net: convert multicast list to list_head
[linux-2.6.git] / drivers / net / bonding / bond_main.c
1 /*
2  * originally based on the dummy device.
3  *
4  * Copyright 1999, Thomas Davis, tadavis@lbl.gov.
5  * Licensed under the GPL. Based on dummy.c, and eql.c devices.
6  *
7  * bonding.c: an Ethernet Bonding driver
8  *
9  * This is useful to talk to a Cisco EtherChannel compatible equipment:
10  *      Cisco 5500
11  *      Sun Trunking (Solaris)
12  *      Alteon AceDirector Trunks
13  *      Linux Bonding
14  *      and probably many L2 switches ...
15  *
16  * How it works:
17  *    ifconfig bond0 ipaddress netmask up
18  *      will setup a network device, with an ip address.  No mac address
19  *      will be assigned at this time.  The hw mac address will come from
20  *      the first slave bonded to the channel.  All slaves will then use
21  *      this hw mac address.
22  *
23  *    ifconfig bond0 down
24  *         will release all slaves, marking them as down.
25  *
26  *    ifenslave bond0 eth0
27  *      will attach eth0 to bond0 as a slave.  eth0 hw mac address will either
28  *      a: be used as initial mac address
29  *      b: if a hw mac address already is there, eth0's hw mac address
30  *         will then be set from bond0.
31  *
32  */
33
34 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
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 <linux/io.h>
57 #include <asm/system.h>
58 #include <asm/dma.h>
59 #include <linux/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 <net/netns/generic.h>
81 #include "bonding.h"
82 #include "bond_3ad.h"
83 #include "bond_alb.h"
84
85 /*---------------------------- Module parameters ----------------------------*/
86
87 /* monitor all links that often (in milliseconds). <=0 disables monitoring */
88 #define BOND_LINK_MON_INTERV    0
89 #define BOND_LINK_ARP_INTERV    0
90
91 static int max_bonds    = BOND_DEFAULT_MAX_BONDS;
92 static int num_grat_arp = 1;
93 static int num_unsol_na = 1;
94 static int miimon       = BOND_LINK_MON_INTERV;
95 static int updelay;
96 static int downdelay;
97 static int use_carrier  = 1;
98 static char *mode;
99 static char *primary;
100 static char *primary_reselect;
101 static char *lacp_rate;
102 static char *ad_select;
103 static char *xmit_hash_policy;
104 static int arp_interval = BOND_LINK_ARP_INTERV;
105 static char *arp_ip_target[BOND_MAX_ARP_TARGETS];
106 static char *arp_validate;
107 static char *fail_over_mac;
108 static struct bond_params bonding_defaults;
109
110 module_param(max_bonds, int, 0);
111 MODULE_PARM_DESC(max_bonds, "Max number of bonded devices");
112 module_param(num_grat_arp, int, 0644);
113 MODULE_PARM_DESC(num_grat_arp, "Number of gratuitous ARP packets to send on failover event");
114 module_param(num_unsol_na, int, 0644);
115 MODULE_PARM_DESC(num_unsol_na, "Number of unsolicited IPv6 Neighbor Advertisements packets to send on failover event");
116 module_param(miimon, int, 0);
117 MODULE_PARM_DESC(miimon, "Link check interval in milliseconds");
118 module_param(updelay, int, 0);
119 MODULE_PARM_DESC(updelay, "Delay before considering link up, in milliseconds");
120 module_param(downdelay, int, 0);
121 MODULE_PARM_DESC(downdelay, "Delay before considering link down, "
122                             "in milliseconds");
123 module_param(use_carrier, int, 0);
124 MODULE_PARM_DESC(use_carrier, "Use netif_carrier_ok (vs MII ioctls) in miimon; "
125                               "0 for off, 1 for on (default)");
126 module_param(mode, charp, 0);
127 MODULE_PARM_DESC(mode, "Mode of operation : 0 for balance-rr, "
128                        "1 for active-backup, 2 for balance-xor, "
129                        "3 for broadcast, 4 for 802.3ad, 5 for balance-tlb, "
130                        "6 for balance-alb");
131 module_param(primary, charp, 0);
132 MODULE_PARM_DESC(primary, "Primary network device to use");
133 module_param(primary_reselect, charp, 0);
134 MODULE_PARM_DESC(primary_reselect, "Reselect primary slave "
135                                    "once it comes up; "
136                                    "0 for always (default), "
137                                    "1 for only if speed of primary is "
138                                    "better, "
139                                    "2 for only on active slave "
140                                    "failure");
141 module_param(lacp_rate, charp, 0);
142 MODULE_PARM_DESC(lacp_rate, "LACPDU tx rate to request from 802.3ad partner "
143                             "(slow/fast)");
144 module_param(ad_select, charp, 0);
145 MODULE_PARM_DESC(ad_select, "803.ad aggregation selection logic: stable (0, default), bandwidth (1), count (2)");
146 module_param(xmit_hash_policy, charp, 0);
147 MODULE_PARM_DESC(xmit_hash_policy, "XOR hashing method: 0 for layer 2 (default)"
148                                    ", 1 for layer 3+4");
149 module_param(arp_interval, int, 0);
150 MODULE_PARM_DESC(arp_interval, "arp interval in milliseconds");
151 module_param_array(arp_ip_target, charp, NULL, 0);
152 MODULE_PARM_DESC(arp_ip_target, "arp targets in n.n.n.n form");
153 module_param(arp_validate, charp, 0);
154 MODULE_PARM_DESC(arp_validate, "validate src/dst of ARP probes: none (default), active, backup or all");
155 module_param(fail_over_mac, charp, 0);
156 MODULE_PARM_DESC(fail_over_mac, "For active-backup, do not set all slaves to the same MAC.  none (default), active or follow");
157
158 /*----------------------------- Global variables ----------------------------*/
159
160 static const char * const version =
161         DRV_DESCRIPTION ": v" DRV_VERSION " (" DRV_RELDATE ")\n";
162
163 int bond_net_id __read_mostly;
164
165 static __be32 arp_target[BOND_MAX_ARP_TARGETS];
166 static int arp_ip_count;
167 static int bond_mode    = BOND_MODE_ROUNDROBIN;
168 static int xmit_hashtype = BOND_XMIT_POLICY_LAYER2;
169 static int lacp_fast;
170
171
172 const struct bond_parm_tbl bond_lacp_tbl[] = {
173 {       "slow",         AD_LACP_SLOW},
174 {       "fast",         AD_LACP_FAST},
175 {       NULL,           -1},
176 };
177
178 const struct bond_parm_tbl bond_mode_tbl[] = {
179 {       "balance-rr",           BOND_MODE_ROUNDROBIN},
180 {       "active-backup",        BOND_MODE_ACTIVEBACKUP},
181 {       "balance-xor",          BOND_MODE_XOR},
182 {       "broadcast",            BOND_MODE_BROADCAST},
183 {       "802.3ad",              BOND_MODE_8023AD},
184 {       "balance-tlb",          BOND_MODE_TLB},
185 {       "balance-alb",          BOND_MODE_ALB},
186 {       NULL,                   -1},
187 };
188
189 const struct bond_parm_tbl xmit_hashtype_tbl[] = {
190 {       "layer2",               BOND_XMIT_POLICY_LAYER2},
191 {       "layer3+4",             BOND_XMIT_POLICY_LAYER34},
192 {       "layer2+3",             BOND_XMIT_POLICY_LAYER23},
193 {       NULL,                   -1},
194 };
195
196 const struct bond_parm_tbl arp_validate_tbl[] = {
197 {       "none",                 BOND_ARP_VALIDATE_NONE},
198 {       "active",               BOND_ARP_VALIDATE_ACTIVE},
199 {       "backup",               BOND_ARP_VALIDATE_BACKUP},
200 {       "all",                  BOND_ARP_VALIDATE_ALL},
201 {       NULL,                   -1},
202 };
203
204 const struct bond_parm_tbl fail_over_mac_tbl[] = {
205 {       "none",                 BOND_FOM_NONE},
206 {       "active",               BOND_FOM_ACTIVE},
207 {       "follow",               BOND_FOM_FOLLOW},
208 {       NULL,                   -1},
209 };
210
211 const struct bond_parm_tbl pri_reselect_tbl[] = {
212 {       "always",               BOND_PRI_RESELECT_ALWAYS},
213 {       "better",               BOND_PRI_RESELECT_BETTER},
214 {       "failure",              BOND_PRI_RESELECT_FAILURE},
215 {       NULL,                   -1},
216 };
217
218 struct bond_parm_tbl ad_select_tbl[] = {
219 {       "stable",       BOND_AD_STABLE},
220 {       "bandwidth",    BOND_AD_BANDWIDTH},
221 {       "count",        BOND_AD_COUNT},
222 {       NULL,           -1},
223 };
224
225 /*-------------------------- Forward declarations ---------------------------*/
226
227 static void bond_send_gratuitous_arp(struct bonding *bond);
228 static int bond_init(struct net_device *bond_dev);
229 static void bond_uninit(struct net_device *bond_dev);
230
231 /*---------------------------- General routines -----------------------------*/
232
233 static const char *bond_mode_name(int mode)
234 {
235         static const char *names[] = {
236                 [BOND_MODE_ROUNDROBIN] = "load balancing (round-robin)",
237                 [BOND_MODE_ACTIVEBACKUP] = "fault-tolerance (active-backup)",
238                 [BOND_MODE_XOR] = "load balancing (xor)",
239                 [BOND_MODE_BROADCAST] = "fault-tolerance (broadcast)",
240                 [BOND_MODE_8023AD] = "IEEE 802.3ad Dynamic link aggregation",
241                 [BOND_MODE_TLB] = "transmit load balancing",
242                 [BOND_MODE_ALB] = "adaptive load balancing",
243         };
244
245         if (mode < 0 || mode > BOND_MODE_ALB)
246                 return "unknown";
247
248         return names[mode];
249 }
250
251 /*---------------------------------- VLAN -----------------------------------*/
252
253 /**
254  * bond_add_vlan - add a new vlan id on bond
255  * @bond: bond that got the notification
256  * @vlan_id: the vlan id to add
257  *
258  * Returns -ENOMEM if allocation failed.
259  */
260 static int bond_add_vlan(struct bonding *bond, unsigned short vlan_id)
261 {
262         struct vlan_entry *vlan;
263
264         pr_debug("bond: %s, vlan id %d\n",
265                  (bond ? bond->dev->name : "None"), vlan_id);
266
267         vlan = kzalloc(sizeof(struct vlan_entry), GFP_KERNEL);
268         if (!vlan)
269                 return -ENOMEM;
270
271         INIT_LIST_HEAD(&vlan->vlan_list);
272         vlan->vlan_id = vlan_id;
273
274         write_lock_bh(&bond->lock);
275
276         list_add_tail(&vlan->vlan_list, &bond->vlan_list);
277
278         write_unlock_bh(&bond->lock);
279
280         pr_debug("added VLAN ID %d on bond %s\n", vlan_id, bond->dev->name);
281
282         return 0;
283 }
284
285 /**
286  * bond_del_vlan - delete a vlan id from bond
287  * @bond: bond that got the notification
288  * @vlan_id: the vlan id to delete
289  *
290  * returns -ENODEV if @vlan_id was not found in @bond.
291  */
292 static int bond_del_vlan(struct bonding *bond, unsigned short vlan_id)
293 {
294         struct vlan_entry *vlan;
295         int res = -ENODEV;
296
297         pr_debug("bond: %s, vlan id %d\n", bond->dev->name, vlan_id);
298
299         write_lock_bh(&bond->lock);
300
301         list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
302                 if (vlan->vlan_id == vlan_id) {
303                         list_del(&vlan->vlan_list);
304
305                         if (bond_is_lb(bond))
306                                 bond_alb_clear_vlan(bond, vlan_id);
307
308                         pr_debug("removed VLAN ID %d from bond %s\n",
309                                  vlan_id, bond->dev->name);
310
311                         kfree(vlan);
312
313                         if (list_empty(&bond->vlan_list) &&
314                             (bond->slave_cnt == 0)) {
315                                 /* Last VLAN removed and no slaves, so
316                                  * restore block on adding VLANs. This will
317                                  * be removed once new slaves that are not
318                                  * VLAN challenged will be added.
319                                  */
320                                 bond->dev->features |= NETIF_F_VLAN_CHALLENGED;
321                         }
322
323                         res = 0;
324                         goto out;
325                 }
326         }
327
328         pr_debug("couldn't find VLAN ID %d in bond %s\n",
329                  vlan_id, bond->dev->name);
330
331 out:
332         write_unlock_bh(&bond->lock);
333         return res;
334 }
335
336 /**
337  * bond_has_challenged_slaves
338  * @bond: the bond we're working on
339  *
340  * Searches the slave list. Returns 1 if a vlan challenged slave
341  * was found, 0 otherwise.
342  *
343  * Assumes bond->lock is held.
344  */
345 static int bond_has_challenged_slaves(struct bonding *bond)
346 {
347         struct slave *slave;
348         int i;
349
350         bond_for_each_slave(bond, slave, i) {
351                 if (slave->dev->features & NETIF_F_VLAN_CHALLENGED) {
352                         pr_debug("found VLAN challenged slave - %s\n",
353                                  slave->dev->name);
354                         return 1;
355                 }
356         }
357
358         pr_debug("no VLAN challenged slaves found\n");
359         return 0;
360 }
361
362 /**
363  * bond_next_vlan - safely skip to the next item in the vlans list.
364  * @bond: the bond we're working on
365  * @curr: item we're advancing from
366  *
367  * Returns %NULL if list is empty, bond->next_vlan if @curr is %NULL,
368  * or @curr->next otherwise (even if it is @curr itself again).
369  *
370  * Caller must hold bond->lock
371  */
372 struct vlan_entry *bond_next_vlan(struct bonding *bond, struct vlan_entry *curr)
373 {
374         struct vlan_entry *next, *last;
375
376         if (list_empty(&bond->vlan_list))
377                 return NULL;
378
379         if (!curr) {
380                 next = list_entry(bond->vlan_list.next,
381                                   struct vlan_entry, vlan_list);
382         } else {
383                 last = list_entry(bond->vlan_list.prev,
384                                   struct vlan_entry, vlan_list);
385                 if (last == curr) {
386                         next = list_entry(bond->vlan_list.next,
387                                           struct vlan_entry, vlan_list);
388                 } else {
389                         next = list_entry(curr->vlan_list.next,
390                                           struct vlan_entry, vlan_list);
391                 }
392         }
393
394         return next;
395 }
396
397 /**
398  * bond_dev_queue_xmit - Prepare skb for xmit.
399  *
400  * @bond: bond device that got this skb for tx.
401  * @skb: hw accel VLAN tagged skb to transmit
402  * @slave_dev: slave that is supposed to xmit this skbuff
403  *
404  * When the bond gets an skb to transmit that is
405  * already hardware accelerated VLAN tagged, and it
406  * needs to relay this skb to a slave that is not
407  * hw accel capable, the skb needs to be "unaccelerated",
408  * i.e. strip the hwaccel tag and re-insert it as part
409  * of the payload.
410  */
411 int bond_dev_queue_xmit(struct bonding *bond, struct sk_buff *skb,
412                         struct net_device *slave_dev)
413 {
414         unsigned short uninitialized_var(vlan_id);
415
416         if (!list_empty(&bond->vlan_list) &&
417             !(slave_dev->features & NETIF_F_HW_VLAN_TX) &&
418             vlan_get_tag(skb, &vlan_id) == 0) {
419                 skb->dev = slave_dev;
420                 skb = vlan_put_tag(skb, vlan_id);
421                 if (!skb) {
422                         /* vlan_put_tag() frees the skb in case of error,
423                          * so return success here so the calling functions
424                          * won't attempt to free is again.
425                          */
426                         return 0;
427                 }
428         } else {
429                 skb->dev = slave_dev;
430         }
431
432         skb->priority = 1;
433         dev_queue_xmit(skb);
434
435         return 0;
436 }
437
438 /*
439  * In the following 3 functions, bond_vlan_rx_register(), bond_vlan_rx_add_vid
440  * and bond_vlan_rx_kill_vid, We don't protect the slave list iteration with a
441  * lock because:
442  * a. This operation is performed in IOCTL context,
443  * b. The operation is protected by the RTNL semaphore in the 8021q code,
444  * c. Holding a lock with BH disabled while directly calling a base driver
445  *    entry point is generally a BAD idea.
446  *
447  * The design of synchronization/protection for this operation in the 8021q
448  * module is good for one or more VLAN devices over a single physical device
449  * and cannot be extended for a teaming solution like bonding, so there is a
450  * potential race condition here where a net device from the vlan group might
451  * be referenced (either by a base driver or the 8021q code) while it is being
452  * removed from the system. However, it turns out we're not making matters
453  * worse, and if it works for regular VLAN usage it will work here too.
454 */
455
456 /**
457  * bond_vlan_rx_register - Propagates registration to slaves
458  * @bond_dev: bonding net device that got called
459  * @grp: vlan group being registered
460  */
461 static void bond_vlan_rx_register(struct net_device *bond_dev,
462                                   struct vlan_group *grp)
463 {
464         struct bonding *bond = netdev_priv(bond_dev);
465         struct slave *slave;
466         int i;
467
468         bond->vlgrp = grp;
469
470         bond_for_each_slave(bond, slave, i) {
471                 struct net_device *slave_dev = slave->dev;
472                 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
473
474                 if ((slave_dev->features & NETIF_F_HW_VLAN_RX) &&
475                     slave_ops->ndo_vlan_rx_register) {
476                         slave_ops->ndo_vlan_rx_register(slave_dev, grp);
477                 }
478         }
479 }
480
481 /**
482  * bond_vlan_rx_add_vid - Propagates adding an id to slaves
483  * @bond_dev: bonding net device that got called
484  * @vid: vlan id being added
485  */
486 static void bond_vlan_rx_add_vid(struct net_device *bond_dev, uint16_t vid)
487 {
488         struct bonding *bond = netdev_priv(bond_dev);
489         struct slave *slave;
490         int i, res;
491
492         bond_for_each_slave(bond, slave, i) {
493                 struct net_device *slave_dev = slave->dev;
494                 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
495
496                 if ((slave_dev->features & NETIF_F_HW_VLAN_FILTER) &&
497                     slave_ops->ndo_vlan_rx_add_vid) {
498                         slave_ops->ndo_vlan_rx_add_vid(slave_dev, vid);
499                 }
500         }
501
502         res = bond_add_vlan(bond, vid);
503         if (res) {
504                 pr_err("%s: Error: Failed to add vlan id %d\n",
505                        bond_dev->name, vid);
506         }
507 }
508
509 /**
510  * bond_vlan_rx_kill_vid - Propagates deleting an id to slaves
511  * @bond_dev: bonding net device that got called
512  * @vid: vlan id being removed
513  */
514 static void bond_vlan_rx_kill_vid(struct net_device *bond_dev, uint16_t vid)
515 {
516         struct bonding *bond = netdev_priv(bond_dev);
517         struct slave *slave;
518         struct net_device *vlan_dev;
519         int i, res;
520
521         bond_for_each_slave(bond, slave, i) {
522                 struct net_device *slave_dev = slave->dev;
523                 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
524
525                 if ((slave_dev->features & NETIF_F_HW_VLAN_FILTER) &&
526                     slave_ops->ndo_vlan_rx_kill_vid) {
527                         /* Save and then restore vlan_dev in the grp array,
528                          * since the slave's driver might clear it.
529                          */
530                         vlan_dev = vlan_group_get_device(bond->vlgrp, vid);
531                         slave_ops->ndo_vlan_rx_kill_vid(slave_dev, vid);
532                         vlan_group_set_device(bond->vlgrp, vid, vlan_dev);
533                 }
534         }
535
536         res = bond_del_vlan(bond, vid);
537         if (res) {
538                 pr_err("%s: Error: Failed to remove vlan id %d\n",
539                        bond_dev->name, vid);
540         }
541 }
542
543 static void bond_add_vlans_on_slave(struct bonding *bond, struct net_device *slave_dev)
544 {
545         struct vlan_entry *vlan;
546         const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
547
548         write_lock_bh(&bond->lock);
549
550         if (list_empty(&bond->vlan_list))
551                 goto out;
552
553         if ((slave_dev->features & NETIF_F_HW_VLAN_RX) &&
554             slave_ops->ndo_vlan_rx_register)
555                 slave_ops->ndo_vlan_rx_register(slave_dev, bond->vlgrp);
556
557         if (!(slave_dev->features & NETIF_F_HW_VLAN_FILTER) ||
558             !(slave_ops->ndo_vlan_rx_add_vid))
559                 goto out;
560
561         list_for_each_entry(vlan, &bond->vlan_list, vlan_list)
562                 slave_ops->ndo_vlan_rx_add_vid(slave_dev, vlan->vlan_id);
563
564 out:
565         write_unlock_bh(&bond->lock);
566 }
567
568 static void bond_del_vlans_from_slave(struct bonding *bond,
569                                       struct net_device *slave_dev)
570 {
571         const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
572         struct vlan_entry *vlan;
573         struct net_device *vlan_dev;
574
575         write_lock_bh(&bond->lock);
576
577         if (list_empty(&bond->vlan_list))
578                 goto out;
579
580         if (!(slave_dev->features & NETIF_F_HW_VLAN_FILTER) ||
581             !(slave_ops->ndo_vlan_rx_kill_vid))
582                 goto unreg;
583
584         list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
585                 /* Save and then restore vlan_dev in the grp array,
586                  * since the slave's driver might clear it.
587                  */
588                 vlan_dev = vlan_group_get_device(bond->vlgrp, vlan->vlan_id);
589                 slave_ops->ndo_vlan_rx_kill_vid(slave_dev, vlan->vlan_id);
590                 vlan_group_set_device(bond->vlgrp, vlan->vlan_id, vlan_dev);
591         }
592
593 unreg:
594         if ((slave_dev->features & NETIF_F_HW_VLAN_RX) &&
595             slave_ops->ndo_vlan_rx_register)
596                 slave_ops->ndo_vlan_rx_register(slave_dev, NULL);
597
598 out:
599         write_unlock_bh(&bond->lock);
600 }
601
602 /*------------------------------- Link status -------------------------------*/
603
604 /*
605  * Set the carrier state for the master according to the state of its
606  * slaves.  If any slaves are up, the master is up.  In 802.3ad mode,
607  * do special 802.3ad magic.
608  *
609  * Returns zero if carrier state does not change, nonzero if it does.
610  */
611 static int bond_set_carrier(struct bonding *bond)
612 {
613         struct slave *slave;
614         int i;
615
616         if (bond->slave_cnt == 0)
617                 goto down;
618
619         if (bond->params.mode == BOND_MODE_8023AD)
620                 return bond_3ad_set_carrier(bond);
621
622         bond_for_each_slave(bond, slave, i) {
623                 if (slave->link == BOND_LINK_UP) {
624                         if (!netif_carrier_ok(bond->dev)) {
625                                 netif_carrier_on(bond->dev);
626                                 return 1;
627                         }
628                         return 0;
629                 }
630         }
631
632 down:
633         if (netif_carrier_ok(bond->dev)) {
634                 netif_carrier_off(bond->dev);
635                 return 1;
636         }
637         return 0;
638 }
639
640 /*
641  * Get link speed and duplex from the slave's base driver
642  * using ethtool. If for some reason the call fails or the
643  * values are invalid, fake speed and duplex to 100/Full
644  * and return error.
645  */
646 static int bond_update_speed_duplex(struct slave *slave)
647 {
648         struct net_device *slave_dev = slave->dev;
649         struct ethtool_cmd etool;
650         int res;
651
652         /* Fake speed and duplex */
653         slave->speed = SPEED_100;
654         slave->duplex = DUPLEX_FULL;
655
656         if (!slave_dev->ethtool_ops || !slave_dev->ethtool_ops->get_settings)
657                 return -1;
658
659         res = slave_dev->ethtool_ops->get_settings(slave_dev, &etool);
660         if (res < 0)
661                 return -1;
662
663         switch (etool.speed) {
664         case SPEED_10:
665         case SPEED_100:
666         case SPEED_1000:
667         case SPEED_10000:
668                 break;
669         default:
670                 return -1;
671         }
672
673         switch (etool.duplex) {
674         case DUPLEX_FULL:
675         case DUPLEX_HALF:
676                 break;
677         default:
678                 return -1;
679         }
680
681         slave->speed = etool.speed;
682         slave->duplex = etool.duplex;
683
684         return 0;
685 }
686
687 /*
688  * if <dev> supports MII link status reporting, check its link status.
689  *
690  * We either do MII/ETHTOOL ioctls, or check netif_carrier_ok(),
691  * depending upon the setting of the use_carrier parameter.
692  *
693  * Return either BMSR_LSTATUS, meaning that the link is up (or we
694  * can't tell and just pretend it is), or 0, meaning that the link is
695  * down.
696  *
697  * If reporting is non-zero, instead of faking link up, return -1 if
698  * both ETHTOOL and MII ioctls fail (meaning the device does not
699  * support them).  If use_carrier is set, return whatever it says.
700  * It'd be nice if there was a good way to tell if a driver supports
701  * netif_carrier, but there really isn't.
702  */
703 static int bond_check_dev_link(struct bonding *bond,
704                                struct net_device *slave_dev, int reporting)
705 {
706         const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
707         int (*ioctl)(struct net_device *, struct ifreq *, int);
708         struct ifreq ifr;
709         struct mii_ioctl_data *mii;
710
711         if (!reporting && !netif_running(slave_dev))
712                 return 0;
713
714         if (bond->params.use_carrier)
715                 return netif_carrier_ok(slave_dev) ? BMSR_LSTATUS : 0;
716
717         /* Try to get link status using Ethtool first. */
718         if (slave_dev->ethtool_ops) {
719                 if (slave_dev->ethtool_ops->get_link) {
720                         u32 link;
721
722                         link = slave_dev->ethtool_ops->get_link(slave_dev);
723
724                         return link ? BMSR_LSTATUS : 0;
725                 }
726         }
727
728         /* Ethtool can't be used, fallback to MII ioctls. */
729         ioctl = slave_ops->ndo_do_ioctl;
730         if (ioctl) {
731                 /* TODO: set pointer to correct ioctl on a per team member */
732                 /*       bases to make this more efficient. that is, once  */
733                 /*       we determine the correct ioctl, we will always    */
734                 /*       call it and not the others for that team          */
735                 /*       member.                                           */
736
737                 /*
738                  * We cannot assume that SIOCGMIIPHY will also read a
739                  * register; not all network drivers (e.g., e100)
740                  * support that.
741                  */
742
743                 /* Yes, the mii is overlaid on the ifreq.ifr_ifru */
744                 strncpy(ifr.ifr_name, slave_dev->name, IFNAMSIZ);
745                 mii = if_mii(&ifr);
746                 if (IOCTL(slave_dev, &ifr, SIOCGMIIPHY) == 0) {
747                         mii->reg_num = MII_BMSR;
748                         if (IOCTL(slave_dev, &ifr, SIOCGMIIREG) == 0)
749                                 return mii->val_out & BMSR_LSTATUS;
750                 }
751         }
752
753         /*
754          * If reporting, report that either there's no dev->do_ioctl,
755          * or both SIOCGMIIREG and get_link failed (meaning that we
756          * cannot report link status).  If not reporting, pretend
757          * we're ok.
758          */
759         return reporting ? -1 : BMSR_LSTATUS;
760 }
761
762 /*----------------------------- Multicast list ------------------------------*/
763
764 /*
765  * Push the promiscuity flag down to appropriate slaves
766  */
767 static int bond_set_promiscuity(struct bonding *bond, int inc)
768 {
769         int err = 0;
770         if (USES_PRIMARY(bond->params.mode)) {
771                 /* write lock already acquired */
772                 if (bond->curr_active_slave) {
773                         err = dev_set_promiscuity(bond->curr_active_slave->dev,
774                                                   inc);
775                 }
776         } else {
777                 struct slave *slave;
778                 int i;
779                 bond_for_each_slave(bond, slave, i) {
780                         err = dev_set_promiscuity(slave->dev, inc);
781                         if (err)
782                                 return err;
783                 }
784         }
785         return err;
786 }
787
788 /*
789  * Push the allmulti flag down to all slaves
790  */
791 static int bond_set_allmulti(struct bonding *bond, int inc)
792 {
793         int err = 0;
794         if (USES_PRIMARY(bond->params.mode)) {
795                 /* write lock already acquired */
796                 if (bond->curr_active_slave) {
797                         err = dev_set_allmulti(bond->curr_active_slave->dev,
798                                                inc);
799                 }
800         } else {
801                 struct slave *slave;
802                 int i;
803                 bond_for_each_slave(bond, slave, i) {
804                         err = dev_set_allmulti(slave->dev, inc);
805                         if (err)
806                                 return err;
807                 }
808         }
809         return err;
810 }
811
812 /*
813  * Add a Multicast address to slaves
814  * according to mode
815  */
816 static void bond_mc_add(struct bonding *bond, void *addr)
817 {
818         if (USES_PRIMARY(bond->params.mode)) {
819                 /* write lock already acquired */
820                 if (bond->curr_active_slave)
821                         dev_mc_add(bond->curr_active_slave->dev, addr);
822         } else {
823                 struct slave *slave;
824                 int i;
825
826                 bond_for_each_slave(bond, slave, i)
827                         dev_mc_add(slave->dev, addr);
828         }
829 }
830
831 /*
832  * Remove a multicast address from slave
833  * according to mode
834  */
835 static void bond_mc_del(struct bonding *bond, void *addr)
836 {
837         if (USES_PRIMARY(bond->params.mode)) {
838                 /* write lock already acquired */
839                 if (bond->curr_active_slave)
840                         dev_mc_del(bond->curr_active_slave->dev, addr);
841         } else {
842                 struct slave *slave;
843                 int i;
844                 bond_for_each_slave(bond, slave, i) {
845                         dev_mc_del(slave->dev, addr);
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         rcu_read_unlock();
869 }
870
871 /*
872  * flush all members of flush->mc_list from device dev->mc_list
873  */
874 static void bond_mc_list_flush(struct net_device *bond_dev,
875                                struct net_device *slave_dev)
876 {
877         struct bonding *bond = netdev_priv(bond_dev);
878         struct netdev_hw_addr *ha;
879
880         netdev_for_each_mc_addr(ha, bond_dev)
881                 dev_mc_del(slave_dev, ha->addr);
882
883         if (bond->params.mode == BOND_MODE_8023AD) {
884                 /* del lacpdu mc addr from mc list */
885                 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
886
887                 dev_mc_del(slave_dev, lacpdu_multicast);
888         }
889 }
890
891 /*--------------------------- Active slave change ---------------------------*/
892
893 /*
894  * Update the mc list and multicast-related flags for the new and
895  * old active slaves (if any) according to the multicast mode, and
896  * promiscuous flags unconditionally.
897  */
898 static void bond_mc_swap(struct bonding *bond, struct slave *new_active,
899                          struct slave *old_active)
900 {
901         struct netdev_hw_addr *ha;
902
903         if (!USES_PRIMARY(bond->params.mode))
904                 /* nothing to do -  mc list is already up-to-date on
905                  * all slaves
906                  */
907                 return;
908
909         if (old_active) {
910                 if (bond->dev->flags & IFF_PROMISC)
911                         dev_set_promiscuity(old_active->dev, -1);
912
913                 if (bond->dev->flags & IFF_ALLMULTI)
914                         dev_set_allmulti(old_active->dev, -1);
915
916                 netdev_for_each_mc_addr(ha, bond->dev)
917                         dev_mc_del(old_active->dev, ha->addr);
918         }
919
920         if (new_active) {
921                 /* FIXME: Signal errors upstream. */
922                 if (bond->dev->flags & IFF_PROMISC)
923                         dev_set_promiscuity(new_active->dev, 1);
924
925                 if (bond->dev->flags & IFF_ALLMULTI)
926                         dev_set_allmulti(new_active->dev, 1);
927
928                 netdev_for_each_mc_addr(ha, bond->dev)
929                         dev_mc_add(new_active->dev, ha->addr);
930                 bond_resend_igmp_join_requests(bond);
931         }
932 }
933
934 /*
935  * bond_do_fail_over_mac
936  *
937  * Perform special MAC address swapping for fail_over_mac settings
938  *
939  * Called with RTNL, bond->lock for read, curr_slave_lock for write_bh.
940  */
941 static void bond_do_fail_over_mac(struct bonding *bond,
942                                   struct slave *new_active,
943                                   struct slave *old_active)
944         __releases(&bond->curr_slave_lock)
945         __releases(&bond->lock)
946         __acquires(&bond->lock)
947         __acquires(&bond->curr_slave_lock)
948 {
949         u8 tmp_mac[ETH_ALEN];
950         struct sockaddr saddr;
951         int rv;
952
953         switch (bond->params.fail_over_mac) {
954         case BOND_FOM_ACTIVE:
955                 if (new_active)
956                         memcpy(bond->dev->dev_addr,  new_active->dev->dev_addr,
957                                new_active->dev->addr_len);
958                 break;
959         case BOND_FOM_FOLLOW:
960                 /*
961                  * if new_active && old_active, swap them
962                  * if just old_active, do nothing (going to no active slave)
963                  * if just new_active, set new_active to bond's MAC
964                  */
965                 if (!new_active)
966                         return;
967
968                 write_unlock_bh(&bond->curr_slave_lock);
969                 read_unlock(&bond->lock);
970
971                 if (old_active) {
972                         memcpy(tmp_mac, new_active->dev->dev_addr, ETH_ALEN);
973                         memcpy(saddr.sa_data, old_active->dev->dev_addr,
974                                ETH_ALEN);
975                         saddr.sa_family = new_active->dev->type;
976                 } else {
977                         memcpy(saddr.sa_data, bond->dev->dev_addr, ETH_ALEN);
978                         saddr.sa_family = bond->dev->type;
979                 }
980
981                 rv = dev_set_mac_address(new_active->dev, &saddr);
982                 if (rv) {
983                         pr_err("%s: Error %d setting MAC of slave %s\n",
984                                bond->dev->name, -rv, new_active->dev->name);
985                         goto out;
986                 }
987
988                 if (!old_active)
989                         goto out;
990
991                 memcpy(saddr.sa_data, tmp_mac, ETH_ALEN);
992                 saddr.sa_family = old_active->dev->type;
993
994                 rv = dev_set_mac_address(old_active->dev, &saddr);
995                 if (rv)
996                         pr_err("%s: Error %d setting MAC of slave %s\n",
997                                bond->dev->name, -rv, new_active->dev->name);
998 out:
999                 read_lock(&bond->lock);
1000                 write_lock_bh(&bond->curr_slave_lock);
1001                 break;
1002         default:
1003                 pr_err("%s: bond_do_fail_over_mac impossible: bad policy %d\n",
1004                        bond->dev->name, bond->params.fail_over_mac);
1005                 break;
1006         }
1007
1008 }
1009
1010 static bool bond_should_change_active(struct bonding *bond)
1011 {
1012         struct slave *prim = bond->primary_slave;
1013         struct slave *curr = bond->curr_active_slave;
1014
1015         if (!prim || !curr || curr->link != BOND_LINK_UP)
1016                 return true;
1017         if (bond->force_primary) {
1018                 bond->force_primary = false;
1019                 return true;
1020         }
1021         if (bond->params.primary_reselect == BOND_PRI_RESELECT_BETTER &&
1022             (prim->speed < curr->speed ||
1023              (prim->speed == curr->speed && prim->duplex <= curr->duplex)))
1024                 return false;
1025         if (bond->params.primary_reselect == BOND_PRI_RESELECT_FAILURE)
1026                 return false;
1027         return true;
1028 }
1029
1030 /**
1031  * find_best_interface - select the best available slave to be the active one
1032  * @bond: our bonding struct
1033  *
1034  * Warning: Caller must hold curr_slave_lock for writing.
1035  */
1036 static struct slave *bond_find_best_slave(struct bonding *bond)
1037 {
1038         struct slave *new_active, *old_active;
1039         struct slave *bestslave = NULL;
1040         int mintime = bond->params.updelay;
1041         int i;
1042
1043         new_active = bond->curr_active_slave;
1044
1045         if (!new_active) { /* there were no active slaves left */
1046                 if (bond->slave_cnt > 0)   /* found one slave */
1047                         new_active = bond->first_slave;
1048                 else
1049                         return NULL; /* still no slave, return NULL */
1050         }
1051
1052         if ((bond->primary_slave) &&
1053             bond->primary_slave->link == BOND_LINK_UP &&
1054             bond_should_change_active(bond)) {
1055                 new_active = bond->primary_slave;
1056         }
1057
1058         /* remember where to stop iterating over the slaves */
1059         old_active = new_active;
1060
1061         bond_for_each_slave_from(bond, new_active, i, old_active) {
1062                 if (new_active->link == BOND_LINK_UP) {
1063                         return new_active;
1064                 } else if (new_active->link == BOND_LINK_BACK &&
1065                            IS_UP(new_active->dev)) {
1066                         /* link up, but waiting for stabilization */
1067                         if (new_active->delay < mintime) {
1068                                 mintime = new_active->delay;
1069                                 bestslave = new_active;
1070                         }
1071                 }
1072         }
1073
1074         return bestslave;
1075 }
1076
1077 /**
1078  * change_active_interface - change the active slave into the specified one
1079  * @bond: our bonding struct
1080  * @new: the new slave to make the active one
1081  *
1082  * Set the new slave to the bond's settings and unset them on the old
1083  * curr_active_slave.
1084  * Setting include flags, mc-list, promiscuity, allmulti, etc.
1085  *
1086  * If @new's link state is %BOND_LINK_BACK we'll set it to %BOND_LINK_UP,
1087  * because it is apparently the best available slave we have, even though its
1088  * updelay hasn't timed out yet.
1089  *
1090  * If new_active is not NULL, caller must hold bond->lock for read and
1091  * curr_slave_lock for write_bh.
1092  */
1093 void bond_change_active_slave(struct bonding *bond, struct slave *new_active)
1094 {
1095         struct slave *old_active = bond->curr_active_slave;
1096
1097         if (old_active == new_active)
1098                 return;
1099
1100         if (new_active) {
1101                 new_active->jiffies = jiffies;
1102
1103                 if (new_active->link == BOND_LINK_BACK) {
1104                         if (USES_PRIMARY(bond->params.mode)) {
1105                                 pr_info("%s: making interface %s the new active one %d ms earlier.\n",
1106                                         bond->dev->name, new_active->dev->name,
1107                                         (bond->params.updelay - new_active->delay) * bond->params.miimon);
1108                         }
1109
1110                         new_active->delay = 0;
1111                         new_active->link = BOND_LINK_UP;
1112
1113                         if (bond->params.mode == BOND_MODE_8023AD)
1114                                 bond_3ad_handle_link_change(new_active, BOND_LINK_UP);
1115
1116                         if (bond_is_lb(bond))
1117                                 bond_alb_handle_link_change(bond, new_active, BOND_LINK_UP);
1118                 } else {
1119                         if (USES_PRIMARY(bond->params.mode)) {
1120                                 pr_info("%s: making interface %s the new active one.\n",
1121                                         bond->dev->name, new_active->dev->name);
1122                         }
1123                 }
1124         }
1125
1126         if (USES_PRIMARY(bond->params.mode))
1127                 bond_mc_swap(bond, new_active, old_active);
1128
1129         if (bond_is_lb(bond)) {
1130                 bond_alb_handle_active_change(bond, new_active);
1131                 if (old_active)
1132                         bond_set_slave_inactive_flags(old_active);
1133                 if (new_active)
1134                         bond_set_slave_active_flags(new_active);
1135         } else {
1136                 bond->curr_active_slave = new_active;
1137         }
1138
1139         if (bond->params.mode == BOND_MODE_ACTIVEBACKUP) {
1140                 if (old_active)
1141                         bond_set_slave_inactive_flags(old_active);
1142
1143                 if (new_active) {
1144                         bond_set_slave_active_flags(new_active);
1145
1146                         if (bond->params.fail_over_mac)
1147                                 bond_do_fail_over_mac(bond, new_active,
1148                                                       old_active);
1149
1150                         bond->send_grat_arp = bond->params.num_grat_arp;
1151                         bond_send_gratuitous_arp(bond);
1152
1153                         bond->send_unsol_na = bond->params.num_unsol_na;
1154                         bond_send_unsolicited_na(bond);
1155
1156                         write_unlock_bh(&bond->curr_slave_lock);
1157                         read_unlock(&bond->lock);
1158
1159                         netdev_bonding_change(bond->dev, NETDEV_BONDING_FAILOVER);
1160
1161                         read_lock(&bond->lock);
1162                         write_lock_bh(&bond->curr_slave_lock);
1163                 }
1164         }
1165 }
1166
1167 /**
1168  * bond_select_active_slave - select a new active slave, if needed
1169  * @bond: our bonding struct
1170  *
1171  * This functions should be called when one of the following occurs:
1172  * - The old curr_active_slave has been released or lost its link.
1173  * - The primary_slave has got its link back.
1174  * - A slave has got its link back and there's no old curr_active_slave.
1175  *
1176  * Caller must hold bond->lock for read and curr_slave_lock for write_bh.
1177  */
1178 void bond_select_active_slave(struct bonding *bond)
1179 {
1180         struct slave *best_slave;
1181         int rv;
1182
1183         best_slave = bond_find_best_slave(bond);
1184         if (best_slave != bond->curr_active_slave) {
1185                 bond_change_active_slave(bond, best_slave);
1186                 rv = bond_set_carrier(bond);
1187                 if (!rv)
1188                         return;
1189
1190                 if (netif_carrier_ok(bond->dev)) {
1191                         pr_info("%s: first active interface up!\n",
1192                                 bond->dev->name);
1193                 } else {
1194                         pr_info("%s: now running without any active interface !\n",
1195                                 bond->dev->name);
1196                 }
1197         }
1198 }
1199
1200 /*--------------------------- slave list handling ---------------------------*/
1201
1202 /*
1203  * This function attaches the slave to the end of list.
1204  *
1205  * bond->lock held for writing by caller.
1206  */
1207 static void bond_attach_slave(struct bonding *bond, struct slave *new_slave)
1208 {
1209         if (bond->first_slave == NULL) { /* attaching the first slave */
1210                 new_slave->next = new_slave;
1211                 new_slave->prev = new_slave;
1212                 bond->first_slave = new_slave;
1213         } else {
1214                 new_slave->next = bond->first_slave;
1215                 new_slave->prev = bond->first_slave->prev;
1216                 new_slave->next->prev = new_slave;
1217                 new_slave->prev->next = new_slave;
1218         }
1219
1220         bond->slave_cnt++;
1221 }
1222
1223 /*
1224  * This function detaches the slave from the list.
1225  * WARNING: no check is made to verify if the slave effectively
1226  * belongs to <bond>.
1227  * Nothing is freed on return, structures are just unchained.
1228  * If any slave pointer in bond was pointing to <slave>,
1229  * it should be changed by the calling function.
1230  *
1231  * bond->lock held for writing by caller.
1232  */
1233 static void bond_detach_slave(struct bonding *bond, struct slave *slave)
1234 {
1235         if (slave->next)
1236                 slave->next->prev = slave->prev;
1237
1238         if (slave->prev)
1239                 slave->prev->next = slave->next;
1240
1241         if (bond->first_slave == slave) { /* slave is the first slave */
1242                 if (bond->slave_cnt > 1) { /* there are more slave */
1243                         bond->first_slave = slave->next;
1244                 } else {
1245                         bond->first_slave = NULL; /* slave was the last one */
1246                 }
1247         }
1248
1249         slave->next = NULL;
1250         slave->prev = NULL;
1251         bond->slave_cnt--;
1252 }
1253
1254 /*---------------------------------- IOCTL ----------------------------------*/
1255
1256 static int bond_sethwaddr(struct net_device *bond_dev,
1257                           struct net_device *slave_dev)
1258 {
1259         pr_debug("bond_dev=%p\n", bond_dev);
1260         pr_debug("slave_dev=%p\n", slave_dev);
1261         pr_debug("slave_dev->addr_len=%d\n", slave_dev->addr_len);
1262         memcpy(bond_dev->dev_addr, slave_dev->dev_addr, slave_dev->addr_len);
1263         return 0;
1264 }
1265
1266 #define BOND_VLAN_FEATURES \
1267         (NETIF_F_VLAN_CHALLENGED | NETIF_F_HW_VLAN_RX | NETIF_F_HW_VLAN_TX | \
1268          NETIF_F_HW_VLAN_FILTER)
1269
1270 /*
1271  * Compute the common dev->feature set available to all slaves.  Some
1272  * feature bits are managed elsewhere, so preserve those feature bits
1273  * on the master device.
1274  */
1275 static int bond_compute_features(struct bonding *bond)
1276 {
1277         struct slave *slave;
1278         struct net_device *bond_dev = bond->dev;
1279         unsigned long features = bond_dev->features;
1280         unsigned long vlan_features = 0;
1281         unsigned short max_hard_header_len = max((u16)ETH_HLEN,
1282                                                 bond_dev->hard_header_len);
1283         int i;
1284
1285         features &= ~(NETIF_F_ALL_CSUM | BOND_VLAN_FEATURES);
1286         features |=  NETIF_F_GSO_MASK | NETIF_F_NO_CSUM;
1287
1288         if (!bond->first_slave)
1289                 goto done;
1290
1291         features &= ~NETIF_F_ONE_FOR_ALL;
1292
1293         vlan_features = bond->first_slave->dev->vlan_features;
1294         bond_for_each_slave(bond, slave, i) {
1295                 features = netdev_increment_features(features,
1296                                                      slave->dev->features,
1297                                                      NETIF_F_ONE_FOR_ALL);
1298                 vlan_features = netdev_increment_features(vlan_features,
1299                                                         slave->dev->vlan_features,
1300                                                         NETIF_F_ONE_FOR_ALL);
1301                 if (slave->dev->hard_header_len > max_hard_header_len)
1302                         max_hard_header_len = slave->dev->hard_header_len;
1303         }
1304
1305 done:
1306         features |= (bond_dev->features & BOND_VLAN_FEATURES);
1307         bond_dev->features = netdev_fix_features(features, NULL);
1308         bond_dev->vlan_features = netdev_fix_features(vlan_features, NULL);
1309         bond_dev->hard_header_len = max_hard_header_len;
1310
1311         return 0;
1312 }
1313
1314 static void bond_setup_by_slave(struct net_device *bond_dev,
1315                                 struct net_device *slave_dev)
1316 {
1317         struct bonding *bond = netdev_priv(bond_dev);
1318
1319         bond_dev->header_ops        = slave_dev->header_ops;
1320
1321         bond_dev->type              = slave_dev->type;
1322         bond_dev->hard_header_len   = slave_dev->hard_header_len;
1323         bond_dev->addr_len          = slave_dev->addr_len;
1324
1325         memcpy(bond_dev->broadcast, slave_dev->broadcast,
1326                 slave_dev->addr_len);
1327         bond->setup_by_slave = 1;
1328 }
1329
1330 /* enslave device <slave> to bond device <master> */
1331 int bond_enslave(struct net_device *bond_dev, struct net_device *slave_dev)
1332 {
1333         struct bonding *bond = netdev_priv(bond_dev);
1334         const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
1335         struct slave *new_slave = NULL;
1336         struct netdev_hw_addr *ha;
1337         struct sockaddr addr;
1338         int link_reporting;
1339         int old_features = bond_dev->features;
1340         int res = 0;
1341
1342         if (!bond->params.use_carrier && slave_dev->ethtool_ops == NULL &&
1343                 slave_ops->ndo_do_ioctl == NULL) {
1344                 pr_warning("%s: Warning: no link monitoring support for %s\n",
1345                            bond_dev->name, slave_dev->name);
1346         }
1347
1348         /* bond must be initialized by bond_open() before enslaving */
1349         if (!(bond_dev->flags & IFF_UP)) {
1350                 pr_warning("%s: master_dev is not up in bond_enslave\n",
1351                            bond_dev->name);
1352         }
1353
1354         /* already enslaved */
1355         if (slave_dev->flags & IFF_SLAVE) {
1356                 pr_debug("Error, Device was already enslaved\n");
1357                 return -EBUSY;
1358         }
1359
1360         /* vlan challenged mutual exclusion */
1361         /* no need to lock since we're protected by rtnl_lock */
1362         if (slave_dev->features & NETIF_F_VLAN_CHALLENGED) {
1363                 pr_debug("%s: NETIF_F_VLAN_CHALLENGED\n", slave_dev->name);
1364                 if (!list_empty(&bond->vlan_list)) {
1365                         pr_err("%s: Error: cannot enslave VLAN challenged slave %s on VLAN enabled bond %s\n",
1366                                bond_dev->name, slave_dev->name, bond_dev->name);
1367                         return -EPERM;
1368                 } else {
1369                         pr_warning("%s: Warning: enslaved VLAN challenged slave %s. Adding VLANs will be blocked as long as %s is part of bond %s\n",
1370                                    bond_dev->name, slave_dev->name,
1371                                    slave_dev->name, bond_dev->name);
1372                         bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
1373                 }
1374         } else {
1375                 pr_debug("%s: ! NETIF_F_VLAN_CHALLENGED\n", slave_dev->name);
1376                 if (bond->slave_cnt == 0) {
1377                         /* First slave, and it is not VLAN challenged,
1378                          * so remove the block of adding VLANs over the bond.
1379                          */
1380                         bond_dev->features &= ~NETIF_F_VLAN_CHALLENGED;
1381                 }
1382         }
1383
1384         /*
1385          * Old ifenslave binaries are no longer supported.  These can
1386          * be identified with moderate accuracy by the state of the slave:
1387          * the current ifenslave will set the interface down prior to
1388          * enslaving it; the old ifenslave will not.
1389          */
1390         if ((slave_dev->flags & IFF_UP)) {
1391                 pr_err("%s is up. This may be due to an out of date ifenslave.\n",
1392                        slave_dev->name);
1393                 res = -EPERM;
1394                 goto err_undo_flags;
1395         }
1396
1397         /* set bonding device ether type by slave - bonding netdevices are
1398          * created with ether_setup, so when the slave type is not ARPHRD_ETHER
1399          * there is a need to override some of the type dependent attribs/funcs.
1400          *
1401          * bond ether type mutual exclusion - don't allow slaves of dissimilar
1402          * ether type (eg ARPHRD_ETHER and ARPHRD_INFINIBAND) share the same bond
1403          */
1404         if (bond->slave_cnt == 0) {
1405                 if (bond_dev->type != slave_dev->type) {
1406                         pr_debug("%s: change device type from %d to %d\n",
1407                                  bond_dev->name,
1408                                  bond_dev->type, slave_dev->type);
1409
1410                         res = netdev_bonding_change(bond_dev,
1411                                                     NETDEV_PRE_TYPE_CHANGE);
1412                         res = notifier_to_errno(res);
1413                         if (res) {
1414                                 pr_err("%s: refused to change device type\n",
1415                                        bond_dev->name);
1416                                 res = -EBUSY;
1417                                 goto err_undo_flags;
1418                         }
1419
1420                         /* Flush unicast and multicast addresses */
1421                         dev_uc_flush(bond_dev);
1422                         dev_mc_flush(bond_dev);
1423
1424                         if (slave_dev->type != ARPHRD_ETHER)
1425                                 bond_setup_by_slave(bond_dev, slave_dev);
1426                         else
1427                                 ether_setup(bond_dev);
1428
1429                         netdev_bonding_change(bond_dev,
1430                                               NETDEV_POST_TYPE_CHANGE);
1431                 }
1432         } else if (bond_dev->type != slave_dev->type) {
1433                 pr_err("%s ether type (%d) is different from other slaves (%d), can not enslave it.\n",
1434                        slave_dev->name,
1435                        slave_dev->type, bond_dev->type);
1436                 res = -EINVAL;
1437                 goto err_undo_flags;
1438         }
1439
1440         if (slave_ops->ndo_set_mac_address == NULL) {
1441                 if (bond->slave_cnt == 0) {
1442                         pr_warning("%s: Warning: The first slave device specified does not support setting the MAC address. Setting fail_over_mac to active.",
1443                                    bond_dev->name);
1444                         bond->params.fail_over_mac = BOND_FOM_ACTIVE;
1445                 } else if (bond->params.fail_over_mac != BOND_FOM_ACTIVE) {
1446                         pr_err("%s: Error: The slave device specified does not support setting the MAC address, but fail_over_mac is not set to active.\n",
1447                                bond_dev->name);
1448                         res = -EOPNOTSUPP;
1449                         goto err_undo_flags;
1450                 }
1451         }
1452
1453         new_slave = kzalloc(sizeof(struct slave), GFP_KERNEL);
1454         if (!new_slave) {
1455                 res = -ENOMEM;
1456                 goto err_undo_flags;
1457         }
1458
1459         /* save slave's original flags before calling
1460          * netdev_set_master and dev_open
1461          */
1462         new_slave->original_flags = slave_dev->flags;
1463
1464         /*
1465          * Save slave's original ("permanent") mac address for modes
1466          * that need it, and for restoring it upon release, and then
1467          * set it to the master's address
1468          */
1469         memcpy(new_slave->perm_hwaddr, slave_dev->dev_addr, ETH_ALEN);
1470
1471         if (!bond->params.fail_over_mac) {
1472                 /*
1473                  * Set slave to master's mac address.  The application already
1474                  * set the master's mac address to that of the first slave
1475                  */
1476                 memcpy(addr.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
1477                 addr.sa_family = slave_dev->type;
1478                 res = dev_set_mac_address(slave_dev, &addr);
1479                 if (res) {
1480                         pr_debug("Error %d calling set_mac_address\n", res);
1481                         goto err_free;
1482                 }
1483         }
1484
1485         res = netdev_set_master(slave_dev, bond_dev);
1486         if (res) {
1487                 pr_debug("Error %d calling netdev_set_master\n", res);
1488                 goto err_restore_mac;
1489         }
1490         /* open the slave since the application closed it */
1491         res = dev_open(slave_dev);
1492         if (res) {
1493                 pr_debug("Opening slave %s failed\n", slave_dev->name);
1494                 goto err_unset_master;
1495         }
1496
1497         new_slave->dev = slave_dev;
1498         slave_dev->priv_flags |= IFF_BONDING;
1499
1500         if (bond_is_lb(bond)) {
1501                 /* bond_alb_init_slave() must be called before all other stages since
1502                  * it might fail and we do not want to have to undo everything
1503                  */
1504                 res = bond_alb_init_slave(bond, new_slave);
1505                 if (res)
1506                         goto err_close;
1507         }
1508
1509         /* If the mode USES_PRIMARY, then the new slave gets the
1510          * master's promisc (and mc) settings only if it becomes the
1511          * curr_active_slave, and that is taken care of later when calling
1512          * bond_change_active()
1513          */
1514         if (!USES_PRIMARY(bond->params.mode)) {
1515                 /* set promiscuity level to new slave */
1516                 if (bond_dev->flags & IFF_PROMISC) {
1517                         res = dev_set_promiscuity(slave_dev, 1);
1518                         if (res)
1519                                 goto err_close;
1520                 }
1521
1522                 /* set allmulti level to new slave */
1523                 if (bond_dev->flags & IFF_ALLMULTI) {
1524                         res = dev_set_allmulti(slave_dev, 1);
1525                         if (res)
1526                                 goto err_close;
1527                 }
1528
1529                 netif_addr_lock_bh(bond_dev);
1530                 /* upload master's mc_list to new slave */
1531                 netdev_for_each_mc_addr(ha, bond_dev)
1532                         dev_mc_add(slave_dev, ha->addr);
1533                 netif_addr_unlock_bh(bond_dev);
1534         }
1535
1536         if (bond->params.mode == BOND_MODE_8023AD) {
1537                 /* add lacpdu mc addr to mc list */
1538                 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
1539
1540                 dev_mc_add(slave_dev, lacpdu_multicast);
1541         }
1542
1543         bond_add_vlans_on_slave(bond, slave_dev);
1544
1545         write_lock_bh(&bond->lock);
1546
1547         bond_attach_slave(bond, new_slave);
1548
1549         new_slave->delay = 0;
1550         new_slave->link_failure_count = 0;
1551
1552         bond_compute_features(bond);
1553
1554         write_unlock_bh(&bond->lock);
1555
1556         read_lock(&bond->lock);
1557
1558         new_slave->last_arp_rx = jiffies;
1559
1560         if (bond->params.miimon && !bond->params.use_carrier) {
1561                 link_reporting = bond_check_dev_link(bond, slave_dev, 1);
1562
1563                 if ((link_reporting == -1) && !bond->params.arp_interval) {
1564                         /*
1565                          * miimon is set but a bonded network driver
1566                          * does not support ETHTOOL/MII and
1567                          * arp_interval is not set.  Note: if
1568                          * use_carrier is enabled, we will never go
1569                          * here (because netif_carrier is always
1570                          * supported); thus, we don't need to change
1571                          * the messages for netif_carrier.
1572                          */
1573                         pr_warning("%s: Warning: MII and ETHTOOL support not available for interface %s, and arp_interval/arp_ip_target module parameters not specified, thus bonding will not detect link failures! see bonding.txt for details.\n",
1574                                bond_dev->name, slave_dev->name);
1575                 } else if (link_reporting == -1) {
1576                         /* unable get link status using mii/ethtool */
1577                         pr_warning("%s: Warning: can't get link status from interface %s; the network driver associated with this interface does not support MII or ETHTOOL link status reporting, thus miimon has no effect on this interface.\n",
1578                                    bond_dev->name, slave_dev->name);
1579                 }
1580         }
1581
1582         /* check for initial state */
1583         if (!bond->params.miimon ||
1584             (bond_check_dev_link(bond, slave_dev, 0) == BMSR_LSTATUS)) {
1585                 if (bond->params.updelay) {
1586                         pr_debug("Initial state of slave_dev is BOND_LINK_BACK\n");
1587                         new_slave->link  = BOND_LINK_BACK;
1588                         new_slave->delay = bond->params.updelay;
1589                 } else {
1590                         pr_debug("Initial state of slave_dev is BOND_LINK_UP\n");
1591                         new_slave->link  = BOND_LINK_UP;
1592                 }
1593                 new_slave->jiffies = jiffies;
1594         } else {
1595                 pr_debug("Initial state of slave_dev is BOND_LINK_DOWN\n");
1596                 new_slave->link  = BOND_LINK_DOWN;
1597         }
1598
1599         if (bond_update_speed_duplex(new_slave) &&
1600             (new_slave->link != BOND_LINK_DOWN)) {
1601                 pr_warning("%s: Warning: failed to get speed and duplex from %s, assumed to be 100Mb/sec and Full.\n",
1602                            bond_dev->name, new_slave->dev->name);
1603
1604                 if (bond->params.mode == BOND_MODE_8023AD) {
1605                         pr_warning("%s: Warning: Operation of 802.3ad mode requires ETHTOOL support in base driver for proper aggregator selection.\n",
1606                                    bond_dev->name);
1607                 }
1608         }
1609
1610         if (USES_PRIMARY(bond->params.mode) && bond->params.primary[0]) {
1611                 /* if there is a primary slave, remember it */
1612                 if (strcmp(bond->params.primary, new_slave->dev->name) == 0) {
1613                         bond->primary_slave = new_slave;
1614                         bond->force_primary = true;
1615                 }
1616         }
1617
1618         write_lock_bh(&bond->curr_slave_lock);
1619
1620         switch (bond->params.mode) {
1621         case BOND_MODE_ACTIVEBACKUP:
1622                 bond_set_slave_inactive_flags(new_slave);
1623                 bond_select_active_slave(bond);
1624                 break;
1625         case BOND_MODE_8023AD:
1626                 /* in 802.3ad mode, the internal mechanism
1627                  * will activate the slaves in the selected
1628                  * aggregator
1629                  */
1630                 bond_set_slave_inactive_flags(new_slave);
1631                 /* if this is the first slave */
1632                 if (bond->slave_cnt == 1) {
1633                         SLAVE_AD_INFO(new_slave).id = 1;
1634                         /* Initialize AD with the number of times that the AD timer is called in 1 second
1635                          * can be called only after the mac address of the bond is set
1636                          */
1637                         bond_3ad_initialize(bond, 1000/AD_TIMER_INTERVAL,
1638                                             bond->params.lacp_fast);
1639                 } else {
1640                         SLAVE_AD_INFO(new_slave).id =
1641                                 SLAVE_AD_INFO(new_slave->prev).id + 1;
1642                 }
1643
1644                 bond_3ad_bind_slave(new_slave);
1645                 break;
1646         case BOND_MODE_TLB:
1647         case BOND_MODE_ALB:
1648                 new_slave->state = BOND_STATE_ACTIVE;
1649                 bond_set_slave_inactive_flags(new_slave);
1650                 bond_select_active_slave(bond);
1651                 break;
1652         default:
1653                 pr_debug("This slave is always active in trunk mode\n");
1654
1655                 /* always active in trunk mode */
1656                 new_slave->state = BOND_STATE_ACTIVE;
1657
1658                 /* In trunking mode there is little meaning to curr_active_slave
1659                  * anyway (it holds no special properties of the bond device),
1660                  * so we can change it without calling change_active_interface()
1661                  */
1662                 if (!bond->curr_active_slave)
1663                         bond->curr_active_slave = new_slave;
1664
1665                 break;
1666         } /* switch(bond_mode) */
1667
1668         write_unlock_bh(&bond->curr_slave_lock);
1669
1670         bond_set_carrier(bond);
1671
1672         read_unlock(&bond->lock);
1673
1674         res = bond_create_slave_symlinks(bond_dev, slave_dev);
1675         if (res)
1676                 goto err_close;
1677
1678         pr_info("%s: enslaving %s as a%s interface with a%s link.\n",
1679                 bond_dev->name, slave_dev->name,
1680                 new_slave->state == BOND_STATE_ACTIVE ? "n active" : " backup",
1681                 new_slave->link != BOND_LINK_DOWN ? "n up" : " down");
1682
1683         /* enslave is successful */
1684         return 0;
1685
1686 /* Undo stages on error */
1687 err_close:
1688         dev_close(slave_dev);
1689
1690 err_unset_master:
1691         netdev_set_master(slave_dev, NULL);
1692
1693 err_restore_mac:
1694         if (!bond->params.fail_over_mac) {
1695                 /* XXX TODO - fom follow mode needs to change master's
1696                  * MAC if this slave's MAC is in use by the bond, or at
1697                  * least print a warning.
1698                  */
1699                 memcpy(addr.sa_data, new_slave->perm_hwaddr, ETH_ALEN);
1700                 addr.sa_family = slave_dev->type;
1701                 dev_set_mac_address(slave_dev, &addr);
1702         }
1703
1704 err_free:
1705         kfree(new_slave);
1706
1707 err_undo_flags:
1708         bond_dev->features = old_features;
1709
1710         return res;
1711 }
1712
1713 /*
1714  * Try to release the slave device <slave> from the bond device <master>
1715  * It is legal to access curr_active_slave without a lock because all the function
1716  * is write-locked.
1717  *
1718  * The rules for slave state should be:
1719  *   for Active/Backup:
1720  *     Active stays on all backups go down
1721  *   for Bonded connections:
1722  *     The first up interface should be left on and all others downed.
1723  */
1724 int bond_release(struct net_device *bond_dev, struct net_device *slave_dev)
1725 {
1726         struct bonding *bond = netdev_priv(bond_dev);
1727         struct slave *slave, *oldcurrent;
1728         struct sockaddr addr;
1729
1730         /* slave is not a slave or master is not master of this slave */
1731         if (!(slave_dev->flags & IFF_SLAVE) ||
1732             (slave_dev->master != bond_dev)) {
1733                 pr_err("%s: Error: cannot release %s.\n",
1734                        bond_dev->name, slave_dev->name);
1735                 return -EINVAL;
1736         }
1737
1738         write_lock_bh(&bond->lock);
1739
1740         slave = bond_get_slave_by_dev(bond, slave_dev);
1741         if (!slave) {
1742                 /* not a slave of this bond */
1743                 pr_info("%s: %s not enslaved\n",
1744                         bond_dev->name, slave_dev->name);
1745                 write_unlock_bh(&bond->lock);
1746                 return -EINVAL;
1747         }
1748
1749         if (!bond->params.fail_over_mac) {
1750                 if (!compare_ether_addr(bond_dev->dev_addr, slave->perm_hwaddr) &&
1751                     bond->slave_cnt > 1)
1752                         pr_warning("%s: Warning: the permanent HWaddr of %s - %pM - is still in use by %s. Set the HWaddr of %s to a different address to avoid conflicts.\n",
1753                                    bond_dev->name, slave_dev->name,
1754                                    slave->perm_hwaddr,
1755                                    bond_dev->name, slave_dev->name);
1756         }
1757
1758         /* Inform AD package of unbinding of slave. */
1759         if (bond->params.mode == BOND_MODE_8023AD) {
1760                 /* must be called before the slave is
1761                  * detached from the list
1762                  */
1763                 bond_3ad_unbind_slave(slave);
1764         }
1765
1766         pr_info("%s: releasing %s interface %s\n",
1767                 bond_dev->name,
1768                 (slave->state == BOND_STATE_ACTIVE) ? "active" : "backup",
1769                 slave_dev->name);
1770
1771         oldcurrent = bond->curr_active_slave;
1772
1773         bond->current_arp_slave = NULL;
1774
1775         /* release the slave from its bond */
1776         bond_detach_slave(bond, slave);
1777
1778         bond_compute_features(bond);
1779
1780         if (bond->primary_slave == slave)
1781                 bond->primary_slave = NULL;
1782
1783         if (oldcurrent == slave)
1784                 bond_change_active_slave(bond, NULL);
1785
1786         if (bond_is_lb(bond)) {
1787                 /* Must be called only after the slave has been
1788                  * detached from the list and the curr_active_slave
1789                  * has been cleared (if our_slave == old_current),
1790                  * but before a new active slave is selected.
1791                  */
1792                 write_unlock_bh(&bond->lock);
1793                 bond_alb_deinit_slave(bond, slave);
1794                 write_lock_bh(&bond->lock);
1795         }
1796
1797         if (oldcurrent == slave) {
1798                 /*
1799                  * Note that we hold RTNL over this sequence, so there
1800                  * is no concern that another slave add/remove event
1801                  * will interfere.
1802                  */
1803                 write_unlock_bh(&bond->lock);
1804                 read_lock(&bond->lock);
1805                 write_lock_bh(&bond->curr_slave_lock);
1806
1807                 bond_select_active_slave(bond);
1808
1809                 write_unlock_bh(&bond->curr_slave_lock);
1810                 read_unlock(&bond->lock);
1811                 write_lock_bh(&bond->lock);
1812         }
1813
1814         if (bond->slave_cnt == 0) {
1815                 bond_set_carrier(bond);
1816
1817                 /* if the last slave was removed, zero the mac address
1818                  * of the master so it will be set by the application
1819                  * to the mac address of the first slave
1820                  */
1821                 memset(bond_dev->dev_addr, 0, bond_dev->addr_len);
1822
1823                 if (list_empty(&bond->vlan_list)) {
1824                         bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
1825                 } else {
1826                         pr_warning("%s: Warning: clearing HW address of %s while it still has VLANs.\n",
1827                                    bond_dev->name, bond_dev->name);
1828                         pr_warning("%s: When re-adding slaves, make sure the bond's HW address matches its VLANs'.\n",
1829                                    bond_dev->name);
1830                 }
1831         } else if ((bond_dev->features & NETIF_F_VLAN_CHALLENGED) &&
1832                    !bond_has_challenged_slaves(bond)) {
1833                 pr_info("%s: last VLAN challenged slave %s left bond %s. VLAN blocking is removed\n",
1834                         bond_dev->name, slave_dev->name, bond_dev->name);
1835                 bond_dev->features &= ~NETIF_F_VLAN_CHALLENGED;
1836         }
1837
1838         write_unlock_bh(&bond->lock);
1839
1840         /* must do this from outside any spinlocks */
1841         bond_destroy_slave_symlinks(bond_dev, slave_dev);
1842
1843         bond_del_vlans_from_slave(bond, slave_dev);
1844
1845         /* If the mode USES_PRIMARY, then we should only remove its
1846          * promisc and mc settings if it was the curr_active_slave, but that was
1847          * already taken care of above when we detached the slave
1848          */
1849         if (!USES_PRIMARY(bond->params.mode)) {
1850                 /* unset promiscuity level from slave */
1851                 if (bond_dev->flags & IFF_PROMISC)
1852                         dev_set_promiscuity(slave_dev, -1);
1853
1854                 /* unset allmulti level from slave */
1855                 if (bond_dev->flags & IFF_ALLMULTI)
1856                         dev_set_allmulti(slave_dev, -1);
1857
1858                 /* flush master's mc_list from slave */
1859                 netif_addr_lock_bh(bond_dev);
1860                 bond_mc_list_flush(bond_dev, slave_dev);
1861                 netif_addr_unlock_bh(bond_dev);
1862         }
1863
1864         netdev_set_master(slave_dev, NULL);
1865
1866         /* close slave before restoring its mac address */
1867         dev_close(slave_dev);
1868
1869         if (bond->params.fail_over_mac != BOND_FOM_ACTIVE) {
1870                 /* restore original ("permanent") mac address */
1871                 memcpy(addr.sa_data, slave->perm_hwaddr, ETH_ALEN);
1872                 addr.sa_family = slave_dev->type;
1873                 dev_set_mac_address(slave_dev, &addr);
1874         }
1875
1876         slave_dev->priv_flags &= ~(IFF_MASTER_8023AD | IFF_MASTER_ALB |
1877                                    IFF_SLAVE_INACTIVE | IFF_BONDING |
1878                                    IFF_SLAVE_NEEDARP);
1879
1880         kfree(slave);
1881
1882         return 0;  /* deletion OK */
1883 }
1884
1885 /*
1886 * First release a slave and than destroy the bond if no more slaves are left.
1887 * Must be under rtnl_lock when this function is called.
1888 */
1889 int  bond_release_and_destroy(struct net_device *bond_dev,
1890                               struct net_device *slave_dev)
1891 {
1892         struct bonding *bond = netdev_priv(bond_dev);
1893         int ret;
1894
1895         ret = bond_release(bond_dev, slave_dev);
1896         if ((ret == 0) && (bond->slave_cnt == 0)) {
1897                 pr_info("%s: destroying bond %s.\n",
1898                         bond_dev->name, bond_dev->name);
1899                 unregister_netdevice(bond_dev);
1900         }
1901         return ret;
1902 }
1903
1904 /*
1905  * This function releases all slaves.
1906  */
1907 static int bond_release_all(struct net_device *bond_dev)
1908 {
1909         struct bonding *bond = netdev_priv(bond_dev);
1910         struct slave *slave;
1911         struct net_device *slave_dev;
1912         struct sockaddr addr;
1913
1914         write_lock_bh(&bond->lock);
1915
1916         netif_carrier_off(bond_dev);
1917
1918         if (bond->slave_cnt == 0)
1919                 goto out;
1920
1921         bond->current_arp_slave = NULL;
1922         bond->primary_slave = NULL;
1923         bond_change_active_slave(bond, NULL);
1924
1925         while ((slave = bond->first_slave) != NULL) {
1926                 /* Inform AD package of unbinding of slave
1927                  * before slave is detached from the list.
1928                  */
1929                 if (bond->params.mode == BOND_MODE_8023AD)
1930                         bond_3ad_unbind_slave(slave);
1931
1932                 slave_dev = slave->dev;
1933                 bond_detach_slave(bond, slave);
1934
1935                 /* now that the slave is detached, unlock and perform
1936                  * all the undo steps that should not be called from
1937                  * within a lock.
1938                  */
1939                 write_unlock_bh(&bond->lock);
1940
1941                 if (bond_is_lb(bond)) {
1942                         /* must be called only after the slave
1943                          * has been detached from the list
1944                          */
1945                         bond_alb_deinit_slave(bond, slave);
1946                 }
1947
1948                 bond_compute_features(bond);
1949
1950                 bond_destroy_slave_symlinks(bond_dev, slave_dev);
1951                 bond_del_vlans_from_slave(bond, slave_dev);
1952
1953                 /* If the mode USES_PRIMARY, then we should only remove its
1954                  * promisc and mc settings if it was the curr_active_slave, but that was
1955                  * already taken care of above when we detached the slave
1956                  */
1957                 if (!USES_PRIMARY(bond->params.mode)) {
1958                         /* unset promiscuity level from slave */
1959                         if (bond_dev->flags & IFF_PROMISC)
1960                                 dev_set_promiscuity(slave_dev, -1);
1961
1962                         /* unset allmulti level from slave */
1963                         if (bond_dev->flags & IFF_ALLMULTI)
1964                                 dev_set_allmulti(slave_dev, -1);
1965
1966                         /* flush master's mc_list from slave */
1967                         netif_addr_lock_bh(bond_dev);
1968                         bond_mc_list_flush(bond_dev, slave_dev);
1969                         netif_addr_unlock_bh(bond_dev);
1970                 }
1971
1972                 netdev_set_master(slave_dev, NULL);
1973
1974                 /* close slave before restoring its mac address */
1975                 dev_close(slave_dev);
1976
1977                 if (!bond->params.fail_over_mac) {
1978                         /* restore original ("permanent") mac address*/
1979                         memcpy(addr.sa_data, slave->perm_hwaddr, ETH_ALEN);
1980                         addr.sa_family = slave_dev->type;
1981                         dev_set_mac_address(slave_dev, &addr);
1982                 }
1983
1984                 slave_dev->priv_flags &= ~(IFF_MASTER_8023AD | IFF_MASTER_ALB |
1985                                            IFF_SLAVE_INACTIVE);
1986
1987                 kfree(slave);
1988
1989                 /* re-acquire the lock before getting the next slave */
1990                 write_lock_bh(&bond->lock);
1991         }
1992
1993         /* zero the mac address of the master so it will be
1994          * set by the application to the mac address of the
1995          * first slave
1996          */
1997         memset(bond_dev->dev_addr, 0, bond_dev->addr_len);
1998
1999         if (list_empty(&bond->vlan_list))
2000                 bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
2001         else {
2002                 pr_warning("%s: Warning: clearing HW address of %s while it still has VLANs.\n",
2003                            bond_dev->name, bond_dev->name);
2004                 pr_warning("%s: When re-adding slaves, make sure the bond's HW address matches its VLANs'.\n",
2005                            bond_dev->name);
2006         }
2007
2008         pr_info("%s: released all slaves\n", bond_dev->name);
2009
2010 out:
2011         write_unlock_bh(&bond->lock);
2012
2013         return 0;
2014 }
2015
2016 /*
2017  * This function changes the active slave to slave <slave_dev>.
2018  * It returns -EINVAL in the following cases.
2019  *  - <slave_dev> is not found in the list.
2020  *  - There is not active slave now.
2021  *  - <slave_dev> is already active.
2022  *  - The link state of <slave_dev> is not BOND_LINK_UP.
2023  *  - <slave_dev> is not running.
2024  * In these cases, this function does nothing.
2025  * In the other cases, current_slave pointer is changed and 0 is returned.
2026  */
2027 static int bond_ioctl_change_active(struct net_device *bond_dev, struct net_device *slave_dev)
2028 {
2029         struct bonding *bond = netdev_priv(bond_dev);
2030         struct slave *old_active = NULL;
2031         struct slave *new_active = NULL;
2032         int res = 0;
2033
2034         if (!USES_PRIMARY(bond->params.mode))
2035                 return -EINVAL;
2036
2037         /* Verify that master_dev is indeed the master of slave_dev */
2038         if (!(slave_dev->flags & IFF_SLAVE) || (slave_dev->master != bond_dev))
2039                 return -EINVAL;
2040
2041         read_lock(&bond->lock);
2042
2043         read_lock(&bond->curr_slave_lock);
2044         old_active = bond->curr_active_slave;
2045         read_unlock(&bond->curr_slave_lock);
2046
2047         new_active = bond_get_slave_by_dev(bond, slave_dev);
2048
2049         /*
2050          * Changing to the current active: do nothing; return success.
2051          */
2052         if (new_active && (new_active == old_active)) {
2053                 read_unlock(&bond->lock);
2054                 return 0;
2055         }
2056
2057         if ((new_active) &&
2058             (old_active) &&
2059             (new_active->link == BOND_LINK_UP) &&
2060             IS_UP(new_active->dev)) {
2061                 write_lock_bh(&bond->curr_slave_lock);
2062                 bond_change_active_slave(bond, new_active);
2063                 write_unlock_bh(&bond->curr_slave_lock);
2064         } else
2065                 res = -EINVAL;
2066
2067         read_unlock(&bond->lock);
2068
2069         return res;
2070 }
2071
2072 static int bond_info_query(struct net_device *bond_dev, struct ifbond *info)
2073 {
2074         struct bonding *bond = netdev_priv(bond_dev);
2075
2076         info->bond_mode = bond->params.mode;
2077         info->miimon = bond->params.miimon;
2078
2079         read_lock(&bond->lock);
2080         info->num_slaves = bond->slave_cnt;
2081         read_unlock(&bond->lock);
2082
2083         return 0;
2084 }
2085
2086 static int bond_slave_info_query(struct net_device *bond_dev, struct ifslave *info)
2087 {
2088         struct bonding *bond = netdev_priv(bond_dev);
2089         struct slave *slave;
2090         int i, res = -ENODEV;
2091
2092         read_lock(&bond->lock);
2093
2094         bond_for_each_slave(bond, slave, i) {
2095                 if (i == (int)info->slave_id) {
2096                         res = 0;
2097                         strcpy(info->slave_name, slave->dev->name);
2098                         info->link = slave->link;
2099                         info->state = slave->state;
2100                         info->link_failure_count = slave->link_failure_count;
2101                         break;
2102                 }
2103         }
2104
2105         read_unlock(&bond->lock);
2106
2107         return res;
2108 }
2109
2110 /*-------------------------------- Monitoring -------------------------------*/
2111
2112
2113 static int bond_miimon_inspect(struct bonding *bond)
2114 {
2115         struct slave *slave;
2116         int i, link_state, commit = 0;
2117         bool ignore_updelay;
2118
2119         ignore_updelay = !bond->curr_active_slave ? true : false;
2120
2121         bond_for_each_slave(bond, slave, i) {
2122                 slave->new_link = BOND_LINK_NOCHANGE;
2123
2124                 link_state = bond_check_dev_link(bond, slave->dev, 0);
2125
2126                 switch (slave->link) {
2127                 case BOND_LINK_UP:
2128                         if (link_state)
2129                                 continue;
2130
2131                         slave->link = BOND_LINK_FAIL;
2132                         slave->delay = bond->params.downdelay;
2133                         if (slave->delay) {
2134                                 pr_info("%s: link status down for %sinterface %s, disabling it in %d ms.\n",
2135                                         bond->dev->name,
2136                                         (bond->params.mode ==
2137                                          BOND_MODE_ACTIVEBACKUP) ?
2138                                         ((slave->state == BOND_STATE_ACTIVE) ?
2139                                          "active " : "backup ") : "",
2140                                         slave->dev->name,
2141                                         bond->params.downdelay * bond->params.miimon);
2142                         }
2143                         /*FALLTHRU*/
2144                 case BOND_LINK_FAIL:
2145                         if (link_state) {
2146                                 /*
2147                                  * recovered before downdelay expired
2148                                  */
2149                                 slave->link = BOND_LINK_UP;
2150                                 slave->jiffies = jiffies;
2151                                 pr_info("%s: link status up again after %d ms for interface %s.\n",
2152                                         bond->dev->name,
2153                                         (bond->params.downdelay - slave->delay) *
2154                                         bond->params.miimon,
2155                                         slave->dev->name);
2156                                 continue;
2157                         }
2158
2159                         if (slave->delay <= 0) {
2160                                 slave->new_link = BOND_LINK_DOWN;
2161                                 commit++;
2162                                 continue;
2163                         }
2164
2165                         slave->delay--;
2166                         break;
2167
2168                 case BOND_LINK_DOWN:
2169                         if (!link_state)
2170                                 continue;
2171
2172                         slave->link = BOND_LINK_BACK;
2173                         slave->delay = bond->params.updelay;
2174
2175                         if (slave->delay) {
2176                                 pr_info("%s: link status up for interface %s, enabling it in %d ms.\n",
2177                                         bond->dev->name, slave->dev->name,
2178                                         ignore_updelay ? 0 :
2179                                         bond->params.updelay *
2180                                         bond->params.miimon);
2181                         }
2182                         /*FALLTHRU*/
2183                 case BOND_LINK_BACK:
2184                         if (!link_state) {
2185                                 slave->link = BOND_LINK_DOWN;
2186                                 pr_info("%s: link status down again after %d ms for interface %s.\n",
2187                                         bond->dev->name,
2188                                         (bond->params.updelay - slave->delay) *
2189                                         bond->params.miimon,
2190                                         slave->dev->name);
2191
2192                                 continue;
2193                         }
2194
2195                         if (ignore_updelay)
2196                                 slave->delay = 0;
2197
2198                         if (slave->delay <= 0) {
2199                                 slave->new_link = BOND_LINK_UP;
2200                                 commit++;
2201                                 ignore_updelay = false;
2202                                 continue;
2203                         }
2204
2205                         slave->delay--;
2206                         break;
2207                 }
2208         }
2209
2210         return commit;
2211 }
2212
2213 static void bond_miimon_commit(struct bonding *bond)
2214 {
2215         struct slave *slave;
2216         int i;
2217
2218         bond_for_each_slave(bond, slave, i) {
2219                 switch (slave->new_link) {
2220                 case BOND_LINK_NOCHANGE:
2221                         continue;
2222
2223                 case BOND_LINK_UP:
2224                         slave->link = BOND_LINK_UP;
2225                         slave->jiffies = jiffies;
2226
2227                         if (bond->params.mode == BOND_MODE_8023AD) {
2228                                 /* prevent it from being the active one */
2229                                 slave->state = BOND_STATE_BACKUP;
2230                         } else if (bond->params.mode != BOND_MODE_ACTIVEBACKUP) {
2231                                 /* make it immediately active */
2232                                 slave->state = BOND_STATE_ACTIVE;
2233                         } else if (slave != bond->primary_slave) {
2234                                 /* prevent it from being the active one */
2235                                 slave->state = BOND_STATE_BACKUP;
2236                         }
2237
2238                         pr_info("%s: link status definitely up for interface %s.\n",
2239                                 bond->dev->name, slave->dev->name);
2240
2241                         /* notify ad that the link status has changed */
2242                         if (bond->params.mode == BOND_MODE_8023AD)
2243                                 bond_3ad_handle_link_change(slave, BOND_LINK_UP);
2244
2245                         if (bond_is_lb(bond))
2246                                 bond_alb_handle_link_change(bond, slave,
2247                                                             BOND_LINK_UP);
2248
2249                         if (!bond->curr_active_slave ||
2250                             (slave == bond->primary_slave))
2251                                 goto do_failover;
2252
2253                         continue;
2254
2255                 case BOND_LINK_DOWN:
2256                         if (slave->link_failure_count < UINT_MAX)
2257                                 slave->link_failure_count++;
2258
2259                         slave->link = BOND_LINK_DOWN;
2260
2261                         if (bond->params.mode == BOND_MODE_ACTIVEBACKUP ||
2262                             bond->params.mode == BOND_MODE_8023AD)
2263                                 bond_set_slave_inactive_flags(slave);
2264
2265                         pr_info("%s: link status definitely down for interface %s, disabling it\n",
2266                                 bond->dev->name, slave->dev->name);
2267
2268                         if (bond->params.mode == BOND_MODE_8023AD)
2269                                 bond_3ad_handle_link_change(slave,
2270                                                             BOND_LINK_DOWN);
2271
2272                         if (bond_is_lb(bond))
2273                                 bond_alb_handle_link_change(bond, slave,
2274                                                             BOND_LINK_DOWN);
2275
2276                         if (slave == bond->curr_active_slave)
2277                                 goto do_failover;
2278
2279                         continue;
2280
2281                 default:
2282                         pr_err("%s: invalid new link %d on slave %s\n",
2283                                bond->dev->name, slave->new_link,
2284                                slave->dev->name);
2285                         slave->new_link = BOND_LINK_NOCHANGE;
2286
2287                         continue;
2288                 }
2289
2290 do_failover:
2291                 ASSERT_RTNL();
2292                 write_lock_bh(&bond->curr_slave_lock);
2293                 bond_select_active_slave(bond);
2294                 write_unlock_bh(&bond->curr_slave_lock);
2295         }
2296
2297         bond_set_carrier(bond);
2298 }
2299
2300 /*
2301  * bond_mii_monitor
2302  *
2303  * Really a wrapper that splits the mii monitor into two phases: an
2304  * inspection, then (if inspection indicates something needs to be done)
2305  * an acquisition of appropriate locks followed by a commit phase to
2306  * implement whatever link state changes are indicated.
2307  */
2308 void bond_mii_monitor(struct work_struct *work)
2309 {
2310         struct bonding *bond = container_of(work, struct bonding,
2311                                             mii_work.work);
2312
2313         read_lock(&bond->lock);
2314         if (bond->kill_timers)
2315                 goto out;
2316
2317         if (bond->slave_cnt == 0)
2318                 goto re_arm;
2319
2320         if (bond->send_grat_arp) {
2321                 read_lock(&bond->curr_slave_lock);
2322                 bond_send_gratuitous_arp(bond);
2323                 read_unlock(&bond->curr_slave_lock);
2324         }
2325
2326         if (bond->send_unsol_na) {
2327                 read_lock(&bond->curr_slave_lock);
2328                 bond_send_unsolicited_na(bond);
2329                 read_unlock(&bond->curr_slave_lock);
2330         }
2331
2332         if (bond_miimon_inspect(bond)) {
2333                 read_unlock(&bond->lock);
2334                 rtnl_lock();
2335                 read_lock(&bond->lock);
2336
2337                 bond_miimon_commit(bond);
2338
2339                 read_unlock(&bond->lock);
2340                 rtnl_unlock();  /* might sleep, hold no other locks */
2341                 read_lock(&bond->lock);
2342         }
2343
2344 re_arm:
2345         if (bond->params.miimon)
2346                 queue_delayed_work(bond->wq, &bond->mii_work,
2347                                    msecs_to_jiffies(bond->params.miimon));
2348 out:
2349         read_unlock(&bond->lock);
2350 }
2351
2352 static __be32 bond_glean_dev_ip(struct net_device *dev)
2353 {
2354         struct in_device *idev;
2355         struct in_ifaddr *ifa;
2356         __be32 addr = 0;
2357
2358         if (!dev)
2359                 return 0;
2360
2361         rcu_read_lock();
2362         idev = __in_dev_get_rcu(dev);
2363         if (!idev)
2364                 goto out;
2365
2366         ifa = idev->ifa_list;
2367         if (!ifa)
2368                 goto out;
2369
2370         addr = ifa->ifa_local;
2371 out:
2372         rcu_read_unlock();
2373         return addr;
2374 }
2375
2376 static int bond_has_this_ip(struct bonding *bond, __be32 ip)
2377 {
2378         struct vlan_entry *vlan;
2379
2380         if (ip == bond->master_ip)
2381                 return 1;
2382
2383         list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
2384                 if (ip == vlan->vlan_ip)
2385                         return 1;
2386         }
2387
2388         return 0;
2389 }
2390
2391 /*
2392  * We go to the (large) trouble of VLAN tagging ARP frames because
2393  * switches in VLAN mode (especially if ports are configured as
2394  * "native" to a VLAN) might not pass non-tagged frames.
2395  */
2396 static void bond_arp_send(struct net_device *slave_dev, int arp_op, __be32 dest_ip, __be32 src_ip, unsigned short vlan_id)
2397 {
2398         struct sk_buff *skb;
2399
2400         pr_debug("arp %d on slave %s: dst %x src %x vid %d\n", arp_op,
2401                  slave_dev->name, dest_ip, src_ip, vlan_id);
2402
2403         skb = arp_create(arp_op, ETH_P_ARP, dest_ip, slave_dev, src_ip,
2404                          NULL, slave_dev->dev_addr, NULL);
2405
2406         if (!skb) {
2407                 pr_err("ARP packet allocation failed\n");
2408                 return;
2409         }
2410         if (vlan_id) {
2411                 skb = vlan_put_tag(skb, vlan_id);
2412                 if (!skb) {
2413                         pr_err("failed to insert VLAN tag\n");
2414                         return;
2415                 }
2416         }
2417         arp_xmit(skb);
2418 }
2419
2420
2421 static void bond_arp_send_all(struct bonding *bond, struct slave *slave)
2422 {
2423         int i, vlan_id, rv;
2424         __be32 *targets = bond->params.arp_targets;
2425         struct vlan_entry *vlan;
2426         struct net_device *vlan_dev;
2427         struct flowi fl;
2428         struct rtable *rt;
2429
2430         for (i = 0; (i < BOND_MAX_ARP_TARGETS); i++) {
2431                 if (!targets[i])
2432                         break;
2433                 pr_debug("basa: target %x\n", targets[i]);
2434                 if (list_empty(&bond->vlan_list)) {
2435                         pr_debug("basa: empty vlan: arp_send\n");
2436                         bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2437                                       bond->master_ip, 0);
2438                         continue;
2439                 }
2440
2441                 /*
2442                  * If VLANs are configured, we do a route lookup to
2443                  * determine which VLAN interface would be used, so we
2444                  * can tag the ARP with the proper VLAN tag.
2445                  */
2446                 memset(&fl, 0, sizeof(fl));
2447                 fl.fl4_dst = targets[i];
2448                 fl.fl4_tos = RTO_ONLINK;
2449
2450                 rv = ip_route_output_key(dev_net(bond->dev), &rt, &fl);
2451                 if (rv) {
2452                         if (net_ratelimit()) {
2453                                 pr_warning("%s: no route to arp_ip_target %pI4\n",
2454                                            bond->dev->name, &fl.fl4_dst);
2455                         }
2456                         continue;
2457                 }
2458
2459                 /*
2460                  * This target is not on a VLAN
2461                  */
2462                 if (rt->u.dst.dev == bond->dev) {
2463                         ip_rt_put(rt);
2464                         pr_debug("basa: rtdev == bond->dev: arp_send\n");
2465                         bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2466                                       bond->master_ip, 0);
2467                         continue;
2468                 }
2469
2470                 vlan_id = 0;
2471                 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
2472                         vlan_dev = vlan_group_get_device(bond->vlgrp, vlan->vlan_id);
2473                         if (vlan_dev == rt->u.dst.dev) {
2474                                 vlan_id = vlan->vlan_id;
2475                                 pr_debug("basa: vlan match on %s %d\n",
2476                                        vlan_dev->name, vlan_id);
2477                                 break;
2478                         }
2479                 }
2480
2481                 if (vlan_id) {
2482                         ip_rt_put(rt);
2483                         bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2484                                       vlan->vlan_ip, vlan_id);
2485                         continue;
2486                 }
2487
2488                 if (net_ratelimit()) {
2489                         pr_warning("%s: no path to arp_ip_target %pI4 via rt.dev %s\n",
2490                                    bond->dev->name, &fl.fl4_dst,
2491                                    rt->u.dst.dev ? rt->u.dst.dev->name : "NULL");
2492                 }
2493                 ip_rt_put(rt);
2494         }
2495 }
2496
2497 /*
2498  * Kick out a gratuitous ARP for an IP on the bonding master plus one
2499  * for each VLAN above us.
2500  *
2501  * Caller must hold curr_slave_lock for read or better
2502  */
2503 static void bond_send_gratuitous_arp(struct bonding *bond)
2504 {
2505         struct slave *slave = bond->curr_active_slave;
2506         struct vlan_entry *vlan;
2507         struct net_device *vlan_dev;
2508
2509         pr_debug("bond_send_grat_arp: bond %s slave %s\n",
2510                  bond->dev->name, slave ? slave->dev->name : "NULL");
2511
2512         if (!slave || !bond->send_grat_arp ||
2513             test_bit(__LINK_STATE_LINKWATCH_PENDING, &slave->dev->state))
2514                 return;
2515
2516         bond->send_grat_arp--;
2517
2518         if (bond->master_ip) {
2519                 bond_arp_send(slave->dev, ARPOP_REPLY, bond->master_ip,
2520                                 bond->master_ip, 0);
2521         }
2522
2523         list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
2524                 vlan_dev = vlan_group_get_device(bond->vlgrp, vlan->vlan_id);
2525                 if (vlan->vlan_ip) {
2526                         bond_arp_send(slave->dev, ARPOP_REPLY, vlan->vlan_ip,
2527                                       vlan->vlan_ip, vlan->vlan_id);
2528                 }
2529         }
2530 }
2531
2532 static void bond_validate_arp(struct bonding *bond, struct slave *slave, __be32 sip, __be32 tip)
2533 {
2534         int i;
2535         __be32 *targets = bond->params.arp_targets;
2536
2537         for (i = 0; (i < BOND_MAX_ARP_TARGETS) && targets[i]; i++) {
2538                 pr_debug("bva: sip %pI4 tip %pI4 t[%d] %pI4 bhti(tip) %d\n",
2539                          &sip, &tip, i, &targets[i],
2540                          bond_has_this_ip(bond, tip));
2541                 if (sip == targets[i]) {
2542                         if (bond_has_this_ip(bond, tip))
2543                                 slave->last_arp_rx = jiffies;
2544                         return;
2545                 }
2546         }
2547 }
2548
2549 static int bond_arp_rcv(struct sk_buff *skb, struct net_device *dev, struct packet_type *pt, struct net_device *orig_dev)
2550 {
2551         struct arphdr *arp;
2552         struct slave *slave;
2553         struct bonding *bond;
2554         unsigned char *arp_ptr;
2555         __be32 sip, tip;
2556
2557         if (dev->priv_flags & IFF_802_1Q_VLAN) {
2558                 /*
2559                  * When using VLANS and bonding, dev and oriv_dev may be
2560                  * incorrect if the physical interface supports VLAN
2561                  * acceleration.  With this change ARP validation now
2562                  * works for hosts only reachable on the VLAN interface.
2563                  */
2564                 dev = vlan_dev_real_dev(dev);
2565                 orig_dev = dev_get_by_index_rcu(dev_net(skb->dev),skb->skb_iif);
2566         }
2567
2568         if (!(dev->priv_flags & IFF_BONDING) || !(dev->flags & IFF_MASTER))
2569                 goto out;
2570
2571         bond = netdev_priv(dev);
2572         read_lock(&bond->lock);
2573
2574         pr_debug("bond_arp_rcv: bond %s skb->dev %s orig_dev %s\n",
2575                  bond->dev->name, skb->dev ? skb->dev->name : "NULL",
2576                  orig_dev ? orig_dev->name : "NULL");
2577
2578         slave = bond_get_slave_by_dev(bond, orig_dev);
2579         if (!slave || !slave_do_arp_validate(bond, slave))
2580                 goto out_unlock;
2581
2582         if (!pskb_may_pull(skb, arp_hdr_len(dev)))
2583                 goto out_unlock;
2584
2585         arp = arp_hdr(skb);
2586         if (arp->ar_hln != dev->addr_len ||
2587             skb->pkt_type == PACKET_OTHERHOST ||
2588             skb->pkt_type == PACKET_LOOPBACK ||
2589             arp->ar_hrd != htons(ARPHRD_ETHER) ||
2590             arp->ar_pro != htons(ETH_P_IP) ||
2591             arp->ar_pln != 4)
2592                 goto out_unlock;
2593
2594         arp_ptr = (unsigned char *)(arp + 1);
2595         arp_ptr += dev->addr_len;
2596         memcpy(&sip, arp_ptr, 4);
2597         arp_ptr += 4 + dev->addr_len;
2598         memcpy(&tip, arp_ptr, 4);
2599
2600         pr_debug("bond_arp_rcv: %s %s/%d av %d sv %d sip %pI4 tip %pI4\n",
2601                  bond->dev->name, slave->dev->name, slave->state,
2602                  bond->params.arp_validate, slave_do_arp_validate(bond, slave),
2603                  &sip, &tip);
2604
2605         /*
2606          * Backup slaves won't see the ARP reply, but do come through
2607          * here for each ARP probe (so we swap the sip/tip to validate
2608          * the probe).  In a "redundant switch, common router" type of
2609          * configuration, the ARP probe will (hopefully) travel from
2610          * the active, through one switch, the router, then the other
2611          * switch before reaching the backup.
2612          */
2613         if (slave->state == BOND_STATE_ACTIVE)
2614                 bond_validate_arp(bond, slave, sip, tip);
2615         else
2616                 bond_validate_arp(bond, slave, tip, sip);
2617
2618 out_unlock:
2619         read_unlock(&bond->lock);
2620 out:
2621         dev_kfree_skb(skb);
2622         return NET_RX_SUCCESS;
2623 }
2624
2625 /*
2626  * this function is called regularly to monitor each slave's link
2627  * ensuring that traffic is being sent and received when arp monitoring
2628  * is used in load-balancing mode. if the adapter has been dormant, then an
2629  * arp is transmitted to generate traffic. see activebackup_arp_monitor for
2630  * arp monitoring in active backup mode.
2631  */
2632 void bond_loadbalance_arp_mon(struct work_struct *work)
2633 {
2634         struct bonding *bond = container_of(work, struct bonding,
2635                                             arp_work.work);
2636         struct slave *slave, *oldcurrent;
2637         int do_failover = 0;
2638         int delta_in_ticks;
2639         int i;
2640
2641         read_lock(&bond->lock);
2642
2643         delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
2644
2645         if (bond->kill_timers)
2646                 goto out;
2647
2648         if (bond->slave_cnt == 0)
2649                 goto re_arm;
2650
2651         read_lock(&bond->curr_slave_lock);
2652         oldcurrent = bond->curr_active_slave;
2653         read_unlock(&bond->curr_slave_lock);
2654
2655         /* see if any of the previous devices are up now (i.e. they have
2656          * xmt and rcv traffic). the curr_active_slave does not come into
2657          * the picture unless it is null. also, slave->jiffies is not needed
2658          * here because we send an arp on each slave and give a slave as
2659          * long as it needs to get the tx/rx within the delta.
2660          * TODO: what about up/down delay in arp mode? it wasn't here before
2661          *       so it can wait
2662          */
2663         bond_for_each_slave(bond, slave, i) {
2664                 if (slave->link != BOND_LINK_UP) {
2665                         if (time_before_eq(jiffies, dev_trans_start(slave->dev) + delta_in_ticks) &&
2666                             time_before_eq(jiffies, slave->dev->last_rx + delta_in_ticks)) {
2667
2668                                 slave->link  = BOND_LINK_UP;
2669                                 slave->state = BOND_STATE_ACTIVE;
2670
2671                                 /* primary_slave has no meaning in round-robin
2672                                  * mode. the window of a slave being up and
2673                                  * curr_active_slave being null after enslaving
2674                                  * is closed.
2675                                  */
2676                                 if (!oldcurrent) {
2677                                         pr_info("%s: link status definitely up for interface %s, ",
2678                                                 bond->dev->name,
2679                                                 slave->dev->name);
2680                                         do_failover = 1;
2681                                 } else {
2682                                         pr_info("%s: interface %s is now up\n",
2683                                                 bond->dev->name,
2684                                                 slave->dev->name);
2685                                 }
2686                         }
2687                 } else {
2688                         /* slave->link == BOND_LINK_UP */
2689
2690                         /* not all switches will respond to an arp request
2691                          * when the source ip is 0, so don't take the link down
2692                          * if we don't know our ip yet
2693                          */
2694                         if (time_after_eq(jiffies, dev_trans_start(slave->dev) + 2*delta_in_ticks) ||
2695                             (time_after_eq(jiffies, slave->dev->last_rx + 2*delta_in_ticks))) {
2696
2697                                 slave->link  = BOND_LINK_DOWN;
2698                                 slave->state = BOND_STATE_BACKUP;
2699
2700                                 if (slave->link_failure_count < UINT_MAX)
2701                                         slave->link_failure_count++;
2702
2703                                 pr_info("%s: interface %s is now down.\n",
2704                                         bond->dev->name,
2705                                         slave->dev->name);
2706
2707                                 if (slave == oldcurrent)
2708                                         do_failover = 1;
2709                         }
2710                 }
2711
2712                 /* note: if switch is in round-robin mode, all links
2713                  * must tx arp to ensure all links rx an arp - otherwise
2714                  * links may oscillate or not come up at all; if switch is
2715                  * in something like xor mode, there is nothing we can
2716                  * do - all replies will be rx'ed on same link causing slaves
2717                  * to be unstable during low/no traffic periods
2718                  */
2719                 if (IS_UP(slave->dev))
2720                         bond_arp_send_all(bond, slave);
2721         }
2722
2723         if (do_failover) {
2724                 write_lock_bh(&bond->curr_slave_lock);
2725
2726                 bond_select_active_slave(bond);
2727
2728                 write_unlock_bh(&bond->curr_slave_lock);
2729         }
2730
2731 re_arm:
2732         if (bond->params.arp_interval)
2733                 queue_delayed_work(bond->wq, &bond->arp_work, delta_in_ticks);
2734 out:
2735         read_unlock(&bond->lock);
2736 }
2737
2738 /*
2739  * Called to inspect slaves for active-backup mode ARP monitor link state
2740  * changes.  Sets new_link in slaves to specify what action should take
2741  * place for the slave.  Returns 0 if no changes are found, >0 if changes
2742  * to link states must be committed.
2743  *
2744  * Called with bond->lock held for read.
2745  */
2746 static int bond_ab_arp_inspect(struct bonding *bond, int delta_in_ticks)
2747 {
2748         struct slave *slave;
2749         int i, commit = 0;
2750
2751         bond_for_each_slave(bond, slave, i) {
2752                 slave->new_link = BOND_LINK_NOCHANGE;
2753
2754                 if (slave->link != BOND_LINK_UP) {
2755                         if (time_before_eq(jiffies, slave_last_rx(bond, slave) +
2756                                            delta_in_ticks)) {
2757                                 slave->new_link = BOND_LINK_UP;
2758                                 commit++;
2759                         }
2760
2761                         continue;
2762                 }
2763
2764                 /*
2765                  * Give slaves 2*delta after being enslaved or made
2766                  * active.  This avoids bouncing, as the last receive
2767                  * times need a full ARP monitor cycle to be updated.
2768                  */
2769                 if (!time_after_eq(jiffies, slave->jiffies +
2770                                    2 * delta_in_ticks))
2771                         continue;
2772
2773                 /*
2774                  * Backup slave is down if:
2775                  * - No current_arp_slave AND
2776                  * - more than 3*delta since last receive AND
2777                  * - the bond has an IP address
2778                  *
2779                  * Note: a non-null current_arp_slave indicates
2780                  * the curr_active_slave went down and we are
2781                  * searching for a new one; under this condition
2782                  * we only take the curr_active_slave down - this
2783                  * gives each slave a chance to tx/rx traffic
2784                  * before being taken out
2785                  */
2786                 if (slave->state == BOND_STATE_BACKUP &&
2787                     !bond->current_arp_slave &&
2788                     time_after(jiffies, slave_last_rx(bond, slave) +
2789                                3 * delta_in_ticks)) {
2790                         slave->new_link = BOND_LINK_DOWN;
2791                         commit++;
2792                 }
2793
2794                 /*
2795                  * Active slave is down if:
2796                  * - more than 2*delta since transmitting OR
2797                  * - (more than 2*delta since receive AND
2798                  *    the bond has an IP address)
2799                  */
2800                 if ((slave->state == BOND_STATE_ACTIVE) &&
2801                     (time_after_eq(jiffies, dev_trans_start(slave->dev) +
2802                                     2 * delta_in_ticks) ||
2803                       (time_after_eq(jiffies, slave_last_rx(bond, slave)
2804                                      + 2 * delta_in_ticks)))) {
2805                         slave->new_link = BOND_LINK_DOWN;
2806                         commit++;
2807                 }
2808         }
2809
2810         return commit;
2811 }
2812
2813 /*
2814  * Called to commit link state changes noted by inspection step of
2815  * active-backup mode ARP monitor.
2816  *
2817  * Called with RTNL and bond->lock for read.
2818  */
2819 static void bond_ab_arp_commit(struct bonding *bond, int delta_in_ticks)
2820 {
2821         struct slave *slave;
2822         int i;
2823
2824         bond_for_each_slave(bond, slave, i) {
2825                 switch (slave->new_link) {
2826                 case BOND_LINK_NOCHANGE:
2827                         continue;
2828
2829                 case BOND_LINK_UP:
2830                         if ((!bond->curr_active_slave &&
2831                              time_before_eq(jiffies,
2832                                             dev_trans_start(slave->dev) +
2833                                             delta_in_ticks)) ||
2834                             bond->curr_active_slave != slave) {
2835                                 slave->link = BOND_LINK_UP;
2836                                 bond->current_arp_slave = NULL;
2837
2838                                 pr_info("%s: link status definitely up for interface %s.\n",
2839                                         bond->dev->name, slave->dev->name);
2840
2841                                 if (!bond->curr_active_slave ||
2842                                     (slave == bond->primary_slave))
2843                                         goto do_failover;
2844
2845                         }
2846
2847                         continue;
2848
2849                 case BOND_LINK_DOWN:
2850                         if (slave->link_failure_count < UINT_MAX)
2851                                 slave->link_failure_count++;
2852
2853                         slave->link = BOND_LINK_DOWN;
2854                         bond_set_slave_inactive_flags(slave);
2855
2856                         pr_info("%s: link status definitely down for interface %s, disabling it\n",
2857                                 bond->dev->name, slave->dev->name);
2858
2859                         if (slave == bond->curr_active_slave) {
2860                                 bond->current_arp_slave = NULL;
2861                                 goto do_failover;
2862                         }
2863
2864                         continue;
2865
2866                 default:
2867                         pr_err("%s: impossible: new_link %d on slave %s\n",
2868                                bond->dev->name, slave->new_link,
2869                                slave->dev->name);
2870                         continue;
2871                 }
2872
2873 do_failover:
2874                 ASSERT_RTNL();
2875                 write_lock_bh(&bond->curr_slave_lock);
2876                 bond_select_active_slave(bond);
2877                 write_unlock_bh(&bond->curr_slave_lock);
2878         }
2879
2880         bond_set_carrier(bond);
2881 }
2882
2883 /*
2884  * Send ARP probes for active-backup mode ARP monitor.
2885  *
2886  * Called with bond->lock held for read.
2887  */
2888 static void bond_ab_arp_probe(struct bonding *bond)
2889 {
2890         struct slave *slave;
2891         int i;
2892
2893         read_lock(&bond->curr_slave_lock);
2894
2895         if (bond->current_arp_slave && bond->curr_active_slave)
2896                 pr_info("PROBE: c_arp %s && cas %s BAD\n",
2897                         bond->current_arp_slave->dev->name,
2898                         bond->curr_active_slave->dev->name);
2899
2900         if (bond->curr_active_slave) {
2901                 bond_arp_send_all(bond, bond->curr_active_slave);
2902                 read_unlock(&bond->curr_slave_lock);
2903                 return;
2904         }
2905
2906         read_unlock(&bond->curr_slave_lock);
2907
2908         /* if we don't have a curr_active_slave, search for the next available
2909          * backup slave from the current_arp_slave and make it the candidate
2910          * for becoming the curr_active_slave
2911          */
2912
2913         if (!bond->current_arp_slave) {
2914                 bond->current_arp_slave = bond->first_slave;
2915                 if (!bond->current_arp_slave)
2916                         return;
2917         }
2918
2919         bond_set_slave_inactive_flags(bond->current_arp_slave);
2920
2921         /* search for next candidate */
2922         bond_for_each_slave_from(bond, slave, i, bond->current_arp_slave->next) {
2923                 if (IS_UP(slave->dev)) {
2924                         slave->link = BOND_LINK_BACK;
2925                         bond_set_slave_active_flags(slave);
2926                         bond_arp_send_all(bond, slave);
2927                         slave->jiffies = jiffies;
2928                         bond->current_arp_slave = slave;
2929                         break;
2930                 }
2931
2932                 /* if the link state is up at this point, we
2933                  * mark it down - this can happen if we have
2934                  * simultaneous link failures and
2935                  * reselect_active_interface doesn't make this
2936                  * one the current slave so it is still marked
2937                  * up when it is actually down
2938                  */
2939                 if (slave->link == BOND_LINK_UP) {
2940                         slave->link = BOND_LINK_DOWN;
2941                         if (slave->link_failure_count < UINT_MAX)
2942                                 slave->link_failure_count++;
2943
2944                         bond_set_slave_inactive_flags(slave);
2945
2946                         pr_info("%s: backup interface %s is now down.\n",
2947                                 bond->dev->name, slave->dev->name);
2948                 }
2949         }
2950 }
2951
2952 void bond_activebackup_arp_mon(struct work_struct *work)
2953 {
2954         struct bonding *bond = container_of(work, struct bonding,
2955                                             arp_work.work);
2956         int delta_in_ticks;
2957
2958         read_lock(&bond->lock);
2959
2960         if (bond->kill_timers)
2961                 goto out;
2962
2963         delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
2964
2965         if (bond->slave_cnt == 0)
2966                 goto re_arm;
2967
2968         if (bond->send_grat_arp) {
2969                 read_lock(&bond->curr_slave_lock);
2970                 bond_send_gratuitous_arp(bond);
2971                 read_unlock(&bond->curr_slave_lock);
2972         }
2973
2974         if (bond->send_unsol_na) {
2975                 read_lock(&bond->curr_slave_lock);
2976                 bond_send_unsolicited_na(bond);
2977                 read_unlock(&bond->curr_slave_lock);
2978         }
2979
2980         if (bond_ab_arp_inspect(bond, delta_in_ticks)) {
2981                 read_unlock(&bond->lock);
2982                 rtnl_lock();
2983                 read_lock(&bond->lock);
2984
2985                 bond_ab_arp_commit(bond, delta_in_ticks);
2986
2987                 read_unlock(&bond->lock);
2988                 rtnl_unlock();
2989                 read_lock(&bond->lock);
2990         }
2991
2992         bond_ab_arp_probe(bond);
2993
2994 re_arm:
2995         if (bond->params.arp_interval)
2996                 queue_delayed_work(bond->wq, &bond->arp_work, delta_in_ticks);
2997 out:
2998         read_unlock(&bond->lock);
2999 }
3000
3001 /*------------------------------ proc/seq_file-------------------------------*/
3002
3003 #ifdef CONFIG_PROC_FS
3004
3005 static void *bond_info_seq_start(struct seq_file *seq, loff_t *pos)
3006         __acquires(&dev_base_lock)
3007         __acquires(&bond->lock)
3008 {
3009         struct bonding *bond = seq->private;
3010         loff_t off = 0;
3011         struct slave *slave;
3012         int i;
3013
3014         /* make sure the bond won't be taken away */
3015         read_lock(&dev_base_lock);
3016         read_lock(&bond->lock);
3017
3018         if (*pos == 0)
3019                 return SEQ_START_TOKEN;
3020
3021         bond_for_each_slave(bond, slave, i) {
3022                 if (++off == *pos)
3023                         return slave;
3024         }
3025
3026         return NULL;
3027 }
3028
3029 static void *bond_info_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3030 {
3031         struct bonding *bond = seq->private;
3032         struct slave *slave = v;
3033
3034         ++*pos;
3035         if (v == SEQ_START_TOKEN)
3036                 return bond->first_slave;
3037
3038         slave = slave->next;
3039
3040         return (slave == bond->first_slave) ? NULL : slave;
3041 }
3042
3043 static void bond_info_seq_stop(struct seq_file *seq, void *v)
3044         __releases(&bond->lock)
3045         __releases(&dev_base_lock)
3046 {
3047         struct bonding *bond = seq->private;
3048
3049         read_unlock(&bond->lock);
3050         read_unlock(&dev_base_lock);
3051 }
3052
3053 static void bond_info_show_master(struct seq_file *seq)
3054 {
3055         struct bonding *bond = seq->private;
3056         struct slave *curr;
3057         int i;
3058
3059         read_lock(&bond->curr_slave_lock);
3060         curr = bond->curr_active_slave;
3061         read_unlock(&bond->curr_slave_lock);
3062
3063         seq_printf(seq, "Bonding Mode: %s",
3064                    bond_mode_name(bond->params.mode));
3065
3066         if (bond->params.mode == BOND_MODE_ACTIVEBACKUP &&
3067             bond->params.fail_over_mac)
3068                 seq_printf(seq, " (fail_over_mac %s)",
3069                    fail_over_mac_tbl[bond->params.fail_over_mac].modename);
3070
3071         seq_printf(seq, "\n");
3072
3073         if (bond->params.mode == BOND_MODE_XOR ||
3074                 bond->params.mode == BOND_MODE_8023AD) {
3075                 seq_printf(seq, "Transmit Hash Policy: %s (%d)\n",
3076                         xmit_hashtype_tbl[bond->params.xmit_policy].modename,
3077                         bond->params.xmit_policy);
3078         }
3079
3080         if (USES_PRIMARY(bond->params.mode)) {
3081                 seq_printf(seq, "Primary Slave: %s",
3082                            (bond->primary_slave) ?
3083                            bond->primary_slave->dev->name : "None");
3084                 if (bond->primary_slave)
3085                         seq_printf(seq, " (primary_reselect %s)",
3086                    pri_reselect_tbl[bond->params.primary_reselect].modename);
3087
3088                 seq_printf(seq, "\nCurrently Active Slave: %s\n",
3089                            (curr) ? curr->dev->name : "None");
3090         }
3091
3092         seq_printf(seq, "MII Status: %s\n", netif_carrier_ok(bond->dev) ?
3093                    "up" : "down");
3094         seq_printf(seq, "MII Polling Interval (ms): %d\n", bond->params.miimon);
3095         seq_printf(seq, "Up Delay (ms): %d\n",
3096                    bond->params.updelay * bond->params.miimon);
3097         seq_printf(seq, "Down Delay (ms): %d\n",
3098                    bond->params.downdelay * bond->params.miimon);
3099
3100
3101         /* ARP information */
3102         if (bond->params.arp_interval > 0) {
3103                 int printed = 0;
3104                 seq_printf(seq, "ARP Polling Interval (ms): %d\n",
3105                                 bond->params.arp_interval);
3106
3107                 seq_printf(seq, "ARP IP target/s (n.n.n.n form):");
3108
3109                 for (i = 0; (i < BOND_MAX_ARP_TARGETS); i++) {
3110                         if (!bond->params.arp_targets[i])
3111                                 break;
3112                         if (printed)
3113                                 seq_printf(seq, ",");
3114                         seq_printf(seq, " %pI4", &bond->params.arp_targets[i]);
3115                         printed = 1;
3116                 }
3117                 seq_printf(seq, "\n");
3118         }
3119
3120         if (bond->params.mode == BOND_MODE_8023AD) {
3121                 struct ad_info ad_info;
3122
3123                 seq_puts(seq, "\n802.3ad info\n");
3124                 seq_printf(seq, "LACP rate: %s\n",
3125                            (bond->params.lacp_fast) ? "fast" : "slow");
3126                 seq_printf(seq, "Aggregator selection policy (ad_select): %s\n",
3127                            ad_select_tbl[bond->params.ad_select].modename);
3128
3129                 if (bond_3ad_get_active_agg_info(bond, &ad_info)) {
3130                         seq_printf(seq, "bond %s has no active aggregator\n",
3131                                    bond->dev->name);
3132                 } else {
3133                         seq_printf(seq, "Active Aggregator Info:\n");
3134
3135                         seq_printf(seq, "\tAggregator ID: %d\n",
3136                                    ad_info.aggregator_id);
3137                         seq_printf(seq, "\tNumber of ports: %d\n",
3138                                    ad_info.ports);
3139                         seq_printf(seq, "\tActor Key: %d\n",
3140                                    ad_info.actor_key);
3141                         seq_printf(seq, "\tPartner Key: %d\n",
3142                                    ad_info.partner_key);
3143                         seq_printf(seq, "\tPartner Mac Address: %pM\n",
3144                                    ad_info.partner_system);
3145                 }
3146         }
3147 }
3148
3149 static void bond_info_show_slave(struct seq_file *seq,
3150                                  const struct slave *slave)
3151 {
3152         struct bonding *bond = seq->private;
3153
3154         seq_printf(seq, "\nSlave Interface: %s\n", slave->dev->name);
3155         seq_printf(seq, "MII Status: %s\n",
3156                    (slave->link == BOND_LINK_UP) ?  "up" : "down");
3157         seq_printf(seq, "Link Failure Count: %u\n",
3158                    slave->link_failure_count);
3159
3160         seq_printf(seq, "Permanent HW addr: %pM\n", slave->perm_hwaddr);
3161
3162         if (bond->params.mode == BOND_MODE_8023AD) {
3163                 const struct aggregator *agg
3164                         = SLAVE_AD_INFO(slave).port.aggregator;
3165
3166                 if (agg)
3167                         seq_printf(seq, "Aggregator ID: %d\n",
3168                                    agg->aggregator_identifier);
3169                 else
3170                         seq_puts(seq, "Aggregator ID: N/A\n");
3171         }
3172 }
3173
3174 static int bond_info_seq_show(struct seq_file *seq, void *v)
3175 {
3176         if (v == SEQ_START_TOKEN) {
3177                 seq_printf(seq, "%s\n", version);
3178                 bond_info_show_master(seq);
3179         } else
3180                 bond_info_show_slave(seq, v);
3181
3182         return 0;
3183 }
3184
3185 static const struct seq_operations bond_info_seq_ops = {
3186         .start = bond_info_seq_start,
3187         .next  = bond_info_seq_next,
3188         .stop  = bond_info_seq_stop,
3189         .show  = bond_info_seq_show,
3190 };
3191
3192 static int bond_info_open(struct inode *inode, struct file *file)
3193 {
3194         struct seq_file *seq;
3195         struct proc_dir_entry *proc;
3196         int res;
3197
3198         res = seq_open(file, &bond_info_seq_ops);
3199         if (!res) {
3200                 /* recover the pointer buried in proc_dir_entry data */
3201                 seq = file->private_data;
3202                 proc = PDE(inode);
3203                 seq->private = proc->data;
3204         }
3205
3206         return res;
3207 }
3208
3209 static const struct file_operations bond_info_fops = {
3210         .owner   = THIS_MODULE,
3211         .open    = bond_info_open,
3212         .read    = seq_read,
3213         .llseek  = seq_lseek,
3214         .release = seq_release,
3215 };
3216
3217 static void bond_create_proc_entry(struct bonding *bond)
3218 {
3219         struct net_device *bond_dev = bond->dev;
3220         struct bond_net *bn = net_generic(dev_net(bond_dev), bond_net_id);
3221
3222         if (bn->proc_dir) {
3223                 bond->proc_entry = proc_create_data(bond_dev->name,
3224                                                     S_IRUGO, bn->proc_dir,
3225                                                     &bond_info_fops, bond);
3226                 if (bond->proc_entry == NULL)
3227                         pr_warning("Warning: Cannot create /proc/net/%s/%s\n",
3228                                    DRV_NAME, bond_dev->name);
3229                 else
3230                         memcpy(bond->proc_file_name, bond_dev->name, IFNAMSIZ);
3231         }
3232 }
3233
3234 static void bond_remove_proc_entry(struct bonding *bond)
3235 {
3236         struct net_device *bond_dev = bond->dev;
3237         struct bond_net *bn = net_generic(dev_net(bond_dev), bond_net_id);
3238
3239         if (bn->proc_dir && bond->proc_entry) {
3240                 remove_proc_entry(bond->proc_file_name, bn->proc_dir);
3241                 memset(bond->proc_file_name, 0, IFNAMSIZ);
3242                 bond->proc_entry = NULL;
3243         }
3244 }
3245
3246 /* Create the bonding directory under /proc/net, if doesn't exist yet.
3247  * Caller must hold rtnl_lock.
3248  */
3249 static void __net_init bond_create_proc_dir(struct bond_net *bn)
3250 {
3251         if (!bn->proc_dir) {
3252                 bn->proc_dir = proc_mkdir(DRV_NAME, bn->net->proc_net);
3253                 if (!bn->proc_dir)
3254                         pr_warning("Warning: cannot create /proc/net/%s\n",
3255                                    DRV_NAME);
3256         }
3257 }
3258
3259 /* Destroy the bonding directory under /proc/net, if empty.
3260  * Caller must hold rtnl_lock.
3261  */
3262 static void __net_exit bond_destroy_proc_dir(struct bond_net *bn)
3263 {
3264         if (bn->proc_dir) {
3265                 remove_proc_entry(DRV_NAME, bn->net->proc_net);
3266                 bn->proc_dir = NULL;
3267         }
3268 }
3269
3270 #else /* !CONFIG_PROC_FS */
3271
3272 static void bond_create_proc_entry(struct bonding *bond)
3273 {
3274 }
3275
3276 static void bond_remove_proc_entry(struct bonding *bond)
3277 {
3278 }
3279
3280 static inline void bond_create_proc_dir(struct bond_net *bn)
3281 {
3282 }
3283
3284 static inline void bond_destroy_proc_dir(struct bond_net *bn)
3285 {
3286 }
3287
3288 #endif /* CONFIG_PROC_FS */
3289
3290
3291 /*-------------------------- netdev event handling --------------------------*/
3292
3293 /*
3294  * Change device name
3295  */
3296 static int bond_event_changename(struct bonding *bond)
3297 {
3298         bond_remove_proc_entry(bond);
3299         bond_create_proc_entry(bond);
3300
3301         return NOTIFY_DONE;
3302 }
3303
3304 static int bond_master_netdev_event(unsigned long event,
3305                                     struct net_device *bond_dev)
3306 {
3307         struct bonding *event_bond = netdev_priv(bond_dev);
3308
3309         switch (event) {
3310         case NETDEV_CHANGENAME:
3311                 return bond_event_changename(event_bond);
3312         default:
3313                 break;
3314         }
3315
3316         return NOTIFY_DONE;
3317 }
3318
3319 static int bond_slave_netdev_event(unsigned long event,
3320                                    struct net_device *slave_dev)
3321 {
3322         struct net_device *bond_dev = slave_dev->master;
3323         struct bonding *bond = netdev_priv(bond_dev);
3324
3325         switch (event) {
3326         case NETDEV_UNREGISTER:
3327                 if (bond_dev) {
3328                         if (bond->setup_by_slave)
3329                                 bond_release_and_destroy(bond_dev, slave_dev);
3330                         else
3331                                 bond_release(bond_dev, slave_dev);
3332                 }
3333                 break;
3334         case NETDEV_CHANGE:
3335                 if (bond->params.mode == BOND_MODE_8023AD || bond_is_lb(bond)) {
3336                         struct slave *slave;
3337
3338                         slave = bond_get_slave_by_dev(bond, slave_dev);
3339                         if (slave) {
3340                                 u16 old_speed = slave->speed;
3341                                 u16 old_duplex = slave->duplex;
3342
3343                                 bond_update_speed_duplex(slave);
3344
3345                                 if (bond_is_lb(bond))
3346                                         break;
3347
3348                                 if (old_speed != slave->speed)
3349                                         bond_3ad_adapter_speed_changed(slave);
3350                                 if (old_duplex != slave->duplex)
3351                                         bond_3ad_adapter_duplex_changed(slave);
3352                         }
3353                 }
3354
3355                 break;
3356         case NETDEV_DOWN:
3357                 /*
3358                  * ... Or is it this?
3359                  */
3360                 break;
3361         case NETDEV_CHANGEMTU:
3362                 /*
3363                  * TODO: Should slaves be allowed to
3364                  * independently alter their MTU?  For
3365                  * an active-backup bond, slaves need
3366                  * not be the same type of device, so
3367                  * MTUs may vary.  For other modes,
3368                  * slaves arguably should have the
3369                  * same MTUs. To do this, we'd need to
3370                  * take over the slave's change_mtu
3371                  * function for the duration of their
3372                  * servitude.
3373                  */
3374                 break;
3375         case NETDEV_CHANGENAME:
3376                 /*
3377                  * TODO: handle changing the primary's name
3378                  */
3379                 break;
3380         case NETDEV_FEAT_CHANGE:
3381                 bond_compute_features(bond);
3382                 break;
3383         default:
3384                 break;
3385         }
3386
3387         return NOTIFY_DONE;
3388 }
3389
3390 /*
3391  * bond_netdev_event: handle netdev notifier chain events.
3392  *
3393  * This function receives events for the netdev chain.  The caller (an
3394  * ioctl handler calling blocking_notifier_call_chain) holds the necessary
3395  * locks for us to safely manipulate the slave devices (RTNL lock,
3396  * dev_probe_lock).
3397  */
3398 static int bond_netdev_event(struct notifier_block *this,
3399                              unsigned long event, void *ptr)
3400 {
3401         struct net_device *event_dev = (struct net_device *)ptr;
3402
3403         pr_debug("event_dev: %s, event: %lx\n",
3404                  event_dev ? event_dev->name : "None",
3405                  event);
3406
3407         if (!(event_dev->priv_flags & IFF_BONDING))
3408                 return NOTIFY_DONE;
3409
3410         if (event_dev->flags & IFF_MASTER) {
3411                 pr_debug("IFF_MASTER\n");
3412                 return bond_master_netdev_event(event, event_dev);
3413         }
3414
3415         if (event_dev->flags & IFF_SLAVE) {
3416                 pr_debug("IFF_SLAVE\n");
3417                 return bond_slave_netdev_event(event, event_dev);
3418         }
3419
3420         return NOTIFY_DONE;
3421 }
3422
3423 /*
3424  * bond_inetaddr_event: handle inetaddr notifier chain events.
3425  *
3426  * We keep track of device IPs primarily to use as source addresses in
3427  * ARP monitor probes (rather than spewing out broadcasts all the time).
3428  *
3429  * We track one IP for the main device (if it has one), plus one per VLAN.
3430  */
3431 static int bond_inetaddr_event(struct notifier_block *this, unsigned long event, void *ptr)
3432 {
3433         struct in_ifaddr *ifa = ptr;
3434         struct net_device *vlan_dev, *event_dev = ifa->ifa_dev->dev;
3435         struct bond_net *bn = net_generic(dev_net(event_dev), bond_net_id);
3436         struct bonding *bond;
3437         struct vlan_entry *vlan;
3438
3439         list_for_each_entry(bond, &bn->dev_list, bond_list) {
3440                 if (bond->dev == event_dev) {
3441                         switch (event) {
3442                         case NETDEV_UP:
3443                                 bond->master_ip = ifa->ifa_local;
3444                                 return NOTIFY_OK;
3445                         case NETDEV_DOWN:
3446                                 bond->master_ip = bond_glean_dev_ip(bond->dev);
3447                                 return NOTIFY_OK;
3448                         default:
3449                                 return NOTIFY_DONE;
3450                         }
3451                 }
3452
3453                 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
3454                         vlan_dev = vlan_group_get_device(bond->vlgrp, vlan->vlan_id);
3455                         if (vlan_dev == event_dev) {
3456                                 switch (event) {
3457                                 case NETDEV_UP:
3458                                         vlan->vlan_ip = ifa->ifa_local;
3459                                         return NOTIFY_OK;
3460                                 case NETDEV_DOWN:
3461                                         vlan->vlan_ip =
3462                                                 bond_glean_dev_ip(vlan_dev);
3463                                         return NOTIFY_OK;
3464                                 default:
3465                                         return NOTIFY_DONE;
3466                                 }
3467                         }
3468                 }
3469         }
3470         return NOTIFY_DONE;
3471 }
3472
3473 static struct notifier_block bond_netdev_notifier = {
3474         .notifier_call = bond_netdev_event,
3475 };
3476
3477 static struct notifier_block bond_inetaddr_notifier = {
3478         .notifier_call = bond_inetaddr_event,
3479 };
3480
3481 /*-------------------------- Packet type handling ---------------------------*/
3482
3483 /* register to receive lacpdus on a bond */
3484 static void bond_register_lacpdu(struct bonding *bond)
3485 {
3486         struct packet_type *pk_type = &(BOND_AD_INFO(bond).ad_pkt_type);
3487
3488         /* initialize packet type */
3489         pk_type->type = PKT_TYPE_LACPDU;
3490         pk_type->dev = bond->dev;
3491         pk_type->func = bond_3ad_lacpdu_recv;
3492
3493         dev_add_pack(pk_type);
3494 }
3495
3496 /* unregister to receive lacpdus on a bond */
3497 static void bond_unregister_lacpdu(struct bonding *bond)
3498 {
3499         dev_remove_pack(&(BOND_AD_INFO(bond).ad_pkt_type));
3500 }
3501
3502 void bond_register_arp(struct bonding *bond)
3503 {
3504         struct packet_type *pt = &bond->arp_mon_pt;
3505
3506         if (pt->type)
3507                 return;
3508
3509         pt->type = htons(ETH_P_ARP);
3510         pt->dev = bond->dev;
3511         pt->func = bond_arp_rcv;
3512         dev_add_pack(pt);
3513 }
3514
3515 void bond_unregister_arp(struct bonding *bond)
3516 {
3517         struct packet_type *pt = &bond->arp_mon_pt;
3518
3519         dev_remove_pack(pt);
3520         pt->type = 0;
3521 }
3522
3523 /*---------------------------- Hashing Policies -----------------------------*/
3524
3525 /*
3526  * Hash for the output device based upon layer 2 and layer 3 data. If
3527  * the packet is not IP mimic bond_xmit_hash_policy_l2()
3528  */
3529 static int bond_xmit_hash_policy_l23(struct sk_buff *skb, int count)
3530 {
3531         struct ethhdr *data = (struct ethhdr *)skb->data;
3532         struct iphdr *iph = ip_hdr(skb);
3533
3534         if (skb->protocol == htons(ETH_P_IP)) {
3535                 return ((ntohl(iph->saddr ^ iph->daddr) & 0xffff) ^
3536                         (data->h_dest[5] ^ data->h_source[5])) % count;
3537         }
3538
3539         return (data->h_dest[5] ^ data->h_source[5]) % count;
3540 }
3541
3542 /*
3543  * Hash for the output device based upon layer 3 and layer 4 data. If
3544  * the packet is a frag or not TCP or UDP, just use layer 3 data.  If it is
3545  * altogether not IP, mimic bond_xmit_hash_policy_l2()
3546  */
3547 static int bond_xmit_hash_policy_l34(struct sk_buff *skb, int count)
3548 {
3549         struct ethhdr *data = (struct ethhdr *)skb->data;
3550         struct iphdr *iph = ip_hdr(skb);
3551         __be16 *layer4hdr = (__be16 *)((u32 *)iph + iph->ihl);
3552         int layer4_xor = 0;
3553
3554         if (skb->protocol == htons(ETH_P_IP)) {
3555                 if (!(iph->frag_off & htons(IP_MF|IP_OFFSET)) &&
3556                     (iph->protocol == IPPROTO_TCP ||
3557                      iph->protocol == IPPROTO_UDP)) {
3558                         layer4_xor = ntohs((*layer4hdr ^ *(layer4hdr + 1)));
3559                 }
3560                 return (layer4_xor ^
3561                         ((ntohl(iph->saddr ^ iph->daddr)) & 0xffff)) % count;
3562
3563         }
3564
3565         return (data->h_dest[5] ^ data->h_source[5]) % count;
3566 }
3567
3568 /*
3569  * Hash for the output device based upon layer 2 data
3570  */
3571 static int bond_xmit_hash_policy_l2(struct sk_buff *skb, int count)
3572 {
3573         struct ethhdr *data = (struct ethhdr *)skb->data;
3574
3575         return (data->h_dest[5] ^ data->h_source[5]) % count;
3576 }
3577
3578 /*-------------------------- Device entry points ----------------------------*/
3579
3580 static int bond_open(struct net_device *bond_dev)
3581 {
3582         struct bonding *bond = netdev_priv(bond_dev);
3583
3584         bond->kill_timers = 0;
3585
3586         if (bond_is_lb(bond)) {
3587                 /* bond_alb_initialize must be called before the timer
3588                  * is started.
3589                  */
3590                 if (bond_alb_initialize(bond, (bond->params.mode == BOND_MODE_ALB))) {
3591                         /* something went wrong - fail the open operation */
3592                         return -ENOMEM;
3593                 }
3594
3595                 INIT_DELAYED_WORK(&bond->alb_work, bond_alb_monitor);
3596                 queue_delayed_work(bond->wq, &bond->alb_work, 0);
3597         }
3598
3599         if (bond->params.miimon) {  /* link check interval, in milliseconds. */
3600                 INIT_DELAYED_WORK(&bond->mii_work, bond_mii_monitor);
3601                 queue_delayed_work(bond->wq, &bond->mii_work, 0);
3602         }
3603
3604         if (bond->params.arp_interval) {  /* arp interval, in milliseconds. */
3605                 if (bond->params.mode == BOND_MODE_ACTIVEBACKUP)
3606                         INIT_DELAYED_WORK(&bond->arp_work,
3607                                           bond_activebackup_arp_mon);
3608                 else
3609                         INIT_DELAYED_WORK(&bond->arp_work,
3610                                           bond_loadbalance_arp_mon);
3611
3612                 queue_delayed_work(bond->wq, &bond->arp_work, 0);
3613                 if (bond->params.arp_validate)
3614                         bond_register_arp(bond);
3615         }
3616
3617         if (bond->params.mode == BOND_MODE_8023AD) {
3618                 INIT_DELAYED_WORK(&bond->ad_work, bond_3ad_state_machine_handler);
3619                 queue_delayed_work(bond->wq, &bond->ad_work, 0);
3620                 /* register to receive LACPDUs */
3621                 bond_register_lacpdu(bond);
3622                 bond_3ad_initiate_agg_selection(bond, 1);
3623         }
3624
3625         return 0;
3626 }
3627
3628 static int bond_close(struct net_device *bond_dev)
3629 {
3630         struct bonding *bond = netdev_priv(bond_dev);
3631
3632         if (bond->params.mode == BOND_MODE_8023AD) {
3633                 /* Unregister the receive of LACPDUs */
3634                 bond_unregister_lacpdu(bond);
3635         }
3636
3637         if (bond->params.arp_validate)
3638                 bond_unregister_arp(bond);
3639
3640         write_lock_bh(&bond->lock);
3641
3642         bond->send_grat_arp = 0;
3643         bond->send_unsol_na = 0;
3644
3645         /* signal timers not to re-arm */
3646         bond->kill_timers = 1;
3647
3648         write_unlock_bh(&bond->lock);
3649
3650         if (bond->params.miimon) {  /* link check interval, in milliseconds. */
3651                 cancel_delayed_work(&bond->mii_work);
3652         }
3653
3654         if (bond->params.arp_interval) {  /* arp interval, in milliseconds. */
3655                 cancel_delayed_work(&bond->arp_work);
3656         }
3657
3658         switch (bond->params.mode) {
3659         case BOND_MODE_8023AD:
3660                 cancel_delayed_work(&bond->ad_work);
3661                 break;
3662         case BOND_MODE_TLB:
3663         case BOND_MODE_ALB:
3664                 cancel_delayed_work(&bond->alb_work);
3665                 break;
3666         default:
3667                 break;
3668         }
3669
3670
3671         if (bond_is_lb(bond)) {
3672                 /* Must be called only after all
3673                  * slaves have been released
3674                  */
3675                 bond_alb_deinitialize(bond);
3676         }
3677
3678         return 0;
3679 }
3680
3681 static struct net_device_stats *bond_get_stats(struct net_device *bond_dev)
3682 {
3683         struct bonding *bond = netdev_priv(bond_dev);
3684         struct net_device_stats *stats = &bond_dev->stats;
3685         struct net_device_stats local_stats;
3686         struct slave *slave;
3687         int i;
3688
3689         memset(&local_stats, 0, sizeof(struct net_device_stats));
3690
3691         read_lock_bh(&bond->lock);
3692
3693         bond_for_each_slave(bond, slave, i) {
3694                 const struct net_device_stats *sstats = dev_get_stats(slave->dev);
3695
3696                 local_stats.rx_packets += sstats->rx_packets;
3697                 local_stats.rx_bytes += sstats->rx_bytes;
3698                 local_stats.rx_errors += sstats->rx_errors;
3699                 local_stats.rx_dropped += sstats->rx_dropped;
3700
3701                 local_stats.tx_packets += sstats->tx_packets;
3702                 local_stats.tx_bytes += sstats->tx_bytes;
3703                 local_stats.tx_errors += sstats->tx_errors;
3704                 local_stats.tx_dropped += sstats->tx_dropped;
3705
3706                 local_stats.multicast += sstats->multicast;
3707                 local_stats.collisions += sstats->collisions;
3708
3709                 local_stats.rx_length_errors += sstats->rx_length_errors;
3710                 local_stats.rx_over_errors += sstats->rx_over_errors;
3711                 local_stats.rx_crc_errors += sstats->rx_crc_errors;
3712                 local_stats.rx_frame_errors += sstats->rx_frame_errors;
3713                 local_stats.rx_fifo_errors += sstats->rx_fifo_errors;
3714                 local_stats.rx_missed_errors += sstats->rx_missed_errors;
3715
3716                 local_stats.tx_aborted_errors += sstats->tx_aborted_errors;
3717                 local_stats.tx_carrier_errors += sstats->tx_carrier_errors;
3718                 local_stats.tx_fifo_errors += sstats->tx_fifo_errors;
3719                 local_stats.tx_heartbeat_errors += sstats->tx_heartbeat_errors;
3720                 local_stats.tx_window_errors += sstats->tx_window_errors;
3721         }
3722
3723         memcpy(stats, &local_stats, sizeof(struct net_device_stats));
3724
3725         read_unlock_bh(&bond->lock);
3726
3727         return stats;
3728 }
3729
3730 static int bond_do_ioctl(struct net_device *bond_dev, struct ifreq *ifr, int cmd)
3731 {
3732         struct net_device *slave_dev = NULL;
3733         struct ifbond k_binfo;
3734         struct ifbond __user *u_binfo = NULL;
3735         struct ifslave k_sinfo;
3736         struct ifslave __user *u_sinfo = NULL;
3737         struct mii_ioctl_data *mii = NULL;
3738         int res = 0;
3739
3740         pr_debug("bond_ioctl: master=%s, cmd=%d\n", bond_dev->name, cmd);
3741
3742         switch (cmd) {
3743         case SIOCGMIIPHY:
3744                 mii = if_mii(ifr);
3745                 if (!mii)
3746                         return -EINVAL;
3747
3748                 mii->phy_id = 0;
3749                 /* Fall Through */
3750         case SIOCGMIIREG:
3751                 /*
3752                  * We do this again just in case we were called by SIOCGMIIREG
3753                  * instead of SIOCGMIIPHY.
3754                  */
3755                 mii = if_mii(ifr);
3756                 if (!mii)
3757                         return -EINVAL;
3758
3759
3760                 if (mii->reg_num == 1) {
3761                         struct bonding *bond = netdev_priv(bond_dev);
3762                         mii->val_out = 0;
3763                         read_lock(&bond->lock);
3764                         read_lock(&bond->curr_slave_lock);
3765                         if (netif_carrier_ok(bond->dev))
3766                                 mii->val_out = BMSR_LSTATUS;
3767
3768                         read_unlock(&bond->curr_slave_lock);
3769                         read_unlock(&bond->lock);
3770                 }
3771
3772                 return 0;
3773         case BOND_INFO_QUERY_OLD:
3774         case SIOCBONDINFOQUERY:
3775                 u_binfo = (struct ifbond __user *)ifr->ifr_data;
3776
3777                 if (copy_from_user(&k_binfo, u_binfo, sizeof(ifbond)))
3778                         return -EFAULT;
3779
3780                 res = bond_info_query(bond_dev, &k_binfo);
3781                 if (res == 0 &&
3782                     copy_to_user(u_binfo, &k_binfo, sizeof(ifbond)))
3783                         return -EFAULT;
3784
3785                 return res;
3786         case BOND_SLAVE_INFO_QUERY_OLD:
3787         case SIOCBONDSLAVEINFOQUERY:
3788                 u_sinfo = (struct ifslave __user *)ifr->ifr_data;
3789
3790                 if (copy_from_user(&k_sinfo, u_sinfo, sizeof(ifslave)))
3791                         return -EFAULT;
3792
3793                 res = bond_slave_info_query(bond_dev, &k_sinfo);
3794                 if (res == 0 &&
3795                     copy_to_user(u_sinfo, &k_sinfo, sizeof(ifslave)))
3796                         return -EFAULT;
3797
3798                 return res;
3799         default:
3800                 /* Go on */
3801                 break;
3802         }
3803
3804         if (!capable(CAP_NET_ADMIN))
3805                 return -EPERM;
3806
3807         slave_dev = dev_get_by_name(dev_net(bond_dev), ifr->ifr_slave);
3808
3809         pr_debug("slave_dev=%p:\n", slave_dev);
3810
3811         if (!slave_dev)
3812                 res = -ENODEV;
3813         else {
3814                 pr_debug("slave_dev->name=%s:\n", slave_dev->name);
3815                 switch (cmd) {
3816                 case BOND_ENSLAVE_OLD:
3817                 case SIOCBONDENSLAVE:
3818                         res = bond_enslave(bond_dev, slave_dev);
3819                         break;
3820                 case BOND_RELEASE_OLD:
3821                 case SIOCBONDRELEASE:
3822                         res = bond_release(bond_dev, slave_dev);
3823                         break;
3824                 case BOND_SETHWADDR_OLD:
3825                 case SIOCBONDSETHWADDR:
3826                         res = bond_sethwaddr(bond_dev, slave_dev);
3827                         break;
3828                 case BOND_CHANGE_ACTIVE_OLD:
3829                 case SIOCBONDCHANGEACTIVE:
3830                         res = bond_ioctl_change_active(bond_dev, slave_dev);
3831                         break;
3832                 default:
3833                         res = -EOPNOTSUPP;
3834                 }
3835
3836                 dev_put(slave_dev);
3837         }
3838
3839         return res;
3840 }
3841
3842 static bool bond_addr_in_mc_list(unsigned char *addr,
3843                                  struct netdev_hw_addr_list *list,
3844                                  int addrlen)
3845 {
3846         struct netdev_hw_addr *ha;
3847
3848         netdev_hw_addr_list_for_each(ha, list)
3849                 if (!memcmp(ha->addr, addr, addrlen))
3850                         return true;
3851
3852         return false;
3853 }
3854
3855 static void bond_set_multicast_list(struct net_device *bond_dev)
3856 {
3857         struct bonding *bond = netdev_priv(bond_dev);
3858         struct netdev_hw_addr *ha;
3859         bool found;
3860
3861         /*
3862          * Do promisc before checking multicast_mode
3863          */
3864         if ((bond_dev->flags & IFF_PROMISC) && !(bond->flags & IFF_PROMISC))
3865                 /*
3866                  * FIXME: Need to handle the error when one of the multi-slaves
3867                  * encounters error.
3868                  */
3869                 bond_set_promiscuity(bond, 1);
3870
3871
3872         if (!(bond_dev->flags & IFF_PROMISC) && (bond->flags & IFF_PROMISC))
3873                 bond_set_promiscuity(bond, -1);
3874
3875
3876         /* set allmulti flag to slaves */
3877         if ((bond_dev->flags & IFF_ALLMULTI) && !(bond->flags & IFF_ALLMULTI))
3878                 /*
3879                  * FIXME: Need to handle the error when one of the multi-slaves
3880                  * encounters error.
3881                  */
3882                 bond_set_allmulti(bond, 1);
3883
3884
3885         if (!(bond_dev->flags & IFF_ALLMULTI) && (bond->flags & IFF_ALLMULTI))
3886                 bond_set_allmulti(bond, -1);
3887
3888
3889         read_lock(&bond->lock);
3890
3891         bond->flags = bond_dev->flags;
3892
3893         /* looking for addresses to add to slaves' mc list */
3894         netdev_for_each_mc_addr(ha, bond_dev) {
3895                 found = bond_addr_in_mc_list(ha->addr, &bond->mc_list,
3896                                              bond_dev->addr_len);
3897                 if (!found)
3898                         bond_mc_add(bond, ha->addr);
3899         }
3900
3901         /* looking for addresses to delete from slaves' list */
3902         netdev_hw_addr_list_for_each(ha, &bond->mc_list) {
3903                 found = bond_addr_in_mc_list(ha->addr, &bond_dev->mc,
3904                                              bond_dev->addr_len);
3905                 if (!found)
3906                         bond_mc_del(bond, ha->addr);
3907         }
3908
3909         /* save master's multicast list */
3910         __hw_addr_flush(&bond->mc_list);
3911         __hw_addr_add_multiple(&bond->mc_list, &bond_dev->mc,
3912                                bond_dev->addr_len, NETDEV_HW_ADDR_T_MULTICAST);
3913
3914         read_unlock(&bond->lock);
3915 }
3916
3917 static int bond_neigh_setup(struct net_device *dev, struct neigh_parms *parms)
3918 {
3919         struct bonding *bond = netdev_priv(dev);
3920         struct slave *slave = bond->first_slave;
3921
3922         if (slave) {
3923                 const struct net_device_ops *slave_ops
3924                         = slave->dev->netdev_ops;
3925                 if (slave_ops->ndo_neigh_setup)
3926                         return slave_ops->ndo_neigh_setup(slave->dev, parms);
3927         }
3928         return 0;
3929 }
3930
3931 /*
3932  * Change the MTU of all of a master's slaves to match the master
3933  */
3934 static int bond_change_mtu(struct net_device *bond_dev, int new_mtu)
3935 {
3936         struct bonding *bond = netdev_priv(bond_dev);
3937         struct slave *slave, *stop_at;
3938         int res = 0;
3939         int i;
3940
3941         pr_debug("bond=%p, name=%s, new_mtu=%d\n", bond,
3942                  (bond_dev ? bond_dev->name : "None"), new_mtu);
3943
3944         /* Can't hold bond->lock with bh disabled here since
3945          * some base drivers panic. On the other hand we can't
3946          * hold bond->lock without bh disabled because we'll
3947          * deadlock. The only solution is to rely on the fact
3948          * that we're under rtnl_lock here, and the slaves
3949          * list won't change. This doesn't solve the problem
3950          * of setting the slave's MTU while it is
3951          * transmitting, but the assumption is that the base
3952          * driver can handle that.
3953          *
3954          * TODO: figure out a way to safely iterate the slaves
3955          * list, but without holding a lock around the actual
3956          * call to the base driver.
3957          */
3958
3959         bond_for_each_slave(bond, slave, i) {
3960                 pr_debug("s %p s->p %p c_m %p\n",
3961                          slave,
3962                          slave->prev,
3963                          slave->dev->netdev_ops->ndo_change_mtu);
3964
3965                 res = dev_set_mtu(slave->dev, new_mtu);
3966
3967                 if (res) {
3968                         /* If we failed to set the slave's mtu to the new value
3969                          * we must abort the operation even in ACTIVE_BACKUP
3970                          * mode, because if we allow the backup slaves to have
3971                          * different mtu values than the active slave we'll
3972                          * need to change their mtu when doing a failover. That
3973                          * means changing their mtu from timer context, which
3974                          * is probably not a good idea.
3975                          */
3976                         pr_debug("err %d %s\n", res, slave->dev->name);
3977                         goto unwind;
3978                 }
3979         }
3980
3981         bond_dev->mtu = new_mtu;
3982
3983         return 0;
3984
3985 unwind:
3986         /* unwind from head to the slave that failed */
3987         stop_at = slave;
3988         bond_for_each_slave_from_to(bond, slave, i, bond->first_slave, stop_at) {
3989                 int tmp_res;
3990
3991                 tmp_res = dev_set_mtu(slave->dev, bond_dev->mtu);
3992                 if (tmp_res) {
3993                         pr_debug("unwind err %d dev %s\n",
3994                                  tmp_res, slave->dev->name);
3995                 }
3996         }
3997
3998         return res;
3999 }
4000
4001 /*
4002  * Change HW address
4003  *
4004  * Note that many devices must be down to change the HW address, and
4005  * downing the master releases all slaves.  We can make bonds full of
4006  * bonding devices to test this, however.
4007  */
4008 static int bond_set_mac_address(struct net_device *bond_dev, void *addr)
4009 {
4010         struct bonding *bond = netdev_priv(bond_dev);
4011         struct sockaddr *sa = addr, tmp_sa;
4012         struct slave *slave, *stop_at;
4013         int res = 0;
4014         int i;
4015
4016         if (bond->params.mode == BOND_MODE_ALB)
4017                 return bond_alb_set_mac_address(bond_dev, addr);
4018
4019
4020         pr_debug("bond=%p, name=%s\n",
4021                  bond, bond_dev ? bond_dev->name : "None");
4022
4023         /*
4024          * If fail_over_mac is set to active, do nothing and return
4025          * success.  Returning an error causes ifenslave to fail.
4026          */
4027         if (bond->params.fail_over_mac == BOND_FOM_ACTIVE)
4028                 return 0;
4029
4030         if (!is_valid_ether_addr(sa->sa_data))
4031                 return -EADDRNOTAVAIL;
4032
4033         /* Can't hold bond->lock with bh disabled here since
4034          * some base drivers panic. On the other hand we can't
4035          * hold bond->lock without bh disabled because we'll
4036          * deadlock. The only solution is to rely on the fact
4037          * that we're under rtnl_lock here, and the slaves
4038          * list won't change. This doesn't solve the problem
4039          * of setting the slave's hw address while it is
4040          * transmitting, but the assumption is that the base
4041          * driver can handle that.
4042          *
4043          * TODO: figure out a way to safely iterate the slaves
4044          * list, but without holding a lock around the actual
4045          * call to the base driver.
4046          */
4047
4048         bond_for_each_slave(bond, slave, i) {
4049                 const struct net_device_ops *slave_ops = slave->dev->netdev_ops;
4050                 pr_debug("slave %p %s\n", slave, slave->dev->name);
4051
4052                 if (slave_ops->ndo_set_mac_address == NULL) {
4053                         res = -EOPNOTSUPP;
4054                         pr_debug("EOPNOTSUPP %s\n", slave->dev->name);
4055                         goto unwind;
4056                 }
4057
4058                 res = dev_set_mac_address(slave->dev, addr);
4059                 if (res) {
4060                         /* TODO: consider downing the slave
4061                          * and retry ?
4062                          * User should expect communications
4063                          * breakage anyway until ARP finish
4064                          * updating, so...
4065                          */
4066                         pr_debug("err %d %s\n", res, slave->dev->name);
4067                         goto unwind;
4068                 }
4069         }
4070
4071         /* success */
4072         memcpy(bond_dev->dev_addr, sa->sa_data, bond_dev->addr_len);
4073         return 0;
4074
4075 unwind:
4076         memcpy(tmp_sa.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
4077         tmp_sa.sa_family = bond_dev->type;
4078
4079         /* unwind from head to the slave that failed */
4080         stop_at = slave;
4081         bond_for_each_slave_from_to(bond, slave, i, bond->first_slave, stop_at) {
4082                 int tmp_res;
4083
4084                 tmp_res = dev_set_mac_address(slave->dev, &tmp_sa);
4085                 if (tmp_res) {
4086                         pr_debug("unwind err %d dev %s\n",
4087                                  tmp_res, slave->dev->name);
4088                 }
4089         }
4090
4091         return res;
4092 }
4093
4094 static int bond_xmit_roundrobin(struct sk_buff *skb, struct net_device *bond_dev)
4095 {
4096         struct bonding *bond = netdev_priv(bond_dev);
4097         struct slave *slave, *start_at;
4098         int i, slave_no, res = 1;
4099
4100         read_lock(&bond->lock);
4101
4102         if (!BOND_IS_OK(bond))
4103                 goto out;
4104
4105         /*
4106          * Concurrent TX may collide on rr_tx_counter; we accept that
4107          * as being rare enough not to justify using an atomic op here
4108          */
4109         slave_no = bond->rr_tx_counter++ % bond->slave_cnt;
4110
4111         bond_for_each_slave(bond, slave, i) {
4112                 slave_no--;
4113                 if (slave_no < 0)
4114                         break;
4115         }
4116
4117         start_at = slave;
4118         bond_for_each_slave_from(bond, slave, i, start_at) {
4119                 if (IS_UP(slave->dev) &&
4120                     (slave->link == BOND_LINK_UP) &&
4121                     (slave->state == BOND_STATE_ACTIVE)) {
4122                         res = bond_dev_queue_xmit(bond, skb, slave->dev);
4123                         break;
4124                 }
4125         }
4126
4127 out:
4128         if (res) {
4129                 /* no suitable interface, frame not sent */
4130                 dev_kfree_skb(skb);
4131         }
4132         read_unlock(&bond->lock);
4133         return NETDEV_TX_OK;
4134 }
4135
4136
4137 /*
4138  * in active-backup mode, we know that bond->curr_active_slave is always valid if
4139  * the bond has a usable interface.
4140  */
4141 static int bond_xmit_activebackup(struct sk_buff *skb, struct net_device *bond_dev)
4142 {
4143         struct bonding *bond = netdev_priv(bond_dev);
4144         int res = 1;
4145
4146         read_lock(&bond->lock);
4147         read_lock(&bond->curr_slave_lock);
4148
4149         if (!BOND_IS_OK(bond))
4150                 goto out;
4151
4152         if (!bond->curr_active_slave)
4153                 goto out;
4154
4155         res = bond_dev_queue_xmit(bond, skb, bond->curr_active_slave->dev);
4156
4157 out:
4158         if (res)
4159                 /* no suitable interface, frame not sent */
4160                 dev_kfree_skb(skb);
4161
4162         read_unlock(&bond->curr_slave_lock);
4163         read_unlock(&bond->lock);
4164         return NETDEV_TX_OK;
4165 }
4166
4167 /*
4168  * In bond_xmit_xor() , we determine the output device by using a pre-
4169  * determined xmit_hash_policy(), If the selected device is not enabled,
4170  * find the next active slave.
4171  */
4172 static int bond_xmit_xor(struct sk_buff *skb, struct net_device *bond_dev)
4173 {
4174         struct bonding *bond = netdev_priv(bond_dev);
4175         struct slave *slave, *start_at;
4176         int slave_no;
4177         int i;
4178         int res = 1;
4179
4180         read_lock(&bond->lock);
4181
4182         if (!BOND_IS_OK(bond))
4183                 goto out;
4184
4185         slave_no = bond->xmit_hash_policy(skb, bond->slave_cnt);
4186
4187         bond_for_each_slave(bond, slave, i) {
4188                 slave_no--;
4189                 if (slave_no < 0)
4190                         break;
4191         }
4192
4193         start_at = slave;
4194
4195         bond_for_each_slave_from(bond, slave, i, start_at) {
4196                 if (IS_UP(slave->dev) &&
4197                     (slave->link == BOND_LINK_UP) &&
4198                     (slave->state == BOND_STATE_ACTIVE)) {
4199                         res = bond_dev_queue_xmit(bond, skb, slave->dev);
4200                         break;
4201                 }
4202         }
4203
4204 out:
4205         if (res) {
4206                 /* no suitable interface, frame not sent */
4207                 dev_kfree_skb(skb);
4208         }
4209         read_unlock(&bond->lock);
4210         return NETDEV_TX_OK;
4211 }
4212
4213 /*
4214  * in broadcast mode, we send everything to all usable interfaces.
4215  */
4216 static int bond_xmit_broadcast(struct sk_buff *skb, struct net_device *bond_dev)
4217 {
4218         struct bonding *bond = netdev_priv(bond_dev);
4219         struct slave *slave, *start_at;
4220         struct net_device *tx_dev = NULL;
4221         int i;
4222         int res = 1;
4223
4224         read_lock(&bond->lock);
4225
4226         if (!BOND_IS_OK(bond))
4227                 goto out;
4228
4229         read_lock(&bond->curr_slave_lock);
4230         start_at = bond->curr_active_slave;
4231         read_unlock(&bond->curr_slave_lock);
4232
4233         if (!start_at)
4234                 goto out;
4235
4236         bond_for_each_slave_from(bond, slave, i, start_at) {
4237                 if (IS_UP(slave->dev) &&
4238                     (slave->link == BOND_LINK_UP) &&
4239                     (slave->state == BOND_STATE_ACTIVE)) {
4240                         if (tx_dev) {
4241                                 struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
4242                                 if (!skb2) {
4243                                         pr_err("%s: Error: bond_xmit_broadcast(): skb_clone() failed\n",
4244                                                bond_dev->name);
4245                                         continue;
4246                                 }
4247
4248                                 res = bond_dev_queue_xmit(bond, skb2, tx_dev);
4249                                 if (res) {
4250                                         dev_kfree_skb(skb2);
4251                                         continue;
4252                                 }
4253                         }
4254                         tx_dev = slave->dev;
4255                 }
4256         }
4257
4258         if (tx_dev)
4259                 res = bond_dev_queue_xmit(bond, skb, tx_dev);
4260
4261 out:
4262         if (res)
4263                 /* no suitable interface, frame not sent */
4264                 dev_kfree_skb(skb);
4265
4266         /* frame sent to all suitable interfaces */
4267         read_unlock(&bond->lock);
4268         return NETDEV_TX_OK;
4269 }
4270
4271 /*------------------------- Device initialization ---------------------------*/
4272
4273 static void bond_set_xmit_hash_policy(struct bonding *bond)
4274 {
4275         switch (bond->params.xmit_policy) {
4276         case BOND_XMIT_POLICY_LAYER23:
4277                 bond->xmit_hash_policy = bond_xmit_hash_policy_l23;
4278                 break;
4279         case BOND_XMIT_POLICY_LAYER34:
4280                 bond->xmit_hash_policy = bond_xmit_hash_policy_l34;
4281                 break;
4282         case BOND_XMIT_POLICY_LAYER2:
4283         default:
4284                 bond->xmit_hash_policy = bond_xmit_hash_policy_l2;
4285                 break;
4286         }
4287 }
4288
4289 static netdev_tx_t bond_start_xmit(struct sk_buff *skb, struct net_device *dev)
4290 {
4291         const struct bonding *bond = netdev_priv(dev);
4292
4293         switch (bond->params.mode) {
4294         case BOND_MODE_ROUNDROBIN:
4295                 return bond_xmit_roundrobin(skb, dev);
4296         case BOND_MODE_ACTIVEBACKUP:
4297                 return bond_xmit_activebackup(skb, dev);
4298         case BOND_MODE_XOR:
4299                 return bond_xmit_xor(skb, dev);
4300         case BOND_MODE_BROADCAST:
4301                 return bond_xmit_broadcast(skb, dev);
4302         case BOND_MODE_8023AD:
4303                 return bond_3ad_xmit_xor(skb, dev);
4304         case BOND_MODE_ALB:
4305         case BOND_MODE_TLB:
4306                 return bond_alb_xmit(skb, dev);
4307         default:
4308                 /* Should never happen, mode already checked */
4309                 pr_err("%s: Error: Unknown bonding mode %d\n",
4310                        dev->name, bond->params.mode);
4311                 WARN_ON_ONCE(1);
4312                 dev_kfree_skb(skb);
4313                 return NETDEV_TX_OK;
4314         }
4315 }
4316
4317
4318 /*
4319  * set bond mode specific net device operations
4320  */
4321 void bond_set_mode_ops(struct bonding *bond, int mode)
4322 {
4323         struct net_device *bond_dev = bond->dev;
4324
4325         switch (mode) {
4326         case BOND_MODE_ROUNDROBIN:
4327                 break;
4328         case BOND_MODE_ACTIVEBACKUP:
4329                 break;
4330         case BOND_MODE_XOR:
4331                 bond_set_xmit_hash_policy(bond);
4332                 break;
4333         case BOND_MODE_BROADCAST:
4334                 break;
4335         case BOND_MODE_8023AD:
4336                 bond_set_master_3ad_flags(bond);
4337                 bond_set_xmit_hash_policy(bond);
4338                 break;
4339         case BOND_MODE_ALB:
4340                 bond_set_master_alb_flags(bond);
4341                 /* FALLTHRU */
4342         case BOND_MODE_TLB:
4343                 break;
4344         default:
4345                 /* Should never happen, mode already checked */
4346                 pr_err("%s: Error: Unknown bonding mode %d\n",
4347                        bond_dev->name, mode);
4348                 break;
4349         }
4350 }
4351
4352 static void bond_ethtool_get_drvinfo(struct net_device *bond_dev,
4353                                     struct ethtool_drvinfo *drvinfo)
4354 {
4355         strncpy(drvinfo->driver, DRV_NAME, 32);
4356         strncpy(drvinfo->version, DRV_VERSION, 32);
4357         snprintf(drvinfo->fw_version, 32, "%d", BOND_ABI_VERSION);
4358 }
4359
4360 static const struct ethtool_ops bond_ethtool_ops = {
4361         .get_drvinfo            = bond_ethtool_get_drvinfo,
4362         .get_link               = ethtool_op_get_link,
4363         .get_tx_csum            = ethtool_op_get_tx_csum,
4364         .get_sg                 = ethtool_op_get_sg,
4365         .get_tso                = ethtool_op_get_tso,
4366         .get_ufo                = ethtool_op_get_ufo,
4367         .get_flags              = ethtool_op_get_flags,
4368 };
4369
4370 static const struct net_device_ops bond_netdev_ops = {
4371         .ndo_init               = bond_init,
4372         .ndo_uninit             = bond_uninit,
4373         .ndo_open               = bond_open,
4374         .ndo_stop               = bond_close,
4375         .ndo_start_xmit         = bond_start_xmit,
4376         .ndo_get_stats          = bond_get_stats,
4377         .ndo_do_ioctl           = bond_do_ioctl,
4378         .ndo_set_multicast_list = bond_set_multicast_list,
4379         .ndo_change_mtu         = bond_change_mtu,
4380         .ndo_set_mac_address    = bond_set_mac_address,
4381         .ndo_neigh_setup        = bond_neigh_setup,
4382         .ndo_vlan_rx_register   = bond_vlan_rx_register,
4383         .ndo_vlan_rx_add_vid    = bond_vlan_rx_add_vid,
4384         .ndo_vlan_rx_kill_vid   = bond_vlan_rx_kill_vid,
4385 };
4386
4387 static void bond_setup(struct net_device *bond_dev)
4388 {
4389         struct bonding *bond = netdev_priv(bond_dev);
4390
4391         /* initialize rwlocks */
4392         rwlock_init(&bond->lock);
4393         rwlock_init(&bond->curr_slave_lock);
4394
4395         bond->params = bonding_defaults;
4396
4397         /* Initialize pointers */
4398         bond->dev = bond_dev;
4399         INIT_LIST_HEAD(&bond->vlan_list);
4400
4401         /* Initialize the device entry points */
4402         ether_setup(bond_dev);
4403         bond_dev->netdev_ops = &bond_netdev_ops;
4404         bond_dev->ethtool_ops = &bond_ethtool_ops;
4405         bond_set_mode_ops(bond, bond->params.mode);
4406
4407         bond_dev->destructor = free_netdev;
4408
4409         /* Initialize the device options */
4410         bond_dev->tx_queue_len = 0;
4411         bond_dev->flags |= IFF_MASTER|IFF_MULTICAST;
4412         bond_dev->priv_flags |= IFF_BONDING;
4413         bond_dev->priv_flags &= ~IFF_XMIT_DST_RELEASE;
4414
4415         if (bond->params.arp_interval)
4416                 bond_dev->priv_flags |= IFF_MASTER_ARPMON;
4417
4418         /* At first, we block adding VLANs. That's the only way to
4419          * prevent problems that occur when adding VLANs over an
4420          * empty bond. The block will be removed once non-challenged
4421          * slaves are enslaved.
4422          */
4423         bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
4424
4425         /* don't acquire bond device's netif_tx_lock when
4426          * transmitting */
4427         bond_dev->features |= NETIF_F_LLTX;
4428
4429         /* By default, we declare the bond to be fully
4430          * VLAN hardware accelerated capable. Special
4431          * care is taken in the various xmit functions
4432          * when there are slaves that are not hw accel
4433          * capable
4434          */
4435         bond_dev->features |= (NETIF_F_HW_VLAN_TX |
4436                                NETIF_F_HW_VLAN_RX |
4437                                NETIF_F_HW_VLAN_FILTER);
4438
4439 }
4440
4441 static void bond_work_cancel_all(struct bonding *bond)
4442 {
4443         write_lock_bh(&bond->lock);
4444         bond->kill_timers = 1;
4445         write_unlock_bh(&bond->lock);
4446
4447         if (bond->params.miimon && delayed_work_pending(&bond->mii_work))
4448                 cancel_delayed_work(&bond->mii_work);
4449
4450         if (bond->params.arp_interval && delayed_work_pending(&bond->arp_work))
4451                 cancel_delayed_work(&bond->arp_work);
4452
4453         if (bond->params.mode == BOND_MODE_ALB &&
4454             delayed_work_pending(&bond->alb_work))
4455                 cancel_delayed_work(&bond->alb_work);
4456
4457         if (bond->params.mode == BOND_MODE_8023AD &&
4458             delayed_work_pending(&bond->ad_work))
4459                 cancel_delayed_work(&bond->ad_work);
4460 }
4461
4462 /*
4463 * Destroy a bonding device.
4464 * Must be under rtnl_lock when this function is called.
4465 */
4466 static void bond_uninit(struct net_device *bond_dev)
4467 {
4468         struct bonding *bond = netdev_priv(bond_dev);
4469
4470         /* Release the bonded slaves */
4471         bond_release_all(bond_dev);
4472
4473         list_del(&bond->bond_list);
4474
4475         bond_work_cancel_all(bond);
4476
4477         bond_remove_proc_entry(bond);
4478
4479         if (bond->wq)
4480                 destroy_workqueue(bond->wq);
4481
4482         __hw_addr_flush(&bond->mc_list);
4483 }
4484
4485 /*------------------------- Module initialization ---------------------------*/
4486
4487 /*
4488  * Convert string input module parms.  Accept either the
4489  * number of the mode or its string name.  A bit complicated because
4490  * some mode names are substrings of other names, and calls from sysfs
4491  * may have whitespace in the name (trailing newlines, for example).
4492  */
4493 int bond_parse_parm(const char *buf, const struct bond_parm_tbl *tbl)
4494 {
4495         int modeint = -1, i, rv;
4496         char *p, modestr[BOND_MAX_MODENAME_LEN + 1] = { 0, };
4497
4498         for (p = (char *)buf; *p; p++)
4499                 if (!(isdigit(*p) || isspace(*p)))
4500                         break;
4501
4502         if (*p)
4503                 rv = sscanf(buf, "%20s", modestr);
4504         else
4505                 rv = sscanf(buf, "%d", &modeint);
4506
4507         if (!rv)
4508                 return -1;
4509
4510         for (i = 0; tbl[i].modename; i++) {
4511                 if (modeint == tbl[i].mode)
4512                         return tbl[i].mode;
4513                 if (strcmp(modestr, tbl[i].modename) == 0)
4514                         return tbl[i].mode;
4515         }
4516
4517         return -1;
4518 }
4519
4520 static int bond_check_params(struct bond_params *params)
4521 {
4522         int arp_validate_value, fail_over_mac_value, primary_reselect_value;
4523
4524         /*
4525          * Convert string parameters.
4526          */
4527         if (mode) {
4528                 bond_mode = bond_parse_parm(mode, bond_mode_tbl);
4529                 if (bond_mode == -1) {
4530                         pr_err("Error: Invalid bonding mode \"%s\"\n",
4531                                mode == NULL ? "NULL" : mode);
4532                         return -EINVAL;
4533                 }
4534         }
4535
4536         if (xmit_hash_policy) {
4537                 if ((bond_mode != BOND_MODE_XOR) &&
4538                     (bond_mode != BOND_MODE_8023AD)) {
4539                         pr_info("xmit_hash_policy param is irrelevant in mode %s\n",
4540                                bond_mode_name(bond_mode));
4541                 } else {
4542                         xmit_hashtype = bond_parse_parm(xmit_hash_policy,
4543                                                         xmit_hashtype_tbl);
4544                         if (xmit_hashtype == -1) {
4545                                 pr_err("Error: Invalid xmit_hash_policy \"%s\"\n",
4546                                        xmit_hash_policy == NULL ? "NULL" :
4547                                        xmit_hash_policy);
4548                                 return -EINVAL;
4549                         }
4550                 }
4551         }
4552
4553         if (lacp_rate) {
4554                 if (bond_mode != BOND_MODE_8023AD) {
4555                         pr_info("lacp_rate param is irrelevant in mode %s\n",
4556                                 bond_mode_name(bond_mode));
4557                 } else {
4558                         lacp_fast = bond_parse_parm(lacp_rate, bond_lacp_tbl);
4559                         if (lacp_fast == -1) {
4560                                 pr_err("Error: Invalid lacp rate \"%s\"\n",
4561                                        lacp_rate == NULL ? "NULL" : lacp_rate);
4562                                 return -EINVAL;
4563                         }
4564                 }
4565         }
4566
4567         if (ad_select) {
4568                 params->ad_select = bond_parse_parm(ad_select, ad_select_tbl);
4569                 if (params->ad_select == -1) {
4570                         pr_err("Error: Invalid ad_select \"%s\"\n",
4571                                ad_select == NULL ? "NULL" : ad_select);
4572                         return -EINVAL;
4573                 }
4574
4575                 if (bond_mode != BOND_MODE_8023AD) {
4576                         pr_warning("ad_select param only affects 802.3ad mode\n");
4577                 }
4578         } else {
4579                 params->ad_select = BOND_AD_STABLE;
4580         }
4581
4582         if (max_bonds < 0) {
4583                 pr_warning("Warning: max_bonds (%d) not in range %d-%d, so it was reset to BOND_DEFAULT_MAX_BONDS (%d)\n",
4584                            max_bonds, 0, INT_MAX, BOND_DEFAULT_MAX_BONDS);
4585                 max_bonds = BOND_DEFAULT_MAX_BONDS;
4586         }
4587
4588         if (miimon < 0) {
4589                 pr_warning("Warning: miimon module parameter (%d), not in range 0-%d, so it was reset to %d\n",
4590                            miimon, INT_MAX, BOND_LINK_MON_INTERV);
4591                 miimon = BOND_LINK_MON_INTERV;
4592         }
4593
4594         if (updelay < 0) {
4595                 pr_warning("Warning: updelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4596                            updelay, INT_MAX);
4597                 updelay = 0;
4598         }
4599
4600         if (downdelay < 0) {
4601                 pr_warning("Warning: downdelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4602                            downdelay, INT_MAX);
4603                 downdelay = 0;
4604         }
4605
4606         if ((use_carrier != 0) && (use_carrier != 1)) {
4607                 pr_warning("Warning: use_carrier module parameter (%d), not of valid value (0/1), so it was set to 1\n",
4608                            use_carrier);
4609                 use_carrier = 1;
4610         }
4611
4612         if (num_grat_arp < 0 || num_grat_arp > 255) {
4613                 pr_warning("Warning: num_grat_arp (%d) not in range 0-255 so it was reset to 1\n",
4614                            num_grat_arp);
4615                 num_grat_arp = 1;
4616         }
4617
4618         if (num_unsol_na < 0 || num_unsol_na > 255) {
4619                 pr_warning("Warning: num_unsol_na (%d) not in range 0-255 so it was reset to 1\n",
4620                            num_unsol_na);
4621                 num_unsol_na = 1;
4622         }
4623
4624         /* reset values for 802.3ad */
4625         if (bond_mode == BOND_MODE_8023AD) {
4626                 if (!miimon) {
4627                         pr_warning("Warning: miimon must be specified, otherwise bonding will not detect link failure, speed and duplex which are essential for 802.3ad operation\n");
4628                         pr_warning("Forcing miimon to 100msec\n");
4629                         miimon = 100;
4630                 }
4631         }
4632
4633         /* reset values for TLB/ALB */
4634         if ((bond_mode == BOND_MODE_TLB) ||
4635             (bond_mode == BOND_MODE_ALB)) {
4636                 if (!miimon) {
4637                         pr_warning("Warning: miimon must be specified, otherwise bonding will not detect link failure and link speed which are essential for TLB/ALB load balancing\n");
4638                         pr_warning("Forcing miimon to 100msec\n");
4639                         miimon = 100;
4640                 }
4641         }
4642
4643         if (bond_mode == BOND_MODE_ALB) {
4644                 pr_notice("In ALB mode you might experience client disconnections upon reconnection of a link if the bonding module updelay parameter (%d msec) is incompatible with the forwarding delay time of the switch\n",
4645                           updelay);
4646         }
4647
4648         if (!miimon) {
4649                 if (updelay || downdelay) {
4650                         /* just warn the user the up/down delay will have
4651                          * no effect since miimon is zero...
4652                          */
4653                         pr_warning("Warning: miimon module parameter not set and updelay (%d) or downdelay (%d) module parameter is set; updelay and downdelay have no effect unless miimon is set\n",
4654                                    updelay, downdelay);
4655                 }
4656         } else {
4657                 /* don't allow arp monitoring */
4658                 if (arp_interval) {
4659                         pr_warning("Warning: miimon (%d) and arp_interval (%d) can't be used simultaneously, disabling ARP monitoring\n",
4660                                    miimon, arp_interval);
4661                         arp_interval = 0;
4662                 }
4663
4664                 if ((updelay % miimon) != 0) {
4665                         pr_warning("Warning: updelay (%d) is not a multiple of miimon (%d), updelay rounded to %d ms\n",
4666                                    updelay, miimon,
4667                                    (updelay / miimon) * miimon);
4668                 }
4669
4670                 updelay /= miimon;
4671
4672                 if ((downdelay % miimon) != 0) {
4673                         pr_warning("Warning: downdelay (%d) is not a multiple of miimon (%d), downdelay rounded to %d ms\n",
4674                                    downdelay, miimon,
4675                                    (downdelay / miimon) * miimon);
4676                 }
4677
4678                 downdelay /= miimon;
4679         }
4680
4681         if (arp_interval < 0) {
4682                 pr_warning("Warning: arp_interval module parameter (%d) , not in range 0-%d, so it was reset to %d\n",
4683                            arp_interval, INT_MAX, BOND_LINK_ARP_INTERV);
4684                 arp_interval = BOND_LINK_ARP_INTERV;
4685         }
4686
4687         for (arp_ip_count = 0;
4688              (arp_ip_count < BOND_MAX_ARP_TARGETS) && arp_ip_target[arp_ip_count];
4689              arp_ip_count++) {
4690                 /* not complete check, but should be good enough to
4691                    catch mistakes */
4692                 if (!isdigit(arp_ip_target[arp_ip_count][0])) {
4693                         pr_warning("Warning: bad arp_ip_target module parameter (%s), ARP monitoring will not be performed\n",
4694                                    arp_ip_target[arp_ip_count]);
4695                         arp_interval = 0;
4696                 } else {
4697                         __be32 ip = in_aton(arp_ip_target[arp_ip_count]);
4698                         arp_target[arp_ip_count] = ip;
4699                 }
4700         }
4701
4702         if (arp_interval && !arp_ip_count) {
4703                 /* don't allow arping if no arp_ip_target given... */
4704                 pr_warning("Warning: arp_interval module parameter (%d) specified without providing an arp_ip_target parameter, arp_interval was reset to 0\n",
4705                            arp_interval);
4706                 arp_interval = 0;
4707         }
4708
4709         if (arp_validate) {
4710                 if (bond_mode != BOND_MODE_ACTIVEBACKUP) {
4711                         pr_err("arp_validate only supported in active-backup mode\n");
4712                         return -EINVAL;
4713                 }
4714                 if (!arp_interval) {
4715                         pr_err("arp_validate requires arp_interval\n");
4716                         return -EINVAL;
4717                 }
4718
4719                 arp_validate_value = bond_parse_parm(arp_validate,
4720                                                      arp_validate_tbl);
4721                 if (arp_validate_value == -1) {
4722                         pr_err("Error: invalid arp_validate \"%s\"\n",
4723                                arp_validate == NULL ? "NULL" : arp_validate);
4724                         return -EINVAL;
4725                 }
4726         } else
4727                 arp_validate_value = 0;
4728
4729         if (miimon) {
4730                 pr_info("MII link monitoring set to %d ms\n", miimon);
4731         } else if (arp_interval) {
4732                 int i;
4733
4734                 pr_info("ARP monitoring set to %d ms, validate %s, with %d target(s):",
4735                         arp_interval,
4736                         arp_validate_tbl[arp_validate_value].modename,
4737                         arp_ip_count);
4738
4739                 for (i = 0; i < arp_ip_count; i++)
4740                         pr_info(" %s", arp_ip_target[i]);
4741
4742                 pr_info("\n");
4743
4744         } else if (max_bonds) {
4745                 /* miimon and arp_interval not set, we need one so things
4746                  * work as expected, see bonding.txt for details
4747                  */
4748                 pr_warning("Warning: either miimon or arp_interval and arp_ip_target module parameters must be specified, otherwise bonding will not detect link failures! see bonding.txt for details.\n");
4749         }
4750
4751         if (primary && !USES_PRIMARY(bond_mode)) {
4752                 /* currently, using a primary only makes sense
4753                  * in active backup, TLB or ALB modes
4754                  */
4755                 pr_warning("Warning: %s primary device specified but has no effect in %s mode\n",
4756                            primary, bond_mode_name(bond_mode));
4757                 primary = NULL;
4758         }
4759
4760         if (primary && primary_reselect) {
4761                 primary_reselect_value = bond_parse_parm(primary_reselect,
4762                                                          pri_reselect_tbl);
4763                 if (primary_reselect_value == -1) {
4764                         pr_err("Error: Invalid primary_reselect \"%s\"\n",
4765                                primary_reselect ==
4766                                         NULL ? "NULL" : primary_reselect);
4767                         return -EINVAL;
4768                 }
4769         } else {
4770                 primary_reselect_value = BOND_PRI_RESELECT_ALWAYS;
4771         }
4772
4773         if (fail_over_mac) {
4774                 fail_over_mac_value = bond_parse_parm(fail_over_mac,
4775                                                       fail_over_mac_tbl);
4776                 if (fail_over_mac_value == -1) {
4777                         pr_err("Error: invalid fail_over_mac \"%s\"\n",
4778                                arp_validate == NULL ? "NULL" : arp_validate);
4779                         return -EINVAL;
4780                 }
4781
4782                 if (bond_mode != BOND_MODE_ACTIVEBACKUP)
4783                         pr_warning("Warning: fail_over_mac only affects active-backup mode.\n");
4784         } else {
4785                 fail_over_mac_value = BOND_FOM_NONE;
4786         }
4787
4788         /* fill params struct with the proper values */
4789         params->mode = bond_mode;
4790         params->xmit_policy = xmit_hashtype;
4791         params->miimon = miimon;
4792         params->num_grat_arp = num_grat_arp;
4793         params->num_unsol_na = num_unsol_na;
4794         params->arp_interval = arp_interval;
4795         params->arp_validate = arp_validate_value;
4796         params->updelay = updelay;
4797         params->downdelay = downdelay;
4798         params->use_carrier = use_carrier;
4799         params->lacp_fast = lacp_fast;
4800         params->primary[0] = 0;
4801         params->primary_reselect = primary_reselect_value;
4802         params->fail_over_mac = fail_over_mac_value;
4803
4804         if (primary) {
4805                 strncpy(params->primary, primary, IFNAMSIZ);
4806                 params->primary[IFNAMSIZ - 1] = 0;
4807         }
4808
4809         memcpy(params->arp_targets, arp_target, sizeof(arp_target));
4810
4811         return 0;
4812 }
4813
4814 static struct lock_class_key bonding_netdev_xmit_lock_key;
4815 static struct lock_class_key bonding_netdev_addr_lock_key;
4816
4817 static void bond_set_lockdep_class_one(struct net_device *dev,
4818                                        struct netdev_queue *txq,
4819                                        void *_unused)
4820 {
4821         lockdep_set_class(&txq->_xmit_lock,
4822                           &bonding_netdev_xmit_lock_key);
4823 }
4824
4825 static void bond_set_lockdep_class(struct net_device *dev)
4826 {
4827         lockdep_set_class(&dev->addr_list_lock,
4828                           &bonding_netdev_addr_lock_key);
4829         netdev_for_each_tx_queue(dev, bond_set_lockdep_class_one, NULL);
4830 }
4831
4832 /*
4833  * Called from registration process
4834  */
4835 static int bond_init(struct net_device *bond_dev)
4836 {
4837         struct bonding *bond = netdev_priv(bond_dev);
4838         struct bond_net *bn = net_generic(dev_net(bond_dev), bond_net_id);
4839
4840         pr_debug("Begin bond_init for %s\n", bond_dev->name);
4841
4842         bond->wq = create_singlethread_workqueue(bond_dev->name);
4843         if (!bond->wq)
4844                 return -ENOMEM;
4845
4846         bond_set_lockdep_class(bond_dev);
4847
4848         netif_carrier_off(bond_dev);
4849
4850         bond_create_proc_entry(bond);
4851         list_add_tail(&bond->bond_list, &bn->dev_list);
4852
4853         bond_prepare_sysfs_group(bond);
4854
4855         __hw_addr_init(&bond->mc_list);
4856         return 0;
4857 }
4858
4859 static int bond_validate(struct nlattr *tb[], struct nlattr *data[])
4860 {
4861         if (tb[IFLA_ADDRESS]) {
4862                 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN)
4863                         return -EINVAL;
4864                 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS])))
4865                         return -EADDRNOTAVAIL;
4866         }
4867         return 0;
4868 }
4869
4870 static struct rtnl_link_ops bond_link_ops __read_mostly = {
4871         .kind           = "bond",
4872         .priv_size      = sizeof(struct bonding),
4873         .setup          = bond_setup,
4874         .validate       = bond_validate,
4875 };
4876
4877 /* Create a new bond based on the specified name and bonding parameters.
4878  * If name is NULL, obtain a suitable "bond%d" name for us.
4879  * Caller must NOT hold rtnl_lock; we need to release it here before we
4880  * set up our sysfs entries.
4881  */
4882 int bond_create(struct net *net, const char *name)
4883 {
4884         struct net_device *bond_dev;
4885         int res;
4886
4887         rtnl_lock();
4888
4889         bond_dev = alloc_netdev(sizeof(struct bonding), name ? name : "",
4890                                 bond_setup);
4891         if (!bond_dev) {
4892                 pr_err("%s: eek! can't alloc netdev!\n", name);
4893                 res = -ENOMEM;
4894                 goto out;
4895         }
4896
4897         dev_net_set(bond_dev, net);
4898         bond_dev->rtnl_link_ops = &bond_link_ops;
4899
4900         if (!name) {
4901                 res = dev_alloc_name(bond_dev, "bond%d");
4902                 if (res < 0)
4903                         goto out_netdev;
4904         }
4905
4906         res = register_netdevice(bond_dev);
4907         if (res < 0)
4908                 goto out_netdev;
4909
4910 out:
4911         rtnl_unlock();
4912         return res;
4913 out_netdev:
4914         free_netdev(bond_dev);
4915         goto out;
4916 }
4917
4918 static int __net_init bond_net_init(struct net *net)
4919 {
4920         struct bond_net *bn = net_generic(net, bond_net_id);
4921
4922         bn->net = net;
4923         INIT_LIST_HEAD(&bn->dev_list);
4924
4925         bond_create_proc_dir(bn);
4926         
4927         return 0;
4928 }
4929
4930 static void __net_exit bond_net_exit(struct net *net)
4931 {
4932         struct bond_net *bn = net_generic(net, bond_net_id);
4933
4934         bond_destroy_proc_dir(bn);
4935 }
4936
4937 static struct pernet_operations bond_net_ops = {
4938         .init = bond_net_init,
4939         .exit = bond_net_exit,
4940         .id   = &bond_net_id,
4941         .size = sizeof(struct bond_net),
4942 };
4943
4944 static int __init bonding_init(void)
4945 {
4946         int i;
4947         int res;
4948
4949         pr_info("%s", version);
4950
4951         res = bond_check_params(&bonding_defaults);
4952         if (res)
4953                 goto out;
4954
4955         res = register_pernet_subsys(&bond_net_ops);
4956         if (res)
4957                 goto out;
4958
4959         res = rtnl_link_register(&bond_link_ops);
4960         if (res)
4961                 goto err_link;
4962
4963         for (i = 0; i < max_bonds; i++) {
4964                 res = bond_create(&init_net, NULL);
4965                 if (res)
4966                         goto err;
4967         }
4968
4969         res = bond_create_sysfs();
4970         if (res)
4971                 goto err;
4972
4973         register_netdevice_notifier(&bond_netdev_notifier);
4974         register_inetaddr_notifier(&bond_inetaddr_notifier);
4975         bond_register_ipv6_notifier();
4976 out:
4977         return res;
4978 err:
4979         rtnl_link_unregister(&bond_link_ops);
4980 err_link:
4981         unregister_pernet_subsys(&bond_net_ops);
4982         goto out;
4983
4984 }
4985
4986 static void __exit bonding_exit(void)
4987 {
4988         unregister_netdevice_notifier(&bond_netdev_notifier);
4989         unregister_inetaddr_notifier(&bond_inetaddr_notifier);
4990         bond_unregister_ipv6_notifier();
4991
4992         bond_destroy_sysfs();
4993
4994         rtnl_link_unregister(&bond_link_ops);
4995         unregister_pernet_subsys(&bond_net_ops);
4996 }
4997
4998 module_init(bonding_init);
4999 module_exit(bonding_exit);
5000 MODULE_LICENSE("GPL");
5001 MODULE_VERSION(DRV_VERSION);
5002 MODULE_DESCRIPTION(DRV_DESCRIPTION ", v" DRV_VERSION);
5003 MODULE_AUTHOR("Thomas Davis, tadavis@lbl.gov and many others");
5004 MODULE_ALIAS_RTNL_LINK("bond");