bonding: allow user-controlled output slave selection
[linux-2.6.git] / drivers / net / bonding / bond_main.c
1 /*
2  * originally based on the dummy device.
3  *
4  * Copyright 1999, Thomas Davis, tadavis@lbl.gov.
5  * Licensed under the GPL. Based on dummy.c, and eql.c devices.
6  *
7  * bonding.c: an Ethernet Bonding driver
8  *
9  * This is useful to talk to a Cisco EtherChannel compatible equipment:
10  *      Cisco 5500
11  *      Sun Trunking (Solaris)
12  *      Alteon AceDirector Trunks
13  *      Linux Bonding
14  *      and probably many L2 switches ...
15  *
16  * How it works:
17  *    ifconfig bond0 ipaddress netmask up
18  *      will setup a network device, with an ip address.  No mac address
19  *      will be assigned at this time.  The hw mac address will come from
20  *      the first slave bonded to the channel.  All slaves will then use
21  *      this hw mac address.
22  *
23  *    ifconfig bond0 down
24  *         will release all slaves, marking them as down.
25  *
26  *    ifenslave bond0 eth0
27  *      will attach eth0 to bond0 as a slave.  eth0 hw mac address will either
28  *      a: be used as initial mac address
29  *      b: if a hw mac address already is there, eth0's hw mac address
30  *         will then be set from bond0.
31  *
32  */
33
34 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
35
36 #include <linux/kernel.h>
37 #include <linux/module.h>
38 #include <linux/types.h>
39 #include <linux/fcntl.h>
40 #include <linux/interrupt.h>
41 #include <linux/ptrace.h>
42 #include <linux/ioport.h>
43 #include <linux/in.h>
44 #include <net/ip.h>
45 #include <linux/ip.h>
46 #include <linux/tcp.h>
47 #include <linux/udp.h>
48 #include <linux/slab.h>
49 #include <linux/string.h>
50 #include <linux/init.h>
51 #include <linux/timer.h>
52 #include <linux/socket.h>
53 #include <linux/ctype.h>
54 #include <linux/inet.h>
55 #include <linux/bitops.h>
56 #include <linux/io.h>
57 #include <asm/system.h>
58 #include <asm/dma.h>
59 #include <linux/uaccess.h>
60 #include <linux/errno.h>
61 #include <linux/netdevice.h>
62 #include <linux/netpoll.h>
63 #include <linux/inetdevice.h>
64 #include <linux/igmp.h>
65 #include <linux/etherdevice.h>
66 #include <linux/skbuff.h>
67 #include <net/sock.h>
68 #include <linux/rtnetlink.h>
69 #include <linux/proc_fs.h>
70 #include <linux/seq_file.h>
71 #include <linux/smp.h>
72 #include <linux/if_ether.h>
73 #include <net/arp.h>
74 #include <linux/mii.h>
75 #include <linux/ethtool.h>
76 #include <linux/if_vlan.h>
77 #include <linux/if_bonding.h>
78 #include <linux/jiffies.h>
79 #include <net/route.h>
80 #include <net/net_namespace.h>
81 #include <net/netns/generic.h>
82 #include "bonding.h"
83 #include "bond_3ad.h"
84 #include "bond_alb.h"
85
86 /*---------------------------- Module parameters ----------------------------*/
87
88 /* monitor all links that often (in milliseconds). <=0 disables monitoring */
89 #define BOND_LINK_MON_INTERV    0
90 #define BOND_LINK_ARP_INTERV    0
91
92 static int max_bonds    = BOND_DEFAULT_MAX_BONDS;
93 static int tx_queues    = BOND_DEFAULT_TX_QUEUES;
94 static int num_grat_arp = 1;
95 static int num_unsol_na = 1;
96 static int miimon       = BOND_LINK_MON_INTERV;
97 static int updelay;
98 static int downdelay;
99 static int use_carrier  = 1;
100 static char *mode;
101 static char *primary;
102 static char *primary_reselect;
103 static char *lacp_rate;
104 static char *ad_select;
105 static char *xmit_hash_policy;
106 static int arp_interval = BOND_LINK_ARP_INTERV;
107 static char *arp_ip_target[BOND_MAX_ARP_TARGETS];
108 static char *arp_validate;
109 static char *fail_over_mac;
110 static int all_slaves_active = 0;
111 static struct bond_params bonding_defaults;
112
113 module_param(max_bonds, int, 0);
114 MODULE_PARM_DESC(max_bonds, "Max number of bonded devices");
115 module_param(tx_queues, int, 0);
116 MODULE_PARM_DESC(tx_queues, "Max number of transmit queues (default = 16)");
117 module_param(num_grat_arp, int, 0644);
118 MODULE_PARM_DESC(num_grat_arp, "Number of gratuitous ARP packets to send on failover event");
119 module_param(num_unsol_na, int, 0644);
120 MODULE_PARM_DESC(num_unsol_na, "Number of unsolicited IPv6 Neighbor Advertisements packets to send on failover event");
121 module_param(miimon, int, 0);
122 MODULE_PARM_DESC(miimon, "Link check interval in milliseconds");
123 module_param(updelay, int, 0);
124 MODULE_PARM_DESC(updelay, "Delay before considering link up, in milliseconds");
125 module_param(downdelay, int, 0);
126 MODULE_PARM_DESC(downdelay, "Delay before considering link down, "
127                             "in milliseconds");
128 module_param(use_carrier, int, 0);
129 MODULE_PARM_DESC(use_carrier, "Use netif_carrier_ok (vs MII ioctls) in miimon; "
130                               "0 for off, 1 for on (default)");
131 module_param(mode, charp, 0);
132 MODULE_PARM_DESC(mode, "Mode of operation : 0 for balance-rr, "
133                        "1 for active-backup, 2 for balance-xor, "
134                        "3 for broadcast, 4 for 802.3ad, 5 for balance-tlb, "
135                        "6 for balance-alb");
136 module_param(primary, charp, 0);
137 MODULE_PARM_DESC(primary, "Primary network device to use");
138 module_param(primary_reselect, charp, 0);
139 MODULE_PARM_DESC(primary_reselect, "Reselect primary slave "
140                                    "once it comes up; "
141                                    "0 for always (default), "
142                                    "1 for only if speed of primary is "
143                                    "better, "
144                                    "2 for only on active slave "
145                                    "failure");
146 module_param(lacp_rate, charp, 0);
147 MODULE_PARM_DESC(lacp_rate, "LACPDU tx rate to request from 802.3ad partner "
148                             "(slow/fast)");
149 module_param(ad_select, charp, 0);
150 MODULE_PARM_DESC(ad_select, "803.ad aggregation selection logic: stable (0, default), bandwidth (1), count (2)");
151 module_param(xmit_hash_policy, charp, 0);
152 MODULE_PARM_DESC(xmit_hash_policy, "XOR hashing method: 0 for layer 2 (default)"
153                                    ", 1 for layer 3+4");
154 module_param(arp_interval, int, 0);
155 MODULE_PARM_DESC(arp_interval, "arp interval in milliseconds");
156 module_param_array(arp_ip_target, charp, NULL, 0);
157 MODULE_PARM_DESC(arp_ip_target, "arp targets in n.n.n.n form");
158 module_param(arp_validate, charp, 0);
159 MODULE_PARM_DESC(arp_validate, "validate src/dst of ARP probes: none (default), active, backup or all");
160 module_param(fail_over_mac, charp, 0);
161 MODULE_PARM_DESC(fail_over_mac, "For active-backup, do not set all slaves to the same MAC.  none (default), active or follow");
162 module_param(all_slaves_active, int, 0);
163 MODULE_PARM_DESC(all_slaves_active, "Keep all frames received on an interface"
164                                      "by setting active flag for all slaves.  "
165                                      "0 for never (default), 1 for always.");
166
167 /*----------------------------- Global variables ----------------------------*/
168
169 static const char * const version =
170         DRV_DESCRIPTION ": v" DRV_VERSION " (" DRV_RELDATE ")\n";
171
172 int bond_net_id __read_mostly;
173
174 static __be32 arp_target[BOND_MAX_ARP_TARGETS];
175 static int arp_ip_count;
176 static int bond_mode    = BOND_MODE_ROUNDROBIN;
177 static int xmit_hashtype = BOND_XMIT_POLICY_LAYER2;
178 static int lacp_fast;
179
180
181 const struct bond_parm_tbl bond_lacp_tbl[] = {
182 {       "slow",         AD_LACP_SLOW},
183 {       "fast",         AD_LACP_FAST},
184 {       NULL,           -1},
185 };
186
187 const struct bond_parm_tbl bond_mode_tbl[] = {
188 {       "balance-rr",           BOND_MODE_ROUNDROBIN},
189 {       "active-backup",        BOND_MODE_ACTIVEBACKUP},
190 {       "balance-xor",          BOND_MODE_XOR},
191 {       "broadcast",            BOND_MODE_BROADCAST},
192 {       "802.3ad",              BOND_MODE_8023AD},
193 {       "balance-tlb",          BOND_MODE_TLB},
194 {       "balance-alb",          BOND_MODE_ALB},
195 {       NULL,                   -1},
196 };
197
198 const struct bond_parm_tbl xmit_hashtype_tbl[] = {
199 {       "layer2",               BOND_XMIT_POLICY_LAYER2},
200 {       "layer3+4",             BOND_XMIT_POLICY_LAYER34},
201 {       "layer2+3",             BOND_XMIT_POLICY_LAYER23},
202 {       NULL,                   -1},
203 };
204
205 const struct bond_parm_tbl arp_validate_tbl[] = {
206 {       "none",                 BOND_ARP_VALIDATE_NONE},
207 {       "active",               BOND_ARP_VALIDATE_ACTIVE},
208 {       "backup",               BOND_ARP_VALIDATE_BACKUP},
209 {       "all",                  BOND_ARP_VALIDATE_ALL},
210 {       NULL,                   -1},
211 };
212
213 const struct bond_parm_tbl fail_over_mac_tbl[] = {
214 {       "none",                 BOND_FOM_NONE},
215 {       "active",               BOND_FOM_ACTIVE},
216 {       "follow",               BOND_FOM_FOLLOW},
217 {       NULL,                   -1},
218 };
219
220 const struct bond_parm_tbl pri_reselect_tbl[] = {
221 {       "always",               BOND_PRI_RESELECT_ALWAYS},
222 {       "better",               BOND_PRI_RESELECT_BETTER},
223 {       "failure",              BOND_PRI_RESELECT_FAILURE},
224 {       NULL,                   -1},
225 };
226
227 struct bond_parm_tbl ad_select_tbl[] = {
228 {       "stable",       BOND_AD_STABLE},
229 {       "bandwidth",    BOND_AD_BANDWIDTH},
230 {       "count",        BOND_AD_COUNT},
231 {       NULL,           -1},
232 };
233
234 /*-------------------------- Forward declarations ---------------------------*/
235
236 static void bond_send_gratuitous_arp(struct bonding *bond);
237 static int bond_init(struct net_device *bond_dev);
238 static void bond_uninit(struct net_device *bond_dev);
239
240 /*---------------------------- General routines -----------------------------*/
241
242 static const char *bond_mode_name(int mode)
243 {
244         static const char *names[] = {
245                 [BOND_MODE_ROUNDROBIN] = "load balancing (round-robin)",
246                 [BOND_MODE_ACTIVEBACKUP] = "fault-tolerance (active-backup)",
247                 [BOND_MODE_XOR] = "load balancing (xor)",
248                 [BOND_MODE_BROADCAST] = "fault-tolerance (broadcast)",
249                 [BOND_MODE_8023AD] = "IEEE 802.3ad Dynamic link aggregation",
250                 [BOND_MODE_TLB] = "transmit load balancing",
251                 [BOND_MODE_ALB] = "adaptive load balancing",
252         };
253
254         if (mode < 0 || mode > BOND_MODE_ALB)
255                 return "unknown";
256
257         return names[mode];
258 }
259
260 /*---------------------------------- VLAN -----------------------------------*/
261
262 /**
263  * bond_add_vlan - add a new vlan id on bond
264  * @bond: bond that got the notification
265  * @vlan_id: the vlan id to add
266  *
267  * Returns -ENOMEM if allocation failed.
268  */
269 static int bond_add_vlan(struct bonding *bond, unsigned short vlan_id)
270 {
271         struct vlan_entry *vlan;
272
273         pr_debug("bond: %s, vlan id %d\n",
274                  (bond ? bond->dev->name : "None"), vlan_id);
275
276         vlan = kzalloc(sizeof(struct vlan_entry), GFP_KERNEL);
277         if (!vlan)
278                 return -ENOMEM;
279
280         INIT_LIST_HEAD(&vlan->vlan_list);
281         vlan->vlan_id = vlan_id;
282
283         write_lock_bh(&bond->lock);
284
285         list_add_tail(&vlan->vlan_list, &bond->vlan_list);
286
287         write_unlock_bh(&bond->lock);
288
289         pr_debug("added VLAN ID %d on bond %s\n", vlan_id, bond->dev->name);
290
291         return 0;
292 }
293
294 /**
295  * bond_del_vlan - delete a vlan id from bond
296  * @bond: bond that got the notification
297  * @vlan_id: the vlan id to delete
298  *
299  * returns -ENODEV if @vlan_id was not found in @bond.
300  */
301 static int bond_del_vlan(struct bonding *bond, unsigned short vlan_id)
302 {
303         struct vlan_entry *vlan;
304         int res = -ENODEV;
305
306         pr_debug("bond: %s, vlan id %d\n", bond->dev->name, vlan_id);
307
308         write_lock_bh(&bond->lock);
309
310         list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
311                 if (vlan->vlan_id == vlan_id) {
312                         list_del(&vlan->vlan_list);
313
314                         if (bond_is_lb(bond))
315                                 bond_alb_clear_vlan(bond, vlan_id);
316
317                         pr_debug("removed VLAN ID %d from bond %s\n",
318                                  vlan_id, bond->dev->name);
319
320                         kfree(vlan);
321
322                         if (list_empty(&bond->vlan_list) &&
323                             (bond->slave_cnt == 0)) {
324                                 /* Last VLAN removed and no slaves, so
325                                  * restore block on adding VLANs. This will
326                                  * be removed once new slaves that are not
327                                  * VLAN challenged will be added.
328                                  */
329                                 bond->dev->features |= NETIF_F_VLAN_CHALLENGED;
330                         }
331
332                         res = 0;
333                         goto out;
334                 }
335         }
336
337         pr_debug("couldn't find VLAN ID %d in bond %s\n",
338                  vlan_id, bond->dev->name);
339
340 out:
341         write_unlock_bh(&bond->lock);
342         return res;
343 }
344
345 /**
346  * bond_has_challenged_slaves
347  * @bond: the bond we're working on
348  *
349  * Searches the slave list. Returns 1 if a vlan challenged slave
350  * was found, 0 otherwise.
351  *
352  * Assumes bond->lock is held.
353  */
354 static int bond_has_challenged_slaves(struct bonding *bond)
355 {
356         struct slave *slave;
357         int i;
358
359         bond_for_each_slave(bond, slave, i) {
360                 if (slave->dev->features & NETIF_F_VLAN_CHALLENGED) {
361                         pr_debug("found VLAN challenged slave - %s\n",
362                                  slave->dev->name);
363                         return 1;
364                 }
365         }
366
367         pr_debug("no VLAN challenged slaves found\n");
368         return 0;
369 }
370
371 /**
372  * bond_next_vlan - safely skip to the next item in the vlans list.
373  * @bond: the bond we're working on
374  * @curr: item we're advancing from
375  *
376  * Returns %NULL if list is empty, bond->next_vlan if @curr is %NULL,
377  * or @curr->next otherwise (even if it is @curr itself again).
378  *
379  * Caller must hold bond->lock
380  */
381 struct vlan_entry *bond_next_vlan(struct bonding *bond, struct vlan_entry *curr)
382 {
383         struct vlan_entry *next, *last;
384
385         if (list_empty(&bond->vlan_list))
386                 return NULL;
387
388         if (!curr) {
389                 next = list_entry(bond->vlan_list.next,
390                                   struct vlan_entry, vlan_list);
391         } else {
392                 last = list_entry(bond->vlan_list.prev,
393                                   struct vlan_entry, vlan_list);
394                 if (last == curr) {
395                         next = list_entry(bond->vlan_list.next,
396                                           struct vlan_entry, vlan_list);
397                 } else {
398                         next = list_entry(curr->vlan_list.next,
399                                           struct vlan_entry, vlan_list);
400                 }
401         }
402
403         return next;
404 }
405
406 /**
407  * bond_dev_queue_xmit - Prepare skb for xmit.
408  *
409  * @bond: bond device that got this skb for tx.
410  * @skb: hw accel VLAN tagged skb to transmit
411  * @slave_dev: slave that is supposed to xmit this skbuff
412  *
413  * When the bond gets an skb to transmit that is
414  * already hardware accelerated VLAN tagged, and it
415  * needs to relay this skb to a slave that is not
416  * hw accel capable, the skb needs to be "unaccelerated",
417  * i.e. strip the hwaccel tag and re-insert it as part
418  * of the payload.
419  */
420 int bond_dev_queue_xmit(struct bonding *bond, struct sk_buff *skb,
421                         struct net_device *slave_dev)
422 {
423         unsigned short uninitialized_var(vlan_id);
424
425         if (!list_empty(&bond->vlan_list) &&
426             !(slave_dev->features & NETIF_F_HW_VLAN_TX) &&
427             vlan_get_tag(skb, &vlan_id) == 0) {
428                 skb->dev = slave_dev;
429                 skb = vlan_put_tag(skb, vlan_id);
430                 if (!skb) {
431                         /* vlan_put_tag() frees the skb in case of error,
432                          * so return success here so the calling functions
433                          * won't attempt to free is again.
434                          */
435                         return 0;
436                 }
437         } else {
438                 skb->dev = slave_dev;
439         }
440
441         skb->priority = 1;
442 #ifdef CONFIG_NET_POLL_CONTROLLER
443         if (unlikely(bond->dev->priv_flags & IFF_IN_NETPOLL)) {
444                 struct netpoll *np = bond->dev->npinfo->netpoll;
445                 slave_dev->npinfo = bond->dev->npinfo;
446                 np->real_dev = np->dev = skb->dev;
447                 slave_dev->priv_flags |= IFF_IN_NETPOLL;
448                 netpoll_send_skb(np, skb);
449                 slave_dev->priv_flags &= ~IFF_IN_NETPOLL;
450                 np->dev = bond->dev;
451         } else
452 #endif
453                 dev_queue_xmit(skb);
454
455         return 0;
456 }
457
458 /*
459  * In the following 3 functions, bond_vlan_rx_register(), bond_vlan_rx_add_vid
460  * and bond_vlan_rx_kill_vid, We don't protect the slave list iteration with a
461  * lock because:
462  * a. This operation is performed in IOCTL context,
463  * b. The operation is protected by the RTNL semaphore in the 8021q code,
464  * c. Holding a lock with BH disabled while directly calling a base driver
465  *    entry point is generally a BAD idea.
466  *
467  * The design of synchronization/protection for this operation in the 8021q
468  * module is good for one or more VLAN devices over a single physical device
469  * and cannot be extended for a teaming solution like bonding, so there is a
470  * potential race condition here where a net device from the vlan group might
471  * be referenced (either by a base driver or the 8021q code) while it is being
472  * removed from the system. However, it turns out we're not making matters
473  * worse, and if it works for regular VLAN usage it will work here too.
474 */
475
476 /**
477  * bond_vlan_rx_register - Propagates registration to slaves
478  * @bond_dev: bonding net device that got called
479  * @grp: vlan group being registered
480  */
481 static void bond_vlan_rx_register(struct net_device *bond_dev,
482                                   struct vlan_group *grp)
483 {
484         struct bonding *bond = netdev_priv(bond_dev);
485         struct slave *slave;
486         int i;
487
488         bond->vlgrp = grp;
489
490         bond_for_each_slave(bond, slave, i) {
491                 struct net_device *slave_dev = slave->dev;
492                 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
493
494                 if ((slave_dev->features & NETIF_F_HW_VLAN_RX) &&
495                     slave_ops->ndo_vlan_rx_register) {
496                         slave_ops->ndo_vlan_rx_register(slave_dev, grp);
497                 }
498         }
499 }
500
501 /**
502  * bond_vlan_rx_add_vid - Propagates adding an id to slaves
503  * @bond_dev: bonding net device that got called
504  * @vid: vlan id being added
505  */
506 static void bond_vlan_rx_add_vid(struct net_device *bond_dev, uint16_t vid)
507 {
508         struct bonding *bond = netdev_priv(bond_dev);
509         struct slave *slave;
510         int i, res;
511
512         bond_for_each_slave(bond, slave, i) {
513                 struct net_device *slave_dev = slave->dev;
514                 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
515
516                 if ((slave_dev->features & NETIF_F_HW_VLAN_FILTER) &&
517                     slave_ops->ndo_vlan_rx_add_vid) {
518                         slave_ops->ndo_vlan_rx_add_vid(slave_dev, vid);
519                 }
520         }
521
522         res = bond_add_vlan(bond, vid);
523         if (res) {
524                 pr_err("%s: Error: Failed to add vlan id %d\n",
525                        bond_dev->name, vid);
526         }
527 }
528
529 /**
530  * bond_vlan_rx_kill_vid - Propagates deleting an id to slaves
531  * @bond_dev: bonding net device that got called
532  * @vid: vlan id being removed
533  */
534 static void bond_vlan_rx_kill_vid(struct net_device *bond_dev, uint16_t vid)
535 {
536         struct bonding *bond = netdev_priv(bond_dev);
537         struct slave *slave;
538         struct net_device *vlan_dev;
539         int i, res;
540
541         bond_for_each_slave(bond, slave, i) {
542                 struct net_device *slave_dev = slave->dev;
543                 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
544
545                 if ((slave_dev->features & NETIF_F_HW_VLAN_FILTER) &&
546                     slave_ops->ndo_vlan_rx_kill_vid) {
547                         /* Save and then restore vlan_dev in the grp array,
548                          * since the slave's driver might clear it.
549                          */
550                         vlan_dev = vlan_group_get_device(bond->vlgrp, vid);
551                         slave_ops->ndo_vlan_rx_kill_vid(slave_dev, vid);
552                         vlan_group_set_device(bond->vlgrp, vid, vlan_dev);
553                 }
554         }
555
556         res = bond_del_vlan(bond, vid);
557         if (res) {
558                 pr_err("%s: Error: Failed to remove vlan id %d\n",
559                        bond_dev->name, vid);
560         }
561 }
562
563 static void bond_add_vlans_on_slave(struct bonding *bond, struct net_device *slave_dev)
564 {
565         struct vlan_entry *vlan;
566         const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
567
568         write_lock_bh(&bond->lock);
569
570         if (list_empty(&bond->vlan_list))
571                 goto out;
572
573         if ((slave_dev->features & NETIF_F_HW_VLAN_RX) &&
574             slave_ops->ndo_vlan_rx_register)
575                 slave_ops->ndo_vlan_rx_register(slave_dev, bond->vlgrp);
576
577         if (!(slave_dev->features & NETIF_F_HW_VLAN_FILTER) ||
578             !(slave_ops->ndo_vlan_rx_add_vid))
579                 goto out;
580
581         list_for_each_entry(vlan, &bond->vlan_list, vlan_list)
582                 slave_ops->ndo_vlan_rx_add_vid(slave_dev, vlan->vlan_id);
583
584 out:
585         write_unlock_bh(&bond->lock);
586 }
587
588 static void bond_del_vlans_from_slave(struct bonding *bond,
589                                       struct net_device *slave_dev)
590 {
591         const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
592         struct vlan_entry *vlan;
593         struct net_device *vlan_dev;
594
595         write_lock_bh(&bond->lock);
596
597         if (list_empty(&bond->vlan_list))
598                 goto out;
599
600         if (!(slave_dev->features & NETIF_F_HW_VLAN_FILTER) ||
601             !(slave_ops->ndo_vlan_rx_kill_vid))
602                 goto unreg;
603
604         list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
605                 /* Save and then restore vlan_dev in the grp array,
606                  * since the slave's driver might clear it.
607                  */
608                 vlan_dev = vlan_group_get_device(bond->vlgrp, vlan->vlan_id);
609                 slave_ops->ndo_vlan_rx_kill_vid(slave_dev, vlan->vlan_id);
610                 vlan_group_set_device(bond->vlgrp, vlan->vlan_id, vlan_dev);
611         }
612
613 unreg:
614         if ((slave_dev->features & NETIF_F_HW_VLAN_RX) &&
615             slave_ops->ndo_vlan_rx_register)
616                 slave_ops->ndo_vlan_rx_register(slave_dev, NULL);
617
618 out:
619         write_unlock_bh(&bond->lock);
620 }
621
622 /*------------------------------- Link status -------------------------------*/
623
624 /*
625  * Set the carrier state for the master according to the state of its
626  * slaves.  If any slaves are up, the master is up.  In 802.3ad mode,
627  * do special 802.3ad magic.
628  *
629  * Returns zero if carrier state does not change, nonzero if it does.
630  */
631 static int bond_set_carrier(struct bonding *bond)
632 {
633         struct slave *slave;
634         int i;
635
636         if (bond->slave_cnt == 0)
637                 goto down;
638
639         if (bond->params.mode == BOND_MODE_8023AD)
640                 return bond_3ad_set_carrier(bond);
641
642         bond_for_each_slave(bond, slave, i) {
643                 if (slave->link == BOND_LINK_UP) {
644                         if (!netif_carrier_ok(bond->dev)) {
645                                 netif_carrier_on(bond->dev);
646                                 return 1;
647                         }
648                         return 0;
649                 }
650         }
651
652 down:
653         if (netif_carrier_ok(bond->dev)) {
654                 netif_carrier_off(bond->dev);
655                 return 1;
656         }
657         return 0;
658 }
659
660 /*
661  * Get link speed and duplex from the slave's base driver
662  * using ethtool. If for some reason the call fails or the
663  * values are invalid, fake speed and duplex to 100/Full
664  * and return error.
665  */
666 static int bond_update_speed_duplex(struct slave *slave)
667 {
668         struct net_device *slave_dev = slave->dev;
669         struct ethtool_cmd etool;
670         int res;
671
672         /* Fake speed and duplex */
673         slave->speed = SPEED_100;
674         slave->duplex = DUPLEX_FULL;
675
676         if (!slave_dev->ethtool_ops || !slave_dev->ethtool_ops->get_settings)
677                 return -1;
678
679         res = slave_dev->ethtool_ops->get_settings(slave_dev, &etool);
680         if (res < 0)
681                 return -1;
682
683         switch (etool.speed) {
684         case SPEED_10:
685         case SPEED_100:
686         case SPEED_1000:
687         case SPEED_10000:
688                 break;
689         default:
690                 return -1;
691         }
692
693         switch (etool.duplex) {
694         case DUPLEX_FULL:
695         case DUPLEX_HALF:
696                 break;
697         default:
698                 return -1;
699         }
700
701         slave->speed = etool.speed;
702         slave->duplex = etool.duplex;
703
704         return 0;
705 }
706
707 /*
708  * if <dev> supports MII link status reporting, check its link status.
709  *
710  * We either do MII/ETHTOOL ioctls, or check netif_carrier_ok(),
711  * depending upon the setting of the use_carrier parameter.
712  *
713  * Return either BMSR_LSTATUS, meaning that the link is up (or we
714  * can't tell and just pretend it is), or 0, meaning that the link is
715  * down.
716  *
717  * If reporting is non-zero, instead of faking link up, return -1 if
718  * both ETHTOOL and MII ioctls fail (meaning the device does not
719  * support them).  If use_carrier is set, return whatever it says.
720  * It'd be nice if there was a good way to tell if a driver supports
721  * netif_carrier, but there really isn't.
722  */
723 static int bond_check_dev_link(struct bonding *bond,
724                                struct net_device *slave_dev, int reporting)
725 {
726         const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
727         int (*ioctl)(struct net_device *, struct ifreq *, int);
728         struct ifreq ifr;
729         struct mii_ioctl_data *mii;
730
731         if (!reporting && !netif_running(slave_dev))
732                 return 0;
733
734         if (bond->params.use_carrier)
735                 return netif_carrier_ok(slave_dev) ? BMSR_LSTATUS : 0;
736
737         /* Try to get link status using Ethtool first. */
738         if (slave_dev->ethtool_ops) {
739                 if (slave_dev->ethtool_ops->get_link) {
740                         u32 link;
741
742                         link = slave_dev->ethtool_ops->get_link(slave_dev);
743
744                         return link ? BMSR_LSTATUS : 0;
745                 }
746         }
747
748         /* Ethtool can't be used, fallback to MII ioctls. */
749         ioctl = slave_ops->ndo_do_ioctl;
750         if (ioctl) {
751                 /* TODO: set pointer to correct ioctl on a per team member */
752                 /*       bases to make this more efficient. that is, once  */
753                 /*       we determine the correct ioctl, we will always    */
754                 /*       call it and not the others for that team          */
755                 /*       member.                                           */
756
757                 /*
758                  * We cannot assume that SIOCGMIIPHY will also read a
759                  * register; not all network drivers (e.g., e100)
760                  * support that.
761                  */
762
763                 /* Yes, the mii is overlaid on the ifreq.ifr_ifru */
764                 strncpy(ifr.ifr_name, slave_dev->name, IFNAMSIZ);
765                 mii = if_mii(&ifr);
766                 if (IOCTL(slave_dev, &ifr, SIOCGMIIPHY) == 0) {
767                         mii->reg_num = MII_BMSR;
768                         if (IOCTL(slave_dev, &ifr, SIOCGMIIREG) == 0)
769                                 return mii->val_out & BMSR_LSTATUS;
770                 }
771         }
772
773         /*
774          * If reporting, report that either there's no dev->do_ioctl,
775          * or both SIOCGMIIREG and get_link failed (meaning that we
776          * cannot report link status).  If not reporting, pretend
777          * we're ok.
778          */
779         return reporting ? -1 : BMSR_LSTATUS;
780 }
781
782 /*----------------------------- Multicast list ------------------------------*/
783
784 /*
785  * Push the promiscuity flag down to appropriate slaves
786  */
787 static int bond_set_promiscuity(struct bonding *bond, int inc)
788 {
789         int err = 0;
790         if (USES_PRIMARY(bond->params.mode)) {
791                 /* write lock already acquired */
792                 if (bond->curr_active_slave) {
793                         err = dev_set_promiscuity(bond->curr_active_slave->dev,
794                                                   inc);
795                 }
796         } else {
797                 struct slave *slave;
798                 int i;
799                 bond_for_each_slave(bond, slave, i) {
800                         err = dev_set_promiscuity(slave->dev, inc);
801                         if (err)
802                                 return err;
803                 }
804         }
805         return err;
806 }
807
808 /*
809  * Push the allmulti flag down to all slaves
810  */
811 static int bond_set_allmulti(struct bonding *bond, int inc)
812 {
813         int err = 0;
814         if (USES_PRIMARY(bond->params.mode)) {
815                 /* write lock already acquired */
816                 if (bond->curr_active_slave) {
817                         err = dev_set_allmulti(bond->curr_active_slave->dev,
818                                                inc);
819                 }
820         } else {
821                 struct slave *slave;
822                 int i;
823                 bond_for_each_slave(bond, slave, i) {
824                         err = dev_set_allmulti(slave->dev, inc);
825                         if (err)
826                                 return err;
827                 }
828         }
829         return err;
830 }
831
832 /*
833  * Add a Multicast address to slaves
834  * according to mode
835  */
836 static void bond_mc_add(struct bonding *bond, void *addr)
837 {
838         if (USES_PRIMARY(bond->params.mode)) {
839                 /* write lock already acquired */
840                 if (bond->curr_active_slave)
841                         dev_mc_add(bond->curr_active_slave->dev, addr);
842         } else {
843                 struct slave *slave;
844                 int i;
845
846                 bond_for_each_slave(bond, slave, i)
847                         dev_mc_add(slave->dev, addr);
848         }
849 }
850
851 /*
852  * Remove a multicast address from slave
853  * according to mode
854  */
855 static void bond_mc_del(struct bonding *bond, void *addr)
856 {
857         if (USES_PRIMARY(bond->params.mode)) {
858                 /* write lock already acquired */
859                 if (bond->curr_active_slave)
860                         dev_mc_del(bond->curr_active_slave->dev, addr);
861         } else {
862                 struct slave *slave;
863                 int i;
864                 bond_for_each_slave(bond, slave, i) {
865                         dev_mc_del(slave->dev, addr);
866                 }
867         }
868 }
869
870
871 /*
872  * Retrieve the list of registered multicast addresses for the bonding
873  * device and retransmit an IGMP JOIN request to the current active
874  * slave.
875  */
876 static void bond_resend_igmp_join_requests(struct bonding *bond)
877 {
878         struct in_device *in_dev;
879         struct ip_mc_list *im;
880
881         rcu_read_lock();
882         in_dev = __in_dev_get_rcu(bond->dev);
883         if (in_dev) {
884                 for (im = in_dev->mc_list; im; im = im->next)
885                         ip_mc_rejoin_group(im);
886         }
887
888         rcu_read_unlock();
889 }
890
891 /*
892  * flush all members of flush->mc_list from device dev->mc_list
893  */
894 static void bond_mc_list_flush(struct net_device *bond_dev,
895                                struct net_device *slave_dev)
896 {
897         struct bonding *bond = netdev_priv(bond_dev);
898         struct netdev_hw_addr *ha;
899
900         netdev_for_each_mc_addr(ha, bond_dev)
901                 dev_mc_del(slave_dev, ha->addr);
902
903         if (bond->params.mode == BOND_MODE_8023AD) {
904                 /* del lacpdu mc addr from mc list */
905                 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
906
907                 dev_mc_del(slave_dev, lacpdu_multicast);
908         }
909 }
910
911 /*--------------------------- Active slave change ---------------------------*/
912
913 /*
914  * Update the mc list and multicast-related flags for the new and
915  * old active slaves (if any) according to the multicast mode, and
916  * promiscuous flags unconditionally.
917  */
918 static void bond_mc_swap(struct bonding *bond, struct slave *new_active,
919                          struct slave *old_active)
920 {
921         struct netdev_hw_addr *ha;
922
923         if (!USES_PRIMARY(bond->params.mode))
924                 /* nothing to do -  mc list is already up-to-date on
925                  * all slaves
926                  */
927                 return;
928
929         if (old_active) {
930                 if (bond->dev->flags & IFF_PROMISC)
931                         dev_set_promiscuity(old_active->dev, -1);
932
933                 if (bond->dev->flags & IFF_ALLMULTI)
934                         dev_set_allmulti(old_active->dev, -1);
935
936                 netdev_for_each_mc_addr(ha, bond->dev)
937                         dev_mc_del(old_active->dev, ha->addr);
938         }
939
940         if (new_active) {
941                 /* FIXME: Signal errors upstream. */
942                 if (bond->dev->flags & IFF_PROMISC)
943                         dev_set_promiscuity(new_active->dev, 1);
944
945                 if (bond->dev->flags & IFF_ALLMULTI)
946                         dev_set_allmulti(new_active->dev, 1);
947
948                 netdev_for_each_mc_addr(ha, bond->dev)
949                         dev_mc_add(new_active->dev, ha->addr);
950                 bond_resend_igmp_join_requests(bond);
951         }
952 }
953
954 /*
955  * bond_do_fail_over_mac
956  *
957  * Perform special MAC address swapping for fail_over_mac settings
958  *
959  * Called with RTNL, bond->lock for read, curr_slave_lock for write_bh.
960  */
961 static void bond_do_fail_over_mac(struct bonding *bond,
962                                   struct slave *new_active,
963                                   struct slave *old_active)
964         __releases(&bond->curr_slave_lock)
965         __releases(&bond->lock)
966         __acquires(&bond->lock)
967         __acquires(&bond->curr_slave_lock)
968 {
969         u8 tmp_mac[ETH_ALEN];
970         struct sockaddr saddr;
971         int rv;
972
973         switch (bond->params.fail_over_mac) {
974         case BOND_FOM_ACTIVE:
975                 if (new_active)
976                         memcpy(bond->dev->dev_addr,  new_active->dev->dev_addr,
977                                new_active->dev->addr_len);
978                 break;
979         case BOND_FOM_FOLLOW:
980                 /*
981                  * if new_active && old_active, swap them
982                  * if just old_active, do nothing (going to no active slave)
983                  * if just new_active, set new_active to bond's MAC
984                  */
985                 if (!new_active)
986                         return;
987
988                 write_unlock_bh(&bond->curr_slave_lock);
989                 read_unlock(&bond->lock);
990
991                 if (old_active) {
992                         memcpy(tmp_mac, new_active->dev->dev_addr, ETH_ALEN);
993                         memcpy(saddr.sa_data, old_active->dev->dev_addr,
994                                ETH_ALEN);
995                         saddr.sa_family = new_active->dev->type;
996                 } else {
997                         memcpy(saddr.sa_data, bond->dev->dev_addr, ETH_ALEN);
998                         saddr.sa_family = bond->dev->type;
999                 }
1000
1001                 rv = dev_set_mac_address(new_active->dev, &saddr);
1002                 if (rv) {
1003                         pr_err("%s: Error %d setting MAC of slave %s\n",
1004                                bond->dev->name, -rv, new_active->dev->name);
1005                         goto out;
1006                 }
1007
1008                 if (!old_active)
1009                         goto out;
1010
1011                 memcpy(saddr.sa_data, tmp_mac, ETH_ALEN);
1012                 saddr.sa_family = old_active->dev->type;
1013
1014                 rv = dev_set_mac_address(old_active->dev, &saddr);
1015                 if (rv)
1016                         pr_err("%s: Error %d setting MAC of slave %s\n",
1017                                bond->dev->name, -rv, new_active->dev->name);
1018 out:
1019                 read_lock(&bond->lock);
1020                 write_lock_bh(&bond->curr_slave_lock);
1021                 break;
1022         default:
1023                 pr_err("%s: bond_do_fail_over_mac impossible: bad policy %d\n",
1024                        bond->dev->name, bond->params.fail_over_mac);
1025                 break;
1026         }
1027
1028 }
1029
1030 static bool bond_should_change_active(struct bonding *bond)
1031 {
1032         struct slave *prim = bond->primary_slave;
1033         struct slave *curr = bond->curr_active_slave;
1034
1035         if (!prim || !curr || curr->link != BOND_LINK_UP)
1036                 return true;
1037         if (bond->force_primary) {
1038                 bond->force_primary = false;
1039                 return true;
1040         }
1041         if (bond->params.primary_reselect == BOND_PRI_RESELECT_BETTER &&
1042             (prim->speed < curr->speed ||
1043              (prim->speed == curr->speed && prim->duplex <= curr->duplex)))
1044                 return false;
1045         if (bond->params.primary_reselect == BOND_PRI_RESELECT_FAILURE)
1046                 return false;
1047         return true;
1048 }
1049
1050 /**
1051  * find_best_interface - select the best available slave to be the active one
1052  * @bond: our bonding struct
1053  *
1054  * Warning: Caller must hold curr_slave_lock for writing.
1055  */
1056 static struct slave *bond_find_best_slave(struct bonding *bond)
1057 {
1058         struct slave *new_active, *old_active;
1059         struct slave *bestslave = NULL;
1060         int mintime = bond->params.updelay;
1061         int i;
1062
1063         new_active = bond->curr_active_slave;
1064
1065         if (!new_active) { /* there were no active slaves left */
1066                 if (bond->slave_cnt > 0)   /* found one slave */
1067                         new_active = bond->first_slave;
1068                 else
1069                         return NULL; /* still no slave, return NULL */
1070         }
1071
1072         if ((bond->primary_slave) &&
1073             bond->primary_slave->link == BOND_LINK_UP &&
1074             bond_should_change_active(bond)) {
1075                 new_active = bond->primary_slave;
1076         }
1077
1078         /* remember where to stop iterating over the slaves */
1079         old_active = new_active;
1080
1081         bond_for_each_slave_from(bond, new_active, i, old_active) {
1082                 if (new_active->link == BOND_LINK_UP) {
1083                         return new_active;
1084                 } else if (new_active->link == BOND_LINK_BACK &&
1085                            IS_UP(new_active->dev)) {
1086                         /* link up, but waiting for stabilization */
1087                         if (new_active->delay < mintime) {
1088                                 mintime = new_active->delay;
1089                                 bestslave = new_active;
1090                         }
1091                 }
1092         }
1093
1094         return bestslave;
1095 }
1096
1097 /**
1098  * change_active_interface - change the active slave into the specified one
1099  * @bond: our bonding struct
1100  * @new: the new slave to make the active one
1101  *
1102  * Set the new slave to the bond's settings and unset them on the old
1103  * curr_active_slave.
1104  * Setting include flags, mc-list, promiscuity, allmulti, etc.
1105  *
1106  * If @new's link state is %BOND_LINK_BACK we'll set it to %BOND_LINK_UP,
1107  * because it is apparently the best available slave we have, even though its
1108  * updelay hasn't timed out yet.
1109  *
1110  * If new_active is not NULL, caller must hold bond->lock for read and
1111  * curr_slave_lock for write_bh.
1112  */
1113 void bond_change_active_slave(struct bonding *bond, struct slave *new_active)
1114 {
1115         struct slave *old_active = bond->curr_active_slave;
1116
1117         if (old_active == new_active)
1118                 return;
1119
1120         if (new_active) {
1121                 new_active->jiffies = jiffies;
1122
1123                 if (new_active->link == BOND_LINK_BACK) {
1124                         if (USES_PRIMARY(bond->params.mode)) {
1125                                 pr_info("%s: making interface %s the new active one %d ms earlier.\n",
1126                                         bond->dev->name, new_active->dev->name,
1127                                         (bond->params.updelay - new_active->delay) * bond->params.miimon);
1128                         }
1129
1130                         new_active->delay = 0;
1131                         new_active->link = BOND_LINK_UP;
1132
1133                         if (bond->params.mode == BOND_MODE_8023AD)
1134                                 bond_3ad_handle_link_change(new_active, BOND_LINK_UP);
1135
1136                         if (bond_is_lb(bond))
1137                                 bond_alb_handle_link_change(bond, new_active, BOND_LINK_UP);
1138                 } else {
1139                         if (USES_PRIMARY(bond->params.mode)) {
1140                                 pr_info("%s: making interface %s the new active one.\n",
1141                                         bond->dev->name, new_active->dev->name);
1142                         }
1143                 }
1144         }
1145
1146         if (USES_PRIMARY(bond->params.mode))
1147                 bond_mc_swap(bond, new_active, old_active);
1148
1149         if (bond_is_lb(bond)) {
1150                 bond_alb_handle_active_change(bond, new_active);
1151                 if (old_active)
1152                         bond_set_slave_inactive_flags(old_active);
1153                 if (new_active)
1154                         bond_set_slave_active_flags(new_active);
1155         } else {
1156                 bond->curr_active_slave = new_active;
1157         }
1158
1159         if (bond->params.mode == BOND_MODE_ACTIVEBACKUP) {
1160                 if (old_active)
1161                         bond_set_slave_inactive_flags(old_active);
1162
1163                 if (new_active) {
1164                         bond_set_slave_active_flags(new_active);
1165
1166                         if (bond->params.fail_over_mac)
1167                                 bond_do_fail_over_mac(bond, new_active,
1168                                                       old_active);
1169
1170                         bond->send_grat_arp = bond->params.num_grat_arp;
1171                         bond_send_gratuitous_arp(bond);
1172
1173                         bond->send_unsol_na = bond->params.num_unsol_na;
1174                         bond_send_unsolicited_na(bond);
1175
1176                         write_unlock_bh(&bond->curr_slave_lock);
1177                         read_unlock(&bond->lock);
1178
1179                         netdev_bonding_change(bond->dev, NETDEV_BONDING_FAILOVER);
1180
1181                         read_lock(&bond->lock);
1182                         write_lock_bh(&bond->curr_slave_lock);
1183                 }
1184         }
1185
1186         /* resend IGMP joins since all were sent on curr_active_slave */
1187         if (bond->params.mode == BOND_MODE_ROUNDROBIN) {
1188                 bond_resend_igmp_join_requests(bond);
1189         }
1190 }
1191
1192 /**
1193  * bond_select_active_slave - select a new active slave, if needed
1194  * @bond: our bonding struct
1195  *
1196  * This functions should be called when one of the following occurs:
1197  * - The old curr_active_slave has been released or lost its link.
1198  * - The primary_slave has got its link back.
1199  * - A slave has got its link back and there's no old curr_active_slave.
1200  *
1201  * Caller must hold bond->lock for read and curr_slave_lock for write_bh.
1202  */
1203 void bond_select_active_slave(struct bonding *bond)
1204 {
1205         struct slave *best_slave;
1206         int rv;
1207
1208         best_slave = bond_find_best_slave(bond);
1209         if (best_slave != bond->curr_active_slave) {
1210                 bond_change_active_slave(bond, best_slave);
1211                 rv = bond_set_carrier(bond);
1212                 if (!rv)
1213                         return;
1214
1215                 if (netif_carrier_ok(bond->dev)) {
1216                         pr_info("%s: first active interface up!\n",
1217                                 bond->dev->name);
1218                 } else {
1219                         pr_info("%s: now running without any active interface !\n",
1220                                 bond->dev->name);
1221                 }
1222         }
1223 }
1224
1225 /*--------------------------- slave list handling ---------------------------*/
1226
1227 /*
1228  * This function attaches the slave to the end of list.
1229  *
1230  * bond->lock held for writing by caller.
1231  */
1232 static void bond_attach_slave(struct bonding *bond, struct slave *new_slave)
1233 {
1234         if (bond->first_slave == NULL) { /* attaching the first slave */
1235                 new_slave->next = new_slave;
1236                 new_slave->prev = new_slave;
1237                 bond->first_slave = new_slave;
1238         } else {
1239                 new_slave->next = bond->first_slave;
1240                 new_slave->prev = bond->first_slave->prev;
1241                 new_slave->next->prev = new_slave;
1242                 new_slave->prev->next = new_slave;
1243         }
1244
1245         bond->slave_cnt++;
1246 }
1247
1248 /*
1249  * This function detaches the slave from the list.
1250  * WARNING: no check is made to verify if the slave effectively
1251  * belongs to <bond>.
1252  * Nothing is freed on return, structures are just unchained.
1253  * If any slave pointer in bond was pointing to <slave>,
1254  * it should be changed by the calling function.
1255  *
1256  * bond->lock held for writing by caller.
1257  */
1258 static void bond_detach_slave(struct bonding *bond, struct slave *slave)
1259 {
1260         if (slave->next)
1261                 slave->next->prev = slave->prev;
1262
1263         if (slave->prev)
1264                 slave->prev->next = slave->next;
1265
1266         if (bond->first_slave == slave) { /* slave is the first slave */
1267                 if (bond->slave_cnt > 1) { /* there are more slave */
1268                         bond->first_slave = slave->next;
1269                 } else {
1270                         bond->first_slave = NULL; /* slave was the last one */
1271                 }
1272         }
1273
1274         slave->next = NULL;
1275         slave->prev = NULL;
1276         bond->slave_cnt--;
1277 }
1278
1279 #ifdef CONFIG_NET_POLL_CONTROLLER
1280 /*
1281  * You must hold read lock on bond->lock before calling this.
1282  */
1283 static bool slaves_support_netpoll(struct net_device *bond_dev)
1284 {
1285         struct bonding *bond = netdev_priv(bond_dev);
1286         struct slave *slave;
1287         int i = 0;
1288         bool ret = true;
1289
1290         bond_for_each_slave(bond, slave, i) {
1291                 if ((slave->dev->priv_flags & IFF_DISABLE_NETPOLL) ||
1292                     !slave->dev->netdev_ops->ndo_poll_controller)
1293                         ret = false;
1294         }
1295         return i != 0 && ret;
1296 }
1297
1298 static void bond_poll_controller(struct net_device *bond_dev)
1299 {
1300         struct net_device *dev = bond_dev->npinfo->netpoll->real_dev;
1301         if (dev != bond_dev)
1302                 netpoll_poll_dev(dev);
1303 }
1304
1305 static void bond_netpoll_cleanup(struct net_device *bond_dev)
1306 {
1307         struct bonding *bond = netdev_priv(bond_dev);
1308         struct slave *slave;
1309         const struct net_device_ops *ops;
1310         int i;
1311
1312         read_lock(&bond->lock);
1313         bond_dev->npinfo = NULL;
1314         bond_for_each_slave(bond, slave, i) {
1315                 if (slave->dev) {
1316                         ops = slave->dev->netdev_ops;
1317                         if (ops->ndo_netpoll_cleanup)
1318                                 ops->ndo_netpoll_cleanup(slave->dev);
1319                         else
1320                                 slave->dev->npinfo = NULL;
1321                 }
1322         }
1323         read_unlock(&bond->lock);
1324 }
1325
1326 #else
1327
1328 static void bond_netpoll_cleanup(struct net_device *bond_dev)
1329 {
1330 }
1331
1332 #endif
1333
1334 /*---------------------------------- IOCTL ----------------------------------*/
1335
1336 static int bond_sethwaddr(struct net_device *bond_dev,
1337                           struct net_device *slave_dev)
1338 {
1339         pr_debug("bond_dev=%p\n", bond_dev);
1340         pr_debug("slave_dev=%p\n", slave_dev);
1341         pr_debug("slave_dev->addr_len=%d\n", slave_dev->addr_len);
1342         memcpy(bond_dev->dev_addr, slave_dev->dev_addr, slave_dev->addr_len);
1343         return 0;
1344 }
1345
1346 #define BOND_VLAN_FEATURES \
1347         (NETIF_F_VLAN_CHALLENGED | NETIF_F_HW_VLAN_RX | NETIF_F_HW_VLAN_TX | \
1348          NETIF_F_HW_VLAN_FILTER)
1349
1350 /*
1351  * Compute the common dev->feature set available to all slaves.  Some
1352  * feature bits are managed elsewhere, so preserve those feature bits
1353  * on the master device.
1354  */
1355 static int bond_compute_features(struct bonding *bond)
1356 {
1357         struct slave *slave;
1358         struct net_device *bond_dev = bond->dev;
1359         unsigned long features = bond_dev->features;
1360         unsigned long vlan_features = 0;
1361         unsigned short max_hard_header_len = max((u16)ETH_HLEN,
1362                                                 bond_dev->hard_header_len);
1363         int i;
1364
1365         features &= ~(NETIF_F_ALL_CSUM | BOND_VLAN_FEATURES);
1366         features |=  NETIF_F_GSO_MASK | NETIF_F_NO_CSUM;
1367
1368         if (!bond->first_slave)
1369                 goto done;
1370
1371         features &= ~NETIF_F_ONE_FOR_ALL;
1372
1373         vlan_features = bond->first_slave->dev->vlan_features;
1374         bond_for_each_slave(bond, slave, i) {
1375                 features = netdev_increment_features(features,
1376                                                      slave->dev->features,
1377                                                      NETIF_F_ONE_FOR_ALL);
1378                 vlan_features = netdev_increment_features(vlan_features,
1379                                                         slave->dev->vlan_features,
1380                                                         NETIF_F_ONE_FOR_ALL);
1381                 if (slave->dev->hard_header_len > max_hard_header_len)
1382                         max_hard_header_len = slave->dev->hard_header_len;
1383         }
1384
1385 done:
1386         features |= (bond_dev->features & BOND_VLAN_FEATURES);
1387         bond_dev->features = netdev_fix_features(features, NULL);
1388         bond_dev->vlan_features = netdev_fix_features(vlan_features, NULL);
1389         bond_dev->hard_header_len = max_hard_header_len;
1390
1391         return 0;
1392 }
1393
1394 static void bond_setup_by_slave(struct net_device *bond_dev,
1395                                 struct net_device *slave_dev)
1396 {
1397         struct bonding *bond = netdev_priv(bond_dev);
1398
1399         bond_dev->header_ops        = slave_dev->header_ops;
1400
1401         bond_dev->type              = slave_dev->type;
1402         bond_dev->hard_header_len   = slave_dev->hard_header_len;
1403         bond_dev->addr_len          = slave_dev->addr_len;
1404
1405         memcpy(bond_dev->broadcast, slave_dev->broadcast,
1406                 slave_dev->addr_len);
1407         bond->setup_by_slave = 1;
1408 }
1409
1410 /* enslave device <slave> to bond device <master> */
1411 int bond_enslave(struct net_device *bond_dev, struct net_device *slave_dev)
1412 {
1413         struct bonding *bond = netdev_priv(bond_dev);
1414         const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
1415         struct slave *new_slave = NULL;
1416         struct netdev_hw_addr *ha;
1417         struct sockaddr addr;
1418         int link_reporting;
1419         int old_features = bond_dev->features;
1420         int res = 0;
1421
1422         if (!bond->params.use_carrier && slave_dev->ethtool_ops == NULL &&
1423                 slave_ops->ndo_do_ioctl == NULL) {
1424                 pr_warning("%s: Warning: no link monitoring support for %s\n",
1425                            bond_dev->name, slave_dev->name);
1426         }
1427
1428         /* bond must be initialized by bond_open() before enslaving */
1429         if (!(bond_dev->flags & IFF_UP)) {
1430                 pr_warning("%s: master_dev is not up in bond_enslave\n",
1431                            bond_dev->name);
1432         }
1433
1434         /* already enslaved */
1435         if (slave_dev->flags & IFF_SLAVE) {
1436                 pr_debug("Error, Device was already enslaved\n");
1437                 return -EBUSY;
1438         }
1439
1440         /* vlan challenged mutual exclusion */
1441         /* no need to lock since we're protected by rtnl_lock */
1442         if (slave_dev->features & NETIF_F_VLAN_CHALLENGED) {
1443                 pr_debug("%s: NETIF_F_VLAN_CHALLENGED\n", slave_dev->name);
1444                 if (!list_empty(&bond->vlan_list)) {
1445                         pr_err("%s: Error: cannot enslave VLAN challenged slave %s on VLAN enabled bond %s\n",
1446                                bond_dev->name, slave_dev->name, bond_dev->name);
1447                         return -EPERM;
1448                 } else {
1449                         pr_warning("%s: Warning: enslaved VLAN challenged slave %s. Adding VLANs will be blocked as long as %s is part of bond %s\n",
1450                                    bond_dev->name, slave_dev->name,
1451                                    slave_dev->name, bond_dev->name);
1452                         bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
1453                 }
1454         } else {
1455                 pr_debug("%s: ! NETIF_F_VLAN_CHALLENGED\n", slave_dev->name);
1456                 if (bond->slave_cnt == 0) {
1457                         /* First slave, and it is not VLAN challenged,
1458                          * so remove the block of adding VLANs over the bond.
1459                          */
1460                         bond_dev->features &= ~NETIF_F_VLAN_CHALLENGED;
1461                 }
1462         }
1463
1464         /*
1465          * Old ifenslave binaries are no longer supported.  These can
1466          * be identified with moderate accuracy by the state of the slave:
1467          * the current ifenslave will set the interface down prior to
1468          * enslaving it; the old ifenslave will not.
1469          */
1470         if ((slave_dev->flags & IFF_UP)) {
1471                 pr_err("%s is up. This may be due to an out of date ifenslave.\n",
1472                        slave_dev->name);
1473                 res = -EPERM;
1474                 goto err_undo_flags;
1475         }
1476
1477         /* set bonding device ether type by slave - bonding netdevices are
1478          * created with ether_setup, so when the slave type is not ARPHRD_ETHER
1479          * there is a need to override some of the type dependent attribs/funcs.
1480          *
1481          * bond ether type mutual exclusion - don't allow slaves of dissimilar
1482          * ether type (eg ARPHRD_ETHER and ARPHRD_INFINIBAND) share the same bond
1483          */
1484         if (bond->slave_cnt == 0) {
1485                 if (bond_dev->type != slave_dev->type) {
1486                         pr_debug("%s: change device type from %d to %d\n",
1487                                  bond_dev->name,
1488                                  bond_dev->type, slave_dev->type);
1489
1490                         res = netdev_bonding_change(bond_dev,
1491                                                     NETDEV_PRE_TYPE_CHANGE);
1492                         res = notifier_to_errno(res);
1493                         if (res) {
1494                                 pr_err("%s: refused to change device type\n",
1495                                        bond_dev->name);
1496                                 res = -EBUSY;
1497                                 goto err_undo_flags;
1498                         }
1499
1500                         /* Flush unicast and multicast addresses */
1501                         dev_uc_flush(bond_dev);
1502                         dev_mc_flush(bond_dev);
1503
1504                         if (slave_dev->type != ARPHRD_ETHER)
1505                                 bond_setup_by_slave(bond_dev, slave_dev);
1506                         else
1507                                 ether_setup(bond_dev);
1508
1509                         netdev_bonding_change(bond_dev,
1510                                               NETDEV_POST_TYPE_CHANGE);
1511                 }
1512         } else if (bond_dev->type != slave_dev->type) {
1513                 pr_err("%s ether type (%d) is different from other slaves (%d), can not enslave it.\n",
1514                        slave_dev->name,
1515                        slave_dev->type, bond_dev->type);
1516                 res = -EINVAL;
1517                 goto err_undo_flags;
1518         }
1519
1520         if (slave_ops->ndo_set_mac_address == NULL) {
1521                 if (bond->slave_cnt == 0) {
1522                         pr_warning("%s: Warning: The first slave device specified does not support setting the MAC address. Setting fail_over_mac to active.",
1523                                    bond_dev->name);
1524                         bond->params.fail_over_mac = BOND_FOM_ACTIVE;
1525                 } else if (bond->params.fail_over_mac != BOND_FOM_ACTIVE) {
1526                         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",
1527                                bond_dev->name);
1528                         res = -EOPNOTSUPP;
1529                         goto err_undo_flags;
1530                 }
1531         }
1532
1533         /* If this is the first slave, then we need to set the master's hardware
1534          * address to be the same as the slave's. */
1535         if (bond->slave_cnt == 0)
1536                 memcpy(bond->dev->dev_addr, slave_dev->dev_addr,
1537                        slave_dev->addr_len);
1538
1539
1540         new_slave = kzalloc(sizeof(struct slave), GFP_KERNEL);
1541         if (!new_slave) {
1542                 res = -ENOMEM;
1543                 goto err_undo_flags;
1544         }
1545
1546         /*
1547          * Set the new_slave's queue_id to be zero.  Queue ID mapping
1548          * is set via sysfs or module option if desired.
1549          */
1550         new_slave->queue_id = 0;
1551
1552         /* Save slave's original mtu and then set it to match the bond */
1553         new_slave->original_mtu = slave_dev->mtu;
1554         res = dev_set_mtu(slave_dev, bond->dev->mtu);
1555         if (res) {
1556                 pr_debug("Error %d calling dev_set_mtu\n", res);
1557                 goto err_free;
1558         }
1559
1560         /*
1561          * Save slave's original ("permanent") mac address for modes
1562          * that need it, and for restoring it upon release, and then
1563          * set it to the master's address
1564          */
1565         memcpy(new_slave->perm_hwaddr, slave_dev->dev_addr, ETH_ALEN);
1566
1567         if (!bond->params.fail_over_mac) {
1568                 /*
1569                  * Set slave to master's mac address.  The application already
1570                  * set the master's mac address to that of the first slave
1571                  */
1572                 memcpy(addr.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
1573                 addr.sa_family = slave_dev->type;
1574                 res = dev_set_mac_address(slave_dev, &addr);
1575                 if (res) {
1576                         pr_debug("Error %d calling set_mac_address\n", res);
1577                         goto err_restore_mtu;
1578                 }
1579         }
1580
1581         res = netdev_set_master(slave_dev, bond_dev);
1582         if (res) {
1583                 pr_debug("Error %d calling netdev_set_master\n", res);
1584                 goto err_restore_mac;
1585         }
1586         /* open the slave since the application closed it */
1587         res = dev_open(slave_dev);
1588         if (res) {
1589                 pr_debug("Opening slave %s failed\n", slave_dev->name);
1590                 goto err_unset_master;
1591         }
1592
1593         new_slave->dev = slave_dev;
1594         slave_dev->priv_flags |= IFF_BONDING;
1595
1596         if (bond_is_lb(bond)) {
1597                 /* bond_alb_init_slave() must be called before all other stages since
1598                  * it might fail and we do not want to have to undo everything
1599                  */
1600                 res = bond_alb_init_slave(bond, new_slave);
1601                 if (res)
1602                         goto err_close;
1603         }
1604
1605         /* If the mode USES_PRIMARY, then the new slave gets the
1606          * master's promisc (and mc) settings only if it becomes the
1607          * curr_active_slave, and that is taken care of later when calling
1608          * bond_change_active()
1609          */
1610         if (!USES_PRIMARY(bond->params.mode)) {
1611                 /* set promiscuity level to new slave */
1612                 if (bond_dev->flags & IFF_PROMISC) {
1613                         res = dev_set_promiscuity(slave_dev, 1);
1614                         if (res)
1615                                 goto err_close;
1616                 }
1617
1618                 /* set allmulti level to new slave */
1619                 if (bond_dev->flags & IFF_ALLMULTI) {
1620                         res = dev_set_allmulti(slave_dev, 1);
1621                         if (res)
1622                                 goto err_close;
1623                 }
1624
1625                 netif_addr_lock_bh(bond_dev);
1626                 /* upload master's mc_list to new slave */
1627                 netdev_for_each_mc_addr(ha, bond_dev)
1628                         dev_mc_add(slave_dev, ha->addr);
1629                 netif_addr_unlock_bh(bond_dev);
1630         }
1631
1632         if (bond->params.mode == BOND_MODE_8023AD) {
1633                 /* add lacpdu mc addr to mc list */
1634                 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
1635
1636                 dev_mc_add(slave_dev, lacpdu_multicast);
1637         }
1638
1639         bond_add_vlans_on_slave(bond, slave_dev);
1640
1641         write_lock_bh(&bond->lock);
1642
1643         bond_attach_slave(bond, new_slave);
1644
1645         new_slave->delay = 0;
1646         new_slave->link_failure_count = 0;
1647
1648         bond_compute_features(bond);
1649
1650         write_unlock_bh(&bond->lock);
1651
1652         read_lock(&bond->lock);
1653
1654         new_slave->last_arp_rx = jiffies;
1655
1656         if (bond->params.miimon && !bond->params.use_carrier) {
1657                 link_reporting = bond_check_dev_link(bond, slave_dev, 1);
1658
1659                 if ((link_reporting == -1) && !bond->params.arp_interval) {
1660                         /*
1661                          * miimon is set but a bonded network driver
1662                          * does not support ETHTOOL/MII and
1663                          * arp_interval is not set.  Note: if
1664                          * use_carrier is enabled, we will never go
1665                          * here (because netif_carrier is always
1666                          * supported); thus, we don't need to change
1667                          * the messages for netif_carrier.
1668                          */
1669                         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",
1670                                bond_dev->name, slave_dev->name);
1671                 } else if (link_reporting == -1) {
1672                         /* unable get link status using mii/ethtool */
1673                         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",
1674                                    bond_dev->name, slave_dev->name);
1675                 }
1676         }
1677
1678         /* check for initial state */
1679         if (!bond->params.miimon ||
1680             (bond_check_dev_link(bond, slave_dev, 0) == BMSR_LSTATUS)) {
1681                 if (bond->params.updelay) {
1682                         pr_debug("Initial state of slave_dev is BOND_LINK_BACK\n");
1683                         new_slave->link  = BOND_LINK_BACK;
1684                         new_slave->delay = bond->params.updelay;
1685                 } else {
1686                         pr_debug("Initial state of slave_dev is BOND_LINK_UP\n");
1687                         new_slave->link  = BOND_LINK_UP;
1688                 }
1689                 new_slave->jiffies = jiffies;
1690         } else {
1691                 pr_debug("Initial state of slave_dev is BOND_LINK_DOWN\n");
1692                 new_slave->link  = BOND_LINK_DOWN;
1693         }
1694
1695         if (bond_update_speed_duplex(new_slave) &&
1696             (new_slave->link != BOND_LINK_DOWN)) {
1697                 pr_warning("%s: Warning: failed to get speed and duplex from %s, assumed to be 100Mb/sec and Full.\n",
1698                            bond_dev->name, new_slave->dev->name);
1699
1700                 if (bond->params.mode == BOND_MODE_8023AD) {
1701                         pr_warning("%s: Warning: Operation of 802.3ad mode requires ETHTOOL support in base driver for proper aggregator selection.\n",
1702                                    bond_dev->name);
1703                 }
1704         }
1705
1706         if (USES_PRIMARY(bond->params.mode) && bond->params.primary[0]) {
1707                 /* if there is a primary slave, remember it */
1708                 if (strcmp(bond->params.primary, new_slave->dev->name) == 0) {
1709                         bond->primary_slave = new_slave;
1710                         bond->force_primary = true;
1711                 }
1712         }
1713
1714         write_lock_bh(&bond->curr_slave_lock);
1715
1716         switch (bond->params.mode) {
1717         case BOND_MODE_ACTIVEBACKUP:
1718                 bond_set_slave_inactive_flags(new_slave);
1719                 bond_select_active_slave(bond);
1720                 break;
1721         case BOND_MODE_8023AD:
1722                 /* in 802.3ad mode, the internal mechanism
1723                  * will activate the slaves in the selected
1724                  * aggregator
1725                  */
1726                 bond_set_slave_inactive_flags(new_slave);
1727                 /* if this is the first slave */
1728                 if (bond->slave_cnt == 1) {
1729                         SLAVE_AD_INFO(new_slave).id = 1;
1730                         /* Initialize AD with the number of times that the AD timer is called in 1 second
1731                          * can be called only after the mac address of the bond is set
1732                          */
1733                         bond_3ad_initialize(bond, 1000/AD_TIMER_INTERVAL,
1734                                             bond->params.lacp_fast);
1735                 } else {
1736                         SLAVE_AD_INFO(new_slave).id =
1737                                 SLAVE_AD_INFO(new_slave->prev).id + 1;
1738                 }
1739
1740                 bond_3ad_bind_slave(new_slave);
1741                 break;
1742         case BOND_MODE_TLB:
1743         case BOND_MODE_ALB:
1744                 new_slave->state = BOND_STATE_ACTIVE;
1745                 bond_set_slave_inactive_flags(new_slave);
1746                 bond_select_active_slave(bond);
1747                 break;
1748         default:
1749                 pr_debug("This slave is always active in trunk mode\n");
1750
1751                 /* always active in trunk mode */
1752                 new_slave->state = BOND_STATE_ACTIVE;
1753
1754                 /* In trunking mode there is little meaning to curr_active_slave
1755                  * anyway (it holds no special properties of the bond device),
1756                  * so we can change it without calling change_active_interface()
1757                  */
1758                 if (!bond->curr_active_slave)
1759                         bond->curr_active_slave = new_slave;
1760
1761                 break;
1762         } /* switch(bond_mode) */
1763
1764         write_unlock_bh(&bond->curr_slave_lock);
1765
1766         bond_set_carrier(bond);
1767
1768 #ifdef CONFIG_NET_POLL_CONTROLLER
1769         if (slaves_support_netpoll(bond_dev)) {
1770                 bond_dev->priv_flags &= ~IFF_DISABLE_NETPOLL;
1771                 if (bond_dev->npinfo)
1772                         slave_dev->npinfo = bond_dev->npinfo;
1773         } else if (!(bond_dev->priv_flags & IFF_DISABLE_NETPOLL)) {
1774                 bond_dev->priv_flags |= IFF_DISABLE_NETPOLL;
1775                 pr_info("New slave device %s does not support netpoll\n",
1776                         slave_dev->name);
1777                 pr_info("Disabling netpoll support for %s\n", bond_dev->name);
1778         }
1779 #endif
1780         read_unlock(&bond->lock);
1781
1782         res = bond_create_slave_symlinks(bond_dev, slave_dev);
1783         if (res)
1784                 goto err_close;
1785
1786         pr_info("%s: enslaving %s as a%s interface with a%s link.\n",
1787                 bond_dev->name, slave_dev->name,
1788                 new_slave->state == BOND_STATE_ACTIVE ? "n active" : " backup",
1789                 new_slave->link != BOND_LINK_DOWN ? "n up" : " down");
1790
1791         /* enslave is successful */
1792         return 0;
1793
1794 /* Undo stages on error */
1795 err_close:
1796         dev_close(slave_dev);
1797
1798 err_unset_master:
1799         netdev_set_master(slave_dev, NULL);
1800
1801 err_restore_mac:
1802         if (!bond->params.fail_over_mac) {
1803                 /* XXX TODO - fom follow mode needs to change master's
1804                  * MAC if this slave's MAC is in use by the bond, or at
1805                  * least print a warning.
1806                  */
1807                 memcpy(addr.sa_data, new_slave->perm_hwaddr, ETH_ALEN);
1808                 addr.sa_family = slave_dev->type;
1809                 dev_set_mac_address(slave_dev, &addr);
1810         }
1811
1812 err_restore_mtu:
1813         dev_set_mtu(slave_dev, new_slave->original_mtu);
1814
1815 err_free:
1816         kfree(new_slave);
1817
1818 err_undo_flags:
1819         bond_dev->features = old_features;
1820
1821         return res;
1822 }
1823
1824 /*
1825  * Try to release the slave device <slave> from the bond device <master>
1826  * It is legal to access curr_active_slave without a lock because all the function
1827  * is write-locked.
1828  *
1829  * The rules for slave state should be:
1830  *   for Active/Backup:
1831  *     Active stays on all backups go down
1832  *   for Bonded connections:
1833  *     The first up interface should be left on and all others downed.
1834  */
1835 int bond_release(struct net_device *bond_dev, struct net_device *slave_dev)
1836 {
1837         struct bonding *bond = netdev_priv(bond_dev);
1838         struct slave *slave, *oldcurrent;
1839         struct sockaddr addr;
1840
1841         /* slave is not a slave or master is not master of this slave */
1842         if (!(slave_dev->flags & IFF_SLAVE) ||
1843             (slave_dev->master != bond_dev)) {
1844                 pr_err("%s: Error: cannot release %s.\n",
1845                        bond_dev->name, slave_dev->name);
1846                 return -EINVAL;
1847         }
1848
1849         netdev_bonding_change(bond_dev, NETDEV_BONDING_DESLAVE);
1850         write_lock_bh(&bond->lock);
1851
1852         slave = bond_get_slave_by_dev(bond, slave_dev);
1853         if (!slave) {
1854                 /* not a slave of this bond */
1855                 pr_info("%s: %s not enslaved\n",
1856                         bond_dev->name, slave_dev->name);
1857                 write_unlock_bh(&bond->lock);
1858                 return -EINVAL;
1859         }
1860
1861         if (!bond->params.fail_over_mac) {
1862                 if (!compare_ether_addr(bond_dev->dev_addr, slave->perm_hwaddr) &&
1863                     bond->slave_cnt > 1)
1864                         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",
1865                                    bond_dev->name, slave_dev->name,
1866                                    slave->perm_hwaddr,
1867                                    bond_dev->name, slave_dev->name);
1868         }
1869
1870         /* Inform AD package of unbinding of slave. */
1871         if (bond->params.mode == BOND_MODE_8023AD) {
1872                 /* must be called before the slave is
1873                  * detached from the list
1874                  */
1875                 bond_3ad_unbind_slave(slave);
1876         }
1877
1878         pr_info("%s: releasing %s interface %s\n",
1879                 bond_dev->name,
1880                 (slave->state == BOND_STATE_ACTIVE) ? "active" : "backup",
1881                 slave_dev->name);
1882
1883         oldcurrent = bond->curr_active_slave;
1884
1885         bond->current_arp_slave = NULL;
1886
1887         /* release the slave from its bond */
1888         bond_detach_slave(bond, slave);
1889
1890         bond_compute_features(bond);
1891
1892         if (bond->primary_slave == slave)
1893                 bond->primary_slave = NULL;
1894
1895         if (oldcurrent == slave)
1896                 bond_change_active_slave(bond, NULL);
1897
1898         if (bond_is_lb(bond)) {
1899                 /* Must be called only after the slave has been
1900                  * detached from the list and the curr_active_slave
1901                  * has been cleared (if our_slave == old_current),
1902                  * but before a new active slave is selected.
1903                  */
1904                 write_unlock_bh(&bond->lock);
1905                 bond_alb_deinit_slave(bond, slave);
1906                 write_lock_bh(&bond->lock);
1907         }
1908
1909         if (oldcurrent == slave) {
1910                 /*
1911                  * Note that we hold RTNL over this sequence, so there
1912                  * is no concern that another slave add/remove event
1913                  * will interfere.
1914                  */
1915                 write_unlock_bh(&bond->lock);
1916                 read_lock(&bond->lock);
1917                 write_lock_bh(&bond->curr_slave_lock);
1918
1919                 bond_select_active_slave(bond);
1920
1921                 write_unlock_bh(&bond->curr_slave_lock);
1922                 read_unlock(&bond->lock);
1923                 write_lock_bh(&bond->lock);
1924         }
1925
1926         if (bond->slave_cnt == 0) {
1927                 bond_set_carrier(bond);
1928
1929                 /* if the last slave was removed, zero the mac address
1930                  * of the master so it will be set by the application
1931                  * to the mac address of the first slave
1932                  */
1933                 memset(bond_dev->dev_addr, 0, bond_dev->addr_len);
1934
1935                 if (list_empty(&bond->vlan_list)) {
1936                         bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
1937                 } else {
1938                         pr_warning("%s: Warning: clearing HW address of %s while it still has VLANs.\n",
1939                                    bond_dev->name, bond_dev->name);
1940                         pr_warning("%s: When re-adding slaves, make sure the bond's HW address matches its VLANs'.\n",
1941                                    bond_dev->name);
1942                 }
1943         } else if ((bond_dev->features & NETIF_F_VLAN_CHALLENGED) &&
1944                    !bond_has_challenged_slaves(bond)) {
1945                 pr_info("%s: last VLAN challenged slave %s left bond %s. VLAN blocking is removed\n",
1946                         bond_dev->name, slave_dev->name, bond_dev->name);
1947                 bond_dev->features &= ~NETIF_F_VLAN_CHALLENGED;
1948         }
1949
1950         write_unlock_bh(&bond->lock);
1951
1952         /* must do this from outside any spinlocks */
1953         bond_destroy_slave_symlinks(bond_dev, slave_dev);
1954
1955         bond_del_vlans_from_slave(bond, slave_dev);
1956
1957         /* If the mode USES_PRIMARY, then we should only remove its
1958          * promisc and mc settings if it was the curr_active_slave, but that was
1959          * already taken care of above when we detached the slave
1960          */
1961         if (!USES_PRIMARY(bond->params.mode)) {
1962                 /* unset promiscuity level from slave */
1963                 if (bond_dev->flags & IFF_PROMISC)
1964                         dev_set_promiscuity(slave_dev, -1);
1965
1966                 /* unset allmulti level from slave */
1967                 if (bond_dev->flags & IFF_ALLMULTI)
1968                         dev_set_allmulti(slave_dev, -1);
1969
1970                 /* flush master's mc_list from slave */
1971                 netif_addr_lock_bh(bond_dev);
1972                 bond_mc_list_flush(bond_dev, slave_dev);
1973                 netif_addr_unlock_bh(bond_dev);
1974         }
1975
1976         netdev_set_master(slave_dev, NULL);
1977
1978 #ifdef CONFIG_NET_POLL_CONTROLLER
1979         read_lock_bh(&bond->lock);
1980         if (slaves_support_netpoll(bond_dev))
1981                 bond_dev->priv_flags &= ~IFF_DISABLE_NETPOLL;
1982         read_unlock_bh(&bond->lock);
1983         if (slave_dev->netdev_ops->ndo_netpoll_cleanup)
1984                 slave_dev->netdev_ops->ndo_netpoll_cleanup(slave_dev);
1985         else
1986                 slave_dev->npinfo = NULL;
1987 #endif
1988
1989         /* close slave before restoring its mac address */
1990         dev_close(slave_dev);
1991
1992         if (bond->params.fail_over_mac != BOND_FOM_ACTIVE) {
1993                 /* restore original ("permanent") mac address */
1994                 memcpy(addr.sa_data, slave->perm_hwaddr, ETH_ALEN);
1995                 addr.sa_family = slave_dev->type;
1996                 dev_set_mac_address(slave_dev, &addr);
1997         }
1998
1999         dev_set_mtu(slave_dev, slave->original_mtu);
2000
2001         slave_dev->priv_flags &= ~(IFF_MASTER_8023AD | IFF_MASTER_ALB |
2002                                    IFF_SLAVE_INACTIVE | IFF_BONDING |
2003                                    IFF_SLAVE_NEEDARP);
2004
2005         kfree(slave);
2006
2007         return 0;  /* deletion OK */
2008 }
2009
2010 /*
2011 * First release a slave and than destroy the bond if no more slaves are left.
2012 * Must be under rtnl_lock when this function is called.
2013 */
2014 int  bond_release_and_destroy(struct net_device *bond_dev,
2015                               struct net_device *slave_dev)
2016 {
2017         struct bonding *bond = netdev_priv(bond_dev);
2018         int ret;
2019
2020         ret = bond_release(bond_dev, slave_dev);
2021         if ((ret == 0) && (bond->slave_cnt == 0)) {
2022                 pr_info("%s: destroying bond %s.\n",
2023                         bond_dev->name, bond_dev->name);
2024                 unregister_netdevice(bond_dev);
2025         }
2026         return ret;
2027 }
2028
2029 /*
2030  * This function releases all slaves.
2031  */
2032 static int bond_release_all(struct net_device *bond_dev)
2033 {
2034         struct bonding *bond = netdev_priv(bond_dev);
2035         struct slave *slave;
2036         struct net_device *slave_dev;
2037         struct sockaddr addr;
2038
2039         write_lock_bh(&bond->lock);
2040
2041         netif_carrier_off(bond_dev);
2042
2043         if (bond->slave_cnt == 0)
2044                 goto out;
2045
2046         bond->current_arp_slave = NULL;
2047         bond->primary_slave = NULL;
2048         bond_change_active_slave(bond, NULL);
2049
2050         while ((slave = bond->first_slave) != NULL) {
2051                 /* Inform AD package of unbinding of slave
2052                  * before slave is detached from the list.
2053                  */
2054                 if (bond->params.mode == BOND_MODE_8023AD)
2055                         bond_3ad_unbind_slave(slave);
2056
2057                 slave_dev = slave->dev;
2058                 bond_detach_slave(bond, slave);
2059
2060                 /* now that the slave is detached, unlock and perform
2061                  * all the undo steps that should not be called from
2062                  * within a lock.
2063                  */
2064                 write_unlock_bh(&bond->lock);
2065
2066                 if (bond_is_lb(bond)) {
2067                         /* must be called only after the slave
2068                          * has been detached from the list
2069                          */
2070                         bond_alb_deinit_slave(bond, slave);
2071                 }
2072
2073                 bond_compute_features(bond);
2074
2075                 bond_destroy_slave_symlinks(bond_dev, slave_dev);
2076                 bond_del_vlans_from_slave(bond, slave_dev);
2077
2078                 /* If the mode USES_PRIMARY, then we should only remove its
2079                  * promisc and mc settings if it was the curr_active_slave, but that was
2080                  * already taken care of above when we detached the slave
2081                  */
2082                 if (!USES_PRIMARY(bond->params.mode)) {
2083                         /* unset promiscuity level from slave */
2084                         if (bond_dev->flags & IFF_PROMISC)
2085                                 dev_set_promiscuity(slave_dev, -1);
2086
2087                         /* unset allmulti level from slave */
2088                         if (bond_dev->flags & IFF_ALLMULTI)
2089                                 dev_set_allmulti(slave_dev, -1);
2090
2091                         /* flush master's mc_list from slave */
2092                         netif_addr_lock_bh(bond_dev);
2093                         bond_mc_list_flush(bond_dev, slave_dev);
2094                         netif_addr_unlock_bh(bond_dev);
2095                 }
2096
2097                 netdev_set_master(slave_dev, NULL);
2098
2099                 /* close slave before restoring its mac address */
2100                 dev_close(slave_dev);
2101
2102                 if (!bond->params.fail_over_mac) {
2103                         /* restore original ("permanent") mac address*/
2104                         memcpy(addr.sa_data, slave->perm_hwaddr, ETH_ALEN);
2105                         addr.sa_family = slave_dev->type;
2106                         dev_set_mac_address(slave_dev, &addr);
2107                 }
2108
2109                 slave_dev->priv_flags &= ~(IFF_MASTER_8023AD | IFF_MASTER_ALB |
2110                                            IFF_SLAVE_INACTIVE);
2111
2112                 kfree(slave);
2113
2114                 /* re-acquire the lock before getting the next slave */
2115                 write_lock_bh(&bond->lock);
2116         }
2117
2118         /* zero the mac address of the master so it will be
2119          * set by the application to the mac address of the
2120          * first slave
2121          */
2122         memset(bond_dev->dev_addr, 0, bond_dev->addr_len);
2123
2124         if (list_empty(&bond->vlan_list))
2125                 bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
2126         else {
2127                 pr_warning("%s: Warning: clearing HW address of %s while it still has VLANs.\n",
2128                            bond_dev->name, bond_dev->name);
2129                 pr_warning("%s: When re-adding slaves, make sure the bond's HW address matches its VLANs'.\n",
2130                            bond_dev->name);
2131         }
2132
2133         pr_info("%s: released all slaves\n", bond_dev->name);
2134
2135 out:
2136         write_unlock_bh(&bond->lock);
2137
2138         return 0;
2139 }
2140
2141 /*
2142  * This function changes the active slave to slave <slave_dev>.
2143  * It returns -EINVAL in the following cases.
2144  *  - <slave_dev> is not found in the list.
2145  *  - There is not active slave now.
2146  *  - <slave_dev> is already active.
2147  *  - The link state of <slave_dev> is not BOND_LINK_UP.
2148  *  - <slave_dev> is not running.
2149  * In these cases, this function does nothing.
2150  * In the other cases, current_slave pointer is changed and 0 is returned.
2151  */
2152 static int bond_ioctl_change_active(struct net_device *bond_dev, struct net_device *slave_dev)
2153 {
2154         struct bonding *bond = netdev_priv(bond_dev);
2155         struct slave *old_active = NULL;
2156         struct slave *new_active = NULL;
2157         int res = 0;
2158
2159         if (!USES_PRIMARY(bond->params.mode))
2160                 return -EINVAL;
2161
2162         /* Verify that master_dev is indeed the master of slave_dev */
2163         if (!(slave_dev->flags & IFF_SLAVE) || (slave_dev->master != bond_dev))
2164                 return -EINVAL;
2165
2166         read_lock(&bond->lock);
2167
2168         read_lock(&bond->curr_slave_lock);
2169         old_active = bond->curr_active_slave;
2170         read_unlock(&bond->curr_slave_lock);
2171
2172         new_active = bond_get_slave_by_dev(bond, slave_dev);
2173
2174         /*
2175          * Changing to the current active: do nothing; return success.
2176          */
2177         if (new_active && (new_active == old_active)) {
2178                 read_unlock(&bond->lock);
2179                 return 0;
2180         }
2181
2182         if ((new_active) &&
2183             (old_active) &&
2184             (new_active->link == BOND_LINK_UP) &&
2185             IS_UP(new_active->dev)) {
2186                 write_lock_bh(&bond->curr_slave_lock);
2187                 bond_change_active_slave(bond, new_active);
2188                 write_unlock_bh(&bond->curr_slave_lock);
2189         } else
2190                 res = -EINVAL;
2191
2192         read_unlock(&bond->lock);
2193
2194         return res;
2195 }
2196
2197 static int bond_info_query(struct net_device *bond_dev, struct ifbond *info)
2198 {
2199         struct bonding *bond = netdev_priv(bond_dev);
2200
2201         info->bond_mode = bond->params.mode;
2202         info->miimon = bond->params.miimon;
2203
2204         read_lock(&bond->lock);
2205         info->num_slaves = bond->slave_cnt;
2206         read_unlock(&bond->lock);
2207
2208         return 0;
2209 }
2210
2211 static int bond_slave_info_query(struct net_device *bond_dev, struct ifslave *info)
2212 {
2213         struct bonding *bond = netdev_priv(bond_dev);
2214         struct slave *slave;
2215         int i, res = -ENODEV;
2216
2217         read_lock(&bond->lock);
2218
2219         bond_for_each_slave(bond, slave, i) {
2220                 if (i == (int)info->slave_id) {
2221                         res = 0;
2222                         strcpy(info->slave_name, slave->dev->name);
2223                         info->link = slave->link;
2224                         info->state = slave->state;
2225                         info->link_failure_count = slave->link_failure_count;
2226                         break;
2227                 }
2228         }
2229
2230         read_unlock(&bond->lock);
2231
2232         return res;
2233 }
2234
2235 /*-------------------------------- Monitoring -------------------------------*/
2236
2237
2238 static int bond_miimon_inspect(struct bonding *bond)
2239 {
2240         struct slave *slave;
2241         int i, link_state, commit = 0;
2242         bool ignore_updelay;
2243
2244         ignore_updelay = !bond->curr_active_slave ? true : false;
2245
2246         bond_for_each_slave(bond, slave, i) {
2247                 slave->new_link = BOND_LINK_NOCHANGE;
2248
2249                 link_state = bond_check_dev_link(bond, slave->dev, 0);
2250
2251                 switch (slave->link) {
2252                 case BOND_LINK_UP:
2253                         if (link_state)
2254                                 continue;
2255
2256                         slave->link = BOND_LINK_FAIL;
2257                         slave->delay = bond->params.downdelay;
2258                         if (slave->delay) {
2259                                 pr_info("%s: link status down for %sinterface %s, disabling it in %d ms.\n",
2260                                         bond->dev->name,
2261                                         (bond->params.mode ==
2262                                          BOND_MODE_ACTIVEBACKUP) ?
2263                                         ((slave->state == BOND_STATE_ACTIVE) ?
2264                                          "active " : "backup ") : "",
2265                                         slave->dev->name,
2266                                         bond->params.downdelay * bond->params.miimon);
2267                         }
2268                         /*FALLTHRU*/
2269                 case BOND_LINK_FAIL:
2270                         if (link_state) {
2271                                 /*
2272                                  * recovered before downdelay expired
2273                                  */
2274                                 slave->link = BOND_LINK_UP;
2275                                 slave->jiffies = jiffies;
2276                                 pr_info("%s: link status up again after %d ms for interface %s.\n",
2277                                         bond->dev->name,
2278                                         (bond->params.downdelay - slave->delay) *
2279                                         bond->params.miimon,
2280                                         slave->dev->name);
2281                                 continue;
2282                         }
2283
2284                         if (slave->delay <= 0) {
2285                                 slave->new_link = BOND_LINK_DOWN;
2286                                 commit++;
2287                                 continue;
2288                         }
2289
2290                         slave->delay--;
2291                         break;
2292
2293                 case BOND_LINK_DOWN:
2294                         if (!link_state)
2295                                 continue;
2296
2297                         slave->link = BOND_LINK_BACK;
2298                         slave->delay = bond->params.updelay;
2299
2300                         if (slave->delay) {
2301                                 pr_info("%s: link status up for interface %s, enabling it in %d ms.\n",
2302                                         bond->dev->name, slave->dev->name,
2303                                         ignore_updelay ? 0 :
2304                                         bond->params.updelay *
2305                                         bond->params.miimon);
2306                         }
2307                         /*FALLTHRU*/
2308                 case BOND_LINK_BACK:
2309                         if (!link_state) {
2310                                 slave->link = BOND_LINK_DOWN;
2311                                 pr_info("%s: link status down again after %d ms for interface %s.\n",
2312                                         bond->dev->name,
2313                                         (bond->params.updelay - slave->delay) *
2314                                         bond->params.miimon,
2315                                         slave->dev->name);
2316
2317                                 continue;
2318                         }
2319
2320                         if (ignore_updelay)
2321                                 slave->delay = 0;
2322
2323                         if (slave->delay <= 0) {
2324                                 slave->new_link = BOND_LINK_UP;
2325                                 commit++;
2326                                 ignore_updelay = false;
2327                                 continue;
2328                         }
2329
2330                         slave->delay--;
2331                         break;
2332                 }
2333         }
2334
2335         return commit;
2336 }
2337
2338 static void bond_miimon_commit(struct bonding *bond)
2339 {
2340         struct slave *slave;
2341         int i;
2342
2343         bond_for_each_slave(bond, slave, i) {
2344                 switch (slave->new_link) {
2345                 case BOND_LINK_NOCHANGE:
2346                         continue;
2347
2348                 case BOND_LINK_UP:
2349                         slave->link = BOND_LINK_UP;
2350                         slave->jiffies = jiffies;
2351
2352                         if (bond->params.mode == BOND_MODE_8023AD) {
2353                                 /* prevent it from being the active one */
2354                                 slave->state = BOND_STATE_BACKUP;
2355                         } else if (bond->params.mode != BOND_MODE_ACTIVEBACKUP) {
2356                                 /* make it immediately active */
2357                                 slave->state = BOND_STATE_ACTIVE;
2358                         } else if (slave != bond->primary_slave) {
2359                                 /* prevent it from being the active one */
2360                                 slave->state = BOND_STATE_BACKUP;
2361                         }
2362
2363                         pr_info("%s: link status definitely up for interface %s.\n",
2364                                 bond->dev->name, slave->dev->name);
2365
2366                         /* notify ad that the link status has changed */
2367                         if (bond->params.mode == BOND_MODE_8023AD)
2368                                 bond_3ad_handle_link_change(slave, BOND_LINK_UP);
2369
2370                         if (bond_is_lb(bond))
2371                                 bond_alb_handle_link_change(bond, slave,
2372                                                             BOND_LINK_UP);
2373
2374                         if (!bond->curr_active_slave ||
2375                             (slave == bond->primary_slave))
2376                                 goto do_failover;
2377
2378                         continue;
2379
2380                 case BOND_LINK_DOWN:
2381                         if (slave->link_failure_count < UINT_MAX)
2382                                 slave->link_failure_count++;
2383
2384                         slave->link = BOND_LINK_DOWN;
2385
2386                         if (bond->params.mode == BOND_MODE_ACTIVEBACKUP ||
2387                             bond->params.mode == BOND_MODE_8023AD)
2388                                 bond_set_slave_inactive_flags(slave);
2389
2390                         pr_info("%s: link status definitely down for interface %s, disabling it\n",
2391                                 bond->dev->name, slave->dev->name);
2392
2393                         if (bond->params.mode == BOND_MODE_8023AD)
2394                                 bond_3ad_handle_link_change(slave,
2395                                                             BOND_LINK_DOWN);
2396
2397                         if (bond_is_lb(bond))
2398                                 bond_alb_handle_link_change(bond, slave,
2399                                                             BOND_LINK_DOWN);
2400
2401                         if (slave == bond->curr_active_slave)
2402                                 goto do_failover;
2403
2404                         continue;
2405
2406                 default:
2407                         pr_err("%s: invalid new link %d on slave %s\n",
2408                                bond->dev->name, slave->new_link,
2409                                slave->dev->name);
2410                         slave->new_link = BOND_LINK_NOCHANGE;
2411
2412                         continue;
2413                 }
2414
2415 do_failover:
2416                 ASSERT_RTNL();
2417                 write_lock_bh(&bond->curr_slave_lock);
2418                 bond_select_active_slave(bond);
2419                 write_unlock_bh(&bond->curr_slave_lock);
2420         }
2421
2422         bond_set_carrier(bond);
2423 }
2424
2425 /*
2426  * bond_mii_monitor
2427  *
2428  * Really a wrapper that splits the mii monitor into two phases: an
2429  * inspection, then (if inspection indicates something needs to be done)
2430  * an acquisition of appropriate locks followed by a commit phase to
2431  * implement whatever link state changes are indicated.
2432  */
2433 void bond_mii_monitor(struct work_struct *work)
2434 {
2435         struct bonding *bond = container_of(work, struct bonding,
2436                                             mii_work.work);
2437
2438         read_lock(&bond->lock);
2439         if (bond->kill_timers)
2440                 goto out;
2441
2442         if (bond->slave_cnt == 0)
2443                 goto re_arm;
2444
2445         if (bond->send_grat_arp) {
2446                 read_lock(&bond->curr_slave_lock);
2447                 bond_send_gratuitous_arp(bond);
2448                 read_unlock(&bond->curr_slave_lock);
2449         }
2450
2451         if (bond->send_unsol_na) {
2452                 read_lock(&bond->curr_slave_lock);
2453                 bond_send_unsolicited_na(bond);
2454                 read_unlock(&bond->curr_slave_lock);
2455         }
2456
2457         if (bond_miimon_inspect(bond)) {
2458                 read_unlock(&bond->lock);
2459                 rtnl_lock();
2460                 read_lock(&bond->lock);
2461
2462                 bond_miimon_commit(bond);
2463
2464                 read_unlock(&bond->lock);
2465                 rtnl_unlock();  /* might sleep, hold no other locks */
2466                 read_lock(&bond->lock);
2467         }
2468
2469 re_arm:
2470         if (bond->params.miimon)
2471                 queue_delayed_work(bond->wq, &bond->mii_work,
2472                                    msecs_to_jiffies(bond->params.miimon));
2473 out:
2474         read_unlock(&bond->lock);
2475 }
2476
2477 static __be32 bond_glean_dev_ip(struct net_device *dev)
2478 {
2479         struct in_device *idev;
2480         struct in_ifaddr *ifa;
2481         __be32 addr = 0;
2482
2483         if (!dev)
2484                 return 0;
2485
2486         rcu_read_lock();
2487         idev = __in_dev_get_rcu(dev);
2488         if (!idev)
2489                 goto out;
2490
2491         ifa = idev->ifa_list;
2492         if (!ifa)
2493                 goto out;
2494
2495         addr = ifa->ifa_local;
2496 out:
2497         rcu_read_unlock();
2498         return addr;
2499 }
2500
2501 static int bond_has_this_ip(struct bonding *bond, __be32 ip)
2502 {
2503         struct vlan_entry *vlan;
2504
2505         if (ip == bond->master_ip)
2506                 return 1;
2507
2508         list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
2509                 if (ip == vlan->vlan_ip)
2510                         return 1;
2511         }
2512
2513         return 0;
2514 }
2515
2516 /*
2517  * We go to the (large) trouble of VLAN tagging ARP frames because
2518  * switches in VLAN mode (especially if ports are configured as
2519  * "native" to a VLAN) might not pass non-tagged frames.
2520  */
2521 static void bond_arp_send(struct net_device *slave_dev, int arp_op, __be32 dest_ip, __be32 src_ip, unsigned short vlan_id)
2522 {
2523         struct sk_buff *skb;
2524
2525         pr_debug("arp %d on slave %s: dst %x src %x vid %d\n", arp_op,
2526                  slave_dev->name, dest_ip, src_ip, vlan_id);
2527
2528         skb = arp_create(arp_op, ETH_P_ARP, dest_ip, slave_dev, src_ip,
2529                          NULL, slave_dev->dev_addr, NULL);
2530
2531         if (!skb) {
2532                 pr_err("ARP packet allocation failed\n");
2533                 return;
2534         }
2535         if (vlan_id) {
2536                 skb = vlan_put_tag(skb, vlan_id);
2537                 if (!skb) {
2538                         pr_err("failed to insert VLAN tag\n");
2539                         return;
2540                 }
2541         }
2542         arp_xmit(skb);
2543 }
2544
2545
2546 static void bond_arp_send_all(struct bonding *bond, struct slave *slave)
2547 {
2548         int i, vlan_id, rv;
2549         __be32 *targets = bond->params.arp_targets;
2550         struct vlan_entry *vlan;
2551         struct net_device *vlan_dev;
2552         struct flowi fl;
2553         struct rtable *rt;
2554
2555         for (i = 0; (i < BOND_MAX_ARP_TARGETS); i++) {
2556                 if (!targets[i])
2557                         break;
2558                 pr_debug("basa: target %x\n", targets[i]);
2559                 if (list_empty(&bond->vlan_list)) {
2560                         pr_debug("basa: empty vlan: arp_send\n");
2561                         bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2562                                       bond->master_ip, 0);
2563                         continue;
2564                 }
2565
2566                 /*
2567                  * If VLANs are configured, we do a route lookup to
2568                  * determine which VLAN interface would be used, so we
2569                  * can tag the ARP with the proper VLAN tag.
2570                  */
2571                 memset(&fl, 0, sizeof(fl));
2572                 fl.fl4_dst = targets[i];
2573                 fl.fl4_tos = RTO_ONLINK;
2574
2575                 rv = ip_route_output_key(dev_net(bond->dev), &rt, &fl);
2576                 if (rv) {
2577                         if (net_ratelimit()) {
2578                                 pr_warning("%s: no route to arp_ip_target %pI4\n",
2579                                            bond->dev->name, &fl.fl4_dst);
2580                         }
2581                         continue;
2582                 }
2583
2584                 /*
2585                  * This target is not on a VLAN
2586                  */
2587                 if (rt->u.dst.dev == bond->dev) {
2588                         ip_rt_put(rt);
2589                         pr_debug("basa: rtdev == bond->dev: arp_send\n");
2590                         bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2591                                       bond->master_ip, 0);
2592                         continue;
2593                 }
2594
2595                 vlan_id = 0;
2596                 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
2597                         vlan_dev = vlan_group_get_device(bond->vlgrp, vlan->vlan_id);
2598                         if (vlan_dev == rt->u.dst.dev) {
2599                                 vlan_id = vlan->vlan_id;
2600                                 pr_debug("basa: vlan match on %s %d\n",
2601                                        vlan_dev->name, vlan_id);
2602                                 break;
2603                         }
2604                 }
2605
2606                 if (vlan_id) {
2607                         ip_rt_put(rt);
2608                         bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2609                                       vlan->vlan_ip, vlan_id);
2610                         continue;
2611                 }
2612
2613                 if (net_ratelimit()) {
2614                         pr_warning("%s: no path to arp_ip_target %pI4 via rt.dev %s\n",
2615                                    bond->dev->name, &fl.fl4_dst,
2616                                    rt->u.dst.dev ? rt->u.dst.dev->name : "NULL");
2617                 }
2618                 ip_rt_put(rt);
2619         }
2620 }
2621
2622 /*
2623  * Kick out a gratuitous ARP for an IP on the bonding master plus one
2624  * for each VLAN above us.
2625  *
2626  * Caller must hold curr_slave_lock for read or better
2627  */
2628 static void bond_send_gratuitous_arp(struct bonding *bond)
2629 {
2630         struct slave *slave = bond->curr_active_slave;
2631         struct vlan_entry *vlan;
2632         struct net_device *vlan_dev;
2633
2634         pr_debug("bond_send_grat_arp: bond %s slave %s\n",
2635                  bond->dev->name, slave ? slave->dev->name : "NULL");
2636
2637         if (!slave || !bond->send_grat_arp ||
2638             test_bit(__LINK_STATE_LINKWATCH_PENDING, &slave->dev->state))
2639                 return;
2640
2641         bond->send_grat_arp--;
2642
2643         if (bond->master_ip) {
2644                 bond_arp_send(slave->dev, ARPOP_REPLY, bond->master_ip,
2645                                 bond->master_ip, 0);
2646         }
2647
2648         list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
2649                 vlan_dev = vlan_group_get_device(bond->vlgrp, vlan->vlan_id);
2650                 if (vlan->vlan_ip) {
2651                         bond_arp_send(slave->dev, ARPOP_REPLY, vlan->vlan_ip,
2652                                       vlan->vlan_ip, vlan->vlan_id);
2653                 }
2654         }
2655 }
2656
2657 static void bond_validate_arp(struct bonding *bond, struct slave *slave, __be32 sip, __be32 tip)
2658 {
2659         int i;
2660         __be32 *targets = bond->params.arp_targets;
2661
2662         for (i = 0; (i < BOND_MAX_ARP_TARGETS) && targets[i]; i++) {
2663                 pr_debug("bva: sip %pI4 tip %pI4 t[%d] %pI4 bhti(tip) %d\n",
2664                          &sip, &tip, i, &targets[i],
2665                          bond_has_this_ip(bond, tip));
2666                 if (sip == targets[i]) {
2667                         if (bond_has_this_ip(bond, tip))
2668                                 slave->last_arp_rx = jiffies;
2669                         return;
2670                 }
2671         }
2672 }
2673
2674 static int bond_arp_rcv(struct sk_buff *skb, struct net_device *dev, struct packet_type *pt, struct net_device *orig_dev)
2675 {
2676         struct arphdr *arp;
2677         struct slave *slave;
2678         struct bonding *bond;
2679         unsigned char *arp_ptr;
2680         __be32 sip, tip;
2681
2682         if (dev->priv_flags & IFF_802_1Q_VLAN) {
2683                 /*
2684                  * When using VLANS and bonding, dev and oriv_dev may be
2685                  * incorrect if the physical interface supports VLAN
2686                  * acceleration.  With this change ARP validation now
2687                  * works for hosts only reachable on the VLAN interface.
2688                  */
2689                 dev = vlan_dev_real_dev(dev);
2690                 orig_dev = dev_get_by_index_rcu(dev_net(skb->dev),skb->skb_iif);
2691         }
2692
2693         if (!(dev->priv_flags & IFF_BONDING) || !(dev->flags & IFF_MASTER))
2694                 goto out;
2695
2696         bond = netdev_priv(dev);
2697         read_lock(&bond->lock);
2698
2699         pr_debug("bond_arp_rcv: bond %s skb->dev %s orig_dev %s\n",
2700                  bond->dev->name, skb->dev ? skb->dev->name : "NULL",
2701                  orig_dev ? orig_dev->name : "NULL");
2702
2703         slave = bond_get_slave_by_dev(bond, orig_dev);
2704         if (!slave || !slave_do_arp_validate(bond, slave))
2705                 goto out_unlock;
2706
2707         if (!pskb_may_pull(skb, arp_hdr_len(dev)))
2708                 goto out_unlock;
2709
2710         arp = arp_hdr(skb);
2711         if (arp->ar_hln != dev->addr_len ||
2712             skb->pkt_type == PACKET_OTHERHOST ||
2713             skb->pkt_type == PACKET_LOOPBACK ||
2714             arp->ar_hrd != htons(ARPHRD_ETHER) ||
2715             arp->ar_pro != htons(ETH_P_IP) ||
2716             arp->ar_pln != 4)
2717                 goto out_unlock;
2718
2719         arp_ptr = (unsigned char *)(arp + 1);
2720         arp_ptr += dev->addr_len;
2721         memcpy(&sip, arp_ptr, 4);
2722         arp_ptr += 4 + dev->addr_len;
2723         memcpy(&tip, arp_ptr, 4);
2724
2725         pr_debug("bond_arp_rcv: %s %s/%d av %d sv %d sip %pI4 tip %pI4\n",
2726                  bond->dev->name, slave->dev->name, slave->state,
2727                  bond->params.arp_validate, slave_do_arp_validate(bond, slave),
2728                  &sip, &tip);
2729
2730         /*
2731          * Backup slaves won't see the ARP reply, but do come through
2732          * here for each ARP probe (so we swap the sip/tip to validate
2733          * the probe).  In a "redundant switch, common router" type of
2734          * configuration, the ARP probe will (hopefully) travel from
2735          * the active, through one switch, the router, then the other
2736          * switch before reaching the backup.
2737          */
2738         if (slave->state == BOND_STATE_ACTIVE)
2739                 bond_validate_arp(bond, slave, sip, tip);
2740         else
2741                 bond_validate_arp(bond, slave, tip, sip);
2742
2743 out_unlock:
2744         read_unlock(&bond->lock);
2745 out:
2746         dev_kfree_skb(skb);
2747         return NET_RX_SUCCESS;
2748 }
2749
2750 /*
2751  * this function is called regularly to monitor each slave's link
2752  * ensuring that traffic is being sent and received when arp monitoring
2753  * is used in load-balancing mode. if the adapter has been dormant, then an
2754  * arp is transmitted to generate traffic. see activebackup_arp_monitor for
2755  * arp monitoring in active backup mode.
2756  */
2757 void bond_loadbalance_arp_mon(struct work_struct *work)
2758 {
2759         struct bonding *bond = container_of(work, struct bonding,
2760                                             arp_work.work);
2761         struct slave *slave, *oldcurrent;
2762         int do_failover = 0;
2763         int delta_in_ticks;
2764         int i;
2765
2766         read_lock(&bond->lock);
2767
2768         delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
2769
2770         if (bond->kill_timers)
2771                 goto out;
2772
2773         if (bond->slave_cnt == 0)
2774                 goto re_arm;
2775
2776         read_lock(&bond->curr_slave_lock);
2777         oldcurrent = bond->curr_active_slave;
2778         read_unlock(&bond->curr_slave_lock);
2779
2780         /* see if any of the previous devices are up now (i.e. they have
2781          * xmt and rcv traffic). the curr_active_slave does not come into
2782          * the picture unless it is null. also, slave->jiffies is not needed
2783          * here because we send an arp on each slave and give a slave as
2784          * long as it needs to get the tx/rx within the delta.
2785          * TODO: what about up/down delay in arp mode? it wasn't here before
2786          *       so it can wait
2787          */
2788         bond_for_each_slave(bond, slave, i) {
2789                 if (slave->link != BOND_LINK_UP) {
2790                         if (time_before_eq(jiffies, dev_trans_start(slave->dev) + delta_in_ticks) &&
2791                             time_before_eq(jiffies, slave->dev->last_rx + delta_in_ticks)) {
2792
2793                                 slave->link  = BOND_LINK_UP;
2794                                 slave->state = BOND_STATE_ACTIVE;
2795
2796                                 /* primary_slave has no meaning in round-robin
2797                                  * mode. the window of a slave being up and
2798                                  * curr_active_slave being null after enslaving
2799                                  * is closed.
2800                                  */
2801                                 if (!oldcurrent) {
2802                                         pr_info("%s: link status definitely up for interface %s, ",
2803                                                 bond->dev->name,
2804                                                 slave->dev->name);
2805                                         do_failover = 1;
2806                                 } else {
2807                                         pr_info("%s: interface %s is now up\n",
2808                                                 bond->dev->name,
2809                                                 slave->dev->name);
2810                                 }
2811                         }
2812                 } else {
2813                         /* slave->link == BOND_LINK_UP */
2814
2815                         /* not all switches will respond to an arp request
2816                          * when the source ip is 0, so don't take the link down
2817                          * if we don't know our ip yet
2818                          */
2819                         if (time_after_eq(jiffies, dev_trans_start(slave->dev) + 2*delta_in_ticks) ||
2820                             (time_after_eq(jiffies, slave->dev->last_rx + 2*delta_in_ticks))) {
2821
2822                                 slave->link  = BOND_LINK_DOWN;
2823                                 slave->state = BOND_STATE_BACKUP;
2824
2825                                 if (slave->link_failure_count < UINT_MAX)
2826                                         slave->link_failure_count++;
2827
2828                                 pr_info("%s: interface %s is now down.\n",
2829                                         bond->dev->name,
2830                                         slave->dev->name);
2831
2832                                 if (slave == oldcurrent)
2833                                         do_failover = 1;
2834                         }
2835                 }
2836
2837                 /* note: if switch is in round-robin mode, all links
2838                  * must tx arp to ensure all links rx an arp - otherwise
2839                  * links may oscillate or not come up at all; if switch is
2840                  * in something like xor mode, there is nothing we can
2841                  * do - all replies will be rx'ed on same link causing slaves
2842                  * to be unstable during low/no traffic periods
2843                  */
2844                 if (IS_UP(slave->dev))
2845                         bond_arp_send_all(bond, slave);
2846         }
2847
2848         if (do_failover) {
2849                 write_lock_bh(&bond->curr_slave_lock);
2850
2851                 bond_select_active_slave(bond);
2852
2853                 write_unlock_bh(&bond->curr_slave_lock);
2854         }
2855
2856 re_arm:
2857         if (bond->params.arp_interval)
2858                 queue_delayed_work(bond->wq, &bond->arp_work, delta_in_ticks);
2859 out:
2860         read_unlock(&bond->lock);
2861 }
2862
2863 /*
2864  * Called to inspect slaves for active-backup mode ARP monitor link state
2865  * changes.  Sets new_link in slaves to specify what action should take
2866  * place for the slave.  Returns 0 if no changes are found, >0 if changes
2867  * to link states must be committed.
2868  *
2869  * Called with bond->lock held for read.
2870  */
2871 static int bond_ab_arp_inspect(struct bonding *bond, int delta_in_ticks)
2872 {
2873         struct slave *slave;
2874         int i, commit = 0;
2875
2876         bond_for_each_slave(bond, slave, i) {
2877                 slave->new_link = BOND_LINK_NOCHANGE;
2878
2879                 if (slave->link != BOND_LINK_UP) {
2880                         if (time_before_eq(jiffies, slave_last_rx(bond, slave) +
2881                                            delta_in_ticks)) {
2882                                 slave->new_link = BOND_LINK_UP;
2883                                 commit++;
2884                         }
2885
2886                         continue;
2887                 }
2888
2889                 /*
2890                  * Give slaves 2*delta after being enslaved or made
2891                  * active.  This avoids bouncing, as the last receive
2892                  * times need a full ARP monitor cycle to be updated.
2893                  */
2894                 if (!time_after_eq(jiffies, slave->jiffies +
2895                                    2 * delta_in_ticks))
2896                         continue;
2897
2898                 /*
2899                  * Backup slave is down if:
2900                  * - No current_arp_slave AND
2901                  * - more than 3*delta since last receive AND
2902                  * - the bond has an IP address
2903                  *
2904                  * Note: a non-null current_arp_slave indicates
2905                  * the curr_active_slave went down and we are
2906                  * searching for a new one; under this condition
2907                  * we only take the curr_active_slave down - this
2908                  * gives each slave a chance to tx/rx traffic
2909                  * before being taken out
2910                  */
2911                 if (slave->state == BOND_STATE_BACKUP &&
2912                     !bond->current_arp_slave &&
2913                     time_after(jiffies, slave_last_rx(bond, slave) +
2914                                3 * delta_in_ticks)) {
2915                         slave->new_link = BOND_LINK_DOWN;
2916                         commit++;
2917                 }
2918
2919                 /*
2920                  * Active slave is down if:
2921                  * - more than 2*delta since transmitting OR
2922                  * - (more than 2*delta since receive AND
2923                  *    the bond has an IP address)
2924                  */
2925                 if ((slave->state == BOND_STATE_ACTIVE) &&
2926                     (time_after_eq(jiffies, dev_trans_start(slave->dev) +
2927                                     2 * delta_in_ticks) ||
2928                       (time_after_eq(jiffies, slave_last_rx(bond, slave)
2929                                      + 2 * delta_in_ticks)))) {
2930                         slave->new_link = BOND_LINK_DOWN;
2931                         commit++;
2932                 }
2933         }
2934
2935         return commit;
2936 }
2937
2938 /*
2939  * Called to commit link state changes noted by inspection step of
2940  * active-backup mode ARP monitor.
2941  *
2942  * Called with RTNL and bond->lock for read.
2943  */
2944 static void bond_ab_arp_commit(struct bonding *bond, int delta_in_ticks)
2945 {
2946         struct slave *slave;
2947         int i;
2948
2949         bond_for_each_slave(bond, slave, i) {
2950                 switch (slave->new_link) {
2951                 case BOND_LINK_NOCHANGE:
2952                         continue;
2953
2954                 case BOND_LINK_UP:
2955                         if ((!bond->curr_active_slave &&
2956                              time_before_eq(jiffies,
2957                                             dev_trans_start(slave->dev) +
2958                                             delta_in_ticks)) ||
2959                             bond->curr_active_slave != slave) {
2960                                 slave->link = BOND_LINK_UP;
2961                                 bond->current_arp_slave = NULL;
2962
2963                                 pr_info("%s: link status definitely up for interface %s.\n",
2964                                         bond->dev->name, slave->dev->name);
2965
2966                                 if (!bond->curr_active_slave ||
2967                                     (slave == bond->primary_slave))
2968                                         goto do_failover;
2969
2970                         }
2971
2972                         continue;
2973
2974                 case BOND_LINK_DOWN:
2975                         if (slave->link_failure_count < UINT_MAX)
2976                                 slave->link_failure_count++;
2977
2978                         slave->link = BOND_LINK_DOWN;
2979                         bond_set_slave_inactive_flags(slave);
2980
2981                         pr_info("%s: link status definitely down for interface %s, disabling it\n",
2982                                 bond->dev->name, slave->dev->name);
2983
2984                         if (slave == bond->curr_active_slave) {
2985                                 bond->current_arp_slave = NULL;
2986                                 goto do_failover;
2987                         }
2988
2989                         continue;
2990
2991                 default:
2992                         pr_err("%s: impossible: new_link %d on slave %s\n",
2993                                bond->dev->name, slave->new_link,
2994                                slave->dev->name);
2995                         continue;
2996                 }
2997
2998 do_failover:
2999                 ASSERT_RTNL();
3000                 write_lock_bh(&bond->curr_slave_lock);
3001                 bond_select_active_slave(bond);
3002                 write_unlock_bh(&bond->curr_slave_lock);
3003         }
3004
3005         bond_set_carrier(bond);
3006 }
3007
3008 /*
3009  * Send ARP probes for active-backup mode ARP monitor.
3010  *
3011  * Called with bond->lock held for read.
3012  */
3013 static void bond_ab_arp_probe(struct bonding *bond)
3014 {
3015         struct slave *slave;
3016         int i;
3017
3018         read_lock(&bond->curr_slave_lock);
3019
3020         if (bond->current_arp_slave && bond->curr_active_slave)
3021                 pr_info("PROBE: c_arp %s && cas %s BAD\n",
3022                         bond->current_arp_slave->dev->name,
3023                         bond->curr_active_slave->dev->name);
3024
3025         if (bond->curr_active_slave) {
3026                 bond_arp_send_all(bond, bond->curr_active_slave);
3027                 read_unlock(&bond->curr_slave_lock);
3028                 return;
3029         }
3030
3031         read_unlock(&bond->curr_slave_lock);
3032
3033         /* if we don't have a curr_active_slave, search for the next available
3034          * backup slave from the current_arp_slave and make it the candidate
3035          * for becoming the curr_active_slave
3036          */
3037
3038         if (!bond->current_arp_slave) {
3039                 bond->current_arp_slave = bond->first_slave;
3040                 if (!bond->current_arp_slave)
3041                         return;
3042         }
3043
3044         bond_set_slave_inactive_flags(bond->current_arp_slave);
3045
3046         /* search for next candidate */
3047         bond_for_each_slave_from(bond, slave, i, bond->current_arp_slave->next) {
3048                 if (IS_UP(slave->dev)) {
3049                         slave->link = BOND_LINK_BACK;
3050                         bond_set_slave_active_flags(slave);
3051                         bond_arp_send_all(bond, slave);
3052                         slave->jiffies = jiffies;
3053                         bond->current_arp_slave = slave;
3054                         break;
3055                 }
3056
3057                 /* if the link state is up at this point, we
3058                  * mark it down - this can happen if we have
3059                  * simultaneous link failures and
3060                  * reselect_active_interface doesn't make this
3061                  * one the current slave so it is still marked
3062                  * up when it is actually down
3063                  */
3064                 if (slave->link == BOND_LINK_UP) {
3065                         slave->link = BOND_LINK_DOWN;
3066                         if (slave->link_failure_count < UINT_MAX)
3067                                 slave->link_failure_count++;
3068
3069                         bond_set_slave_inactive_flags(slave);
3070
3071                         pr_info("%s: backup interface %s is now down.\n",
3072                                 bond->dev->name, slave->dev->name);
3073                 }
3074         }
3075 }
3076
3077 void bond_activebackup_arp_mon(struct work_struct *work)
3078 {
3079         struct bonding *bond = container_of(work, struct bonding,
3080                                             arp_work.work);
3081         int delta_in_ticks;
3082
3083         read_lock(&bond->lock);
3084
3085         if (bond->kill_timers)
3086                 goto out;
3087
3088         delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
3089
3090         if (bond->slave_cnt == 0)
3091                 goto re_arm;
3092
3093         if (bond->send_grat_arp) {
3094                 read_lock(&bond->curr_slave_lock);
3095                 bond_send_gratuitous_arp(bond);
3096                 read_unlock(&bond->curr_slave_lock);
3097         }
3098
3099         if (bond->send_unsol_na) {
3100                 read_lock(&bond->curr_slave_lock);
3101                 bond_send_unsolicited_na(bond);
3102                 read_unlock(&bond->curr_slave_lock);
3103         }
3104
3105         if (bond_ab_arp_inspect(bond, delta_in_ticks)) {
3106                 read_unlock(&bond->lock);
3107                 rtnl_lock();
3108                 read_lock(&bond->lock);
3109
3110                 bond_ab_arp_commit(bond, delta_in_ticks);
3111
3112                 read_unlock(&bond->lock);
3113                 rtnl_unlock();
3114                 read_lock(&bond->lock);
3115         }
3116
3117         bond_ab_arp_probe(bond);
3118
3119 re_arm:
3120         if (bond->params.arp_interval)
3121                 queue_delayed_work(bond->wq, &bond->arp_work, delta_in_ticks);
3122 out:
3123         read_unlock(&bond->lock);
3124 }
3125
3126 /*------------------------------ proc/seq_file-------------------------------*/
3127
3128 #ifdef CONFIG_PROC_FS
3129
3130 static void *bond_info_seq_start(struct seq_file *seq, loff_t *pos)
3131         __acquires(&dev_base_lock)
3132         __acquires(&bond->lock)
3133 {
3134         struct bonding *bond = seq->private;
3135         loff_t off = 0;
3136         struct slave *slave;
3137         int i;
3138
3139         /* make sure the bond won't be taken away */
3140         read_lock(&dev_base_lock);
3141         read_lock(&bond->lock);
3142
3143         if (*pos == 0)
3144                 return SEQ_START_TOKEN;
3145
3146         bond_for_each_slave(bond, slave, i) {
3147                 if (++off == *pos)
3148                         return slave;
3149         }
3150
3151         return NULL;
3152 }
3153
3154 static void *bond_info_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3155 {
3156         struct bonding *bond = seq->private;
3157         struct slave *slave = v;
3158
3159         ++*pos;
3160         if (v == SEQ_START_TOKEN)
3161                 return bond->first_slave;
3162
3163         slave = slave->next;
3164
3165         return (slave == bond->first_slave) ? NULL : slave;
3166 }
3167
3168 static void bond_info_seq_stop(struct seq_file *seq, void *v)
3169         __releases(&bond->lock)
3170         __releases(&dev_base_lock)
3171 {
3172         struct bonding *bond = seq->private;
3173
3174         read_unlock(&bond->lock);
3175         read_unlock(&dev_base_lock);
3176 }
3177
3178 static void bond_info_show_master(struct seq_file *seq)
3179 {
3180         struct bonding *bond = seq->private;
3181         struct slave *curr;
3182         int i;
3183
3184         read_lock(&bond->curr_slave_lock);
3185         curr = bond->curr_active_slave;
3186         read_unlock(&bond->curr_slave_lock);
3187
3188         seq_printf(seq, "Bonding Mode: %s",
3189                    bond_mode_name(bond->params.mode));
3190
3191         if (bond->params.mode == BOND_MODE_ACTIVEBACKUP &&
3192             bond->params.fail_over_mac)
3193                 seq_printf(seq, " (fail_over_mac %s)",
3194                    fail_over_mac_tbl[bond->params.fail_over_mac].modename);
3195
3196         seq_printf(seq, "\n");
3197
3198         if (bond->params.mode == BOND_MODE_XOR ||
3199                 bond->params.mode == BOND_MODE_8023AD) {
3200                 seq_printf(seq, "Transmit Hash Policy: %s (%d)\n",
3201                         xmit_hashtype_tbl[bond->params.xmit_policy].modename,
3202                         bond->params.xmit_policy);
3203         }
3204
3205         if (USES_PRIMARY(bond->params.mode)) {
3206                 seq_printf(seq, "Primary Slave: %s",
3207                            (bond->primary_slave) ?
3208                            bond->primary_slave->dev->name : "None");
3209                 if (bond->primary_slave)
3210                         seq_printf(seq, " (primary_reselect %s)",
3211                    pri_reselect_tbl[bond->params.primary_reselect].modename);
3212
3213                 seq_printf(seq, "\nCurrently Active Slave: %s\n",
3214                            (curr) ? curr->dev->name : "None");
3215         }
3216
3217         seq_printf(seq, "MII Status: %s\n", netif_carrier_ok(bond->dev) ?
3218                    "up" : "down");
3219         seq_printf(seq, "MII Polling Interval (ms): %d\n", bond->params.miimon);
3220         seq_printf(seq, "Up Delay (ms): %d\n",
3221                    bond->params.updelay * bond->params.miimon);
3222         seq_printf(seq, "Down Delay (ms): %d\n",
3223                    bond->params.downdelay * bond->params.miimon);
3224
3225
3226         /* ARP information */
3227         if (bond->params.arp_interval > 0) {
3228                 int printed = 0;
3229                 seq_printf(seq, "ARP Polling Interval (ms): %d\n",
3230                                 bond->params.arp_interval);
3231
3232                 seq_printf(seq, "ARP IP target/s (n.n.n.n form):");
3233
3234                 for (i = 0; (i < BOND_MAX_ARP_TARGETS); i++) {
3235                         if (!bond->params.arp_targets[i])
3236                                 break;
3237                         if (printed)
3238                                 seq_printf(seq, ",");
3239                         seq_printf(seq, " %pI4", &bond->params.arp_targets[i]);
3240                         printed = 1;
3241                 }
3242                 seq_printf(seq, "\n");
3243         }
3244
3245         if (bond->params.mode == BOND_MODE_8023AD) {
3246                 struct ad_info ad_info;
3247
3248                 seq_puts(seq, "\n802.3ad info\n");
3249                 seq_printf(seq, "LACP rate: %s\n",
3250                            (bond->params.lacp_fast) ? "fast" : "slow");
3251                 seq_printf(seq, "Aggregator selection policy (ad_select): %s\n",
3252                            ad_select_tbl[bond->params.ad_select].modename);
3253
3254                 if (bond_3ad_get_active_agg_info(bond, &ad_info)) {
3255                         seq_printf(seq, "bond %s has no active aggregator\n",
3256                                    bond->dev->name);
3257                 } else {
3258                         seq_printf(seq, "Active Aggregator Info:\n");
3259
3260                         seq_printf(seq, "\tAggregator ID: %d\n",
3261                                    ad_info.aggregator_id);
3262                         seq_printf(seq, "\tNumber of ports: %d\n",
3263                                    ad_info.ports);
3264                         seq_printf(seq, "\tActor Key: %d\n",
3265                                    ad_info.actor_key);
3266                         seq_printf(seq, "\tPartner Key: %d\n",
3267                                    ad_info.partner_key);
3268                         seq_printf(seq, "\tPartner Mac Address: %pM\n",
3269                                    ad_info.partner_system);
3270                 }
3271         }
3272 }
3273
3274 static void bond_info_show_slave(struct seq_file *seq,
3275                                  const struct slave *slave)
3276 {
3277         struct bonding *bond = seq->private;
3278
3279         seq_printf(seq, "\nSlave Interface: %s\n", slave->dev->name);
3280         seq_printf(seq, "MII Status: %s\n",
3281                    (slave->link == BOND_LINK_UP) ?  "up" : "down");
3282         seq_printf(seq, "Link Failure Count: %u\n",
3283                    slave->link_failure_count);
3284
3285         seq_printf(seq, "Permanent HW addr: %pM\n", slave->perm_hwaddr);
3286
3287         if (bond->params.mode == BOND_MODE_8023AD) {
3288                 const struct aggregator *agg
3289                         = SLAVE_AD_INFO(slave).port.aggregator;
3290
3291                 if (agg)
3292                         seq_printf(seq, "Aggregator ID: %d\n",
3293                                    agg->aggregator_identifier);
3294                 else
3295                         seq_puts(seq, "Aggregator ID: N/A\n");
3296         }
3297         seq_printf(seq, "Slave queue ID: %d\n", slave->queue_id);
3298 }
3299
3300 static int bond_info_seq_show(struct seq_file *seq, void *v)
3301 {
3302         if (v == SEQ_START_TOKEN) {
3303                 seq_printf(seq, "%s\n", version);
3304                 bond_info_show_master(seq);
3305         } else
3306                 bond_info_show_slave(seq, v);
3307
3308         return 0;
3309 }
3310
3311 static const struct seq_operations bond_info_seq_ops = {
3312         .start = bond_info_seq_start,
3313         .next  = bond_info_seq_next,
3314         .stop  = bond_info_seq_stop,
3315         .show  = bond_info_seq_show,
3316 };
3317
3318 static int bond_info_open(struct inode *inode, struct file *file)
3319 {
3320         struct seq_file *seq;
3321         struct proc_dir_entry *proc;
3322         int res;
3323
3324         res = seq_open(file, &bond_info_seq_ops);
3325         if (!res) {
3326                 /* recover the pointer buried in proc_dir_entry data */
3327                 seq = file->private_data;
3328                 proc = PDE(inode);
3329                 seq->private = proc->data;
3330         }
3331
3332         return res;
3333 }
3334
3335 static const struct file_operations bond_info_fops = {
3336         .owner   = THIS_MODULE,
3337         .open    = bond_info_open,
3338         .read    = seq_read,
3339         .llseek  = seq_lseek,
3340         .release = seq_release,
3341 };
3342
3343 static void bond_create_proc_entry(struct bonding *bond)
3344 {
3345         struct net_device *bond_dev = bond->dev;
3346         struct bond_net *bn = net_generic(dev_net(bond_dev), bond_net_id);
3347
3348         if (bn->proc_dir) {
3349                 bond->proc_entry = proc_create_data(bond_dev->name,
3350                                                     S_IRUGO, bn->proc_dir,
3351                                                     &bond_info_fops, bond);
3352                 if (bond->proc_entry == NULL)
3353                         pr_warning("Warning: Cannot create /proc/net/%s/%s\n",
3354                                    DRV_NAME, bond_dev->name);
3355                 else
3356                         memcpy(bond->proc_file_name, bond_dev->name, IFNAMSIZ);
3357         }
3358 }
3359
3360 static void bond_remove_proc_entry(struct bonding *bond)
3361 {
3362         struct net_device *bond_dev = bond->dev;
3363         struct bond_net *bn = net_generic(dev_net(bond_dev), bond_net_id);
3364
3365         if (bn->proc_dir && bond->proc_entry) {
3366                 remove_proc_entry(bond->proc_file_name, bn->proc_dir);
3367                 memset(bond->proc_file_name, 0, IFNAMSIZ);
3368                 bond->proc_entry = NULL;
3369         }
3370 }
3371
3372 /* Create the bonding directory under /proc/net, if doesn't exist yet.
3373  * Caller must hold rtnl_lock.
3374  */
3375 static void __net_init bond_create_proc_dir(struct bond_net *bn)
3376 {
3377         if (!bn->proc_dir) {
3378                 bn->proc_dir = proc_mkdir(DRV_NAME, bn->net->proc_net);
3379                 if (!bn->proc_dir)
3380                         pr_warning("Warning: cannot create /proc/net/%s\n",
3381                                    DRV_NAME);
3382         }
3383 }
3384
3385 /* Destroy the bonding directory under /proc/net, if empty.
3386  * Caller must hold rtnl_lock.
3387  */
3388 static void __net_exit bond_destroy_proc_dir(struct bond_net *bn)
3389 {
3390         if (bn->proc_dir) {
3391                 remove_proc_entry(DRV_NAME, bn->net->proc_net);
3392                 bn->proc_dir = NULL;
3393         }
3394 }
3395
3396 #else /* !CONFIG_PROC_FS */
3397
3398 static void bond_create_proc_entry(struct bonding *bond)
3399 {
3400 }
3401
3402 static void bond_remove_proc_entry(struct bonding *bond)
3403 {
3404 }
3405
3406 static inline void bond_create_proc_dir(struct bond_net *bn)
3407 {
3408 }
3409
3410 static inline void bond_destroy_proc_dir(struct bond_net *bn)
3411 {
3412 }
3413
3414 #endif /* CONFIG_PROC_FS */
3415
3416
3417 /*-------------------------- netdev event handling --------------------------*/
3418
3419 /*
3420  * Change device name
3421  */
3422 static int bond_event_changename(struct bonding *bond)
3423 {
3424         bond_remove_proc_entry(bond);
3425         bond_create_proc_entry(bond);
3426
3427         return NOTIFY_DONE;
3428 }
3429
3430 static int bond_master_netdev_event(unsigned long event,
3431                                     struct net_device *bond_dev)
3432 {
3433         struct bonding *event_bond = netdev_priv(bond_dev);
3434
3435         switch (event) {
3436         case NETDEV_CHANGENAME:
3437                 return bond_event_changename(event_bond);
3438         default:
3439                 break;
3440         }
3441
3442         return NOTIFY_DONE;
3443 }
3444
3445 static int bond_slave_netdev_event(unsigned long event,
3446                                    struct net_device *slave_dev)
3447 {
3448         struct net_device *bond_dev = slave_dev->master;
3449         struct bonding *bond = netdev_priv(bond_dev);
3450
3451         switch (event) {
3452         case NETDEV_UNREGISTER:
3453                 if (bond_dev) {
3454                         if (bond->setup_by_slave)
3455                                 bond_release_and_destroy(bond_dev, slave_dev);
3456                         else
3457                                 bond_release(bond_dev, slave_dev);
3458                 }
3459                 break;
3460         case NETDEV_CHANGE:
3461                 if (bond->params.mode == BOND_MODE_8023AD || bond_is_lb(bond)) {
3462                         struct slave *slave;
3463
3464                         slave = bond_get_slave_by_dev(bond, slave_dev);
3465                         if (slave) {
3466                                 u16 old_speed = slave->speed;
3467                                 u16 old_duplex = slave->duplex;
3468
3469                                 bond_update_speed_duplex(slave);
3470
3471                                 if (bond_is_lb(bond))
3472                                         break;
3473
3474                                 if (old_speed != slave->speed)
3475                                         bond_3ad_adapter_speed_changed(slave);
3476                                 if (old_duplex != slave->duplex)
3477                                         bond_3ad_adapter_duplex_changed(slave);
3478                         }
3479                 }
3480
3481                 break;
3482         case NETDEV_DOWN:
3483                 /*
3484                  * ... Or is it this?
3485                  */
3486                 break;
3487         case NETDEV_CHANGEMTU:
3488                 /*
3489                  * TODO: Should slaves be allowed to
3490                  * independently alter their MTU?  For
3491                  * an active-backup bond, slaves need
3492                  * not be the same type of device, so
3493                  * MTUs may vary.  For other modes,
3494                  * slaves arguably should have the
3495                  * same MTUs. To do this, we'd need to
3496                  * take over the slave's change_mtu
3497                  * function for the duration of their
3498                  * servitude.
3499                  */
3500                 break;
3501         case NETDEV_CHANGENAME:
3502                 /*
3503                  * TODO: handle changing the primary's name
3504                  */
3505                 break;
3506         case NETDEV_FEAT_CHANGE:
3507                 bond_compute_features(bond);
3508                 break;
3509         default:
3510                 break;
3511         }
3512
3513         return NOTIFY_DONE;
3514 }
3515
3516 /*
3517  * bond_netdev_event: handle netdev notifier chain events.
3518  *
3519  * This function receives events for the netdev chain.  The caller (an
3520  * ioctl handler calling blocking_notifier_call_chain) holds the necessary
3521  * locks for us to safely manipulate the slave devices (RTNL lock,
3522  * dev_probe_lock).
3523  */
3524 static int bond_netdev_event(struct notifier_block *this,
3525                              unsigned long event, void *ptr)
3526 {
3527         struct net_device *event_dev = (struct net_device *)ptr;
3528
3529         pr_debug("event_dev: %s, event: %lx\n",
3530                  event_dev ? event_dev->name : "None",
3531                  event);
3532
3533         if (!(event_dev->priv_flags & IFF_BONDING))
3534                 return NOTIFY_DONE;
3535
3536         if (event_dev->flags & IFF_MASTER) {
3537                 pr_debug("IFF_MASTER\n");
3538                 return bond_master_netdev_event(event, event_dev);
3539         }
3540
3541         if (event_dev->flags & IFF_SLAVE) {
3542                 pr_debug("IFF_SLAVE\n");
3543                 return bond_slave_netdev_event(event, event_dev);
3544         }
3545
3546         return NOTIFY_DONE;
3547 }
3548
3549 /*
3550  * bond_inetaddr_event: handle inetaddr notifier chain events.
3551  *
3552  * We keep track of device IPs primarily to use as source addresses in
3553  * ARP monitor probes (rather than spewing out broadcasts all the time).
3554  *
3555  * We track one IP for the main device (if it has one), plus one per VLAN.
3556  */
3557 static int bond_inetaddr_event(struct notifier_block *this, unsigned long event, void *ptr)
3558 {
3559         struct in_ifaddr *ifa = ptr;
3560         struct net_device *vlan_dev, *event_dev = ifa->ifa_dev->dev;
3561         struct bond_net *bn = net_generic(dev_net(event_dev), bond_net_id);
3562         struct bonding *bond;
3563         struct vlan_entry *vlan;
3564
3565         list_for_each_entry(bond, &bn->dev_list, bond_list) {
3566                 if (bond->dev == event_dev) {
3567                         switch (event) {
3568                         case NETDEV_UP:
3569                                 bond->master_ip = ifa->ifa_local;
3570                                 return NOTIFY_OK;
3571                         case NETDEV_DOWN:
3572                                 bond->master_ip = bond_glean_dev_ip(bond->dev);
3573                                 return NOTIFY_OK;
3574                         default:
3575                                 return NOTIFY_DONE;
3576                         }
3577                 }
3578
3579                 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
3580                         vlan_dev = vlan_group_get_device(bond->vlgrp, vlan->vlan_id);
3581                         if (vlan_dev == event_dev) {
3582                                 switch (event) {
3583                                 case NETDEV_UP:
3584                                         vlan->vlan_ip = ifa->ifa_local;
3585                                         return NOTIFY_OK;
3586                                 case NETDEV_DOWN:
3587                                         vlan->vlan_ip =
3588                                                 bond_glean_dev_ip(vlan_dev);
3589                                         return NOTIFY_OK;
3590                                 default:
3591                                         return NOTIFY_DONE;
3592                                 }
3593                         }
3594                 }
3595         }
3596         return NOTIFY_DONE;
3597 }
3598
3599 static struct notifier_block bond_netdev_notifier = {
3600         .notifier_call = bond_netdev_event,
3601 };
3602
3603 static struct notifier_block bond_inetaddr_notifier = {
3604         .notifier_call = bond_inetaddr_event,
3605 };
3606
3607 /*-------------------------- Packet type handling ---------------------------*/
3608
3609 /* register to receive lacpdus on a bond */
3610 static void bond_register_lacpdu(struct bonding *bond)
3611 {
3612         struct packet_type *pk_type = &(BOND_AD_INFO(bond).ad_pkt_type);
3613
3614         /* initialize packet type */
3615         pk_type->type = PKT_TYPE_LACPDU;
3616         pk_type->dev = bond->dev;
3617         pk_type->func = bond_3ad_lacpdu_recv;
3618
3619         dev_add_pack(pk_type);
3620 }
3621
3622 /* unregister to receive lacpdus on a bond */
3623 static void bond_unregister_lacpdu(struct bonding *bond)
3624 {
3625         dev_remove_pack(&(BOND_AD_INFO(bond).ad_pkt_type));
3626 }
3627
3628 void bond_register_arp(struct bonding *bond)
3629 {
3630         struct packet_type *pt = &bond->arp_mon_pt;
3631
3632         if (pt->type)
3633                 return;
3634
3635         pt->type = htons(ETH_P_ARP);
3636         pt->dev = bond->dev;
3637         pt->func = bond_arp_rcv;
3638         dev_add_pack(pt);
3639 }
3640
3641 void bond_unregister_arp(struct bonding *bond)
3642 {
3643         struct packet_type *pt = &bond->arp_mon_pt;
3644
3645         dev_remove_pack(pt);
3646         pt->type = 0;
3647 }
3648
3649 /*---------------------------- Hashing Policies -----------------------------*/
3650
3651 /*
3652  * Hash for the output device based upon layer 2 and layer 3 data. If
3653  * the packet is not IP mimic bond_xmit_hash_policy_l2()
3654  */
3655 static int bond_xmit_hash_policy_l23(struct sk_buff *skb, int count)
3656 {
3657         struct ethhdr *data = (struct ethhdr *)skb->data;
3658         struct iphdr *iph = ip_hdr(skb);
3659
3660         if (skb->protocol == htons(ETH_P_IP)) {
3661                 return ((ntohl(iph->saddr ^ iph->daddr) & 0xffff) ^
3662                         (data->h_dest[5] ^ data->h_source[5])) % count;
3663         }
3664
3665         return (data->h_dest[5] ^ data->h_source[5]) % count;
3666 }
3667
3668 /*
3669  * Hash for the output device based upon layer 3 and layer 4 data. If
3670  * the packet is a frag or not TCP or UDP, just use layer 3 data.  If it is
3671  * altogether not IP, mimic bond_xmit_hash_policy_l2()
3672  */
3673 static int bond_xmit_hash_policy_l34(struct sk_buff *skb, int count)
3674 {
3675         struct ethhdr *data = (struct ethhdr *)skb->data;
3676         struct iphdr *iph = ip_hdr(skb);
3677         __be16 *layer4hdr = (__be16 *)((u32 *)iph + iph->ihl);
3678         int layer4_xor = 0;
3679
3680         if (skb->protocol == htons(ETH_P_IP)) {
3681                 if (!(iph->frag_off & htons(IP_MF|IP_OFFSET)) &&
3682                     (iph->protocol == IPPROTO_TCP ||
3683                      iph->protocol == IPPROTO_UDP)) {
3684                         layer4_xor = ntohs((*layer4hdr ^ *(layer4hdr + 1)));
3685                 }
3686                 return (layer4_xor ^
3687                         ((ntohl(iph->saddr ^ iph->daddr)) & 0xffff)) % count;
3688
3689         }
3690
3691         return (data->h_dest[5] ^ data->h_source[5]) % count;
3692 }
3693
3694 /*
3695  * Hash for the output device based upon layer 2 data
3696  */
3697 static int bond_xmit_hash_policy_l2(struct sk_buff *skb, int count)
3698 {
3699         struct ethhdr *data = (struct ethhdr *)skb->data;
3700
3701         return (data->h_dest[5] ^ data->h_source[5]) % count;
3702 }
3703
3704 /*-------------------------- Device entry points ----------------------------*/
3705
3706 static int bond_open(struct net_device *bond_dev)
3707 {
3708         struct bonding *bond = netdev_priv(bond_dev);
3709
3710         bond->kill_timers = 0;
3711
3712         if (bond_is_lb(bond)) {
3713                 /* bond_alb_initialize must be called before the timer
3714                  * is started.
3715                  */
3716                 if (bond_alb_initialize(bond, (bond->params.mode == BOND_MODE_ALB))) {
3717                         /* something went wrong - fail the open operation */
3718                         return -ENOMEM;
3719                 }
3720
3721                 INIT_DELAYED_WORK(&bond->alb_work, bond_alb_monitor);
3722                 queue_delayed_work(bond->wq, &bond->alb_work, 0);
3723         }
3724
3725         if (bond->params.miimon) {  /* link check interval, in milliseconds. */
3726                 INIT_DELAYED_WORK(&bond->mii_work, bond_mii_monitor);
3727                 queue_delayed_work(bond->wq, &bond->mii_work, 0);
3728         }
3729
3730         if (bond->params.arp_interval) {  /* arp interval, in milliseconds. */
3731                 if (bond->params.mode == BOND_MODE_ACTIVEBACKUP)
3732                         INIT_DELAYED_WORK(&bond->arp_work,
3733                                           bond_activebackup_arp_mon);
3734                 else
3735                         INIT_DELAYED_WORK(&bond->arp_work,
3736                                           bond_loadbalance_arp_mon);
3737
3738                 queue_delayed_work(bond->wq, &bond->arp_work, 0);
3739                 if (bond->params.arp_validate)
3740                         bond_register_arp(bond);
3741         }
3742
3743         if (bond->params.mode == BOND_MODE_8023AD) {
3744                 INIT_DELAYED_WORK(&bond->ad_work, bond_3ad_state_machine_handler);
3745                 queue_delayed_work(bond->wq, &bond->ad_work, 0);
3746                 /* register to receive LACPDUs */
3747                 bond_register_lacpdu(bond);
3748                 bond_3ad_initiate_agg_selection(bond, 1);
3749         }
3750
3751         return 0;
3752 }
3753
3754 static int bond_close(struct net_device *bond_dev)
3755 {
3756         struct bonding *bond = netdev_priv(bond_dev);
3757
3758         if (bond->params.mode == BOND_MODE_8023AD) {
3759                 /* Unregister the receive of LACPDUs */
3760                 bond_unregister_lacpdu(bond);
3761         }
3762
3763         if (bond->params.arp_validate)
3764                 bond_unregister_arp(bond);
3765
3766         write_lock_bh(&bond->lock);
3767
3768         bond->send_grat_arp = 0;
3769         bond->send_unsol_na = 0;
3770
3771         /* signal timers not to re-arm */
3772         bond->kill_timers = 1;
3773
3774         write_unlock_bh(&bond->lock);
3775
3776         if (bond->params.miimon) {  /* link check interval, in milliseconds. */
3777                 cancel_delayed_work(&bond->mii_work);
3778         }
3779
3780         if (bond->params.arp_interval) {  /* arp interval, in milliseconds. */
3781                 cancel_delayed_work(&bond->arp_work);
3782         }
3783
3784         switch (bond->params.mode) {
3785         case BOND_MODE_8023AD:
3786                 cancel_delayed_work(&bond->ad_work);
3787                 break;
3788         case BOND_MODE_TLB:
3789         case BOND_MODE_ALB:
3790                 cancel_delayed_work(&bond->alb_work);
3791                 break;
3792         default:
3793                 break;
3794         }
3795
3796
3797         if (bond_is_lb(bond)) {
3798                 /* Must be called only after all
3799                  * slaves have been released
3800                  */
3801                 bond_alb_deinitialize(bond);
3802         }
3803
3804         return 0;
3805 }
3806
3807 static struct net_device_stats *bond_get_stats(struct net_device *bond_dev)
3808 {
3809         struct bonding *bond = netdev_priv(bond_dev);
3810         struct net_device_stats *stats = &bond_dev->stats;
3811         struct net_device_stats local_stats;
3812         struct slave *slave;
3813         int i;
3814
3815         memset(&local_stats, 0, sizeof(struct net_device_stats));
3816
3817         read_lock_bh(&bond->lock);
3818
3819         bond_for_each_slave(bond, slave, i) {
3820                 const struct net_device_stats *sstats = dev_get_stats(slave->dev);
3821
3822                 local_stats.rx_packets += sstats->rx_packets;
3823                 local_stats.rx_bytes += sstats->rx_bytes;
3824                 local_stats.rx_errors += sstats->rx_errors;
3825                 local_stats.rx_dropped += sstats->rx_dropped;
3826
3827                 local_stats.tx_packets += sstats->tx_packets;
3828                 local_stats.tx_bytes += sstats->tx_bytes;
3829                 local_stats.tx_errors += sstats->tx_errors;
3830                 local_stats.tx_dropped += sstats->tx_dropped;
3831
3832                 local_stats.multicast += sstats->multicast;
3833                 local_stats.collisions += sstats->collisions;
3834
3835                 local_stats.rx_length_errors += sstats->rx_length_errors;
3836                 local_stats.rx_over_errors += sstats->rx_over_errors;
3837                 local_stats.rx_crc_errors += sstats->rx_crc_errors;
3838                 local_stats.rx_frame_errors += sstats->rx_frame_errors;
3839                 local_stats.rx_fifo_errors += sstats->rx_fifo_errors;
3840                 local_stats.rx_missed_errors += sstats->rx_missed_errors;
3841
3842                 local_stats.tx_aborted_errors += sstats->tx_aborted_errors;
3843                 local_stats.tx_carrier_errors += sstats->tx_carrier_errors;
3844                 local_stats.tx_fifo_errors += sstats->tx_fifo_errors;
3845                 local_stats.tx_heartbeat_errors += sstats->tx_heartbeat_errors;
3846                 local_stats.tx_window_errors += sstats->tx_window_errors;
3847         }
3848
3849         memcpy(stats, &local_stats, sizeof(struct net_device_stats));
3850
3851         read_unlock_bh(&bond->lock);
3852
3853         return stats;
3854 }
3855
3856 static int bond_do_ioctl(struct net_device *bond_dev, struct ifreq *ifr, int cmd)
3857 {
3858         struct net_device *slave_dev = NULL;
3859         struct ifbond k_binfo;
3860         struct ifbond __user *u_binfo = NULL;
3861         struct ifslave k_sinfo;
3862         struct ifslave __user *u_sinfo = NULL;
3863         struct mii_ioctl_data *mii = NULL;
3864         int res = 0;
3865
3866         pr_debug("bond_ioctl: master=%s, cmd=%d\n", bond_dev->name, cmd);
3867
3868         switch (cmd) {
3869         case SIOCGMIIPHY:
3870                 mii = if_mii(ifr);
3871                 if (!mii)
3872                         return -EINVAL;
3873
3874                 mii->phy_id = 0;
3875                 /* Fall Through */
3876         case SIOCGMIIREG:
3877                 /*
3878                  * We do this again just in case we were called by SIOCGMIIREG
3879                  * instead of SIOCGMIIPHY.
3880                  */
3881                 mii = if_mii(ifr);
3882                 if (!mii)
3883                         return -EINVAL;
3884
3885
3886                 if (mii->reg_num == 1) {
3887                         struct bonding *bond = netdev_priv(bond_dev);
3888                         mii->val_out = 0;
3889                         read_lock(&bond->lock);
3890                         read_lock(&bond->curr_slave_lock);
3891                         if (netif_carrier_ok(bond->dev))
3892                                 mii->val_out = BMSR_LSTATUS;
3893
3894                         read_unlock(&bond->curr_slave_lock);
3895                         read_unlock(&bond->lock);
3896                 }
3897
3898                 return 0;
3899         case BOND_INFO_QUERY_OLD:
3900         case SIOCBONDINFOQUERY:
3901                 u_binfo = (struct ifbond __user *)ifr->ifr_data;
3902
3903                 if (copy_from_user(&k_binfo, u_binfo, sizeof(ifbond)))
3904                         return -EFAULT;
3905
3906                 res = bond_info_query(bond_dev, &k_binfo);
3907                 if (res == 0 &&
3908                     copy_to_user(u_binfo, &k_binfo, sizeof(ifbond)))
3909                         return -EFAULT;
3910
3911                 return res;
3912         case BOND_SLAVE_INFO_QUERY_OLD:
3913         case SIOCBONDSLAVEINFOQUERY:
3914                 u_sinfo = (struct ifslave __user *)ifr->ifr_data;
3915
3916                 if (copy_from_user(&k_sinfo, u_sinfo, sizeof(ifslave)))
3917                         return -EFAULT;
3918
3919                 res = bond_slave_info_query(bond_dev, &k_sinfo);
3920                 if (res == 0 &&
3921                     copy_to_user(u_sinfo, &k_sinfo, sizeof(ifslave)))
3922                         return -EFAULT;
3923
3924                 return res;
3925         default:
3926                 /* Go on */
3927                 break;
3928         }
3929
3930         if (!capable(CAP_NET_ADMIN))
3931                 return -EPERM;
3932
3933         slave_dev = dev_get_by_name(dev_net(bond_dev), ifr->ifr_slave);
3934
3935         pr_debug("slave_dev=%p:\n", slave_dev);
3936
3937         if (!slave_dev)
3938                 res = -ENODEV;
3939         else {
3940                 pr_debug("slave_dev->name=%s:\n", slave_dev->name);
3941                 switch (cmd) {
3942                 case BOND_ENSLAVE_OLD:
3943                 case SIOCBONDENSLAVE:
3944                         res = bond_enslave(bond_dev, slave_dev);
3945                         break;
3946                 case BOND_RELEASE_OLD:
3947                 case SIOCBONDRELEASE:
3948                         res = bond_release(bond_dev, slave_dev);
3949                         break;
3950                 case BOND_SETHWADDR_OLD:
3951                 case SIOCBONDSETHWADDR:
3952                         res = bond_sethwaddr(bond_dev, slave_dev);
3953                         break;
3954                 case BOND_CHANGE_ACTIVE_OLD:
3955                 case SIOCBONDCHANGEACTIVE:
3956                         res = bond_ioctl_change_active(bond_dev, slave_dev);
3957                         break;
3958                 default:
3959                         res = -EOPNOTSUPP;
3960                 }
3961
3962                 dev_put(slave_dev);
3963         }
3964
3965         return res;
3966 }
3967
3968 static bool bond_addr_in_mc_list(unsigned char *addr,
3969                                  struct netdev_hw_addr_list *list,
3970                                  int addrlen)
3971 {
3972         struct netdev_hw_addr *ha;
3973
3974         netdev_hw_addr_list_for_each(ha, list)
3975                 if (!memcmp(ha->addr, addr, addrlen))
3976                         return true;
3977
3978         return false;
3979 }
3980
3981 static void bond_set_multicast_list(struct net_device *bond_dev)
3982 {
3983         struct bonding *bond = netdev_priv(bond_dev);
3984         struct netdev_hw_addr *ha;
3985         bool found;
3986
3987         /*
3988          * Do promisc before checking multicast_mode
3989          */
3990         if ((bond_dev->flags & IFF_PROMISC) && !(bond->flags & IFF_PROMISC))
3991                 /*
3992                  * FIXME: Need to handle the error when one of the multi-slaves
3993                  * encounters error.
3994                  */
3995                 bond_set_promiscuity(bond, 1);
3996
3997
3998         if (!(bond_dev->flags & IFF_PROMISC) && (bond->flags & IFF_PROMISC))
3999                 bond_set_promiscuity(bond, -1);
4000
4001
4002         /* set allmulti flag to slaves */
4003         if ((bond_dev->flags & IFF_ALLMULTI) && !(bond->flags & IFF_ALLMULTI))
4004                 /*
4005                  * FIXME: Need to handle the error when one of the multi-slaves
4006                  * encounters error.
4007                  */
4008                 bond_set_allmulti(bond, 1);
4009
4010
4011         if (!(bond_dev->flags & IFF_ALLMULTI) && (bond->flags & IFF_ALLMULTI))
4012                 bond_set_allmulti(bond, -1);
4013
4014
4015         read_lock(&bond->lock);
4016
4017         bond->flags = bond_dev->flags;
4018
4019         /* looking for addresses to add to slaves' mc list */
4020         netdev_for_each_mc_addr(ha, bond_dev) {
4021                 found = bond_addr_in_mc_list(ha->addr, &bond->mc_list,
4022                                              bond_dev->addr_len);
4023                 if (!found)
4024                         bond_mc_add(bond, ha->addr);
4025         }
4026
4027         /* looking for addresses to delete from slaves' list */
4028         netdev_hw_addr_list_for_each(ha, &bond->mc_list) {
4029                 found = bond_addr_in_mc_list(ha->addr, &bond_dev->mc,
4030                                              bond_dev->addr_len);
4031                 if (!found)
4032                         bond_mc_del(bond, ha->addr);
4033         }
4034
4035         /* save master's multicast list */
4036         __hw_addr_flush(&bond->mc_list);
4037         __hw_addr_add_multiple(&bond->mc_list, &bond_dev->mc,
4038                                bond_dev->addr_len, NETDEV_HW_ADDR_T_MULTICAST);
4039
4040         read_unlock(&bond->lock);
4041 }
4042
4043 static int bond_neigh_setup(struct net_device *dev, struct neigh_parms *parms)
4044 {
4045         struct bonding *bond = netdev_priv(dev);
4046         struct slave *slave = bond->first_slave;
4047
4048         if (slave) {
4049                 const struct net_device_ops *slave_ops
4050                         = slave->dev->netdev_ops;
4051                 if (slave_ops->ndo_neigh_setup)
4052                         return slave_ops->ndo_neigh_setup(slave->dev, parms);
4053         }
4054         return 0;
4055 }
4056
4057 /*
4058  * Change the MTU of all of a master's slaves to match the master
4059  */
4060 static int bond_change_mtu(struct net_device *bond_dev, int new_mtu)
4061 {
4062         struct bonding *bond = netdev_priv(bond_dev);
4063         struct slave *slave, *stop_at;
4064         int res = 0;
4065         int i;
4066
4067         pr_debug("bond=%p, name=%s, new_mtu=%d\n", bond,
4068                  (bond_dev ? bond_dev->name : "None"), new_mtu);
4069
4070         /* Can't hold bond->lock with bh disabled here since
4071          * some base drivers panic. On the other hand we can't
4072          * hold bond->lock without bh disabled because we'll
4073          * deadlock. The only solution is to rely on the fact
4074          * that we're under rtnl_lock here, and the slaves
4075          * list won't change. This doesn't solve the problem
4076          * of setting the slave's MTU while it is
4077          * transmitting, but the assumption is that the base
4078          * driver can handle that.
4079          *
4080          * TODO: figure out a way to safely iterate the slaves
4081          * list, but without holding a lock around the actual
4082          * call to the base driver.
4083          */
4084
4085         bond_for_each_slave(bond, slave, i) {
4086                 pr_debug("s %p s->p %p c_m %p\n",
4087                          slave,
4088                          slave->prev,
4089                          slave->dev->netdev_ops->ndo_change_mtu);
4090
4091                 res = dev_set_mtu(slave->dev, new_mtu);
4092
4093                 if (res) {
4094                         /* If we failed to set the slave's mtu to the new value
4095                          * we must abort the operation even in ACTIVE_BACKUP
4096                          * mode, because if we allow the backup slaves to have
4097                          * different mtu values than the active slave we'll
4098                          * need to change their mtu when doing a failover. That
4099                          * means changing their mtu from timer context, which
4100                          * is probably not a good idea.
4101                          */
4102                         pr_debug("err %d %s\n", res, slave->dev->name);
4103                         goto unwind;
4104                 }
4105         }
4106
4107         bond_dev->mtu = new_mtu;
4108
4109         return 0;
4110
4111 unwind:
4112         /* unwind from head to the slave that failed */
4113         stop_at = slave;
4114         bond_for_each_slave_from_to(bond, slave, i, bond->first_slave, stop_at) {
4115                 int tmp_res;
4116
4117                 tmp_res = dev_set_mtu(slave->dev, bond_dev->mtu);
4118                 if (tmp_res) {
4119                         pr_debug("unwind err %d dev %s\n",
4120                                  tmp_res, slave->dev->name);
4121                 }
4122         }
4123
4124         return res;
4125 }
4126
4127 /*
4128  * Change HW address
4129  *
4130  * Note that many devices must be down to change the HW address, and
4131  * downing the master releases all slaves.  We can make bonds full of
4132  * bonding devices to test this, however.
4133  */
4134 static int bond_set_mac_address(struct net_device *bond_dev, void *addr)
4135 {
4136         struct bonding *bond = netdev_priv(bond_dev);
4137         struct sockaddr *sa = addr, tmp_sa;
4138         struct slave *slave, *stop_at;
4139         int res = 0;
4140         int i;
4141
4142         if (bond->params.mode == BOND_MODE_ALB)
4143                 return bond_alb_set_mac_address(bond_dev, addr);
4144
4145
4146         pr_debug("bond=%p, name=%s\n",
4147                  bond, bond_dev ? bond_dev->name : "None");
4148
4149         /*
4150          * If fail_over_mac is set to active, do nothing and return
4151          * success.  Returning an error causes ifenslave to fail.
4152          */
4153         if (bond->params.fail_over_mac == BOND_FOM_ACTIVE)
4154                 return 0;
4155
4156         if (!is_valid_ether_addr(sa->sa_data))
4157                 return -EADDRNOTAVAIL;
4158
4159         /* Can't hold bond->lock with bh disabled here since
4160          * some base drivers panic. On the other hand we can't
4161          * hold bond->lock without bh disabled because we'll
4162          * deadlock. The only solution is to rely on the fact
4163          * that we're under rtnl_lock here, and the slaves
4164          * list won't change. This doesn't solve the problem
4165          * of setting the slave's hw address while it is
4166          * transmitting, but the assumption is that the base
4167          * driver can handle that.
4168          *
4169          * TODO: figure out a way to safely iterate the slaves
4170          * list, but without holding a lock around the actual
4171          * call to the base driver.
4172          */
4173
4174         bond_for_each_slave(bond, slave, i) {
4175                 const struct net_device_ops *slave_ops = slave->dev->netdev_ops;
4176                 pr_debug("slave %p %s\n", slave, slave->dev->name);
4177
4178                 if (slave_ops->ndo_set_mac_address == NULL) {
4179                         res = -EOPNOTSUPP;
4180                         pr_debug("EOPNOTSUPP %s\n", slave->dev->name);
4181                         goto unwind;
4182                 }
4183
4184                 res = dev_set_mac_address(slave->dev, addr);
4185                 if (res) {
4186                         /* TODO: consider downing the slave
4187                          * and retry ?
4188                          * User should expect communications
4189                          * breakage anyway until ARP finish
4190                          * updating, so...
4191                          */
4192                         pr_debug("err %d %s\n", res, slave->dev->name);
4193                         goto unwind;
4194                 }
4195         }
4196
4197         /* success */
4198         memcpy(bond_dev->dev_addr, sa->sa_data, bond_dev->addr_len);
4199         return 0;
4200
4201 unwind:
4202         memcpy(tmp_sa.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
4203         tmp_sa.sa_family = bond_dev->type;
4204
4205         /* unwind from head to the slave that failed */
4206         stop_at = slave;
4207         bond_for_each_slave_from_to(bond, slave, i, bond->first_slave, stop_at) {
4208                 int tmp_res;
4209
4210                 tmp_res = dev_set_mac_address(slave->dev, &tmp_sa);
4211                 if (tmp_res) {
4212                         pr_debug("unwind err %d dev %s\n",
4213                                  tmp_res, slave->dev->name);
4214                 }
4215         }
4216
4217         return res;
4218 }
4219
4220 static int bond_xmit_roundrobin(struct sk_buff *skb, struct net_device *bond_dev)
4221 {
4222         struct bonding *bond = netdev_priv(bond_dev);
4223         struct slave *slave, *start_at;
4224         int i, slave_no, res = 1;
4225         struct iphdr *iph = ip_hdr(skb);
4226
4227         read_lock(&bond->lock);
4228
4229         if (!BOND_IS_OK(bond))
4230                 goto out;
4231         /*
4232          * Start with the curr_active_slave that joined the bond as the
4233          * default for sending IGMP traffic.  For failover purposes one
4234          * needs to maintain some consistency for the interface that will
4235          * send the join/membership reports.  The curr_active_slave found
4236          * will send all of this type of traffic.
4237          */
4238         if ((iph->protocol == IPPROTO_IGMP) &&
4239             (skb->protocol == htons(ETH_P_IP))) {
4240
4241                 read_lock(&bond->curr_slave_lock);
4242                 slave = bond->curr_active_slave;
4243                 read_unlock(&bond->curr_slave_lock);
4244
4245                 if (!slave)
4246                         goto out;
4247         } else {
4248                 /*
4249                  * Concurrent TX may collide on rr_tx_counter; we accept
4250                  * that as being rare enough not to justify using an
4251                  * atomic op here.
4252                  */
4253                 slave_no = bond->rr_tx_counter++ % bond->slave_cnt;
4254
4255                 bond_for_each_slave(bond, slave, i) {
4256                         slave_no--;
4257                         if (slave_no < 0)
4258                                 break;
4259                 }
4260         }
4261
4262         start_at = slave;
4263         bond_for_each_slave_from(bond, slave, i, start_at) {
4264                 if (IS_UP(slave->dev) &&
4265                     (slave->link == BOND_LINK_UP) &&
4266                     (slave->state == BOND_STATE_ACTIVE)) {
4267                         res = bond_dev_queue_xmit(bond, skb, slave->dev);
4268                         break;
4269                 }
4270         }
4271
4272 out:
4273         if (res) {
4274                 /* no suitable interface, frame not sent */
4275                 dev_kfree_skb(skb);
4276         }
4277         read_unlock(&bond->lock);
4278         return NETDEV_TX_OK;
4279 }
4280
4281
4282 /*
4283  * in active-backup mode, we know that bond->curr_active_slave is always valid if
4284  * the bond has a usable interface.
4285  */
4286 static int bond_xmit_activebackup(struct sk_buff *skb, struct net_device *bond_dev)
4287 {
4288         struct bonding *bond = netdev_priv(bond_dev);
4289         int res = 1;
4290
4291         read_lock(&bond->lock);
4292         read_lock(&bond->curr_slave_lock);
4293
4294         if (!BOND_IS_OK(bond))
4295                 goto out;
4296
4297         if (!bond->curr_active_slave)
4298                 goto out;
4299
4300         res = bond_dev_queue_xmit(bond, skb, bond->curr_active_slave->dev);
4301
4302 out:
4303         if (res)
4304                 /* no suitable interface, frame not sent */
4305                 dev_kfree_skb(skb);
4306
4307         read_unlock(&bond->curr_slave_lock);
4308         read_unlock(&bond->lock);
4309         return NETDEV_TX_OK;
4310 }
4311
4312 /*
4313  * In bond_xmit_xor() , we determine the output device by using a pre-
4314  * determined xmit_hash_policy(), If the selected device is not enabled,
4315  * find the next active slave.
4316  */
4317 static int bond_xmit_xor(struct sk_buff *skb, struct net_device *bond_dev)
4318 {
4319         struct bonding *bond = netdev_priv(bond_dev);
4320         struct slave *slave, *start_at;
4321         int slave_no;
4322         int i;
4323         int res = 1;
4324
4325         read_lock(&bond->lock);
4326
4327         if (!BOND_IS_OK(bond))
4328                 goto out;
4329
4330         slave_no = bond->xmit_hash_policy(skb, bond->slave_cnt);
4331
4332         bond_for_each_slave(bond, slave, i) {
4333                 slave_no--;
4334                 if (slave_no < 0)
4335                         break;
4336         }
4337
4338         start_at = slave;
4339
4340         bond_for_each_slave_from(bond, slave, i, start_at) {
4341                 if (IS_UP(slave->dev) &&
4342                     (slave->link == BOND_LINK_UP) &&
4343                     (slave->state == BOND_STATE_ACTIVE)) {
4344                         res = bond_dev_queue_xmit(bond, skb, slave->dev);
4345                         break;
4346                 }
4347         }
4348
4349 out:
4350         if (res) {
4351                 /* no suitable interface, frame not sent */
4352                 dev_kfree_skb(skb);
4353         }
4354         read_unlock(&bond->lock);
4355         return NETDEV_TX_OK;
4356 }
4357
4358 /*
4359  * in broadcast mode, we send everything to all usable interfaces.
4360  */
4361 static int bond_xmit_broadcast(struct sk_buff *skb, struct net_device *bond_dev)
4362 {
4363         struct bonding *bond = netdev_priv(bond_dev);
4364         struct slave *slave, *start_at;
4365         struct net_device *tx_dev = NULL;
4366         int i;
4367         int res = 1;
4368
4369         read_lock(&bond->lock);
4370
4371         if (!BOND_IS_OK(bond))
4372                 goto out;
4373
4374         read_lock(&bond->curr_slave_lock);
4375         start_at = bond->curr_active_slave;
4376         read_unlock(&bond->curr_slave_lock);
4377
4378         if (!start_at)
4379                 goto out;
4380
4381         bond_for_each_slave_from(bond, slave, i, start_at) {
4382                 if (IS_UP(slave->dev) &&
4383                     (slave->link == BOND_LINK_UP) &&
4384                     (slave->state == BOND_STATE_ACTIVE)) {
4385                         if (tx_dev) {
4386                                 struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
4387                                 if (!skb2) {
4388                                         pr_err("%s: Error: bond_xmit_broadcast(): skb_clone() failed\n",
4389                                                bond_dev->name);
4390                                         continue;
4391                                 }
4392
4393                                 res = bond_dev_queue_xmit(bond, skb2, tx_dev);
4394                                 if (res) {
4395                                         dev_kfree_skb(skb2);
4396                                         continue;
4397                                 }
4398                         }
4399                         tx_dev = slave->dev;
4400                 }
4401         }
4402
4403         if (tx_dev)
4404                 res = bond_dev_queue_xmit(bond, skb, tx_dev);
4405
4406 out:
4407         if (res)
4408                 /* no suitable interface, frame not sent */
4409                 dev_kfree_skb(skb);
4410
4411         /* frame sent to all suitable interfaces */
4412         read_unlock(&bond->lock);
4413         return NETDEV_TX_OK;
4414 }
4415
4416 /*------------------------- Device initialization ---------------------------*/
4417
4418 static void bond_set_xmit_hash_policy(struct bonding *bond)
4419 {
4420         switch (bond->params.xmit_policy) {
4421         case BOND_XMIT_POLICY_LAYER23:
4422                 bond->xmit_hash_policy = bond_xmit_hash_policy_l23;
4423                 break;
4424         case BOND_XMIT_POLICY_LAYER34:
4425                 bond->xmit_hash_policy = bond_xmit_hash_policy_l34;
4426                 break;
4427         case BOND_XMIT_POLICY_LAYER2:
4428         default:
4429                 bond->xmit_hash_policy = bond_xmit_hash_policy_l2;
4430                 break;
4431         }
4432 }
4433
4434 /*
4435  * Lookup the slave that corresponds to a qid
4436  */
4437 static inline int bond_slave_override(struct bonding *bond,
4438                                       struct sk_buff *skb)
4439 {
4440         int i, res = 1;
4441         struct slave *slave = NULL;
4442         struct slave *check_slave;
4443
4444         read_lock(&bond->lock);
4445
4446         if (!BOND_IS_OK(bond) || !skb->queue_mapping)
4447                 goto out;
4448
4449         /* Find out if any slaves have the same mapping as this skb. */
4450         bond_for_each_slave(bond, check_slave, i) {
4451                 if (check_slave->queue_id == skb->queue_mapping) {
4452                         slave = check_slave;
4453                         break;
4454                 }
4455         }
4456
4457         /* If the slave isn't UP, use default transmit policy. */
4458         if (slave && slave->queue_id && IS_UP(slave->dev) &&
4459             (slave->link == BOND_LINK_UP)) {
4460                 res = bond_dev_queue_xmit(bond, skb, slave->dev);
4461         }
4462
4463 out:
4464         read_unlock(&bond->lock);
4465         return res;
4466 }
4467
4468 static u16 bond_select_queue(struct net_device *dev, struct sk_buff *skb)
4469 {
4470         /*
4471          * This helper function exists to help dev_pick_tx get the correct
4472          * destination queue.  Using a helper function skips the a call to
4473          * skb_tx_hash and will put the skbs in the queue we expect on their
4474          * way down to the bonding driver.
4475          */
4476         return skb->queue_mapping;
4477 }
4478
4479 static netdev_tx_t bond_start_xmit(struct sk_buff *skb, struct net_device *dev)
4480 {
4481         struct bonding *bond = netdev_priv(dev);
4482
4483         if (TX_QUEUE_OVERRIDE(bond->params.mode)) {
4484                 if (!bond_slave_override(bond, skb))
4485                         return NETDEV_TX_OK;
4486         }
4487
4488         switch (bond->params.mode) {
4489         case BOND_MODE_ROUNDROBIN:
4490                 return bond_xmit_roundrobin(skb, dev);
4491         case BOND_MODE_ACTIVEBACKUP:
4492                 return bond_xmit_activebackup(skb, dev);
4493         case BOND_MODE_XOR:
4494                 return bond_xmit_xor(skb, dev);
4495         case BOND_MODE_BROADCAST:
4496                 return bond_xmit_broadcast(skb, dev);
4497         case BOND_MODE_8023AD:
4498                 return bond_3ad_xmit_xor(skb, dev);
4499         case BOND_MODE_ALB:
4500         case BOND_MODE_TLB:
4501                 return bond_alb_xmit(skb, dev);
4502         default:
4503                 /* Should never happen, mode already checked */
4504                 pr_err("%s: Error: Unknown bonding mode %d\n",
4505                        dev->name, bond->params.mode);
4506                 WARN_ON_ONCE(1);
4507                 dev_kfree_skb(skb);
4508                 return NETDEV_TX_OK;
4509         }
4510 }
4511
4512
4513 /*
4514  * set bond mode specific net device operations
4515  */
4516 void bond_set_mode_ops(struct bonding *bond, int mode)
4517 {
4518         struct net_device *bond_dev = bond->dev;
4519
4520         switch (mode) {
4521         case BOND_MODE_ROUNDROBIN:
4522                 break;
4523         case BOND_MODE_ACTIVEBACKUP:
4524                 break;
4525         case BOND_MODE_XOR:
4526                 bond_set_xmit_hash_policy(bond);
4527                 break;
4528         case BOND_MODE_BROADCAST:
4529                 break;
4530         case BOND_MODE_8023AD:
4531                 bond_set_master_3ad_flags(bond);
4532                 bond_set_xmit_hash_policy(bond);
4533                 break;
4534         case BOND_MODE_ALB:
4535                 bond_set_master_alb_flags(bond);
4536                 /* FALLTHRU */
4537         case BOND_MODE_TLB:
4538                 break;
4539         default:
4540                 /* Should never happen, mode already checked */
4541                 pr_err("%s: Error: Unknown bonding mode %d\n",
4542                        bond_dev->name, mode);
4543                 break;
4544         }
4545 }
4546
4547 static void bond_ethtool_get_drvinfo(struct net_device *bond_dev,
4548                                     struct ethtool_drvinfo *drvinfo)
4549 {
4550         strncpy(drvinfo->driver, DRV_NAME, 32);
4551         strncpy(drvinfo->version, DRV_VERSION, 32);
4552         snprintf(drvinfo->fw_version, 32, "%d", BOND_ABI_VERSION);
4553 }
4554
4555 static const struct ethtool_ops bond_ethtool_ops = {
4556         .get_drvinfo            = bond_ethtool_get_drvinfo,
4557         .get_link               = ethtool_op_get_link,
4558         .get_tx_csum            = ethtool_op_get_tx_csum,
4559         .get_sg                 = ethtool_op_get_sg,
4560         .get_tso                = ethtool_op_get_tso,
4561         .get_ufo                = ethtool_op_get_ufo,
4562         .get_flags              = ethtool_op_get_flags,
4563 };
4564
4565 static const struct net_device_ops bond_netdev_ops = {
4566         .ndo_init               = bond_init,
4567         .ndo_uninit             = bond_uninit,
4568         .ndo_open               = bond_open,
4569         .ndo_stop               = bond_close,
4570         .ndo_start_xmit         = bond_start_xmit,
4571         .ndo_select_queue       = bond_select_queue,
4572         .ndo_get_stats          = bond_get_stats,
4573         .ndo_do_ioctl           = bond_do_ioctl,
4574         .ndo_set_multicast_list = bond_set_multicast_list,
4575         .ndo_change_mtu         = bond_change_mtu,
4576         .ndo_set_mac_address    = bond_set_mac_address,
4577         .ndo_neigh_setup        = bond_neigh_setup,
4578         .ndo_vlan_rx_register   = bond_vlan_rx_register,
4579         .ndo_vlan_rx_add_vid    = bond_vlan_rx_add_vid,
4580         .ndo_vlan_rx_kill_vid   = bond_vlan_rx_kill_vid,
4581 #ifdef CONFIG_NET_POLL_CONTROLLER
4582         .ndo_netpoll_cleanup    = bond_netpoll_cleanup,
4583         .ndo_poll_controller    = bond_poll_controller,
4584 #endif
4585 };
4586
4587 static void bond_destructor(struct net_device *bond_dev)
4588 {
4589         struct bonding *bond = netdev_priv(bond_dev);
4590         if (bond->wq)
4591                 destroy_workqueue(bond->wq);
4592         free_netdev(bond_dev);
4593 }
4594
4595 static void bond_setup(struct net_device *bond_dev)
4596 {
4597         struct bonding *bond = netdev_priv(bond_dev);
4598
4599         /* initialize rwlocks */
4600         rwlock_init(&bond->lock);
4601         rwlock_init(&bond->curr_slave_lock);
4602
4603         bond->params = bonding_defaults;
4604
4605         /* Initialize pointers */
4606         bond->dev = bond_dev;
4607         INIT_LIST_HEAD(&bond->vlan_list);
4608
4609         /* Initialize the device entry points */
4610         ether_setup(bond_dev);
4611         bond_dev->netdev_ops = &bond_netdev_ops;
4612         bond_dev->ethtool_ops = &bond_ethtool_ops;
4613         bond_set_mode_ops(bond, bond->params.mode);
4614
4615         bond_dev->destructor = bond_destructor;
4616
4617         /* Initialize the device options */
4618         bond_dev->tx_queue_len = 0;
4619         bond_dev->flags |= IFF_MASTER|IFF_MULTICAST;
4620         bond_dev->priv_flags |= IFF_BONDING;
4621         bond_dev->priv_flags &= ~IFF_XMIT_DST_RELEASE;
4622
4623         if (bond->params.arp_interval)
4624                 bond_dev->priv_flags |= IFF_MASTER_ARPMON;
4625
4626         /* At first, we block adding VLANs. That's the only way to
4627          * prevent problems that occur when adding VLANs over an
4628          * empty bond. The block will be removed once non-challenged
4629          * slaves are enslaved.
4630          */
4631         bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
4632
4633         /* don't acquire bond device's netif_tx_lock when
4634          * transmitting */
4635         bond_dev->features |= NETIF_F_LLTX;
4636
4637         /* By default, we declare the bond to be fully
4638          * VLAN hardware accelerated capable. Special
4639          * care is taken in the various xmit functions
4640          * when there are slaves that are not hw accel
4641          * capable
4642          */
4643         bond_dev->features |= (NETIF_F_HW_VLAN_TX |
4644                                NETIF_F_HW_VLAN_RX |
4645                                NETIF_F_HW_VLAN_FILTER);
4646
4647 }
4648
4649 static void bond_work_cancel_all(struct bonding *bond)
4650 {
4651         write_lock_bh(&bond->lock);
4652         bond->kill_timers = 1;
4653         write_unlock_bh(&bond->lock);
4654
4655         if (bond->params.miimon && delayed_work_pending(&bond->mii_work))
4656                 cancel_delayed_work(&bond->mii_work);
4657
4658         if (bond->params.arp_interval && delayed_work_pending(&bond->arp_work))
4659                 cancel_delayed_work(&bond->arp_work);
4660
4661         if (bond->params.mode == BOND_MODE_ALB &&
4662             delayed_work_pending(&bond->alb_work))
4663                 cancel_delayed_work(&bond->alb_work);
4664
4665         if (bond->params.mode == BOND_MODE_8023AD &&
4666             delayed_work_pending(&bond->ad_work))
4667                 cancel_delayed_work(&bond->ad_work);
4668 }
4669
4670 /*
4671 * Destroy a bonding device.
4672 * Must be under rtnl_lock when this function is called.
4673 */
4674 static void bond_uninit(struct net_device *bond_dev)
4675 {
4676         struct bonding *bond = netdev_priv(bond_dev);
4677
4678         bond_netpoll_cleanup(bond_dev);
4679
4680         /* Release the bonded slaves */
4681         bond_release_all(bond_dev);
4682
4683         list_del(&bond->bond_list);
4684
4685         bond_work_cancel_all(bond);
4686
4687         bond_remove_proc_entry(bond);
4688
4689         __hw_addr_flush(&bond->mc_list);
4690 }
4691
4692 /*------------------------- Module initialization ---------------------------*/
4693
4694 /*
4695  * Convert string input module parms.  Accept either the
4696  * number of the mode or its string name.  A bit complicated because
4697  * some mode names are substrings of other names, and calls from sysfs
4698  * may have whitespace in the name (trailing newlines, for example).
4699  */
4700 int bond_parse_parm(const char *buf, const struct bond_parm_tbl *tbl)
4701 {
4702         int modeint = -1, i, rv;
4703         char *p, modestr[BOND_MAX_MODENAME_LEN + 1] = { 0, };
4704
4705         for (p = (char *)buf; *p; p++)
4706                 if (!(isdigit(*p) || isspace(*p)))
4707                         break;
4708
4709         if (*p)
4710                 rv = sscanf(buf, "%20s", modestr);
4711         else
4712                 rv = sscanf(buf, "%d", &modeint);
4713
4714         if (!rv)
4715                 return -1;
4716
4717         for (i = 0; tbl[i].modename; i++) {
4718                 if (modeint == tbl[i].mode)
4719                         return tbl[i].mode;
4720                 if (strcmp(modestr, tbl[i].modename) == 0)
4721                         return tbl[i].mode;
4722         }
4723
4724         return -1;
4725 }
4726
4727 static int bond_check_params(struct bond_params *params)
4728 {
4729         int arp_validate_value, fail_over_mac_value, primary_reselect_value;
4730
4731         /*
4732          * Convert string parameters.
4733          */
4734         if (mode) {
4735                 bond_mode = bond_parse_parm(mode, bond_mode_tbl);
4736                 if (bond_mode == -1) {
4737                         pr_err("Error: Invalid bonding mode \"%s\"\n",
4738                                mode == NULL ? "NULL" : mode);
4739                         return -EINVAL;
4740                 }
4741         }
4742
4743         if (xmit_hash_policy) {
4744                 if ((bond_mode != BOND_MODE_XOR) &&
4745                     (bond_mode != BOND_MODE_8023AD)) {
4746                         pr_info("xmit_hash_policy param is irrelevant in mode %s\n",
4747                                bond_mode_name(bond_mode));
4748                 } else {
4749                         xmit_hashtype = bond_parse_parm(xmit_hash_policy,
4750                                                         xmit_hashtype_tbl);
4751                         if (xmit_hashtype == -1) {
4752                                 pr_err("Error: Invalid xmit_hash_policy \"%s\"\n",
4753                                        xmit_hash_policy == NULL ? "NULL" :
4754                                        xmit_hash_policy);
4755                                 return -EINVAL;
4756                         }
4757                 }
4758         }
4759
4760         if (lacp_rate) {
4761                 if (bond_mode != BOND_MODE_8023AD) {
4762                         pr_info("lacp_rate param is irrelevant in mode %s\n",
4763                                 bond_mode_name(bond_mode));
4764                 } else {
4765                         lacp_fast = bond_parse_parm(lacp_rate, bond_lacp_tbl);
4766                         if (lacp_fast == -1) {
4767                                 pr_err("Error: Invalid lacp rate \"%s\"\n",
4768                                        lacp_rate == NULL ? "NULL" : lacp_rate);
4769                                 return -EINVAL;
4770                         }
4771                 }
4772         }
4773
4774         if (ad_select) {
4775                 params->ad_select = bond_parse_parm(ad_select, ad_select_tbl);
4776                 if (params->ad_select == -1) {
4777                         pr_err("Error: Invalid ad_select \"%s\"\n",
4778                                ad_select == NULL ? "NULL" : ad_select);
4779                         return -EINVAL;
4780                 }
4781
4782                 if (bond_mode != BOND_MODE_8023AD) {
4783                         pr_warning("ad_select param only affects 802.3ad mode\n");
4784                 }
4785         } else {
4786                 params->ad_select = BOND_AD_STABLE;
4787         }
4788
4789         if (max_bonds < 0) {
4790                 pr_warning("Warning: max_bonds (%d) not in range %d-%d, so it was reset to BOND_DEFAULT_MAX_BONDS (%d)\n",
4791                            max_bonds, 0, INT_MAX, BOND_DEFAULT_MAX_BONDS);
4792                 max_bonds = BOND_DEFAULT_MAX_BONDS;
4793         }
4794
4795         if (miimon < 0) {
4796                 pr_warning("Warning: miimon module parameter (%d), not in range 0-%d, so it was reset to %d\n",
4797                            miimon, INT_MAX, BOND_LINK_MON_INTERV);
4798                 miimon = BOND_LINK_MON_INTERV;
4799         }
4800
4801         if (updelay < 0) {
4802                 pr_warning("Warning: updelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4803                            updelay, INT_MAX);
4804                 updelay = 0;
4805         }
4806
4807         if (downdelay < 0) {
4808                 pr_warning("Warning: downdelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4809                            downdelay, INT_MAX);
4810                 downdelay = 0;
4811         }
4812
4813         if ((use_carrier != 0) && (use_carrier != 1)) {
4814                 pr_warning("Warning: use_carrier module parameter (%d), not of valid value (0/1), so it was set to 1\n",
4815                            use_carrier);
4816                 use_carrier = 1;
4817         }
4818
4819         if (num_grat_arp < 0 || num_grat_arp > 255) {
4820                 pr_warning("Warning: num_grat_arp (%d) not in range 0-255 so it was reset to 1\n",
4821                            num_grat_arp);
4822                 num_grat_arp = 1;
4823         }
4824
4825         if (num_unsol_na < 0 || num_unsol_na > 255) {
4826                 pr_warning("Warning: num_unsol_na (%d) not in range 0-255 so it was reset to 1\n",
4827                            num_unsol_na);
4828                 num_unsol_na = 1;
4829         }
4830
4831         /* reset values for 802.3ad */
4832         if (bond_mode == BOND_MODE_8023AD) {
4833                 if (!miimon) {
4834                         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");
4835                         pr_warning("Forcing miimon to 100msec\n");
4836                         miimon = 100;
4837                 }
4838         }
4839
4840         if (tx_queues < 1 || tx_queues > 255) {
4841                 pr_warning("Warning: tx_queues (%d) should be between "
4842                            "1 and 255, resetting to %d\n",
4843                            tx_queues, BOND_DEFAULT_TX_QUEUES);
4844                 tx_queues = BOND_DEFAULT_TX_QUEUES;
4845         }
4846
4847         if ((all_slaves_active != 0) && (all_slaves_active != 1)) {
4848                 pr_warning("Warning: all_slaves_active module parameter (%d), "
4849                            "not of valid value (0/1), so it was set to "
4850                            "0\n", all_slaves_active);
4851                 all_slaves_active = 0;
4852         }
4853
4854         /* reset values for TLB/ALB */
4855         if ((bond_mode == BOND_MODE_TLB) ||
4856             (bond_mode == BOND_MODE_ALB)) {
4857                 if (!miimon) {
4858                         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");
4859                         pr_warning("Forcing miimon to 100msec\n");
4860                         miimon = 100;
4861                 }
4862         }
4863
4864         if (bond_mode == BOND_MODE_ALB) {
4865                 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",
4866                           updelay);
4867         }
4868
4869         if (!miimon) {
4870                 if (updelay || downdelay) {
4871                         /* just warn the user the up/down delay will have
4872                          * no effect since miimon is zero...
4873                          */
4874                         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",
4875                                    updelay, downdelay);
4876                 }
4877         } else {
4878                 /* don't allow arp monitoring */
4879                 if (arp_interval) {
4880                         pr_warning("Warning: miimon (%d) and arp_interval (%d) can't be used simultaneously, disabling ARP monitoring\n",
4881                                    miimon, arp_interval);
4882                         arp_interval = 0;
4883                 }
4884
4885                 if ((updelay % miimon) != 0) {
4886                         pr_warning("Warning: updelay (%d) is not a multiple of miimon (%d), updelay rounded to %d ms\n",
4887                                    updelay, miimon,
4888                                    (updelay / miimon) * miimon);
4889                 }
4890
4891                 updelay /= miimon;
4892
4893                 if ((downdelay % miimon) != 0) {
4894                         pr_warning("Warning: downdelay (%d) is not a multiple of miimon (%d), downdelay rounded to %d ms\n",
4895                                    downdelay, miimon,
4896                                    (downdelay / miimon) * miimon);
4897                 }
4898
4899                 downdelay /= miimon;
4900         }
4901
4902         if (arp_interval < 0) {
4903                 pr_warning("Warning: arp_interval module parameter (%d) , not in range 0-%d, so it was reset to %d\n",
4904                            arp_interval, INT_MAX, BOND_LINK_ARP_INTERV);
4905                 arp_interval = BOND_LINK_ARP_INTERV;
4906         }
4907
4908         for (arp_ip_count = 0;
4909              (arp_ip_count < BOND_MAX_ARP_TARGETS) && arp_ip_target[arp_ip_count];
4910              arp_ip_count++) {
4911                 /* not complete check, but should be good enough to
4912                    catch mistakes */
4913                 if (!isdigit(arp_ip_target[arp_ip_count][0])) {
4914                         pr_warning("Warning: bad arp_ip_target module parameter (%s), ARP monitoring will not be performed\n",
4915                                    arp_ip_target[arp_ip_count]);
4916                         arp_interval = 0;
4917                 } else {
4918                         __be32 ip = in_aton(arp_ip_target[arp_ip_count]);
4919                         arp_target[arp_ip_count] = ip;
4920                 }
4921         }
4922
4923         if (arp_interval && !arp_ip_count) {
4924                 /* don't allow arping if no arp_ip_target given... */
4925                 pr_warning("Warning: arp_interval module parameter (%d) specified without providing an arp_ip_target parameter, arp_interval was reset to 0\n",
4926                            arp_interval);
4927                 arp_interval = 0;
4928         }
4929
4930         if (arp_validate) {
4931                 if (bond_mode != BOND_MODE_ACTIVEBACKUP) {
4932                         pr_err("arp_validate only supported in active-backup mode\n");
4933                         return -EINVAL;
4934                 }
4935                 if (!arp_interval) {
4936                         pr_err("arp_validate requires arp_interval\n");
4937                         return -EINVAL;
4938                 }
4939
4940                 arp_validate_value = bond_parse_parm(arp_validate,
4941                                                      arp_validate_tbl);
4942                 if (arp_validate_value == -1) {
4943                         pr_err("Error: invalid arp_validate \"%s\"\n",
4944                                arp_validate == NULL ? "NULL" : arp_validate);
4945                         return -EINVAL;
4946                 }
4947         } else
4948                 arp_validate_value = 0;
4949
4950         if (miimon) {
4951                 pr_info("MII link monitoring set to %d ms\n", miimon);
4952         } else if (arp_interval) {
4953                 int i;
4954
4955                 pr_info("ARP monitoring set to %d ms, validate %s, with %d target(s):",
4956                         arp_interval,
4957                         arp_validate_tbl[arp_validate_value].modename,
4958                         arp_ip_count);
4959
4960                 for (i = 0; i < arp_ip_count; i++)
4961                         pr_info(" %s", arp_ip_target[i]);
4962
4963                 pr_info("\n");
4964
4965         } else if (max_bonds) {
4966                 /* miimon and arp_interval not set, we need one so things
4967                  * work as expected, see bonding.txt for details
4968                  */
4969                 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");
4970         }
4971
4972         if (primary && !USES_PRIMARY(bond_mode)) {
4973                 /* currently, using a primary only makes sense
4974                  * in active backup, TLB or ALB modes
4975                  */
4976                 pr_warning("Warning: %s primary device specified but has no effect in %s mode\n",
4977                            primary, bond_mode_name(bond_mode));
4978                 primary = NULL;
4979         }
4980
4981         if (primary && primary_reselect) {
4982                 primary_reselect_value = bond_parse_parm(primary_reselect,
4983                                                          pri_reselect_tbl);
4984                 if (primary_reselect_value == -1) {
4985                         pr_err("Error: Invalid primary_reselect \"%s\"\n",
4986                                primary_reselect ==
4987                                         NULL ? "NULL" : primary_reselect);
4988                         return -EINVAL;
4989                 }
4990         } else {
4991                 primary_reselect_value = BOND_PRI_RESELECT_ALWAYS;
4992         }
4993
4994         if (fail_over_mac) {
4995                 fail_over_mac_value = bond_parse_parm(fail_over_mac,
4996                                                       fail_over_mac_tbl);
4997                 if (fail_over_mac_value == -1) {
4998                         pr_err("Error: invalid fail_over_mac \"%s\"\n",
4999                                arp_validate == NULL ? "NULL" : arp_validate);
5000                         return -EINVAL;
5001                 }
5002
5003                 if (bond_mode != BOND_MODE_ACTIVEBACKUP)
5004                         pr_warning("Warning: fail_over_mac only affects active-backup mode.\n");
5005         } else {
5006                 fail_over_mac_value = BOND_FOM_NONE;
5007         }
5008
5009         /* fill params struct with the proper values */
5010         params->mode = bond_mode;
5011         params->xmit_policy = xmit_hashtype;
5012         params->miimon = miimon;
5013         params->num_grat_arp = num_grat_arp;
5014         params->num_unsol_na = num_unsol_na;
5015         params->arp_interval = arp_interval;
5016         params->arp_validate = arp_validate_value;
5017         params->updelay = updelay;
5018         params->downdelay = downdelay;
5019         params->use_carrier = use_carrier;
5020         params->lacp_fast = lacp_fast;
5021         params->primary[0] = 0;
5022         params->primary_reselect = primary_reselect_value;
5023         params->fail_over_mac = fail_over_mac_value;
5024         params->tx_queues = tx_queues;
5025         params->all_slaves_active = all_slaves_active;
5026
5027         if (primary) {
5028                 strncpy(params->primary, primary, IFNAMSIZ);
5029                 params->primary[IFNAMSIZ - 1] = 0;
5030         }
5031
5032         memcpy(params->arp_targets, arp_target, sizeof(arp_target));
5033
5034         return 0;
5035 }
5036
5037 static struct lock_class_key bonding_netdev_xmit_lock_key;
5038 static struct lock_class_key bonding_netdev_addr_lock_key;
5039
5040 static void bond_set_lockdep_class_one(struct net_device *dev,
5041                                        struct netdev_queue *txq,
5042                                        void *_unused)
5043 {
5044         lockdep_set_class(&txq->_xmit_lock,
5045                           &bonding_netdev_xmit_lock_key);
5046 }
5047
5048 static void bond_set_lockdep_class(struct net_device *dev)
5049 {
5050         lockdep_set_class(&dev->addr_list_lock,
5051                           &bonding_netdev_addr_lock_key);
5052         netdev_for_each_tx_queue(dev, bond_set_lockdep_class_one, NULL);
5053 }
5054
5055 /*
5056  * Called from registration process
5057  */
5058 static int bond_init(struct net_device *bond_dev)
5059 {
5060         struct bonding *bond = netdev_priv(bond_dev);
5061         struct bond_net *bn = net_generic(dev_net(bond_dev), bond_net_id);
5062
5063         pr_debug("Begin bond_init for %s\n", bond_dev->name);
5064
5065         bond->wq = create_singlethread_workqueue(bond_dev->name);
5066         if (!bond->wq)
5067                 return -ENOMEM;
5068
5069         bond_set_lockdep_class(bond_dev);
5070
5071         netif_carrier_off(bond_dev);
5072
5073         bond_create_proc_entry(bond);
5074         list_add_tail(&bond->bond_list, &bn->dev_list);
5075
5076         bond_prepare_sysfs_group(bond);
5077
5078         __hw_addr_init(&bond->mc_list);
5079         return 0;
5080 }
5081
5082 static int bond_validate(struct nlattr *tb[], struct nlattr *data[])
5083 {
5084         if (tb[IFLA_ADDRESS]) {
5085                 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN)
5086                         return -EINVAL;
5087                 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS])))
5088                         return -EADDRNOTAVAIL;
5089         }
5090         return 0;
5091 }
5092
5093 static struct rtnl_link_ops bond_link_ops __read_mostly = {
5094         .kind           = "bond",
5095         .priv_size      = sizeof(struct bonding),
5096         .setup          = bond_setup,
5097         .validate       = bond_validate,
5098 };
5099
5100 /* Create a new bond based on the specified name and bonding parameters.
5101  * If name is NULL, obtain a suitable "bond%d" name for us.
5102  * Caller must NOT hold rtnl_lock; we need to release it here before we
5103  * set up our sysfs entries.
5104  */
5105 int bond_create(struct net *net, const char *name)
5106 {
5107         struct net_device *bond_dev;
5108         int res;
5109
5110         rtnl_lock();
5111
5112         bond_dev = alloc_netdev_mq(sizeof(struct bonding), name ? name : "",
5113                                 bond_setup, tx_queues);
5114         if (!bond_dev) {
5115                 pr_err("%s: eek! can't alloc netdev!\n", name);
5116                 rtnl_unlock();
5117                 return -ENOMEM;
5118         }
5119
5120         dev_net_set(bond_dev, net);
5121         bond_dev->rtnl_link_ops = &bond_link_ops;
5122
5123         if (!name) {
5124                 res = dev_alloc_name(bond_dev, "bond%d");
5125                 if (res < 0)
5126                         goto out;
5127         }
5128
5129         res = register_netdevice(bond_dev);
5130
5131 out:
5132         rtnl_unlock();
5133         if (res < 0)
5134                 bond_destructor(bond_dev);
5135         return res;
5136 }
5137
5138 static int __net_init bond_net_init(struct net *net)
5139 {
5140         struct bond_net *bn = net_generic(net, bond_net_id);
5141
5142         bn->net = net;
5143         INIT_LIST_HEAD(&bn->dev_list);
5144
5145         bond_create_proc_dir(bn);
5146         
5147         return 0;
5148 }
5149
5150 static void __net_exit bond_net_exit(struct net *net)
5151 {
5152         struct bond_net *bn = net_generic(net, bond_net_id);
5153
5154         bond_destroy_proc_dir(bn);
5155 }
5156
5157 static struct pernet_operations bond_net_ops = {
5158         .init = bond_net_init,
5159         .exit = bond_net_exit,
5160         .id   = &bond_net_id,
5161         .size = sizeof(struct bond_net),
5162 };
5163
5164 static int __init bonding_init(void)
5165 {
5166         int i;
5167         int res;
5168
5169         pr_info("%s", version);
5170
5171         res = bond_check_params(&bonding_defaults);
5172         if (res)
5173                 goto out;
5174
5175         res = register_pernet_subsys(&bond_net_ops);
5176         if (res)
5177                 goto out;
5178
5179         res = rtnl_link_register(&bond_link_ops);
5180         if (res)
5181                 goto err_link;
5182
5183         for (i = 0; i < max_bonds; i++) {
5184                 res = bond_create(&init_net, NULL);
5185                 if (res)
5186                         goto err;
5187         }
5188
5189         res = bond_create_sysfs();
5190         if (res)
5191                 goto err;
5192
5193         register_netdevice_notifier(&bond_netdev_notifier);
5194         register_inetaddr_notifier(&bond_inetaddr_notifier);
5195         bond_register_ipv6_notifier();
5196 out:
5197         return res;
5198 err:
5199         rtnl_link_unregister(&bond_link_ops);
5200 err_link:
5201         unregister_pernet_subsys(&bond_net_ops);
5202         goto out;
5203
5204 }
5205
5206 static void __exit bonding_exit(void)
5207 {
5208         unregister_netdevice_notifier(&bond_netdev_notifier);
5209         unregister_inetaddr_notifier(&bond_inetaddr_notifier);
5210         bond_unregister_ipv6_notifier();
5211
5212         bond_destroy_sysfs();
5213
5214         rtnl_link_unregister(&bond_link_ops);
5215         unregister_pernet_subsys(&bond_net_ops);
5216 }
5217
5218 module_init(bonding_init);
5219 module_exit(bonding_exit);
5220 MODULE_LICENSE("GPL");
5221 MODULE_VERSION(DRV_VERSION);
5222 MODULE_DESCRIPTION(DRV_DESCRIPTION ", v" DRV_VERSION);
5223 MODULE_AUTHOR("Thomas Davis, tadavis@lbl.gov and many others");
5224 MODULE_ALIAS_RTNL_LINK("bond");