bonding: deliver netdev event for fail-over under the active-backup mode
[linux-2.6.git] / drivers / net / bonding / bond_main.c
1 /*
2  * originally based on the dummy device.
3  *
4  * Copyright 1999, Thomas Davis, tadavis@lbl.gov.
5  * Licensed under the GPL. Based on dummy.c, and eql.c devices.
6  *
7  * bonding.c: an Ethernet Bonding driver
8  *
9  * This is useful to talk to a Cisco EtherChannel compatible equipment:
10  *      Cisco 5500
11  *      Sun Trunking (Solaris)
12  *      Alteon AceDirector Trunks
13  *      Linux Bonding
14  *      and probably many L2 switches ...
15  *
16  * How it works:
17  *    ifconfig bond0 ipaddress netmask up
18  *      will setup a network device, with an ip address.  No mac address
19  *      will be assigned at this time.  The hw mac address will come from
20  *      the first slave bonded to the channel.  All slaves will then use
21  *      this hw mac address.
22  *
23  *    ifconfig bond0 down
24  *         will release all slaves, marking them as down.
25  *
26  *    ifenslave bond0 eth0
27  *      will attach eth0 to bond0 as a slave.  eth0 hw mac address will either
28  *      a: be used as initial mac address
29  *      b: if a hw mac address already is there, eth0's hw mac address
30  *         will then be set from bond0.
31  *
32  */
33
34 //#define BONDING_DEBUG 1
35
36 #include <linux/kernel.h>
37 #include <linux/module.h>
38 #include <linux/types.h>
39 #include <linux/fcntl.h>
40 #include <linux/interrupt.h>
41 #include <linux/ptrace.h>
42 #include <linux/ioport.h>
43 #include <linux/in.h>
44 #include <net/ip.h>
45 #include <linux/ip.h>
46 #include <linux/tcp.h>
47 #include <linux/udp.h>
48 #include <linux/slab.h>
49 #include <linux/string.h>
50 #include <linux/init.h>
51 #include <linux/timer.h>
52 #include <linux/socket.h>
53 #include <linux/ctype.h>
54 #include <linux/inet.h>
55 #include <linux/bitops.h>
56 #include <asm/system.h>
57 #include <asm/io.h>
58 #include <asm/dma.h>
59 #include <asm/uaccess.h>
60 #include <linux/errno.h>
61 #include <linux/netdevice.h>
62 #include <linux/inetdevice.h>
63 #include <linux/igmp.h>
64 #include <linux/etherdevice.h>
65 #include <linux/skbuff.h>
66 #include <net/sock.h>
67 #include <linux/rtnetlink.h>
68 #include <linux/proc_fs.h>
69 #include <linux/seq_file.h>
70 #include <linux/smp.h>
71 #include <linux/if_ether.h>
72 #include <net/arp.h>
73 #include <linux/mii.h>
74 #include <linux/ethtool.h>
75 #include <linux/if_vlan.h>
76 #include <linux/if_bonding.h>
77 #include <linux/jiffies.h>
78 #include <net/route.h>
79 #include <net/net_namespace.h>
80 #include "bonding.h"
81 #include "bond_3ad.h"
82 #include "bond_alb.h"
83
84 /*---------------------------- Module parameters ----------------------------*/
85
86 /* monitor all links that often (in milliseconds). <=0 disables monitoring */
87 #define BOND_LINK_MON_INTERV    0
88 #define BOND_LINK_ARP_INTERV    0
89
90 static int max_bonds    = BOND_DEFAULT_MAX_BONDS;
91 static int num_grat_arp = 1;
92 static int miimon       = BOND_LINK_MON_INTERV;
93 static int updelay      = 0;
94 static int downdelay    = 0;
95 static int use_carrier  = 1;
96 static char *mode       = NULL;
97 static char *primary    = NULL;
98 static char *lacp_rate  = NULL;
99 static char *xmit_hash_policy = NULL;
100 static int arp_interval = BOND_LINK_ARP_INTERV;
101 static char *arp_ip_target[BOND_MAX_ARP_TARGETS] = { NULL, };
102 static char *arp_validate = NULL;
103 static char *fail_over_mac = NULL;
104 struct bond_params bonding_defaults;
105
106 module_param(max_bonds, int, 0);
107 MODULE_PARM_DESC(max_bonds, "Max number of bonded devices");
108 module_param(num_grat_arp, int, 0644);
109 MODULE_PARM_DESC(num_grat_arp, "Number of gratuitous ARP packets to send on failover event");
110 module_param(miimon, int, 0);
111 MODULE_PARM_DESC(miimon, "Link check interval in milliseconds");
112 module_param(updelay, int, 0);
113 MODULE_PARM_DESC(updelay, "Delay before considering link up, in milliseconds");
114 module_param(downdelay, int, 0);
115 MODULE_PARM_DESC(downdelay, "Delay before considering link down, "
116                             "in milliseconds");
117 module_param(use_carrier, int, 0);
118 MODULE_PARM_DESC(use_carrier, "Use netif_carrier_ok (vs MII ioctls) in miimon; "
119                               "0 for off, 1 for on (default)");
120 module_param(mode, charp, 0);
121 MODULE_PARM_DESC(mode, "Mode of operation : 0 for balance-rr, "
122                        "1 for active-backup, 2 for balance-xor, "
123                        "3 for broadcast, 4 for 802.3ad, 5 for balance-tlb, "
124                        "6 for balance-alb");
125 module_param(primary, charp, 0);
126 MODULE_PARM_DESC(primary, "Primary network device to use");
127 module_param(lacp_rate, charp, 0);
128 MODULE_PARM_DESC(lacp_rate, "LACPDU tx rate to request from 802.3ad partner "
129                             "(slow/fast)");
130 module_param(xmit_hash_policy, charp, 0);
131 MODULE_PARM_DESC(xmit_hash_policy, "XOR hashing method: 0 for layer 2 (default)"
132                                    ", 1 for layer 3+4");
133 module_param(arp_interval, int, 0);
134 MODULE_PARM_DESC(arp_interval, "arp interval in milliseconds");
135 module_param_array(arp_ip_target, charp, NULL, 0);
136 MODULE_PARM_DESC(arp_ip_target, "arp targets in n.n.n.n form");
137 module_param(arp_validate, charp, 0);
138 MODULE_PARM_DESC(arp_validate, "validate src/dst of ARP probes: none (default), active, backup or all");
139 module_param(fail_over_mac, charp, 0);
140 MODULE_PARM_DESC(fail_over_mac, "For active-backup, do not set all slaves to the same MAC.  none (default), active or follow");
141
142 /*----------------------------- Global variables ----------------------------*/
143
144 static const char * const version =
145         DRV_DESCRIPTION ": v" DRV_VERSION " (" DRV_RELDATE ")\n";
146
147 LIST_HEAD(bond_dev_list);
148
149 #ifdef CONFIG_PROC_FS
150 static struct proc_dir_entry *bond_proc_dir = NULL;
151 #endif
152
153 extern struct rw_semaphore bonding_rwsem;
154 static __be32 arp_target[BOND_MAX_ARP_TARGETS] = { 0, } ;
155 static int arp_ip_count = 0;
156 static int bond_mode    = BOND_MODE_ROUNDROBIN;
157 static int xmit_hashtype= BOND_XMIT_POLICY_LAYER2;
158 static int lacp_fast    = 0;
159
160
161 struct bond_parm_tbl bond_lacp_tbl[] = {
162 {       "slow",         AD_LACP_SLOW},
163 {       "fast",         AD_LACP_FAST},
164 {       NULL,           -1},
165 };
166
167 struct bond_parm_tbl bond_mode_tbl[] = {
168 {       "balance-rr",           BOND_MODE_ROUNDROBIN},
169 {       "active-backup",        BOND_MODE_ACTIVEBACKUP},
170 {       "balance-xor",          BOND_MODE_XOR},
171 {       "broadcast",            BOND_MODE_BROADCAST},
172 {       "802.3ad",              BOND_MODE_8023AD},
173 {       "balance-tlb",          BOND_MODE_TLB},
174 {       "balance-alb",          BOND_MODE_ALB},
175 {       NULL,                   -1},
176 };
177
178 struct bond_parm_tbl xmit_hashtype_tbl[] = {
179 {       "layer2",               BOND_XMIT_POLICY_LAYER2},
180 {       "layer3+4",             BOND_XMIT_POLICY_LAYER34},
181 {       "layer2+3",             BOND_XMIT_POLICY_LAYER23},
182 {       NULL,                   -1},
183 };
184
185 struct bond_parm_tbl arp_validate_tbl[] = {
186 {       "none",                 BOND_ARP_VALIDATE_NONE},
187 {       "active",               BOND_ARP_VALIDATE_ACTIVE},
188 {       "backup",               BOND_ARP_VALIDATE_BACKUP},
189 {       "all",                  BOND_ARP_VALIDATE_ALL},
190 {       NULL,                   -1},
191 };
192
193 struct bond_parm_tbl fail_over_mac_tbl[] = {
194 {       "none",                 BOND_FOM_NONE},
195 {       "active",               BOND_FOM_ACTIVE},
196 {       "follow",               BOND_FOM_FOLLOW},
197 {       NULL,                   -1},
198 };
199
200 /*-------------------------- Forward declarations ---------------------------*/
201
202 static void bond_send_gratuitous_arp(struct bonding *bond);
203 static void bond_deinit(struct net_device *bond_dev);
204
205 /*---------------------------- General routines -----------------------------*/
206
207 static const char *bond_mode_name(int mode)
208 {
209         switch (mode) {
210         case BOND_MODE_ROUNDROBIN :
211                 return "load balancing (round-robin)";
212         case BOND_MODE_ACTIVEBACKUP :
213                 return "fault-tolerance (active-backup)";
214         case BOND_MODE_XOR :
215                 return "load balancing (xor)";
216         case BOND_MODE_BROADCAST :
217                 return "fault-tolerance (broadcast)";
218         case BOND_MODE_8023AD:
219                 return "IEEE 802.3ad Dynamic link aggregation";
220         case BOND_MODE_TLB:
221                 return "transmit load balancing";
222         case BOND_MODE_ALB:
223                 return "adaptive load balancing";
224         default:
225                 return "unknown";
226         }
227 }
228
229 /*---------------------------------- VLAN -----------------------------------*/
230
231 /**
232  * bond_add_vlan - add a new vlan id on bond
233  * @bond: bond that got the notification
234  * @vlan_id: the vlan id to add
235  *
236  * Returns -ENOMEM if allocation failed.
237  */
238 static int bond_add_vlan(struct bonding *bond, unsigned short vlan_id)
239 {
240         struct vlan_entry *vlan;
241
242         dprintk("bond: %s, vlan id %d\n",
243                 (bond ? bond->dev->name: "None"), vlan_id);
244
245         vlan = kmalloc(sizeof(struct vlan_entry), GFP_KERNEL);
246         if (!vlan) {
247                 return -ENOMEM;
248         }
249
250         INIT_LIST_HEAD(&vlan->vlan_list);
251         vlan->vlan_id = vlan_id;
252         vlan->vlan_ip = 0;
253
254         write_lock_bh(&bond->lock);
255
256         list_add_tail(&vlan->vlan_list, &bond->vlan_list);
257
258         write_unlock_bh(&bond->lock);
259
260         dprintk("added VLAN ID %d on bond %s\n", vlan_id, bond->dev->name);
261
262         return 0;
263 }
264
265 /**
266  * bond_del_vlan - delete a vlan id from bond
267  * @bond: bond that got the notification
268  * @vlan_id: the vlan id to delete
269  *
270  * returns -ENODEV if @vlan_id was not found in @bond.
271  */
272 static int bond_del_vlan(struct bonding *bond, unsigned short vlan_id)
273 {
274         struct vlan_entry *vlan;
275         int res = -ENODEV;
276
277         dprintk("bond: %s, vlan id %d\n", bond->dev->name, vlan_id);
278
279         write_lock_bh(&bond->lock);
280
281         list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
282                 if (vlan->vlan_id == vlan_id) {
283                         list_del(&vlan->vlan_list);
284
285                         if ((bond->params.mode == BOND_MODE_TLB) ||
286                             (bond->params.mode == BOND_MODE_ALB)) {
287                                 bond_alb_clear_vlan(bond, vlan_id);
288                         }
289
290                         dprintk("removed VLAN ID %d from bond %s\n", vlan_id,
291                                 bond->dev->name);
292
293                         kfree(vlan);
294
295                         if (list_empty(&bond->vlan_list) &&
296                             (bond->slave_cnt == 0)) {
297                                 /* Last VLAN removed and no slaves, so
298                                  * restore block on adding VLANs. This will
299                                  * be removed once new slaves that are not
300                                  * VLAN challenged will be added.
301                                  */
302                                 bond->dev->features |= NETIF_F_VLAN_CHALLENGED;
303                         }
304
305                         res = 0;
306                         goto out;
307                 }
308         }
309
310         dprintk("couldn't find VLAN ID %d in bond %s\n", vlan_id,
311                 bond->dev->name);
312
313 out:
314         write_unlock_bh(&bond->lock);
315         return res;
316 }
317
318 /**
319  * bond_has_challenged_slaves
320  * @bond: the bond we're working on
321  *
322  * Searches the slave list. Returns 1 if a vlan challenged slave
323  * was found, 0 otherwise.
324  *
325  * Assumes bond->lock is held.
326  */
327 static int bond_has_challenged_slaves(struct bonding *bond)
328 {
329         struct slave *slave;
330         int i;
331
332         bond_for_each_slave(bond, slave, i) {
333                 if (slave->dev->features & NETIF_F_VLAN_CHALLENGED) {
334                         dprintk("found VLAN challenged slave - %s\n",
335                                 slave->dev->name);
336                         return 1;
337                 }
338         }
339
340         dprintk("no VLAN challenged slaves found\n");
341         return 0;
342 }
343
344 /**
345  * bond_next_vlan - safely skip to the next item in the vlans list.
346  * @bond: the bond we're working on
347  * @curr: item we're advancing from
348  *
349  * Returns %NULL if list is empty, bond->next_vlan if @curr is %NULL,
350  * or @curr->next otherwise (even if it is @curr itself again).
351  * 
352  * Caller must hold bond->lock
353  */
354 struct vlan_entry *bond_next_vlan(struct bonding *bond, struct vlan_entry *curr)
355 {
356         struct vlan_entry *next, *last;
357
358         if (list_empty(&bond->vlan_list)) {
359                 return NULL;
360         }
361
362         if (!curr) {
363                 next = list_entry(bond->vlan_list.next,
364                                   struct vlan_entry, vlan_list);
365         } else {
366                 last = list_entry(bond->vlan_list.prev,
367                                   struct vlan_entry, vlan_list);
368                 if (last == curr) {
369                         next = list_entry(bond->vlan_list.next,
370                                           struct vlan_entry, vlan_list);
371                 } else {
372                         next = list_entry(curr->vlan_list.next,
373                                           struct vlan_entry, vlan_list);
374                 }
375         }
376
377         return next;
378 }
379
380 /**
381  * bond_dev_queue_xmit - Prepare skb for xmit.
382  * 
383  * @bond: bond device that got this skb for tx.
384  * @skb: hw accel VLAN tagged skb to transmit
385  * @slave_dev: slave that is supposed to xmit this skbuff
386  * 
387  * When the bond gets an skb to transmit that is
388  * already hardware accelerated VLAN tagged, and it
389  * needs to relay this skb to a slave that is not
390  * hw accel capable, the skb needs to be "unaccelerated",
391  * i.e. strip the hwaccel tag and re-insert it as part
392  * of the payload.
393  */
394 int bond_dev_queue_xmit(struct bonding *bond, struct sk_buff *skb, struct net_device *slave_dev)
395 {
396         unsigned short uninitialized_var(vlan_id);
397
398         if (!list_empty(&bond->vlan_list) &&
399             !(slave_dev->features & NETIF_F_HW_VLAN_TX) &&
400             vlan_get_tag(skb, &vlan_id) == 0) {
401                 skb->dev = slave_dev;
402                 skb = vlan_put_tag(skb, vlan_id);
403                 if (!skb) {
404                         /* vlan_put_tag() frees the skb in case of error,
405                          * so return success here so the calling functions
406                          * won't attempt to free is again.
407                          */
408                         return 0;
409                 }
410         } else {
411                 skb->dev = slave_dev;
412         }
413
414         skb->priority = 1;
415         dev_queue_xmit(skb);
416
417         return 0;
418 }
419
420 /*
421  * In the following 3 functions, bond_vlan_rx_register(), bond_vlan_rx_add_vid
422  * and bond_vlan_rx_kill_vid, We don't protect the slave list iteration with a
423  * lock because:
424  * a. This operation is performed in IOCTL context,
425  * b. The operation is protected by the RTNL semaphore in the 8021q code,
426  * c. Holding a lock with BH disabled while directly calling a base driver
427  *    entry point is generally a BAD idea.
428  * 
429  * The design of synchronization/protection for this operation in the 8021q
430  * module is good for one or more VLAN devices over a single physical device
431  * and cannot be extended for a teaming solution like bonding, so there is a
432  * potential race condition here where a net device from the vlan group might
433  * be referenced (either by a base driver or the 8021q code) while it is being
434  * removed from the system. However, it turns out we're not making matters
435  * worse, and if it works for regular VLAN usage it will work here too.
436 */
437
438 /**
439  * bond_vlan_rx_register - Propagates registration to slaves
440  * @bond_dev: bonding net device that got called
441  * @grp: vlan group being registered
442  */
443 static void bond_vlan_rx_register(struct net_device *bond_dev, struct vlan_group *grp)
444 {
445         struct bonding *bond = bond_dev->priv;
446         struct slave *slave;
447         int i;
448
449         bond->vlgrp = grp;
450
451         bond_for_each_slave(bond, slave, i) {
452                 struct net_device *slave_dev = slave->dev;
453
454                 if ((slave_dev->features & NETIF_F_HW_VLAN_RX) &&
455                     slave_dev->vlan_rx_register) {
456                         slave_dev->vlan_rx_register(slave_dev, grp);
457                 }
458         }
459 }
460
461 /**
462  * bond_vlan_rx_add_vid - Propagates adding an id to slaves
463  * @bond_dev: bonding net device that got called
464  * @vid: vlan id being added
465  */
466 static void bond_vlan_rx_add_vid(struct net_device *bond_dev, uint16_t vid)
467 {
468         struct bonding *bond = bond_dev->priv;
469         struct slave *slave;
470         int i, res;
471
472         bond_for_each_slave(bond, slave, i) {
473                 struct net_device *slave_dev = slave->dev;
474
475                 if ((slave_dev->features & NETIF_F_HW_VLAN_FILTER) &&
476                     slave_dev->vlan_rx_add_vid) {
477                         slave_dev->vlan_rx_add_vid(slave_dev, vid);
478                 }
479         }
480
481         res = bond_add_vlan(bond, vid);
482         if (res) {
483                 printk(KERN_ERR DRV_NAME
484                        ": %s: Error: Failed to add vlan id %d\n",
485                        bond_dev->name, vid);
486         }
487 }
488
489 /**
490  * bond_vlan_rx_kill_vid - Propagates deleting an id to slaves
491  * @bond_dev: bonding net device that got called
492  * @vid: vlan id being removed
493  */
494 static void bond_vlan_rx_kill_vid(struct net_device *bond_dev, uint16_t vid)
495 {
496         struct bonding *bond = bond_dev->priv;
497         struct slave *slave;
498         struct net_device *vlan_dev;
499         int i, res;
500
501         bond_for_each_slave(bond, slave, i) {
502                 struct net_device *slave_dev = slave->dev;
503
504                 if ((slave_dev->features & NETIF_F_HW_VLAN_FILTER) &&
505                     slave_dev->vlan_rx_kill_vid) {
506                         /* Save and then restore vlan_dev in the grp array,
507                          * since the slave's driver might clear it.
508                          */
509                         vlan_dev = vlan_group_get_device(bond->vlgrp, vid);
510                         slave_dev->vlan_rx_kill_vid(slave_dev, vid);
511                         vlan_group_set_device(bond->vlgrp, vid, vlan_dev);
512                 }
513         }
514
515         res = bond_del_vlan(bond, vid);
516         if (res) {
517                 printk(KERN_ERR DRV_NAME
518                        ": %s: Error: Failed to remove vlan id %d\n",
519                        bond_dev->name, vid);
520         }
521 }
522
523 static void bond_add_vlans_on_slave(struct bonding *bond, struct net_device *slave_dev)
524 {
525         struct vlan_entry *vlan;
526
527         write_lock_bh(&bond->lock);
528
529         if (list_empty(&bond->vlan_list)) {
530                 goto out;
531         }
532
533         if ((slave_dev->features & NETIF_F_HW_VLAN_RX) &&
534             slave_dev->vlan_rx_register) {
535                 slave_dev->vlan_rx_register(slave_dev, bond->vlgrp);
536         }
537
538         if (!(slave_dev->features & NETIF_F_HW_VLAN_FILTER) ||
539             !(slave_dev->vlan_rx_add_vid)) {
540                 goto out;
541         }
542
543         list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
544                 slave_dev->vlan_rx_add_vid(slave_dev, vlan->vlan_id);
545         }
546
547 out:
548         write_unlock_bh(&bond->lock);
549 }
550
551 static void bond_del_vlans_from_slave(struct bonding *bond, struct net_device *slave_dev)
552 {
553         struct vlan_entry *vlan;
554         struct net_device *vlan_dev;
555
556         write_lock_bh(&bond->lock);
557
558         if (list_empty(&bond->vlan_list)) {
559                 goto out;
560         }
561
562         if (!(slave_dev->features & NETIF_F_HW_VLAN_FILTER) ||
563             !(slave_dev->vlan_rx_kill_vid)) {
564                 goto unreg;
565         }
566
567         list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
568                 /* Save and then restore vlan_dev in the grp array,
569                  * since the slave's driver might clear it.
570                  */
571                 vlan_dev = vlan_group_get_device(bond->vlgrp, vlan->vlan_id);
572                 slave_dev->vlan_rx_kill_vid(slave_dev, vlan->vlan_id);
573                 vlan_group_set_device(bond->vlgrp, vlan->vlan_id, vlan_dev);
574         }
575
576 unreg:
577         if ((slave_dev->features & NETIF_F_HW_VLAN_RX) &&
578             slave_dev->vlan_rx_register) {
579                 slave_dev->vlan_rx_register(slave_dev, NULL);
580         }
581
582 out:
583         write_unlock_bh(&bond->lock);
584 }
585
586 /*------------------------------- Link status -------------------------------*/
587
588 /*
589  * Set the carrier state for the master according to the state of its
590  * slaves.  If any slaves are up, the master is up.  In 802.3ad mode,
591  * do special 802.3ad magic.
592  *
593  * Returns zero if carrier state does not change, nonzero if it does.
594  */
595 static int bond_set_carrier(struct bonding *bond)
596 {
597         struct slave *slave;
598         int i;
599
600         if (bond->slave_cnt == 0)
601                 goto down;
602
603         if (bond->params.mode == BOND_MODE_8023AD)
604                 return bond_3ad_set_carrier(bond);
605
606         bond_for_each_slave(bond, slave, i) {
607                 if (slave->link == BOND_LINK_UP) {
608                         if (!netif_carrier_ok(bond->dev)) {
609                                 netif_carrier_on(bond->dev);
610                                 return 1;
611                         }
612                         return 0;
613                 }
614         }
615
616 down:
617         if (netif_carrier_ok(bond->dev)) {
618                 netif_carrier_off(bond->dev);
619                 return 1;
620         }
621         return 0;
622 }
623
624 /*
625  * Get link speed and duplex from the slave's base driver
626  * using ethtool. If for some reason the call fails or the
627  * values are invalid, fake speed and duplex to 100/Full
628  * and return error.
629  */
630 static int bond_update_speed_duplex(struct slave *slave)
631 {
632         struct net_device *slave_dev = slave->dev;
633         struct ethtool_cmd etool;
634         int res;
635
636         /* Fake speed and duplex */
637         slave->speed = SPEED_100;
638         slave->duplex = DUPLEX_FULL;
639
640         if (!slave_dev->ethtool_ops || !slave_dev->ethtool_ops->get_settings)
641                 return -1;
642
643         res = slave_dev->ethtool_ops->get_settings(slave_dev, &etool);
644         if (res < 0)
645                 return -1;
646
647         switch (etool.speed) {
648         case SPEED_10:
649         case SPEED_100:
650         case SPEED_1000:
651         case SPEED_10000:
652                 break;
653         default:
654                 return -1;
655         }
656
657         switch (etool.duplex) {
658         case DUPLEX_FULL:
659         case DUPLEX_HALF:
660                 break;
661         default:
662                 return -1;
663         }
664
665         slave->speed = etool.speed;
666         slave->duplex = etool.duplex;
667
668         return 0;
669 }
670
671 /*
672  * if <dev> supports MII link status reporting, check its link status.
673  *
674  * We either do MII/ETHTOOL ioctls, or check netif_carrier_ok(),
675  * depening upon the setting of the use_carrier parameter.
676  *
677  * Return either BMSR_LSTATUS, meaning that the link is up (or we
678  * can't tell and just pretend it is), or 0, meaning that the link is
679  * down.
680  *
681  * If reporting is non-zero, instead of faking link up, return -1 if
682  * both ETHTOOL and MII ioctls fail (meaning the device does not
683  * support them).  If use_carrier is set, return whatever it says.
684  * It'd be nice if there was a good way to tell if a driver supports
685  * netif_carrier, but there really isn't.
686  */
687 static int bond_check_dev_link(struct bonding *bond, struct net_device *slave_dev, int reporting)
688 {
689         static int (* ioctl)(struct net_device *, struct ifreq *, int);
690         struct ifreq ifr;
691         struct mii_ioctl_data *mii;
692
693         if (bond->params.use_carrier) {
694                 return netif_carrier_ok(slave_dev) ? BMSR_LSTATUS : 0;
695         }
696
697         ioctl = slave_dev->do_ioctl;
698         if (ioctl) {
699                 /* TODO: set pointer to correct ioctl on a per team member */
700                 /*       bases to make this more efficient. that is, once  */
701                 /*       we determine the correct ioctl, we will always    */
702                 /*       call it and not the others for that team          */
703                 /*       member.                                           */
704
705                 /*
706                  * We cannot assume that SIOCGMIIPHY will also read a
707                  * register; not all network drivers (e.g., e100)
708                  * support that.
709                  */
710
711                 /* Yes, the mii is overlaid on the ifreq.ifr_ifru */
712                 strncpy(ifr.ifr_name, slave_dev->name, IFNAMSIZ);
713                 mii = if_mii(&ifr);
714                 if (IOCTL(slave_dev, &ifr, SIOCGMIIPHY) == 0) {
715                         mii->reg_num = MII_BMSR;
716                         if (IOCTL(slave_dev, &ifr, SIOCGMIIREG) == 0) {
717                                 return (mii->val_out & BMSR_LSTATUS);
718                         }
719                 }
720         }
721
722         /*
723          * Some drivers cache ETHTOOL_GLINK for a period of time so we only
724          * attempt to get link status from it if the above MII ioctls fail.
725          */
726         if (slave_dev->ethtool_ops) {
727                 if (slave_dev->ethtool_ops->get_link) {
728                         u32 link;
729
730                         link = slave_dev->ethtool_ops->get_link(slave_dev);
731
732                         return link ? BMSR_LSTATUS : 0;
733                 }
734         }
735
736         /*
737          * If reporting, report that either there's no dev->do_ioctl,
738          * or both SIOCGMIIREG and get_link failed (meaning that we
739          * cannot report link status).  If not reporting, pretend
740          * we're ok.
741          */
742         return (reporting ? -1 : BMSR_LSTATUS);
743 }
744
745 /*----------------------------- Multicast list ------------------------------*/
746
747 /*
748  * Returns 0 if dmi1 and dmi2 are the same, non-0 otherwise
749  */
750 static inline int bond_is_dmi_same(struct dev_mc_list *dmi1, struct dev_mc_list *dmi2)
751 {
752         return memcmp(dmi1->dmi_addr, dmi2->dmi_addr, dmi1->dmi_addrlen) == 0 &&
753                         dmi1->dmi_addrlen == dmi2->dmi_addrlen;
754 }
755
756 /*
757  * returns dmi entry if found, NULL otherwise
758  */
759 static struct dev_mc_list *bond_mc_list_find_dmi(struct dev_mc_list *dmi, struct dev_mc_list *mc_list)
760 {
761         struct dev_mc_list *idmi;
762
763         for (idmi = mc_list; idmi; idmi = idmi->next) {
764                 if (bond_is_dmi_same(dmi, idmi)) {
765                         return idmi;
766                 }
767         }
768
769         return NULL;
770 }
771
772 /*
773  * Push the promiscuity flag down to appropriate slaves
774  */
775 static void bond_set_promiscuity(struct bonding *bond, int inc)
776 {
777         if (USES_PRIMARY(bond->params.mode)) {
778                 /* write lock already acquired */
779                 if (bond->curr_active_slave) {
780                         dev_set_promiscuity(bond->curr_active_slave->dev, inc);
781                 }
782         } else {
783                 struct slave *slave;
784                 int i;
785                 bond_for_each_slave(bond, slave, i) {
786                         dev_set_promiscuity(slave->dev, inc);
787                 }
788         }
789 }
790
791 /*
792  * Push the allmulti flag down to all slaves
793  */
794 static void bond_set_allmulti(struct bonding *bond, int inc)
795 {
796         if (USES_PRIMARY(bond->params.mode)) {
797                 /* write lock already acquired */
798                 if (bond->curr_active_slave) {
799                         dev_set_allmulti(bond->curr_active_slave->dev, inc);
800                 }
801         } else {
802                 struct slave *slave;
803                 int i;
804                 bond_for_each_slave(bond, slave, i) {
805                         dev_set_allmulti(slave->dev, inc);
806                 }
807         }
808 }
809
810 /*
811  * Add a Multicast address to slaves
812  * according to mode
813  */
814 static void bond_mc_add(struct bonding *bond, void *addr, int alen)
815 {
816         if (USES_PRIMARY(bond->params.mode)) {
817                 /* write lock already acquired */
818                 if (bond->curr_active_slave) {
819                         dev_mc_add(bond->curr_active_slave->dev, addr, alen, 0);
820                 }
821         } else {
822                 struct slave *slave;
823                 int i;
824                 bond_for_each_slave(bond, slave, i) {
825                         dev_mc_add(slave->dev, addr, alen, 0);
826                 }
827         }
828 }
829
830 /*
831  * Remove a multicast address from slave
832  * according to mode
833  */
834 static void bond_mc_delete(struct bonding *bond, void *addr, int alen)
835 {
836         if (USES_PRIMARY(bond->params.mode)) {
837                 /* write lock already acquired */
838                 if (bond->curr_active_slave) {
839                         dev_mc_delete(bond->curr_active_slave->dev, addr, alen, 0);
840                 }
841         } else {
842                 struct slave *slave;
843                 int i;
844                 bond_for_each_slave(bond, slave, i) {
845                         dev_mc_delete(slave->dev, addr, alen, 0);
846                 }
847         }
848 }
849
850
851 /*
852  * Retrieve the list of registered multicast addresses for the bonding
853  * device and retransmit an IGMP JOIN request to the current active
854  * slave.
855  */
856 static void bond_resend_igmp_join_requests(struct bonding *bond)
857 {
858         struct in_device *in_dev;
859         struct ip_mc_list *im;
860
861         rcu_read_lock();
862         in_dev = __in_dev_get_rcu(bond->dev);
863         if (in_dev) {
864                 for (im = in_dev->mc_list; im; im = im->next) {
865                         ip_mc_rejoin_group(im);
866                 }
867         }
868
869         rcu_read_unlock();
870 }
871
872 /*
873  * Totally destroys the mc_list in bond
874  */
875 static void bond_mc_list_destroy(struct bonding *bond)
876 {
877         struct dev_mc_list *dmi;
878
879         dmi = bond->mc_list;
880         while (dmi) {
881                 bond->mc_list = dmi->next;
882                 kfree(dmi);
883                 dmi = bond->mc_list;
884         }
885         bond->mc_list = NULL;
886 }
887
888 /*
889  * Copy all the Multicast addresses from src to the bonding device dst
890  */
891 static int bond_mc_list_copy(struct dev_mc_list *mc_list, struct bonding *bond,
892                              gfp_t gfp_flag)
893 {
894         struct dev_mc_list *dmi, *new_dmi;
895
896         for (dmi = mc_list; dmi; dmi = dmi->next) {
897                 new_dmi = kmalloc(sizeof(struct dev_mc_list), gfp_flag);
898
899                 if (!new_dmi) {
900                         /* FIXME: Potential memory leak !!! */
901                         return -ENOMEM;
902                 }
903
904                 new_dmi->next = bond->mc_list;
905                 bond->mc_list = new_dmi;
906                 new_dmi->dmi_addrlen = dmi->dmi_addrlen;
907                 memcpy(new_dmi->dmi_addr, dmi->dmi_addr, dmi->dmi_addrlen);
908                 new_dmi->dmi_users = dmi->dmi_users;
909                 new_dmi->dmi_gusers = dmi->dmi_gusers;
910         }
911
912         return 0;
913 }
914
915 /*
916  * flush all members of flush->mc_list from device dev->mc_list
917  */
918 static void bond_mc_list_flush(struct net_device *bond_dev, struct net_device *slave_dev)
919 {
920         struct bonding *bond = bond_dev->priv;
921         struct dev_mc_list *dmi;
922
923         for (dmi = bond_dev->mc_list; dmi; dmi = dmi->next) {
924                 dev_mc_delete(slave_dev, dmi->dmi_addr, dmi->dmi_addrlen, 0);
925         }
926
927         if (bond->params.mode == BOND_MODE_8023AD) {
928                 /* del lacpdu mc addr from mc list */
929                 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
930
931                 dev_mc_delete(slave_dev, lacpdu_multicast, ETH_ALEN, 0);
932         }
933 }
934
935 /*--------------------------- Active slave change ---------------------------*/
936
937 /*
938  * Update the mc list and multicast-related flags for the new and
939  * old active slaves (if any) according to the multicast mode, and
940  * promiscuous flags unconditionally.
941  */
942 static void bond_mc_swap(struct bonding *bond, struct slave *new_active, struct slave *old_active)
943 {
944         struct dev_mc_list *dmi;
945
946         if (!USES_PRIMARY(bond->params.mode)) {
947                 /* nothing to do -  mc list is already up-to-date on
948                  * all slaves
949                  */
950                 return;
951         }
952
953         if (old_active) {
954                 if (bond->dev->flags & IFF_PROMISC) {
955                         dev_set_promiscuity(old_active->dev, -1);
956                 }
957
958                 if (bond->dev->flags & IFF_ALLMULTI) {
959                         dev_set_allmulti(old_active->dev, -1);
960                 }
961
962                 for (dmi = bond->dev->mc_list; dmi; dmi = dmi->next) {
963                         dev_mc_delete(old_active->dev, dmi->dmi_addr, dmi->dmi_addrlen, 0);
964                 }
965         }
966
967         if (new_active) {
968                 if (bond->dev->flags & IFF_PROMISC) {
969                         dev_set_promiscuity(new_active->dev, 1);
970                 }
971
972                 if (bond->dev->flags & IFF_ALLMULTI) {
973                         dev_set_allmulti(new_active->dev, 1);
974                 }
975
976                 for (dmi = bond->dev->mc_list; dmi; dmi = dmi->next) {
977                         dev_mc_add(new_active->dev, dmi->dmi_addr, dmi->dmi_addrlen, 0);
978                 }
979                 bond_resend_igmp_join_requests(bond);
980         }
981 }
982
983 /*
984  * bond_do_fail_over_mac
985  *
986  * Perform special MAC address swapping for fail_over_mac settings
987  *
988  * Called with RTNL, bond->lock for read, curr_slave_lock for write_bh.
989  */
990 static void bond_do_fail_over_mac(struct bonding *bond,
991                                   struct slave *new_active,
992                                   struct slave *old_active)
993 {
994         u8 tmp_mac[ETH_ALEN];
995         struct sockaddr saddr;
996         int rv;
997
998         switch (bond->params.fail_over_mac) {
999         case BOND_FOM_ACTIVE:
1000                 if (new_active)
1001                         memcpy(bond->dev->dev_addr,  new_active->dev->dev_addr,
1002                                new_active->dev->addr_len);
1003                 break;
1004         case BOND_FOM_FOLLOW:
1005                 /*
1006                  * if new_active && old_active, swap them
1007                  * if just old_active, do nothing (going to no active slave)
1008                  * if just new_active, set new_active to bond's MAC
1009                  */
1010                 if (!new_active)
1011                         return;
1012
1013                 write_unlock_bh(&bond->curr_slave_lock);
1014                 read_unlock(&bond->lock);
1015
1016                 if (old_active) {
1017                         memcpy(tmp_mac, new_active->dev->dev_addr, ETH_ALEN);
1018                         memcpy(saddr.sa_data, old_active->dev->dev_addr,
1019                                ETH_ALEN);
1020                         saddr.sa_family = new_active->dev->type;
1021                 } else {
1022                         memcpy(saddr.sa_data, bond->dev->dev_addr, ETH_ALEN);
1023                         saddr.sa_family = bond->dev->type;
1024                 }
1025
1026                 rv = dev_set_mac_address(new_active->dev, &saddr);
1027                 if (rv) {
1028                         printk(KERN_ERR DRV_NAME
1029                                ": %s: Error %d setting MAC of slave %s\n",
1030                                bond->dev->name, -rv, new_active->dev->name);
1031                         goto out;
1032                 }
1033
1034                 if (!old_active)
1035                         goto out;
1036
1037                 memcpy(saddr.sa_data, tmp_mac, ETH_ALEN);
1038                 saddr.sa_family = old_active->dev->type;
1039
1040                 rv = dev_set_mac_address(old_active->dev, &saddr);
1041                 if (rv)
1042                         printk(KERN_ERR DRV_NAME
1043                                ": %s: Error %d setting MAC of slave %s\n",
1044                                bond->dev->name, -rv, new_active->dev->name);
1045 out:
1046                 read_lock(&bond->lock);
1047                 write_lock_bh(&bond->curr_slave_lock);
1048                 break;
1049         default:
1050                 printk(KERN_ERR DRV_NAME
1051                        ": %s: bond_do_fail_over_mac impossible: bad policy %d\n",
1052                        bond->dev->name, bond->params.fail_over_mac);
1053                 break;
1054         }
1055
1056 }
1057
1058
1059 /**
1060  * find_best_interface - select the best available slave to be the active one
1061  * @bond: our bonding struct
1062  *
1063  * Warning: Caller must hold curr_slave_lock for writing.
1064  */
1065 static struct slave *bond_find_best_slave(struct bonding *bond)
1066 {
1067         struct slave *new_active, *old_active;
1068         struct slave *bestslave = NULL;
1069         int mintime = bond->params.updelay;
1070         int i;
1071
1072         new_active = old_active = bond->curr_active_slave;
1073
1074         if (!new_active) { /* there were no active slaves left */
1075                 if (bond->slave_cnt > 0) {  /* found one slave */
1076                         new_active = bond->first_slave;
1077                 } else {
1078                         return NULL; /* still no slave, return NULL */
1079                 }
1080         }
1081
1082         /* first try the primary link; if arping, a link must tx/rx traffic
1083          * before it can be considered the curr_active_slave - also, we would skip
1084          * slaves between the curr_active_slave and primary_slave that may be up
1085          * and able to arp
1086          */
1087         if ((bond->primary_slave) &&
1088             (!bond->params.arp_interval) &&
1089             (IS_UP(bond->primary_slave->dev))) {
1090                 new_active = bond->primary_slave;
1091         }
1092
1093         /* remember where to stop iterating over the slaves */
1094         old_active = new_active;
1095
1096         bond_for_each_slave_from(bond, new_active, i, old_active) {
1097                 if (IS_UP(new_active->dev)) {
1098                         if (new_active->link == BOND_LINK_UP) {
1099                                 return new_active;
1100                         } else if (new_active->link == BOND_LINK_BACK) {
1101                                 /* link up, but waiting for stabilization */
1102                                 if (new_active->delay < mintime) {
1103                                         mintime = new_active->delay;
1104                                         bestslave = new_active;
1105                                 }
1106                         }
1107                 }
1108         }
1109
1110         return bestslave;
1111 }
1112
1113 /**
1114  * change_active_interface - change the active slave into the specified one
1115  * @bond: our bonding struct
1116  * @new: the new slave to make the active one
1117  *
1118  * Set the new slave to the bond's settings and unset them on the old
1119  * curr_active_slave.
1120  * Setting include flags, mc-list, promiscuity, allmulti, etc.
1121  *
1122  * If @new's link state is %BOND_LINK_BACK we'll set it to %BOND_LINK_UP,
1123  * because it is apparently the best available slave we have, even though its
1124  * updelay hasn't timed out yet.
1125  *
1126  * If new_active is not NULL, caller must hold bond->lock for read and
1127  * curr_slave_lock for write_bh.
1128  */
1129 void bond_change_active_slave(struct bonding *bond, struct slave *new_active)
1130 {
1131         struct slave *old_active = bond->curr_active_slave;
1132
1133         if (old_active == new_active) {
1134                 return;
1135         }
1136
1137         if (new_active) {
1138                 new_active->jiffies = jiffies;
1139
1140                 if (new_active->link == BOND_LINK_BACK) {
1141                         if (USES_PRIMARY(bond->params.mode)) {
1142                                 printk(KERN_INFO DRV_NAME
1143                                        ": %s: making interface %s the new "
1144                                        "active one %d ms earlier.\n",
1145                                        bond->dev->name, new_active->dev->name,
1146                                        (bond->params.updelay - new_active->delay) * bond->params.miimon);
1147                         }
1148
1149                         new_active->delay = 0;
1150                         new_active->link = BOND_LINK_UP;
1151
1152                         if (bond->params.mode == BOND_MODE_8023AD) {
1153                                 bond_3ad_handle_link_change(new_active, BOND_LINK_UP);
1154                         }
1155
1156                         if ((bond->params.mode == BOND_MODE_TLB) ||
1157                             (bond->params.mode == BOND_MODE_ALB)) {
1158                                 bond_alb_handle_link_change(bond, new_active, BOND_LINK_UP);
1159                         }
1160                 } else {
1161                         if (USES_PRIMARY(bond->params.mode)) {
1162                                 printk(KERN_INFO DRV_NAME
1163                                        ": %s: making interface %s the new "
1164                                        "active one.\n",
1165                                        bond->dev->name, new_active->dev->name);
1166                         }
1167                 }
1168         }
1169
1170         if (USES_PRIMARY(bond->params.mode)) {
1171                 bond_mc_swap(bond, new_active, old_active);
1172         }
1173
1174         if ((bond->params.mode == BOND_MODE_TLB) ||
1175             (bond->params.mode == BOND_MODE_ALB)) {
1176                 bond_alb_handle_active_change(bond, new_active);
1177                 if (old_active)
1178                         bond_set_slave_inactive_flags(old_active);
1179                 if (new_active)
1180                         bond_set_slave_active_flags(new_active);
1181         } else {
1182                 bond->curr_active_slave = new_active;
1183         }
1184
1185         if (bond->params.mode == BOND_MODE_ACTIVEBACKUP) {
1186                 if (old_active) {
1187                         bond_set_slave_inactive_flags(old_active);
1188                 }
1189
1190                 if (new_active) {
1191                         bond_set_slave_active_flags(new_active);
1192
1193                         if (bond->params.fail_over_mac)
1194                                 bond_do_fail_over_mac(bond, new_active,
1195                                                       old_active);
1196
1197                         bond->send_grat_arp = bond->params.num_grat_arp;
1198                         if (!test_bit(__LINK_STATE_LINKWATCH_PENDING,
1199                                         &bond->curr_active_slave->dev->state)) {
1200                                 bond_send_gratuitous_arp(bond);
1201                                 bond->send_grat_arp--;
1202                         } else {
1203                                 dprintk("delaying gratuitous arp on %s\n",
1204                                         bond->curr_active_slave->dev->name);
1205                         }
1206
1207                         write_unlock_bh(&bond->curr_slave_lock);
1208                         read_unlock(&bond->lock);
1209
1210                         netdev_bonding_change(bond->dev);
1211
1212                         read_lock(&bond->lock);
1213                         write_lock_bh(&bond->curr_slave_lock);
1214                 }
1215         }
1216 }
1217
1218 /**
1219  * bond_select_active_slave - select a new active slave, if needed
1220  * @bond: our bonding struct
1221  *
1222  * This functions shoud be called when one of the following occurs:
1223  * - The old curr_active_slave has been released or lost its link.
1224  * - The primary_slave has got its link back.
1225  * - A slave has got its link back and there's no old curr_active_slave.
1226  *
1227  * Caller must hold bond->lock for read and curr_slave_lock for write_bh.
1228  */
1229 void bond_select_active_slave(struct bonding *bond)
1230 {
1231         struct slave *best_slave;
1232         int rv;
1233
1234         best_slave = bond_find_best_slave(bond);
1235         if (best_slave != bond->curr_active_slave) {
1236                 bond_change_active_slave(bond, best_slave);
1237                 rv = bond_set_carrier(bond);
1238                 if (!rv)
1239                         return;
1240
1241                 if (netif_carrier_ok(bond->dev)) {
1242                         printk(KERN_INFO DRV_NAME
1243                                ": %s: first active interface up!\n",
1244                                bond->dev->name);
1245                 } else {
1246                         printk(KERN_INFO DRV_NAME ": %s: "
1247                                "now running without any active interface !\n",
1248                                bond->dev->name);
1249                 }
1250         }
1251 }
1252
1253 /*--------------------------- slave list handling ---------------------------*/
1254
1255 /*
1256  * This function attaches the slave to the end of list.
1257  *
1258  * bond->lock held for writing by caller.
1259  */
1260 static void bond_attach_slave(struct bonding *bond, struct slave *new_slave)
1261 {
1262         if (bond->first_slave == NULL) { /* attaching the first slave */
1263                 new_slave->next = new_slave;
1264                 new_slave->prev = new_slave;
1265                 bond->first_slave = new_slave;
1266         } else {
1267                 new_slave->next = bond->first_slave;
1268                 new_slave->prev = bond->first_slave->prev;
1269                 new_slave->next->prev = new_slave;
1270                 new_slave->prev->next = new_slave;
1271         }
1272
1273         bond->slave_cnt++;
1274 }
1275
1276 /*
1277  * This function detaches the slave from the list.
1278  * WARNING: no check is made to verify if the slave effectively
1279  * belongs to <bond>.
1280  * Nothing is freed on return, structures are just unchained.
1281  * If any slave pointer in bond was pointing to <slave>,
1282  * it should be changed by the calling function.
1283  *
1284  * bond->lock held for writing by caller.
1285  */
1286 static void bond_detach_slave(struct bonding *bond, struct slave *slave)
1287 {
1288         if (slave->next) {
1289                 slave->next->prev = slave->prev;
1290         }
1291
1292         if (slave->prev) {
1293                 slave->prev->next = slave->next;
1294         }
1295
1296         if (bond->first_slave == slave) { /* slave is the first slave */
1297                 if (bond->slave_cnt > 1) { /* there are more slave */
1298                         bond->first_slave = slave->next;
1299                 } else {
1300                         bond->first_slave = NULL; /* slave was the last one */
1301                 }
1302         }
1303
1304         slave->next = NULL;
1305         slave->prev = NULL;
1306         bond->slave_cnt--;
1307 }
1308
1309 /*---------------------------------- IOCTL ----------------------------------*/
1310
1311 static int bond_sethwaddr(struct net_device *bond_dev,
1312                           struct net_device *slave_dev)
1313 {
1314         dprintk("bond_dev=%p\n", bond_dev);
1315         dprintk("slave_dev=%p\n", slave_dev);
1316         dprintk("slave_dev->addr_len=%d\n", slave_dev->addr_len);
1317         memcpy(bond_dev->dev_addr, slave_dev->dev_addr, slave_dev->addr_len);
1318         return 0;
1319 }
1320
1321 #define BOND_VLAN_FEATURES \
1322         (NETIF_F_VLAN_CHALLENGED | NETIF_F_HW_VLAN_RX | NETIF_F_HW_VLAN_TX | \
1323          NETIF_F_HW_VLAN_FILTER)
1324
1325 /* 
1326  * Compute the common dev->feature set available to all slaves.  Some
1327  * feature bits are managed elsewhere, so preserve those feature bits
1328  * on the master device.
1329  */
1330 static int bond_compute_features(struct bonding *bond)
1331 {
1332         struct slave *slave;
1333         struct net_device *bond_dev = bond->dev;
1334         unsigned long features = bond_dev->features;
1335         unsigned short max_hard_header_len = max((u16)ETH_HLEN,
1336                                                 bond_dev->hard_header_len);
1337         int i;
1338
1339         features &= ~(NETIF_F_ALL_CSUM | BOND_VLAN_FEATURES);
1340         features |= NETIF_F_SG | NETIF_F_FRAGLIST | NETIF_F_HIGHDMA |
1341                     NETIF_F_GSO_MASK | NETIF_F_NO_CSUM;
1342
1343         bond_for_each_slave(bond, slave, i) {
1344                 features = netdev_compute_features(features,
1345                                                    slave->dev->features);
1346                 if (slave->dev->hard_header_len > max_hard_header_len)
1347                         max_hard_header_len = slave->dev->hard_header_len;
1348         }
1349
1350         features |= (bond_dev->features & BOND_VLAN_FEATURES);
1351         bond_dev->features = features;
1352         bond_dev->hard_header_len = max_hard_header_len;
1353
1354         return 0;
1355 }
1356
1357
1358 static void bond_setup_by_slave(struct net_device *bond_dev,
1359                                 struct net_device *slave_dev)
1360 {
1361         struct bonding *bond = bond_dev->priv;
1362
1363         bond_dev->neigh_setup           = slave_dev->neigh_setup;
1364         bond_dev->header_ops            = slave_dev->header_ops;
1365
1366         bond_dev->type              = slave_dev->type;
1367         bond_dev->hard_header_len   = slave_dev->hard_header_len;
1368         bond_dev->addr_len          = slave_dev->addr_len;
1369
1370         memcpy(bond_dev->broadcast, slave_dev->broadcast,
1371                 slave_dev->addr_len);
1372         bond->setup_by_slave = 1;
1373 }
1374
1375 /* enslave device <slave> to bond device <master> */
1376 int bond_enslave(struct net_device *bond_dev, struct net_device *slave_dev)
1377 {
1378         struct bonding *bond = bond_dev->priv;
1379         struct slave *new_slave = NULL;
1380         struct dev_mc_list *dmi;
1381         struct sockaddr addr;
1382         int link_reporting;
1383         int old_features = bond_dev->features;
1384         int res = 0;
1385
1386         if (!bond->params.use_carrier && slave_dev->ethtool_ops == NULL &&
1387                 slave_dev->do_ioctl == NULL) {
1388                 printk(KERN_WARNING DRV_NAME
1389                        ": %s: Warning: no link monitoring support for %s\n",
1390                        bond_dev->name, slave_dev->name);
1391         }
1392
1393         /* bond must be initialized by bond_open() before enslaving */
1394         if (!(bond_dev->flags & IFF_UP)) {
1395                 printk(KERN_WARNING DRV_NAME
1396                         " %s: master_dev is not up in bond_enslave\n",
1397                         bond_dev->name);
1398         }
1399
1400         /* already enslaved */
1401         if (slave_dev->flags & IFF_SLAVE) {
1402                 dprintk("Error, Device was already enslaved\n");
1403                 return -EBUSY;
1404         }
1405
1406         /* vlan challenged mutual exclusion */
1407         /* no need to lock since we're protected by rtnl_lock */
1408         if (slave_dev->features & NETIF_F_VLAN_CHALLENGED) {
1409                 dprintk("%s: NETIF_F_VLAN_CHALLENGED\n", slave_dev->name);
1410                 if (!list_empty(&bond->vlan_list)) {
1411                         printk(KERN_ERR DRV_NAME
1412                                ": %s: Error: cannot enslave VLAN "
1413                                "challenged slave %s on VLAN enabled "
1414                                "bond %s\n", bond_dev->name, slave_dev->name,
1415                                bond_dev->name);
1416                         return -EPERM;
1417                 } else {
1418                         printk(KERN_WARNING DRV_NAME
1419                                ": %s: Warning: enslaved VLAN challenged "
1420                                "slave %s. Adding VLANs will be blocked as "
1421                                "long as %s is part of bond %s\n",
1422                                bond_dev->name, slave_dev->name, slave_dev->name,
1423                                bond_dev->name);
1424                         bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
1425                 }
1426         } else {
1427                 dprintk("%s: ! NETIF_F_VLAN_CHALLENGED\n", slave_dev->name);
1428                 if (bond->slave_cnt == 0) {
1429                         /* First slave, and it is not VLAN challenged,
1430                          * so remove the block of adding VLANs over the bond.
1431                          */
1432                         bond_dev->features &= ~NETIF_F_VLAN_CHALLENGED;
1433                 }
1434         }
1435
1436         /*
1437          * Old ifenslave binaries are no longer supported.  These can
1438          * be identified with moderate accurary by the state of the slave:
1439          * the current ifenslave will set the interface down prior to
1440          * enslaving it; the old ifenslave will not.
1441          */
1442         if ((slave_dev->flags & IFF_UP)) {
1443                 printk(KERN_ERR DRV_NAME ": %s is up. "
1444                        "This may be due to an out of date ifenslave.\n",
1445                        slave_dev->name);
1446                 res = -EPERM;
1447                 goto err_undo_flags;
1448         }
1449
1450         /* set bonding device ether type by slave - bonding netdevices are
1451          * created with ether_setup, so when the slave type is not ARPHRD_ETHER
1452          * there is a need to override some of the type dependent attribs/funcs.
1453          *
1454          * bond ether type mutual exclusion - don't allow slaves of dissimilar
1455          * ether type (eg ARPHRD_ETHER and ARPHRD_INFINIBAND) share the same bond
1456          */
1457         if (bond->slave_cnt == 0) {
1458                 if (slave_dev->type != ARPHRD_ETHER)
1459                         bond_setup_by_slave(bond_dev, slave_dev);
1460         } else if (bond_dev->type != slave_dev->type) {
1461                 printk(KERN_ERR DRV_NAME ": %s ether type (%d) is different "
1462                         "from other slaves (%d), can not enslave it.\n",
1463                         slave_dev->name,
1464                         slave_dev->type, bond_dev->type);
1465                         res = -EINVAL;
1466                         goto err_undo_flags;
1467         }
1468
1469         if (slave_dev->set_mac_address == NULL) {
1470                 if (bond->slave_cnt == 0) {
1471                         printk(KERN_WARNING DRV_NAME
1472                                ": %s: Warning: The first slave device "
1473                                "specified does not support setting the MAC "
1474                                "address. Setting fail_over_mac to active.",
1475                                bond_dev->name);
1476                         bond->params.fail_over_mac = BOND_FOM_ACTIVE;
1477                 } else if (bond->params.fail_over_mac != BOND_FOM_ACTIVE) {
1478                         printk(KERN_ERR DRV_NAME
1479                                 ": %s: Error: The slave device specified "
1480                                 "does not support setting the MAC address, "
1481                                 "but fail_over_mac is not set to active.\n"
1482                                 , bond_dev->name);
1483                         res = -EOPNOTSUPP;
1484                         goto err_undo_flags;
1485                 }
1486         }
1487
1488         new_slave = kzalloc(sizeof(struct slave), GFP_KERNEL);
1489         if (!new_slave) {
1490                 res = -ENOMEM;
1491                 goto err_undo_flags;
1492         }
1493
1494         /* save slave's original flags before calling
1495          * netdev_set_master and dev_open
1496          */
1497         new_slave->original_flags = slave_dev->flags;
1498
1499         /*
1500          * Save slave's original ("permanent") mac address for modes
1501          * that need it, and for restoring it upon release, and then
1502          * set it to the master's address
1503          */
1504         memcpy(new_slave->perm_hwaddr, slave_dev->dev_addr, ETH_ALEN);
1505
1506         if (!bond->params.fail_over_mac) {
1507                 /*
1508                  * Set slave to master's mac address.  The application already
1509                  * set the master's mac address to that of the first slave
1510                  */
1511                 memcpy(addr.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
1512                 addr.sa_family = slave_dev->type;
1513                 res = dev_set_mac_address(slave_dev, &addr);
1514                 if (res) {
1515                         dprintk("Error %d calling set_mac_address\n", res);
1516                         goto err_free;
1517                 }
1518         }
1519
1520         res = netdev_set_master(slave_dev, bond_dev);
1521         if (res) {
1522                 dprintk("Error %d calling netdev_set_master\n", res);
1523                 goto err_restore_mac;
1524         }
1525         /* open the slave since the application closed it */
1526         res = dev_open(slave_dev);
1527         if (res) {
1528                 dprintk("Openning slave %s failed\n", slave_dev->name);
1529                 goto err_unset_master;
1530         }
1531
1532         new_slave->dev = slave_dev;
1533         slave_dev->priv_flags |= IFF_BONDING;
1534
1535         if ((bond->params.mode == BOND_MODE_TLB) ||
1536             (bond->params.mode == BOND_MODE_ALB)) {
1537                 /* bond_alb_init_slave() must be called before all other stages since
1538                  * it might fail and we do not want to have to undo everything
1539                  */
1540                 res = bond_alb_init_slave(bond, new_slave);
1541                 if (res) {
1542                         goto err_close;
1543                 }
1544         }
1545
1546         /* If the mode USES_PRIMARY, then the new slave gets the
1547          * master's promisc (and mc) settings only if it becomes the
1548          * curr_active_slave, and that is taken care of later when calling
1549          * bond_change_active()
1550          */
1551         if (!USES_PRIMARY(bond->params.mode)) {
1552                 /* set promiscuity level to new slave */
1553                 if (bond_dev->flags & IFF_PROMISC) {
1554                         dev_set_promiscuity(slave_dev, 1);
1555                 }
1556
1557                 /* set allmulti level to new slave */
1558                 if (bond_dev->flags & IFF_ALLMULTI) {
1559                         dev_set_allmulti(slave_dev, 1);
1560                 }
1561
1562                 netif_tx_lock_bh(bond_dev);
1563                 /* upload master's mc_list to new slave */
1564                 for (dmi = bond_dev->mc_list; dmi; dmi = dmi->next) {
1565                         dev_mc_add (slave_dev, dmi->dmi_addr, dmi->dmi_addrlen, 0);
1566                 }
1567                 netif_tx_unlock_bh(bond_dev);
1568         }
1569
1570         if (bond->params.mode == BOND_MODE_8023AD) {
1571                 /* add lacpdu mc addr to mc list */
1572                 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
1573
1574                 dev_mc_add(slave_dev, lacpdu_multicast, ETH_ALEN, 0);
1575         }
1576
1577         bond_add_vlans_on_slave(bond, slave_dev);
1578
1579         write_lock_bh(&bond->lock);
1580
1581         bond_attach_slave(bond, new_slave);
1582
1583         new_slave->delay = 0;
1584         new_slave->link_failure_count = 0;
1585
1586         bond_compute_features(bond);
1587
1588         write_unlock_bh(&bond->lock);
1589
1590         read_lock(&bond->lock);
1591
1592         new_slave->last_arp_rx = jiffies;
1593
1594         if (bond->params.miimon && !bond->params.use_carrier) {
1595                 link_reporting = bond_check_dev_link(bond, slave_dev, 1);
1596
1597                 if ((link_reporting == -1) && !bond->params.arp_interval) {
1598                         /*
1599                          * miimon is set but a bonded network driver
1600                          * does not support ETHTOOL/MII and
1601                          * arp_interval is not set.  Note: if
1602                          * use_carrier is enabled, we will never go
1603                          * here (because netif_carrier is always
1604                          * supported); thus, we don't need to change
1605                          * the messages for netif_carrier.
1606                          */
1607                         printk(KERN_WARNING DRV_NAME
1608                                ": %s: Warning: MII and ETHTOOL support not "
1609                                "available for interface %s, and "
1610                                "arp_interval/arp_ip_target module parameters "
1611                                "not specified, thus bonding will not detect "
1612                                "link failures! see bonding.txt for details.\n",
1613                                bond_dev->name, slave_dev->name);
1614                 } else if (link_reporting == -1) {
1615                         /* unable get link status using mii/ethtool */
1616                         printk(KERN_WARNING DRV_NAME
1617                                ": %s: Warning: can't get link status from "
1618                                "interface %s; the network driver associated "
1619                                "with this interface does not support MII or "
1620                                "ETHTOOL link status reporting, thus miimon "
1621                                "has no effect on this interface.\n",
1622                                bond_dev->name, slave_dev->name);
1623                 }
1624         }
1625
1626         /* check for initial state */
1627         if (!bond->params.miimon ||
1628             (bond_check_dev_link(bond, slave_dev, 0) == BMSR_LSTATUS)) {
1629                 if (bond->params.updelay) {
1630                         dprintk("Initial state of slave_dev is "
1631                                 "BOND_LINK_BACK\n");
1632                         new_slave->link  = BOND_LINK_BACK;
1633                         new_slave->delay = bond->params.updelay;
1634                 } else {
1635                         dprintk("Initial state of slave_dev is "
1636                                 "BOND_LINK_UP\n");
1637                         new_slave->link  = BOND_LINK_UP;
1638                 }
1639                 new_slave->jiffies = jiffies;
1640         } else {
1641                 dprintk("Initial state of slave_dev is "
1642                         "BOND_LINK_DOWN\n");
1643                 new_slave->link  = BOND_LINK_DOWN;
1644         }
1645
1646         if (bond_update_speed_duplex(new_slave) &&
1647             (new_slave->link != BOND_LINK_DOWN)) {
1648                 printk(KERN_WARNING DRV_NAME
1649                        ": %s: Warning: failed to get speed and duplex from %s, "
1650                        "assumed to be 100Mb/sec and Full.\n",
1651                        bond_dev->name, new_slave->dev->name);
1652
1653                 if (bond->params.mode == BOND_MODE_8023AD) {
1654                         printk(KERN_WARNING DRV_NAME
1655                                ": %s: Warning: Operation of 802.3ad mode requires ETHTOOL "
1656                                "support in base driver for proper aggregator "
1657                                "selection.\n", bond_dev->name);
1658                 }
1659         }
1660
1661         if (USES_PRIMARY(bond->params.mode) && bond->params.primary[0]) {
1662                 /* if there is a primary slave, remember it */
1663                 if (strcmp(bond->params.primary, new_slave->dev->name) == 0) {
1664                         bond->primary_slave = new_slave;
1665                 }
1666         }
1667
1668         write_lock_bh(&bond->curr_slave_lock);
1669
1670         switch (bond->params.mode) {
1671         case BOND_MODE_ACTIVEBACKUP:
1672                 bond_set_slave_inactive_flags(new_slave);
1673                 bond_select_active_slave(bond);
1674                 break;
1675         case BOND_MODE_8023AD:
1676                 /* in 802.3ad mode, the internal mechanism
1677                  * will activate the slaves in the selected
1678                  * aggregator
1679                  */
1680                 bond_set_slave_inactive_flags(new_slave);
1681                 /* if this is the first slave */
1682                 if (bond->slave_cnt == 1) {
1683                         SLAVE_AD_INFO(new_slave).id = 1;
1684                         /* Initialize AD with the number of times that the AD timer is called in 1 second
1685                          * can be called only after the mac address of the bond is set
1686                          */
1687                         bond_3ad_initialize(bond, 1000/AD_TIMER_INTERVAL,
1688                                             bond->params.lacp_fast);
1689                 } else {
1690                         SLAVE_AD_INFO(new_slave).id =
1691                                 SLAVE_AD_INFO(new_slave->prev).id + 1;
1692                 }
1693
1694                 bond_3ad_bind_slave(new_slave);
1695                 break;
1696         case BOND_MODE_TLB:
1697         case BOND_MODE_ALB:
1698                 new_slave->state = BOND_STATE_ACTIVE;
1699                 bond_set_slave_inactive_flags(new_slave);
1700                 break;
1701         default:
1702                 dprintk("This slave is always active in trunk mode\n");
1703
1704                 /* always active in trunk mode */
1705                 new_slave->state = BOND_STATE_ACTIVE;
1706
1707                 /* In trunking mode there is little meaning to curr_active_slave
1708                  * anyway (it holds no special properties of the bond device),
1709                  * so we can change it without calling change_active_interface()
1710                  */
1711                 if (!bond->curr_active_slave) {
1712                         bond->curr_active_slave = new_slave;
1713                 }
1714                 break;
1715         } /* switch(bond_mode) */
1716
1717         write_unlock_bh(&bond->curr_slave_lock);
1718
1719         bond_set_carrier(bond);
1720
1721         read_unlock(&bond->lock);
1722
1723         res = bond_create_slave_symlinks(bond_dev, slave_dev);
1724         if (res)
1725                 goto err_close;
1726
1727         printk(KERN_INFO DRV_NAME
1728                ": %s: enslaving %s as a%s interface with a%s link.\n",
1729                bond_dev->name, slave_dev->name,
1730                new_slave->state == BOND_STATE_ACTIVE ? "n active" : " backup",
1731                new_slave->link != BOND_LINK_DOWN ? "n up" : " down");
1732
1733         /* enslave is successful */
1734         return 0;
1735
1736 /* Undo stages on error */
1737 err_close:
1738         dev_close(slave_dev);
1739
1740 err_unset_master:
1741         netdev_set_master(slave_dev, NULL);
1742
1743 err_restore_mac:
1744         if (!bond->params.fail_over_mac) {
1745                 /* XXX TODO - fom follow mode needs to change master's
1746                  * MAC if this slave's MAC is in use by the bond, or at
1747                  * least print a warning.
1748                  */
1749                 memcpy(addr.sa_data, new_slave->perm_hwaddr, ETH_ALEN);
1750                 addr.sa_family = slave_dev->type;
1751                 dev_set_mac_address(slave_dev, &addr);
1752         }
1753
1754 err_free:
1755         kfree(new_slave);
1756
1757 err_undo_flags:
1758         bond_dev->features = old_features;
1759  
1760         return res;
1761 }
1762
1763 /*
1764  * Try to release the slave device <slave> from the bond device <master>
1765  * It is legal to access curr_active_slave without a lock because all the function
1766  * is write-locked.
1767  *
1768  * The rules for slave state should be:
1769  *   for Active/Backup:
1770  *     Active stays on all backups go down
1771  *   for Bonded connections:
1772  *     The first up interface should be left on and all others downed.
1773  */
1774 int bond_release(struct net_device *bond_dev, struct net_device *slave_dev)
1775 {
1776         struct bonding *bond = bond_dev->priv;
1777         struct slave *slave, *oldcurrent;
1778         struct sockaddr addr;
1779         int mac_addr_differ;
1780         DECLARE_MAC_BUF(mac);
1781
1782         /* slave is not a slave or master is not master of this slave */
1783         if (!(slave_dev->flags & IFF_SLAVE) ||
1784             (slave_dev->master != bond_dev)) {
1785                 printk(KERN_ERR DRV_NAME
1786                        ": %s: Error: cannot release %s.\n",
1787                        bond_dev->name, slave_dev->name);
1788                 return -EINVAL;
1789         }
1790
1791         write_lock_bh(&bond->lock);
1792
1793         slave = bond_get_slave_by_dev(bond, slave_dev);
1794         if (!slave) {
1795                 /* not a slave of this bond */
1796                 printk(KERN_INFO DRV_NAME
1797                        ": %s: %s not enslaved\n",
1798                        bond_dev->name, slave_dev->name);
1799                 write_unlock_bh(&bond->lock);
1800                 return -EINVAL;
1801         }
1802
1803         if (!bond->params.fail_over_mac) {
1804                 mac_addr_differ = memcmp(bond_dev->dev_addr, slave->perm_hwaddr,
1805                                          ETH_ALEN);
1806                 if (!mac_addr_differ && (bond->slave_cnt > 1))
1807                         printk(KERN_WARNING DRV_NAME
1808                                ": %s: Warning: the permanent HWaddr of %s - "
1809                                "%s - is still in use by %s. "
1810                                "Set the HWaddr of %s to a different address "
1811                                "to avoid conflicts.\n",
1812                                bond_dev->name, slave_dev->name,
1813                                print_mac(mac, slave->perm_hwaddr),
1814                                bond_dev->name, slave_dev->name);
1815         }
1816
1817         /* Inform AD package of unbinding of slave. */
1818         if (bond->params.mode == BOND_MODE_8023AD) {
1819                 /* must be called before the slave is
1820                  * detached from the list
1821                  */
1822                 bond_3ad_unbind_slave(slave);
1823         }
1824
1825         printk(KERN_INFO DRV_NAME
1826                ": %s: releasing %s interface %s\n",
1827                bond_dev->name,
1828                (slave->state == BOND_STATE_ACTIVE)
1829                ? "active" : "backup",
1830                slave_dev->name);
1831
1832         oldcurrent = bond->curr_active_slave;
1833
1834         bond->current_arp_slave = NULL;
1835
1836         /* release the slave from its bond */
1837         bond_detach_slave(bond, slave);
1838
1839         bond_compute_features(bond);
1840
1841         if (bond->primary_slave == slave) {
1842                 bond->primary_slave = NULL;
1843         }
1844
1845         if (oldcurrent == slave) {
1846                 bond_change_active_slave(bond, NULL);
1847         }
1848
1849         if ((bond->params.mode == BOND_MODE_TLB) ||
1850             (bond->params.mode == BOND_MODE_ALB)) {
1851                 /* Must be called only after the slave has been
1852                  * detached from the list and the curr_active_slave
1853                  * has been cleared (if our_slave == old_current),
1854                  * but before a new active slave is selected.
1855                  */
1856                 write_unlock_bh(&bond->lock);
1857                 bond_alb_deinit_slave(bond, slave);
1858                 write_lock_bh(&bond->lock);
1859         }
1860
1861         if (oldcurrent == slave) {
1862                 /*
1863                  * Note that we hold RTNL over this sequence, so there
1864                  * is no concern that another slave add/remove event
1865                  * will interfere.
1866                  */
1867                 write_unlock_bh(&bond->lock);
1868                 read_lock(&bond->lock);
1869                 write_lock_bh(&bond->curr_slave_lock);
1870
1871                 bond_select_active_slave(bond);
1872
1873                 write_unlock_bh(&bond->curr_slave_lock);
1874                 read_unlock(&bond->lock);
1875                 write_lock_bh(&bond->lock);
1876         }
1877
1878         if (bond->slave_cnt == 0) {
1879                 bond_set_carrier(bond);
1880
1881                 /* if the last slave was removed, zero the mac address
1882                  * of the master so it will be set by the application
1883                  * to the mac address of the first slave
1884                  */
1885                 memset(bond_dev->dev_addr, 0, bond_dev->addr_len);
1886
1887                 if (list_empty(&bond->vlan_list)) {
1888                         bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
1889                 } else {
1890                         printk(KERN_WARNING DRV_NAME
1891                                ": %s: Warning: clearing HW address of %s while it "
1892                                "still has VLANs.\n",
1893                                bond_dev->name, bond_dev->name);
1894                         printk(KERN_WARNING DRV_NAME
1895                                ": %s: When re-adding slaves, make sure the bond's "
1896                                "HW address matches its VLANs'.\n",
1897                                bond_dev->name);
1898                 }
1899         } else if ((bond_dev->features & NETIF_F_VLAN_CHALLENGED) &&
1900                    !bond_has_challenged_slaves(bond)) {
1901                 printk(KERN_INFO DRV_NAME
1902                        ": %s: last VLAN challenged slave %s "
1903                        "left bond %s. VLAN blocking is removed\n",
1904                        bond_dev->name, slave_dev->name, bond_dev->name);
1905                 bond_dev->features &= ~NETIF_F_VLAN_CHALLENGED;
1906         }
1907
1908         write_unlock_bh(&bond->lock);
1909
1910         /* must do this from outside any spinlocks */
1911         bond_destroy_slave_symlinks(bond_dev, slave_dev);
1912
1913         bond_del_vlans_from_slave(bond, slave_dev);
1914
1915         /* If the mode USES_PRIMARY, then we should only remove its
1916          * promisc and mc settings if it was the curr_active_slave, but that was
1917          * already taken care of above when we detached the slave
1918          */
1919         if (!USES_PRIMARY(bond->params.mode)) {
1920                 /* unset promiscuity level from slave */
1921                 if (bond_dev->flags & IFF_PROMISC) {
1922                         dev_set_promiscuity(slave_dev, -1);
1923                 }
1924
1925                 /* unset allmulti level from slave */
1926                 if (bond_dev->flags & IFF_ALLMULTI) {
1927                         dev_set_allmulti(slave_dev, -1);
1928                 }
1929
1930                 /* flush master's mc_list from slave */
1931                 netif_tx_lock_bh(bond_dev);
1932                 bond_mc_list_flush(bond_dev, slave_dev);
1933                 netif_tx_unlock_bh(bond_dev);
1934         }
1935
1936         netdev_set_master(slave_dev, NULL);
1937
1938         /* close slave before restoring its mac address */
1939         dev_close(slave_dev);
1940
1941         if (bond->params.fail_over_mac != BOND_FOM_ACTIVE) {
1942                 /* restore original ("permanent") mac address */
1943                 memcpy(addr.sa_data, slave->perm_hwaddr, ETH_ALEN);
1944                 addr.sa_family = slave_dev->type;
1945                 dev_set_mac_address(slave_dev, &addr);
1946         }
1947
1948         slave_dev->priv_flags &= ~(IFF_MASTER_8023AD | IFF_MASTER_ALB |
1949                                    IFF_SLAVE_INACTIVE | IFF_BONDING |
1950                                    IFF_SLAVE_NEEDARP);
1951
1952         kfree(slave);
1953
1954         return 0;  /* deletion OK */
1955 }
1956
1957 /*
1958 * Destroy a bonding device.
1959 * Must be under rtnl_lock when this function is called.
1960 */
1961 void bond_destroy(struct bonding *bond)
1962 {
1963         bond_deinit(bond->dev);
1964         bond_destroy_sysfs_entry(bond);
1965         unregister_netdevice(bond->dev);
1966 }
1967
1968 /*
1969 * First release a slave and than destroy the bond if no more slaves iare left.
1970 * Must be under rtnl_lock when this function is called.
1971 */
1972 int  bond_release_and_destroy(struct net_device *bond_dev, struct net_device *slave_dev)
1973 {
1974         struct bonding *bond = bond_dev->priv;
1975         int ret;
1976
1977         ret = bond_release(bond_dev, slave_dev);
1978         if ((ret == 0) && (bond->slave_cnt == 0)) {
1979                 printk(KERN_INFO DRV_NAME ": %s: destroying bond %s.\n",
1980                        bond_dev->name, bond_dev->name);
1981                 bond_destroy(bond);
1982         }
1983         return ret;
1984 }
1985
1986 /*
1987  * This function releases all slaves.
1988  */
1989 static int bond_release_all(struct net_device *bond_dev)
1990 {
1991         struct bonding *bond = bond_dev->priv;
1992         struct slave *slave;
1993         struct net_device *slave_dev;
1994         struct sockaddr addr;
1995
1996         write_lock_bh(&bond->lock);
1997
1998         netif_carrier_off(bond_dev);
1999
2000         if (bond->slave_cnt == 0) {
2001                 goto out;
2002         }
2003
2004         bond->current_arp_slave = NULL;
2005         bond->primary_slave = NULL;
2006         bond_change_active_slave(bond, NULL);
2007
2008         while ((slave = bond->first_slave) != NULL) {
2009                 /* Inform AD package of unbinding of slave
2010                  * before slave is detached from the list.
2011                  */
2012                 if (bond->params.mode == BOND_MODE_8023AD) {
2013                         bond_3ad_unbind_slave(slave);
2014                 }
2015
2016                 slave_dev = slave->dev;
2017                 bond_detach_slave(bond, slave);
2018
2019                 /* now that the slave is detached, unlock and perform
2020                  * all the undo steps that should not be called from
2021                  * within a lock.
2022                  */
2023                 write_unlock_bh(&bond->lock);
2024
2025                 if ((bond->params.mode == BOND_MODE_TLB) ||
2026                     (bond->params.mode == BOND_MODE_ALB)) {
2027                         /* must be called only after the slave
2028                          * has been detached from the list
2029                          */
2030                         bond_alb_deinit_slave(bond, slave);
2031                 }
2032
2033                 bond_compute_features(bond);
2034
2035                 bond_destroy_slave_symlinks(bond_dev, slave_dev);
2036                 bond_del_vlans_from_slave(bond, slave_dev);
2037
2038                 /* If the mode USES_PRIMARY, then we should only remove its
2039                  * promisc and mc settings if it was the curr_active_slave, but that was
2040                  * already taken care of above when we detached the slave
2041                  */
2042                 if (!USES_PRIMARY(bond->params.mode)) {
2043                         /* unset promiscuity level from slave */
2044                         if (bond_dev->flags & IFF_PROMISC) {
2045                                 dev_set_promiscuity(slave_dev, -1);
2046                         }
2047
2048                         /* unset allmulti level from slave */
2049                         if (bond_dev->flags & IFF_ALLMULTI) {
2050                                 dev_set_allmulti(slave_dev, -1);
2051                         }
2052
2053                         /* flush master's mc_list from slave */
2054                         netif_tx_lock_bh(bond_dev);
2055                         bond_mc_list_flush(bond_dev, slave_dev);
2056                         netif_tx_unlock_bh(bond_dev);
2057                 }
2058
2059                 netdev_set_master(slave_dev, NULL);
2060
2061                 /* close slave before restoring its mac address */
2062                 dev_close(slave_dev);
2063
2064                 if (!bond->params.fail_over_mac) {
2065                         /* restore original ("permanent") mac address*/
2066                         memcpy(addr.sa_data, slave->perm_hwaddr, ETH_ALEN);
2067                         addr.sa_family = slave_dev->type;
2068                         dev_set_mac_address(slave_dev, &addr);
2069                 }
2070
2071                 slave_dev->priv_flags &= ~(IFF_MASTER_8023AD | IFF_MASTER_ALB |
2072                                            IFF_SLAVE_INACTIVE);
2073
2074                 kfree(slave);
2075
2076                 /* re-acquire the lock before getting the next slave */
2077                 write_lock_bh(&bond->lock);
2078         }
2079
2080         /* zero the mac address of the master so it will be
2081          * set by the application to the mac address of the
2082          * first slave
2083          */
2084         memset(bond_dev->dev_addr, 0, bond_dev->addr_len);
2085
2086         if (list_empty(&bond->vlan_list)) {
2087                 bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
2088         } else {
2089                 printk(KERN_WARNING DRV_NAME
2090                        ": %s: Warning: clearing HW address of %s while it "
2091                        "still has VLANs.\n",
2092                        bond_dev->name, bond_dev->name);
2093                 printk(KERN_WARNING DRV_NAME
2094                        ": %s: When re-adding slaves, make sure the bond's "
2095                        "HW address matches its VLANs'.\n",
2096                        bond_dev->name);
2097         }
2098
2099         printk(KERN_INFO DRV_NAME
2100                ": %s: released all slaves\n",
2101                bond_dev->name);
2102
2103 out:
2104         write_unlock_bh(&bond->lock);
2105
2106         return 0;
2107 }
2108
2109 /*
2110  * This function changes the active slave to slave <slave_dev>.
2111  * It returns -EINVAL in the following cases.
2112  *  - <slave_dev> is not found in the list.
2113  *  - There is not active slave now.
2114  *  - <slave_dev> is already active.
2115  *  - The link state of <slave_dev> is not BOND_LINK_UP.
2116  *  - <slave_dev> is not running.
2117  * In these cases, this fuction does nothing.
2118  * In the other cases, currnt_slave pointer is changed and 0 is returned.
2119  */
2120 static int bond_ioctl_change_active(struct net_device *bond_dev, struct net_device *slave_dev)
2121 {
2122         struct bonding *bond = bond_dev->priv;
2123         struct slave *old_active = NULL;
2124         struct slave *new_active = NULL;
2125         int res = 0;
2126
2127         if (!USES_PRIMARY(bond->params.mode)) {
2128                 return -EINVAL;
2129         }
2130
2131         /* Verify that master_dev is indeed the master of slave_dev */
2132         if (!(slave_dev->flags & IFF_SLAVE) ||
2133             (slave_dev->master != bond_dev)) {
2134                 return -EINVAL;
2135         }
2136
2137         read_lock(&bond->lock);
2138
2139         read_lock(&bond->curr_slave_lock);
2140         old_active = bond->curr_active_slave;
2141         read_unlock(&bond->curr_slave_lock);
2142
2143         new_active = bond_get_slave_by_dev(bond, slave_dev);
2144
2145         /*
2146          * Changing to the current active: do nothing; return success.
2147          */
2148         if (new_active && (new_active == old_active)) {
2149                 read_unlock(&bond->lock);
2150                 return 0;
2151         }
2152
2153         if ((new_active) &&
2154             (old_active) &&
2155             (new_active->link == BOND_LINK_UP) &&
2156             IS_UP(new_active->dev)) {
2157                 write_lock_bh(&bond->curr_slave_lock);
2158                 bond_change_active_slave(bond, new_active);
2159                 write_unlock_bh(&bond->curr_slave_lock);
2160         } else {
2161                 res = -EINVAL;
2162         }
2163
2164         read_unlock(&bond->lock);
2165
2166         return res;
2167 }
2168
2169 static int bond_info_query(struct net_device *bond_dev, struct ifbond *info)
2170 {
2171         struct bonding *bond = bond_dev->priv;
2172
2173         info->bond_mode = bond->params.mode;
2174         info->miimon = bond->params.miimon;
2175
2176         read_lock(&bond->lock);
2177         info->num_slaves = bond->slave_cnt;
2178         read_unlock(&bond->lock);
2179
2180         return 0;
2181 }
2182
2183 static int bond_slave_info_query(struct net_device *bond_dev, struct ifslave *info)
2184 {
2185         struct bonding *bond = bond_dev->priv;
2186         struct slave *slave;
2187         int i, found = 0;
2188
2189         if (info->slave_id < 0) {
2190                 return -ENODEV;
2191         }
2192
2193         read_lock(&bond->lock);
2194
2195         bond_for_each_slave(bond, slave, i) {
2196                 if (i == (int)info->slave_id) {
2197                         found = 1;
2198                         break;
2199                 }
2200         }
2201
2202         read_unlock(&bond->lock);
2203
2204         if (found) {
2205                 strcpy(info->slave_name, slave->dev->name);
2206                 info->link = slave->link;
2207                 info->state = slave->state;
2208                 info->link_failure_count = slave->link_failure_count;
2209         } else {
2210                 return -ENODEV;
2211         }
2212
2213         return 0;
2214 }
2215
2216 /*-------------------------------- Monitoring -------------------------------*/
2217
2218 /*
2219  * if !have_locks, return nonzero if a failover is necessary.  if
2220  * have_locks, do whatever failover activities are needed.
2221  *
2222  * This is to separate the inspection and failover steps for locking
2223  * purposes; failover requires rtnl, but acquiring it for every
2224  * inspection is undesirable, so a wrapper first does inspection, and
2225  * the acquires the necessary locks and calls again to perform
2226  * failover if needed.  Since all locks are dropped, a complete
2227  * restart is needed between calls.
2228  */
2229 static int __bond_mii_monitor(struct bonding *bond, int have_locks)
2230 {
2231         struct slave *slave, *oldcurrent;
2232         int do_failover = 0;
2233         int i;
2234
2235         if (bond->slave_cnt == 0)
2236                 goto out;
2237
2238         /* we will try to read the link status of each of our slaves, and
2239          * set their IFF_RUNNING flag appropriately. For each slave not
2240          * supporting MII status, we won't do anything so that a user-space
2241          * program could monitor the link itself if needed.
2242          */
2243
2244         if (bond->send_grat_arp) {
2245                 if (bond->curr_active_slave && test_bit(__LINK_STATE_LINKWATCH_PENDING,
2246                                 &bond->curr_active_slave->dev->state))
2247                         dprintk("Needs to send gratuitous arp but not yet\n");
2248                 else {
2249                         dprintk("sending delayed gratuitous arp on on %s\n",
2250                                 bond->curr_active_slave->dev->name);
2251                         bond_send_gratuitous_arp(bond);
2252                         bond->send_grat_arp--;
2253                 }
2254         }
2255         read_lock(&bond->curr_slave_lock);
2256         oldcurrent = bond->curr_active_slave;
2257         read_unlock(&bond->curr_slave_lock);
2258
2259         bond_for_each_slave(bond, slave, i) {
2260                 struct net_device *slave_dev = slave->dev;
2261                 int link_state;
2262                 u16 old_speed = slave->speed;
2263                 u8 old_duplex = slave->duplex;
2264
2265                 link_state = bond_check_dev_link(bond, slave_dev, 0);
2266
2267                 switch (slave->link) {
2268                 case BOND_LINK_UP:      /* the link was up */
2269                         if (link_state == BMSR_LSTATUS) {
2270                                 if (!oldcurrent) {
2271                                         if (!have_locks)
2272                                                 return 1;
2273                                         do_failover = 1;
2274                                 }
2275                                 break;
2276                         } else { /* link going down */
2277                                 slave->link  = BOND_LINK_FAIL;
2278                                 slave->delay = bond->params.downdelay;
2279
2280                                 if (slave->link_failure_count < UINT_MAX) {
2281                                         slave->link_failure_count++;
2282                                 }
2283
2284                                 if (bond->params.downdelay) {
2285                                         printk(KERN_INFO DRV_NAME
2286                                                ": %s: link status down for %s "
2287                                                "interface %s, disabling it in "
2288                                                "%d ms.\n",
2289                                                bond->dev->name,
2290                                                IS_UP(slave_dev)
2291                                                ? ((bond->params.mode == BOND_MODE_ACTIVEBACKUP)
2292                                                   ? ((slave == oldcurrent)
2293                                                      ? "active " : "backup ")
2294                                                   : "")
2295                                                : "idle ",
2296                                                slave_dev->name,
2297                                                bond->params.downdelay * bond->params.miimon);
2298                                 }
2299                         }
2300                         /* no break ! fall through the BOND_LINK_FAIL test to
2301                            ensure proper action to be taken
2302                         */
2303                 case BOND_LINK_FAIL:    /* the link has just gone down */
2304                         if (link_state != BMSR_LSTATUS) {
2305                                 /* link stays down */
2306                                 if (slave->delay <= 0) {
2307                                         if (!have_locks)
2308                                                 return 1;
2309
2310                                         /* link down for too long time */
2311                                         slave->link = BOND_LINK_DOWN;
2312
2313                                         /* in active/backup mode, we must
2314                                          * completely disable this interface
2315                                          */
2316                                         if ((bond->params.mode == BOND_MODE_ACTIVEBACKUP) ||
2317                                             (bond->params.mode == BOND_MODE_8023AD)) {
2318                                                 bond_set_slave_inactive_flags(slave);
2319                                         }
2320
2321                                         printk(KERN_INFO DRV_NAME
2322                                                ": %s: link status definitely "
2323                                                "down for interface %s, "
2324                                                "disabling it\n",
2325                                                bond->dev->name,
2326                                                slave_dev->name);
2327
2328                                         /* notify ad that the link status has changed */
2329                                         if (bond->params.mode == BOND_MODE_8023AD) {
2330                                                 bond_3ad_handle_link_change(slave, BOND_LINK_DOWN);
2331                                         }
2332
2333                                         if ((bond->params.mode == BOND_MODE_TLB) ||
2334                                             (bond->params.mode == BOND_MODE_ALB)) {
2335                                                 bond_alb_handle_link_change(bond, slave, BOND_LINK_DOWN);
2336                                         }
2337
2338                                         if (slave == oldcurrent) {
2339                                                 do_failover = 1;
2340                                         }
2341                                 } else {
2342                                         slave->delay--;
2343                                 }
2344                         } else {
2345                                 /* link up again */
2346                                 slave->link  = BOND_LINK_UP;
2347                                 slave->jiffies = jiffies;
2348                                 printk(KERN_INFO DRV_NAME
2349                                        ": %s: link status up again after %d "
2350                                        "ms for interface %s.\n",
2351                                        bond->dev->name,
2352                                        (bond->params.downdelay - slave->delay) * bond->params.miimon,
2353                                        slave_dev->name);
2354                         }
2355                         break;
2356                 case BOND_LINK_DOWN:    /* the link was down */
2357                         if (link_state != BMSR_LSTATUS) {
2358                                 /* the link stays down, nothing more to do */
2359                                 break;
2360                         } else {        /* link going up */
2361                                 slave->link  = BOND_LINK_BACK;
2362                                 slave->delay = bond->params.updelay;
2363
2364                                 if (bond->params.updelay) {
2365                                         /* if updelay == 0, no need to
2366                                            advertise about a 0 ms delay */
2367                                         printk(KERN_INFO DRV_NAME
2368                                                ": %s: link status up for "
2369                                                "interface %s, enabling it "
2370                                                "in %d ms.\n",
2371                                                bond->dev->name,
2372                                                slave_dev->name,
2373                                                bond->params.updelay * bond->params.miimon);
2374                                 }
2375                         }
2376                         /* no break ! fall through the BOND_LINK_BACK state in
2377                            case there's something to do.
2378                         */
2379                 case BOND_LINK_BACK:    /* the link has just come back */
2380                         if (link_state != BMSR_LSTATUS) {
2381                                 /* link down again */
2382                                 slave->link  = BOND_LINK_DOWN;
2383
2384                                 printk(KERN_INFO DRV_NAME
2385                                        ": %s: link status down again after %d "
2386                                        "ms for interface %s.\n",
2387                                        bond->dev->name,
2388                                        (bond->params.updelay - slave->delay) * bond->params.miimon,
2389                                        slave_dev->name);
2390                         } else {
2391                                 /* link stays up */
2392                                 if (slave->delay == 0) {
2393                                         if (!have_locks)
2394                                                 return 1;
2395
2396                                         /* now the link has been up for long time enough */
2397                                         slave->link = BOND_LINK_UP;
2398                                         slave->jiffies = jiffies;
2399
2400                                         if (bond->params.mode == BOND_MODE_8023AD) {
2401                                                 /* prevent it from being the active one */
2402                                                 slave->state = BOND_STATE_BACKUP;
2403                                         } else if (bond->params.mode != BOND_MODE_ACTIVEBACKUP) {
2404                                                 /* make it immediately active */
2405                                                 slave->state = BOND_STATE_ACTIVE;
2406                                         } else if (slave != bond->primary_slave) {
2407                                                 /* prevent it from being the active one */
2408                                                 slave->state = BOND_STATE_BACKUP;
2409                                         }
2410
2411                                         printk(KERN_INFO DRV_NAME
2412                                                ": %s: link status definitely "
2413                                                "up for interface %s.\n",
2414                                                bond->dev->name,
2415                                                slave_dev->name);
2416
2417                                         /* notify ad that the link status has changed */
2418                                         if (bond->params.mode == BOND_MODE_8023AD) {
2419                                                 bond_3ad_handle_link_change(slave, BOND_LINK_UP);
2420                                         }
2421
2422                                         if ((bond->params.mode == BOND_MODE_TLB) ||
2423                                             (bond->params.mode == BOND_MODE_ALB)) {
2424                                                 bond_alb_handle_link_change(bond, slave, BOND_LINK_UP);
2425                                         }
2426
2427                                         if ((!oldcurrent) ||
2428                                             (slave == bond->primary_slave)) {
2429                                                 do_failover = 1;
2430                                         }
2431                                 } else {
2432                                         slave->delay--;
2433                                 }
2434                         }
2435                         break;
2436                 default:
2437                         /* Should not happen */
2438                         printk(KERN_ERR DRV_NAME
2439                                ": %s: Error: %s Illegal value (link=%d)\n",
2440                                bond->dev->name,
2441                                slave->dev->name,
2442                                slave->link);
2443                         goto out;
2444                 } /* end of switch (slave->link) */
2445
2446                 bond_update_speed_duplex(slave);
2447
2448                 if (bond->params.mode == BOND_MODE_8023AD) {
2449                         if (old_speed != slave->speed) {
2450                                 bond_3ad_adapter_speed_changed(slave);
2451                         }
2452
2453                         if (old_duplex != slave->duplex) {
2454                                 bond_3ad_adapter_duplex_changed(slave);
2455                         }
2456                 }
2457
2458         } /* end of for */
2459
2460         if (do_failover) {
2461                 ASSERT_RTNL();
2462
2463                 write_lock_bh(&bond->curr_slave_lock);
2464
2465                 bond_select_active_slave(bond);
2466
2467                 write_unlock_bh(&bond->curr_slave_lock);
2468
2469         } else
2470                 bond_set_carrier(bond);
2471
2472 out:
2473         return 0;
2474 }
2475
2476 /*
2477  * bond_mii_monitor
2478  *
2479  * Really a wrapper that splits the mii monitor into two phases: an
2480  * inspection, then (if inspection indicates something needs to be
2481  * done) an acquisition of appropriate locks followed by another pass
2482  * to implement whatever link state changes are indicated.
2483  */
2484 void bond_mii_monitor(struct work_struct *work)
2485 {
2486         struct bonding *bond = container_of(work, struct bonding,
2487                                             mii_work.work);
2488         unsigned long delay;
2489
2490         read_lock(&bond->lock);
2491         if (bond->kill_timers) {
2492                 read_unlock(&bond->lock);
2493                 return;
2494         }
2495         if (__bond_mii_monitor(bond, 0)) {
2496                 read_unlock(&bond->lock);
2497                 rtnl_lock();
2498                 read_lock(&bond->lock);
2499                 __bond_mii_monitor(bond, 1);
2500                 read_unlock(&bond->lock);
2501                 rtnl_unlock();  /* might sleep, hold no other locks */
2502                 read_lock(&bond->lock);
2503         }
2504
2505         delay = msecs_to_jiffies(bond->params.miimon);
2506         read_unlock(&bond->lock);
2507         queue_delayed_work(bond->wq, &bond->mii_work, delay);
2508 }
2509
2510 static __be32 bond_glean_dev_ip(struct net_device *dev)
2511 {
2512         struct in_device *idev;
2513         struct in_ifaddr *ifa;
2514         __be32 addr = 0;
2515
2516         if (!dev)
2517                 return 0;
2518
2519         rcu_read_lock();
2520         idev = __in_dev_get_rcu(dev);
2521         if (!idev)
2522                 goto out;
2523
2524         ifa = idev->ifa_list;
2525         if (!ifa)
2526                 goto out;
2527
2528         addr = ifa->ifa_local;
2529 out:
2530         rcu_read_unlock();
2531         return addr;
2532 }
2533
2534 static int bond_has_this_ip(struct bonding *bond, __be32 ip)
2535 {
2536         struct vlan_entry *vlan;
2537
2538         if (ip == bond->master_ip)
2539                 return 1;
2540
2541         list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
2542                 if (ip == vlan->vlan_ip)
2543                         return 1;
2544         }
2545
2546         return 0;
2547 }
2548
2549 /*
2550  * We go to the (large) trouble of VLAN tagging ARP frames because
2551  * switches in VLAN mode (especially if ports are configured as
2552  * "native" to a VLAN) might not pass non-tagged frames.
2553  */
2554 static void bond_arp_send(struct net_device *slave_dev, int arp_op, __be32 dest_ip, __be32 src_ip, unsigned short vlan_id)
2555 {
2556         struct sk_buff *skb;
2557
2558         dprintk("arp %d on slave %s: dst %x src %x vid %d\n", arp_op,
2559                slave_dev->name, dest_ip, src_ip, vlan_id);
2560                
2561         skb = arp_create(arp_op, ETH_P_ARP, dest_ip, slave_dev, src_ip,
2562                          NULL, slave_dev->dev_addr, NULL);
2563
2564         if (!skb) {
2565                 printk(KERN_ERR DRV_NAME ": ARP packet allocation failed\n");
2566                 return;
2567         }
2568         if (vlan_id) {
2569                 skb = vlan_put_tag(skb, vlan_id);
2570                 if (!skb) {
2571                         printk(KERN_ERR DRV_NAME ": failed to insert VLAN tag\n");
2572                         return;
2573                 }
2574         }
2575         arp_xmit(skb);
2576 }
2577
2578
2579 static void bond_arp_send_all(struct bonding *bond, struct slave *slave)
2580 {
2581         int i, vlan_id, rv;
2582         __be32 *targets = bond->params.arp_targets;
2583         struct vlan_entry *vlan;
2584         struct net_device *vlan_dev;
2585         struct flowi fl;
2586         struct rtable *rt;
2587
2588         for (i = 0; (i < BOND_MAX_ARP_TARGETS); i++) {
2589                 if (!targets[i])
2590                         continue;
2591                 dprintk("basa: target %x\n", targets[i]);
2592                 if (list_empty(&bond->vlan_list)) {
2593                         dprintk("basa: empty vlan: arp_send\n");
2594                         bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2595                                       bond->master_ip, 0);
2596                         continue;
2597                 }
2598
2599                 /*
2600                  * If VLANs are configured, we do a route lookup to
2601                  * determine which VLAN interface would be used, so we
2602                  * can tag the ARP with the proper VLAN tag.
2603                  */
2604                 memset(&fl, 0, sizeof(fl));
2605                 fl.fl4_dst = targets[i];
2606                 fl.fl4_tos = RTO_ONLINK;
2607
2608                 rv = ip_route_output_key(&init_net, &rt, &fl);
2609                 if (rv) {
2610                         if (net_ratelimit()) {
2611                                 printk(KERN_WARNING DRV_NAME
2612                              ": %s: no route to arp_ip_target %u.%u.%u.%u\n",
2613                                        bond->dev->name, NIPQUAD(fl.fl4_dst));
2614                         }
2615                         continue;
2616                 }
2617
2618                 /*
2619                  * This target is not on a VLAN
2620                  */
2621                 if (rt->u.dst.dev == bond->dev) {
2622                         ip_rt_put(rt);
2623                         dprintk("basa: rtdev == bond->dev: arp_send\n");
2624                         bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2625                                       bond->master_ip, 0);
2626                         continue;
2627                 }
2628
2629                 vlan_id = 0;
2630                 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
2631                         vlan_dev = vlan_group_get_device(bond->vlgrp, vlan->vlan_id);
2632                         if (vlan_dev == rt->u.dst.dev) {
2633                                 vlan_id = vlan->vlan_id;
2634                                 dprintk("basa: vlan match on %s %d\n",
2635                                        vlan_dev->name, vlan_id);
2636                                 break;
2637                         }
2638                 }
2639
2640                 if (vlan_id) {
2641                         ip_rt_put(rt);
2642                         bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2643                                       vlan->vlan_ip, vlan_id);
2644                         continue;
2645                 }
2646
2647                 if (net_ratelimit()) {
2648                         printk(KERN_WARNING DRV_NAME
2649                ": %s: no path to arp_ip_target %u.%u.%u.%u via rt.dev %s\n",
2650                                bond->dev->name, NIPQUAD(fl.fl4_dst),
2651                                rt->u.dst.dev ? rt->u.dst.dev->name : "NULL");
2652                 }
2653                 ip_rt_put(rt);
2654         }
2655 }
2656
2657 /*
2658  * Kick out a gratuitous ARP for an IP on the bonding master plus one
2659  * for each VLAN above us.
2660  */
2661 static void bond_send_gratuitous_arp(struct bonding *bond)
2662 {
2663         struct slave *slave = bond->curr_active_slave;
2664         struct vlan_entry *vlan;
2665         struct net_device *vlan_dev;
2666
2667         dprintk("bond_send_grat_arp: bond %s slave %s\n", bond->dev->name,
2668                                 slave ? slave->dev->name : "NULL");
2669         if (!slave)
2670                 return;
2671
2672         if (bond->master_ip) {
2673                 bond_arp_send(slave->dev, ARPOP_REPLY, bond->master_ip,
2674                                 bond->master_ip, 0);
2675         }
2676
2677         list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
2678                 vlan_dev = vlan_group_get_device(bond->vlgrp, vlan->vlan_id);
2679                 if (vlan->vlan_ip) {
2680                         bond_arp_send(slave->dev, ARPOP_REPLY, vlan->vlan_ip,
2681                                       vlan->vlan_ip, vlan->vlan_id);
2682                 }
2683         }
2684 }
2685
2686 static void bond_validate_arp(struct bonding *bond, struct slave *slave, __be32 sip, __be32 tip)
2687 {
2688         int i;
2689         __be32 *targets = bond->params.arp_targets;
2690
2691         targets = bond->params.arp_targets;
2692         for (i = 0; (i < BOND_MAX_ARP_TARGETS) && targets[i]; i++) {
2693                 dprintk("bva: sip %u.%u.%u.%u tip %u.%u.%u.%u t[%d] "
2694                         "%u.%u.%u.%u bhti(tip) %d\n",
2695                        NIPQUAD(sip), NIPQUAD(tip), i, NIPQUAD(targets[i]),
2696                        bond_has_this_ip(bond, tip));
2697                 if (sip == targets[i]) {
2698                         if (bond_has_this_ip(bond, tip))
2699                                 slave->last_arp_rx = jiffies;
2700                         return;
2701                 }
2702         }
2703 }
2704
2705 static int bond_arp_rcv(struct sk_buff *skb, struct net_device *dev, struct packet_type *pt, struct net_device *orig_dev)
2706 {
2707         struct arphdr *arp;
2708         struct slave *slave;
2709         struct bonding *bond;
2710         unsigned char *arp_ptr;
2711         __be32 sip, tip;
2712
2713         if (dev_net(dev) != &init_net)
2714                 goto out;
2715
2716         if (!(dev->priv_flags & IFF_BONDING) || !(dev->flags & IFF_MASTER))
2717                 goto out;
2718
2719         bond = dev->priv;
2720         read_lock(&bond->lock);
2721
2722         dprintk("bond_arp_rcv: bond %s skb->dev %s orig_dev %s\n",
2723                 bond->dev->name, skb->dev ? skb->dev->name : "NULL",
2724                 orig_dev ? orig_dev->name : "NULL");
2725
2726         slave = bond_get_slave_by_dev(bond, orig_dev);
2727         if (!slave || !slave_do_arp_validate(bond, slave))
2728                 goto out_unlock;
2729
2730         if (!pskb_may_pull(skb, arp_hdr_len(dev)))
2731                 goto out_unlock;
2732
2733         arp = arp_hdr(skb);
2734         if (arp->ar_hln != dev->addr_len ||
2735             skb->pkt_type == PACKET_OTHERHOST ||
2736             skb->pkt_type == PACKET_LOOPBACK ||
2737             arp->ar_hrd != htons(ARPHRD_ETHER) ||
2738             arp->ar_pro != htons(ETH_P_IP) ||
2739             arp->ar_pln != 4)
2740                 goto out_unlock;
2741
2742         arp_ptr = (unsigned char *)(arp + 1);
2743         arp_ptr += dev->addr_len;
2744         memcpy(&sip, arp_ptr, 4);
2745         arp_ptr += 4 + dev->addr_len;
2746         memcpy(&tip, arp_ptr, 4);
2747
2748         dprintk("bond_arp_rcv: %s %s/%d av %d sv %d sip %u.%u.%u.%u"
2749                 " tip %u.%u.%u.%u\n", bond->dev->name, slave->dev->name,
2750                 slave->state, bond->params.arp_validate,
2751                 slave_do_arp_validate(bond, slave), NIPQUAD(sip), NIPQUAD(tip));
2752
2753         /*
2754          * Backup slaves won't see the ARP reply, but do come through
2755          * here for each ARP probe (so we swap the sip/tip to validate
2756          * the probe).  In a "redundant switch, common router" type of
2757          * configuration, the ARP probe will (hopefully) travel from
2758          * the active, through one switch, the router, then the other
2759          * switch before reaching the backup.
2760          */
2761         if (slave->state == BOND_STATE_ACTIVE)
2762                 bond_validate_arp(bond, slave, sip, tip);
2763         else
2764                 bond_validate_arp(bond, slave, tip, sip);
2765
2766 out_unlock:
2767         read_unlock(&bond->lock);
2768 out:
2769         dev_kfree_skb(skb);
2770         return NET_RX_SUCCESS;
2771 }
2772
2773 /*
2774  * this function is called regularly to monitor each slave's link
2775  * ensuring that traffic is being sent and received when arp monitoring
2776  * is used in load-balancing mode. if the adapter has been dormant, then an
2777  * arp is transmitted to generate traffic. see activebackup_arp_monitor for
2778  * arp monitoring in active backup mode.
2779  */
2780 void bond_loadbalance_arp_mon(struct work_struct *work)
2781 {
2782         struct bonding *bond = container_of(work, struct bonding,
2783                                             arp_work.work);
2784         struct slave *slave, *oldcurrent;
2785         int do_failover = 0;
2786         int delta_in_ticks;
2787         int i;
2788
2789         read_lock(&bond->lock);
2790
2791         delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
2792
2793         if (bond->kill_timers) {
2794                 goto out;
2795         }
2796
2797         if (bond->slave_cnt == 0) {
2798                 goto re_arm;
2799         }
2800
2801         read_lock(&bond->curr_slave_lock);
2802         oldcurrent = bond->curr_active_slave;
2803         read_unlock(&bond->curr_slave_lock);
2804
2805         /* see if any of the previous devices are up now (i.e. they have
2806          * xmt and rcv traffic). the curr_active_slave does not come into
2807          * the picture unless it is null. also, slave->jiffies is not needed
2808          * here because we send an arp on each slave and give a slave as
2809          * long as it needs to get the tx/rx within the delta.
2810          * TODO: what about up/down delay in arp mode? it wasn't here before
2811          *       so it can wait
2812          */
2813         bond_for_each_slave(bond, slave, i) {
2814                 if (slave->link != BOND_LINK_UP) {
2815                         if (time_before_eq(jiffies, slave->dev->trans_start + delta_in_ticks) &&
2816                             time_before_eq(jiffies, slave->dev->last_rx + delta_in_ticks)) {
2817
2818                                 slave->link  = BOND_LINK_UP;
2819                                 slave->state = BOND_STATE_ACTIVE;
2820
2821                                 /* primary_slave has no meaning in round-robin
2822                                  * mode. the window of a slave being up and
2823                                  * curr_active_slave being null after enslaving
2824                                  * is closed.
2825                                  */
2826                                 if (!oldcurrent) {
2827                                         printk(KERN_INFO DRV_NAME
2828                                                ": %s: link status definitely "
2829                                                "up for interface %s, ",
2830                                                bond->dev->name,
2831                                                slave->dev->name);
2832                                         do_failover = 1;
2833                                 } else {
2834                                         printk(KERN_INFO DRV_NAME
2835                                                ": %s: interface %s is now up\n",
2836                                                bond->dev->name,
2837                                                slave->dev->name);
2838                                 }
2839                         }
2840                 } else {
2841                         /* slave->link == BOND_LINK_UP */
2842
2843                         /* not all switches will respond to an arp request
2844                          * when the source ip is 0, so don't take the link down
2845                          * if we don't know our ip yet
2846                          */
2847                         if (time_after_eq(jiffies, slave->dev->trans_start + 2*delta_in_ticks) ||
2848                             (time_after_eq(jiffies, slave->dev->last_rx + 2*delta_in_ticks))) {
2849
2850                                 slave->link  = BOND_LINK_DOWN;
2851                                 slave->state = BOND_STATE_BACKUP;
2852
2853                                 if (slave->link_failure_count < UINT_MAX) {
2854                                         slave->link_failure_count++;
2855                                 }
2856
2857                                 printk(KERN_INFO DRV_NAME
2858                                        ": %s: interface %s is now down.\n",
2859                                        bond->dev->name,
2860                                        slave->dev->name);
2861
2862                                 if (slave == oldcurrent) {
2863                                         do_failover = 1;
2864                                 }
2865                         }
2866                 }
2867
2868                 /* note: if switch is in round-robin mode, all links
2869                  * must tx arp to ensure all links rx an arp - otherwise
2870                  * links may oscillate or not come up at all; if switch is
2871                  * in something like xor mode, there is nothing we can
2872                  * do - all replies will be rx'ed on same link causing slaves
2873                  * to be unstable during low/no traffic periods
2874                  */
2875                 if (IS_UP(slave->dev)) {
2876                         bond_arp_send_all(bond, slave);
2877                 }
2878         }
2879
2880         if (do_failover) {
2881                 write_lock_bh(&bond->curr_slave_lock);
2882
2883                 bond_select_active_slave(bond);
2884
2885                 write_unlock_bh(&bond->curr_slave_lock);
2886         }
2887
2888 re_arm:
2889         if (bond->params.arp_interval)
2890                 queue_delayed_work(bond->wq, &bond->arp_work, delta_in_ticks);
2891 out:
2892         read_unlock(&bond->lock);
2893 }
2894
2895 /*
2896  * Called to inspect slaves for active-backup mode ARP monitor link state
2897  * changes.  Sets new_link in slaves to specify what action should take
2898  * place for the slave.  Returns 0 if no changes are found, >0 if changes
2899  * to link states must be committed.
2900  *
2901  * Called with bond->lock held for read.
2902  */
2903 static int bond_ab_arp_inspect(struct bonding *bond, int delta_in_ticks)
2904 {
2905         struct slave *slave;
2906         int i, commit = 0;
2907
2908         bond_for_each_slave(bond, slave, i) {
2909                 slave->new_link = BOND_LINK_NOCHANGE;
2910
2911                 if (slave->link != BOND_LINK_UP) {
2912                         if (time_before_eq(jiffies, slave_last_rx(bond, slave) +
2913                                            delta_in_ticks)) {
2914                                 slave->new_link = BOND_LINK_UP;
2915                                 commit++;
2916                         }
2917
2918                         continue;
2919                 }
2920
2921                 /*
2922                  * Give slaves 2*delta after being enslaved or made
2923                  * active.  This avoids bouncing, as the last receive
2924                  * times need a full ARP monitor cycle to be updated.
2925                  */
2926                 if (!time_after_eq(jiffies, slave->jiffies +
2927                                    2 * delta_in_ticks))
2928                         continue;
2929
2930                 /*
2931                  * Backup slave is down if:
2932                  * - No current_arp_slave AND
2933                  * - more than 3*delta since last receive AND
2934                  * - the bond has an IP address
2935                  *
2936                  * Note: a non-null current_arp_slave indicates
2937                  * the curr_active_slave went down and we are
2938                  * searching for a new one; under this condition
2939                  * we only take the curr_active_slave down - this
2940                  * gives each slave a chance to tx/rx traffic
2941                  * before being taken out
2942                  */
2943                 if (slave->state == BOND_STATE_BACKUP &&
2944                     !bond->current_arp_slave &&
2945                     time_after(jiffies, slave_last_rx(bond, slave) +
2946                                3 * delta_in_ticks)) {
2947                         slave->new_link = BOND_LINK_DOWN;
2948                         commit++;
2949                 }
2950
2951                 /*
2952                  * Active slave is down if:
2953                  * - more than 2*delta since transmitting OR
2954                  * - (more than 2*delta since receive AND
2955                  *    the bond has an IP address)
2956                  */
2957                 if ((slave->state == BOND_STATE_ACTIVE) &&
2958                     (time_after_eq(jiffies, slave->dev->trans_start +
2959                                     2 * delta_in_ticks) ||
2960                       (time_after_eq(jiffies, slave_last_rx(bond, slave)
2961                                      + 2 * delta_in_ticks)))) {
2962                         slave->new_link = BOND_LINK_DOWN;
2963                         commit++;
2964                 }
2965         }
2966
2967         read_lock(&bond->curr_slave_lock);
2968
2969         /*
2970          * Trigger a commit if the primary option setting has changed.
2971          */
2972         if (bond->primary_slave &&
2973             (bond->primary_slave != bond->curr_active_slave) &&
2974             (bond->primary_slave->link == BOND_LINK_UP))
2975                 commit++;
2976
2977         read_unlock(&bond->curr_slave_lock);
2978
2979         return commit;
2980 }
2981
2982 /*
2983  * Called to commit link state changes noted by inspection step of
2984  * active-backup mode ARP monitor.
2985  *
2986  * Called with RTNL and bond->lock for read.
2987  */
2988 static void bond_ab_arp_commit(struct bonding *bond, int delta_in_ticks)
2989 {
2990         struct slave *slave;
2991         int i;
2992
2993         bond_for_each_slave(bond, slave, i) {
2994                 switch (slave->new_link) {
2995                 case BOND_LINK_NOCHANGE:
2996                         continue;
2997
2998                 case BOND_LINK_UP:
2999                         write_lock_bh(&bond->curr_slave_lock);
3000
3001                         if (!bond->curr_active_slave &&
3002                             time_before_eq(jiffies, slave->dev->trans_start +
3003                                            delta_in_ticks)) {
3004                                 slave->link = BOND_LINK_UP;
3005                                 bond_change_active_slave(bond, slave);
3006                                 bond->current_arp_slave = NULL;
3007
3008                                 printk(KERN_INFO DRV_NAME
3009                                        ": %s: %s is up and now the "
3010                                        "active interface\n",
3011                                        bond->dev->name, slave->dev->name);
3012
3013                         } else if (bond->curr_active_slave != slave) {
3014                                 /* this slave has just come up but we
3015                                  * already have a current slave; this can
3016                                  * also happen if bond_enslave adds a new
3017                                  * slave that is up while we are searching
3018                                  * for a new slave
3019                                  */
3020                                 slave->link = BOND_LINK_UP;
3021                                 bond_set_slave_inactive_flags(slave);
3022                                 bond->current_arp_slave = NULL;
3023
3024                                 printk(KERN_INFO DRV_NAME
3025                                        ": %s: backup interface %s is now up\n",
3026                                        bond->dev->name, slave->dev->name);
3027                         }
3028
3029                         write_unlock_bh(&bond->curr_slave_lock);
3030
3031                         break;
3032
3033                 case BOND_LINK_DOWN:
3034                         if (slave->link_failure_count < UINT_MAX)
3035                                 slave->link_failure_count++;
3036
3037                         slave->link = BOND_LINK_DOWN;
3038
3039                         if (slave == bond->curr_active_slave) {
3040                                 printk(KERN_INFO DRV_NAME
3041                                        ": %s: link status down for active "
3042                                        "interface %s, disabling it\n",
3043                                        bond->dev->name, slave->dev->name);
3044
3045                                 bond_set_slave_inactive_flags(slave);
3046
3047                                 write_lock_bh(&bond->curr_slave_lock);
3048
3049                                 bond_select_active_slave(bond);
3050                                 if (bond->curr_active_slave)
3051                                         bond->curr_active_slave->jiffies =
3052                                                 jiffies;
3053
3054                                 write_unlock_bh(&bond->curr_slave_lock);
3055
3056                                 bond->current_arp_slave = NULL;
3057
3058                         } else if (slave->state == BOND_STATE_BACKUP) {
3059                                 printk(KERN_INFO DRV_NAME
3060                                        ": %s: backup interface %s is now down\n",
3061                                        bond->dev->name, slave->dev->name);
3062
3063                                 bond_set_slave_inactive_flags(slave);
3064                         }
3065                         break;
3066
3067                 default:
3068                         printk(KERN_ERR DRV_NAME
3069                                ": %s: impossible: new_link %d on slave %s\n",
3070                                bond->dev->name, slave->new_link,
3071                                slave->dev->name);
3072                 }
3073         }
3074
3075         /*
3076          * No race with changes to primary via sysfs, as we hold rtnl.
3077          */
3078         if (bond->primary_slave &&
3079             (bond->primary_slave != bond->curr_active_slave) &&
3080             (bond->primary_slave->link == BOND_LINK_UP)) {
3081                 write_lock_bh(&bond->curr_slave_lock);
3082                 bond_change_active_slave(bond, bond->primary_slave);
3083                 write_unlock_bh(&bond->curr_slave_lock);
3084         }
3085
3086         bond_set_carrier(bond);
3087 }
3088
3089 /*
3090  * Send ARP probes for active-backup mode ARP monitor.
3091  *
3092  * Called with bond->lock held for read.
3093  */
3094 static void bond_ab_arp_probe(struct bonding *bond)
3095 {
3096         struct slave *slave;
3097         int i;
3098
3099         read_lock(&bond->curr_slave_lock);
3100
3101         if (bond->current_arp_slave && bond->curr_active_slave)
3102                 printk("PROBE: c_arp %s && cas %s BAD\n",
3103                        bond->current_arp_slave->dev->name,
3104                        bond->curr_active_slave->dev->name);
3105
3106         if (bond->curr_active_slave) {
3107                 bond_arp_send_all(bond, bond->curr_active_slave);
3108                 read_unlock(&bond->curr_slave_lock);
3109                 return;
3110         }
3111
3112         read_unlock(&bond->curr_slave_lock);
3113
3114         /* if we don't have a curr_active_slave, search for the next available
3115          * backup slave from the current_arp_slave and make it the candidate
3116          * for becoming the curr_active_slave
3117          */
3118
3119         if (!bond->current_arp_slave) {
3120                 bond->current_arp_slave = bond->first_slave;
3121                 if (!bond->current_arp_slave)
3122                         return;
3123         }
3124
3125         bond_set_slave_inactive_flags(bond->current_arp_slave);
3126
3127         /* search for next candidate */
3128         bond_for_each_slave_from(bond, slave, i, bond->current_arp_slave->next) {
3129                 if (IS_UP(slave->dev)) {
3130                         slave->link = BOND_LINK_BACK;
3131                         bond_set_slave_active_flags(slave);
3132                         bond_arp_send_all(bond, slave);
3133                         slave->jiffies = jiffies;
3134                         bond->current_arp_slave = slave;
3135                         break;
3136                 }
3137
3138                 /* if the link state is up at this point, we
3139                  * mark it down - this can happen if we have
3140                  * simultaneous link failures and
3141                  * reselect_active_interface doesn't make this
3142                  * one the current slave so it is still marked
3143                  * up when it is actually down
3144                  */
3145                 if (slave->link == BOND_LINK_UP) {
3146                         slave->link = BOND_LINK_DOWN;
3147                         if (slave->link_failure_count < UINT_MAX)
3148                                 slave->link_failure_count++;
3149
3150                         bond_set_slave_inactive_flags(slave);
3151
3152                         printk(KERN_INFO DRV_NAME
3153                                ": %s: backup interface %s is now down.\n",
3154                                bond->dev->name, slave->dev->name);
3155                 }
3156         }
3157 }
3158
3159 void bond_activebackup_arp_mon(struct work_struct *work)
3160 {
3161         struct bonding *bond = container_of(work, struct bonding,
3162                                             arp_work.work);
3163         int delta_in_ticks;
3164
3165         read_lock(&bond->lock);
3166
3167         if (bond->kill_timers)
3168                 goto out;
3169
3170         delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
3171
3172         if (bond->slave_cnt == 0)
3173                 goto re_arm;
3174
3175         if (bond_ab_arp_inspect(bond, delta_in_ticks)) {
3176                 read_unlock(&bond->lock);
3177                 rtnl_lock();
3178                 read_lock(&bond->lock);
3179
3180                 bond_ab_arp_commit(bond, delta_in_ticks);
3181
3182                 read_unlock(&bond->lock);
3183                 rtnl_unlock();
3184                 read_lock(&bond->lock);
3185         }
3186
3187         bond_ab_arp_probe(bond);
3188
3189 re_arm:
3190         if (bond->params.arp_interval) {
3191                 queue_delayed_work(bond->wq, &bond->arp_work, delta_in_ticks);
3192         }
3193 out:
3194         read_unlock(&bond->lock);
3195 }
3196
3197 /*------------------------------ proc/seq_file-------------------------------*/
3198
3199 #ifdef CONFIG_PROC_FS
3200
3201 static void *bond_info_seq_start(struct seq_file *seq, loff_t *pos)
3202 {
3203         struct bonding *bond = seq->private;
3204         loff_t off = 0;
3205         struct slave *slave;
3206         int i;
3207
3208         /* make sure the bond won't be taken away */
3209         read_lock(&dev_base_lock);
3210         read_lock(&bond->lock);
3211
3212         if (*pos == 0) {
3213                 return SEQ_START_TOKEN;
3214         }
3215
3216         bond_for_each_slave(bond, slave, i) {
3217                 if (++off == *pos) {
3218                         return slave;
3219                 }
3220         }
3221
3222         return NULL;
3223 }
3224
3225 static void *bond_info_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3226 {
3227         struct bonding *bond = seq->private;
3228         struct slave *slave = v;
3229
3230         ++*pos;
3231         if (v == SEQ_START_TOKEN) {
3232                 return bond->first_slave;
3233         }
3234
3235         slave = slave->next;
3236
3237         return (slave == bond->first_slave) ? NULL : slave;
3238 }
3239
3240 static void bond_info_seq_stop(struct seq_file *seq, void *v)
3241 {
3242         struct bonding *bond = seq->private;
3243
3244         read_unlock(&bond->lock);
3245         read_unlock(&dev_base_lock);
3246 }
3247
3248 static void bond_info_show_master(struct seq_file *seq)
3249 {
3250         struct bonding *bond = seq->private;
3251         struct slave *curr;
3252         int i;
3253         u32 target;
3254
3255         read_lock(&bond->curr_slave_lock);
3256         curr = bond->curr_active_slave;
3257         read_unlock(&bond->curr_slave_lock);
3258
3259         seq_printf(seq, "Bonding Mode: %s",
3260                    bond_mode_name(bond->params.mode));
3261
3262         if (bond->params.mode == BOND_MODE_ACTIVEBACKUP &&
3263             bond->params.fail_over_mac)
3264                 seq_printf(seq, " (fail_over_mac %s)",
3265                    fail_over_mac_tbl[bond->params.fail_over_mac].modename);
3266
3267         seq_printf(seq, "\n");
3268
3269         if (bond->params.mode == BOND_MODE_XOR ||
3270                 bond->params.mode == BOND_MODE_8023AD) {
3271                 seq_printf(seq, "Transmit Hash Policy: %s (%d)\n",
3272                         xmit_hashtype_tbl[bond->params.xmit_policy].modename,
3273                         bond->params.xmit_policy);
3274         }
3275
3276         if (USES_PRIMARY(bond->params.mode)) {
3277                 seq_printf(seq, "Primary Slave: %s\n",
3278                            (bond->primary_slave) ?
3279                            bond->primary_slave->dev->name : "None");
3280
3281                 seq_printf(seq, "Currently Active Slave: %s\n",
3282                            (curr) ? curr->dev->name : "None");
3283         }
3284
3285         seq_printf(seq, "MII Status: %s\n", netif_carrier_ok(bond->dev) ?
3286                    "up" : "down");
3287         seq_printf(seq, "MII Polling Interval (ms): %d\n", bond->params.miimon);
3288         seq_printf(seq, "Up Delay (ms): %d\n",
3289                    bond->params.updelay * bond->params.miimon);
3290         seq_printf(seq, "Down Delay (ms): %d\n",
3291                    bond->params.downdelay * bond->params.miimon);
3292
3293
3294         /* ARP information */
3295         if(bond->params.arp_interval > 0) {
3296                 int printed=0;
3297                 seq_printf(seq, "ARP Polling Interval (ms): %d\n",
3298                                 bond->params.arp_interval);
3299
3300                 seq_printf(seq, "ARP IP target/s (n.n.n.n form):");
3301
3302                 for(i = 0; (i < BOND_MAX_ARP_TARGETS) ;i++) {
3303                         if (!bond->params.arp_targets[i])
3304                                 continue;
3305                         if (printed)
3306                                 seq_printf(seq, ",");
3307                         target = ntohl(bond->params.arp_targets[i]);
3308                         seq_printf(seq, " %d.%d.%d.%d", HIPQUAD(target));
3309                         printed = 1;
3310                 }
3311                 seq_printf(seq, "\n");
3312         }
3313
3314         if (bond->params.mode == BOND_MODE_8023AD) {
3315                 struct ad_info ad_info;
3316                 DECLARE_MAC_BUF(mac);
3317
3318                 seq_puts(seq, "\n802.3ad info\n");
3319                 seq_printf(seq, "LACP rate: %s\n",
3320                            (bond->params.lacp_fast) ? "fast" : "slow");
3321
3322                 if (bond_3ad_get_active_agg_info(bond, &ad_info)) {
3323                         seq_printf(seq, "bond %s has no active aggregator\n",
3324                                    bond->dev->name);
3325                 } else {
3326                         seq_printf(seq, "Active Aggregator Info:\n");
3327
3328                         seq_printf(seq, "\tAggregator ID: %d\n",
3329                                    ad_info.aggregator_id);
3330                         seq_printf(seq, "\tNumber of ports: %d\n",
3331                                    ad_info.ports);
3332                         seq_printf(seq, "\tActor Key: %d\n",
3333                                    ad_info.actor_key);
3334                         seq_printf(seq, "\tPartner Key: %d\n",
3335                                    ad_info.partner_key);
3336                         seq_printf(seq, "\tPartner Mac Address: %s\n",
3337                                    print_mac(mac, ad_info.partner_system));
3338                 }
3339         }
3340 }
3341
3342 static void bond_info_show_slave(struct seq_file *seq, const struct slave *slave)
3343 {
3344         struct bonding *bond = seq->private;
3345         DECLARE_MAC_BUF(mac);
3346
3347         seq_printf(seq, "\nSlave Interface: %s\n", slave->dev->name);
3348         seq_printf(seq, "MII Status: %s\n",
3349                    (slave->link == BOND_LINK_UP) ?  "up" : "down");
3350         seq_printf(seq, "Link Failure Count: %u\n",
3351                    slave->link_failure_count);
3352
3353         seq_printf(seq,
3354                    "Permanent HW addr: %s\n",
3355                    print_mac(mac, slave->perm_hwaddr));
3356
3357         if (bond->params.mode == BOND_MODE_8023AD) {
3358                 const struct aggregator *agg
3359                         = SLAVE_AD_INFO(slave).port.aggregator;
3360
3361                 if (agg) {
3362                         seq_printf(seq, "Aggregator ID: %d\n",
3363                                    agg->aggregator_identifier);
3364                 } else {
3365                         seq_puts(seq, "Aggregator ID: N/A\n");
3366                 }
3367         }
3368 }
3369
3370 static int bond_info_seq_show(struct seq_file *seq, void *v)
3371 {
3372         if (v == SEQ_START_TOKEN) {
3373                 seq_printf(seq, "%s\n", version);
3374                 bond_info_show_master(seq);
3375         } else {
3376                 bond_info_show_slave(seq, v);
3377         }
3378
3379         return 0;
3380 }
3381
3382 static struct seq_operations bond_info_seq_ops = {
3383         .start = bond_info_seq_start,
3384         .next  = bond_info_seq_next,
3385         .stop  = bond_info_seq_stop,
3386         .show  = bond_info_seq_show,
3387 };
3388
3389 static int bond_info_open(struct inode *inode, struct file *file)
3390 {
3391         struct seq_file *seq;
3392         struct proc_dir_entry *proc;
3393         int res;
3394
3395         res = seq_open(file, &bond_info_seq_ops);
3396         if (!res) {
3397                 /* recover the pointer buried in proc_dir_entry data */
3398                 seq = file->private_data;
3399                 proc = PDE(inode);
3400                 seq->private = proc->data;
3401         }
3402
3403         return res;
3404 }
3405
3406 static const struct file_operations bond_info_fops = {
3407         .owner   = THIS_MODULE,
3408         .open    = bond_info_open,
3409         .read    = seq_read,
3410         .llseek  = seq_lseek,
3411         .release = seq_release,
3412 };
3413
3414 static int bond_create_proc_entry(struct bonding *bond)
3415 {
3416         struct net_device *bond_dev = bond->dev;
3417
3418         if (bond_proc_dir) {
3419                 bond->proc_entry = proc_create_data(bond_dev->name,
3420                                                     S_IRUGO, bond_proc_dir,
3421                                                     &bond_info_fops, bond);
3422                 if (bond->proc_entry == NULL) {
3423                         printk(KERN_WARNING DRV_NAME
3424                                ": Warning: Cannot create /proc/net/%s/%s\n",
3425                                DRV_NAME, bond_dev->name);
3426                 } else {
3427                         memcpy(bond->proc_file_name, bond_dev->name, IFNAMSIZ);
3428                 }
3429         }
3430
3431         return 0;
3432 }
3433
3434 static void bond_remove_proc_entry(struct bonding *bond)
3435 {
3436         if (bond_proc_dir && bond->proc_entry) {
3437                 remove_proc_entry(bond->proc_file_name, bond_proc_dir);
3438                 memset(bond->proc_file_name, 0, IFNAMSIZ);
3439                 bond->proc_entry = NULL;
3440         }
3441 }
3442
3443 /* Create the bonding directory under /proc/net, if doesn't exist yet.
3444  * Caller must hold rtnl_lock.
3445  */
3446 static void bond_create_proc_dir(void)
3447 {
3448         int len = strlen(DRV_NAME);
3449
3450         for (bond_proc_dir = init_net.proc_net->subdir; bond_proc_dir;
3451              bond_proc_dir = bond_proc_dir->next) {
3452                 if ((bond_proc_dir->namelen == len) &&
3453                     !memcmp(bond_proc_dir->name, DRV_NAME, len)) {
3454                         break;
3455                 }
3456         }
3457
3458         if (!bond_proc_dir) {
3459                 bond_proc_dir = proc_mkdir(DRV_NAME, init_net.proc_net);
3460                 if (bond_proc_dir) {
3461                         bond_proc_dir->owner = THIS_MODULE;
3462                 } else {
3463                         printk(KERN_WARNING DRV_NAME
3464                                 ": Warning: cannot create /proc/net/%s\n",
3465                                 DRV_NAME);
3466                 }
3467         }
3468 }
3469
3470 /* Destroy the bonding directory under /proc/net, if empty.
3471  * Caller must hold rtnl_lock.
3472  */
3473 static void bond_destroy_proc_dir(void)
3474 {
3475         struct proc_dir_entry *de;
3476
3477         if (!bond_proc_dir) {
3478                 return;
3479         }
3480
3481         /* verify that the /proc dir is empty */
3482         for (de = bond_proc_dir->subdir; de; de = de->next) {
3483                 /* ignore . and .. */
3484                 if (*(de->name) != '.') {
3485                         break;
3486                 }
3487         }
3488
3489         if (de) {
3490                 if (bond_proc_dir->owner == THIS_MODULE) {
3491                         bond_proc_dir->owner = NULL;
3492                 }
3493         } else {
3494                 remove_proc_entry(DRV_NAME, init_net.proc_net);
3495                 bond_proc_dir = NULL;
3496         }
3497 }
3498 #endif /* CONFIG_PROC_FS */
3499
3500 /*-------------------------- netdev event handling --------------------------*/
3501
3502 /*
3503  * Change device name
3504  */
3505 static int bond_event_changename(struct bonding *bond)
3506 {
3507 #ifdef CONFIG_PROC_FS
3508         bond_remove_proc_entry(bond);
3509         bond_create_proc_entry(bond);
3510 #endif
3511         down_write(&(bonding_rwsem));
3512         bond_destroy_sysfs_entry(bond);
3513         bond_create_sysfs_entry(bond);
3514         up_write(&(bonding_rwsem));
3515         return NOTIFY_DONE;
3516 }
3517
3518 static int bond_master_netdev_event(unsigned long event, struct net_device *bond_dev)
3519 {
3520         struct bonding *event_bond = bond_dev->priv;
3521
3522         switch (event) {
3523         case NETDEV_CHANGENAME:
3524                 return bond_event_changename(event_bond);
3525         case NETDEV_UNREGISTER:
3526                 bond_release_all(event_bond->dev);
3527                 break;
3528         default:
3529                 break;
3530         }
3531
3532         return NOTIFY_DONE;
3533 }
3534
3535 static int bond_slave_netdev_event(unsigned long event, struct net_device *slave_dev)
3536 {
3537         struct net_device *bond_dev = slave_dev->master;
3538         struct bonding *bond = bond_dev->priv;
3539
3540         switch (event) {
3541         case NETDEV_UNREGISTER:
3542                 if (bond_dev) {
3543                         if (bond->setup_by_slave)
3544                                 bond_release_and_destroy(bond_dev, slave_dev);
3545                         else
3546                                 bond_release(bond_dev, slave_dev);
3547                 }
3548                 break;
3549         case NETDEV_CHANGE:
3550                 /*
3551                  * TODO: is this what we get if somebody
3552                  * sets up a hierarchical bond, then rmmod's
3553                  * one of the slave bonding devices?
3554                  */
3555                 break;
3556         case NETDEV_DOWN:
3557                 /*
3558                  * ... Or is it this?
3559                  */
3560                 break;
3561         case NETDEV_CHANGEMTU:
3562                 /*
3563                  * TODO: Should slaves be allowed to
3564                  * independently alter their MTU?  For
3565                  * an active-backup bond, slaves need
3566                  * not be the same type of device, so
3567                  * MTUs may vary.  For other modes,
3568                  * slaves arguably should have the
3569                  * same MTUs. To do this, we'd need to
3570                  * take over the slave's change_mtu
3571                  * function for the duration of their
3572                  * servitude.
3573                  */
3574                 break;
3575         case NETDEV_CHANGENAME:
3576                 /*
3577                  * TODO: handle changing the primary's name
3578                  */
3579                 break;
3580         case NETDEV_FEAT_CHANGE:
3581                 bond_compute_features(bond);
3582                 break;
3583         default:
3584                 break;
3585         }
3586
3587         return NOTIFY_DONE;
3588 }
3589
3590 /*
3591  * bond_netdev_event: handle netdev notifier chain events.
3592  *
3593  * This function receives events for the netdev chain.  The caller (an
3594  * ioctl handler calling blocking_notifier_call_chain) holds the necessary
3595  * locks for us to safely manipulate the slave devices (RTNL lock,
3596  * dev_probe_lock).
3597  */
3598 static int bond_netdev_event(struct notifier_block *this, unsigned long event, void *ptr)
3599 {
3600         struct net_device *event_dev = (struct net_device *)ptr;
3601
3602         if (dev_net(event_dev) != &init_net)
3603                 return NOTIFY_DONE;
3604
3605         dprintk("event_dev: %s, event: %lx\n",
3606                 (event_dev ? event_dev->name : "None"),
3607                 event);
3608
3609         if (!(event_dev->priv_flags & IFF_BONDING))
3610                 return NOTIFY_DONE;
3611
3612         if (event_dev->flags & IFF_MASTER) {
3613                 dprintk("IFF_MASTER\n");
3614                 return bond_master_netdev_event(event, event_dev);
3615         }
3616
3617         if (event_dev->flags & IFF_SLAVE) {
3618                 dprintk("IFF_SLAVE\n");
3619                 return bond_slave_netdev_event(event, event_dev);
3620         }
3621
3622         return NOTIFY_DONE;
3623 }
3624
3625 /*
3626  * bond_inetaddr_event: handle inetaddr notifier chain events.
3627  *
3628  * We keep track of device IPs primarily to use as source addresses in
3629  * ARP monitor probes (rather than spewing out broadcasts all the time).
3630  *
3631  * We track one IP for the main device (if it has one), plus one per VLAN.
3632  */
3633 static int bond_inetaddr_event(struct notifier_block *this, unsigned long event, void *ptr)
3634 {
3635         struct in_ifaddr *ifa = ptr;
3636         struct net_device *vlan_dev, *event_dev = ifa->ifa_dev->dev;
3637         struct bonding *bond;
3638         struct vlan_entry *vlan;
3639
3640         if (dev_net(ifa->ifa_dev->dev) != &init_net)
3641                 return NOTIFY_DONE;
3642
3643         list_for_each_entry(bond, &bond_dev_list, bond_list) {
3644                 if (bond->dev == event_dev) {
3645                         switch (event) {
3646                         case NETDEV_UP:
3647                                 bond->master_ip = ifa->ifa_local;
3648                                 return NOTIFY_OK;
3649                         case NETDEV_DOWN:
3650                                 bond->master_ip = bond_glean_dev_ip(bond->dev);
3651                                 return NOTIFY_OK;
3652                         default:
3653                                 return NOTIFY_DONE;
3654                         }
3655                 }
3656
3657                 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
3658                         vlan_dev = vlan_group_get_device(bond->vlgrp, vlan->vlan_id);
3659                         if (vlan_dev == event_dev) {
3660                                 switch (event) {
3661                                 case NETDEV_UP:
3662                                         vlan->vlan_ip = ifa->ifa_local;
3663                                         return NOTIFY_OK;
3664                                 case NETDEV_DOWN:
3665                                         vlan->vlan_ip =
3666                                                 bond_glean_dev_ip(vlan_dev);
3667                                         return NOTIFY_OK;
3668                                 default:
3669                                         return NOTIFY_DONE;
3670                                 }
3671                         }
3672                 }
3673         }
3674         return NOTIFY_DONE;
3675 }
3676
3677 static struct notifier_block bond_netdev_notifier = {
3678         .notifier_call = bond_netdev_event,
3679 };
3680
3681 static struct notifier_block bond_inetaddr_notifier = {
3682         .notifier_call = bond_inetaddr_event,
3683 };
3684
3685 /*-------------------------- Packet type handling ---------------------------*/
3686
3687 /* register to receive lacpdus on a bond */
3688 static void bond_register_lacpdu(struct bonding *bond)
3689 {
3690         struct packet_type *pk_type = &(BOND_AD_INFO(bond).ad_pkt_type);
3691
3692         /* initialize packet type */
3693         pk_type->type = PKT_TYPE_LACPDU;
3694         pk_type->dev = bond->dev;
3695         pk_type->func = bond_3ad_lacpdu_recv;
3696
3697         dev_add_pack(pk_type);
3698 }
3699
3700 /* unregister to receive lacpdus on a bond */
3701 static void bond_unregister_lacpdu(struct bonding *bond)
3702 {
3703         dev_remove_pack(&(BOND_AD_INFO(bond).ad_pkt_type));
3704 }
3705
3706 void bond_register_arp(struct bonding *bond)
3707 {
3708         struct packet_type *pt = &bond->arp_mon_pt;
3709
3710         if (pt->type)
3711                 return;
3712
3713         pt->type = htons(ETH_P_ARP);
3714         pt->dev = bond->dev;
3715         pt->func = bond_arp_rcv;
3716         dev_add_pack(pt);
3717 }
3718
3719 void bond_unregister_arp(struct bonding *bond)
3720 {
3721         struct packet_type *pt = &bond->arp_mon_pt;
3722
3723         dev_remove_pack(pt);
3724         pt->type = 0;
3725 }
3726
3727 /*---------------------------- Hashing Policies -----------------------------*/
3728
3729 /*
3730  * Hash for the output device based upon layer 2 and layer 3 data. If
3731  * the packet is not IP mimic bond_xmit_hash_policy_l2()
3732  */
3733 static int bond_xmit_hash_policy_l23(struct sk_buff *skb,
3734                                      struct net_device *bond_dev, int count)
3735 {
3736         struct ethhdr *data = (struct ethhdr *)skb->data;
3737         struct iphdr *iph = ip_hdr(skb);
3738
3739         if (skb->protocol == __constant_htons(ETH_P_IP)) {
3740                 return ((ntohl(iph->saddr ^ iph->daddr) & 0xffff) ^
3741                         (data->h_dest[5] ^ bond_dev->dev_addr[5])) % count;
3742         }
3743
3744         return (data->h_dest[5] ^ bond_dev->dev_addr[5]) % count;
3745 }
3746
3747 /*
3748  * Hash for the output device based upon layer 3 and layer 4 data. If
3749  * the packet is a frag or not TCP or UDP, just use layer 3 data.  If it is
3750  * altogether not IP, mimic bond_xmit_hash_policy_l2()
3751  */
3752 static int bond_xmit_hash_policy_l34(struct sk_buff *skb,
3753                                     struct net_device *bond_dev, int count)
3754 {
3755         struct ethhdr *data = (struct ethhdr *)skb->data;
3756         struct iphdr *iph = ip_hdr(skb);
3757         __be16 *layer4hdr = (__be16 *)((u32 *)iph + iph->ihl);
3758         int layer4_xor = 0;
3759
3760         if (skb->protocol == __constant_htons(ETH_P_IP)) {
3761                 if (!(iph->frag_off & __constant_htons(IP_MF|IP_OFFSET)) &&
3762                     (iph->protocol == IPPROTO_TCP ||
3763                      iph->protocol == IPPROTO_UDP)) {
3764                         layer4_xor = ntohs((*layer4hdr ^ *(layer4hdr + 1)));
3765                 }
3766                 return (layer4_xor ^
3767                         ((ntohl(iph->saddr ^ iph->daddr)) & 0xffff)) % count;
3768
3769         }
3770
3771         return (data->h_dest[5] ^ bond_dev->dev_addr[5]) % count;
3772 }
3773
3774 /*
3775  * Hash for the output device based upon layer 2 data
3776  */
3777 static int bond_xmit_hash_policy_l2(struct sk_buff *skb,
3778                                    struct net_device *bond_dev, int count)
3779 {
3780         struct ethhdr *data = (struct ethhdr *)skb->data;
3781
3782         return (data->h_dest[5] ^ bond_dev->dev_addr[5]) % count;
3783 }
3784
3785 /*-------------------------- Device entry points ----------------------------*/
3786
3787 static int bond_open(struct net_device *bond_dev)
3788 {
3789         struct bonding *bond = bond_dev->priv;
3790
3791         bond->kill_timers = 0;
3792
3793         if ((bond->params.mode == BOND_MODE_TLB) ||
3794             (bond->params.mode == BOND_MODE_ALB)) {
3795                 /* bond_alb_initialize must be called before the timer
3796                  * is started.
3797                  */
3798                 if (bond_alb_initialize(bond, (bond->params.mode == BOND_MODE_ALB))) {
3799                         /* something went wrong - fail the open operation */
3800                         return -1;
3801                 }
3802
3803                 INIT_DELAYED_WORK(&bond->alb_work, bond_alb_monitor);
3804                 queue_delayed_work(bond->wq, &bond->alb_work, 0);
3805         }
3806
3807         if (bond->params.miimon) {  /* link check interval, in milliseconds. */
3808                 INIT_DELAYED_WORK(&bond->mii_work, bond_mii_monitor);
3809                 queue_delayed_work(bond->wq, &bond->mii_work, 0);
3810         }
3811
3812         if (bond->params.arp_interval) {  /* arp interval, in milliseconds. */
3813                 if (bond->params.mode == BOND_MODE_ACTIVEBACKUP)
3814                         INIT_DELAYED_WORK(&bond->arp_work,
3815                                           bond_activebackup_arp_mon);
3816                 else
3817                         INIT_DELAYED_WORK(&bond->arp_work,
3818                                           bond_loadbalance_arp_mon);
3819
3820                 queue_delayed_work(bond->wq, &bond->arp_work, 0);
3821                 if (bond->params.arp_validate)
3822                         bond_register_arp(bond);
3823         }
3824
3825         if (bond->params.mode == BOND_MODE_8023AD) {
3826                 INIT_DELAYED_WORK(&bond->ad_work, bond_3ad_state_machine_handler);
3827                 queue_delayed_work(bond->wq, &bond->ad_work, 0);
3828                 /* register to receive LACPDUs */
3829                 bond_register_lacpdu(bond);
3830         }
3831
3832         return 0;
3833 }
3834
3835 static int bond_close(struct net_device *bond_dev)
3836 {
3837         struct bonding *bond = bond_dev->priv;
3838
3839         if (bond->params.mode == BOND_MODE_8023AD) {
3840                 /* Unregister the receive of LACPDUs */
3841                 bond_unregister_lacpdu(bond);
3842         }
3843
3844         if (bond->params.arp_validate)
3845                 bond_unregister_arp(bond);
3846
3847         write_lock_bh(&bond->lock);
3848
3849
3850         /* signal timers not to re-arm */
3851         bond->kill_timers = 1;
3852
3853         write_unlock_bh(&bond->lock);
3854
3855         if (bond->params.miimon) {  /* link check interval, in milliseconds. */
3856                 cancel_delayed_work(&bond->mii_work);
3857         }
3858
3859         if (bond->params.arp_interval) {  /* arp interval, in milliseconds. */
3860                 cancel_delayed_work(&bond->arp_work);
3861         }
3862
3863         switch (bond->params.mode) {
3864         case BOND_MODE_8023AD:
3865                 cancel_delayed_work(&bond->ad_work);
3866                 break;
3867         case BOND_MODE_TLB:
3868         case BOND_MODE_ALB:
3869                 cancel_delayed_work(&bond->alb_work);
3870                 break;
3871         default:
3872                 break;
3873         }
3874
3875
3876         if ((bond->params.mode == BOND_MODE_TLB) ||
3877             (bond->params.mode == BOND_MODE_ALB)) {
3878                 /* Must be called only after all
3879                  * slaves have been released
3880                  */
3881                 bond_alb_deinitialize(bond);
3882         }
3883
3884         return 0;
3885 }
3886
3887 static struct net_device_stats *bond_get_stats(struct net_device *bond_dev)
3888 {
3889         struct bonding *bond = bond_dev->priv;
3890         struct net_device_stats *stats = &(bond->stats), *sstats;
3891         struct net_device_stats local_stats;
3892         struct slave *slave;
3893         int i;
3894
3895         memset(&local_stats, 0, sizeof(struct net_device_stats));
3896
3897         read_lock_bh(&bond->lock);
3898
3899         bond_for_each_slave(bond, slave, i) {
3900                 sstats = slave->dev->get_stats(slave->dev);
3901                 local_stats.rx_packets += sstats->rx_packets;
3902                 local_stats.rx_bytes += sstats->rx_bytes;
3903                 local_stats.rx_errors += sstats->rx_errors;
3904                 local_stats.rx_dropped += sstats->rx_dropped;
3905
3906                 local_stats.tx_packets += sstats->tx_packets;
3907                 local_stats.tx_bytes += sstats->tx_bytes;
3908                 local_stats.tx_errors += sstats->tx_errors;
3909                 local_stats.tx_dropped += sstats->tx_dropped;
3910
3911                 local_stats.multicast += sstats->multicast;
3912                 local_stats.collisions += sstats->collisions;
3913
3914                 local_stats.rx_length_errors += sstats->rx_length_errors;
3915                 local_stats.rx_over_errors += sstats->rx_over_errors;
3916                 local_stats.rx_crc_errors += sstats->rx_crc_errors;
3917                 local_stats.rx_frame_errors += sstats->rx_frame_errors;
3918                 local_stats.rx_fifo_errors += sstats->rx_fifo_errors;
3919                 local_stats.rx_missed_errors += sstats->rx_missed_errors;
3920
3921                 local_stats.tx_aborted_errors += sstats->tx_aborted_errors;
3922                 local_stats.tx_carrier_errors += sstats->tx_carrier_errors;
3923                 local_stats.tx_fifo_errors += sstats->tx_fifo_errors;
3924                 local_stats.tx_heartbeat_errors += sstats->tx_heartbeat_errors;
3925                 local_stats.tx_window_errors += sstats->tx_window_errors;
3926         }
3927
3928         memcpy(stats, &local_stats, sizeof(struct net_device_stats));
3929
3930         read_unlock_bh(&bond->lock);
3931
3932         return stats;
3933 }
3934
3935 static int bond_do_ioctl(struct net_device *bond_dev, struct ifreq *ifr, int cmd)
3936 {
3937         struct net_device *slave_dev = NULL;
3938         struct ifbond k_binfo;
3939         struct ifbond __user *u_binfo = NULL;
3940         struct ifslave k_sinfo;
3941         struct ifslave __user *u_sinfo = NULL;
3942         struct mii_ioctl_data *mii = NULL;
3943         int res = 0;
3944
3945         dprintk("bond_ioctl: master=%s, cmd=%d\n",
3946                 bond_dev->name, cmd);
3947
3948         switch (cmd) {
3949         case SIOCGMIIPHY:
3950                 mii = if_mii(ifr);
3951                 if (!mii) {
3952                         return -EINVAL;
3953                 }
3954                 mii->phy_id = 0;
3955                 /* Fall Through */
3956         case SIOCGMIIREG:
3957                 /*
3958                  * We do this again just in case we were called by SIOCGMIIREG
3959                  * instead of SIOCGMIIPHY.
3960                  */
3961                 mii = if_mii(ifr);
3962                 if (!mii) {
3963                         return -EINVAL;
3964                 }
3965
3966                 if (mii->reg_num == 1) {
3967                         struct bonding *bond = bond_dev->priv;
3968                         mii->val_out = 0;
3969                         read_lock(&bond->lock);
3970                         read_lock(&bond->curr_slave_lock);
3971                         if (netif_carrier_ok(bond->dev)) {
3972                                 mii->val_out = BMSR_LSTATUS;
3973                         }
3974                         read_unlock(&bond->curr_slave_lock);
3975                         read_unlock(&bond->lock);
3976                 }
3977
3978                 return 0;
3979         case BOND_INFO_QUERY_OLD:
3980         case SIOCBONDINFOQUERY:
3981                 u_binfo = (struct ifbond __user *)ifr->ifr_data;
3982
3983                 if (copy_from_user(&k_binfo, u_binfo, sizeof(ifbond))) {
3984                         return -EFAULT;
3985                 }
3986
3987                 res = bond_info_query(bond_dev, &k_binfo);
3988                 if (res == 0) {
3989                         if (copy_to_user(u_binfo, &k_binfo, sizeof(ifbond))) {
3990                                 return -EFAULT;
3991                         }
3992                 }
3993
3994                 return res;
3995         case BOND_SLAVE_INFO_QUERY_OLD:
3996         case SIOCBONDSLAVEINFOQUERY:
3997                 u_sinfo = (struct ifslave __user *)ifr->ifr_data;
3998
3999                 if (copy_from_user(&k_sinfo, u_sinfo, sizeof(ifslave))) {
4000                         return -EFAULT;
4001                 }
4002
4003                 res = bond_slave_info_query(bond_dev, &k_sinfo);
4004                 if (res == 0) {
4005                         if (copy_to_user(u_sinfo, &k_sinfo, sizeof(ifslave))) {
4006                                 return -EFAULT;
4007                         }
4008                 }
4009
4010                 return res;
4011         default:
4012                 /* Go on */
4013                 break;
4014         }
4015
4016         if (!capable(CAP_NET_ADMIN)) {
4017                 return -EPERM;
4018         }
4019
4020         down_write(&(bonding_rwsem));
4021         slave_dev = dev_get_by_name(&init_net, ifr->ifr_slave);
4022
4023         dprintk("slave_dev=%p: \n", slave_dev);
4024
4025         if (!slave_dev) {
4026                 res = -ENODEV;
4027         } else {
4028                 dprintk("slave_dev->name=%s: \n", slave_dev->name);
4029                 switch (cmd) {
4030                 case BOND_ENSLAVE_OLD:
4031                 case SIOCBONDENSLAVE:
4032                         res = bond_enslave(bond_dev, slave_dev);
4033                         break;
4034                 case BOND_RELEASE_OLD:
4035                 case SIOCBONDRELEASE:
4036                         res = bond_release(bond_dev, slave_dev);
4037                         break;
4038                 case BOND_SETHWADDR_OLD:
4039                 case SIOCBONDSETHWADDR:
4040                         res = bond_sethwaddr(bond_dev, slave_dev);
4041                         break;
4042                 case BOND_CHANGE_ACTIVE_OLD:
4043                 case SIOCBONDCHANGEACTIVE:
4044                         res = bond_ioctl_change_active(bond_dev, slave_dev);
4045                         break;
4046                 default:
4047                         res = -EOPNOTSUPP;
4048                 }
4049
4050                 dev_put(slave_dev);
4051         }
4052
4053         up_write(&(bonding_rwsem));
4054         return res;
4055 }
4056
4057 static void bond_set_multicast_list(struct net_device *bond_dev)
4058 {
4059         struct bonding *bond = bond_dev->priv;
4060         struct dev_mc_list *dmi;
4061
4062         /*
4063          * Do promisc before checking multicast_mode
4064          */
4065         if ((bond_dev->flags & IFF_PROMISC) && !(bond->flags & IFF_PROMISC)) {
4066                 bond_set_promiscuity(bond, 1);
4067         }
4068
4069         if (!(bond_dev->flags & IFF_PROMISC) && (bond->flags & IFF_PROMISC)) {
4070                 bond_set_promiscuity(bond, -1);
4071         }
4072
4073         /* set allmulti flag to slaves */
4074         if ((bond_dev->flags & IFF_ALLMULTI) && !(bond->flags & IFF_ALLMULTI)) {
4075                 bond_set_allmulti(bond, 1);
4076         }
4077
4078         if (!(bond_dev->flags & IFF_ALLMULTI) && (bond->flags & IFF_ALLMULTI)) {
4079                 bond_set_allmulti(bond, -1);
4080         }
4081
4082         read_lock(&bond->lock);
4083
4084         bond->flags = bond_dev->flags;
4085
4086         /* looking for addresses to add to slaves' mc list */
4087         for (dmi = bond_dev->mc_list; dmi; dmi = dmi->next) {
4088                 if (!bond_mc_list_find_dmi(dmi, bond->mc_list)) {
4089                         bond_mc_add(bond, dmi->dmi_addr, dmi->dmi_addrlen);
4090                 }
4091         }
4092
4093         /* looking for addresses to delete from slaves' list */
4094         for (dmi = bond->mc_list; dmi; dmi = dmi->next) {
4095                 if (!bond_mc_list_find_dmi(dmi, bond_dev->mc_list)) {
4096                         bond_mc_delete(bond, dmi->dmi_addr, dmi->dmi_addrlen);
4097                 }
4098         }
4099
4100         /* save master's multicast list */
4101         bond_mc_list_destroy(bond);
4102         bond_mc_list_copy(bond_dev->mc_list, bond, GFP_ATOMIC);
4103
4104         read_unlock(&bond->lock);
4105 }
4106
4107 /*
4108  * Change the MTU of all of a master's slaves to match the master
4109  */
4110 static int bond_change_mtu(struct net_device *bond_dev, int new_mtu)
4111 {
4112         struct bonding *bond = bond_dev->priv;
4113         struct slave *slave, *stop_at;
4114         int res = 0;
4115         int i;
4116
4117         dprintk("bond=%p, name=%s, new_mtu=%d\n", bond,
4118                 (bond_dev ? bond_dev->name : "None"), new_mtu);
4119
4120         /* Can't hold bond->lock with bh disabled here since
4121          * some base drivers panic. On the other hand we can't
4122          * hold bond->lock without bh disabled because we'll
4123          * deadlock. The only solution is to rely on the fact
4124          * that we're under rtnl_lock here, and the slaves
4125          * list won't change. This doesn't solve the problem
4126          * of setting the slave's MTU while it is
4127          * transmitting, but the assumption is that the base
4128          * driver can handle that.
4129          *
4130          * TODO: figure out a way to safely iterate the slaves
4131          * list, but without holding a lock around the actual
4132          * call to the base driver.
4133          */
4134
4135         bond_for_each_slave(bond, slave, i) {
4136                 dprintk("s %p s->p %p c_m %p\n", slave,
4137                         slave->prev, slave->dev->change_mtu);
4138
4139                 res = dev_set_mtu(slave->dev, new_mtu);
4140
4141                 if (res) {
4142                         /* If we failed to set the slave's mtu to the new value
4143                          * we must abort the operation even in ACTIVE_BACKUP
4144                          * mode, because if we allow the backup slaves to have
4145                          * different mtu values than the active slave we'll
4146                          * need to change their mtu when doing a failover. That
4147                          * means changing their mtu from timer context, which
4148                          * is probably not a good idea.
4149                          */
4150                         dprintk("err %d %s\n", res, slave->dev->name);
4151                         goto unwind;
4152                 }
4153         }
4154
4155         bond_dev->mtu = new_mtu;
4156
4157         return 0;
4158
4159 unwind:
4160         /* unwind from head to the slave that failed */
4161         stop_at = slave;
4162         bond_for_each_slave_from_to(bond, slave, i, bond->first_slave, stop_at) {
4163                 int tmp_res;
4164
4165                 tmp_res = dev_set_mtu(slave->dev, bond_dev->mtu);
4166                 if (tmp_res) {
4167                         dprintk("unwind err %d dev %s\n", tmp_res,
4168                                 slave->dev->name);
4169                 }
4170         }
4171
4172         return res;
4173 }
4174
4175 /*
4176  * Change HW address
4177  *
4178  * Note that many devices must be down to change the HW address, and
4179  * downing the master releases all slaves.  We can make bonds full of
4180  * bonding devices to test this, however.
4181  */
4182 static int bond_set_mac_address(struct net_device *bond_dev, void *addr)
4183 {
4184         struct bonding *bond = bond_dev->priv;
4185         struct sockaddr *sa = addr, tmp_sa;
4186         struct slave *slave, *stop_at;
4187         int res = 0;
4188         int i;
4189
4190         dprintk("bond=%p, name=%s\n", bond, (bond_dev ? bond_dev->name : "None"));
4191
4192         /*
4193          * If fail_over_mac is set to active, do nothing and return
4194          * success.  Returning an error causes ifenslave to fail.
4195          */
4196         if (bond->params.fail_over_mac == BOND_FOM_ACTIVE)
4197                 return 0;
4198
4199         if (!is_valid_ether_addr(sa->sa_data)) {
4200                 return -EADDRNOTAVAIL;
4201         }
4202
4203         /* Can't hold bond->lock with bh disabled here since
4204          * some base drivers panic. On the other hand we can't
4205          * hold bond->lock without bh disabled because we'll
4206          * deadlock. The only solution is to rely on the fact
4207          * that we're under rtnl_lock here, and the slaves
4208          * list won't change. This doesn't solve the problem
4209          * of setting the slave's hw address while it is
4210          * transmitting, but the assumption is that the base
4211          * driver can handle that.
4212          *
4213          * TODO: figure out a way to safely iterate the slaves
4214          * list, but without holding a lock around the actual
4215          * call to the base driver.
4216          */
4217
4218         bond_for_each_slave(bond, slave, i) {
4219                 dprintk("slave %p %s\n", slave, slave->dev->name);
4220
4221                 if (slave->dev->set_mac_address == NULL) {
4222                         res = -EOPNOTSUPP;
4223                         dprintk("EOPNOTSUPP %s\n", slave->dev->name);
4224                         goto unwind;
4225                 }
4226
4227                 res = dev_set_mac_address(slave->dev, addr);
4228                 if (res) {
4229                         /* TODO: consider downing the slave
4230                          * and retry ?
4231                          * User should expect communications
4232                          * breakage anyway until ARP finish
4233                          * updating, so...
4234                          */
4235                         dprintk("err %d %s\n", res, slave->dev->name);
4236                         goto unwind;
4237                 }
4238         }
4239
4240         /* success */
4241         memcpy(bond_dev->dev_addr, sa->sa_data, bond_dev->addr_len);
4242         return 0;
4243
4244 unwind:
4245         memcpy(tmp_sa.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
4246         tmp_sa.sa_family = bond_dev->type;
4247
4248         /* unwind from head to the slave that failed */
4249         stop_at = slave;
4250         bond_for_each_slave_from_to(bond, slave, i, bond->first_slave, stop_at) {
4251                 int tmp_res;
4252
4253                 tmp_res = dev_set_mac_address(slave->dev, &tmp_sa);
4254                 if (tmp_res) {
4255                         dprintk("unwind err %d dev %s\n", tmp_res,
4256                                 slave->dev->name);
4257                 }
4258         }
4259
4260         return res;
4261 }
4262
4263 static int bond_xmit_roundrobin(struct sk_buff *skb, struct net_device *bond_dev)
4264 {
4265         struct bonding *bond = bond_dev->priv;
4266         struct slave *slave, *start_at;
4267         int i, slave_no, res = 1;
4268
4269         read_lock(&bond->lock);
4270
4271         if (!BOND_IS_OK(bond)) {
4272                 goto out;
4273         }
4274
4275         /*
4276          * Concurrent TX may collide on rr_tx_counter; we accept that
4277          * as being rare enough not to justify using an atomic op here
4278          */
4279         slave_no = bond->rr_tx_counter++ % bond->slave_cnt;
4280
4281         bond_for_each_slave(bond, slave, i) {
4282                 slave_no--;
4283                 if (slave_no < 0) {
4284                         break;
4285                 }
4286         }
4287
4288         start_at = slave;
4289         bond_for_each_slave_from(bond, slave, i, start_at) {
4290                 if (IS_UP(slave->dev) &&
4291                     (slave->link == BOND_LINK_UP) &&
4292                     (slave->state == BOND_STATE_ACTIVE)) {
4293                         res = bond_dev_queue_xmit(bond, skb, slave->dev);
4294                         break;
4295                 }
4296         }
4297
4298 out:
4299         if (res) {
4300                 /* no suitable interface, frame not sent */
4301                 dev_kfree_skb(skb);
4302         }
4303         read_unlock(&bond->lock);
4304         return 0;
4305 }
4306
4307
4308 /*
4309  * in active-backup mode, we know that bond->curr_active_slave is always valid if
4310  * the bond has a usable interface.
4311  */
4312 static int bond_xmit_activebackup(struct sk_buff *skb, struct net_device *bond_dev)
4313 {
4314         struct bonding *bond = bond_dev->priv;
4315         int res = 1;
4316
4317         read_lock(&bond->lock);
4318         read_lock(&bond->curr_slave_lock);
4319
4320         if (!BOND_IS_OK(bond)) {
4321                 goto out;
4322         }
4323
4324         if (!bond->curr_active_slave)
4325                 goto out;
4326
4327         res = bond_dev_queue_xmit(bond, skb, bond->curr_active_slave->dev);
4328
4329 out:
4330         if (res) {
4331                 /* no suitable interface, frame not sent */
4332                 dev_kfree_skb(skb);
4333         }
4334         read_unlock(&bond->curr_slave_lock);
4335         read_unlock(&bond->lock);
4336         return 0;
4337 }
4338
4339 /*
4340  * In bond_xmit_xor() , we determine the output device by using a pre-
4341  * determined xmit_hash_policy(), If the selected device is not enabled,
4342  * find the next active slave.
4343  */
4344 static int bond_xmit_xor(struct sk_buff *skb, struct net_device *bond_dev)
4345 {
4346         struct bonding *bond = bond_dev->priv;
4347         struct slave *slave, *start_at;
4348         int slave_no;
4349         int i;
4350         int res = 1;
4351
4352         read_lock(&bond->lock);
4353
4354         if (!BOND_IS_OK(bond)) {
4355                 goto out;
4356         }
4357
4358         slave_no = bond->xmit_hash_policy(skb, bond_dev, bond->slave_cnt);
4359
4360         bond_for_each_slave(bond, slave, i) {
4361                 slave_no--;
4362                 if (slave_no < 0) {
4363                         break;
4364                 }
4365         }
4366
4367         start_at = slave;
4368
4369         bond_for_each_slave_from(bond, slave, i, start_at) {
4370                 if (IS_UP(slave->dev) &&
4371                     (slave->link == BOND_LINK_UP) &&
4372                     (slave->state == BOND_STATE_ACTIVE)) {
4373                         res = bond_dev_queue_xmit(bond, skb, slave->dev);
4374                         break;
4375                 }
4376         }
4377
4378 out:
4379         if (res) {
4380                 /* no suitable interface, frame not sent */
4381                 dev_kfree_skb(skb);
4382         }
4383         read_unlock(&bond->lock);
4384         return 0;
4385 }
4386
4387 /*
4388  * in broadcast mode, we send everything to all usable interfaces.
4389  */
4390 static int bond_xmit_broadcast(struct sk_buff *skb, struct net_device *bond_dev)
4391 {
4392         struct bonding *bond = bond_dev->priv;
4393         struct slave *slave, *start_at;
4394         struct net_device *tx_dev = NULL;
4395         int i;
4396         int res = 1;
4397
4398         read_lock(&bond->lock);
4399
4400         if (!BOND_IS_OK(bond)) {
4401                 goto out;
4402         }
4403
4404         read_lock(&bond->curr_slave_lock);
4405         start_at = bond->curr_active_slave;
4406         read_unlock(&bond->curr_slave_lock);
4407
4408         if (!start_at) {
4409                 goto out;
4410         }
4411
4412         bond_for_each_slave_from(bond, slave, i, start_at) {
4413                 if (IS_UP(slave->dev) &&
4414                     (slave->link == BOND_LINK_UP) &&
4415                     (slave->state == BOND_STATE_ACTIVE)) {
4416                         if (tx_dev) {
4417                                 struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
4418                                 if (!skb2) {
4419                                         printk(KERN_ERR DRV_NAME
4420                                                ": %s: Error: bond_xmit_broadcast(): "
4421                                                "skb_clone() failed\n",
4422                                                bond_dev->name);
4423                                         continue;
4424                                 }
4425
4426                                 res = bond_dev_queue_xmit(bond, skb2, tx_dev);
4427                                 if (res) {
4428                                         dev_kfree_skb(skb2);
4429                                         continue;
4430                                 }
4431                         }
4432                         tx_dev = slave->dev;
4433                 }
4434         }
4435
4436         if (tx_dev) {
4437                 res = bond_dev_queue_xmit(bond, skb, tx_dev);
4438         }
4439
4440 out:
4441         if (res) {
4442                 /* no suitable interface, frame not sent */
4443                 dev_kfree_skb(skb);
4444         }
4445         /* frame sent to all suitable interfaces */
4446         read_unlock(&bond->lock);
4447         return 0;
4448 }
4449
4450 /*------------------------- Device initialization ---------------------------*/
4451
4452 static void bond_set_xmit_hash_policy(struct bonding *bond)
4453 {
4454         switch (bond->params.xmit_policy) {
4455         case BOND_XMIT_POLICY_LAYER23:
4456                 bond->xmit_hash_policy = bond_xmit_hash_policy_l23;
4457                 break;
4458         case BOND_XMIT_POLICY_LAYER34:
4459                 bond->xmit_hash_policy = bond_xmit_hash_policy_l34;
4460                 break;
4461         case BOND_XMIT_POLICY_LAYER2:
4462         default:
4463                 bond->xmit_hash_policy = bond_xmit_hash_policy_l2;
4464                 break;
4465         }
4466 }
4467
4468 /*
4469  * set bond mode specific net device operations
4470  */
4471 void bond_set_mode_ops(struct bonding *bond, int mode)
4472 {
4473         struct net_device *bond_dev = bond->dev;
4474
4475         switch (mode) {
4476         case BOND_MODE_ROUNDROBIN:
4477                 bond_dev->hard_start_xmit = bond_xmit_roundrobin;
4478                 break;
4479         case BOND_MODE_ACTIVEBACKUP:
4480                 bond_dev->hard_start_xmit = bond_xmit_activebackup;
4481                 break;
4482         case BOND_MODE_XOR:
4483                 bond_dev->hard_start_xmit = bond_xmit_xor;
4484                 bond_set_xmit_hash_policy(bond);
4485                 break;
4486         case BOND_MODE_BROADCAST:
4487                 bond_dev->hard_start_xmit = bond_xmit_broadcast;
4488                 break;
4489         case BOND_MODE_8023AD:
4490                 bond_set_master_3ad_flags(bond);
4491                 bond_dev->hard_start_xmit = bond_3ad_xmit_xor;
4492                 bond_set_xmit_hash_policy(bond);
4493                 break;
4494         case BOND_MODE_ALB:
4495                 bond_set_master_alb_flags(bond);
4496                 /* FALLTHRU */
4497         case BOND_MODE_TLB:
4498                 bond_dev->hard_start_xmit = bond_alb_xmit;
4499                 bond_dev->set_mac_address = bond_alb_set_mac_address;
4500                 break;
4501         default:
4502                 /* Should never happen, mode already checked */
4503                 printk(KERN_ERR DRV_NAME
4504                        ": %s: Error: Unknown bonding mode %d\n",
4505                        bond_dev->name,
4506                        mode);
4507                 break;
4508         }
4509 }
4510
4511 static void bond_ethtool_get_drvinfo(struct net_device *bond_dev,
4512                                     struct ethtool_drvinfo *drvinfo)
4513 {
4514         strncpy(drvinfo->driver, DRV_NAME, 32);
4515         strncpy(drvinfo->version, DRV_VERSION, 32);
4516         snprintf(drvinfo->fw_version, 32, "%d", BOND_ABI_VERSION);
4517 }
4518
4519 static const struct ethtool_ops bond_ethtool_ops = {
4520         .get_drvinfo            = bond_ethtool_get_drvinfo,
4521 };
4522
4523 /*
4524  * Does not allocate but creates a /proc entry.
4525  * Allowed to fail.
4526  */
4527 static int bond_init(struct net_device *bond_dev, struct bond_params *params)
4528 {
4529         struct bonding *bond = bond_dev->priv;
4530
4531         dprintk("Begin bond_init for %s\n", bond_dev->name);
4532
4533         /* initialize rwlocks */
4534         rwlock_init(&bond->lock);
4535         rwlock_init(&bond->curr_slave_lock);
4536
4537         bond->params = *params; /* copy params struct */
4538
4539         bond->wq = create_singlethread_workqueue(bond_dev->name);
4540         if (!bond->wq)
4541                 return -ENOMEM;
4542
4543         /* Initialize pointers */
4544         bond->first_slave = NULL;
4545         bond->curr_active_slave = NULL;
4546         bond->current_arp_slave = NULL;
4547         bond->primary_slave = NULL;
4548         bond->dev = bond_dev;
4549         bond->send_grat_arp = 0;
4550         bond->setup_by_slave = 0;
4551         INIT_LIST_HEAD(&bond->vlan_list);
4552
4553         /* Initialize the device entry points */
4554         bond_dev->open = bond_open;
4555         bond_dev->stop = bond_close;
4556         bond_dev->get_stats = bond_get_stats;
4557         bond_dev->do_ioctl = bond_do_ioctl;
4558         bond_dev->ethtool_ops = &bond_ethtool_ops;
4559         bond_dev->set_multicast_list = bond_set_multicast_list;
4560         bond_dev->change_mtu = bond_change_mtu;
4561         bond_dev->set_mac_address = bond_set_mac_address;
4562         bond_dev->validate_addr = NULL;
4563
4564         bond_set_mode_ops(bond, bond->params.mode);
4565
4566         bond_dev->destructor = free_netdev;
4567
4568         /* Initialize the device options */
4569         bond_dev->tx_queue_len = 0;
4570         bond_dev->flags |= IFF_MASTER|IFF_MULTICAST;
4571         bond_dev->priv_flags |= IFF_BONDING;
4572
4573         /* At first, we block adding VLANs. That's the only way to
4574          * prevent problems that occur when adding VLANs over an
4575          * empty bond. The block will be removed once non-challenged
4576          * slaves are enslaved.
4577          */
4578         bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
4579
4580         /* don't acquire bond device's netif_tx_lock when
4581          * transmitting */
4582         bond_dev->features |= NETIF_F_LLTX;
4583
4584         /* By default, we declare the bond to be fully
4585          * VLAN hardware accelerated capable. Special
4586          * care is taken in the various xmit functions
4587          * when there are slaves that are not hw accel
4588          * capable
4589          */
4590         bond_dev->vlan_rx_register = bond_vlan_rx_register;
4591         bond_dev->vlan_rx_add_vid  = bond_vlan_rx_add_vid;
4592         bond_dev->vlan_rx_kill_vid = bond_vlan_rx_kill_vid;
4593         bond_dev->features |= (NETIF_F_HW_VLAN_TX |
4594                                NETIF_F_HW_VLAN_RX |
4595                                NETIF_F_HW_VLAN_FILTER);
4596
4597 #ifdef CONFIG_PROC_FS
4598         bond_create_proc_entry(bond);
4599 #endif
4600         list_add_tail(&bond->bond_list, &bond_dev_list);
4601
4602         return 0;
4603 }
4604
4605 /* De-initialize device specific data.
4606  * Caller must hold rtnl_lock.
4607  */
4608 static void bond_deinit(struct net_device *bond_dev)
4609 {
4610         struct bonding *bond = bond_dev->priv;
4611
4612         list_del(&bond->bond_list);
4613
4614 #ifdef CONFIG_PROC_FS
4615         bond_remove_proc_entry(bond);
4616 #endif
4617 }
4618
4619 static void bond_work_cancel_all(struct bonding *bond)
4620 {
4621         write_lock_bh(&bond->lock);
4622         bond->kill_timers = 1;
4623         write_unlock_bh(&bond->lock);
4624
4625         if (bond->params.miimon && delayed_work_pending(&bond->mii_work))
4626                 cancel_delayed_work(&bond->mii_work);
4627
4628         if (bond->params.arp_interval && delayed_work_pending(&bond->arp_work))
4629                 cancel_delayed_work(&bond->arp_work);
4630
4631         if (bond->params.mode == BOND_MODE_ALB &&
4632             delayed_work_pending(&bond->alb_work))
4633                 cancel_delayed_work(&bond->alb_work);
4634
4635         if (bond->params.mode == BOND_MODE_8023AD &&
4636             delayed_work_pending(&bond->ad_work))
4637                 cancel_delayed_work(&bond->ad_work);
4638 }
4639
4640 /* Unregister and free all bond devices.
4641  * Caller must hold rtnl_lock.
4642  */
4643 static void bond_free_all(void)
4644 {
4645         struct bonding *bond, *nxt;
4646
4647         list_for_each_entry_safe(bond, nxt, &bond_dev_list, bond_list) {
4648                 struct net_device *bond_dev = bond->dev;
4649
4650                 bond_work_cancel_all(bond);
4651                 netif_tx_lock_bh(bond_dev);
4652                 bond_mc_list_destroy(bond);
4653                 netif_tx_unlock_bh(bond_dev);
4654                 /* Release the bonded slaves */
4655                 bond_release_all(bond_dev);
4656                 bond_destroy(bond);
4657         }
4658
4659 #ifdef CONFIG_PROC_FS
4660         bond_destroy_proc_dir();
4661 #endif
4662 }
4663
4664 /*------------------------- Module initialization ---------------------------*/
4665
4666 /*
4667  * Convert string input module parms.  Accept either the
4668  * number of the mode or its string name.  A bit complicated because
4669  * some mode names are substrings of other names, and calls from sysfs
4670  * may have whitespace in the name (trailing newlines, for example).
4671  */
4672 int bond_parse_parm(const char *buf, struct bond_parm_tbl *tbl)
4673 {
4674         int mode = -1, i, rv;
4675         char *p, modestr[BOND_MAX_MODENAME_LEN + 1] = { 0, };
4676
4677         for (p = (char *)buf; *p; p++)
4678                 if (!(isdigit(*p) || isspace(*p)))
4679                         break;
4680
4681         if (*p)
4682                 rv = sscanf(buf, "%20s", modestr);
4683         else
4684                 rv = sscanf(buf, "%d", &mode);
4685
4686         if (!rv)
4687                 return -1;
4688
4689         for (i = 0; tbl[i].modename; i++) {
4690                 if (mode == tbl[i].mode)
4691                         return tbl[i].mode;
4692                 if (strcmp(modestr, tbl[i].modename) == 0)
4693                         return tbl[i].mode;
4694         }
4695
4696         return -1;
4697 }
4698
4699 static int bond_check_params(struct bond_params *params)
4700 {
4701         int arp_validate_value, fail_over_mac_value;
4702
4703         /*
4704          * Convert string parameters.
4705          */
4706         if (mode) {
4707                 bond_mode = bond_parse_parm(mode, bond_mode_tbl);
4708                 if (bond_mode == -1) {
4709                         printk(KERN_ERR DRV_NAME
4710                                ": Error: Invalid bonding mode \"%s\"\n",
4711                                mode == NULL ? "NULL" : mode);
4712                         return -EINVAL;
4713                 }
4714         }
4715
4716         if (xmit_hash_policy) {
4717                 if ((bond_mode != BOND_MODE_XOR) &&
4718                     (bond_mode != BOND_MODE_8023AD)) {
4719                         printk(KERN_INFO DRV_NAME
4720                                ": xor_mode param is irrelevant in mode %s\n",
4721                                bond_mode_name(bond_mode));
4722                 } else {
4723                         xmit_hashtype = bond_parse_parm(xmit_hash_policy,
4724                                                         xmit_hashtype_tbl);
4725                         if (xmit_hashtype == -1) {
4726                                 printk(KERN_ERR DRV_NAME
4727                                 ": Error: Invalid xmit_hash_policy \"%s\"\n",
4728                                 xmit_hash_policy == NULL ? "NULL" :
4729                                        xmit_hash_policy);
4730                                 return -EINVAL;
4731                         }
4732                 }
4733         }
4734
4735         if (lacp_rate) {
4736                 if (bond_mode != BOND_MODE_8023AD) {
4737                         printk(KERN_INFO DRV_NAME
4738                                ": lacp_rate param is irrelevant in mode %s\n",
4739                                bond_mode_name(bond_mode));
4740                 } else {
4741                         lacp_fast = bond_parse_parm(lacp_rate, bond_lacp_tbl);
4742                         if (lacp_fast == -1) {
4743                                 printk(KERN_ERR DRV_NAME
4744                                        ": Error: Invalid lacp rate \"%s\"\n",
4745                                        lacp_rate == NULL ? "NULL" : lacp_rate);
4746                                 return -EINVAL;
4747                         }
4748                 }
4749         }
4750
4751         if (max_bonds < 1 || max_bonds > INT_MAX) {
4752                 printk(KERN_WARNING DRV_NAME
4753                        ": Warning: max_bonds (%d) not in range %d-%d, so it "
4754                        "was reset to BOND_DEFAULT_MAX_BONDS (%d)\n",
4755                        max_bonds, 1, INT_MAX, BOND_DEFAULT_MAX_BONDS);
4756                 max_bonds = BOND_DEFAULT_MAX_BONDS;
4757         }
4758
4759         if (miimon < 0) {
4760                 printk(KERN_WARNING DRV_NAME
4761                        ": Warning: miimon module parameter (%d), "
4762                        "not in range 0-%d, so it was reset to %d\n",
4763                        miimon, INT_MAX, BOND_LINK_MON_INTERV);
4764                 miimon = BOND_LINK_MON_INTERV;
4765         }
4766
4767         if (updelay < 0) {
4768                 printk(KERN_WARNING DRV_NAME
4769                        ": Warning: updelay module parameter (%d), "
4770                        "not in range 0-%d, so it was reset to 0\n",
4771                        updelay, INT_MAX);
4772                 updelay = 0;
4773         }
4774
4775         if (downdelay < 0) {
4776                 printk(KERN_WARNING DRV_NAME
4777                        ": Warning: downdelay module parameter (%d), "
4778                        "not in range 0-%d, so it was reset to 0\n",
4779                        downdelay, INT_MAX);
4780                 downdelay = 0;
4781         }
4782
4783         if ((use_carrier != 0) && (use_carrier != 1)) {
4784                 printk(KERN_WARNING DRV_NAME
4785                        ": Warning: use_carrier module parameter (%d), "
4786                        "not of valid value (0/1), so it was set to 1\n",
4787                        use_carrier);
4788                 use_carrier = 1;
4789         }
4790
4791         if (num_grat_arp < 0 || num_grat_arp > 255) {
4792                 printk(KERN_WARNING DRV_NAME
4793                        ": Warning: num_grat_arp (%d) not in range 0-255 so it "
4794                        "was reset to 1 \n", num_grat_arp);
4795                 num_grat_arp = 1;
4796         }
4797
4798         /* reset values for 802.3ad */
4799         if (bond_mode == BOND_MODE_8023AD) {
4800                 if (!miimon) {
4801                         printk(KERN_WARNING DRV_NAME
4802                                ": Warning: miimon must be specified, "
4803                                "otherwise bonding will not detect link "
4804                                "failure, speed and duplex which are "
4805                                "essential for 802.3ad operation\n");
4806                         printk(KERN_WARNING "Forcing miimon to 100msec\n");
4807                         miimon = 100;
4808                 }
4809         }
4810
4811         /* reset values for TLB/ALB */
4812         if ((bond_mode == BOND_MODE_TLB) ||
4813             (bond_mode == BOND_MODE_ALB)) {
4814                 if (!miimon) {
4815                         printk(KERN_WARNING DRV_NAME
4816                                ": Warning: miimon must be specified, "
4817                                "otherwise bonding will not detect link "
4818                                "failure and link speed which are essential "
4819                                "for TLB/ALB load balancing\n");
4820                         printk(KERN_WARNING "Forcing miimon to 100msec\n");
4821                         miimon = 100;
4822                 }
4823         }
4824
4825         if (bond_mode == BOND_MODE_ALB) {
4826                 printk(KERN_NOTICE DRV_NAME
4827                        ": In ALB mode you might experience client "
4828                        "disconnections upon reconnection of a link if the "
4829                        "bonding module updelay parameter (%d msec) is "
4830                        "incompatible with the forwarding delay time of the "
4831                        "switch\n",
4832                        updelay);
4833         }
4834
4835         if (!miimon) {
4836                 if (updelay || downdelay) {
4837                         /* just warn the user the up/down delay will have
4838                          * no effect since miimon is zero...
4839                          */
4840                         printk(KERN_WARNING DRV_NAME
4841                                ": Warning: miimon module parameter not set "
4842                                "and updelay (%d) or downdelay (%d) module "
4843                                "parameter is set; updelay and downdelay have "
4844                                "no effect unless miimon is set\n",
4845                                updelay, downdelay);
4846                 }
4847         } else {
4848                 /* don't allow arp monitoring */
4849                 if (arp_interval) {
4850                         printk(KERN_WARNING DRV_NAME
4851                                ": Warning: miimon (%d) and arp_interval (%d) "
4852                                "can't be used simultaneously, disabling ARP "
4853                                "monitoring\n",
4854                                miimon, arp_interval);
4855                         arp_interval = 0;
4856                 }
4857
4858                 if ((updelay % miimon) != 0) {
4859                         printk(KERN_WARNING DRV_NAME
4860                                ": Warning: updelay (%d) is not a multiple "
4861                                "of miimon (%d), updelay rounded to %d ms\n",
4862                                updelay, miimon, (updelay / miimon) * miimon);
4863                 }
4864
4865                 updelay /= miimon;
4866
4867                 if ((downdelay % miimon) != 0) {
4868                         printk(KERN_WARNING DRV_NAME
4869                                ": Warning: downdelay (%d) is not a multiple "
4870                                "of miimon (%d), downdelay rounded to %d ms\n",
4871                                downdelay, miimon,
4872                                (downdelay / miimon) * miimon);
4873                 }
4874
4875                 downdelay /= miimon;
4876         }
4877
4878         if (arp_interval < 0) {
4879                 printk(KERN_WARNING DRV_NAME
4880                        ": Warning: arp_interval module parameter (%d) "
4881                        ", not in range 0-%d, so it was reset to %d\n",
4882                        arp_interval, INT_MAX, BOND_LINK_ARP_INTERV);
4883                 arp_interval = BOND_LINK_ARP_INTERV;
4884         }
4885
4886         for (arp_ip_count = 0;
4887              (arp_ip_count < BOND_MAX_ARP_TARGETS) && arp_ip_target[arp_ip_count];
4888              arp_ip_count++) {
4889                 /* not complete check, but should be good enough to
4890                    catch mistakes */
4891                 if (!isdigit(arp_ip_target[arp_ip_count][0])) {
4892                         printk(KERN_WARNING DRV_NAME
4893                                ": Warning: bad arp_ip_target module parameter "
4894                                "(%s), ARP monitoring will not be performed\n",
4895                                arp_ip_target[arp_ip_count]);
4896                         arp_interval = 0;
4897                 } else {
4898                         __be32 ip = in_aton(arp_ip_target[arp_ip_count]);
4899                         arp_target[arp_ip_count] = ip;
4900                 }
4901         }
4902
4903         if (arp_interval && !arp_ip_count) {
4904                 /* don't allow arping if no arp_ip_target given... */
4905                 printk(KERN_WARNING DRV_NAME
4906                        ": Warning: arp_interval module parameter (%d) "
4907                        "specified without providing an arp_ip_target "
4908                        "parameter, arp_interval was reset to 0\n",
4909                        arp_interval);
4910                 arp_interval = 0;
4911         }
4912
4913         if (arp_validate) {
4914                 if (bond_mode != BOND_MODE_ACTIVEBACKUP) {
4915                         printk(KERN_ERR DRV_NAME
4916                ": arp_validate only supported in active-backup mode\n");
4917                         return -EINVAL;
4918                 }
4919                 if (!arp_interval) {
4920                         printk(KERN_ERR DRV_NAME
4921                                ": arp_validate requires arp_interval\n");
4922                         return -EINVAL;
4923                 }
4924
4925                 arp_validate_value = bond_parse_parm(arp_validate,
4926                                                      arp_validate_tbl);
4927                 if (arp_validate_value == -1) {
4928                         printk(KERN_ERR DRV_NAME
4929                                ": Error: invalid arp_validate \"%s\"\n",
4930                                arp_validate == NULL ? "NULL" : arp_validate);
4931                         return -EINVAL;
4932                 }
4933         } else
4934                 arp_validate_value = 0;
4935
4936         if (miimon) {
4937                 printk(KERN_INFO DRV_NAME
4938                        ": MII link monitoring set to %d ms\n",
4939                        miimon);
4940         } else if (arp_interval) {
4941                 int i;
4942
4943                 printk(KERN_INFO DRV_NAME
4944                        ": ARP monitoring set to %d ms, validate %s, with %d target(s):",
4945                        arp_interval,
4946                        arp_validate_tbl[arp_validate_value].modename,
4947                        arp_ip_count);
4948
4949                 for (i = 0; i < arp_ip_count; i++)
4950                         printk (" %s", arp_ip_target[i]);
4951
4952                 printk("\n");
4953
4954         } else {
4955                 /* miimon and arp_interval not set, we need one so things
4956                  * work as expected, see bonding.txt for details
4957                  */
4958                 printk(KERN_WARNING DRV_NAME
4959                        ": Warning: either miimon or arp_interval and "
4960                        "arp_ip_target module parameters must be specified, "
4961                        "otherwise bonding will not detect link failures! see "
4962                        "bonding.txt for details.\n");
4963         }
4964
4965         if (primary && !USES_PRIMARY(bond_mode)) {
4966                 /* currently, using a primary only makes sense
4967                  * in active backup, TLB or ALB modes
4968                  */
4969                 printk(KERN_WARNING DRV_NAME
4970                        ": Warning: %s primary device specified but has no "
4971                        "effect in %s mode\n",
4972                        primary, bond_mode_name(bond_mode));
4973                 primary = NULL;
4974         }
4975
4976         if (fail_over_mac) {
4977                 fail_over_mac_value = bond_parse_parm(fail_over_mac,
4978                                                       fail_over_mac_tbl);
4979                 if (fail_over_mac_value == -1) {
4980                         printk(KERN_ERR DRV_NAME
4981                                ": Error: invalid fail_over_mac \"%s\"\n",
4982                                arp_validate == NULL ? "NULL" : arp_validate);
4983                         return -EINVAL;
4984                 }
4985
4986                 if (bond_mode != BOND_MODE_ACTIVEBACKUP)
4987                         printk(KERN_WARNING DRV_NAME
4988                                ": Warning: fail_over_mac only affects "
4989                                "active-backup mode.\n");
4990         } else {
4991                 fail_over_mac_value = BOND_FOM_NONE;
4992         }
4993
4994         /* fill params struct with the proper values */
4995         params->mode = bond_mode;
4996         params->xmit_policy = xmit_hashtype;
4997         params->miimon = miimon;
4998         params->num_grat_arp = num_grat_arp;
4999         params->arp_interval = arp_interval;
5000         params->arp_validate = arp_validate_value;
5001         params->updelay = updelay;
5002         params->downdelay = downdelay;
5003         params->use_carrier = use_carrier;
5004         params->lacp_fast = lacp_fast;
5005         params->primary[0] = 0;
5006         params->fail_over_mac = fail_over_mac_value;
5007
5008         if (primary) {
5009                 strncpy(params->primary, primary, IFNAMSIZ);
5010                 params->primary[IFNAMSIZ - 1] = 0;
5011         }
5012
5013         memcpy(params->arp_targets, arp_target, sizeof(arp_target));
5014
5015         return 0;
5016 }
5017
5018 static struct lock_class_key bonding_netdev_xmit_lock_key;
5019
5020 /* Create a new bond based on the specified name and bonding parameters.
5021  * If name is NULL, obtain a suitable "bond%d" name for us.
5022  * Caller must NOT hold rtnl_lock; we need to release it here before we
5023  * set up our sysfs entries.
5024  */
5025 int bond_create(char *name, struct bond_params *params)
5026 {
5027         struct net_device *bond_dev;
5028         struct bonding *bond;
5029         int res;
5030
5031         rtnl_lock();
5032         down_write(&bonding_rwsem);
5033
5034         /* Check to see if the bond already exists. */
5035         if (name) {
5036                 list_for_each_entry(bond, &bond_dev_list, bond_list)
5037                         if (strnicmp(bond->dev->name, name, IFNAMSIZ) == 0) {
5038                                 printk(KERN_ERR DRV_NAME
5039                                ": cannot add bond %s; it already exists\n",
5040                                        name);
5041                                 res = -EPERM;
5042                                 goto out_rtnl;
5043                         }
5044         }
5045
5046         bond_dev = alloc_netdev(sizeof(struct bonding), name ? name : "",
5047                                 ether_setup);
5048         if (!bond_dev) {
5049                 printk(KERN_ERR DRV_NAME
5050                        ": %s: eek! can't alloc netdev!\n",
5051                        name);
5052                 res = -ENOMEM;
5053                 goto out_rtnl;
5054         }
5055
5056         if (!name) {
5057                 res = dev_alloc_name(bond_dev, "bond%d");
5058                 if (res < 0)
5059                         goto out_netdev;
5060         }
5061
5062         /* bond_init() must be called after dev_alloc_name() (for the
5063          * /proc files), but before register_netdevice(), because we
5064          * need to set function pointers.
5065          */
5066
5067         res = bond_init(bond_dev, params);
5068         if (res < 0) {
5069                 goto out_netdev;
5070         }
5071
5072         res = register_netdevice(bond_dev);
5073         if (res < 0) {
5074                 goto out_bond;
5075         }
5076
5077         lockdep_set_class(&bond_dev->_xmit_lock, &bonding_netdev_xmit_lock_key);
5078
5079         netif_carrier_off(bond_dev);
5080
5081         up_write(&bonding_rwsem);
5082         rtnl_unlock(); /* allows sysfs registration of net device */
5083         res = bond_create_sysfs_entry(bond_dev->priv);
5084         if (res < 0) {
5085                 rtnl_lock();
5086                 down_write(&bonding_rwsem);
5087                 bond_deinit(bond_dev);
5088                 unregister_netdevice(bond_dev);
5089                 goto out_rtnl;
5090         }
5091
5092         return 0;
5093
5094 out_bond:
5095         bond_deinit(bond_dev);
5096 out_netdev:
5097         free_netdev(bond_dev);
5098 out_rtnl:
5099         up_write(&bonding_rwsem);
5100         rtnl_unlock();
5101         return res;
5102 }
5103
5104 static int __init bonding_init(void)
5105 {
5106         int i;
5107         int res;
5108         struct bonding *bond;
5109
5110         printk(KERN_INFO "%s", version);
5111
5112         res = bond_check_params(&bonding_defaults);
5113         if (res) {
5114                 goto out;
5115         }
5116
5117 #ifdef CONFIG_PROC_FS
5118         bond_create_proc_dir();
5119 #endif
5120
5121         init_rwsem(&bonding_rwsem);
5122
5123         for (i = 0; i < max_bonds; i++) {
5124                 res = bond_create(NULL, &bonding_defaults);
5125                 if (res)
5126                         goto err;
5127         }
5128
5129         res = bond_create_sysfs();
5130         if (res)
5131                 goto err;
5132
5133         register_netdevice_notifier(&bond_netdev_notifier);
5134         register_inetaddr_notifier(&bond_inetaddr_notifier);
5135
5136         goto out;
5137 err:
5138         list_for_each_entry(bond, &bond_dev_list, bond_list) {
5139                 bond_work_cancel_all(bond);
5140                 destroy_workqueue(bond->wq);
5141         }
5142
5143         bond_destroy_sysfs();
5144
5145         rtnl_lock();
5146         bond_free_all();
5147         rtnl_unlock();
5148 out:
5149         return res;
5150
5151 }
5152
5153 static void __exit bonding_exit(void)
5154 {
5155         unregister_netdevice_notifier(&bond_netdev_notifier);
5156         unregister_inetaddr_notifier(&bond_inetaddr_notifier);
5157
5158         bond_destroy_sysfs();
5159
5160         rtnl_lock();
5161         bond_free_all();
5162         rtnl_unlock();
5163 }
5164
5165 module_init(bonding_init);
5166 module_exit(bonding_exit);
5167 MODULE_LICENSE("GPL");
5168 MODULE_VERSION(DRV_VERSION);
5169 MODULE_DESCRIPTION(DRV_DESCRIPTION ", v" DRV_VERSION);
5170 MODULE_AUTHOR("Thomas Davis, tadavis@lbl.gov and many others");
5171 MODULE_SUPPORTED_DEVICE("most ethernet devices");
5172
5173 /*
5174  * Local variables:
5175  *  c-indent-level: 8
5176  *  c-basic-offset: 8
5177  *  tab-width: 8
5178  * End:
5179  */
5180