e41b3e57260c057a7269d63d8ea72d3700245541
[linux-2.6.git] / drivers / net / bonding / bond_main.c
1 /*
2  * originally based on the dummy device.
3  *
4  * Copyright 1999, Thomas Davis, tadavis@lbl.gov.
5  * Licensed under the GPL. Based on dummy.c, and eql.c devices.
6  *
7  * bonding.c: an Ethernet Bonding driver
8  *
9  * This is useful to talk to a Cisco EtherChannel compatible equipment:
10  *      Cisco 5500
11  *      Sun Trunking (Solaris)
12  *      Alteon AceDirector Trunks
13  *      Linux Bonding
14  *      and probably many L2 switches ...
15  *
16  * How it works:
17  *    ifconfig bond0 ipaddress netmask up
18  *      will setup a network device, with an ip address.  No mac address
19  *      will be assigned at this time.  The hw mac address will come from
20  *      the first slave bonded to the channel.  All slaves will then use
21  *      this hw mac address.
22  *
23  *    ifconfig bond0 down
24  *         will release all slaves, marking them as down.
25  *
26  *    ifenslave bond0 eth0
27  *      will attach eth0 to bond0 as a slave.  eth0 hw mac address will either
28  *      a: be used as initial mac address
29  *      b: if a hw mac address already is there, eth0's hw mac address
30  *         will then be set from bond0.
31  *
32  */
33
34 //#define BONDING_DEBUG 1
35
36 #include <linux/kernel.h>
37 #include <linux/module.h>
38 #include <linux/types.h>
39 #include <linux/fcntl.h>
40 #include <linux/interrupt.h>
41 #include <linux/ptrace.h>
42 #include <linux/ioport.h>
43 #include <linux/in.h>
44 #include <net/ip.h>
45 #include <linux/ip.h>
46 #include <linux/tcp.h>
47 #include <linux/udp.h>
48 #include <linux/slab.h>
49 #include <linux/string.h>
50 #include <linux/init.h>
51 #include <linux/timer.h>
52 #include <linux/socket.h>
53 #include <linux/ctype.h>
54 #include <linux/inet.h>
55 #include <linux/bitops.h>
56 #include <asm/system.h>
57 #include <asm/io.h>
58 #include <asm/dma.h>
59 #include <asm/uaccess.h>
60 #include <linux/errno.h>
61 #include <linux/netdevice.h>
62 #include <linux/inetdevice.h>
63 #include <linux/igmp.h>
64 #include <linux/etherdevice.h>
65 #include <linux/skbuff.h>
66 #include <net/sock.h>
67 #include <linux/rtnetlink.h>
68 #include <linux/proc_fs.h>
69 #include <linux/seq_file.h>
70 #include <linux/smp.h>
71 #include <linux/if_ether.h>
72 #include <net/arp.h>
73 #include <linux/mii.h>
74 #include <linux/ethtool.h>
75 #include <linux/if_vlan.h>
76 #include <linux/if_bonding.h>
77 #include <linux/jiffies.h>
78 #include <net/route.h>
79 #include <net/net_namespace.h>
80 #include "bonding.h"
81 #include "bond_3ad.h"
82 #include "bond_alb.h"
83
84 /*---------------------------- Module parameters ----------------------------*/
85
86 /* monitor all links that often (in milliseconds). <=0 disables monitoring */
87 #define BOND_LINK_MON_INTERV    0
88 #define BOND_LINK_ARP_INTERV    0
89
90 static int max_bonds    = BOND_DEFAULT_MAX_BONDS;
91 static int miimon       = BOND_LINK_MON_INTERV;
92 static int updelay      = 0;
93 static int downdelay    = 0;
94 static int use_carrier  = 1;
95 static char *mode       = NULL;
96 static char *primary    = NULL;
97 static char *lacp_rate  = NULL;
98 static char *xmit_hash_policy = NULL;
99 static int arp_interval = BOND_LINK_ARP_INTERV;
100 static char *arp_ip_target[BOND_MAX_ARP_TARGETS] = { NULL, };
101 static char *arp_validate = NULL;
102 static int fail_over_mac = 0;
103 struct bond_params bonding_defaults;
104
105 module_param(max_bonds, int, 0);
106 MODULE_PARM_DESC(max_bonds, "Max number of bonded devices");
107 module_param(miimon, int, 0);
108 MODULE_PARM_DESC(miimon, "Link check interval in milliseconds");
109 module_param(updelay, int, 0);
110 MODULE_PARM_DESC(updelay, "Delay before considering link up, in milliseconds");
111 module_param(downdelay, int, 0);
112 MODULE_PARM_DESC(downdelay, "Delay before considering link down, "
113                             "in milliseconds");
114 module_param(use_carrier, int, 0);
115 MODULE_PARM_DESC(use_carrier, "Use netif_carrier_ok (vs MII ioctls) in miimon; "
116                               "0 for off, 1 for on (default)");
117 module_param(mode, charp, 0);
118 MODULE_PARM_DESC(mode, "Mode of operation : 0 for balance-rr, "
119                        "1 for active-backup, 2 for balance-xor, "
120                        "3 for broadcast, 4 for 802.3ad, 5 for balance-tlb, "
121                        "6 for balance-alb");
122 module_param(primary, charp, 0);
123 MODULE_PARM_DESC(primary, "Primary network device to use");
124 module_param(lacp_rate, charp, 0);
125 MODULE_PARM_DESC(lacp_rate, "LACPDU tx rate to request from 802.3ad partner "
126                             "(slow/fast)");
127 module_param(xmit_hash_policy, charp, 0);
128 MODULE_PARM_DESC(xmit_hash_policy, "XOR hashing method: 0 for layer 2 (default)"
129                                    ", 1 for layer 3+4");
130 module_param(arp_interval, int, 0);
131 MODULE_PARM_DESC(arp_interval, "arp interval in milliseconds");
132 module_param_array(arp_ip_target, charp, NULL, 0);
133 MODULE_PARM_DESC(arp_ip_target, "arp targets in n.n.n.n form");
134 module_param(arp_validate, charp, 0);
135 MODULE_PARM_DESC(arp_validate, "validate src/dst of ARP probes: none (default), active, backup or all");
136 module_param(fail_over_mac, int, 0);
137 MODULE_PARM_DESC(fail_over_mac, "For active-backup, do not set all slaves to the same MAC.  0 of off (default), 1 for on.");
138
139 /*----------------------------- Global variables ----------------------------*/
140
141 static const char * const version =
142         DRV_DESCRIPTION ": v" DRV_VERSION " (" DRV_RELDATE ")\n";
143
144 LIST_HEAD(bond_dev_list);
145
146 #ifdef CONFIG_PROC_FS
147 static struct proc_dir_entry *bond_proc_dir = NULL;
148 #endif
149
150 extern struct rw_semaphore bonding_rwsem;
151 static __be32 arp_target[BOND_MAX_ARP_TARGETS] = { 0, } ;
152 static int arp_ip_count = 0;
153 static int bond_mode    = BOND_MODE_ROUNDROBIN;
154 static int xmit_hashtype= BOND_XMIT_POLICY_LAYER2;
155 static int lacp_fast    = 0;
156
157
158 struct bond_parm_tbl bond_lacp_tbl[] = {
159 {       "slow",         AD_LACP_SLOW},
160 {       "fast",         AD_LACP_FAST},
161 {       NULL,           -1},
162 };
163
164 struct bond_parm_tbl bond_mode_tbl[] = {
165 {       "balance-rr",           BOND_MODE_ROUNDROBIN},
166 {       "active-backup",        BOND_MODE_ACTIVEBACKUP},
167 {       "balance-xor",          BOND_MODE_XOR},
168 {       "broadcast",            BOND_MODE_BROADCAST},
169 {       "802.3ad",              BOND_MODE_8023AD},
170 {       "balance-tlb",          BOND_MODE_TLB},
171 {       "balance-alb",          BOND_MODE_ALB},
172 {       NULL,                   -1},
173 };
174
175 struct bond_parm_tbl xmit_hashtype_tbl[] = {
176 {       "layer2",               BOND_XMIT_POLICY_LAYER2},
177 {       "layer3+4",             BOND_XMIT_POLICY_LAYER34},
178 {       "layer2+3",             BOND_XMIT_POLICY_LAYER23},
179 {       NULL,                   -1},
180 };
181
182 struct bond_parm_tbl arp_validate_tbl[] = {
183 {       "none",                 BOND_ARP_VALIDATE_NONE},
184 {       "active",               BOND_ARP_VALIDATE_ACTIVE},
185 {       "backup",               BOND_ARP_VALIDATE_BACKUP},
186 {       "all",                  BOND_ARP_VALIDATE_ALL},
187 {       NULL,                   -1},
188 };
189
190 /*-------------------------- Forward declarations ---------------------------*/
191
192 static void bond_send_gratuitous_arp(struct bonding *bond);
193 static void bond_deinit(struct net_device *bond_dev);
194
195 /*---------------------------- General routines -----------------------------*/
196
197 static const char *bond_mode_name(int mode)
198 {
199         switch (mode) {
200         case BOND_MODE_ROUNDROBIN :
201                 return "load balancing (round-robin)";
202         case BOND_MODE_ACTIVEBACKUP :
203                 return "fault-tolerance (active-backup)";
204         case BOND_MODE_XOR :
205                 return "load balancing (xor)";
206         case BOND_MODE_BROADCAST :
207                 return "fault-tolerance (broadcast)";
208         case BOND_MODE_8023AD:
209                 return "IEEE 802.3ad Dynamic link aggregation";
210         case BOND_MODE_TLB:
211                 return "transmit load balancing";
212         case BOND_MODE_ALB:
213                 return "adaptive load balancing";
214         default:
215                 return "unknown";
216         }
217 }
218
219 /*---------------------------------- VLAN -----------------------------------*/
220
221 /**
222  * bond_add_vlan - add a new vlan id on bond
223  * @bond: bond that got the notification
224  * @vlan_id: the vlan id to add
225  *
226  * Returns -ENOMEM if allocation failed.
227  */
228 static int bond_add_vlan(struct bonding *bond, unsigned short vlan_id)
229 {
230         struct vlan_entry *vlan;
231
232         dprintk("bond: %s, vlan id %d\n",
233                 (bond ? bond->dev->name: "None"), vlan_id);
234
235         vlan = kmalloc(sizeof(struct vlan_entry), GFP_KERNEL);
236         if (!vlan) {
237                 return -ENOMEM;
238         }
239
240         INIT_LIST_HEAD(&vlan->vlan_list);
241         vlan->vlan_id = vlan_id;
242         vlan->vlan_ip = 0;
243
244         write_lock_bh(&bond->lock);
245
246         list_add_tail(&vlan->vlan_list, &bond->vlan_list);
247
248         write_unlock_bh(&bond->lock);
249
250         dprintk("added VLAN ID %d on bond %s\n", vlan_id, bond->dev->name);
251
252         return 0;
253 }
254
255 /**
256  * bond_del_vlan - delete a vlan id from bond
257  * @bond: bond that got the notification
258  * @vlan_id: the vlan id to delete
259  *
260  * returns -ENODEV if @vlan_id was not found in @bond.
261  */
262 static int bond_del_vlan(struct bonding *bond, unsigned short vlan_id)
263 {
264         struct vlan_entry *vlan, *next;
265         int res = -ENODEV;
266
267         dprintk("bond: %s, vlan id %d\n", bond->dev->name, vlan_id);
268
269         write_lock_bh(&bond->lock);
270
271         list_for_each_entry_safe(vlan, next, &bond->vlan_list, vlan_list) {
272                 if (vlan->vlan_id == vlan_id) {
273                         list_del(&vlan->vlan_list);
274
275                         if ((bond->params.mode == BOND_MODE_TLB) ||
276                             (bond->params.mode == BOND_MODE_ALB)) {
277                                 bond_alb_clear_vlan(bond, vlan_id);
278                         }
279
280                         dprintk("removed VLAN ID %d from bond %s\n", vlan_id,
281                                 bond->dev->name);
282
283                         kfree(vlan);
284
285                         if (list_empty(&bond->vlan_list) &&
286                             (bond->slave_cnt == 0)) {
287                                 /* Last VLAN removed and no slaves, so
288                                  * restore block on adding VLANs. This will
289                                  * be removed once new slaves that are not
290                                  * VLAN challenged will be added.
291                                  */
292                                 bond->dev->features |= NETIF_F_VLAN_CHALLENGED;
293                         }
294
295                         res = 0;
296                         goto out;
297                 }
298         }
299
300         dprintk("couldn't find VLAN ID %d in bond %s\n", vlan_id,
301                 bond->dev->name);
302
303 out:
304         write_unlock_bh(&bond->lock);
305         return res;
306 }
307
308 /**
309  * bond_has_challenged_slaves
310  * @bond: the bond we're working on
311  *
312  * Searches the slave list. Returns 1 if a vlan challenged slave
313  * was found, 0 otherwise.
314  *
315  * Assumes bond->lock is held.
316  */
317 static int bond_has_challenged_slaves(struct bonding *bond)
318 {
319         struct slave *slave;
320         int i;
321
322         bond_for_each_slave(bond, slave, i) {
323                 if (slave->dev->features & NETIF_F_VLAN_CHALLENGED) {
324                         dprintk("found VLAN challenged slave - %s\n",
325                                 slave->dev->name);
326                         return 1;
327                 }
328         }
329
330         dprintk("no VLAN challenged slaves found\n");
331         return 0;
332 }
333
334 /**
335  * bond_next_vlan - safely skip to the next item in the vlans list.
336  * @bond: the bond we're working on
337  * @curr: item we're advancing from
338  *
339  * Returns %NULL if list is empty, bond->next_vlan if @curr is %NULL,
340  * or @curr->next otherwise (even if it is @curr itself again).
341  * 
342  * Caller must hold bond->lock
343  */
344 struct vlan_entry *bond_next_vlan(struct bonding *bond, struct vlan_entry *curr)
345 {
346         struct vlan_entry *next, *last;
347
348         if (list_empty(&bond->vlan_list)) {
349                 return NULL;
350         }
351
352         if (!curr) {
353                 next = list_entry(bond->vlan_list.next,
354                                   struct vlan_entry, vlan_list);
355         } else {
356                 last = list_entry(bond->vlan_list.prev,
357                                   struct vlan_entry, vlan_list);
358                 if (last == curr) {
359                         next = list_entry(bond->vlan_list.next,
360                                           struct vlan_entry, vlan_list);
361                 } else {
362                         next = list_entry(curr->vlan_list.next,
363                                           struct vlan_entry, vlan_list);
364                 }
365         }
366
367         return next;
368 }
369
370 /**
371  * bond_dev_queue_xmit - Prepare skb for xmit.
372  * 
373  * @bond: bond device that got this skb for tx.
374  * @skb: hw accel VLAN tagged skb to transmit
375  * @slave_dev: slave that is supposed to xmit this skbuff
376  * 
377  * When the bond gets an skb to transmit that is
378  * already hardware accelerated VLAN tagged, and it
379  * needs to relay this skb to a slave that is not
380  * hw accel capable, the skb needs to be "unaccelerated",
381  * i.e. strip the hwaccel tag and re-insert it as part
382  * of the payload.
383  */
384 int bond_dev_queue_xmit(struct bonding *bond, struct sk_buff *skb, struct net_device *slave_dev)
385 {
386         unsigned short uninitialized_var(vlan_id);
387
388         if (!list_empty(&bond->vlan_list) &&
389             !(slave_dev->features & NETIF_F_HW_VLAN_TX) &&
390             vlan_get_tag(skb, &vlan_id) == 0) {
391                 skb->dev = slave_dev;
392                 skb = vlan_put_tag(skb, vlan_id);
393                 if (!skb) {
394                         /* vlan_put_tag() frees the skb in case of error,
395                          * so return success here so the calling functions
396                          * won't attempt to free is again.
397                          */
398                         return 0;
399                 }
400         } else {
401                 skb->dev = slave_dev;
402         }
403
404         skb->priority = 1;
405         dev_queue_xmit(skb);
406
407         return 0;
408 }
409
410 /*
411  * In the following 3 functions, bond_vlan_rx_register(), bond_vlan_rx_add_vid
412  * and bond_vlan_rx_kill_vid, We don't protect the slave list iteration with a
413  * lock because:
414  * a. This operation is performed in IOCTL context,
415  * b. The operation is protected by the RTNL semaphore in the 8021q code,
416  * c. Holding a lock with BH disabled while directly calling a base driver
417  *    entry point is generally a BAD idea.
418  * 
419  * The design of synchronization/protection for this operation in the 8021q
420  * module is good for one or more VLAN devices over a single physical device
421  * and cannot be extended for a teaming solution like bonding, so there is a
422  * potential race condition here where a net device from the vlan group might
423  * be referenced (either by a base driver or the 8021q code) while it is being
424  * removed from the system. However, it turns out we're not making matters
425  * worse, and if it works for regular VLAN usage it will work here too.
426 */
427
428 /**
429  * bond_vlan_rx_register - Propagates registration to slaves
430  * @bond_dev: bonding net device that got called
431  * @grp: vlan group being registered
432  */
433 static void bond_vlan_rx_register(struct net_device *bond_dev, struct vlan_group *grp)
434 {
435         struct bonding *bond = bond_dev->priv;
436         struct slave *slave;
437         int i;
438
439         bond->vlgrp = grp;
440
441         bond_for_each_slave(bond, slave, i) {
442                 struct net_device *slave_dev = slave->dev;
443
444                 if ((slave_dev->features & NETIF_F_HW_VLAN_RX) &&
445                     slave_dev->vlan_rx_register) {
446                         slave_dev->vlan_rx_register(slave_dev, grp);
447                 }
448         }
449 }
450
451 /**
452  * bond_vlan_rx_add_vid - Propagates adding an id to slaves
453  * @bond_dev: bonding net device that got called
454  * @vid: vlan id being added
455  */
456 static void bond_vlan_rx_add_vid(struct net_device *bond_dev, uint16_t vid)
457 {
458         struct bonding *bond = bond_dev->priv;
459         struct slave *slave;
460         int i, res;
461
462         bond_for_each_slave(bond, slave, i) {
463                 struct net_device *slave_dev = slave->dev;
464
465                 if ((slave_dev->features & NETIF_F_HW_VLAN_FILTER) &&
466                     slave_dev->vlan_rx_add_vid) {
467                         slave_dev->vlan_rx_add_vid(slave_dev, vid);
468                 }
469         }
470
471         res = bond_add_vlan(bond, vid);
472         if (res) {
473                 printk(KERN_ERR DRV_NAME
474                        ": %s: Error: Failed to add vlan id %d\n",
475                        bond_dev->name, vid);
476         }
477 }
478
479 /**
480  * bond_vlan_rx_kill_vid - Propagates deleting an id to slaves
481  * @bond_dev: bonding net device that got called
482  * @vid: vlan id being removed
483  */
484 static void bond_vlan_rx_kill_vid(struct net_device *bond_dev, uint16_t vid)
485 {
486         struct bonding *bond = bond_dev->priv;
487         struct slave *slave;
488         struct net_device *vlan_dev;
489         int i, res;
490
491         bond_for_each_slave(bond, slave, i) {
492                 struct net_device *slave_dev = slave->dev;
493
494                 if ((slave_dev->features & NETIF_F_HW_VLAN_FILTER) &&
495                     slave_dev->vlan_rx_kill_vid) {
496                         /* Save and then restore vlan_dev in the grp array,
497                          * since the slave's driver might clear it.
498                          */
499                         vlan_dev = vlan_group_get_device(bond->vlgrp, vid);
500                         slave_dev->vlan_rx_kill_vid(slave_dev, vid);
501                         vlan_group_set_device(bond->vlgrp, vid, vlan_dev);
502                 }
503         }
504
505         res = bond_del_vlan(bond, vid);
506         if (res) {
507                 printk(KERN_ERR DRV_NAME
508                        ": %s: Error: Failed to remove vlan id %d\n",
509                        bond_dev->name, vid);
510         }
511 }
512
513 static void bond_add_vlans_on_slave(struct bonding *bond, struct net_device *slave_dev)
514 {
515         struct vlan_entry *vlan;
516
517         write_lock_bh(&bond->lock);
518
519         if (list_empty(&bond->vlan_list)) {
520                 goto out;
521         }
522
523         if ((slave_dev->features & NETIF_F_HW_VLAN_RX) &&
524             slave_dev->vlan_rx_register) {
525                 slave_dev->vlan_rx_register(slave_dev, bond->vlgrp);
526         }
527
528         if (!(slave_dev->features & NETIF_F_HW_VLAN_FILTER) ||
529             !(slave_dev->vlan_rx_add_vid)) {
530                 goto out;
531         }
532
533         list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
534                 slave_dev->vlan_rx_add_vid(slave_dev, vlan->vlan_id);
535         }
536
537 out:
538         write_unlock_bh(&bond->lock);
539 }
540
541 static void bond_del_vlans_from_slave(struct bonding *bond, struct net_device *slave_dev)
542 {
543         struct vlan_entry *vlan;
544         struct net_device *vlan_dev;
545
546         write_lock_bh(&bond->lock);
547
548         if (list_empty(&bond->vlan_list)) {
549                 goto out;
550         }
551
552         if (!(slave_dev->features & NETIF_F_HW_VLAN_FILTER) ||
553             !(slave_dev->vlan_rx_kill_vid)) {
554                 goto unreg;
555         }
556
557         list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
558                 /* Save and then restore vlan_dev in the grp array,
559                  * since the slave's driver might clear it.
560                  */
561                 vlan_dev = vlan_group_get_device(bond->vlgrp, vlan->vlan_id);
562                 slave_dev->vlan_rx_kill_vid(slave_dev, vlan->vlan_id);
563                 vlan_group_set_device(bond->vlgrp, vlan->vlan_id, vlan_dev);
564         }
565
566 unreg:
567         if ((slave_dev->features & NETIF_F_HW_VLAN_RX) &&
568             slave_dev->vlan_rx_register) {
569                 slave_dev->vlan_rx_register(slave_dev, NULL);
570         }
571
572 out:
573         write_unlock_bh(&bond->lock);
574 }
575
576 /*------------------------------- Link status -------------------------------*/
577
578 /*
579  * Set the carrier state for the master according to the state of its
580  * slaves.  If any slaves are up, the master is up.  In 802.3ad mode,
581  * do special 802.3ad magic.
582  *
583  * Returns zero if carrier state does not change, nonzero if it does.
584  */
585 static int bond_set_carrier(struct bonding *bond)
586 {
587         struct slave *slave;
588         int i;
589
590         if (bond->slave_cnt == 0)
591                 goto down;
592
593         if (bond->params.mode == BOND_MODE_8023AD)
594                 return bond_3ad_set_carrier(bond);
595
596         bond_for_each_slave(bond, slave, i) {
597                 if (slave->link == BOND_LINK_UP) {
598                         if (!netif_carrier_ok(bond->dev)) {
599                                 netif_carrier_on(bond->dev);
600                                 return 1;
601                         }
602                         return 0;
603                 }
604         }
605
606 down:
607         if (netif_carrier_ok(bond->dev)) {
608                 netif_carrier_off(bond->dev);
609                 return 1;
610         }
611         return 0;
612 }
613
614 /*
615  * Get link speed and duplex from the slave's base driver
616  * using ethtool. If for some reason the call fails or the
617  * values are invalid, fake speed and duplex to 100/Full
618  * and return error.
619  */
620 static int bond_update_speed_duplex(struct slave *slave)
621 {
622         struct net_device *slave_dev = slave->dev;
623         struct ethtool_cmd etool;
624         int res;
625
626         /* Fake speed and duplex */
627         slave->speed = SPEED_100;
628         slave->duplex = DUPLEX_FULL;
629
630         if (!slave_dev->ethtool_ops || !slave_dev->ethtool_ops->get_settings)
631                 return -1;
632
633         res = slave_dev->ethtool_ops->get_settings(slave_dev, &etool);
634         if (res < 0)
635                 return -1;
636
637         switch (etool.speed) {
638         case SPEED_10:
639         case SPEED_100:
640         case SPEED_1000:
641         case SPEED_10000:
642                 break;
643         default:
644                 return -1;
645         }
646
647         switch (etool.duplex) {
648         case DUPLEX_FULL:
649         case DUPLEX_HALF:
650                 break;
651         default:
652                 return -1;
653         }
654
655         slave->speed = etool.speed;
656         slave->duplex = etool.duplex;
657
658         return 0;
659 }
660
661 /*
662  * if <dev> supports MII link status reporting, check its link status.
663  *
664  * We either do MII/ETHTOOL ioctls, or check netif_carrier_ok(),
665  * depening upon the setting of the use_carrier parameter.
666  *
667  * Return either BMSR_LSTATUS, meaning that the link is up (or we
668  * can't tell and just pretend it is), or 0, meaning that the link is
669  * down.
670  *
671  * If reporting is non-zero, instead of faking link up, return -1 if
672  * both ETHTOOL and MII ioctls fail (meaning the device does not
673  * support them).  If use_carrier is set, return whatever it says.
674  * It'd be nice if there was a good way to tell if a driver supports
675  * netif_carrier, but there really isn't.
676  */
677 static int bond_check_dev_link(struct bonding *bond, struct net_device *slave_dev, int reporting)
678 {
679         static int (* ioctl)(struct net_device *, struct ifreq *, int);
680         struct ifreq ifr;
681         struct mii_ioctl_data *mii;
682
683         if (bond->params.use_carrier) {
684                 return netif_carrier_ok(slave_dev) ? BMSR_LSTATUS : 0;
685         }
686
687         ioctl = slave_dev->do_ioctl;
688         if (ioctl) {
689                 /* TODO: set pointer to correct ioctl on a per team member */
690                 /*       bases to make this more efficient. that is, once  */
691                 /*       we determine the correct ioctl, we will always    */
692                 /*       call it and not the others for that team          */
693                 /*       member.                                           */
694
695                 /*
696                  * We cannot assume that SIOCGMIIPHY will also read a
697                  * register; not all network drivers (e.g., e100)
698                  * support that.
699                  */
700
701                 /* Yes, the mii is overlaid on the ifreq.ifr_ifru */
702                 strncpy(ifr.ifr_name, slave_dev->name, IFNAMSIZ);
703                 mii = if_mii(&ifr);
704                 if (IOCTL(slave_dev, &ifr, SIOCGMIIPHY) == 0) {
705                         mii->reg_num = MII_BMSR;
706                         if (IOCTL(slave_dev, &ifr, SIOCGMIIREG) == 0) {
707                                 return (mii->val_out & BMSR_LSTATUS);
708                         }
709                 }
710         }
711
712         /*
713          * Some drivers cache ETHTOOL_GLINK for a period of time so we only
714          * attempt to get link status from it if the above MII ioctls fail.
715          */
716         if (slave_dev->ethtool_ops) {
717                 if (slave_dev->ethtool_ops->get_link) {
718                         u32 link;
719
720                         link = slave_dev->ethtool_ops->get_link(slave_dev);
721
722                         return link ? BMSR_LSTATUS : 0;
723                 }
724         }
725
726         /*
727          * If reporting, report that either there's no dev->do_ioctl,
728          * or both SIOCGMIIREG and get_link failed (meaning that we
729          * cannot report link status).  If not reporting, pretend
730          * we're ok.
731          */
732         return (reporting ? -1 : BMSR_LSTATUS);
733 }
734
735 /*----------------------------- Multicast list ------------------------------*/
736
737 /*
738  * Returns 0 if dmi1 and dmi2 are the same, non-0 otherwise
739  */
740 static inline int bond_is_dmi_same(struct dev_mc_list *dmi1, struct dev_mc_list *dmi2)
741 {
742         return memcmp(dmi1->dmi_addr, dmi2->dmi_addr, dmi1->dmi_addrlen) == 0 &&
743                         dmi1->dmi_addrlen == dmi2->dmi_addrlen;
744 }
745
746 /*
747  * returns dmi entry if found, NULL otherwise
748  */
749 static struct dev_mc_list *bond_mc_list_find_dmi(struct dev_mc_list *dmi, struct dev_mc_list *mc_list)
750 {
751         struct dev_mc_list *idmi;
752
753         for (idmi = mc_list; idmi; idmi = idmi->next) {
754                 if (bond_is_dmi_same(dmi, idmi)) {
755                         return idmi;
756                 }
757         }
758
759         return NULL;
760 }
761
762 /*
763  * Push the promiscuity flag down to appropriate slaves
764  */
765 static void bond_set_promiscuity(struct bonding *bond, int inc)
766 {
767         if (USES_PRIMARY(bond->params.mode)) {
768                 /* write lock already acquired */
769                 if (bond->curr_active_slave) {
770                         dev_set_promiscuity(bond->curr_active_slave->dev, inc);
771                 }
772         } else {
773                 struct slave *slave;
774                 int i;
775                 bond_for_each_slave(bond, slave, i) {
776                         dev_set_promiscuity(slave->dev, inc);
777                 }
778         }
779 }
780
781 /*
782  * Push the allmulti flag down to all slaves
783  */
784 static void bond_set_allmulti(struct bonding *bond, int inc)
785 {
786         if (USES_PRIMARY(bond->params.mode)) {
787                 /* write lock already acquired */
788                 if (bond->curr_active_slave) {
789                         dev_set_allmulti(bond->curr_active_slave->dev, inc);
790                 }
791         } else {
792                 struct slave *slave;
793                 int i;
794                 bond_for_each_slave(bond, slave, i) {
795                         dev_set_allmulti(slave->dev, inc);
796                 }
797         }
798 }
799
800 /*
801  * Add a Multicast address to slaves
802  * according to mode
803  */
804 static void bond_mc_add(struct bonding *bond, void *addr, int alen)
805 {
806         if (USES_PRIMARY(bond->params.mode)) {
807                 /* write lock already acquired */
808                 if (bond->curr_active_slave) {
809                         dev_mc_add(bond->curr_active_slave->dev, addr, alen, 0);
810                 }
811         } else {
812                 struct slave *slave;
813                 int i;
814                 bond_for_each_slave(bond, slave, i) {
815                         dev_mc_add(slave->dev, addr, alen, 0);
816                 }
817         }
818 }
819
820 /*
821  * Remove a multicast address from slave
822  * according to mode
823  */
824 static void bond_mc_delete(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_delete(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_delete(slave->dev, addr, alen, 0);
836                 }
837         }
838 }
839
840
841 /*
842  * Retrieve the list of registered multicast addresses for the bonding
843  * device and retransmit an IGMP JOIN request to the current active
844  * slave.
845  */
846 static void bond_resend_igmp_join_requests(struct bonding *bond)
847 {
848         struct in_device *in_dev;
849         struct ip_mc_list *im;
850
851         rcu_read_lock();
852         in_dev = __in_dev_get_rcu(bond->dev);
853         if (in_dev) {
854                 for (im = in_dev->mc_list; im; im = im->next) {
855                         ip_mc_rejoin_group(im);
856                 }
857         }
858
859         rcu_read_unlock();
860 }
861
862 /*
863  * Totally destroys the mc_list in bond
864  */
865 static void bond_mc_list_destroy(struct bonding *bond)
866 {
867         struct dev_mc_list *dmi;
868
869         dmi = bond->mc_list;
870         while (dmi) {
871                 bond->mc_list = dmi->next;
872                 kfree(dmi);
873                 dmi = bond->mc_list;
874         }
875         bond->mc_list = NULL;
876 }
877
878 /*
879  * Copy all the Multicast addresses from src to the bonding device dst
880  */
881 static int bond_mc_list_copy(struct dev_mc_list *mc_list, struct bonding *bond,
882                              gfp_t gfp_flag)
883 {
884         struct dev_mc_list *dmi, *new_dmi;
885
886         for (dmi = mc_list; dmi; dmi = dmi->next) {
887                 new_dmi = kmalloc(sizeof(struct dev_mc_list), gfp_flag);
888
889                 if (!new_dmi) {
890                         /* FIXME: Potential memory leak !!! */
891                         return -ENOMEM;
892                 }
893
894                 new_dmi->next = bond->mc_list;
895                 bond->mc_list = new_dmi;
896                 new_dmi->dmi_addrlen = dmi->dmi_addrlen;
897                 memcpy(new_dmi->dmi_addr, dmi->dmi_addr, dmi->dmi_addrlen);
898                 new_dmi->dmi_users = dmi->dmi_users;
899                 new_dmi->dmi_gusers = dmi->dmi_gusers;
900         }
901
902         return 0;
903 }
904
905 /*
906  * flush all members of flush->mc_list from device dev->mc_list
907  */
908 static void bond_mc_list_flush(struct net_device *bond_dev, struct net_device *slave_dev)
909 {
910         struct bonding *bond = bond_dev->priv;
911         struct dev_mc_list *dmi;
912
913         for (dmi = bond_dev->mc_list; dmi; dmi = dmi->next) {
914                 dev_mc_delete(slave_dev, dmi->dmi_addr, dmi->dmi_addrlen, 0);
915         }
916
917         if (bond->params.mode == BOND_MODE_8023AD) {
918                 /* del lacpdu mc addr from mc list */
919                 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
920
921                 dev_mc_delete(slave_dev, lacpdu_multicast, ETH_ALEN, 0);
922         }
923 }
924
925 /*--------------------------- Active slave change ---------------------------*/
926
927 /*
928  * Update the mc list and multicast-related flags for the new and
929  * old active slaves (if any) according to the multicast mode, and
930  * promiscuous flags unconditionally.
931  */
932 static void bond_mc_swap(struct bonding *bond, struct slave *new_active, struct slave *old_active)
933 {
934         struct dev_mc_list *dmi;
935
936         if (!USES_PRIMARY(bond->params.mode)) {
937                 /* nothing to do -  mc list is already up-to-date on
938                  * all slaves
939                  */
940                 return;
941         }
942
943         if (old_active) {
944                 if (bond->dev->flags & IFF_PROMISC) {
945                         dev_set_promiscuity(old_active->dev, -1);
946                 }
947
948                 if (bond->dev->flags & IFF_ALLMULTI) {
949                         dev_set_allmulti(old_active->dev, -1);
950                 }
951
952                 for (dmi = bond->dev->mc_list; dmi; dmi = dmi->next) {
953                         dev_mc_delete(old_active->dev, dmi->dmi_addr, dmi->dmi_addrlen, 0);
954                 }
955         }
956
957         if (new_active) {
958                 if (bond->dev->flags & IFF_PROMISC) {
959                         dev_set_promiscuity(new_active->dev, 1);
960                 }
961
962                 if (bond->dev->flags & IFF_ALLMULTI) {
963                         dev_set_allmulti(new_active->dev, 1);
964                 }
965
966                 for (dmi = bond->dev->mc_list; dmi; dmi = dmi->next) {
967                         dev_mc_add(new_active->dev, dmi->dmi_addr, dmi->dmi_addrlen, 0);
968                 }
969                 bond_resend_igmp_join_requests(bond);
970         }
971 }
972
973 /**
974  * find_best_interface - select the best available slave to be the active one
975  * @bond: our bonding struct
976  *
977  * Warning: Caller must hold curr_slave_lock for writing.
978  */
979 static struct slave *bond_find_best_slave(struct bonding *bond)
980 {
981         struct slave *new_active, *old_active;
982         struct slave *bestslave = NULL;
983         int mintime = bond->params.updelay;
984         int i;
985
986         new_active = old_active = bond->curr_active_slave;
987
988         if (!new_active) { /* there were no active slaves left */
989                 if (bond->slave_cnt > 0) {  /* found one slave */
990                         new_active = bond->first_slave;
991                 } else {
992                         return NULL; /* still no slave, return NULL */
993                 }
994         }
995
996         /* first try the primary link; if arping, a link must tx/rx traffic
997          * before it can be considered the curr_active_slave - also, we would skip
998          * slaves between the curr_active_slave and primary_slave that may be up
999          * and able to arp
1000          */
1001         if ((bond->primary_slave) &&
1002             (!bond->params.arp_interval) &&
1003             (IS_UP(bond->primary_slave->dev))) {
1004                 new_active = bond->primary_slave;
1005         }
1006
1007         /* remember where to stop iterating over the slaves */
1008         old_active = new_active;
1009
1010         bond_for_each_slave_from(bond, new_active, i, old_active) {
1011                 if (IS_UP(new_active->dev)) {
1012                         if (new_active->link == BOND_LINK_UP) {
1013                                 return new_active;
1014                         } else if (new_active->link == BOND_LINK_BACK) {
1015                                 /* link up, but waiting for stabilization */
1016                                 if (new_active->delay < mintime) {
1017                                         mintime = new_active->delay;
1018                                         bestslave = new_active;
1019                                 }
1020                         }
1021                 }
1022         }
1023
1024         return bestslave;
1025 }
1026
1027 /**
1028  * change_active_interface - change the active slave into the specified one
1029  * @bond: our bonding struct
1030  * @new: the new slave to make the active one
1031  *
1032  * Set the new slave to the bond's settings and unset them on the old
1033  * curr_active_slave.
1034  * Setting include flags, mc-list, promiscuity, allmulti, etc.
1035  *
1036  * If @new's link state is %BOND_LINK_BACK we'll set it to %BOND_LINK_UP,
1037  * because it is apparently the best available slave we have, even though its
1038  * updelay hasn't timed out yet.
1039  *
1040  * Warning: Caller must hold curr_slave_lock for writing.
1041  */
1042 void bond_change_active_slave(struct bonding *bond, struct slave *new_active)
1043 {
1044         struct slave *old_active = bond->curr_active_slave;
1045
1046         if (old_active == new_active) {
1047                 return;
1048         }
1049
1050         if (new_active) {
1051                 if (new_active->link == BOND_LINK_BACK) {
1052                         if (USES_PRIMARY(bond->params.mode)) {
1053                                 printk(KERN_INFO DRV_NAME
1054                                        ": %s: making interface %s the new "
1055                                        "active one %d ms earlier.\n",
1056                                        bond->dev->name, new_active->dev->name,
1057                                        (bond->params.updelay - new_active->delay) * bond->params.miimon);
1058                         }
1059
1060                         new_active->delay = 0;
1061                         new_active->link = BOND_LINK_UP;
1062                         new_active->jiffies = jiffies;
1063
1064                         if (bond->params.mode == BOND_MODE_8023AD) {
1065                                 bond_3ad_handle_link_change(new_active, BOND_LINK_UP);
1066                         }
1067
1068                         if ((bond->params.mode == BOND_MODE_TLB) ||
1069                             (bond->params.mode == BOND_MODE_ALB)) {
1070                                 bond_alb_handle_link_change(bond, new_active, BOND_LINK_UP);
1071                         }
1072                 } else {
1073                         if (USES_PRIMARY(bond->params.mode)) {
1074                                 printk(KERN_INFO DRV_NAME
1075                                        ": %s: making interface %s the new "
1076                                        "active one.\n",
1077                                        bond->dev->name, new_active->dev->name);
1078                         }
1079                 }
1080         }
1081
1082         if (USES_PRIMARY(bond->params.mode)) {
1083                 bond_mc_swap(bond, new_active, old_active);
1084         }
1085
1086         if ((bond->params.mode == BOND_MODE_TLB) ||
1087             (bond->params.mode == BOND_MODE_ALB)) {
1088                 bond_alb_handle_active_change(bond, new_active);
1089                 if (old_active)
1090                         bond_set_slave_inactive_flags(old_active);
1091                 if (new_active)
1092                         bond_set_slave_active_flags(new_active);
1093         } else {
1094                 bond->curr_active_slave = new_active;
1095         }
1096
1097         if (bond->params.mode == BOND_MODE_ACTIVEBACKUP) {
1098                 if (old_active) {
1099                         bond_set_slave_inactive_flags(old_active);
1100                 }
1101
1102                 if (new_active) {
1103                         bond_set_slave_active_flags(new_active);
1104                 }
1105
1106                 /* when bonding does not set the slave MAC address, the bond MAC
1107                  * address is the one of the active slave.
1108                  */
1109                 if (new_active && bond->params.fail_over_mac)
1110                         memcpy(bond->dev->dev_addr,  new_active->dev->dev_addr,
1111                                 new_active->dev->addr_len);
1112                 if (bond->curr_active_slave &&
1113                         test_bit(__LINK_STATE_LINKWATCH_PENDING,
1114                                         &bond->curr_active_slave->dev->state)) {
1115                         dprintk("delaying gratuitous arp on %s\n",
1116                                 bond->curr_active_slave->dev->name);
1117                         bond->send_grat_arp = 1;
1118                 } else
1119                         bond_send_gratuitous_arp(bond);
1120         }
1121 }
1122
1123 /**
1124  * bond_select_active_slave - select a new active slave, if needed
1125  * @bond: our bonding struct
1126  *
1127  * This functions shoud be called when one of the following occurs:
1128  * - The old curr_active_slave has been released or lost its link.
1129  * - The primary_slave has got its link back.
1130  * - A slave has got its link back and there's no old curr_active_slave.
1131  *
1132  * Warning: Caller must hold curr_slave_lock for writing.
1133  */
1134 void bond_select_active_slave(struct bonding *bond)
1135 {
1136         struct slave *best_slave;
1137         int rv;
1138
1139         best_slave = bond_find_best_slave(bond);
1140         if (best_slave != bond->curr_active_slave) {
1141                 bond_change_active_slave(bond, best_slave);
1142                 rv = bond_set_carrier(bond);
1143                 if (!rv)
1144                         return;
1145
1146                 if (netif_carrier_ok(bond->dev)) {
1147                         printk(KERN_INFO DRV_NAME
1148                                ": %s: first active interface up!\n",
1149                                bond->dev->name);
1150                 } else {
1151                         printk(KERN_INFO DRV_NAME ": %s: "
1152                                "now running without any active interface !\n",
1153                                bond->dev->name);
1154                 }
1155         }
1156 }
1157
1158 /*--------------------------- slave list handling ---------------------------*/
1159
1160 /*
1161  * This function attaches the slave to the end of list.
1162  *
1163  * bond->lock held for writing by caller.
1164  */
1165 static void bond_attach_slave(struct bonding *bond, struct slave *new_slave)
1166 {
1167         if (bond->first_slave == NULL) { /* attaching the first slave */
1168                 new_slave->next = new_slave;
1169                 new_slave->prev = new_slave;
1170                 bond->first_slave = new_slave;
1171         } else {
1172                 new_slave->next = bond->first_slave;
1173                 new_slave->prev = bond->first_slave->prev;
1174                 new_slave->next->prev = new_slave;
1175                 new_slave->prev->next = new_slave;
1176         }
1177
1178         bond->slave_cnt++;
1179 }
1180
1181 /*
1182  * This function detaches the slave from the list.
1183  * WARNING: no check is made to verify if the slave effectively
1184  * belongs to <bond>.
1185  * Nothing is freed on return, structures are just unchained.
1186  * If any slave pointer in bond was pointing to <slave>,
1187  * it should be changed by the calling function.
1188  *
1189  * bond->lock held for writing by caller.
1190  */
1191 static void bond_detach_slave(struct bonding *bond, struct slave *slave)
1192 {
1193         if (slave->next) {
1194                 slave->next->prev = slave->prev;
1195         }
1196
1197         if (slave->prev) {
1198                 slave->prev->next = slave->next;
1199         }
1200
1201         if (bond->first_slave == slave) { /* slave is the first slave */
1202                 if (bond->slave_cnt > 1) { /* there are more slave */
1203                         bond->first_slave = slave->next;
1204                 } else {
1205                         bond->first_slave = NULL; /* slave was the last one */
1206                 }
1207         }
1208
1209         slave->next = NULL;
1210         slave->prev = NULL;
1211         bond->slave_cnt--;
1212 }
1213
1214 /*---------------------------------- IOCTL ----------------------------------*/
1215
1216 static int bond_sethwaddr(struct net_device *bond_dev,
1217                           struct net_device *slave_dev)
1218 {
1219         dprintk("bond_dev=%p\n", bond_dev);
1220         dprintk("slave_dev=%p\n", slave_dev);
1221         dprintk("slave_dev->addr_len=%d\n", slave_dev->addr_len);
1222         memcpy(bond_dev->dev_addr, slave_dev->dev_addr, slave_dev->addr_len);
1223         return 0;
1224 }
1225
1226 #define BOND_VLAN_FEATURES \
1227         (NETIF_F_VLAN_CHALLENGED | NETIF_F_HW_VLAN_RX | NETIF_F_HW_VLAN_TX | \
1228          NETIF_F_HW_VLAN_FILTER)
1229
1230 /* 
1231  * Compute the common dev->feature set available to all slaves.  Some
1232  * feature bits are managed elsewhere, so preserve those feature bits
1233  * on the master device.
1234  */
1235 static int bond_compute_features(struct bonding *bond)
1236 {
1237         struct slave *slave;
1238         struct net_device *bond_dev = bond->dev;
1239         unsigned long features = bond_dev->features;
1240         unsigned short max_hard_header_len = max((u16)ETH_HLEN,
1241                                                 bond_dev->hard_header_len);
1242         int i;
1243
1244         features &= ~(NETIF_F_ALL_CSUM | BOND_VLAN_FEATURES);
1245         features |= NETIF_F_SG | NETIF_F_FRAGLIST | NETIF_F_HIGHDMA |
1246                     NETIF_F_GSO_MASK | NETIF_F_NO_CSUM;
1247
1248         bond_for_each_slave(bond, slave, i) {
1249                 features = netdev_compute_features(features,
1250                                                    slave->dev->features);
1251                 if (slave->dev->hard_header_len > max_hard_header_len)
1252                         max_hard_header_len = slave->dev->hard_header_len;
1253         }
1254
1255         features |= (bond_dev->features & BOND_VLAN_FEATURES);
1256         bond_dev->features = features;
1257         bond_dev->hard_header_len = max_hard_header_len;
1258
1259         return 0;
1260 }
1261
1262
1263 static void bond_setup_by_slave(struct net_device *bond_dev,
1264                                 struct net_device *slave_dev)
1265 {
1266         struct bonding *bond = bond_dev->priv;
1267
1268         bond_dev->neigh_setup           = slave_dev->neigh_setup;
1269         bond_dev->header_ops            = slave_dev->header_ops;
1270
1271         bond_dev->type              = slave_dev->type;
1272         bond_dev->hard_header_len   = slave_dev->hard_header_len;
1273         bond_dev->addr_len          = slave_dev->addr_len;
1274
1275         memcpy(bond_dev->broadcast, slave_dev->broadcast,
1276                 slave_dev->addr_len);
1277         bond->setup_by_slave = 1;
1278 }
1279
1280 /* enslave device <slave> to bond device <master> */
1281 int bond_enslave(struct net_device *bond_dev, struct net_device *slave_dev)
1282 {
1283         struct bonding *bond = bond_dev->priv;
1284         struct slave *new_slave = NULL;
1285         struct dev_mc_list *dmi;
1286         struct sockaddr addr;
1287         int link_reporting;
1288         int old_features = bond_dev->features;
1289         int res = 0;
1290
1291         if (!bond->params.use_carrier && slave_dev->ethtool_ops == NULL &&
1292                 slave_dev->do_ioctl == NULL) {
1293                 printk(KERN_WARNING DRV_NAME
1294                        ": %s: Warning: no link monitoring support for %s\n",
1295                        bond_dev->name, slave_dev->name);
1296         }
1297
1298         /* bond must be initialized by bond_open() before enslaving */
1299         if (!(bond_dev->flags & IFF_UP)) {
1300                 printk(KERN_WARNING DRV_NAME
1301                         " %s: master_dev is not up in bond_enslave\n",
1302                         bond_dev->name);
1303         }
1304
1305         /* already enslaved */
1306         if (slave_dev->flags & IFF_SLAVE) {
1307                 dprintk("Error, Device was already enslaved\n");
1308                 return -EBUSY;
1309         }
1310
1311         /* vlan challenged mutual exclusion */
1312         /* no need to lock since we're protected by rtnl_lock */
1313         if (slave_dev->features & NETIF_F_VLAN_CHALLENGED) {
1314                 dprintk("%s: NETIF_F_VLAN_CHALLENGED\n", slave_dev->name);
1315                 if (!list_empty(&bond->vlan_list)) {
1316                         printk(KERN_ERR DRV_NAME
1317                                ": %s: Error: cannot enslave VLAN "
1318                                "challenged slave %s on VLAN enabled "
1319                                "bond %s\n", bond_dev->name, slave_dev->name,
1320                                bond_dev->name);
1321                         return -EPERM;
1322                 } else {
1323                         printk(KERN_WARNING DRV_NAME
1324                                ": %s: Warning: enslaved VLAN challenged "
1325                                "slave %s. Adding VLANs will be blocked as "
1326                                "long as %s is part of bond %s\n",
1327                                bond_dev->name, slave_dev->name, slave_dev->name,
1328                                bond_dev->name);
1329                         bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
1330                 }
1331         } else {
1332                 dprintk("%s: ! NETIF_F_VLAN_CHALLENGED\n", slave_dev->name);
1333                 if (bond->slave_cnt == 0) {
1334                         /* First slave, and it is not VLAN challenged,
1335                          * so remove the block of adding VLANs over the bond.
1336                          */
1337                         bond_dev->features &= ~NETIF_F_VLAN_CHALLENGED;
1338                 }
1339         }
1340
1341         /*
1342          * Old ifenslave binaries are no longer supported.  These can
1343          * be identified with moderate accurary by the state of the slave:
1344          * the current ifenslave will set the interface down prior to
1345          * enslaving it; the old ifenslave will not.
1346          */
1347         if ((slave_dev->flags & IFF_UP)) {
1348                 printk(KERN_ERR DRV_NAME ": %s is up. "
1349                        "This may be due to an out of date ifenslave.\n",
1350                        slave_dev->name);
1351                 res = -EPERM;
1352                 goto err_undo_flags;
1353         }
1354
1355         /* set bonding device ether type by slave - bonding netdevices are
1356          * created with ether_setup, so when the slave type is not ARPHRD_ETHER
1357          * there is a need to override some of the type dependent attribs/funcs.
1358          *
1359          * bond ether type mutual exclusion - don't allow slaves of dissimilar
1360          * ether type (eg ARPHRD_ETHER and ARPHRD_INFINIBAND) share the same bond
1361          */
1362         if (bond->slave_cnt == 0) {
1363                 if (slave_dev->type != ARPHRD_ETHER)
1364                         bond_setup_by_slave(bond_dev, slave_dev);
1365         } else if (bond_dev->type != slave_dev->type) {
1366                 printk(KERN_ERR DRV_NAME ": %s ether type (%d) is different "
1367                         "from other slaves (%d), can not enslave it.\n",
1368                         slave_dev->name,
1369                         slave_dev->type, bond_dev->type);
1370                         res = -EINVAL;
1371                         goto err_undo_flags;
1372         }
1373
1374         if (slave_dev->set_mac_address == NULL) {
1375                 if (bond->slave_cnt == 0) {
1376                         printk(KERN_WARNING DRV_NAME
1377                                ": %s: Warning: The first slave device "
1378                                "specified does not support setting the MAC "
1379                                "address. Enabling the fail_over_mac option.",
1380                                bond_dev->name);
1381                         bond->params.fail_over_mac = 1;
1382                 } else if (!bond->params.fail_over_mac) {
1383                         printk(KERN_ERR DRV_NAME
1384                                 ": %s: Error: The slave device specified "
1385                                 "does not support setting the MAC address, "
1386                                 "but fail_over_mac is not enabled.\n"
1387                                 , bond_dev->name);
1388                         res = -EOPNOTSUPP;
1389                         goto err_undo_flags;
1390                 }
1391         }
1392
1393         new_slave = kzalloc(sizeof(struct slave), GFP_KERNEL);
1394         if (!new_slave) {
1395                 res = -ENOMEM;
1396                 goto err_undo_flags;
1397         }
1398
1399         /* save slave's original flags before calling
1400          * netdev_set_master and dev_open
1401          */
1402         new_slave->original_flags = slave_dev->flags;
1403
1404         /*
1405          * Save slave's original ("permanent") mac address for modes
1406          * that need it, and for restoring it upon release, and then
1407          * set it to the master's address
1408          */
1409         memcpy(new_slave->perm_hwaddr, slave_dev->dev_addr, ETH_ALEN);
1410
1411         if (!bond->params.fail_over_mac) {
1412                 /*
1413                  * Set slave to master's mac address.  The application already
1414                  * set the master's mac address to that of the first slave
1415                  */
1416                 memcpy(addr.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
1417                 addr.sa_family = slave_dev->type;
1418                 res = dev_set_mac_address(slave_dev, &addr);
1419                 if (res) {
1420                         dprintk("Error %d calling set_mac_address\n", res);
1421                         goto err_free;
1422                 }
1423         }
1424
1425         res = netdev_set_master(slave_dev, bond_dev);
1426         if (res) {
1427                 dprintk("Error %d calling netdev_set_master\n", res);
1428                 goto err_close;
1429         }
1430         /* open the slave since the application closed it */
1431         res = dev_open(slave_dev);
1432         if (res) {
1433                 dprintk("Openning slave %s failed\n", slave_dev->name);
1434                 goto err_restore_mac;
1435         }
1436
1437         new_slave->dev = slave_dev;
1438         slave_dev->priv_flags |= IFF_BONDING;
1439
1440         if ((bond->params.mode == BOND_MODE_TLB) ||
1441             (bond->params.mode == BOND_MODE_ALB)) {
1442                 /* bond_alb_init_slave() must be called before all other stages since
1443                  * it might fail and we do not want to have to undo everything
1444                  */
1445                 res = bond_alb_init_slave(bond, new_slave);
1446                 if (res) {
1447                         goto err_unset_master;
1448                 }
1449         }
1450
1451         /* If the mode USES_PRIMARY, then the new slave gets the
1452          * master's promisc (and mc) settings only if it becomes the
1453          * curr_active_slave, and that is taken care of later when calling
1454          * bond_change_active()
1455          */
1456         if (!USES_PRIMARY(bond->params.mode)) {
1457                 /* set promiscuity level to new slave */
1458                 if (bond_dev->flags & IFF_PROMISC) {
1459                         dev_set_promiscuity(slave_dev, 1);
1460                 }
1461
1462                 /* set allmulti level to new slave */
1463                 if (bond_dev->flags & IFF_ALLMULTI) {
1464                         dev_set_allmulti(slave_dev, 1);
1465                 }
1466
1467                 netif_tx_lock_bh(bond_dev);
1468                 /* upload master's mc_list to new slave */
1469                 for (dmi = bond_dev->mc_list; dmi; dmi = dmi->next) {
1470                         dev_mc_add (slave_dev, dmi->dmi_addr, dmi->dmi_addrlen, 0);
1471                 }
1472                 netif_tx_unlock_bh(bond_dev);
1473         }
1474
1475         if (bond->params.mode == BOND_MODE_8023AD) {
1476                 /* add lacpdu mc addr to mc list */
1477                 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
1478
1479                 dev_mc_add(slave_dev, lacpdu_multicast, ETH_ALEN, 0);
1480         }
1481
1482         bond_add_vlans_on_slave(bond, slave_dev);
1483
1484         write_lock_bh(&bond->lock);
1485
1486         bond_attach_slave(bond, new_slave);
1487
1488         new_slave->delay = 0;
1489         new_slave->link_failure_count = 0;
1490
1491         bond_compute_features(bond);
1492
1493         new_slave->last_arp_rx = jiffies;
1494
1495         if (bond->params.miimon && !bond->params.use_carrier) {
1496                 link_reporting = bond_check_dev_link(bond, slave_dev, 1);
1497
1498                 if ((link_reporting == -1) && !bond->params.arp_interval) {
1499                         /*
1500                          * miimon is set but a bonded network driver
1501                          * does not support ETHTOOL/MII and
1502                          * arp_interval is not set.  Note: if
1503                          * use_carrier is enabled, we will never go
1504                          * here (because netif_carrier is always
1505                          * supported); thus, we don't need to change
1506                          * the messages for netif_carrier.
1507                          */
1508                         printk(KERN_WARNING DRV_NAME
1509                                ": %s: Warning: MII and ETHTOOL support not "
1510                                "available for interface %s, and "
1511                                "arp_interval/arp_ip_target module parameters "
1512                                "not specified, thus bonding will not detect "
1513                                "link failures! see bonding.txt for details.\n",
1514                                bond_dev->name, slave_dev->name);
1515                 } else if (link_reporting == -1) {
1516                         /* unable get link status using mii/ethtool */
1517                         printk(KERN_WARNING DRV_NAME
1518                                ": %s: Warning: can't get link status from "
1519                                "interface %s; the network driver associated "
1520                                "with this interface does not support MII or "
1521                                "ETHTOOL link status reporting, thus miimon "
1522                                "has no effect on this interface.\n",
1523                                bond_dev->name, slave_dev->name);
1524                 }
1525         }
1526
1527         /* check for initial state */
1528         if (!bond->params.miimon ||
1529             (bond_check_dev_link(bond, slave_dev, 0) == BMSR_LSTATUS)) {
1530                 if (bond->params.updelay) {
1531                         dprintk("Initial state of slave_dev is "
1532                                 "BOND_LINK_BACK\n");
1533                         new_slave->link  = BOND_LINK_BACK;
1534                         new_slave->delay = bond->params.updelay;
1535                 } else {
1536                         dprintk("Initial state of slave_dev is "
1537                                 "BOND_LINK_UP\n");
1538                         new_slave->link  = BOND_LINK_UP;
1539                 }
1540                 new_slave->jiffies = jiffies;
1541         } else {
1542                 dprintk("Initial state of slave_dev is "
1543                         "BOND_LINK_DOWN\n");
1544                 new_slave->link  = BOND_LINK_DOWN;
1545         }
1546
1547         if (bond_update_speed_duplex(new_slave) &&
1548             (new_slave->link != BOND_LINK_DOWN)) {
1549                 printk(KERN_WARNING DRV_NAME
1550                        ": %s: Warning: failed to get speed and duplex from %s, "
1551                        "assumed to be 100Mb/sec and Full.\n",
1552                        bond_dev->name, new_slave->dev->name);
1553
1554                 if (bond->params.mode == BOND_MODE_8023AD) {
1555                         printk(KERN_WARNING DRV_NAME
1556                                ": %s: Warning: Operation of 802.3ad mode requires ETHTOOL "
1557                                "support in base driver for proper aggregator "
1558                                "selection.\n", bond_dev->name);
1559                 }
1560         }
1561
1562         if (USES_PRIMARY(bond->params.mode) && bond->params.primary[0]) {
1563                 /* if there is a primary slave, remember it */
1564                 if (strcmp(bond->params.primary, new_slave->dev->name) == 0) {
1565                         bond->primary_slave = new_slave;
1566                 }
1567         }
1568
1569         switch (bond->params.mode) {
1570         case BOND_MODE_ACTIVEBACKUP:
1571                 bond_set_slave_inactive_flags(new_slave);
1572                 bond_select_active_slave(bond);
1573                 break;
1574         case BOND_MODE_8023AD:
1575                 /* in 802.3ad mode, the internal mechanism
1576                  * will activate the slaves in the selected
1577                  * aggregator
1578                  */
1579                 bond_set_slave_inactive_flags(new_slave);
1580                 /* if this is the first slave */
1581                 if (bond->slave_cnt == 1) {
1582                         SLAVE_AD_INFO(new_slave).id = 1;
1583                         /* Initialize AD with the number of times that the AD timer is called in 1 second
1584                          * can be called only after the mac address of the bond is set
1585                          */
1586                         bond_3ad_initialize(bond, 1000/AD_TIMER_INTERVAL,
1587                                             bond->params.lacp_fast);
1588                 } else {
1589                         SLAVE_AD_INFO(new_slave).id =
1590                                 SLAVE_AD_INFO(new_slave->prev).id + 1;
1591                 }
1592
1593                 bond_3ad_bind_slave(new_slave);
1594                 break;
1595         case BOND_MODE_TLB:
1596         case BOND_MODE_ALB:
1597                 new_slave->state = BOND_STATE_ACTIVE;
1598                 bond_set_slave_inactive_flags(new_slave);
1599                 break;
1600         default:
1601                 dprintk("This slave is always active in trunk mode\n");
1602
1603                 /* always active in trunk mode */
1604                 new_slave->state = BOND_STATE_ACTIVE;
1605
1606                 /* In trunking mode there is little meaning to curr_active_slave
1607                  * anyway (it holds no special properties of the bond device),
1608                  * so we can change it without calling change_active_interface()
1609                  */
1610                 if (!bond->curr_active_slave) {
1611                         bond->curr_active_slave = new_slave;
1612                 }
1613                 break;
1614         } /* switch(bond_mode) */
1615
1616         bond_set_carrier(bond);
1617
1618         write_unlock_bh(&bond->lock);
1619
1620         res = bond_create_slave_symlinks(bond_dev, slave_dev);
1621         if (res)
1622                 goto err_unset_master;
1623
1624         printk(KERN_INFO DRV_NAME
1625                ": %s: enslaving %s as a%s interface with a%s link.\n",
1626                bond_dev->name, slave_dev->name,
1627                new_slave->state == BOND_STATE_ACTIVE ? "n active" : " backup",
1628                new_slave->link != BOND_LINK_DOWN ? "n up" : " down");
1629
1630         /* enslave is successful */
1631         return 0;
1632
1633 /* Undo stages on error */
1634 err_unset_master:
1635         netdev_set_master(slave_dev, NULL);
1636
1637 err_close:
1638         dev_close(slave_dev);
1639
1640 err_restore_mac:
1641         if (!bond->params.fail_over_mac) {
1642                 memcpy(addr.sa_data, new_slave->perm_hwaddr, ETH_ALEN);
1643                 addr.sa_family = slave_dev->type;
1644                 dev_set_mac_address(slave_dev, &addr);
1645         }
1646
1647 err_free:
1648         kfree(new_slave);
1649
1650 err_undo_flags:
1651         bond_dev->features = old_features;
1652  
1653         return res;
1654 }
1655
1656 /*
1657  * Try to release the slave device <slave> from the bond device <master>
1658  * It is legal to access curr_active_slave without a lock because all the function
1659  * is write-locked.
1660  *
1661  * The rules for slave state should be:
1662  *   for Active/Backup:
1663  *     Active stays on all backups go down
1664  *   for Bonded connections:
1665  *     The first up interface should be left on and all others downed.
1666  */
1667 int bond_release(struct net_device *bond_dev, struct net_device *slave_dev)
1668 {
1669         struct bonding *bond = bond_dev->priv;
1670         struct slave *slave, *oldcurrent;
1671         struct sockaddr addr;
1672         int mac_addr_differ;
1673         DECLARE_MAC_BUF(mac);
1674
1675         /* slave is not a slave or master is not master of this slave */
1676         if (!(slave_dev->flags & IFF_SLAVE) ||
1677             (slave_dev->master != bond_dev)) {
1678                 printk(KERN_ERR DRV_NAME
1679                        ": %s: Error: cannot release %s.\n",
1680                        bond_dev->name, slave_dev->name);
1681                 return -EINVAL;
1682         }
1683
1684         write_lock_bh(&bond->lock);
1685
1686         slave = bond_get_slave_by_dev(bond, slave_dev);
1687         if (!slave) {
1688                 /* not a slave of this bond */
1689                 printk(KERN_INFO DRV_NAME
1690                        ": %s: %s not enslaved\n",
1691                        bond_dev->name, slave_dev->name);
1692                 write_unlock_bh(&bond->lock);
1693                 return -EINVAL;
1694         }
1695
1696         mac_addr_differ = memcmp(bond_dev->dev_addr,
1697                                  slave->perm_hwaddr,
1698                                  ETH_ALEN);
1699         if (!mac_addr_differ && (bond->slave_cnt > 1)) {
1700                 printk(KERN_WARNING DRV_NAME
1701                        ": %s: Warning: the permanent HWaddr of %s - "
1702                        "%s - is still in use by %s. "
1703                        "Set the HWaddr of %s to a different address "
1704                        "to avoid conflicts.\n",
1705                        bond_dev->name,
1706                        slave_dev->name,
1707                        print_mac(mac, slave->perm_hwaddr),
1708                        bond_dev->name,
1709                        slave_dev->name);
1710         }
1711
1712         /* Inform AD package of unbinding of slave. */
1713         if (bond->params.mode == BOND_MODE_8023AD) {
1714                 /* must be called before the slave is
1715                  * detached from the list
1716                  */
1717                 bond_3ad_unbind_slave(slave);
1718         }
1719
1720         printk(KERN_INFO DRV_NAME
1721                ": %s: releasing %s interface %s\n",
1722                bond_dev->name,
1723                (slave->state == BOND_STATE_ACTIVE)
1724                ? "active" : "backup",
1725                slave_dev->name);
1726
1727         oldcurrent = bond->curr_active_slave;
1728
1729         bond->current_arp_slave = NULL;
1730
1731         /* release the slave from its bond */
1732         bond_detach_slave(bond, slave);
1733
1734         bond_compute_features(bond);
1735
1736         if (bond->primary_slave == slave) {
1737                 bond->primary_slave = NULL;
1738         }
1739
1740         if (oldcurrent == slave) {
1741                 bond_change_active_slave(bond, NULL);
1742         }
1743
1744         if ((bond->params.mode == BOND_MODE_TLB) ||
1745             (bond->params.mode == BOND_MODE_ALB)) {
1746                 /* Must be called only after the slave has been
1747                  * detached from the list and the curr_active_slave
1748                  * has been cleared (if our_slave == old_current),
1749                  * but before a new active slave is selected.
1750                  */
1751                 write_unlock_bh(&bond->lock);
1752                 bond_alb_deinit_slave(bond, slave);
1753                 write_lock_bh(&bond->lock);
1754         }
1755
1756         if (oldcurrent == slave) {
1757                 /*
1758                  * Note that we hold RTNL over this sequence, so there
1759                  * is no concern that another slave add/remove event
1760                  * will interfere.
1761                  */
1762                 write_unlock_bh(&bond->lock);
1763                 read_lock(&bond->lock);
1764                 write_lock_bh(&bond->curr_slave_lock);
1765
1766                 bond_select_active_slave(bond);
1767
1768                 write_unlock_bh(&bond->curr_slave_lock);
1769                 read_unlock(&bond->lock);
1770                 write_lock_bh(&bond->lock);
1771         }
1772
1773         if (bond->slave_cnt == 0) {
1774                 bond_set_carrier(bond);
1775
1776                 /* if the last slave was removed, zero the mac address
1777                  * of the master so it will be set by the application
1778                  * to the mac address of the first slave
1779                  */
1780                 memset(bond_dev->dev_addr, 0, bond_dev->addr_len);
1781
1782                 if (list_empty(&bond->vlan_list)) {
1783                         bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
1784                 } else {
1785                         printk(KERN_WARNING DRV_NAME
1786                                ": %s: Warning: clearing HW address of %s while it "
1787                                "still has VLANs.\n",
1788                                bond_dev->name, bond_dev->name);
1789                         printk(KERN_WARNING DRV_NAME
1790                                ": %s: When re-adding slaves, make sure the bond's "
1791                                "HW address matches its VLANs'.\n",
1792                                bond_dev->name);
1793                 }
1794         } else if ((bond_dev->features & NETIF_F_VLAN_CHALLENGED) &&
1795                    !bond_has_challenged_slaves(bond)) {
1796                 printk(KERN_INFO DRV_NAME
1797                        ": %s: last VLAN challenged slave %s "
1798                        "left bond %s. VLAN blocking is removed\n",
1799                        bond_dev->name, slave_dev->name, bond_dev->name);
1800                 bond_dev->features &= ~NETIF_F_VLAN_CHALLENGED;
1801         }
1802
1803         write_unlock_bh(&bond->lock);
1804
1805         /* must do this from outside any spinlocks */
1806         bond_destroy_slave_symlinks(bond_dev, slave_dev);
1807
1808         bond_del_vlans_from_slave(bond, slave_dev);
1809
1810         /* If the mode USES_PRIMARY, then we should only remove its
1811          * promisc and mc settings if it was the curr_active_slave, but that was
1812          * already taken care of above when we detached the slave
1813          */
1814         if (!USES_PRIMARY(bond->params.mode)) {
1815                 /* unset promiscuity level from slave */
1816                 if (bond_dev->flags & IFF_PROMISC) {
1817                         dev_set_promiscuity(slave_dev, -1);
1818                 }
1819
1820                 /* unset allmulti level from slave */
1821                 if (bond_dev->flags & IFF_ALLMULTI) {
1822                         dev_set_allmulti(slave_dev, -1);
1823                 }
1824
1825                 /* flush master's mc_list from slave */
1826                 netif_tx_lock_bh(bond_dev);
1827                 bond_mc_list_flush(bond_dev, slave_dev);
1828                 netif_tx_unlock_bh(bond_dev);
1829         }
1830
1831         netdev_set_master(slave_dev, NULL);
1832
1833         /* close slave before restoring its mac address */
1834         dev_close(slave_dev);
1835
1836         if (!bond->params.fail_over_mac) {
1837                 /* restore original ("permanent") mac address */
1838                 memcpy(addr.sa_data, slave->perm_hwaddr, ETH_ALEN);
1839                 addr.sa_family = slave_dev->type;
1840                 dev_set_mac_address(slave_dev, &addr);
1841         }
1842
1843         slave_dev->priv_flags &= ~(IFF_MASTER_8023AD | IFF_MASTER_ALB |
1844                                    IFF_SLAVE_INACTIVE | IFF_BONDING |
1845                                    IFF_SLAVE_NEEDARP);
1846
1847         kfree(slave);
1848
1849         return 0;  /* deletion OK */
1850 }
1851
1852 /*
1853 * Destroy a bonding device.
1854 * Must be under rtnl_lock when this function is called.
1855 */
1856 void bond_destroy(struct bonding *bond)
1857 {
1858         bond_deinit(bond->dev);
1859         bond_destroy_sysfs_entry(bond);
1860         unregister_netdevice(bond->dev);
1861 }
1862
1863 /*
1864 * First release a slave and than destroy the bond if no more slaves iare left.
1865 * Must be under rtnl_lock when this function is called.
1866 */
1867 int  bond_release_and_destroy(struct net_device *bond_dev, struct net_device *slave_dev)
1868 {
1869         struct bonding *bond = bond_dev->priv;
1870         int ret;
1871
1872         ret = bond_release(bond_dev, slave_dev);
1873         if ((ret == 0) && (bond->slave_cnt == 0)) {
1874                 printk(KERN_INFO DRV_NAME ": %s: destroying bond %s.\n",
1875                        bond_dev->name, bond_dev->name);
1876                 bond_destroy(bond);
1877         }
1878         return ret;
1879 }
1880
1881 /*
1882  * This function releases all slaves.
1883  */
1884 static int bond_release_all(struct net_device *bond_dev)
1885 {
1886         struct bonding *bond = bond_dev->priv;
1887         struct slave *slave;
1888         struct net_device *slave_dev;
1889         struct sockaddr addr;
1890
1891         write_lock_bh(&bond->lock);
1892
1893         netif_carrier_off(bond_dev);
1894
1895         if (bond->slave_cnt == 0) {
1896                 goto out;
1897         }
1898
1899         bond->current_arp_slave = NULL;
1900         bond->primary_slave = NULL;
1901         bond_change_active_slave(bond, NULL);
1902
1903         while ((slave = bond->first_slave) != NULL) {
1904                 /* Inform AD package of unbinding of slave
1905                  * before slave is detached from the list.
1906                  */
1907                 if (bond->params.mode == BOND_MODE_8023AD) {
1908                         bond_3ad_unbind_slave(slave);
1909                 }
1910
1911                 slave_dev = slave->dev;
1912                 bond_detach_slave(bond, slave);
1913
1914                 /* now that the slave is detached, unlock and perform
1915                  * all the undo steps that should not be called from
1916                  * within a lock.
1917                  */
1918                 write_unlock_bh(&bond->lock);
1919
1920                 if ((bond->params.mode == BOND_MODE_TLB) ||
1921                     (bond->params.mode == BOND_MODE_ALB)) {
1922                         /* must be called only after the slave
1923                          * has been detached from the list
1924                          */
1925                         bond_alb_deinit_slave(bond, slave);
1926                 }
1927
1928                 bond_compute_features(bond);
1929
1930                 bond_destroy_slave_symlinks(bond_dev, slave_dev);
1931                 bond_del_vlans_from_slave(bond, slave_dev);
1932
1933                 /* If the mode USES_PRIMARY, then we should only remove its
1934                  * promisc and mc settings if it was the curr_active_slave, but that was
1935                  * already taken care of above when we detached the slave
1936                  */
1937                 if (!USES_PRIMARY(bond->params.mode)) {
1938                         /* unset promiscuity level from slave */
1939                         if (bond_dev->flags & IFF_PROMISC) {
1940                                 dev_set_promiscuity(slave_dev, -1);
1941                         }
1942
1943                         /* unset allmulti level from slave */
1944                         if (bond_dev->flags & IFF_ALLMULTI) {
1945                                 dev_set_allmulti(slave_dev, -1);
1946                         }
1947
1948                         /* flush master's mc_list from slave */
1949                         netif_tx_lock_bh(bond_dev);
1950                         bond_mc_list_flush(bond_dev, slave_dev);
1951                         netif_tx_unlock_bh(bond_dev);
1952                 }
1953
1954                 netdev_set_master(slave_dev, NULL);
1955
1956                 /* close slave before restoring its mac address */
1957                 dev_close(slave_dev);
1958
1959                 if (!bond->params.fail_over_mac) {
1960                         /* restore original ("permanent") mac address*/
1961                         memcpy(addr.sa_data, slave->perm_hwaddr, ETH_ALEN);
1962                         addr.sa_family = slave_dev->type;
1963                         dev_set_mac_address(slave_dev, &addr);
1964                 }
1965
1966                 slave_dev->priv_flags &= ~(IFF_MASTER_8023AD | IFF_MASTER_ALB |
1967                                            IFF_SLAVE_INACTIVE);
1968
1969                 kfree(slave);
1970
1971                 /* re-acquire the lock before getting the next slave */
1972                 write_lock_bh(&bond->lock);
1973         }
1974
1975         /* zero the mac address of the master so it will be
1976          * set by the application to the mac address of the
1977          * first slave
1978          */
1979         memset(bond_dev->dev_addr, 0, bond_dev->addr_len);
1980
1981         if (list_empty(&bond->vlan_list)) {
1982                 bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
1983         } else {
1984                 printk(KERN_WARNING DRV_NAME
1985                        ": %s: Warning: clearing HW address of %s while it "
1986                        "still has VLANs.\n",
1987                        bond_dev->name, bond_dev->name);
1988                 printk(KERN_WARNING DRV_NAME
1989                        ": %s: When re-adding slaves, make sure the bond's "
1990                        "HW address matches its VLANs'.\n",
1991                        bond_dev->name);
1992         }
1993
1994         printk(KERN_INFO DRV_NAME
1995                ": %s: released all slaves\n",
1996                bond_dev->name);
1997
1998 out:
1999         write_unlock_bh(&bond->lock);
2000
2001         return 0;
2002 }
2003
2004 /*
2005  * This function changes the active slave to slave <slave_dev>.
2006  * It returns -EINVAL in the following cases.
2007  *  - <slave_dev> is not found in the list.
2008  *  - There is not active slave now.
2009  *  - <slave_dev> is already active.
2010  *  - The link state of <slave_dev> is not BOND_LINK_UP.
2011  *  - <slave_dev> is not running.
2012  * In these cases, this fuction does nothing.
2013  * In the other cases, currnt_slave pointer is changed and 0 is returned.
2014  */
2015 static int bond_ioctl_change_active(struct net_device *bond_dev, struct net_device *slave_dev)
2016 {
2017         struct bonding *bond = bond_dev->priv;
2018         struct slave *old_active = NULL;
2019         struct slave *new_active = NULL;
2020         int res = 0;
2021
2022         if (!USES_PRIMARY(bond->params.mode)) {
2023                 return -EINVAL;
2024         }
2025
2026         /* Verify that master_dev is indeed the master of slave_dev */
2027         if (!(slave_dev->flags & IFF_SLAVE) ||
2028             (slave_dev->master != bond_dev)) {
2029                 return -EINVAL;
2030         }
2031
2032         read_lock(&bond->lock);
2033
2034         read_lock(&bond->curr_slave_lock);
2035         old_active = bond->curr_active_slave;
2036         read_unlock(&bond->curr_slave_lock);
2037
2038         new_active = bond_get_slave_by_dev(bond, slave_dev);
2039
2040         /*
2041          * Changing to the current active: do nothing; return success.
2042          */
2043         if (new_active && (new_active == old_active)) {
2044                 read_unlock(&bond->lock);
2045                 return 0;
2046         }
2047
2048         if ((new_active) &&
2049             (old_active) &&
2050             (new_active->link == BOND_LINK_UP) &&
2051             IS_UP(new_active->dev)) {
2052                 write_lock_bh(&bond->curr_slave_lock);
2053                 bond_change_active_slave(bond, new_active);
2054                 write_unlock_bh(&bond->curr_slave_lock);
2055         } else {
2056                 res = -EINVAL;
2057         }
2058
2059         read_unlock(&bond->lock);
2060
2061         return res;
2062 }
2063
2064 static int bond_info_query(struct net_device *bond_dev, struct ifbond *info)
2065 {
2066         struct bonding *bond = bond_dev->priv;
2067
2068         info->bond_mode = bond->params.mode;
2069         info->miimon = bond->params.miimon;
2070
2071         read_lock(&bond->lock);
2072         info->num_slaves = bond->slave_cnt;
2073         read_unlock(&bond->lock);
2074
2075         return 0;
2076 }
2077
2078 static int bond_slave_info_query(struct net_device *bond_dev, struct ifslave *info)
2079 {
2080         struct bonding *bond = bond_dev->priv;
2081         struct slave *slave;
2082         int i, found = 0;
2083
2084         if (info->slave_id < 0) {
2085                 return -ENODEV;
2086         }
2087
2088         read_lock(&bond->lock);
2089
2090         bond_for_each_slave(bond, slave, i) {
2091                 if (i == (int)info->slave_id) {
2092                         found = 1;
2093                         break;
2094                 }
2095         }
2096
2097         read_unlock(&bond->lock);
2098
2099         if (found) {
2100                 strcpy(info->slave_name, slave->dev->name);
2101                 info->link = slave->link;
2102                 info->state = slave->state;
2103                 info->link_failure_count = slave->link_failure_count;
2104         } else {
2105                 return -ENODEV;
2106         }
2107
2108         return 0;
2109 }
2110
2111 /*-------------------------------- Monitoring -------------------------------*/
2112
2113 /*
2114  * if !have_locks, return nonzero if a failover is necessary.  if
2115  * have_locks, do whatever failover activities are needed.
2116  *
2117  * This is to separate the inspection and failover steps for locking
2118  * purposes; failover requires rtnl, but acquiring it for every
2119  * inspection is undesirable, so a wrapper first does inspection, and
2120  * the acquires the necessary locks and calls again to perform
2121  * failover if needed.  Since all locks are dropped, a complete
2122  * restart is needed between calls.
2123  */
2124 static int __bond_mii_monitor(struct bonding *bond, int have_locks)
2125 {
2126         struct slave *slave, *oldcurrent;
2127         int do_failover = 0;
2128         int i;
2129
2130         if (bond->slave_cnt == 0)
2131                 goto out;
2132
2133         /* we will try to read the link status of each of our slaves, and
2134          * set their IFF_RUNNING flag appropriately. For each slave not
2135          * supporting MII status, we won't do anything so that a user-space
2136          * program could monitor the link itself if needed.
2137          */
2138
2139         if (bond->send_grat_arp) {
2140                 if (bond->curr_active_slave && test_bit(__LINK_STATE_LINKWATCH_PENDING,
2141                                 &bond->curr_active_slave->dev->state))
2142                         dprintk("Needs to send gratuitous arp but not yet\n");
2143                 else {
2144                         dprintk("sending delayed gratuitous arp on on %s\n",
2145                                 bond->curr_active_slave->dev->name);
2146                         bond_send_gratuitous_arp(bond);
2147                         bond->send_grat_arp = 0;
2148                 }
2149         }
2150         read_lock(&bond->curr_slave_lock);
2151         oldcurrent = bond->curr_active_slave;
2152         read_unlock(&bond->curr_slave_lock);
2153
2154         bond_for_each_slave(bond, slave, i) {
2155                 struct net_device *slave_dev = slave->dev;
2156                 int link_state;
2157                 u16 old_speed = slave->speed;
2158                 u8 old_duplex = slave->duplex;
2159
2160                 link_state = bond_check_dev_link(bond, slave_dev, 0);
2161
2162                 switch (slave->link) {
2163                 case BOND_LINK_UP:      /* the link was up */
2164                         if (link_state == BMSR_LSTATUS) {
2165                                 if (!oldcurrent) {
2166                                         if (!have_locks)
2167                                                 return 1;
2168                                         do_failover = 1;
2169                                 }
2170                                 break;
2171                         } else { /* link going down */
2172                                 slave->link  = BOND_LINK_FAIL;
2173                                 slave->delay = bond->params.downdelay;
2174
2175                                 if (slave->link_failure_count < UINT_MAX) {
2176                                         slave->link_failure_count++;
2177                                 }
2178
2179                                 if (bond->params.downdelay) {
2180                                         printk(KERN_INFO DRV_NAME
2181                                                ": %s: link status down for %s "
2182                                                "interface %s, disabling it in "
2183                                                "%d ms.\n",
2184                                                bond->dev->name,
2185                                                IS_UP(slave_dev)
2186                                                ? ((bond->params.mode == BOND_MODE_ACTIVEBACKUP)
2187                                                   ? ((slave == oldcurrent)
2188                                                      ? "active " : "backup ")
2189                                                   : "")
2190                                                : "idle ",
2191                                                slave_dev->name,
2192                                                bond->params.downdelay * bond->params.miimon);
2193                                 }
2194                         }
2195                         /* no break ! fall through the BOND_LINK_FAIL test to
2196                            ensure proper action to be taken
2197                         */
2198                 case BOND_LINK_FAIL:    /* the link has just gone down */
2199                         if (link_state != BMSR_LSTATUS) {
2200                                 /* link stays down */
2201                                 if (slave->delay <= 0) {
2202                                         if (!have_locks)
2203                                                 return 1;
2204
2205                                         /* link down for too long time */
2206                                         slave->link = BOND_LINK_DOWN;
2207
2208                                         /* in active/backup mode, we must
2209                                          * completely disable this interface
2210                                          */
2211                                         if ((bond->params.mode == BOND_MODE_ACTIVEBACKUP) ||
2212                                             (bond->params.mode == BOND_MODE_8023AD)) {
2213                                                 bond_set_slave_inactive_flags(slave);
2214                                         }
2215
2216                                         printk(KERN_INFO DRV_NAME
2217                                                ": %s: link status definitely "
2218                                                "down for interface %s, "
2219                                                "disabling it\n",
2220                                                bond->dev->name,
2221                                                slave_dev->name);
2222
2223                                         /* notify ad that the link status has changed */
2224                                         if (bond->params.mode == BOND_MODE_8023AD) {
2225                                                 bond_3ad_handle_link_change(slave, BOND_LINK_DOWN);
2226                                         }
2227
2228                                         if ((bond->params.mode == BOND_MODE_TLB) ||
2229                                             (bond->params.mode == BOND_MODE_ALB)) {
2230                                                 bond_alb_handle_link_change(bond, slave, BOND_LINK_DOWN);
2231                                         }
2232
2233                                         if (slave == oldcurrent) {
2234                                                 do_failover = 1;
2235                                         }
2236                                 } else {
2237                                         slave->delay--;
2238                                 }
2239                         } else {
2240                                 /* link up again */
2241                                 slave->link  = BOND_LINK_UP;
2242                                 slave->jiffies = jiffies;
2243                                 printk(KERN_INFO DRV_NAME
2244                                        ": %s: link status up again after %d "
2245                                        "ms for interface %s.\n",
2246                                        bond->dev->name,
2247                                        (bond->params.downdelay - slave->delay) * bond->params.miimon,
2248                                        slave_dev->name);
2249                         }
2250                         break;
2251                 case BOND_LINK_DOWN:    /* the link was down */
2252                         if (link_state != BMSR_LSTATUS) {
2253                                 /* the link stays down, nothing more to do */
2254                                 break;
2255                         } else {        /* link going up */
2256                                 slave->link  = BOND_LINK_BACK;
2257                                 slave->delay = bond->params.updelay;
2258
2259                                 if (bond->params.updelay) {
2260                                         /* if updelay == 0, no need to
2261                                            advertise about a 0 ms delay */
2262                                         printk(KERN_INFO DRV_NAME
2263                                                ": %s: link status up for "
2264                                                "interface %s, enabling it "
2265                                                "in %d ms.\n",
2266                                                bond->dev->name,
2267                                                slave_dev->name,
2268                                                bond->params.updelay * bond->params.miimon);
2269                                 }
2270                         }
2271                         /* no break ! fall through the BOND_LINK_BACK state in
2272                            case there's something to do.
2273                         */
2274                 case BOND_LINK_BACK:    /* the link has just come back */
2275                         if (link_state != BMSR_LSTATUS) {
2276                                 /* link down again */
2277                                 slave->link  = BOND_LINK_DOWN;
2278
2279                                 printk(KERN_INFO DRV_NAME
2280                                        ": %s: link status down again after %d "
2281                                        "ms for interface %s.\n",
2282                                        bond->dev->name,
2283                                        (bond->params.updelay - slave->delay) * bond->params.miimon,
2284                                        slave_dev->name);
2285                         } else {
2286                                 /* link stays up */
2287                                 if (slave->delay == 0) {
2288                                         if (!have_locks)
2289                                                 return 1;
2290
2291                                         /* now the link has been up for long time enough */
2292                                         slave->link = BOND_LINK_UP;
2293                                         slave->jiffies = jiffies;
2294
2295                                         if (bond->params.mode == BOND_MODE_8023AD) {
2296                                                 /* prevent it from being the active one */
2297                                                 slave->state = BOND_STATE_BACKUP;
2298                                         } else if (bond->params.mode != BOND_MODE_ACTIVEBACKUP) {
2299                                                 /* make it immediately active */
2300                                                 slave->state = BOND_STATE_ACTIVE;
2301                                         } else if (slave != bond->primary_slave) {
2302                                                 /* prevent it from being the active one */
2303                                                 slave->state = BOND_STATE_BACKUP;
2304                                         }
2305
2306                                         printk(KERN_INFO DRV_NAME
2307                                                ": %s: link status definitely "
2308                                                "up for interface %s.\n",
2309                                                bond->dev->name,
2310                                                slave_dev->name);
2311
2312                                         /* notify ad that the link status has changed */
2313                                         if (bond->params.mode == BOND_MODE_8023AD) {
2314                                                 bond_3ad_handle_link_change(slave, BOND_LINK_UP);
2315                                         }
2316
2317                                         if ((bond->params.mode == BOND_MODE_TLB) ||
2318                                             (bond->params.mode == BOND_MODE_ALB)) {
2319                                                 bond_alb_handle_link_change(bond, slave, BOND_LINK_UP);
2320                                         }
2321
2322                                         if ((!oldcurrent) ||
2323                                             (slave == bond->primary_slave)) {
2324                                                 do_failover = 1;
2325                                         }
2326                                 } else {
2327                                         slave->delay--;
2328                                 }
2329                         }
2330                         break;
2331                 default:
2332                         /* Should not happen */
2333                         printk(KERN_ERR DRV_NAME
2334                                ": %s: Error: %s Illegal value (link=%d)\n",
2335                                bond->dev->name,
2336                                slave->dev->name,
2337                                slave->link);
2338                         goto out;
2339                 } /* end of switch (slave->link) */
2340
2341                 bond_update_speed_duplex(slave);
2342
2343                 if (bond->params.mode == BOND_MODE_8023AD) {
2344                         if (old_speed != slave->speed) {
2345                                 bond_3ad_adapter_speed_changed(slave);
2346                         }
2347
2348                         if (old_duplex != slave->duplex) {
2349                                 bond_3ad_adapter_duplex_changed(slave);
2350                         }
2351                 }
2352
2353         } /* end of for */
2354
2355         if (do_failover) {
2356                 ASSERT_RTNL();
2357
2358                 write_lock_bh(&bond->curr_slave_lock);
2359
2360                 bond_select_active_slave(bond);
2361
2362                 write_unlock_bh(&bond->curr_slave_lock);
2363
2364         } else
2365                 bond_set_carrier(bond);
2366
2367 out:
2368         return 0;
2369 }
2370
2371 /*
2372  * bond_mii_monitor
2373  *
2374  * Really a wrapper that splits the mii monitor into two phases: an
2375  * inspection, then (if inspection indicates something needs to be
2376  * done) an acquisition of appropriate locks followed by another pass
2377  * to implement whatever link state changes are indicated.
2378  */
2379 void bond_mii_monitor(struct work_struct *work)
2380 {
2381         struct bonding *bond = container_of(work, struct bonding,
2382                                             mii_work.work);
2383         unsigned long delay;
2384
2385         read_lock(&bond->lock);
2386         if (bond->kill_timers) {
2387                 read_unlock(&bond->lock);
2388                 return;
2389         }
2390         if (__bond_mii_monitor(bond, 0)) {
2391                 read_unlock(&bond->lock);
2392                 rtnl_lock();
2393                 read_lock(&bond->lock);
2394                 __bond_mii_monitor(bond, 1);
2395                 read_unlock(&bond->lock);
2396                 rtnl_unlock();  /* might sleep, hold no other locks */
2397                 read_lock(&bond->lock);
2398         }
2399
2400         delay = ((bond->params.miimon * HZ) / 1000) ? : 1;
2401         read_unlock(&bond->lock);
2402         queue_delayed_work(bond->wq, &bond->mii_work, delay);
2403 }
2404
2405 static __be32 bond_glean_dev_ip(struct net_device *dev)
2406 {
2407         struct in_device *idev;
2408         struct in_ifaddr *ifa;
2409         __be32 addr = 0;
2410
2411         if (!dev)
2412                 return 0;
2413
2414         rcu_read_lock();
2415         idev = __in_dev_get_rcu(dev);
2416         if (!idev)
2417                 goto out;
2418
2419         ifa = idev->ifa_list;
2420         if (!ifa)
2421                 goto out;
2422
2423         addr = ifa->ifa_local;
2424 out:
2425         rcu_read_unlock();
2426         return addr;
2427 }
2428
2429 static int bond_has_ip(struct bonding *bond)
2430 {
2431         struct vlan_entry *vlan, *vlan_next;
2432
2433         if (bond->master_ip)
2434                 return 1;
2435
2436         if (list_empty(&bond->vlan_list))
2437                 return 0;
2438
2439         list_for_each_entry_safe(vlan, vlan_next, &bond->vlan_list,
2440                                  vlan_list) {
2441                 if (vlan->vlan_ip)
2442                         return 1;
2443         }
2444
2445         return 0;
2446 }
2447
2448 static int bond_has_this_ip(struct bonding *bond, __be32 ip)
2449 {
2450         struct vlan_entry *vlan, *vlan_next;
2451
2452         if (ip == bond->master_ip)
2453                 return 1;
2454
2455         if (list_empty(&bond->vlan_list))
2456                 return 0;
2457
2458         list_for_each_entry_safe(vlan, vlan_next, &bond->vlan_list,
2459                                  vlan_list) {
2460                 if (ip == vlan->vlan_ip)
2461                         return 1;
2462         }
2463
2464         return 0;
2465 }
2466
2467 /*
2468  * We go to the (large) trouble of VLAN tagging ARP frames because
2469  * switches in VLAN mode (especially if ports are configured as
2470  * "native" to a VLAN) might not pass non-tagged frames.
2471  */
2472 static void bond_arp_send(struct net_device *slave_dev, int arp_op, __be32 dest_ip, __be32 src_ip, unsigned short vlan_id)
2473 {
2474         struct sk_buff *skb;
2475
2476         dprintk("arp %d on slave %s: dst %x src %x vid %d\n", arp_op,
2477                slave_dev->name, dest_ip, src_ip, vlan_id);
2478                
2479         skb = arp_create(arp_op, ETH_P_ARP, dest_ip, slave_dev, src_ip,
2480                          NULL, slave_dev->dev_addr, NULL);
2481
2482         if (!skb) {
2483                 printk(KERN_ERR DRV_NAME ": ARP packet allocation failed\n");
2484                 return;
2485         }
2486         if (vlan_id) {
2487                 skb = vlan_put_tag(skb, vlan_id);
2488                 if (!skb) {
2489                         printk(KERN_ERR DRV_NAME ": failed to insert VLAN tag\n");
2490                         return;
2491                 }
2492         }
2493         arp_xmit(skb);
2494 }
2495
2496
2497 static void bond_arp_send_all(struct bonding *bond, struct slave *slave)
2498 {
2499         int i, vlan_id, rv;
2500         __be32 *targets = bond->params.arp_targets;
2501         struct vlan_entry *vlan, *vlan_next;
2502         struct net_device *vlan_dev;
2503         struct flowi fl;
2504         struct rtable *rt;
2505
2506         for (i = 0; (i < BOND_MAX_ARP_TARGETS); i++) {
2507                 if (!targets[i])
2508                         continue;
2509                 dprintk("basa: target %x\n", targets[i]);
2510                 if (list_empty(&bond->vlan_list)) {
2511                         dprintk("basa: empty vlan: arp_send\n");
2512                         bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2513                                       bond->master_ip, 0);
2514                         continue;
2515                 }
2516
2517                 /*
2518                  * If VLANs are configured, we do a route lookup to
2519                  * determine which VLAN interface would be used, so we
2520                  * can tag the ARP with the proper VLAN tag.
2521                  */
2522                 memset(&fl, 0, sizeof(fl));
2523                 fl.fl4_dst = targets[i];
2524                 fl.fl4_tos = RTO_ONLINK;
2525
2526                 rv = ip_route_output_key(&init_net, &rt, &fl);
2527                 if (rv) {
2528                         if (net_ratelimit()) {
2529                                 printk(KERN_WARNING DRV_NAME
2530                              ": %s: no route to arp_ip_target %u.%u.%u.%u\n",
2531                                        bond->dev->name, NIPQUAD(fl.fl4_dst));
2532                         }
2533                         continue;
2534                 }
2535
2536                 /*
2537                  * This target is not on a VLAN
2538                  */
2539                 if (rt->u.dst.dev == bond->dev) {
2540                         ip_rt_put(rt);
2541                         dprintk("basa: rtdev == bond->dev: arp_send\n");
2542                         bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2543                                       bond->master_ip, 0);
2544                         continue;
2545                 }
2546
2547                 vlan_id = 0;
2548                 list_for_each_entry_safe(vlan, vlan_next, &bond->vlan_list,
2549                                          vlan_list) {
2550                         vlan_dev = vlan_group_get_device(bond->vlgrp, vlan->vlan_id);
2551                         if (vlan_dev == rt->u.dst.dev) {
2552                                 vlan_id = vlan->vlan_id;
2553                                 dprintk("basa: vlan match on %s %d\n",
2554                                        vlan_dev->name, vlan_id);
2555                                 break;
2556                         }
2557                 }
2558
2559                 if (vlan_id) {
2560                         ip_rt_put(rt);
2561                         bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2562                                       vlan->vlan_ip, vlan_id);
2563                         continue;
2564                 }
2565
2566                 if (net_ratelimit()) {
2567                         printk(KERN_WARNING DRV_NAME
2568                ": %s: no path to arp_ip_target %u.%u.%u.%u via rt.dev %s\n",
2569                                bond->dev->name, NIPQUAD(fl.fl4_dst),
2570                                rt->u.dst.dev ? rt->u.dst.dev->name : "NULL");
2571                 }
2572                 ip_rt_put(rt);
2573         }
2574 }
2575
2576 /*
2577  * Kick out a gratuitous ARP for an IP on the bonding master plus one
2578  * for each VLAN above us.
2579  */
2580 static void bond_send_gratuitous_arp(struct bonding *bond)
2581 {
2582         struct slave *slave = bond->curr_active_slave;
2583         struct vlan_entry *vlan;
2584         struct net_device *vlan_dev;
2585
2586         dprintk("bond_send_grat_arp: bond %s slave %s\n", bond->dev->name,
2587                                 slave ? slave->dev->name : "NULL");
2588         if (!slave)
2589                 return;
2590
2591         if (bond->master_ip) {
2592                 bond_arp_send(slave->dev, ARPOP_REPLY, bond->master_ip,
2593                                 bond->master_ip, 0);
2594         }
2595
2596         list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
2597                 vlan_dev = vlan_group_get_device(bond->vlgrp, vlan->vlan_id);
2598                 if (vlan->vlan_ip) {
2599                         bond_arp_send(slave->dev, ARPOP_REPLY, vlan->vlan_ip,
2600                                       vlan->vlan_ip, vlan->vlan_id);
2601                 }
2602         }
2603 }
2604
2605 static void bond_validate_arp(struct bonding *bond, struct slave *slave, __be32 sip, __be32 tip)
2606 {
2607         int i;
2608         __be32 *targets = bond->params.arp_targets;
2609
2610         targets = bond->params.arp_targets;
2611         for (i = 0; (i < BOND_MAX_ARP_TARGETS) && targets[i]; i++) {
2612                 dprintk("bva: sip %u.%u.%u.%u tip %u.%u.%u.%u t[%d] "
2613                         "%u.%u.%u.%u bhti(tip) %d\n",
2614                        NIPQUAD(sip), NIPQUAD(tip), i, NIPQUAD(targets[i]),
2615                        bond_has_this_ip(bond, tip));
2616                 if (sip == targets[i]) {
2617                         if (bond_has_this_ip(bond, tip))
2618                                 slave->last_arp_rx = jiffies;
2619                         return;
2620                 }
2621         }
2622 }
2623
2624 static int bond_arp_rcv(struct sk_buff *skb, struct net_device *dev, struct packet_type *pt, struct net_device *orig_dev)
2625 {
2626         struct arphdr *arp;
2627         struct slave *slave;
2628         struct bonding *bond;
2629         unsigned char *arp_ptr;
2630         __be32 sip, tip;
2631
2632         if (dev_net(dev) != &init_net)
2633                 goto out;
2634
2635         if (!(dev->priv_flags & IFF_BONDING) || !(dev->flags & IFF_MASTER))
2636                 goto out;
2637
2638         bond = dev->priv;
2639         read_lock(&bond->lock);
2640
2641         dprintk("bond_arp_rcv: bond %s skb->dev %s orig_dev %s\n",
2642                 bond->dev->name, skb->dev ? skb->dev->name : "NULL",
2643                 orig_dev ? orig_dev->name : "NULL");
2644
2645         slave = bond_get_slave_by_dev(bond, orig_dev);
2646         if (!slave || !slave_do_arp_validate(bond, slave))
2647                 goto out_unlock;
2648
2649         if (!pskb_may_pull(skb, arp_hdr_len(dev)))
2650                 goto out_unlock;
2651
2652         arp = arp_hdr(skb);
2653         if (arp->ar_hln != dev->addr_len ||
2654             skb->pkt_type == PACKET_OTHERHOST ||
2655             skb->pkt_type == PACKET_LOOPBACK ||
2656             arp->ar_hrd != htons(ARPHRD_ETHER) ||
2657             arp->ar_pro != htons(ETH_P_IP) ||
2658             arp->ar_pln != 4)
2659                 goto out_unlock;
2660
2661         arp_ptr = (unsigned char *)(arp + 1);
2662         arp_ptr += dev->addr_len;
2663         memcpy(&sip, arp_ptr, 4);
2664         arp_ptr += 4 + dev->addr_len;
2665         memcpy(&tip, arp_ptr, 4);
2666
2667         dprintk("bond_arp_rcv: %s %s/%d av %d sv %d sip %u.%u.%u.%u"
2668                 " tip %u.%u.%u.%u\n", bond->dev->name, slave->dev->name,
2669                 slave->state, bond->params.arp_validate,
2670                 slave_do_arp_validate(bond, slave), NIPQUAD(sip), NIPQUAD(tip));
2671
2672         /*
2673          * Backup slaves won't see the ARP reply, but do come through
2674          * here for each ARP probe (so we swap the sip/tip to validate
2675          * the probe).  In a "redundant switch, common router" type of
2676          * configuration, the ARP probe will (hopefully) travel from
2677          * the active, through one switch, the router, then the other
2678          * switch before reaching the backup.
2679          */
2680         if (slave->state == BOND_STATE_ACTIVE)
2681                 bond_validate_arp(bond, slave, sip, tip);
2682         else
2683                 bond_validate_arp(bond, slave, tip, sip);
2684
2685 out_unlock:
2686         read_unlock(&bond->lock);
2687 out:
2688         dev_kfree_skb(skb);
2689         return NET_RX_SUCCESS;
2690 }
2691
2692 /*
2693  * this function is called regularly to monitor each slave's link
2694  * ensuring that traffic is being sent and received when arp monitoring
2695  * is used in load-balancing mode. if the adapter has been dormant, then an
2696  * arp is transmitted to generate traffic. see activebackup_arp_monitor for
2697  * arp monitoring in active backup mode.
2698  */
2699 void bond_loadbalance_arp_mon(struct work_struct *work)
2700 {
2701         struct bonding *bond = container_of(work, struct bonding,
2702                                             arp_work.work);
2703         struct slave *slave, *oldcurrent;
2704         int do_failover = 0;
2705         int delta_in_ticks;
2706         int i;
2707
2708         read_lock(&bond->lock);
2709
2710         delta_in_ticks = (bond->params.arp_interval * HZ) / 1000;
2711
2712         if (bond->kill_timers) {
2713                 goto out;
2714         }
2715
2716         if (bond->slave_cnt == 0) {
2717                 goto re_arm;
2718         }
2719
2720         read_lock(&bond->curr_slave_lock);
2721         oldcurrent = bond->curr_active_slave;
2722         read_unlock(&bond->curr_slave_lock);
2723
2724         /* see if any of the previous devices are up now (i.e. they have
2725          * xmt and rcv traffic). the curr_active_slave does not come into
2726          * the picture unless it is null. also, slave->jiffies is not needed
2727          * here because we send an arp on each slave and give a slave as
2728          * long as it needs to get the tx/rx within the delta.
2729          * TODO: what about up/down delay in arp mode? it wasn't here before
2730          *       so it can wait
2731          */
2732         bond_for_each_slave(bond, slave, i) {
2733                 if (slave->link != BOND_LINK_UP) {
2734                         if (time_before_eq(jiffies, slave->dev->trans_start + delta_in_ticks) &&
2735                             time_before_eq(jiffies, slave->dev->last_rx + delta_in_ticks)) {
2736
2737                                 slave->link  = BOND_LINK_UP;
2738                                 slave->state = BOND_STATE_ACTIVE;
2739
2740                                 /* primary_slave has no meaning in round-robin
2741                                  * mode. the window of a slave being up and
2742                                  * curr_active_slave being null after enslaving
2743                                  * is closed.
2744                                  */
2745                                 if (!oldcurrent) {
2746                                         printk(KERN_INFO DRV_NAME
2747                                                ": %s: link status definitely "
2748                                                "up for interface %s, ",
2749                                                bond->dev->name,
2750                                                slave->dev->name);
2751                                         do_failover = 1;
2752                                 } else {
2753                                         printk(KERN_INFO DRV_NAME
2754                                                ": %s: interface %s is now up\n",
2755                                                bond->dev->name,
2756                                                slave->dev->name);
2757                                 }
2758                         }
2759                 } else {
2760                         /* slave->link == BOND_LINK_UP */
2761
2762                         /* not all switches will respond to an arp request
2763                          * when the source ip is 0, so don't take the link down
2764                          * if we don't know our ip yet
2765                          */
2766                         if (time_after_eq(jiffies, slave->dev->trans_start + 2*delta_in_ticks) ||
2767                             (time_after_eq(jiffies, slave->dev->last_rx + 2*delta_in_ticks) &&
2768                              bond_has_ip(bond))) {
2769
2770                                 slave->link  = BOND_LINK_DOWN;
2771                                 slave->state = BOND_STATE_BACKUP;
2772
2773                                 if (slave->link_failure_count < UINT_MAX) {
2774                                         slave->link_failure_count++;
2775                                 }
2776
2777                                 printk(KERN_INFO DRV_NAME
2778                                        ": %s: interface %s is now down.\n",
2779                                        bond->dev->name,
2780                                        slave->dev->name);
2781
2782                                 if (slave == oldcurrent) {
2783                                         do_failover = 1;
2784                                 }
2785                         }
2786                 }
2787
2788                 /* note: if switch is in round-robin mode, all links
2789                  * must tx arp to ensure all links rx an arp - otherwise
2790                  * links may oscillate or not come up at all; if switch is
2791                  * in something like xor mode, there is nothing we can
2792                  * do - all replies will be rx'ed on same link causing slaves
2793                  * to be unstable during low/no traffic periods
2794                  */
2795                 if (IS_UP(slave->dev)) {
2796                         bond_arp_send_all(bond, slave);
2797                 }
2798         }
2799
2800         if (do_failover) {
2801                 write_lock_bh(&bond->curr_slave_lock);
2802
2803                 bond_select_active_slave(bond);
2804
2805                 write_unlock_bh(&bond->curr_slave_lock);
2806         }
2807
2808 re_arm:
2809         if (bond->params.arp_interval)
2810                 queue_delayed_work(bond->wq, &bond->arp_work, delta_in_ticks);
2811 out:
2812         read_unlock(&bond->lock);
2813 }
2814
2815 /*
2816  * When using arp monitoring in active-backup mode, this function is
2817  * called to determine if any backup slaves have went down or a new
2818  * current slave needs to be found.
2819  * The backup slaves never generate traffic, they are considered up by merely
2820  * receiving traffic. If the current slave goes down, each backup slave will
2821  * be given the opportunity to tx/rx an arp before being taken down - this
2822  * prevents all slaves from being taken down due to the current slave not
2823  * sending any traffic for the backups to receive. The arps are not necessarily
2824  * necessary, any tx and rx traffic will keep the current slave up. While any
2825  * rx traffic will keep the backup slaves up, the current slave is responsible
2826  * for generating traffic to keep them up regardless of any other traffic they
2827  * may have received.
2828  * see loadbalance_arp_monitor for arp monitoring in load balancing mode
2829  */
2830 void bond_activebackup_arp_mon(struct work_struct *work)
2831 {
2832         struct bonding *bond = container_of(work, struct bonding,
2833                                             arp_work.work);
2834         struct slave *slave;
2835         int delta_in_ticks;
2836         int i;
2837
2838         read_lock(&bond->lock);
2839
2840         delta_in_ticks = (bond->params.arp_interval * HZ) / 1000;
2841
2842         if (bond->kill_timers) {
2843                 goto out;
2844         }
2845
2846         if (bond->slave_cnt == 0) {
2847                 goto re_arm;
2848         }
2849
2850         /* determine if any slave has come up or any backup slave has
2851          * gone down
2852          * TODO: what about up/down delay in arp mode? it wasn't here before
2853          *       so it can wait
2854          */
2855         bond_for_each_slave(bond, slave, i) {
2856                 if (slave->link != BOND_LINK_UP) {
2857                         if (time_before_eq(jiffies,
2858                             slave_last_rx(bond, slave) + delta_in_ticks)) {
2859
2860                                 slave->link = BOND_LINK_UP;
2861
2862                                 write_lock_bh(&bond->curr_slave_lock);
2863
2864                                 if ((!bond->curr_active_slave) &&
2865                                     time_before_eq(jiffies, slave->dev->trans_start + delta_in_ticks)) {
2866                                         bond_change_active_slave(bond, slave);
2867                                         bond->current_arp_slave = NULL;
2868                                 } else if (bond->curr_active_slave != slave) {
2869                                         /* this slave has just come up but we
2870                                          * already have a current slave; this
2871                                          * can also happen if bond_enslave adds
2872                                          * a new slave that is up while we are
2873                                          * searching for a new slave
2874                                          */
2875                                         bond_set_slave_inactive_flags(slave);
2876                                         bond->current_arp_slave = NULL;
2877                                 }
2878
2879                                 bond_set_carrier(bond);
2880
2881                                 if (slave == bond->curr_active_slave) {
2882                                         printk(KERN_INFO DRV_NAME
2883                                                ": %s: %s is up and now the "
2884                                                "active interface\n",
2885                                                bond->dev->name,
2886                                                slave->dev->name);
2887                                         netif_carrier_on(bond->dev);
2888                                 } else {
2889                                         printk(KERN_INFO DRV_NAME
2890                                                ": %s: backup interface %s is "
2891                                                "now up\n",
2892                                                bond->dev->name,
2893                                                slave->dev->name);
2894                                 }
2895
2896                                 write_unlock_bh(&bond->curr_slave_lock);
2897                         }
2898                 } else {
2899                         read_lock(&bond->curr_slave_lock);
2900
2901                         if ((slave != bond->curr_active_slave) &&
2902                             (!bond->current_arp_slave) &&
2903                             (time_after_eq(jiffies, slave_last_rx(bond, slave) + 3*delta_in_ticks) &&
2904                              bond_has_ip(bond))) {
2905                                 /* a backup slave has gone down; three times
2906                                  * the delta allows the current slave to be
2907                                  * taken out before the backup slave.
2908                                  * note: a non-null current_arp_slave indicates
2909                                  * the curr_active_slave went down and we are
2910                                  * searching for a new one; under this
2911                                  * condition we only take the curr_active_slave
2912                                  * down - this gives each slave a chance to
2913                                  * tx/rx traffic before being taken out
2914                                  */
2915
2916                                 read_unlock(&bond->curr_slave_lock);
2917
2918                                 slave->link  = BOND_LINK_DOWN;
2919
2920                                 if (slave->link_failure_count < UINT_MAX) {
2921                                         slave->link_failure_count++;
2922                                 }
2923
2924                                 bond_set_slave_inactive_flags(slave);
2925
2926                                 printk(KERN_INFO DRV_NAME
2927                                        ": %s: backup interface %s is now down\n",
2928                                        bond->dev->name,
2929                                        slave->dev->name);
2930                         } else {
2931                                 read_unlock(&bond->curr_slave_lock);
2932                         }
2933                 }
2934         }
2935
2936         read_lock(&bond->curr_slave_lock);
2937         slave = bond->curr_active_slave;
2938         read_unlock(&bond->curr_slave_lock);
2939
2940         if (slave) {
2941                 /* if we have sent traffic in the past 2*arp_intervals but
2942                  * haven't xmit and rx traffic in that time interval, select
2943                  * a different slave. slave->jiffies is only updated when
2944                  * a slave first becomes the curr_active_slave - not necessarily
2945                  * after every arp; this ensures the slave has a full 2*delta
2946                  * before being taken out. if a primary is being used, check
2947                  * if it is up and needs to take over as the curr_active_slave
2948                  */
2949                 if ((time_after_eq(jiffies, slave->dev->trans_start + 2*delta_in_ticks) ||
2950                         (time_after_eq(jiffies, slave_last_rx(bond, slave) + 2*delta_in_ticks) &&
2951                          bond_has_ip(bond))) &&
2952                         time_after_eq(jiffies, slave->jiffies + 2*delta_in_ticks)) {
2953
2954                         slave->link  = BOND_LINK_DOWN;
2955
2956                         if (slave->link_failure_count < UINT_MAX) {
2957                                 slave->link_failure_count++;
2958                         }
2959
2960                         printk(KERN_INFO DRV_NAME
2961                                ": %s: link status down for active interface "
2962                                "%s, disabling it\n",
2963                                bond->dev->name,
2964                                slave->dev->name);
2965
2966                         write_lock_bh(&bond->curr_slave_lock);
2967
2968                         bond_select_active_slave(bond);
2969                         slave = bond->curr_active_slave;
2970
2971                         write_unlock_bh(&bond->curr_slave_lock);
2972
2973                         bond->current_arp_slave = slave;
2974
2975                         if (slave) {
2976                                 slave->jiffies = jiffies;
2977                         }
2978                 } else if ((bond->primary_slave) &&
2979                            (bond->primary_slave != slave) &&
2980                            (bond->primary_slave->link == BOND_LINK_UP)) {
2981                         /* at this point, slave is the curr_active_slave */
2982                         printk(KERN_INFO DRV_NAME
2983                                ": %s: changing from interface %s to primary "
2984                                "interface %s\n",
2985                                bond->dev->name,
2986                                slave->dev->name,
2987                                bond->primary_slave->dev->name);
2988
2989                         /* primary is up so switch to it */
2990                         write_lock_bh(&bond->curr_slave_lock);
2991                         bond_change_active_slave(bond, bond->primary_slave);
2992                         write_unlock_bh(&bond->curr_slave_lock);
2993
2994                         slave = bond->primary_slave;
2995                         slave->jiffies = jiffies;
2996                 } else {
2997                         bond->current_arp_slave = NULL;
2998                 }
2999
3000                 /* the current slave must tx an arp to ensure backup slaves
3001                  * rx traffic
3002                  */
3003                 if (slave && bond_has_ip(bond)) {
3004                         bond_arp_send_all(bond, slave);
3005                 }
3006         }
3007
3008         /* if we don't have a curr_active_slave, search for the next available
3009          * backup slave from the current_arp_slave and make it the candidate
3010          * for becoming the curr_active_slave
3011          */
3012         if (!slave) {
3013                 if (!bond->current_arp_slave) {
3014                         bond->current_arp_slave = bond->first_slave;
3015                 }
3016
3017                 if (bond->current_arp_slave) {
3018                         bond_set_slave_inactive_flags(bond->current_arp_slave);
3019
3020                         /* search for next candidate */
3021                         bond_for_each_slave_from(bond, slave, i, bond->current_arp_slave->next) {
3022                                 if (IS_UP(slave->dev)) {
3023                                         slave->link = BOND_LINK_BACK;
3024                                         bond_set_slave_active_flags(slave);
3025                                         bond_arp_send_all(bond, slave);
3026                                         slave->jiffies = jiffies;
3027                                         bond->current_arp_slave = slave;
3028                                         break;
3029                                 }
3030
3031                                 /* if the link state is up at this point, we
3032                                  * mark it down - this can happen if we have
3033                                  * simultaneous link failures and
3034                                  * reselect_active_interface doesn't make this
3035                                  * one the current slave so it is still marked
3036                                  * up when it is actually down
3037                                  */
3038                                 if (slave->link == BOND_LINK_UP) {
3039                                         slave->link  = BOND_LINK_DOWN;
3040                                         if (slave->link_failure_count < UINT_MAX) {
3041                                                 slave->link_failure_count++;
3042                                         }
3043
3044                                         bond_set_slave_inactive_flags(slave);
3045
3046                                         printk(KERN_INFO DRV_NAME
3047                                                ": %s: backup interface %s is "
3048                                                "now down.\n",
3049                                                bond->dev->name,
3050                                                slave->dev->name);
3051                                 }
3052                         }
3053                 }
3054         }
3055
3056 re_arm:
3057         if (bond->params.arp_interval) {
3058                 queue_delayed_work(bond->wq, &bond->arp_work, delta_in_ticks);
3059         }
3060 out:
3061         read_unlock(&bond->lock);
3062 }
3063
3064 /*------------------------------ proc/seq_file-------------------------------*/
3065
3066 #ifdef CONFIG_PROC_FS
3067
3068 static void *bond_info_seq_start(struct seq_file *seq, loff_t *pos)
3069 {
3070         struct bonding *bond = seq->private;
3071         loff_t off = 0;
3072         struct slave *slave;
3073         int i;
3074
3075         /* make sure the bond won't be taken away */
3076         read_lock(&dev_base_lock);
3077         read_lock(&bond->lock);
3078
3079         if (*pos == 0) {
3080                 return SEQ_START_TOKEN;
3081         }
3082
3083         bond_for_each_slave(bond, slave, i) {
3084                 if (++off == *pos) {
3085                         return slave;
3086                 }
3087         }
3088
3089         return NULL;
3090 }
3091
3092 static void *bond_info_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3093 {
3094         struct bonding *bond = seq->private;
3095         struct slave *slave = v;
3096
3097         ++*pos;
3098         if (v == SEQ_START_TOKEN) {
3099                 return bond->first_slave;
3100         }
3101
3102         slave = slave->next;
3103
3104         return (slave == bond->first_slave) ? NULL : slave;
3105 }
3106
3107 static void bond_info_seq_stop(struct seq_file *seq, void *v)
3108 {
3109         struct bonding *bond = seq->private;
3110
3111         read_unlock(&bond->lock);
3112         read_unlock(&dev_base_lock);
3113 }
3114
3115 static void bond_info_show_master(struct seq_file *seq)
3116 {
3117         struct bonding *bond = seq->private;
3118         struct slave *curr;
3119         int i;
3120         u32 target;
3121
3122         read_lock(&bond->curr_slave_lock);
3123         curr = bond->curr_active_slave;
3124         read_unlock(&bond->curr_slave_lock);
3125
3126         seq_printf(seq, "Bonding Mode: %s",
3127                    bond_mode_name(bond->params.mode));
3128
3129         if (bond->params.mode == BOND_MODE_ACTIVEBACKUP &&
3130             bond->params.fail_over_mac)
3131                 seq_printf(seq, " (fail_over_mac)");
3132
3133         seq_printf(seq, "\n");
3134
3135         if (bond->params.mode == BOND_MODE_XOR ||
3136                 bond->params.mode == BOND_MODE_8023AD) {
3137                 seq_printf(seq, "Transmit Hash Policy: %s (%d)\n",
3138                         xmit_hashtype_tbl[bond->params.xmit_policy].modename,
3139                         bond->params.xmit_policy);
3140         }
3141
3142         if (USES_PRIMARY(bond->params.mode)) {
3143                 seq_printf(seq, "Primary Slave: %s\n",
3144                            (bond->primary_slave) ?
3145                            bond->primary_slave->dev->name : "None");
3146
3147                 seq_printf(seq, "Currently Active Slave: %s\n",
3148                            (curr) ? curr->dev->name : "None");
3149         }
3150
3151         seq_printf(seq, "MII Status: %s\n", netif_carrier_ok(bond->dev) ?
3152                    "up" : "down");
3153         seq_printf(seq, "MII Polling Interval (ms): %d\n", bond->params.miimon);
3154         seq_printf(seq, "Up Delay (ms): %d\n",
3155                    bond->params.updelay * bond->params.miimon);
3156         seq_printf(seq, "Down Delay (ms): %d\n",
3157                    bond->params.downdelay * bond->params.miimon);
3158
3159
3160         /* ARP information */
3161         if(bond->params.arp_interval > 0) {
3162                 int printed=0;
3163                 seq_printf(seq, "ARP Polling Interval (ms): %d\n",
3164                                 bond->params.arp_interval);
3165
3166                 seq_printf(seq, "ARP IP target/s (n.n.n.n form):");
3167
3168                 for(i = 0; (i < BOND_MAX_ARP_TARGETS) ;i++) {
3169                         if (!bond->params.arp_targets[i])
3170                                 continue;
3171                         if (printed)
3172                                 seq_printf(seq, ",");
3173                         target = ntohl(bond->params.arp_targets[i]);
3174                         seq_printf(seq, " %d.%d.%d.%d", HIPQUAD(target));
3175                         printed = 1;
3176                 }
3177                 seq_printf(seq, "\n");
3178         }
3179
3180         if (bond->params.mode == BOND_MODE_8023AD) {
3181                 struct ad_info ad_info;
3182                 DECLARE_MAC_BUF(mac);
3183
3184                 seq_puts(seq, "\n802.3ad info\n");
3185                 seq_printf(seq, "LACP rate: %s\n",
3186                            (bond->params.lacp_fast) ? "fast" : "slow");
3187
3188                 if (bond_3ad_get_active_agg_info(bond, &ad_info)) {
3189                         seq_printf(seq, "bond %s has no active aggregator\n",
3190                                    bond->dev->name);
3191                 } else {
3192                         seq_printf(seq, "Active Aggregator Info:\n");
3193
3194                         seq_printf(seq, "\tAggregator ID: %d\n",
3195                                    ad_info.aggregator_id);
3196                         seq_printf(seq, "\tNumber of ports: %d\n",
3197                                    ad_info.ports);
3198                         seq_printf(seq, "\tActor Key: %d\n",
3199                                    ad_info.actor_key);
3200                         seq_printf(seq, "\tPartner Key: %d\n",
3201                                    ad_info.partner_key);
3202                         seq_printf(seq, "\tPartner Mac Address: %s\n",
3203                                    print_mac(mac, ad_info.partner_system));
3204                 }
3205         }
3206 }
3207
3208 static void bond_info_show_slave(struct seq_file *seq, const struct slave *slave)
3209 {
3210         struct bonding *bond = seq->private;
3211         DECLARE_MAC_BUF(mac);
3212
3213         seq_printf(seq, "\nSlave Interface: %s\n", slave->dev->name);
3214         seq_printf(seq, "MII Status: %s\n",
3215                    (slave->link == BOND_LINK_UP) ?  "up" : "down");
3216         seq_printf(seq, "Link Failure Count: %u\n",
3217                    slave->link_failure_count);
3218
3219         seq_printf(seq,
3220                    "Permanent HW addr: %s\n",
3221                    print_mac(mac, slave->perm_hwaddr));
3222
3223         if (bond->params.mode == BOND_MODE_8023AD) {
3224                 const struct aggregator *agg
3225                         = SLAVE_AD_INFO(slave).port.aggregator;
3226
3227                 if (agg) {
3228                         seq_printf(seq, "Aggregator ID: %d\n",
3229                                    agg->aggregator_identifier);
3230                 } else {
3231                         seq_puts(seq, "Aggregator ID: N/A\n");
3232                 }
3233         }
3234 }
3235
3236 static int bond_info_seq_show(struct seq_file *seq, void *v)
3237 {
3238         if (v == SEQ_START_TOKEN) {
3239                 seq_printf(seq, "%s\n", version);
3240                 bond_info_show_master(seq);
3241         } else {
3242                 bond_info_show_slave(seq, v);
3243         }
3244
3245         return 0;
3246 }
3247
3248 static struct seq_operations bond_info_seq_ops = {
3249         .start = bond_info_seq_start,
3250         .next  = bond_info_seq_next,
3251         .stop  = bond_info_seq_stop,
3252         .show  = bond_info_seq_show,
3253 };
3254
3255 static int bond_info_open(struct inode *inode, struct file *file)
3256 {
3257         struct seq_file *seq;
3258         struct proc_dir_entry *proc;
3259         int res;
3260
3261         res = seq_open(file, &bond_info_seq_ops);
3262         if (!res) {
3263                 /* recover the pointer buried in proc_dir_entry data */
3264                 seq = file->private_data;
3265                 proc = PDE(inode);
3266                 seq->private = proc->data;
3267         }
3268
3269         return res;
3270 }
3271
3272 static const struct file_operations bond_info_fops = {
3273         .owner   = THIS_MODULE,
3274         .open    = bond_info_open,
3275         .read    = seq_read,
3276         .llseek  = seq_lseek,
3277         .release = seq_release,
3278 };
3279
3280 static int bond_create_proc_entry(struct bonding *bond)
3281 {
3282         struct net_device *bond_dev = bond->dev;
3283
3284         if (bond_proc_dir) {
3285                 bond->proc_entry = proc_create_data(bond_dev->name,
3286                                                     S_IRUGO, bond_proc_dir,
3287                                                     &bond_info_fops, bond);
3288                 if (bond->proc_entry == NULL) {
3289                         printk(KERN_WARNING DRV_NAME
3290                                ": Warning: Cannot create /proc/net/%s/%s\n",
3291                                DRV_NAME, bond_dev->name);
3292                 } else {
3293                         memcpy(bond->proc_file_name, bond_dev->name, IFNAMSIZ);
3294                 }
3295         }
3296
3297         return 0;
3298 }
3299
3300 static void bond_remove_proc_entry(struct bonding *bond)
3301 {
3302         if (bond_proc_dir && bond->proc_entry) {
3303                 remove_proc_entry(bond->proc_file_name, bond_proc_dir);
3304                 memset(bond->proc_file_name, 0, IFNAMSIZ);
3305                 bond->proc_entry = NULL;
3306         }
3307 }
3308
3309 /* Create the bonding directory under /proc/net, if doesn't exist yet.
3310  * Caller must hold rtnl_lock.
3311  */
3312 static void bond_create_proc_dir(void)
3313 {
3314         int len = strlen(DRV_NAME);
3315
3316         for (bond_proc_dir = init_net.proc_net->subdir; bond_proc_dir;
3317              bond_proc_dir = bond_proc_dir->next) {
3318                 if ((bond_proc_dir->namelen == len) &&
3319                     !memcmp(bond_proc_dir->name, DRV_NAME, len)) {
3320                         break;
3321                 }
3322         }
3323
3324         if (!bond_proc_dir) {
3325                 bond_proc_dir = proc_mkdir(DRV_NAME, init_net.proc_net);
3326                 if (bond_proc_dir) {
3327                         bond_proc_dir->owner = THIS_MODULE;
3328                 } else {
3329                         printk(KERN_WARNING DRV_NAME
3330                                 ": Warning: cannot create /proc/net/%s\n",
3331                                 DRV_NAME);
3332                 }
3333         }
3334 }
3335
3336 /* Destroy the bonding directory under /proc/net, if empty.
3337  * Caller must hold rtnl_lock.
3338  */
3339 static void bond_destroy_proc_dir(void)
3340 {
3341         struct proc_dir_entry *de;
3342
3343         if (!bond_proc_dir) {
3344                 return;
3345         }
3346
3347         /* verify that the /proc dir is empty */
3348         for (de = bond_proc_dir->subdir; de; de = de->next) {
3349                 /* ignore . and .. */
3350                 if (*(de->name) != '.') {
3351                         break;
3352                 }
3353         }
3354
3355         if (de) {
3356                 if (bond_proc_dir->owner == THIS_MODULE) {
3357                         bond_proc_dir->owner = NULL;
3358                 }
3359         } else {
3360                 remove_proc_entry(DRV_NAME, init_net.proc_net);
3361                 bond_proc_dir = NULL;
3362         }
3363 }
3364 #endif /* CONFIG_PROC_FS */
3365
3366 /*-------------------------- netdev event handling --------------------------*/
3367
3368 /*
3369  * Change device name
3370  */
3371 static int bond_event_changename(struct bonding *bond)
3372 {
3373 #ifdef CONFIG_PROC_FS
3374         bond_remove_proc_entry(bond);
3375         bond_create_proc_entry(bond);
3376 #endif
3377         down_write(&(bonding_rwsem));
3378         bond_destroy_sysfs_entry(bond);
3379         bond_create_sysfs_entry(bond);
3380         up_write(&(bonding_rwsem));
3381         return NOTIFY_DONE;
3382 }
3383
3384 static int bond_master_netdev_event(unsigned long event, struct net_device *bond_dev)
3385 {
3386         struct bonding *event_bond = bond_dev->priv;
3387
3388         switch (event) {
3389         case NETDEV_CHANGENAME:
3390                 return bond_event_changename(event_bond);
3391         case NETDEV_UNREGISTER:
3392                 bond_release_all(event_bond->dev);
3393                 break;
3394         default:
3395                 break;
3396         }
3397
3398         return NOTIFY_DONE;
3399 }
3400
3401 static int bond_slave_netdev_event(unsigned long event, struct net_device *slave_dev)
3402 {
3403         struct net_device *bond_dev = slave_dev->master;
3404         struct bonding *bond = bond_dev->priv;
3405
3406         switch (event) {
3407         case NETDEV_UNREGISTER:
3408                 if (bond_dev) {
3409                         if (bond->setup_by_slave)
3410                                 bond_release_and_destroy(bond_dev, slave_dev);
3411                         else
3412                                 bond_release(bond_dev, slave_dev);
3413                 }
3414                 break;
3415         case NETDEV_CHANGE:
3416                 /*
3417                  * TODO: is this what we get if somebody
3418                  * sets up a hierarchical bond, then rmmod's
3419                  * one of the slave bonding devices?
3420                  */
3421                 break;
3422         case NETDEV_DOWN:
3423                 /*
3424                  * ... Or is it this?
3425                  */
3426                 break;
3427         case NETDEV_CHANGEMTU:
3428                 /*
3429                  * TODO: Should slaves be allowed to
3430                  * independently alter their MTU?  For
3431                  * an active-backup bond, slaves need
3432                  * not be the same type of device, so
3433                  * MTUs may vary.  For other modes,
3434                  * slaves arguably should have the
3435                  * same MTUs. To do this, we'd need to
3436                  * take over the slave's change_mtu
3437                  * function for the duration of their
3438                  * servitude.
3439                  */
3440                 break;
3441         case NETDEV_CHANGENAME:
3442                 /*
3443                  * TODO: handle changing the primary's name
3444                  */
3445                 break;
3446         case NETDEV_FEAT_CHANGE:
3447                 bond_compute_features(bond);
3448                 break;
3449         default:
3450                 break;
3451         }
3452
3453         return NOTIFY_DONE;
3454 }
3455
3456 /*
3457  * bond_netdev_event: handle netdev notifier chain events.
3458  *
3459  * This function receives events for the netdev chain.  The caller (an
3460  * ioctl handler calling blocking_notifier_call_chain) holds the necessary
3461  * locks for us to safely manipulate the slave devices (RTNL lock,
3462  * dev_probe_lock).
3463  */
3464 static int bond_netdev_event(struct notifier_block *this, unsigned long event, void *ptr)
3465 {
3466         struct net_device *event_dev = (struct net_device *)ptr;
3467
3468         if (dev_net(event_dev) != &init_net)
3469                 return NOTIFY_DONE;
3470
3471         dprintk("event_dev: %s, event: %lx\n",
3472                 (event_dev ? event_dev->name : "None"),
3473                 event);
3474
3475         if (!(event_dev->priv_flags & IFF_BONDING))
3476                 return NOTIFY_DONE;
3477
3478         if (event_dev->flags & IFF_MASTER) {
3479                 dprintk("IFF_MASTER\n");
3480                 return bond_master_netdev_event(event, event_dev);
3481         }
3482
3483         if (event_dev->flags & IFF_SLAVE) {
3484                 dprintk("IFF_SLAVE\n");
3485                 return bond_slave_netdev_event(event, event_dev);
3486         }
3487
3488         return NOTIFY_DONE;
3489 }
3490
3491 /*
3492  * bond_inetaddr_event: handle inetaddr notifier chain events.
3493  *
3494  * We keep track of device IPs primarily to use as source addresses in
3495  * ARP monitor probes (rather than spewing out broadcasts all the time).
3496  *
3497  * We track one IP for the main device (if it has one), plus one per VLAN.
3498  */
3499 static int bond_inetaddr_event(struct notifier_block *this, unsigned long event, void *ptr)
3500 {
3501         struct in_ifaddr *ifa = ptr;
3502         struct net_device *vlan_dev, *event_dev = ifa->ifa_dev->dev;
3503         struct bonding *bond, *bond_next;
3504         struct vlan_entry *vlan, *vlan_next;
3505
3506         if (dev_net(ifa->ifa_dev->dev) != &init_net)
3507                 return NOTIFY_DONE;
3508
3509         list_for_each_entry_safe(bond, bond_next, &bond_dev_list, bond_list) {
3510                 if (bond->dev == event_dev) {
3511                         switch (event) {
3512                         case NETDEV_UP:
3513                                 bond->master_ip = ifa->ifa_local;
3514                                 return NOTIFY_OK;
3515                         case NETDEV_DOWN:
3516                                 bond->master_ip = bond_glean_dev_ip(bond->dev);
3517                                 return NOTIFY_OK;
3518                         default:
3519                                 return NOTIFY_DONE;
3520                         }
3521                 }
3522
3523                 if (list_empty(&bond->vlan_list))
3524                         continue;
3525
3526                 list_for_each_entry_safe(vlan, vlan_next, &bond->vlan_list,
3527                                          vlan_list) {
3528                         vlan_dev = vlan_group_get_device(bond->vlgrp, vlan->vlan_id);
3529                         if (vlan_dev == event_dev) {
3530                                 switch (event) {
3531                                 case NETDEV_UP:
3532                                         vlan->vlan_ip = ifa->ifa_local;
3533                                         return NOTIFY_OK;
3534                                 case NETDEV_DOWN:
3535                                         vlan->vlan_ip =
3536                                                 bond_glean_dev_ip(vlan_dev);
3537                                         return NOTIFY_OK;
3538                                 default:
3539                                         return NOTIFY_DONE;
3540                                 }
3541                         }
3542                 }
3543         }
3544         return NOTIFY_DONE;
3545 }
3546
3547 static struct notifier_block bond_netdev_notifier = {
3548         .notifier_call = bond_netdev_event,
3549 };
3550
3551 static struct notifier_block bond_inetaddr_notifier = {
3552         .notifier_call = bond_inetaddr_event,
3553 };
3554
3555 /*-------------------------- Packet type handling ---------------------------*/
3556
3557 /* register to receive lacpdus on a bond */
3558 static void bond_register_lacpdu(struct bonding *bond)
3559 {
3560         struct packet_type *pk_type = &(BOND_AD_INFO(bond).ad_pkt_type);
3561
3562         /* initialize packet type */
3563         pk_type->type = PKT_TYPE_LACPDU;
3564         pk_type->dev = bond->dev;
3565         pk_type->func = bond_3ad_lacpdu_recv;
3566
3567         dev_add_pack(pk_type);
3568 }
3569
3570 /* unregister to receive lacpdus on a bond */
3571 static void bond_unregister_lacpdu(struct bonding *bond)
3572 {
3573         dev_remove_pack(&(BOND_AD_INFO(bond).ad_pkt_type));
3574 }
3575
3576 void bond_register_arp(struct bonding *bond)
3577 {
3578         struct packet_type *pt = &bond->arp_mon_pt;
3579
3580         if (pt->type)
3581                 return;
3582
3583         pt->type = htons(ETH_P_ARP);
3584         pt->dev = bond->dev;
3585         pt->func = bond_arp_rcv;
3586         dev_add_pack(pt);
3587 }
3588
3589 void bond_unregister_arp(struct bonding *bond)
3590 {
3591         struct packet_type *pt = &bond->arp_mon_pt;
3592
3593         dev_remove_pack(pt);
3594         pt->type = 0;
3595 }
3596
3597 /*---------------------------- Hashing Policies -----------------------------*/
3598
3599 /*
3600  * Hash for the output device based upon layer 2 and layer 3 data. If
3601  * the packet is not IP mimic bond_xmit_hash_policy_l2()
3602  */
3603 static int bond_xmit_hash_policy_l23(struct sk_buff *skb,
3604                                      struct net_device *bond_dev, int count)
3605 {
3606         struct ethhdr *data = (struct ethhdr *)skb->data;
3607         struct iphdr *iph = ip_hdr(skb);
3608
3609         if (skb->protocol == __constant_htons(ETH_P_IP)) {
3610                 return ((ntohl(iph->saddr ^ iph->daddr) & 0xffff) ^
3611                         (data->h_dest[5] ^ bond_dev->dev_addr[5])) % count;
3612         }
3613
3614         return (data->h_dest[5] ^ bond_dev->dev_addr[5]) % count;
3615 }
3616
3617 /*
3618  * Hash for the output device based upon layer 3 and layer 4 data. If
3619  * the packet is a frag or not TCP or UDP, just use layer 3 data.  If it is
3620  * altogether not IP, mimic bond_xmit_hash_policy_l2()
3621  */
3622 static int bond_xmit_hash_policy_l34(struct sk_buff *skb,
3623                                     struct net_device *bond_dev, int count)
3624 {
3625         struct ethhdr *data = (struct ethhdr *)skb->data;
3626         struct iphdr *iph = ip_hdr(skb);
3627         __be16 *layer4hdr = (__be16 *)((u32 *)iph + iph->ihl);
3628         int layer4_xor = 0;
3629
3630         if (skb->protocol == __constant_htons(ETH_P_IP)) {
3631                 if (!(iph->frag_off & __constant_htons(IP_MF|IP_OFFSET)) &&
3632                     (iph->protocol == IPPROTO_TCP ||
3633                      iph->protocol == IPPROTO_UDP)) {
3634                         layer4_xor = ntohs((*layer4hdr ^ *(layer4hdr + 1)));
3635                 }
3636                 return (layer4_xor ^
3637                         ((ntohl(iph->saddr ^ iph->daddr)) & 0xffff)) % count;
3638
3639         }
3640
3641         return (data->h_dest[5] ^ bond_dev->dev_addr[5]) % count;
3642 }
3643
3644 /*
3645  * Hash for the output device based upon layer 2 data
3646  */
3647 static int bond_xmit_hash_policy_l2(struct sk_buff *skb,
3648                                    struct net_device *bond_dev, int count)
3649 {
3650         struct ethhdr *data = (struct ethhdr *)skb->data;
3651
3652         return (data->h_dest[5] ^ bond_dev->dev_addr[5]) % count;
3653 }
3654
3655 /*-------------------------- Device entry points ----------------------------*/
3656
3657 static int bond_open(struct net_device *bond_dev)
3658 {
3659         struct bonding *bond = bond_dev->priv;
3660
3661         bond->kill_timers = 0;
3662
3663         if ((bond->params.mode == BOND_MODE_TLB) ||
3664             (bond->params.mode == BOND_MODE_ALB)) {
3665                 /* bond_alb_initialize must be called before the timer
3666                  * is started.
3667                  */
3668                 if (bond_alb_initialize(bond, (bond->params.mode == BOND_MODE_ALB))) {
3669                         /* something went wrong - fail the open operation */
3670                         return -1;
3671                 }
3672
3673                 INIT_DELAYED_WORK(&bond->alb_work, bond_alb_monitor);
3674                 queue_delayed_work(bond->wq, &bond->alb_work, 0);
3675         }
3676
3677         if (bond->params.miimon) {  /* link check interval, in milliseconds. */
3678                 INIT_DELAYED_WORK(&bond->mii_work, bond_mii_monitor);
3679                 queue_delayed_work(bond->wq, &bond->mii_work, 0);
3680         }
3681
3682         if (bond->params.arp_interval) {  /* arp interval, in milliseconds. */
3683                 if (bond->params.mode == BOND_MODE_ACTIVEBACKUP)
3684                         INIT_DELAYED_WORK(&bond->arp_work,
3685                                           bond_activebackup_arp_mon);
3686                 else
3687                         INIT_DELAYED_WORK(&bond->arp_work,
3688                                           bond_loadbalance_arp_mon);
3689
3690                 queue_delayed_work(bond->wq, &bond->arp_work, 0);
3691                 if (bond->params.arp_validate)
3692                         bond_register_arp(bond);
3693         }
3694
3695         if (bond->params.mode == BOND_MODE_8023AD) {
3696                 INIT_DELAYED_WORK(&bond->ad_work, bond_3ad_state_machine_handler);
3697                 queue_delayed_work(bond->wq, &bond->ad_work, 0);
3698                 /* register to receive LACPDUs */
3699                 bond_register_lacpdu(bond);
3700         }
3701
3702         return 0;
3703 }
3704
3705 static int bond_close(struct net_device *bond_dev)
3706 {
3707         struct bonding *bond = bond_dev->priv;
3708
3709         if (bond->params.mode == BOND_MODE_8023AD) {
3710                 /* Unregister the receive of LACPDUs */
3711                 bond_unregister_lacpdu(bond);
3712         }
3713
3714         if (bond->params.arp_validate)
3715                 bond_unregister_arp(bond);
3716
3717         write_lock_bh(&bond->lock);
3718
3719
3720         /* signal timers not to re-arm */
3721         bond->kill_timers = 1;
3722
3723         write_unlock_bh(&bond->lock);
3724
3725         if (bond->params.miimon) {  /* link check interval, in milliseconds. */
3726                 cancel_delayed_work(&bond->mii_work);
3727         }
3728
3729         if (bond->params.arp_interval) {  /* arp interval, in milliseconds. */
3730                 cancel_delayed_work(&bond->arp_work);
3731         }
3732
3733         switch (bond->params.mode) {
3734         case BOND_MODE_8023AD:
3735                 cancel_delayed_work(&bond->ad_work);
3736                 break;
3737         case BOND_MODE_TLB:
3738         case BOND_MODE_ALB:
3739                 cancel_delayed_work(&bond->alb_work);
3740                 break;
3741         default:
3742                 break;
3743         }
3744
3745
3746         if ((bond->params.mode == BOND_MODE_TLB) ||
3747             (bond->params.mode == BOND_MODE_ALB)) {
3748                 /* Must be called only after all
3749                  * slaves have been released
3750                  */
3751                 bond_alb_deinitialize(bond);
3752         }
3753
3754         return 0;
3755 }
3756
3757 static struct net_device_stats *bond_get_stats(struct net_device *bond_dev)
3758 {
3759         struct bonding *bond = bond_dev->priv;
3760         struct net_device_stats *stats = &(bond->stats), *sstats;
3761         struct net_device_stats local_stats;
3762         struct slave *slave;
3763         int i;
3764
3765         memset(&local_stats, 0, sizeof(struct net_device_stats));
3766
3767         read_lock_bh(&bond->lock);
3768
3769         bond_for_each_slave(bond, slave, i) {
3770                 sstats = slave->dev->get_stats(slave->dev);
3771                 local_stats.rx_packets += sstats->rx_packets;
3772                 local_stats.rx_bytes += sstats->rx_bytes;
3773                 local_stats.rx_errors += sstats->rx_errors;
3774                 local_stats.rx_dropped += sstats->rx_dropped;
3775
3776                 local_stats.tx_packets += sstats->tx_packets;
3777                 local_stats.tx_bytes += sstats->tx_bytes;
3778                 local_stats.tx_errors += sstats->tx_errors;
3779                 local_stats.tx_dropped += sstats->tx_dropped;
3780
3781                 local_stats.multicast += sstats->multicast;
3782                 local_stats.collisions += sstats->collisions;
3783
3784                 local_stats.rx_length_errors += sstats->rx_length_errors;
3785                 local_stats.rx_over_errors += sstats->rx_over_errors;
3786                 local_stats.rx_crc_errors += sstats->rx_crc_errors;
3787                 local_stats.rx_frame_errors += sstats->rx_frame_errors;
3788                 local_stats.rx_fifo_errors += sstats->rx_fifo_errors;
3789                 local_stats.rx_missed_errors += sstats->rx_missed_errors;
3790
3791                 local_stats.tx_aborted_errors += sstats->tx_aborted_errors;
3792                 local_stats.tx_carrier_errors += sstats->tx_carrier_errors;
3793                 local_stats.tx_fifo_errors += sstats->tx_fifo_errors;
3794                 local_stats.tx_heartbeat_errors += sstats->tx_heartbeat_errors;
3795                 local_stats.tx_window_errors += sstats->tx_window_errors;
3796         }
3797
3798         memcpy(stats, &local_stats, sizeof(struct net_device_stats));
3799
3800         read_unlock_bh(&bond->lock);
3801
3802         return stats;
3803 }
3804
3805 static int bond_do_ioctl(struct net_device *bond_dev, struct ifreq *ifr, int cmd)
3806 {
3807         struct net_device *slave_dev = NULL;
3808         struct ifbond k_binfo;
3809         struct ifbond __user *u_binfo = NULL;
3810         struct ifslave k_sinfo;
3811         struct ifslave __user *u_sinfo = NULL;
3812         struct mii_ioctl_data *mii = NULL;
3813         int res = 0;
3814
3815         dprintk("bond_ioctl: master=%s, cmd=%d\n",
3816                 bond_dev->name, cmd);
3817
3818         switch (cmd) {
3819         case SIOCGMIIPHY:
3820                 mii = if_mii(ifr);
3821                 if (!mii) {
3822                         return -EINVAL;
3823                 }
3824                 mii->phy_id = 0;
3825                 /* Fall Through */
3826         case SIOCGMIIREG:
3827                 /*
3828                  * We do this again just in case we were called by SIOCGMIIREG
3829                  * instead of SIOCGMIIPHY.
3830                  */
3831                 mii = if_mii(ifr);
3832                 if (!mii) {
3833                         return -EINVAL;
3834                 }
3835
3836                 if (mii->reg_num == 1) {
3837                         struct bonding *bond = bond_dev->priv;
3838                         mii->val_out = 0;
3839                         read_lock(&bond->lock);
3840                         read_lock(&bond->curr_slave_lock);
3841                         if (netif_carrier_ok(bond->dev)) {
3842                                 mii->val_out = BMSR_LSTATUS;
3843                         }
3844                         read_unlock(&bond->curr_slave_lock);
3845                         read_unlock(&bond->lock);
3846                 }
3847
3848                 return 0;
3849         case BOND_INFO_QUERY_OLD:
3850         case SIOCBONDINFOQUERY:
3851                 u_binfo = (struct ifbond __user *)ifr->ifr_data;
3852
3853                 if (copy_from_user(&k_binfo, u_binfo, sizeof(ifbond))) {
3854                         return -EFAULT;
3855                 }
3856
3857                 res = bond_info_query(bond_dev, &k_binfo);
3858                 if (res == 0) {
3859                         if (copy_to_user(u_binfo, &k_binfo, sizeof(ifbond))) {
3860                                 return -EFAULT;
3861                         }
3862                 }
3863
3864                 return res;
3865         case BOND_SLAVE_INFO_QUERY_OLD:
3866         case SIOCBONDSLAVEINFOQUERY:
3867                 u_sinfo = (struct ifslave __user *)ifr->ifr_data;
3868
3869                 if (copy_from_user(&k_sinfo, u_sinfo, sizeof(ifslave))) {
3870                         return -EFAULT;
3871                 }
3872
3873                 res = bond_slave_info_query(bond_dev, &k_sinfo);
3874                 if (res == 0) {
3875                         if (copy_to_user(u_sinfo, &k_sinfo, sizeof(ifslave))) {
3876                                 return -EFAULT;
3877                         }
3878                 }
3879
3880                 return res;
3881         default:
3882                 /* Go on */
3883                 break;
3884         }
3885
3886         if (!capable(CAP_NET_ADMIN)) {
3887                 return -EPERM;
3888         }
3889
3890         down_write(&(bonding_rwsem));
3891         slave_dev = dev_get_by_name(&init_net, ifr->ifr_slave);
3892
3893         dprintk("slave_dev=%p: \n", slave_dev);
3894
3895         if (!slave_dev) {
3896                 res = -ENODEV;
3897         } else {
3898                 dprintk("slave_dev->name=%s: \n", slave_dev->name);
3899                 switch (cmd) {
3900                 case BOND_ENSLAVE_OLD:
3901                 case SIOCBONDENSLAVE:
3902                         res = bond_enslave(bond_dev, slave_dev);
3903                         break;
3904                 case BOND_RELEASE_OLD:
3905                 case SIOCBONDRELEASE:
3906                         res = bond_release(bond_dev, slave_dev);
3907                         break;
3908                 case BOND_SETHWADDR_OLD:
3909                 case SIOCBONDSETHWADDR:
3910                         res = bond_sethwaddr(bond_dev, slave_dev);
3911                         break;
3912                 case BOND_CHANGE_ACTIVE_OLD:
3913                 case SIOCBONDCHANGEACTIVE:
3914                         res = bond_ioctl_change_active(bond_dev, slave_dev);
3915                         break;
3916                 default:
3917                         res = -EOPNOTSUPP;
3918                 }
3919
3920                 dev_put(slave_dev);
3921         }
3922
3923         up_write(&(bonding_rwsem));
3924         return res;
3925 }
3926
3927 static void bond_set_multicast_list(struct net_device *bond_dev)
3928 {
3929         struct bonding *bond = bond_dev->priv;
3930         struct dev_mc_list *dmi;
3931
3932         /*
3933          * Do promisc before checking multicast_mode
3934          */
3935         if ((bond_dev->flags & IFF_PROMISC) && !(bond->flags & IFF_PROMISC)) {
3936                 bond_set_promiscuity(bond, 1);
3937         }
3938
3939         if (!(bond_dev->flags & IFF_PROMISC) && (bond->flags & IFF_PROMISC)) {
3940                 bond_set_promiscuity(bond, -1);
3941         }
3942
3943         /* set allmulti flag to slaves */
3944         if ((bond_dev->flags & IFF_ALLMULTI) && !(bond->flags & IFF_ALLMULTI)) {
3945                 bond_set_allmulti(bond, 1);
3946         }
3947
3948         if (!(bond_dev->flags & IFF_ALLMULTI) && (bond->flags & IFF_ALLMULTI)) {
3949                 bond_set_allmulti(bond, -1);
3950         }
3951
3952         read_lock(&bond->lock);
3953
3954         bond->flags = bond_dev->flags;
3955
3956         /* looking for addresses to add to slaves' mc list */
3957         for (dmi = bond_dev->mc_list; dmi; dmi = dmi->next) {
3958                 if (!bond_mc_list_find_dmi(dmi, bond->mc_list)) {
3959                         bond_mc_add(bond, dmi->dmi_addr, dmi->dmi_addrlen);
3960                 }
3961         }
3962
3963         /* looking for addresses to delete from slaves' list */
3964         for (dmi = bond->mc_list; dmi; dmi = dmi->next) {
3965                 if (!bond_mc_list_find_dmi(dmi, bond_dev->mc_list)) {
3966                         bond_mc_delete(bond, dmi->dmi_addr, dmi->dmi_addrlen);
3967                 }
3968         }
3969
3970         /* save master's multicast list */
3971         bond_mc_list_destroy(bond);
3972         bond_mc_list_copy(bond_dev->mc_list, bond, GFP_ATOMIC);
3973
3974         read_unlock(&bond->lock);
3975 }
3976
3977 /*
3978  * Change the MTU of all of a master's slaves to match the master
3979  */
3980 static int bond_change_mtu(struct net_device *bond_dev, int new_mtu)
3981 {
3982         struct bonding *bond = bond_dev->priv;
3983         struct slave *slave, *stop_at;
3984         int res = 0;
3985         int i;
3986
3987         dprintk("bond=%p, name=%s, new_mtu=%d\n", bond,
3988                 (bond_dev ? bond_dev->name : "None"), new_mtu);
3989
3990         /* Can't hold bond->lock with bh disabled here since
3991          * some base drivers panic. On the other hand we can't
3992          * hold bond->lock without bh disabled because we'll
3993          * deadlock. The only solution is to rely on the fact
3994          * that we're under rtnl_lock here, and the slaves
3995          * list won't change. This doesn't solve the problem
3996          * of setting the slave's MTU while it is
3997          * transmitting, but the assumption is that the base
3998          * driver can handle that.
3999          *
4000          * TODO: figure out a way to safely iterate the slaves
4001          * list, but without holding a lock around the actual
4002          * call to the base driver.
4003          */
4004
4005         bond_for_each_slave(bond, slave, i) {
4006                 dprintk("s %p s->p %p c_m %p\n", slave,
4007                         slave->prev, slave->dev->change_mtu);
4008
4009                 res = dev_set_mtu(slave->dev, new_mtu);
4010
4011                 if (res) {
4012                         /* If we failed to set the slave's mtu to the new value
4013                          * we must abort the operation even in ACTIVE_BACKUP
4014                          * mode, because if we allow the backup slaves to have
4015                          * different mtu values than the active slave we'll
4016                          * need to change their mtu when doing a failover. That
4017                          * means changing their mtu from timer context, which
4018                          * is probably not a good idea.
4019                          */
4020                         dprintk("err %d %s\n", res, slave->dev->name);
4021                         goto unwind;
4022                 }
4023         }
4024
4025         bond_dev->mtu = new_mtu;
4026
4027         return 0;
4028
4029 unwind:
4030         /* unwind from head to the slave that failed */
4031         stop_at = slave;
4032         bond_for_each_slave_from_to(bond, slave, i, bond->first_slave, stop_at) {
4033                 int tmp_res;
4034
4035                 tmp_res = dev_set_mtu(slave->dev, bond_dev->mtu);
4036                 if (tmp_res) {
4037                         dprintk("unwind err %d dev %s\n", tmp_res,
4038                                 slave->dev->name);
4039                 }
4040         }
4041
4042         return res;
4043 }
4044
4045 /*
4046  * Change HW address
4047  *
4048  * Note that many devices must be down to change the HW address, and
4049  * downing the master releases all slaves.  We can make bonds full of
4050  * bonding devices to test this, however.
4051  */
4052 static int bond_set_mac_address(struct net_device *bond_dev, void *addr)
4053 {
4054         struct bonding *bond = bond_dev->priv;
4055         struct sockaddr *sa = addr, tmp_sa;
4056         struct slave *slave, *stop_at;
4057         int res = 0;
4058         int i;
4059
4060         dprintk("bond=%p, name=%s\n", bond, (bond_dev ? bond_dev->name : "None"));
4061
4062         /*
4063          * If fail_over_mac is enabled, do nothing and return success.
4064          * Returning an error causes ifenslave to fail.
4065          */
4066         if (bond->params.fail_over_mac)
4067                 return 0;
4068
4069         if (!is_valid_ether_addr(sa->sa_data)) {
4070                 return -EADDRNOTAVAIL;
4071         }
4072
4073         /* Can't hold bond->lock with bh disabled here since
4074          * some base drivers panic. On the other hand we can't
4075          * hold bond->lock without bh disabled because we'll
4076          * deadlock. The only solution is to rely on the fact
4077          * that we're under rtnl_lock here, and the slaves
4078          * list won't change. This doesn't solve the problem
4079          * of setting the slave's hw address while it is
4080          * transmitting, but the assumption is that the base
4081          * driver can handle that.
4082          *
4083          * TODO: figure out a way to safely iterate the slaves
4084          * list, but without holding a lock around the actual
4085          * call to the base driver.
4086          */
4087
4088         bond_for_each_slave(bond, slave, i) {
4089                 dprintk("slave %p %s\n", slave, slave->dev->name);
4090
4091                 if (slave->dev->set_mac_address == NULL) {
4092                         res = -EOPNOTSUPP;
4093                         dprintk("EOPNOTSUPP %s\n", slave->dev->name);
4094                         goto unwind;
4095                 }
4096
4097                 res = dev_set_mac_address(slave->dev, addr);
4098                 if (res) {
4099                         /* TODO: consider downing the slave
4100                          * and retry ?
4101                          * User should expect communications
4102                          * breakage anyway until ARP finish
4103                          * updating, so...
4104                          */
4105                         dprintk("err %d %s\n", res, slave->dev->name);
4106                         goto unwind;
4107                 }
4108         }
4109
4110         /* success */
4111         memcpy(bond_dev->dev_addr, sa->sa_data, bond_dev->addr_len);
4112         return 0;
4113
4114 unwind:
4115         memcpy(tmp_sa.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
4116         tmp_sa.sa_family = bond_dev->type;
4117
4118         /* unwind from head to the slave that failed */
4119         stop_at = slave;
4120         bond_for_each_slave_from_to(bond, slave, i, bond->first_slave, stop_at) {
4121                 int tmp_res;
4122
4123                 tmp_res = dev_set_mac_address(slave->dev, &tmp_sa);
4124                 if (tmp_res) {
4125                         dprintk("unwind err %d dev %s\n", tmp_res,
4126                                 slave->dev->name);
4127                 }
4128         }
4129
4130         return res;
4131 }
4132
4133 static int bond_xmit_roundrobin(struct sk_buff *skb, struct net_device *bond_dev)
4134 {
4135         struct bonding *bond = bond_dev->priv;
4136         struct slave *slave, *start_at;
4137         int i, slave_no, res = 1;
4138
4139         read_lock(&bond->lock);
4140
4141         if (!BOND_IS_OK(bond)) {
4142                 goto out;
4143         }
4144
4145         /*
4146          * Concurrent TX may collide on rr_tx_counter; we accept that
4147          * as being rare enough not to justify using an atomic op here
4148          */
4149         slave_no = bond->rr_tx_counter++ % bond->slave_cnt;
4150
4151         bond_for_each_slave(bond, slave, i) {
4152                 slave_no--;
4153                 if (slave_no < 0) {
4154                         break;
4155                 }
4156         }
4157
4158         start_at = slave;
4159         bond_for_each_slave_from(bond, slave, i, start_at) {
4160                 if (IS_UP(slave->dev) &&
4161                     (slave->link == BOND_LINK_UP) &&
4162                     (slave->state == BOND_STATE_ACTIVE)) {
4163                         res = bond_dev_queue_xmit(bond, skb, slave->dev);
4164                         break;
4165                 }
4166         }
4167
4168 out:
4169         if (res) {
4170                 /* no suitable interface, frame not sent */
4171                 dev_kfree_skb(skb);
4172         }
4173         read_unlock(&bond->lock);
4174         return 0;
4175 }
4176
4177
4178 /*
4179  * in active-backup mode, we know that bond->curr_active_slave is always valid if
4180  * the bond has a usable interface.
4181  */
4182 static int bond_xmit_activebackup(struct sk_buff *skb, struct net_device *bond_dev)
4183 {
4184         struct bonding *bond = bond_dev->priv;
4185         int res = 1;
4186
4187         read_lock(&bond->lock);
4188         read_lock(&bond->curr_slave_lock);
4189
4190         if (!BOND_IS_OK(bond)) {
4191                 goto out;
4192         }
4193
4194         if (!bond->curr_active_slave)
4195                 goto out;
4196
4197         res = bond_dev_queue_xmit(bond, skb, bond->curr_active_slave->dev);
4198
4199 out:
4200         if (res) {
4201                 /* no suitable interface, frame not sent */
4202                 dev_kfree_skb(skb);
4203         }
4204         read_unlock(&bond->curr_slave_lock);
4205         read_unlock(&bond->lock);
4206         return 0;
4207 }
4208
4209 /*
4210  * In bond_xmit_xor() , we determine the output device by using a pre-
4211  * determined xmit_hash_policy(), If the selected device is not enabled,
4212  * find the next active slave.
4213  */
4214 static int bond_xmit_xor(struct sk_buff *skb, struct net_device *bond_dev)
4215 {
4216         struct bonding *bond = bond_dev->priv;
4217         struct slave *slave, *start_at;
4218         int slave_no;
4219         int i;
4220         int res = 1;
4221
4222         read_lock(&bond->lock);
4223
4224         if (!BOND_IS_OK(bond)) {
4225                 goto out;
4226         }
4227
4228         slave_no = bond->xmit_hash_policy(skb, bond_dev, bond->slave_cnt);
4229
4230         bond_for_each_slave(bond, slave, i) {
4231                 slave_no--;
4232                 if (slave_no < 0) {
4233                         break;
4234                 }
4235         }
4236
4237         start_at = slave;
4238
4239         bond_for_each_slave_from(bond, slave, i, start_at) {
4240                 if (IS_UP(slave->dev) &&
4241                     (slave->link == BOND_LINK_UP) &&
4242                     (slave->state == BOND_STATE_ACTIVE)) {
4243                         res = bond_dev_queue_xmit(bond, skb, slave->dev);
4244                         break;
4245                 }
4246         }
4247
4248 out:
4249         if (res) {
4250                 /* no suitable interface, frame not sent */
4251                 dev_kfree_skb(skb);
4252         }
4253         read_unlock(&bond->lock);
4254         return 0;
4255 }
4256
4257 /*
4258  * in broadcast mode, we send everything to all usable interfaces.
4259  */
4260 static int bond_xmit_broadcast(struct sk_buff *skb, struct net_device *bond_dev)
4261 {
4262         struct bonding *bond = bond_dev->priv;
4263         struct slave *slave, *start_at;
4264         struct net_device *tx_dev = NULL;
4265         int i;
4266         int res = 1;
4267
4268         read_lock(&bond->lock);
4269
4270         if (!BOND_IS_OK(bond)) {
4271                 goto out;
4272         }
4273
4274         read_lock(&bond->curr_slave_lock);
4275         start_at = bond->curr_active_slave;
4276         read_unlock(&bond->curr_slave_lock);
4277
4278         if (!start_at) {
4279                 goto out;
4280         }
4281
4282         bond_for_each_slave_from(bond, slave, i, start_at) {
4283                 if (IS_UP(slave->dev) &&
4284                     (slave->link == BOND_LINK_UP) &&
4285                     (slave->state == BOND_STATE_ACTIVE)) {
4286                         if (tx_dev) {
4287                                 struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
4288                                 if (!skb2) {
4289                                         printk(KERN_ERR DRV_NAME
4290                                                ": %s: Error: bond_xmit_broadcast(): "
4291                                                "skb_clone() failed\n",
4292                                                bond_dev->name);
4293                                         continue;
4294                                 }
4295
4296                                 res = bond_dev_queue_xmit(bond, skb2, tx_dev);
4297                                 if (res) {
4298                                         dev_kfree_skb(skb2);
4299                                         continue;
4300                                 }
4301                         }
4302                         tx_dev = slave->dev;
4303                 }
4304         }
4305
4306         if (tx_dev) {
4307                 res = bond_dev_queue_xmit(bond, skb, tx_dev);
4308         }
4309
4310 out:
4311         if (res) {
4312                 /* no suitable interface, frame not sent */
4313                 dev_kfree_skb(skb);
4314         }
4315         /* frame sent to all suitable interfaces */
4316         read_unlock(&bond->lock);
4317         return 0;
4318 }
4319
4320 /*------------------------- Device initialization ---------------------------*/
4321
4322 static void bond_set_xmit_hash_policy(struct bonding *bond)
4323 {
4324         switch (bond->params.xmit_policy) {
4325         case BOND_XMIT_POLICY_LAYER23:
4326                 bond->xmit_hash_policy = bond_xmit_hash_policy_l23;
4327                 break;
4328         case BOND_XMIT_POLICY_LAYER34:
4329                 bond->xmit_hash_policy = bond_xmit_hash_policy_l34;
4330                 break;
4331         case BOND_XMIT_POLICY_LAYER2:
4332         default:
4333                 bond->xmit_hash_policy = bond_xmit_hash_policy_l2;
4334                 break;
4335         }
4336 }
4337
4338 /*
4339  * set bond mode specific net device operations
4340  */
4341 void bond_set_mode_ops(struct bonding *bond, int mode)
4342 {
4343         struct net_device *bond_dev = bond->dev;
4344
4345         switch (mode) {
4346         case BOND_MODE_ROUNDROBIN:
4347                 bond_dev->hard_start_xmit = bond_xmit_roundrobin;
4348                 break;
4349         case BOND_MODE_ACTIVEBACKUP:
4350                 bond_dev->hard_start_xmit = bond_xmit_activebackup;
4351                 break;
4352         case BOND_MODE_XOR:
4353                 bond_dev->hard_start_xmit = bond_xmit_xor;
4354                 bond_set_xmit_hash_policy(bond);
4355                 break;
4356         case BOND_MODE_BROADCAST:
4357                 bond_dev->hard_start_xmit = bond_xmit_broadcast;
4358                 break;
4359         case BOND_MODE_8023AD:
4360                 bond_set_master_3ad_flags(bond);
4361                 bond_dev->hard_start_xmit = bond_3ad_xmit_xor;
4362                 bond_set_xmit_hash_policy(bond);
4363                 break;
4364         case BOND_MODE_ALB:
4365                 bond_set_master_alb_flags(bond);
4366                 /* FALLTHRU */
4367         case BOND_MODE_TLB:
4368                 bond_dev->hard_start_xmit = bond_alb_xmit;
4369                 bond_dev->set_mac_address = bond_alb_set_mac_address;
4370                 break;
4371         default:
4372                 /* Should never happen, mode already checked */
4373                 printk(KERN_ERR DRV_NAME
4374                        ": %s: Error: Unknown bonding mode %d\n",
4375                        bond_dev->name,
4376                        mode);
4377                 break;
4378         }
4379 }
4380
4381 static void bond_ethtool_get_drvinfo(struct net_device *bond_dev,
4382                                     struct ethtool_drvinfo *drvinfo)
4383 {
4384         strncpy(drvinfo->driver, DRV_NAME, 32);
4385         strncpy(drvinfo->version, DRV_VERSION, 32);
4386         snprintf(drvinfo->fw_version, 32, "%d", BOND_ABI_VERSION);
4387 }
4388
4389 static const struct ethtool_ops bond_ethtool_ops = {
4390         .get_drvinfo            = bond_ethtool_get_drvinfo,
4391 };
4392
4393 /*
4394  * Does not allocate but creates a /proc entry.
4395  * Allowed to fail.
4396  */
4397 static int bond_init(struct net_device *bond_dev, struct bond_params *params)
4398 {
4399         struct bonding *bond = bond_dev->priv;
4400
4401         dprintk("Begin bond_init for %s\n", bond_dev->name);
4402
4403         /* initialize rwlocks */
4404         rwlock_init(&bond->lock);
4405         rwlock_init(&bond->curr_slave_lock);
4406
4407         bond->params = *params; /* copy params struct */
4408
4409         bond->wq = create_singlethread_workqueue(bond_dev->name);
4410         if (!bond->wq)
4411                 return -ENOMEM;
4412
4413         /* Initialize pointers */
4414         bond->first_slave = NULL;
4415         bond->curr_active_slave = NULL;
4416         bond->current_arp_slave = NULL;
4417         bond->primary_slave = NULL;
4418         bond->dev = bond_dev;
4419         bond->send_grat_arp = 0;
4420         bond->setup_by_slave = 0;
4421         INIT_LIST_HEAD(&bond->vlan_list);
4422
4423         /* Initialize the device entry points */
4424         bond_dev->open = bond_open;
4425         bond_dev->stop = bond_close;
4426         bond_dev->get_stats = bond_get_stats;
4427         bond_dev->do_ioctl = bond_do_ioctl;
4428         bond_dev->ethtool_ops = &bond_ethtool_ops;
4429         bond_dev->set_multicast_list = bond_set_multicast_list;
4430         bond_dev->change_mtu = bond_change_mtu;
4431         bond_dev->set_mac_address = bond_set_mac_address;
4432         bond_dev->validate_addr = NULL;
4433
4434         bond_set_mode_ops(bond, bond->params.mode);
4435
4436         bond_dev->destructor = free_netdev;
4437
4438         /* Initialize the device options */
4439         bond_dev->tx_queue_len = 0;
4440         bond_dev->flags |= IFF_MASTER|IFF_MULTICAST;
4441         bond_dev->priv_flags |= IFF_BONDING;
4442
4443         /* At first, we block adding VLANs. That's the only way to
4444          * prevent problems that occur when adding VLANs over an
4445          * empty bond. The block will be removed once non-challenged
4446          * slaves are enslaved.
4447          */
4448         bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
4449
4450         /* don't acquire bond device's netif_tx_lock when
4451          * transmitting */
4452         bond_dev->features |= NETIF_F_LLTX;
4453
4454         /* By default, we declare the bond to be fully
4455          * VLAN hardware accelerated capable. Special
4456          * care is taken in the various xmit functions
4457          * when there are slaves that are not hw accel
4458          * capable
4459          */
4460         bond_dev->vlan_rx_register = bond_vlan_rx_register;
4461         bond_dev->vlan_rx_add_vid  = bond_vlan_rx_add_vid;
4462         bond_dev->vlan_rx_kill_vid = bond_vlan_rx_kill_vid;
4463         bond_dev->features |= (NETIF_F_HW_VLAN_TX |
4464                                NETIF_F_HW_VLAN_RX |
4465                                NETIF_F_HW_VLAN_FILTER);
4466
4467 #ifdef CONFIG_PROC_FS
4468         bond_create_proc_entry(bond);
4469 #endif
4470         list_add_tail(&bond->bond_list, &bond_dev_list);
4471
4472         return 0;
4473 }
4474
4475 /* De-initialize device specific data.
4476  * Caller must hold rtnl_lock.
4477  */
4478 static void bond_deinit(struct net_device *bond_dev)
4479 {
4480         struct bonding *bond = bond_dev->priv;
4481
4482         list_del(&bond->bond_list);
4483
4484 #ifdef CONFIG_PROC_FS
4485         bond_remove_proc_entry(bond);
4486 #endif
4487 }
4488
4489 static void bond_work_cancel_all(struct bonding *bond)
4490 {
4491         write_lock_bh(&bond->lock);
4492         bond->kill_timers = 1;
4493         write_unlock_bh(&bond->lock);
4494
4495         if (bond->params.miimon && delayed_work_pending(&bond->mii_work))
4496                 cancel_delayed_work(&bond->mii_work);
4497
4498         if (bond->params.arp_interval && delayed_work_pending(&bond->arp_work))
4499                 cancel_delayed_work(&bond->arp_work);
4500
4501         if (bond->params.mode == BOND_MODE_ALB &&
4502             delayed_work_pending(&bond->alb_work))
4503                 cancel_delayed_work(&bond->alb_work);
4504
4505         if (bond->params.mode == BOND_MODE_8023AD &&
4506             delayed_work_pending(&bond->ad_work))
4507                 cancel_delayed_work(&bond->ad_work);
4508 }
4509
4510 /* Unregister and free all bond devices.
4511  * Caller must hold rtnl_lock.
4512  */
4513 static void bond_free_all(void)
4514 {
4515         struct bonding *bond, *nxt;
4516
4517         list_for_each_entry_safe(bond, nxt, &bond_dev_list, bond_list) {
4518                 struct net_device *bond_dev = bond->dev;
4519
4520                 bond_work_cancel_all(bond);
4521                 netif_tx_lock_bh(bond_dev);
4522                 bond_mc_list_destroy(bond);
4523                 netif_tx_unlock_bh(bond_dev);
4524                 /* Release the bonded slaves */
4525                 bond_release_all(bond_dev);
4526                 bond_destroy(bond);
4527         }
4528
4529 #ifdef CONFIG_PROC_FS
4530         bond_destroy_proc_dir();
4531 #endif
4532 }
4533
4534 /*------------------------- Module initialization ---------------------------*/
4535
4536 /*
4537  * Convert string input module parms.  Accept either the
4538  * number of the mode or its string name.  A bit complicated because
4539  * some mode names are substrings of other names, and calls from sysfs
4540  * may have whitespace in the name (trailing newlines, for example).
4541  */
4542 int bond_parse_parm(const char *buf, struct bond_parm_tbl *tbl)
4543 {
4544         int mode = -1, i, rv;
4545         char *p, modestr[BOND_MAX_MODENAME_LEN + 1] = { 0, };
4546
4547         for (p = (char *)buf; *p; p++)
4548                 if (!(isdigit(*p) || isspace(*p)))
4549                         break;
4550
4551         if (*p)
4552                 rv = sscanf(buf, "%20s", modestr);
4553         else
4554                 rv = sscanf(buf, "%d", &mode);
4555
4556         if (!rv)
4557                 return -1;
4558
4559         for (i = 0; tbl[i].modename; i++) {
4560                 if (mode == tbl[i].mode)
4561                         return tbl[i].mode;
4562                 if (strcmp(modestr, tbl[i].modename) == 0)
4563                         return tbl[i].mode;
4564         }
4565
4566         return -1;
4567 }
4568
4569 static int bond_check_params(struct bond_params *params)
4570 {
4571         int arp_validate_value;
4572
4573         /*
4574          * Convert string parameters.
4575          */
4576         if (mode) {
4577                 bond_mode = bond_parse_parm(mode, bond_mode_tbl);
4578                 if (bond_mode == -1) {
4579                         printk(KERN_ERR DRV_NAME
4580                                ": Error: Invalid bonding mode \"%s\"\n",
4581                                mode == NULL ? "NULL" : mode);
4582                         return -EINVAL;
4583                 }
4584         }
4585
4586         if (xmit_hash_policy) {
4587                 if ((bond_mode != BOND_MODE_XOR) &&
4588                     (bond_mode != BOND_MODE_8023AD)) {
4589                         printk(KERN_INFO DRV_NAME
4590                                ": xor_mode param is irrelevant in mode %s\n",
4591                                bond_mode_name(bond_mode));
4592                 } else {
4593                         xmit_hashtype = bond_parse_parm(xmit_hash_policy,
4594                                                         xmit_hashtype_tbl);
4595                         if (xmit_hashtype == -1) {
4596                                 printk(KERN_ERR DRV_NAME
4597                                 ": Error: Invalid xmit_hash_policy \"%s\"\n",
4598                                 xmit_hash_policy == NULL ? "NULL" :
4599                                        xmit_hash_policy);
4600                                 return -EINVAL;
4601                         }
4602                 }
4603         }
4604
4605         if (lacp_rate) {
4606                 if (bond_mode != BOND_MODE_8023AD) {
4607                         printk(KERN_INFO DRV_NAME
4608                                ": lacp_rate param is irrelevant in mode %s\n",
4609                                bond_mode_name(bond_mode));
4610                 } else {
4611                         lacp_fast = bond_parse_parm(lacp_rate, bond_lacp_tbl);
4612                         if (lacp_fast == -1) {
4613                                 printk(KERN_ERR DRV_NAME
4614                                        ": Error: Invalid lacp rate \"%s\"\n",
4615                                        lacp_rate == NULL ? "NULL" : lacp_rate);
4616                                 return -EINVAL;
4617                         }
4618                 }
4619         }
4620
4621         if (max_bonds < 1 || max_bonds > INT_MAX) {
4622                 printk(KERN_WARNING DRV_NAME
4623                        ": Warning: max_bonds (%d) not in range %d-%d, so it "
4624                        "was reset to BOND_DEFAULT_MAX_BONDS (%d)\n",
4625                        max_bonds, 1, INT_MAX, BOND_DEFAULT_MAX_BONDS);
4626                 max_bonds = BOND_DEFAULT_MAX_BONDS;
4627         }
4628
4629         if (miimon < 0) {
4630                 printk(KERN_WARNING DRV_NAME
4631                        ": Warning: miimon module parameter (%d), "
4632                        "not in range 0-%d, so it was reset to %d\n",
4633                        miimon, INT_MAX, BOND_LINK_MON_INTERV);
4634                 miimon = BOND_LINK_MON_INTERV;
4635         }
4636
4637         if (updelay < 0) {
4638                 printk(KERN_WARNING DRV_NAME
4639                        ": Warning: updelay module parameter (%d), "
4640                        "not in range 0-%d, so it was reset to 0\n",
4641                        updelay, INT_MAX);
4642                 updelay = 0;
4643         }
4644
4645         if (downdelay < 0) {
4646                 printk(KERN_WARNING DRV_NAME
4647                        ": Warning: downdelay module parameter (%d), "
4648                        "not in range 0-%d, so it was reset to 0\n",
4649                        downdelay, INT_MAX);
4650                 downdelay = 0;
4651         }
4652
4653         if ((use_carrier != 0) && (use_carrier != 1)) {
4654                 printk(KERN_WARNING DRV_NAME
4655                        ": Warning: use_carrier module parameter (%d), "
4656                        "not of valid value (0/1), so it was set to 1\n",
4657                        use_carrier);
4658                 use_carrier = 1;
4659         }
4660
4661         /* reset values for 802.3ad */
4662         if (bond_mode == BOND_MODE_8023AD) {
4663                 if (!miimon) {
4664                         printk(KERN_WARNING DRV_NAME
4665                                ": Warning: miimon must be specified, "
4666                                "otherwise bonding will not detect link "
4667                                "failure, speed and duplex which are "
4668                                "essential for 802.3ad operation\n");
4669                         printk(KERN_WARNING "Forcing miimon to 100msec\n");
4670                         miimon = 100;
4671                 }
4672         }
4673
4674         /* reset values for TLB/ALB */
4675         if ((bond_mode == BOND_MODE_TLB) ||
4676             (bond_mode == BOND_MODE_ALB)) {
4677                 if (!miimon) {
4678                         printk(KERN_WARNING DRV_NAME
4679                                ": Warning: miimon must be specified, "
4680                                "otherwise bonding will not detect link "
4681                                "failure and link speed which are essential "
4682                                "for TLB/ALB load balancing\n");
4683                         printk(KERN_WARNING "Forcing miimon to 100msec\n");
4684                         miimon = 100;
4685                 }
4686         }
4687
4688         if (bond_mode == BOND_MODE_ALB) {
4689                 printk(KERN_NOTICE DRV_NAME
4690                        ": In ALB mode you might experience client "
4691                        "disconnections upon reconnection of a link if the "
4692                        "bonding module updelay parameter (%d msec) is "
4693                        "incompatible with the forwarding delay time of the "
4694                        "switch\n",
4695                        updelay);
4696         }
4697
4698         if (!miimon) {
4699                 if (updelay || downdelay) {
4700                         /* just warn the user the up/down delay will have
4701                          * no effect since miimon is zero...
4702                          */
4703                         printk(KERN_WARNING DRV_NAME
4704                                ": Warning: miimon module parameter not set "
4705                                "and updelay (%d) or downdelay (%d) module "
4706                                "parameter is set; updelay and downdelay have "
4707                                "no effect unless miimon is set\n",
4708                                updelay, downdelay);
4709                 }
4710         } else {
4711                 /* don't allow arp monitoring */
4712                 if (arp_interval) {
4713                         printk(KERN_WARNING DRV_NAME
4714                                ": Warning: miimon (%d) and arp_interval (%d) "
4715                                "can't be used simultaneously, disabling ARP "
4716                                "monitoring\n",
4717                                miimon, arp_interval);
4718                         arp_interval = 0;
4719                 }
4720
4721                 if ((updelay % miimon) != 0) {
4722                         printk(KERN_WARNING DRV_NAME
4723                                ": Warning: updelay (%d) is not a multiple "
4724                                "of miimon (%d), updelay rounded to %d ms\n",
4725                                updelay, miimon, (updelay / miimon) * miimon);
4726                 }
4727
4728                 updelay /= miimon;
4729
4730                 if ((downdelay % miimon) != 0) {
4731                         printk(KERN_WARNING DRV_NAME
4732                                ": Warning: downdelay (%d) is not a multiple "
4733                                "of miimon (%d), downdelay rounded to %d ms\n",
4734                                downdelay, miimon,
4735                                (downdelay / miimon) * miimon);
4736                 }
4737
4738                 downdelay /= miimon;
4739         }
4740
4741         if (arp_interval < 0) {
4742                 printk(KERN_WARNING DRV_NAME
4743                        ": Warning: arp_interval module parameter (%d) "
4744                        ", not in range 0-%d, so it was reset to %d\n",
4745                        arp_interval, INT_MAX, BOND_LINK_ARP_INTERV);
4746                 arp_interval = BOND_LINK_ARP_INTERV;
4747         }
4748
4749         for (arp_ip_count = 0;
4750              (arp_ip_count < BOND_MAX_ARP_TARGETS) && arp_ip_target[arp_ip_count];
4751              arp_ip_count++) {
4752                 /* not complete check, but should be good enough to
4753                    catch mistakes */
4754                 if (!isdigit(arp_ip_target[arp_ip_count][0])) {
4755                         printk(KERN_WARNING DRV_NAME
4756                                ": Warning: bad arp_ip_target module parameter "
4757                                "(%s), ARP monitoring will not be performed\n",
4758                                arp_ip_target[arp_ip_count]);
4759                         arp_interval = 0;
4760                 } else {
4761                         __be32 ip = in_aton(arp_ip_target[arp_ip_count]);
4762                         arp_target[arp_ip_count] = ip;
4763                 }
4764         }
4765
4766         if (arp_interval && !arp_ip_count) {
4767                 /* don't allow arping if no arp_ip_target given... */
4768                 printk(KERN_WARNING DRV_NAME
4769                        ": Warning: arp_interval module parameter (%d) "
4770                        "specified without providing an arp_ip_target "
4771                        "parameter, arp_interval was reset to 0\n",
4772                        arp_interval);
4773                 arp_interval = 0;
4774         }
4775
4776         if (arp_validate) {
4777                 if (bond_mode != BOND_MODE_ACTIVEBACKUP) {
4778                         printk(KERN_ERR DRV_NAME
4779                ": arp_validate only supported in active-backup mode\n");
4780                         return -EINVAL;
4781                 }
4782                 if (!arp_interval) {
4783                         printk(KERN_ERR DRV_NAME
4784                                ": arp_validate requires arp_interval\n");
4785                         return -EINVAL;
4786                 }
4787
4788                 arp_validate_value = bond_parse_parm(arp_validate,
4789                                                      arp_validate_tbl);
4790                 if (arp_validate_value == -1) {
4791                         printk(KERN_ERR DRV_NAME
4792                                ": Error: invalid arp_validate \"%s\"\n",
4793                                arp_validate == NULL ? "NULL" : arp_validate);
4794                         return -EINVAL;
4795                 }
4796         } else
4797                 arp_validate_value = 0;
4798
4799         if (miimon) {
4800                 printk(KERN_INFO DRV_NAME
4801                        ": MII link monitoring set to %d ms\n",
4802                        miimon);
4803         } else if (arp_interval) {
4804                 int i;
4805
4806                 printk(KERN_INFO DRV_NAME
4807                        ": ARP monitoring set to %d ms, validate %s, with %d target(s):",
4808                        arp_interval,
4809                        arp_validate_tbl[arp_validate_value].modename,
4810                        arp_ip_count);
4811
4812                 for (i = 0; i < arp_ip_count; i++)
4813                         printk (" %s", arp_ip_target[i]);
4814
4815                 printk("\n");
4816
4817         } else {
4818                 /* miimon and arp_interval not set, we need one so things
4819                  * work as expected, see bonding.txt for details
4820                  */
4821                 printk(KERN_WARNING DRV_NAME
4822                        ": Warning: either miimon or arp_interval and "
4823                        "arp_ip_target module parameters must be specified, "
4824                        "otherwise bonding will not detect link failures! see "
4825                        "bonding.txt for details.\n");
4826         }
4827
4828         if (primary && !USES_PRIMARY(bond_mode)) {
4829                 /* currently, using a primary only makes sense
4830                  * in active backup, TLB or ALB modes
4831                  */
4832                 printk(KERN_WARNING DRV_NAME
4833                        ": Warning: %s primary device specified but has no "
4834                        "effect in %s mode\n",
4835                        primary, bond_mode_name(bond_mode));
4836                 primary = NULL;
4837         }
4838
4839         if (fail_over_mac && (bond_mode != BOND_MODE_ACTIVEBACKUP))
4840                 printk(KERN_WARNING DRV_NAME
4841                        ": Warning: fail_over_mac only affects "
4842                        "active-backup mode.\n");
4843
4844         /* fill params struct with the proper values */
4845         params->mode = bond_mode;
4846         params->xmit_policy = xmit_hashtype;
4847         params->miimon = miimon;
4848         params->arp_interval = arp_interval;
4849         params->arp_validate = arp_validate_value;
4850         params->updelay = updelay;
4851         params->downdelay = downdelay;
4852         params->use_carrier = use_carrier;
4853         params->lacp_fast = lacp_fast;
4854         params->primary[0] = 0;
4855         params->fail_over_mac = fail_over_mac;
4856
4857         if (primary) {
4858                 strncpy(params->primary, primary, IFNAMSIZ);
4859                 params->primary[IFNAMSIZ - 1] = 0;
4860         }
4861
4862         memcpy(params->arp_targets, arp_target, sizeof(arp_target));
4863
4864         return 0;
4865 }
4866
4867 static struct lock_class_key bonding_netdev_xmit_lock_key;
4868
4869 /* Create a new bond based on the specified name and bonding parameters.
4870  * If name is NULL, obtain a suitable "bond%d" name for us.
4871  * Caller must NOT hold rtnl_lock; we need to release it here before we
4872  * set up our sysfs entries.
4873  */
4874 int bond_create(char *name, struct bond_params *params, struct bonding **newbond)
4875 {
4876         struct net_device *bond_dev;
4877         struct bonding *bond, *nxt;
4878         int res;
4879
4880         rtnl_lock();
4881         down_write(&bonding_rwsem);
4882
4883         /* Check to see if the bond already exists. */
4884         if (name) {
4885                 list_for_each_entry_safe(bond, nxt, &bond_dev_list, bond_list)
4886                         if (strnicmp(bond->dev->name, name, IFNAMSIZ) == 0) {
4887                                 printk(KERN_ERR DRV_NAME
4888                                ": cannot add bond %s; it already exists\n",
4889                                        name);
4890                                 res = -EPERM;
4891                                 goto out_rtnl;
4892                         }
4893         }
4894
4895         bond_dev = alloc_netdev(sizeof(struct bonding), name ? name : "",
4896                                 ether_setup);
4897         if (!bond_dev) {
4898                 printk(KERN_ERR DRV_NAME
4899                        ": %s: eek! can't alloc netdev!\n",
4900                        name);
4901                 res = -ENOMEM;
4902                 goto out_rtnl;
4903         }
4904
4905         if (!name) {
4906                 res = dev_alloc_name(bond_dev, "bond%d");
4907                 if (res < 0)
4908                         goto out_netdev;
4909         }
4910
4911         /* bond_init() must be called after dev_alloc_name() (for the
4912          * /proc files), but before register_netdevice(), because we
4913          * need to set function pointers.
4914          */
4915
4916         res = bond_init(bond_dev, params);
4917         if (res < 0) {
4918                 goto out_netdev;
4919         }
4920
4921         res = register_netdevice(bond_dev);
4922         if (res < 0) {
4923                 goto out_bond;
4924         }
4925
4926         lockdep_set_class(&bond_dev->_xmit_lock, &bonding_netdev_xmit_lock_key);
4927
4928         if (newbond)
4929                 *newbond = bond_dev->priv;
4930
4931         netif_carrier_off(bond_dev);
4932
4933         up_write(&bonding_rwsem);
4934         rtnl_unlock(); /* allows sysfs registration of net device */
4935         res = bond_create_sysfs_entry(bond_dev->priv);
4936         if (res < 0) {
4937                 rtnl_lock();
4938                 down_write(&bonding_rwsem);
4939                 bond_deinit(bond_dev);
4940                 unregister_netdevice(bond_dev);
4941                 goto out_rtnl;
4942         }
4943
4944         return 0;
4945
4946 out_bond:
4947         bond_deinit(bond_dev);
4948 out_netdev:
4949         free_netdev(bond_dev);
4950 out_rtnl:
4951         up_write(&bonding_rwsem);
4952         rtnl_unlock();
4953         return res;
4954 }
4955
4956 static int __init bonding_init(void)
4957 {
4958         int i;
4959         int res;
4960         struct bonding *bond, *nxt;
4961
4962         printk(KERN_INFO "%s", version);
4963
4964         res = bond_check_params(&bonding_defaults);
4965         if (res) {
4966                 goto out;
4967         }
4968
4969 #ifdef CONFIG_PROC_FS
4970         bond_create_proc_dir();
4971 #endif
4972
4973         init_rwsem(&bonding_rwsem);
4974
4975         for (i = 0; i < max_bonds; i++) {
4976                 res = bond_create(NULL, &bonding_defaults, NULL);
4977                 if (res)
4978                         goto err;
4979         }
4980
4981         res = bond_create_sysfs();
4982         if (res)
4983                 goto err;
4984
4985         register_netdevice_notifier(&bond_netdev_notifier);
4986         register_inetaddr_notifier(&bond_inetaddr_notifier);
4987
4988         goto out;
4989 err:
4990         list_for_each_entry_safe(bond, nxt, &bond_dev_list, bond_list) {
4991                 bond_work_cancel_all(bond);
4992                 destroy_workqueue(bond->wq);
4993         }
4994
4995         bond_destroy_sysfs();
4996
4997         rtnl_lock();
4998         bond_free_all();
4999         rtnl_unlock();
5000 out:
5001         return res;
5002
5003 }
5004
5005 static void __exit bonding_exit(void)
5006 {
5007         unregister_netdevice_notifier(&bond_netdev_notifier);
5008         unregister_inetaddr_notifier(&bond_inetaddr_notifier);
5009
5010         bond_destroy_sysfs();
5011
5012         rtnl_lock();
5013         bond_free_all();
5014         rtnl_unlock();
5015 }
5016
5017 module_init(bonding_init);
5018 module_exit(bonding_exit);
5019 MODULE_LICENSE("GPL");
5020 MODULE_VERSION(DRV_VERSION);
5021 MODULE_DESCRIPTION(DRV_DESCRIPTION ", v" DRV_VERSION);
5022 MODULE_AUTHOR("Thomas Davis, tadavis@lbl.gov and many others");
5023 MODULE_SUPPORTED_DEVICE("most ethernet devices");
5024
5025 /*
5026  * Local variables:
5027  *  c-indent-level: 8
5028  *  c-basic-offset: 8
5029  *  tab-width: 8
5030  * End:
5031  */
5032