]> 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
bonding: IGMP handling cleanup
[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_bh(&bond->lock);
497 bond->vlgrp = grp;
498 write_unlock_bh(&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
877 rcu_read_lock();
878 in_dev = __in_dev_get_rcu(dev);
879 if (in_dev)
880 ip_mc_rejoin_groups(in_dev);
881 rcu_read_unlock();
882}
883
884/*
885 * Retrieve the list of registered multicast addresses for the bonding
886 * device and retransmit an IGMP JOIN request to the current active
887 * slave.
888 */
889static void bond_resend_igmp_join_requests(struct bonding *bond)
890{
891 struct net_device *vlan_dev;
892 struct vlan_entry *vlan;
893
894 read_lock(&bond->lock);
895
896 /* rejoin all groups on bond device */
897 __bond_resend_igmp_join_requests(bond->dev);
898
899 /* rejoin all groups on vlan devices */
900 if (bond->vlgrp) {
901 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
902 vlan_dev = vlan_group_get_device(bond->vlgrp,
903 vlan->vlan_id);
904 if (vlan_dev)
905 __bond_resend_igmp_join_requests(vlan_dev);
906 }
907 }
908
909 if (--bond->igmp_retrans > 0)
910 queue_delayed_work(bond->wq, &bond->mcast_work, HZ/5);
911
912 read_unlock(&bond->lock);
913}
914
915static void bond_resend_igmp_join_requests_delayed(struct work_struct *work)
916{
917 struct bonding *bond = container_of(work, struct bonding,
918 mcast_work.work);
919 bond_resend_igmp_join_requests(bond);
920}
921
922/*
923 * flush all members of flush->mc_list from device dev->mc_list
924 */
925static void bond_mc_list_flush(struct net_device *bond_dev,
926 struct net_device *slave_dev)
927{
928 struct bonding *bond = netdev_priv(bond_dev);
929 struct netdev_hw_addr *ha;
930
931 netdev_for_each_mc_addr(ha, bond_dev)
932 dev_mc_del(slave_dev, ha->addr);
933
934 if (bond->params.mode == BOND_MODE_8023AD) {
935 /* del lacpdu mc addr from mc list */
936 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
937
938 dev_mc_del(slave_dev, lacpdu_multicast);
939 }
940}
941
942/*--------------------------- Active slave change ---------------------------*/
943
944/*
945 * Update the mc list and multicast-related flags for the new and
946 * old active slaves (if any) according to the multicast mode, and
947 * promiscuous flags unconditionally.
948 */
949static void bond_mc_swap(struct bonding *bond, struct slave *new_active,
950 struct slave *old_active)
951{
952 struct netdev_hw_addr *ha;
953
954 if (!USES_PRIMARY(bond->params.mode))
955 /* nothing to do - mc list is already up-to-date on
956 * all slaves
957 */
958 return;
959
960 if (old_active) {
961 if (bond->dev->flags & IFF_PROMISC)
962 dev_set_promiscuity(old_active->dev, -1);
963
964 if (bond->dev->flags & IFF_ALLMULTI)
965 dev_set_allmulti(old_active->dev, -1);
966
967 netdev_for_each_mc_addr(ha, bond->dev)
968 dev_mc_del(old_active->dev, ha->addr);
969 }
970
971 if (new_active) {
972 /* FIXME: Signal errors upstream. */
973 if (bond->dev->flags & IFF_PROMISC)
974 dev_set_promiscuity(new_active->dev, 1);
975
976 if (bond->dev->flags & IFF_ALLMULTI)
977 dev_set_allmulti(new_active->dev, 1);
978
979 netdev_for_each_mc_addr(ha, bond->dev)
980 dev_mc_add(new_active->dev, ha->addr);
981 }
982}
983
984/*
985 * bond_do_fail_over_mac
986 *
987 * Perform special MAC address swapping for fail_over_mac settings
988 *
989 * Called with RTNL, bond->lock for read, curr_slave_lock for write_bh.
990 */
991static void bond_do_fail_over_mac(struct bonding *bond,
992 struct slave *new_active,
993 struct slave *old_active)
994 __releases(&bond->curr_slave_lock)
995 __releases(&bond->lock)
996 __acquires(&bond->lock)
997 __acquires(&bond->curr_slave_lock)
998{
999 u8 tmp_mac[ETH_ALEN];
1000 struct sockaddr saddr;
1001 int rv;
1002
1003 switch (bond->params.fail_over_mac) {
1004 case BOND_FOM_ACTIVE:
1005 if (new_active)
1006 memcpy(bond->dev->dev_addr, new_active->dev->dev_addr,
1007 new_active->dev->addr_len);
1008 break;
1009 case BOND_FOM_FOLLOW:
1010 /*
1011 * if new_active && old_active, swap them
1012 * if just old_active, do nothing (going to no active slave)
1013 * if just new_active, set new_active to bond's MAC
1014 */
1015 if (!new_active)
1016 return;
1017
1018 write_unlock_bh(&bond->curr_slave_lock);
1019 read_unlock(&bond->lock);
1020
1021 if (old_active) {
1022 memcpy(tmp_mac, new_active->dev->dev_addr, ETH_ALEN);
1023 memcpy(saddr.sa_data, old_active->dev->dev_addr,
1024 ETH_ALEN);
1025 saddr.sa_family = new_active->dev->type;
1026 } else {
1027 memcpy(saddr.sa_data, bond->dev->dev_addr, ETH_ALEN);
1028 saddr.sa_family = bond->dev->type;
1029 }
1030
1031 rv = dev_set_mac_address(new_active->dev, &saddr);
1032 if (rv) {
1033 pr_err("%s: Error %d setting MAC of slave %s\n",
1034 bond->dev->name, -rv, new_active->dev->name);
1035 goto out;
1036 }
1037
1038 if (!old_active)
1039 goto out;
1040
1041 memcpy(saddr.sa_data, tmp_mac, ETH_ALEN);
1042 saddr.sa_family = old_active->dev->type;
1043
1044 rv = dev_set_mac_address(old_active->dev, &saddr);
1045 if (rv)
1046 pr_err("%s: Error %d setting MAC of slave %s\n",
1047 bond->dev->name, -rv, new_active->dev->name);
1048out:
1049 read_lock(&bond->lock);
1050 write_lock_bh(&bond->curr_slave_lock);
1051 break;
1052 default:
1053 pr_err("%s: bond_do_fail_over_mac impossible: bad policy %d\n",
1054 bond->dev->name, bond->params.fail_over_mac);
1055 break;
1056 }
1057
1058}
1059
1060static bool bond_should_change_active(struct bonding *bond)
1061{
1062 struct slave *prim = bond->primary_slave;
1063 struct slave *curr = bond->curr_active_slave;
1064
1065 if (!prim || !curr || curr->link != BOND_LINK_UP)
1066 return true;
1067 if (bond->force_primary) {
1068 bond->force_primary = false;
1069 return true;
1070 }
1071 if (bond->params.primary_reselect == BOND_PRI_RESELECT_BETTER &&
1072 (prim->speed < curr->speed ||
1073 (prim->speed == curr->speed && prim->duplex <= curr->duplex)))
1074 return false;
1075 if (bond->params.primary_reselect == BOND_PRI_RESELECT_FAILURE)
1076 return false;
1077 return true;
1078}
1079
1080/**
1081 * find_best_interface - select the best available slave to be the active one
1082 * @bond: our bonding struct
1083 *
1084 * Warning: Caller must hold curr_slave_lock for writing.
1085 */
1086static struct slave *bond_find_best_slave(struct bonding *bond)
1087{
1088 struct slave *new_active, *old_active;
1089 struct slave *bestslave = NULL;
1090 int mintime = bond->params.updelay;
1091 int i;
1092
1093 new_active = bond->curr_active_slave;
1094
1095 if (!new_active) { /* there were no active slaves left */
1096 if (bond->slave_cnt > 0) /* found one slave */
1097 new_active = bond->first_slave;
1098 else
1099 return NULL; /* still no slave, return NULL */
1100 }
1101
1102 if ((bond->primary_slave) &&
1103 bond->primary_slave->link == BOND_LINK_UP &&
1104 bond_should_change_active(bond)) {
1105 new_active = bond->primary_slave;
1106 }
1107
1108 /* remember where to stop iterating over the slaves */
1109 old_active = new_active;
1110
1111 bond_for_each_slave_from(bond, new_active, i, old_active) {
1112 if (new_active->link == BOND_LINK_UP) {
1113 return new_active;
1114 } else if (new_active->link == BOND_LINK_BACK &&
1115 IS_UP(new_active->dev)) {
1116 /* link up, but waiting for stabilization */
1117 if (new_active->delay < mintime) {
1118 mintime = new_active->delay;
1119 bestslave = new_active;
1120 }
1121 }
1122 }
1123
1124 return bestslave;
1125}
1126
1127/**
1128 * change_active_interface - change the active slave into the specified one
1129 * @bond: our bonding struct
1130 * @new: the new slave to make the active one
1131 *
1132 * Set the new slave to the bond's settings and unset them on the old
1133 * curr_active_slave.
1134 * Setting include flags, mc-list, promiscuity, allmulti, etc.
1135 *
1136 * If @new's link state is %BOND_LINK_BACK we'll set it to %BOND_LINK_UP,
1137 * because it is apparently the best available slave we have, even though its
1138 * updelay hasn't timed out yet.
1139 *
1140 * If new_active is not NULL, caller must hold bond->lock for read and
1141 * curr_slave_lock for write_bh.
1142 */
1143void bond_change_active_slave(struct bonding *bond, struct slave *new_active)
1144{
1145 struct slave *old_active = bond->curr_active_slave;
1146
1147 if (old_active == new_active)
1148 return;
1149
1150 if (new_active) {
1151 new_active->jiffies = jiffies;
1152
1153 if (new_active->link == BOND_LINK_BACK) {
1154 if (USES_PRIMARY(bond->params.mode)) {
1155 pr_info("%s: making interface %s the new active one %d ms earlier.\n",
1156 bond->dev->name, new_active->dev->name,
1157 (bond->params.updelay - new_active->delay) * bond->params.miimon);
1158 }
1159
1160 new_active->delay = 0;
1161 new_active->link = BOND_LINK_UP;
1162
1163 if (bond->params.mode == BOND_MODE_8023AD)
1164 bond_3ad_handle_link_change(new_active, BOND_LINK_UP);
1165
1166 if (bond_is_lb(bond))
1167 bond_alb_handle_link_change(bond, new_active, BOND_LINK_UP);
1168 } else {
1169 if (USES_PRIMARY(bond->params.mode)) {
1170 pr_info("%s: making interface %s the new active one.\n",
1171 bond->dev->name, new_active->dev->name);
1172 }
1173 }
1174 }
1175
1176 if (USES_PRIMARY(bond->params.mode))
1177 bond_mc_swap(bond, new_active, old_active);
1178
1179 if (bond_is_lb(bond)) {
1180 bond_alb_handle_active_change(bond, new_active);
1181 if (old_active)
1182 bond_set_slave_inactive_flags(old_active);
1183 if (new_active)
1184 bond_set_slave_active_flags(new_active);
1185 } else {
1186 bond->curr_active_slave = new_active;
1187 }
1188
1189 if (bond->params.mode == BOND_MODE_ACTIVEBACKUP) {
1190 if (old_active)
1191 bond_set_slave_inactive_flags(old_active);
1192
1193 if (new_active) {
1194 bond_set_slave_active_flags(new_active);
1195
1196 if (bond->params.fail_over_mac)
1197 bond_do_fail_over_mac(bond, new_active,
1198 old_active);
1199
1200 bond->send_grat_arp = bond->params.num_grat_arp;
1201 bond_send_gratuitous_arp(bond);
1202
1203 bond->send_unsol_na = bond->params.num_unsol_na;
1204 bond_send_unsolicited_na(bond);
1205
1206 write_unlock_bh(&bond->curr_slave_lock);
1207 read_unlock(&bond->lock);
1208
1209 netdev_bonding_change(bond->dev, NETDEV_BONDING_FAILOVER);
1210
1211 read_lock(&bond->lock);
1212 write_lock_bh(&bond->curr_slave_lock);
1213 }
1214 }
1215
1216 /* resend IGMP joins since active slave has changed or
1217 * all were sent on curr_active_slave */
1218 if ((USES_PRIMARY(bond->params.mode) && new_active) ||
1219 bond->params.mode == BOND_MODE_ROUNDROBIN) {
1220 bond->igmp_retrans = bond->params.resend_igmp;
1221 queue_delayed_work(bond->wq, &bond->mcast_work, 0);
1222 }
1223}
1224
1225/**
1226 * bond_select_active_slave - select a new active slave, if needed
1227 * @bond: our bonding struct
1228 *
1229 * This functions should be called when one of the following occurs:
1230 * - The old curr_active_slave has been released or lost its link.
1231 * - The primary_slave has got its link back.
1232 * - A slave has got its link back and there's no old curr_active_slave.
1233 *
1234 * Caller must hold bond->lock for read and curr_slave_lock for write_bh.
1235 */
1236void bond_select_active_slave(struct bonding *bond)
1237{
1238 struct slave *best_slave;
1239 int rv;
1240
1241 best_slave = bond_find_best_slave(bond);
1242 if (best_slave != bond->curr_active_slave) {
1243 bond_change_active_slave(bond, best_slave);
1244 rv = bond_set_carrier(bond);
1245 if (!rv)
1246 return;
1247
1248 if (netif_carrier_ok(bond->dev)) {
1249 pr_info("%s: first active interface up!\n",
1250 bond->dev->name);
1251 } else {
1252 pr_info("%s: now running without any active interface !\n",
1253 bond->dev->name);
1254 }
1255 }
1256}
1257
1258/*--------------------------- slave list handling ---------------------------*/
1259
1260/*
1261 * This function attaches the slave to the end of list.
1262 *
1263 * bond->lock held for writing by caller.
1264 */
1265static void bond_attach_slave(struct bonding *bond, struct slave *new_slave)
1266{
1267 if (bond->first_slave == NULL) { /* attaching the first slave */
1268 new_slave->next = new_slave;
1269 new_slave->prev = new_slave;
1270 bond->first_slave = new_slave;
1271 } else {
1272 new_slave->next = bond->first_slave;
1273 new_slave->prev = bond->first_slave->prev;
1274 new_slave->next->prev = new_slave;
1275 new_slave->prev->next = new_slave;
1276 }
1277
1278 bond->slave_cnt++;
1279}
1280
1281/*
1282 * This function detaches the slave from the list.
1283 * WARNING: no check is made to verify if the slave effectively
1284 * belongs to <bond>.
1285 * Nothing is freed on return, structures are just unchained.
1286 * If any slave pointer in bond was pointing to <slave>,
1287 * it should be changed by the calling function.
1288 *
1289 * bond->lock held for writing by caller.
1290 */
1291static void bond_detach_slave(struct bonding *bond, struct slave *slave)
1292{
1293 if (slave->next)
1294 slave->next->prev = slave->prev;
1295
1296 if (slave->prev)
1297 slave->prev->next = slave->next;
1298
1299 if (bond->first_slave == slave) { /* slave is the first slave */
1300 if (bond->slave_cnt > 1) { /* there are more slave */
1301 bond->first_slave = slave->next;
1302 } else {
1303 bond->first_slave = NULL; /* slave was the last one */
1304 }
1305 }
1306
1307 slave->next = NULL;
1308 slave->prev = NULL;
1309 bond->slave_cnt--;
1310}
1311
1312#ifdef CONFIG_NET_POLL_CONTROLLER
1313/*
1314 * You must hold read lock on bond->lock before calling this.
1315 */
1316static bool slaves_support_netpoll(struct net_device *bond_dev)
1317{
1318 struct bonding *bond = netdev_priv(bond_dev);
1319 struct slave *slave;
1320 int i = 0;
1321 bool ret = true;
1322
1323 bond_for_each_slave(bond, slave, i) {
1324 if ((slave->dev->priv_flags & IFF_DISABLE_NETPOLL) ||
1325 !slave->dev->netdev_ops->ndo_poll_controller)
1326 ret = false;
1327 }
1328 return i != 0 && ret;
1329}
1330
1331static void bond_poll_controller(struct net_device *bond_dev)
1332{
1333 struct bonding *bond = netdev_priv(bond_dev);
1334 struct slave *slave;
1335 int i;
1336
1337 bond_for_each_slave(bond, slave, i) {
1338 if (slave->dev && IS_UP(slave->dev))
1339 netpoll_poll_dev(slave->dev);
1340 }
1341}
1342
1343static void bond_netpoll_cleanup(struct net_device *bond_dev)
1344{
1345 struct bonding *bond = netdev_priv(bond_dev);
1346 struct slave *slave;
1347 const struct net_device_ops *ops;
1348 int i;
1349
1350 read_lock(&bond->lock);
1351 bond_dev->npinfo = NULL;
1352 bond_for_each_slave(bond, slave, i) {
1353 if (slave->dev) {
1354 ops = slave->dev->netdev_ops;
1355 if (ops->ndo_netpoll_cleanup)
1356 ops->ndo_netpoll_cleanup(slave->dev);
1357 else
1358 slave->dev->npinfo = NULL;
1359 }
1360 }
1361 read_unlock(&bond->lock);
1362}
1363
1364#else
1365
1366static void bond_netpoll_cleanup(struct net_device *bond_dev)
1367{
1368}
1369
1370#endif
1371
1372/*---------------------------------- IOCTL ----------------------------------*/
1373
1374static int bond_sethwaddr(struct net_device *bond_dev,
1375 struct net_device *slave_dev)
1376{
1377 pr_debug("bond_dev=%p\n", bond_dev);
1378 pr_debug("slave_dev=%p\n", slave_dev);
1379 pr_debug("slave_dev->addr_len=%d\n", slave_dev->addr_len);
1380 memcpy(bond_dev->dev_addr, slave_dev->dev_addr, slave_dev->addr_len);
1381 return 0;
1382}
1383
1384#define BOND_VLAN_FEATURES \
1385 (NETIF_F_VLAN_CHALLENGED | NETIF_F_HW_VLAN_RX | NETIF_F_HW_VLAN_TX | \
1386 NETIF_F_HW_VLAN_FILTER)
1387
1388/*
1389 * Compute the common dev->feature set available to all slaves. Some
1390 * feature bits are managed elsewhere, so preserve those feature bits
1391 * on the master device.
1392 */
1393static int bond_compute_features(struct bonding *bond)
1394{
1395 struct slave *slave;
1396 struct net_device *bond_dev = bond->dev;
1397 unsigned long features = bond_dev->features;
1398 unsigned long vlan_features = 0;
1399 unsigned short max_hard_header_len = max((u16)ETH_HLEN,
1400 bond_dev->hard_header_len);
1401 int i;
1402
1403 features &= ~(NETIF_F_ALL_CSUM | BOND_VLAN_FEATURES);
1404 features |= NETIF_F_GSO_MASK | NETIF_F_NO_CSUM;
1405
1406 if (!bond->first_slave)
1407 goto done;
1408
1409 features &= ~NETIF_F_ONE_FOR_ALL;
1410
1411 vlan_features = bond->first_slave->dev->vlan_features;
1412 bond_for_each_slave(bond, slave, i) {
1413 features = netdev_increment_features(features,
1414 slave->dev->features,
1415 NETIF_F_ONE_FOR_ALL);
1416 vlan_features = netdev_increment_features(vlan_features,
1417 slave->dev->vlan_features,
1418 NETIF_F_ONE_FOR_ALL);
1419 if (slave->dev->hard_header_len > max_hard_header_len)
1420 max_hard_header_len = slave->dev->hard_header_len;
1421 }
1422
1423done:
1424 features |= (bond_dev->features & BOND_VLAN_FEATURES);
1425 bond_dev->features = netdev_fix_features(features, NULL);
1426 bond_dev->vlan_features = netdev_fix_features(vlan_features, NULL);
1427 bond_dev->hard_header_len = max_hard_header_len;
1428
1429 return 0;
1430}
1431
1432static void bond_setup_by_slave(struct net_device *bond_dev,
1433 struct net_device *slave_dev)
1434{
1435 struct bonding *bond = netdev_priv(bond_dev);
1436
1437 bond_dev->header_ops = slave_dev->header_ops;
1438
1439 bond_dev->type = slave_dev->type;
1440 bond_dev->hard_header_len = slave_dev->hard_header_len;
1441 bond_dev->addr_len = slave_dev->addr_len;
1442
1443 memcpy(bond_dev->broadcast, slave_dev->broadcast,
1444 slave_dev->addr_len);
1445 bond->setup_by_slave = 1;
1446}
1447
1448/* enslave device <slave> to bond device <master> */
1449int bond_enslave(struct net_device *bond_dev, struct net_device *slave_dev)
1450{
1451 struct bonding *bond = netdev_priv(bond_dev);
1452 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
1453 struct slave *new_slave = NULL;
1454 struct netdev_hw_addr *ha;
1455 struct sockaddr addr;
1456 int link_reporting;
1457 int old_features = bond_dev->features;
1458 int res = 0;
1459
1460 if (!bond->params.use_carrier && slave_dev->ethtool_ops == NULL &&
1461 slave_ops->ndo_do_ioctl == NULL) {
1462 pr_warning("%s: Warning: no link monitoring support for %s\n",
1463 bond_dev->name, slave_dev->name);
1464 }
1465
1466 /* bond must be initialized by bond_open() before enslaving */
1467 if (!(bond_dev->flags & IFF_UP)) {
1468 pr_warning("%s: master_dev is not up in bond_enslave\n",
1469 bond_dev->name);
1470 }
1471
1472 /* already enslaved */
1473 if (slave_dev->flags & IFF_SLAVE) {
1474 pr_debug("Error, Device was already enslaved\n");
1475 return -EBUSY;
1476 }
1477
1478 /* vlan challenged mutual exclusion */
1479 /* no need to lock since we're protected by rtnl_lock */
1480 if (slave_dev->features & NETIF_F_VLAN_CHALLENGED) {
1481 pr_debug("%s: NETIF_F_VLAN_CHALLENGED\n", slave_dev->name);
1482 if (bond->vlgrp) {
1483 pr_err("%s: Error: cannot enslave VLAN challenged slave %s on VLAN enabled bond %s\n",
1484 bond_dev->name, slave_dev->name, bond_dev->name);
1485 return -EPERM;
1486 } else {
1487 pr_warning("%s: Warning: enslaved VLAN challenged slave %s. Adding VLANs will be blocked as long as %s is part of bond %s\n",
1488 bond_dev->name, slave_dev->name,
1489 slave_dev->name, bond_dev->name);
1490 bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
1491 }
1492 } else {
1493 pr_debug("%s: ! NETIF_F_VLAN_CHALLENGED\n", slave_dev->name);
1494 if (bond->slave_cnt == 0) {
1495 /* First slave, and it is not VLAN challenged,
1496 * so remove the block of adding VLANs over the bond.
1497 */
1498 bond_dev->features &= ~NETIF_F_VLAN_CHALLENGED;
1499 }
1500 }
1501
1502 /*
1503 * Old ifenslave binaries are no longer supported. These can
1504 * be identified with moderate accuracy by the state of the slave:
1505 * the current ifenslave will set the interface down prior to
1506 * enslaving it; the old ifenslave will not.
1507 */
1508 if ((slave_dev->flags & IFF_UP)) {
1509 pr_err("%s is up. This may be due to an out of date ifenslave.\n",
1510 slave_dev->name);
1511 res = -EPERM;
1512 goto err_undo_flags;
1513 }
1514
1515 /* set bonding device ether type by slave - bonding netdevices are
1516 * created with ether_setup, so when the slave type is not ARPHRD_ETHER
1517 * there is a need to override some of the type dependent attribs/funcs.
1518 *
1519 * bond ether type mutual exclusion - don't allow slaves of dissimilar
1520 * ether type (eg ARPHRD_ETHER and ARPHRD_INFINIBAND) share the same bond
1521 */
1522 if (bond->slave_cnt == 0) {
1523 if (bond_dev->type != slave_dev->type) {
1524 pr_debug("%s: change device type from %d to %d\n",
1525 bond_dev->name,
1526 bond_dev->type, slave_dev->type);
1527
1528 res = netdev_bonding_change(bond_dev,
1529 NETDEV_PRE_TYPE_CHANGE);
1530 res = notifier_to_errno(res);
1531 if (res) {
1532 pr_err("%s: refused to change device type\n",
1533 bond_dev->name);
1534 res = -EBUSY;
1535 goto err_undo_flags;
1536 }
1537
1538 /* Flush unicast and multicast addresses */
1539 dev_uc_flush(bond_dev);
1540 dev_mc_flush(bond_dev);
1541
1542 if (slave_dev->type != ARPHRD_ETHER)
1543 bond_setup_by_slave(bond_dev, slave_dev);
1544 else
1545 ether_setup(bond_dev);
1546
1547 netdev_bonding_change(bond_dev,
1548 NETDEV_POST_TYPE_CHANGE);
1549 }
1550 } else if (bond_dev->type != slave_dev->type) {
1551 pr_err("%s ether type (%d) is different from other slaves (%d), can not enslave it.\n",
1552 slave_dev->name,
1553 slave_dev->type, bond_dev->type);
1554 res = -EINVAL;
1555 goto err_undo_flags;
1556 }
1557
1558 if (slave_ops->ndo_set_mac_address == NULL) {
1559 if (bond->slave_cnt == 0) {
1560 pr_warning("%s: Warning: The first slave device specified does not support setting the MAC address. Setting fail_over_mac to active.",
1561 bond_dev->name);
1562 bond->params.fail_over_mac = BOND_FOM_ACTIVE;
1563 } else if (bond->params.fail_over_mac != BOND_FOM_ACTIVE) {
1564 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",
1565 bond_dev->name);
1566 res = -EOPNOTSUPP;
1567 goto err_undo_flags;
1568 }
1569 }
1570
1571 /* If this is the first slave, then we need to set the master's hardware
1572 * address to be the same as the slave's. */
1573 if (bond->slave_cnt == 0)
1574 memcpy(bond->dev->dev_addr, slave_dev->dev_addr,
1575 slave_dev->addr_len);
1576
1577
1578 new_slave = kzalloc(sizeof(struct slave), GFP_KERNEL);
1579 if (!new_slave) {
1580 res = -ENOMEM;
1581 goto err_undo_flags;
1582 }
1583
1584 /*
1585 * Set the new_slave's queue_id to be zero. Queue ID mapping
1586 * is set via sysfs or module option if desired.
1587 */
1588 new_slave->queue_id = 0;
1589
1590 /* Save slave's original mtu and then set it to match the bond */
1591 new_slave->original_mtu = slave_dev->mtu;
1592 res = dev_set_mtu(slave_dev, bond->dev->mtu);
1593 if (res) {
1594 pr_debug("Error %d calling dev_set_mtu\n", res);
1595 goto err_free;
1596 }
1597
1598 /*
1599 * Save slave's original ("permanent") mac address for modes
1600 * that need it, and for restoring it upon release, and then
1601 * set it to the master's address
1602 */
1603 memcpy(new_slave->perm_hwaddr, slave_dev->dev_addr, ETH_ALEN);
1604
1605 if (!bond->params.fail_over_mac) {
1606 /*
1607 * Set slave to master's mac address. The application already
1608 * set the master's mac address to that of the first slave
1609 */
1610 memcpy(addr.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
1611 addr.sa_family = slave_dev->type;
1612 res = dev_set_mac_address(slave_dev, &addr);
1613 if (res) {
1614 pr_debug("Error %d calling set_mac_address\n", res);
1615 goto err_restore_mtu;
1616 }
1617 }
1618
1619 res = netdev_set_master(slave_dev, bond_dev);
1620 if (res) {
1621 pr_debug("Error %d calling netdev_set_master\n", res);
1622 goto err_restore_mac;
1623 }
1624 /* open the slave since the application closed it */
1625 res = dev_open(slave_dev);
1626 if (res) {
1627 pr_debug("Opening slave %s failed\n", slave_dev->name);
1628 goto err_unset_master;
1629 }
1630
1631 new_slave->dev = slave_dev;
1632 slave_dev->priv_flags |= IFF_BONDING;
1633
1634 if (bond_is_lb(bond)) {
1635 /* bond_alb_init_slave() must be called before all other stages since
1636 * it might fail and we do not want to have to undo everything
1637 */
1638 res = bond_alb_init_slave(bond, new_slave);
1639 if (res)
1640 goto err_close;
1641 }
1642
1643 /* If the mode USES_PRIMARY, then the new slave gets the
1644 * master's promisc (and mc) settings only if it becomes the
1645 * curr_active_slave, and that is taken care of later when calling
1646 * bond_change_active()
1647 */
1648 if (!USES_PRIMARY(bond->params.mode)) {
1649 /* set promiscuity level to new slave */
1650 if (bond_dev->flags & IFF_PROMISC) {
1651 res = dev_set_promiscuity(slave_dev, 1);
1652 if (res)
1653 goto err_close;
1654 }
1655
1656 /* set allmulti level to new slave */
1657 if (bond_dev->flags & IFF_ALLMULTI) {
1658 res = dev_set_allmulti(slave_dev, 1);
1659 if (res)
1660 goto err_close;
1661 }
1662
1663 netif_addr_lock_bh(bond_dev);
1664 /* upload master's mc_list to new slave */
1665 netdev_for_each_mc_addr(ha, bond_dev)
1666 dev_mc_add(slave_dev, ha->addr);
1667 netif_addr_unlock_bh(bond_dev);
1668 }
1669
1670 if (bond->params.mode == BOND_MODE_8023AD) {
1671 /* add lacpdu mc addr to mc list */
1672 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
1673
1674 dev_mc_add(slave_dev, lacpdu_multicast);
1675 }
1676
1677 bond_add_vlans_on_slave(bond, slave_dev);
1678
1679 write_lock_bh(&bond->lock);
1680
1681 bond_attach_slave(bond, new_slave);
1682
1683 new_slave->delay = 0;
1684 new_slave->link_failure_count = 0;
1685
1686 bond_compute_features(bond);
1687
1688 write_unlock_bh(&bond->lock);
1689
1690 read_lock(&bond->lock);
1691
1692 new_slave->last_arp_rx = jiffies;
1693
1694 if (bond->params.miimon && !bond->params.use_carrier) {
1695 link_reporting = bond_check_dev_link(bond, slave_dev, 1);
1696
1697 if ((link_reporting == -1) && !bond->params.arp_interval) {
1698 /*
1699 * miimon is set but a bonded network driver
1700 * does not support ETHTOOL/MII and
1701 * arp_interval is not set. Note: if
1702 * use_carrier is enabled, we will never go
1703 * here (because netif_carrier is always
1704 * supported); thus, we don't need to change
1705 * the messages for netif_carrier.
1706 */
1707 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",
1708 bond_dev->name, slave_dev->name);
1709 } else if (link_reporting == -1) {
1710 /* unable get link status using mii/ethtool */
1711 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",
1712 bond_dev->name, slave_dev->name);
1713 }
1714 }
1715
1716 /* check for initial state */
1717 if (!bond->params.miimon ||
1718 (bond_check_dev_link(bond, slave_dev, 0) == BMSR_LSTATUS)) {
1719 if (bond->params.updelay) {
1720 pr_debug("Initial state of slave_dev is BOND_LINK_BACK\n");
1721 new_slave->link = BOND_LINK_BACK;
1722 new_slave->delay = bond->params.updelay;
1723 } else {
1724 pr_debug("Initial state of slave_dev is BOND_LINK_UP\n");
1725 new_slave->link = BOND_LINK_UP;
1726 }
1727 new_slave->jiffies = jiffies;
1728 } else {
1729 pr_debug("Initial state of slave_dev is BOND_LINK_DOWN\n");
1730 new_slave->link = BOND_LINK_DOWN;
1731 }
1732
1733 if (bond_update_speed_duplex(new_slave) &&
1734 (new_slave->link != BOND_LINK_DOWN)) {
1735 pr_warning("%s: Warning: failed to get speed and duplex from %s, assumed to be 100Mb/sec and Full.\n",
1736 bond_dev->name, new_slave->dev->name);
1737
1738 if (bond->params.mode == BOND_MODE_8023AD) {
1739 pr_warning("%s: Warning: Operation of 802.3ad mode requires ETHTOOL support in base driver for proper aggregator selection.\n",
1740 bond_dev->name);
1741 }
1742 }
1743
1744 if (USES_PRIMARY(bond->params.mode) && bond->params.primary[0]) {
1745 /* if there is a primary slave, remember it */
1746 if (strcmp(bond->params.primary, new_slave->dev->name) == 0) {
1747 bond->primary_slave = new_slave;
1748 bond->force_primary = true;
1749 }
1750 }
1751
1752 write_lock_bh(&bond->curr_slave_lock);
1753
1754 switch (bond->params.mode) {
1755 case BOND_MODE_ACTIVEBACKUP:
1756 bond_set_slave_inactive_flags(new_slave);
1757 bond_select_active_slave(bond);
1758 break;
1759 case BOND_MODE_8023AD:
1760 /* in 802.3ad mode, the internal mechanism
1761 * will activate the slaves in the selected
1762 * aggregator
1763 */
1764 bond_set_slave_inactive_flags(new_slave);
1765 /* if this is the first slave */
1766 if (bond->slave_cnt == 1) {
1767 SLAVE_AD_INFO(new_slave).id = 1;
1768 /* Initialize AD with the number of times that the AD timer is called in 1 second
1769 * can be called only after the mac address of the bond is set
1770 */
1771 bond_3ad_initialize(bond, 1000/AD_TIMER_INTERVAL,
1772 bond->params.lacp_fast);
1773 } else {
1774 SLAVE_AD_INFO(new_slave).id =
1775 SLAVE_AD_INFO(new_slave->prev).id + 1;
1776 }
1777
1778 bond_3ad_bind_slave(new_slave);
1779 break;
1780 case BOND_MODE_TLB:
1781 case BOND_MODE_ALB:
1782 new_slave->state = BOND_STATE_ACTIVE;
1783 bond_set_slave_inactive_flags(new_slave);
1784 bond_select_active_slave(bond);
1785 break;
1786 default:
1787 pr_debug("This slave is always active in trunk mode\n");
1788
1789 /* always active in trunk mode */
1790 new_slave->state = BOND_STATE_ACTIVE;
1791
1792 /* In trunking mode there is little meaning to curr_active_slave
1793 * anyway (it holds no special properties of the bond device),
1794 * so we can change it without calling change_active_interface()
1795 */
1796 if (!bond->curr_active_slave)
1797 bond->curr_active_slave = new_slave;
1798
1799 break;
1800 } /* switch(bond_mode) */
1801
1802 write_unlock_bh(&bond->curr_slave_lock);
1803
1804 bond_set_carrier(bond);
1805
1806#ifdef CONFIG_NET_POLL_CONTROLLER
1807 if (slaves_support_netpoll(bond_dev)) {
1808 bond_dev->priv_flags &= ~IFF_DISABLE_NETPOLL;
1809 if (bond_dev->npinfo)
1810 slave_dev->npinfo = bond_dev->npinfo;
1811 } else if (!(bond_dev->priv_flags & IFF_DISABLE_NETPOLL)) {
1812 bond_dev->priv_flags |= IFF_DISABLE_NETPOLL;
1813 pr_info("New slave device %s does not support netpoll\n",
1814 slave_dev->name);
1815 pr_info("Disabling netpoll support for %s\n", bond_dev->name);
1816 }
1817#endif
1818 read_unlock(&bond->lock);
1819
1820 res = bond_create_slave_symlinks(bond_dev, slave_dev);
1821 if (res)
1822 goto err_close;
1823
1824 pr_info("%s: enslaving %s as a%s interface with a%s link.\n",
1825 bond_dev->name, slave_dev->name,
1826 new_slave->state == BOND_STATE_ACTIVE ? "n active" : " backup",
1827 new_slave->link != BOND_LINK_DOWN ? "n up" : " down");
1828
1829 /* enslave is successful */
1830 return 0;
1831
1832/* Undo stages on error */
1833err_close:
1834 dev_close(slave_dev);
1835
1836err_unset_master:
1837 netdev_set_master(slave_dev, NULL);
1838
1839err_restore_mac:
1840 if (!bond->params.fail_over_mac) {
1841 /* XXX TODO - fom follow mode needs to change master's
1842 * MAC if this slave's MAC is in use by the bond, or at
1843 * least print a warning.
1844 */
1845 memcpy(addr.sa_data, new_slave->perm_hwaddr, ETH_ALEN);
1846 addr.sa_family = slave_dev->type;
1847 dev_set_mac_address(slave_dev, &addr);
1848 }
1849
1850err_restore_mtu:
1851 dev_set_mtu(slave_dev, new_slave->original_mtu);
1852
1853err_free:
1854 kfree(new_slave);
1855
1856err_undo_flags:
1857 bond_dev->features = old_features;
1858
1859 return res;
1860}
1861
1862/*
1863 * Try to release the slave device <slave> from the bond device <master>
1864 * It is legal to access curr_active_slave without a lock because all the function
1865 * is write-locked.
1866 *
1867 * The rules for slave state should be:
1868 * for Active/Backup:
1869 * Active stays on all backups go down
1870 * for Bonded connections:
1871 * The first up interface should be left on and all others downed.
1872 */
1873int bond_release(struct net_device *bond_dev, struct net_device *slave_dev)
1874{
1875 struct bonding *bond = netdev_priv(bond_dev);
1876 struct slave *slave, *oldcurrent;
1877 struct sockaddr addr;
1878
1879 /* slave is not a slave or master is not master of this slave */
1880 if (!(slave_dev->flags & IFF_SLAVE) ||
1881 (slave_dev->master != bond_dev)) {
1882 pr_err("%s: Error: cannot release %s.\n",
1883 bond_dev->name, slave_dev->name);
1884 return -EINVAL;
1885 }
1886
1887 block_netpoll_tx();
1888 netdev_bonding_change(bond_dev, NETDEV_BONDING_DESLAVE);
1889 write_lock_bh(&bond->lock);
1890
1891 slave = bond_get_slave_by_dev(bond, slave_dev);
1892 if (!slave) {
1893 /* not a slave of this bond */
1894 pr_info("%s: %s not enslaved\n",
1895 bond_dev->name, slave_dev->name);
1896 write_unlock_bh(&bond->lock);
1897 unblock_netpoll_tx();
1898 return -EINVAL;
1899 }
1900
1901 if (!bond->params.fail_over_mac) {
1902 if (!compare_ether_addr(bond_dev->dev_addr, slave->perm_hwaddr) &&
1903 bond->slave_cnt > 1)
1904 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",
1905 bond_dev->name, slave_dev->name,
1906 slave->perm_hwaddr,
1907 bond_dev->name, slave_dev->name);
1908 }
1909
1910 /* Inform AD package of unbinding of slave. */
1911 if (bond->params.mode == BOND_MODE_8023AD) {
1912 /* must be called before the slave is
1913 * detached from the list
1914 */
1915 bond_3ad_unbind_slave(slave);
1916 }
1917
1918 pr_info("%s: releasing %s interface %s\n",
1919 bond_dev->name,
1920 (slave->state == BOND_STATE_ACTIVE) ? "active" : "backup",
1921 slave_dev->name);
1922
1923 oldcurrent = bond->curr_active_slave;
1924
1925 bond->current_arp_slave = NULL;
1926
1927 /* release the slave from its bond */
1928 bond_detach_slave(bond, slave);
1929
1930 bond_compute_features(bond);
1931
1932 if (bond->primary_slave == slave)
1933 bond->primary_slave = NULL;
1934
1935 if (oldcurrent == slave)
1936 bond_change_active_slave(bond, NULL);
1937
1938 if (bond_is_lb(bond)) {
1939 /* Must be called only after the slave has been
1940 * detached from the list and the curr_active_slave
1941 * has been cleared (if our_slave == old_current),
1942 * but before a new active slave is selected.
1943 */
1944 write_unlock_bh(&bond->lock);
1945 bond_alb_deinit_slave(bond, slave);
1946 write_lock_bh(&bond->lock);
1947 }
1948
1949 if (oldcurrent == slave) {
1950 /*
1951 * Note that we hold RTNL over this sequence, so there
1952 * is no concern that another slave add/remove event
1953 * will interfere.
1954 */
1955 write_unlock_bh(&bond->lock);
1956 read_lock(&bond->lock);
1957 write_lock_bh(&bond->curr_slave_lock);
1958
1959 bond_select_active_slave(bond);
1960
1961 write_unlock_bh(&bond->curr_slave_lock);
1962 read_unlock(&bond->lock);
1963 write_lock_bh(&bond->lock);
1964 }
1965
1966 if (bond->slave_cnt == 0) {
1967 bond_set_carrier(bond);
1968
1969 /* if the last slave was removed, zero the mac address
1970 * of the master so it will be set by the application
1971 * to the mac address of the first slave
1972 */
1973 memset(bond_dev->dev_addr, 0, bond_dev->addr_len);
1974
1975 if (!bond->vlgrp) {
1976 bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
1977 } else {
1978 pr_warning("%s: Warning: clearing HW address of %s while it still has VLANs.\n",
1979 bond_dev->name, bond_dev->name);
1980 pr_warning("%s: When re-adding slaves, make sure the bond's HW address matches its VLANs'.\n",
1981 bond_dev->name);
1982 }
1983 } else if ((bond_dev->features & NETIF_F_VLAN_CHALLENGED) &&
1984 !bond_has_challenged_slaves(bond)) {
1985 pr_info("%s: last VLAN challenged slave %s left bond %s. VLAN blocking is removed\n",
1986 bond_dev->name, slave_dev->name, bond_dev->name);
1987 bond_dev->features &= ~NETIF_F_VLAN_CHALLENGED;
1988 }
1989
1990 write_unlock_bh(&bond->lock);
1991 unblock_netpoll_tx();
1992
1993 /* must do this from outside any spinlocks */
1994 bond_destroy_slave_symlinks(bond_dev, slave_dev);
1995
1996 bond_del_vlans_from_slave(bond, slave_dev);
1997
1998 /* If the mode USES_PRIMARY, then we should only remove its
1999 * promisc and mc settings if it was the curr_active_slave, but that was
2000 * already taken care of above when we detached the slave
2001 */
2002 if (!USES_PRIMARY(bond->params.mode)) {
2003 /* unset promiscuity level from slave */
2004 if (bond_dev->flags & IFF_PROMISC)
2005 dev_set_promiscuity(slave_dev, -1);
2006
2007 /* unset allmulti level from slave */
2008 if (bond_dev->flags & IFF_ALLMULTI)
2009 dev_set_allmulti(slave_dev, -1);
2010
2011 /* flush master's mc_list from slave */
2012 netif_addr_lock_bh(bond_dev);
2013 bond_mc_list_flush(bond_dev, slave_dev);
2014 netif_addr_unlock_bh(bond_dev);
2015 }
2016
2017 netdev_set_master(slave_dev, NULL);
2018
2019#ifdef CONFIG_NET_POLL_CONTROLLER
2020 read_lock_bh(&bond->lock);
2021
2022 if (slaves_support_netpoll(bond_dev))
2023 bond_dev->priv_flags &= ~IFF_DISABLE_NETPOLL;
2024 read_unlock_bh(&bond->lock);
2025 if (slave_dev->netdev_ops->ndo_netpoll_cleanup)
2026 slave_dev->netdev_ops->ndo_netpoll_cleanup(slave_dev);
2027 else
2028 slave_dev->npinfo = NULL;
2029#endif
2030
2031 /* close slave before restoring its mac address */
2032 dev_close(slave_dev);
2033
2034 if (bond->params.fail_over_mac != BOND_FOM_ACTIVE) {
2035 /* restore original ("permanent") mac address */
2036 memcpy(addr.sa_data, slave->perm_hwaddr, ETH_ALEN);
2037 addr.sa_family = slave_dev->type;
2038 dev_set_mac_address(slave_dev, &addr);
2039 }
2040
2041 dev_set_mtu(slave_dev, slave->original_mtu);
2042
2043 slave_dev->priv_flags &= ~(IFF_MASTER_8023AD | IFF_MASTER_ALB |
2044 IFF_SLAVE_INACTIVE | IFF_BONDING |
2045 IFF_SLAVE_NEEDARP);
2046
2047 kfree(slave);
2048
2049 return 0; /* deletion OK */
2050}
2051
2052/*
2053* First release a slave and than destroy the bond if no more slaves are left.
2054* Must be under rtnl_lock when this function is called.
2055*/
2056static int bond_release_and_destroy(struct net_device *bond_dev,
2057 struct net_device *slave_dev)
2058{
2059 struct bonding *bond = netdev_priv(bond_dev);
2060 int ret;
2061
2062 ret = bond_release(bond_dev, slave_dev);
2063 if ((ret == 0) && (bond->slave_cnt == 0)) {
2064 pr_info("%s: destroying bond %s.\n",
2065 bond_dev->name, bond_dev->name);
2066 unregister_netdevice(bond_dev);
2067 }
2068 return ret;
2069}
2070
2071/*
2072 * This function releases all slaves.
2073 */
2074static int bond_release_all(struct net_device *bond_dev)
2075{
2076 struct bonding *bond = netdev_priv(bond_dev);
2077 struct slave *slave;
2078 struct net_device *slave_dev;
2079 struct sockaddr addr;
2080
2081 write_lock_bh(&bond->lock);
2082
2083 netif_carrier_off(bond_dev);
2084
2085 if (bond->slave_cnt == 0)
2086 goto out;
2087
2088 bond->current_arp_slave = NULL;
2089 bond->primary_slave = NULL;
2090 bond_change_active_slave(bond, NULL);
2091
2092 while ((slave = bond->first_slave) != NULL) {
2093 /* Inform AD package of unbinding of slave
2094 * before slave is detached from the list.
2095 */
2096 if (bond->params.mode == BOND_MODE_8023AD)
2097 bond_3ad_unbind_slave(slave);
2098
2099 slave_dev = slave->dev;
2100 bond_detach_slave(bond, slave);
2101
2102 /* now that the slave is detached, unlock and perform
2103 * all the undo steps that should not be called from
2104 * within a lock.
2105 */
2106 write_unlock_bh(&bond->lock);
2107
2108 if (bond_is_lb(bond)) {
2109 /* must be called only after the slave
2110 * has been detached from the list
2111 */
2112 bond_alb_deinit_slave(bond, slave);
2113 }
2114
2115 bond_compute_features(bond);
2116
2117 bond_destroy_slave_symlinks(bond_dev, slave_dev);
2118 bond_del_vlans_from_slave(bond, slave_dev);
2119
2120 /* If the mode USES_PRIMARY, then we should only remove its
2121 * promisc and mc settings if it was the curr_active_slave, but that was
2122 * already taken care of above when we detached the slave
2123 */
2124 if (!USES_PRIMARY(bond->params.mode)) {
2125 /* unset promiscuity level from slave */
2126 if (bond_dev->flags & IFF_PROMISC)
2127 dev_set_promiscuity(slave_dev, -1);
2128
2129 /* unset allmulti level from slave */
2130 if (bond_dev->flags & IFF_ALLMULTI)
2131 dev_set_allmulti(slave_dev, -1);
2132
2133 /* flush master's mc_list from slave */
2134 netif_addr_lock_bh(bond_dev);
2135 bond_mc_list_flush(bond_dev, slave_dev);
2136 netif_addr_unlock_bh(bond_dev);
2137 }
2138
2139 netdev_set_master(slave_dev, NULL);
2140
2141 /* close slave before restoring its mac address */
2142 dev_close(slave_dev);
2143
2144 if (!bond->params.fail_over_mac) {
2145 /* restore original ("permanent") mac address*/
2146 memcpy(addr.sa_data, slave->perm_hwaddr, ETH_ALEN);
2147 addr.sa_family = slave_dev->type;
2148 dev_set_mac_address(slave_dev, &addr);
2149 }
2150
2151 slave_dev->priv_flags &= ~(IFF_MASTER_8023AD | IFF_MASTER_ALB |
2152 IFF_SLAVE_INACTIVE);
2153
2154 kfree(slave);
2155
2156 /* re-acquire the lock before getting the next slave */
2157 write_lock_bh(&bond->lock);
2158 }
2159
2160 /* zero the mac address of the master so it will be
2161 * set by the application to the mac address of the
2162 * first slave
2163 */
2164 memset(bond_dev->dev_addr, 0, bond_dev->addr_len);
2165
2166 if (!bond->vlgrp) {
2167 bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
2168 } else {
2169 pr_warning("%s: Warning: clearing HW address of %s while it still has VLANs.\n",
2170 bond_dev->name, bond_dev->name);
2171 pr_warning("%s: When re-adding slaves, make sure the bond's HW address matches its VLANs'.\n",
2172 bond_dev->name);
2173 }
2174
2175 pr_info("%s: released all slaves\n", bond_dev->name);
2176
2177out:
2178 write_unlock_bh(&bond->lock);
2179 return 0;
2180}
2181
2182/*
2183 * This function changes the active slave to slave <slave_dev>.
2184 * It returns -EINVAL in the following cases.
2185 * - <slave_dev> is not found in the list.
2186 * - There is not active slave now.
2187 * - <slave_dev> is already active.
2188 * - The link state of <slave_dev> is not BOND_LINK_UP.
2189 * - <slave_dev> is not running.
2190 * In these cases, this function does nothing.
2191 * In the other cases, current_slave pointer is changed and 0 is returned.
2192 */
2193static int bond_ioctl_change_active(struct net_device *bond_dev, struct net_device *slave_dev)
2194{
2195 struct bonding *bond = netdev_priv(bond_dev);
2196 struct slave *old_active = NULL;
2197 struct slave *new_active = NULL;
2198 int res = 0;
2199
2200 if (!USES_PRIMARY(bond->params.mode))
2201 return -EINVAL;
2202
2203 /* Verify that master_dev is indeed the master of slave_dev */
2204 if (!(slave_dev->flags & IFF_SLAVE) || (slave_dev->master != bond_dev))
2205 return -EINVAL;
2206
2207 read_lock(&bond->lock);
2208
2209 read_lock(&bond->curr_slave_lock);
2210 old_active = bond->curr_active_slave;
2211 read_unlock(&bond->curr_slave_lock);
2212
2213 new_active = bond_get_slave_by_dev(bond, slave_dev);
2214
2215 /*
2216 * Changing to the current active: do nothing; return success.
2217 */
2218 if (new_active && (new_active == old_active)) {
2219 read_unlock(&bond->lock);
2220 return 0;
2221 }
2222
2223 if ((new_active) &&
2224 (old_active) &&
2225 (new_active->link == BOND_LINK_UP) &&
2226 IS_UP(new_active->dev)) {
2227 block_netpoll_tx();
2228 write_lock_bh(&bond->curr_slave_lock);
2229 bond_change_active_slave(bond, new_active);
2230 write_unlock_bh(&bond->curr_slave_lock);
2231 unblock_netpoll_tx();
2232 } else
2233 res = -EINVAL;
2234
2235 read_unlock(&bond->lock);
2236
2237 return res;
2238}
2239
2240static int bond_info_query(struct net_device *bond_dev, struct ifbond *info)
2241{
2242 struct bonding *bond = netdev_priv(bond_dev);
2243
2244 info->bond_mode = bond->params.mode;
2245 info->miimon = bond->params.miimon;
2246
2247 read_lock(&bond->lock);
2248 info->num_slaves = bond->slave_cnt;
2249 read_unlock(&bond->lock);
2250
2251 return 0;
2252}
2253
2254static int bond_slave_info_query(struct net_device *bond_dev, struct ifslave *info)
2255{
2256 struct bonding *bond = netdev_priv(bond_dev);
2257 struct slave *slave;
2258 int i, res = -ENODEV;
2259
2260 read_lock(&bond->lock);
2261
2262 bond_for_each_slave(bond, slave, i) {
2263 if (i == (int)info->slave_id) {
2264 res = 0;
2265 strcpy(info->slave_name, slave->dev->name);
2266 info->link = slave->link;
2267 info->state = slave->state;
2268 info->link_failure_count = slave->link_failure_count;
2269 break;
2270 }
2271 }
2272
2273 read_unlock(&bond->lock);
2274
2275 return res;
2276}
2277
2278/*-------------------------------- Monitoring -------------------------------*/
2279
2280
2281static int bond_miimon_inspect(struct bonding *bond)
2282{
2283 struct slave *slave;
2284 int i, link_state, commit = 0;
2285 bool ignore_updelay;
2286
2287 ignore_updelay = !bond->curr_active_slave ? true : false;
2288
2289 bond_for_each_slave(bond, slave, i) {
2290 slave->new_link = BOND_LINK_NOCHANGE;
2291
2292 link_state = bond_check_dev_link(bond, slave->dev, 0);
2293
2294 switch (slave->link) {
2295 case BOND_LINK_UP:
2296 if (link_state)
2297 continue;
2298
2299 slave->link = BOND_LINK_FAIL;
2300 slave->delay = bond->params.downdelay;
2301 if (slave->delay) {
2302 pr_info("%s: link status down for %sinterface %s, disabling it in %d ms.\n",
2303 bond->dev->name,
2304 (bond->params.mode ==
2305 BOND_MODE_ACTIVEBACKUP) ?
2306 ((slave->state == BOND_STATE_ACTIVE) ?
2307 "active " : "backup ") : "",
2308 slave->dev->name,
2309 bond->params.downdelay * bond->params.miimon);
2310 }
2311 /*FALLTHRU*/
2312 case BOND_LINK_FAIL:
2313 if (link_state) {
2314 /*
2315 * recovered before downdelay expired
2316 */
2317 slave->link = BOND_LINK_UP;
2318 slave->jiffies = jiffies;
2319 pr_info("%s: link status up again after %d ms for interface %s.\n",
2320 bond->dev->name,
2321 (bond->params.downdelay - slave->delay) *
2322 bond->params.miimon,
2323 slave->dev->name);
2324 continue;
2325 }
2326
2327 if (slave->delay <= 0) {
2328 slave->new_link = BOND_LINK_DOWN;
2329 commit++;
2330 continue;
2331 }
2332
2333 slave->delay--;
2334 break;
2335
2336 case BOND_LINK_DOWN:
2337 if (!link_state)
2338 continue;
2339
2340 slave->link = BOND_LINK_BACK;
2341 slave->delay = bond->params.updelay;
2342
2343 if (slave->delay) {
2344 pr_info("%s: link status up for interface %s, enabling it in %d ms.\n",
2345 bond->dev->name, slave->dev->name,
2346 ignore_updelay ? 0 :
2347 bond->params.updelay *
2348 bond->params.miimon);
2349 }
2350 /*FALLTHRU*/
2351 case BOND_LINK_BACK:
2352 if (!link_state) {
2353 slave->link = BOND_LINK_DOWN;
2354 pr_info("%s: link status down again after %d ms for interface %s.\n",
2355 bond->dev->name,
2356 (bond->params.updelay - slave->delay) *
2357 bond->params.miimon,
2358 slave->dev->name);
2359
2360 continue;
2361 }
2362
2363 if (ignore_updelay)
2364 slave->delay = 0;
2365
2366 if (slave->delay <= 0) {
2367 slave->new_link = BOND_LINK_UP;
2368 commit++;
2369 ignore_updelay = false;
2370 continue;
2371 }
2372
2373 slave->delay--;
2374 break;
2375 }
2376 }
2377
2378 return commit;
2379}
2380
2381static void bond_miimon_commit(struct bonding *bond)
2382{
2383 struct slave *slave;
2384 int i;
2385
2386 bond_for_each_slave(bond, slave, i) {
2387 switch (slave->new_link) {
2388 case BOND_LINK_NOCHANGE:
2389 continue;
2390
2391 case BOND_LINK_UP:
2392 slave->link = BOND_LINK_UP;
2393 slave->jiffies = jiffies;
2394
2395 if (bond->params.mode == BOND_MODE_8023AD) {
2396 /* prevent it from being the active one */
2397 slave->state = BOND_STATE_BACKUP;
2398 } else if (bond->params.mode != BOND_MODE_ACTIVEBACKUP) {
2399 /* make it immediately active */
2400 slave->state = BOND_STATE_ACTIVE;
2401 } else if (slave != bond->primary_slave) {
2402 /* prevent it from being the active one */
2403 slave->state = BOND_STATE_BACKUP;
2404 }
2405
2406 bond_update_speed_duplex(slave);
2407
2408 pr_info("%s: link status definitely up for interface %s, %d Mbps %s duplex.\n",
2409 bond->dev->name, slave->dev->name,
2410 slave->speed, slave->duplex ? "full" : "half");
2411
2412 /* notify ad that the link status has changed */
2413 if (bond->params.mode == BOND_MODE_8023AD)
2414 bond_3ad_handle_link_change(slave, BOND_LINK_UP);
2415
2416 if (bond_is_lb(bond))
2417 bond_alb_handle_link_change(bond, slave,
2418 BOND_LINK_UP);
2419
2420 if (!bond->curr_active_slave ||
2421 (slave == bond->primary_slave))
2422 goto do_failover;
2423
2424 continue;
2425
2426 case BOND_LINK_DOWN:
2427 if (slave->link_failure_count < UINT_MAX)
2428 slave->link_failure_count++;
2429
2430 slave->link = BOND_LINK_DOWN;
2431
2432 if (bond->params.mode == BOND_MODE_ACTIVEBACKUP ||
2433 bond->params.mode == BOND_MODE_8023AD)
2434 bond_set_slave_inactive_flags(slave);
2435
2436 pr_info("%s: link status definitely down for interface %s, disabling it\n",
2437 bond->dev->name, slave->dev->name);
2438
2439 if (bond->params.mode == BOND_MODE_8023AD)
2440 bond_3ad_handle_link_change(slave,
2441 BOND_LINK_DOWN);
2442
2443 if (bond_is_lb(bond))
2444 bond_alb_handle_link_change(bond, slave,
2445 BOND_LINK_DOWN);
2446
2447 if (slave == bond->curr_active_slave)
2448 goto do_failover;
2449
2450 continue;
2451
2452 default:
2453 pr_err("%s: invalid new link %d on slave %s\n",
2454 bond->dev->name, slave->new_link,
2455 slave->dev->name);
2456 slave->new_link = BOND_LINK_NOCHANGE;
2457
2458 continue;
2459 }
2460
2461do_failover:
2462 ASSERT_RTNL();
2463 block_netpoll_tx();
2464 write_lock_bh(&bond->curr_slave_lock);
2465 bond_select_active_slave(bond);
2466 write_unlock_bh(&bond->curr_slave_lock);
2467 unblock_netpoll_tx();
2468 }
2469
2470 bond_set_carrier(bond);
2471}
2472
2473/*
2474 * bond_mii_monitor
2475 *
2476 * Really a wrapper that splits the mii monitor into two phases: an
2477 * inspection, then (if inspection indicates something needs to be done)
2478 * an acquisition of appropriate locks followed by a commit phase to
2479 * implement whatever link state changes are indicated.
2480 */
2481void bond_mii_monitor(struct work_struct *work)
2482{
2483 struct bonding *bond = container_of(work, struct bonding,
2484 mii_work.work);
2485
2486 read_lock(&bond->lock);
2487 if (bond->kill_timers)
2488 goto out;
2489
2490 if (bond->slave_cnt == 0)
2491 goto re_arm;
2492
2493 if (bond->send_grat_arp) {
2494 read_lock(&bond->curr_slave_lock);
2495 bond_send_gratuitous_arp(bond);
2496 read_unlock(&bond->curr_slave_lock);
2497 }
2498
2499 if (bond->send_unsol_na) {
2500 read_lock(&bond->curr_slave_lock);
2501 bond_send_unsolicited_na(bond);
2502 read_unlock(&bond->curr_slave_lock);
2503 }
2504
2505 if (bond_miimon_inspect(bond)) {
2506 read_unlock(&bond->lock);
2507 rtnl_lock();
2508 read_lock(&bond->lock);
2509
2510 bond_miimon_commit(bond);
2511
2512 read_unlock(&bond->lock);
2513 rtnl_unlock(); /* might sleep, hold no other locks */
2514 read_lock(&bond->lock);
2515 }
2516
2517re_arm:
2518 if (bond->params.miimon)
2519 queue_delayed_work(bond->wq, &bond->mii_work,
2520 msecs_to_jiffies(bond->params.miimon));
2521out:
2522 read_unlock(&bond->lock);
2523}
2524
2525static __be32 bond_glean_dev_ip(struct net_device *dev)
2526{
2527 struct in_device *idev;
2528 struct in_ifaddr *ifa;
2529 __be32 addr = 0;
2530
2531 if (!dev)
2532 return 0;
2533
2534 rcu_read_lock();
2535 idev = __in_dev_get_rcu(dev);
2536 if (!idev)
2537 goto out;
2538
2539 ifa = idev->ifa_list;
2540 if (!ifa)
2541 goto out;
2542
2543 addr = ifa->ifa_local;
2544out:
2545 rcu_read_unlock();
2546 return addr;
2547}
2548
2549static int bond_has_this_ip(struct bonding *bond, __be32 ip)
2550{
2551 struct vlan_entry *vlan;
2552
2553 if (ip == bond->master_ip)
2554 return 1;
2555
2556 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
2557 if (ip == vlan->vlan_ip)
2558 return 1;
2559 }
2560
2561 return 0;
2562}
2563
2564/*
2565 * We go to the (large) trouble of VLAN tagging ARP frames because
2566 * switches in VLAN mode (especially if ports are configured as
2567 * "native" to a VLAN) might not pass non-tagged frames.
2568 */
2569static void bond_arp_send(struct net_device *slave_dev, int arp_op, __be32 dest_ip, __be32 src_ip, unsigned short vlan_id)
2570{
2571 struct sk_buff *skb;
2572
2573 pr_debug("arp %d on slave %s: dst %x src %x vid %d\n", arp_op,
2574 slave_dev->name, dest_ip, src_ip, vlan_id);
2575
2576 skb = arp_create(arp_op, ETH_P_ARP, dest_ip, slave_dev, src_ip,
2577 NULL, slave_dev->dev_addr, NULL);
2578
2579 if (!skb) {
2580 pr_err("ARP packet allocation failed\n");
2581 return;
2582 }
2583 if (vlan_id) {
2584 skb = vlan_put_tag(skb, vlan_id);
2585 if (!skb) {
2586 pr_err("failed to insert VLAN tag\n");
2587 return;
2588 }
2589 }
2590 arp_xmit(skb);
2591}
2592
2593
2594static void bond_arp_send_all(struct bonding *bond, struct slave *slave)
2595{
2596 int i, vlan_id, rv;
2597 __be32 *targets = bond->params.arp_targets;
2598 struct vlan_entry *vlan;
2599 struct net_device *vlan_dev;
2600 struct flowi fl;
2601 struct rtable *rt;
2602
2603 for (i = 0; (i < BOND_MAX_ARP_TARGETS); i++) {
2604 if (!targets[i])
2605 break;
2606 pr_debug("basa: target %x\n", targets[i]);
2607 if (!bond->vlgrp) {
2608 pr_debug("basa: empty vlan: arp_send\n");
2609 bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2610 bond->master_ip, 0);
2611 continue;
2612 }
2613
2614 /*
2615 * If VLANs are configured, we do a route lookup to
2616 * determine which VLAN interface would be used, so we
2617 * can tag the ARP with the proper VLAN tag.
2618 */
2619 memset(&fl, 0, sizeof(fl));
2620 fl.fl4_dst = targets[i];
2621 fl.fl4_tos = RTO_ONLINK;
2622
2623 rv = ip_route_output_key(dev_net(bond->dev), &rt, &fl);
2624 if (rv) {
2625 if (net_ratelimit()) {
2626 pr_warning("%s: no route to arp_ip_target %pI4\n",
2627 bond->dev->name, &fl.fl4_dst);
2628 }
2629 continue;
2630 }
2631
2632 /*
2633 * This target is not on a VLAN
2634 */
2635 if (rt->dst.dev == bond->dev) {
2636 ip_rt_put(rt);
2637 pr_debug("basa: rtdev == bond->dev: arp_send\n");
2638 bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2639 bond->master_ip, 0);
2640 continue;
2641 }
2642
2643 vlan_id = 0;
2644 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
2645 vlan_dev = vlan_group_get_device(bond->vlgrp, vlan->vlan_id);
2646 if (vlan_dev == rt->dst.dev) {
2647 vlan_id = vlan->vlan_id;
2648 pr_debug("basa: vlan match on %s %d\n",
2649 vlan_dev->name, vlan_id);
2650 break;
2651 }
2652 }
2653
2654 if (vlan_id) {
2655 ip_rt_put(rt);
2656 bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2657 vlan->vlan_ip, vlan_id);
2658 continue;
2659 }
2660
2661 if (net_ratelimit()) {
2662 pr_warning("%s: no path to arp_ip_target %pI4 via rt.dev %s\n",
2663 bond->dev->name, &fl.fl4_dst,
2664 rt->dst.dev ? rt->dst.dev->name : "NULL");
2665 }
2666 ip_rt_put(rt);
2667 }
2668}
2669
2670/*
2671 * Kick out a gratuitous ARP for an IP on the bonding master plus one
2672 * for each VLAN above us.
2673 *
2674 * Caller must hold curr_slave_lock for read or better
2675 */
2676static void bond_send_gratuitous_arp(struct bonding *bond)
2677{
2678 struct slave *slave = bond->curr_active_slave;
2679 struct vlan_entry *vlan;
2680 struct net_device *vlan_dev;
2681
2682 pr_debug("bond_send_grat_arp: bond %s slave %s\n",
2683 bond->dev->name, slave ? slave->dev->name : "NULL");
2684
2685 if (!slave || !bond->send_grat_arp ||
2686 test_bit(__LINK_STATE_LINKWATCH_PENDING, &slave->dev->state))
2687 return;
2688
2689 bond->send_grat_arp--;
2690
2691 if (bond->master_ip) {
2692 bond_arp_send(slave->dev, ARPOP_REPLY, bond->master_ip,
2693 bond->master_ip, 0);
2694 }
2695
2696 if (!bond->vlgrp)
2697 return;
2698
2699 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
2700 vlan_dev = vlan_group_get_device(bond->vlgrp, vlan->vlan_id);
2701 if (vlan->vlan_ip) {
2702 bond_arp_send(slave->dev, ARPOP_REPLY, vlan->vlan_ip,
2703 vlan->vlan_ip, vlan->vlan_id);
2704 }
2705 }
2706}
2707
2708static void bond_validate_arp(struct bonding *bond, struct slave *slave, __be32 sip, __be32 tip)
2709{
2710 int i;
2711 __be32 *targets = bond->params.arp_targets;
2712
2713 for (i = 0; (i < BOND_MAX_ARP_TARGETS) && targets[i]; i++) {
2714 pr_debug("bva: sip %pI4 tip %pI4 t[%d] %pI4 bhti(tip) %d\n",
2715 &sip, &tip, i, &targets[i],
2716 bond_has_this_ip(bond, tip));
2717 if (sip == targets[i]) {
2718 if (bond_has_this_ip(bond, tip))
2719 slave->last_arp_rx = jiffies;
2720 return;
2721 }
2722 }
2723}
2724
2725static int bond_arp_rcv(struct sk_buff *skb, struct net_device *dev, struct packet_type *pt, struct net_device *orig_dev)
2726{
2727 struct arphdr *arp;
2728 struct slave *slave;
2729 struct bonding *bond;
2730 unsigned char *arp_ptr;
2731 __be32 sip, tip;
2732
2733 if (dev->priv_flags & IFF_802_1Q_VLAN) {
2734 /*
2735 * When using VLANS and bonding, dev and oriv_dev may be
2736 * incorrect if the physical interface supports VLAN
2737 * acceleration. With this change ARP validation now
2738 * works for hosts only reachable on the VLAN interface.
2739 */
2740 dev = vlan_dev_real_dev(dev);
2741 orig_dev = dev_get_by_index_rcu(dev_net(skb->dev),skb->skb_iif);
2742 }
2743
2744 if (!(dev->priv_flags & IFF_BONDING) || !(dev->flags & IFF_MASTER))
2745 goto out;
2746
2747 bond = netdev_priv(dev);
2748 read_lock(&bond->lock);
2749
2750 pr_debug("bond_arp_rcv: bond %s skb->dev %s orig_dev %s\n",
2751 bond->dev->name, skb->dev ? skb->dev->name : "NULL",
2752 orig_dev ? orig_dev->name : "NULL");
2753
2754 slave = bond_get_slave_by_dev(bond, orig_dev);
2755 if (!slave || !slave_do_arp_validate(bond, slave))
2756 goto out_unlock;
2757
2758 if (!pskb_may_pull(skb, arp_hdr_len(dev)))
2759 goto out_unlock;
2760
2761 arp = arp_hdr(skb);
2762 if (arp->ar_hln != dev->addr_len ||
2763 skb->pkt_type == PACKET_OTHERHOST ||
2764 skb->pkt_type == PACKET_LOOPBACK ||
2765 arp->ar_hrd != htons(ARPHRD_ETHER) ||
2766 arp->ar_pro != htons(ETH_P_IP) ||
2767 arp->ar_pln != 4)
2768 goto out_unlock;
2769
2770 arp_ptr = (unsigned char *)(arp + 1);
2771 arp_ptr += dev->addr_len;
2772 memcpy(&sip, arp_ptr, 4);
2773 arp_ptr += 4 + dev->addr_len;
2774 memcpy(&tip, arp_ptr, 4);
2775
2776 pr_debug("bond_arp_rcv: %s %s/%d av %d sv %d sip %pI4 tip %pI4\n",
2777 bond->dev->name, slave->dev->name, slave->state,
2778 bond->params.arp_validate, slave_do_arp_validate(bond, slave),
2779 &sip, &tip);
2780
2781 /*
2782 * Backup slaves won't see the ARP reply, but do come through
2783 * here for each ARP probe (so we swap the sip/tip to validate
2784 * the probe). In a "redundant switch, common router" type of
2785 * configuration, the ARP probe will (hopefully) travel from
2786 * the active, through one switch, the router, then the other
2787 * switch before reaching the backup.
2788 */
2789 if (slave->state == BOND_STATE_ACTIVE)
2790 bond_validate_arp(bond, slave, sip, tip);
2791 else
2792 bond_validate_arp(bond, slave, tip, sip);
2793
2794out_unlock:
2795 read_unlock(&bond->lock);
2796out:
2797 dev_kfree_skb(skb);
2798 return NET_RX_SUCCESS;
2799}
2800
2801/*
2802 * this function is called regularly to monitor each slave's link
2803 * ensuring that traffic is being sent and received when arp monitoring
2804 * is used in load-balancing mode. if the adapter has been dormant, then an
2805 * arp is transmitted to generate traffic. see activebackup_arp_monitor for
2806 * arp monitoring in active backup mode.
2807 */
2808void bond_loadbalance_arp_mon(struct work_struct *work)
2809{
2810 struct bonding *bond = container_of(work, struct bonding,
2811 arp_work.work);
2812 struct slave *slave, *oldcurrent;
2813 int do_failover = 0;
2814 int delta_in_ticks;
2815 int i;
2816
2817 read_lock(&bond->lock);
2818
2819 delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
2820
2821 if (bond->kill_timers)
2822 goto out;
2823
2824 if (bond->slave_cnt == 0)
2825 goto re_arm;
2826
2827 read_lock(&bond->curr_slave_lock);
2828 oldcurrent = bond->curr_active_slave;
2829 read_unlock(&bond->curr_slave_lock);
2830
2831 /* see if any of the previous devices are up now (i.e. they have
2832 * xmt and rcv traffic). the curr_active_slave does not come into
2833 * the picture unless it is null. also, slave->jiffies is not needed
2834 * here because we send an arp on each slave and give a slave as
2835 * long as it needs to get the tx/rx within the delta.
2836 * TODO: what about up/down delay in arp mode? it wasn't here before
2837 * so it can wait
2838 */
2839 bond_for_each_slave(bond, slave, i) {
2840 unsigned long trans_start = dev_trans_start(slave->dev);
2841
2842 if (slave->link != BOND_LINK_UP) {
2843 if (time_in_range(jiffies,
2844 trans_start - delta_in_ticks,
2845 trans_start + delta_in_ticks) &&
2846 time_in_range(jiffies,
2847 slave->dev->last_rx - delta_in_ticks,
2848 slave->dev->last_rx + delta_in_ticks)) {
2849
2850 slave->link = BOND_LINK_UP;
2851 slave->state = BOND_STATE_ACTIVE;
2852
2853 /* primary_slave has no meaning in round-robin
2854 * mode. the window of a slave being up and
2855 * curr_active_slave being null after enslaving
2856 * is closed.
2857 */
2858 if (!oldcurrent) {
2859 pr_info("%s: link status definitely up for interface %s, ",
2860 bond->dev->name,
2861 slave->dev->name);
2862 do_failover = 1;
2863 } else {
2864 pr_info("%s: interface %s is now up\n",
2865 bond->dev->name,
2866 slave->dev->name);
2867 }
2868 }
2869 } else {
2870 /* slave->link == BOND_LINK_UP */
2871
2872 /* not all switches will respond to an arp request
2873 * when the source ip is 0, so don't take the link down
2874 * if we don't know our ip yet
2875 */
2876 if (!time_in_range(jiffies,
2877 trans_start - delta_in_ticks,
2878 trans_start + 2 * delta_in_ticks) ||
2879 !time_in_range(jiffies,
2880 slave->dev->last_rx - delta_in_ticks,
2881 slave->dev->last_rx + 2 * delta_in_ticks)) {
2882
2883 slave->link = BOND_LINK_DOWN;
2884 slave->state = BOND_STATE_BACKUP;
2885
2886 if (slave->link_failure_count < UINT_MAX)
2887 slave->link_failure_count++;
2888
2889 pr_info("%s: interface %s is now down.\n",
2890 bond->dev->name,
2891 slave->dev->name);
2892
2893 if (slave == oldcurrent)
2894 do_failover = 1;
2895 }
2896 }
2897
2898 /* note: if switch is in round-robin mode, all links
2899 * must tx arp to ensure all links rx an arp - otherwise
2900 * links may oscillate or not come up at all; if switch is
2901 * in something like xor mode, there is nothing we can
2902 * do - all replies will be rx'ed on same link causing slaves
2903 * to be unstable during low/no traffic periods
2904 */
2905 if (IS_UP(slave->dev))
2906 bond_arp_send_all(bond, slave);
2907 }
2908
2909 if (do_failover) {
2910 block_netpoll_tx();
2911 write_lock_bh(&bond->curr_slave_lock);
2912
2913 bond_select_active_slave(bond);
2914
2915 write_unlock_bh(&bond->curr_slave_lock);
2916 unblock_netpoll_tx();
2917 }
2918
2919re_arm:
2920 if (bond->params.arp_interval)
2921 queue_delayed_work(bond->wq, &bond->arp_work, delta_in_ticks);
2922out:
2923 read_unlock(&bond->lock);
2924}
2925
2926/*
2927 * Called to inspect slaves for active-backup mode ARP monitor link state
2928 * changes. Sets new_link in slaves to specify what action should take
2929 * place for the slave. Returns 0 if no changes are found, >0 if changes
2930 * to link states must be committed.
2931 *
2932 * Called with bond->lock held for read.
2933 */
2934static int bond_ab_arp_inspect(struct bonding *bond, int delta_in_ticks)
2935{
2936 struct slave *slave;
2937 int i, commit = 0;
2938 unsigned long trans_start;
2939
2940 bond_for_each_slave(bond, slave, i) {
2941 slave->new_link = BOND_LINK_NOCHANGE;
2942
2943 if (slave->link != BOND_LINK_UP) {
2944 if (time_in_range(jiffies,
2945 slave_last_rx(bond, slave) - delta_in_ticks,
2946 slave_last_rx(bond, slave) + delta_in_ticks)) {
2947
2948 slave->new_link = BOND_LINK_UP;
2949 commit++;
2950 }
2951
2952 continue;
2953 }
2954
2955 /*
2956 * Give slaves 2*delta after being enslaved or made
2957 * active. This avoids bouncing, as the last receive
2958 * times need a full ARP monitor cycle to be updated.
2959 */
2960 if (time_in_range(jiffies,
2961 slave->jiffies - delta_in_ticks,
2962 slave->jiffies + 2 * delta_in_ticks))
2963 continue;
2964
2965 /*
2966 * Backup slave is down if:
2967 * - No current_arp_slave AND
2968 * - more than 3*delta since last receive AND
2969 * - the bond has an IP address
2970 *
2971 * Note: a non-null current_arp_slave indicates
2972 * the curr_active_slave went down and we are
2973 * searching for a new one; under this condition
2974 * we only take the curr_active_slave down - this
2975 * gives each slave a chance to tx/rx traffic
2976 * before being taken out
2977 */
2978 if (slave->state == BOND_STATE_BACKUP &&
2979 !bond->current_arp_slave &&
2980 !time_in_range(jiffies,
2981 slave_last_rx(bond, slave) - delta_in_ticks,
2982 slave_last_rx(bond, slave) + 3 * delta_in_ticks)) {
2983
2984 slave->new_link = BOND_LINK_DOWN;
2985 commit++;
2986 }
2987
2988 /*
2989 * Active slave is down if:
2990 * - more than 2*delta since transmitting OR
2991 * - (more than 2*delta since receive AND
2992 * the bond has an IP address)
2993 */
2994 trans_start = dev_trans_start(slave->dev);
2995 if ((slave->state == BOND_STATE_ACTIVE) &&
2996 (!time_in_range(jiffies,
2997 trans_start - delta_in_ticks,
2998 trans_start + 2 * delta_in_ticks) ||
2999 !time_in_range(jiffies,
3000 slave_last_rx(bond, slave) - delta_in_ticks,
3001 slave_last_rx(bond, slave) + 2 * delta_in_ticks))) {
3002
3003 slave->new_link = BOND_LINK_DOWN;
3004 commit++;
3005 }
3006 }
3007
3008 return commit;
3009}
3010
3011/*
3012 * Called to commit link state changes noted by inspection step of
3013 * active-backup mode ARP monitor.
3014 *
3015 * Called with RTNL and bond->lock for read.
3016 */
3017static void bond_ab_arp_commit(struct bonding *bond, int delta_in_ticks)
3018{
3019 struct slave *slave;
3020 int i;
3021 unsigned long trans_start;
3022
3023 bond_for_each_slave(bond, slave, i) {
3024 switch (slave->new_link) {
3025 case BOND_LINK_NOCHANGE:
3026 continue;
3027
3028 case BOND_LINK_UP:
3029 trans_start = dev_trans_start(slave->dev);
3030 if ((!bond->curr_active_slave &&
3031 time_in_range(jiffies,
3032 trans_start - delta_in_ticks,
3033 trans_start + delta_in_ticks)) ||
3034 bond->curr_active_slave != slave) {
3035 slave->link = BOND_LINK_UP;
3036 bond->current_arp_slave = NULL;
3037
3038 pr_info("%s: link status definitely up for interface %s.\n",
3039 bond->dev->name, slave->dev->name);
3040
3041 if (!bond->curr_active_slave ||
3042 (slave == bond->primary_slave))
3043 goto do_failover;
3044
3045 }
3046
3047 continue;
3048
3049 case BOND_LINK_DOWN:
3050 if (slave->link_failure_count < UINT_MAX)
3051 slave->link_failure_count++;
3052
3053 slave->link = BOND_LINK_DOWN;
3054 bond_set_slave_inactive_flags(slave);
3055
3056 pr_info("%s: link status definitely down for interface %s, disabling it\n",
3057 bond->dev->name, slave->dev->name);
3058
3059 if (slave == bond->curr_active_slave) {
3060 bond->current_arp_slave = NULL;
3061 goto do_failover;
3062 }
3063
3064 continue;
3065
3066 default:
3067 pr_err("%s: impossible: new_link %d on slave %s\n",
3068 bond->dev->name, slave->new_link,
3069 slave->dev->name);
3070 continue;
3071 }
3072
3073do_failover:
3074 ASSERT_RTNL();
3075 block_netpoll_tx();
3076 write_lock_bh(&bond->curr_slave_lock);
3077 bond_select_active_slave(bond);
3078 write_unlock_bh(&bond->curr_slave_lock);
3079 unblock_netpoll_tx();
3080 }
3081
3082 bond_set_carrier(bond);
3083}
3084
3085/*
3086 * Send ARP probes for active-backup mode ARP monitor.
3087 *
3088 * Called with bond->lock held for read.
3089 */
3090static void bond_ab_arp_probe(struct bonding *bond)
3091{
3092 struct slave *slave;
3093 int i;
3094
3095 read_lock(&bond->curr_slave_lock);
3096
3097 if (bond->current_arp_slave && bond->curr_active_slave)
3098 pr_info("PROBE: c_arp %s && cas %s BAD\n",
3099 bond->current_arp_slave->dev->name,
3100 bond->curr_active_slave->dev->name);
3101
3102 if (bond->curr_active_slave) {
3103 bond_arp_send_all(bond, bond->curr_active_slave);
3104 read_unlock(&bond->curr_slave_lock);
3105 return;
3106 }
3107
3108 read_unlock(&bond->curr_slave_lock);
3109
3110 /* if we don't have a curr_active_slave, search for the next available
3111 * backup slave from the current_arp_slave and make it the candidate
3112 * for becoming the curr_active_slave
3113 */
3114
3115 if (!bond->current_arp_slave) {
3116 bond->current_arp_slave = bond->first_slave;
3117 if (!bond->current_arp_slave)
3118 return;
3119 }
3120
3121 bond_set_slave_inactive_flags(bond->current_arp_slave);
3122
3123 /* search for next candidate */
3124 bond_for_each_slave_from(bond, slave, i, bond->current_arp_slave->next) {
3125 if (IS_UP(slave->dev)) {
3126 slave->link = BOND_LINK_BACK;
3127 bond_set_slave_active_flags(slave);
3128 bond_arp_send_all(bond, slave);
3129 slave->jiffies = jiffies;
3130 bond->current_arp_slave = slave;
3131 break;
3132 }
3133
3134 /* if the link state is up at this point, we
3135 * mark it down - this can happen if we have
3136 * simultaneous link failures and
3137 * reselect_active_interface doesn't make this
3138 * one the current slave so it is still marked
3139 * up when it is actually down
3140 */
3141 if (slave->link == BOND_LINK_UP) {
3142 slave->link = BOND_LINK_DOWN;
3143 if (slave->link_failure_count < UINT_MAX)
3144 slave->link_failure_count++;
3145
3146 bond_set_slave_inactive_flags(slave);
3147
3148 pr_info("%s: backup interface %s is now down.\n",
3149 bond->dev->name, slave->dev->name);
3150 }
3151 }
3152}
3153
3154void bond_activebackup_arp_mon(struct work_struct *work)
3155{
3156 struct bonding *bond = container_of(work, struct bonding,
3157 arp_work.work);
3158 int delta_in_ticks;
3159
3160 read_lock(&bond->lock);
3161
3162 if (bond->kill_timers)
3163 goto out;
3164
3165 delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
3166
3167 if (bond->slave_cnt == 0)
3168 goto re_arm;
3169
3170 if (bond->send_grat_arp) {
3171 read_lock(&bond->curr_slave_lock);
3172 bond_send_gratuitous_arp(bond);
3173 read_unlock(&bond->curr_slave_lock);
3174 }
3175
3176 if (bond->send_unsol_na) {
3177 read_lock(&bond->curr_slave_lock);
3178 bond_send_unsolicited_na(bond);
3179 read_unlock(&bond->curr_slave_lock);
3180 }
3181
3182 if (bond_ab_arp_inspect(bond, delta_in_ticks)) {
3183 read_unlock(&bond->lock);
3184 rtnl_lock();
3185 read_lock(&bond->lock);
3186
3187 bond_ab_arp_commit(bond, delta_in_ticks);
3188
3189 read_unlock(&bond->lock);
3190 rtnl_unlock();
3191 read_lock(&bond->lock);
3192 }
3193
3194 bond_ab_arp_probe(bond);
3195
3196re_arm:
3197 if (bond->params.arp_interval)
3198 queue_delayed_work(bond->wq, &bond->arp_work, delta_in_ticks);
3199out:
3200 read_unlock(&bond->lock);
3201}
3202
3203/*------------------------------ proc/seq_file-------------------------------*/
3204
3205#ifdef CONFIG_PROC_FS
3206
3207static void *bond_info_seq_start(struct seq_file *seq, loff_t *pos)
3208 __acquires(RCU)
3209 __acquires(&bond->lock)
3210{
3211 struct bonding *bond = seq->private;
3212 loff_t off = 0;
3213 struct slave *slave;
3214 int i;
3215
3216 /* make sure the bond won't be taken away */
3217 rcu_read_lock();
3218 read_lock(&bond->lock);
3219
3220 if (*pos == 0)
3221 return SEQ_START_TOKEN;
3222
3223 bond_for_each_slave(bond, slave, i) {
3224 if (++off == *pos)
3225 return slave;
3226 }
3227
3228 return NULL;
3229}
3230
3231static void *bond_info_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3232{
3233 struct bonding *bond = seq->private;
3234 struct slave *slave = v;
3235
3236 ++*pos;
3237 if (v == SEQ_START_TOKEN)
3238 return bond->first_slave;
3239
3240 slave = slave->next;
3241
3242 return (slave == bond->first_slave) ? NULL : slave;
3243}
3244
3245static void bond_info_seq_stop(struct seq_file *seq, void *v)
3246 __releases(&bond->lock)
3247 __releases(RCU)
3248{
3249 struct bonding *bond = seq->private;
3250
3251 read_unlock(&bond->lock);
3252 rcu_read_unlock();
3253}
3254
3255static void bond_info_show_master(struct seq_file *seq)
3256{
3257 struct bonding *bond = seq->private;
3258 struct slave *curr;
3259 int i;
3260
3261 read_lock(&bond->curr_slave_lock);
3262 curr = bond->curr_active_slave;
3263 read_unlock(&bond->curr_slave_lock);
3264
3265 seq_printf(seq, "Bonding Mode: %s",
3266 bond_mode_name(bond->params.mode));
3267
3268 if (bond->params.mode == BOND_MODE_ACTIVEBACKUP &&
3269 bond->params.fail_over_mac)
3270 seq_printf(seq, " (fail_over_mac %s)",
3271 fail_over_mac_tbl[bond->params.fail_over_mac].modename);
3272
3273 seq_printf(seq, "\n");
3274
3275 if (bond->params.mode == BOND_MODE_XOR ||
3276 bond->params.mode == BOND_MODE_8023AD) {
3277 seq_printf(seq, "Transmit Hash Policy: %s (%d)\n",
3278 xmit_hashtype_tbl[bond->params.xmit_policy].modename,
3279 bond->params.xmit_policy);
3280 }
3281
3282 if (USES_PRIMARY(bond->params.mode)) {
3283 seq_printf(seq, "Primary Slave: %s",
3284 (bond->primary_slave) ?
3285 bond->primary_slave->dev->name : "None");
3286 if (bond->primary_slave)
3287 seq_printf(seq, " (primary_reselect %s)",
3288 pri_reselect_tbl[bond->params.primary_reselect].modename);
3289
3290 seq_printf(seq, "\nCurrently Active Slave: %s\n",
3291 (curr) ? curr->dev->name : "None");
3292 }
3293
3294 seq_printf(seq, "MII Status: %s\n", netif_carrier_ok(bond->dev) ?
3295 "up" : "down");
3296 seq_printf(seq, "MII Polling Interval (ms): %d\n", bond->params.miimon);
3297 seq_printf(seq, "Up Delay (ms): %d\n",
3298 bond->params.updelay * bond->params.miimon);
3299 seq_printf(seq, "Down Delay (ms): %d\n",
3300 bond->params.downdelay * bond->params.miimon);
3301
3302
3303 /* ARP information */
3304 if (bond->params.arp_interval > 0) {
3305 int printed = 0;
3306 seq_printf(seq, "ARP Polling Interval (ms): %d\n",
3307 bond->params.arp_interval);
3308
3309 seq_printf(seq, "ARP IP target/s (n.n.n.n form):");
3310
3311 for (i = 0; (i < BOND_MAX_ARP_TARGETS); i++) {
3312 if (!bond->params.arp_targets[i])
3313 break;
3314 if (printed)
3315 seq_printf(seq, ",");
3316 seq_printf(seq, " %pI4", &bond->params.arp_targets[i]);
3317 printed = 1;
3318 }
3319 seq_printf(seq, "\n");
3320 }
3321
3322 if (bond->params.mode == BOND_MODE_8023AD) {
3323 struct ad_info ad_info;
3324
3325 seq_puts(seq, "\n802.3ad info\n");
3326 seq_printf(seq, "LACP rate: %s\n",
3327 (bond->params.lacp_fast) ? "fast" : "slow");
3328 seq_printf(seq, "Aggregator selection policy (ad_select): %s\n",
3329 ad_select_tbl[bond->params.ad_select].modename);
3330
3331 if (bond_3ad_get_active_agg_info(bond, &ad_info)) {
3332 seq_printf(seq, "bond %s has no active aggregator\n",
3333 bond->dev->name);
3334 } else {
3335 seq_printf(seq, "Active Aggregator Info:\n");
3336
3337 seq_printf(seq, "\tAggregator ID: %d\n",
3338 ad_info.aggregator_id);
3339 seq_printf(seq, "\tNumber of ports: %d\n",
3340 ad_info.ports);
3341 seq_printf(seq, "\tActor Key: %d\n",
3342 ad_info.actor_key);
3343 seq_printf(seq, "\tPartner Key: %d\n",
3344 ad_info.partner_key);
3345 seq_printf(seq, "\tPartner Mac Address: %pM\n",
3346 ad_info.partner_system);
3347 }
3348 }
3349}
3350
3351static void bond_info_show_slave(struct seq_file *seq,
3352 const struct slave *slave)
3353{
3354 struct bonding *bond = seq->private;
3355
3356 seq_printf(seq, "\nSlave Interface: %s\n", slave->dev->name);
3357 seq_printf(seq, "MII Status: %s\n",
3358 (slave->link == BOND_LINK_UP) ? "up" : "down");
3359 seq_printf(seq, "Speed: %d Mbps\n", slave->speed);
3360 seq_printf(seq, "Duplex: %s\n", slave->duplex ? "full" : "half");
3361 seq_printf(seq, "Link Failure Count: %u\n",
3362 slave->link_failure_count);
3363
3364 seq_printf(seq, "Permanent HW addr: %pM\n", slave->perm_hwaddr);
3365
3366 if (bond->params.mode == BOND_MODE_8023AD) {
3367 const struct aggregator *agg
3368 = SLAVE_AD_INFO(slave).port.aggregator;
3369
3370 if (agg)
3371 seq_printf(seq, "Aggregator ID: %d\n",
3372 agg->aggregator_identifier);
3373 else
3374 seq_puts(seq, "Aggregator ID: N/A\n");
3375 }
3376 seq_printf(seq, "Slave queue ID: %d\n", slave->queue_id);
3377}
3378
3379static int bond_info_seq_show(struct seq_file *seq, void *v)
3380{
3381 if (v == SEQ_START_TOKEN) {
3382 seq_printf(seq, "%s\n", version);
3383 bond_info_show_master(seq);
3384 } else
3385 bond_info_show_slave(seq, v);
3386
3387 return 0;
3388}
3389
3390static const struct seq_operations bond_info_seq_ops = {
3391 .start = bond_info_seq_start,
3392 .next = bond_info_seq_next,
3393 .stop = bond_info_seq_stop,
3394 .show = bond_info_seq_show,
3395};
3396
3397static int bond_info_open(struct inode *inode, struct file *file)
3398{
3399 struct seq_file *seq;
3400 struct proc_dir_entry *proc;
3401 int res;
3402
3403 res = seq_open(file, &bond_info_seq_ops);
3404 if (!res) {
3405 /* recover the pointer buried in proc_dir_entry data */
3406 seq = file->private_data;
3407 proc = PDE(inode);
3408 seq->private = proc->data;
3409 }
3410
3411 return res;
3412}
3413
3414static const struct file_operations bond_info_fops = {
3415 .owner = THIS_MODULE,
3416 .open = bond_info_open,
3417 .read = seq_read,
3418 .llseek = seq_lseek,
3419 .release = seq_release,
3420};
3421
3422static void bond_create_proc_entry(struct bonding *bond)
3423{
3424 struct net_device *bond_dev = bond->dev;
3425 struct bond_net *bn = net_generic(dev_net(bond_dev), bond_net_id);
3426
3427 if (bn->proc_dir) {
3428 bond->proc_entry = proc_create_data(bond_dev->name,
3429 S_IRUGO, bn->proc_dir,
3430 &bond_info_fops, bond);
3431 if (bond->proc_entry == NULL)
3432 pr_warning("Warning: Cannot create /proc/net/%s/%s\n",
3433 DRV_NAME, bond_dev->name);
3434 else
3435 memcpy(bond->proc_file_name, bond_dev->name, IFNAMSIZ);
3436 }
3437}
3438
3439static void bond_remove_proc_entry(struct bonding *bond)
3440{
3441 struct net_device *bond_dev = bond->dev;
3442 struct bond_net *bn = net_generic(dev_net(bond_dev), bond_net_id);
3443
3444 if (bn->proc_dir && bond->proc_entry) {
3445 remove_proc_entry(bond->proc_file_name, bn->proc_dir);
3446 memset(bond->proc_file_name, 0, IFNAMSIZ);
3447 bond->proc_entry = NULL;
3448 }
3449}
3450
3451/* Create the bonding directory under /proc/net, if doesn't exist yet.
3452 * Caller must hold rtnl_lock.
3453 */
3454static void __net_init bond_create_proc_dir(struct bond_net *bn)
3455{
3456 if (!bn->proc_dir) {
3457 bn->proc_dir = proc_mkdir(DRV_NAME, bn->net->proc_net);
3458 if (!bn->proc_dir)
3459 pr_warning("Warning: cannot create /proc/net/%s\n",
3460 DRV_NAME);
3461 }
3462}
3463
3464/* Destroy the bonding directory under /proc/net, if empty.
3465 * Caller must hold rtnl_lock.
3466 */
3467static void __net_exit bond_destroy_proc_dir(struct bond_net *bn)
3468{
3469 if (bn->proc_dir) {
3470 remove_proc_entry(DRV_NAME, bn->net->proc_net);
3471 bn->proc_dir = NULL;
3472 }
3473}
3474
3475#else /* !CONFIG_PROC_FS */
3476
3477static void bond_create_proc_entry(struct bonding *bond)
3478{
3479}
3480
3481static void bond_remove_proc_entry(struct bonding *bond)
3482{
3483}
3484
3485static inline void bond_create_proc_dir(struct bond_net *bn)
3486{
3487}
3488
3489static inline void bond_destroy_proc_dir(struct bond_net *bn)
3490{
3491}
3492
3493#endif /* CONFIG_PROC_FS */
3494
3495
3496/*-------------------------- netdev event handling --------------------------*/
3497
3498/*
3499 * Change device name
3500 */
3501static int bond_event_changename(struct bonding *bond)
3502{
3503 bond_remove_proc_entry(bond);
3504 bond_create_proc_entry(bond);
3505
3506 return NOTIFY_DONE;
3507}
3508
3509static int bond_master_netdev_event(unsigned long event,
3510 struct net_device *bond_dev)
3511{
3512 struct bonding *event_bond = netdev_priv(bond_dev);
3513
3514 switch (event) {
3515 case NETDEV_CHANGENAME:
3516 return bond_event_changename(event_bond);
3517 default:
3518 break;
3519 }
3520
3521 return NOTIFY_DONE;
3522}
3523
3524static int bond_slave_netdev_event(unsigned long event,
3525 struct net_device *slave_dev)
3526{
3527 struct net_device *bond_dev = slave_dev->master;
3528 struct bonding *bond = netdev_priv(bond_dev);
3529
3530 switch (event) {
3531 case NETDEV_UNREGISTER:
3532 if (bond_dev) {
3533 if (bond->setup_by_slave)
3534 bond_release_and_destroy(bond_dev, slave_dev);
3535 else
3536 bond_release(bond_dev, slave_dev);
3537 }
3538 break;
3539 case NETDEV_CHANGE:
3540 if (bond->params.mode == BOND_MODE_8023AD || bond_is_lb(bond)) {
3541 struct slave *slave;
3542
3543 slave = bond_get_slave_by_dev(bond, slave_dev);
3544 if (slave) {
3545 u16 old_speed = slave->speed;
3546 u16 old_duplex = slave->duplex;
3547
3548 bond_update_speed_duplex(slave);
3549
3550 if (bond_is_lb(bond))
3551 break;
3552
3553 if (old_speed != slave->speed)
3554 bond_3ad_adapter_speed_changed(slave);
3555 if (old_duplex != slave->duplex)
3556 bond_3ad_adapter_duplex_changed(slave);
3557 }
3558 }
3559
3560 break;
3561 case NETDEV_DOWN:
3562 /*
3563 * ... Or is it this?
3564 */
3565 break;
3566 case NETDEV_CHANGEMTU:
3567 /*
3568 * TODO: Should slaves be allowed to
3569 * independently alter their MTU? For
3570 * an active-backup bond, slaves need
3571 * not be the same type of device, so
3572 * MTUs may vary. For other modes,
3573 * slaves arguably should have the
3574 * same MTUs. To do this, we'd need to
3575 * take over the slave's change_mtu
3576 * function for the duration of their
3577 * servitude.
3578 */
3579 break;
3580 case NETDEV_CHANGENAME:
3581 /*
3582 * TODO: handle changing the primary's name
3583 */
3584 break;
3585 case NETDEV_FEAT_CHANGE:
3586 bond_compute_features(bond);
3587 break;
3588 default:
3589 break;
3590 }
3591
3592 return NOTIFY_DONE;
3593}
3594
3595/*
3596 * bond_netdev_event: handle netdev notifier chain events.
3597 *
3598 * This function receives events for the netdev chain. The caller (an
3599 * ioctl handler calling blocking_notifier_call_chain) holds the necessary
3600 * locks for us to safely manipulate the slave devices (RTNL lock,
3601 * dev_probe_lock).
3602 */
3603static int bond_netdev_event(struct notifier_block *this,
3604 unsigned long event, void *ptr)
3605{
3606 struct net_device *event_dev = (struct net_device *)ptr;
3607
3608 pr_debug("event_dev: %s, event: %lx\n",
3609 event_dev ? event_dev->name : "None",
3610 event);
3611
3612 if (!(event_dev->priv_flags & IFF_BONDING))
3613 return NOTIFY_DONE;
3614
3615 if (event_dev->flags & IFF_MASTER) {
3616 pr_debug("IFF_MASTER\n");
3617 return bond_master_netdev_event(event, event_dev);
3618 }
3619
3620 if (event_dev->flags & IFF_SLAVE) {
3621 pr_debug("IFF_SLAVE\n");
3622 return bond_slave_netdev_event(event, event_dev);
3623 }
3624
3625 return NOTIFY_DONE;
3626}
3627
3628/*
3629 * bond_inetaddr_event: handle inetaddr notifier chain events.
3630 *
3631 * We keep track of device IPs primarily to use as source addresses in
3632 * ARP monitor probes (rather than spewing out broadcasts all the time).
3633 *
3634 * We track one IP for the main device (if it has one), plus one per VLAN.
3635 */
3636static int bond_inetaddr_event(struct notifier_block *this, unsigned long event, void *ptr)
3637{
3638 struct in_ifaddr *ifa = ptr;
3639 struct net_device *vlan_dev, *event_dev = ifa->ifa_dev->dev;
3640 struct bond_net *bn = net_generic(dev_net(event_dev), bond_net_id);
3641 struct bonding *bond;
3642 struct vlan_entry *vlan;
3643
3644 list_for_each_entry(bond, &bn->dev_list, bond_list) {
3645 if (bond->dev == event_dev) {
3646 switch (event) {
3647 case NETDEV_UP:
3648 bond->master_ip = ifa->ifa_local;
3649 return NOTIFY_OK;
3650 case NETDEV_DOWN:
3651 bond->master_ip = bond_glean_dev_ip(bond->dev);
3652 return NOTIFY_OK;
3653 default:
3654 return NOTIFY_DONE;
3655 }
3656 }
3657
3658 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
3659 if (!bond->vlgrp)
3660 continue;
3661 vlan_dev = vlan_group_get_device(bond->vlgrp, vlan->vlan_id);
3662 if (vlan_dev == event_dev) {
3663 switch (event) {
3664 case NETDEV_UP:
3665 vlan->vlan_ip = ifa->ifa_local;
3666 return NOTIFY_OK;
3667 case NETDEV_DOWN:
3668 vlan->vlan_ip =
3669 bond_glean_dev_ip(vlan_dev);
3670 return NOTIFY_OK;
3671 default:
3672 return NOTIFY_DONE;
3673 }
3674 }
3675 }
3676 }
3677 return NOTIFY_DONE;
3678}
3679
3680static struct notifier_block bond_netdev_notifier = {
3681 .notifier_call = bond_netdev_event,
3682};
3683
3684static struct notifier_block bond_inetaddr_notifier = {
3685 .notifier_call = bond_inetaddr_event,
3686};
3687
3688/*-------------------------- Packet type handling ---------------------------*/
3689
3690/* register to receive lacpdus on a bond */
3691static void bond_register_lacpdu(struct bonding *bond)
3692{
3693 struct packet_type *pk_type = &(BOND_AD_INFO(bond).ad_pkt_type);
3694
3695 /* initialize packet type */
3696 pk_type->type = PKT_TYPE_LACPDU;
3697 pk_type->dev = bond->dev;
3698 pk_type->func = bond_3ad_lacpdu_recv;
3699
3700 dev_add_pack(pk_type);
3701}
3702
3703/* unregister to receive lacpdus on a bond */
3704static void bond_unregister_lacpdu(struct bonding *bond)
3705{
3706 dev_remove_pack(&(BOND_AD_INFO(bond).ad_pkt_type));
3707}
3708
3709void bond_register_arp(struct bonding *bond)
3710{
3711 struct packet_type *pt = &bond->arp_mon_pt;
3712
3713 if (pt->type)
3714 return;
3715
3716 pt->type = htons(ETH_P_ARP);
3717 pt->dev = bond->dev;
3718 pt->func = bond_arp_rcv;
3719 dev_add_pack(pt);
3720}
3721
3722void bond_unregister_arp(struct bonding *bond)
3723{
3724 struct packet_type *pt = &bond->arp_mon_pt;
3725
3726 dev_remove_pack(pt);
3727 pt->type = 0;
3728}
3729
3730/*---------------------------- Hashing Policies -----------------------------*/
3731
3732/*
3733 * Hash for the output device based upon layer 2 and layer 3 data. If
3734 * the packet is not IP mimic bond_xmit_hash_policy_l2()
3735 */
3736static int bond_xmit_hash_policy_l23(struct sk_buff *skb, int count)
3737{
3738 struct ethhdr *data = (struct ethhdr *)skb->data;
3739 struct iphdr *iph = ip_hdr(skb);
3740
3741 if (skb->protocol == htons(ETH_P_IP)) {
3742 return ((ntohl(iph->saddr ^ iph->daddr) & 0xffff) ^
3743 (data->h_dest[5] ^ data->h_source[5])) % count;
3744 }
3745
3746 return (data->h_dest[5] ^ data->h_source[5]) % count;
3747}
3748
3749/*
3750 * Hash for the output device based upon layer 3 and layer 4 data. If
3751 * the packet is a frag or not TCP or UDP, just use layer 3 data. If it is
3752 * altogether not IP, mimic bond_xmit_hash_policy_l2()
3753 */
3754static int bond_xmit_hash_policy_l34(struct sk_buff *skb, int count)
3755{
3756 struct ethhdr *data = (struct ethhdr *)skb->data;
3757 struct iphdr *iph = ip_hdr(skb);
3758 __be16 *layer4hdr = (__be16 *)((u32 *)iph + iph->ihl);
3759 int layer4_xor = 0;
3760
3761 if (skb->protocol == htons(ETH_P_IP)) {
3762 if (!(iph->frag_off & htons(IP_MF|IP_OFFSET)) &&
3763 (iph->protocol == IPPROTO_TCP ||
3764 iph->protocol == IPPROTO_UDP)) {
3765 layer4_xor = ntohs((*layer4hdr ^ *(layer4hdr + 1)));
3766 }
3767 return (layer4_xor ^
3768 ((ntohl(iph->saddr ^ iph->daddr)) & 0xffff)) % count;
3769
3770 }
3771
3772 return (data->h_dest[5] ^ data->h_source[5]) % count;
3773}
3774
3775/*
3776 * Hash for the output device based upon layer 2 data
3777 */
3778static int bond_xmit_hash_policy_l2(struct sk_buff *skb, int count)
3779{
3780 struct ethhdr *data = (struct ethhdr *)skb->data;
3781
3782 return (data->h_dest[5] ^ data->h_source[5]) % count;
3783}
3784
3785/*-------------------------- Device entry points ----------------------------*/
3786
3787static int bond_open(struct net_device *bond_dev)
3788{
3789 struct bonding *bond = netdev_priv(bond_dev);
3790
3791 bond->kill_timers = 0;
3792
3793 INIT_DELAYED_WORK(&bond->mcast_work, bond_resend_igmp_join_requests_delayed);
3794
3795 if (bond_is_lb(bond)) {
3796 /* bond_alb_initialize must be called before the timer
3797 * is started.
3798 */
3799 if (bond_alb_initialize(bond, (bond->params.mode == BOND_MODE_ALB))) {
3800 /* something went wrong - fail the open operation */
3801 return -ENOMEM;
3802 }
3803
3804 INIT_DELAYED_WORK(&bond->alb_work, bond_alb_monitor);
3805 queue_delayed_work(bond->wq, &bond->alb_work, 0);
3806 }
3807
3808 if (bond->params.miimon) { /* link check interval, in milliseconds. */
3809 INIT_DELAYED_WORK(&bond->mii_work, bond_mii_monitor);
3810 queue_delayed_work(bond->wq, &bond->mii_work, 0);
3811 }
3812
3813 if (bond->params.arp_interval) { /* arp interval, in milliseconds. */
3814 if (bond->params.mode == BOND_MODE_ACTIVEBACKUP)
3815 INIT_DELAYED_WORK(&bond->arp_work,
3816 bond_activebackup_arp_mon);
3817 else
3818 INIT_DELAYED_WORK(&bond->arp_work,
3819 bond_loadbalance_arp_mon);
3820
3821 queue_delayed_work(bond->wq, &bond->arp_work, 0);
3822 if (bond->params.arp_validate)
3823 bond_register_arp(bond);
3824 }
3825
3826 if (bond->params.mode == BOND_MODE_8023AD) {
3827 INIT_DELAYED_WORK(&bond->ad_work, bond_3ad_state_machine_handler);
3828 queue_delayed_work(bond->wq, &bond->ad_work, 0);
3829 /* register to receive LACPDUs */
3830 bond_register_lacpdu(bond);
3831 bond_3ad_initiate_agg_selection(bond, 1);
3832 }
3833
3834 return 0;
3835}
3836
3837static int bond_close(struct net_device *bond_dev)
3838{
3839 struct bonding *bond = netdev_priv(bond_dev);
3840
3841 if (bond->params.mode == BOND_MODE_8023AD) {
3842 /* Unregister the receive of LACPDUs */
3843 bond_unregister_lacpdu(bond);
3844 }
3845
3846 if (bond->params.arp_validate)
3847 bond_unregister_arp(bond);
3848
3849 write_lock_bh(&bond->lock);
3850
3851 bond->send_grat_arp = 0;
3852 bond->send_unsol_na = 0;
3853
3854 /* signal timers not to re-arm */
3855 bond->kill_timers = 1;
3856
3857 write_unlock_bh(&bond->lock);
3858
3859 if (bond->params.miimon) { /* link check interval, in milliseconds. */
3860 cancel_delayed_work(&bond->mii_work);
3861 }
3862
3863 if (bond->params.arp_interval) { /* arp interval, in milliseconds. */
3864 cancel_delayed_work(&bond->arp_work);
3865 }
3866
3867 switch (bond->params.mode) {
3868 case BOND_MODE_8023AD:
3869 cancel_delayed_work(&bond->ad_work);
3870 break;
3871 case BOND_MODE_TLB:
3872 case BOND_MODE_ALB:
3873 cancel_delayed_work(&bond->alb_work);
3874 break;
3875 default:
3876 break;
3877 }
3878
3879 if (delayed_work_pending(&bond->mcast_work))
3880 cancel_delayed_work(&bond->mcast_work);
3881
3882 if (bond_is_lb(bond)) {
3883 /* Must be called only after all
3884 * slaves have been released
3885 */
3886 bond_alb_deinitialize(bond);
3887 }
3888
3889 return 0;
3890}
3891
3892static struct rtnl_link_stats64 *bond_get_stats(struct net_device *bond_dev,
3893 struct rtnl_link_stats64 *stats)
3894{
3895 struct bonding *bond = netdev_priv(bond_dev);
3896 struct rtnl_link_stats64 temp;
3897 struct slave *slave;
3898 int i;
3899
3900 memset(stats, 0, sizeof(*stats));
3901
3902 read_lock_bh(&bond->lock);
3903
3904 bond_for_each_slave(bond, slave, i) {
3905 const struct rtnl_link_stats64 *sstats =
3906 dev_get_stats(slave->dev, &temp);
3907
3908 stats->rx_packets += sstats->rx_packets;
3909 stats->rx_bytes += sstats->rx_bytes;
3910 stats->rx_errors += sstats->rx_errors;
3911 stats->rx_dropped += sstats->rx_dropped;
3912
3913 stats->tx_packets += sstats->tx_packets;
3914 stats->tx_bytes += sstats->tx_bytes;
3915 stats->tx_errors += sstats->tx_errors;
3916 stats->tx_dropped += sstats->tx_dropped;
3917
3918 stats->multicast += sstats->multicast;
3919 stats->collisions += sstats->collisions;
3920
3921 stats->rx_length_errors += sstats->rx_length_errors;
3922 stats->rx_over_errors += sstats->rx_over_errors;
3923 stats->rx_crc_errors += sstats->rx_crc_errors;
3924 stats->rx_frame_errors += sstats->rx_frame_errors;
3925 stats->rx_fifo_errors += sstats->rx_fifo_errors;
3926 stats->rx_missed_errors += sstats->rx_missed_errors;
3927
3928 stats->tx_aborted_errors += sstats->tx_aborted_errors;
3929 stats->tx_carrier_errors += sstats->tx_carrier_errors;
3930 stats->tx_fifo_errors += sstats->tx_fifo_errors;
3931 stats->tx_heartbeat_errors += sstats->tx_heartbeat_errors;
3932 stats->tx_window_errors += sstats->tx_window_errors;
3933 }
3934
3935 read_unlock_bh(&bond->lock);
3936
3937 return stats;
3938}
3939
3940static int bond_do_ioctl(struct net_device *bond_dev, struct ifreq *ifr, int cmd)
3941{
3942 struct net_device *slave_dev = NULL;
3943 struct ifbond k_binfo;
3944 struct ifbond __user *u_binfo = NULL;
3945 struct ifslave k_sinfo;
3946 struct ifslave __user *u_sinfo = NULL;
3947 struct mii_ioctl_data *mii = NULL;
3948 int res = 0;
3949
3950 pr_debug("bond_ioctl: master=%s, cmd=%d\n", bond_dev->name, cmd);
3951
3952 switch (cmd) {
3953 case SIOCGMIIPHY:
3954 mii = if_mii(ifr);
3955 if (!mii)
3956 return -EINVAL;
3957
3958 mii->phy_id = 0;
3959 /* Fall Through */
3960 case SIOCGMIIREG:
3961 /*
3962 * We do this again just in case we were called by SIOCGMIIREG
3963 * instead of SIOCGMIIPHY.
3964 */
3965 mii = if_mii(ifr);
3966 if (!mii)
3967 return -EINVAL;
3968
3969
3970 if (mii->reg_num == 1) {
3971 struct bonding *bond = netdev_priv(bond_dev);
3972 mii->val_out = 0;
3973 read_lock(&bond->lock);
3974 read_lock(&bond->curr_slave_lock);
3975 if (netif_carrier_ok(bond->dev))
3976 mii->val_out = BMSR_LSTATUS;
3977
3978 read_unlock(&bond->curr_slave_lock);
3979 read_unlock(&bond->lock);
3980 }
3981
3982 return 0;
3983 case BOND_INFO_QUERY_OLD:
3984 case SIOCBONDINFOQUERY:
3985 u_binfo = (struct ifbond __user *)ifr->ifr_data;
3986
3987 if (copy_from_user(&k_binfo, u_binfo, sizeof(ifbond)))
3988 return -EFAULT;
3989
3990 res = bond_info_query(bond_dev, &k_binfo);
3991 if (res == 0 &&
3992 copy_to_user(u_binfo, &k_binfo, sizeof(ifbond)))
3993 return -EFAULT;
3994
3995 return res;
3996 case BOND_SLAVE_INFO_QUERY_OLD:
3997 case SIOCBONDSLAVEINFOQUERY:
3998 u_sinfo = (struct ifslave __user *)ifr->ifr_data;
3999
4000 if (copy_from_user(&k_sinfo, u_sinfo, sizeof(ifslave)))
4001 return -EFAULT;
4002
4003 res = bond_slave_info_query(bond_dev, &k_sinfo);
4004 if (res == 0 &&
4005 copy_to_user(u_sinfo, &k_sinfo, sizeof(ifslave)))
4006 return -EFAULT;
4007
4008 return res;
4009 default:
4010 /* Go on */
4011 break;
4012 }
4013
4014 if (!capable(CAP_NET_ADMIN))
4015 return -EPERM;
4016
4017 slave_dev = dev_get_by_name(dev_net(bond_dev), ifr->ifr_slave);
4018
4019 pr_debug("slave_dev=%p:\n", slave_dev);
4020
4021 if (!slave_dev)
4022 res = -ENODEV;
4023 else {
4024 pr_debug("slave_dev->name=%s:\n", slave_dev->name);
4025 switch (cmd) {
4026 case BOND_ENSLAVE_OLD:
4027 case SIOCBONDENSLAVE:
4028 res = bond_enslave(bond_dev, slave_dev);
4029 break;
4030 case BOND_RELEASE_OLD:
4031 case SIOCBONDRELEASE:
4032 res = bond_release(bond_dev, slave_dev);
4033 break;
4034 case BOND_SETHWADDR_OLD:
4035 case SIOCBONDSETHWADDR:
4036 res = bond_sethwaddr(bond_dev, slave_dev);
4037 break;
4038 case BOND_CHANGE_ACTIVE_OLD:
4039 case SIOCBONDCHANGEACTIVE:
4040 res = bond_ioctl_change_active(bond_dev, slave_dev);
4041 break;
4042 default:
4043 res = -EOPNOTSUPP;
4044 }
4045
4046 dev_put(slave_dev);
4047 }
4048
4049 return res;
4050}
4051
4052static bool bond_addr_in_mc_list(unsigned char *addr,
4053 struct netdev_hw_addr_list *list,
4054 int addrlen)
4055{
4056 struct netdev_hw_addr *ha;
4057
4058 netdev_hw_addr_list_for_each(ha, list)
4059 if (!memcmp(ha->addr, addr, addrlen))
4060 return true;
4061
4062 return false;
4063}
4064
4065static void bond_set_multicast_list(struct net_device *bond_dev)
4066{
4067 struct bonding *bond = netdev_priv(bond_dev);
4068 struct netdev_hw_addr *ha;
4069 bool found;
4070
4071 /*
4072 * Do promisc before checking multicast_mode
4073 */
4074 if ((bond_dev->flags & IFF_PROMISC) && !(bond->flags & IFF_PROMISC))
4075 /*
4076 * FIXME: Need to handle the error when one of the multi-slaves
4077 * encounters error.
4078 */
4079 bond_set_promiscuity(bond, 1);
4080
4081
4082 if (!(bond_dev->flags & IFF_PROMISC) && (bond->flags & IFF_PROMISC))
4083 bond_set_promiscuity(bond, -1);
4084
4085
4086 /* set allmulti flag to slaves */
4087 if ((bond_dev->flags & IFF_ALLMULTI) && !(bond->flags & IFF_ALLMULTI))
4088 /*
4089 * FIXME: Need to handle the error when one of the multi-slaves
4090 * encounters error.
4091 */
4092 bond_set_allmulti(bond, 1);
4093
4094
4095 if (!(bond_dev->flags & IFF_ALLMULTI) && (bond->flags & IFF_ALLMULTI))
4096 bond_set_allmulti(bond, -1);
4097
4098
4099 read_lock(&bond->lock);
4100
4101 bond->flags = bond_dev->flags;
4102
4103 /* looking for addresses to add to slaves' mc list */
4104 netdev_for_each_mc_addr(ha, bond_dev) {
4105 found = bond_addr_in_mc_list(ha->addr, &bond->mc_list,
4106 bond_dev->addr_len);
4107 if (!found)
4108 bond_mc_add(bond, ha->addr);
4109 }
4110
4111 /* looking for addresses to delete from slaves' list */
4112 netdev_hw_addr_list_for_each(ha, &bond->mc_list) {
4113 found = bond_addr_in_mc_list(ha->addr, &bond_dev->mc,
4114 bond_dev->addr_len);
4115 if (!found)
4116 bond_mc_del(bond, ha->addr);
4117 }
4118
4119 /* save master's multicast list */
4120 __hw_addr_flush(&bond->mc_list);
4121 __hw_addr_add_multiple(&bond->mc_list, &bond_dev->mc,
4122 bond_dev->addr_len, NETDEV_HW_ADDR_T_MULTICAST);
4123
4124 read_unlock(&bond->lock);
4125}
4126
4127static int bond_neigh_setup(struct net_device *dev, struct neigh_parms *parms)
4128{
4129 struct bonding *bond = netdev_priv(dev);
4130 struct slave *slave = bond->first_slave;
4131
4132 if (slave) {
4133 const struct net_device_ops *slave_ops
4134 = slave->dev->netdev_ops;
4135 if (slave_ops->ndo_neigh_setup)
4136 return slave_ops->ndo_neigh_setup(slave->dev, parms);
4137 }
4138 return 0;
4139}
4140
4141/*
4142 * Change the MTU of all of a master's slaves to match the master
4143 */
4144static int bond_change_mtu(struct net_device *bond_dev, int new_mtu)
4145{
4146 struct bonding *bond = netdev_priv(bond_dev);
4147 struct slave *slave, *stop_at;
4148 int res = 0;
4149 int i;
4150
4151 pr_debug("bond=%p, name=%s, new_mtu=%d\n", bond,
4152 (bond_dev ? bond_dev->name : "None"), new_mtu);
4153
4154 /* Can't hold bond->lock with bh disabled here since
4155 * some base drivers panic. On the other hand we can't
4156 * hold bond->lock without bh disabled because we'll
4157 * deadlock. The only solution is to rely on the fact
4158 * that we're under rtnl_lock here, and the slaves
4159 * list won't change. This doesn't solve the problem
4160 * of setting the slave's MTU while it is
4161 * transmitting, but the assumption is that the base
4162 * driver can handle that.
4163 *
4164 * TODO: figure out a way to safely iterate the slaves
4165 * list, but without holding a lock around the actual
4166 * call to the base driver.
4167 */
4168
4169 bond_for_each_slave(bond, slave, i) {
4170 pr_debug("s %p s->p %p c_m %p\n",
4171 slave,
4172 slave->prev,
4173 slave->dev->netdev_ops->ndo_change_mtu);
4174
4175 res = dev_set_mtu(slave->dev, new_mtu);
4176
4177 if (res) {
4178 /* If we failed to set the slave's mtu to the new value
4179 * we must abort the operation even in ACTIVE_BACKUP
4180 * mode, because if we allow the backup slaves to have
4181 * different mtu values than the active slave we'll
4182 * need to change their mtu when doing a failover. That
4183 * means changing their mtu from timer context, which
4184 * is probably not a good idea.
4185 */
4186 pr_debug("err %d %s\n", res, slave->dev->name);
4187 goto unwind;
4188 }
4189 }
4190
4191 bond_dev->mtu = new_mtu;
4192
4193 return 0;
4194
4195unwind:
4196 /* unwind from head to the slave that failed */
4197 stop_at = slave;
4198 bond_for_each_slave_from_to(bond, slave, i, bond->first_slave, stop_at) {
4199 int tmp_res;
4200
4201 tmp_res = dev_set_mtu(slave->dev, bond_dev->mtu);
4202 if (tmp_res) {
4203 pr_debug("unwind err %d dev %s\n",
4204 tmp_res, slave->dev->name);
4205 }
4206 }
4207
4208 return res;
4209}
4210
4211/*
4212 * Change HW address
4213 *
4214 * Note that many devices must be down to change the HW address, and
4215 * downing the master releases all slaves. We can make bonds full of
4216 * bonding devices to test this, however.
4217 */
4218static int bond_set_mac_address(struct net_device *bond_dev, void *addr)
4219{
4220 struct bonding *bond = netdev_priv(bond_dev);
4221 struct sockaddr *sa = addr, tmp_sa;
4222 struct slave *slave, *stop_at;
4223 int res = 0;
4224 int i;
4225
4226 if (bond->params.mode == BOND_MODE_ALB)
4227 return bond_alb_set_mac_address(bond_dev, addr);
4228
4229
4230 pr_debug("bond=%p, name=%s\n",
4231 bond, bond_dev ? bond_dev->name : "None");
4232
4233 /*
4234 * If fail_over_mac is set to active, do nothing and return
4235 * success. Returning an error causes ifenslave to fail.
4236 */
4237 if (bond->params.fail_over_mac == BOND_FOM_ACTIVE)
4238 return 0;
4239
4240 if (!is_valid_ether_addr(sa->sa_data))
4241 return -EADDRNOTAVAIL;
4242
4243 /* Can't hold bond->lock with bh disabled here since
4244 * some base drivers panic. On the other hand we can't
4245 * hold bond->lock without bh disabled because we'll
4246 * deadlock. The only solution is to rely on the fact
4247 * that we're under rtnl_lock here, and the slaves
4248 * list won't change. This doesn't solve the problem
4249 * of setting the slave's hw address while it is
4250 * transmitting, but the assumption is that the base
4251 * driver can handle that.
4252 *
4253 * TODO: figure out a way to safely iterate the slaves
4254 * list, but without holding a lock around the actual
4255 * call to the base driver.
4256 */
4257
4258 bond_for_each_slave(bond, slave, i) {
4259 const struct net_device_ops *slave_ops = slave->dev->netdev_ops;
4260 pr_debug("slave %p %s\n", slave, slave->dev->name);
4261
4262 if (slave_ops->ndo_set_mac_address == NULL) {
4263 res = -EOPNOTSUPP;
4264 pr_debug("EOPNOTSUPP %s\n", slave->dev->name);
4265 goto unwind;
4266 }
4267
4268 res = dev_set_mac_address(slave->dev, addr);
4269 if (res) {
4270 /* TODO: consider downing the slave
4271 * and retry ?
4272 * User should expect communications
4273 * breakage anyway until ARP finish
4274 * updating, so...
4275 */
4276 pr_debug("err %d %s\n", res, slave->dev->name);
4277 goto unwind;
4278 }
4279 }
4280
4281 /* success */
4282 memcpy(bond_dev->dev_addr, sa->sa_data, bond_dev->addr_len);
4283 return 0;
4284
4285unwind:
4286 memcpy(tmp_sa.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
4287 tmp_sa.sa_family = bond_dev->type;
4288
4289 /* unwind from head to the slave that failed */
4290 stop_at = slave;
4291 bond_for_each_slave_from_to(bond, slave, i, bond->first_slave, stop_at) {
4292 int tmp_res;
4293
4294 tmp_res = dev_set_mac_address(slave->dev, &tmp_sa);
4295 if (tmp_res) {
4296 pr_debug("unwind err %d dev %s\n",
4297 tmp_res, slave->dev->name);
4298 }
4299 }
4300
4301 return res;
4302}
4303
4304static int bond_xmit_roundrobin(struct sk_buff *skb, struct net_device *bond_dev)
4305{
4306 struct bonding *bond = netdev_priv(bond_dev);
4307 struct slave *slave, *start_at;
4308 int i, slave_no, res = 1;
4309 struct iphdr *iph = ip_hdr(skb);
4310
4311 read_lock(&bond->lock);
4312
4313 if (!BOND_IS_OK(bond))
4314 goto out;
4315 /*
4316 * Start with the curr_active_slave that joined the bond as the
4317 * default for sending IGMP traffic. For failover purposes one
4318 * needs to maintain some consistency for the interface that will
4319 * send the join/membership reports. The curr_active_slave found
4320 * will send all of this type of traffic.
4321 */
4322 if ((iph->protocol == IPPROTO_IGMP) &&
4323 (skb->protocol == htons(ETH_P_IP))) {
4324
4325 read_lock(&bond->curr_slave_lock);
4326 slave = bond->curr_active_slave;
4327 read_unlock(&bond->curr_slave_lock);
4328
4329 if (!slave)
4330 goto out;
4331 } else {
4332 /*
4333 * Concurrent TX may collide on rr_tx_counter; we accept
4334 * that as being rare enough not to justify using an
4335 * atomic op here.
4336 */
4337 slave_no = bond->rr_tx_counter++ % bond->slave_cnt;
4338
4339 bond_for_each_slave(bond, slave, i) {
4340 slave_no--;
4341 if (slave_no < 0)
4342 break;
4343 }
4344 }
4345
4346 start_at = slave;
4347 bond_for_each_slave_from(bond, slave, i, start_at) {
4348 if (IS_UP(slave->dev) &&
4349 (slave->link == BOND_LINK_UP) &&
4350 (slave->state == BOND_STATE_ACTIVE)) {
4351 res = bond_dev_queue_xmit(bond, skb, slave->dev);
4352 break;
4353 }
4354 }
4355
4356out:
4357 if (res) {
4358 /* no suitable interface, frame not sent */
4359 dev_kfree_skb(skb);
4360 }
4361 read_unlock(&bond->lock);
4362 return NETDEV_TX_OK;
4363}
4364
4365
4366/*
4367 * in active-backup mode, we know that bond->curr_active_slave is always valid if
4368 * the bond has a usable interface.
4369 */
4370static int bond_xmit_activebackup(struct sk_buff *skb, struct net_device *bond_dev)
4371{
4372 struct bonding *bond = netdev_priv(bond_dev);
4373 int res = 1;
4374
4375 read_lock(&bond->lock);
4376 read_lock(&bond->curr_slave_lock);
4377
4378 if (!BOND_IS_OK(bond))
4379 goto out;
4380
4381 if (!bond->curr_active_slave)
4382 goto out;
4383
4384 res = bond_dev_queue_xmit(bond, skb, bond->curr_active_slave->dev);
4385
4386out:
4387 if (res)
4388 /* no suitable interface, frame not sent */
4389 dev_kfree_skb(skb);
4390
4391 read_unlock(&bond->curr_slave_lock);
4392 read_unlock(&bond->lock);
4393 return NETDEV_TX_OK;
4394}
4395
4396/*
4397 * In bond_xmit_xor() , we determine the output device by using a pre-
4398 * determined xmit_hash_policy(), If the selected device is not enabled,
4399 * find the next active slave.
4400 */
4401static int bond_xmit_xor(struct sk_buff *skb, struct net_device *bond_dev)
4402{
4403 struct bonding *bond = netdev_priv(bond_dev);
4404 struct slave *slave, *start_at;
4405 int slave_no;
4406 int i;
4407 int res = 1;
4408
4409 read_lock(&bond->lock);
4410
4411 if (!BOND_IS_OK(bond))
4412 goto out;
4413
4414 slave_no = bond->xmit_hash_policy(skb, bond->slave_cnt);
4415
4416 bond_for_each_slave(bond, slave, i) {
4417 slave_no--;
4418 if (slave_no < 0)
4419 break;
4420 }
4421
4422 start_at = slave;
4423
4424 bond_for_each_slave_from(bond, slave, i, start_at) {
4425 if (IS_UP(slave->dev) &&
4426 (slave->link == BOND_LINK_UP) &&
4427 (slave->state == BOND_STATE_ACTIVE)) {
4428 res = bond_dev_queue_xmit(bond, skb, slave->dev);
4429 break;
4430 }
4431 }
4432
4433out:
4434 if (res) {
4435 /* no suitable interface, frame not sent */
4436 dev_kfree_skb(skb);
4437 }
4438 read_unlock(&bond->lock);
4439 return NETDEV_TX_OK;
4440}
4441
4442/*
4443 * in broadcast mode, we send everything to all usable interfaces.
4444 */
4445static int bond_xmit_broadcast(struct sk_buff *skb, struct net_device *bond_dev)
4446{
4447 struct bonding *bond = netdev_priv(bond_dev);
4448 struct slave *slave, *start_at;
4449 struct net_device *tx_dev = NULL;
4450 int i;
4451 int res = 1;
4452
4453 read_lock(&bond->lock);
4454
4455 if (!BOND_IS_OK(bond))
4456 goto out;
4457
4458 read_lock(&bond->curr_slave_lock);
4459 start_at = bond->curr_active_slave;
4460 read_unlock(&bond->curr_slave_lock);
4461
4462 if (!start_at)
4463 goto out;
4464
4465 bond_for_each_slave_from(bond, slave, i, start_at) {
4466 if (IS_UP(slave->dev) &&
4467 (slave->link == BOND_LINK_UP) &&
4468 (slave->state == BOND_STATE_ACTIVE)) {
4469 if (tx_dev) {
4470 struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
4471 if (!skb2) {
4472 pr_err("%s: Error: bond_xmit_broadcast(): skb_clone() failed\n",
4473 bond_dev->name);
4474 continue;
4475 }
4476
4477 res = bond_dev_queue_xmit(bond, skb2, tx_dev);
4478 if (res) {
4479 dev_kfree_skb(skb2);
4480 continue;
4481 }
4482 }
4483 tx_dev = slave->dev;
4484 }
4485 }
4486
4487 if (tx_dev)
4488 res = bond_dev_queue_xmit(bond, skb, tx_dev);
4489
4490out:
4491 if (res)
4492 /* no suitable interface, frame not sent */
4493 dev_kfree_skb(skb);
4494
4495 /* frame sent to all suitable interfaces */
4496 read_unlock(&bond->lock);
4497 return NETDEV_TX_OK;
4498}
4499
4500/*------------------------- Device initialization ---------------------------*/
4501
4502static void bond_set_xmit_hash_policy(struct bonding *bond)
4503{
4504 switch (bond->params.xmit_policy) {
4505 case BOND_XMIT_POLICY_LAYER23:
4506 bond->xmit_hash_policy = bond_xmit_hash_policy_l23;
4507 break;
4508 case BOND_XMIT_POLICY_LAYER34:
4509 bond->xmit_hash_policy = bond_xmit_hash_policy_l34;
4510 break;
4511 case BOND_XMIT_POLICY_LAYER2:
4512 default:
4513 bond->xmit_hash_policy = bond_xmit_hash_policy_l2;
4514 break;
4515 }
4516}
4517
4518/*
4519 * Lookup the slave that corresponds to a qid
4520 */
4521static inline int bond_slave_override(struct bonding *bond,
4522 struct sk_buff *skb)
4523{
4524 int i, res = 1;
4525 struct slave *slave = NULL;
4526 struct slave *check_slave;
4527
4528 read_lock(&bond->lock);
4529
4530 if (!BOND_IS_OK(bond) || !skb->queue_mapping)
4531 goto out;
4532
4533 /* Find out if any slaves have the same mapping as this skb. */
4534 bond_for_each_slave(bond, check_slave, i) {
4535 if (check_slave->queue_id == skb->queue_mapping) {
4536 slave = check_slave;
4537 break;
4538 }
4539 }
4540
4541 /* If the slave isn't UP, use default transmit policy. */
4542 if (slave && slave->queue_id && IS_UP(slave->dev) &&
4543 (slave->link == BOND_LINK_UP)) {
4544 res = bond_dev_queue_xmit(bond, skb, slave->dev);
4545 }
4546
4547out:
4548 read_unlock(&bond->lock);
4549 return res;
4550}
4551
4552static u16 bond_select_queue(struct net_device *dev, struct sk_buff *skb)
4553{
4554 /*
4555 * This helper function exists to help dev_pick_tx get the correct
4556 * destination queue. Using a helper function skips the a call to
4557 * skb_tx_hash and will put the skbs in the queue we expect on their
4558 * way down to the bonding driver.
4559 */
4560 return skb->queue_mapping;
4561}
4562
4563static netdev_tx_t bond_start_xmit(struct sk_buff *skb, struct net_device *dev)
4564{
4565 struct bonding *bond = netdev_priv(dev);
4566
4567 /*
4568 * If we risk deadlock from transmitting this in the
4569 * netpoll path, tell netpoll to queue the frame for later tx
4570 */
4571 if (is_netpoll_tx_blocked(dev))
4572 return NETDEV_TX_BUSY;
4573
4574 if (TX_QUEUE_OVERRIDE(bond->params.mode)) {
4575 if (!bond_slave_override(bond, skb))
4576 return NETDEV_TX_OK;
4577 }
4578
4579 switch (bond->params.mode) {
4580 case BOND_MODE_ROUNDROBIN:
4581 return bond_xmit_roundrobin(skb, dev);
4582 case BOND_MODE_ACTIVEBACKUP:
4583 return bond_xmit_activebackup(skb, dev);
4584 case BOND_MODE_XOR:
4585 return bond_xmit_xor(skb, dev);
4586 case BOND_MODE_BROADCAST:
4587 return bond_xmit_broadcast(skb, dev);
4588 case BOND_MODE_8023AD:
4589 return bond_3ad_xmit_xor(skb, dev);
4590 case BOND_MODE_ALB:
4591 case BOND_MODE_TLB:
4592 return bond_alb_xmit(skb, dev);
4593 default:
4594 /* Should never happen, mode already checked */
4595 pr_err("%s: Error: Unknown bonding mode %d\n",
4596 dev->name, bond->params.mode);
4597 WARN_ON_ONCE(1);
4598 dev_kfree_skb(skb);
4599 return NETDEV_TX_OK;
4600 }
4601}
4602
4603
4604/*
4605 * set bond mode specific net device operations
4606 */
4607void bond_set_mode_ops(struct bonding *bond, int mode)
4608{
4609 struct net_device *bond_dev = bond->dev;
4610
4611 switch (mode) {
4612 case BOND_MODE_ROUNDROBIN:
4613 break;
4614 case BOND_MODE_ACTIVEBACKUP:
4615 break;
4616 case BOND_MODE_XOR:
4617 bond_set_xmit_hash_policy(bond);
4618 break;
4619 case BOND_MODE_BROADCAST:
4620 break;
4621 case BOND_MODE_8023AD:
4622 bond_set_master_3ad_flags(bond);
4623 bond_set_xmit_hash_policy(bond);
4624 break;
4625 case BOND_MODE_ALB:
4626 bond_set_master_alb_flags(bond);
4627 /* FALLTHRU */
4628 case BOND_MODE_TLB:
4629 break;
4630 default:
4631 /* Should never happen, mode already checked */
4632 pr_err("%s: Error: Unknown bonding mode %d\n",
4633 bond_dev->name, mode);
4634 break;
4635 }
4636}
4637
4638static void bond_ethtool_get_drvinfo(struct net_device *bond_dev,
4639 struct ethtool_drvinfo *drvinfo)
4640{
4641 strncpy(drvinfo->driver, DRV_NAME, 32);
4642 strncpy(drvinfo->version, DRV_VERSION, 32);
4643 snprintf(drvinfo->fw_version, 32, "%d", BOND_ABI_VERSION);
4644}
4645
4646static const struct ethtool_ops bond_ethtool_ops = {
4647 .get_drvinfo = bond_ethtool_get_drvinfo,
4648 .get_link = ethtool_op_get_link,
4649 .get_tx_csum = ethtool_op_get_tx_csum,
4650 .get_sg = ethtool_op_get_sg,
4651 .get_tso = ethtool_op_get_tso,
4652 .get_ufo = ethtool_op_get_ufo,
4653 .get_flags = ethtool_op_get_flags,
4654};
4655
4656static const struct net_device_ops bond_netdev_ops = {
4657 .ndo_init = bond_init,
4658 .ndo_uninit = bond_uninit,
4659 .ndo_open = bond_open,
4660 .ndo_stop = bond_close,
4661 .ndo_start_xmit = bond_start_xmit,
4662 .ndo_select_queue = bond_select_queue,
4663 .ndo_get_stats64 = bond_get_stats,
4664 .ndo_do_ioctl = bond_do_ioctl,
4665 .ndo_set_multicast_list = bond_set_multicast_list,
4666 .ndo_change_mtu = bond_change_mtu,
4667 .ndo_set_mac_address = bond_set_mac_address,
4668 .ndo_neigh_setup = bond_neigh_setup,
4669 .ndo_vlan_rx_register = bond_vlan_rx_register,
4670 .ndo_vlan_rx_add_vid = bond_vlan_rx_add_vid,
4671 .ndo_vlan_rx_kill_vid = bond_vlan_rx_kill_vid,
4672#ifdef CONFIG_NET_POLL_CONTROLLER
4673 .ndo_netpoll_cleanup = bond_netpoll_cleanup,
4674 .ndo_poll_controller = bond_poll_controller,
4675#endif
4676};
4677
4678static void bond_destructor(struct net_device *bond_dev)
4679{
4680 struct bonding *bond = netdev_priv(bond_dev);
4681 if (bond->wq)
4682 destroy_workqueue(bond->wq);
4683 free_netdev(bond_dev);
4684}
4685
4686static void bond_setup(struct net_device *bond_dev)
4687{
4688 struct bonding *bond = netdev_priv(bond_dev);
4689
4690 /* initialize rwlocks */
4691 rwlock_init(&bond->lock);
4692 rwlock_init(&bond->curr_slave_lock);
4693
4694 bond->params = bonding_defaults;
4695
4696 /* Initialize pointers */
4697 bond->dev = bond_dev;
4698 INIT_LIST_HEAD(&bond->vlan_list);
4699
4700 /* Initialize the device entry points */
4701 ether_setup(bond_dev);
4702 bond_dev->netdev_ops = &bond_netdev_ops;
4703 bond_dev->ethtool_ops = &bond_ethtool_ops;
4704 bond_set_mode_ops(bond, bond->params.mode);
4705
4706 bond_dev->destructor = bond_destructor;
4707
4708 /* Initialize the device options */
4709 bond_dev->tx_queue_len = 0;
4710 bond_dev->flags |= IFF_MASTER|IFF_MULTICAST;
4711 bond_dev->priv_flags |= IFF_BONDING;
4712 bond_dev->priv_flags &= ~IFF_XMIT_DST_RELEASE;
4713
4714 if (bond->params.arp_interval)
4715 bond_dev->priv_flags |= IFF_MASTER_ARPMON;
4716
4717 /* At first, we block adding VLANs. That's the only way to
4718 * prevent problems that occur when adding VLANs over an
4719 * empty bond. The block will be removed once non-challenged
4720 * slaves are enslaved.
4721 */
4722 bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
4723
4724 /* don't acquire bond device's netif_tx_lock when
4725 * transmitting */
4726 bond_dev->features |= NETIF_F_LLTX;
4727
4728 /* By default, we declare the bond to be fully
4729 * VLAN hardware accelerated capable. Special
4730 * care is taken in the various xmit functions
4731 * when there are slaves that are not hw accel
4732 * capable
4733 */
4734 bond_dev->features |= (NETIF_F_HW_VLAN_TX |
4735 NETIF_F_HW_VLAN_RX |
4736 NETIF_F_HW_VLAN_FILTER);
4737
4738 /* By default, we enable GRO on bonding devices.
4739 * Actual support requires lowlevel drivers are GRO ready.
4740 */
4741 bond_dev->features |= NETIF_F_GRO;
4742}
4743
4744static void bond_work_cancel_all(struct bonding *bond)
4745{
4746 write_lock_bh(&bond->lock);
4747 bond->kill_timers = 1;
4748 write_unlock_bh(&bond->lock);
4749
4750 if (bond->params.miimon && delayed_work_pending(&bond->mii_work))
4751 cancel_delayed_work(&bond->mii_work);
4752
4753 if (bond->params.arp_interval && delayed_work_pending(&bond->arp_work))
4754 cancel_delayed_work(&bond->arp_work);
4755
4756 if (bond->params.mode == BOND_MODE_ALB &&
4757 delayed_work_pending(&bond->alb_work))
4758 cancel_delayed_work(&bond->alb_work);
4759
4760 if (bond->params.mode == BOND_MODE_8023AD &&
4761 delayed_work_pending(&bond->ad_work))
4762 cancel_delayed_work(&bond->ad_work);
4763
4764 if (delayed_work_pending(&bond->mcast_work))
4765 cancel_delayed_work(&bond->mcast_work);
4766}
4767
4768/*
4769* Destroy a bonding device.
4770* Must be under rtnl_lock when this function is called.
4771*/
4772static void bond_uninit(struct net_device *bond_dev)
4773{
4774 struct bonding *bond = netdev_priv(bond_dev);
4775 struct vlan_entry *vlan, *tmp;
4776
4777 bond_netpoll_cleanup(bond_dev);
4778
4779 /* Release the bonded slaves */
4780 bond_release_all(bond_dev);
4781
4782 list_del(&bond->bond_list);
4783
4784 bond_work_cancel_all(bond);
4785
4786 bond_remove_proc_entry(bond);
4787
4788 __hw_addr_flush(&bond->mc_list);
4789
4790 list_for_each_entry_safe(vlan, tmp, &bond->vlan_list, vlan_list) {
4791 list_del(&vlan->vlan_list);
4792 kfree(vlan);
4793 }
4794}
4795
4796/*------------------------- Module initialization ---------------------------*/
4797
4798/*
4799 * Convert string input module parms. Accept either the
4800 * number of the mode or its string name. A bit complicated because
4801 * some mode names are substrings of other names, and calls from sysfs
4802 * may have whitespace in the name (trailing newlines, for example).
4803 */
4804int bond_parse_parm(const char *buf, const struct bond_parm_tbl *tbl)
4805{
4806 int modeint = -1, i, rv;
4807 char *p, modestr[BOND_MAX_MODENAME_LEN + 1] = { 0, };
4808
4809 for (p = (char *)buf; *p; p++)
4810 if (!(isdigit(*p) || isspace(*p)))
4811 break;
4812
4813 if (*p)
4814 rv = sscanf(buf, "%20s", modestr);
4815 else
4816 rv = sscanf(buf, "%d", &modeint);
4817
4818 if (!rv)
4819 return -1;
4820
4821 for (i = 0; tbl[i].modename; i++) {
4822 if (modeint == tbl[i].mode)
4823 return tbl[i].mode;
4824 if (strcmp(modestr, tbl[i].modename) == 0)
4825 return tbl[i].mode;
4826 }
4827
4828 return -1;
4829}
4830
4831static int bond_check_params(struct bond_params *params)
4832{
4833 int arp_validate_value, fail_over_mac_value, primary_reselect_value;
4834
4835 /*
4836 * Convert string parameters.
4837 */
4838 if (mode) {
4839 bond_mode = bond_parse_parm(mode, bond_mode_tbl);
4840 if (bond_mode == -1) {
4841 pr_err("Error: Invalid bonding mode \"%s\"\n",
4842 mode == NULL ? "NULL" : mode);
4843 return -EINVAL;
4844 }
4845 }
4846
4847 if (xmit_hash_policy) {
4848 if ((bond_mode != BOND_MODE_XOR) &&
4849 (bond_mode != BOND_MODE_8023AD)) {
4850 pr_info("xmit_hash_policy param is irrelevant in mode %s\n",
4851 bond_mode_name(bond_mode));
4852 } else {
4853 xmit_hashtype = bond_parse_parm(xmit_hash_policy,
4854 xmit_hashtype_tbl);
4855 if (xmit_hashtype == -1) {
4856 pr_err("Error: Invalid xmit_hash_policy \"%s\"\n",
4857 xmit_hash_policy == NULL ? "NULL" :
4858 xmit_hash_policy);
4859 return -EINVAL;
4860 }
4861 }
4862 }
4863
4864 if (lacp_rate) {
4865 if (bond_mode != BOND_MODE_8023AD) {
4866 pr_info("lacp_rate param is irrelevant in mode %s\n",
4867 bond_mode_name(bond_mode));
4868 } else {
4869 lacp_fast = bond_parse_parm(lacp_rate, bond_lacp_tbl);
4870 if (lacp_fast == -1) {
4871 pr_err("Error: Invalid lacp rate \"%s\"\n",
4872 lacp_rate == NULL ? "NULL" : lacp_rate);
4873 return -EINVAL;
4874 }
4875 }
4876 }
4877
4878 if (ad_select) {
4879 params->ad_select = bond_parse_parm(ad_select, ad_select_tbl);
4880 if (params->ad_select == -1) {
4881 pr_err("Error: Invalid ad_select \"%s\"\n",
4882 ad_select == NULL ? "NULL" : ad_select);
4883 return -EINVAL;
4884 }
4885
4886 if (bond_mode != BOND_MODE_8023AD) {
4887 pr_warning("ad_select param only affects 802.3ad mode\n");
4888 }
4889 } else {
4890 params->ad_select = BOND_AD_STABLE;
4891 }
4892
4893 if (max_bonds < 0) {
4894 pr_warning("Warning: max_bonds (%d) not in range %d-%d, so it was reset to BOND_DEFAULT_MAX_BONDS (%d)\n",
4895 max_bonds, 0, INT_MAX, BOND_DEFAULT_MAX_BONDS);
4896 max_bonds = BOND_DEFAULT_MAX_BONDS;
4897 }
4898
4899 if (miimon < 0) {
4900 pr_warning("Warning: miimon module parameter (%d), not in range 0-%d, so it was reset to %d\n",
4901 miimon, INT_MAX, BOND_LINK_MON_INTERV);
4902 miimon = BOND_LINK_MON_INTERV;
4903 }
4904
4905 if (updelay < 0) {
4906 pr_warning("Warning: updelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4907 updelay, INT_MAX);
4908 updelay = 0;
4909 }
4910
4911 if (downdelay < 0) {
4912 pr_warning("Warning: downdelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4913 downdelay, INT_MAX);
4914 downdelay = 0;
4915 }
4916
4917 if ((use_carrier != 0) && (use_carrier != 1)) {
4918 pr_warning("Warning: use_carrier module parameter (%d), not of valid value (0/1), so it was set to 1\n",
4919 use_carrier);
4920 use_carrier = 1;
4921 }
4922
4923 if (num_grat_arp < 0 || num_grat_arp > 255) {
4924 pr_warning("Warning: num_grat_arp (%d) not in range 0-255 so it was reset to 1\n",
4925 num_grat_arp);
4926 num_grat_arp = 1;
4927 }
4928
4929 if (num_unsol_na < 0 || num_unsol_na > 255) {
4930 pr_warning("Warning: num_unsol_na (%d) not in range 0-255 so it was reset to 1\n",
4931 num_unsol_na);
4932 num_unsol_na = 1;
4933 }
4934
4935 /* reset values for 802.3ad */
4936 if (bond_mode == BOND_MODE_8023AD) {
4937 if (!miimon) {
4938 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");
4939 pr_warning("Forcing miimon to 100msec\n");
4940 miimon = 100;
4941 }
4942 }
4943
4944 if (tx_queues < 1 || tx_queues > 255) {
4945 pr_warning("Warning: tx_queues (%d) should be between "
4946 "1 and 255, resetting to %d\n",
4947 tx_queues, BOND_DEFAULT_TX_QUEUES);
4948 tx_queues = BOND_DEFAULT_TX_QUEUES;
4949 }
4950
4951 if ((all_slaves_active != 0) && (all_slaves_active != 1)) {
4952 pr_warning("Warning: all_slaves_active module parameter (%d), "
4953 "not of valid value (0/1), so it was set to "
4954 "0\n", all_slaves_active);
4955 all_slaves_active = 0;
4956 }
4957
4958 if (resend_igmp < 0 || resend_igmp > 255) {
4959 pr_warning("Warning: resend_igmp (%d) should be between "
4960 "0 and 255, resetting to %d\n",
4961 resend_igmp, BOND_DEFAULT_RESEND_IGMP);
4962 resend_igmp = BOND_DEFAULT_RESEND_IGMP;
4963 }
4964
4965 /* reset values for TLB/ALB */
4966 if ((bond_mode == BOND_MODE_TLB) ||
4967 (bond_mode == BOND_MODE_ALB)) {
4968 if (!miimon) {
4969 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");
4970 pr_warning("Forcing miimon to 100msec\n");
4971 miimon = 100;
4972 }
4973 }
4974
4975 if (bond_mode == BOND_MODE_ALB) {
4976 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",
4977 updelay);
4978 }
4979
4980 if (!miimon) {
4981 if (updelay || downdelay) {
4982 /* just warn the user the up/down delay will have
4983 * no effect since miimon is zero...
4984 */
4985 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",
4986 updelay, downdelay);
4987 }
4988 } else {
4989 /* don't allow arp monitoring */
4990 if (arp_interval) {
4991 pr_warning("Warning: miimon (%d) and arp_interval (%d) can't be used simultaneously, disabling ARP monitoring\n",
4992 miimon, arp_interval);
4993 arp_interval = 0;
4994 }
4995
4996 if ((updelay % miimon) != 0) {
4997 pr_warning("Warning: updelay (%d) is not a multiple of miimon (%d), updelay rounded to %d ms\n",
4998 updelay, miimon,
4999 (updelay / miimon) * miimon);
5000 }
5001
5002 updelay /= miimon;
5003
5004 if ((downdelay % miimon) != 0) {
5005 pr_warning("Warning: downdelay (%d) is not a multiple of miimon (%d), downdelay rounded to %d ms\n",
5006 downdelay, miimon,
5007 (downdelay / miimon) * miimon);
5008 }
5009
5010 downdelay /= miimon;
5011 }
5012
5013 if (arp_interval < 0) {
5014 pr_warning("Warning: arp_interval module parameter (%d) , not in range 0-%d, so it was reset to %d\n",
5015 arp_interval, INT_MAX, BOND_LINK_ARP_INTERV);
5016 arp_interval = BOND_LINK_ARP_INTERV;
5017 }
5018
5019 for (arp_ip_count = 0;
5020 (arp_ip_count < BOND_MAX_ARP_TARGETS) && arp_ip_target[arp_ip_count];
5021 arp_ip_count++) {
5022 /* not complete check, but should be good enough to
5023 catch mistakes */
5024 if (!isdigit(arp_ip_target[arp_ip_count][0])) {
5025 pr_warning("Warning: bad arp_ip_target module parameter (%s), ARP monitoring will not be performed\n",
5026 arp_ip_target[arp_ip_count]);
5027 arp_interval = 0;
5028 } else {
5029 __be32 ip = in_aton(arp_ip_target[arp_ip_count]);
5030 arp_target[arp_ip_count] = ip;
5031 }
5032 }
5033
5034 if (arp_interval && !arp_ip_count) {
5035 /* don't allow arping if no arp_ip_target given... */
5036 pr_warning("Warning: arp_interval module parameter (%d) specified without providing an arp_ip_target parameter, arp_interval was reset to 0\n",
5037 arp_interval);
5038 arp_interval = 0;
5039 }
5040
5041 if (arp_validate) {
5042 if (bond_mode != BOND_MODE_ACTIVEBACKUP) {
5043 pr_err("arp_validate only supported in active-backup mode\n");
5044 return -EINVAL;
5045 }
5046 if (!arp_interval) {
5047 pr_err("arp_validate requires arp_interval\n");
5048 return -EINVAL;
5049 }
5050
5051 arp_validate_value = bond_parse_parm(arp_validate,
5052 arp_validate_tbl);
5053 if (arp_validate_value == -1) {
5054 pr_err("Error: invalid arp_validate \"%s\"\n",
5055 arp_validate == NULL ? "NULL" : arp_validate);
5056 return -EINVAL;
5057 }
5058 } else
5059 arp_validate_value = 0;
5060
5061 if (miimon) {
5062 pr_info("MII link monitoring set to %d ms\n", miimon);
5063 } else if (arp_interval) {
5064 int i;
5065
5066 pr_info("ARP monitoring set to %d ms, validate %s, with %d target(s):",
5067 arp_interval,
5068 arp_validate_tbl[arp_validate_value].modename,
5069 arp_ip_count);
5070
5071 for (i = 0; i < arp_ip_count; i++)
5072 pr_info(" %s", arp_ip_target[i]);
5073
5074 pr_info("\n");
5075
5076 } else if (max_bonds) {
5077 /* miimon and arp_interval not set, we need one so things
5078 * work as expected, see bonding.txt for details
5079 */
5080 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");
5081 }
5082
5083 if (primary && !USES_PRIMARY(bond_mode)) {
5084 /* currently, using a primary only makes sense
5085 * in active backup, TLB or ALB modes
5086 */
5087 pr_warning("Warning: %s primary device specified but has no effect in %s mode\n",
5088 primary, bond_mode_name(bond_mode));
5089 primary = NULL;
5090 }
5091
5092 if (primary && primary_reselect) {
5093 primary_reselect_value = bond_parse_parm(primary_reselect,
5094 pri_reselect_tbl);
5095 if (primary_reselect_value == -1) {
5096 pr_err("Error: Invalid primary_reselect \"%s\"\n",
5097 primary_reselect ==
5098 NULL ? "NULL" : primary_reselect);
5099 return -EINVAL;
5100 }
5101 } else {
5102 primary_reselect_value = BOND_PRI_RESELECT_ALWAYS;
5103 }
5104
5105 if (fail_over_mac) {
5106 fail_over_mac_value = bond_parse_parm(fail_over_mac,
5107 fail_over_mac_tbl);
5108 if (fail_over_mac_value == -1) {
5109 pr_err("Error: invalid fail_over_mac \"%s\"\n",
5110 arp_validate == NULL ? "NULL" : arp_validate);
5111 return -EINVAL;
5112 }
5113
5114 if (bond_mode != BOND_MODE_ACTIVEBACKUP)
5115 pr_warning("Warning: fail_over_mac only affects active-backup mode.\n");
5116 } else {
5117 fail_over_mac_value = BOND_FOM_NONE;
5118 }
5119
5120 /* fill params struct with the proper values */
5121 params->mode = bond_mode;
5122 params->xmit_policy = xmit_hashtype;
5123 params->miimon = miimon;
5124 params->num_grat_arp = num_grat_arp;
5125 params->num_unsol_na = num_unsol_na;
5126 params->arp_interval = arp_interval;
5127 params->arp_validate = arp_validate_value;
5128 params->updelay = updelay;
5129 params->downdelay = downdelay;
5130 params->use_carrier = use_carrier;
5131 params->lacp_fast = lacp_fast;
5132 params->primary[0] = 0;
5133 params->primary_reselect = primary_reselect_value;
5134 params->fail_over_mac = fail_over_mac_value;
5135 params->tx_queues = tx_queues;
5136 params->all_slaves_active = all_slaves_active;
5137 params->resend_igmp = resend_igmp;
5138
5139 if (primary) {
5140 strncpy(params->primary, primary, IFNAMSIZ);
5141 params->primary[IFNAMSIZ - 1] = 0;
5142 }
5143
5144 memcpy(params->arp_targets, arp_target, sizeof(arp_target));
5145
5146 return 0;
5147}
5148
5149static struct lock_class_key bonding_netdev_xmit_lock_key;
5150static struct lock_class_key bonding_netdev_addr_lock_key;
5151
5152static void bond_set_lockdep_class_one(struct net_device *dev,
5153 struct netdev_queue *txq,
5154 void *_unused)
5155{
5156 lockdep_set_class(&txq->_xmit_lock,
5157 &bonding_netdev_xmit_lock_key);
5158}
5159
5160static void bond_set_lockdep_class(struct net_device *dev)
5161{
5162 lockdep_set_class(&dev->addr_list_lock,
5163 &bonding_netdev_addr_lock_key);
5164 netdev_for_each_tx_queue(dev, bond_set_lockdep_class_one, NULL);
5165}
5166
5167/*
5168 * Called from registration process
5169 */
5170static int bond_init(struct net_device *bond_dev)
5171{
5172 struct bonding *bond = netdev_priv(bond_dev);
5173 struct bond_net *bn = net_generic(dev_net(bond_dev), bond_net_id);
5174
5175 pr_debug("Begin bond_init for %s\n", bond_dev->name);
5176
5177 bond->wq = create_singlethread_workqueue(bond_dev->name);
5178 if (!bond->wq)
5179 return -ENOMEM;
5180
5181 bond_set_lockdep_class(bond_dev);
5182
5183 netif_carrier_off(bond_dev);
5184
5185 bond_create_proc_entry(bond);
5186 list_add_tail(&bond->bond_list, &bn->dev_list);
5187
5188 bond_prepare_sysfs_group(bond);
5189
5190 __hw_addr_init(&bond->mc_list);
5191 return 0;
5192}
5193
5194static int bond_validate(struct nlattr *tb[], struct nlattr *data[])
5195{
5196 if (tb[IFLA_ADDRESS]) {
5197 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN)
5198 return -EINVAL;
5199 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS])))
5200 return -EADDRNOTAVAIL;
5201 }
5202 return 0;
5203}
5204
5205static struct rtnl_link_ops bond_link_ops __read_mostly = {
5206 .kind = "bond",
5207 .priv_size = sizeof(struct bonding),
5208 .setup = bond_setup,
5209 .validate = bond_validate,
5210};
5211
5212/* Create a new bond based on the specified name and bonding parameters.
5213 * If name is NULL, obtain a suitable "bond%d" name for us.
5214 * Caller must NOT hold rtnl_lock; we need to release it here before we
5215 * set up our sysfs entries.
5216 */
5217int bond_create(struct net *net, const char *name)
5218{
5219 struct net_device *bond_dev;
5220 int res;
5221
5222 rtnl_lock();
5223
5224 bond_dev = alloc_netdev_mq(sizeof(struct bonding), name ? name : "",
5225 bond_setup, tx_queues);
5226 if (!bond_dev) {
5227 pr_err("%s: eek! can't alloc netdev!\n", name);
5228 rtnl_unlock();
5229 return -ENOMEM;
5230 }
5231
5232 dev_net_set(bond_dev, net);
5233 bond_dev->rtnl_link_ops = &bond_link_ops;
5234
5235 if (!name) {
5236 res = dev_alloc_name(bond_dev, "bond%d");
5237 if (res < 0)
5238 goto out;
5239 } else {
5240 /*
5241 * If we're given a name to register
5242 * we need to ensure that its not already
5243 * registered
5244 */
5245 res = -EEXIST;
5246 if (__dev_get_by_name(net, name) != NULL)
5247 goto out;
5248 }
5249
5250 res = register_netdevice(bond_dev);
5251
5252out:
5253 rtnl_unlock();
5254 if (res < 0)
5255 bond_destructor(bond_dev);
5256 return res;
5257}
5258
5259static int __net_init bond_net_init(struct net *net)
5260{
5261 struct bond_net *bn = net_generic(net, bond_net_id);
5262
5263 bn->net = net;
5264 INIT_LIST_HEAD(&bn->dev_list);
5265
5266 bond_create_proc_dir(bn);
5267
5268 return 0;
5269}
5270
5271static void __net_exit bond_net_exit(struct net *net)
5272{
5273 struct bond_net *bn = net_generic(net, bond_net_id);
5274
5275 bond_destroy_proc_dir(bn);
5276}
5277
5278static struct pernet_operations bond_net_ops = {
5279 .init = bond_net_init,
5280 .exit = bond_net_exit,
5281 .id = &bond_net_id,
5282 .size = sizeof(struct bond_net),
5283};
5284
5285static int __init bonding_init(void)
5286{
5287 int i;
5288 int res;
5289
5290 pr_info("%s", version);
5291
5292 res = bond_check_params(&bonding_defaults);
5293 if (res)
5294 goto out;
5295
5296#ifdef CONFIG_NET_POLL_CONTROLLER
5297 if (!alloc_cpumask_var(&netpoll_block_tx, GFP_KERNEL)) {
5298 res = -ENOMEM;
5299 goto out;
5300 }
5301#endif
5302
5303 res = register_pernet_subsys(&bond_net_ops);
5304 if (res)
5305 goto out;
5306
5307 res = rtnl_link_register(&bond_link_ops);
5308 if (res)
5309 goto err_link;
5310
5311 for (i = 0; i < max_bonds; i++) {
5312 res = bond_create(&init_net, NULL);
5313 if (res)
5314 goto err;
5315 }
5316
5317 res = bond_create_sysfs();
5318 if (res)
5319 goto err;
5320
5321
5322 register_netdevice_notifier(&bond_netdev_notifier);
5323 register_inetaddr_notifier(&bond_inetaddr_notifier);
5324 bond_register_ipv6_notifier();
5325out:
5326 return res;
5327err:
5328 rtnl_link_unregister(&bond_link_ops);
5329err_link:
5330 unregister_pernet_subsys(&bond_net_ops);
5331#ifdef CONFIG_NET_POLL_CONTROLLER
5332 free_cpumask_var(netpoll_block_tx);
5333#endif
5334 goto out;
5335
5336}
5337
5338static void __exit bonding_exit(void)
5339{
5340 unregister_netdevice_notifier(&bond_netdev_notifier);
5341 unregister_inetaddr_notifier(&bond_inetaddr_notifier);
5342 bond_unregister_ipv6_notifier();
5343
5344 bond_destroy_sysfs();
5345
5346 rtnl_link_unregister(&bond_link_ops);
5347 unregister_pernet_subsys(&bond_net_ops);
5348
5349#ifdef CONFIG_NET_POLL_CONTROLLER
5350 free_cpumask_var(netpoll_block_tx);
5351#endif
5352}
5353
5354module_init(bonding_init);
5355module_exit(bonding_exit);
5356MODULE_LICENSE("GPL");
5357MODULE_VERSION(DRV_VERSION);
5358MODULE_DESCRIPTION(DRV_DESCRIPTION ", v" DRV_VERSION);
5359MODULE_AUTHOR("Thomas Davis, tadavis@lbl.gov and many others");
5360MODULE_ALIAS_RTNL_LINK("bond");
Clone of davem's net-next-2.6 tree