bonding: fix miimon failure counter
[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 BONDING_DEBUG 1
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 <asm/system.h>
57 #include <asm/io.h>
58 #include <asm/dma.h>
59 #include <asm/uaccess.h>
60 #include <linux/errno.h>
61 #include <linux/netdevice.h>
62 #include <linux/inetdevice.h>
63 #include <linux/igmp.h>
64 #include <linux/etherdevice.h>
65 #include <linux/skbuff.h>
66 #include <net/sock.h>
67 #include <linux/rtnetlink.h>
68 #include <linux/proc_fs.h>
69 #include <linux/seq_file.h>
70 #include <linux/smp.h>
71 #include <linux/if_ether.h>
72 #include <net/arp.h>
73 #include <linux/mii.h>
74 #include <linux/ethtool.h>
75 #include <linux/if_vlan.h>
76 #include <linux/if_bonding.h>
77 #include <linux/jiffies.h>
78 #include <net/route.h>
79 #include <net/net_namespace.h>
80 #include "bonding.h"
81 #include "bond_3ad.h"
82 #include "bond_alb.h"
83
84 /*---------------------------- Module parameters ----------------------------*/
85
86 /* monitor all links that often (in milliseconds). <=0 disables monitoring */
87 #define BOND_LINK_MON_INTERV    0
88 #define BOND_LINK_ARP_INTERV    0
89
90 static int max_bonds    = BOND_DEFAULT_MAX_BONDS;
91 static int num_grat_arp = 1;
92 static int miimon       = BOND_LINK_MON_INTERV;
93 static int updelay      = 0;
94 static int downdelay    = 0;
95 static int use_carrier  = 1;
96 static char *mode       = NULL;
97 static char *primary    = NULL;
98 static char *lacp_rate  = NULL;
99 static char *xmit_hash_policy = NULL;
100 static int arp_interval = BOND_LINK_ARP_INTERV;
101 static char *arp_ip_target[BOND_MAX_ARP_TARGETS] = { NULL, };
102 static char *arp_validate = NULL;
103 static char *fail_over_mac = NULL;
104 struct bond_params bonding_defaults;
105
106 module_param(max_bonds, int, 0);
107 MODULE_PARM_DESC(max_bonds, "Max number of bonded devices");
108 module_param(num_grat_arp, int, 0644);
109 MODULE_PARM_DESC(num_grat_arp, "Number of gratuitous ARP packets to send on failover event");
110 module_param(miimon, int, 0);
111 MODULE_PARM_DESC(miimon, "Link check interval in milliseconds");
112 module_param(updelay, int, 0);
113 MODULE_PARM_DESC(updelay, "Delay before considering link up, in milliseconds");
114 module_param(downdelay, int, 0);
115 MODULE_PARM_DESC(downdelay, "Delay before considering link down, "
116                             "in milliseconds");
117 module_param(use_carrier, int, 0);
118 MODULE_PARM_DESC(use_carrier, "Use netif_carrier_ok (vs MII ioctls) in miimon; "
119                               "0 for off, 1 for on (default)");
120 module_param(mode, charp, 0);
121 MODULE_PARM_DESC(mode, "Mode of operation : 0 for balance-rr, "
122                        "1 for active-backup, 2 for balance-xor, "
123                        "3 for broadcast, 4 for 802.3ad, 5 for balance-tlb, "
124                        "6 for balance-alb");
125 module_param(primary, charp, 0);
126 MODULE_PARM_DESC(primary, "Primary network device to use");
127 module_param(lacp_rate, charp, 0);
128 MODULE_PARM_DESC(lacp_rate, "LACPDU tx rate to request from 802.3ad partner "
129                             "(slow/fast)");
130 module_param(xmit_hash_policy, charp, 0);
131 MODULE_PARM_DESC(xmit_hash_policy, "XOR hashing method: 0 for layer 2 (default)"
132                                    ", 1 for layer 3+4");
133 module_param(arp_interval, int, 0);
134 MODULE_PARM_DESC(arp_interval, "arp interval in milliseconds");
135 module_param_array(arp_ip_target, charp, NULL, 0);
136 MODULE_PARM_DESC(arp_ip_target, "arp targets in n.n.n.n form");
137 module_param(arp_validate, charp, 0);
138 MODULE_PARM_DESC(arp_validate, "validate src/dst of ARP probes: none (default), active, backup or all");
139 module_param(fail_over_mac, charp, 0);
140 MODULE_PARM_DESC(fail_over_mac, "For active-backup, do not set all slaves to the same MAC.  none (default), active or follow");
141
142 /*----------------------------- Global variables ----------------------------*/
143
144 static const char * const version =
145         DRV_DESCRIPTION ": v" DRV_VERSION " (" DRV_RELDATE ")\n";
146
147 LIST_HEAD(bond_dev_list);
148
149 #ifdef CONFIG_PROC_FS
150 static struct proc_dir_entry *bond_proc_dir = NULL;
151 #endif
152
153 extern struct rw_semaphore bonding_rwsem;
154 static __be32 arp_target[BOND_MAX_ARP_TARGETS] = { 0, } ;
155 static int arp_ip_count = 0;
156 static int bond_mode    = BOND_MODE_ROUNDROBIN;
157 static int xmit_hashtype= BOND_XMIT_POLICY_LAYER2;
158 static int lacp_fast    = 0;
159
160
161 struct bond_parm_tbl bond_lacp_tbl[] = {
162 {       "slow",         AD_LACP_SLOW},
163 {       "fast",         AD_LACP_FAST},
164 {       NULL,           -1},
165 };
166
167 struct bond_parm_tbl bond_mode_tbl[] = {
168 {       "balance-rr",           BOND_MODE_ROUNDROBIN},
169 {       "active-backup",        BOND_MODE_ACTIVEBACKUP},
170 {       "balance-xor",          BOND_MODE_XOR},
171 {       "broadcast",            BOND_MODE_BROADCAST},
172 {       "802.3ad",              BOND_MODE_8023AD},
173 {       "balance-tlb",          BOND_MODE_TLB},
174 {       "balance-alb",          BOND_MODE_ALB},
175 {       NULL,                   -1},
176 };
177
178 struct bond_parm_tbl xmit_hashtype_tbl[] = {
179 {       "layer2",               BOND_XMIT_POLICY_LAYER2},
180 {       "layer3+4",             BOND_XMIT_POLICY_LAYER34},
181 {       "layer2+3",             BOND_XMIT_POLICY_LAYER23},
182 {       NULL,                   -1},
183 };
184
185 struct bond_parm_tbl arp_validate_tbl[] = {
186 {       "none",                 BOND_ARP_VALIDATE_NONE},
187 {       "active",               BOND_ARP_VALIDATE_ACTIVE},
188 {       "backup",               BOND_ARP_VALIDATE_BACKUP},
189 {       "all",                  BOND_ARP_VALIDATE_ALL},
190 {       NULL,                   -1},
191 };
192
193 struct bond_parm_tbl fail_over_mac_tbl[] = {
194 {       "none",                 BOND_FOM_NONE},
195 {       "active",               BOND_FOM_ACTIVE},
196 {       "follow",               BOND_FOM_FOLLOW},
197 {       NULL,                   -1},
198 };
199
200 /*-------------------------- Forward declarations ---------------------------*/
201
202 static void bond_send_gratuitous_arp(struct bonding *bond);
203 static void bond_deinit(struct net_device *bond_dev);
204
205 /*---------------------------- General routines -----------------------------*/
206
207 static const char *bond_mode_name(int mode)
208 {
209         switch (mode) {
210         case BOND_MODE_ROUNDROBIN :
211                 return "load balancing (round-robin)";
212         case BOND_MODE_ACTIVEBACKUP :
213                 return "fault-tolerance (active-backup)";
214         case BOND_MODE_XOR :
215                 return "load balancing (xor)";
216         case BOND_MODE_BROADCAST :
217                 return "fault-tolerance (broadcast)";
218         case BOND_MODE_8023AD:
219                 return "IEEE 802.3ad Dynamic link aggregation";
220         case BOND_MODE_TLB:
221                 return "transmit load balancing";
222         case BOND_MODE_ALB:
223                 return "adaptive load balancing";
224         default:
225                 return "unknown";
226         }
227 }
228
229 /*---------------------------------- VLAN -----------------------------------*/
230
231 /**
232  * bond_add_vlan - add a new vlan id on bond
233  * @bond: bond that got the notification
234  * @vlan_id: the vlan id to add
235  *
236  * Returns -ENOMEM if allocation failed.
237  */
238 static int bond_add_vlan(struct bonding *bond, unsigned short vlan_id)
239 {
240         struct vlan_entry *vlan;
241
242         dprintk("bond: %s, vlan id %d\n",
243                 (bond ? bond->dev->name: "None"), vlan_id);
244
245         vlan = kmalloc(sizeof(struct vlan_entry), GFP_KERNEL);
246         if (!vlan) {
247                 return -ENOMEM;
248         }
249
250         INIT_LIST_HEAD(&vlan->vlan_list);
251         vlan->vlan_id = vlan_id;
252         vlan->vlan_ip = 0;
253
254         write_lock_bh(&bond->lock);
255
256         list_add_tail(&vlan->vlan_list, &bond->vlan_list);
257
258         write_unlock_bh(&bond->lock);
259
260         dprintk("added VLAN ID %d on bond %s\n", vlan_id, bond->dev->name);
261
262         return 0;
263 }
264
265 /**
266  * bond_del_vlan - delete a vlan id from bond
267  * @bond: bond that got the notification
268  * @vlan_id: the vlan id to delete
269  *
270  * returns -ENODEV if @vlan_id was not found in @bond.
271  */
272 static int bond_del_vlan(struct bonding *bond, unsigned short vlan_id)
273 {
274         struct vlan_entry *vlan;
275         int res = -ENODEV;
276
277         dprintk("bond: %s, vlan id %d\n", bond->dev->name, vlan_id);
278
279         write_lock_bh(&bond->lock);
280
281         list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
282                 if (vlan->vlan_id == vlan_id) {
283                         list_del(&vlan->vlan_list);
284
285                         if ((bond->params.mode == BOND_MODE_TLB) ||
286                             (bond->params.mode == BOND_MODE_ALB)) {
287                                 bond_alb_clear_vlan(bond, vlan_id);
288                         }
289
290                         dprintk("removed VLAN ID %d from bond %s\n", vlan_id,
291                                 bond->dev->name);
292
293                         kfree(vlan);
294
295                         if (list_empty(&bond->vlan_list) &&
296                             (bond->slave_cnt == 0)) {
297                                 /* Last VLAN removed and no slaves, so
298                                  * restore block on adding VLANs. This will
299                                  * be removed once new slaves that are not
300                                  * VLAN challenged will be added.
301                                  */
302                                 bond->dev->features |= NETIF_F_VLAN_CHALLENGED;
303                         }
304
305                         res = 0;
306                         goto out;
307                 }
308         }
309
310         dprintk("couldn't find VLAN ID %d in bond %s\n", vlan_id,
311                 bond->dev->name);
312
313 out:
314         write_unlock_bh(&bond->lock);
315         return res;
316 }
317
318 /**
319  * bond_has_challenged_slaves
320  * @bond: the bond we're working on
321  *
322  * Searches the slave list. Returns 1 if a vlan challenged slave
323  * was found, 0 otherwise.
324  *
325  * Assumes bond->lock is held.
326  */
327 static int bond_has_challenged_slaves(struct bonding *bond)
328 {
329         struct slave *slave;
330         int i;
331
332         bond_for_each_slave(bond, slave, i) {
333                 if (slave->dev->features & NETIF_F_VLAN_CHALLENGED) {
334                         dprintk("found VLAN challenged slave - %s\n",
335                                 slave->dev->name);
336                         return 1;
337                 }
338         }
339
340         dprintk("no VLAN challenged slaves found\n");
341         return 0;
342 }
343
344 /**
345  * bond_next_vlan - safely skip to the next item in the vlans list.
346  * @bond: the bond we're working on
347  * @curr: item we're advancing from
348  *
349  * Returns %NULL if list is empty, bond->next_vlan if @curr is %NULL,
350  * or @curr->next otherwise (even if it is @curr itself again).
351  * 
352  * Caller must hold bond->lock
353  */
354 struct vlan_entry *bond_next_vlan(struct bonding *bond, struct vlan_entry *curr)
355 {
356         struct vlan_entry *next, *last;
357
358         if (list_empty(&bond->vlan_list)) {
359                 return NULL;
360         }
361
362         if (!curr) {
363                 next = list_entry(bond->vlan_list.next,
364                                   struct vlan_entry, vlan_list);
365         } else {
366                 last = list_entry(bond->vlan_list.prev,
367                                   struct vlan_entry, vlan_list);
368                 if (last == curr) {
369                         next = list_entry(bond->vlan_list.next,
370                                           struct vlan_entry, vlan_list);
371                 } else {
372                         next = list_entry(curr->vlan_list.next,
373                                           struct vlan_entry, vlan_list);
374                 }
375         }
376
377         return next;
378 }
379
380 /**
381  * bond_dev_queue_xmit - Prepare skb for xmit.
382  * 
383  * @bond: bond device that got this skb for tx.
384  * @skb: hw accel VLAN tagged skb to transmit
385  * @slave_dev: slave that is supposed to xmit this skbuff
386  * 
387  * When the bond gets an skb to transmit that is
388  * already hardware accelerated VLAN tagged, and it
389  * needs to relay this skb to a slave that is not
390  * hw accel capable, the skb needs to be "unaccelerated",
391  * i.e. strip the hwaccel tag and re-insert it as part
392  * of the payload.
393  */
394 int bond_dev_queue_xmit(struct bonding *bond, struct sk_buff *skb, struct net_device *slave_dev)
395 {
396         unsigned short uninitialized_var(vlan_id);
397
398         if (!list_empty(&bond->vlan_list) &&
399             !(slave_dev->features & NETIF_F_HW_VLAN_TX) &&
400             vlan_get_tag(skb, &vlan_id) == 0) {
401                 skb->dev = slave_dev;
402                 skb = vlan_put_tag(skb, vlan_id);
403                 if (!skb) {
404                         /* vlan_put_tag() frees the skb in case of error,
405                          * so return success here so the calling functions
406                          * won't attempt to free is again.
407                          */
408                         return 0;
409                 }
410         } else {
411                 skb->dev = slave_dev;
412         }
413
414         skb->priority = 1;
415         dev_queue_xmit(skb);
416
417         return 0;
418 }
419
420 /*
421  * In the following 3 functions, bond_vlan_rx_register(), bond_vlan_rx_add_vid
422  * and bond_vlan_rx_kill_vid, We don't protect the slave list iteration with a
423  * lock because:
424  * a. This operation is performed in IOCTL context,
425  * b. The operation is protected by the RTNL semaphore in the 8021q code,
426  * c. Holding a lock with BH disabled while directly calling a base driver
427  *    entry point is generally a BAD idea.
428  * 
429  * The design of synchronization/protection for this operation in the 8021q
430  * module is good for one or more VLAN devices over a single physical device
431  * and cannot be extended for a teaming solution like bonding, so there is a
432  * potential race condition here where a net device from the vlan group might
433  * be referenced (either by a base driver or the 8021q code) while it is being
434  * removed from the system. However, it turns out we're not making matters
435  * worse, and if it works for regular VLAN usage it will work here too.
436 */
437
438 /**
439  * bond_vlan_rx_register - Propagates registration to slaves
440  * @bond_dev: bonding net device that got called
441  * @grp: vlan group being registered
442  */
443 static void bond_vlan_rx_register(struct net_device *bond_dev, struct vlan_group *grp)
444 {
445         struct bonding *bond = bond_dev->priv;
446         struct slave *slave;
447         int i;
448
449         bond->vlgrp = grp;
450
451         bond_for_each_slave(bond, slave, i) {
452                 struct net_device *slave_dev = slave->dev;
453
454                 if ((slave_dev->features & NETIF_F_HW_VLAN_RX) &&
455                     slave_dev->vlan_rx_register) {
456                         slave_dev->vlan_rx_register(slave_dev, grp);
457                 }
458         }
459 }
460
461 /**
462  * bond_vlan_rx_add_vid - Propagates adding an id to slaves
463  * @bond_dev: bonding net device that got called
464  * @vid: vlan id being added
465  */
466 static void bond_vlan_rx_add_vid(struct net_device *bond_dev, uint16_t vid)
467 {
468         struct bonding *bond = bond_dev->priv;
469         struct slave *slave;
470         int i, res;
471
472         bond_for_each_slave(bond, slave, i) {
473                 struct net_device *slave_dev = slave->dev;
474
475                 if ((slave_dev->features & NETIF_F_HW_VLAN_FILTER) &&
476                     slave_dev->vlan_rx_add_vid) {
477                         slave_dev->vlan_rx_add_vid(slave_dev, vid);
478                 }
479         }
480
481         res = bond_add_vlan(bond, vid);
482         if (res) {
483                 printk(KERN_ERR DRV_NAME
484                        ": %s: Error: Failed to add vlan id %d\n",
485                        bond_dev->name, vid);
486         }
487 }
488
489 /**
490  * bond_vlan_rx_kill_vid - Propagates deleting an id to slaves
491  * @bond_dev: bonding net device that got called
492  * @vid: vlan id being removed
493  */
494 static void bond_vlan_rx_kill_vid(struct net_device *bond_dev, uint16_t vid)
495 {
496         struct bonding *bond = bond_dev->priv;
497         struct slave *slave;
498         struct net_device *vlan_dev;
499         int i, res;
500
501         bond_for_each_slave(bond, slave, i) {
502                 struct net_device *slave_dev = slave->dev;
503
504                 if ((slave_dev->features & NETIF_F_HW_VLAN_FILTER) &&
505                     slave_dev->vlan_rx_kill_vid) {
506                         /* Save and then restore vlan_dev in the grp array,
507                          * since the slave's driver might clear it.
508                          */
509                         vlan_dev = vlan_group_get_device(bond->vlgrp, vid);
510                         slave_dev->vlan_rx_kill_vid(slave_dev, vid);
511                         vlan_group_set_device(bond->vlgrp, vid, vlan_dev);
512                 }
513         }
514
515         res = bond_del_vlan(bond, vid);
516         if (res) {
517                 printk(KERN_ERR DRV_NAME
518                        ": %s: Error: Failed to remove vlan id %d\n",
519                        bond_dev->name, vid);
520         }
521 }
522
523 static void bond_add_vlans_on_slave(struct bonding *bond, struct net_device *slave_dev)
524 {
525         struct vlan_entry *vlan;
526
527         write_lock_bh(&bond->lock);
528
529         if (list_empty(&bond->vlan_list)) {
530                 goto out;
531         }
532
533         if ((slave_dev->features & NETIF_F_HW_VLAN_RX) &&
534             slave_dev->vlan_rx_register) {
535                 slave_dev->vlan_rx_register(slave_dev, bond->vlgrp);
536         }
537
538         if (!(slave_dev->features & NETIF_F_HW_VLAN_FILTER) ||
539             !(slave_dev->vlan_rx_add_vid)) {
540                 goto out;
541         }
542
543         list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
544                 slave_dev->vlan_rx_add_vid(slave_dev, vlan->vlan_id);
545         }
546
547 out:
548         write_unlock_bh(&bond->lock);
549 }
550
551 static void bond_del_vlans_from_slave(struct bonding *bond, struct net_device *slave_dev)
552 {
553         struct vlan_entry *vlan;
554         struct net_device *vlan_dev;
555
556         write_lock_bh(&bond->lock);
557
558         if (list_empty(&bond->vlan_list)) {
559                 goto out;
560         }
561
562         if (!(slave_dev->features & NETIF_F_HW_VLAN_FILTER) ||
563             !(slave_dev->vlan_rx_kill_vid)) {
564                 goto unreg;
565         }
566
567         list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
568                 /* Save and then restore vlan_dev in the grp array,
569                  * since the slave's driver might clear it.
570                  */
571                 vlan_dev = vlan_group_get_device(bond->vlgrp, vlan->vlan_id);
572                 slave_dev->vlan_rx_kill_vid(slave_dev, vlan->vlan_id);
573                 vlan_group_set_device(bond->vlgrp, vlan->vlan_id, vlan_dev);
574         }
575
576 unreg:
577         if ((slave_dev->features & NETIF_F_HW_VLAN_RX) &&
578             slave_dev->vlan_rx_register) {
579                 slave_dev->vlan_rx_register(slave_dev, NULL);
580         }
581
582 out:
583         write_unlock_bh(&bond->lock);
584 }
585
586 /*------------------------------- Link status -------------------------------*/
587
588 /*
589  * Set the carrier state for the master according to the state of its
590  * slaves.  If any slaves are up, the master is up.  In 802.3ad mode,
591  * do special 802.3ad magic.
592  *
593  * Returns zero if carrier state does not change, nonzero if it does.
594  */
595 static int bond_set_carrier(struct bonding *bond)
596 {
597         struct slave *slave;
598         int i;
599
600         if (bond->slave_cnt == 0)
601                 goto down;
602
603         if (bond->params.mode == BOND_MODE_8023AD)
604                 return bond_3ad_set_carrier(bond);
605
606         bond_for_each_slave(bond, slave, i) {
607                 if (slave->link == BOND_LINK_UP) {
608                         if (!netif_carrier_ok(bond->dev)) {
609                                 netif_carrier_on(bond->dev);
610                                 return 1;
611                         }
612                         return 0;
613                 }
614         }
615
616 down:
617         if (netif_carrier_ok(bond->dev)) {
618                 netif_carrier_off(bond->dev);
619                 return 1;
620         }
621         return 0;
622 }
623
624 /*
625  * Get link speed and duplex from the slave's base driver
626  * using ethtool. If for some reason the call fails or the
627  * values are invalid, fake speed and duplex to 100/Full
628  * and return error.
629  */
630 static int bond_update_speed_duplex(struct slave *slave)
631 {
632         struct net_device *slave_dev = slave->dev;
633         struct ethtool_cmd etool;
634         int res;
635
636         /* Fake speed and duplex */
637         slave->speed = SPEED_100;
638         slave->duplex = DUPLEX_FULL;
639
640         if (!slave_dev->ethtool_ops || !slave_dev->ethtool_ops->get_settings)
641                 return -1;
642
643         res = slave_dev->ethtool_ops->get_settings(slave_dev, &etool);
644         if (res < 0)
645                 return -1;
646
647         switch (etool.speed) {
648         case SPEED_10:
649         case SPEED_100:
650         case SPEED_1000:
651         case SPEED_10000:
652                 break;
653         default:
654                 return -1;
655         }
656
657         switch (etool.duplex) {
658         case DUPLEX_FULL:
659         case DUPLEX_HALF:
660                 break;
661         default:
662                 return -1;
663         }
664
665         slave->speed = etool.speed;
666         slave->duplex = etool.duplex;
667
668         return 0;
669 }
670
671 /*
672  * if <dev> supports MII link status reporting, check its link status.
673  *
674  * We either do MII/ETHTOOL ioctls, or check netif_carrier_ok(),
675  * depening upon the setting of the use_carrier parameter.
676  *
677  * Return either BMSR_LSTATUS, meaning that the link is up (or we
678  * can't tell and just pretend it is), or 0, meaning that the link is
679  * down.
680  *
681  * If reporting is non-zero, instead of faking link up, return -1 if
682  * both ETHTOOL and MII ioctls fail (meaning the device does not
683  * support them).  If use_carrier is set, return whatever it says.
684  * It'd be nice if there was a good way to tell if a driver supports
685  * netif_carrier, but there really isn't.
686  */
687 static int bond_check_dev_link(struct bonding *bond, struct net_device *slave_dev, int reporting)
688 {
689         static int (* ioctl)(struct net_device *, struct ifreq *, int);
690         struct ifreq ifr;
691         struct mii_ioctl_data *mii;
692
693         if (bond->params.use_carrier) {
694                 return netif_carrier_ok(slave_dev) ? BMSR_LSTATUS : 0;
695         }
696
697         ioctl = slave_dev->do_ioctl;
698         if (ioctl) {
699                 /* TODO: set pointer to correct ioctl on a per team member */
700                 /*       bases to make this more efficient. that is, once  */
701                 /*       we determine the correct ioctl, we will always    */
702                 /*       call it and not the others for that team          */
703                 /*       member.                                           */
704
705                 /*
706                  * We cannot assume that SIOCGMIIPHY will also read a
707                  * register; not all network drivers (e.g., e100)
708                  * support that.
709                  */
710
711                 /* Yes, the mii is overlaid on the ifreq.ifr_ifru */
712                 strncpy(ifr.ifr_name, slave_dev->name, IFNAMSIZ);
713                 mii = if_mii(&ifr);
714                 if (IOCTL(slave_dev, &ifr, SIOCGMIIPHY) == 0) {
715                         mii->reg_num = MII_BMSR;
716                         if (IOCTL(slave_dev, &ifr, SIOCGMIIREG) == 0) {
717                                 return (mii->val_out & BMSR_LSTATUS);
718                         }
719                 }
720         }
721
722         /*
723          * Some drivers cache ETHTOOL_GLINK for a period of time so we only
724          * attempt to get link status from it if the above MII ioctls fail.
725          */
726         if (slave_dev->ethtool_ops) {
727                 if (slave_dev->ethtool_ops->get_link) {
728                         u32 link;
729
730                         link = slave_dev->ethtool_ops->get_link(slave_dev);
731
732                         return link ? BMSR_LSTATUS : 0;
733                 }
734         }
735
736         /*
737          * If reporting, report that either there's no dev->do_ioctl,
738          * or both SIOCGMIIREG and get_link failed (meaning that we
739          * cannot report link status).  If not reporting, pretend
740          * we're ok.
741          */
742         return (reporting ? -1 : BMSR_LSTATUS);
743 }
744
745 /*----------------------------- Multicast list ------------------------------*/
746
747 /*
748  * Returns 0 if dmi1 and dmi2 are the same, non-0 otherwise
749  */
750 static inline int bond_is_dmi_same(struct dev_mc_list *dmi1, struct dev_mc_list *dmi2)
751 {
752         return memcmp(dmi1->dmi_addr, dmi2->dmi_addr, dmi1->dmi_addrlen) == 0 &&
753                         dmi1->dmi_addrlen == dmi2->dmi_addrlen;
754 }
755
756 /*
757  * returns dmi entry if found, NULL otherwise
758  */
759 static struct dev_mc_list *bond_mc_list_find_dmi(struct dev_mc_list *dmi, struct dev_mc_list *mc_list)
760 {
761         struct dev_mc_list *idmi;
762
763         for (idmi = mc_list; idmi; idmi = idmi->next) {
764                 if (bond_is_dmi_same(dmi, idmi)) {
765                         return idmi;
766                 }
767         }
768
769         return NULL;
770 }
771
772 /*
773  * Push the promiscuity flag down to appropriate slaves
774  */
775 static int bond_set_promiscuity(struct bonding *bond, int inc)
776 {
777         int err = 0;
778         if (USES_PRIMARY(bond->params.mode)) {
779                 /* write lock already acquired */
780                 if (bond->curr_active_slave) {
781                         err = dev_set_promiscuity(bond->curr_active_slave->dev,
782                                                   inc);
783                 }
784         } else {
785                 struct slave *slave;
786                 int i;
787                 bond_for_each_slave(bond, slave, i) {
788                         err = dev_set_promiscuity(slave->dev, inc);
789                         if (err)
790                                 return err;
791                 }
792         }
793         return err;
794 }
795
796 /*
797  * Push the allmulti flag down to all slaves
798  */
799 static int bond_set_allmulti(struct bonding *bond, int inc)
800 {
801         int err = 0;
802         if (USES_PRIMARY(bond->params.mode)) {
803                 /* write lock already acquired */
804                 if (bond->curr_active_slave) {
805                         err = dev_set_allmulti(bond->curr_active_slave->dev,
806                                                inc);
807                 }
808         } else {
809                 struct slave *slave;
810                 int i;
811                 bond_for_each_slave(bond, slave, i) {
812                         err = dev_set_allmulti(slave->dev, inc);
813                         if (err)
814                                 return err;
815                 }
816         }
817         return err;
818 }
819
820 /*
821  * Add a Multicast address to slaves
822  * according to mode
823  */
824 static void bond_mc_add(struct bonding *bond, void *addr, int alen)
825 {
826         if (USES_PRIMARY(bond->params.mode)) {
827                 /* write lock already acquired */
828                 if (bond->curr_active_slave) {
829                         dev_mc_add(bond->curr_active_slave->dev, addr, alen, 0);
830                 }
831         } else {
832                 struct slave *slave;
833                 int i;
834                 bond_for_each_slave(bond, slave, i) {
835                         dev_mc_add(slave->dev, addr, alen, 0);
836                 }
837         }
838 }
839
840 /*
841  * Remove a multicast address from slave
842  * according to mode
843  */
844 static void bond_mc_delete(struct bonding *bond, void *addr, int alen)
845 {
846         if (USES_PRIMARY(bond->params.mode)) {
847                 /* write lock already acquired */
848                 if (bond->curr_active_slave) {
849                         dev_mc_delete(bond->curr_active_slave->dev, addr, alen, 0);
850                 }
851         } else {
852                 struct slave *slave;
853                 int i;
854                 bond_for_each_slave(bond, slave, i) {
855                         dev_mc_delete(slave->dev, addr, alen, 0);
856                 }
857         }
858 }
859
860
861 /*
862  * Retrieve the list of registered multicast addresses for the bonding
863  * device and retransmit an IGMP JOIN request to the current active
864  * slave.
865  */
866 static void bond_resend_igmp_join_requests(struct bonding *bond)
867 {
868         struct in_device *in_dev;
869         struct ip_mc_list *im;
870
871         rcu_read_lock();
872         in_dev = __in_dev_get_rcu(bond->dev);
873         if (in_dev) {
874                 for (im = in_dev->mc_list; im; im = im->next) {
875                         ip_mc_rejoin_group(im);
876                 }
877         }
878
879         rcu_read_unlock();
880 }
881
882 /*
883  * Totally destroys the mc_list in bond
884  */
885 static void bond_mc_list_destroy(struct bonding *bond)
886 {
887         struct dev_mc_list *dmi;
888
889         dmi = bond->mc_list;
890         while (dmi) {
891                 bond->mc_list = dmi->next;
892                 kfree(dmi);
893                 dmi = bond->mc_list;
894         }
895         bond->mc_list = NULL;
896 }
897
898 /*
899  * Copy all the Multicast addresses from src to the bonding device dst
900  */
901 static int bond_mc_list_copy(struct dev_mc_list *mc_list, struct bonding *bond,
902                              gfp_t gfp_flag)
903 {
904         struct dev_mc_list *dmi, *new_dmi;
905
906         for (dmi = mc_list; dmi; dmi = dmi->next) {
907                 new_dmi = kmalloc(sizeof(struct dev_mc_list), gfp_flag);
908
909                 if (!new_dmi) {
910                         /* FIXME: Potential memory leak !!! */
911                         return -ENOMEM;
912                 }
913
914                 new_dmi->next = bond->mc_list;
915                 bond->mc_list = new_dmi;
916                 new_dmi->dmi_addrlen = dmi->dmi_addrlen;
917                 memcpy(new_dmi->dmi_addr, dmi->dmi_addr, dmi->dmi_addrlen);
918                 new_dmi->dmi_users = dmi->dmi_users;
919                 new_dmi->dmi_gusers = dmi->dmi_gusers;
920         }
921
922         return 0;
923 }
924
925 /*
926  * flush all members of flush->mc_list from device dev->mc_list
927  */
928 static void bond_mc_list_flush(struct net_device *bond_dev, struct net_device *slave_dev)
929 {
930         struct bonding *bond = bond_dev->priv;
931         struct dev_mc_list *dmi;
932
933         for (dmi = bond_dev->mc_list; dmi; dmi = dmi->next) {
934                 dev_mc_delete(slave_dev, dmi->dmi_addr, dmi->dmi_addrlen, 0);
935         }
936
937         if (bond->params.mode == BOND_MODE_8023AD) {
938                 /* del lacpdu mc addr from mc list */
939                 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
940
941                 dev_mc_delete(slave_dev, lacpdu_multicast, ETH_ALEN, 0);
942         }
943 }
944
945 /*--------------------------- Active slave change ---------------------------*/
946
947 /*
948  * Update the mc list and multicast-related flags for the new and
949  * old active slaves (if any) according to the multicast mode, and
950  * promiscuous flags unconditionally.
951  */
952 static void bond_mc_swap(struct bonding *bond, struct slave *new_active, struct slave *old_active)
953 {
954         struct dev_mc_list *dmi;
955
956         if (!USES_PRIMARY(bond->params.mode)) {
957                 /* nothing to do -  mc list is already up-to-date on
958                  * all slaves
959                  */
960                 return;
961         }
962
963         if (old_active) {
964                 if (bond->dev->flags & IFF_PROMISC) {
965                         dev_set_promiscuity(old_active->dev, -1);
966                 }
967
968                 if (bond->dev->flags & IFF_ALLMULTI) {
969                         dev_set_allmulti(old_active->dev, -1);
970                 }
971
972                 for (dmi = bond->dev->mc_list; dmi; dmi = dmi->next) {
973                         dev_mc_delete(old_active->dev, dmi->dmi_addr, dmi->dmi_addrlen, 0);
974                 }
975         }
976
977         if (new_active) {
978                 /* FIXME: Signal errors upstream. */
979                 if (bond->dev->flags & IFF_PROMISC) {
980                         dev_set_promiscuity(new_active->dev, 1);
981                 }
982
983                 if (bond->dev->flags & IFF_ALLMULTI) {
984                         dev_set_allmulti(new_active->dev, 1);
985                 }
986
987                 for (dmi = bond->dev->mc_list; dmi; dmi = dmi->next) {
988                         dev_mc_add(new_active->dev, dmi->dmi_addr, dmi->dmi_addrlen, 0);
989                 }
990                 bond_resend_igmp_join_requests(bond);
991         }
992 }
993
994 /*
995  * bond_do_fail_over_mac
996  *
997  * Perform special MAC address swapping for fail_over_mac settings
998  *
999  * Called with RTNL, bond->lock for read, curr_slave_lock for write_bh.
1000  */
1001 static void bond_do_fail_over_mac(struct bonding *bond,
1002                                   struct slave *new_active,
1003                                   struct slave *old_active)
1004 {
1005         u8 tmp_mac[ETH_ALEN];
1006         struct sockaddr saddr;
1007         int rv;
1008
1009         switch (bond->params.fail_over_mac) {
1010         case BOND_FOM_ACTIVE:
1011                 if (new_active)
1012                         memcpy(bond->dev->dev_addr,  new_active->dev->dev_addr,
1013                                new_active->dev->addr_len);
1014                 break;
1015         case BOND_FOM_FOLLOW:
1016                 /*
1017                  * if new_active && old_active, swap them
1018                  * if just old_active, do nothing (going to no active slave)
1019                  * if just new_active, set new_active to bond's MAC
1020                  */
1021                 if (!new_active)
1022                         return;
1023
1024                 write_unlock_bh(&bond->curr_slave_lock);
1025                 read_unlock(&bond->lock);
1026
1027                 if (old_active) {
1028                         memcpy(tmp_mac, new_active->dev->dev_addr, ETH_ALEN);
1029                         memcpy(saddr.sa_data, old_active->dev->dev_addr,
1030                                ETH_ALEN);
1031                         saddr.sa_family = new_active->dev->type;
1032                 } else {
1033                         memcpy(saddr.sa_data, bond->dev->dev_addr, ETH_ALEN);
1034                         saddr.sa_family = bond->dev->type;
1035                 }
1036
1037                 rv = dev_set_mac_address(new_active->dev, &saddr);
1038                 if (rv) {
1039                         printk(KERN_ERR DRV_NAME
1040                                ": %s: Error %d setting MAC of slave %s\n",
1041                                bond->dev->name, -rv, new_active->dev->name);
1042                         goto out;
1043                 }
1044
1045                 if (!old_active)
1046                         goto out;
1047
1048                 memcpy(saddr.sa_data, tmp_mac, ETH_ALEN);
1049                 saddr.sa_family = old_active->dev->type;
1050
1051                 rv = dev_set_mac_address(old_active->dev, &saddr);
1052                 if (rv)
1053                         printk(KERN_ERR DRV_NAME
1054                                ": %s: Error %d setting MAC of slave %s\n",
1055                                bond->dev->name, -rv, new_active->dev->name);
1056 out:
1057                 read_lock(&bond->lock);
1058                 write_lock_bh(&bond->curr_slave_lock);
1059                 break;
1060         default:
1061                 printk(KERN_ERR DRV_NAME
1062                        ": %s: bond_do_fail_over_mac impossible: bad policy %d\n",
1063                        bond->dev->name, bond->params.fail_over_mac);
1064                 break;
1065         }
1066
1067 }
1068
1069
1070 /**
1071  * find_best_interface - select the best available slave to be the active one
1072  * @bond: our bonding struct
1073  *
1074  * Warning: Caller must hold curr_slave_lock for writing.
1075  */
1076 static struct slave *bond_find_best_slave(struct bonding *bond)
1077 {
1078         struct slave *new_active, *old_active;
1079         struct slave *bestslave = NULL;
1080         int mintime = bond->params.updelay;
1081         int i;
1082
1083         new_active = old_active = bond->curr_active_slave;
1084
1085         if (!new_active) { /* there were no active slaves left */
1086                 if (bond->slave_cnt > 0) {  /* found one slave */
1087                         new_active = bond->first_slave;
1088                 } else {
1089                         return NULL; /* still no slave, return NULL */
1090                 }
1091         }
1092
1093         /* first try the primary link; if arping, a link must tx/rx traffic
1094          * before it can be considered the curr_active_slave - also, we would skip
1095          * slaves between the curr_active_slave and primary_slave that may be up
1096          * and able to arp
1097          */
1098         if ((bond->primary_slave) &&
1099             (!bond->params.arp_interval) &&
1100             (IS_UP(bond->primary_slave->dev))) {
1101                 new_active = bond->primary_slave;
1102         }
1103
1104         /* remember where to stop iterating over the slaves */
1105         old_active = new_active;
1106
1107         bond_for_each_slave_from(bond, new_active, i, old_active) {
1108                 if (IS_UP(new_active->dev)) {
1109                         if (new_active->link == BOND_LINK_UP) {
1110                                 return new_active;
1111                         } else if (new_active->link == BOND_LINK_BACK) {
1112                                 /* link up, but waiting for stabilization */
1113                                 if (new_active->delay < mintime) {
1114                                         mintime = new_active->delay;
1115                                         bestslave = new_active;
1116                                 }
1117                         }
1118                 }
1119         }
1120
1121         return bestslave;
1122 }
1123
1124 /**
1125  * change_active_interface - change the active slave into the specified one
1126  * @bond: our bonding struct
1127  * @new: the new slave to make the active one
1128  *
1129  * Set the new slave to the bond's settings and unset them on the old
1130  * curr_active_slave.
1131  * Setting include flags, mc-list, promiscuity, allmulti, etc.
1132  *
1133  * If @new's link state is %BOND_LINK_BACK we'll set it to %BOND_LINK_UP,
1134  * because it is apparently the best available slave we have, even though its
1135  * updelay hasn't timed out yet.
1136  *
1137  * If new_active is not NULL, caller must hold bond->lock for read and
1138  * curr_slave_lock for write_bh.
1139  */
1140 void bond_change_active_slave(struct bonding *bond, struct slave *new_active)
1141 {
1142         struct slave *old_active = bond->curr_active_slave;
1143
1144         if (old_active == new_active) {
1145                 return;
1146         }
1147
1148         if (new_active) {
1149                 new_active->jiffies = jiffies;
1150
1151                 if (new_active->link == BOND_LINK_BACK) {
1152                         if (USES_PRIMARY(bond->params.mode)) {
1153                                 printk(KERN_INFO DRV_NAME
1154                                        ": %s: making interface %s the new "
1155                                        "active one %d ms earlier.\n",
1156                                        bond->dev->name, new_active->dev->name,
1157                                        (bond->params.updelay - new_active->delay) * bond->params.miimon);
1158                         }
1159
1160                         new_active->delay = 0;
1161                         new_active->link = BOND_LINK_UP;
1162
1163                         if (bond->params.mode == BOND_MODE_8023AD) {
1164                                 bond_3ad_handle_link_change(new_active, BOND_LINK_UP);
1165                         }
1166
1167                         if ((bond->params.mode == BOND_MODE_TLB) ||
1168                             (bond->params.mode == BOND_MODE_ALB)) {
1169                                 bond_alb_handle_link_change(bond, new_active, BOND_LINK_UP);
1170                         }
1171                 } else {
1172                         if (USES_PRIMARY(bond->params.mode)) {
1173                                 printk(KERN_INFO DRV_NAME
1174                                        ": %s: making interface %s the new "
1175                                        "active one.\n",
1176                                        bond->dev->name, new_active->dev->name);
1177                         }
1178                 }
1179         }
1180
1181         if (USES_PRIMARY(bond->params.mode)) {
1182                 bond_mc_swap(bond, new_active, old_active);
1183         }
1184
1185         if ((bond->params.mode == BOND_MODE_TLB) ||
1186             (bond->params.mode == BOND_MODE_ALB)) {
1187                 bond_alb_handle_active_change(bond, new_active);
1188                 if (old_active)
1189                         bond_set_slave_inactive_flags(old_active);
1190                 if (new_active)
1191                         bond_set_slave_active_flags(new_active);
1192         } else {
1193                 bond->curr_active_slave = new_active;
1194         }
1195
1196         if (bond->params.mode == BOND_MODE_ACTIVEBACKUP) {
1197                 if (old_active) {
1198                         bond_set_slave_inactive_flags(old_active);
1199                 }
1200
1201                 if (new_active) {
1202                         bond_set_slave_active_flags(new_active);
1203
1204                         if (bond->params.fail_over_mac)
1205                                 bond_do_fail_over_mac(bond, new_active,
1206                                                       old_active);
1207
1208                         bond->send_grat_arp = bond->params.num_grat_arp;
1209                         bond_send_gratuitous_arp(bond);
1210
1211                         write_unlock_bh(&bond->curr_slave_lock);
1212                         read_unlock(&bond->lock);
1213
1214                         netdev_bonding_change(bond->dev);
1215
1216                         read_lock(&bond->lock);
1217                         write_lock_bh(&bond->curr_slave_lock);
1218                 }
1219         }
1220 }
1221
1222 /**
1223  * bond_select_active_slave - select a new active slave, if needed
1224  * @bond: our bonding struct
1225  *
1226  * This functions shoud be called when one of the following occurs:
1227  * - The old curr_active_slave has been released or lost its link.
1228  * - The primary_slave has got its link back.
1229  * - A slave has got its link back and there's no old curr_active_slave.
1230  *
1231  * Caller must hold bond->lock for read and curr_slave_lock for write_bh.
1232  */
1233 void bond_select_active_slave(struct bonding *bond)
1234 {
1235         struct slave *best_slave;
1236         int rv;
1237
1238         best_slave = bond_find_best_slave(bond);
1239         if (best_slave != bond->curr_active_slave) {
1240                 bond_change_active_slave(bond, best_slave);
1241                 rv = bond_set_carrier(bond);
1242                 if (!rv)
1243                         return;
1244
1245                 if (netif_carrier_ok(bond->dev)) {
1246                         printk(KERN_INFO DRV_NAME
1247                                ": %s: first active interface up!\n",
1248                                bond->dev->name);
1249                 } else {
1250                         printk(KERN_INFO DRV_NAME ": %s: "
1251                                "now running without any active interface !\n",
1252                                bond->dev->name);
1253                 }
1254         }
1255 }
1256
1257 /*--------------------------- slave list handling ---------------------------*/
1258
1259 /*
1260  * This function attaches the slave to the end of list.
1261  *
1262  * bond->lock held for writing by caller.
1263  */
1264 static void bond_attach_slave(struct bonding *bond, struct slave *new_slave)
1265 {
1266         if (bond->first_slave == NULL) { /* attaching the first slave */
1267                 new_slave->next = new_slave;
1268                 new_slave->prev = new_slave;
1269                 bond->first_slave = new_slave;
1270         } else {
1271                 new_slave->next = bond->first_slave;
1272                 new_slave->prev = bond->first_slave->prev;
1273                 new_slave->next->prev = new_slave;
1274                 new_slave->prev->next = new_slave;
1275         }
1276
1277         bond->slave_cnt++;
1278 }
1279
1280 /*
1281  * This function detaches the slave from the list.
1282  * WARNING: no check is made to verify if the slave effectively
1283  * belongs to <bond>.
1284  * Nothing is freed on return, structures are just unchained.
1285  * If any slave pointer in bond was pointing to <slave>,
1286  * it should be changed by the calling function.
1287  *
1288  * bond->lock held for writing by caller.
1289  */
1290 static void bond_detach_slave(struct bonding *bond, struct slave *slave)
1291 {
1292         if (slave->next) {
1293                 slave->next->prev = slave->prev;
1294         }
1295
1296         if (slave->prev) {
1297                 slave->prev->next = slave->next;
1298         }
1299
1300         if (bond->first_slave == slave) { /* slave is the first slave */
1301                 if (bond->slave_cnt > 1) { /* there are more slave */
1302                         bond->first_slave = slave->next;
1303                 } else {
1304                         bond->first_slave = NULL; /* slave was the last one */
1305                 }
1306         }
1307
1308         slave->next = NULL;
1309         slave->prev = NULL;
1310         bond->slave_cnt--;
1311 }
1312
1313 /*---------------------------------- IOCTL ----------------------------------*/
1314
1315 static int bond_sethwaddr(struct net_device *bond_dev,
1316                           struct net_device *slave_dev)
1317 {
1318         dprintk("bond_dev=%p\n", bond_dev);
1319         dprintk("slave_dev=%p\n", slave_dev);
1320         dprintk("slave_dev->addr_len=%d\n", slave_dev->addr_len);
1321         memcpy(bond_dev->dev_addr, slave_dev->dev_addr, slave_dev->addr_len);
1322         return 0;
1323 }
1324
1325 #define BOND_VLAN_FEATURES \
1326         (NETIF_F_VLAN_CHALLENGED | NETIF_F_HW_VLAN_RX | NETIF_F_HW_VLAN_TX | \
1327          NETIF_F_HW_VLAN_FILTER)
1328
1329 /* 
1330  * Compute the common dev->feature set available to all slaves.  Some
1331  * feature bits are managed elsewhere, so preserve those feature bits
1332  * on the master device.
1333  */
1334 static int bond_compute_features(struct bonding *bond)
1335 {
1336         struct slave *slave;
1337         struct net_device *bond_dev = bond->dev;
1338         unsigned long features = bond_dev->features;
1339         unsigned short max_hard_header_len = max((u16)ETH_HLEN,
1340                                                 bond_dev->hard_header_len);
1341         int i;
1342
1343         features &= ~(NETIF_F_ALL_CSUM | BOND_VLAN_FEATURES);
1344         features |=  NETIF_F_GSO_MASK | NETIF_F_NO_CSUM;
1345
1346         if (!bond->first_slave)
1347                 goto done;
1348
1349         features &= ~NETIF_F_ONE_FOR_ALL;
1350
1351         bond_for_each_slave(bond, slave, i) {
1352                 features = netdev_increment_features(features,
1353                                                      slave->dev->features,
1354                                                      NETIF_F_ONE_FOR_ALL);
1355                 if (slave->dev->hard_header_len > max_hard_header_len)
1356                         max_hard_header_len = slave->dev->hard_header_len;
1357         }
1358
1359 done:
1360         features |= (bond_dev->features & BOND_VLAN_FEATURES);
1361         bond_dev->features = netdev_fix_features(features, NULL);
1362         bond_dev->hard_header_len = max_hard_header_len;
1363
1364         return 0;
1365 }
1366
1367
1368 static void bond_setup_by_slave(struct net_device *bond_dev,
1369                                 struct net_device *slave_dev)
1370 {
1371         struct bonding *bond = bond_dev->priv;
1372
1373         bond_dev->neigh_setup           = slave_dev->neigh_setup;
1374         bond_dev->header_ops            = slave_dev->header_ops;
1375
1376         bond_dev->type              = slave_dev->type;
1377         bond_dev->hard_header_len   = slave_dev->hard_header_len;
1378         bond_dev->addr_len          = slave_dev->addr_len;
1379
1380         memcpy(bond_dev->broadcast, slave_dev->broadcast,
1381                 slave_dev->addr_len);
1382         bond->setup_by_slave = 1;
1383 }
1384
1385 /* enslave device <slave> to bond device <master> */
1386 int bond_enslave(struct net_device *bond_dev, struct net_device *slave_dev)
1387 {
1388         struct bonding *bond = bond_dev->priv;
1389         struct slave *new_slave = NULL;
1390         struct dev_mc_list *dmi;
1391         struct sockaddr addr;
1392         int link_reporting;
1393         int old_features = bond_dev->features;
1394         int res = 0;
1395
1396         if (!bond->params.use_carrier && slave_dev->ethtool_ops == NULL &&
1397                 slave_dev->do_ioctl == NULL) {
1398                 printk(KERN_WARNING DRV_NAME
1399                        ": %s: Warning: no link monitoring support for %s\n",
1400                        bond_dev->name, slave_dev->name);
1401         }
1402
1403         /* bond must be initialized by bond_open() before enslaving */
1404         if (!(bond_dev->flags & IFF_UP)) {
1405                 printk(KERN_WARNING DRV_NAME
1406                         " %s: master_dev is not up in bond_enslave\n",
1407                         bond_dev->name);
1408         }
1409
1410         /* already enslaved */
1411         if (slave_dev->flags & IFF_SLAVE) {
1412                 dprintk("Error, Device was already enslaved\n");
1413                 return -EBUSY;
1414         }
1415
1416         /* vlan challenged mutual exclusion */
1417         /* no need to lock since we're protected by rtnl_lock */
1418         if (slave_dev->features & NETIF_F_VLAN_CHALLENGED) {
1419                 dprintk("%s: NETIF_F_VLAN_CHALLENGED\n", slave_dev->name);
1420                 if (!list_empty(&bond->vlan_list)) {
1421                         printk(KERN_ERR DRV_NAME
1422                                ": %s: Error: cannot enslave VLAN "
1423                                "challenged slave %s on VLAN enabled "
1424                                "bond %s\n", bond_dev->name, slave_dev->name,
1425                                bond_dev->name);
1426                         return -EPERM;
1427                 } else {
1428                         printk(KERN_WARNING DRV_NAME
1429                                ": %s: Warning: enslaved VLAN challenged "
1430                                "slave %s. Adding VLANs will be blocked as "
1431                                "long as %s is part of bond %s\n",
1432                                bond_dev->name, slave_dev->name, slave_dev->name,
1433                                bond_dev->name);
1434                         bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
1435                 }
1436         } else {
1437                 dprintk("%s: ! NETIF_F_VLAN_CHALLENGED\n", slave_dev->name);
1438                 if (bond->slave_cnt == 0) {
1439                         /* First slave, and it is not VLAN challenged,
1440                          * so remove the block of adding VLANs over the bond.
1441                          */
1442                         bond_dev->features &= ~NETIF_F_VLAN_CHALLENGED;
1443                 }
1444         }
1445
1446         /*
1447          * Old ifenslave binaries are no longer supported.  These can
1448          * be identified with moderate accurary by the state of the slave:
1449          * the current ifenslave will set the interface down prior to
1450          * enslaving it; the old ifenslave will not.
1451          */
1452         if ((slave_dev->flags & IFF_UP)) {
1453                 printk(KERN_ERR DRV_NAME ": %s is up. "
1454                        "This may be due to an out of date ifenslave.\n",
1455                        slave_dev->name);
1456                 res = -EPERM;
1457                 goto err_undo_flags;
1458         }
1459
1460         /* set bonding device ether type by slave - bonding netdevices are
1461          * created with ether_setup, so when the slave type is not ARPHRD_ETHER
1462          * there is a need to override some of the type dependent attribs/funcs.
1463          *
1464          * bond ether type mutual exclusion - don't allow slaves of dissimilar
1465          * ether type (eg ARPHRD_ETHER and ARPHRD_INFINIBAND) share the same bond
1466          */
1467         if (bond->slave_cnt == 0) {
1468                 if (slave_dev->type != ARPHRD_ETHER)
1469                         bond_setup_by_slave(bond_dev, slave_dev);
1470         } else if (bond_dev->type != slave_dev->type) {
1471                 printk(KERN_ERR DRV_NAME ": %s ether type (%d) is different "
1472                         "from other slaves (%d), can not enslave it.\n",
1473                         slave_dev->name,
1474                         slave_dev->type, bond_dev->type);
1475                         res = -EINVAL;
1476                         goto err_undo_flags;
1477         }
1478
1479         if (slave_dev->set_mac_address == NULL) {
1480                 if (bond->slave_cnt == 0) {
1481                         printk(KERN_WARNING DRV_NAME
1482                                ": %s: Warning: The first slave device "
1483                                "specified does not support setting the MAC "
1484                                "address. Setting fail_over_mac to active.",
1485                                bond_dev->name);
1486                         bond->params.fail_over_mac = BOND_FOM_ACTIVE;
1487                 } else if (bond->params.fail_over_mac != BOND_FOM_ACTIVE) {
1488                         printk(KERN_ERR DRV_NAME
1489                                 ": %s: Error: The slave device specified "
1490                                 "does not support setting the MAC address, "
1491                                 "but fail_over_mac is not set to active.\n"
1492                                 , bond_dev->name);
1493                         res = -EOPNOTSUPP;
1494                         goto err_undo_flags;
1495                 }
1496         }
1497
1498         new_slave = kzalloc(sizeof(struct slave), GFP_KERNEL);
1499         if (!new_slave) {
1500                 res = -ENOMEM;
1501                 goto err_undo_flags;
1502         }
1503
1504         /* save slave's original flags before calling
1505          * netdev_set_master and dev_open
1506          */
1507         new_slave->original_flags = slave_dev->flags;
1508
1509         /*
1510          * Save slave's original ("permanent") mac address for modes
1511          * that need it, and for restoring it upon release, and then
1512          * set it to the master's address
1513          */
1514         memcpy(new_slave->perm_hwaddr, slave_dev->dev_addr, ETH_ALEN);
1515
1516         if (!bond->params.fail_over_mac) {
1517                 /*
1518                  * Set slave to master's mac address.  The application already
1519                  * set the master's mac address to that of the first slave
1520                  */
1521                 memcpy(addr.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
1522                 addr.sa_family = slave_dev->type;
1523                 res = dev_set_mac_address(slave_dev, &addr);
1524                 if (res) {
1525                         dprintk("Error %d calling set_mac_address\n", res);
1526                         goto err_free;
1527                 }
1528         }
1529
1530         res = netdev_set_master(slave_dev, bond_dev);
1531         if (res) {
1532                 dprintk("Error %d calling netdev_set_master\n", res);
1533                 goto err_restore_mac;
1534         }
1535         /* open the slave since the application closed it */
1536         res = dev_open(slave_dev);
1537         if (res) {
1538                 dprintk("Openning slave %s failed\n", slave_dev->name);
1539                 goto err_unset_master;
1540         }
1541
1542         new_slave->dev = slave_dev;
1543         slave_dev->priv_flags |= IFF_BONDING;
1544
1545         if ((bond->params.mode == BOND_MODE_TLB) ||
1546             (bond->params.mode == BOND_MODE_ALB)) {
1547                 /* bond_alb_init_slave() must be called before all other stages since
1548                  * it might fail and we do not want to have to undo everything
1549                  */
1550                 res = bond_alb_init_slave(bond, new_slave);
1551                 if (res) {
1552                         goto err_close;
1553                 }
1554         }
1555
1556         /* If the mode USES_PRIMARY, then the new slave gets the
1557          * master's promisc (and mc) settings only if it becomes the
1558          * curr_active_slave, and that is taken care of later when calling
1559          * bond_change_active()
1560          */
1561         if (!USES_PRIMARY(bond->params.mode)) {
1562                 /* set promiscuity level to new slave */
1563                 if (bond_dev->flags & IFF_PROMISC) {
1564                         res = dev_set_promiscuity(slave_dev, 1);
1565                         if (res)
1566                                 goto err_close;
1567                 }
1568
1569                 /* set allmulti level to new slave */
1570                 if (bond_dev->flags & IFF_ALLMULTI) {
1571                         res = dev_set_allmulti(slave_dev, 1);
1572                         if (res)
1573                                 goto err_close;
1574                 }
1575
1576                 netif_addr_lock_bh(bond_dev);
1577                 /* upload master's mc_list to new slave */
1578                 for (dmi = bond_dev->mc_list; dmi; dmi = dmi->next) {
1579                         dev_mc_add (slave_dev, dmi->dmi_addr, dmi->dmi_addrlen, 0);
1580                 }
1581                 netif_addr_unlock_bh(bond_dev);
1582         }
1583
1584         if (bond->params.mode == BOND_MODE_8023AD) {
1585                 /* add lacpdu mc addr to mc list */
1586                 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
1587
1588                 dev_mc_add(slave_dev, lacpdu_multicast, ETH_ALEN, 0);
1589         }
1590
1591         bond_add_vlans_on_slave(bond, slave_dev);
1592
1593         write_lock_bh(&bond->lock);
1594
1595         bond_attach_slave(bond, new_slave);
1596
1597         new_slave->delay = 0;
1598         new_slave->link_failure_count = 0;
1599
1600         bond_compute_features(bond);
1601
1602         write_unlock_bh(&bond->lock);
1603
1604         read_lock(&bond->lock);
1605
1606         new_slave->last_arp_rx = jiffies;
1607
1608         if (bond->params.miimon && !bond->params.use_carrier) {
1609                 link_reporting = bond_check_dev_link(bond, slave_dev, 1);
1610
1611                 if ((link_reporting == -1) && !bond->params.arp_interval) {
1612                         /*
1613                          * miimon is set but a bonded network driver
1614                          * does not support ETHTOOL/MII and
1615                          * arp_interval is not set.  Note: if
1616                          * use_carrier is enabled, we will never go
1617                          * here (because netif_carrier is always
1618                          * supported); thus, we don't need to change
1619                          * the messages for netif_carrier.
1620                          */
1621                         printk(KERN_WARNING DRV_NAME
1622                                ": %s: Warning: MII and ETHTOOL support not "
1623                                "available for interface %s, and "
1624                                "arp_interval/arp_ip_target module parameters "
1625                                "not specified, thus bonding will not detect "
1626                                "link failures! see bonding.txt for details.\n",
1627                                bond_dev->name, slave_dev->name);
1628                 } else if (link_reporting == -1) {
1629                         /* unable get link status using mii/ethtool */
1630                         printk(KERN_WARNING DRV_NAME
1631                                ": %s: Warning: can't get link status from "
1632                                "interface %s; the network driver associated "
1633                                "with this interface does not support MII or "
1634                                "ETHTOOL link status reporting, thus miimon "
1635                                "has no effect on this interface.\n",
1636                                bond_dev->name, slave_dev->name);
1637                 }
1638         }
1639
1640         /* check for initial state */
1641         if (!bond->params.miimon ||
1642             (bond_check_dev_link(bond, slave_dev, 0) == BMSR_LSTATUS)) {
1643                 if (bond->params.updelay) {
1644                         dprintk("Initial state of slave_dev is "
1645                                 "BOND_LINK_BACK\n");
1646                         new_slave->link  = BOND_LINK_BACK;
1647                         new_slave->delay = bond->params.updelay;
1648                 } else {
1649                         dprintk("Initial state of slave_dev is "
1650                                 "BOND_LINK_UP\n");
1651                         new_slave->link  = BOND_LINK_UP;
1652                 }
1653                 new_slave->jiffies = jiffies;
1654         } else {
1655                 dprintk("Initial state of slave_dev is "
1656                         "BOND_LINK_DOWN\n");
1657                 new_slave->link  = BOND_LINK_DOWN;
1658         }
1659
1660         if (bond_update_speed_duplex(new_slave) &&
1661             (new_slave->link != BOND_LINK_DOWN)) {
1662                 printk(KERN_WARNING DRV_NAME
1663                        ": %s: Warning: failed to get speed and duplex from %s, "
1664                        "assumed to be 100Mb/sec and Full.\n",
1665                        bond_dev->name, new_slave->dev->name);
1666
1667                 if (bond->params.mode == BOND_MODE_8023AD) {
1668                         printk(KERN_WARNING DRV_NAME
1669                                ": %s: Warning: Operation of 802.3ad mode requires ETHTOOL "
1670                                "support in base driver for proper aggregator "
1671                                "selection.\n", bond_dev->name);
1672                 }
1673         }
1674
1675         if (USES_PRIMARY(bond->params.mode) && bond->params.primary[0]) {
1676                 /* if there is a primary slave, remember it */
1677                 if (strcmp(bond->params.primary, new_slave->dev->name) == 0) {
1678                         bond->primary_slave = new_slave;
1679                 }
1680         }
1681
1682         write_lock_bh(&bond->curr_slave_lock);
1683
1684         switch (bond->params.mode) {
1685         case BOND_MODE_ACTIVEBACKUP:
1686                 bond_set_slave_inactive_flags(new_slave);
1687                 bond_select_active_slave(bond);
1688                 break;
1689         case BOND_MODE_8023AD:
1690                 /* in 802.3ad mode, the internal mechanism
1691                  * will activate the slaves in the selected
1692                  * aggregator
1693                  */
1694                 bond_set_slave_inactive_flags(new_slave);
1695                 /* if this is the first slave */
1696                 if (bond->slave_cnt == 1) {
1697                         SLAVE_AD_INFO(new_slave).id = 1;
1698                         /* Initialize AD with the number of times that the AD timer is called in 1 second
1699                          * can be called only after the mac address of the bond is set
1700                          */
1701                         bond_3ad_initialize(bond, 1000/AD_TIMER_INTERVAL,
1702                                             bond->params.lacp_fast);
1703                 } else {
1704                         SLAVE_AD_INFO(new_slave).id =
1705                                 SLAVE_AD_INFO(new_slave->prev).id + 1;
1706                 }
1707
1708                 bond_3ad_bind_slave(new_slave);
1709                 break;
1710         case BOND_MODE_TLB:
1711         case BOND_MODE_ALB:
1712                 new_slave->state = BOND_STATE_ACTIVE;
1713                 bond_set_slave_inactive_flags(new_slave);
1714                 break;
1715         default:
1716                 dprintk("This slave is always active in trunk mode\n");
1717
1718                 /* always active in trunk mode */
1719                 new_slave->state = BOND_STATE_ACTIVE;
1720
1721                 /* In trunking mode there is little meaning to curr_active_slave
1722                  * anyway (it holds no special properties of the bond device),
1723                  * so we can change it without calling change_active_interface()
1724                  */
1725                 if (!bond->curr_active_slave) {
1726                         bond->curr_active_slave = new_slave;
1727                 }
1728                 break;
1729         } /* switch(bond_mode) */
1730
1731         write_unlock_bh(&bond->curr_slave_lock);
1732
1733         bond_set_carrier(bond);
1734
1735         read_unlock(&bond->lock);
1736
1737         res = bond_create_slave_symlinks(bond_dev, slave_dev);
1738         if (res)
1739                 goto err_close;
1740
1741         printk(KERN_INFO DRV_NAME
1742                ": %s: enslaving %s as a%s interface with a%s link.\n",
1743                bond_dev->name, slave_dev->name,
1744                new_slave->state == BOND_STATE_ACTIVE ? "n active" : " backup",
1745                new_slave->link != BOND_LINK_DOWN ? "n up" : " down");
1746
1747         /* enslave is successful */
1748         return 0;
1749
1750 /* Undo stages on error */
1751 err_close:
1752         dev_close(slave_dev);
1753
1754 err_unset_master:
1755         netdev_set_master(slave_dev, NULL);
1756
1757 err_restore_mac:
1758         if (!bond->params.fail_over_mac) {
1759                 /* XXX TODO - fom follow mode needs to change master's
1760                  * MAC if this slave's MAC is in use by the bond, or at
1761                  * least print a warning.
1762                  */
1763                 memcpy(addr.sa_data, new_slave->perm_hwaddr, ETH_ALEN);
1764                 addr.sa_family = slave_dev->type;
1765                 dev_set_mac_address(slave_dev, &addr);
1766         }
1767
1768 err_free:
1769         kfree(new_slave);
1770
1771 err_undo_flags:
1772         bond_dev->features = old_features;
1773  
1774         return res;
1775 }
1776
1777 /*
1778  * Try to release the slave device <slave> from the bond device <master>
1779  * It is legal to access curr_active_slave without a lock because all the function
1780  * is write-locked.
1781  *
1782  * The rules for slave state should be:
1783  *   for Active/Backup:
1784  *     Active stays on all backups go down
1785  *   for Bonded connections:
1786  *     The first up interface should be left on and all others downed.
1787  */
1788 int bond_release(struct net_device *bond_dev, struct net_device *slave_dev)
1789 {
1790         struct bonding *bond = bond_dev->priv;
1791         struct slave *slave, *oldcurrent;
1792         struct sockaddr addr;
1793         int mac_addr_differ;
1794         DECLARE_MAC_BUF(mac);
1795
1796         /* slave is not a slave or master is not master of this slave */
1797         if (!(slave_dev->flags & IFF_SLAVE) ||
1798             (slave_dev->master != bond_dev)) {
1799                 printk(KERN_ERR DRV_NAME
1800                        ": %s: Error: cannot release %s.\n",
1801                        bond_dev->name, slave_dev->name);
1802                 return -EINVAL;
1803         }
1804
1805         write_lock_bh(&bond->lock);
1806
1807         slave = bond_get_slave_by_dev(bond, slave_dev);
1808         if (!slave) {
1809                 /* not a slave of this bond */
1810                 printk(KERN_INFO DRV_NAME
1811                        ": %s: %s not enslaved\n",
1812                        bond_dev->name, slave_dev->name);
1813                 write_unlock_bh(&bond->lock);
1814                 return -EINVAL;
1815         }
1816
1817         if (!bond->params.fail_over_mac) {
1818                 mac_addr_differ = memcmp(bond_dev->dev_addr, slave->perm_hwaddr,
1819                                          ETH_ALEN);
1820                 if (!mac_addr_differ && (bond->slave_cnt > 1))
1821                         printk(KERN_WARNING DRV_NAME
1822                                ": %s: Warning: the permanent HWaddr of %s - "
1823                                "%s - is still in use by %s. "
1824                                "Set the HWaddr of %s to a different address "
1825                                "to avoid conflicts.\n",
1826                                bond_dev->name, slave_dev->name,
1827                                print_mac(mac, slave->perm_hwaddr),
1828                                bond_dev->name, slave_dev->name);
1829         }
1830
1831         /* Inform AD package of unbinding of slave. */
1832         if (bond->params.mode == BOND_MODE_8023AD) {
1833                 /* must be called before the slave is
1834                  * detached from the list
1835                  */
1836                 bond_3ad_unbind_slave(slave);
1837         }
1838
1839         printk(KERN_INFO DRV_NAME
1840                ": %s: releasing %s interface %s\n",
1841                bond_dev->name,
1842                (slave->state == BOND_STATE_ACTIVE)
1843                ? "active" : "backup",
1844                slave_dev->name);
1845
1846         oldcurrent = bond->curr_active_slave;
1847
1848         bond->current_arp_slave = NULL;
1849
1850         /* release the slave from its bond */
1851         bond_detach_slave(bond, slave);
1852
1853         bond_compute_features(bond);
1854
1855         if (bond->primary_slave == slave) {
1856                 bond->primary_slave = NULL;
1857         }
1858
1859         if (oldcurrent == slave) {
1860                 bond_change_active_slave(bond, NULL);
1861         }
1862
1863         if ((bond->params.mode == BOND_MODE_TLB) ||
1864             (bond->params.mode == BOND_MODE_ALB)) {
1865                 /* Must be called only after the slave has been
1866                  * detached from the list and the curr_active_slave
1867                  * has been cleared (if our_slave == old_current),
1868                  * but before a new active slave is selected.
1869                  */
1870                 write_unlock_bh(&bond->lock);
1871                 bond_alb_deinit_slave(bond, slave);
1872                 write_lock_bh(&bond->lock);
1873         }
1874
1875         if (oldcurrent == slave) {
1876                 /*
1877                  * Note that we hold RTNL over this sequence, so there
1878                  * is no concern that another slave add/remove event
1879                  * will interfere.
1880                  */
1881                 write_unlock_bh(&bond->lock);
1882                 read_lock(&bond->lock);
1883                 write_lock_bh(&bond->curr_slave_lock);
1884
1885                 bond_select_active_slave(bond);
1886
1887                 write_unlock_bh(&bond->curr_slave_lock);
1888                 read_unlock(&bond->lock);
1889                 write_lock_bh(&bond->lock);
1890         }
1891
1892         if (bond->slave_cnt == 0) {
1893                 bond_set_carrier(bond);
1894
1895                 /* if the last slave was removed, zero the mac address
1896                  * of the master so it will be set by the application
1897                  * to the mac address of the first slave
1898                  */
1899                 memset(bond_dev->dev_addr, 0, bond_dev->addr_len);
1900
1901                 if (list_empty(&bond->vlan_list)) {
1902                         bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
1903                 } else {
1904                         printk(KERN_WARNING DRV_NAME
1905                                ": %s: Warning: clearing HW address of %s while it "
1906                                "still has VLANs.\n",
1907                                bond_dev->name, bond_dev->name);
1908                         printk(KERN_WARNING DRV_NAME
1909                                ": %s: When re-adding slaves, make sure the bond's "
1910                                "HW address matches its VLANs'.\n",
1911                                bond_dev->name);
1912                 }
1913         } else if ((bond_dev->features & NETIF_F_VLAN_CHALLENGED) &&
1914                    !bond_has_challenged_slaves(bond)) {
1915                 printk(KERN_INFO DRV_NAME
1916                        ": %s: last VLAN challenged slave %s "
1917                        "left bond %s. VLAN blocking is removed\n",
1918                        bond_dev->name, slave_dev->name, bond_dev->name);
1919                 bond_dev->features &= ~NETIF_F_VLAN_CHALLENGED;
1920         }
1921
1922         write_unlock_bh(&bond->lock);
1923
1924         /* must do this from outside any spinlocks */
1925         bond_destroy_slave_symlinks(bond_dev, slave_dev);
1926
1927         bond_del_vlans_from_slave(bond, slave_dev);
1928
1929         /* If the mode USES_PRIMARY, then we should only remove its
1930          * promisc and mc settings if it was the curr_active_slave, but that was
1931          * already taken care of above when we detached the slave
1932          */
1933         if (!USES_PRIMARY(bond->params.mode)) {
1934                 /* unset promiscuity level from slave */
1935                 if (bond_dev->flags & IFF_PROMISC) {
1936                         dev_set_promiscuity(slave_dev, -1);
1937                 }
1938
1939                 /* unset allmulti level from slave */
1940                 if (bond_dev->flags & IFF_ALLMULTI) {
1941                         dev_set_allmulti(slave_dev, -1);
1942                 }
1943
1944                 /* flush master's mc_list from slave */
1945                 netif_addr_lock_bh(bond_dev);
1946                 bond_mc_list_flush(bond_dev, slave_dev);
1947                 netif_addr_unlock_bh(bond_dev);
1948         }
1949
1950         netdev_set_master(slave_dev, NULL);
1951
1952         /* close slave before restoring its mac address */
1953         dev_close(slave_dev);
1954
1955         if (bond->params.fail_over_mac != BOND_FOM_ACTIVE) {
1956                 /* restore original ("permanent") mac address */
1957                 memcpy(addr.sa_data, slave->perm_hwaddr, ETH_ALEN);
1958                 addr.sa_family = slave_dev->type;
1959                 dev_set_mac_address(slave_dev, &addr);
1960         }
1961
1962         slave_dev->priv_flags &= ~(IFF_MASTER_8023AD | IFF_MASTER_ALB |
1963                                    IFF_SLAVE_INACTIVE | IFF_BONDING |
1964                                    IFF_SLAVE_NEEDARP);
1965
1966         kfree(slave);
1967
1968         return 0;  /* deletion OK */
1969 }
1970
1971 /*
1972 * Destroy a bonding device.
1973 * Must be under rtnl_lock when this function is called.
1974 */
1975 void bond_destroy(struct bonding *bond)
1976 {
1977         bond_deinit(bond->dev);
1978         bond_destroy_sysfs_entry(bond);
1979         unregister_netdevice(bond->dev);
1980 }
1981
1982 /*
1983 * First release a slave and than destroy the bond if no more slaves iare left.
1984 * Must be under rtnl_lock when this function is called.
1985 */
1986 int  bond_release_and_destroy(struct net_device *bond_dev, struct net_device *slave_dev)
1987 {
1988         struct bonding *bond = bond_dev->priv;
1989         int ret;
1990
1991         ret = bond_release(bond_dev, slave_dev);
1992         if ((ret == 0) && (bond->slave_cnt == 0)) {
1993                 printk(KERN_INFO DRV_NAME ": %s: destroying bond %s.\n",
1994                        bond_dev->name, bond_dev->name);
1995                 bond_destroy(bond);
1996         }
1997         return ret;
1998 }
1999
2000 /*
2001  * This function releases all slaves.
2002  */
2003 static int bond_release_all(struct net_device *bond_dev)
2004 {
2005         struct bonding *bond = bond_dev->priv;
2006         struct slave *slave;
2007         struct net_device *slave_dev;
2008         struct sockaddr addr;
2009
2010         write_lock_bh(&bond->lock);
2011
2012         netif_carrier_off(bond_dev);
2013
2014         if (bond->slave_cnt == 0) {
2015                 goto out;
2016         }
2017
2018         bond->current_arp_slave = NULL;
2019         bond->primary_slave = NULL;
2020         bond_change_active_slave(bond, NULL);
2021
2022         while ((slave = bond->first_slave) != NULL) {
2023                 /* Inform AD package of unbinding of slave
2024                  * before slave is detached from the list.
2025                  */
2026                 if (bond->params.mode == BOND_MODE_8023AD) {
2027                         bond_3ad_unbind_slave(slave);
2028                 }
2029
2030                 slave_dev = slave->dev;
2031                 bond_detach_slave(bond, slave);
2032
2033                 /* now that the slave is detached, unlock and perform
2034                  * all the undo steps that should not be called from
2035                  * within a lock.
2036                  */
2037                 write_unlock_bh(&bond->lock);
2038
2039                 if ((bond->params.mode == BOND_MODE_TLB) ||
2040                     (bond->params.mode == BOND_MODE_ALB)) {
2041                         /* must be called only after the slave
2042                          * has been detached from the list
2043                          */
2044                         bond_alb_deinit_slave(bond, slave);
2045                 }
2046
2047                 bond_compute_features(bond);
2048
2049                 bond_destroy_slave_symlinks(bond_dev, slave_dev);
2050                 bond_del_vlans_from_slave(bond, slave_dev);
2051
2052                 /* If the mode USES_PRIMARY, then we should only remove its
2053                  * promisc and mc settings if it was the curr_active_slave, but that was
2054                  * already taken care of above when we detached the slave
2055                  */
2056                 if (!USES_PRIMARY(bond->params.mode)) {
2057                         /* unset promiscuity level from slave */
2058                         if (bond_dev->flags & IFF_PROMISC) {
2059                                 dev_set_promiscuity(slave_dev, -1);
2060                         }
2061
2062                         /* unset allmulti level from slave */
2063                         if (bond_dev->flags & IFF_ALLMULTI) {
2064                                 dev_set_allmulti(slave_dev, -1);
2065                         }
2066
2067                         /* flush master's mc_list from slave */
2068                         netif_addr_lock_bh(bond_dev);
2069                         bond_mc_list_flush(bond_dev, slave_dev);
2070                         netif_addr_unlock_bh(bond_dev);
2071                 }
2072
2073                 netdev_set_master(slave_dev, NULL);
2074
2075                 /* close slave before restoring its mac address */
2076                 dev_close(slave_dev);
2077
2078                 if (!bond->params.fail_over_mac) {
2079                         /* restore original ("permanent") mac address*/
2080                         memcpy(addr.sa_data, slave->perm_hwaddr, ETH_ALEN);
2081                         addr.sa_family = slave_dev->type;
2082                         dev_set_mac_address(slave_dev, &addr);
2083                 }
2084
2085                 slave_dev->priv_flags &= ~(IFF_MASTER_8023AD | IFF_MASTER_ALB |
2086                                            IFF_SLAVE_INACTIVE);
2087
2088                 kfree(slave);
2089
2090                 /* re-acquire the lock before getting the next slave */
2091                 write_lock_bh(&bond->lock);
2092         }
2093
2094         /* zero the mac address of the master so it will be
2095          * set by the application to the mac address of the
2096          * first slave
2097          */
2098         memset(bond_dev->dev_addr, 0, bond_dev->addr_len);
2099
2100         if (list_empty(&bond->vlan_list)) {
2101                 bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
2102         } else {
2103                 printk(KERN_WARNING DRV_NAME
2104                        ": %s: Warning: clearing HW address of %s while it "
2105                        "still has VLANs.\n",
2106                        bond_dev->name, bond_dev->name);
2107                 printk(KERN_WARNING DRV_NAME
2108                        ": %s: When re-adding slaves, make sure the bond's "
2109                        "HW address matches its VLANs'.\n",
2110                        bond_dev->name);
2111         }
2112
2113         printk(KERN_INFO DRV_NAME
2114                ": %s: released all slaves\n",
2115                bond_dev->name);
2116
2117 out:
2118         write_unlock_bh(&bond->lock);
2119
2120         return 0;
2121 }
2122
2123 /*
2124  * This function changes the active slave to slave <slave_dev>.
2125  * It returns -EINVAL in the following cases.
2126  *  - <slave_dev> is not found in the list.
2127  *  - There is not active slave now.
2128  *  - <slave_dev> is already active.
2129  *  - The link state of <slave_dev> is not BOND_LINK_UP.
2130  *  - <slave_dev> is not running.
2131  * In these cases, this fuction does nothing.
2132  * In the other cases, currnt_slave pointer is changed and 0 is returned.
2133  */
2134 static int bond_ioctl_change_active(struct net_device *bond_dev, struct net_device *slave_dev)
2135 {
2136         struct bonding *bond = bond_dev->priv;
2137         struct slave *old_active = NULL;
2138         struct slave *new_active = NULL;
2139         int res = 0;
2140
2141         if (!USES_PRIMARY(bond->params.mode)) {
2142                 return -EINVAL;
2143         }
2144
2145         /* Verify that master_dev is indeed the master of slave_dev */
2146         if (!(slave_dev->flags & IFF_SLAVE) ||
2147             (slave_dev->master != bond_dev)) {
2148                 return -EINVAL;
2149         }
2150
2151         read_lock(&bond->lock);
2152
2153         read_lock(&bond->curr_slave_lock);
2154         old_active = bond->curr_active_slave;
2155         read_unlock(&bond->curr_slave_lock);
2156
2157         new_active = bond_get_slave_by_dev(bond, slave_dev);
2158
2159         /*
2160          * Changing to the current active: do nothing; return success.
2161          */
2162         if (new_active && (new_active == old_active)) {
2163                 read_unlock(&bond->lock);
2164                 return 0;
2165         }
2166
2167         if ((new_active) &&
2168             (old_active) &&
2169             (new_active->link == BOND_LINK_UP) &&
2170             IS_UP(new_active->dev)) {
2171                 write_lock_bh(&bond->curr_slave_lock);
2172                 bond_change_active_slave(bond, new_active);
2173                 write_unlock_bh(&bond->curr_slave_lock);
2174         } else {
2175                 res = -EINVAL;
2176         }
2177
2178         read_unlock(&bond->lock);
2179
2180         return res;
2181 }
2182
2183 static int bond_info_query(struct net_device *bond_dev, struct ifbond *info)
2184 {
2185         struct bonding *bond = bond_dev->priv;
2186
2187         info->bond_mode = bond->params.mode;
2188         info->miimon = bond->params.miimon;
2189
2190         read_lock(&bond->lock);
2191         info->num_slaves = bond->slave_cnt;
2192         read_unlock(&bond->lock);
2193
2194         return 0;
2195 }
2196
2197 static int bond_slave_info_query(struct net_device *bond_dev, struct ifslave *info)
2198 {
2199         struct bonding *bond = bond_dev->priv;
2200         struct slave *slave;
2201         int i, found = 0;
2202
2203         if (info->slave_id < 0) {
2204                 return -ENODEV;
2205         }
2206
2207         read_lock(&bond->lock);
2208
2209         bond_for_each_slave(bond, slave, i) {
2210                 if (i == (int)info->slave_id) {
2211                         found = 1;
2212                         break;
2213                 }
2214         }
2215
2216         read_unlock(&bond->lock);
2217
2218         if (found) {
2219                 strcpy(info->slave_name, slave->dev->name);
2220                 info->link = slave->link;
2221                 info->state = slave->state;
2222                 info->link_failure_count = slave->link_failure_count;
2223         } else {
2224                 return -ENODEV;
2225         }
2226
2227         return 0;
2228 }
2229
2230 /*-------------------------------- Monitoring -------------------------------*/
2231
2232
2233 static int bond_miimon_inspect(struct bonding *bond)
2234 {
2235         struct slave *slave;
2236         int i, link_state, commit = 0;
2237
2238         bond_for_each_slave(bond, slave, i) {
2239                 slave->new_link = BOND_LINK_NOCHANGE;
2240
2241                 link_state = bond_check_dev_link(bond, slave->dev, 0);
2242
2243                 switch (slave->link) {
2244                 case BOND_LINK_UP:
2245                         if (link_state)
2246                                 continue;
2247
2248                         slave->link = BOND_LINK_FAIL;
2249                         slave->delay = bond->params.downdelay;
2250                         if (slave->delay) {
2251                                 printk(KERN_INFO DRV_NAME
2252                                        ": %s: link status down for %s"
2253                                        "interface %s, disabling it in %d ms.\n",
2254                                        bond->dev->name,
2255                                        (bond->params.mode ==
2256                                         BOND_MODE_ACTIVEBACKUP) ?
2257                                        ((slave->state == BOND_STATE_ACTIVE) ?
2258                                         "active " : "backup ") : "",
2259                                        slave->dev->name,
2260                                        bond->params.downdelay * bond->params.miimon);
2261                         }
2262                         /*FALLTHRU*/
2263                 case BOND_LINK_FAIL:
2264                         if (link_state) {
2265                                 /*
2266                                  * recovered before downdelay expired
2267                                  */
2268                                 slave->link = BOND_LINK_UP;
2269                                 slave->jiffies = jiffies;
2270                                 printk(KERN_INFO DRV_NAME
2271                                        ": %s: link status up again after %d "
2272                                        "ms for interface %s.\n",
2273                                        bond->dev->name,
2274                                        (bond->params.downdelay - slave->delay) *
2275                                        bond->params.miimon,
2276                                        slave->dev->name);
2277                                 continue;
2278                         }
2279
2280                         if (slave->delay <= 0) {
2281                                 slave->new_link = BOND_LINK_DOWN;
2282                                 commit++;
2283                                 continue;
2284                         }
2285
2286                         slave->delay--;
2287                         break;
2288
2289                 case BOND_LINK_DOWN:
2290                         if (!link_state)
2291                                 continue;
2292
2293                         slave->link = BOND_LINK_BACK;
2294                         slave->delay = bond->params.updelay;
2295
2296                         if (slave->delay) {
2297                                 printk(KERN_INFO DRV_NAME
2298                                        ": %s: link status up for "
2299                                        "interface %s, enabling it in %d ms.\n",
2300                                        bond->dev->name, slave->dev->name,
2301                                        bond->params.updelay *
2302                                        bond->params.miimon);
2303                         }
2304                         /*FALLTHRU*/
2305                 case BOND_LINK_BACK:
2306                         if (!link_state) {
2307                                 slave->link = BOND_LINK_DOWN;
2308                                 printk(KERN_INFO DRV_NAME
2309                                        ": %s: link status down again after %d "
2310                                        "ms for interface %s.\n",
2311                                        bond->dev->name,
2312                                        (bond->params.updelay - slave->delay) *
2313                                        bond->params.miimon,
2314                                        slave->dev->name);
2315
2316                                 continue;
2317                         }
2318
2319                         if (slave->delay <= 0) {
2320                                 slave->new_link = BOND_LINK_UP;
2321                                 commit++;
2322                                 continue;
2323                         }
2324
2325                         slave->delay--;
2326                         break;
2327                 }
2328         }
2329
2330         return commit;
2331 }
2332
2333 static void bond_miimon_commit(struct bonding *bond)
2334 {
2335         struct slave *slave;
2336         int i;
2337
2338         bond_for_each_slave(bond, slave, i) {
2339                 switch (slave->new_link) {
2340                 case BOND_LINK_NOCHANGE:
2341                         continue;
2342
2343                 case BOND_LINK_UP:
2344                         slave->link = BOND_LINK_UP;
2345                         slave->jiffies = jiffies;
2346
2347                         if (bond->params.mode == BOND_MODE_8023AD) {
2348                                 /* prevent it from being the active one */
2349                                 slave->state = BOND_STATE_BACKUP;
2350                         } else if (bond->params.mode != BOND_MODE_ACTIVEBACKUP) {
2351                                 /* make it immediately active */
2352                                 slave->state = BOND_STATE_ACTIVE;
2353                         } else if (slave != bond->primary_slave) {
2354                                 /* prevent it from being the active one */
2355                                 slave->state = BOND_STATE_BACKUP;
2356                         }
2357
2358                         printk(KERN_INFO DRV_NAME
2359                                ": %s: link status definitely "
2360                                "up for interface %s.\n",
2361                                bond->dev->name, slave->dev->name);
2362
2363                         /* notify ad that the link status has changed */
2364                         if (bond->params.mode == BOND_MODE_8023AD)
2365                                 bond_3ad_handle_link_change(slave, BOND_LINK_UP);
2366
2367                         if ((bond->params.mode == BOND_MODE_TLB) ||
2368                             (bond->params.mode == BOND_MODE_ALB))
2369                                 bond_alb_handle_link_change(bond, slave,
2370                                                             BOND_LINK_UP);
2371
2372                         if (!bond->curr_active_slave ||
2373                             (slave == bond->primary_slave))
2374                                 goto do_failover;
2375
2376                         continue;
2377
2378                 case BOND_LINK_DOWN:
2379                         if (slave->link_failure_count < UINT_MAX)
2380                                 slave->link_failure_count++;
2381
2382                         slave->link = BOND_LINK_DOWN;
2383
2384                         if (bond->params.mode == BOND_MODE_ACTIVEBACKUP ||
2385                             bond->params.mode == BOND_MODE_8023AD)
2386                                 bond_set_slave_inactive_flags(slave);
2387
2388                         printk(KERN_INFO DRV_NAME
2389                                ": %s: link status definitely down for "
2390                                "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->params.mode == BOND_MODE_TLB ||
2398                             bond->params.mode == BOND_MODE_ALB)
2399                                 bond_alb_handle_link_change(bond, slave,
2400                                                             BOND_LINK_DOWN);
2401
2402                         if (slave == bond->curr_active_slave)
2403                                 goto do_failover;
2404
2405                         continue;
2406
2407                 default:
2408                         printk(KERN_ERR DRV_NAME
2409                                ": %s: invalid new link %d on slave %s\n",
2410                                bond->dev->name, slave->new_link,
2411                                slave->dev->name);
2412                         slave->new_link = BOND_LINK_NOCHANGE;
2413
2414                         continue;
2415                 }
2416
2417 do_failover:
2418                 ASSERT_RTNL();
2419                 write_lock_bh(&bond->curr_slave_lock);
2420                 bond_select_active_slave(bond);
2421                 write_unlock_bh(&bond->curr_slave_lock);
2422         }
2423
2424         bond_set_carrier(bond);
2425 }
2426
2427 /*
2428  * bond_mii_monitor
2429  *
2430  * Really a wrapper that splits the mii monitor into two phases: an
2431  * inspection, then (if inspection indicates something needs to be done)
2432  * an acquisition of appropriate locks followed by a commit phase to
2433  * implement whatever link state changes are indicated.
2434  */
2435 void bond_mii_monitor(struct work_struct *work)
2436 {
2437         struct bonding *bond = container_of(work, struct bonding,
2438                                             mii_work.work);
2439
2440         read_lock(&bond->lock);
2441         if (bond->kill_timers)
2442                 goto out;
2443
2444         if (bond->slave_cnt == 0)
2445                 goto re_arm;
2446
2447         if (bond->send_grat_arp) {
2448                 read_lock(&bond->curr_slave_lock);
2449                 bond_send_gratuitous_arp(bond);
2450                 read_unlock(&bond->curr_slave_lock);
2451         }
2452
2453         if (bond_miimon_inspect(bond)) {
2454                 read_unlock(&bond->lock);
2455                 rtnl_lock();
2456                 read_lock(&bond->lock);
2457
2458                 bond_miimon_commit(bond);
2459
2460                 read_unlock(&bond->lock);
2461                 rtnl_unlock();  /* might sleep, hold no other locks */
2462                 read_lock(&bond->lock);
2463         }
2464
2465 re_arm:
2466         if (bond->params.miimon)
2467                 queue_delayed_work(bond->wq, &bond->mii_work,
2468                                    msecs_to_jiffies(bond->params.miimon));
2469 out:
2470         read_unlock(&bond->lock);
2471 }
2472
2473 static __be32 bond_glean_dev_ip(struct net_device *dev)
2474 {
2475         struct in_device *idev;
2476         struct in_ifaddr *ifa;
2477         __be32 addr = 0;
2478
2479         if (!dev)
2480                 return 0;
2481
2482         rcu_read_lock();
2483         idev = __in_dev_get_rcu(dev);
2484         if (!idev)
2485                 goto out;
2486
2487         ifa = idev->ifa_list;
2488         if (!ifa)
2489                 goto out;
2490
2491         addr = ifa->ifa_local;
2492 out:
2493         rcu_read_unlock();
2494         return addr;
2495 }
2496
2497 static int bond_has_this_ip(struct bonding *bond, __be32 ip)
2498 {
2499         struct vlan_entry *vlan;
2500
2501         if (ip == bond->master_ip)
2502                 return 1;
2503
2504         list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
2505                 if (ip == vlan->vlan_ip)
2506                         return 1;
2507         }
2508
2509         return 0;
2510 }
2511
2512 /*
2513  * We go to the (large) trouble of VLAN tagging ARP frames because
2514  * switches in VLAN mode (especially if ports are configured as
2515  * "native" to a VLAN) might not pass non-tagged frames.
2516  */
2517 static void bond_arp_send(struct net_device *slave_dev, int arp_op, __be32 dest_ip, __be32 src_ip, unsigned short vlan_id)
2518 {
2519         struct sk_buff *skb;
2520
2521         dprintk("arp %d on slave %s: dst %x src %x vid %d\n", arp_op,
2522                slave_dev->name, dest_ip, src_ip, vlan_id);
2523                
2524         skb = arp_create(arp_op, ETH_P_ARP, dest_ip, slave_dev, src_ip,
2525                          NULL, slave_dev->dev_addr, NULL);
2526
2527         if (!skb) {
2528                 printk(KERN_ERR DRV_NAME ": ARP packet allocation failed\n");
2529                 return;
2530         }
2531         if (vlan_id) {
2532                 skb = vlan_put_tag(skb, vlan_id);
2533                 if (!skb) {
2534                         printk(KERN_ERR DRV_NAME ": failed to insert VLAN tag\n");
2535                         return;
2536                 }
2537         }
2538         arp_xmit(skb);
2539 }
2540
2541
2542 static void bond_arp_send_all(struct bonding *bond, struct slave *slave)
2543 {
2544         int i, vlan_id, rv;
2545         __be32 *targets = bond->params.arp_targets;
2546         struct vlan_entry *vlan;
2547         struct net_device *vlan_dev;
2548         struct flowi fl;
2549         struct rtable *rt;
2550
2551         for (i = 0; (i < BOND_MAX_ARP_TARGETS); i++) {
2552                 if (!targets[i])
2553                         continue;
2554                 dprintk("basa: target %x\n", targets[i]);
2555                 if (list_empty(&bond->vlan_list)) {
2556                         dprintk("basa: empty vlan: arp_send\n");
2557                         bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2558                                       bond->master_ip, 0);
2559                         continue;
2560                 }
2561
2562                 /*
2563                  * If VLANs are configured, we do a route lookup to
2564                  * determine which VLAN interface would be used, so we
2565                  * can tag the ARP with the proper VLAN tag.
2566                  */
2567                 memset(&fl, 0, sizeof(fl));
2568                 fl.fl4_dst = targets[i];
2569                 fl.fl4_tos = RTO_ONLINK;
2570
2571                 rv = ip_route_output_key(&init_net, &rt, &fl);
2572                 if (rv) {
2573                         if (net_ratelimit()) {
2574                                 printk(KERN_WARNING DRV_NAME
2575                              ": %s: no route to arp_ip_target %u.%u.%u.%u\n",
2576                                        bond->dev->name, NIPQUAD(fl.fl4_dst));
2577                         }
2578                         continue;
2579                 }
2580
2581                 /*
2582                  * This target is not on a VLAN
2583                  */
2584                 if (rt->u.dst.dev == bond->dev) {
2585                         ip_rt_put(rt);
2586                         dprintk("basa: rtdev == bond->dev: arp_send\n");
2587                         bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2588                                       bond->master_ip, 0);
2589                         continue;
2590                 }
2591
2592                 vlan_id = 0;
2593                 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
2594                         vlan_dev = vlan_group_get_device(bond->vlgrp, vlan->vlan_id);
2595                         if (vlan_dev == rt->u.dst.dev) {
2596                                 vlan_id = vlan->vlan_id;
2597                                 dprintk("basa: vlan match on %s %d\n",
2598                                        vlan_dev->name, vlan_id);
2599                                 break;
2600                         }
2601                 }
2602
2603                 if (vlan_id) {
2604                         ip_rt_put(rt);
2605                         bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2606                                       vlan->vlan_ip, vlan_id);
2607                         continue;
2608                 }
2609
2610                 if (net_ratelimit()) {
2611                         printk(KERN_WARNING DRV_NAME
2612                ": %s: no path to arp_ip_target %u.%u.%u.%u via rt.dev %s\n",
2613                                bond->dev->name, NIPQUAD(fl.fl4_dst),
2614                                rt->u.dst.dev ? rt->u.dst.dev->name : "NULL");
2615                 }
2616                 ip_rt_put(rt);
2617         }
2618 }
2619
2620 /*
2621  * Kick out a gratuitous ARP for an IP on the bonding master plus one
2622  * for each VLAN above us.
2623  *
2624  * Caller must hold curr_slave_lock for read or better
2625  */
2626 static void bond_send_gratuitous_arp(struct bonding *bond)
2627 {
2628         struct slave *slave = bond->curr_active_slave;
2629         struct vlan_entry *vlan;
2630         struct net_device *vlan_dev;
2631
2632         dprintk("bond_send_grat_arp: bond %s slave %s\n", bond->dev->name,
2633                                 slave ? slave->dev->name : "NULL");
2634
2635         if (!slave || !bond->send_grat_arp ||
2636             test_bit(__LINK_STATE_LINKWATCH_PENDING, &slave->dev->state))
2637                 return;
2638
2639         bond->send_grat_arp--;
2640
2641         if (bond->master_ip) {
2642                 bond_arp_send(slave->dev, ARPOP_REPLY, bond->master_ip,
2643                                 bond->master_ip, 0);
2644         }
2645
2646         list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
2647                 vlan_dev = vlan_group_get_device(bond->vlgrp, vlan->vlan_id);
2648                 if (vlan->vlan_ip) {
2649                         bond_arp_send(slave->dev, ARPOP_REPLY, vlan->vlan_ip,
2650                                       vlan->vlan_ip, vlan->vlan_id);
2651                 }
2652         }
2653 }
2654
2655 static void bond_validate_arp(struct bonding *bond, struct slave *slave, __be32 sip, __be32 tip)
2656 {
2657         int i;
2658         __be32 *targets = bond->params.arp_targets;
2659
2660         targets = bond->params.arp_targets;
2661         for (i = 0; (i < BOND_MAX_ARP_TARGETS) && targets[i]; i++) {
2662                 dprintk("bva: sip %u.%u.%u.%u tip %u.%u.%u.%u t[%d] "
2663                         "%u.%u.%u.%u bhti(tip) %d\n",
2664                        NIPQUAD(sip), NIPQUAD(tip), i, NIPQUAD(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_net(dev) != &init_net)
2683                 goto out;
2684
2685         if (!(dev->priv_flags & IFF_BONDING) || !(dev->flags & IFF_MASTER))
2686                 goto out;
2687
2688         bond = dev->priv;
2689         read_lock(&bond->lock);
2690
2691         dprintk("bond_arp_rcv: bond %s skb->dev %s orig_dev %s\n",
2692                 bond->dev->name, skb->dev ? skb->dev->name : "NULL",
2693                 orig_dev ? orig_dev->name : "NULL");
2694
2695         slave = bond_get_slave_by_dev(bond, orig_dev);
2696         if (!slave || !slave_do_arp_validate(bond, slave))
2697                 goto out_unlock;
2698
2699         if (!pskb_may_pull(skb, arp_hdr_len(dev)))
2700                 goto out_unlock;
2701
2702         arp = arp_hdr(skb);
2703         if (arp->ar_hln != dev->addr_len ||
2704             skb->pkt_type == PACKET_OTHERHOST ||
2705             skb->pkt_type == PACKET_LOOPBACK ||
2706             arp->ar_hrd != htons(ARPHRD_ETHER) ||
2707             arp->ar_pro != htons(ETH_P_IP) ||
2708             arp->ar_pln != 4)
2709                 goto out_unlock;
2710
2711         arp_ptr = (unsigned char *)(arp + 1);
2712         arp_ptr += dev->addr_len;
2713         memcpy(&sip, arp_ptr, 4);
2714         arp_ptr += 4 + dev->addr_len;
2715         memcpy(&tip, arp_ptr, 4);
2716
2717         dprintk("bond_arp_rcv: %s %s/%d av %d sv %d sip %u.%u.%u.%u"
2718                 " tip %u.%u.%u.%u\n", bond->dev->name, slave->dev->name,
2719                 slave->state, bond->params.arp_validate,
2720                 slave_do_arp_validate(bond, slave), NIPQUAD(sip), NIPQUAD(tip));
2721
2722         /*
2723          * Backup slaves won't see the ARP reply, but do come through
2724          * here for each ARP probe (so we swap the sip/tip to validate
2725          * the probe).  In a "redundant switch, common router" type of
2726          * configuration, the ARP probe will (hopefully) travel from
2727          * the active, through one switch, the router, then the other
2728          * switch before reaching the backup.
2729          */
2730         if (slave->state == BOND_STATE_ACTIVE)
2731                 bond_validate_arp(bond, slave, sip, tip);
2732         else
2733                 bond_validate_arp(bond, slave, tip, sip);
2734
2735 out_unlock:
2736         read_unlock(&bond->lock);
2737 out:
2738         dev_kfree_skb(skb);
2739         return NET_RX_SUCCESS;
2740 }
2741
2742 /*
2743  * this function is called regularly to monitor each slave's link
2744  * ensuring that traffic is being sent and received when arp monitoring
2745  * is used in load-balancing mode. if the adapter has been dormant, then an
2746  * arp is transmitted to generate traffic. see activebackup_arp_monitor for
2747  * arp monitoring in active backup mode.
2748  */
2749 void bond_loadbalance_arp_mon(struct work_struct *work)
2750 {
2751         struct bonding *bond = container_of(work, struct bonding,
2752                                             arp_work.work);
2753         struct slave *slave, *oldcurrent;
2754         int do_failover = 0;
2755         int delta_in_ticks;
2756         int i;
2757
2758         read_lock(&bond->lock);
2759
2760         delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
2761
2762         if (bond->kill_timers) {
2763                 goto out;
2764         }
2765
2766         if (bond->slave_cnt == 0) {
2767                 goto re_arm;
2768         }
2769
2770         read_lock(&bond->curr_slave_lock);
2771         oldcurrent = bond->curr_active_slave;
2772         read_unlock(&bond->curr_slave_lock);
2773
2774         /* see if any of the previous devices are up now (i.e. they have
2775          * xmt and rcv traffic). the curr_active_slave does not come into
2776          * the picture unless it is null. also, slave->jiffies is not needed
2777          * here because we send an arp on each slave and give a slave as
2778          * long as it needs to get the tx/rx within the delta.
2779          * TODO: what about up/down delay in arp mode? it wasn't here before
2780          *       so it can wait
2781          */
2782         bond_for_each_slave(bond, slave, i) {
2783                 if (slave->link != BOND_LINK_UP) {
2784                         if (time_before_eq(jiffies, slave->dev->trans_start + delta_in_ticks) &&
2785                             time_before_eq(jiffies, slave->dev->last_rx + delta_in_ticks)) {
2786
2787                                 slave->link  = BOND_LINK_UP;
2788                                 slave->state = BOND_STATE_ACTIVE;
2789
2790                                 /* primary_slave has no meaning in round-robin
2791                                  * mode. the window of a slave being up and
2792                                  * curr_active_slave being null after enslaving
2793                                  * is closed.
2794                                  */
2795                                 if (!oldcurrent) {
2796                                         printk(KERN_INFO DRV_NAME
2797                                                ": %s: link status definitely "
2798                                                "up for interface %s, ",
2799                                                bond->dev->name,
2800                                                slave->dev->name);
2801                                         do_failover = 1;
2802                                 } else {
2803                                         printk(KERN_INFO DRV_NAME
2804                                                ": %s: interface %s is now up\n",
2805                                                bond->dev->name,
2806                                                slave->dev->name);
2807                                 }
2808                         }
2809                 } else {
2810                         /* slave->link == BOND_LINK_UP */
2811
2812                         /* not all switches will respond to an arp request
2813                          * when the source ip is 0, so don't take the link down
2814                          * if we don't know our ip yet
2815                          */
2816                         if (time_after_eq(jiffies, slave->dev->trans_start + 2*delta_in_ticks) ||
2817                             (time_after_eq(jiffies, slave->dev->last_rx + 2*delta_in_ticks))) {
2818
2819                                 slave->link  = BOND_LINK_DOWN;
2820                                 slave->state = BOND_STATE_BACKUP;
2821
2822                                 if (slave->link_failure_count < UINT_MAX) {
2823                                         slave->link_failure_count++;
2824                                 }
2825
2826                                 printk(KERN_INFO DRV_NAME
2827                                        ": %s: interface %s is now down.\n",
2828                                        bond->dev->name,
2829                                        slave->dev->name);
2830
2831                                 if (slave == oldcurrent) {
2832                                         do_failover = 1;
2833                                 }
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
2849         if (do_failover) {
2850                 write_lock_bh(&bond->curr_slave_lock);
2851
2852                 bond_select_active_slave(bond);
2853
2854                 write_unlock_bh(&bond->curr_slave_lock);
2855         }
2856
2857 re_arm:
2858         if (bond->params.arp_interval)
2859                 queue_delayed_work(bond->wq, &bond->arp_work, delta_in_ticks);
2860 out:
2861         read_unlock(&bond->lock);
2862 }
2863
2864 /*
2865  * Called to inspect slaves for active-backup mode ARP monitor link state
2866  * changes.  Sets new_link in slaves to specify what action should take
2867  * place for the slave.  Returns 0 if no changes are found, >0 if changes
2868  * to link states must be committed.
2869  *
2870  * Called with bond->lock held for read.
2871  */
2872 static int bond_ab_arp_inspect(struct bonding *bond, int delta_in_ticks)
2873 {
2874         struct slave *slave;
2875         int i, commit = 0;
2876
2877         bond_for_each_slave(bond, slave, i) {
2878                 slave->new_link = BOND_LINK_NOCHANGE;
2879
2880                 if (slave->link != BOND_LINK_UP) {
2881                         if (time_before_eq(jiffies, slave_last_rx(bond, slave) +
2882                                            delta_in_ticks)) {
2883                                 slave->new_link = BOND_LINK_UP;
2884                                 commit++;
2885                         }
2886
2887                         continue;
2888                 }
2889
2890                 /*
2891                  * Give slaves 2*delta after being enslaved or made
2892                  * active.  This avoids bouncing, as the last receive
2893                  * times need a full ARP monitor cycle to be updated.
2894                  */
2895                 if (!time_after_eq(jiffies, slave->jiffies +
2896                                    2 * delta_in_ticks))
2897                         continue;
2898
2899                 /*
2900                  * Backup slave is down if:
2901                  * - No current_arp_slave AND
2902                  * - more than 3*delta since last receive AND
2903                  * - the bond has an IP address
2904                  *
2905                  * Note: a non-null current_arp_slave indicates
2906                  * the curr_active_slave went down and we are
2907                  * searching for a new one; under this condition
2908                  * we only take the curr_active_slave down - this
2909                  * gives each slave a chance to tx/rx traffic
2910                  * before being taken out
2911                  */
2912                 if (slave->state == BOND_STATE_BACKUP &&
2913                     !bond->current_arp_slave &&
2914                     time_after(jiffies, slave_last_rx(bond, slave) +
2915                                3 * delta_in_ticks)) {
2916                         slave->new_link = BOND_LINK_DOWN;
2917                         commit++;
2918                 }
2919
2920                 /*
2921                  * Active slave is down if:
2922                  * - more than 2*delta since transmitting OR
2923                  * - (more than 2*delta since receive AND
2924                  *    the bond has an IP address)
2925                  */
2926                 if ((slave->state == BOND_STATE_ACTIVE) &&
2927                     (time_after_eq(jiffies, slave->dev->trans_start +
2928                                     2 * delta_in_ticks) ||
2929                       (time_after_eq(jiffies, slave_last_rx(bond, slave)
2930                                      + 2 * delta_in_ticks)))) {
2931                         slave->new_link = BOND_LINK_DOWN;
2932                         commit++;
2933                 }
2934         }
2935
2936         read_lock(&bond->curr_slave_lock);
2937
2938         /*
2939          * Trigger a commit if the primary option setting has changed.
2940          */
2941         if (bond->primary_slave &&
2942             (bond->primary_slave != bond->curr_active_slave) &&
2943             (bond->primary_slave->link == BOND_LINK_UP))
2944                 commit++;
2945
2946         read_unlock(&bond->curr_slave_lock);
2947
2948         return commit;
2949 }
2950
2951 /*
2952  * Called to commit link state changes noted by inspection step of
2953  * active-backup mode ARP monitor.
2954  *
2955  * Called with RTNL and bond->lock for read.
2956  */
2957 static void bond_ab_arp_commit(struct bonding *bond, int delta_in_ticks)
2958 {
2959         struct slave *slave;
2960         int i;
2961
2962         bond_for_each_slave(bond, slave, i) {
2963                 switch (slave->new_link) {
2964                 case BOND_LINK_NOCHANGE:
2965                         continue;
2966
2967                 case BOND_LINK_UP:
2968                         write_lock_bh(&bond->curr_slave_lock);
2969
2970                         if (!bond->curr_active_slave &&
2971                             time_before_eq(jiffies, slave->dev->trans_start +
2972                                            delta_in_ticks)) {
2973                                 slave->link = BOND_LINK_UP;
2974                                 bond_change_active_slave(bond, slave);
2975                                 bond->current_arp_slave = NULL;
2976
2977                                 printk(KERN_INFO DRV_NAME
2978                                        ": %s: %s is up and now the "
2979                                        "active interface\n",
2980                                        bond->dev->name, slave->dev->name);
2981
2982                         } else if (bond->curr_active_slave != slave) {
2983                                 /* this slave has just come up but we
2984                                  * already have a current slave; this can
2985                                  * also happen if bond_enslave adds a new
2986                                  * slave that is up while we are searching
2987                                  * for a new slave
2988                                  */
2989                                 slave->link = BOND_LINK_UP;
2990                                 bond_set_slave_inactive_flags(slave);
2991                                 bond->current_arp_slave = NULL;
2992
2993                                 printk(KERN_INFO DRV_NAME
2994                                        ": %s: backup interface %s is now up\n",
2995                                        bond->dev->name, slave->dev->name);
2996                         }
2997
2998                         write_unlock_bh(&bond->curr_slave_lock);
2999
3000                         break;
3001
3002                 case BOND_LINK_DOWN:
3003                         if (slave->link_failure_count < UINT_MAX)
3004                                 slave->link_failure_count++;
3005
3006                         slave->link = BOND_LINK_DOWN;
3007
3008                         if (slave == bond->curr_active_slave) {
3009                                 printk(KERN_INFO DRV_NAME
3010                                        ": %s: link status down for active "
3011                                        "interface %s, disabling it\n",
3012                                        bond->dev->name, slave->dev->name);
3013
3014                                 bond_set_slave_inactive_flags(slave);
3015
3016                                 write_lock_bh(&bond->curr_slave_lock);
3017
3018                                 bond_select_active_slave(bond);
3019                                 if (bond->curr_active_slave)
3020                                         bond->curr_active_slave->jiffies =
3021                                                 jiffies;
3022
3023                                 write_unlock_bh(&bond->curr_slave_lock);
3024
3025                                 bond->current_arp_slave = NULL;
3026
3027                         } else if (slave->state == BOND_STATE_BACKUP) {
3028                                 printk(KERN_INFO DRV_NAME
3029                                        ": %s: backup interface %s is now down\n",
3030                                        bond->dev->name, slave->dev->name);
3031
3032                                 bond_set_slave_inactive_flags(slave);
3033                         }
3034                         break;
3035
3036                 default:
3037                         printk(KERN_ERR DRV_NAME
3038                                ": %s: impossible: new_link %d on slave %s\n",
3039                                bond->dev->name, slave->new_link,
3040                                slave->dev->name);
3041                 }
3042         }
3043
3044         /*
3045          * No race with changes to primary via sysfs, as we hold rtnl.
3046          */
3047         if (bond->primary_slave &&
3048             (bond->primary_slave != bond->curr_active_slave) &&
3049             (bond->primary_slave->link == BOND_LINK_UP)) {
3050                 write_lock_bh(&bond->curr_slave_lock);
3051                 bond_change_active_slave(bond, bond->primary_slave);
3052                 write_unlock_bh(&bond->curr_slave_lock);
3053         }
3054
3055         bond_set_carrier(bond);
3056 }
3057
3058 /*
3059  * Send ARP probes for active-backup mode ARP monitor.
3060  *
3061  * Called with bond->lock held for read.
3062  */
3063 static void bond_ab_arp_probe(struct bonding *bond)
3064 {
3065         struct slave *slave;
3066         int i;
3067
3068         read_lock(&bond->curr_slave_lock);
3069
3070         if (bond->current_arp_slave && bond->curr_active_slave)
3071                 printk("PROBE: c_arp %s && cas %s BAD\n",
3072                        bond->current_arp_slave->dev->name,
3073                        bond->curr_active_slave->dev->name);
3074
3075         if (bond->curr_active_slave) {
3076                 bond_arp_send_all(bond, bond->curr_active_slave);
3077                 read_unlock(&bond->curr_slave_lock);
3078                 return;
3079         }
3080
3081         read_unlock(&bond->curr_slave_lock);
3082
3083         /* if we don't have a curr_active_slave, search for the next available
3084          * backup slave from the current_arp_slave and make it the candidate
3085          * for becoming the curr_active_slave
3086          */
3087
3088         if (!bond->current_arp_slave) {
3089                 bond->current_arp_slave = bond->first_slave;
3090                 if (!bond->current_arp_slave)
3091                         return;
3092         }
3093
3094         bond_set_slave_inactive_flags(bond->current_arp_slave);
3095
3096         /* search for next candidate */
3097         bond_for_each_slave_from(bond, slave, i, bond->current_arp_slave->next) {
3098                 if (IS_UP(slave->dev)) {
3099                         slave->link = BOND_LINK_BACK;
3100                         bond_set_slave_active_flags(slave);
3101                         bond_arp_send_all(bond, slave);
3102                         slave->jiffies = jiffies;
3103                         bond->current_arp_slave = slave;
3104                         break;
3105                 }
3106
3107                 /* if the link state is up at this point, we
3108                  * mark it down - this can happen if we have
3109                  * simultaneous link failures and
3110                  * reselect_active_interface doesn't make this
3111                  * one the current slave so it is still marked
3112                  * up when it is actually down
3113                  */
3114                 if (slave->link == BOND_LINK_UP) {
3115                         slave->link = BOND_LINK_DOWN;
3116                         if (slave->link_failure_count < UINT_MAX)
3117                                 slave->link_failure_count++;
3118
3119                         bond_set_slave_inactive_flags(slave);
3120
3121                         printk(KERN_INFO DRV_NAME
3122                                ": %s: backup interface %s is now down.\n",
3123                                bond->dev->name, slave->dev->name);
3124                 }
3125         }
3126 }
3127
3128 void bond_activebackup_arp_mon(struct work_struct *work)
3129 {
3130         struct bonding *bond = container_of(work, struct bonding,
3131                                             arp_work.work);
3132         int delta_in_ticks;
3133
3134         read_lock(&bond->lock);
3135
3136         if (bond->kill_timers)
3137                 goto out;
3138
3139         delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
3140
3141         if (bond->slave_cnt == 0)
3142                 goto re_arm;
3143
3144         if (bond->send_grat_arp) {
3145                 read_lock(&bond->curr_slave_lock);
3146                 bond_send_gratuitous_arp(bond);
3147                 read_unlock(&bond->curr_slave_lock);
3148         }
3149
3150         if (bond_ab_arp_inspect(bond, delta_in_ticks)) {
3151                 read_unlock(&bond->lock);
3152                 rtnl_lock();
3153                 read_lock(&bond->lock);
3154
3155                 bond_ab_arp_commit(bond, delta_in_ticks);
3156
3157                 read_unlock(&bond->lock);
3158                 rtnl_unlock();
3159                 read_lock(&bond->lock);
3160         }
3161
3162         bond_ab_arp_probe(bond);
3163
3164 re_arm:
3165         if (bond->params.arp_interval) {
3166                 queue_delayed_work(bond->wq, &bond->arp_work, delta_in_ticks);
3167         }
3168 out:
3169         read_unlock(&bond->lock);
3170 }
3171
3172 /*------------------------------ proc/seq_file-------------------------------*/
3173
3174 #ifdef CONFIG_PROC_FS
3175
3176 static void *bond_info_seq_start(struct seq_file *seq, loff_t *pos)
3177 {
3178         struct bonding *bond = seq->private;
3179         loff_t off = 0;
3180         struct slave *slave;
3181         int i;
3182
3183         /* make sure the bond won't be taken away */
3184         read_lock(&dev_base_lock);
3185         read_lock(&bond->lock);
3186
3187         if (*pos == 0) {
3188                 return SEQ_START_TOKEN;
3189         }
3190
3191         bond_for_each_slave(bond, slave, i) {
3192                 if (++off == *pos) {
3193                         return slave;
3194                 }
3195         }
3196
3197         return NULL;
3198 }
3199
3200 static void *bond_info_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3201 {
3202         struct bonding *bond = seq->private;
3203         struct slave *slave = v;
3204
3205         ++*pos;
3206         if (v == SEQ_START_TOKEN) {
3207                 return bond->first_slave;
3208         }
3209
3210         slave = slave->next;
3211
3212         return (slave == bond->first_slave) ? NULL : slave;
3213 }
3214
3215 static void bond_info_seq_stop(struct seq_file *seq, void *v)
3216 {
3217         struct bonding *bond = seq->private;
3218
3219         read_unlock(&bond->lock);
3220         read_unlock(&dev_base_lock);
3221 }
3222
3223 static void bond_info_show_master(struct seq_file *seq)
3224 {
3225         struct bonding *bond = seq->private;
3226         struct slave *curr;
3227         int i;
3228         u32 target;
3229
3230         read_lock(&bond->curr_slave_lock);
3231         curr = bond->curr_active_slave;
3232         read_unlock(&bond->curr_slave_lock);
3233
3234         seq_printf(seq, "Bonding Mode: %s",
3235                    bond_mode_name(bond->params.mode));
3236
3237         if (bond->params.mode == BOND_MODE_ACTIVEBACKUP &&
3238             bond->params.fail_over_mac)
3239                 seq_printf(seq, " (fail_over_mac %s)",
3240                    fail_over_mac_tbl[bond->params.fail_over_mac].modename);
3241
3242         seq_printf(seq, "\n");
3243
3244         if (bond->params.mode == BOND_MODE_XOR ||
3245                 bond->params.mode == BOND_MODE_8023AD) {
3246                 seq_printf(seq, "Transmit Hash Policy: %s (%d)\n",
3247                         xmit_hashtype_tbl[bond->params.xmit_policy].modename,
3248                         bond->params.xmit_policy);
3249         }
3250
3251         if (USES_PRIMARY(bond->params.mode)) {
3252                 seq_printf(seq, "Primary Slave: %s\n",
3253                            (bond->primary_slave) ?
3254                            bond->primary_slave->dev->name : "None");
3255
3256                 seq_printf(seq, "Currently Active Slave: %s\n",
3257                            (curr) ? curr->dev->name : "None");
3258         }
3259
3260         seq_printf(seq, "MII Status: %s\n", netif_carrier_ok(bond->dev) ?
3261                    "up" : "down");
3262         seq_printf(seq, "MII Polling Interval (ms): %d\n", bond->params.miimon);
3263         seq_printf(seq, "Up Delay (ms): %d\n",
3264                    bond->params.updelay * bond->params.miimon);
3265         seq_printf(seq, "Down Delay (ms): %d\n",
3266                    bond->params.downdelay * bond->params.miimon);
3267
3268
3269         /* ARP information */
3270         if(bond->params.arp_interval > 0) {
3271                 int printed=0;
3272                 seq_printf(seq, "ARP Polling Interval (ms): %d\n",
3273                                 bond->params.arp_interval);
3274
3275                 seq_printf(seq, "ARP IP target/s (n.n.n.n form):");
3276
3277                 for(i = 0; (i < BOND_MAX_ARP_TARGETS) ;i++) {
3278                         if (!bond->params.arp_targets[i])
3279                                 continue;
3280                         if (printed)
3281                                 seq_printf(seq, ",");
3282                         target = ntohl(bond->params.arp_targets[i]);
3283                         seq_printf(seq, " %d.%d.%d.%d", HIPQUAD(target));
3284                         printed = 1;
3285                 }
3286                 seq_printf(seq, "\n");
3287         }
3288
3289         if (bond->params.mode == BOND_MODE_8023AD) {
3290                 struct ad_info ad_info;
3291                 DECLARE_MAC_BUF(mac);
3292
3293                 seq_puts(seq, "\n802.3ad info\n");
3294                 seq_printf(seq, "LACP rate: %s\n",
3295                            (bond->params.lacp_fast) ? "fast" : "slow");
3296
3297                 if (bond_3ad_get_active_agg_info(bond, &ad_info)) {
3298                         seq_printf(seq, "bond %s has no active aggregator\n",
3299                                    bond->dev->name);
3300                 } else {
3301                         seq_printf(seq, "Active Aggregator Info:\n");
3302
3303                         seq_printf(seq, "\tAggregator ID: %d\n",
3304                                    ad_info.aggregator_id);
3305                         seq_printf(seq, "\tNumber of ports: %d\n",
3306                                    ad_info.ports);
3307                         seq_printf(seq, "\tActor Key: %d\n",
3308                                    ad_info.actor_key);
3309                         seq_printf(seq, "\tPartner Key: %d\n",
3310                                    ad_info.partner_key);
3311                         seq_printf(seq, "\tPartner Mac Address: %s\n",
3312                                    print_mac(mac, ad_info.partner_system));
3313                 }
3314         }
3315 }
3316
3317 static void bond_info_show_slave(struct seq_file *seq, const struct slave *slave)
3318 {
3319         struct bonding *bond = seq->private;
3320         DECLARE_MAC_BUF(mac);
3321
3322         seq_printf(seq, "\nSlave Interface: %s\n", slave->dev->name);
3323         seq_printf(seq, "MII Status: %s\n",
3324                    (slave->link == BOND_LINK_UP) ?  "up" : "down");
3325         seq_printf(seq, "Link Failure Count: %u\n",
3326                    slave->link_failure_count);
3327
3328         seq_printf(seq,
3329                    "Permanent HW addr: %s\n",
3330                    print_mac(mac, slave->perm_hwaddr));
3331
3332         if (bond->params.mode == BOND_MODE_8023AD) {
3333                 const struct aggregator *agg
3334                         = SLAVE_AD_INFO(slave).port.aggregator;
3335
3336                 if (agg) {
3337                         seq_printf(seq, "Aggregator ID: %d\n",
3338                                    agg->aggregator_identifier);
3339                 } else {
3340                         seq_puts(seq, "Aggregator ID: N/A\n");
3341                 }
3342         }
3343 }
3344
3345 static int bond_info_seq_show(struct seq_file *seq, void *v)
3346 {
3347         if (v == SEQ_START_TOKEN) {
3348                 seq_printf(seq, "%s\n", version);
3349                 bond_info_show_master(seq);
3350         } else {
3351                 bond_info_show_slave(seq, v);
3352         }
3353
3354         return 0;
3355 }
3356
3357 static struct seq_operations bond_info_seq_ops = {
3358         .start = bond_info_seq_start,
3359         .next  = bond_info_seq_next,
3360         .stop  = bond_info_seq_stop,
3361         .show  = bond_info_seq_show,
3362 };
3363
3364 static int bond_info_open(struct inode *inode, struct file *file)
3365 {
3366         struct seq_file *seq;
3367         struct proc_dir_entry *proc;
3368         int res;
3369
3370         res = seq_open(file, &bond_info_seq_ops);
3371         if (!res) {
3372                 /* recover the pointer buried in proc_dir_entry data */
3373                 seq = file->private_data;
3374                 proc = PDE(inode);
3375                 seq->private = proc->data;
3376         }
3377
3378         return res;
3379 }
3380
3381 static const struct file_operations bond_info_fops = {
3382         .owner   = THIS_MODULE,
3383         .open    = bond_info_open,
3384         .read    = seq_read,
3385         .llseek  = seq_lseek,
3386         .release = seq_release,
3387 };
3388
3389 static int bond_create_proc_entry(struct bonding *bond)
3390 {
3391         struct net_device *bond_dev = bond->dev;
3392
3393         if (bond_proc_dir) {
3394                 bond->proc_entry = proc_create_data(bond_dev->name,
3395                                                     S_IRUGO, bond_proc_dir,
3396                                                     &bond_info_fops, bond);
3397                 if (bond->proc_entry == NULL) {
3398                         printk(KERN_WARNING DRV_NAME
3399                                ": Warning: Cannot create /proc/net/%s/%s\n",
3400                                DRV_NAME, bond_dev->name);
3401                 } else {
3402                         memcpy(bond->proc_file_name, bond_dev->name, IFNAMSIZ);
3403                 }
3404         }
3405
3406         return 0;
3407 }
3408
3409 static void bond_remove_proc_entry(struct bonding *bond)
3410 {
3411         if (bond_proc_dir && bond->proc_entry) {
3412                 remove_proc_entry(bond->proc_file_name, bond_proc_dir);
3413                 memset(bond->proc_file_name, 0, IFNAMSIZ);
3414                 bond->proc_entry = NULL;
3415         }
3416 }
3417
3418 /* Create the bonding directory under /proc/net, if doesn't exist yet.
3419  * Caller must hold rtnl_lock.
3420  */
3421 static void bond_create_proc_dir(void)
3422 {
3423         int len = strlen(DRV_NAME);
3424
3425         for (bond_proc_dir = init_net.proc_net->subdir; bond_proc_dir;
3426              bond_proc_dir = bond_proc_dir->next) {
3427                 if ((bond_proc_dir->namelen == len) &&
3428                     !memcmp(bond_proc_dir->name, DRV_NAME, len)) {
3429                         break;
3430                 }
3431         }
3432
3433         if (!bond_proc_dir) {
3434                 bond_proc_dir = proc_mkdir(DRV_NAME, init_net.proc_net);
3435                 if (bond_proc_dir) {
3436                         bond_proc_dir->owner = THIS_MODULE;
3437                 } else {
3438                         printk(KERN_WARNING DRV_NAME
3439                                 ": Warning: cannot create /proc/net/%s\n",
3440                                 DRV_NAME);
3441                 }
3442         }
3443 }
3444
3445 /* Destroy the bonding directory under /proc/net, if empty.
3446  * Caller must hold rtnl_lock.
3447  */
3448 static void bond_destroy_proc_dir(void)
3449 {
3450         struct proc_dir_entry *de;
3451
3452         if (!bond_proc_dir) {
3453                 return;
3454         }
3455
3456         /* verify that the /proc dir is empty */
3457         for (de = bond_proc_dir->subdir; de; de = de->next) {
3458                 /* ignore . and .. */
3459                 if (*(de->name) != '.') {
3460                         break;
3461                 }
3462         }
3463
3464         if (de) {
3465                 if (bond_proc_dir->owner == THIS_MODULE) {
3466                         bond_proc_dir->owner = NULL;
3467                 }
3468         } else {
3469                 remove_proc_entry(DRV_NAME, init_net.proc_net);
3470                 bond_proc_dir = NULL;
3471         }
3472 }
3473 #endif /* CONFIG_PROC_FS */
3474
3475 /*-------------------------- netdev event handling --------------------------*/
3476
3477 /*
3478  * Change device name
3479  */
3480 static int bond_event_changename(struct bonding *bond)
3481 {
3482 #ifdef CONFIG_PROC_FS
3483         bond_remove_proc_entry(bond);
3484         bond_create_proc_entry(bond);
3485 #endif
3486         down_write(&(bonding_rwsem));
3487         bond_destroy_sysfs_entry(bond);
3488         bond_create_sysfs_entry(bond);
3489         up_write(&(bonding_rwsem));
3490         return NOTIFY_DONE;
3491 }
3492
3493 static int bond_master_netdev_event(unsigned long event, struct net_device *bond_dev)
3494 {
3495         struct bonding *event_bond = bond_dev->priv;
3496
3497         switch (event) {
3498         case NETDEV_CHANGENAME:
3499                 return bond_event_changename(event_bond);
3500         case NETDEV_UNREGISTER:
3501                 bond_release_all(event_bond->dev);
3502                 break;
3503         default:
3504                 break;
3505         }
3506
3507         return NOTIFY_DONE;
3508 }
3509
3510 static int bond_slave_netdev_event(unsigned long event, struct net_device *slave_dev)
3511 {
3512         struct net_device *bond_dev = slave_dev->master;
3513         struct bonding *bond = bond_dev->priv;
3514
3515         switch (event) {
3516         case NETDEV_UNREGISTER:
3517                 if (bond_dev) {
3518                         if (bond->setup_by_slave)
3519                                 bond_release_and_destroy(bond_dev, slave_dev);
3520                         else
3521                                 bond_release(bond_dev, slave_dev);
3522                 }
3523                 break;
3524         case NETDEV_CHANGE:
3525                 /*
3526                  * TODO: is this what we get if somebody
3527                  * sets up a hierarchical bond, then rmmod's
3528                  * one of the slave bonding devices?
3529                  */
3530                 break;
3531         case NETDEV_DOWN:
3532                 /*
3533                  * ... Or is it this?
3534                  */
3535                 break;
3536         case NETDEV_CHANGEMTU:
3537                 /*
3538                  * TODO: Should slaves be allowed to
3539                  * independently alter their MTU?  For
3540                  * an active-backup bond, slaves need
3541                  * not be the same type of device, so
3542                  * MTUs may vary.  For other modes,
3543                  * slaves arguably should have the
3544                  * same MTUs. To do this, we'd need to
3545                  * take over the slave's change_mtu
3546                  * function for the duration of their
3547                  * servitude.
3548                  */
3549                 break;
3550         case NETDEV_CHANGENAME:
3551                 /*
3552                  * TODO: handle changing the primary's name
3553                  */
3554                 break;
3555         case NETDEV_FEAT_CHANGE:
3556                 bond_compute_features(bond);
3557                 break;
3558         default:
3559                 break;
3560         }
3561
3562         return NOTIFY_DONE;
3563 }
3564
3565 /*
3566  * bond_netdev_event: handle netdev notifier chain events.
3567  *
3568  * This function receives events for the netdev chain.  The caller (an
3569  * ioctl handler calling blocking_notifier_call_chain) holds the necessary
3570  * locks for us to safely manipulate the slave devices (RTNL lock,
3571  * dev_probe_lock).
3572  */
3573 static int bond_netdev_event(struct notifier_block *this, unsigned long event, void *ptr)
3574 {
3575         struct net_device *event_dev = (struct net_device *)ptr;
3576
3577         if (dev_net(event_dev) != &init_net)
3578                 return NOTIFY_DONE;
3579
3580         dprintk("event_dev: %s, event: %lx\n",
3581                 (event_dev ? event_dev->name : "None"),
3582                 event);
3583
3584         if (!(event_dev->priv_flags & IFF_BONDING))
3585                 return NOTIFY_DONE;
3586
3587         if (event_dev->flags & IFF_MASTER) {
3588                 dprintk("IFF_MASTER\n");
3589                 return bond_master_netdev_event(event, event_dev);
3590         }
3591
3592         if (event_dev->flags & IFF_SLAVE) {
3593                 dprintk("IFF_SLAVE\n");
3594                 return bond_slave_netdev_event(event, event_dev);
3595         }
3596
3597         return NOTIFY_DONE;
3598 }
3599
3600 /*
3601  * bond_inetaddr_event: handle inetaddr notifier chain events.
3602  *
3603  * We keep track of device IPs primarily to use as source addresses in
3604  * ARP monitor probes (rather than spewing out broadcasts all the time).
3605  *
3606  * We track one IP for the main device (if it has one), plus one per VLAN.
3607  */
3608 static int bond_inetaddr_event(struct notifier_block *this, unsigned long event, void *ptr)
3609 {
3610         struct in_ifaddr *ifa = ptr;
3611         struct net_device *vlan_dev, *event_dev = ifa->ifa_dev->dev;
3612         struct bonding *bond;
3613         struct vlan_entry *vlan;
3614
3615         if (dev_net(ifa->ifa_dev->dev) != &init_net)
3616                 return NOTIFY_DONE;
3617
3618         list_for_each_entry(bond, &bond_dev_list, bond_list) {
3619                 if (bond->dev == event_dev) {
3620                         switch (event) {
3621                         case NETDEV_UP:
3622                                 bond->master_ip = ifa->ifa_local;
3623                                 return NOTIFY_OK;
3624                         case NETDEV_DOWN:
3625                                 bond->master_ip = bond_glean_dev_ip(bond->dev);
3626                                 return NOTIFY_OK;
3627                         default:
3628                                 return NOTIFY_DONE;
3629                         }
3630                 }
3631
3632                 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
3633                         vlan_dev = vlan_group_get_device(bond->vlgrp, vlan->vlan_id);
3634                         if (vlan_dev == event_dev) {
3635                                 switch (event) {
3636                                 case NETDEV_UP:
3637                                         vlan->vlan_ip = ifa->ifa_local;
3638                                         return NOTIFY_OK;
3639                                 case NETDEV_DOWN:
3640                                         vlan->vlan_ip =
3641                                                 bond_glean_dev_ip(vlan_dev);
3642                                         return NOTIFY_OK;
3643                                 default:
3644                                         return NOTIFY_DONE;
3645                                 }
3646                         }
3647                 }
3648         }
3649         return NOTIFY_DONE;
3650 }
3651
3652 static struct notifier_block bond_netdev_notifier = {
3653         .notifier_call = bond_netdev_event,
3654 };
3655
3656 static struct notifier_block bond_inetaddr_notifier = {
3657         .notifier_call = bond_inetaddr_event,
3658 };
3659
3660 /*-------------------------- Packet type handling ---------------------------*/
3661
3662 /* register to receive lacpdus on a bond */
3663 static void bond_register_lacpdu(struct bonding *bond)
3664 {
3665         struct packet_type *pk_type = &(BOND_AD_INFO(bond).ad_pkt_type);
3666
3667         /* initialize packet type */
3668         pk_type->type = PKT_TYPE_LACPDU;
3669         pk_type->dev = bond->dev;
3670         pk_type->func = bond_3ad_lacpdu_recv;
3671
3672         dev_add_pack(pk_type);
3673 }
3674
3675 /* unregister to receive lacpdus on a bond */
3676 static void bond_unregister_lacpdu(struct bonding *bond)
3677 {
3678         dev_remove_pack(&(BOND_AD_INFO(bond).ad_pkt_type));
3679 }
3680
3681 void bond_register_arp(struct bonding *bond)
3682 {
3683         struct packet_type *pt = &bond->arp_mon_pt;
3684
3685         if (pt->type)
3686                 return;
3687
3688         pt->type = htons(ETH_P_ARP);
3689         pt->dev = bond->dev;
3690         pt->func = bond_arp_rcv;
3691         dev_add_pack(pt);
3692 }
3693
3694 void bond_unregister_arp(struct bonding *bond)
3695 {
3696         struct packet_type *pt = &bond->arp_mon_pt;
3697
3698         dev_remove_pack(pt);
3699         pt->type = 0;
3700 }
3701
3702 /*---------------------------- Hashing Policies -----------------------------*/
3703
3704 /*
3705  * Hash for the output device based upon layer 2 and layer 3 data. If
3706  * the packet is not IP mimic bond_xmit_hash_policy_l2()
3707  */
3708 static int bond_xmit_hash_policy_l23(struct sk_buff *skb,
3709                                      struct net_device *bond_dev, int count)
3710 {
3711         struct ethhdr *data = (struct ethhdr *)skb->data;
3712         struct iphdr *iph = ip_hdr(skb);
3713
3714         if (skb->protocol == htons(ETH_P_IP)) {
3715                 return ((ntohl(iph->saddr ^ iph->daddr) & 0xffff) ^
3716                         (data->h_dest[5] ^ bond_dev->dev_addr[5])) % count;
3717         }
3718
3719         return (data->h_dest[5] ^ bond_dev->dev_addr[5]) % count;
3720 }
3721
3722 /*
3723  * Hash for the output device based upon layer 3 and layer 4 data. If
3724  * the packet is a frag or not TCP or UDP, just use layer 3 data.  If it is
3725  * altogether not IP, mimic bond_xmit_hash_policy_l2()
3726  */
3727 static int bond_xmit_hash_policy_l34(struct sk_buff *skb,
3728                                     struct net_device *bond_dev, int count)
3729 {
3730         struct ethhdr *data = (struct ethhdr *)skb->data;
3731         struct iphdr *iph = ip_hdr(skb);
3732         __be16 *layer4hdr = (__be16 *)((u32 *)iph + iph->ihl);
3733         int layer4_xor = 0;
3734
3735         if (skb->protocol == htons(ETH_P_IP)) {
3736                 if (!(iph->frag_off & htons(IP_MF|IP_OFFSET)) &&
3737                     (iph->protocol == IPPROTO_TCP ||
3738                      iph->protocol == IPPROTO_UDP)) {
3739                         layer4_xor = ntohs((*layer4hdr ^ *(layer4hdr + 1)));
3740                 }
3741                 return (layer4_xor ^
3742                         ((ntohl(iph->saddr ^ iph->daddr)) & 0xffff)) % count;
3743
3744         }
3745
3746         return (data->h_dest[5] ^ bond_dev->dev_addr[5]) % count;
3747 }
3748
3749 /*
3750  * Hash for the output device based upon layer 2 data
3751  */
3752 static int bond_xmit_hash_policy_l2(struct sk_buff *skb,
3753                                    struct net_device *bond_dev, int count)
3754 {
3755         struct ethhdr *data = (struct ethhdr *)skb->data;
3756
3757         return (data->h_dest[5] ^ bond_dev->dev_addr[5]) % count;
3758 }
3759
3760 /*-------------------------- Device entry points ----------------------------*/
3761
3762 static int bond_open(struct net_device *bond_dev)
3763 {
3764         struct bonding *bond = bond_dev->priv;
3765
3766         bond->kill_timers = 0;
3767
3768         if ((bond->params.mode == BOND_MODE_TLB) ||
3769             (bond->params.mode == BOND_MODE_ALB)) {
3770                 /* bond_alb_initialize must be called before the timer
3771                  * is started.
3772                  */
3773                 if (bond_alb_initialize(bond, (bond->params.mode == BOND_MODE_ALB))) {
3774                         /* something went wrong - fail the open operation */
3775                         return -1;
3776                 }
3777
3778                 INIT_DELAYED_WORK(&bond->alb_work, bond_alb_monitor);
3779                 queue_delayed_work(bond->wq, &bond->alb_work, 0);
3780         }
3781
3782         if (bond->params.miimon) {  /* link check interval, in milliseconds. */
3783                 INIT_DELAYED_WORK(&bond->mii_work, bond_mii_monitor);
3784                 queue_delayed_work(bond->wq, &bond->mii_work, 0);
3785         }
3786
3787         if (bond->params.arp_interval) {  /* arp interval, in milliseconds. */
3788                 if (bond->params.mode == BOND_MODE_ACTIVEBACKUP)
3789                         INIT_DELAYED_WORK(&bond->arp_work,
3790                                           bond_activebackup_arp_mon);
3791                 else
3792                         INIT_DELAYED_WORK(&bond->arp_work,
3793                                           bond_loadbalance_arp_mon);
3794
3795                 queue_delayed_work(bond->wq, &bond->arp_work, 0);
3796                 if (bond->params.arp_validate)
3797                         bond_register_arp(bond);
3798         }
3799
3800         if (bond->params.mode == BOND_MODE_8023AD) {
3801                 INIT_DELAYED_WORK(&bond->ad_work, bond_3ad_state_machine_handler);
3802                 queue_delayed_work(bond->wq, &bond->ad_work, 0);
3803                 /* register to receive LACPDUs */
3804                 bond_register_lacpdu(bond);
3805         }
3806
3807         return 0;
3808 }
3809
3810 static int bond_close(struct net_device *bond_dev)
3811 {
3812         struct bonding *bond = bond_dev->priv;
3813
3814         if (bond->params.mode == BOND_MODE_8023AD) {
3815                 /* Unregister the receive of LACPDUs */
3816                 bond_unregister_lacpdu(bond);
3817         }
3818
3819         if (bond->params.arp_validate)
3820                 bond_unregister_arp(bond);
3821
3822         write_lock_bh(&bond->lock);
3823
3824         bond->send_grat_arp = 0;
3825
3826         /* signal timers not to re-arm */
3827         bond->kill_timers = 1;
3828
3829         write_unlock_bh(&bond->lock);
3830
3831         if (bond->params.miimon) {  /* link check interval, in milliseconds. */
3832                 cancel_delayed_work(&bond->mii_work);
3833         }
3834
3835         if (bond->params.arp_interval) {  /* arp interval, in milliseconds. */
3836                 cancel_delayed_work(&bond->arp_work);
3837         }
3838
3839         switch (bond->params.mode) {
3840         case BOND_MODE_8023AD:
3841                 cancel_delayed_work(&bond->ad_work);
3842                 break;
3843         case BOND_MODE_TLB:
3844         case BOND_MODE_ALB:
3845                 cancel_delayed_work(&bond->alb_work);
3846                 break;
3847         default:
3848                 break;
3849         }
3850
3851
3852         if ((bond->params.mode == BOND_MODE_TLB) ||
3853             (bond->params.mode == BOND_MODE_ALB)) {
3854                 /* Must be called only after all
3855                  * slaves have been released
3856                  */
3857                 bond_alb_deinitialize(bond);
3858         }
3859
3860         return 0;
3861 }
3862
3863 static struct net_device_stats *bond_get_stats(struct net_device *bond_dev)
3864 {
3865         struct bonding *bond = bond_dev->priv;
3866         struct net_device_stats *stats = &(bond->stats), *sstats;
3867         struct net_device_stats local_stats;
3868         struct slave *slave;
3869         int i;
3870
3871         memset(&local_stats, 0, sizeof(struct net_device_stats));
3872
3873         read_lock_bh(&bond->lock);
3874
3875         bond_for_each_slave(bond, slave, i) {
3876                 sstats = slave->dev->get_stats(slave->dev);
3877                 local_stats.rx_packets += sstats->rx_packets;
3878                 local_stats.rx_bytes += sstats->rx_bytes;
3879                 local_stats.rx_errors += sstats->rx_errors;
3880                 local_stats.rx_dropped += sstats->rx_dropped;
3881
3882                 local_stats.tx_packets += sstats->tx_packets;
3883                 local_stats.tx_bytes += sstats->tx_bytes;
3884                 local_stats.tx_errors += sstats->tx_errors;
3885                 local_stats.tx_dropped += sstats->tx_dropped;
3886
3887                 local_stats.multicast += sstats->multicast;
3888                 local_stats.collisions += sstats->collisions;
3889
3890                 local_stats.rx_length_errors += sstats->rx_length_errors;
3891                 local_stats.rx_over_errors += sstats->rx_over_errors;
3892                 local_stats.rx_crc_errors += sstats->rx_crc_errors;
3893                 local_stats.rx_frame_errors += sstats->rx_frame_errors;
3894                 local_stats.rx_fifo_errors += sstats->rx_fifo_errors;
3895                 local_stats.rx_missed_errors += sstats->rx_missed_errors;
3896
3897                 local_stats.tx_aborted_errors += sstats->tx_aborted_errors;
3898                 local_stats.tx_carrier_errors += sstats->tx_carrier_errors;
3899                 local_stats.tx_fifo_errors += sstats->tx_fifo_errors;
3900                 local_stats.tx_heartbeat_errors += sstats->tx_heartbeat_errors;
3901                 local_stats.tx_window_errors += sstats->tx_window_errors;
3902         }
3903
3904         memcpy(stats, &local_stats, sizeof(struct net_device_stats));
3905
3906         read_unlock_bh(&bond->lock);
3907
3908         return stats;
3909 }
3910
3911 static int bond_do_ioctl(struct net_device *bond_dev, struct ifreq *ifr, int cmd)
3912 {
3913         struct net_device *slave_dev = NULL;
3914         struct ifbond k_binfo;
3915         struct ifbond __user *u_binfo = NULL;
3916         struct ifslave k_sinfo;
3917         struct ifslave __user *u_sinfo = NULL;
3918         struct mii_ioctl_data *mii = NULL;
3919         int res = 0;
3920
3921         dprintk("bond_ioctl: master=%s, cmd=%d\n",
3922                 bond_dev->name, cmd);
3923
3924         switch (cmd) {
3925         case SIOCGMIIPHY:
3926                 mii = if_mii(ifr);
3927                 if (!mii) {
3928                         return -EINVAL;
3929                 }
3930                 mii->phy_id = 0;
3931                 /* Fall Through */
3932         case SIOCGMIIREG:
3933                 /*
3934                  * We do this again just in case we were called by SIOCGMIIREG
3935                  * instead of SIOCGMIIPHY.
3936                  */
3937                 mii = if_mii(ifr);
3938                 if (!mii) {
3939                         return -EINVAL;
3940                 }
3941
3942                 if (mii->reg_num == 1) {
3943                         struct bonding *bond = bond_dev->priv;
3944                         mii->val_out = 0;
3945                         read_lock(&bond->lock);
3946                         read_lock(&bond->curr_slave_lock);
3947                         if (netif_carrier_ok(bond->dev)) {
3948                                 mii->val_out = BMSR_LSTATUS;
3949                         }
3950                         read_unlock(&bond->curr_slave_lock);
3951                         read_unlock(&bond->lock);
3952                 }
3953
3954                 return 0;
3955         case BOND_INFO_QUERY_OLD:
3956         case SIOCBONDINFOQUERY:
3957                 u_binfo = (struct ifbond __user *)ifr->ifr_data;
3958
3959                 if (copy_from_user(&k_binfo, u_binfo, sizeof(ifbond))) {
3960                         return -EFAULT;
3961                 }
3962
3963                 res = bond_info_query(bond_dev, &k_binfo);
3964                 if (res == 0) {
3965                         if (copy_to_user(u_binfo, &k_binfo, sizeof(ifbond))) {
3966                                 return -EFAULT;
3967                         }
3968                 }
3969
3970                 return res;
3971         case BOND_SLAVE_INFO_QUERY_OLD:
3972         case SIOCBONDSLAVEINFOQUERY:
3973                 u_sinfo = (struct ifslave __user *)ifr->ifr_data;
3974
3975                 if (copy_from_user(&k_sinfo, u_sinfo, sizeof(ifslave))) {
3976                         return -EFAULT;
3977                 }
3978
3979                 res = bond_slave_info_query(bond_dev, &k_sinfo);
3980                 if (res == 0) {
3981                         if (copy_to_user(u_sinfo, &k_sinfo, sizeof(ifslave))) {
3982                                 return -EFAULT;
3983                         }
3984                 }
3985
3986                 return res;
3987         default:
3988                 /* Go on */
3989                 break;
3990         }
3991
3992         if (!capable(CAP_NET_ADMIN)) {
3993                 return -EPERM;
3994         }
3995
3996         down_write(&(bonding_rwsem));
3997         slave_dev = dev_get_by_name(&init_net, ifr->ifr_slave);
3998
3999         dprintk("slave_dev=%p: \n", slave_dev);
4000
4001         if (!slave_dev) {
4002                 res = -ENODEV;
4003         } else {
4004                 dprintk("slave_dev->name=%s: \n", slave_dev->name);
4005                 switch (cmd) {
4006                 case BOND_ENSLAVE_OLD:
4007                 case SIOCBONDENSLAVE:
4008                         res = bond_enslave(bond_dev, slave_dev);
4009                         break;
4010                 case BOND_RELEASE_OLD:
4011                 case SIOCBONDRELEASE:
4012                         res = bond_release(bond_dev, slave_dev);
4013                         break;
4014                 case BOND_SETHWADDR_OLD:
4015                 case SIOCBONDSETHWADDR:
4016                         res = bond_sethwaddr(bond_dev, slave_dev);
4017                         break;
4018                 case BOND_CHANGE_ACTIVE_OLD:
4019                 case SIOCBONDCHANGEACTIVE:
4020                         res = bond_ioctl_change_active(bond_dev, slave_dev);
4021                         break;
4022                 default:
4023                         res = -EOPNOTSUPP;
4024                 }
4025
4026                 dev_put(slave_dev);
4027         }
4028
4029         up_write(&(bonding_rwsem));
4030         return res;
4031 }
4032
4033 static void bond_set_multicast_list(struct net_device *bond_dev)
4034 {
4035         struct bonding *bond = bond_dev->priv;
4036         struct dev_mc_list *dmi;
4037
4038         /*
4039          * Do promisc before checking multicast_mode
4040          */
4041         if ((bond_dev->flags & IFF_PROMISC) && !(bond->flags & IFF_PROMISC)) {
4042                 /*
4043                  * FIXME: Need to handle the error when one of the multi-slaves
4044                  * encounters error.
4045                  */
4046                 bond_set_promiscuity(bond, 1);
4047         }
4048
4049         if (!(bond_dev->flags & IFF_PROMISC) && (bond->flags & IFF_PROMISC)) {
4050                 bond_set_promiscuity(bond, -1);
4051         }
4052
4053         /* set allmulti flag to slaves */
4054         if ((bond_dev->flags & IFF_ALLMULTI) && !(bond->flags & IFF_ALLMULTI)) {
4055                 /*
4056                  * FIXME: Need to handle the error when one of the multi-slaves
4057                  * encounters error.
4058                  */
4059                 bond_set_allmulti(bond, 1);
4060         }
4061
4062         if (!(bond_dev->flags & IFF_ALLMULTI) && (bond->flags & IFF_ALLMULTI)) {
4063                 bond_set_allmulti(bond, -1);
4064         }
4065
4066         read_lock(&bond->lock);
4067
4068         bond->flags = bond_dev->flags;
4069
4070         /* looking for addresses to add to slaves' mc list */
4071         for (dmi = bond_dev->mc_list; dmi; dmi = dmi->next) {
4072                 if (!bond_mc_list_find_dmi(dmi, bond->mc_list)) {
4073                         bond_mc_add(bond, dmi->dmi_addr, dmi->dmi_addrlen);
4074                 }
4075         }
4076
4077         /* looking for addresses to delete from slaves' list */
4078         for (dmi = bond->mc_list; dmi; dmi = dmi->next) {
4079                 if (!bond_mc_list_find_dmi(dmi, bond_dev->mc_list)) {
4080                         bond_mc_delete(bond, dmi->dmi_addr, dmi->dmi_addrlen);
4081                 }
4082         }
4083
4084         /* save master's multicast list */
4085         bond_mc_list_destroy(bond);
4086         bond_mc_list_copy(bond_dev->mc_list, bond, GFP_ATOMIC);
4087
4088         read_unlock(&bond->lock);
4089 }
4090
4091 /*
4092  * Change the MTU of all of a master's slaves to match the master
4093  */
4094 static int bond_change_mtu(struct net_device *bond_dev, int new_mtu)
4095 {
4096         struct bonding *bond = bond_dev->priv;
4097         struct slave *slave, *stop_at;
4098         int res = 0;
4099         int i;
4100
4101         dprintk("bond=%p, name=%s, new_mtu=%d\n", bond,
4102                 (bond_dev ? bond_dev->name : "None"), new_mtu);
4103
4104         /* Can't hold bond->lock with bh disabled here since
4105          * some base drivers panic. On the other hand we can't
4106          * hold bond->lock without bh disabled because we'll
4107          * deadlock. The only solution is to rely on the fact
4108          * that we're under rtnl_lock here, and the slaves
4109          * list won't change. This doesn't solve the problem
4110          * of setting the slave's MTU while it is
4111          * transmitting, but the assumption is that the base
4112          * driver can handle that.
4113          *
4114          * TODO: figure out a way to safely iterate the slaves
4115          * list, but without holding a lock around the actual
4116          * call to the base driver.
4117          */
4118
4119         bond_for_each_slave(bond, slave, i) {
4120                 dprintk("s %p s->p %p c_m %p\n", slave,
4121                         slave->prev, slave->dev->change_mtu);
4122
4123                 res = dev_set_mtu(slave->dev, new_mtu);
4124
4125                 if (res) {
4126                         /* If we failed to set the slave's mtu to the new value
4127                          * we must abort the operation even in ACTIVE_BACKUP
4128                          * mode, because if we allow the backup slaves to have
4129                          * different mtu values than the active slave we'll
4130                          * need to change their mtu when doing a failover. That
4131                          * means changing their mtu from timer context, which
4132                          * is probably not a good idea.
4133                          */
4134                         dprintk("err %d %s\n", res, slave->dev->name);
4135                         goto unwind;
4136                 }
4137         }
4138
4139         bond_dev->mtu = new_mtu;
4140
4141         return 0;
4142
4143 unwind:
4144         /* unwind from head to the slave that failed */
4145         stop_at = slave;
4146         bond_for_each_slave_from_to(bond, slave, i, bond->first_slave, stop_at) {
4147                 int tmp_res;
4148
4149                 tmp_res = dev_set_mtu(slave->dev, bond_dev->mtu);
4150                 if (tmp_res) {
4151                         dprintk("unwind err %d dev %s\n", tmp_res,
4152                                 slave->dev->name);
4153                 }
4154         }
4155
4156         return res;
4157 }
4158
4159 /*
4160  * Change HW address
4161  *
4162  * Note that many devices must be down to change the HW address, and
4163  * downing the master releases all slaves.  We can make bonds full of
4164  * bonding devices to test this, however.
4165  */
4166 static int bond_set_mac_address(struct net_device *bond_dev, void *addr)
4167 {
4168         struct bonding *bond = bond_dev->priv;
4169         struct sockaddr *sa = addr, tmp_sa;
4170         struct slave *slave, *stop_at;
4171         int res = 0;
4172         int i;
4173
4174         dprintk("bond=%p, name=%s\n", bond, (bond_dev ? bond_dev->name : "None"));
4175
4176         /*
4177          * If fail_over_mac is set to active, do nothing and return
4178          * success.  Returning an error causes ifenslave to fail.
4179          */
4180         if (bond->params.fail_over_mac == BOND_FOM_ACTIVE)
4181                 return 0;
4182
4183         if (!is_valid_ether_addr(sa->sa_data)) {
4184                 return -EADDRNOTAVAIL;
4185         }
4186
4187         /* Can't hold bond->lock with bh disabled here since
4188          * some base drivers panic. On the other hand we can't
4189          * hold bond->lock without bh disabled because we'll
4190          * deadlock. The only solution is to rely on the fact
4191          * that we're under rtnl_lock here, and the slaves
4192          * list won't change. This doesn't solve the problem
4193          * of setting the slave's hw address while it is
4194          * transmitting, but the assumption is that the base
4195          * driver can handle that.
4196          *
4197          * TODO: figure out a way to safely iterate the slaves
4198          * list, but without holding a lock around the actual
4199          * call to the base driver.
4200          */
4201
4202         bond_for_each_slave(bond, slave, i) {
4203                 dprintk("slave %p %s\n", slave, slave->dev->name);
4204
4205                 if (slave->dev->set_mac_address == NULL) {
4206                         res = -EOPNOTSUPP;
4207                         dprintk("EOPNOTSUPP %s\n", slave->dev->name);
4208                         goto unwind;
4209                 }
4210
4211                 res = dev_set_mac_address(slave->dev, addr);
4212                 if (res) {
4213                         /* TODO: consider downing the slave
4214                          * and retry ?
4215                          * User should expect communications
4216                          * breakage anyway until ARP finish
4217                          * updating, so...
4218                          */
4219                         dprintk("err %d %s\n", res, slave->dev->name);
4220                         goto unwind;
4221                 }
4222         }
4223
4224         /* success */
4225         memcpy(bond_dev->dev_addr, sa->sa_data, bond_dev->addr_len);
4226         return 0;
4227
4228 unwind:
4229         memcpy(tmp_sa.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
4230         tmp_sa.sa_family = bond_dev->type;
4231
4232         /* unwind from head to the slave that failed */
4233         stop_at = slave;
4234         bond_for_each_slave_from_to(bond, slave, i, bond->first_slave, stop_at) {
4235                 int tmp_res;
4236
4237                 tmp_res = dev_set_mac_address(slave->dev, &tmp_sa);
4238                 if (tmp_res) {
4239                         dprintk("unwind err %d dev %s\n", tmp_res,
4240                                 slave->dev->name);
4241                 }
4242         }
4243
4244         return res;
4245 }
4246
4247 static int bond_xmit_roundrobin(struct sk_buff *skb, struct net_device *bond_dev)
4248 {
4249         struct bonding *bond = bond_dev->priv;
4250         struct slave *slave, *start_at;
4251         int i, slave_no, res = 1;
4252
4253         read_lock(&bond->lock);
4254
4255         if (!BOND_IS_OK(bond)) {
4256                 goto out;
4257         }
4258
4259         /*
4260          * Concurrent TX may collide on rr_tx_counter; we accept that
4261          * as being rare enough not to justify using an atomic op here
4262          */
4263         slave_no = bond->rr_tx_counter++ % bond->slave_cnt;
4264
4265         bond_for_each_slave(bond, slave, i) {
4266                 slave_no--;
4267                 if (slave_no < 0) {
4268                         break;
4269                 }
4270         }
4271
4272         start_at = slave;
4273         bond_for_each_slave_from(bond, slave, i, start_at) {
4274                 if (IS_UP(slave->dev) &&
4275                     (slave->link == BOND_LINK_UP) &&
4276                     (slave->state == BOND_STATE_ACTIVE)) {
4277                         res = bond_dev_queue_xmit(bond, skb, slave->dev);
4278                         break;
4279                 }
4280         }
4281
4282 out:
4283         if (res) {
4284                 /* no suitable interface, frame not sent */
4285                 dev_kfree_skb(skb);
4286         }
4287         read_unlock(&bond->lock);
4288         return 0;
4289 }
4290
4291
4292 /*
4293  * in active-backup mode, we know that bond->curr_active_slave is always valid if
4294  * the bond has a usable interface.
4295  */
4296 static int bond_xmit_activebackup(struct sk_buff *skb, struct net_device *bond_dev)
4297 {
4298         struct bonding *bond = bond_dev->priv;
4299         int res = 1;
4300
4301         read_lock(&bond->lock);
4302         read_lock(&bond->curr_slave_lock);
4303
4304         if (!BOND_IS_OK(bond)) {
4305                 goto out;
4306         }
4307
4308         if (!bond->curr_active_slave)
4309                 goto out;
4310
4311         res = bond_dev_queue_xmit(bond, skb, bond->curr_active_slave->dev);
4312
4313 out:
4314         if (res) {
4315                 /* no suitable interface, frame not sent */
4316                 dev_kfree_skb(skb);
4317         }
4318         read_unlock(&bond->curr_slave_lock);
4319         read_unlock(&bond->lock);
4320         return 0;
4321 }
4322
4323 /*
4324  * In bond_xmit_xor() , we determine the output device by using a pre-
4325  * determined xmit_hash_policy(), If the selected device is not enabled,
4326  * find the next active slave.
4327  */
4328 static int bond_xmit_xor(struct sk_buff *skb, struct net_device *bond_dev)
4329 {
4330         struct bonding *bond = bond_dev->priv;
4331         struct slave *slave, *start_at;
4332         int slave_no;
4333         int i;
4334         int res = 1;
4335
4336         read_lock(&bond->lock);
4337
4338         if (!BOND_IS_OK(bond)) {
4339                 goto out;
4340         }
4341
4342         slave_no = bond->xmit_hash_policy(skb, bond_dev, bond->slave_cnt);
4343
4344         bond_for_each_slave(bond, slave, i) {
4345                 slave_no--;
4346                 if (slave_no < 0) {
4347                         break;
4348                 }
4349         }
4350
4351         start_at = slave;
4352
4353         bond_for_each_slave_from(bond, slave, i, start_at) {
4354                 if (IS_UP(slave->dev) &&
4355                     (slave->link == BOND_LINK_UP) &&
4356                     (slave->state == BOND_STATE_ACTIVE)) {
4357                         res = bond_dev_queue_xmit(bond, skb, slave->dev);
4358                         break;
4359                 }
4360         }
4361
4362 out:
4363         if (res) {
4364                 /* no suitable interface, frame not sent */
4365                 dev_kfree_skb(skb);
4366         }
4367         read_unlock(&bond->lock);
4368         return 0;
4369 }
4370
4371 /*
4372  * in broadcast mode, we send everything to all usable interfaces.
4373  */
4374 static int bond_xmit_broadcast(struct sk_buff *skb, struct net_device *bond_dev)
4375 {
4376         struct bonding *bond = bond_dev->priv;
4377         struct slave *slave, *start_at;
4378         struct net_device *tx_dev = NULL;
4379         int i;
4380         int res = 1;
4381
4382         read_lock(&bond->lock);
4383
4384         if (!BOND_IS_OK(bond)) {
4385                 goto out;
4386         }
4387
4388         read_lock(&bond->curr_slave_lock);
4389         start_at = bond->curr_active_slave;
4390         read_unlock(&bond->curr_slave_lock);
4391
4392         if (!start_at) {
4393                 goto out;
4394         }
4395
4396         bond_for_each_slave_from(bond, slave, i, start_at) {
4397                 if (IS_UP(slave->dev) &&
4398                     (slave->link == BOND_LINK_UP) &&
4399                     (slave->state == BOND_STATE_ACTIVE)) {
4400                         if (tx_dev) {
4401                                 struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
4402                                 if (!skb2) {
4403                                         printk(KERN_ERR DRV_NAME
4404                                                ": %s: Error: bond_xmit_broadcast(): "
4405                                                "skb_clone() failed\n",
4406                                                bond_dev->name);
4407                                         continue;
4408                                 }
4409
4410                                 res = bond_dev_queue_xmit(bond, skb2, tx_dev);
4411                                 if (res) {
4412                                         dev_kfree_skb(skb2);
4413                                         continue;
4414                                 }
4415                         }
4416                         tx_dev = slave->dev;
4417                 }
4418         }
4419
4420         if (tx_dev) {
4421                 res = bond_dev_queue_xmit(bond, skb, tx_dev);
4422         }
4423
4424 out:
4425         if (res) {
4426                 /* no suitable interface, frame not sent */
4427                 dev_kfree_skb(skb);
4428         }
4429         /* frame sent to all suitable interfaces */
4430         read_unlock(&bond->lock);
4431         return 0;
4432 }
4433
4434 /*------------------------- Device initialization ---------------------------*/
4435
4436 static void bond_set_xmit_hash_policy(struct bonding *bond)
4437 {
4438         switch (bond->params.xmit_policy) {
4439         case BOND_XMIT_POLICY_LAYER23:
4440                 bond->xmit_hash_policy = bond_xmit_hash_policy_l23;
4441                 break;
4442         case BOND_XMIT_POLICY_LAYER34:
4443                 bond->xmit_hash_policy = bond_xmit_hash_policy_l34;
4444                 break;
4445         case BOND_XMIT_POLICY_LAYER2:
4446         default:
4447                 bond->xmit_hash_policy = bond_xmit_hash_policy_l2;
4448                 break;
4449         }
4450 }
4451
4452 /*
4453  * set bond mode specific net device operations
4454  */
4455 void bond_set_mode_ops(struct bonding *bond, int mode)
4456 {
4457         struct net_device *bond_dev = bond->dev;
4458
4459         switch (mode) {
4460         case BOND_MODE_ROUNDROBIN:
4461                 bond_dev->hard_start_xmit = bond_xmit_roundrobin;
4462                 break;
4463         case BOND_MODE_ACTIVEBACKUP:
4464                 bond_dev->hard_start_xmit = bond_xmit_activebackup;
4465                 break;
4466         case BOND_MODE_XOR:
4467                 bond_dev->hard_start_xmit = bond_xmit_xor;
4468                 bond_set_xmit_hash_policy(bond);
4469                 break;
4470         case BOND_MODE_BROADCAST:
4471                 bond_dev->hard_start_xmit = bond_xmit_broadcast;
4472                 break;
4473         case BOND_MODE_8023AD:
4474                 bond_set_master_3ad_flags(bond);
4475                 bond_dev->hard_start_xmit = bond_3ad_xmit_xor;
4476                 bond_set_xmit_hash_policy(bond);
4477                 break;
4478         case BOND_MODE_ALB:
4479                 bond_set_master_alb_flags(bond);
4480                 /* FALLTHRU */
4481         case BOND_MODE_TLB:
4482                 bond_dev->hard_start_xmit = bond_alb_xmit;
4483                 bond_dev->set_mac_address = bond_alb_set_mac_address;
4484                 break;
4485         default:
4486                 /* Should never happen, mode already checked */
4487                 printk(KERN_ERR DRV_NAME
4488                        ": %s: Error: Unknown bonding mode %d\n",
4489                        bond_dev->name,
4490                        mode);
4491                 break;
4492         }
4493 }
4494
4495 static void bond_ethtool_get_drvinfo(struct net_device *bond_dev,
4496                                     struct ethtool_drvinfo *drvinfo)
4497 {
4498         strncpy(drvinfo->driver, DRV_NAME, 32);
4499         strncpy(drvinfo->version, DRV_VERSION, 32);
4500         snprintf(drvinfo->fw_version, 32, "%d", BOND_ABI_VERSION);
4501 }
4502
4503 static const struct ethtool_ops bond_ethtool_ops = {
4504         .get_drvinfo            = bond_ethtool_get_drvinfo,
4505         .get_link               = ethtool_op_get_link,
4506         .get_tx_csum            = ethtool_op_get_tx_csum,
4507         .get_sg                 = ethtool_op_get_sg,
4508         .get_tso                = ethtool_op_get_tso,
4509         .get_ufo                = ethtool_op_get_ufo,
4510         .get_flags              = ethtool_op_get_flags,
4511 };
4512
4513 /*
4514  * Does not allocate but creates a /proc entry.
4515  * Allowed to fail.
4516  */
4517 static int bond_init(struct net_device *bond_dev, struct bond_params *params)
4518 {
4519         struct bonding *bond = bond_dev->priv;
4520
4521         dprintk("Begin bond_init for %s\n", bond_dev->name);
4522
4523         /* initialize rwlocks */
4524         rwlock_init(&bond->lock);
4525         rwlock_init(&bond->curr_slave_lock);
4526
4527         bond->params = *params; /* copy params struct */
4528
4529         bond->wq = create_singlethread_workqueue(bond_dev->name);
4530         if (!bond->wq)
4531                 return -ENOMEM;
4532
4533         /* Initialize pointers */
4534         bond->first_slave = NULL;
4535         bond->curr_active_slave = NULL;
4536         bond->current_arp_slave = NULL;
4537         bond->primary_slave = NULL;
4538         bond->dev = bond_dev;
4539         bond->send_grat_arp = 0;
4540         bond->setup_by_slave = 0;
4541         INIT_LIST_HEAD(&bond->vlan_list);
4542
4543         /* Initialize the device entry points */
4544         bond_dev->open = bond_open;
4545         bond_dev->stop = bond_close;
4546         bond_dev->get_stats = bond_get_stats;
4547         bond_dev->do_ioctl = bond_do_ioctl;
4548         bond_dev->ethtool_ops = &bond_ethtool_ops;
4549         bond_dev->set_multicast_list = bond_set_multicast_list;
4550         bond_dev->change_mtu = bond_change_mtu;
4551         bond_dev->set_mac_address = bond_set_mac_address;
4552         bond_dev->validate_addr = NULL;
4553
4554         bond_set_mode_ops(bond, bond->params.mode);
4555
4556         bond_dev->destructor = free_netdev;
4557
4558         /* Initialize the device options */
4559         bond_dev->tx_queue_len = 0;
4560         bond_dev->flags |= IFF_MASTER|IFF_MULTICAST;
4561         bond_dev->priv_flags |= IFF_BONDING;
4562
4563         /* At first, we block adding VLANs. That's the only way to
4564          * prevent problems that occur when adding VLANs over an
4565          * empty bond. The block will be removed once non-challenged
4566          * slaves are enslaved.
4567          */
4568         bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
4569
4570         /* don't acquire bond device's netif_tx_lock when
4571          * transmitting */
4572         bond_dev->features |= NETIF_F_LLTX;
4573
4574         /* By default, we declare the bond to be fully
4575          * VLAN hardware accelerated capable. Special
4576          * care is taken in the various xmit functions
4577          * when there are slaves that are not hw accel
4578          * capable
4579          */
4580         bond_dev->vlan_rx_register = bond_vlan_rx_register;
4581         bond_dev->vlan_rx_add_vid  = bond_vlan_rx_add_vid;
4582         bond_dev->vlan_rx_kill_vid = bond_vlan_rx_kill_vid;
4583         bond_dev->features |= (NETIF_F_HW_VLAN_TX |
4584                                NETIF_F_HW_VLAN_RX |
4585                                NETIF_F_HW_VLAN_FILTER);
4586
4587 #ifdef CONFIG_PROC_FS
4588         bond_create_proc_entry(bond);
4589 #endif
4590         list_add_tail(&bond->bond_list, &bond_dev_list);
4591
4592         return 0;
4593 }
4594
4595 /* De-initialize device specific data.
4596  * Caller must hold rtnl_lock.
4597  */
4598 static void bond_deinit(struct net_device *bond_dev)
4599 {
4600         struct bonding *bond = bond_dev->priv;
4601
4602         list_del(&bond->bond_list);
4603
4604 #ifdef CONFIG_PROC_FS
4605         bond_remove_proc_entry(bond);
4606 #endif
4607 }
4608
4609 static void bond_work_cancel_all(struct bonding *bond)
4610 {
4611         write_lock_bh(&bond->lock);
4612         bond->kill_timers = 1;
4613         write_unlock_bh(&bond->lock);
4614
4615         if (bond->params.miimon && delayed_work_pending(&bond->mii_work))
4616                 cancel_delayed_work(&bond->mii_work);
4617
4618         if (bond->params.arp_interval && delayed_work_pending(&bond->arp_work))
4619                 cancel_delayed_work(&bond->arp_work);
4620
4621         if (bond->params.mode == BOND_MODE_ALB &&
4622             delayed_work_pending(&bond->alb_work))
4623                 cancel_delayed_work(&bond->alb_work);
4624
4625         if (bond->params.mode == BOND_MODE_8023AD &&
4626             delayed_work_pending(&bond->ad_work))
4627                 cancel_delayed_work(&bond->ad_work);
4628 }
4629
4630 /* Unregister and free all bond devices.
4631  * Caller must hold rtnl_lock.
4632  */
4633 static void bond_free_all(void)
4634 {
4635         struct bonding *bond, *nxt;
4636
4637         list_for_each_entry_safe(bond, nxt, &bond_dev_list, bond_list) {
4638                 struct net_device *bond_dev = bond->dev;
4639
4640                 bond_work_cancel_all(bond);
4641                 netif_addr_lock_bh(bond_dev);
4642                 bond_mc_list_destroy(bond);
4643                 netif_addr_unlock_bh(bond_dev);
4644                 /* Release the bonded slaves */
4645                 bond_release_all(bond_dev);
4646                 bond_destroy(bond);
4647         }
4648
4649 #ifdef CONFIG_PROC_FS
4650         bond_destroy_proc_dir();
4651 #endif
4652 }
4653
4654 /*------------------------- Module initialization ---------------------------*/
4655
4656 /*
4657  * Convert string input module parms.  Accept either the
4658  * number of the mode or its string name.  A bit complicated because
4659  * some mode names are substrings of other names, and calls from sysfs
4660  * may have whitespace in the name (trailing newlines, for example).
4661  */
4662 int bond_parse_parm(const char *buf, struct bond_parm_tbl *tbl)
4663 {
4664         int mode = -1, i, rv;
4665         char *p, modestr[BOND_MAX_MODENAME_LEN + 1] = { 0, };
4666
4667         for (p = (char *)buf; *p; p++)
4668                 if (!(isdigit(*p) || isspace(*p)))
4669                         break;
4670
4671         if (*p)
4672                 rv = sscanf(buf, "%20s", modestr);
4673         else
4674                 rv = sscanf(buf, "%d", &mode);
4675
4676         if (!rv)
4677                 return -1;
4678
4679         for (i = 0; tbl[i].modename; i++) {
4680                 if (mode == tbl[i].mode)
4681                         return tbl[i].mode;
4682                 if (strcmp(modestr, tbl[i].modename) == 0)
4683                         return tbl[i].mode;
4684         }
4685
4686         return -1;
4687 }
4688
4689 static int bond_check_params(struct bond_params *params)
4690 {
4691         int arp_validate_value, fail_over_mac_value;
4692
4693         /*
4694          * Convert string parameters.
4695          */
4696         if (mode) {
4697                 bond_mode = bond_parse_parm(mode, bond_mode_tbl);
4698                 if (bond_mode == -1) {
4699                         printk(KERN_ERR DRV_NAME
4700                                ": Error: Invalid bonding mode \"%s\"\n",
4701                                mode == NULL ? "NULL" : mode);
4702                         return -EINVAL;
4703                 }
4704         }
4705
4706         if (xmit_hash_policy) {
4707                 if ((bond_mode != BOND_MODE_XOR) &&
4708                     (bond_mode != BOND_MODE_8023AD)) {
4709                         printk(KERN_INFO DRV_NAME
4710                                ": xor_mode param is irrelevant in mode %s\n",
4711                                bond_mode_name(bond_mode));
4712                 } else {
4713                         xmit_hashtype = bond_parse_parm(xmit_hash_policy,
4714                                                         xmit_hashtype_tbl);
4715                         if (xmit_hashtype == -1) {
4716                                 printk(KERN_ERR DRV_NAME
4717                                 ": Error: Invalid xmit_hash_policy \"%s\"\n",
4718                                 xmit_hash_policy == NULL ? "NULL" :
4719                                        xmit_hash_policy);
4720                                 return -EINVAL;
4721                         }
4722                 }
4723         }
4724
4725         if (lacp_rate) {
4726                 if (bond_mode != BOND_MODE_8023AD) {
4727                         printk(KERN_INFO DRV_NAME
4728                                ": lacp_rate param is irrelevant in mode %s\n",
4729                                bond_mode_name(bond_mode));
4730                 } else {
4731                         lacp_fast = bond_parse_parm(lacp_rate, bond_lacp_tbl);
4732                         if (lacp_fast == -1) {
4733                                 printk(KERN_ERR DRV_NAME
4734                                        ": Error: Invalid lacp rate \"%s\"\n",
4735                                        lacp_rate == NULL ? "NULL" : lacp_rate);
4736                                 return -EINVAL;
4737                         }
4738                 }
4739         }
4740
4741         if (max_bonds < 0 || max_bonds > INT_MAX) {
4742                 printk(KERN_WARNING DRV_NAME
4743                        ": Warning: max_bonds (%d) not in range %d-%d, so it "
4744                        "was reset to BOND_DEFAULT_MAX_BONDS (%d)\n",
4745                        max_bonds, 0, INT_MAX, BOND_DEFAULT_MAX_BONDS);
4746                 max_bonds = BOND_DEFAULT_MAX_BONDS;
4747         }
4748
4749         if (miimon < 0) {
4750                 printk(KERN_WARNING DRV_NAME
4751                        ": Warning: miimon module parameter (%d), "
4752                        "not in range 0-%d, so it was reset to %d\n",
4753                        miimon, INT_MAX, BOND_LINK_MON_INTERV);
4754                 miimon = BOND_LINK_MON_INTERV;
4755         }
4756
4757         if (updelay < 0) {
4758                 printk(KERN_WARNING DRV_NAME
4759                        ": Warning: updelay module parameter (%d), "
4760                        "not in range 0-%d, so it was reset to 0\n",
4761                        updelay, INT_MAX);
4762                 updelay = 0;
4763         }
4764
4765         if (downdelay < 0) {
4766                 printk(KERN_WARNING DRV_NAME
4767                        ": Warning: downdelay module parameter (%d), "
4768                        "not in range 0-%d, so it was reset to 0\n",
4769                        downdelay, INT_MAX);
4770                 downdelay = 0;
4771         }
4772
4773         if ((use_carrier != 0) && (use_carrier != 1)) {
4774                 printk(KERN_WARNING DRV_NAME
4775                        ": Warning: use_carrier module parameter (%d), "
4776                        "not of valid value (0/1), so it was set to 1\n",
4777                        use_carrier);
4778                 use_carrier = 1;
4779         }
4780
4781         if (num_grat_arp < 0 || num_grat_arp > 255) {
4782                 printk(KERN_WARNING DRV_NAME
4783                        ": Warning: num_grat_arp (%d) not in range 0-255 so it "
4784                        "was reset to 1 \n", num_grat_arp);
4785                 num_grat_arp = 1;
4786         }
4787
4788         /* reset values for 802.3ad */
4789         if (bond_mode == BOND_MODE_8023AD) {
4790                 if (!miimon) {
4791                         printk(KERN_WARNING DRV_NAME
4792                                ": Warning: miimon must be specified, "
4793                                "otherwise bonding will not detect link "
4794                                "failure, speed and duplex which are "
4795                                "essential for 802.3ad operation\n");
4796                         printk(KERN_WARNING "Forcing miimon to 100msec\n");
4797                         miimon = 100;
4798                 }
4799         }
4800
4801         /* reset values for TLB/ALB */
4802         if ((bond_mode == BOND_MODE_TLB) ||
4803             (bond_mode == BOND_MODE_ALB)) {
4804                 if (!miimon) {
4805                         printk(KERN_WARNING DRV_NAME
4806                                ": Warning: miimon must be specified, "
4807                                "otherwise bonding will not detect link "
4808                                "failure and link speed which are essential "
4809                                "for TLB/ALB load balancing\n");
4810                         printk(KERN_WARNING "Forcing miimon to 100msec\n");
4811                         miimon = 100;
4812                 }
4813         }
4814
4815         if (bond_mode == BOND_MODE_ALB) {
4816                 printk(KERN_NOTICE DRV_NAME
4817                        ": In ALB mode you might experience client "
4818                        "disconnections upon reconnection of a link if the "
4819                        "bonding module updelay parameter (%d msec) is "
4820                        "incompatible with the forwarding delay time of the "
4821                        "switch\n",
4822                        updelay);
4823         }
4824
4825         if (!miimon) {
4826                 if (updelay || downdelay) {
4827                         /* just warn the user the up/down delay will have
4828                          * no effect since miimon is zero...
4829                          */
4830                         printk(KERN_WARNING DRV_NAME
4831                                ": Warning: miimon module parameter not set "
4832                                "and updelay (%d) or downdelay (%d) module "
4833                                "parameter is set; updelay and downdelay have "
4834                                "no effect unless miimon is set\n",
4835                                updelay, downdelay);
4836                 }
4837         } else {
4838                 /* don't allow arp monitoring */
4839                 if (arp_interval) {
4840                         printk(KERN_WARNING DRV_NAME
4841                                ": Warning: miimon (%d) and arp_interval (%d) "
4842                                "can't be used simultaneously, disabling ARP "
4843                                "monitoring\n",
4844                                miimon, arp_interval);
4845                         arp_interval = 0;
4846                 }
4847
4848                 if ((updelay % miimon) != 0) {
4849                         printk(KERN_WARNING DRV_NAME
4850                                ": Warning: updelay (%d) is not a multiple "
4851                                "of miimon (%d), updelay rounded to %d ms\n",
4852                                updelay, miimon, (updelay / miimon) * miimon);
4853                 }
4854
4855                 updelay /= miimon;
4856
4857                 if ((downdelay % miimon) != 0) {
4858                         printk(KERN_WARNING DRV_NAME
4859                                ": Warning: downdelay (%d) is not a multiple "
4860                                "of miimon (%d), downdelay rounded to %d ms\n",
4861                                downdelay, miimon,
4862                                (downdelay / miimon) * miimon);
4863                 }
4864
4865                 downdelay /= miimon;
4866         }
4867
4868         if (arp_interval < 0) {
4869                 printk(KERN_WARNING DRV_NAME
4870                        ": Warning: arp_interval module parameter (%d) "
4871                        ", not in range 0-%d, so it was reset to %d\n",
4872                        arp_interval, INT_MAX, BOND_LINK_ARP_INTERV);
4873                 arp_interval = BOND_LINK_ARP_INTERV;
4874         }
4875
4876         for (arp_ip_count = 0;
4877              (arp_ip_count < BOND_MAX_ARP_TARGETS) && arp_ip_target[arp_ip_count];
4878              arp_ip_count++) {
4879                 /* not complete check, but should be good enough to
4880                    catch mistakes */
4881                 if (!isdigit(arp_ip_target[arp_ip_count][0])) {
4882                         printk(KERN_WARNING DRV_NAME
4883                                ": Warning: bad arp_ip_target module parameter "
4884                                "(%s), ARP monitoring will not be performed\n",
4885                                arp_ip_target[arp_ip_count]);
4886                         arp_interval = 0;
4887                 } else {
4888                         __be32 ip = in_aton(arp_ip_target[arp_ip_count]);
4889                         arp_target[arp_ip_count] = ip;
4890                 }
4891         }
4892
4893         if (arp_interval && !arp_ip_count) {
4894                 /* don't allow arping if no arp_ip_target given... */
4895                 printk(KERN_WARNING DRV_NAME
4896                        ": Warning: arp_interval module parameter (%d) "
4897                        "specified without providing an arp_ip_target "
4898                        "parameter, arp_interval was reset to 0\n",
4899                        arp_interval);
4900                 arp_interval = 0;
4901         }
4902
4903         if (arp_validate) {
4904                 if (bond_mode != BOND_MODE_ACTIVEBACKUP) {
4905                         printk(KERN_ERR DRV_NAME
4906                ": arp_validate only supported in active-backup mode\n");
4907                         return -EINVAL;
4908                 }
4909                 if (!arp_interval) {
4910                         printk(KERN_ERR DRV_NAME
4911                                ": arp_validate requires arp_interval\n");
4912                         return -EINVAL;
4913                 }
4914
4915                 arp_validate_value = bond_parse_parm(arp_validate,
4916                                                      arp_validate_tbl);
4917                 if (arp_validate_value == -1) {
4918                         printk(KERN_ERR DRV_NAME
4919                                ": Error: invalid arp_validate \"%s\"\n",
4920                                arp_validate == NULL ? "NULL" : arp_validate);
4921                         return -EINVAL;
4922                 }
4923         } else
4924                 arp_validate_value = 0;
4925
4926         if (miimon) {
4927                 printk(KERN_INFO DRV_NAME
4928                        ": MII link monitoring set to %d ms\n",
4929                        miimon);
4930         } else if (arp_interval) {
4931                 int i;
4932
4933                 printk(KERN_INFO DRV_NAME
4934                        ": ARP monitoring set to %d ms, validate %s, with %d target(s):",
4935                        arp_interval,
4936                        arp_validate_tbl[arp_validate_value].modename,
4937                        arp_ip_count);
4938
4939                 for (i = 0; i < arp_ip_count; i++)
4940                         printk (" %s", arp_ip_target[i]);
4941
4942                 printk("\n");
4943
4944         } else if (max_bonds) {
4945                 /* miimon and arp_interval not set, we need one so things
4946                  * work as expected, see bonding.txt for details
4947                  */
4948                 printk(KERN_WARNING DRV_NAME
4949                        ": Warning: either miimon or arp_interval and "
4950                        "arp_ip_target module parameters must be specified, "
4951                        "otherwise bonding will not detect link failures! see "
4952                        "bonding.txt for details.\n");
4953         }
4954
4955         if (primary && !USES_PRIMARY(bond_mode)) {
4956                 /* currently, using a primary only makes sense
4957                  * in active backup, TLB or ALB modes
4958                  */
4959                 printk(KERN_WARNING DRV_NAME
4960                        ": Warning: %s primary device specified but has no "
4961                        "effect in %s mode\n",
4962                        primary, bond_mode_name(bond_mode));
4963                 primary = NULL;
4964         }
4965
4966         if (fail_over_mac) {
4967                 fail_over_mac_value = bond_parse_parm(fail_over_mac,
4968                                                       fail_over_mac_tbl);
4969                 if (fail_over_mac_value == -1) {
4970                         printk(KERN_ERR DRV_NAME
4971                                ": Error: invalid fail_over_mac \"%s\"\n",
4972                                arp_validate == NULL ? "NULL" : arp_validate);
4973                         return -EINVAL;
4974                 }
4975
4976                 if (bond_mode != BOND_MODE_ACTIVEBACKUP)
4977                         printk(KERN_WARNING DRV_NAME
4978                                ": Warning: fail_over_mac only affects "
4979                                "active-backup mode.\n");
4980         } else {
4981                 fail_over_mac_value = BOND_FOM_NONE;
4982         }
4983
4984         /* fill params struct with the proper values */
4985         params->mode = bond_mode;
4986         params->xmit_policy = xmit_hashtype;
4987         params->miimon = miimon;
4988         params->num_grat_arp = num_grat_arp;
4989         params->arp_interval = arp_interval;
4990         params->arp_validate = arp_validate_value;
4991         params->updelay = updelay;
4992         params->downdelay = downdelay;
4993         params->use_carrier = use_carrier;
4994         params->lacp_fast = lacp_fast;
4995         params->primary[0] = 0;
4996         params->fail_over_mac = fail_over_mac_value;
4997
4998         if (primary) {
4999                 strncpy(params->primary, primary, IFNAMSIZ);
5000                 params->primary[IFNAMSIZ - 1] = 0;
5001         }
5002
5003         memcpy(params->arp_targets, arp_target, sizeof(arp_target));
5004
5005         return 0;
5006 }
5007
5008 static struct lock_class_key bonding_netdev_xmit_lock_key;
5009 static struct lock_class_key bonding_netdev_addr_lock_key;
5010
5011 static void bond_set_lockdep_class_one(struct net_device *dev,
5012                                        struct netdev_queue *txq,
5013                                        void *_unused)
5014 {
5015         lockdep_set_class(&txq->_xmit_lock,
5016                           &bonding_netdev_xmit_lock_key);
5017 }
5018
5019 static void bond_set_lockdep_class(struct net_device *dev)
5020 {
5021         lockdep_set_class(&dev->addr_list_lock,
5022                           &bonding_netdev_addr_lock_key);
5023         netdev_for_each_tx_queue(dev, bond_set_lockdep_class_one, NULL);
5024 }
5025
5026 /* Create a new bond based on the specified name and bonding parameters.
5027  * If name is NULL, obtain a suitable "bond%d" name for us.
5028  * Caller must NOT hold rtnl_lock; we need to release it here before we
5029  * set up our sysfs entries.
5030  */
5031 int bond_create(char *name, struct bond_params *params)
5032 {
5033         struct net_device *bond_dev;
5034         struct bonding *bond;
5035         int res;
5036
5037         rtnl_lock();
5038         down_write(&bonding_rwsem);
5039
5040         /* Check to see if the bond already exists. */
5041         if (name) {
5042                 list_for_each_entry(bond, &bond_dev_list, bond_list)
5043                         if (strnicmp(bond->dev->name, name, IFNAMSIZ) == 0) {
5044                                 printk(KERN_ERR DRV_NAME
5045                                ": cannot add bond %s; it already exists\n",
5046                                        name);
5047                                 res = -EPERM;
5048                                 goto out_rtnl;
5049                         }
5050         }
5051
5052         bond_dev = alloc_netdev(sizeof(struct bonding), name ? name : "",
5053                                 ether_setup);
5054         if (!bond_dev) {
5055                 printk(KERN_ERR DRV_NAME
5056                        ": %s: eek! can't alloc netdev!\n",
5057                        name);
5058                 res = -ENOMEM;
5059                 goto out_rtnl;
5060         }
5061
5062         if (!name) {
5063                 res = dev_alloc_name(bond_dev, "bond%d");
5064                 if (res < 0)
5065                         goto out_netdev;
5066         }
5067
5068         /* bond_init() must be called after dev_alloc_name() (for the
5069          * /proc files), but before register_netdevice(), because we
5070          * need to set function pointers.
5071          */
5072
5073         res = bond_init(bond_dev, params);
5074         if (res < 0) {
5075                 goto out_netdev;
5076         }
5077
5078         res = register_netdevice(bond_dev);
5079         if (res < 0) {
5080                 goto out_bond;
5081         }
5082
5083         bond_set_lockdep_class(bond_dev);
5084
5085         netif_carrier_off(bond_dev);
5086
5087         up_write(&bonding_rwsem);
5088         rtnl_unlock(); /* allows sysfs registration of net device */
5089         res = bond_create_sysfs_entry(bond_dev->priv);
5090         if (res < 0) {
5091                 rtnl_lock();
5092                 down_write(&bonding_rwsem);
5093                 bond_deinit(bond_dev);
5094                 unregister_netdevice(bond_dev);
5095                 goto out_rtnl;
5096         }
5097
5098         return 0;
5099
5100 out_bond:
5101         bond_deinit(bond_dev);
5102 out_netdev:
5103         free_netdev(bond_dev);
5104 out_rtnl:
5105         up_write(&bonding_rwsem);
5106         rtnl_unlock();
5107         return res;
5108 }
5109
5110 static int __init bonding_init(void)
5111 {
5112         int i;
5113         int res;
5114         struct bonding *bond;
5115
5116         printk(KERN_INFO "%s", version);
5117
5118         res = bond_check_params(&bonding_defaults);
5119         if (res) {
5120                 goto out;
5121         }
5122
5123 #ifdef CONFIG_PROC_FS
5124         bond_create_proc_dir();
5125 #endif
5126
5127         init_rwsem(&bonding_rwsem);
5128
5129         for (i = 0; i < max_bonds; i++) {
5130                 res = bond_create(NULL, &bonding_defaults);
5131                 if (res)
5132                         goto err;
5133         }
5134
5135         res = bond_create_sysfs();
5136         if (res)
5137                 goto err;
5138
5139         register_netdevice_notifier(&bond_netdev_notifier);
5140         register_inetaddr_notifier(&bond_inetaddr_notifier);
5141
5142         goto out;
5143 err:
5144         list_for_each_entry(bond, &bond_dev_list, bond_list) {
5145                 bond_work_cancel_all(bond);
5146                 destroy_workqueue(bond->wq);
5147         }
5148
5149         bond_destroy_sysfs();
5150
5151         rtnl_lock();
5152         bond_free_all();
5153         rtnl_unlock();
5154 out:
5155         return res;
5156
5157 }
5158
5159 static void __exit bonding_exit(void)
5160 {
5161         unregister_netdevice_notifier(&bond_netdev_notifier);
5162         unregister_inetaddr_notifier(&bond_inetaddr_notifier);
5163
5164         bond_destroy_sysfs();
5165
5166         rtnl_lock();
5167         bond_free_all();
5168         rtnl_unlock();
5169 }
5170
5171 module_init(bonding_init);
5172 module_exit(bonding_exit);
5173 MODULE_LICENSE("GPL");
5174 MODULE_VERSION(DRV_VERSION);
5175 MODULE_DESCRIPTION(DRV_DESCRIPTION ", v" DRV_VERSION);
5176 MODULE_AUTHOR("Thomas Davis, tadavis@lbl.gov and many others");
5177 MODULE_SUPPORTED_DEVICE("most ethernet devices");
5178
5179 /*
5180  * Local variables:
5181  *  c-indent-level: 8
5182  *  c-basic-offset: 8
5183  *  tab-width: 8
5184  * End:
5185  */
5186