Merge branch 'master' of master.kernel.org:/pub/scm/linux/kernel/git/davem/net-2.6
[linux-2.6] / drivers / net / bonding / bond_main.c
1 /*
2  * originally based on the dummy device.
3  *
4  * Copyright 1999, Thomas Davis, tadavis@lbl.gov.
5  * Licensed under the GPL. Based on dummy.c, and eql.c devices.
6  *
7  * bonding.c: an Ethernet Bonding driver
8  *
9  * This is useful to talk to a Cisco EtherChannel compatible equipment:
10  *      Cisco 5500
11  *      Sun Trunking (Solaris)
12  *      Alteon AceDirector Trunks
13  *      Linux Bonding
14  *      and probably many L2 switches ...
15  *
16  * How it works:
17  *    ifconfig bond0 ipaddress netmask up
18  *      will setup a network device, with an ip address.  No mac address
19  *      will be assigned at this time.  The hw mac address will come from
20  *      the first slave bonded to the channel.  All slaves will then use
21  *      this hw mac address.
22  *
23  *    ifconfig bond0 down
24  *         will release all slaves, marking them as down.
25  *
26  *    ifenslave bond0 eth0
27  *      will attach eth0 to bond0 as a slave.  eth0 hw mac address will either
28  *      a: be used as initial mac address
29  *      b: if a hw mac address already is there, eth0's hw mac address
30  *         will then be set from bond0.
31  *
32  */
33
34 //#define BONDING_DEBUG 1
35
36 #include <linux/kernel.h>
37 #include <linux/module.h>
38 #include <linux/types.h>
39 #include <linux/fcntl.h>
40 #include <linux/interrupt.h>
41 #include <linux/ptrace.h>
42 #include <linux/ioport.h>
43 #include <linux/in.h>
44 #include <net/ip.h>
45 #include <linux/ip.h>
46 #include <linux/tcp.h>
47 #include <linux/udp.h>
48 #include <linux/slab.h>
49 #include <linux/string.h>
50 #include <linux/init.h>
51 #include <linux/timer.h>
52 #include <linux/socket.h>
53 #include <linux/ctype.h>
54 #include <linux/inet.h>
55 #include <linux/bitops.h>
56 #include <asm/system.h>
57 #include <asm/io.h>
58 #include <asm/dma.h>
59 #include <asm/uaccess.h>
60 #include <linux/errno.h>
61 #include <linux/netdevice.h>
62 #include <linux/inetdevice.h>
63 #include <linux/igmp.h>
64 #include <linux/etherdevice.h>
65 #include <linux/skbuff.h>
66 #include <net/sock.h>
67 #include <linux/rtnetlink.h>
68 #include <linux/proc_fs.h>
69 #include <linux/seq_file.h>
70 #include <linux/smp.h>
71 #include <linux/if_ether.h>
72 #include <net/arp.h>
73 #include <linux/mii.h>
74 #include <linux/ethtool.h>
75 #include <linux/if_vlan.h>
76 #include <linux/if_bonding.h>
77 #include <linux/jiffies.h>
78 #include <net/route.h>
79 #include <net/net_namespace.h>
80 #include "bonding.h"
81 #include "bond_3ad.h"
82 #include "bond_alb.h"
83
84 /*---------------------------- Module parameters ----------------------------*/
85
86 /* monitor all links that often (in milliseconds). <=0 disables monitoring */
87 #define BOND_LINK_MON_INTERV    0
88 #define BOND_LINK_ARP_INTERV    0
89
90 static int max_bonds    = BOND_DEFAULT_MAX_BONDS;
91 static int num_grat_arp = 1;
92 static int miimon       = BOND_LINK_MON_INTERV;
93 static int updelay      = 0;
94 static int downdelay    = 0;
95 static int use_carrier  = 1;
96 static char *mode       = NULL;
97 static char *primary    = NULL;
98 static char *lacp_rate  = NULL;
99 static char *xmit_hash_policy = NULL;
100 static int arp_interval = BOND_LINK_ARP_INTERV;
101 static char *arp_ip_target[BOND_MAX_ARP_TARGETS] = { NULL, };
102 static char *arp_validate = NULL;
103 static char *fail_over_mac = NULL;
104 struct bond_params bonding_defaults;
105
106 module_param(max_bonds, int, 0);
107 MODULE_PARM_DESC(max_bonds, "Max number of bonded devices");
108 module_param(num_grat_arp, int, 0644);
109 MODULE_PARM_DESC(num_grat_arp, "Number of gratuitous ARP packets to send on failover event");
110 module_param(miimon, int, 0);
111 MODULE_PARM_DESC(miimon, "Link check interval in milliseconds");
112 module_param(updelay, int, 0);
113 MODULE_PARM_DESC(updelay, "Delay before considering link up, in milliseconds");
114 module_param(downdelay, int, 0);
115 MODULE_PARM_DESC(downdelay, "Delay before considering link down, "
116                             "in milliseconds");
117 module_param(use_carrier, int, 0);
118 MODULE_PARM_DESC(use_carrier, "Use netif_carrier_ok (vs MII ioctls) in miimon; "
119                               "0 for off, 1 for on (default)");
120 module_param(mode, charp, 0);
121 MODULE_PARM_DESC(mode, "Mode of operation : 0 for balance-rr, "
122                        "1 for active-backup, 2 for balance-xor, "
123                        "3 for broadcast, 4 for 802.3ad, 5 for balance-tlb, "
124                        "6 for balance-alb");
125 module_param(primary, charp, 0);
126 MODULE_PARM_DESC(primary, "Primary network device to use");
127 module_param(lacp_rate, charp, 0);
128 MODULE_PARM_DESC(lacp_rate, "LACPDU tx rate to request from 802.3ad partner "
129                             "(slow/fast)");
130 module_param(xmit_hash_policy, charp, 0);
131 MODULE_PARM_DESC(xmit_hash_policy, "XOR hashing method: 0 for layer 2 (default)"
132                                    ", 1 for layer 3+4");
133 module_param(arp_interval, int, 0);
134 MODULE_PARM_DESC(arp_interval, "arp interval in milliseconds");
135 module_param_array(arp_ip_target, charp, NULL, 0);
136 MODULE_PARM_DESC(arp_ip_target, "arp targets in n.n.n.n form");
137 module_param(arp_validate, charp, 0);
138 MODULE_PARM_DESC(arp_validate, "validate src/dst of ARP probes: none (default), active, backup or all");
139 module_param(fail_over_mac, charp, 0);
140 MODULE_PARM_DESC(fail_over_mac, "For active-backup, do not set all slaves to the same MAC.  none (default), active or follow");
141
142 /*----------------------------- Global variables ----------------------------*/
143
144 static const char * const version =
145         DRV_DESCRIPTION ": v" DRV_VERSION " (" DRV_RELDATE ")\n";
146
147 LIST_HEAD(bond_dev_list);
148
149 #ifdef CONFIG_PROC_FS
150 static struct proc_dir_entry *bond_proc_dir = NULL;
151 #endif
152
153 extern struct rw_semaphore bonding_rwsem;
154 static __be32 arp_target[BOND_MAX_ARP_TARGETS] = { 0, } ;
155 static int arp_ip_count = 0;
156 static int bond_mode    = BOND_MODE_ROUNDROBIN;
157 static int xmit_hashtype= BOND_XMIT_POLICY_LAYER2;
158 static int lacp_fast    = 0;
159
160
161 struct bond_parm_tbl bond_lacp_tbl[] = {
162 {       "slow",         AD_LACP_SLOW},
163 {       "fast",         AD_LACP_FAST},
164 {       NULL,           -1},
165 };
166
167 struct bond_parm_tbl bond_mode_tbl[] = {
168 {       "balance-rr",           BOND_MODE_ROUNDROBIN},
169 {       "active-backup",        BOND_MODE_ACTIVEBACKUP},
170 {       "balance-xor",          BOND_MODE_XOR},
171 {       "broadcast",            BOND_MODE_BROADCAST},
172 {       "802.3ad",              BOND_MODE_8023AD},
173 {       "balance-tlb",          BOND_MODE_TLB},
174 {       "balance-alb",          BOND_MODE_ALB},
175 {       NULL,                   -1},
176 };
177
178 struct bond_parm_tbl xmit_hashtype_tbl[] = {
179 {       "layer2",               BOND_XMIT_POLICY_LAYER2},
180 {       "layer3+4",             BOND_XMIT_POLICY_LAYER34},
181 {       "layer2+3",             BOND_XMIT_POLICY_LAYER23},
182 {       NULL,                   -1},
183 };
184
185 struct bond_parm_tbl arp_validate_tbl[] = {
186 {       "none",                 BOND_ARP_VALIDATE_NONE},
187 {       "active",               BOND_ARP_VALIDATE_ACTIVE},
188 {       "backup",               BOND_ARP_VALIDATE_BACKUP},
189 {       "all",                  BOND_ARP_VALIDATE_ALL},
190 {       NULL,                   -1},
191 };
192
193 struct bond_parm_tbl fail_over_mac_tbl[] = {
194 {       "none",                 BOND_FOM_NONE},
195 {       "active",               BOND_FOM_ACTIVE},
196 {       "follow",               BOND_FOM_FOLLOW},
197 {       NULL,                   -1},
198 };
199
200 /*-------------------------- Forward declarations ---------------------------*/
201
202 static void bond_send_gratuitous_arp(struct bonding *bond);
203 static void bond_deinit(struct net_device *bond_dev);
204
205 /*---------------------------- General routines -----------------------------*/
206
207 static const char *bond_mode_name(int mode)
208 {
209         switch (mode) {
210         case BOND_MODE_ROUNDROBIN :
211                 return "load balancing (round-robin)";
212         case BOND_MODE_ACTIVEBACKUP :
213                 return "fault-tolerance (active-backup)";
214         case BOND_MODE_XOR :
215                 return "load balancing (xor)";
216         case BOND_MODE_BROADCAST :
217                 return "fault-tolerance (broadcast)";
218         case BOND_MODE_8023AD:
219                 return "IEEE 802.3ad Dynamic link aggregation";
220         case BOND_MODE_TLB:
221                 return "transmit load balancing";
222         case BOND_MODE_ALB:
223                 return "adaptive load balancing";
224         default:
225                 return "unknown";
226         }
227 }
228
229 /*---------------------------------- VLAN -----------------------------------*/
230
231 /**
232  * bond_add_vlan - add a new vlan id on bond
233  * @bond: bond that got the notification
234  * @vlan_id: the vlan id to add
235  *
236  * Returns -ENOMEM if allocation failed.
237  */
238 static int bond_add_vlan(struct bonding *bond, unsigned short vlan_id)
239 {
240         struct vlan_entry *vlan;
241
242         dprintk("bond: %s, vlan id %d\n",
243                 (bond ? bond->dev->name: "None"), vlan_id);
244
245         vlan = kmalloc(sizeof(struct vlan_entry), GFP_KERNEL);
246         if (!vlan) {
247                 return -ENOMEM;
248         }
249
250         INIT_LIST_HEAD(&vlan->vlan_list);
251         vlan->vlan_id = vlan_id;
252         vlan->vlan_ip = 0;
253
254         write_lock_bh(&bond->lock);
255
256         list_add_tail(&vlan->vlan_list, &bond->vlan_list);
257
258         write_unlock_bh(&bond->lock);
259
260         dprintk("added VLAN ID %d on bond %s\n", vlan_id, bond->dev->name);
261
262         return 0;
263 }
264
265 /**
266  * bond_del_vlan - delete a vlan id from bond
267  * @bond: bond that got the notification
268  * @vlan_id: the vlan id to delete
269  *
270  * returns -ENODEV if @vlan_id was not found in @bond.
271  */
272 static int bond_del_vlan(struct bonding *bond, unsigned short vlan_id)
273 {
274         struct vlan_entry *vlan;
275         int res = -ENODEV;
276
277         dprintk("bond: %s, vlan id %d\n", bond->dev->name, vlan_id);
278
279         write_lock_bh(&bond->lock);
280
281         list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
282                 if (vlan->vlan_id == vlan_id) {
283                         list_del(&vlan->vlan_list);
284
285                         if ((bond->params.mode == BOND_MODE_TLB) ||
286                             (bond->params.mode == BOND_MODE_ALB)) {
287                                 bond_alb_clear_vlan(bond, vlan_id);
288                         }
289
290                         dprintk("removed VLAN ID %d from bond %s\n", vlan_id,
291                                 bond->dev->name);
292
293                         kfree(vlan);
294
295                         if (list_empty(&bond->vlan_list) &&
296                             (bond->slave_cnt == 0)) {
297                                 /* Last VLAN removed and no slaves, so
298                                  * restore block on adding VLANs. This will
299                                  * be removed once new slaves that are not
300                                  * VLAN challenged will be added.
301                                  */
302                                 bond->dev->features |= NETIF_F_VLAN_CHALLENGED;
303                         }
304
305                         res = 0;
306                         goto out;
307                 }
308         }
309
310         dprintk("couldn't find VLAN ID %d in bond %s\n", vlan_id,
311                 bond->dev->name);
312
313 out:
314         write_unlock_bh(&bond->lock);
315         return res;
316 }
317
318 /**
319  * bond_has_challenged_slaves
320  * @bond: the bond we're working on
321  *
322  * Searches the slave list. Returns 1 if a vlan challenged slave
323  * was found, 0 otherwise.
324  *
325  * Assumes bond->lock is held.
326  */
327 static int bond_has_challenged_slaves(struct bonding *bond)
328 {
329         struct slave *slave;
330         int i;
331
332         bond_for_each_slave(bond, slave, i) {
333                 if (slave->dev->features & NETIF_F_VLAN_CHALLENGED) {
334                         dprintk("found VLAN challenged slave - %s\n",
335                                 slave->dev->name);
336                         return 1;
337                 }
338         }
339
340         dprintk("no VLAN challenged slaves found\n");
341         return 0;
342 }
343
344 /**
345  * bond_next_vlan - safely skip to the next item in the vlans list.
346  * @bond: the bond we're working on
347  * @curr: item we're advancing from
348  *
349  * Returns %NULL if list is empty, bond->next_vlan if @curr is %NULL,
350  * or @curr->next otherwise (even if it is @curr itself again).
351  * 
352  * Caller must hold bond->lock
353  */
354 struct vlan_entry *bond_next_vlan(struct bonding *bond, struct vlan_entry *curr)
355 {
356         struct vlan_entry *next, *last;
357
358         if (list_empty(&bond->vlan_list)) {
359                 return NULL;
360         }
361
362         if (!curr) {
363                 next = list_entry(bond->vlan_list.next,
364                                   struct vlan_entry, vlan_list);
365         } else {
366                 last = list_entry(bond->vlan_list.prev,
367                                   struct vlan_entry, vlan_list);
368                 if (last == curr) {
369                         next = list_entry(bond->vlan_list.next,
370                                           struct vlan_entry, vlan_list);
371                 } else {
372                         next = list_entry(curr->vlan_list.next,
373                                           struct vlan_entry, vlan_list);
374                 }
375         }
376
377         return next;
378 }
379
380 /**
381  * bond_dev_queue_xmit - Prepare skb for xmit.
382  * 
383  * @bond: bond device that got this skb for tx.
384  * @skb: hw accel VLAN tagged skb to transmit
385  * @slave_dev: slave that is supposed to xmit this skbuff
386  * 
387  * When the bond gets an skb to transmit that is
388  * already hardware accelerated VLAN tagged, and it
389  * needs to relay this skb to a slave that is not
390  * hw accel capable, the skb needs to be "unaccelerated",
391  * i.e. strip the hwaccel tag and re-insert it as part
392  * of the payload.
393  */
394 int bond_dev_queue_xmit(struct bonding *bond, struct sk_buff *skb, struct net_device *slave_dev)
395 {
396         unsigned short uninitialized_var(vlan_id);
397
398         if (!list_empty(&bond->vlan_list) &&
399             !(slave_dev->features & NETIF_F_HW_VLAN_TX) &&
400             vlan_get_tag(skb, &vlan_id) == 0) {
401                 skb->dev = slave_dev;
402                 skb = vlan_put_tag(skb, vlan_id);
403                 if (!skb) {
404                         /* vlan_put_tag() frees the skb in case of error,
405                          * so return success here so the calling functions
406                          * won't attempt to free is again.
407                          */
408                         return 0;
409                 }
410         } else {
411                 skb->dev = slave_dev;
412         }
413
414         skb->priority = 1;
415         dev_queue_xmit(skb);
416
417         return 0;
418 }
419
420 /*
421  * In the following 3 functions, bond_vlan_rx_register(), bond_vlan_rx_add_vid
422  * and bond_vlan_rx_kill_vid, We don't protect the slave list iteration with a
423  * lock because:
424  * a. This operation is performed in IOCTL context,
425  * b. The operation is protected by the RTNL semaphore in the 8021q code,
426  * c. Holding a lock with BH disabled while directly calling a base driver
427  *    entry point is generally a BAD idea.
428  * 
429  * The design of synchronization/protection for this operation in the 8021q
430  * module is good for one or more VLAN devices over a single physical device
431  * and cannot be extended for a teaming solution like bonding, so there is a
432  * potential race condition here where a net device from the vlan group might
433  * be referenced (either by a base driver or the 8021q code) while it is being
434  * removed from the system. However, it turns out we're not making matters
435  * worse, and if it works for regular VLAN usage it will work here too.
436 */
437
438 /**
439  * bond_vlan_rx_register - Propagates registration to slaves
440  * @bond_dev: bonding net device that got called
441  * @grp: vlan group being registered
442  */
443 static void bond_vlan_rx_register(struct net_device *bond_dev, struct vlan_group *grp)
444 {
445         struct bonding *bond = bond_dev->priv;
446         struct slave *slave;
447         int i;
448
449         bond->vlgrp = grp;
450
451         bond_for_each_slave(bond, slave, i) {
452                 struct net_device *slave_dev = slave->dev;
453
454                 if ((slave_dev->features & NETIF_F_HW_VLAN_RX) &&
455                     slave_dev->vlan_rx_register) {
456                         slave_dev->vlan_rx_register(slave_dev, grp);
457                 }
458         }
459 }
460
461 /**
462  * bond_vlan_rx_add_vid - Propagates adding an id to slaves
463  * @bond_dev: bonding net device that got called
464  * @vid: vlan id being added
465  */
466 static void bond_vlan_rx_add_vid(struct net_device *bond_dev, uint16_t vid)
467 {
468         struct bonding *bond = bond_dev->priv;
469         struct slave *slave;
470         int i, res;
471
472         bond_for_each_slave(bond, slave, i) {
473                 struct net_device *slave_dev = slave->dev;
474
475                 if ((slave_dev->features & NETIF_F_HW_VLAN_FILTER) &&
476                     slave_dev->vlan_rx_add_vid) {
477                         slave_dev->vlan_rx_add_vid(slave_dev, vid);
478                 }
479         }
480
481         res = bond_add_vlan(bond, vid);
482         if (res) {
483                 printk(KERN_ERR DRV_NAME
484                        ": %s: Error: Failed to add vlan id %d\n",
485                        bond_dev->name, vid);
486         }
487 }
488
489 /**
490  * bond_vlan_rx_kill_vid - Propagates deleting an id to slaves
491  * @bond_dev: bonding net device that got called
492  * @vid: vlan id being removed
493  */
494 static void bond_vlan_rx_kill_vid(struct net_device *bond_dev, uint16_t vid)
495 {
496         struct bonding *bond = bond_dev->priv;
497         struct slave *slave;
498         struct net_device *vlan_dev;
499         int i, res;
500
501         bond_for_each_slave(bond, slave, i) {
502                 struct net_device *slave_dev = slave->dev;
503
504                 if ((slave_dev->features & NETIF_F_HW_VLAN_FILTER) &&
505                     slave_dev->vlan_rx_kill_vid) {
506                         /* Save and then restore vlan_dev in the grp array,
507                          * since the slave's driver might clear it.
508                          */
509                         vlan_dev = vlan_group_get_device(bond->vlgrp, vid);
510                         slave_dev->vlan_rx_kill_vid(slave_dev, vid);
511                         vlan_group_set_device(bond->vlgrp, vid, vlan_dev);
512                 }
513         }
514
515         res = bond_del_vlan(bond, vid);
516         if (res) {
517                 printk(KERN_ERR DRV_NAME
518                        ": %s: Error: Failed to remove vlan id %d\n",
519                        bond_dev->name, vid);
520         }
521 }
522
523 static void bond_add_vlans_on_slave(struct bonding *bond, struct net_device *slave_dev)
524 {
525         struct vlan_entry *vlan;
526
527         write_lock_bh(&bond->lock);
528
529         if (list_empty(&bond->vlan_list)) {
530                 goto out;
531         }
532
533         if ((slave_dev->features & NETIF_F_HW_VLAN_RX) &&
534             slave_dev->vlan_rx_register) {
535                 slave_dev->vlan_rx_register(slave_dev, bond->vlgrp);
536         }
537
538         if (!(slave_dev->features & NETIF_F_HW_VLAN_FILTER) ||
539             !(slave_dev->vlan_rx_add_vid)) {
540                 goto out;
541         }
542
543         list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
544                 slave_dev->vlan_rx_add_vid(slave_dev, vlan->vlan_id);
545         }
546
547 out:
548         write_unlock_bh(&bond->lock);
549 }
550
551 static void bond_del_vlans_from_slave(struct bonding *bond, struct net_device *slave_dev)
552 {
553         struct vlan_entry *vlan;
554         struct net_device *vlan_dev;
555
556         write_lock_bh(&bond->lock);
557
558         if (list_empty(&bond->vlan_list)) {
559                 goto out;
560         }
561
562         if (!(slave_dev->features & NETIF_F_HW_VLAN_FILTER) ||
563             !(slave_dev->vlan_rx_kill_vid)) {
564                 goto unreg;
565         }
566
567         list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
568                 /* Save and then restore vlan_dev in the grp array,
569                  * since the slave's driver might clear it.
570                  */
571                 vlan_dev = vlan_group_get_device(bond->vlgrp, vlan->vlan_id);
572                 slave_dev->vlan_rx_kill_vid(slave_dev, vlan->vlan_id);
573                 vlan_group_set_device(bond->vlgrp, vlan->vlan_id, vlan_dev);
574         }
575
576 unreg:
577         if ((slave_dev->features & NETIF_F_HW_VLAN_RX) &&
578             slave_dev->vlan_rx_register) {
579                 slave_dev->vlan_rx_register(slave_dev, NULL);
580         }
581
582 out:
583         write_unlock_bh(&bond->lock);
584 }
585
586 /*------------------------------- Link status -------------------------------*/
587
588 /*
589  * Set the carrier state for the master according to the state of its
590  * slaves.  If any slaves are up, the master is up.  In 802.3ad mode,
591  * do special 802.3ad magic.
592  *
593  * Returns zero if carrier state does not change, nonzero if it does.
594  */
595 static int bond_set_carrier(struct bonding *bond)
596 {
597         struct slave *slave;
598         int i;
599
600         if (bond->slave_cnt == 0)
601                 goto down;
602
603         if (bond->params.mode == BOND_MODE_8023AD)
604                 return bond_3ad_set_carrier(bond);
605
606         bond_for_each_slave(bond, slave, i) {
607                 if (slave->link == BOND_LINK_UP) {
608                         if (!netif_carrier_ok(bond->dev)) {
609                                 netif_carrier_on(bond->dev);
610                                 return 1;
611                         }
612                         return 0;
613                 }
614         }
615
616 down:
617         if (netif_carrier_ok(bond->dev)) {
618                 netif_carrier_off(bond->dev);
619                 return 1;
620         }
621         return 0;
622 }
623
624 /*
625  * Get link speed and duplex from the slave's base driver
626  * using ethtool. If for some reason the call fails or the
627  * values are invalid, fake speed and duplex to 100/Full
628  * and return error.
629  */
630 static int bond_update_speed_duplex(struct slave *slave)
631 {
632         struct net_device *slave_dev = slave->dev;
633         struct ethtool_cmd etool;
634         int res;
635
636         /* Fake speed and duplex */
637         slave->speed = SPEED_100;
638         slave->duplex = DUPLEX_FULL;
639
640         if (!slave_dev->ethtool_ops || !slave_dev->ethtool_ops->get_settings)
641                 return -1;
642
643         res = slave_dev->ethtool_ops->get_settings(slave_dev, &etool);
644         if (res < 0)
645                 return -1;
646
647         switch (etool.speed) {
648         case SPEED_10:
649         case SPEED_100:
650         case SPEED_1000:
651         case SPEED_10000:
652                 break;
653         default:
654                 return -1;
655         }
656
657         switch (etool.duplex) {
658         case DUPLEX_FULL:
659         case DUPLEX_HALF:
660                 break;
661         default:
662                 return -1;
663         }
664
665         slave->speed = etool.speed;
666         slave->duplex = etool.duplex;
667
668         return 0;
669 }
670
671 /*
672  * if <dev> supports MII link status reporting, check its link status.
673  *
674  * We either do MII/ETHTOOL ioctls, or check netif_carrier_ok(),
675  * depening upon the setting of the use_carrier parameter.
676  *
677  * Return either BMSR_LSTATUS, meaning that the link is up (or we
678  * can't tell and just pretend it is), or 0, meaning that the link is
679  * down.
680  *
681  * If reporting is non-zero, instead of faking link up, return -1 if
682  * both ETHTOOL and MII ioctls fail (meaning the device does not
683  * support them).  If use_carrier is set, return whatever it says.
684  * It'd be nice if there was a good way to tell if a driver supports
685  * netif_carrier, but there really isn't.
686  */
687 static int bond_check_dev_link(struct bonding *bond, struct net_device *slave_dev, int reporting)
688 {
689         static int (* ioctl)(struct net_device *, struct ifreq *, int);
690         struct ifreq ifr;
691         struct mii_ioctl_data *mii;
692
693         if (bond->params.use_carrier) {
694                 return netif_carrier_ok(slave_dev) ? BMSR_LSTATUS : 0;
695         }
696
697         ioctl = slave_dev->do_ioctl;
698         if (ioctl) {
699                 /* TODO: set pointer to correct ioctl on a per team member */
700                 /*       bases to make this more efficient. that is, once  */
701                 /*       we determine the correct ioctl, we will always    */
702                 /*       call it and not the others for that team          */
703                 /*       member.                                           */
704
705                 /*
706                  * We cannot assume that SIOCGMIIPHY will also read a
707                  * register; not all network drivers (e.g., e100)
708                  * support that.
709                  */
710
711                 /* Yes, the mii is overlaid on the ifreq.ifr_ifru */
712                 strncpy(ifr.ifr_name, slave_dev->name, IFNAMSIZ);
713                 mii = if_mii(&ifr);
714                 if (IOCTL(slave_dev, &ifr, SIOCGMIIPHY) == 0) {
715                         mii->reg_num = MII_BMSR;
716                         if (IOCTL(slave_dev, &ifr, SIOCGMIIREG) == 0) {
717                                 return (mii->val_out & BMSR_LSTATUS);
718                         }
719                 }
720         }
721
722         /*
723          * Some drivers cache ETHTOOL_GLINK for a period of time so we only
724          * attempt to get link status from it if the above MII ioctls fail.
725          */
726         if (slave_dev->ethtool_ops) {
727                 if (slave_dev->ethtool_ops->get_link) {
728                         u32 link;
729
730                         link = slave_dev->ethtool_ops->get_link(slave_dev);
731
732                         return link ? BMSR_LSTATUS : 0;
733                 }
734         }
735
736         /*
737          * If reporting, report that either there's no dev->do_ioctl,
738          * or both SIOCGMIIREG and get_link failed (meaning that we
739          * cannot report link status).  If not reporting, pretend
740          * we're ok.
741          */
742         return (reporting ? -1 : BMSR_LSTATUS);
743 }
744
745 /*----------------------------- Multicast list ------------------------------*/
746
747 /*
748  * Returns 0 if dmi1 and dmi2 are the same, non-0 otherwise
749  */
750 static inline int bond_is_dmi_same(struct dev_mc_list *dmi1, struct dev_mc_list *dmi2)
751 {
752         return memcmp(dmi1->dmi_addr, dmi2->dmi_addr, dmi1->dmi_addrlen) == 0 &&
753                         dmi1->dmi_addrlen == dmi2->dmi_addrlen;
754 }
755
756 /*
757  * returns dmi entry if found, NULL otherwise
758  */
759 static struct dev_mc_list *bond_mc_list_find_dmi(struct dev_mc_list *dmi, struct dev_mc_list *mc_list)
760 {
761         struct dev_mc_list *idmi;
762
763         for (idmi = mc_list; idmi; idmi = idmi->next) {
764                 if (bond_is_dmi_same(dmi, idmi)) {
765                         return idmi;
766                 }
767         }
768
769         return NULL;
770 }
771
772 /*
773  * Push the promiscuity flag down to appropriate slaves
774  */
775 static 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                         bond_send_gratuitous_arp(bond);
1199
1200                         write_unlock_bh(&bond->curr_slave_lock);
1201                         read_unlock(&bond->lock);
1202
1203                         netdev_bonding_change(bond->dev);
1204
1205                         read_lock(&bond->lock);
1206                         write_lock_bh(&bond->curr_slave_lock);
1207                 }
1208         }
1209 }
1210
1211 /**
1212  * bond_select_active_slave - select a new active slave, if needed
1213  * @bond: our bonding struct
1214  *
1215  * This functions shoud be called when one of the following occurs:
1216  * - The old curr_active_slave has been released or lost its link.
1217  * - The primary_slave has got its link back.
1218  * - A slave has got its link back and there's no old curr_active_slave.
1219  *
1220  * Caller must hold bond->lock for read and curr_slave_lock for write_bh.
1221  */
1222 void bond_select_active_slave(struct bonding *bond)
1223 {
1224         struct slave *best_slave;
1225         int rv;
1226
1227         best_slave = bond_find_best_slave(bond);
1228         if (best_slave != bond->curr_active_slave) {
1229                 bond_change_active_slave(bond, best_slave);
1230                 rv = bond_set_carrier(bond);
1231                 if (!rv)
1232                         return;
1233
1234                 if (netif_carrier_ok(bond->dev)) {
1235                         printk(KERN_INFO DRV_NAME
1236                                ": %s: first active interface up!\n",
1237                                bond->dev->name);
1238                 } else {
1239                         printk(KERN_INFO DRV_NAME ": %s: "
1240                                "now running without any active interface !\n",
1241                                bond->dev->name);
1242                 }
1243         }
1244 }
1245
1246 /*--------------------------- slave list handling ---------------------------*/
1247
1248 /*
1249  * This function attaches the slave to the end of list.
1250  *
1251  * bond->lock held for writing by caller.
1252  */
1253 static void bond_attach_slave(struct bonding *bond, struct slave *new_slave)
1254 {
1255         if (bond->first_slave == NULL) { /* attaching the first slave */
1256                 new_slave->next = new_slave;
1257                 new_slave->prev = new_slave;
1258                 bond->first_slave = new_slave;
1259         } else {
1260                 new_slave->next = bond->first_slave;
1261                 new_slave->prev = bond->first_slave->prev;
1262                 new_slave->next->prev = new_slave;
1263                 new_slave->prev->next = new_slave;
1264         }
1265
1266         bond->slave_cnt++;
1267 }
1268
1269 /*
1270  * This function detaches the slave from the list.
1271  * WARNING: no check is made to verify if the slave effectively
1272  * belongs to <bond>.
1273  * Nothing is freed on return, structures are just unchained.
1274  * If any slave pointer in bond was pointing to <slave>,
1275  * it should be changed by the calling function.
1276  *
1277  * bond->lock held for writing by caller.
1278  */
1279 static void bond_detach_slave(struct bonding *bond, struct slave *slave)
1280 {
1281         if (slave->next) {
1282                 slave->next->prev = slave->prev;
1283         }
1284
1285         if (slave->prev) {
1286                 slave->prev->next = slave->next;
1287         }
1288
1289         if (bond->first_slave == slave) { /* slave is the first slave */
1290                 if (bond->slave_cnt > 1) { /* there are more slave */
1291                         bond->first_slave = slave->next;
1292                 } else {
1293                         bond->first_slave = NULL; /* slave was the last one */
1294                 }
1295         }
1296
1297         slave->next = NULL;
1298         slave->prev = NULL;
1299         bond->slave_cnt--;
1300 }
1301
1302 /*---------------------------------- IOCTL ----------------------------------*/
1303
1304 static int bond_sethwaddr(struct net_device *bond_dev,
1305                           struct net_device *slave_dev)
1306 {
1307         dprintk("bond_dev=%p\n", bond_dev);
1308         dprintk("slave_dev=%p\n", slave_dev);
1309         dprintk("slave_dev->addr_len=%d\n", slave_dev->addr_len);
1310         memcpy(bond_dev->dev_addr, slave_dev->dev_addr, slave_dev->addr_len);
1311         return 0;
1312 }
1313
1314 #define BOND_VLAN_FEATURES \
1315         (NETIF_F_VLAN_CHALLENGED | NETIF_F_HW_VLAN_RX | NETIF_F_HW_VLAN_TX | \
1316          NETIF_F_HW_VLAN_FILTER)
1317
1318 /* 
1319  * Compute the common dev->feature set available to all slaves.  Some
1320  * feature bits are managed elsewhere, so preserve those feature bits
1321  * on the master device.
1322  */
1323 static int bond_compute_features(struct bonding *bond)
1324 {
1325         struct slave *slave;
1326         struct net_device *bond_dev = bond->dev;
1327         unsigned long features = bond_dev->features;
1328         unsigned short max_hard_header_len = max((u16)ETH_HLEN,
1329                                                 bond_dev->hard_header_len);
1330         int i;
1331
1332         features &= ~(NETIF_F_ALL_CSUM | BOND_VLAN_FEATURES);
1333         features |= NETIF_F_SG | NETIF_F_FRAGLIST | NETIF_F_HIGHDMA |
1334                     NETIF_F_GSO_MASK | NETIF_F_NO_CSUM;
1335
1336         bond_for_each_slave(bond, slave, i) {
1337                 features = netdev_compute_features(features,
1338                                                    slave->dev->features);
1339                 if (slave->dev->hard_header_len > max_hard_header_len)
1340                         max_hard_header_len = slave->dev->hard_header_len;
1341         }
1342
1343         features |= (bond_dev->features & BOND_VLAN_FEATURES);
1344         bond_dev->features = features;
1345         bond_dev->hard_header_len = max_hard_header_len;
1346
1347         return 0;
1348 }
1349
1350
1351 static void bond_setup_by_slave(struct net_device *bond_dev,
1352                                 struct net_device *slave_dev)
1353 {
1354         struct bonding *bond = bond_dev->priv;
1355
1356         bond_dev->neigh_setup           = slave_dev->neigh_setup;
1357         bond_dev->header_ops            = slave_dev->header_ops;
1358
1359         bond_dev->type              = slave_dev->type;
1360         bond_dev->hard_header_len   = slave_dev->hard_header_len;
1361         bond_dev->addr_len          = slave_dev->addr_len;
1362
1363         memcpy(bond_dev->broadcast, slave_dev->broadcast,
1364                 slave_dev->addr_len);
1365         bond->setup_by_slave = 1;
1366 }
1367
1368 /* enslave device <slave> to bond device <master> */
1369 int bond_enslave(struct net_device *bond_dev, struct net_device *slave_dev)
1370 {
1371         struct bonding *bond = bond_dev->priv;
1372         struct slave *new_slave = NULL;
1373         struct dev_mc_list *dmi;
1374         struct sockaddr addr;
1375         int link_reporting;
1376         int old_features = bond_dev->features;
1377         int res = 0;
1378
1379         if (!bond->params.use_carrier && slave_dev->ethtool_ops == NULL &&
1380                 slave_dev->do_ioctl == NULL) {
1381                 printk(KERN_WARNING DRV_NAME
1382                        ": %s: Warning: no link monitoring support for %s\n",
1383                        bond_dev->name, slave_dev->name);
1384         }
1385
1386         /* bond must be initialized by bond_open() before enslaving */
1387         if (!(bond_dev->flags & IFF_UP)) {
1388                 printk(KERN_WARNING DRV_NAME
1389                         " %s: master_dev is not up in bond_enslave\n",
1390                         bond_dev->name);
1391         }
1392
1393         /* already enslaved */
1394         if (slave_dev->flags & IFF_SLAVE) {
1395                 dprintk("Error, Device was already enslaved\n");
1396                 return -EBUSY;
1397         }
1398
1399         /* vlan challenged mutual exclusion */
1400         /* no need to lock since we're protected by rtnl_lock */
1401         if (slave_dev->features & NETIF_F_VLAN_CHALLENGED) {
1402                 dprintk("%s: NETIF_F_VLAN_CHALLENGED\n", slave_dev->name);
1403                 if (!list_empty(&bond->vlan_list)) {
1404                         printk(KERN_ERR DRV_NAME
1405                                ": %s: Error: cannot enslave VLAN "
1406                                "challenged slave %s on VLAN enabled "
1407                                "bond %s\n", bond_dev->name, slave_dev->name,
1408                                bond_dev->name);
1409                         return -EPERM;
1410                 } else {
1411                         printk(KERN_WARNING DRV_NAME
1412                                ": %s: Warning: enslaved VLAN challenged "
1413                                "slave %s. Adding VLANs will be blocked as "
1414                                "long as %s is part of bond %s\n",
1415                                bond_dev->name, slave_dev->name, slave_dev->name,
1416                                bond_dev->name);
1417                         bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
1418                 }
1419         } else {
1420                 dprintk("%s: ! NETIF_F_VLAN_CHALLENGED\n", slave_dev->name);
1421                 if (bond->slave_cnt == 0) {
1422                         /* First slave, and it is not VLAN challenged,
1423                          * so remove the block of adding VLANs over the bond.
1424                          */
1425                         bond_dev->features &= ~NETIF_F_VLAN_CHALLENGED;
1426                 }
1427         }
1428
1429         /*
1430          * Old ifenslave binaries are no longer supported.  These can
1431          * be identified with moderate accurary by the state of the slave:
1432          * the current ifenslave will set the interface down prior to
1433          * enslaving it; the old ifenslave will not.
1434          */
1435         if ((slave_dev->flags & IFF_UP)) {
1436                 printk(KERN_ERR DRV_NAME ": %s is up. "
1437                        "This may be due to an out of date ifenslave.\n",
1438                        slave_dev->name);
1439                 res = -EPERM;
1440                 goto err_undo_flags;
1441         }
1442
1443         /* set bonding device ether type by slave - bonding netdevices are
1444          * created with ether_setup, so when the slave type is not ARPHRD_ETHER
1445          * there is a need to override some of the type dependent attribs/funcs.
1446          *
1447          * bond ether type mutual exclusion - don't allow slaves of dissimilar
1448          * ether type (eg ARPHRD_ETHER and ARPHRD_INFINIBAND) share the same bond
1449          */
1450         if (bond->slave_cnt == 0) {
1451                 if (slave_dev->type != ARPHRD_ETHER)
1452                         bond_setup_by_slave(bond_dev, slave_dev);
1453         } else if (bond_dev->type != slave_dev->type) {
1454                 printk(KERN_ERR DRV_NAME ": %s ether type (%d) is different "
1455                         "from other slaves (%d), can not enslave it.\n",
1456                         slave_dev->name,
1457                         slave_dev->type, bond_dev->type);
1458                         res = -EINVAL;
1459                         goto err_undo_flags;
1460         }
1461
1462         if (slave_dev->set_mac_address == NULL) {
1463                 if (bond->slave_cnt == 0) {
1464                         printk(KERN_WARNING DRV_NAME
1465                                ": %s: Warning: The first slave device "
1466                                "specified does not support setting the MAC "
1467                                "address. Setting fail_over_mac to active.",
1468                                bond_dev->name);
1469                         bond->params.fail_over_mac = BOND_FOM_ACTIVE;
1470                 } else if (bond->params.fail_over_mac != BOND_FOM_ACTIVE) {
1471                         printk(KERN_ERR DRV_NAME
1472                                 ": %s: Error: The slave device specified "
1473                                 "does not support setting the MAC address, "
1474                                 "but fail_over_mac is not set to active.\n"
1475                                 , bond_dev->name);
1476                         res = -EOPNOTSUPP;
1477                         goto err_undo_flags;
1478                 }
1479         }
1480
1481         new_slave = kzalloc(sizeof(struct slave), GFP_KERNEL);
1482         if (!new_slave) {
1483                 res = -ENOMEM;
1484                 goto err_undo_flags;
1485         }
1486
1487         /* save slave's original flags before calling
1488          * netdev_set_master and dev_open
1489          */
1490         new_slave->original_flags = slave_dev->flags;
1491
1492         /*
1493          * Save slave's original ("permanent") mac address for modes
1494          * that need it, and for restoring it upon release, and then
1495          * set it to the master's address
1496          */
1497         memcpy(new_slave->perm_hwaddr, slave_dev->dev_addr, ETH_ALEN);
1498
1499         if (!bond->params.fail_over_mac) {
1500                 /*
1501                  * Set slave to master's mac address.  The application already
1502                  * set the master's mac address to that of the first slave
1503                  */
1504                 memcpy(addr.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
1505                 addr.sa_family = slave_dev->type;
1506                 res = dev_set_mac_address(slave_dev, &addr);
1507                 if (res) {
1508                         dprintk("Error %d calling set_mac_address\n", res);
1509                         goto err_free;
1510                 }
1511         }
1512
1513         res = netdev_set_master(slave_dev, bond_dev);
1514         if (res) {
1515                 dprintk("Error %d calling netdev_set_master\n", res);
1516                 goto err_restore_mac;
1517         }
1518         /* open the slave since the application closed it */
1519         res = dev_open(slave_dev);
1520         if (res) {
1521                 dprintk("Openning slave %s failed\n", slave_dev->name);
1522                 goto err_unset_master;
1523         }
1524
1525         new_slave->dev = slave_dev;
1526         slave_dev->priv_flags |= IFF_BONDING;
1527
1528         if ((bond->params.mode == BOND_MODE_TLB) ||
1529             (bond->params.mode == BOND_MODE_ALB)) {
1530                 /* bond_alb_init_slave() must be called before all other stages since
1531                  * it might fail and we do not want to have to undo everything
1532                  */
1533                 res = bond_alb_init_slave(bond, new_slave);
1534                 if (res) {
1535                         goto err_close;
1536                 }
1537         }
1538
1539         /* If the mode USES_PRIMARY, then the new slave gets the
1540          * master's promisc (and mc) settings only if it becomes the
1541          * curr_active_slave, and that is taken care of later when calling
1542          * bond_change_active()
1543          */
1544         if (!USES_PRIMARY(bond->params.mode)) {
1545                 /* set promiscuity level to new slave */
1546                 if (bond_dev->flags & IFF_PROMISC) {
1547                         dev_set_promiscuity(slave_dev, 1);
1548                 }
1549
1550                 /* set allmulti level to new slave */
1551                 if (bond_dev->flags & IFF_ALLMULTI) {
1552                         dev_set_allmulti(slave_dev, 1);
1553                 }
1554
1555                 netif_tx_lock_bh(bond_dev);
1556                 /* upload master's mc_list to new slave */
1557                 for (dmi = bond_dev->mc_list; dmi; dmi = dmi->next) {
1558                         dev_mc_add (slave_dev, dmi->dmi_addr, dmi->dmi_addrlen, 0);
1559                 }
1560                 netif_tx_unlock_bh(bond_dev);
1561         }
1562
1563         if (bond->params.mode == BOND_MODE_8023AD) {
1564                 /* add lacpdu mc addr to mc list */
1565                 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
1566
1567                 dev_mc_add(slave_dev, lacpdu_multicast, ETH_ALEN, 0);
1568         }
1569
1570         bond_add_vlans_on_slave(bond, slave_dev);
1571
1572         write_lock_bh(&bond->lock);
1573
1574         bond_attach_slave(bond, new_slave);
1575
1576         new_slave->delay = 0;
1577         new_slave->link_failure_count = 0;
1578
1579         bond_compute_features(bond);
1580
1581         write_unlock_bh(&bond->lock);
1582
1583         read_lock(&bond->lock);
1584
1585         new_slave->last_arp_rx = jiffies;
1586
1587         if (bond->params.miimon && !bond->params.use_carrier) {
1588                 link_reporting = bond_check_dev_link(bond, slave_dev, 1);
1589
1590                 if ((link_reporting == -1) && !bond->params.arp_interval) {
1591                         /*
1592                          * miimon is set but a bonded network driver
1593                          * does not support ETHTOOL/MII and
1594                          * arp_interval is not set.  Note: if
1595                          * use_carrier is enabled, we will never go
1596                          * here (because netif_carrier is always
1597                          * supported); thus, we don't need to change
1598                          * the messages for netif_carrier.
1599                          */
1600                         printk(KERN_WARNING DRV_NAME
1601                                ": %s: Warning: MII and ETHTOOL support not "
1602                                "available for interface %s, and "
1603                                "arp_interval/arp_ip_target module parameters "
1604                                "not specified, thus bonding will not detect "
1605                                "link failures! see bonding.txt for details.\n",
1606                                bond_dev->name, slave_dev->name);
1607                 } else if (link_reporting == -1) {
1608                         /* unable get link status using mii/ethtool */
1609                         printk(KERN_WARNING DRV_NAME
1610                                ": %s: Warning: can't get link status from "
1611                                "interface %s; the network driver associated "
1612                                "with this interface does not support MII or "
1613                                "ETHTOOL link status reporting, thus miimon "
1614                                "has no effect on this interface.\n",
1615                                bond_dev->name, slave_dev->name);
1616                 }
1617         }
1618
1619         /* check for initial state */
1620         if (!bond->params.miimon ||
1621             (bond_check_dev_link(bond, slave_dev, 0) == BMSR_LSTATUS)) {
1622                 if (bond->params.updelay) {
1623                         dprintk("Initial state of slave_dev is "
1624                                 "BOND_LINK_BACK\n");
1625                         new_slave->link  = BOND_LINK_BACK;
1626                         new_slave->delay = bond->params.updelay;
1627                 } else {
1628                         dprintk("Initial state of slave_dev is "
1629                                 "BOND_LINK_UP\n");
1630                         new_slave->link  = BOND_LINK_UP;
1631                 }
1632                 new_slave->jiffies = jiffies;
1633         } else {
1634                 dprintk("Initial state of slave_dev is "
1635                         "BOND_LINK_DOWN\n");
1636                 new_slave->link  = BOND_LINK_DOWN;
1637         }
1638
1639         if (bond_update_speed_duplex(new_slave) &&
1640             (new_slave->link != BOND_LINK_DOWN)) {
1641                 printk(KERN_WARNING DRV_NAME
1642                        ": %s: Warning: failed to get speed and duplex from %s, "
1643                        "assumed to be 100Mb/sec and Full.\n",
1644                        bond_dev->name, new_slave->dev->name);
1645
1646                 if (bond->params.mode == BOND_MODE_8023AD) {
1647                         printk(KERN_WARNING DRV_NAME
1648                                ": %s: Warning: Operation of 802.3ad mode requires ETHTOOL "
1649                                "support in base driver for proper aggregator "
1650                                "selection.\n", bond_dev->name);
1651                 }
1652         }
1653
1654         if (USES_PRIMARY(bond->params.mode) && bond->params.primary[0]) {
1655                 /* if there is a primary slave, remember it */
1656                 if (strcmp(bond->params.primary, new_slave->dev->name) == 0) {
1657                         bond->primary_slave = new_slave;
1658                 }
1659         }
1660
1661         write_lock_bh(&bond->curr_slave_lock);
1662
1663         switch (bond->params.mode) {
1664         case BOND_MODE_ACTIVEBACKUP:
1665                 bond_set_slave_inactive_flags(new_slave);
1666                 bond_select_active_slave(bond);
1667                 break;
1668         case BOND_MODE_8023AD:
1669                 /* in 802.3ad mode, the internal mechanism
1670                  * will activate the slaves in the selected
1671                  * aggregator
1672                  */
1673                 bond_set_slave_inactive_flags(new_slave);
1674                 /* if this is the first slave */
1675                 if (bond->slave_cnt == 1) {
1676                         SLAVE_AD_INFO(new_slave).id = 1;
1677                         /* Initialize AD with the number of times that the AD timer is called in 1 second
1678                          * can be called only after the mac address of the bond is set
1679                          */
1680                         bond_3ad_initialize(bond, 1000/AD_TIMER_INTERVAL,
1681                                             bond->params.lacp_fast);
1682                 } else {
1683                         SLAVE_AD_INFO(new_slave).id =
1684                                 SLAVE_AD_INFO(new_slave->prev).id + 1;
1685                 }
1686
1687                 bond_3ad_bind_slave(new_slave);
1688                 break;
1689         case BOND_MODE_TLB:
1690         case BOND_MODE_ALB:
1691                 new_slave->state = BOND_STATE_ACTIVE;
1692                 bond_set_slave_inactive_flags(new_slave);
1693                 break;
1694         default:
1695                 dprintk("This slave is always active in trunk mode\n");
1696
1697                 /* always active in trunk mode */
1698                 new_slave->state = BOND_STATE_ACTIVE;
1699
1700                 /* In trunking mode there is little meaning to curr_active_slave
1701                  * anyway (it holds no special properties of the bond device),
1702                  * so we can change it without calling change_active_interface()
1703                  */
1704                 if (!bond->curr_active_slave) {
1705                         bond->curr_active_slave = new_slave;
1706                 }
1707                 break;
1708         } /* switch(bond_mode) */
1709
1710         write_unlock_bh(&bond->curr_slave_lock);
1711
1712         bond_set_carrier(bond);
1713
1714         read_unlock(&bond->lock);
1715
1716         res = bond_create_slave_symlinks(bond_dev, slave_dev);
1717         if (res)
1718                 goto err_close;
1719
1720         printk(KERN_INFO DRV_NAME
1721                ": %s: enslaving %s as a%s interface with a%s link.\n",
1722                bond_dev->name, slave_dev->name,
1723                new_slave->state == BOND_STATE_ACTIVE ? "n active" : " backup",
1724                new_slave->link != BOND_LINK_DOWN ? "n up" : " down");
1725
1726         /* enslave is successful */
1727         return 0;
1728
1729 /* Undo stages on error */
1730 err_close:
1731         dev_close(slave_dev);
1732
1733 err_unset_master:
1734         netdev_set_master(slave_dev, NULL);
1735
1736 err_restore_mac:
1737         if (!bond->params.fail_over_mac) {
1738                 /* XXX TODO - fom follow mode needs to change master's
1739                  * MAC if this slave's MAC is in use by the bond, or at
1740                  * least print a warning.
1741                  */
1742                 memcpy(addr.sa_data, new_slave->perm_hwaddr, ETH_ALEN);
1743                 addr.sa_family = slave_dev->type;
1744                 dev_set_mac_address(slave_dev, &addr);
1745         }
1746
1747 err_free:
1748         kfree(new_slave);
1749
1750 err_undo_flags:
1751         bond_dev->features = old_features;
1752  
1753         return res;
1754 }
1755
1756 /*
1757  * Try to release the slave device <slave> from the bond device <master>
1758  * It is legal to access curr_active_slave without a lock because all the function
1759  * is write-locked.
1760  *
1761  * The rules for slave state should be:
1762  *   for Active/Backup:
1763  *     Active stays on all backups go down
1764  *   for Bonded connections:
1765  *     The first up interface should be left on and all others downed.
1766  */
1767 int bond_release(struct net_device *bond_dev, struct net_device *slave_dev)
1768 {
1769         struct bonding *bond = bond_dev->priv;
1770         struct slave *slave, *oldcurrent;
1771         struct sockaddr addr;
1772         int mac_addr_differ;
1773         DECLARE_MAC_BUF(mac);
1774
1775         /* slave is not a slave or master is not master of this slave */
1776         if (!(slave_dev->flags & IFF_SLAVE) ||
1777             (slave_dev->master != bond_dev)) {
1778                 printk(KERN_ERR DRV_NAME
1779                        ": %s: Error: cannot release %s.\n",
1780                        bond_dev->name, slave_dev->name);
1781                 return -EINVAL;
1782         }
1783
1784         write_lock_bh(&bond->lock);
1785
1786         slave = bond_get_slave_by_dev(bond, slave_dev);
1787         if (!slave) {
1788                 /* not a slave of this bond */
1789                 printk(KERN_INFO DRV_NAME
1790                        ": %s: %s not enslaved\n",
1791                        bond_dev->name, slave_dev->name);
1792                 write_unlock_bh(&bond->lock);
1793                 return -EINVAL;
1794         }
1795
1796         if (!bond->params.fail_over_mac) {
1797                 mac_addr_differ = memcmp(bond_dev->dev_addr, slave->perm_hwaddr,
1798                                          ETH_ALEN);
1799                 if (!mac_addr_differ && (bond->slave_cnt > 1))
1800                         printk(KERN_WARNING DRV_NAME
1801                                ": %s: Warning: the permanent HWaddr of %s - "
1802                                "%s - is still in use by %s. "
1803                                "Set the HWaddr of %s to a different address "
1804                                "to avoid conflicts.\n",
1805                                bond_dev->name, slave_dev->name,
1806                                print_mac(mac, slave->perm_hwaddr),
1807                                bond_dev->name, slave_dev->name);
1808         }
1809
1810         /* Inform AD package of unbinding of slave. */
1811         if (bond->params.mode == BOND_MODE_8023AD) {
1812                 /* must be called before the slave is
1813                  * detached from the list
1814                  */
1815                 bond_3ad_unbind_slave(slave);
1816         }
1817
1818         printk(KERN_INFO DRV_NAME
1819                ": %s: releasing %s interface %s\n",
1820                bond_dev->name,
1821                (slave->state == BOND_STATE_ACTIVE)
1822                ? "active" : "backup",
1823                slave_dev->name);
1824
1825         oldcurrent = bond->curr_active_slave;
1826
1827         bond->current_arp_slave = NULL;
1828
1829         /* release the slave from its bond */
1830         bond_detach_slave(bond, slave);
1831
1832         bond_compute_features(bond);
1833
1834         if (bond->primary_slave == slave) {
1835                 bond->primary_slave = NULL;
1836         }
1837
1838         if (oldcurrent == slave) {
1839                 bond_change_active_slave(bond, NULL);
1840         }
1841
1842         if ((bond->params.mode == BOND_MODE_TLB) ||
1843             (bond->params.mode == BOND_MODE_ALB)) {
1844                 /* Must be called only after the slave has been
1845                  * detached from the list and the curr_active_slave
1846                  * has been cleared (if our_slave == old_current),
1847                  * but before a new active slave is selected.
1848                  */
1849                 write_unlock_bh(&bond->lock);
1850                 bond_alb_deinit_slave(bond, slave);
1851                 write_lock_bh(&bond->lock);
1852         }
1853
1854         if (oldcurrent == slave) {
1855                 /*
1856                  * Note that we hold RTNL over this sequence, so there
1857                  * is no concern that another slave add/remove event
1858                  * will interfere.
1859                  */
1860                 write_unlock_bh(&bond->lock);
1861                 read_lock(&bond->lock);
1862                 write_lock_bh(&bond->curr_slave_lock);
1863
1864                 bond_select_active_slave(bond);
1865
1866                 write_unlock_bh(&bond->curr_slave_lock);
1867                 read_unlock(&bond->lock);
1868                 write_lock_bh(&bond->lock);
1869         }
1870
1871         if (bond->slave_cnt == 0) {
1872                 bond_set_carrier(bond);
1873
1874                 /* if the last slave was removed, zero the mac address
1875                  * of the master so it will be set by the application
1876                  * to the mac address of the first slave
1877                  */
1878                 memset(bond_dev->dev_addr, 0, bond_dev->addr_len);
1879
1880                 if (list_empty(&bond->vlan_list)) {
1881                         bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
1882                 } else {
1883                         printk(KERN_WARNING DRV_NAME
1884                                ": %s: Warning: clearing HW address of %s while it "
1885                                "still has VLANs.\n",
1886                                bond_dev->name, bond_dev->name);
1887                         printk(KERN_WARNING DRV_NAME
1888                                ": %s: When re-adding slaves, make sure the bond's "
1889                                "HW address matches its VLANs'.\n",
1890                                bond_dev->name);
1891                 }
1892         } else if ((bond_dev->features & NETIF_F_VLAN_CHALLENGED) &&
1893                    !bond_has_challenged_slaves(bond)) {
1894                 printk(KERN_INFO DRV_NAME
1895                        ": %s: last VLAN challenged slave %s "
1896                        "left bond %s. VLAN blocking is removed\n",
1897                        bond_dev->name, slave_dev->name, bond_dev->name);
1898                 bond_dev->features &= ~NETIF_F_VLAN_CHALLENGED;
1899         }
1900
1901         write_unlock_bh(&bond->lock);
1902
1903         /* must do this from outside any spinlocks */
1904         bond_destroy_slave_symlinks(bond_dev, slave_dev);
1905
1906         bond_del_vlans_from_slave(bond, slave_dev);
1907
1908         /* If the mode USES_PRIMARY, then we should only remove its
1909          * promisc and mc settings if it was the curr_active_slave, but that was
1910          * already taken care of above when we detached the slave
1911          */
1912         if (!USES_PRIMARY(bond->params.mode)) {
1913                 /* unset promiscuity level from slave */
1914                 if (bond_dev->flags & IFF_PROMISC) {
1915                         dev_set_promiscuity(slave_dev, -1);
1916                 }
1917
1918                 /* unset allmulti level from slave */
1919                 if (bond_dev->flags & IFF_ALLMULTI) {
1920                         dev_set_allmulti(slave_dev, -1);
1921                 }
1922
1923                 /* flush master's mc_list from slave */
1924                 netif_tx_lock_bh(bond_dev);
1925                 bond_mc_list_flush(bond_dev, slave_dev);
1926                 netif_tx_unlock_bh(bond_dev);
1927         }
1928
1929         netdev_set_master(slave_dev, NULL);
1930
1931         /* close slave before restoring its mac address */
1932         dev_close(slave_dev);
1933
1934         if (bond->params.fail_over_mac != BOND_FOM_ACTIVE) {
1935                 /* restore original ("permanent") mac address */
1936                 memcpy(addr.sa_data, slave->perm_hwaddr, ETH_ALEN);
1937                 addr.sa_family = slave_dev->type;
1938                 dev_set_mac_address(slave_dev, &addr);
1939         }
1940
1941         slave_dev->priv_flags &= ~(IFF_MASTER_8023AD | IFF_MASTER_ALB |
1942                                    IFF_SLAVE_INACTIVE | IFF_BONDING |
1943                                    IFF_SLAVE_NEEDARP);
1944
1945         kfree(slave);
1946
1947         return 0;  /* deletion OK */
1948 }
1949
1950 /*
1951 * Destroy a bonding device.
1952 * Must be under rtnl_lock when this function is called.
1953 */
1954 void bond_destroy(struct bonding *bond)
1955 {
1956         bond_deinit(bond->dev);
1957         bond_destroy_sysfs_entry(bond);
1958         unregister_netdevice(bond->dev);
1959 }
1960
1961 /*
1962 * First release a slave and than destroy the bond if no more slaves iare left.
1963 * Must be under rtnl_lock when this function is called.
1964 */
1965 int  bond_release_and_destroy(struct net_device *bond_dev, struct net_device *slave_dev)
1966 {
1967         struct bonding *bond = bond_dev->priv;
1968         int ret;
1969
1970         ret = bond_release(bond_dev, slave_dev);
1971         if ((ret == 0) && (bond->slave_cnt == 0)) {
1972                 printk(KERN_INFO DRV_NAME ": %s: destroying bond %s.\n",
1973                        bond_dev->name, bond_dev->name);
1974                 bond_destroy(bond);
1975         }
1976         return ret;
1977 }
1978
1979 /*
1980  * This function releases all slaves.
1981  */
1982 static int bond_release_all(struct net_device *bond_dev)
1983 {
1984         struct bonding *bond = bond_dev->priv;
1985         struct slave *slave;
1986         struct net_device *slave_dev;
1987         struct sockaddr addr;
1988
1989         write_lock_bh(&bond->lock);
1990
1991         netif_carrier_off(bond_dev);
1992
1993         if (bond->slave_cnt == 0) {
1994                 goto out;
1995         }
1996
1997         bond->current_arp_slave = NULL;
1998         bond->primary_slave = NULL;
1999         bond_change_active_slave(bond, NULL);
2000
2001         while ((slave = bond->first_slave) != NULL) {
2002                 /* Inform AD package of unbinding of slave
2003                  * before slave is detached from the list.
2004                  */
2005                 if (bond->params.mode == BOND_MODE_8023AD) {
2006                         bond_3ad_unbind_slave(slave);
2007                 }
2008
2009                 slave_dev = slave->dev;
2010                 bond_detach_slave(bond, slave);
2011
2012                 /* now that the slave is detached, unlock and perform
2013                  * all the undo steps that should not be called from
2014                  * within a lock.
2015                  */
2016                 write_unlock_bh(&bond->lock);
2017
2018                 if ((bond->params.mode == BOND_MODE_TLB) ||
2019                     (bond->params.mode == BOND_MODE_ALB)) {
2020                         /* must be called only after the slave
2021                          * has been detached from the list
2022                          */
2023                         bond_alb_deinit_slave(bond, slave);
2024                 }
2025
2026                 bond_compute_features(bond);
2027
2028                 bond_destroy_slave_symlinks(bond_dev, slave_dev);
2029                 bond_del_vlans_from_slave(bond, slave_dev);
2030
2031                 /* If the mode USES_PRIMARY, then we should only remove its
2032                  * promisc and mc settings if it was the curr_active_slave, but that was
2033                  * already taken care of above when we detached the slave
2034                  */
2035                 if (!USES_PRIMARY(bond->params.mode)) {
2036                         /* unset promiscuity level from slave */
2037                         if (bond_dev->flags & IFF_PROMISC) {
2038                                 dev_set_promiscuity(slave_dev, -1);
2039                         }
2040
2041                         /* unset allmulti level from slave */
2042                         if (bond_dev->flags & IFF_ALLMULTI) {
2043                                 dev_set_allmulti(slave_dev, -1);
2044                         }
2045
2046                         /* flush master's mc_list from slave */
2047                         netif_tx_lock_bh(bond_dev);
2048                         bond_mc_list_flush(bond_dev, slave_dev);
2049                         netif_tx_unlock_bh(bond_dev);
2050                 }
2051
2052                 netdev_set_master(slave_dev, NULL);
2053
2054                 /* close slave before restoring its mac address */
2055                 dev_close(slave_dev);
2056
2057                 if (!bond->params.fail_over_mac) {
2058                         /* restore original ("permanent") mac address*/
2059                         memcpy(addr.sa_data, slave->perm_hwaddr, ETH_ALEN);
2060                         addr.sa_family = slave_dev->type;
2061                         dev_set_mac_address(slave_dev, &addr);
2062                 }
2063
2064                 slave_dev->priv_flags &= ~(IFF_MASTER_8023AD | IFF_MASTER_ALB |
2065                                            IFF_SLAVE_INACTIVE);
2066
2067                 kfree(slave);
2068
2069                 /* re-acquire the lock before getting the next slave */
2070                 write_lock_bh(&bond->lock);
2071         }
2072
2073         /* zero the mac address of the master so it will be
2074          * set by the application to the mac address of the
2075          * first slave
2076          */
2077         memset(bond_dev->dev_addr, 0, bond_dev->addr_len);
2078
2079         if (list_empty(&bond->vlan_list)) {
2080                 bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
2081         } else {
2082                 printk(KERN_WARNING DRV_NAME
2083                        ": %s: Warning: clearing HW address of %s while it "
2084                        "still has VLANs.\n",
2085                        bond_dev->name, bond_dev->name);
2086                 printk(KERN_WARNING DRV_NAME
2087                        ": %s: When re-adding slaves, make sure the bond's "
2088                        "HW address matches its VLANs'.\n",
2089                        bond_dev->name);
2090         }
2091
2092         printk(KERN_INFO DRV_NAME
2093                ": %s: released all slaves\n",
2094                bond_dev->name);
2095
2096 out:
2097         write_unlock_bh(&bond->lock);
2098
2099         return 0;
2100 }
2101
2102 /*
2103  * This function changes the active slave to slave <slave_dev>.
2104  * It returns -EINVAL in the following cases.
2105  *  - <slave_dev> is not found in the list.
2106  *  - There is not active slave now.
2107  *  - <slave_dev> is already active.
2108  *  - The link state of <slave_dev> is not BOND_LINK_UP.
2109  *  - <slave_dev> is not running.
2110  * In these cases, this fuction does nothing.
2111  * In the other cases, currnt_slave pointer is changed and 0 is returned.
2112  */
2113 static int bond_ioctl_change_active(struct net_device *bond_dev, struct net_device *slave_dev)
2114 {
2115         struct bonding *bond = bond_dev->priv;
2116         struct slave *old_active = NULL;
2117         struct slave *new_active = NULL;
2118         int res = 0;
2119
2120         if (!USES_PRIMARY(bond->params.mode)) {
2121                 return -EINVAL;
2122         }
2123
2124         /* Verify that master_dev is indeed the master of slave_dev */
2125         if (!(slave_dev->flags & IFF_SLAVE) ||
2126             (slave_dev->master != bond_dev)) {
2127                 return -EINVAL;
2128         }
2129
2130         read_lock(&bond->lock);
2131
2132         read_lock(&bond->curr_slave_lock);
2133         old_active = bond->curr_active_slave;
2134         read_unlock(&bond->curr_slave_lock);
2135
2136         new_active = bond_get_slave_by_dev(bond, slave_dev);
2137
2138         /*
2139          * Changing to the current active: do nothing; return success.
2140          */
2141         if (new_active && (new_active == old_active)) {
2142                 read_unlock(&bond->lock);
2143                 return 0;
2144         }
2145
2146         if ((new_active) &&
2147             (old_active) &&
2148             (new_active->link == BOND_LINK_UP) &&
2149             IS_UP(new_active->dev)) {
2150                 write_lock_bh(&bond->curr_slave_lock);
2151                 bond_change_active_slave(bond, new_active);
2152                 write_unlock_bh(&bond->curr_slave_lock);
2153         } else {
2154                 res = -EINVAL;
2155         }
2156
2157         read_unlock(&bond->lock);
2158
2159         return res;
2160 }
2161
2162 static int bond_info_query(struct net_device *bond_dev, struct ifbond *info)
2163 {
2164         struct bonding *bond = bond_dev->priv;
2165
2166         info->bond_mode = bond->params.mode;
2167         info->miimon = bond->params.miimon;
2168
2169         read_lock(&bond->lock);
2170         info->num_slaves = bond->slave_cnt;
2171         read_unlock(&bond->lock);
2172
2173         return 0;
2174 }
2175
2176 static int bond_slave_info_query(struct net_device *bond_dev, struct ifslave *info)
2177 {
2178         struct bonding *bond = bond_dev->priv;
2179         struct slave *slave;
2180         int i, found = 0;
2181
2182         if (info->slave_id < 0) {
2183                 return -ENODEV;
2184         }
2185
2186         read_lock(&bond->lock);
2187
2188         bond_for_each_slave(bond, slave, i) {
2189                 if (i == (int)info->slave_id) {
2190                         found = 1;
2191                         break;
2192                 }
2193         }
2194
2195         read_unlock(&bond->lock);
2196
2197         if (found) {
2198                 strcpy(info->slave_name, slave->dev->name);
2199                 info->link = slave->link;
2200                 info->state = slave->state;
2201                 info->link_failure_count = slave->link_failure_count;
2202         } else {
2203                 return -ENODEV;
2204         }
2205
2206         return 0;
2207 }
2208
2209 /*-------------------------------- Monitoring -------------------------------*/
2210
2211 /*
2212  * if !have_locks, return nonzero if a failover is necessary.  if
2213  * have_locks, do whatever failover activities are needed.
2214  *
2215  * This is to separate the inspection and failover steps for locking
2216  * purposes; failover requires rtnl, but acquiring it for every
2217  * inspection is undesirable, so a wrapper first does inspection, and
2218  * the acquires the necessary locks and calls again to perform
2219  * failover if needed.  Since all locks are dropped, a complete
2220  * restart is needed between calls.
2221  */
2222 static int __bond_mii_monitor(struct bonding *bond, int have_locks)
2223 {
2224         struct slave *slave, *oldcurrent;
2225         int do_failover = 0;
2226         int i;
2227
2228         if (bond->slave_cnt == 0)
2229                 goto out;
2230
2231         /* we will try to read the link status of each of our slaves, and
2232          * set their IFF_RUNNING flag appropriately. For each slave not
2233          * supporting MII status, we won't do anything so that a user-space
2234          * program could monitor the link itself if needed.
2235          */
2236
2237         read_lock(&bond->curr_slave_lock);
2238         oldcurrent = bond->curr_active_slave;
2239         read_unlock(&bond->curr_slave_lock);
2240
2241         bond_for_each_slave(bond, slave, i) {
2242                 struct net_device *slave_dev = slave->dev;
2243                 int link_state;
2244                 u16 old_speed = slave->speed;
2245                 u8 old_duplex = slave->duplex;
2246
2247                 link_state = bond_check_dev_link(bond, slave_dev, 0);
2248
2249                 switch (slave->link) {
2250                 case BOND_LINK_UP:      /* the link was up */
2251                         if (link_state == BMSR_LSTATUS) {
2252                                 if (!oldcurrent) {
2253                                         if (!have_locks)
2254                                                 return 1;
2255                                         do_failover = 1;
2256                                 }
2257                                 break;
2258                         } else { /* link going down */
2259                                 slave->link  = BOND_LINK_FAIL;
2260                                 slave->delay = bond->params.downdelay;
2261
2262                                 if (slave->link_failure_count < UINT_MAX) {
2263                                         slave->link_failure_count++;
2264                                 }
2265
2266                                 if (bond->params.downdelay) {
2267                                         printk(KERN_INFO DRV_NAME
2268                                                ": %s: link status down for %s "
2269                                                "interface %s, disabling it in "
2270                                                "%d ms.\n",
2271                                                bond->dev->name,
2272                                                IS_UP(slave_dev)
2273                                                ? ((bond->params.mode == BOND_MODE_ACTIVEBACKUP)
2274                                                   ? ((slave == oldcurrent)
2275                                                      ? "active " : "backup ")
2276                                                   : "")
2277                                                : "idle ",
2278                                                slave_dev->name,
2279                                                bond->params.downdelay * bond->params.miimon);
2280                                 }
2281                         }
2282                         /* no break ! fall through the BOND_LINK_FAIL test to
2283                            ensure proper action to be taken
2284                         */
2285                 case BOND_LINK_FAIL:    /* the link has just gone down */
2286                         if (link_state != BMSR_LSTATUS) {
2287                                 /* link stays down */
2288                                 if (slave->delay <= 0) {
2289                                         if (!have_locks)
2290                                                 return 1;
2291
2292                                         /* link down for too long time */
2293                                         slave->link = BOND_LINK_DOWN;
2294
2295                                         /* in active/backup mode, we must
2296                                          * completely disable this interface
2297                                          */
2298                                         if ((bond->params.mode == BOND_MODE_ACTIVEBACKUP) ||
2299                                             (bond->params.mode == BOND_MODE_8023AD)) {
2300                                                 bond_set_slave_inactive_flags(slave);
2301                                         }
2302
2303                                         printk(KERN_INFO DRV_NAME
2304                                                ": %s: link status definitely "
2305                                                "down for interface %s, "
2306                                                "disabling it\n",
2307                                                bond->dev->name,
2308                                                slave_dev->name);
2309
2310                                         /* notify ad that the link status has changed */
2311                                         if (bond->params.mode == BOND_MODE_8023AD) {
2312                                                 bond_3ad_handle_link_change(slave, BOND_LINK_DOWN);
2313                                         }
2314
2315                                         if ((bond->params.mode == BOND_MODE_TLB) ||
2316                                             (bond->params.mode == BOND_MODE_ALB)) {
2317                                                 bond_alb_handle_link_change(bond, slave, BOND_LINK_DOWN);
2318                                         }
2319
2320                                         if (slave == oldcurrent) {
2321                                                 do_failover = 1;
2322                                         }
2323                                 } else {
2324                                         slave->delay--;
2325                                 }
2326                         } else {
2327                                 /* link up again */
2328                                 slave->link  = BOND_LINK_UP;
2329                                 slave->jiffies = jiffies;
2330                                 printk(KERN_INFO DRV_NAME
2331                                        ": %s: link status up again after %d "
2332                                        "ms for interface %s.\n",
2333                                        bond->dev->name,
2334                                        (bond->params.downdelay - slave->delay) * bond->params.miimon,
2335                                        slave_dev->name);
2336                         }
2337                         break;
2338                 case BOND_LINK_DOWN:    /* the link was down */
2339                         if (link_state != BMSR_LSTATUS) {
2340                                 /* the link stays down, nothing more to do */
2341                                 break;
2342                         } else {        /* link going up */
2343                                 slave->link  = BOND_LINK_BACK;
2344                                 slave->delay = bond->params.updelay;
2345
2346                                 if (bond->params.updelay) {
2347                                         /* if updelay == 0, no need to
2348                                            advertise about a 0 ms delay */
2349                                         printk(KERN_INFO DRV_NAME
2350                                                ": %s: link status up for "
2351                                                "interface %s, enabling it "
2352                                                "in %d ms.\n",
2353                                                bond->dev->name,
2354                                                slave_dev->name,
2355                                                bond->params.updelay * bond->params.miimon);
2356                                 }
2357                         }
2358                         /* no break ! fall through the BOND_LINK_BACK state in
2359                            case there's something to do.
2360                         */
2361                 case BOND_LINK_BACK:    /* the link has just come back */
2362                         if (link_state != BMSR_LSTATUS) {
2363                                 /* link down again */
2364                                 slave->link  = BOND_LINK_DOWN;
2365
2366                                 printk(KERN_INFO DRV_NAME
2367                                        ": %s: link status down again after %d "
2368                                        "ms for interface %s.\n",
2369                                        bond->dev->name,
2370                                        (bond->params.updelay - slave->delay) * bond->params.miimon,
2371                                        slave_dev->name);
2372                         } else {
2373                                 /* link stays up */
2374                                 if (slave->delay == 0) {
2375                                         if (!have_locks)
2376                                                 return 1;
2377
2378                                         /* now the link has been up for long time enough */
2379                                         slave->link = BOND_LINK_UP;
2380                                         slave->jiffies = jiffies;
2381
2382                                         if (bond->params.mode == BOND_MODE_8023AD) {
2383                                                 /* prevent it from being the active one */
2384                                                 slave->state = BOND_STATE_BACKUP;
2385                                         } else if (bond->params.mode != BOND_MODE_ACTIVEBACKUP) {
2386                                                 /* make it immediately active */
2387                                                 slave->state = BOND_STATE_ACTIVE;
2388                                         } else if (slave != bond->primary_slave) {
2389                                                 /* prevent it from being the active one */
2390                                                 slave->state = BOND_STATE_BACKUP;
2391                                         }
2392
2393                                         printk(KERN_INFO DRV_NAME
2394                                                ": %s: link status definitely "
2395                                                "up for interface %s.\n",
2396                                                bond->dev->name,
2397                                                slave_dev->name);
2398
2399                                         /* notify ad that the link status has changed */
2400                                         if (bond->params.mode == BOND_MODE_8023AD) {
2401                                                 bond_3ad_handle_link_change(slave, BOND_LINK_UP);
2402                                         }
2403
2404                                         if ((bond->params.mode == BOND_MODE_TLB) ||
2405                                             (bond->params.mode == BOND_MODE_ALB)) {
2406                                                 bond_alb_handle_link_change(bond, slave, BOND_LINK_UP);
2407                                         }
2408
2409                                         if ((!oldcurrent) ||
2410                                             (slave == bond->primary_slave)) {
2411                                                 do_failover = 1;
2412                                         }
2413                                 } else {
2414                                         slave->delay--;
2415                                 }
2416                         }
2417                         break;
2418                 default:
2419                         /* Should not happen */
2420                         printk(KERN_ERR DRV_NAME
2421                                ": %s: Error: %s Illegal value (link=%d)\n",
2422                                bond->dev->name,
2423                                slave->dev->name,
2424                                slave->link);
2425                         goto out;
2426                 } /* end of switch (slave->link) */
2427
2428                 bond_update_speed_duplex(slave);
2429
2430                 if (bond->params.mode == BOND_MODE_8023AD) {
2431                         if (old_speed != slave->speed) {
2432                                 bond_3ad_adapter_speed_changed(slave);
2433                         }
2434
2435                         if (old_duplex != slave->duplex) {
2436                                 bond_3ad_adapter_duplex_changed(slave);
2437                         }
2438                 }
2439
2440         } /* end of for */
2441
2442         if (do_failover) {
2443                 ASSERT_RTNL();
2444
2445                 write_lock_bh(&bond->curr_slave_lock);
2446
2447                 bond_select_active_slave(bond);
2448
2449                 write_unlock_bh(&bond->curr_slave_lock);
2450
2451         } else
2452                 bond_set_carrier(bond);
2453
2454 out:
2455         return 0;
2456 }
2457
2458 /*
2459  * bond_mii_monitor
2460  *
2461  * Really a wrapper that splits the mii monitor into two phases: an
2462  * inspection, then (if inspection indicates something needs to be
2463  * done) an acquisition of appropriate locks followed by another pass
2464  * to implement whatever link state changes are indicated.
2465  */
2466 void bond_mii_monitor(struct work_struct *work)
2467 {
2468         struct bonding *bond = container_of(work, struct bonding,
2469                                             mii_work.work);
2470         unsigned long delay;
2471
2472         read_lock(&bond->lock);
2473         if (bond->kill_timers) {
2474                 read_unlock(&bond->lock);
2475                 return;
2476         }
2477
2478         if (bond->send_grat_arp) {
2479                 read_lock(&bond->curr_slave_lock);
2480                 bond_send_gratuitous_arp(bond);
2481                 read_unlock(&bond->curr_slave_lock);
2482         }
2483
2484         if (__bond_mii_monitor(bond, 0)) {
2485                 read_unlock(&bond->lock);
2486                 rtnl_lock();
2487                 read_lock(&bond->lock);
2488                 __bond_mii_monitor(bond, 1);
2489                 read_unlock(&bond->lock);
2490                 rtnl_unlock();  /* might sleep, hold no other locks */
2491                 read_lock(&bond->lock);
2492         }
2493
2494         delay = msecs_to_jiffies(bond->params.miimon);
2495         read_unlock(&bond->lock);
2496         queue_delayed_work(bond->wq, &bond->mii_work, delay);
2497 }
2498
2499 static __be32 bond_glean_dev_ip(struct net_device *dev)
2500 {
2501         struct in_device *idev;
2502         struct in_ifaddr *ifa;
2503         __be32 addr = 0;
2504
2505         if (!dev)
2506                 return 0;
2507
2508         rcu_read_lock();
2509         idev = __in_dev_get_rcu(dev);
2510         if (!idev)
2511                 goto out;
2512
2513         ifa = idev->ifa_list;
2514         if (!ifa)
2515                 goto out;
2516
2517         addr = ifa->ifa_local;
2518 out:
2519         rcu_read_unlock();
2520         return addr;
2521 }
2522
2523 static int bond_has_this_ip(struct bonding *bond, __be32 ip)
2524 {
2525         struct vlan_entry *vlan;
2526
2527         if (ip == bond->master_ip)
2528                 return 1;
2529
2530         list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
2531                 if (ip == vlan->vlan_ip)
2532                         return 1;
2533         }
2534
2535         return 0;
2536 }
2537
2538 /*
2539  * We go to the (large) trouble of VLAN tagging ARP frames because
2540  * switches in VLAN mode (especially if ports are configured as
2541  * "native" to a VLAN) might not pass non-tagged frames.
2542  */
2543 static void bond_arp_send(struct net_device *slave_dev, int arp_op, __be32 dest_ip, __be32 src_ip, unsigned short vlan_id)
2544 {
2545         struct sk_buff *skb;
2546
2547         dprintk("arp %d on slave %s: dst %x src %x vid %d\n", arp_op,
2548                slave_dev->name, dest_ip, src_ip, vlan_id);
2549                
2550         skb = arp_create(arp_op, ETH_P_ARP, dest_ip, slave_dev, src_ip,
2551                          NULL, slave_dev->dev_addr, NULL);
2552
2553         if (!skb) {
2554                 printk(KERN_ERR DRV_NAME ": ARP packet allocation failed\n");
2555                 return;
2556         }
2557         if (vlan_id) {
2558                 skb = vlan_put_tag(skb, vlan_id);
2559                 if (!skb) {
2560                         printk(KERN_ERR DRV_NAME ": failed to insert VLAN tag\n");
2561                         return;
2562                 }
2563         }
2564         arp_xmit(skb);
2565 }
2566
2567
2568 static void bond_arp_send_all(struct bonding *bond, struct slave *slave)
2569 {
2570         int i, vlan_id, rv;
2571         __be32 *targets = bond->params.arp_targets;
2572         struct vlan_entry *vlan;
2573         struct net_device *vlan_dev;
2574         struct flowi fl;
2575         struct rtable *rt;
2576
2577         for (i = 0; (i < BOND_MAX_ARP_TARGETS); i++) {
2578                 if (!targets[i])
2579                         continue;
2580                 dprintk("basa: target %x\n", targets[i]);
2581                 if (list_empty(&bond->vlan_list)) {
2582                         dprintk("basa: empty vlan: arp_send\n");
2583                         bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2584                                       bond->master_ip, 0);
2585                         continue;
2586                 }
2587
2588                 /*
2589                  * If VLANs are configured, we do a route lookup to
2590                  * determine which VLAN interface would be used, so we
2591                  * can tag the ARP with the proper VLAN tag.
2592                  */
2593                 memset(&fl, 0, sizeof(fl));
2594                 fl.fl4_dst = targets[i];
2595                 fl.fl4_tos = RTO_ONLINK;
2596
2597                 rv = ip_route_output_key(&init_net, &rt, &fl);
2598                 if (rv) {
2599                         if (net_ratelimit()) {
2600                                 printk(KERN_WARNING DRV_NAME
2601                              ": %s: no route to arp_ip_target %u.%u.%u.%u\n",
2602                                        bond->dev->name, NIPQUAD(fl.fl4_dst));
2603                         }
2604                         continue;
2605                 }
2606
2607                 /*
2608                  * This target is not on a VLAN
2609                  */
2610                 if (rt->u.dst.dev == bond->dev) {
2611                         ip_rt_put(rt);
2612                         dprintk("basa: rtdev == bond->dev: arp_send\n");
2613                         bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2614                                       bond->master_ip, 0);
2615                         continue;
2616                 }
2617
2618                 vlan_id = 0;
2619                 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
2620                         vlan_dev = vlan_group_get_device(bond->vlgrp, vlan->vlan_id);
2621                         if (vlan_dev == rt->u.dst.dev) {
2622                                 vlan_id = vlan->vlan_id;
2623                                 dprintk("basa: vlan match on %s %d\n",
2624                                        vlan_dev->name, vlan_id);
2625                                 break;
2626                         }
2627                 }
2628
2629                 if (vlan_id) {
2630                         ip_rt_put(rt);
2631                         bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2632                                       vlan->vlan_ip, vlan_id);
2633                         continue;
2634                 }
2635
2636                 if (net_ratelimit()) {
2637                         printk(KERN_WARNING DRV_NAME
2638                ": %s: no path to arp_ip_target %u.%u.%u.%u via rt.dev %s\n",
2639                                bond->dev->name, NIPQUAD(fl.fl4_dst),
2640                                rt->u.dst.dev ? rt->u.dst.dev->name : "NULL");
2641                 }
2642                 ip_rt_put(rt);
2643         }
2644 }
2645
2646 /*
2647  * Kick out a gratuitous ARP for an IP on the bonding master plus one
2648  * for each VLAN above us.
2649  *
2650  * Caller must hold curr_slave_lock for read or better
2651  */
2652 static void bond_send_gratuitous_arp(struct bonding *bond)
2653 {
2654         struct slave *slave = bond->curr_active_slave;
2655         struct vlan_entry *vlan;
2656         struct net_device *vlan_dev;
2657
2658         dprintk("bond_send_grat_arp: bond %s slave %s\n", bond->dev->name,
2659                                 slave ? slave->dev->name : "NULL");
2660
2661         if (!slave || !bond->send_grat_arp ||
2662             test_bit(__LINK_STATE_LINKWATCH_PENDING, &slave->dev->state))
2663                 return;
2664
2665         bond->send_grat_arp--;
2666
2667         if (bond->master_ip) {
2668                 bond_arp_send(slave->dev, ARPOP_REPLY, bond->master_ip,
2669                                 bond->master_ip, 0);
2670         }
2671
2672         list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
2673                 vlan_dev = vlan_group_get_device(bond->vlgrp, vlan->vlan_id);
2674                 if (vlan->vlan_ip) {
2675                         bond_arp_send(slave->dev, ARPOP_REPLY, vlan->vlan_ip,
2676                                       vlan->vlan_ip, vlan->vlan_id);
2677                 }
2678         }
2679 }
2680
2681 static void bond_validate_arp(struct bonding *bond, struct slave *slave, __be32 sip, __be32 tip)
2682 {
2683         int i;
2684         __be32 *targets = bond->params.arp_targets;
2685
2686         targets = bond->params.arp_targets;
2687         for (i = 0; (i < BOND_MAX_ARP_TARGETS) && targets[i]; i++) {
2688                 dprintk("bva: sip %u.%u.%u.%u tip %u.%u.%u.%u t[%d] "
2689                         "%u.%u.%u.%u bhti(tip) %d\n",
2690                        NIPQUAD(sip), NIPQUAD(tip), i, NIPQUAD(targets[i]),
2691                        bond_has_this_ip(bond, tip));
2692                 if (sip == targets[i]) {
2693                         if (bond_has_this_ip(bond, tip))
2694                                 slave->last_arp_rx = jiffies;
2695                         return;
2696                 }
2697         }
2698 }
2699
2700 static int bond_arp_rcv(struct sk_buff *skb, struct net_device *dev, struct packet_type *pt, struct net_device *orig_dev)
2701 {
2702         struct arphdr *arp;
2703         struct slave *slave;
2704         struct bonding *bond;
2705         unsigned char *arp_ptr;
2706         __be32 sip, tip;
2707
2708         if (dev_net(dev) != &init_net)
2709                 goto out;
2710
2711         if (!(dev->priv_flags & IFF_BONDING) || !(dev->flags & IFF_MASTER))
2712                 goto out;
2713
2714         bond = dev->priv;
2715         read_lock(&bond->lock);
2716
2717         dprintk("bond_arp_rcv: bond %s skb->dev %s orig_dev %s\n",
2718                 bond->dev->name, skb->dev ? skb->dev->name : "NULL",
2719                 orig_dev ? orig_dev->name : "NULL");
2720
2721         slave = bond_get_slave_by_dev(bond, orig_dev);
2722         if (!slave || !slave_do_arp_validate(bond, slave))
2723                 goto out_unlock;
2724
2725         if (!pskb_may_pull(skb, arp_hdr_len(dev)))
2726                 goto out_unlock;
2727
2728         arp = arp_hdr(skb);
2729         if (arp->ar_hln != dev->addr_len ||
2730             skb->pkt_type == PACKET_OTHERHOST ||
2731             skb->pkt_type == PACKET_LOOPBACK ||
2732             arp->ar_hrd != htons(ARPHRD_ETHER) ||
2733             arp->ar_pro != htons(ETH_P_IP) ||
2734             arp->ar_pln != 4)
2735                 goto out_unlock;
2736
2737         arp_ptr = (unsigned char *)(arp + 1);
2738         arp_ptr += dev->addr_len;
2739         memcpy(&sip, arp_ptr, 4);
2740         arp_ptr += 4 + dev->addr_len;
2741         memcpy(&tip, arp_ptr, 4);
2742
2743         dprintk("bond_arp_rcv: %s %s/%d av %d sv %d sip %u.%u.%u.%u"
2744                 " tip %u.%u.%u.%u\n", bond->dev->name, slave->dev->name,
2745                 slave->state, bond->params.arp_validate,
2746                 slave_do_arp_validate(bond, slave), NIPQUAD(sip), NIPQUAD(tip));
2747
2748         /*
2749          * Backup slaves won't see the ARP reply, but do come through
2750          * here for each ARP probe (so we swap the sip/tip to validate
2751          * the probe).  In a "redundant switch, common router" type of
2752          * configuration, the ARP probe will (hopefully) travel from
2753          * the active, through one switch, the router, then the other
2754          * switch before reaching the backup.
2755          */
2756         if (slave->state == BOND_STATE_ACTIVE)
2757                 bond_validate_arp(bond, slave, sip, tip);
2758         else
2759                 bond_validate_arp(bond, slave, tip, sip);
2760
2761 out_unlock:
2762         read_unlock(&bond->lock);
2763 out:
2764         dev_kfree_skb(skb);
2765         return NET_RX_SUCCESS;
2766 }
2767
2768 /*
2769  * this function is called regularly to monitor each slave's link
2770  * ensuring that traffic is being sent and received when arp monitoring
2771  * is used in load-balancing mode. if the adapter has been dormant, then an
2772  * arp is transmitted to generate traffic. see activebackup_arp_monitor for
2773  * arp monitoring in active backup mode.
2774  */
2775 void bond_loadbalance_arp_mon(struct work_struct *work)
2776 {
2777         struct bonding *bond = container_of(work, struct bonding,
2778                                             arp_work.work);
2779         struct slave *slave, *oldcurrent;
2780         int do_failover = 0;
2781         int delta_in_ticks;
2782         int i;
2783
2784         read_lock(&bond->lock);
2785
2786         delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
2787
2788         if (bond->kill_timers) {
2789                 goto out;
2790         }
2791
2792         if (bond->slave_cnt == 0) {
2793                 goto re_arm;
2794         }
2795
2796         read_lock(&bond->curr_slave_lock);
2797         oldcurrent = bond->curr_active_slave;
2798         read_unlock(&bond->curr_slave_lock);
2799
2800         /* see if any of the previous devices are up now (i.e. they have
2801          * xmt and rcv traffic). the curr_active_slave does not come into
2802          * the picture unless it is null. also, slave->jiffies is not needed
2803          * here because we send an arp on each slave and give a slave as
2804          * long as it needs to get the tx/rx within the delta.
2805          * TODO: what about up/down delay in arp mode? it wasn't here before
2806          *       so it can wait
2807          */
2808         bond_for_each_slave(bond, slave, i) {
2809                 if (slave->link != BOND_LINK_UP) {
2810                         if (time_before_eq(jiffies, slave->dev->trans_start + delta_in_ticks) &&
2811                             time_before_eq(jiffies, slave->dev->last_rx + delta_in_ticks)) {
2812
2813                                 slave->link  = BOND_LINK_UP;
2814                                 slave->state = BOND_STATE_ACTIVE;
2815
2816                                 /* primary_slave has no meaning in round-robin
2817                                  * mode. the window of a slave being up and
2818                                  * curr_active_slave being null after enslaving
2819                                  * is closed.
2820                                  */
2821                                 if (!oldcurrent) {
2822                                         printk(KERN_INFO DRV_NAME
2823                                                ": %s: link status definitely "
2824                                                "up for interface %s, ",
2825                                                bond->dev->name,
2826                                                slave->dev->name);
2827                                         do_failover = 1;
2828                                 } else {
2829                                         printk(KERN_INFO DRV_NAME
2830                                                ": %s: interface %s is now up\n",
2831                                                bond->dev->name,
2832                                                slave->dev->name);
2833                                 }
2834                         }
2835                 } else {
2836                         /* slave->link == BOND_LINK_UP */
2837
2838                         /* not all switches will respond to an arp request
2839                          * when the source ip is 0, so don't take the link down
2840                          * if we don't know our ip yet
2841                          */
2842                         if (time_after_eq(jiffies, slave->dev->trans_start + 2*delta_in_ticks) ||
2843                             (time_after_eq(jiffies, slave->dev->last_rx + 2*delta_in_ticks))) {
2844
2845                                 slave->link  = BOND_LINK_DOWN;
2846                                 slave->state = BOND_STATE_BACKUP;
2847
2848                                 if (slave->link_failure_count < UINT_MAX) {
2849                                         slave->link_failure_count++;
2850                                 }
2851
2852                                 printk(KERN_INFO DRV_NAME
2853                                        ": %s: interface %s is now down.\n",
2854                                        bond->dev->name,
2855                                        slave->dev->name);
2856
2857                                 if (slave == oldcurrent) {
2858                                         do_failover = 1;
2859                                 }
2860                         }
2861                 }
2862
2863                 /* note: if switch is in round-robin mode, all links
2864                  * must tx arp to ensure all links rx an arp - otherwise
2865                  * links may oscillate or not come up at all; if switch is
2866                  * in something like xor mode, there is nothing we can
2867                  * do - all replies will be rx'ed on same link causing slaves
2868                  * to be unstable during low/no traffic periods
2869                  */
2870                 if (IS_UP(slave->dev)) {
2871                         bond_arp_send_all(bond, slave);
2872                 }
2873         }
2874
2875         if (do_failover) {
2876                 write_lock_bh(&bond->curr_slave_lock);
2877
2878                 bond_select_active_slave(bond);
2879
2880                 write_unlock_bh(&bond->curr_slave_lock);
2881         }
2882
2883 re_arm:
2884         if (bond->params.arp_interval)
2885                 queue_delayed_work(bond->wq, &bond->arp_work, delta_in_ticks);
2886 out:
2887         read_unlock(&bond->lock);
2888 }
2889
2890 /*
2891  * Called to inspect slaves for active-backup mode ARP monitor link state
2892  * changes.  Sets new_link in slaves to specify what action should take
2893  * place for the slave.  Returns 0 if no changes are found, >0 if changes
2894  * to link states must be committed.
2895  *
2896  * Called with bond->lock held for read.
2897  */
2898 static int bond_ab_arp_inspect(struct bonding *bond, int delta_in_ticks)
2899 {
2900         struct slave *slave;
2901         int i, commit = 0;
2902
2903         bond_for_each_slave(bond, slave, i) {
2904                 slave->new_link = BOND_LINK_NOCHANGE;
2905
2906                 if (slave->link != BOND_LINK_UP) {
2907                         if (time_before_eq(jiffies, slave_last_rx(bond, slave) +
2908                                            delta_in_ticks)) {
2909                                 slave->new_link = BOND_LINK_UP;
2910                                 commit++;
2911                         }
2912
2913                         continue;
2914                 }
2915
2916                 /*
2917                  * Give slaves 2*delta after being enslaved or made
2918                  * active.  This avoids bouncing, as the last receive
2919                  * times need a full ARP monitor cycle to be updated.
2920                  */
2921                 if (!time_after_eq(jiffies, slave->jiffies +
2922                                    2 * delta_in_ticks))
2923                         continue;
2924
2925                 /*
2926                  * Backup slave is down if:
2927                  * - No current_arp_slave AND
2928                  * - more than 3*delta since last receive AND
2929                  * - the bond has an IP address
2930                  *
2931                  * Note: a non-null current_arp_slave indicates
2932                  * the curr_active_slave went down and we are
2933                  * searching for a new one; under this condition
2934                  * we only take the curr_active_slave down - this
2935                  * gives each slave a chance to tx/rx traffic
2936                  * before being taken out
2937                  */
2938                 if (slave->state == BOND_STATE_BACKUP &&
2939                     !bond->current_arp_slave &&
2940                     time_after(jiffies, slave_last_rx(bond, slave) +
2941                                3 * delta_in_ticks)) {
2942                         slave->new_link = BOND_LINK_DOWN;
2943                         commit++;
2944                 }
2945
2946                 /*
2947                  * Active slave is down if:
2948                  * - more than 2*delta since transmitting OR
2949                  * - (more than 2*delta since receive AND
2950                  *    the bond has an IP address)
2951                  */
2952                 if ((slave->state == BOND_STATE_ACTIVE) &&
2953                     (time_after_eq(jiffies, slave->dev->trans_start +
2954                                     2 * delta_in_ticks) ||
2955                       (time_after_eq(jiffies, slave_last_rx(bond, slave)
2956                                      + 2 * delta_in_ticks)))) {
2957                         slave->new_link = BOND_LINK_DOWN;
2958                         commit++;
2959                 }
2960         }
2961
2962         read_lock(&bond->curr_slave_lock);
2963
2964         /*
2965          * Trigger a commit if the primary option setting has changed.
2966          */
2967         if (bond->primary_slave &&
2968             (bond->primary_slave != bond->curr_active_slave) &&
2969             (bond->primary_slave->link == BOND_LINK_UP))
2970                 commit++;
2971
2972         read_unlock(&bond->curr_slave_lock);
2973
2974         return commit;
2975 }
2976
2977 /*
2978  * Called to commit link state changes noted by inspection step of
2979  * active-backup mode ARP monitor.
2980  *
2981  * Called with RTNL and bond->lock for read.
2982  */
2983 static void bond_ab_arp_commit(struct bonding *bond, int delta_in_ticks)
2984 {
2985         struct slave *slave;
2986         int i;
2987
2988         bond_for_each_slave(bond, slave, i) {
2989                 switch (slave->new_link) {
2990                 case BOND_LINK_NOCHANGE:
2991                         continue;
2992
2993                 case BOND_LINK_UP:
2994                         write_lock_bh(&bond->curr_slave_lock);
2995
2996                         if (!bond->curr_active_slave &&
2997                             time_before_eq(jiffies, slave->dev->trans_start +
2998                                            delta_in_ticks)) {
2999                                 slave->link = BOND_LINK_UP;
3000                                 bond_change_active_slave(bond, slave);
3001                                 bond->current_arp_slave = NULL;
3002
3003                                 printk(KERN_INFO DRV_NAME
3004                                        ": %s: %s is up and now the "
3005                                        "active interface\n",
3006                                        bond->dev->name, slave->dev->name);
3007
3008                         } else if (bond->curr_active_slave != slave) {
3009                                 /* this slave has just come up but we
3010                                  * already have a current slave; this can
3011                                  * also happen if bond_enslave adds a new
3012                                  * slave that is up while we are searching
3013                                  * for a new slave
3014                                  */
3015                                 slave->link = BOND_LINK_UP;
3016                                 bond_set_slave_inactive_flags(slave);
3017                                 bond->current_arp_slave = NULL;
3018
3019                                 printk(KERN_INFO DRV_NAME
3020                                        ": %s: backup interface %s is now up\n",
3021                                        bond->dev->name, slave->dev->name);
3022                         }
3023
3024                         write_unlock_bh(&bond->curr_slave_lock);
3025
3026                         break;
3027
3028                 case BOND_LINK_DOWN:
3029                         if (slave->link_failure_count < UINT_MAX)
3030                                 slave->link_failure_count++;
3031
3032                         slave->link = BOND_LINK_DOWN;
3033
3034                         if (slave == bond->curr_active_slave) {
3035                                 printk(KERN_INFO DRV_NAME
3036                                        ": %s: link status down for active "
3037                                        "interface %s, disabling it\n",
3038                                        bond->dev->name, slave->dev->name);
3039
3040                                 bond_set_slave_inactive_flags(slave);
3041
3042                                 write_lock_bh(&bond->curr_slave_lock);
3043
3044                                 bond_select_active_slave(bond);
3045                                 if (bond->curr_active_slave)
3046                                         bond->curr_active_slave->jiffies =
3047                                                 jiffies;
3048
3049                                 write_unlock_bh(&bond->curr_slave_lock);
3050
3051                                 bond->current_arp_slave = NULL;
3052
3053                         } else if (slave->state == BOND_STATE_BACKUP) {
3054                                 printk(KERN_INFO DRV_NAME
3055                                        ": %s: backup interface %s is now down\n",
3056                                        bond->dev->name, slave->dev->name);
3057
3058                                 bond_set_slave_inactive_flags(slave);
3059                         }
3060                         break;
3061
3062                 default:
3063                         printk(KERN_ERR DRV_NAME
3064                                ": %s: impossible: new_link %d on slave %s\n",
3065                                bond->dev->name, slave->new_link,
3066                                slave->dev->name);
3067                 }
3068         }
3069
3070         /*
3071          * No race with changes to primary via sysfs, as we hold rtnl.
3072          */
3073         if (bond->primary_slave &&
3074             (bond->primary_slave != bond->curr_active_slave) &&
3075             (bond->primary_slave->link == BOND_LINK_UP)) {
3076                 write_lock_bh(&bond->curr_slave_lock);
3077                 bond_change_active_slave(bond, bond->primary_slave);
3078                 write_unlock_bh(&bond->curr_slave_lock);
3079         }
3080
3081         bond_set_carrier(bond);
3082 }
3083
3084 /*
3085  * Send ARP probes for active-backup mode ARP monitor.
3086  *
3087  * Called with bond->lock held for read.
3088  */
3089 static void bond_ab_arp_probe(struct bonding *bond)
3090 {
3091         struct slave *slave;
3092         int i;
3093
3094         read_lock(&bond->curr_slave_lock);
3095
3096         if (bond->current_arp_slave && bond->curr_active_slave)
3097                 printk("PROBE: c_arp %s && cas %s BAD\n",
3098                        bond->current_arp_slave->dev->name,
3099                        bond->curr_active_slave->dev->name);
3100
3101         if (bond->curr_active_slave) {
3102                 bond_arp_send_all(bond, bond->curr_active_slave);
3103                 read_unlock(&bond->curr_slave_lock);
3104                 return;
3105         }
3106
3107         read_unlock(&bond->curr_slave_lock);
3108
3109         /* if we don't have a curr_active_slave, search for the next available
3110          * backup slave from the current_arp_slave and make it the candidate
3111          * for becoming the curr_active_slave
3112          */
3113
3114         if (!bond->current_arp_slave) {
3115                 bond->current_arp_slave = bond->first_slave;
3116                 if (!bond->current_arp_slave)
3117                         return;
3118         }
3119
3120         bond_set_slave_inactive_flags(bond->current_arp_slave);
3121
3122         /* search for next candidate */
3123         bond_for_each_slave_from(bond, slave, i, bond->current_arp_slave->next) {
3124                 if (IS_UP(slave->dev)) {
3125                         slave->link = BOND_LINK_BACK;
3126                         bond_set_slave_active_flags(slave);
3127                         bond_arp_send_all(bond, slave);
3128                         slave->jiffies = jiffies;
3129                         bond->current_arp_slave = slave;
3130                         break;
3131                 }
3132
3133                 /* if the link state is up at this point, we
3134                  * mark it down - this can happen if we have
3135                  * simultaneous link failures and
3136                  * reselect_active_interface doesn't make this
3137                  * one the current slave so it is still marked
3138                  * up when it is actually down
3139                  */
3140                 if (slave->link == BOND_LINK_UP) {
3141                         slave->link = BOND_LINK_DOWN;
3142                         if (slave->link_failure_count < UINT_MAX)
3143                                 slave->link_failure_count++;
3144
3145                         bond_set_slave_inactive_flags(slave);
3146
3147                         printk(KERN_INFO DRV_NAME
3148                                ": %s: backup interface %s is now down.\n",
3149                                bond->dev->name, slave->dev->name);
3150                 }
3151         }
3152 }
3153
3154 void bond_activebackup_arp_mon(struct work_struct *work)
3155 {
3156         struct bonding *bond = container_of(work, struct bonding,
3157                                             arp_work.work);
3158         int delta_in_ticks;
3159
3160         read_lock(&bond->lock);
3161
3162         if (bond->kill_timers)
3163                 goto out;
3164
3165         delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
3166
3167         if (bond->slave_cnt == 0)
3168                 goto re_arm;
3169
3170         if (bond->send_grat_arp) {
3171                 read_lock(&bond->curr_slave_lock);
3172                 bond_send_gratuitous_arp(bond);
3173                 read_unlock(&bond->curr_slave_lock);
3174         }
3175
3176         if (bond_ab_arp_inspect(bond, delta_in_ticks)) {
3177                 read_unlock(&bond->lock);
3178                 rtnl_lock();
3179                 read_lock(&bond->lock);
3180
3181                 bond_ab_arp_commit(bond, delta_in_ticks);
3182
3183                 read_unlock(&bond->lock);
3184                 rtnl_unlock();
3185                 read_lock(&bond->lock);
3186         }
3187
3188         bond_ab_arp_probe(bond);
3189
3190 re_arm:
3191         if (bond->params.arp_interval) {
3192                 queue_delayed_work(bond->wq, &bond->arp_work, delta_in_ticks);
3193         }
3194 out:
3195         read_unlock(&bond->lock);
3196 }
3197
3198 /*------------------------------ proc/seq_file-------------------------------*/
3199
3200 #ifdef CONFIG_PROC_FS
3201
3202 static void *bond_info_seq_start(struct seq_file *seq, loff_t *pos)
3203 {
3204         struct bonding *bond = seq->private;
3205         loff_t off = 0;
3206         struct slave *slave;
3207         int i;
3208
3209         /* make sure the bond won't be taken away */
3210         read_lock(&dev_base_lock);
3211         read_lock(&bond->lock);
3212
3213         if (*pos == 0) {
3214                 return SEQ_START_TOKEN;
3215         }
3216
3217         bond_for_each_slave(bond, slave, i) {
3218                 if (++off == *pos) {
3219                         return slave;
3220                 }
3221         }
3222
3223         return NULL;
3224 }
3225
3226 static void *bond_info_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3227 {
3228         struct bonding *bond = seq->private;
3229         struct slave *slave = v;
3230
3231         ++*pos;
3232         if (v == SEQ_START_TOKEN) {
3233                 return bond->first_slave;
3234         }
3235
3236         slave = slave->next;
3237
3238         return (slave == bond->first_slave) ? NULL : slave;
3239 }
3240
3241 static void bond_info_seq_stop(struct seq_file *seq, void *v)
3242 {
3243         struct bonding *bond = seq->private;
3244
3245         read_unlock(&bond->lock);
3246         read_unlock(&dev_base_lock);
3247 }
3248
3249 static void bond_info_show_master(struct seq_file *seq)
3250 {
3251         struct bonding *bond = seq->private;
3252         struct slave *curr;
3253         int i;
3254         u32 target;
3255
3256         read_lock(&bond->curr_slave_lock);
3257         curr = bond->curr_active_slave;
3258         read_unlock(&bond->curr_slave_lock);
3259
3260         seq_printf(seq, "Bonding Mode: %s",
3261                    bond_mode_name(bond->params.mode));
3262
3263         if (bond->params.mode == BOND_MODE_ACTIVEBACKUP &&
3264             bond->params.fail_over_mac)
3265                 seq_printf(seq, " (fail_over_mac %s)",
3266                    fail_over_mac_tbl[bond->params.fail_over_mac].modename);
3267
3268         seq_printf(seq, "\n");
3269
3270         if (bond->params.mode == BOND_MODE_XOR ||
3271                 bond->params.mode == BOND_MODE_8023AD) {
3272                 seq_printf(seq, "Transmit Hash Policy: %s (%d)\n",
3273                         xmit_hashtype_tbl[bond->params.xmit_policy].modename,
3274                         bond->params.xmit_policy);
3275         }
3276
3277         if (USES_PRIMARY(bond->params.mode)) {
3278                 seq_printf(seq, "Primary Slave: %s\n",
3279                            (bond->primary_slave) ?
3280                            bond->primary_slave->dev->name : "None");
3281
3282                 seq_printf(seq, "Currently Active Slave: %s\n",
3283                            (curr) ? curr->dev->name : "None");
3284         }
3285
3286         seq_printf(seq, "MII Status: %s\n", netif_carrier_ok(bond->dev) ?
3287                    "up" : "down");
3288         seq_printf(seq, "MII Polling Interval (ms): %d\n", bond->params.miimon);
3289         seq_printf(seq, "Up Delay (ms): %d\n",
3290                    bond->params.updelay * bond->params.miimon);
3291         seq_printf(seq, "Down Delay (ms): %d\n",
3292                    bond->params.downdelay * bond->params.miimon);
3293
3294
3295         /* ARP information */
3296         if(bond->params.arp_interval > 0) {
3297                 int printed=0;
3298                 seq_printf(seq, "ARP Polling Interval (ms): %d\n",
3299                                 bond->params.arp_interval);
3300
3301                 seq_printf(seq, "ARP IP target/s (n.n.n.n form):");
3302
3303                 for(i = 0; (i < BOND_MAX_ARP_TARGETS) ;i++) {
3304                         if (!bond->params.arp_targets[i])
3305                                 continue;
3306                         if (printed)
3307                                 seq_printf(seq, ",");
3308                         target = ntohl(bond->params.arp_targets[i]);
3309                         seq_printf(seq, " %d.%d.%d.%d", HIPQUAD(target));
3310                         printed = 1;
3311                 }
3312                 seq_printf(seq, "\n");
3313         }
3314
3315         if (bond->params.mode == BOND_MODE_8023AD) {
3316                 struct ad_info ad_info;
3317                 DECLARE_MAC_BUF(mac);
3318
3319                 seq_puts(seq, "\n802.3ad info\n");
3320                 seq_printf(seq, "LACP rate: %s\n",
3321                            (bond->params.lacp_fast) ? "fast" : "slow");
3322
3323                 if (bond_3ad_get_active_agg_info(bond, &ad_info)) {
3324                         seq_printf(seq, "bond %s has no active aggregator\n",
3325                                    bond->dev->name);
3326                 } else {
3327                         seq_printf(seq, "Active Aggregator Info:\n");
3328
3329                         seq_printf(seq, "\tAggregator ID: %d\n",
3330                                    ad_info.aggregator_id);
3331                         seq_printf(seq, "\tNumber of ports: %d\n",
3332                                    ad_info.ports);
3333                         seq_printf(seq, "\tActor Key: %d\n",
3334                                    ad_info.actor_key);
3335                         seq_printf(seq, "\tPartner Key: %d\n",
3336                                    ad_info.partner_key);
3337                         seq_printf(seq, "\tPartner Mac Address: %s\n",
3338                                    print_mac(mac, ad_info.partner_system));
3339                 }
3340         }
3341 }
3342
3343 static void bond_info_show_slave(struct seq_file *seq, const struct slave *slave)
3344 {
3345         struct bonding *bond = seq->private;
3346         DECLARE_MAC_BUF(mac);
3347
3348         seq_printf(seq, "\nSlave Interface: %s\n", slave->dev->name);
3349         seq_printf(seq, "MII Status: %s\n",
3350                    (slave->link == BOND_LINK_UP) ?  "up" : "down");
3351         seq_printf(seq, "Link Failure Count: %u\n",
3352                    slave->link_failure_count);
3353
3354         seq_printf(seq,
3355                    "Permanent HW addr: %s\n",
3356                    print_mac(mac, slave->perm_hwaddr));
3357
3358         if (bond->params.mode == BOND_MODE_8023AD) {
3359                 const struct aggregator *agg
3360                         = SLAVE_AD_INFO(slave).port.aggregator;
3361
3362                 if (agg) {
3363                         seq_printf(seq, "Aggregator ID: %d\n",
3364                                    agg->aggregator_identifier);
3365                 } else {
3366                         seq_puts(seq, "Aggregator ID: N/A\n");
3367                 }
3368         }
3369 }
3370
3371 static int bond_info_seq_show(struct seq_file *seq, void *v)
3372 {
3373         if (v == SEQ_START_TOKEN) {
3374                 seq_printf(seq, "%s\n", version);
3375                 bond_info_show_master(seq);
3376         } else {
3377                 bond_info_show_slave(seq, v);
3378         }
3379
3380         return 0;
3381 }
3382
3383 static struct seq_operations bond_info_seq_ops = {
3384         .start = bond_info_seq_start,
3385         .next  = bond_info_seq_next,
3386         .stop  = bond_info_seq_stop,
3387         .show  = bond_info_seq_show,
3388 };
3389
3390 static int bond_info_open(struct inode *inode, struct file *file)
3391 {
3392         struct seq_file *seq;
3393         struct proc_dir_entry *proc;
3394         int res;
3395
3396         res = seq_open(file, &bond_info_seq_ops);
3397         if (!res) {
3398                 /* recover the pointer buried in proc_dir_entry data */
3399                 seq = file->private_data;
3400                 proc = PDE(inode);
3401                 seq->private = proc->data;
3402         }
3403
3404         return res;
3405 }
3406
3407 static const struct file_operations bond_info_fops = {
3408         .owner   = THIS_MODULE,
3409         .open    = bond_info_open,
3410         .read    = seq_read,
3411         .llseek  = seq_lseek,
3412         .release = seq_release,
3413 };
3414
3415 static int bond_create_proc_entry(struct bonding *bond)
3416 {
3417         struct net_device *bond_dev = bond->dev;
3418
3419         if (bond_proc_dir) {
3420                 bond->proc_entry = proc_create_data(bond_dev->name,
3421                                                     S_IRUGO, bond_proc_dir,
3422                                                     &bond_info_fops, bond);
3423                 if (bond->proc_entry == NULL) {
3424                         printk(KERN_WARNING DRV_NAME
3425                                ": Warning: Cannot create /proc/net/%s/%s\n",
3426                                DRV_NAME, bond_dev->name);
3427                 } else {
3428                         memcpy(bond->proc_file_name, bond_dev->name, IFNAMSIZ);
3429                 }
3430         }
3431
3432         return 0;
3433 }
3434
3435 static void bond_remove_proc_entry(struct bonding *bond)
3436 {
3437         if (bond_proc_dir && bond->proc_entry) {
3438                 remove_proc_entry(bond->proc_file_name, bond_proc_dir);
3439                 memset(bond->proc_file_name, 0, IFNAMSIZ);
3440                 bond->proc_entry = NULL;
3441         }
3442 }
3443
3444 /* Create the bonding directory under /proc/net, if doesn't exist yet.
3445  * Caller must hold rtnl_lock.
3446  */
3447 static void bond_create_proc_dir(void)
3448 {
3449         int len = strlen(DRV_NAME);
3450
3451         for (bond_proc_dir = init_net.proc_net->subdir; bond_proc_dir;
3452              bond_proc_dir = bond_proc_dir->next) {
3453                 if ((bond_proc_dir->namelen == len) &&
3454                     !memcmp(bond_proc_dir->name, DRV_NAME, len)) {
3455                         break;
3456                 }
3457         }
3458
3459         if (!bond_proc_dir) {
3460                 bond_proc_dir = proc_mkdir(DRV_NAME, init_net.proc_net);
3461                 if (bond_proc_dir) {
3462                         bond_proc_dir->owner = THIS_MODULE;
3463                 } else {
3464                         printk(KERN_WARNING DRV_NAME
3465                                 ": Warning: cannot create /proc/net/%s\n",
3466                                 DRV_NAME);
3467                 }
3468         }
3469 }
3470
3471 /* Destroy the bonding directory under /proc/net, if empty.
3472  * Caller must hold rtnl_lock.
3473  */
3474 static void bond_destroy_proc_dir(void)
3475 {
3476         struct proc_dir_entry *de;
3477
3478         if (!bond_proc_dir) {
3479                 return;
3480         }
3481
3482         /* verify that the /proc dir is empty */
3483         for (de = bond_proc_dir->subdir; de; de = de->next) {
3484                 /* ignore . and .. */
3485                 if (*(de->name) != '.') {
3486                         break;
3487                 }
3488         }
3489
3490         if (de) {
3491                 if (bond_proc_dir->owner == THIS_MODULE) {
3492                         bond_proc_dir->owner = NULL;
3493                 }
3494         } else {
3495                 remove_proc_entry(DRV_NAME, init_net.proc_net);
3496                 bond_proc_dir = NULL;
3497         }
3498 }
3499 #endif /* CONFIG_PROC_FS */
3500
3501 /*-------------------------- netdev event handling --------------------------*/
3502
3503 /*
3504  * Change device name
3505  */
3506 static int bond_event_changename(struct bonding *bond)
3507 {
3508 #ifdef CONFIG_PROC_FS
3509         bond_remove_proc_entry(bond);
3510         bond_create_proc_entry(bond);
3511 #endif
3512         down_write(&(bonding_rwsem));
3513         bond_destroy_sysfs_entry(bond);
3514         bond_create_sysfs_entry(bond);
3515         up_write(&(bonding_rwsem));
3516         return NOTIFY_DONE;
3517 }
3518
3519 static int bond_master_netdev_event(unsigned long event, struct net_device *bond_dev)
3520 {
3521         struct bonding *event_bond = bond_dev->priv;
3522
3523         switch (event) {
3524         case NETDEV_CHANGENAME:
3525                 return bond_event_changename(event_bond);
3526         case NETDEV_UNREGISTER:
3527                 bond_release_all(event_bond->dev);
3528                 break;
3529         default:
3530                 break;
3531         }
3532
3533         return NOTIFY_DONE;
3534 }
3535
3536 static int bond_slave_netdev_event(unsigned long event, struct net_device *slave_dev)
3537 {
3538         struct net_device *bond_dev = slave_dev->master;
3539         struct bonding *bond = bond_dev->priv;
3540
3541         switch (event) {
3542         case NETDEV_UNREGISTER:
3543                 if (bond_dev) {
3544                         if (bond->setup_by_slave)
3545                                 bond_release_and_destroy(bond_dev, slave_dev);
3546                         else
3547                                 bond_release(bond_dev, slave_dev);
3548                 }
3549                 break;
3550         case NETDEV_CHANGE:
3551                 /*
3552                  * TODO: is this what we get if somebody
3553                  * sets up a hierarchical bond, then rmmod's
3554                  * one of the slave bonding devices?
3555                  */
3556                 break;
3557         case NETDEV_DOWN:
3558                 /*
3559                  * ... Or is it this?
3560                  */
3561                 break;
3562         case NETDEV_CHANGEMTU:
3563                 /*
3564                  * TODO: Should slaves be allowed to
3565                  * independently alter their MTU?  For
3566                  * an active-backup bond, slaves need
3567                  * not be the same type of device, so
3568                  * MTUs may vary.  For other modes,
3569                  * slaves arguably should have the
3570                  * same MTUs. To do this, we'd need to
3571                  * take over the slave's change_mtu
3572                  * function for the duration of their
3573                  * servitude.
3574                  */
3575                 break;
3576         case NETDEV_CHANGENAME:
3577                 /*
3578                  * TODO: handle changing the primary's name
3579                  */
3580                 break;
3581         case NETDEV_FEAT_CHANGE:
3582                 bond_compute_features(bond);
3583                 break;
3584         default:
3585                 break;
3586         }
3587
3588         return NOTIFY_DONE;
3589 }
3590
3591 /*
3592  * bond_netdev_event: handle netdev notifier chain events.
3593  *
3594  * This function receives events for the netdev chain.  The caller (an
3595  * ioctl handler calling blocking_notifier_call_chain) holds the necessary
3596  * locks for us to safely manipulate the slave devices (RTNL lock,
3597  * dev_probe_lock).
3598  */
3599 static int bond_netdev_event(struct notifier_block *this, unsigned long event, void *ptr)
3600 {
3601         struct net_device *event_dev = (struct net_device *)ptr;
3602
3603         if (dev_net(event_dev) != &init_net)
3604                 return NOTIFY_DONE;
3605
3606         dprintk("event_dev: %s, event: %lx\n",
3607                 (event_dev ? event_dev->name : "None"),
3608                 event);
3609
3610         if (!(event_dev->priv_flags & IFF_BONDING))
3611                 return NOTIFY_DONE;
3612
3613         if (event_dev->flags & IFF_MASTER) {
3614                 dprintk("IFF_MASTER\n");
3615                 return bond_master_netdev_event(event, event_dev);
3616         }
3617
3618         if (event_dev->flags & IFF_SLAVE) {
3619                 dprintk("IFF_SLAVE\n");
3620                 return bond_slave_netdev_event(event, event_dev);
3621         }
3622
3623         return NOTIFY_DONE;
3624 }
3625
3626 /*
3627  * bond_inetaddr_event: handle inetaddr notifier chain events.
3628  *
3629  * We keep track of device IPs primarily to use as source addresses in
3630  * ARP monitor probes (rather than spewing out broadcasts all the time).
3631  *
3632  * We track one IP for the main device (if it has one), plus one per VLAN.
3633  */
3634 static int bond_inetaddr_event(struct notifier_block *this, unsigned long event, void *ptr)
3635 {
3636         struct in_ifaddr *ifa = ptr;
3637         struct net_device *vlan_dev, *event_dev = ifa->ifa_dev->dev;
3638         struct bonding *bond;
3639         struct vlan_entry *vlan;
3640
3641         if (dev_net(ifa->ifa_dev->dev) != &init_net)
3642                 return NOTIFY_DONE;
3643
3644         list_for_each_entry(bond, &bond_dev_list, bond_list) {
3645                 if (bond->dev == event_dev) {
3646                         switch (event) {
3647                         case NETDEV_UP:
3648                                 bond->master_ip = ifa->ifa_local;
3649                                 return NOTIFY_OK;
3650                         case NETDEV_DOWN:
3651                                 bond->master_ip = bond_glean_dev_ip(bond->dev);
3652                                 return NOTIFY_OK;
3653                         default:
3654                                 return NOTIFY_DONE;
3655                         }
3656                 }
3657
3658                 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
3659                         vlan_dev = vlan_group_get_device(bond->vlgrp, vlan->vlan_id);
3660                         if (vlan_dev == event_dev) {
3661                                 switch (event) {
3662                                 case NETDEV_UP:
3663                                         vlan->vlan_ip = ifa->ifa_local;
3664                                         return NOTIFY_OK;
3665                                 case NETDEV_DOWN:
3666                                         vlan->vlan_ip =
3667                                                 bond_glean_dev_ip(vlan_dev);
3668                                         return NOTIFY_OK;
3669                                 default:
3670                                         return NOTIFY_DONE;
3671                                 }
3672                         }
3673                 }
3674         }
3675         return NOTIFY_DONE;
3676 }
3677
3678 static struct notifier_block bond_netdev_notifier = {
3679         .notifier_call = bond_netdev_event,
3680 };
3681
3682 static struct notifier_block bond_inetaddr_notifier = {
3683         .notifier_call = bond_inetaddr_event,
3684 };
3685
3686 /*-------------------------- Packet type handling ---------------------------*/
3687
3688 /* register to receive lacpdus on a bond */
3689 static void bond_register_lacpdu(struct bonding *bond)
3690 {
3691         struct packet_type *pk_type = &(BOND_AD_INFO(bond).ad_pkt_type);
3692
3693         /* initialize packet type */
3694         pk_type->type = PKT_TYPE_LACPDU;
3695         pk_type->dev = bond->dev;
3696         pk_type->func = bond_3ad_lacpdu_recv;
3697
3698         dev_add_pack(pk_type);
3699 }
3700
3701 /* unregister to receive lacpdus on a bond */
3702 static void bond_unregister_lacpdu(struct bonding *bond)
3703 {
3704         dev_remove_pack(&(BOND_AD_INFO(bond).ad_pkt_type));
3705 }
3706
3707 void bond_register_arp(struct bonding *bond)
3708 {
3709         struct packet_type *pt = &bond->arp_mon_pt;
3710
3711         if (pt->type)
3712                 return;
3713
3714         pt->type = htons(ETH_P_ARP);
3715         pt->dev = bond->dev;
3716         pt->func = bond_arp_rcv;
3717         dev_add_pack(pt);
3718 }
3719
3720 void bond_unregister_arp(struct bonding *bond)
3721 {
3722         struct packet_type *pt = &bond->arp_mon_pt;
3723
3724         dev_remove_pack(pt);
3725         pt->type = 0;
3726 }
3727
3728 /*---------------------------- Hashing Policies -----------------------------*/
3729
3730 /*
3731  * Hash for the output device based upon layer 2 and layer 3 data. If
3732  * the packet is not IP mimic bond_xmit_hash_policy_l2()
3733  */
3734 static int bond_xmit_hash_policy_l23(struct sk_buff *skb,
3735                                      struct net_device *bond_dev, int count)
3736 {
3737         struct ethhdr *data = (struct ethhdr *)skb->data;
3738         struct iphdr *iph = ip_hdr(skb);
3739
3740         if (skb->protocol == __constant_htons(ETH_P_IP)) {
3741                 return ((ntohl(iph->saddr ^ iph->daddr) & 0xffff) ^
3742                         (data->h_dest[5] ^ bond_dev->dev_addr[5])) % count;
3743         }
3744
3745         return (data->h_dest[5] ^ bond_dev->dev_addr[5]) % count;
3746 }
3747
3748 /*
3749  * Hash for the output device based upon layer 3 and layer 4 data. If
3750  * the packet is a frag or not TCP or UDP, just use layer 3 data.  If it is
3751  * altogether not IP, mimic bond_xmit_hash_policy_l2()
3752  */
3753 static int bond_xmit_hash_policy_l34(struct sk_buff *skb,
3754                                     struct net_device *bond_dev, int count)
3755 {
3756         struct ethhdr *data = (struct ethhdr *)skb->data;
3757         struct iphdr *iph = ip_hdr(skb);
3758         __be16 *layer4hdr = (__be16 *)((u32 *)iph + iph->ihl);
3759         int layer4_xor = 0;
3760
3761         if (skb->protocol == __constant_htons(ETH_P_IP)) {
3762                 if (!(iph->frag_off & __constant_htons(IP_MF|IP_OFFSET)) &&
3763                     (iph->protocol == IPPROTO_TCP ||
3764                      iph->protocol == IPPROTO_UDP)) {
3765                         layer4_xor = ntohs((*layer4hdr ^ *(layer4hdr + 1)));
3766                 }
3767                 return (layer4_xor ^
3768                         ((ntohl(iph->saddr ^ iph->daddr)) & 0xffff)) % count;
3769
3770         }
3771
3772         return (data->h_dest[5] ^ bond_dev->dev_addr[5]) % count;
3773 }
3774
3775 /*
3776  * Hash for the output device based upon layer 2 data
3777  */
3778 static int bond_xmit_hash_policy_l2(struct sk_buff *skb,
3779                                    struct net_device *bond_dev, int count)
3780 {
3781         struct ethhdr *data = (struct ethhdr *)skb->data;
3782
3783         return (data->h_dest[5] ^ bond_dev->dev_addr[5]) % count;
3784 }
3785
3786 /*-------------------------- Device entry points ----------------------------*/
3787
3788 static int bond_open(struct net_device *bond_dev)
3789 {
3790         struct bonding *bond = bond_dev->priv;
3791
3792         bond->kill_timers = 0;
3793
3794         if ((bond->params.mode == BOND_MODE_TLB) ||
3795             (bond->params.mode == BOND_MODE_ALB)) {
3796                 /* bond_alb_initialize must be called before the timer
3797                  * is started.
3798                  */
3799                 if (bond_alb_initialize(bond, (bond->params.mode == BOND_MODE_ALB))) {
3800                         /* something went wrong - fail the open operation */
3801                         return -1;
3802                 }
3803
3804                 INIT_DELAYED_WORK(&bond->alb_work, bond_alb_monitor);
3805                 queue_delayed_work(bond->wq, &bond->alb_work, 0);
3806         }
3807
3808         if (bond->params.miimon) {  /* link check interval, in milliseconds. */
3809                 INIT_DELAYED_WORK(&bond->mii_work, bond_mii_monitor);
3810                 queue_delayed_work(bond->wq, &bond->mii_work, 0);
3811         }
3812
3813         if (bond->params.arp_interval) {  /* arp interval, in milliseconds. */
3814                 if (bond->params.mode == BOND_MODE_ACTIVEBACKUP)
3815                         INIT_DELAYED_WORK(&bond->arp_work,
3816                                           bond_activebackup_arp_mon);
3817                 else
3818                         INIT_DELAYED_WORK(&bond->arp_work,
3819                                           bond_loadbalance_arp_mon);
3820
3821                 queue_delayed_work(bond->wq, &bond->arp_work, 0);
3822                 if (bond->params.arp_validate)
3823                         bond_register_arp(bond);
3824         }
3825
3826         if (bond->params.mode == BOND_MODE_8023AD) {
3827                 INIT_DELAYED_WORK(&bond->ad_work, bond_3ad_state_machine_handler);
3828                 queue_delayed_work(bond->wq, &bond->ad_work, 0);
3829                 /* register to receive LACPDUs */
3830                 bond_register_lacpdu(bond);
3831         }
3832
3833         return 0;
3834 }
3835
3836 static int bond_close(struct net_device *bond_dev)
3837 {
3838         struct bonding *bond = bond_dev->priv;
3839
3840         if (bond->params.mode == BOND_MODE_8023AD) {
3841                 /* Unregister the receive of LACPDUs */
3842                 bond_unregister_lacpdu(bond);
3843         }
3844
3845         if (bond->params.arp_validate)
3846                 bond_unregister_arp(bond);
3847
3848         write_lock_bh(&bond->lock);
3849
3850         bond->send_grat_arp = 0;
3851
3852         /* signal timers not to re-arm */
3853         bond->kill_timers = 1;
3854
3855         write_unlock_bh(&bond->lock);
3856
3857         if (bond->params.miimon) {  /* link check interval, in milliseconds. */
3858                 cancel_delayed_work(&bond->mii_work);
3859         }
3860
3861         if (bond->params.arp_interval) {  /* arp interval, in milliseconds. */
3862                 cancel_delayed_work(&bond->arp_work);
3863         }
3864
3865         switch (bond->params.mode) {
3866         case BOND_MODE_8023AD:
3867                 cancel_delayed_work(&bond->ad_work);
3868                 break;
3869         case BOND_MODE_TLB:
3870         case BOND_MODE_ALB:
3871                 cancel_delayed_work(&bond->alb_work);
3872                 break;
3873         default:
3874                 break;
3875         }
3876
3877
3878         if ((bond->params.mode == BOND_MODE_TLB) ||
3879             (bond->params.mode == BOND_MODE_ALB)) {
3880                 /* Must be called only after all
3881                  * slaves have been released
3882                  */
3883                 bond_alb_deinitialize(bond);
3884         }
3885
3886         return 0;
3887 }
3888
3889 static struct net_device_stats *bond_get_stats(struct net_device *bond_dev)
3890 {
3891         struct bonding *bond = bond_dev->priv;
3892         struct net_device_stats *stats = &(bond->stats), *sstats;
3893         struct net_device_stats local_stats;
3894         struct slave *slave;
3895         int i;
3896
3897         memset(&local_stats, 0, sizeof(struct net_device_stats));
3898
3899         read_lock_bh(&bond->lock);
3900
3901         bond_for_each_slave(bond, slave, i) {
3902                 sstats = slave->dev->get_stats(slave->dev);
3903                 local_stats.rx_packets += sstats->rx_packets;
3904                 local_stats.rx_bytes += sstats->rx_bytes;
3905                 local_stats.rx_errors += sstats->rx_errors;
3906                 local_stats.rx_dropped += sstats->rx_dropped;
3907
3908                 local_stats.tx_packets += sstats->tx_packets;
3909                 local_stats.tx_bytes += sstats->tx_bytes;
3910                 local_stats.tx_errors += sstats->tx_errors;
3911                 local_stats.tx_dropped += sstats->tx_dropped;
3912
3913                 local_stats.multicast += sstats->multicast;
3914                 local_stats.collisions += sstats->collisions;
3915
3916                 local_stats.rx_length_errors += sstats->rx_length_errors;
3917                 local_stats.rx_over_errors += sstats->rx_over_errors;
3918                 local_stats.rx_crc_errors += sstats->rx_crc_errors;
3919                 local_stats.rx_frame_errors += sstats->rx_frame_errors;
3920                 local_stats.rx_fifo_errors += sstats->rx_fifo_errors;
3921                 local_stats.rx_missed_errors += sstats->rx_missed_errors;
3922
3923                 local_stats.tx_aborted_errors += sstats->tx_aborted_errors;
3924                 local_stats.tx_carrier_errors += sstats->tx_carrier_errors;
3925                 local_stats.tx_fifo_errors += sstats->tx_fifo_errors;
3926                 local_stats.tx_heartbeat_errors += sstats->tx_heartbeat_errors;
3927                 local_stats.tx_window_errors += sstats->tx_window_errors;
3928         }
3929
3930         memcpy(stats, &local_stats, sizeof(struct net_device_stats));
3931
3932         read_unlock_bh(&bond->lock);
3933
3934         return stats;
3935 }
3936
3937 static int bond_do_ioctl(struct net_device *bond_dev, struct ifreq *ifr, int cmd)
3938 {
3939         struct net_device *slave_dev = NULL;
3940         struct ifbond k_binfo;
3941         struct ifbond __user *u_binfo = NULL;
3942         struct ifslave k_sinfo;
3943         struct ifslave __user *u_sinfo = NULL;
3944         struct mii_ioctl_data *mii = NULL;
3945         int res = 0;
3946
3947         dprintk("bond_ioctl: master=%s, cmd=%d\n",
3948                 bond_dev->name, cmd);
3949
3950         switch (cmd) {
3951         case SIOCGMIIPHY:
3952                 mii = if_mii(ifr);
3953                 if (!mii) {
3954                         return -EINVAL;
3955                 }
3956                 mii->phy_id = 0;
3957                 /* Fall Through */
3958         case SIOCGMIIREG:
3959                 /*
3960                  * We do this again just in case we were called by SIOCGMIIREG
3961                  * instead of SIOCGMIIPHY.
3962                  */
3963                 mii = if_mii(ifr);
3964                 if (!mii) {
3965                         return -EINVAL;
3966                 }
3967
3968                 if (mii->reg_num == 1) {
3969                         struct bonding *bond = bond_dev->priv;
3970                         mii->val_out = 0;
3971                         read_lock(&bond->lock);
3972                         read_lock(&bond->curr_slave_lock);
3973                         if (netif_carrier_ok(bond->dev)) {
3974                                 mii->val_out = BMSR_LSTATUS;
3975                         }
3976                         read_unlock(&bond->curr_slave_lock);
3977                         read_unlock(&bond->lock);
3978                 }
3979
3980                 return 0;
3981         case BOND_INFO_QUERY_OLD:
3982         case SIOCBONDINFOQUERY:
3983                 u_binfo = (struct ifbond __user *)ifr->ifr_data;
3984
3985                 if (copy_from_user(&k_binfo, u_binfo, sizeof(ifbond))) {
3986                         return -EFAULT;
3987                 }
3988
3989                 res = bond_info_query(bond_dev, &k_binfo);
3990                 if (res == 0) {
3991                         if (copy_to_user(u_binfo, &k_binfo, sizeof(ifbond))) {
3992                                 return -EFAULT;
3993                         }
3994                 }
3995
3996                 return res;
3997         case BOND_SLAVE_INFO_QUERY_OLD:
3998         case SIOCBONDSLAVEINFOQUERY:
3999                 u_sinfo = (struct ifslave __user *)ifr->ifr_data;
4000
4001                 if (copy_from_user(&k_sinfo, u_sinfo, sizeof(ifslave))) {
4002                         return -EFAULT;
4003                 }
4004
4005                 res = bond_slave_info_query(bond_dev, &k_sinfo);
4006                 if (res == 0) {
4007                         if (copy_to_user(u_sinfo, &k_sinfo, sizeof(ifslave))) {
4008                                 return -EFAULT;
4009                         }
4010                 }
4011
4012                 return res;
4013         default:
4014                 /* Go on */
4015                 break;
4016         }
4017
4018         if (!capable(CAP_NET_ADMIN)) {
4019                 return -EPERM;
4020         }
4021
4022         down_write(&(bonding_rwsem));
4023         slave_dev = dev_get_by_name(&init_net, ifr->ifr_slave);
4024
4025         dprintk("slave_dev=%p: \n", slave_dev);
4026
4027         if (!slave_dev) {
4028                 res = -ENODEV;
4029         } else {
4030                 dprintk("slave_dev->name=%s: \n", slave_dev->name);
4031                 switch (cmd) {
4032                 case BOND_ENSLAVE_OLD:
4033                 case SIOCBONDENSLAVE:
4034                         res = bond_enslave(bond_dev, slave_dev);
4035                         break;
4036                 case BOND_RELEASE_OLD:
4037                 case SIOCBONDRELEASE:
4038                         res = bond_release(bond_dev, slave_dev);
4039                         break;
4040                 case BOND_SETHWADDR_OLD:
4041                 case SIOCBONDSETHWADDR:
4042                         res = bond_sethwaddr(bond_dev, slave_dev);
4043                         break;
4044                 case BOND_CHANGE_ACTIVE_OLD:
4045                 case SIOCBONDCHANGEACTIVE:
4046                         res = bond_ioctl_change_active(bond_dev, slave_dev);
4047                         break;
4048                 default:
4049                         res = -EOPNOTSUPP;
4050                 }
4051
4052                 dev_put(slave_dev);
4053         }
4054
4055         up_write(&(bonding_rwsem));
4056         return res;
4057 }
4058
4059 static void bond_set_multicast_list(struct net_device *bond_dev)
4060 {
4061         struct bonding *bond = bond_dev->priv;
4062         struct dev_mc_list *dmi;
4063
4064         /*
4065          * Do promisc before checking multicast_mode
4066          */
4067         if ((bond_dev->flags & IFF_PROMISC) && !(bond->flags & IFF_PROMISC)) {
4068                 bond_set_promiscuity(bond, 1);
4069         }
4070
4071         if (!(bond_dev->flags & IFF_PROMISC) && (bond->flags & IFF_PROMISC)) {
4072                 bond_set_promiscuity(bond, -1);
4073         }
4074
4075         /* set allmulti flag to slaves */
4076         if ((bond_dev->flags & IFF_ALLMULTI) && !(bond->flags & IFF_ALLMULTI)) {
4077                 bond_set_allmulti(bond, 1);
4078         }
4079
4080         if (!(bond_dev->flags & IFF_ALLMULTI) && (bond->flags & IFF_ALLMULTI)) {
4081                 bond_set_allmulti(bond, -1);
4082         }
4083
4084         read_lock(&bond->lock);
4085
4086         bond->flags = bond_dev->flags;
4087
4088         /* looking for addresses to add to slaves' mc list */
4089         for (dmi = bond_dev->mc_list; dmi; dmi = dmi->next) {
4090                 if (!bond_mc_list_find_dmi(dmi, bond->mc_list)) {
4091                         bond_mc_add(bond, dmi->dmi_addr, dmi->dmi_addrlen);
4092                 }
4093         }
4094
4095         /* looking for addresses to delete from slaves' list */
4096         for (dmi = bond->mc_list; dmi; dmi = dmi->next) {
4097                 if (!bond_mc_list_find_dmi(dmi, bond_dev->mc_list)) {
4098                         bond_mc_delete(bond, dmi->dmi_addr, dmi->dmi_addrlen);
4099                 }
4100         }
4101
4102         /* save master's multicast list */
4103         bond_mc_list_destroy(bond);
4104         bond_mc_list_copy(bond_dev->mc_list, bond, GFP_ATOMIC);
4105
4106         read_unlock(&bond->lock);
4107 }
4108
4109 /*
4110  * Change the MTU of all of a master's slaves to match the master
4111  */
4112 static int bond_change_mtu(struct net_device *bond_dev, int new_mtu)
4113 {
4114         struct bonding *bond = bond_dev->priv;
4115         struct slave *slave, *stop_at;
4116         int res = 0;
4117         int i;
4118
4119         dprintk("bond=%p, name=%s, new_mtu=%d\n", bond,
4120                 (bond_dev ? bond_dev->name : "None"), new_mtu);
4121
4122         /* Can't hold bond->lock with bh disabled here since
4123          * some base drivers panic. On the other hand we can't
4124          * hold bond->lock without bh disabled because we'll
4125          * deadlock. The only solution is to rely on the fact
4126          * that we're under rtnl_lock here, and the slaves
4127          * list won't change. This doesn't solve the problem
4128          * of setting the slave's MTU while it is
4129          * transmitting, but the assumption is that the base
4130          * driver can handle that.
4131          *
4132          * TODO: figure out a way to safely iterate the slaves
4133          * list, but without holding a lock around the actual
4134          * call to the base driver.
4135          */
4136
4137         bond_for_each_slave(bond, slave, i) {
4138                 dprintk("s %p s->p %p c_m %p\n", slave,
4139                         slave->prev, slave->dev->change_mtu);
4140
4141                 res = dev_set_mtu(slave->dev, new_mtu);
4142
4143                 if (res) {
4144                         /* If we failed to set the slave's mtu to the new value
4145                          * we must abort the operation even in ACTIVE_BACKUP
4146                          * mode, because if we allow the backup slaves to have
4147                          * different mtu values than the active slave we'll
4148                          * need to change their mtu when doing a failover. That
4149                          * means changing their mtu from timer context, which
4150                          * is probably not a good idea.
4151                          */
4152                         dprintk("err %d %s\n", res, slave->dev->name);
4153                         goto unwind;
4154                 }
4155         }
4156
4157         bond_dev->mtu = new_mtu;
4158
4159         return 0;
4160
4161 unwind:
4162         /* unwind from head to the slave that failed */
4163         stop_at = slave;
4164         bond_for_each_slave_from_to(bond, slave, i, bond->first_slave, stop_at) {
4165                 int tmp_res;
4166
4167                 tmp_res = dev_set_mtu(slave->dev, bond_dev->mtu);
4168                 if (tmp_res) {
4169                         dprintk("unwind err %d dev %s\n", tmp_res,
4170                                 slave->dev->name);
4171                 }
4172         }
4173
4174         return res;
4175 }
4176
4177 /*
4178  * Change HW address
4179  *
4180  * Note that many devices must be down to change the HW address, and
4181  * downing the master releases all slaves.  We can make bonds full of
4182  * bonding devices to test this, however.
4183  */
4184 static int bond_set_mac_address(struct net_device *bond_dev, void *addr)
4185 {
4186         struct bonding *bond = bond_dev->priv;
4187         struct sockaddr *sa = addr, tmp_sa;
4188         struct slave *slave, *stop_at;
4189         int res = 0;
4190         int i;
4191
4192         dprintk("bond=%p, name=%s\n", bond, (bond_dev ? bond_dev->name : "None"));
4193
4194         /*
4195          * If fail_over_mac is set to active, do nothing and return
4196          * success.  Returning an error causes ifenslave to fail.
4197          */
4198         if (bond->params.fail_over_mac == BOND_FOM_ACTIVE)
4199                 return 0;
4200
4201         if (!is_valid_ether_addr(sa->sa_data)) {
4202                 return -EADDRNOTAVAIL;
4203         }
4204
4205         /* Can't hold bond->lock with bh disabled here since
4206          * some base drivers panic. On the other hand we can't
4207          * hold bond->lock without bh disabled because we'll
4208          * deadlock. The only solution is to rely on the fact
4209          * that we're under rtnl_lock here, and the slaves
4210          * list won't change. This doesn't solve the problem
4211          * of setting the slave's hw address while it is
4212          * transmitting, but the assumption is that the base
4213          * driver can handle that.
4214          *
4215          * TODO: figure out a way to safely iterate the slaves
4216          * list, but without holding a lock around the actual
4217          * call to the base driver.
4218          */
4219
4220         bond_for_each_slave(bond, slave, i) {
4221                 dprintk("slave %p %s\n", slave, slave->dev->name);
4222
4223                 if (slave->dev->set_mac_address == NULL) {
4224                         res = -EOPNOTSUPP;
4225                         dprintk("EOPNOTSUPP %s\n", slave->dev->name);
4226                         goto unwind;
4227                 }
4228
4229                 res = dev_set_mac_address(slave->dev, addr);
4230                 if (res) {
4231                         /* TODO: consider downing the slave
4232                          * and retry ?
4233                          * User should expect communications
4234                          * breakage anyway until ARP finish
4235                          * updating, so...
4236                          */
4237                         dprintk("err %d %s\n", res, slave->dev->name);
4238                         goto unwind;
4239                 }
4240         }
4241
4242         /* success */
4243         memcpy(bond_dev->dev_addr, sa->sa_data, bond_dev->addr_len);
4244         return 0;
4245
4246 unwind:
4247         memcpy(tmp_sa.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
4248         tmp_sa.sa_family = bond_dev->type;
4249
4250         /* unwind from head to the slave that failed */
4251         stop_at = slave;
4252         bond_for_each_slave_from_to(bond, slave, i, bond->first_slave, stop_at) {
4253                 int tmp_res;
4254
4255                 tmp_res = dev_set_mac_address(slave->dev, &tmp_sa);
4256                 if (tmp_res) {
4257                         dprintk("unwind err %d dev %s\n", tmp_res,
4258                                 slave->dev->name);
4259                 }
4260         }
4261
4262         return res;
4263 }
4264
4265 static int bond_xmit_roundrobin(struct sk_buff *skb, struct net_device *bond_dev)
4266 {
4267         struct bonding *bond = bond_dev->priv;
4268         struct slave *slave, *start_at;
4269         int i, slave_no, res = 1;
4270
4271         read_lock(&bond->lock);
4272
4273         if (!BOND_IS_OK(bond)) {
4274                 goto out;
4275         }
4276
4277         /*
4278          * Concurrent TX may collide on rr_tx_counter; we accept that
4279          * as being rare enough not to justify using an atomic op here
4280          */
4281         slave_no = bond->rr_tx_counter++ % bond->slave_cnt;
4282
4283         bond_for_each_slave(bond, slave, i) {
4284                 slave_no--;
4285                 if (slave_no < 0) {
4286                         break;
4287                 }
4288         }
4289
4290         start_at = slave;
4291         bond_for_each_slave_from(bond, slave, i, start_at) {
4292                 if (IS_UP(slave->dev) &&
4293                     (slave->link == BOND_LINK_UP) &&
4294                     (slave->state == BOND_STATE_ACTIVE)) {
4295                         res = bond_dev_queue_xmit(bond, skb, slave->dev);
4296                         break;
4297                 }
4298         }
4299
4300 out:
4301         if (res) {
4302                 /* no suitable interface, frame not sent */
4303                 dev_kfree_skb(skb);
4304         }
4305         read_unlock(&bond->lock);
4306         return 0;
4307 }
4308
4309
4310 /*
4311  * in active-backup mode, we know that bond->curr_active_slave is always valid if
4312  * the bond has a usable interface.
4313  */
4314 static int bond_xmit_activebackup(struct sk_buff *skb, struct net_device *bond_dev)
4315 {
4316         struct bonding *bond = bond_dev->priv;
4317         int res = 1;
4318
4319         read_lock(&bond->lock);
4320         read_lock(&bond->curr_slave_lock);
4321
4322         if (!BOND_IS_OK(bond)) {
4323                 goto out;
4324         }
4325
4326         if (!bond->curr_active_slave)
4327                 goto out;
4328
4329         res = bond_dev_queue_xmit(bond, skb, bond->curr_active_slave->dev);
4330
4331 out:
4332         if (res) {
4333                 /* no suitable interface, frame not sent */
4334                 dev_kfree_skb(skb);
4335         }
4336         read_unlock(&bond->curr_slave_lock);
4337         read_unlock(&bond->lock);
4338         return 0;
4339 }
4340
4341 /*
4342  * In bond_xmit_xor() , we determine the output device by using a pre-
4343  * determined xmit_hash_policy(), If the selected device is not enabled,
4344  * find the next active slave.
4345  */
4346 static int bond_xmit_xor(struct sk_buff *skb, struct net_device *bond_dev)
4347 {
4348         struct bonding *bond = bond_dev->priv;
4349         struct slave *slave, *start_at;
4350         int slave_no;
4351         int i;
4352         int res = 1;
4353
4354         read_lock(&bond->lock);
4355
4356         if (!BOND_IS_OK(bond)) {
4357                 goto out;
4358         }
4359
4360         slave_no = bond->xmit_hash_policy(skb, bond_dev, bond->slave_cnt);
4361
4362         bond_for_each_slave(bond, slave, i) {
4363                 slave_no--;
4364                 if (slave_no < 0) {
4365                         break;
4366                 }
4367         }
4368
4369         start_at = slave;
4370
4371         bond_for_each_slave_from(bond, slave, i, start_at) {
4372                 if (IS_UP(slave->dev) &&
4373                     (slave->link == BOND_LINK_UP) &&
4374                     (slave->state == BOND_STATE_ACTIVE)) {
4375                         res = bond_dev_queue_xmit(bond, skb, slave->dev);
4376                         break;
4377                 }
4378         }
4379
4380 out:
4381         if (res) {
4382                 /* no suitable interface, frame not sent */
4383                 dev_kfree_skb(skb);
4384         }
4385         read_unlock(&bond->lock);
4386         return 0;
4387 }
4388
4389 /*
4390  * in broadcast mode, we send everything to all usable interfaces.
4391  */
4392 static int bond_xmit_broadcast(struct sk_buff *skb, struct net_device *bond_dev)
4393 {
4394         struct bonding *bond = bond_dev->priv;
4395         struct slave *slave, *start_at;
4396         struct net_device *tx_dev = NULL;
4397         int i;
4398         int res = 1;
4399
4400         read_lock(&bond->lock);
4401
4402         if (!BOND_IS_OK(bond)) {
4403                 goto out;
4404         }
4405
4406         read_lock(&bond->curr_slave_lock);
4407         start_at = bond->curr_active_slave;
4408         read_unlock(&bond->curr_slave_lock);
4409
4410         if (!start_at) {
4411                 goto out;
4412         }
4413
4414         bond_for_each_slave_from(bond, slave, i, start_at) {
4415                 if (IS_UP(slave->dev) &&
4416                     (slave->link == BOND_LINK_UP) &&
4417                     (slave->state == BOND_STATE_ACTIVE)) {
4418                         if (tx_dev) {
4419                                 struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
4420                                 if (!skb2) {
4421                                         printk(KERN_ERR DRV_NAME
4422                                                ": %s: Error: bond_xmit_broadcast(): "
4423                                                "skb_clone() failed\n",
4424                                                bond_dev->name);
4425                                         continue;
4426                                 }
4427
4428                                 res = bond_dev_queue_xmit(bond, skb2, tx_dev);
4429                                 if (res) {
4430                                         dev_kfree_skb(skb2);
4431                                         continue;
4432                                 }
4433                         }
4434                         tx_dev = slave->dev;
4435                 }
4436         }
4437
4438         if (tx_dev) {
4439                 res = bond_dev_queue_xmit(bond, skb, tx_dev);
4440         }
4441
4442 out:
4443         if (res) {
4444                 /* no suitable interface, frame not sent */
4445                 dev_kfree_skb(skb);
4446         }
4447         /* frame sent to all suitable interfaces */
4448         read_unlock(&bond->lock);
4449         return 0;
4450 }
4451
4452 /*------------------------- Device initialization ---------------------------*/
4453
4454 static void bond_set_xmit_hash_policy(struct bonding *bond)
4455 {
4456         switch (bond->params.xmit_policy) {
4457         case BOND_XMIT_POLICY_LAYER23:
4458                 bond->xmit_hash_policy = bond_xmit_hash_policy_l23;
4459                 break;
4460         case BOND_XMIT_POLICY_LAYER34:
4461                 bond->xmit_hash_policy = bond_xmit_hash_policy_l34;
4462                 break;
4463         case BOND_XMIT_POLICY_LAYER2:
4464         default:
4465                 bond->xmit_hash_policy = bond_xmit_hash_policy_l2;
4466                 break;
4467         }
4468 }
4469
4470 /*
4471  * set bond mode specific net device operations
4472  */
4473 void bond_set_mode_ops(struct bonding *bond, int mode)
4474 {
4475         struct net_device *bond_dev = bond->dev;
4476
4477         switch (mode) {
4478         case BOND_MODE_ROUNDROBIN:
4479                 bond_dev->hard_start_xmit = bond_xmit_roundrobin;
4480                 break;
4481         case BOND_MODE_ACTIVEBACKUP:
4482                 bond_dev->hard_start_xmit = bond_xmit_activebackup;
4483                 break;
4484         case BOND_MODE_XOR:
4485                 bond_dev->hard_start_xmit = bond_xmit_xor;
4486                 bond_set_xmit_hash_policy(bond);
4487                 break;
4488         case BOND_MODE_BROADCAST:
4489                 bond_dev->hard_start_xmit = bond_xmit_broadcast;
4490                 break;
4491         case BOND_MODE_8023AD:
4492                 bond_set_master_3ad_flags(bond);
4493                 bond_dev->hard_start_xmit = bond_3ad_xmit_xor;
4494                 bond_set_xmit_hash_policy(bond);
4495                 break;
4496         case BOND_MODE_ALB:
4497                 bond_set_master_alb_flags(bond);
4498                 /* FALLTHRU */
4499         case BOND_MODE_TLB:
4500                 bond_dev->hard_start_xmit = bond_alb_xmit;
4501                 bond_dev->set_mac_address = bond_alb_set_mac_address;
4502                 break;
4503         default:
4504                 /* Should never happen, mode already checked */
4505                 printk(KERN_ERR DRV_NAME
4506                        ": %s: Error: Unknown bonding mode %d\n",
4507                        bond_dev->name,
4508                        mode);
4509                 break;
4510         }
4511 }
4512
4513 static void bond_ethtool_get_drvinfo(struct net_device *bond_dev,
4514                                     struct ethtool_drvinfo *drvinfo)
4515 {
4516         strncpy(drvinfo->driver, DRV_NAME, 32);
4517         strncpy(drvinfo->version, DRV_VERSION, 32);
4518         snprintf(drvinfo->fw_version, 32, "%d", BOND_ABI_VERSION);
4519 }
4520
4521 static const struct ethtool_ops bond_ethtool_ops = {
4522         .get_drvinfo            = bond_ethtool_get_drvinfo,
4523 };
4524
4525 /*
4526  * Does not allocate but creates a /proc entry.
4527  * Allowed to fail.
4528  */
4529 static int bond_init(struct net_device *bond_dev, struct bond_params *params)
4530 {
4531         struct bonding *bond = bond_dev->priv;
4532
4533         dprintk("Begin bond_init for %s\n", bond_dev->name);
4534
4535         /* initialize rwlocks */
4536         rwlock_init(&bond->lock);
4537         rwlock_init(&bond->curr_slave_lock);
4538
4539         bond->params = *params; /* copy params struct */
4540
4541         bond->wq = create_singlethread_workqueue(bond_dev->name);
4542         if (!bond->wq)
4543                 return -ENOMEM;
4544
4545         /* Initialize pointers */
4546         bond->first_slave = NULL;
4547         bond->curr_active_slave = NULL;
4548         bond->current_arp_slave = NULL;
4549         bond->primary_slave = NULL;
4550         bond->dev = bond_dev;
4551         bond->send_grat_arp = 0;
4552         bond->setup_by_slave = 0;
4553         INIT_LIST_HEAD(&bond->vlan_list);
4554
4555         /* Initialize the device entry points */
4556         bond_dev->open = bond_open;
4557         bond_dev->stop = bond_close;
4558         bond_dev->get_stats = bond_get_stats;
4559         bond_dev->do_ioctl = bond_do_ioctl;
4560         bond_dev->ethtool_ops = &bond_ethtool_ops;
4561         bond_dev->set_multicast_list = bond_set_multicast_list;
4562         bond_dev->change_mtu = bond_change_mtu;
4563         bond_dev->set_mac_address = bond_set_mac_address;
4564         bond_dev->validate_addr = NULL;
4565
4566         bond_set_mode_ops(bond, bond->params.mode);
4567
4568         bond_dev->destructor = free_netdev;
4569
4570         /* Initialize the device options */
4571         bond_dev->tx_queue_len = 0;
4572         bond_dev->flags |= IFF_MASTER|IFF_MULTICAST;
4573         bond_dev->priv_flags |= IFF_BONDING;
4574
4575         /* At first, we block adding VLANs. That's the only way to
4576          * prevent problems that occur when adding VLANs over an
4577          * empty bond. The block will be removed once non-challenged
4578          * slaves are enslaved.
4579          */
4580         bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
4581
4582         /* don't acquire bond device's netif_tx_lock when
4583          * transmitting */
4584         bond_dev->features |= NETIF_F_LLTX;
4585
4586         /* By default, we declare the bond to be fully
4587          * VLAN hardware accelerated capable. Special
4588          * care is taken in the various xmit functions
4589          * when there are slaves that are not hw accel
4590          * capable
4591          */
4592         bond_dev->vlan_rx_register = bond_vlan_rx_register;
4593         bond_dev->vlan_rx_add_vid  = bond_vlan_rx_add_vid;
4594         bond_dev->vlan_rx_kill_vid = bond_vlan_rx_kill_vid;
4595         bond_dev->features |= (NETIF_F_HW_VLAN_TX |
4596                                NETIF_F_HW_VLAN_RX |
4597                                NETIF_F_HW_VLAN_FILTER);
4598
4599 #ifdef CONFIG_PROC_FS
4600         bond_create_proc_entry(bond);
4601 #endif
4602         list_add_tail(&bond->bond_list, &bond_dev_list);
4603
4604         return 0;
4605 }
4606
4607 /* De-initialize device specific data.
4608  * Caller must hold rtnl_lock.
4609  */
4610 static void bond_deinit(struct net_device *bond_dev)
4611 {
4612         struct bonding *bond = bond_dev->priv;
4613
4614         list_del(&bond->bond_list);
4615
4616 #ifdef CONFIG_PROC_FS
4617         bond_remove_proc_entry(bond);
4618 #endif
4619 }
4620
4621 static void bond_work_cancel_all(struct bonding *bond)
4622 {
4623         write_lock_bh(&bond->lock);
4624         bond->kill_timers = 1;
4625         write_unlock_bh(&bond->lock);
4626
4627         if (bond->params.miimon && delayed_work_pending(&bond->mii_work))
4628                 cancel_delayed_work(&bond->mii_work);
4629
4630         if (bond->params.arp_interval && delayed_work_pending(&bond->arp_work))
4631                 cancel_delayed_work(&bond->arp_work);
4632
4633         if (bond->params.mode == BOND_MODE_ALB &&
4634             delayed_work_pending(&bond->alb_work))
4635                 cancel_delayed_work(&bond->alb_work);
4636
4637         if (bond->params.mode == BOND_MODE_8023AD &&
4638             delayed_work_pending(&bond->ad_work))
4639                 cancel_delayed_work(&bond->ad_work);
4640 }
4641
4642 /* Unregister and free all bond devices.
4643  * Caller must hold rtnl_lock.
4644  */
4645 static void bond_free_all(void)
4646 {
4647         struct bonding *bond, *nxt;
4648
4649         list_for_each_entry_safe(bond, nxt, &bond_dev_list, bond_list) {
4650                 struct net_device *bond_dev = bond->dev;
4651
4652                 bond_work_cancel_all(bond);
4653                 netif_tx_lock_bh(bond_dev);
4654                 bond_mc_list_destroy(bond);
4655                 netif_tx_unlock_bh(bond_dev);
4656                 /* Release the bonded slaves */
4657                 bond_release_all(bond_dev);
4658                 bond_destroy(bond);
4659         }
4660
4661 #ifdef CONFIG_PROC_FS
4662         bond_destroy_proc_dir();
4663 #endif
4664 }
4665
4666 /*------------------------- Module initialization ---------------------------*/
4667
4668 /*
4669  * Convert string input module parms.  Accept either the
4670  * number of the mode or its string name.  A bit complicated because
4671  * some mode names are substrings of other names, and calls from sysfs
4672  * may have whitespace in the name (trailing newlines, for example).
4673  */
4674 int bond_parse_parm(const char *buf, struct bond_parm_tbl *tbl)
4675 {
4676         int mode = -1, i, rv;
4677         char *p, modestr[BOND_MAX_MODENAME_LEN + 1] = { 0, };
4678
4679         for (p = (char *)buf; *p; p++)
4680                 if (!(isdigit(*p) || isspace(*p)))
4681                         break;
4682
4683         if (*p)
4684                 rv = sscanf(buf, "%20s", modestr);
4685         else
4686                 rv = sscanf(buf, "%d", &mode);
4687
4688         if (!rv)
4689                 return -1;
4690
4691         for (i = 0; tbl[i].modename; i++) {
4692                 if (mode == tbl[i].mode)
4693                         return tbl[i].mode;
4694                 if (strcmp(modestr, tbl[i].modename) == 0)
4695                         return tbl[i].mode;
4696         }
4697
4698         return -1;
4699 }
4700
4701 static int bond_check_params(struct bond_params *params)
4702 {
4703         int arp_validate_value, fail_over_mac_value;
4704
4705         /*
4706          * Convert string parameters.
4707          */
4708         if (mode) {
4709                 bond_mode = bond_parse_parm(mode, bond_mode_tbl);
4710                 if (bond_mode == -1) {
4711                         printk(KERN_ERR DRV_NAME
4712                                ": Error: Invalid bonding mode \"%s\"\n",
4713                                mode == NULL ? "NULL" : mode);
4714                         return -EINVAL;
4715                 }
4716         }
4717
4718         if (xmit_hash_policy) {
4719                 if ((bond_mode != BOND_MODE_XOR) &&
4720                     (bond_mode != BOND_MODE_8023AD)) {
4721                         printk(KERN_INFO DRV_NAME
4722                                ": xor_mode param is irrelevant in mode %s\n",
4723                                bond_mode_name(bond_mode));
4724                 } else {
4725                         xmit_hashtype = bond_parse_parm(xmit_hash_policy,
4726                                                         xmit_hashtype_tbl);
4727                         if (xmit_hashtype == -1) {
4728                                 printk(KERN_ERR DRV_NAME
4729                                 ": Error: Invalid xmit_hash_policy \"%s\"\n",
4730                                 xmit_hash_policy == NULL ? "NULL" :
4731                                        xmit_hash_policy);
4732                                 return -EINVAL;
4733                         }
4734                 }
4735         }
4736
4737         if (lacp_rate) {
4738                 if (bond_mode != BOND_MODE_8023AD) {
4739                         printk(KERN_INFO DRV_NAME
4740                                ": lacp_rate param is irrelevant in mode %s\n",
4741                                bond_mode_name(bond_mode));
4742                 } else {
4743                         lacp_fast = bond_parse_parm(lacp_rate, bond_lacp_tbl);
4744                         if (lacp_fast == -1) {
4745                                 printk(KERN_ERR DRV_NAME
4746                                        ": Error: Invalid lacp rate \"%s\"\n",
4747                                        lacp_rate == NULL ? "NULL" : lacp_rate);
4748                                 return -EINVAL;
4749                         }
4750                 }
4751         }
4752
4753         if (max_bonds < 0 || max_bonds > INT_MAX) {
4754                 printk(KERN_WARNING DRV_NAME
4755                        ": Warning: max_bonds (%d) not in range %d-%d, so it "
4756                        "was reset to BOND_DEFAULT_MAX_BONDS (%d)\n",
4757                        max_bonds, 0, INT_MAX, BOND_DEFAULT_MAX_BONDS);
4758                 max_bonds = BOND_DEFAULT_MAX_BONDS;
4759         }
4760
4761         if (miimon < 0) {
4762                 printk(KERN_WARNING DRV_NAME
4763                        ": Warning: miimon module parameter (%d), "
4764                        "not in range 0-%d, so it was reset to %d\n",
4765                        miimon, INT_MAX, BOND_LINK_MON_INTERV);
4766                 miimon = BOND_LINK_MON_INTERV;
4767         }
4768
4769         if (updelay < 0) {
4770                 printk(KERN_WARNING DRV_NAME
4771                        ": Warning: updelay module parameter (%d), "
4772                        "not in range 0-%d, so it was reset to 0\n",
4773                        updelay, INT_MAX);
4774                 updelay = 0;
4775         }
4776
4777         if (downdelay < 0) {
4778                 printk(KERN_WARNING DRV_NAME
4779                        ": Warning: downdelay module parameter (%d), "
4780                        "not in range 0-%d, so it was reset to 0\n",
4781                        downdelay, INT_MAX);
4782                 downdelay = 0;
4783         }
4784
4785         if ((use_carrier != 0) && (use_carrier != 1)) {
4786                 printk(KERN_WARNING DRV_NAME
4787                        ": Warning: use_carrier module parameter (%d), "
4788                        "not of valid value (0/1), so it was set to 1\n",
4789                        use_carrier);
4790                 use_carrier = 1;
4791         }
4792
4793         if (num_grat_arp < 0 || num_grat_arp > 255) {
4794                 printk(KERN_WARNING DRV_NAME
4795                        ": Warning: num_grat_arp (%d) not in range 0-255 so it "
4796                        "was reset to 1 \n", num_grat_arp);
4797                 num_grat_arp = 1;
4798         }
4799
4800         /* reset values for 802.3ad */
4801         if (bond_mode == BOND_MODE_8023AD) {
4802                 if (!miimon) {
4803                         printk(KERN_WARNING DRV_NAME
4804                                ": Warning: miimon must be specified, "
4805                                "otherwise bonding will not detect link "
4806                                "failure, speed and duplex which are "
4807                                "essential for 802.3ad operation\n");
4808                         printk(KERN_WARNING "Forcing miimon to 100msec\n");
4809                         miimon = 100;
4810                 }
4811         }
4812
4813         /* reset values for TLB/ALB */
4814         if ((bond_mode == BOND_MODE_TLB) ||
4815             (bond_mode == BOND_MODE_ALB)) {
4816                 if (!miimon) {
4817                         printk(KERN_WARNING DRV_NAME
4818                                ": Warning: miimon must be specified, "
4819                                "otherwise bonding will not detect link "
4820                                "failure and link speed which are essential "
4821                                "for TLB/ALB load balancing\n");
4822                         printk(KERN_WARNING "Forcing miimon to 100msec\n");
4823                         miimon = 100;
4824                 }
4825         }
4826
4827         if (bond_mode == BOND_MODE_ALB) {
4828                 printk(KERN_NOTICE DRV_NAME
4829                        ": In ALB mode you might experience client "
4830                        "disconnections upon reconnection of a link if the "
4831                        "bonding module updelay parameter (%d msec) is "
4832                        "incompatible with the forwarding delay time of the "
4833                        "switch\n",
4834                        updelay);
4835         }
4836
4837         if (!miimon) {
4838                 if (updelay || downdelay) {
4839                         /* just warn the user the up/down delay will have
4840                          * no effect since miimon is zero...
4841                          */
4842                         printk(KERN_WARNING DRV_NAME
4843                                ": Warning: miimon module parameter not set "
4844                                "and updelay (%d) or downdelay (%d) module "
4845                                "parameter is set; updelay and downdelay have "
4846                                "no effect unless miimon is set\n",
4847                                updelay, downdelay);
4848                 }
4849         } else {
4850                 /* don't allow arp monitoring */
4851                 if (arp_interval) {
4852                         printk(KERN_WARNING DRV_NAME
4853                                ": Warning: miimon (%d) and arp_interval (%d) "
4854                                "can't be used simultaneously, disabling ARP "
4855                                "monitoring\n",
4856                                miimon, arp_interval);
4857                         arp_interval = 0;
4858                 }
4859
4860                 if ((updelay % miimon) != 0) {
4861                         printk(KERN_WARNING DRV_NAME
4862                                ": Warning: updelay (%d) is not a multiple "
4863                                "of miimon (%d), updelay rounded to %d ms\n",
4864                                updelay, miimon, (updelay / miimon) * miimon);
4865                 }
4866
4867                 updelay /= miimon;
4868
4869                 if ((downdelay % miimon) != 0) {
4870                         printk(KERN_WARNING DRV_NAME
4871                                ": Warning: downdelay (%d) is not a multiple "
4872                                "of miimon (%d), downdelay rounded to %d ms\n",
4873                                downdelay, miimon,
4874                                (downdelay / miimon) * miimon);
4875                 }
4876
4877                 downdelay /= miimon;
4878         }
4879
4880         if (arp_interval < 0) {
4881                 printk(KERN_WARNING DRV_NAME
4882                        ": Warning: arp_interval module parameter (%d) "
4883                        ", not in range 0-%d, so it was reset to %d\n",
4884                        arp_interval, INT_MAX, BOND_LINK_ARP_INTERV);
4885                 arp_interval = BOND_LINK_ARP_INTERV;
4886         }
4887
4888         for (arp_ip_count = 0;
4889              (arp_ip_count < BOND_MAX_ARP_TARGETS) && arp_ip_target[arp_ip_count];
4890              arp_ip_count++) {
4891                 /* not complete check, but should be good enough to
4892                    catch mistakes */
4893                 if (!isdigit(arp_ip_target[arp_ip_count][0])) {
4894                         printk(KERN_WARNING DRV_NAME
4895                                ": Warning: bad arp_ip_target module parameter "
4896                                "(%s), ARP monitoring will not be performed\n",
4897                                arp_ip_target[arp_ip_count]);
4898                         arp_interval = 0;
4899                 } else {
4900                         __be32 ip = in_aton(arp_ip_target[arp_ip_count]);
4901                         arp_target[arp_ip_count] = ip;
4902                 }
4903         }
4904
4905         if (arp_interval && !arp_ip_count) {
4906                 /* don't allow arping if no arp_ip_target given... */
4907                 printk(KERN_WARNING DRV_NAME
4908                        ": Warning: arp_interval module parameter (%d) "
4909                        "specified without providing an arp_ip_target "
4910                        "parameter, arp_interval was reset to 0\n",
4911                        arp_interval);
4912                 arp_interval = 0;
4913         }
4914
4915         if (arp_validate) {
4916                 if (bond_mode != BOND_MODE_ACTIVEBACKUP) {
4917                         printk(KERN_ERR DRV_NAME
4918                ": arp_validate only supported in active-backup mode\n");
4919                         return -EINVAL;
4920                 }
4921                 if (!arp_interval) {
4922                         printk(KERN_ERR DRV_NAME
4923                                ": arp_validate requires arp_interval\n");
4924                         return -EINVAL;
4925                 }
4926
4927                 arp_validate_value = bond_parse_parm(arp_validate,
4928                                                      arp_validate_tbl);
4929                 if (arp_validate_value == -1) {
4930                         printk(KERN_ERR DRV_NAME
4931                                ": Error: invalid arp_validate \"%s\"\n",
4932                                arp_validate == NULL ? "NULL" : arp_validate);
4933                         return -EINVAL;
4934                 }
4935         } else
4936                 arp_validate_value = 0;
4937
4938         if (miimon) {
4939                 printk(KERN_INFO DRV_NAME
4940                        ": MII link monitoring set to %d ms\n",
4941                        miimon);
4942         } else if (arp_interval) {
4943                 int i;
4944
4945                 printk(KERN_INFO DRV_NAME
4946                        ": ARP monitoring set to %d ms, validate %s, with %d target(s):",
4947                        arp_interval,
4948                        arp_validate_tbl[arp_validate_value].modename,
4949                        arp_ip_count);
4950
4951                 for (i = 0; i < arp_ip_count; i++)
4952                         printk (" %s", arp_ip_target[i]);
4953
4954                 printk("\n");
4955
4956         } else if (max_bonds) {
4957                 /* miimon and arp_interval not set, we need one so things
4958                  * work as expected, see bonding.txt for details
4959                  */
4960                 printk(KERN_WARNING DRV_NAME
4961                        ": Warning: either miimon or arp_interval and "
4962                        "arp_ip_target module parameters must be specified, "
4963                        "otherwise bonding will not detect link failures! see "
4964                        "bonding.txt for details.\n");
4965         }
4966
4967         if (primary && !USES_PRIMARY(bond_mode)) {
4968                 /* currently, using a primary only makes sense
4969                  * in active backup, TLB or ALB modes
4970                  */
4971                 printk(KERN_WARNING DRV_NAME
4972                        ": Warning: %s primary device specified but has no "
4973                        "effect in %s mode\n",
4974                        primary, bond_mode_name(bond_mode));
4975                 primary = NULL;
4976         }
4977
4978         if (fail_over_mac) {
4979                 fail_over_mac_value = bond_parse_parm(fail_over_mac,
4980                                                       fail_over_mac_tbl);
4981                 if (fail_over_mac_value == -1) {
4982                         printk(KERN_ERR DRV_NAME
4983                                ": Error: invalid fail_over_mac \"%s\"\n",
4984                                arp_validate == NULL ? "NULL" : arp_validate);
4985                         return -EINVAL;
4986                 }
4987
4988                 if (bond_mode != BOND_MODE_ACTIVEBACKUP)
4989                         printk(KERN_WARNING DRV_NAME
4990                                ": Warning: fail_over_mac only affects "
4991                                "active-backup mode.\n");
4992         } else {
4993                 fail_over_mac_value = BOND_FOM_NONE;
4994         }
4995
4996         /* fill params struct with the proper values */
4997         params->mode = bond_mode;
4998         params->xmit_policy = xmit_hashtype;
4999         params->miimon = miimon;
5000         params->num_grat_arp = num_grat_arp;
5001         params->arp_interval = arp_interval;
5002         params->arp_validate = arp_validate_value;
5003         params->updelay = updelay;
5004         params->downdelay = downdelay;
5005         params->use_carrier = use_carrier;
5006         params->lacp_fast = lacp_fast;
5007         params->primary[0] = 0;
5008         params->fail_over_mac = fail_over_mac_value;
5009
5010         if (primary) {
5011                 strncpy(params->primary, primary, IFNAMSIZ);
5012                 params->primary[IFNAMSIZ - 1] = 0;
5013         }
5014
5015         memcpy(params->arp_targets, arp_target, sizeof(arp_target));
5016
5017         return 0;
5018 }
5019
5020 static struct lock_class_key bonding_netdev_xmit_lock_key;
5021
5022 /* Create a new bond based on the specified name and bonding parameters.
5023  * If name is NULL, obtain a suitable "bond%d" name for us.
5024  * Caller must NOT hold rtnl_lock; we need to release it here before we
5025  * set up our sysfs entries.
5026  */
5027 int bond_create(char *name, struct bond_params *params)
5028 {
5029         struct net_device *bond_dev;
5030         struct bonding *bond;
5031         int res;
5032
5033         rtnl_lock();
5034         down_write(&bonding_rwsem);
5035
5036         /* Check to see if the bond already exists. */
5037         if (name) {
5038                 list_for_each_entry(bond, &bond_dev_list, bond_list)
5039                         if (strnicmp(bond->dev->name, name, IFNAMSIZ) == 0) {
5040                                 printk(KERN_ERR DRV_NAME
5041                                ": cannot add bond %s; it already exists\n",
5042                                        name);
5043                                 res = -EPERM;
5044                                 goto out_rtnl;
5045                         }
5046         }
5047
5048         bond_dev = alloc_netdev(sizeof(struct bonding), name ? name : "",
5049                                 ether_setup);
5050         if (!bond_dev) {
5051                 printk(KERN_ERR DRV_NAME
5052                        ": %s: eek! can't alloc netdev!\n",
5053                        name);
5054                 res = -ENOMEM;
5055                 goto out_rtnl;
5056         }
5057
5058         if (!name) {
5059                 res = dev_alloc_name(bond_dev, "bond%d");
5060                 if (res < 0)
5061                         goto out_netdev;
5062         }
5063
5064         /* bond_init() must be called after dev_alloc_name() (for the
5065          * /proc files), but before register_netdevice(), because we
5066          * need to set function pointers.
5067          */
5068
5069         res = bond_init(bond_dev, params);
5070         if (res < 0) {
5071                 goto out_netdev;
5072         }
5073
5074         res = register_netdevice(bond_dev);
5075         if (res < 0) {
5076                 goto out_bond;
5077         }
5078
5079         lockdep_set_class(&bond_dev->_xmit_lock, &bonding_netdev_xmit_lock_key);
5080
5081         netif_carrier_off(bond_dev);
5082
5083         up_write(&bonding_rwsem);
5084         rtnl_unlock(); /* allows sysfs registration of net device */
5085         res = bond_create_sysfs_entry(bond_dev->priv);
5086         if (res < 0) {
5087                 rtnl_lock();
5088                 down_write(&bonding_rwsem);
5089                 bond_deinit(bond_dev);
5090                 unregister_netdevice(bond_dev);
5091                 goto out_rtnl;
5092         }
5093
5094         return 0;
5095
5096 out_bond:
5097         bond_deinit(bond_dev);
5098 out_netdev:
5099         free_netdev(bond_dev);
5100 out_rtnl:
5101         up_write(&bonding_rwsem);
5102         rtnl_unlock();
5103         return res;
5104 }
5105
5106 static int __init bonding_init(void)
5107 {
5108         int i;
5109         int res;
5110         struct bonding *bond;
5111
5112         printk(KERN_INFO "%s", version);
5113
5114         res = bond_check_params(&bonding_defaults);
5115         if (res) {
5116                 goto out;
5117         }
5118
5119 #ifdef CONFIG_PROC_FS
5120         bond_create_proc_dir();
5121 #endif
5122
5123         init_rwsem(&bonding_rwsem);
5124
5125         for (i = 0; i < max_bonds; i++) {
5126                 res = bond_create(NULL, &bonding_defaults);
5127                 if (res)
5128                         goto err;
5129         }
5130
5131         res = bond_create_sysfs();
5132         if (res)
5133                 goto err;
5134
5135         register_netdevice_notifier(&bond_netdev_notifier);
5136         register_inetaddr_notifier(&bond_inetaddr_notifier);
5137
5138         goto out;
5139 err:
5140         list_for_each_entry(bond, &bond_dev_list, bond_list) {
5141                 bond_work_cancel_all(bond);
5142                 destroy_workqueue(bond->wq);
5143         }
5144
5145         bond_destroy_sysfs();
5146
5147         rtnl_lock();
5148         bond_free_all();
5149         rtnl_unlock();
5150 out:
5151         return res;
5152
5153 }
5154
5155 static void __exit bonding_exit(void)
5156 {
5157         unregister_netdevice_notifier(&bond_netdev_notifier);
5158         unregister_inetaddr_notifier(&bond_inetaddr_notifier);
5159
5160         bond_destroy_sysfs();
5161
5162         rtnl_lock();
5163         bond_free_all();
5164         rtnl_unlock();
5165 }
5166
5167 module_init(bonding_init);
5168 module_exit(bonding_exit);
5169 MODULE_LICENSE("GPL");
5170 MODULE_VERSION(DRV_VERSION);
5171 MODULE_DESCRIPTION(DRV_DESCRIPTION ", v" DRV_VERSION);
5172 MODULE_AUTHOR("Thomas Davis, tadavis@lbl.gov and many others");
5173 MODULE_SUPPORTED_DEVICE("most ethernet devices");
5174
5175 /*
5176  * Local variables:
5177  *  c-indent-level: 8
5178  *  c-basic-offset: 8
5179  *  tab-width: 8
5180  * End:
5181  */
5182