Merge branch 'master' of master.kernel.org:/pub/scm/linux/kernel/git/davem/net-2.6
[linux-2.6] / 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
1795         /* slave is not a slave or master is not master of this slave */
1796         if (!(slave_dev->flags & IFF_SLAVE) ||
1797             (slave_dev->master != bond_dev)) {
1798                 printk(KERN_ERR DRV_NAME
1799                        ": %s: Error: cannot release %s.\n",
1800                        bond_dev->name, slave_dev->name);
1801                 return -EINVAL;
1802         }
1803
1804         write_lock_bh(&bond->lock);
1805
1806         slave = bond_get_slave_by_dev(bond, slave_dev);
1807         if (!slave) {
1808                 /* not a slave of this bond */
1809                 printk(KERN_INFO DRV_NAME
1810                        ": %s: %s not enslaved\n",
1811                        bond_dev->name, slave_dev->name);
1812                 write_unlock_bh(&bond->lock);
1813                 return -EINVAL;
1814         }
1815
1816         if (!bond->params.fail_over_mac) {
1817                 mac_addr_differ = memcmp(bond_dev->dev_addr, slave->perm_hwaddr,
1818                                          ETH_ALEN);
1819                 if (!mac_addr_differ && (bond->slave_cnt > 1))
1820                         printk(KERN_WARNING DRV_NAME
1821                                ": %s: Warning: the permanent HWaddr of %s - "
1822                                "%pM - is still in use by %s. "
1823                                "Set the HWaddr of %s to a different address "
1824                                "to avoid conflicts.\n",
1825                                bond_dev->name, slave_dev->name,
1826                                slave->perm_hwaddr,
1827                                bond_dev->name, slave_dev->name);
1828         }
1829
1830         /* Inform AD package of unbinding of slave. */
1831         if (bond->params.mode == BOND_MODE_8023AD) {
1832                 /* must be called before the slave is
1833                  * detached from the list
1834                  */
1835                 bond_3ad_unbind_slave(slave);
1836         }
1837
1838         printk(KERN_INFO DRV_NAME
1839                ": %s: releasing %s interface %s\n",
1840                bond_dev->name,
1841                (slave->state == BOND_STATE_ACTIVE)
1842                ? "active" : "backup",
1843                slave_dev->name);
1844
1845         oldcurrent = bond->curr_active_slave;
1846
1847         bond->current_arp_slave = NULL;
1848
1849         /* release the slave from its bond */
1850         bond_detach_slave(bond, slave);
1851
1852         bond_compute_features(bond);
1853
1854         if (bond->primary_slave == slave) {
1855                 bond->primary_slave = NULL;
1856         }
1857
1858         if (oldcurrent == slave) {
1859                 bond_change_active_slave(bond, NULL);
1860         }
1861
1862         if ((bond->params.mode == BOND_MODE_TLB) ||
1863             (bond->params.mode == BOND_MODE_ALB)) {
1864                 /* Must be called only after the slave has been
1865                  * detached from the list and the curr_active_slave
1866                  * has been cleared (if our_slave == old_current),
1867                  * but before a new active slave is selected.
1868                  */
1869                 write_unlock_bh(&bond->lock);
1870                 bond_alb_deinit_slave(bond, slave);
1871                 write_lock_bh(&bond->lock);
1872         }
1873
1874         if (oldcurrent == slave) {
1875                 /*
1876                  * Note that we hold RTNL over this sequence, so there
1877                  * is no concern that another slave add/remove event
1878                  * will interfere.
1879                  */
1880                 write_unlock_bh(&bond->lock);
1881                 read_lock(&bond->lock);
1882                 write_lock_bh(&bond->curr_slave_lock);
1883
1884                 bond_select_active_slave(bond);
1885
1886                 write_unlock_bh(&bond->curr_slave_lock);
1887                 read_unlock(&bond->lock);
1888                 write_lock_bh(&bond->lock);
1889         }
1890
1891         if (bond->slave_cnt == 0) {
1892                 bond_set_carrier(bond);
1893
1894                 /* if the last slave was removed, zero the mac address
1895                  * of the master so it will be set by the application
1896                  * to the mac address of the first slave
1897                  */
1898                 memset(bond_dev->dev_addr, 0, bond_dev->addr_len);
1899
1900                 if (list_empty(&bond->vlan_list)) {
1901                         bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
1902                 } else {
1903                         printk(KERN_WARNING DRV_NAME
1904                                ": %s: Warning: clearing HW address of %s while it "
1905                                "still has VLANs.\n",
1906                                bond_dev->name, bond_dev->name);
1907                         printk(KERN_WARNING DRV_NAME
1908                                ": %s: When re-adding slaves, make sure the bond's "
1909                                "HW address matches its VLANs'.\n",
1910                                bond_dev->name);
1911                 }
1912         } else if ((bond_dev->features & NETIF_F_VLAN_CHALLENGED) &&
1913                    !bond_has_challenged_slaves(bond)) {
1914                 printk(KERN_INFO DRV_NAME
1915                        ": %s: last VLAN challenged slave %s "
1916                        "left bond %s. VLAN blocking is removed\n",
1917                        bond_dev->name, slave_dev->name, bond_dev->name);
1918                 bond_dev->features &= ~NETIF_F_VLAN_CHALLENGED;
1919         }
1920
1921         write_unlock_bh(&bond->lock);
1922
1923         /* must do this from outside any spinlocks */
1924         bond_destroy_slave_symlinks(bond_dev, slave_dev);
1925
1926         bond_del_vlans_from_slave(bond, slave_dev);
1927
1928         /* If the mode USES_PRIMARY, then we should only remove its
1929          * promisc and mc settings if it was the curr_active_slave, but that was
1930          * already taken care of above when we detached the slave
1931          */
1932         if (!USES_PRIMARY(bond->params.mode)) {
1933                 /* unset promiscuity level from slave */
1934                 if (bond_dev->flags & IFF_PROMISC) {
1935                         dev_set_promiscuity(slave_dev, -1);
1936                 }
1937
1938                 /* unset allmulti level from slave */
1939                 if (bond_dev->flags & IFF_ALLMULTI) {
1940                         dev_set_allmulti(slave_dev, -1);
1941                 }
1942
1943                 /* flush master's mc_list from slave */
1944                 netif_addr_lock_bh(bond_dev);
1945                 bond_mc_list_flush(bond_dev, slave_dev);
1946                 netif_addr_unlock_bh(bond_dev);
1947         }
1948
1949         netdev_set_master(slave_dev, NULL);
1950
1951         /* close slave before restoring its mac address */
1952         dev_close(slave_dev);
1953
1954         if (bond->params.fail_over_mac != BOND_FOM_ACTIVE) {
1955                 /* restore original ("permanent") mac address */
1956                 memcpy(addr.sa_data, slave->perm_hwaddr, ETH_ALEN);
1957                 addr.sa_family = slave_dev->type;
1958                 dev_set_mac_address(slave_dev, &addr);
1959         }
1960
1961         slave_dev->priv_flags &= ~(IFF_MASTER_8023AD | IFF_MASTER_ALB |
1962                                    IFF_SLAVE_INACTIVE | IFF_BONDING |
1963                                    IFF_SLAVE_NEEDARP);
1964
1965         kfree(slave);
1966
1967         return 0;  /* deletion OK */
1968 }
1969
1970 /*
1971 * Destroy a bonding device.
1972 * Must be under rtnl_lock when this function is called.
1973 */
1974 void bond_destroy(struct bonding *bond)
1975 {
1976         bond_deinit(bond->dev);
1977         bond_destroy_sysfs_entry(bond);
1978         unregister_netdevice(bond->dev);
1979 }
1980
1981 static void bond_destructor(struct net_device *bond_dev)
1982 {
1983         struct bonding *bond = bond_dev->priv;
1984
1985         if (bond->wq)
1986                 destroy_workqueue(bond->wq);
1987
1988         netif_addr_lock_bh(bond_dev);
1989         bond_mc_list_destroy(bond);
1990         netif_addr_unlock_bh(bond_dev);
1991
1992         free_netdev(bond_dev);
1993 }
1994
1995 /*
1996 * First release a slave and than destroy the bond if no more slaves iare left.
1997 * Must be under rtnl_lock when this function is called.
1998 */
1999 int  bond_release_and_destroy(struct net_device *bond_dev, struct net_device *slave_dev)
2000 {
2001         struct bonding *bond = bond_dev->priv;
2002         int ret;
2003
2004         ret = bond_release(bond_dev, slave_dev);
2005         if ((ret == 0) && (bond->slave_cnt == 0)) {
2006                 printk(KERN_INFO DRV_NAME ": %s: destroying bond %s.\n",
2007                        bond_dev->name, bond_dev->name);
2008                 bond_destroy(bond);
2009         }
2010         return ret;
2011 }
2012
2013 /*
2014  * This function releases all slaves.
2015  */
2016 static int bond_release_all(struct net_device *bond_dev)
2017 {
2018         struct bonding *bond = bond_dev->priv;
2019         struct slave *slave;
2020         struct net_device *slave_dev;
2021         struct sockaddr addr;
2022
2023         write_lock_bh(&bond->lock);
2024
2025         netif_carrier_off(bond_dev);
2026
2027         if (bond->slave_cnt == 0) {
2028                 goto out;
2029         }
2030
2031         bond->current_arp_slave = NULL;
2032         bond->primary_slave = NULL;
2033         bond_change_active_slave(bond, NULL);
2034
2035         while ((slave = bond->first_slave) != NULL) {
2036                 /* Inform AD package of unbinding of slave
2037                  * before slave is detached from the list.
2038                  */
2039                 if (bond->params.mode == BOND_MODE_8023AD) {
2040                         bond_3ad_unbind_slave(slave);
2041                 }
2042
2043                 slave_dev = slave->dev;
2044                 bond_detach_slave(bond, slave);
2045
2046                 /* now that the slave is detached, unlock and perform
2047                  * all the undo steps that should not be called from
2048                  * within a lock.
2049                  */
2050                 write_unlock_bh(&bond->lock);
2051
2052                 if ((bond->params.mode == BOND_MODE_TLB) ||
2053                     (bond->params.mode == BOND_MODE_ALB)) {
2054                         /* must be called only after the slave
2055                          * has been detached from the list
2056                          */
2057                         bond_alb_deinit_slave(bond, slave);
2058                 }
2059
2060                 bond_compute_features(bond);
2061
2062                 bond_destroy_slave_symlinks(bond_dev, slave_dev);
2063                 bond_del_vlans_from_slave(bond, slave_dev);
2064
2065                 /* If the mode USES_PRIMARY, then we should only remove its
2066                  * promisc and mc settings if it was the curr_active_slave, but that was
2067                  * already taken care of above when we detached the slave
2068                  */
2069                 if (!USES_PRIMARY(bond->params.mode)) {
2070                         /* unset promiscuity level from slave */
2071                         if (bond_dev->flags & IFF_PROMISC) {
2072                                 dev_set_promiscuity(slave_dev, -1);
2073                         }
2074
2075                         /* unset allmulti level from slave */
2076                         if (bond_dev->flags & IFF_ALLMULTI) {
2077                                 dev_set_allmulti(slave_dev, -1);
2078                         }
2079
2080                         /* flush master's mc_list from slave */
2081                         netif_addr_lock_bh(bond_dev);
2082                         bond_mc_list_flush(bond_dev, slave_dev);
2083                         netif_addr_unlock_bh(bond_dev);
2084                 }
2085
2086                 netdev_set_master(slave_dev, NULL);
2087
2088                 /* close slave before restoring its mac address */
2089                 dev_close(slave_dev);
2090
2091                 if (!bond->params.fail_over_mac) {
2092                         /* restore original ("permanent") mac address*/
2093                         memcpy(addr.sa_data, slave->perm_hwaddr, ETH_ALEN);
2094                         addr.sa_family = slave_dev->type;
2095                         dev_set_mac_address(slave_dev, &addr);
2096                 }
2097
2098                 slave_dev->priv_flags &= ~(IFF_MASTER_8023AD | IFF_MASTER_ALB |
2099                                            IFF_SLAVE_INACTIVE);
2100
2101                 kfree(slave);
2102
2103                 /* re-acquire the lock before getting the next slave */
2104                 write_lock_bh(&bond->lock);
2105         }
2106
2107         /* zero the mac address of the master so it will be
2108          * set by the application to the mac address of the
2109          * first slave
2110          */
2111         memset(bond_dev->dev_addr, 0, bond_dev->addr_len);
2112
2113         if (list_empty(&bond->vlan_list)) {
2114                 bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
2115         } else {
2116                 printk(KERN_WARNING DRV_NAME
2117                        ": %s: Warning: clearing HW address of %s while it "
2118                        "still has VLANs.\n",
2119                        bond_dev->name, bond_dev->name);
2120                 printk(KERN_WARNING DRV_NAME
2121                        ": %s: When re-adding slaves, make sure the bond's "
2122                        "HW address matches its VLANs'.\n",
2123                        bond_dev->name);
2124         }
2125
2126         printk(KERN_INFO DRV_NAME
2127                ": %s: released all slaves\n",
2128                bond_dev->name);
2129
2130 out:
2131         write_unlock_bh(&bond->lock);
2132
2133         return 0;
2134 }
2135
2136 /*
2137  * This function changes the active slave to slave <slave_dev>.
2138  * It returns -EINVAL in the following cases.
2139  *  - <slave_dev> is not found in the list.
2140  *  - There is not active slave now.
2141  *  - <slave_dev> is already active.
2142  *  - The link state of <slave_dev> is not BOND_LINK_UP.
2143  *  - <slave_dev> is not running.
2144  * In these cases, this fuction does nothing.
2145  * In the other cases, currnt_slave pointer is changed and 0 is returned.
2146  */
2147 static int bond_ioctl_change_active(struct net_device *bond_dev, struct net_device *slave_dev)
2148 {
2149         struct bonding *bond = bond_dev->priv;
2150         struct slave *old_active = NULL;
2151         struct slave *new_active = NULL;
2152         int res = 0;
2153
2154         if (!USES_PRIMARY(bond->params.mode)) {
2155                 return -EINVAL;
2156         }
2157
2158         /* Verify that master_dev is indeed the master of slave_dev */
2159         if (!(slave_dev->flags & IFF_SLAVE) ||
2160             (slave_dev->master != bond_dev)) {
2161                 return -EINVAL;
2162         }
2163
2164         read_lock(&bond->lock);
2165
2166         read_lock(&bond->curr_slave_lock);
2167         old_active = bond->curr_active_slave;
2168         read_unlock(&bond->curr_slave_lock);
2169
2170         new_active = bond_get_slave_by_dev(bond, slave_dev);
2171
2172         /*
2173          * Changing to the current active: do nothing; return success.
2174          */
2175         if (new_active && (new_active == old_active)) {
2176                 read_unlock(&bond->lock);
2177                 return 0;
2178         }
2179
2180         if ((new_active) &&
2181             (old_active) &&
2182             (new_active->link == BOND_LINK_UP) &&
2183             IS_UP(new_active->dev)) {
2184                 write_lock_bh(&bond->curr_slave_lock);
2185                 bond_change_active_slave(bond, new_active);
2186                 write_unlock_bh(&bond->curr_slave_lock);
2187         } else {
2188                 res = -EINVAL;
2189         }
2190
2191         read_unlock(&bond->lock);
2192
2193         return res;
2194 }
2195
2196 static int bond_info_query(struct net_device *bond_dev, struct ifbond *info)
2197 {
2198         struct bonding *bond = bond_dev->priv;
2199
2200         info->bond_mode = bond->params.mode;
2201         info->miimon = bond->params.miimon;
2202
2203         read_lock(&bond->lock);
2204         info->num_slaves = bond->slave_cnt;
2205         read_unlock(&bond->lock);
2206
2207         return 0;
2208 }
2209
2210 static int bond_slave_info_query(struct net_device *bond_dev, struct ifslave *info)
2211 {
2212         struct bonding *bond = bond_dev->priv;
2213         struct slave *slave;
2214         int i, found = 0;
2215
2216         if (info->slave_id < 0) {
2217                 return -ENODEV;
2218         }
2219
2220         read_lock(&bond->lock);
2221
2222         bond_for_each_slave(bond, slave, i) {
2223                 if (i == (int)info->slave_id) {
2224                         found = 1;
2225                         break;
2226                 }
2227         }
2228
2229         read_unlock(&bond->lock);
2230
2231         if (found) {
2232                 strcpy(info->slave_name, slave->dev->name);
2233                 info->link = slave->link;
2234                 info->state = slave->state;
2235                 info->link_failure_count = slave->link_failure_count;
2236         } else {
2237                 return -ENODEV;
2238         }
2239
2240         return 0;
2241 }
2242
2243 /*-------------------------------- Monitoring -------------------------------*/
2244
2245
2246 static int bond_miimon_inspect(struct bonding *bond)
2247 {
2248         struct slave *slave;
2249         int i, link_state, commit = 0;
2250
2251         bond_for_each_slave(bond, slave, i) {
2252                 slave->new_link = BOND_LINK_NOCHANGE;
2253
2254                 link_state = bond_check_dev_link(bond, slave->dev, 0);
2255
2256                 switch (slave->link) {
2257                 case BOND_LINK_UP:
2258                         if (link_state)
2259                                 continue;
2260
2261                         slave->link = BOND_LINK_FAIL;
2262                         slave->delay = bond->params.downdelay;
2263                         if (slave->delay) {
2264                                 printk(KERN_INFO DRV_NAME
2265                                        ": %s: link status down for %s"
2266                                        "interface %s, disabling it in %d ms.\n",
2267                                        bond->dev->name,
2268                                        (bond->params.mode ==
2269                                         BOND_MODE_ACTIVEBACKUP) ?
2270                                        ((slave->state == BOND_STATE_ACTIVE) ?
2271                                         "active " : "backup ") : "",
2272                                        slave->dev->name,
2273                                        bond->params.downdelay * bond->params.miimon);
2274                         }
2275                         /*FALLTHRU*/
2276                 case BOND_LINK_FAIL:
2277                         if (link_state) {
2278                                 /*
2279                                  * recovered before downdelay expired
2280                                  */
2281                                 slave->link = BOND_LINK_UP;
2282                                 slave->jiffies = jiffies;
2283                                 printk(KERN_INFO DRV_NAME
2284                                        ": %s: link status up again after %d "
2285                                        "ms for interface %s.\n",
2286                                        bond->dev->name,
2287                                        (bond->params.downdelay - slave->delay) *
2288                                        bond->params.miimon,
2289                                        slave->dev->name);
2290                                 continue;
2291                         }
2292
2293                         if (slave->delay <= 0) {
2294                                 slave->new_link = BOND_LINK_DOWN;
2295                                 commit++;
2296                                 continue;
2297                         }
2298
2299                         slave->delay--;
2300                         break;
2301
2302                 case BOND_LINK_DOWN:
2303                         if (!link_state)
2304                                 continue;
2305
2306                         slave->link = BOND_LINK_BACK;
2307                         slave->delay = bond->params.updelay;
2308
2309                         if (slave->delay) {
2310                                 printk(KERN_INFO DRV_NAME
2311                                        ": %s: link status up for "
2312                                        "interface %s, enabling it in %d ms.\n",
2313                                        bond->dev->name, slave->dev->name,
2314                                        bond->params.updelay *
2315                                        bond->params.miimon);
2316                         }
2317                         /*FALLTHRU*/
2318                 case BOND_LINK_BACK:
2319                         if (!link_state) {
2320                                 slave->link = BOND_LINK_DOWN;
2321                                 printk(KERN_INFO DRV_NAME
2322                                        ": %s: link status down again after %d "
2323                                        "ms for interface %s.\n",
2324                                        bond->dev->name,
2325                                        (bond->params.updelay - slave->delay) *
2326                                        bond->params.miimon,
2327                                        slave->dev->name);
2328
2329                                 continue;
2330                         }
2331
2332                         if (slave->delay <= 0) {
2333                                 slave->new_link = BOND_LINK_UP;
2334                                 commit++;
2335                                 continue;
2336                         }
2337
2338                         slave->delay--;
2339                         break;
2340                 }
2341         }
2342
2343         return commit;
2344 }
2345
2346 static void bond_miimon_commit(struct bonding *bond)
2347 {
2348         struct slave *slave;
2349         int i;
2350
2351         bond_for_each_slave(bond, slave, i) {
2352                 switch (slave->new_link) {
2353                 case BOND_LINK_NOCHANGE:
2354                         continue;
2355
2356                 case BOND_LINK_UP:
2357                         slave->link = BOND_LINK_UP;
2358                         slave->jiffies = jiffies;
2359
2360                         if (bond->params.mode == BOND_MODE_8023AD) {
2361                                 /* prevent it from being the active one */
2362                                 slave->state = BOND_STATE_BACKUP;
2363                         } else if (bond->params.mode != BOND_MODE_ACTIVEBACKUP) {
2364                                 /* make it immediately active */
2365                                 slave->state = BOND_STATE_ACTIVE;
2366                         } else if (slave != bond->primary_slave) {
2367                                 /* prevent it from being the active one */
2368                                 slave->state = BOND_STATE_BACKUP;
2369                         }
2370
2371                         printk(KERN_INFO DRV_NAME
2372                                ": %s: link status definitely "
2373                                "up for interface %s.\n",
2374                                bond->dev->name, slave->dev->name);
2375
2376                         /* notify ad that the link status has changed */
2377                         if (bond->params.mode == BOND_MODE_8023AD)
2378                                 bond_3ad_handle_link_change(slave, BOND_LINK_UP);
2379
2380                         if ((bond->params.mode == BOND_MODE_TLB) ||
2381                             (bond->params.mode == BOND_MODE_ALB))
2382                                 bond_alb_handle_link_change(bond, slave,
2383                                                             BOND_LINK_UP);
2384
2385                         if (!bond->curr_active_slave ||
2386                             (slave == bond->primary_slave))
2387                                 goto do_failover;
2388
2389                         continue;
2390
2391                 case BOND_LINK_DOWN:
2392                         if (slave->link_failure_count < UINT_MAX)
2393                                 slave->link_failure_count++;
2394
2395                         slave->link = BOND_LINK_DOWN;
2396
2397                         if (bond->params.mode == BOND_MODE_ACTIVEBACKUP ||
2398                             bond->params.mode == BOND_MODE_8023AD)
2399                                 bond_set_slave_inactive_flags(slave);
2400
2401                         printk(KERN_INFO DRV_NAME
2402                                ": %s: link status definitely down for "
2403                                "interface %s, disabling it\n",
2404                                bond->dev->name, slave->dev->name);
2405
2406                         if (bond->params.mode == BOND_MODE_8023AD)
2407                                 bond_3ad_handle_link_change(slave,
2408                                                             BOND_LINK_DOWN);
2409
2410                         if (bond->params.mode == BOND_MODE_TLB ||
2411                             bond->params.mode == BOND_MODE_ALB)
2412                                 bond_alb_handle_link_change(bond, slave,
2413                                                             BOND_LINK_DOWN);
2414
2415                         if (slave == bond->curr_active_slave)
2416                                 goto do_failover;
2417
2418                         continue;
2419
2420                 default:
2421                         printk(KERN_ERR DRV_NAME
2422                                ": %s: invalid new link %d on slave %s\n",
2423                                bond->dev->name, slave->new_link,
2424                                slave->dev->name);
2425                         slave->new_link = BOND_LINK_NOCHANGE;
2426
2427                         continue;
2428                 }
2429
2430 do_failover:
2431                 ASSERT_RTNL();
2432                 write_lock_bh(&bond->curr_slave_lock);
2433                 bond_select_active_slave(bond);
2434                 write_unlock_bh(&bond->curr_slave_lock);
2435         }
2436
2437         bond_set_carrier(bond);
2438 }
2439
2440 /*
2441  * bond_mii_monitor
2442  *
2443  * Really a wrapper that splits the mii monitor into two phases: an
2444  * inspection, then (if inspection indicates something needs to be done)
2445  * an acquisition of appropriate locks followed by a commit phase to
2446  * implement whatever link state changes are indicated.
2447  */
2448 void bond_mii_monitor(struct work_struct *work)
2449 {
2450         struct bonding *bond = container_of(work, struct bonding,
2451                                             mii_work.work);
2452
2453         read_lock(&bond->lock);
2454         if (bond->kill_timers)
2455                 goto out;
2456
2457         if (bond->slave_cnt == 0)
2458                 goto re_arm;
2459
2460         if (bond->send_grat_arp) {
2461                 read_lock(&bond->curr_slave_lock);
2462                 bond_send_gratuitous_arp(bond);
2463                 read_unlock(&bond->curr_slave_lock);
2464         }
2465
2466         if (bond_miimon_inspect(bond)) {
2467                 read_unlock(&bond->lock);
2468                 rtnl_lock();
2469                 read_lock(&bond->lock);
2470
2471                 bond_miimon_commit(bond);
2472
2473                 read_unlock(&bond->lock);
2474                 rtnl_unlock();  /* might sleep, hold no other locks */
2475                 read_lock(&bond->lock);
2476         }
2477
2478 re_arm:
2479         if (bond->params.miimon)
2480                 queue_delayed_work(bond->wq, &bond->mii_work,
2481                                    msecs_to_jiffies(bond->params.miimon));
2482 out:
2483         read_unlock(&bond->lock);
2484 }
2485
2486 static __be32 bond_glean_dev_ip(struct net_device *dev)
2487 {
2488         struct in_device *idev;
2489         struct in_ifaddr *ifa;
2490         __be32 addr = 0;
2491
2492         if (!dev)
2493                 return 0;
2494
2495         rcu_read_lock();
2496         idev = __in_dev_get_rcu(dev);
2497         if (!idev)
2498                 goto out;
2499
2500         ifa = idev->ifa_list;
2501         if (!ifa)
2502                 goto out;
2503
2504         addr = ifa->ifa_local;
2505 out:
2506         rcu_read_unlock();
2507         return addr;
2508 }
2509
2510 static int bond_has_this_ip(struct bonding *bond, __be32 ip)
2511 {
2512         struct vlan_entry *vlan;
2513
2514         if (ip == bond->master_ip)
2515                 return 1;
2516
2517         list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
2518                 if (ip == vlan->vlan_ip)
2519                         return 1;
2520         }
2521
2522         return 0;
2523 }
2524
2525 /*
2526  * We go to the (large) trouble of VLAN tagging ARP frames because
2527  * switches in VLAN mode (especially if ports are configured as
2528  * "native" to a VLAN) might not pass non-tagged frames.
2529  */
2530 static void bond_arp_send(struct net_device *slave_dev, int arp_op, __be32 dest_ip, __be32 src_ip, unsigned short vlan_id)
2531 {
2532         struct sk_buff *skb;
2533
2534         dprintk("arp %d on slave %s: dst %x src %x vid %d\n", arp_op,
2535                slave_dev->name, dest_ip, src_ip, vlan_id);
2536                
2537         skb = arp_create(arp_op, ETH_P_ARP, dest_ip, slave_dev, src_ip,
2538                          NULL, slave_dev->dev_addr, NULL);
2539
2540         if (!skb) {
2541                 printk(KERN_ERR DRV_NAME ": ARP packet allocation failed\n");
2542                 return;
2543         }
2544         if (vlan_id) {
2545                 skb = vlan_put_tag(skb, vlan_id);
2546                 if (!skb) {
2547                         printk(KERN_ERR DRV_NAME ": failed to insert VLAN tag\n");
2548                         return;
2549                 }
2550         }
2551         arp_xmit(skb);
2552 }
2553
2554
2555 static void bond_arp_send_all(struct bonding *bond, struct slave *slave)
2556 {
2557         int i, vlan_id, rv;
2558         __be32 *targets = bond->params.arp_targets;
2559         struct vlan_entry *vlan;
2560         struct net_device *vlan_dev;
2561         struct flowi fl;
2562         struct rtable *rt;
2563
2564         for (i = 0; (i < BOND_MAX_ARP_TARGETS); i++) {
2565                 if (!targets[i])
2566                         continue;
2567                 dprintk("basa: target %x\n", targets[i]);
2568                 if (list_empty(&bond->vlan_list)) {
2569                         dprintk("basa: empty vlan: arp_send\n");
2570                         bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2571                                       bond->master_ip, 0);
2572                         continue;
2573                 }
2574
2575                 /*
2576                  * If VLANs are configured, we do a route lookup to
2577                  * determine which VLAN interface would be used, so we
2578                  * can tag the ARP with the proper VLAN tag.
2579                  */
2580                 memset(&fl, 0, sizeof(fl));
2581                 fl.fl4_dst = targets[i];
2582                 fl.fl4_tos = RTO_ONLINK;
2583
2584                 rv = ip_route_output_key(&init_net, &rt, &fl);
2585                 if (rv) {
2586                         if (net_ratelimit()) {
2587                                 printk(KERN_WARNING DRV_NAME
2588                              ": %s: no route to arp_ip_target %pI4\n",
2589                                        bond->dev->name, &fl.fl4_dst);
2590                         }
2591                         continue;
2592                 }
2593
2594                 /*
2595                  * This target is not on a VLAN
2596                  */
2597                 if (rt->u.dst.dev == bond->dev) {
2598                         ip_rt_put(rt);
2599                         dprintk("basa: rtdev == bond->dev: arp_send\n");
2600                         bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2601                                       bond->master_ip, 0);
2602                         continue;
2603                 }
2604
2605                 vlan_id = 0;
2606                 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
2607                         vlan_dev = vlan_group_get_device(bond->vlgrp, vlan->vlan_id);
2608                         if (vlan_dev == rt->u.dst.dev) {
2609                                 vlan_id = vlan->vlan_id;
2610                                 dprintk("basa: vlan match on %s %d\n",
2611                                        vlan_dev->name, vlan_id);
2612                                 break;
2613                         }
2614                 }
2615
2616                 if (vlan_id) {
2617                         ip_rt_put(rt);
2618                         bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2619                                       vlan->vlan_ip, vlan_id);
2620                         continue;
2621                 }
2622
2623                 if (net_ratelimit()) {
2624                         printk(KERN_WARNING DRV_NAME
2625                ": %s: no path to arp_ip_target %pI4 via rt.dev %s\n",
2626                                bond->dev->name, &fl.fl4_dst,
2627                                rt->u.dst.dev ? rt->u.dst.dev->name : "NULL");
2628                 }
2629                 ip_rt_put(rt);
2630         }
2631 }
2632
2633 /*
2634  * Kick out a gratuitous ARP for an IP on the bonding master plus one
2635  * for each VLAN above us.
2636  *
2637  * Caller must hold curr_slave_lock for read or better
2638  */
2639 static void bond_send_gratuitous_arp(struct bonding *bond)
2640 {
2641         struct slave *slave = bond->curr_active_slave;
2642         struct vlan_entry *vlan;
2643         struct net_device *vlan_dev;
2644
2645         dprintk("bond_send_grat_arp: bond %s slave %s\n", bond->dev->name,
2646                                 slave ? slave->dev->name : "NULL");
2647
2648         if (!slave || !bond->send_grat_arp ||
2649             test_bit(__LINK_STATE_LINKWATCH_PENDING, &slave->dev->state))
2650                 return;
2651
2652         bond->send_grat_arp--;
2653
2654         if (bond->master_ip) {
2655                 bond_arp_send(slave->dev, ARPOP_REPLY, bond->master_ip,
2656                                 bond->master_ip, 0);
2657         }
2658
2659         list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
2660                 vlan_dev = vlan_group_get_device(bond->vlgrp, vlan->vlan_id);
2661                 if (vlan->vlan_ip) {
2662                         bond_arp_send(slave->dev, ARPOP_REPLY, vlan->vlan_ip,
2663                                       vlan->vlan_ip, vlan->vlan_id);
2664                 }
2665         }
2666 }
2667
2668 static void bond_validate_arp(struct bonding *bond, struct slave *slave, __be32 sip, __be32 tip)
2669 {
2670         int i;
2671         __be32 *targets = bond->params.arp_targets;
2672
2673         targets = bond->params.arp_targets;
2674         for (i = 0; (i < BOND_MAX_ARP_TARGETS) && targets[i]; i++) {
2675                 dprintk("bva: sip %pI4 tip %pI4 t[%d] %pI4 bhti(tip) %d\n",
2676                         &sip, &tip, i, &targets[i], bond_has_this_ip(bond, tip));
2677                 if (sip == targets[i]) {
2678                         if (bond_has_this_ip(bond, tip))
2679                                 slave->last_arp_rx = jiffies;
2680                         return;
2681                 }
2682         }
2683 }
2684
2685 static int bond_arp_rcv(struct sk_buff *skb, struct net_device *dev, struct packet_type *pt, struct net_device *orig_dev)
2686 {
2687         struct arphdr *arp;
2688         struct slave *slave;
2689         struct bonding *bond;
2690         unsigned char *arp_ptr;
2691         __be32 sip, tip;
2692
2693         if (dev_net(dev) != &init_net)
2694                 goto out;
2695
2696         if (!(dev->priv_flags & IFF_BONDING) || !(dev->flags & IFF_MASTER))
2697                 goto out;
2698
2699         bond = dev->priv;
2700         read_lock(&bond->lock);
2701
2702         dprintk("bond_arp_rcv: bond %s skb->dev %s orig_dev %s\n",
2703                 bond->dev->name, skb->dev ? skb->dev->name : "NULL",
2704                 orig_dev ? orig_dev->name : "NULL");
2705
2706         slave = bond_get_slave_by_dev(bond, orig_dev);
2707         if (!slave || !slave_do_arp_validate(bond, slave))
2708                 goto out_unlock;
2709
2710         if (!pskb_may_pull(skb, arp_hdr_len(dev)))
2711                 goto out_unlock;
2712
2713         arp = arp_hdr(skb);
2714         if (arp->ar_hln != dev->addr_len ||
2715             skb->pkt_type == PACKET_OTHERHOST ||
2716             skb->pkt_type == PACKET_LOOPBACK ||
2717             arp->ar_hrd != htons(ARPHRD_ETHER) ||
2718             arp->ar_pro != htons(ETH_P_IP) ||
2719             arp->ar_pln != 4)
2720                 goto out_unlock;
2721
2722         arp_ptr = (unsigned char *)(arp + 1);
2723         arp_ptr += dev->addr_len;
2724         memcpy(&sip, arp_ptr, 4);
2725         arp_ptr += 4 + dev->addr_len;
2726         memcpy(&tip, arp_ptr, 4);
2727
2728         dprintk("bond_arp_rcv: %s %s/%d av %d sv %d sip %pI4 tip %pI4\n",
2729                 bond->dev->name, slave->dev->name, slave->state,
2730                 bond->params.arp_validate, slave_do_arp_validate(bond, slave),
2731                 &sip, &tip);
2732
2733         /*
2734          * Backup slaves won't see the ARP reply, but do come through
2735          * here for each ARP probe (so we swap the sip/tip to validate
2736          * the probe).  In a "redundant switch, common router" type of
2737          * configuration, the ARP probe will (hopefully) travel from
2738          * the active, through one switch, the router, then the other
2739          * switch before reaching the backup.
2740          */
2741         if (slave->state == BOND_STATE_ACTIVE)
2742                 bond_validate_arp(bond, slave, sip, tip);
2743         else
2744                 bond_validate_arp(bond, slave, tip, sip);
2745
2746 out_unlock:
2747         read_unlock(&bond->lock);
2748 out:
2749         dev_kfree_skb(skb);
2750         return NET_RX_SUCCESS;
2751 }
2752
2753 /*
2754  * this function is called regularly to monitor each slave's link
2755  * ensuring that traffic is being sent and received when arp monitoring
2756  * is used in load-balancing mode. if the adapter has been dormant, then an
2757  * arp is transmitted to generate traffic. see activebackup_arp_monitor for
2758  * arp monitoring in active backup mode.
2759  */
2760 void bond_loadbalance_arp_mon(struct work_struct *work)
2761 {
2762         struct bonding *bond = container_of(work, struct bonding,
2763                                             arp_work.work);
2764         struct slave *slave, *oldcurrent;
2765         int do_failover = 0;
2766         int delta_in_ticks;
2767         int i;
2768
2769         read_lock(&bond->lock);
2770
2771         delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
2772
2773         if (bond->kill_timers) {
2774                 goto out;
2775         }
2776
2777         if (bond->slave_cnt == 0) {
2778                 goto re_arm;
2779         }
2780
2781         read_lock(&bond->curr_slave_lock);
2782         oldcurrent = bond->curr_active_slave;
2783         read_unlock(&bond->curr_slave_lock);
2784
2785         /* see if any of the previous devices are up now (i.e. they have
2786          * xmt and rcv traffic). the curr_active_slave does not come into
2787          * the picture unless it is null. also, slave->jiffies is not needed
2788          * here because we send an arp on each slave and give a slave as
2789          * long as it needs to get the tx/rx within the delta.
2790          * TODO: what about up/down delay in arp mode? it wasn't here before
2791          *       so it can wait
2792          */
2793         bond_for_each_slave(bond, slave, i) {
2794                 if (slave->link != BOND_LINK_UP) {
2795                         if (time_before_eq(jiffies, slave->dev->trans_start + delta_in_ticks) &&
2796                             time_before_eq(jiffies, slave->dev->last_rx + delta_in_ticks)) {
2797
2798                                 slave->link  = BOND_LINK_UP;
2799                                 slave->state = BOND_STATE_ACTIVE;
2800
2801                                 /* primary_slave has no meaning in round-robin
2802                                  * mode. the window of a slave being up and
2803                                  * curr_active_slave being null after enslaving
2804                                  * is closed.
2805                                  */
2806                                 if (!oldcurrent) {
2807                                         printk(KERN_INFO DRV_NAME
2808                                                ": %s: link status definitely "
2809                                                "up for interface %s, ",
2810                                                bond->dev->name,
2811                                                slave->dev->name);
2812                                         do_failover = 1;
2813                                 } else {
2814                                         printk(KERN_INFO DRV_NAME
2815                                                ": %s: interface %s is now up\n",
2816                                                bond->dev->name,
2817                                                slave->dev->name);
2818                                 }
2819                         }
2820                 } else {
2821                         /* slave->link == BOND_LINK_UP */
2822
2823                         /* not all switches will respond to an arp request
2824                          * when the source ip is 0, so don't take the link down
2825                          * if we don't know our ip yet
2826                          */
2827                         if (time_after_eq(jiffies, slave->dev->trans_start + 2*delta_in_ticks) ||
2828                             (time_after_eq(jiffies, slave->dev->last_rx + 2*delta_in_ticks))) {
2829
2830                                 slave->link  = BOND_LINK_DOWN;
2831                                 slave->state = BOND_STATE_BACKUP;
2832
2833                                 if (slave->link_failure_count < UINT_MAX) {
2834                                         slave->link_failure_count++;
2835                                 }
2836
2837                                 printk(KERN_INFO DRV_NAME
2838                                        ": %s: interface %s is now down.\n",
2839                                        bond->dev->name,
2840                                        slave->dev->name);
2841
2842                                 if (slave == oldcurrent) {
2843                                         do_failover = 1;
2844                                 }
2845                         }
2846                 }
2847
2848                 /* note: if switch is in round-robin mode, all links
2849                  * must tx arp to ensure all links rx an arp - otherwise
2850                  * links may oscillate or not come up at all; if switch is
2851                  * in something like xor mode, there is nothing we can
2852                  * do - all replies will be rx'ed on same link causing slaves
2853                  * to be unstable during low/no traffic periods
2854                  */
2855                 if (IS_UP(slave->dev)) {
2856                         bond_arp_send_all(bond, slave);
2857                 }
2858         }
2859
2860         if (do_failover) {
2861                 write_lock_bh(&bond->curr_slave_lock);
2862
2863                 bond_select_active_slave(bond);
2864
2865                 write_unlock_bh(&bond->curr_slave_lock);
2866         }
2867
2868 re_arm:
2869         if (bond->params.arp_interval)
2870                 queue_delayed_work(bond->wq, &bond->arp_work, delta_in_ticks);
2871 out:
2872         read_unlock(&bond->lock);
2873 }
2874
2875 /*
2876  * Called to inspect slaves for active-backup mode ARP monitor link state
2877  * changes.  Sets new_link in slaves to specify what action should take
2878  * place for the slave.  Returns 0 if no changes are found, >0 if changes
2879  * to link states must be committed.
2880  *
2881  * Called with bond->lock held for read.
2882  */
2883 static int bond_ab_arp_inspect(struct bonding *bond, int delta_in_ticks)
2884 {
2885         struct slave *slave;
2886         int i, commit = 0;
2887
2888         bond_for_each_slave(bond, slave, i) {
2889                 slave->new_link = BOND_LINK_NOCHANGE;
2890
2891                 if (slave->link != BOND_LINK_UP) {
2892                         if (time_before_eq(jiffies, slave_last_rx(bond, slave) +
2893                                            delta_in_ticks)) {
2894                                 slave->new_link = BOND_LINK_UP;
2895                                 commit++;
2896                         }
2897
2898                         continue;
2899                 }
2900
2901                 /*
2902                  * Give slaves 2*delta after being enslaved or made
2903                  * active.  This avoids bouncing, as the last receive
2904                  * times need a full ARP monitor cycle to be updated.
2905                  */
2906                 if (!time_after_eq(jiffies, slave->jiffies +
2907                                    2 * delta_in_ticks))
2908                         continue;
2909
2910                 /*
2911                  * Backup slave is down if:
2912                  * - No current_arp_slave AND
2913                  * - more than 3*delta since last receive AND
2914                  * - the bond has an IP address
2915                  *
2916                  * Note: a non-null current_arp_slave indicates
2917                  * the curr_active_slave went down and we are
2918                  * searching for a new one; under this condition
2919                  * we only take the curr_active_slave down - this
2920                  * gives each slave a chance to tx/rx traffic
2921                  * before being taken out
2922                  */
2923                 if (slave->state == BOND_STATE_BACKUP &&
2924                     !bond->current_arp_slave &&
2925                     time_after(jiffies, slave_last_rx(bond, slave) +
2926                                3 * delta_in_ticks)) {
2927                         slave->new_link = BOND_LINK_DOWN;
2928                         commit++;
2929                 }
2930
2931                 /*
2932                  * Active slave is down if:
2933                  * - more than 2*delta since transmitting OR
2934                  * - (more than 2*delta since receive AND
2935                  *    the bond has an IP address)
2936                  */
2937                 if ((slave->state == BOND_STATE_ACTIVE) &&
2938                     (time_after_eq(jiffies, slave->dev->trans_start +
2939                                     2 * delta_in_ticks) ||
2940                       (time_after_eq(jiffies, slave_last_rx(bond, slave)
2941                                      + 2 * delta_in_ticks)))) {
2942                         slave->new_link = BOND_LINK_DOWN;
2943                         commit++;
2944                 }
2945         }
2946
2947         read_lock(&bond->curr_slave_lock);
2948
2949         /*
2950          * Trigger a commit if the primary option setting has changed.
2951          */
2952         if (bond->primary_slave &&
2953             (bond->primary_slave != bond->curr_active_slave) &&
2954             (bond->primary_slave->link == BOND_LINK_UP))
2955                 commit++;
2956
2957         read_unlock(&bond->curr_slave_lock);
2958
2959         return commit;
2960 }
2961
2962 /*
2963  * Called to commit link state changes noted by inspection step of
2964  * active-backup mode ARP monitor.
2965  *
2966  * Called with RTNL and bond->lock for read.
2967  */
2968 static void bond_ab_arp_commit(struct bonding *bond, int delta_in_ticks)
2969 {
2970         struct slave *slave;
2971         int i;
2972
2973         bond_for_each_slave(bond, slave, i) {
2974                 switch (slave->new_link) {
2975                 case BOND_LINK_NOCHANGE:
2976                         continue;
2977
2978                 case BOND_LINK_UP:
2979                         write_lock_bh(&bond->curr_slave_lock);
2980
2981                         if (!bond->curr_active_slave &&
2982                             time_before_eq(jiffies, slave->dev->trans_start +
2983                                            delta_in_ticks)) {
2984                                 slave->link = BOND_LINK_UP;
2985                                 bond_change_active_slave(bond, slave);
2986                                 bond->current_arp_slave = NULL;
2987
2988                                 printk(KERN_INFO DRV_NAME
2989                                        ": %s: %s is up and now the "
2990                                        "active interface\n",
2991                                        bond->dev->name, slave->dev->name);
2992
2993                         } else if (bond->curr_active_slave != slave) {
2994                                 /* this slave has just come up but we
2995                                  * already have a current slave; this can
2996                                  * also happen if bond_enslave adds a new
2997                                  * slave that is up while we are searching
2998                                  * for a new slave
2999                                  */
3000                                 slave->link = BOND_LINK_UP;
3001                                 bond_set_slave_inactive_flags(slave);
3002                                 bond->current_arp_slave = NULL;
3003
3004                                 printk(KERN_INFO DRV_NAME
3005                                        ": %s: backup interface %s is now up\n",
3006                                        bond->dev->name, slave->dev->name);
3007                         }
3008
3009                         write_unlock_bh(&bond->curr_slave_lock);
3010
3011                         break;
3012
3013                 case BOND_LINK_DOWN:
3014                         if (slave->link_failure_count < UINT_MAX)
3015                                 slave->link_failure_count++;
3016
3017                         slave->link = BOND_LINK_DOWN;
3018
3019                         if (slave == bond->curr_active_slave) {
3020                                 printk(KERN_INFO DRV_NAME
3021                                        ": %s: link status down for active "
3022                                        "interface %s, disabling it\n",
3023                                        bond->dev->name, slave->dev->name);
3024
3025                                 bond_set_slave_inactive_flags(slave);
3026
3027                                 write_lock_bh(&bond->curr_slave_lock);
3028
3029                                 bond_select_active_slave(bond);
3030                                 if (bond->curr_active_slave)
3031                                         bond->curr_active_slave->jiffies =
3032                                                 jiffies;
3033
3034                                 write_unlock_bh(&bond->curr_slave_lock);
3035
3036                                 bond->current_arp_slave = NULL;
3037
3038                         } else if (slave->state == BOND_STATE_BACKUP) {
3039                                 printk(KERN_INFO DRV_NAME
3040                                        ": %s: backup interface %s is now down\n",
3041                                        bond->dev->name, slave->dev->name);
3042
3043                                 bond_set_slave_inactive_flags(slave);
3044                         }
3045                         break;
3046
3047                 default:
3048                         printk(KERN_ERR DRV_NAME
3049                                ": %s: impossible: new_link %d on slave %s\n",
3050                                bond->dev->name, slave->new_link,
3051                                slave->dev->name);
3052                 }
3053         }
3054
3055         /*
3056          * No race with changes to primary via sysfs, as we hold rtnl.
3057          */
3058         if (bond->primary_slave &&
3059             (bond->primary_slave != bond->curr_active_slave) &&
3060             (bond->primary_slave->link == BOND_LINK_UP)) {
3061                 write_lock_bh(&bond->curr_slave_lock);
3062                 bond_change_active_slave(bond, bond->primary_slave);
3063                 write_unlock_bh(&bond->curr_slave_lock);
3064         }
3065
3066         bond_set_carrier(bond);
3067 }
3068
3069 /*
3070  * Send ARP probes for active-backup mode ARP monitor.
3071  *
3072  * Called with bond->lock held for read.
3073  */
3074 static void bond_ab_arp_probe(struct bonding *bond)
3075 {
3076         struct slave *slave;
3077         int i;
3078
3079         read_lock(&bond->curr_slave_lock);
3080
3081         if (bond->current_arp_slave && bond->curr_active_slave)
3082                 printk("PROBE: c_arp %s && cas %s BAD\n",
3083                        bond->current_arp_slave->dev->name,
3084                        bond->curr_active_slave->dev->name);
3085
3086         if (bond->curr_active_slave) {
3087                 bond_arp_send_all(bond, bond->curr_active_slave);
3088                 read_unlock(&bond->curr_slave_lock);
3089                 return;
3090         }
3091
3092         read_unlock(&bond->curr_slave_lock);
3093
3094         /* if we don't have a curr_active_slave, search for the next available
3095          * backup slave from the current_arp_slave and make it the candidate
3096          * for becoming the curr_active_slave
3097          */
3098
3099         if (!bond->current_arp_slave) {
3100                 bond->current_arp_slave = bond->first_slave;
3101                 if (!bond->current_arp_slave)
3102                         return;
3103         }
3104
3105         bond_set_slave_inactive_flags(bond->current_arp_slave);
3106
3107         /* search for next candidate */
3108         bond_for_each_slave_from(bond, slave, i, bond->current_arp_slave->next) {
3109                 if (IS_UP(slave->dev)) {
3110                         slave->link = BOND_LINK_BACK;
3111                         bond_set_slave_active_flags(slave);
3112                         bond_arp_send_all(bond, slave);
3113                         slave->jiffies = jiffies;
3114                         bond->current_arp_slave = slave;
3115                         break;
3116                 }
3117
3118                 /* if the link state is up at this point, we
3119                  * mark it down - this can happen if we have
3120                  * simultaneous link failures and
3121                  * reselect_active_interface doesn't make this
3122                  * one the current slave so it is still marked
3123                  * up when it is actually down
3124                  */
3125                 if (slave->link == BOND_LINK_UP) {
3126                         slave->link = BOND_LINK_DOWN;
3127                         if (slave->link_failure_count < UINT_MAX)
3128                                 slave->link_failure_count++;
3129
3130                         bond_set_slave_inactive_flags(slave);
3131
3132                         printk(KERN_INFO DRV_NAME
3133                                ": %s: backup interface %s is now down.\n",
3134                                bond->dev->name, slave->dev->name);
3135                 }
3136         }
3137 }
3138
3139 void bond_activebackup_arp_mon(struct work_struct *work)
3140 {
3141         struct bonding *bond = container_of(work, struct bonding,
3142                                             arp_work.work);
3143         int delta_in_ticks;
3144
3145         read_lock(&bond->lock);
3146
3147         if (bond->kill_timers)
3148                 goto out;
3149
3150         delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
3151
3152         if (bond->slave_cnt == 0)
3153                 goto re_arm;
3154
3155         if (bond->send_grat_arp) {
3156                 read_lock(&bond->curr_slave_lock);
3157                 bond_send_gratuitous_arp(bond);
3158                 read_unlock(&bond->curr_slave_lock);
3159         }
3160
3161         if (bond_ab_arp_inspect(bond, delta_in_ticks)) {
3162                 read_unlock(&bond->lock);
3163                 rtnl_lock();
3164                 read_lock(&bond->lock);
3165
3166                 bond_ab_arp_commit(bond, delta_in_ticks);
3167
3168                 read_unlock(&bond->lock);
3169                 rtnl_unlock();
3170                 read_lock(&bond->lock);
3171         }
3172
3173         bond_ab_arp_probe(bond);
3174
3175 re_arm:
3176         if (bond->params.arp_interval) {
3177                 queue_delayed_work(bond->wq, &bond->arp_work, delta_in_ticks);
3178         }
3179 out:
3180         read_unlock(&bond->lock);
3181 }
3182
3183 /*------------------------------ proc/seq_file-------------------------------*/
3184
3185 #ifdef CONFIG_PROC_FS
3186
3187 static void *bond_info_seq_start(struct seq_file *seq, loff_t *pos)
3188 {
3189         struct bonding *bond = seq->private;
3190         loff_t off = 0;
3191         struct slave *slave;
3192         int i;
3193
3194         /* make sure the bond won't be taken away */
3195         read_lock(&dev_base_lock);
3196         read_lock(&bond->lock);
3197
3198         if (*pos == 0) {
3199                 return SEQ_START_TOKEN;
3200         }
3201
3202         bond_for_each_slave(bond, slave, i) {
3203                 if (++off == *pos) {
3204                         return slave;
3205                 }
3206         }
3207
3208         return NULL;
3209 }
3210
3211 static void *bond_info_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3212 {
3213         struct bonding *bond = seq->private;
3214         struct slave *slave = v;
3215
3216         ++*pos;
3217         if (v == SEQ_START_TOKEN) {
3218                 return bond->first_slave;
3219         }
3220
3221         slave = slave->next;
3222
3223         return (slave == bond->first_slave) ? NULL : slave;
3224 }
3225
3226 static void bond_info_seq_stop(struct seq_file *seq, void *v)
3227 {
3228         struct bonding *bond = seq->private;
3229
3230         read_unlock(&bond->lock);
3231         read_unlock(&dev_base_lock);
3232 }
3233
3234 static void bond_info_show_master(struct seq_file *seq)
3235 {
3236         struct bonding *bond = seq->private;
3237         struct slave *curr;
3238         int i;
3239
3240         read_lock(&bond->curr_slave_lock);
3241         curr = bond->curr_active_slave;
3242         read_unlock(&bond->curr_slave_lock);
3243
3244         seq_printf(seq, "Bonding Mode: %s",
3245                    bond_mode_name(bond->params.mode));
3246
3247         if (bond->params.mode == BOND_MODE_ACTIVEBACKUP &&
3248             bond->params.fail_over_mac)
3249                 seq_printf(seq, " (fail_over_mac %s)",
3250                    fail_over_mac_tbl[bond->params.fail_over_mac].modename);
3251
3252         seq_printf(seq, "\n");
3253
3254         if (bond->params.mode == BOND_MODE_XOR ||
3255                 bond->params.mode == BOND_MODE_8023AD) {
3256                 seq_printf(seq, "Transmit Hash Policy: %s (%d)\n",
3257                         xmit_hashtype_tbl[bond->params.xmit_policy].modename,
3258                         bond->params.xmit_policy);
3259         }
3260
3261         if (USES_PRIMARY(bond->params.mode)) {
3262                 seq_printf(seq, "Primary Slave: %s\n",
3263                            (bond->primary_slave) ?
3264                            bond->primary_slave->dev->name : "None");
3265
3266                 seq_printf(seq, "Currently Active Slave: %s\n",
3267                            (curr) ? curr->dev->name : "None");
3268         }
3269
3270         seq_printf(seq, "MII Status: %s\n", netif_carrier_ok(bond->dev) ?
3271                    "up" : "down");
3272         seq_printf(seq, "MII Polling Interval (ms): %d\n", bond->params.miimon);
3273         seq_printf(seq, "Up Delay (ms): %d\n",
3274                    bond->params.updelay * bond->params.miimon);
3275         seq_printf(seq, "Down Delay (ms): %d\n",
3276                    bond->params.downdelay * bond->params.miimon);
3277
3278
3279         /* ARP information */
3280         if(bond->params.arp_interval > 0) {
3281                 int printed=0;
3282                 seq_printf(seq, "ARP Polling Interval (ms): %d\n",
3283                                 bond->params.arp_interval);
3284
3285                 seq_printf(seq, "ARP IP target/s (n.n.n.n form):");
3286
3287                 for(i = 0; (i < BOND_MAX_ARP_TARGETS) ;i++) {
3288                         if (!bond->params.arp_targets[i])
3289                                 continue;
3290                         if (printed)
3291                                 seq_printf(seq, ",");
3292                         seq_printf(seq, " %pI4", &bond->params.arp_targets[i]);
3293                         printed = 1;
3294                 }
3295                 seq_printf(seq, "\n");
3296         }
3297
3298         if (bond->params.mode == BOND_MODE_8023AD) {
3299                 struct ad_info ad_info;
3300
3301                 seq_puts(seq, "\n802.3ad info\n");
3302                 seq_printf(seq, "LACP rate: %s\n",
3303                            (bond->params.lacp_fast) ? "fast" : "slow");
3304
3305                 if (bond_3ad_get_active_agg_info(bond, &ad_info)) {
3306                         seq_printf(seq, "bond %s has no active aggregator\n",
3307                                    bond->dev->name);
3308                 } else {
3309                         seq_printf(seq, "Active Aggregator Info:\n");
3310
3311                         seq_printf(seq, "\tAggregator ID: %d\n",
3312                                    ad_info.aggregator_id);
3313                         seq_printf(seq, "\tNumber of ports: %d\n",
3314                                    ad_info.ports);
3315                         seq_printf(seq, "\tActor Key: %d\n",
3316                                    ad_info.actor_key);
3317                         seq_printf(seq, "\tPartner Key: %d\n",
3318                                    ad_info.partner_key);
3319                         seq_printf(seq, "\tPartner Mac Address: %pM\n",
3320                                    ad_info.partner_system);
3321                 }
3322         }
3323 }
3324
3325 static void bond_info_show_slave(struct seq_file *seq, const struct slave *slave)
3326 {
3327         struct bonding *bond = seq->private;
3328
3329         seq_printf(seq, "\nSlave Interface: %s\n", slave->dev->name);
3330         seq_printf(seq, "MII Status: %s\n",
3331                    (slave->link == BOND_LINK_UP) ?  "up" : "down");
3332         seq_printf(seq, "Link Failure Count: %u\n",
3333                    slave->link_failure_count);
3334
3335         seq_printf(seq, "Permanent HW addr: %pM\n", slave->perm_hwaddr);
3336
3337         if (bond->params.mode == BOND_MODE_8023AD) {
3338                 const struct aggregator *agg
3339                         = SLAVE_AD_INFO(slave).port.aggregator;
3340
3341                 if (agg) {
3342                         seq_printf(seq, "Aggregator ID: %d\n",
3343                                    agg->aggregator_identifier);
3344                 } else {
3345                         seq_puts(seq, "Aggregator ID: N/A\n");
3346                 }
3347         }
3348 }
3349
3350 static int bond_info_seq_show(struct seq_file *seq, void *v)
3351 {
3352         if (v == SEQ_START_TOKEN) {
3353                 seq_printf(seq, "%s\n", version);
3354                 bond_info_show_master(seq);
3355         } else {
3356                 bond_info_show_slave(seq, v);
3357         }
3358
3359         return 0;
3360 }
3361
3362 static struct seq_operations bond_info_seq_ops = {
3363         .start = bond_info_seq_start,
3364         .next  = bond_info_seq_next,
3365         .stop  = bond_info_seq_stop,
3366         .show  = bond_info_seq_show,
3367 };
3368
3369 static int bond_info_open(struct inode *inode, struct file *file)
3370 {
3371         struct seq_file *seq;
3372         struct proc_dir_entry *proc;
3373         int res;
3374
3375         res = seq_open(file, &bond_info_seq_ops);
3376         if (!res) {
3377                 /* recover the pointer buried in proc_dir_entry data */
3378                 seq = file->private_data;
3379                 proc = PDE(inode);
3380                 seq->private = proc->data;
3381         }
3382
3383         return res;
3384 }
3385
3386 static const struct file_operations bond_info_fops = {
3387         .owner   = THIS_MODULE,
3388         .open    = bond_info_open,
3389         .read    = seq_read,
3390         .llseek  = seq_lseek,
3391         .release = seq_release,
3392 };
3393
3394 static int bond_create_proc_entry(struct bonding *bond)
3395 {
3396         struct net_device *bond_dev = bond->dev;
3397
3398         if (bond_proc_dir) {
3399                 bond->proc_entry = proc_create_data(bond_dev->name,
3400                                                     S_IRUGO, bond_proc_dir,
3401                                                     &bond_info_fops, bond);
3402                 if (bond->proc_entry == NULL) {
3403                         printk(KERN_WARNING DRV_NAME
3404                                ": Warning: Cannot create /proc/net/%s/%s\n",
3405                                DRV_NAME, bond_dev->name);
3406                 } else {
3407                         memcpy(bond->proc_file_name, bond_dev->name, IFNAMSIZ);
3408                 }
3409         }
3410
3411         return 0;
3412 }
3413
3414 static void bond_remove_proc_entry(struct bonding *bond)
3415 {
3416         if (bond_proc_dir && bond->proc_entry) {
3417                 remove_proc_entry(bond->proc_file_name, bond_proc_dir);
3418                 memset(bond->proc_file_name, 0, IFNAMSIZ);
3419                 bond->proc_entry = NULL;
3420         }
3421 }
3422
3423 /* Create the bonding directory under /proc/net, if doesn't exist yet.
3424  * Caller must hold rtnl_lock.
3425  */
3426 static void bond_create_proc_dir(void)
3427 {
3428         int len = strlen(DRV_NAME);
3429
3430         for (bond_proc_dir = init_net.proc_net->subdir; bond_proc_dir;
3431              bond_proc_dir = bond_proc_dir->next) {
3432                 if ((bond_proc_dir->namelen == len) &&
3433                     !memcmp(bond_proc_dir->name, DRV_NAME, len)) {
3434                         break;
3435                 }
3436         }
3437
3438         if (!bond_proc_dir) {
3439                 bond_proc_dir = proc_mkdir(DRV_NAME, init_net.proc_net);
3440                 if (bond_proc_dir) {
3441                         bond_proc_dir->owner = THIS_MODULE;
3442                 } else {
3443                         printk(KERN_WARNING DRV_NAME
3444                                 ": Warning: cannot create /proc/net/%s\n",
3445                                 DRV_NAME);
3446                 }
3447         }
3448 }
3449
3450 /* Destroy the bonding directory under /proc/net, if empty.
3451  * Caller must hold rtnl_lock.
3452  */
3453 static void bond_destroy_proc_dir(void)
3454 {
3455         struct proc_dir_entry *de;
3456
3457         if (!bond_proc_dir) {
3458                 return;
3459         }
3460
3461         /* verify that the /proc dir is empty */
3462         for (de = bond_proc_dir->subdir; de; de = de->next) {
3463                 /* ignore . and .. */
3464                 if (*(de->name) != '.') {
3465                         break;
3466                 }
3467         }
3468
3469         if (de) {
3470                 if (bond_proc_dir->owner == THIS_MODULE) {
3471                         bond_proc_dir->owner = NULL;
3472                 }
3473         } else {
3474                 remove_proc_entry(DRV_NAME, init_net.proc_net);
3475                 bond_proc_dir = NULL;
3476         }
3477 }
3478 #endif /* CONFIG_PROC_FS */
3479
3480 /*-------------------------- netdev event handling --------------------------*/
3481
3482 /*
3483  * Change device name
3484  */
3485 static int bond_event_changename(struct bonding *bond)
3486 {
3487 #ifdef CONFIG_PROC_FS
3488         bond_remove_proc_entry(bond);
3489         bond_create_proc_entry(bond);
3490 #endif
3491         down_write(&(bonding_rwsem));
3492         bond_destroy_sysfs_entry(bond);
3493         bond_create_sysfs_entry(bond);
3494         up_write(&(bonding_rwsem));
3495         return NOTIFY_DONE;
3496 }
3497
3498 static int bond_master_netdev_event(unsigned long event, struct net_device *bond_dev)
3499 {
3500         struct bonding *event_bond = bond_dev->priv;
3501
3502         switch (event) {
3503         case NETDEV_CHANGENAME:
3504                 return bond_event_changename(event_bond);
3505         case NETDEV_UNREGISTER:
3506                 bond_release_all(event_bond->dev);
3507                 break;
3508         default:
3509                 break;
3510         }
3511
3512         return NOTIFY_DONE;
3513 }
3514
3515 static int bond_slave_netdev_event(unsigned long event, struct net_device *slave_dev)
3516 {
3517         struct net_device *bond_dev = slave_dev->master;
3518         struct bonding *bond = bond_dev->priv;
3519
3520         switch (event) {
3521         case NETDEV_UNREGISTER:
3522                 if (bond_dev) {
3523                         if (bond->setup_by_slave)
3524                                 bond_release_and_destroy(bond_dev, slave_dev);
3525                         else
3526                                 bond_release(bond_dev, slave_dev);
3527                 }
3528                 break;
3529         case NETDEV_CHANGE:
3530                 /*
3531                  * TODO: is this what we get if somebody
3532                  * sets up a hierarchical bond, then rmmod's
3533                  * one of the slave bonding devices?
3534                  */
3535                 break;
3536         case NETDEV_DOWN:
3537                 /*
3538                  * ... Or is it this?
3539                  */
3540                 break;
3541         case NETDEV_CHANGEMTU:
3542                 /*
3543                  * TODO: Should slaves be allowed to
3544                  * independently alter their MTU?  For
3545                  * an active-backup bond, slaves need
3546                  * not be the same type of device, so
3547                  * MTUs may vary.  For other modes,
3548                  * slaves arguably should have the
3549                  * same MTUs. To do this, we'd need to
3550                  * take over the slave's change_mtu
3551                  * function for the duration of their
3552                  * servitude.
3553                  */
3554                 break;
3555         case NETDEV_CHANGENAME:
3556                 /*
3557                  * TODO: handle changing the primary's name
3558                  */
3559                 break;
3560         case NETDEV_FEAT_CHANGE:
3561                 bond_compute_features(bond);
3562                 break;
3563         default:
3564                 break;
3565         }
3566
3567         return NOTIFY_DONE;
3568 }
3569
3570 /*
3571  * bond_netdev_event: handle netdev notifier chain events.
3572  *
3573  * This function receives events for the netdev chain.  The caller (an
3574  * ioctl handler calling blocking_notifier_call_chain) holds the necessary
3575  * locks for us to safely manipulate the slave devices (RTNL lock,
3576  * dev_probe_lock).
3577  */
3578 static int bond_netdev_event(struct notifier_block *this, unsigned long event, void *ptr)
3579 {
3580         struct net_device *event_dev = (struct net_device *)ptr;
3581
3582         if (dev_net(event_dev) != &init_net)
3583                 return NOTIFY_DONE;
3584
3585         dprintk("event_dev: %s, event: %lx\n",
3586                 (event_dev ? event_dev->name : "None"),
3587                 event);
3588
3589         if (!(event_dev->priv_flags & IFF_BONDING))
3590                 return NOTIFY_DONE;
3591
3592         if (event_dev->flags & IFF_MASTER) {
3593                 dprintk("IFF_MASTER\n");
3594                 return bond_master_netdev_event(event, event_dev);
3595         }
3596
3597         if (event_dev->flags & IFF_SLAVE) {
3598                 dprintk("IFF_SLAVE\n");
3599                 return bond_slave_netdev_event(event, event_dev);
3600         }
3601
3602         return NOTIFY_DONE;
3603 }
3604
3605 /*
3606  * bond_inetaddr_event: handle inetaddr notifier chain events.
3607  *
3608  * We keep track of device IPs primarily to use as source addresses in
3609  * ARP monitor probes (rather than spewing out broadcasts all the time).
3610  *
3611  * We track one IP for the main device (if it has one), plus one per VLAN.
3612  */
3613 static int bond_inetaddr_event(struct notifier_block *this, unsigned long event, void *ptr)
3614 {
3615         struct in_ifaddr *ifa = ptr;
3616         struct net_device *vlan_dev, *event_dev = ifa->ifa_dev->dev;
3617         struct bonding *bond;
3618         struct vlan_entry *vlan;
3619
3620         if (dev_net(ifa->ifa_dev->dev) != &init_net)
3621                 return NOTIFY_DONE;
3622
3623         list_for_each_entry(bond, &bond_dev_list, bond_list) {
3624                 if (bond->dev == event_dev) {
3625                         switch (event) {
3626                         case NETDEV_UP:
3627                                 bond->master_ip = ifa->ifa_local;
3628                                 return NOTIFY_OK;
3629                         case NETDEV_DOWN:
3630                                 bond->master_ip = bond_glean_dev_ip(bond->dev);
3631                                 return NOTIFY_OK;
3632                         default:
3633                                 return NOTIFY_DONE;
3634                         }
3635                 }
3636
3637                 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
3638                         vlan_dev = vlan_group_get_device(bond->vlgrp, vlan->vlan_id);
3639                         if (vlan_dev == event_dev) {
3640                                 switch (event) {
3641                                 case NETDEV_UP:
3642                                         vlan->vlan_ip = ifa->ifa_local;
3643                                         return NOTIFY_OK;
3644                                 case NETDEV_DOWN:
3645                                         vlan->vlan_ip =
3646                                                 bond_glean_dev_ip(vlan_dev);
3647                                         return NOTIFY_OK;
3648                                 default:
3649                                         return NOTIFY_DONE;
3650                                 }
3651                         }
3652                 }
3653         }
3654         return NOTIFY_DONE;
3655 }
3656
3657 static struct notifier_block bond_netdev_notifier = {
3658         .notifier_call = bond_netdev_event,
3659 };
3660
3661 static struct notifier_block bond_inetaddr_notifier = {
3662         .notifier_call = bond_inetaddr_event,
3663 };
3664
3665 /*-------------------------- Packet type handling ---------------------------*/
3666
3667 /* register to receive lacpdus on a bond */
3668 static void bond_register_lacpdu(struct bonding *bond)
3669 {
3670         struct packet_type *pk_type = &(BOND_AD_INFO(bond).ad_pkt_type);
3671
3672         /* initialize packet type */
3673         pk_type->type = PKT_TYPE_LACPDU;
3674         pk_type->dev = bond->dev;
3675         pk_type->func = bond_3ad_lacpdu_recv;
3676
3677         dev_add_pack(pk_type);
3678 }
3679
3680 /* unregister to receive lacpdus on a bond */
3681 static void bond_unregister_lacpdu(struct bonding *bond)
3682 {
3683         dev_remove_pack(&(BOND_AD_INFO(bond).ad_pkt_type));
3684 }
3685
3686 void bond_register_arp(struct bonding *bond)
3687 {
3688         struct packet_type *pt = &bond->arp_mon_pt;
3689
3690         if (pt->type)
3691                 return;
3692
3693         pt->type = htons(ETH_P_ARP);
3694         pt->dev = bond->dev;
3695         pt->func = bond_arp_rcv;
3696         dev_add_pack(pt);
3697 }
3698
3699 void bond_unregister_arp(struct bonding *bond)
3700 {
3701         struct packet_type *pt = &bond->arp_mon_pt;
3702
3703         dev_remove_pack(pt);
3704         pt->type = 0;
3705 }
3706
3707 /*---------------------------- Hashing Policies -----------------------------*/
3708
3709 /*
3710  * Hash for the output device based upon layer 2 and layer 3 data. If
3711  * the packet is not IP mimic bond_xmit_hash_policy_l2()
3712  */
3713 static int bond_xmit_hash_policy_l23(struct sk_buff *skb,
3714                                      struct net_device *bond_dev, int count)
3715 {
3716         struct ethhdr *data = (struct ethhdr *)skb->data;
3717         struct iphdr *iph = ip_hdr(skb);
3718
3719         if (skb->protocol == htons(ETH_P_IP)) {
3720                 return ((ntohl(iph->saddr ^ iph->daddr) & 0xffff) ^
3721                         (data->h_dest[5] ^ bond_dev->dev_addr[5])) % count;
3722         }
3723
3724         return (data->h_dest[5] ^ bond_dev->dev_addr[5]) % count;
3725 }
3726
3727 /*
3728  * Hash for the output device based upon layer 3 and layer 4 data. If
3729  * the packet is a frag or not TCP or UDP, just use layer 3 data.  If it is
3730  * altogether not IP, mimic bond_xmit_hash_policy_l2()
3731  */
3732 static int bond_xmit_hash_policy_l34(struct sk_buff *skb,
3733                                     struct net_device *bond_dev, int count)
3734 {
3735         struct ethhdr *data = (struct ethhdr *)skb->data;
3736         struct iphdr *iph = ip_hdr(skb);
3737         __be16 *layer4hdr = (__be16 *)((u32 *)iph + iph->ihl);
3738         int layer4_xor = 0;
3739
3740         if (skb->protocol == htons(ETH_P_IP)) {
3741                 if (!(iph->frag_off & htons(IP_MF|IP_OFFSET)) &&
3742                     (iph->protocol == IPPROTO_TCP ||
3743                      iph->protocol == IPPROTO_UDP)) {
3744                         layer4_xor = ntohs((*layer4hdr ^ *(layer4hdr + 1)));
3745                 }
3746                 return (layer4_xor ^
3747                         ((ntohl(iph->saddr ^ iph->daddr)) & 0xffff)) % count;
3748
3749         }
3750
3751         return (data->h_dest[5] ^ bond_dev->dev_addr[5]) % count;
3752 }
3753
3754 /*
3755  * Hash for the output device based upon layer 2 data
3756  */
3757 static int bond_xmit_hash_policy_l2(struct sk_buff *skb,
3758                                    struct net_device *bond_dev, int count)
3759 {
3760         struct ethhdr *data = (struct ethhdr *)skb->data;
3761
3762         return (data->h_dest[5] ^ bond_dev->dev_addr[5]) % count;
3763 }
3764
3765 /*-------------------------- Device entry points ----------------------------*/
3766
3767 static int bond_open(struct net_device *bond_dev)
3768 {
3769         struct bonding *bond = bond_dev->priv;
3770
3771         bond->kill_timers = 0;
3772
3773         if ((bond->params.mode == BOND_MODE_TLB) ||
3774             (bond->params.mode == BOND_MODE_ALB)) {
3775                 /* bond_alb_initialize must be called before the timer
3776                  * is started.
3777                  */
3778                 if (bond_alb_initialize(bond, (bond->params.mode == BOND_MODE_ALB))) {
3779                         /* something went wrong - fail the open operation */
3780                         return -1;
3781                 }
3782
3783                 INIT_DELAYED_WORK(&bond->alb_work, bond_alb_monitor);
3784                 queue_delayed_work(bond->wq, &bond->alb_work, 0);
3785         }
3786
3787         if (bond->params.miimon) {  /* link check interval, in milliseconds. */
3788                 INIT_DELAYED_WORK(&bond->mii_work, bond_mii_monitor);
3789                 queue_delayed_work(bond->wq, &bond->mii_work, 0);
3790         }
3791
3792         if (bond->params.arp_interval) {  /* arp interval, in milliseconds. */
3793                 if (bond->params.mode == BOND_MODE_ACTIVEBACKUP)
3794                         INIT_DELAYED_WORK(&bond->arp_work,
3795                                           bond_activebackup_arp_mon);
3796                 else
3797                         INIT_DELAYED_WORK(&bond->arp_work,
3798                                           bond_loadbalance_arp_mon);
3799
3800                 queue_delayed_work(bond->wq, &bond->arp_work, 0);
3801                 if (bond->params.arp_validate)
3802                         bond_register_arp(bond);
3803         }
3804
3805         if (bond->params.mode == BOND_MODE_8023AD) {
3806                 INIT_DELAYED_WORK(&bond->ad_work, bond_3ad_state_machine_handler);
3807                 queue_delayed_work(bond->wq, &bond->ad_work, 0);
3808                 /* register to receive LACPDUs */
3809                 bond_register_lacpdu(bond);
3810         }
3811
3812         return 0;
3813 }
3814
3815 static int bond_close(struct net_device *bond_dev)
3816 {
3817         struct bonding *bond = bond_dev->priv;
3818
3819         if (bond->params.mode == BOND_MODE_8023AD) {
3820                 /* Unregister the receive of LACPDUs */
3821                 bond_unregister_lacpdu(bond);
3822         }
3823
3824         if (bond->params.arp_validate)
3825                 bond_unregister_arp(bond);
3826
3827         write_lock_bh(&bond->lock);
3828
3829         bond->send_grat_arp = 0;
3830
3831         /* signal timers not to re-arm */
3832         bond->kill_timers = 1;
3833
3834         write_unlock_bh(&bond->lock);
3835
3836         if (bond->params.miimon) {  /* link check interval, in milliseconds. */
3837                 cancel_delayed_work(&bond->mii_work);
3838         }
3839
3840         if (bond->params.arp_interval) {  /* arp interval, in milliseconds. */
3841                 cancel_delayed_work(&bond->arp_work);
3842         }
3843
3844         switch (bond->params.mode) {
3845         case BOND_MODE_8023AD:
3846                 cancel_delayed_work(&bond->ad_work);
3847                 break;
3848         case BOND_MODE_TLB:
3849         case BOND_MODE_ALB:
3850                 cancel_delayed_work(&bond->alb_work);
3851                 break;
3852         default:
3853                 break;
3854         }
3855
3856
3857         if ((bond->params.mode == BOND_MODE_TLB) ||
3858             (bond->params.mode == BOND_MODE_ALB)) {
3859                 /* Must be called only after all
3860                  * slaves have been released
3861                  */
3862                 bond_alb_deinitialize(bond);
3863         }
3864
3865         return 0;
3866 }
3867
3868 static struct net_device_stats *bond_get_stats(struct net_device *bond_dev)
3869 {
3870         struct bonding *bond = bond_dev->priv;
3871         struct net_device_stats *stats = &(bond->stats), *sstats;
3872         struct net_device_stats local_stats;
3873         struct slave *slave;
3874         int i;
3875
3876         memset(&local_stats, 0, sizeof(struct net_device_stats));
3877
3878         read_lock_bh(&bond->lock);
3879
3880         bond_for_each_slave(bond, slave, i) {
3881                 sstats = slave->dev->get_stats(slave->dev);
3882                 local_stats.rx_packets += sstats->rx_packets;
3883                 local_stats.rx_bytes += sstats->rx_bytes;
3884                 local_stats.rx_errors += sstats->rx_errors;
3885                 local_stats.rx_dropped += sstats->rx_dropped;
3886
3887                 local_stats.tx_packets += sstats->tx_packets;
3888                 local_stats.tx_bytes += sstats->tx_bytes;
3889                 local_stats.tx_errors += sstats->tx_errors;
3890                 local_stats.tx_dropped += sstats->tx_dropped;
3891
3892                 local_stats.multicast += sstats->multicast;
3893                 local_stats.collisions += sstats->collisions;
3894
3895                 local_stats.rx_length_errors += sstats->rx_length_errors;
3896                 local_stats.rx_over_errors += sstats->rx_over_errors;
3897                 local_stats.rx_crc_errors += sstats->rx_crc_errors;
3898                 local_stats.rx_frame_errors += sstats->rx_frame_errors;
3899                 local_stats.rx_fifo_errors += sstats->rx_fifo_errors;
3900                 local_stats.rx_missed_errors += sstats->rx_missed_errors;
3901
3902                 local_stats.tx_aborted_errors += sstats->tx_aborted_errors;
3903                 local_stats.tx_carrier_errors += sstats->tx_carrier_errors;
3904                 local_stats.tx_fifo_errors += sstats->tx_fifo_errors;
3905                 local_stats.tx_heartbeat_errors += sstats->tx_heartbeat_errors;
3906                 local_stats.tx_window_errors += sstats->tx_window_errors;
3907         }
3908
3909         memcpy(stats, &local_stats, sizeof(struct net_device_stats));
3910
3911         read_unlock_bh(&bond->lock);
3912
3913         return stats;
3914 }
3915
3916 static int bond_do_ioctl(struct net_device *bond_dev, struct ifreq *ifr, int cmd)
3917 {
3918         struct net_device *slave_dev = NULL;
3919         struct ifbond k_binfo;
3920         struct ifbond __user *u_binfo = NULL;
3921         struct ifslave k_sinfo;
3922         struct ifslave __user *u_sinfo = NULL;
3923         struct mii_ioctl_data *mii = NULL;
3924         int res = 0;
3925
3926         dprintk("bond_ioctl: master=%s, cmd=%d\n",
3927                 bond_dev->name, cmd);
3928
3929         switch (cmd) {
3930         case SIOCGMIIPHY:
3931                 mii = if_mii(ifr);
3932                 if (!mii) {
3933                         return -EINVAL;
3934                 }
3935                 mii->phy_id = 0;
3936                 /* Fall Through */
3937         case SIOCGMIIREG:
3938                 /*
3939                  * We do this again just in case we were called by SIOCGMIIREG
3940                  * instead of SIOCGMIIPHY.
3941                  */
3942                 mii = if_mii(ifr);
3943                 if (!mii) {
3944                         return -EINVAL;
3945                 }
3946
3947                 if (mii->reg_num == 1) {
3948                         struct bonding *bond = bond_dev->priv;
3949                         mii->val_out = 0;
3950                         read_lock(&bond->lock);
3951                         read_lock(&bond->curr_slave_lock);
3952                         if (netif_carrier_ok(bond->dev)) {
3953                                 mii->val_out = BMSR_LSTATUS;
3954                         }
3955                         read_unlock(&bond->curr_slave_lock);
3956                         read_unlock(&bond->lock);
3957                 }
3958
3959                 return 0;
3960         case BOND_INFO_QUERY_OLD:
3961         case SIOCBONDINFOQUERY:
3962                 u_binfo = (struct ifbond __user *)ifr->ifr_data;
3963
3964                 if (copy_from_user(&k_binfo, u_binfo, sizeof(ifbond))) {
3965                         return -EFAULT;
3966                 }
3967
3968                 res = bond_info_query(bond_dev, &k_binfo);
3969                 if (res == 0) {
3970                         if (copy_to_user(u_binfo, &k_binfo, sizeof(ifbond))) {
3971                                 return -EFAULT;
3972                         }
3973                 }
3974
3975                 return res;
3976         case BOND_SLAVE_INFO_QUERY_OLD:
3977         case SIOCBONDSLAVEINFOQUERY:
3978                 u_sinfo = (struct ifslave __user *)ifr->ifr_data;
3979
3980                 if (copy_from_user(&k_sinfo, u_sinfo, sizeof(ifslave))) {
3981                         return -EFAULT;
3982                 }
3983
3984                 res = bond_slave_info_query(bond_dev, &k_sinfo);
3985                 if (res == 0) {
3986                         if (copy_to_user(u_sinfo, &k_sinfo, sizeof(ifslave))) {
3987                                 return -EFAULT;
3988                         }
3989                 }
3990
3991                 return res;
3992         default:
3993                 /* Go on */
3994                 break;
3995         }
3996
3997         if (!capable(CAP_NET_ADMIN)) {
3998                 return -EPERM;
3999         }
4000
4001         down_write(&(bonding_rwsem));
4002         slave_dev = dev_get_by_name(&init_net, ifr->ifr_slave);
4003
4004         dprintk("slave_dev=%p: \n", slave_dev);
4005
4006         if (!slave_dev) {
4007                 res = -ENODEV;
4008         } else {
4009                 dprintk("slave_dev->name=%s: \n", slave_dev->name);
4010                 switch (cmd) {
4011                 case BOND_ENSLAVE_OLD:
4012                 case SIOCBONDENSLAVE:
4013                         res = bond_enslave(bond_dev, slave_dev);
4014                         break;
4015                 case BOND_RELEASE_OLD:
4016                 case SIOCBONDRELEASE:
4017                         res = bond_release(bond_dev, slave_dev);
4018                         break;
4019                 case BOND_SETHWADDR_OLD:
4020                 case SIOCBONDSETHWADDR:
4021                         res = bond_sethwaddr(bond_dev, slave_dev);
4022                         break;
4023                 case BOND_CHANGE_ACTIVE_OLD:
4024                 case SIOCBONDCHANGEACTIVE:
4025                         res = bond_ioctl_change_active(bond_dev, slave_dev);
4026                         break;
4027                 default:
4028                         res = -EOPNOTSUPP;
4029                 }
4030
4031                 dev_put(slave_dev);
4032         }
4033
4034         up_write(&(bonding_rwsem));
4035         return res;
4036 }
4037
4038 static void bond_set_multicast_list(struct net_device *bond_dev)
4039 {
4040         struct bonding *bond = bond_dev->priv;
4041         struct dev_mc_list *dmi;
4042
4043         /*
4044          * Do promisc before checking multicast_mode
4045          */
4046         if ((bond_dev->flags & IFF_PROMISC) && !(bond->flags & IFF_PROMISC)) {
4047                 /*
4048                  * FIXME: Need to handle the error when one of the multi-slaves
4049                  * encounters error.
4050                  */
4051                 bond_set_promiscuity(bond, 1);
4052         }
4053
4054         if (!(bond_dev->flags & IFF_PROMISC) && (bond->flags & IFF_PROMISC)) {
4055                 bond_set_promiscuity(bond, -1);
4056         }
4057
4058         /* set allmulti flag to slaves */
4059         if ((bond_dev->flags & IFF_ALLMULTI) && !(bond->flags & IFF_ALLMULTI)) {
4060                 /*
4061                  * FIXME: Need to handle the error when one of the multi-slaves
4062                  * encounters error.
4063                  */
4064                 bond_set_allmulti(bond, 1);
4065         }
4066
4067         if (!(bond_dev->flags & IFF_ALLMULTI) && (bond->flags & IFF_ALLMULTI)) {
4068                 bond_set_allmulti(bond, -1);
4069         }
4070
4071         read_lock(&bond->lock);
4072
4073         bond->flags = bond_dev->flags;
4074
4075         /* looking for addresses to add to slaves' mc list */
4076         for (dmi = bond_dev->mc_list; dmi; dmi = dmi->next) {
4077                 if (!bond_mc_list_find_dmi(dmi, bond->mc_list)) {
4078                         bond_mc_add(bond, dmi->dmi_addr, dmi->dmi_addrlen);
4079                 }
4080         }
4081
4082         /* looking for addresses to delete from slaves' list */
4083         for (dmi = bond->mc_list; dmi; dmi = dmi->next) {
4084                 if (!bond_mc_list_find_dmi(dmi, bond_dev->mc_list)) {
4085                         bond_mc_delete(bond, dmi->dmi_addr, dmi->dmi_addrlen);
4086                 }
4087         }
4088
4089         /* save master's multicast list */
4090         bond_mc_list_destroy(bond);
4091         bond_mc_list_copy(bond_dev->mc_list, bond, GFP_ATOMIC);
4092
4093         read_unlock(&bond->lock);
4094 }
4095
4096 /*
4097  * Change the MTU of all of a master's slaves to match the master
4098  */
4099 static int bond_change_mtu(struct net_device *bond_dev, int new_mtu)
4100 {
4101         struct bonding *bond = bond_dev->priv;
4102         struct slave *slave, *stop_at;
4103         int res = 0;
4104         int i;
4105
4106         dprintk("bond=%p, name=%s, new_mtu=%d\n", bond,
4107                 (bond_dev ? bond_dev->name : "None"), new_mtu);
4108
4109         /* Can't hold bond->lock with bh disabled here since
4110          * some base drivers panic. On the other hand we can't
4111          * hold bond->lock without bh disabled because we'll
4112          * deadlock. The only solution is to rely on the fact
4113          * that we're under rtnl_lock here, and the slaves
4114          * list won't change. This doesn't solve the problem
4115          * of setting the slave's MTU while it is
4116          * transmitting, but the assumption is that the base
4117          * driver can handle that.
4118          *
4119          * TODO: figure out a way to safely iterate the slaves
4120          * list, but without holding a lock around the actual
4121          * call to the base driver.
4122          */
4123
4124         bond_for_each_slave(bond, slave, i) {
4125                 dprintk("s %p s->p %p c_m %p\n", slave,
4126                         slave->prev, slave->dev->change_mtu);
4127
4128                 res = dev_set_mtu(slave->dev, new_mtu);
4129
4130                 if (res) {
4131                         /* If we failed to set the slave's mtu to the new value
4132                          * we must abort the operation even in ACTIVE_BACKUP
4133                          * mode, because if we allow the backup slaves to have
4134                          * different mtu values than the active slave we'll
4135                          * need to change their mtu when doing a failover. That
4136                          * means changing their mtu from timer context, which
4137                          * is probably not a good idea.
4138                          */
4139                         dprintk("err %d %s\n", res, slave->dev->name);
4140                         goto unwind;
4141                 }
4142         }
4143
4144         bond_dev->mtu = new_mtu;
4145
4146         return 0;
4147
4148 unwind:
4149         /* unwind from head to the slave that failed */
4150         stop_at = slave;
4151         bond_for_each_slave_from_to(bond, slave, i, bond->first_slave, stop_at) {
4152                 int tmp_res;
4153
4154                 tmp_res = dev_set_mtu(slave->dev, bond_dev->mtu);
4155                 if (tmp_res) {
4156                         dprintk("unwind err %d dev %s\n", tmp_res,
4157                                 slave->dev->name);
4158                 }
4159         }
4160
4161         return res;
4162 }
4163
4164 /*
4165  * Change HW address
4166  *
4167  * Note that many devices must be down to change the HW address, and
4168  * downing the master releases all slaves.  We can make bonds full of
4169  * bonding devices to test this, however.
4170  */
4171 static int bond_set_mac_address(struct net_device *bond_dev, void *addr)
4172 {
4173         struct bonding *bond = bond_dev->priv;
4174         struct sockaddr *sa = addr, tmp_sa;
4175         struct slave *slave, *stop_at;
4176         int res = 0;
4177         int i;
4178
4179         dprintk("bond=%p, name=%s\n", bond, (bond_dev ? bond_dev->name : "None"));
4180
4181         /*
4182          * If fail_over_mac is set to active, do nothing and return
4183          * success.  Returning an error causes ifenslave to fail.
4184          */
4185         if (bond->params.fail_over_mac == BOND_FOM_ACTIVE)
4186                 return 0;
4187
4188         if (!is_valid_ether_addr(sa->sa_data)) {
4189                 return -EADDRNOTAVAIL;
4190         }
4191
4192         /* Can't hold bond->lock with bh disabled here since
4193          * some base drivers panic. On the other hand we can't
4194          * hold bond->lock without bh disabled because we'll
4195          * deadlock. The only solution is to rely on the fact
4196          * that we're under rtnl_lock here, and the slaves
4197          * list won't change. This doesn't solve the problem
4198          * of setting the slave's hw address while it is
4199          * transmitting, but the assumption is that the base
4200          * driver can handle that.
4201          *
4202          * TODO: figure out a way to safely iterate the slaves
4203          * list, but without holding a lock around the actual
4204          * call to the base driver.
4205          */
4206
4207         bond_for_each_slave(bond, slave, i) {
4208                 dprintk("slave %p %s\n", slave, slave->dev->name);
4209
4210                 if (slave->dev->set_mac_address == NULL) {
4211                         res = -EOPNOTSUPP;
4212                         dprintk("EOPNOTSUPP %s\n", slave->dev->name);
4213                         goto unwind;
4214                 }
4215
4216                 res = dev_set_mac_address(slave->dev, addr);
4217                 if (res) {
4218                         /* TODO: consider downing the slave
4219                          * and retry ?
4220                          * User should expect communications
4221                          * breakage anyway until ARP finish
4222                          * updating, so...
4223                          */
4224                         dprintk("err %d %s\n", res, slave->dev->name);
4225                         goto unwind;
4226                 }
4227         }
4228
4229         /* success */
4230         memcpy(bond_dev->dev_addr, sa->sa_data, bond_dev->addr_len);
4231         return 0;
4232
4233 unwind:
4234         memcpy(tmp_sa.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
4235         tmp_sa.sa_family = bond_dev->type;
4236
4237         /* unwind from head to the slave that failed */
4238         stop_at = slave;
4239         bond_for_each_slave_from_to(bond, slave, i, bond->first_slave, stop_at) {
4240                 int tmp_res;
4241
4242                 tmp_res = dev_set_mac_address(slave->dev, &tmp_sa);
4243                 if (tmp_res) {
4244                         dprintk("unwind err %d dev %s\n", tmp_res,
4245                                 slave->dev->name);
4246                 }
4247         }
4248
4249         return res;
4250 }
4251
4252 static int bond_xmit_roundrobin(struct sk_buff *skb, struct net_device *bond_dev)
4253 {
4254         struct bonding *bond = bond_dev->priv;
4255         struct slave *slave, *start_at;
4256         int i, slave_no, res = 1;
4257
4258         read_lock(&bond->lock);
4259
4260         if (!BOND_IS_OK(bond)) {
4261                 goto out;
4262         }
4263
4264         /*
4265          * Concurrent TX may collide on rr_tx_counter; we accept that
4266          * as being rare enough not to justify using an atomic op here
4267          */
4268         slave_no = bond->rr_tx_counter++ % bond->slave_cnt;
4269
4270         bond_for_each_slave(bond, slave, i) {
4271                 slave_no--;
4272                 if (slave_no < 0) {
4273                         break;
4274                 }
4275         }
4276
4277         start_at = slave;
4278         bond_for_each_slave_from(bond, slave, i, start_at) {
4279                 if (IS_UP(slave->dev) &&
4280                     (slave->link == BOND_LINK_UP) &&
4281                     (slave->state == BOND_STATE_ACTIVE)) {
4282                         res = bond_dev_queue_xmit(bond, skb, slave->dev);
4283                         break;
4284                 }
4285         }
4286
4287 out:
4288         if (res) {
4289                 /* no suitable interface, frame not sent */
4290                 dev_kfree_skb(skb);
4291         }
4292         read_unlock(&bond->lock);
4293         return 0;
4294 }
4295
4296
4297 /*
4298  * in active-backup mode, we know that bond->curr_active_slave is always valid if
4299  * the bond has a usable interface.
4300  */
4301 static int bond_xmit_activebackup(struct sk_buff *skb, struct net_device *bond_dev)
4302 {
4303         struct bonding *bond = bond_dev->priv;
4304         int res = 1;
4305
4306         read_lock(&bond->lock);
4307         read_lock(&bond->curr_slave_lock);
4308
4309         if (!BOND_IS_OK(bond)) {
4310                 goto out;
4311         }
4312
4313         if (!bond->curr_active_slave)
4314                 goto out;
4315
4316         res = bond_dev_queue_xmit(bond, skb, bond->curr_active_slave->dev);
4317
4318 out:
4319         if (res) {
4320                 /* no suitable interface, frame not sent */
4321                 dev_kfree_skb(skb);
4322         }
4323         read_unlock(&bond->curr_slave_lock);
4324         read_unlock(&bond->lock);
4325         return 0;
4326 }
4327
4328 /*
4329  * In bond_xmit_xor() , we determine the output device by using a pre-
4330  * determined xmit_hash_policy(), If the selected device is not enabled,
4331  * find the next active slave.
4332  */
4333 static int bond_xmit_xor(struct sk_buff *skb, struct net_device *bond_dev)
4334 {
4335         struct bonding *bond = bond_dev->priv;
4336         struct slave *slave, *start_at;
4337         int slave_no;
4338         int i;
4339         int res = 1;
4340
4341         read_lock(&bond->lock);
4342
4343         if (!BOND_IS_OK(bond)) {
4344                 goto out;
4345         }
4346
4347         slave_no = bond->xmit_hash_policy(skb, bond_dev, bond->slave_cnt);
4348
4349         bond_for_each_slave(bond, slave, i) {
4350                 slave_no--;
4351                 if (slave_no < 0) {
4352                         break;
4353                 }
4354         }
4355
4356         start_at = slave;
4357
4358         bond_for_each_slave_from(bond, slave, i, start_at) {
4359                 if (IS_UP(slave->dev) &&
4360                     (slave->link == BOND_LINK_UP) &&
4361                     (slave->state == BOND_STATE_ACTIVE)) {
4362                         res = bond_dev_queue_xmit(bond, skb, slave->dev);
4363                         break;
4364                 }
4365         }
4366
4367 out:
4368         if (res) {
4369                 /* no suitable interface, frame not sent */
4370                 dev_kfree_skb(skb);
4371         }
4372         read_unlock(&bond->lock);
4373         return 0;
4374 }
4375
4376 /*
4377  * in broadcast mode, we send everything to all usable interfaces.
4378  */
4379 static int bond_xmit_broadcast(struct sk_buff *skb, struct net_device *bond_dev)
4380 {
4381         struct bonding *bond = bond_dev->priv;
4382         struct slave *slave, *start_at;
4383         struct net_device *tx_dev = NULL;
4384         int i;
4385         int res = 1;
4386
4387         read_lock(&bond->lock);
4388
4389         if (!BOND_IS_OK(bond)) {
4390                 goto out;
4391         }
4392
4393         read_lock(&bond->curr_slave_lock);
4394         start_at = bond->curr_active_slave;
4395         read_unlock(&bond->curr_slave_lock);
4396
4397         if (!start_at) {
4398                 goto out;
4399         }
4400
4401         bond_for_each_slave_from(bond, slave, i, start_at) {
4402                 if (IS_UP(slave->dev) &&
4403                     (slave->link == BOND_LINK_UP) &&
4404                     (slave->state == BOND_STATE_ACTIVE)) {
4405                         if (tx_dev) {
4406                                 struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
4407                                 if (!skb2) {
4408                                         printk(KERN_ERR DRV_NAME
4409                                                ": %s: Error: bond_xmit_broadcast(): "
4410                                                "skb_clone() failed\n",
4411                                                bond_dev->name);
4412                                         continue;
4413                                 }
4414
4415                                 res = bond_dev_queue_xmit(bond, skb2, tx_dev);
4416                                 if (res) {
4417                                         dev_kfree_skb(skb2);
4418                                         continue;
4419                                 }
4420                         }
4421                         tx_dev = slave->dev;
4422                 }
4423         }
4424
4425         if (tx_dev) {
4426                 res = bond_dev_queue_xmit(bond, skb, tx_dev);
4427         }
4428
4429 out:
4430         if (res) {
4431                 /* no suitable interface, frame not sent */
4432                 dev_kfree_skb(skb);
4433         }
4434         /* frame sent to all suitable interfaces */
4435         read_unlock(&bond->lock);
4436         return 0;
4437 }
4438
4439 /*------------------------- Device initialization ---------------------------*/
4440
4441 static void bond_set_xmit_hash_policy(struct bonding *bond)
4442 {
4443         switch (bond->params.xmit_policy) {
4444         case BOND_XMIT_POLICY_LAYER23:
4445                 bond->xmit_hash_policy = bond_xmit_hash_policy_l23;
4446                 break;
4447         case BOND_XMIT_POLICY_LAYER34:
4448                 bond->xmit_hash_policy = bond_xmit_hash_policy_l34;
4449                 break;
4450         case BOND_XMIT_POLICY_LAYER2:
4451         default:
4452                 bond->xmit_hash_policy = bond_xmit_hash_policy_l2;
4453                 break;
4454         }
4455 }
4456
4457 /*
4458  * set bond mode specific net device operations
4459  */
4460 void bond_set_mode_ops(struct bonding *bond, int mode)
4461 {
4462         struct net_device *bond_dev = bond->dev;
4463
4464         switch (mode) {
4465         case BOND_MODE_ROUNDROBIN:
4466                 bond_dev->hard_start_xmit = bond_xmit_roundrobin;
4467                 break;
4468         case BOND_MODE_ACTIVEBACKUP:
4469                 bond_dev->hard_start_xmit = bond_xmit_activebackup;
4470                 break;
4471         case BOND_MODE_XOR:
4472                 bond_dev->hard_start_xmit = bond_xmit_xor;
4473                 bond_set_xmit_hash_policy(bond);
4474                 break;
4475         case BOND_MODE_BROADCAST:
4476                 bond_dev->hard_start_xmit = bond_xmit_broadcast;
4477                 break;
4478         case BOND_MODE_8023AD:
4479                 bond_set_master_3ad_flags(bond);
4480                 bond_dev->hard_start_xmit = bond_3ad_xmit_xor;
4481                 bond_set_xmit_hash_policy(bond);
4482                 break;
4483         case BOND_MODE_ALB:
4484                 bond_set_master_alb_flags(bond);
4485                 /* FALLTHRU */
4486         case BOND_MODE_TLB:
4487                 bond_dev->hard_start_xmit = bond_alb_xmit;
4488                 bond_dev->set_mac_address = bond_alb_set_mac_address;
4489                 break;
4490         default:
4491                 /* Should never happen, mode already checked */
4492                 printk(KERN_ERR DRV_NAME
4493                        ": %s: Error: Unknown bonding mode %d\n",
4494                        bond_dev->name,
4495                        mode);
4496                 break;
4497         }
4498 }
4499
4500 static void bond_ethtool_get_drvinfo(struct net_device *bond_dev,
4501                                     struct ethtool_drvinfo *drvinfo)
4502 {
4503         strncpy(drvinfo->driver, DRV_NAME, 32);
4504         strncpy(drvinfo->version, DRV_VERSION, 32);
4505         snprintf(drvinfo->fw_version, 32, "%d", BOND_ABI_VERSION);
4506 }
4507
4508 static const struct ethtool_ops bond_ethtool_ops = {
4509         .get_drvinfo            = bond_ethtool_get_drvinfo,
4510         .get_link               = ethtool_op_get_link,
4511         .get_tx_csum            = ethtool_op_get_tx_csum,
4512         .get_sg                 = ethtool_op_get_sg,
4513         .get_tso                = ethtool_op_get_tso,
4514         .get_ufo                = ethtool_op_get_ufo,
4515         .get_flags              = ethtool_op_get_flags,
4516 };
4517
4518 /*
4519  * Does not allocate but creates a /proc entry.
4520  * Allowed to fail.
4521  */
4522 static int bond_init(struct net_device *bond_dev, struct bond_params *params)
4523 {
4524         struct bonding *bond = bond_dev->priv;
4525
4526         dprintk("Begin bond_init for %s\n", bond_dev->name);
4527
4528         /* initialize rwlocks */
4529         rwlock_init(&bond->lock);
4530         rwlock_init(&bond->curr_slave_lock);
4531
4532         bond->params = *params; /* copy params struct */
4533
4534         bond->wq = create_singlethread_workqueue(bond_dev->name);
4535         if (!bond->wq)
4536                 return -ENOMEM;
4537
4538         /* Initialize pointers */
4539         bond->first_slave = NULL;
4540         bond->curr_active_slave = NULL;
4541         bond->current_arp_slave = NULL;
4542         bond->primary_slave = NULL;
4543         bond->dev = bond_dev;
4544         bond->send_grat_arp = 0;
4545         bond->setup_by_slave = 0;
4546         INIT_LIST_HEAD(&bond->vlan_list);
4547
4548         /* Initialize the device entry points */
4549         bond_dev->open = bond_open;
4550         bond_dev->stop = bond_close;
4551         bond_dev->get_stats = bond_get_stats;
4552         bond_dev->do_ioctl = bond_do_ioctl;
4553         bond_dev->ethtool_ops = &bond_ethtool_ops;
4554         bond_dev->set_multicast_list = bond_set_multicast_list;
4555         bond_dev->change_mtu = bond_change_mtu;
4556         bond_dev->set_mac_address = bond_set_mac_address;
4557         bond_dev->validate_addr = NULL;
4558
4559         bond_set_mode_ops(bond, bond->params.mode);
4560
4561         bond_dev->destructor = bond_destructor;
4562
4563         /* Initialize the device options */
4564         bond_dev->tx_queue_len = 0;
4565         bond_dev->flags |= IFF_MASTER|IFF_MULTICAST;
4566         bond_dev->priv_flags |= IFF_BONDING;
4567         if (bond->params.arp_interval)
4568                 bond_dev->priv_flags |= IFF_MASTER_ARPMON;
4569
4570         /* At first, we block adding VLANs. That's the only way to
4571          * prevent problems that occur when adding VLANs over an
4572          * empty bond. The block will be removed once non-challenged
4573          * slaves are enslaved.
4574          */
4575         bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
4576
4577         /* don't acquire bond device's netif_tx_lock when
4578          * transmitting */
4579         bond_dev->features |= NETIF_F_LLTX;
4580
4581         /* By default, we declare the bond to be fully
4582          * VLAN hardware accelerated capable. Special
4583          * care is taken in the various xmit functions
4584          * when there are slaves that are not hw accel
4585          * capable
4586          */
4587         bond_dev->vlan_rx_register = bond_vlan_rx_register;
4588         bond_dev->vlan_rx_add_vid  = bond_vlan_rx_add_vid;
4589         bond_dev->vlan_rx_kill_vid = bond_vlan_rx_kill_vid;
4590         bond_dev->features |= (NETIF_F_HW_VLAN_TX |
4591                                NETIF_F_HW_VLAN_RX |
4592                                NETIF_F_HW_VLAN_FILTER);
4593
4594 #ifdef CONFIG_PROC_FS
4595         bond_create_proc_entry(bond);
4596 #endif
4597         list_add_tail(&bond->bond_list, &bond_dev_list);
4598
4599         return 0;
4600 }
4601
4602 static void bond_work_cancel_all(struct bonding *bond)
4603 {
4604         write_lock_bh(&bond->lock);
4605         bond->kill_timers = 1;
4606         write_unlock_bh(&bond->lock);
4607
4608         if (bond->params.miimon && delayed_work_pending(&bond->mii_work))
4609                 cancel_delayed_work(&bond->mii_work);
4610
4611         if (bond->params.arp_interval && delayed_work_pending(&bond->arp_work))
4612                 cancel_delayed_work(&bond->arp_work);
4613
4614         if (bond->params.mode == BOND_MODE_ALB &&
4615             delayed_work_pending(&bond->alb_work))
4616                 cancel_delayed_work(&bond->alb_work);
4617
4618         if (bond->params.mode == BOND_MODE_8023AD &&
4619             delayed_work_pending(&bond->ad_work))
4620                 cancel_delayed_work(&bond->ad_work);
4621 }
4622
4623 /* De-initialize device specific data.
4624  * Caller must hold rtnl_lock.
4625  */
4626 static void bond_deinit(struct net_device *bond_dev)
4627 {
4628         struct bonding *bond = bond_dev->priv;
4629
4630         list_del(&bond->bond_list);
4631
4632         bond_work_cancel_all(bond);
4633
4634 #ifdef CONFIG_PROC_FS
4635         bond_remove_proc_entry(bond);
4636 #endif
4637 }
4638
4639 /* Unregister and free all bond devices.
4640  * Caller must hold rtnl_lock.
4641  */
4642 static void bond_free_all(void)
4643 {
4644         struct bonding *bond, *nxt;
4645
4646         list_for_each_entry_safe(bond, nxt, &bond_dev_list, bond_list) {
4647                 struct net_device *bond_dev = bond->dev;
4648
4649                 bond_work_cancel_all(bond);
4650                 /* Release the bonded slaves */
4651                 bond_release_all(bond_dev);
4652                 bond_destroy(bond);
4653         }
4654
4655 #ifdef CONFIG_PROC_FS
4656         bond_destroy_proc_dir();
4657 #endif
4658 }
4659
4660 /*------------------------- Module initialization ---------------------------*/
4661
4662 /*
4663  * Convert string input module parms.  Accept either the
4664  * number of the mode or its string name.  A bit complicated because
4665  * some mode names are substrings of other names, and calls from sysfs
4666  * may have whitespace in the name (trailing newlines, for example).
4667  */
4668 int bond_parse_parm(const char *buf, struct bond_parm_tbl *tbl)
4669 {
4670         int mode = -1, i, rv;
4671         char *p, modestr[BOND_MAX_MODENAME_LEN + 1] = { 0, };
4672
4673         for (p = (char *)buf; *p; p++)
4674                 if (!(isdigit(*p) || isspace(*p)))
4675                         break;
4676
4677         if (*p)
4678                 rv = sscanf(buf, "%20s", modestr);
4679         else
4680                 rv = sscanf(buf, "%d", &mode);
4681
4682         if (!rv)
4683                 return -1;
4684
4685         for (i = 0; tbl[i].modename; i++) {
4686                 if (mode == tbl[i].mode)
4687                         return tbl[i].mode;
4688                 if (strcmp(modestr, tbl[i].modename) == 0)
4689                         return tbl[i].mode;
4690         }
4691
4692         return -1;
4693 }
4694
4695 static int bond_check_params(struct bond_params *params)
4696 {
4697         int arp_validate_value, fail_over_mac_value;
4698
4699         /*
4700          * Convert string parameters.
4701          */
4702         if (mode) {
4703                 bond_mode = bond_parse_parm(mode, bond_mode_tbl);
4704                 if (bond_mode == -1) {
4705                         printk(KERN_ERR DRV_NAME
4706                                ": Error: Invalid bonding mode \"%s\"\n",
4707                                mode == NULL ? "NULL" : mode);
4708                         return -EINVAL;
4709                 }
4710         }
4711
4712         if (xmit_hash_policy) {
4713                 if ((bond_mode != BOND_MODE_XOR) &&
4714                     (bond_mode != BOND_MODE_8023AD)) {
4715                         printk(KERN_INFO DRV_NAME
4716                                ": xor_mode param is irrelevant in mode %s\n",
4717                                bond_mode_name(bond_mode));
4718                 } else {
4719                         xmit_hashtype = bond_parse_parm(xmit_hash_policy,
4720                                                         xmit_hashtype_tbl);
4721                         if (xmit_hashtype == -1) {
4722                                 printk(KERN_ERR DRV_NAME
4723                                 ": Error: Invalid xmit_hash_policy \"%s\"\n",
4724                                 xmit_hash_policy == NULL ? "NULL" :
4725                                        xmit_hash_policy);
4726                                 return -EINVAL;
4727                         }
4728                 }
4729         }
4730
4731         if (lacp_rate) {
4732                 if (bond_mode != BOND_MODE_8023AD) {
4733                         printk(KERN_INFO DRV_NAME
4734                                ": lacp_rate param is irrelevant in mode %s\n",
4735                                bond_mode_name(bond_mode));
4736                 } else {
4737                         lacp_fast = bond_parse_parm(lacp_rate, bond_lacp_tbl);
4738                         if (lacp_fast == -1) {
4739                                 printk(KERN_ERR DRV_NAME
4740                                        ": Error: Invalid lacp rate \"%s\"\n",
4741                                        lacp_rate == NULL ? "NULL" : lacp_rate);
4742                                 return -EINVAL;
4743                         }
4744                 }
4745         }
4746
4747         if (max_bonds < 0 || max_bonds > INT_MAX) {
4748                 printk(KERN_WARNING DRV_NAME
4749                        ": Warning: max_bonds (%d) not in range %d-%d, so it "
4750                        "was reset to BOND_DEFAULT_MAX_BONDS (%d)\n",
4751                        max_bonds, 0, INT_MAX, BOND_DEFAULT_MAX_BONDS);
4752                 max_bonds = BOND_DEFAULT_MAX_BONDS;
4753         }
4754
4755         if (miimon < 0) {
4756                 printk(KERN_WARNING DRV_NAME
4757                        ": Warning: miimon module parameter (%d), "
4758                        "not in range 0-%d, so it was reset to %d\n",
4759                        miimon, INT_MAX, BOND_LINK_MON_INTERV);
4760                 miimon = BOND_LINK_MON_INTERV;
4761         }
4762
4763         if (updelay < 0) {
4764                 printk(KERN_WARNING DRV_NAME
4765                        ": Warning: updelay module parameter (%d), "
4766                        "not in range 0-%d, so it was reset to 0\n",
4767                        updelay, INT_MAX);
4768                 updelay = 0;
4769         }
4770
4771         if (downdelay < 0) {
4772                 printk(KERN_WARNING DRV_NAME
4773                        ": Warning: downdelay module parameter (%d), "
4774                        "not in range 0-%d, so it was reset to 0\n",
4775                        downdelay, INT_MAX);
4776                 downdelay = 0;
4777         }
4778
4779         if ((use_carrier != 0) && (use_carrier != 1)) {
4780                 printk(KERN_WARNING DRV_NAME
4781                        ": Warning: use_carrier module parameter (%d), "
4782                        "not of valid value (0/1), so it was set to 1\n",
4783                        use_carrier);
4784                 use_carrier = 1;
4785         }
4786
4787         if (num_grat_arp < 0 || num_grat_arp > 255) {
4788                 printk(KERN_WARNING DRV_NAME
4789                        ": Warning: num_grat_arp (%d) not in range 0-255 so it "
4790                        "was reset to 1 \n", num_grat_arp);
4791                 num_grat_arp = 1;
4792         }
4793
4794         /* reset values for 802.3ad */
4795         if (bond_mode == BOND_MODE_8023AD) {
4796                 if (!miimon) {
4797                         printk(KERN_WARNING DRV_NAME
4798                                ": Warning: miimon must be specified, "
4799                                "otherwise bonding will not detect link "
4800                                "failure, speed and duplex which are "
4801                                "essential for 802.3ad operation\n");
4802                         printk(KERN_WARNING "Forcing miimon to 100msec\n");
4803                         miimon = 100;
4804                 }
4805         }
4806
4807         /* reset values for TLB/ALB */
4808         if ((bond_mode == BOND_MODE_TLB) ||
4809             (bond_mode == BOND_MODE_ALB)) {
4810                 if (!miimon) {
4811                         printk(KERN_WARNING DRV_NAME
4812                                ": Warning: miimon must be specified, "
4813                                "otherwise bonding will not detect link "
4814                                "failure and link speed which are essential "
4815                                "for TLB/ALB load balancing\n");
4816                         printk(KERN_WARNING "Forcing miimon to 100msec\n");
4817                         miimon = 100;
4818                 }
4819         }
4820
4821         if (bond_mode == BOND_MODE_ALB) {
4822                 printk(KERN_NOTICE DRV_NAME
4823                        ": In ALB mode you might experience client "
4824                        "disconnections upon reconnection of a link if the "
4825                        "bonding module updelay parameter (%d msec) is "
4826                        "incompatible with the forwarding delay time of the "
4827                        "switch\n",
4828                        updelay);
4829         }
4830
4831         if (!miimon) {
4832                 if (updelay || downdelay) {
4833                         /* just warn the user the up/down delay will have
4834                          * no effect since miimon is zero...
4835                          */
4836                         printk(KERN_WARNING DRV_NAME
4837                                ": Warning: miimon module parameter not set "
4838                                "and updelay (%d) or downdelay (%d) module "
4839                                "parameter is set; updelay and downdelay have "
4840                                "no effect unless miimon is set\n",
4841                                updelay, downdelay);
4842                 }
4843         } else {
4844                 /* don't allow arp monitoring */
4845                 if (arp_interval) {
4846                         printk(KERN_WARNING DRV_NAME
4847                                ": Warning: miimon (%d) and arp_interval (%d) "
4848                                "can't be used simultaneously, disabling ARP "
4849                                "monitoring\n",
4850                                miimon, arp_interval);
4851                         arp_interval = 0;
4852                 }
4853
4854                 if ((updelay % miimon) != 0) {
4855                         printk(KERN_WARNING DRV_NAME
4856                                ": Warning: updelay (%d) is not a multiple "
4857                                "of miimon (%d), updelay rounded to %d ms\n",
4858                                updelay, miimon, (updelay / miimon) * miimon);
4859                 }
4860
4861                 updelay /= miimon;
4862
4863                 if ((downdelay % miimon) != 0) {
4864                         printk(KERN_WARNING DRV_NAME
4865                                ": Warning: downdelay (%d) is not a multiple "
4866                                "of miimon (%d), downdelay rounded to %d ms\n",
4867                                downdelay, miimon,
4868                                (downdelay / miimon) * miimon);
4869                 }
4870
4871                 downdelay /= miimon;
4872         }
4873
4874         if (arp_interval < 0) {
4875                 printk(KERN_WARNING DRV_NAME
4876                        ": Warning: arp_interval module parameter (%d) "
4877                        ", not in range 0-%d, so it was reset to %d\n",
4878                        arp_interval, INT_MAX, BOND_LINK_ARP_INTERV);
4879                 arp_interval = BOND_LINK_ARP_INTERV;
4880         }
4881
4882         for (arp_ip_count = 0;
4883              (arp_ip_count < BOND_MAX_ARP_TARGETS) && arp_ip_target[arp_ip_count];
4884              arp_ip_count++) {
4885                 /* not complete check, but should be good enough to
4886                    catch mistakes */
4887                 if (!isdigit(arp_ip_target[arp_ip_count][0])) {
4888                         printk(KERN_WARNING DRV_NAME
4889                                ": Warning: bad arp_ip_target module parameter "
4890                                "(%s), ARP monitoring will not be performed\n",
4891                                arp_ip_target[arp_ip_count]);
4892                         arp_interval = 0;
4893                 } else {
4894                         __be32 ip = in_aton(arp_ip_target[arp_ip_count]);
4895                         arp_target[arp_ip_count] = ip;
4896                 }
4897         }
4898
4899         if (arp_interval && !arp_ip_count) {
4900                 /* don't allow arping if no arp_ip_target given... */
4901                 printk(KERN_WARNING DRV_NAME
4902                        ": Warning: arp_interval module parameter (%d) "
4903                        "specified without providing an arp_ip_target "
4904                        "parameter, arp_interval was reset to 0\n",
4905                        arp_interval);
4906                 arp_interval = 0;
4907         }
4908
4909         if (arp_validate) {
4910                 if (bond_mode != BOND_MODE_ACTIVEBACKUP) {
4911                         printk(KERN_ERR DRV_NAME
4912                ": arp_validate only supported in active-backup mode\n");
4913                         return -EINVAL;
4914                 }
4915                 if (!arp_interval) {
4916                         printk(KERN_ERR DRV_NAME
4917                                ": arp_validate requires arp_interval\n");
4918                         return -EINVAL;
4919                 }
4920
4921                 arp_validate_value = bond_parse_parm(arp_validate,
4922                                                      arp_validate_tbl);
4923                 if (arp_validate_value == -1) {
4924                         printk(KERN_ERR DRV_NAME
4925                                ": Error: invalid arp_validate \"%s\"\n",
4926                                arp_validate == NULL ? "NULL" : arp_validate);
4927                         return -EINVAL;
4928                 }
4929         } else
4930                 arp_validate_value = 0;
4931
4932         if (miimon) {
4933                 printk(KERN_INFO DRV_NAME
4934                        ": MII link monitoring set to %d ms\n",
4935                        miimon);
4936         } else if (arp_interval) {
4937                 int i;
4938
4939                 printk(KERN_INFO DRV_NAME
4940                        ": ARP monitoring set to %d ms, validate %s, with %d target(s):",
4941                        arp_interval,
4942                        arp_validate_tbl[arp_validate_value].modename,
4943                        arp_ip_count);
4944
4945                 for (i = 0; i < arp_ip_count; i++)
4946                         printk (" %s", arp_ip_target[i]);
4947
4948                 printk("\n");
4949
4950         } else if (max_bonds) {
4951                 /* miimon and arp_interval not set, we need one so things
4952                  * work as expected, see bonding.txt for details
4953                  */
4954                 printk(KERN_WARNING DRV_NAME
4955                        ": Warning: either miimon or arp_interval and "
4956                        "arp_ip_target module parameters must be specified, "
4957                        "otherwise bonding will not detect link failures! see "
4958                        "bonding.txt for details.\n");
4959         }
4960
4961         if (primary && !USES_PRIMARY(bond_mode)) {
4962                 /* currently, using a primary only makes sense
4963                  * in active backup, TLB or ALB modes
4964                  */
4965                 printk(KERN_WARNING DRV_NAME
4966                        ": Warning: %s primary device specified but has no "
4967                        "effect in %s mode\n",
4968                        primary, bond_mode_name(bond_mode));
4969                 primary = NULL;
4970         }
4971
4972         if (fail_over_mac) {
4973                 fail_over_mac_value = bond_parse_parm(fail_over_mac,
4974                                                       fail_over_mac_tbl);
4975                 if (fail_over_mac_value == -1) {
4976                         printk(KERN_ERR DRV_NAME
4977                                ": Error: invalid fail_over_mac \"%s\"\n",
4978                                arp_validate == NULL ? "NULL" : arp_validate);
4979                         return -EINVAL;
4980                 }
4981
4982                 if (bond_mode != BOND_MODE_ACTIVEBACKUP)
4983                         printk(KERN_WARNING DRV_NAME
4984                                ": Warning: fail_over_mac only affects "
4985                                "active-backup mode.\n");
4986         } else {
4987                 fail_over_mac_value = BOND_FOM_NONE;
4988         }
4989
4990         /* fill params struct with the proper values */
4991         params->mode = bond_mode;
4992         params->xmit_policy = xmit_hashtype;
4993         params->miimon = miimon;
4994         params->num_grat_arp = num_grat_arp;
4995         params->arp_interval = arp_interval;
4996         params->arp_validate = arp_validate_value;
4997         params->updelay = updelay;
4998         params->downdelay = downdelay;
4999         params->use_carrier = use_carrier;
5000         params->lacp_fast = lacp_fast;
5001         params->primary[0] = 0;
5002         params->fail_over_mac = fail_over_mac_value;
5003
5004         if (primary) {
5005                 strncpy(params->primary, primary, IFNAMSIZ);
5006                 params->primary[IFNAMSIZ - 1] = 0;
5007         }
5008
5009         memcpy(params->arp_targets, arp_target, sizeof(arp_target));
5010
5011         return 0;
5012 }
5013
5014 static struct lock_class_key bonding_netdev_xmit_lock_key;
5015 static struct lock_class_key bonding_netdev_addr_lock_key;
5016
5017 static void bond_set_lockdep_class_one(struct net_device *dev,
5018                                        struct netdev_queue *txq,
5019                                        void *_unused)
5020 {
5021         lockdep_set_class(&txq->_xmit_lock,
5022                           &bonding_netdev_xmit_lock_key);
5023 }
5024
5025 static void bond_set_lockdep_class(struct net_device *dev)
5026 {
5027         lockdep_set_class(&dev->addr_list_lock,
5028                           &bonding_netdev_addr_lock_key);
5029         netdev_for_each_tx_queue(dev, bond_set_lockdep_class_one, NULL);
5030 }
5031
5032 /* Create a new bond based on the specified name and bonding parameters.
5033  * If name is NULL, obtain a suitable "bond%d" name for us.
5034  * Caller must NOT hold rtnl_lock; we need to release it here before we
5035  * set up our sysfs entries.
5036  */
5037 int bond_create(char *name, struct bond_params *params)
5038 {
5039         struct net_device *bond_dev;
5040         struct bonding *bond;
5041         int res;
5042
5043         rtnl_lock();
5044         down_write(&bonding_rwsem);
5045
5046         /* Check to see if the bond already exists. */
5047         if (name) {
5048                 list_for_each_entry(bond, &bond_dev_list, bond_list)
5049                         if (strnicmp(bond->dev->name, name, IFNAMSIZ) == 0) {
5050                                 printk(KERN_ERR DRV_NAME
5051                                ": cannot add bond %s; it already exists\n",
5052                                        name);
5053                                 res = -EPERM;
5054                                 goto out_rtnl;
5055                         }
5056         }
5057
5058         bond_dev = alloc_netdev(sizeof(struct bonding), name ? name : "",
5059                                 ether_setup);
5060         if (!bond_dev) {
5061                 printk(KERN_ERR DRV_NAME
5062                        ": %s: eek! can't alloc netdev!\n",
5063                        name);
5064                 res = -ENOMEM;
5065                 goto out_rtnl;
5066         }
5067
5068         if (!name) {
5069                 res = dev_alloc_name(bond_dev, "bond%d");
5070                 if (res < 0)
5071                         goto out_netdev;
5072         }
5073
5074         /* bond_init() must be called after dev_alloc_name() (for the
5075          * /proc files), but before register_netdevice(), because we
5076          * need to set function pointers.
5077          */
5078
5079         res = bond_init(bond_dev, params);
5080         if (res < 0) {
5081                 goto out_netdev;
5082         }
5083
5084         res = register_netdevice(bond_dev);
5085         if (res < 0) {
5086                 goto out_bond;
5087         }
5088
5089         bond_set_lockdep_class(bond_dev);
5090
5091         netif_carrier_off(bond_dev);
5092
5093         up_write(&bonding_rwsem);
5094         rtnl_unlock(); /* allows sysfs registration of net device */
5095         res = bond_create_sysfs_entry(bond_dev->priv);
5096         if (res < 0) {
5097                 rtnl_lock();
5098                 down_write(&bonding_rwsem);
5099                 bond_deinit(bond_dev);
5100                 unregister_netdevice(bond_dev);
5101                 goto out_rtnl;
5102         }
5103
5104         return 0;
5105
5106 out_bond:
5107         bond_deinit(bond_dev);
5108 out_netdev:
5109         free_netdev(bond_dev);
5110 out_rtnl:
5111         up_write(&bonding_rwsem);
5112         rtnl_unlock();
5113         return res;
5114 }
5115
5116 static int __init bonding_init(void)
5117 {
5118         int i;
5119         int res;
5120         struct bonding *bond;
5121
5122         printk(KERN_INFO "%s", version);
5123
5124         res = bond_check_params(&bonding_defaults);
5125         if (res) {
5126                 goto out;
5127         }
5128
5129 #ifdef CONFIG_PROC_FS
5130         bond_create_proc_dir();
5131 #endif
5132
5133         init_rwsem(&bonding_rwsem);
5134
5135         for (i = 0; i < max_bonds; i++) {
5136                 res = bond_create(NULL, &bonding_defaults);
5137                 if (res)
5138                         goto err;
5139         }
5140
5141         res = bond_create_sysfs();
5142         if (res)
5143                 goto err;
5144
5145         register_netdevice_notifier(&bond_netdev_notifier);
5146         register_inetaddr_notifier(&bond_inetaddr_notifier);
5147
5148         goto out;
5149 err:
5150         list_for_each_entry(bond, &bond_dev_list, bond_list) {
5151                 bond_work_cancel_all(bond);
5152                 destroy_workqueue(bond->wq);
5153         }
5154
5155         bond_destroy_sysfs();
5156
5157         rtnl_lock();
5158         bond_free_all();
5159         rtnl_unlock();
5160 out:
5161         return res;
5162
5163 }
5164
5165 static void __exit bonding_exit(void)
5166 {
5167         unregister_netdevice_notifier(&bond_netdev_notifier);
5168         unregister_inetaddr_notifier(&bond_inetaddr_notifier);
5169
5170         bond_destroy_sysfs();
5171
5172         rtnl_lock();
5173         bond_free_all();
5174         rtnl_unlock();
5175 }
5176
5177 module_init(bonding_init);
5178 module_exit(bonding_exit);
5179 MODULE_LICENSE("GPL");
5180 MODULE_VERSION(DRV_VERSION);
5181 MODULE_DESCRIPTION(DRV_DESCRIPTION ", v" DRV_VERSION);
5182 MODULE_AUTHOR("Thomas Davis, tadavis@lbl.gov and many others");
5183 MODULE_SUPPORTED_DEVICE("most ethernet devices");
5184
5185 /*
5186  * Local variables:
5187  *  c-indent-level: 8
5188  *  c-basic-offset: 8
5189  *  tab-width: 8
5190  * End:
5191  */
5192