode %d\n",
4600 dev->name, bond->params.mode);
4601 WARN_ON_ONCE(1);
4602 dev_kfree_skb(skb);
4603 return NETDEV_TX_OK;
4604 }
4605}
4606
4607
4608/*
4609 * set bond mode specific net device operations
4610 */
4611void bond_set_mode_ops(struct bonding *bond, int mode)
4612{
4613 struct net_device *bond_dev = bond->dev;
4614
4615 switch (mode) {
4616 case BOND_MODE_ROUNDROBIN:
4617 break;
4618 case BOND_MODE_ACTIVEBACKUP:
4619 break;
4620 case BOND_MODE_XOR:
4621 bond_set_xmit_hash_policy(bond);
4622 break;
4623 case BOND_MODE_BROADCAST:
4624 break;
4625 case BOND_MODE_8023AD:
4626 bond_set_master_3ad_flags(bond);
4627 bond_set_xmit_hash_policy(bond);
4628 break;
4629 case BOND_MODE_ALB:
4630 bond_set_master_alb_flags(bond);
4631 /* FALLTHRU */
4632 case BOND_MODE_TLB:
4633 break;
4634 default:
4635 /* Should never happen, mode already checked */
4636 pr_err("%s: Error: Unknown bonding mode %d\n",
4637 bond_dev->name, mode);
4638 break;
4639 }
4640}
4641
4642static void bond_ethtool_get_drvinfo(struct net_device *bond_dev,
4643 struct ethtool_drvinfo *drvinfo)
4644{
4645 strncpy(drvinfo->driver, DRV_NAME, 32);
4646 strncpy(drvinfo->version, DRV_VERSION, 32);
4647 snprintf(drvinfo->fw_version, 32, "%d", BOND_ABI_VERSION);
4648}
4649
4650static const struct ethtool_ops bond_ethtool_ops = {
4651 .get_drvinfo = bond_ethtool_get_drvinfo,
4652 .get_link = ethtool_op_get_link,
4653 .get_tx_csum = ethtool_op_get_tx_csum,
4654 .get_sg = ethtool_op_get_sg,
4655 .get_tso = ethtool_op_get_tso,
4656 .get_ufo = ethtool_op_get_ufo,
4657 .get_flags = ethtool_op_get_flags,
4658};
4659
4660static const struct net_device_ops bond_netdev_ops = {
4661 .ndo_init = bond_init,
4662 .ndo_uninit = bond_uninit,
4663 .ndo_open = bond_open,
4664 .ndo_stop = bond_close,
4665 .ndo_start_xmit = bond_start_xmit,
4666 .ndo_select_queue = bond_select_queue,
4667 .ndo_get_stats64 = bond_get_stats,
4668 .ndo_do_ioctl = bond_do_ioctl,
4669 .ndo_set_multicast_list = bond_set_multicast_list,
4670 .ndo_change_mtu = bond_change_mtu,
4671 .ndo_set_mac_address = bond_set_mac_address,
4672 .ndo_neigh_setup = bond_neigh_setup,
4673 .ndo_vlan_rx_register = bond_vlan_rx_register,
4674 .ndo_vlan_rx_add_vid = bond_vlan_rx_add_vid,
4675 .ndo_vlan_rx_kill_vid = bond_vlan_rx_kill_vid,
4676#ifdef CONFIG_NET_POLL_CONTROLLER
4677 .ndo_netpoll_cleanup = bond_netpoll_cleanup,
4678 .ndo_poll_controller = bond_poll_controller,
4679#endif
4680};
4681
4682static void bond_destructor(struct net_device *bond_dev)
4683{
4684 struct bonding *bond = netdev_priv(bond_dev);
4685 if (bond->wq)
4686 destroy_workqueue(bond->wq);
4687 free_netdev(bond_dev);
4688}
4689
4690static void bond_setup(struct net_device *bond_dev)
4691{
4692 struct bonding *bond = netdev_priv(bond_dev);
4693
4694 /* initialize rwlocks */
4695 rwlock_init(&bond->lock);
4696 rwlock_init(&bond->curr_slave_lock);
4697
4698 bond->params = bonding_defaults;
4699
4700 /* Initialize pointers */
4701 bond->dev = bond_dev;
4702 INIT_LIST_HEAD(&bond->vlan_list);
4703
4704 /* Initialize the device entry points */
4705 ether_setup(bond_dev);
4706 bond_dev->netdev_ops = &bond_netdev_ops;
4707 bond_dev->ethtool_ops = &bond_ethtool_ops;
4708 bond_set_mode_ops(bond, bond->params.mode);
4709
4710 bond_dev->destructor = bond_destructor;
4711
4712 /* Initialize the device options */
4713 bond_dev->tx_queue_len = 0;
4714 bond_dev->flags |= IFF_MASTER|IFF_MULTICAST;
4715 bond_dev->priv_flags |= IFF_BONDING;
4716 bond_dev->priv_flags &= ~IFF_XMIT_DST_RELEASE;
4717
4718 if (bond->params.arp_interval)
4719 bond_dev->priv_flags |= IFF_MASTER_ARPMON;
4720
4721 /* At first, we block adding VLANs. That's the only way to
4722 * prevent problems that occur when adding VLANs over an
4723 * empty bond. The block will be removed once non-challenged
4724 * slaves are enslaved.
4725 */
4726 bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
4727
4728 /* don't acquire bond device's netif_tx_lock when
4729 * transmitting */
4730 bond_dev->features |= NETIF_F_LLTX;
4731
4732 /* By default, we declare the bond to be fully
4733 * VLAN hardware accelerated capable. Special
4734 * care is taken in the various xmit functions
4735 * when there are slaves that are not hw accel
4736 * capable
4737 */
4738 bond_dev->features |= (NETIF_F_HW_VLAN_TX |
4739 NETIF_F_HW_VLAN_RX |
4740 NETIF_F_HW_VLAN_FILTER);
4741
4742 /* By default, we enable GRO on bonding devices.
4743 * Actual support requires lowlevel drivers are GRO ready.
4744 */
4745 bond_dev->features |= NETIF_F_GRO;
4746}
4747
4748static void bond_work_cancel_all(struct bonding *bond)
4749{
4750 write_lock_bh(&bond->lock);
4751 bond->kill_timers = 1;
4752 write_unlock_bh(&bond->lock);
4753
4754 if (bond->params.miimon && delayed_work_pending(&bond->mii_work))
4755 cancel_delayed_work(&bond->mii_work);
4756
4757 if (bond->params.arp_interval && delayed_work_pending(&bond->arp_work))
4758 cancel_delayed_work(&bond->arp_work);
4759
4760 if (bond->params.mode == BOND_MODE_ALB &&
4761 delayed_work_pending(&bond->alb_work))
4762 cancel_delayed_work(&bond->alb_work);
4763
4764 if (bond->params.mode == BOND_MODE_8023AD &&
4765 delayed_work_pending(&bond->ad_work))
4766 cancel_delayed_work(&bond->ad_work);
4767
4768 if (delayed_work_pending(&bond->mcast_work))
4769 cancel_delayed_work(&bond->mcast_work);
4770}
4771
4772/*
4773* Destroy a bonding device.
4774* Must be under rtnl_lock when this function is called.
4775*/
4776static void bond_uninit(struct net_device *bond_dev)
4777{
4778 struct bonding *bond = netdev_priv(bond_dev);
4779 struct vlan_entry *vlan, *tmp;
4780
4781 bond_netpoll_cleanup(bond_dev);
4782
4783 /* Release the bonded slaves */
4784 bond_release_all(bond_dev);
4785
4786 list_del(&bond->bond_list);
4787
4788 bond_work_cancel_all(bond);
4789
4790 bond_remove_proc_entry(bond);
4791
4792 __hw_addr_flush(&bond->mc_list);
4793
4794 list_for_each_entry_safe(vlan, tmp, &bond->vlan_list, vlan_list) {
4795 list_del(&vlan->vlan_list);
4796 kfree(vlan);
4797 }
4798}
4799
4800/*------------------------- Module initialization ---------------------------*/
4801
4802/*
4803 * Convert string input module parms. Accept either the
4804 * number of the mode or its string name. A bit complicated because
4805 * some mode names are substrings of other names, and calls from sysfs
4806 * may have whitespace in the name (trailing newlines, for example).
4807 */
4808int bond_parse_parm(const char *buf, const struct bond_parm_tbl *tbl)
4809{
4810 int modeint = -1, i, rv;
4811 char *p, modestr[BOND_MAX_MODENAME_LEN + 1] = { 0, };
4812
4813 for (p = (char *)buf; *p; p++)
4814 if (!(isdigit(*p) || isspace(*p)))
4815 break;
4816
4817 if (*p)
4818 rv = sscanf(buf, "%20s", modestr);
4819 else
4820 rv = sscanf(buf, "%d", &modeint);
4821
4822 if (!rv)
4823 return -1;
4824
4825 for (i = 0; tbl[i].modename; i++) {
4826 if (modeint == tbl[i].mode)
4827 return tbl[i].mode;
4828 if (strcmp(modestr, tbl[i].modename) == 0)
4829 return tbl[i].mode;
4830 }
4831
4832 return -1;
4833}
4834
4835static int bond_check_params(struct bond_params *params)
4836{
4837 int arp_validate_value, fail_over_mac_value, primary_reselect_value;
4838
4839 /*
4840 * Convert string parameters.
4841 */
4842 if (mode) {
4843 bond_mode = bond_parse_parm(mode, bond_mode_tbl);
4844 if (bond_mode == -1) {
4845 pr_err("Error: Invalid bonding mode \"%s\"\n",
4846 mode == NULL ? "NULL" : mode);
4847 return -EINVAL;
4848 }
4849 }
4850
4851 if (xmit_hash_policy) {
4852 if ((bond_mode != BOND_MODE_XOR) &&
4853 (bond_mode != BOND_MODE_8023AD)) {
4854 pr_info("xmit_hash_policy param is irrelevant in mode %s\n",
4855 bond_mode_name(bond_mode));
4856 } else {
4857 xmit_hashtype = bond_parse_parm(xmit_hash_policy,
4858 xmit_hashtype_tbl);
4859 if (xmit_hashtype == -1) {
4860 pr_err("Error: Invalid xmit_hash_policy \"%s\"\n",
4861 xmit_hash_policy == NULL ? "NULL" :
4862 xmit_hash_policy);
4863 return -EINVAL;
4864 }
4865 }
4866 }
4867
4868 if (lacp_rate) {
4869 if (bond_mode != BOND_MODE_8023AD) {
4870 pr_info("lacp_rate param is irrelevant in mode %s\n",
4871 bond_mode_name(bond_mode));
4872 } else {
4873 lacp_fast = bond_parse_parm(lacp_rate, bond_lacp_tbl);
4874 if (lacp_fast == -1) {
4875 pr_err("Error: Invalid lacp rate \"%s\"\n",
4876 lacp_rate == NULL ? "NULL" : lacp_rate);
4877 return -EINVAL;
4878 }
4879 }
4880 }
4881
4882 if (ad_select) {
4883 params->ad_select = bond_parse_parm(ad_select, ad_select_tbl);
4884 if (params->ad_select == -1) {
4885 pr_err("Error: Invalid ad_select \"%s\"\n",
4886 ad_select == NULL ? "NULL" : ad_select);
4887 return -EINVAL;
4888 }
4889
4890 if (bond_mode != BOND_MODE_8023AD) {
4891 pr_warning("ad_select param only affects 802.3ad mode\n");
4892 }
4893 } else {
4894 params->ad_select = BOND_AD_STABLE;
4895 }
4896
4897 if (max_bonds < 0) {
4898 pr_warning("Warning: max_bonds (%d) not in range %d-%d, so it was reset to BOND_DEFAULT_MAX_BONDS (%d)\n",
4899 max_bonds, 0, INT_MAX, BOND_DEFAULT_MAX_BONDS);
4900 max_bonds = BOND_DEFAULT_MAX_BONDS;
4901 }
4902
4903 if (miimon < 0) {
4904 pr_warning("Warning: miimon module parameter (%d), not in range 0-%d, so it was reset to %d\n",
4905 miimon, INT_MAX, BOND_LINK_MON_INTERV);
4906 miimon = BOND_LINK_MON_INTERV;
4907 }
4908
4909 if (updelay < 0) {
4910 pr_warning("Warning: updelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4911 updelay, INT_MAX);
4912 updelay = 0;
4913 }
4914
4915 if (downdelay < 0) {
4916 pr_warning("Warning: downdelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4917 downdelay, INT_MAX);
4918 downdelay = 0;
4919 }
4920
4921 if ((use_carrier != 0) && (use_carrier != 1)) {
4922 pr_warning("Warning: use_carrier module parameter (%d), not of valid value (0/1), so it was set to 1\n",
4923 use_carrier);
4924 use_carrier = 1;
4925 }
4926
4927 if (num_grat_arp < 0 || num_grat_arp > 255) {
4928 pr_warning("Warning: num_grat_arp (%d) not in range 0-255 so it was reset to 1\n",
4929 num_grat_arp);
4930 num_grat_arp = 1;
4931 }
4932
4933 if (num_unsol_na < 0 || num_unsol_na > 255) {
4934 pr_warning("Warning: num_unsol_na (%d) not in range 0-255 so it was reset to 1\n",
4935 num_unsol_na);
4936 num_unsol_na = 1;
4937 }
4938
4939 /* reset values for 802.3ad */
4940 if (bond_mode == BOND_MODE_8023AD) {
4941 if (!miimon) {
4942 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");
4943 pr_warning("Forcing miimon to 100msec\n");
4944 miimon = 100;
4945 }
4946 }
4947
4948 if (tx_queues < 1 || tx_queues > 255) {
4949 pr_warning("Warning: tx_queues (%d) should be between "
4950 "1 and 255, resetting to %d\n",
4951 tx_queues, BOND_DEFAULT_TX_QUEUES);
4952 tx_queues = BOND_DEFAULT_TX_QUEUES;
4953 }
4954
4955 if ((all_slaves_active != 0) && (all_slaves_active != 1)) {
4956 pr_warning("Warning: all_slaves_active module parameter (%d), "
4957 "not of valid value (0/1), so it was set to "
4958 "0\n", all_slaves_active);
4959 all_slaves_active = 0;
4960 }
4961
4962 if (resend_igmp < 0 || resend_igmp > 255) {
4963 pr_warning("Warning: resend_igmp (%d) should be between "
4964 "0 and 255, resetting to %d\n",
4965 resend_igmp, BOND_DEFAULT_RESEND_IGMP);
4966 resend_igmp = BOND_DEFAULT_RESEND_IGMP;
4967 }
4968
4969 /* reset values for TLB/ALB */
4970 if ((bond_mode == BOND_MODE_TLB) ||
4971 (bond_mode == BOND_MODE_ALB)) {
4972 if (!miimon) {
4973 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");
4974 pr_warning("Forcing miimon to 100msec\n");
4975 miimon = 100;
4976 }
4977 }
4978
4979 if (bond_mode == BOND_MODE_ALB) {
4980 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",
4981 updelay);
4982 }
4983
4984 if (!miimon) {
4985 if (updelay || downdelay) {
4986 /* just warn the user the up/down delay will have
4987 * no effect since miimon is zero...
4988 */
4989 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",
4990 updelay, downdelay);
4991 }
4992 } else {
4993 /* don't allow arp monitoring */
4994 if (arp_interval) {
4995 pr_warning("Warning: miimon (%d) and arp_interval (%d) can't be used simultaneously, disabling ARP monitoring\n",
4996 miimon, arp_interval);
4997 arp_interval = 0;
4998 }
4999
5000 if ((updelay % miimon) != 0) {
5001 pr_warning("Warning: updelay (%d) is not a multiple of miimon (%d), updelay rounded to %d ms\n",
5002 updelay, miimon,
5003 (updelay / miimon) * miimon);
5004 }
5005
5006 updelay /= miimon;
5007
5008 if ((downdelay % miimon) != 0) {
5009 pr_warning("Warning: downdelay (%d) is not a multiple of miimon (%d), downdelay rounded to %d ms\n",
5010 downdelay, miimon,
5011 (downdelay / miimon) * miimon);
5012 }
5013
5014 downdelay /= miimon;
5015 }
5016
5017 if (arp_interval < 0) {
5018 pr_warning("Warning: arp_interval module parameter (%d) , not in range 0-%d, so it was reset to %d\n",
5019 arp_interval, INT_MAX, BOND_LINK_ARP_INTERV);
5020 arp_interval = BOND_LINK_ARP_INTERV;
5021 }
5022
5023 for (arp_ip_count = 0;
5024 (arp_ip_count < BOND_MAX_ARP_TARGETS) && arp_ip_target[arp_ip_count];
5025 arp_ip_count++) {
5026 /* not complete check, but should be good enough to
5027 catch mistakes */
5028 if (!isdigit(arp_ip_target[arp_ip_count][0])) {
5029 pr_warning("Warning: bad arp_ip_target module parameter (%s), ARP monitoring will not be performed\n",
5030 arp_ip_target[arp_ip_count]);
5031 arp_interval = 0;
5032 } else {
5033 __be32 ip = in_aton(arp_ip_target[arp_ip_count]);
5034 arp_target[arp_ip_count] = ip;
5035 }
5036 }
5037
5038 if (arp_interval && !arp_ip_count) {
5039 /* don't allow arping if no arp_ip_target given... */
5040 pr_warning("Warning: arp_interval module parameter (%d) specified without providing an arp_ip_target parameter, arp_interval was reset to 0\n",
5041 arp_interval);
5042 arp_interval = 0;
5043 }
5044
5045 if (arp_validate) {
5046 if (bond_mode != BOND_MODE_ACTIVEBACKUP) {
5047 pr_err("arp_validate only supported in active-backup mode\n");
5048 return -EINVAL;
5049 }
5050 if (!arp_interval) {
5051 pr_err("arp_validate requires arp_interval\n");
5052 return -EINVAL;
5053 }
5054
5055 arp_validate_value = bond_parse_parm(arp_validate,
5056 arp_validate_tbl);
5057 if (arp_validate_value == -1) {
5058 pr_err("Error: invalid arp_validate \"%s\"\n",
5059 arp_validate == NULL ? "NULL" : arp_validate);
5060 return -EINVAL;
5061 }
5062 } else
5063 arp_validate_value = 0;
5064
5065 if (miimon) {
5066 pr_info("MII link monitoring set to %d ms\n", miimon);
5067 } else if (arp_interval) {
5068 int i;
5069
5070 pr_info("ARP monitoring set to %d ms, validate %s, with %d target(s):",
5071 arp_interval,
5072 arp_validate_tbl[arp_validate_value].modename,
5073 arp_ip_count);
5074
5075 for (i = 0; i < arp_ip_count; i++)
5076 pr_info(" %s", arp_ip_target[i]);
5077
5078 pr_info("\n");
5079
5080 } else if (max_bonds) {
5081 /* miimon and arp_interval not set, we need one so things
5082 * work as expected, see bonding.txt for details
5083 */
5084 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");
5085 }
5086
5087 if (primary && !USES_PRIMARY(bond_mode)) {
5088 /* currently, using a primary only makes sense
5089 * in active backup, TLB or ALB modes
5090 */
5091 pr_warning("Warning: %s primary device specified but has no effect in %s mode\n",
5092 primary, bond_mode_name(bond_mode));
5093 primary = NULL;
5094 }
5095
5096 if (primary && primary_reselect) {
5097 primary_reselect_value = bond_parse_parm(primary_reselect,
5098 pri_reselect_tbl);
5099 if (primary_reselect_value == -1) {
5100 pr_err("Error: Invalid primary_reselect \"%s\"\n",
5101 primary_reselect ==
5102 NULL ? "NULL" : primary_reselect);
5103 return -EINVAL;
5104 }
5105 } else {
5106 primary_reselect_value = BOND_PRI_RESELECT_ALWAYS;
5107 }
5108
5109 if (fail_over_mac) {
5110 fail_over_mac_value = bond_parse_parm(fail_over_mac,
5111 fail_over_mac_tbl);
5112 if (fail_over_mac_value == -1) {
5113 pr_err("Error: invalid fail_over_mac \"%s\"\n",
5114 arp_validate == NULL ? "NULL" : arp_validate);
5115 return -EINVAL;
5116 }
5117
5118 if (bond_mode != BOND_MODE_ACTIVEBACKUP)
5119 pr_warning("Warning: fail_over_mac only affects active-backup mode.\n");
5120 } else {
5121 fail_over_mac_value = BOND_FOM_NONE;
5122 }
5123
5124 /* fill params struct with the proper values */
5125 params->mode = bond_mode;
5126 params->xmit_policy = xmit_hashtype;
5127 params->miimon = miimon;
5128 params->num_grat_arp = num_grat_arp;
5129 params->num_unsol_na = num_unsol_na;
5130 params->arp_interval = arp_interval;
5131 params->arp_validate = arp_validate_value;
5132 params->updelay = updelay;
5133 params->downdelay = downdelay;
5134 params->use_carrier = use_carrier;
5135 params->lacp_fast = lacp_fast;
5136 params->primary[0] = 0;
5137 params->primary_reselect = primary_reselect_value;
5138 params->fail_over_mac = fail_over_mac_value;
5139 params->tx_queues = tx_queues;
5140 params->all_slaves_active = all_slaves_active;
5141 params->resend_igmp = resend_igmp;
5142
5143 if (primary) {
5144 strncpy(params->primary, primary, IFNAMSIZ);
5145 params->primary[IFNAMSIZ - 1] = 0;
5146 }
5147
5148 memcpy(params->arp_targets, arp_target, sizeof(arp_target));
5149
5150 return 0;
5151}
5152
5153static struct lock_class_key bonding_netdev_xmit_lock_key;
5154static struct lock_class_key bonding_netdev_addr_lock_key;
5155
5156static void bond_set_lockdep_class_one(struct net_device *dev,
5157 struct netdev_queue *txq,
5158 void *_unused)
5159{
5160 lockdep_set_class(&txq->_xmit_lock,
5161 &bonding_netdev_xmit_lock_key);
5162}
5163
5164static void bond_set_lockdep_class(struct net_device *dev)
5165{
5166 lockdep_set_class(&dev->addr_list_lock,
5167 &bonding_netdev_addr_lock_key);
5168 netdev_for_each_tx_queue(dev, bond_set_lockdep_class_one, NULL);
5169}
5170
5171/*
5172 * Called from registration process
5173 */
5174static int bond_init(struct net_device *bond_dev)
5175{
5176 struct bonding *bond = netdev_priv(bond_dev);
5177 struct bond_net *bn = net_generic(dev_net(bond_dev), bond_net_id);
5178
5179 pr_debug("Begin bond_init for %s\n", bond_dev->name);
5180
5181 bond->wq = create_singlethread_workqueue(bond_dev->name);
5182 if (!bond->wq)
5183 return -ENOMEM;
5184
5185 bond_set_lockdep_class(bond_dev);
5186
5187 netif_carrier_off(bond_dev);
5188
5189 bond_create_proc_entry(bond);
5190 list_add_tail(&bond->bond_list, &bn->dev_list);
5191
5192 bond_prepare_sysfs_group(bond);
5193
5194 __hw_addr_init(&bond->mc_list);
5195 return 0;
5196}
5197
5198static int bond_validate(struct nlattr *tb[], struct nlattr *data[])
5199{
5200 if (tb[IFLA_ADDRESS]) {
5201 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN)
5202 return -EINVAL;
5203 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS])))
5204 return -EADDRNOTAVAIL;
5205 }
5206 return 0;
5207}
5208
5209static struct rtnl_link_ops bond_link_ops __read_mostly = {
5210 .kind = "bond",
5211 .priv_size = sizeof(struct bonding),
5212 .setup = bond_setup,
5213 .validate = bond_validate,
5214};
5215
5216/* Create a new bond based on the specified name and bonding parameters.
5217 * If name is NULL, obtain a suitable "bond%d" name for us.
5218 * Caller must NOT hold rtnl_lock; we need to release it here before we
5219 * set up our sysfs entries.
5220 */
5221int bond_create(struct net *net, const char *name)
5222{
5223 struct net_device *bond_dev;
5224 int res;
5225
5226 rtnl_lock();
5227
5228 bond_dev = alloc_netdev_mq(sizeof(struct bonding), name ? name : "",
5229 bond_setup, tx_queues);
5230 if (!bond_dev) {
5231 pr_err("%s: eek! can't alloc netdev!\n", name);
5232 rtnl_unlock();
5233 return -ENOMEM;
5234 }
5235
5236 dev_net_set(bond_dev, net);
5237 bond_dev->rtnl_link_ops = &bond_link_ops;
5238
5239 if (!name) {
5240 res = dev_alloc_name(bond_dev, "bond%d");
5241 if (res < 0)
5242 goto out;
5243 } else {
5244 /*
5245 * If we're given a name to register
5246 * we need to ensure that its not already
5247 * registered
5248 */
5249 res = -EEXIST;
5250 if (__dev_get_by_name(net, name) != NULL)
5251 goto out;
5252 }
5253
5254 res = register_netdevice(bond_dev);
5255
5256out:
5257 rtnl_unlock();
5258 if (res < 0)
5259 bond_destructor(bond_dev);
5260 return res;
5261}
5262
5263static int __net_init bond_net_init(struct net *net)
5264{
5265 struct bond_net *bn = net_generic(net, bond_net_id);
5266
5267 bn->net = net;
5268 INIT_LIST_HEAD(&bn->dev_list);
5269
5270 bond_create_proc_dir(bn);
5271
5272 return 0;
5273}
5274
5275static void __net_exit bond_net_exit(struct net *net)
5276{
5277 struct bond_net *bn = net_generic(net, bond_net_id);
5278
5279 bond_destroy_proc_dir(bn);
5280}
5281
5282static struct pernet_operations bond_net_ops = {
5283 .init = bond_net_init,
5284 .exit = bond_net_exit,
5285 .id = &bond_net_id,
5286 .size = sizeof(struct bond_net),
5287};
5288
5289static int __init bonding_init(void)
5290{
5291 int i;
5292 int res;
5293
5294 pr_info("%s", version);
5295
5296 res = bond_check_params(&bonding_defaults);
5297 if (res)
5298 goto out;
5299
5300#ifdef CONFIG_NET_POLL_CONTROLLER
5301 if (!alloc_cpumask_var(&netpoll_block_tx, GFP_KERNEL)) {
5302 res = -ENOMEM;
5303 goto out;
5304 }
5305#endif
5306
5307 res = register_pernet_subsys(&bond_net_ops);
5308 if (res)
5309 goto out;
5310
5311 res = rtnl_link_register(&bond_link_ops);
5312 if (res)
5313 goto err_link;
5314
5315 for (i = 0; i < max_bonds; i++) {
5316 res = bond_create(&init_net, NULL);
5317 if (res)
5318 goto err;
5319 }
5320
5321 res = bond_create_sysfs();
5322 if (res)
5323 goto err;
5324
5325
5326 register_netdevice_notifier(&bond_netdev_notifier);
5327 register_inetaddr_notifier(&bond_inetaddr_notifier);
5328 bond_register_ipv6_notifier();
5329out:
5330 return res;
5331err:
5332 rtnl_link_unregister(&bond_link_ops);
5333err_link:
5334 unregister_pernet_subsys(&bond_net_ops);
5335#ifdef CONFIG_NET_POLL_CONTROLLER
5336 free_cpumask_var(netpoll_block_tx);
5337#endif
5338 goto out;
5339
5340}
5341
5342static void __exit bonding_exit(void)
5343{
5344 unregister_netdevice_notifier(&bond_netdev_notifier);
5345 unregister_inetaddr_notifier(&bond_inetaddr_notifier);
5346 bond_unregister_ipv6_notifier();
5347
5348 bond_destroy_sysfs();
5349
5350 rtnl_link_unregister(&bond_link_ops);
5351 unregister_pernet_subsys(&bond_net_ops);
5352
5353#ifdef CONFIG_NET_POLL_CONTROLLER
5354 free_cpumask_var(netpoll_block_tx);
5355#endif
5356}
5357
5358module_init(bonding_init);
5359module_exit(bonding_exit);
5360MODULE_LICENSE("GPL");
5361MODULE_VERSION(DRV_VERSION);
5362MODULE_DESCRIPTION(DRV_DESCRIPTION ", v" DRV_VERSION);
5363MODULE_AUTHOR("Thomas Davis, tadavis@lbl.gov and many others");
5364MODULE_ALIAS_RTNL_LINK("bond");
Clone of davem's net-next-2.6 tree