bonding: enable gro by default
[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/netpoll.h>
63 #include <linux/inetdevice.h>
64 #include <linux/igmp.h>
65 #include <linux/etherdevice.h>
66 #include <linux/skbuff.h>
67 #include <net/sock.h>
68 #include <linux/rtnetlink.h>
69 #include <linux/proc_fs.h>
70 #include <linux/seq_file.h>
71 #include <linux/smp.h>
72 #include <linux/if_ether.h>
73 #include <net/arp.h>
74 #include <linux/mii.h>
75 #include <linux/ethtool.h>
76 #include <linux/if_vlan.h>
77 #include <linux/if_bonding.h>
78 #include <linux/jiffies.h>
79 #include <net/route.h>
80 #include <net/net_namespace.h>
81 #include <net/netns/generic.h>
82 #include "bonding.h"
83 #include "bond_3ad.h"
84 #include "bond_alb.h"
85
86 /*---------------------------- Module parameters ----------------------------*/
87
88 /* monitor all links that often (in milliseconds). <=0 disables monitoring */
89 #define BOND_LINK_MON_INTERV    0
90 #define BOND_LINK_ARP_INTERV    0
91
92 static int max_bonds    = BOND_DEFAULT_MAX_BONDS;
93 static int tx_queues    = BOND_DEFAULT_TX_QUEUES;
94 static int num_grat_arp = 1;
95 static int num_unsol_na = 1;
96 static int miimon       = BOND_LINK_MON_INTERV;
97 static int updelay;
98 static int downdelay;
99 static int use_carrier  = 1;
100 static char *mode;
101 static char *primary;
102 static char *primary_reselect;
103 static char *lacp_rate;
104 static char *ad_select;
105 static char *xmit_hash_policy;
106 static int arp_interval = BOND_LINK_ARP_INTERV;
107 static char *arp_ip_target[BOND_MAX_ARP_TARGETS];
108 static char *arp_validate;
109 static char *fail_over_mac;
110 static int all_slaves_active = 0;
111 static struct bond_params bonding_defaults;
112
113 module_param(max_bonds, int, 0);
114 MODULE_PARM_DESC(max_bonds, "Max number of bonded devices");
115 module_param(tx_queues, int, 0);
116 MODULE_PARM_DESC(tx_queues, "Max number of transmit queues (default = 16)");
117 module_param(num_grat_arp, int, 0644);
118 MODULE_PARM_DESC(num_grat_arp, "Number of gratuitous ARP packets to send on failover event");
119 module_param(num_unsol_na, int, 0644);
120 MODULE_PARM_DESC(num_unsol_na, "Number of unsolicited IPv6 Neighbor Advertisements packets to send on failover event");
121 module_param(miimon, int, 0);
122 MODULE_PARM_DESC(miimon, "Link check interval in milliseconds");
123 module_param(updelay, int, 0);
124 MODULE_PARM_DESC(updelay, "Delay before considering link up, in milliseconds");
125 module_param(downdelay, int, 0);
126 MODULE_PARM_DESC(downdelay, "Delay before considering link down, "
127                             "in milliseconds");
128 module_param(use_carrier, int, 0);
129 MODULE_PARM_DESC(use_carrier, "Use netif_carrier_ok (vs MII ioctls) in miimon; "
130                               "0 for off, 1 for on (default)");
131 module_param(mode, charp, 0);
132 MODULE_PARM_DESC(mode, "Mode of operation : 0 for balance-rr, "
133                        "1 for active-backup, 2 for balance-xor, "
134                        "3 for broadcast, 4 for 802.3ad, 5 for balance-tlb, "
135                        "6 for balance-alb");
136 module_param(primary, charp, 0);
137 MODULE_PARM_DESC(primary, "Primary network device to use");
138 module_param(primary_reselect, charp, 0);
139 MODULE_PARM_DESC(primary_reselect, "Reselect primary slave "
140                                    "once it comes up; "
141                                    "0 for always (default), "
142                                    "1 for only if speed of primary is "
143                                    "better, "
144                                    "2 for only on active slave "
145                                    "failure");
146 module_param(lacp_rate, charp, 0);
147 MODULE_PARM_DESC(lacp_rate, "LACPDU tx rate to request from 802.3ad partner "
148                             "(slow/fast)");
149 module_param(ad_select, charp, 0);
150 MODULE_PARM_DESC(ad_select, "803.ad aggregation selection logic: stable (0, default), bandwidth (1), count (2)");
151 module_param(xmit_hash_policy, charp, 0);
152 MODULE_PARM_DESC(xmit_hash_policy, "XOR hashing method: 0 for layer 2 (default)"
153                                    ", 1 for layer 3+4");
154 module_param(arp_interval, int, 0);
155 MODULE_PARM_DESC(arp_interval, "arp interval in milliseconds");
156 module_param_array(arp_ip_target, charp, NULL, 0);
157 MODULE_PARM_DESC(arp_ip_target, "arp targets in n.n.n.n form");
158 module_param(arp_validate, charp, 0);
159 MODULE_PARM_DESC(arp_validate, "validate src/dst of ARP probes: none (default), active, backup or all");
160 module_param(fail_over_mac, charp, 0);
161 MODULE_PARM_DESC(fail_over_mac, "For active-backup, do not set all slaves to the same MAC.  none (default), active or follow");
162 module_param(all_slaves_active, int, 0);
163 MODULE_PARM_DESC(all_slaves_active, "Keep all frames received on an interface"
164                                      "by setting active flag for all slaves.  "
165                                      "0 for never (default), 1 for always.");
166
167 /*----------------------------- Global variables ----------------------------*/
168
169 static const char * const version =
170         DRV_DESCRIPTION ": v" DRV_VERSION " (" DRV_RELDATE ")\n";
171
172 int bond_net_id __read_mostly;
173
174 static __be32 arp_target[BOND_MAX_ARP_TARGETS];
175 static int arp_ip_count;
176 static int bond_mode    = BOND_MODE_ROUNDROBIN;
177 static int xmit_hashtype = BOND_XMIT_POLICY_LAYER2;
178 static int lacp_fast;
179 #ifdef CONFIG_NET_POLL_CONTROLLER
180 static int disable_netpoll = 1;
181 #endif
182
183 const struct bond_parm_tbl bond_lacp_tbl[] = {
184 {       "slow",         AD_LACP_SLOW},
185 {       "fast",         AD_LACP_FAST},
186 {       NULL,           -1},
187 };
188
189 const struct bond_parm_tbl bond_mode_tbl[] = {
190 {       "balance-rr",           BOND_MODE_ROUNDROBIN},
191 {       "active-backup",        BOND_MODE_ACTIVEBACKUP},
192 {       "balance-xor",          BOND_MODE_XOR},
193 {       "broadcast",            BOND_MODE_BROADCAST},
194 {       "802.3ad",              BOND_MODE_8023AD},
195 {       "balance-tlb",          BOND_MODE_TLB},
196 {       "balance-alb",          BOND_MODE_ALB},
197 {       NULL,                   -1},
198 };
199
200 const struct bond_parm_tbl xmit_hashtype_tbl[] = {
201 {       "layer2",               BOND_XMIT_POLICY_LAYER2},
202 {       "layer3+4",             BOND_XMIT_POLICY_LAYER34},
203 {       "layer2+3",             BOND_XMIT_POLICY_LAYER23},
204 {       NULL,                   -1},
205 };
206
207 const struct bond_parm_tbl arp_validate_tbl[] = {
208 {       "none",                 BOND_ARP_VALIDATE_NONE},
209 {       "active",               BOND_ARP_VALIDATE_ACTIVE},
210 {       "backup",               BOND_ARP_VALIDATE_BACKUP},
211 {       "all",                  BOND_ARP_VALIDATE_ALL},
212 {       NULL,                   -1},
213 };
214
215 const struct bond_parm_tbl fail_over_mac_tbl[] = {
216 {       "none",                 BOND_FOM_NONE},
217 {       "active",               BOND_FOM_ACTIVE},
218 {       "follow",               BOND_FOM_FOLLOW},
219 {       NULL,                   -1},
220 };
221
222 const struct bond_parm_tbl pri_reselect_tbl[] = {
223 {       "always",               BOND_PRI_RESELECT_ALWAYS},
224 {       "better",               BOND_PRI_RESELECT_BETTER},
225 {       "failure",              BOND_PRI_RESELECT_FAILURE},
226 {       NULL,                   -1},
227 };
228
229 struct bond_parm_tbl ad_select_tbl[] = {
230 {       "stable",       BOND_AD_STABLE},
231 {       "bandwidth",    BOND_AD_BANDWIDTH},
232 {       "count",        BOND_AD_COUNT},
233 {       NULL,           -1},
234 };
235
236 /*-------------------------- Forward declarations ---------------------------*/
237
238 static void bond_send_gratuitous_arp(struct bonding *bond);
239 static int bond_init(struct net_device *bond_dev);
240 static void bond_uninit(struct net_device *bond_dev);
241
242 /*---------------------------- General routines -----------------------------*/
243
244 static const char *bond_mode_name(int mode)
245 {
246         static const char *names[] = {
247                 [BOND_MODE_ROUNDROBIN] = "load balancing (round-robin)",
248                 [BOND_MODE_ACTIVEBACKUP] = "fault-tolerance (active-backup)",
249                 [BOND_MODE_XOR] = "load balancing (xor)",
250                 [BOND_MODE_BROADCAST] = "fault-tolerance (broadcast)",
251                 [BOND_MODE_8023AD] = "IEEE 802.3ad Dynamic link aggregation",
252                 [BOND_MODE_TLB] = "transmit load balancing",
253                 [BOND_MODE_ALB] = "adaptive load balancing",
254         };
255
256         if (mode < 0 || mode > BOND_MODE_ALB)
257                 return "unknown";
258
259         return names[mode];
260 }
261
262 /*---------------------------------- VLAN -----------------------------------*/
263
264 /**
265  * bond_add_vlan - add a new vlan id on bond
266  * @bond: bond that got the notification
267  * @vlan_id: the vlan id to add
268  *
269  * Returns -ENOMEM if allocation failed.
270  */
271 static int bond_add_vlan(struct bonding *bond, unsigned short vlan_id)
272 {
273         struct vlan_entry *vlan;
274
275         pr_debug("bond: %s, vlan id %d\n",
276                  (bond ? bond->dev->name : "None"), vlan_id);
277
278         vlan = kzalloc(sizeof(struct vlan_entry), GFP_KERNEL);
279         if (!vlan)
280                 return -ENOMEM;
281
282         INIT_LIST_HEAD(&vlan->vlan_list);
283         vlan->vlan_id = vlan_id;
284
285         write_lock_bh(&bond->lock);
286
287         list_add_tail(&vlan->vlan_list, &bond->vlan_list);
288
289         write_unlock_bh(&bond->lock);
290
291         pr_debug("added VLAN ID %d on bond %s\n", vlan_id, bond->dev->name);
292
293         return 0;
294 }
295
296 /**
297  * bond_del_vlan - delete a vlan id from bond
298  * @bond: bond that got the notification
299  * @vlan_id: the vlan id to delete
300  *
301  * returns -ENODEV if @vlan_id was not found in @bond.
302  */
303 static int bond_del_vlan(struct bonding *bond, unsigned short vlan_id)
304 {
305         struct vlan_entry *vlan;
306         int res = -ENODEV;
307
308         pr_debug("bond: %s, vlan id %d\n", bond->dev->name, vlan_id);
309
310         write_lock_bh(&bond->lock);
311
312         list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
313                 if (vlan->vlan_id == vlan_id) {
314                         list_del(&vlan->vlan_list);
315
316                         if (bond_is_lb(bond))
317                                 bond_alb_clear_vlan(bond, vlan_id);
318
319                         pr_debug("removed VLAN ID %d from bond %s\n",
320                                  vlan_id, bond->dev->name);
321
322                         kfree(vlan);
323
324                         if (list_empty(&bond->vlan_list) &&
325                             (bond->slave_cnt == 0)) {
326                                 /* Last VLAN removed and no slaves, so
327                                  * restore block on adding VLANs. This will
328                                  * be removed once new slaves that are not
329                                  * VLAN challenged will be added.
330                                  */
331                                 bond->dev->features |= NETIF_F_VLAN_CHALLENGED;
332                         }
333
334                         res = 0;
335                         goto out;
336                 }
337         }
338
339         pr_debug("couldn't find VLAN ID %d in bond %s\n",
340                  vlan_id, bond->dev->name);
341
342 out:
343         write_unlock_bh(&bond->lock);
344         return res;
345 }
346
347 /**
348  * bond_has_challenged_slaves
349  * @bond: the bond we're working on
350  *
351  * Searches the slave list. Returns 1 if a vlan challenged slave
352  * was found, 0 otherwise.
353  *
354  * Assumes bond->lock is held.
355  */
356 static int bond_has_challenged_slaves(struct bonding *bond)
357 {
358         struct slave *slave;
359         int i;
360
361         bond_for_each_slave(bond, slave, i) {
362                 if (slave->dev->features & NETIF_F_VLAN_CHALLENGED) {
363                         pr_debug("found VLAN challenged slave - %s\n",
364                                  slave->dev->name);
365                         return 1;
366                 }
367         }
368
369         pr_debug("no VLAN challenged slaves found\n");
370         return 0;
371 }
372
373 /**
374  * bond_next_vlan - safely skip to the next item in the vlans list.
375  * @bond: the bond we're working on
376  * @curr: item we're advancing from
377  *
378  * Returns %NULL if list is empty, bond->next_vlan if @curr is %NULL,
379  * or @curr->next otherwise (even if it is @curr itself again).
380  *
381  * Caller must hold bond->lock
382  */
383 struct vlan_entry *bond_next_vlan(struct bonding *bond, struct vlan_entry *curr)
384 {
385         struct vlan_entry *next, *last;
386
387         if (list_empty(&bond->vlan_list))
388                 return NULL;
389
390         if (!curr) {
391                 next = list_entry(bond->vlan_list.next,
392                                   struct vlan_entry, vlan_list);
393         } else {
394                 last = list_entry(bond->vlan_list.prev,
395                                   struct vlan_entry, vlan_list);
396                 if (last == curr) {
397                         next = list_entry(bond->vlan_list.next,
398                                           struct vlan_entry, vlan_list);
399                 } else {
400                         next = list_entry(curr->vlan_list.next,
401                                           struct vlan_entry, vlan_list);
402                 }
403         }
404
405         return next;
406 }
407
408 /**
409  * bond_dev_queue_xmit - Prepare skb for xmit.
410  *
411  * @bond: bond device that got this skb for tx.
412  * @skb: hw accel VLAN tagged skb to transmit
413  * @slave_dev: slave that is supposed to xmit this skbuff
414  *
415  * When the bond gets an skb to transmit that is
416  * already hardware accelerated VLAN tagged, and it
417  * needs to relay this skb to a slave that is not
418  * hw accel capable, the skb needs to be "unaccelerated",
419  * i.e. strip the hwaccel tag and re-insert it as part
420  * of the payload.
421  */
422 int bond_dev_queue_xmit(struct bonding *bond, struct sk_buff *skb,
423                         struct net_device *slave_dev)
424 {
425         unsigned short uninitialized_var(vlan_id);
426
427         /* Test vlan_list not vlgrp to catch and handle 802.1p tags */
428         if (!list_empty(&bond->vlan_list) &&
429             !(slave_dev->features & NETIF_F_HW_VLAN_TX) &&
430             vlan_get_tag(skb, &vlan_id) == 0) {
431                 skb->dev = slave_dev;
432                 skb = vlan_put_tag(skb, vlan_id);
433                 if (!skb) {
434                         /* vlan_put_tag() frees the skb in case of error,
435                          * so return success here so the calling functions
436                          * won't attempt to free is again.
437                          */
438                         return 0;
439                 }
440         } else {
441                 skb->dev = slave_dev;
442         }
443
444         skb->priority = 1;
445 #ifdef CONFIG_NET_POLL_CONTROLLER
446         if (unlikely(bond->dev->priv_flags & IFF_IN_NETPOLL)) {
447                 struct netpoll *np = bond->dev->npinfo->netpoll;
448                 slave_dev->npinfo = bond->dev->npinfo;
449                 np->real_dev = np->dev = skb->dev;
450                 slave_dev->priv_flags |= IFF_IN_NETPOLL;
451                 netpoll_send_skb(np, skb);
452                 slave_dev->priv_flags &= ~IFF_IN_NETPOLL;
453                 np->dev = bond->dev;
454         } else
455 #endif
456                 dev_queue_xmit(skb);
457
458         return 0;
459 }
460
461 /*
462  * In the following 3 functions, bond_vlan_rx_register(), bond_vlan_rx_add_vid
463  * and bond_vlan_rx_kill_vid, We don't protect the slave list iteration with a
464  * lock because:
465  * a. This operation is performed in IOCTL context,
466  * b. The operation is protected by the RTNL semaphore in the 8021q code,
467  * c. Holding a lock with BH disabled while directly calling a base driver
468  *    entry point is generally a BAD idea.
469  *
470  * The design of synchronization/protection for this operation in the 8021q
471  * module is good for one or more VLAN devices over a single physical device
472  * and cannot be extended for a teaming solution like bonding, so there is a
473  * potential race condition here where a net device from the vlan group might
474  * be referenced (either by a base driver or the 8021q code) while it is being
475  * removed from the system. However, it turns out we're not making matters
476  * worse, and if it works for regular VLAN usage it will work here too.
477 */
478
479 /**
480  * bond_vlan_rx_register - Propagates registration to slaves
481  * @bond_dev: bonding net device that got called
482  * @grp: vlan group being registered
483  */
484 static void bond_vlan_rx_register(struct net_device *bond_dev,
485                                   struct vlan_group *grp)
486 {
487         struct bonding *bond = netdev_priv(bond_dev);
488         struct slave *slave;
489         int i;
490
491         write_lock(&bond->lock);
492         bond->vlgrp = grp;
493         write_unlock(&bond->lock);
494
495         bond_for_each_slave(bond, slave, i) {
496                 struct net_device *slave_dev = slave->dev;
497                 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
498
499                 if ((slave_dev->features & NETIF_F_HW_VLAN_RX) &&
500                     slave_ops->ndo_vlan_rx_register) {
501                         slave_ops->ndo_vlan_rx_register(slave_dev, grp);
502                 }
503         }
504 }
505
506 /**
507  * bond_vlan_rx_add_vid - Propagates adding an id to slaves
508  * @bond_dev: bonding net device that got called
509  * @vid: vlan id being added
510  */
511 static void bond_vlan_rx_add_vid(struct net_device *bond_dev, uint16_t vid)
512 {
513         struct bonding *bond = netdev_priv(bond_dev);
514         struct slave *slave;
515         int i, res;
516
517         bond_for_each_slave(bond, slave, i) {
518                 struct net_device *slave_dev = slave->dev;
519                 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
520
521                 if ((slave_dev->features & NETIF_F_HW_VLAN_FILTER) &&
522                     slave_ops->ndo_vlan_rx_add_vid) {
523                         slave_ops->ndo_vlan_rx_add_vid(slave_dev, vid);
524                 }
525         }
526
527         res = bond_add_vlan(bond, vid);
528         if (res) {
529                 pr_err("%s: Error: Failed to add vlan id %d\n",
530                        bond_dev->name, vid);
531         }
532 }
533
534 /**
535  * bond_vlan_rx_kill_vid - Propagates deleting an id to slaves
536  * @bond_dev: bonding net device that got called
537  * @vid: vlan id being removed
538  */
539 static void bond_vlan_rx_kill_vid(struct net_device *bond_dev, uint16_t vid)
540 {
541         struct bonding *bond = netdev_priv(bond_dev);
542         struct slave *slave;
543         struct net_device *vlan_dev;
544         int i, res;
545
546         bond_for_each_slave(bond, slave, i) {
547                 struct net_device *slave_dev = slave->dev;
548                 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
549
550                 if ((slave_dev->features & NETIF_F_HW_VLAN_FILTER) &&
551                     slave_ops->ndo_vlan_rx_kill_vid) {
552                         /* Save and then restore vlan_dev in the grp array,
553                          * since the slave's driver might clear it.
554                          */
555                         vlan_dev = vlan_group_get_device(bond->vlgrp, vid);
556                         slave_ops->ndo_vlan_rx_kill_vid(slave_dev, vid);
557                         vlan_group_set_device(bond->vlgrp, vid, vlan_dev);
558                 }
559         }
560
561         res = bond_del_vlan(bond, vid);
562         if (res) {
563                 pr_err("%s: Error: Failed to remove vlan id %d\n",
564                        bond_dev->name, vid);
565         }
566 }
567
568 static void bond_add_vlans_on_slave(struct bonding *bond, struct net_device *slave_dev)
569 {
570         struct vlan_entry *vlan;
571         const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
572
573         if (!bond->vlgrp)
574                 return;
575
576         if ((slave_dev->features & NETIF_F_HW_VLAN_RX) &&
577             slave_ops->ndo_vlan_rx_register)
578                 slave_ops->ndo_vlan_rx_register(slave_dev, bond->vlgrp);
579
580         if (!(slave_dev->features & NETIF_F_HW_VLAN_FILTER) ||
581             !(slave_ops->ndo_vlan_rx_add_vid))
582                 return;
583
584         list_for_each_entry(vlan, &bond->vlan_list, vlan_list)
585                 slave_ops->ndo_vlan_rx_add_vid(slave_dev, vlan->vlan_id);
586 }
587
588 static void bond_del_vlans_from_slave(struct bonding *bond,
589                                       struct net_device *slave_dev)
590 {
591         const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
592         struct vlan_entry *vlan;
593         struct net_device *vlan_dev;
594
595         if (!bond->vlgrp)
596                 return;
597
598         if (!(slave_dev->features & NETIF_F_HW_VLAN_FILTER) ||
599             !(slave_ops->ndo_vlan_rx_kill_vid))
600                 goto unreg;
601
602         list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
603                 if (!vlan->vlan_id)
604                         continue;
605                 /* Save and then restore vlan_dev in the grp array,
606                  * since the slave's driver might clear it.
607                  */
608                 vlan_dev = vlan_group_get_device(bond->vlgrp, vlan->vlan_id);
609                 slave_ops->ndo_vlan_rx_kill_vid(slave_dev, vlan->vlan_id);
610                 vlan_group_set_device(bond->vlgrp, vlan->vlan_id, vlan_dev);
611         }
612
613 unreg:
614         if ((slave_dev->features & NETIF_F_HW_VLAN_RX) &&
615             slave_ops->ndo_vlan_rx_register)
616                 slave_ops->ndo_vlan_rx_register(slave_dev, NULL);
617 }
618
619 /*------------------------------- Link status -------------------------------*/
620
621 /*
622  * Set the carrier state for the master according to the state of its
623  * slaves.  If any slaves are up, the master is up.  In 802.3ad mode,
624  * do special 802.3ad magic.
625  *
626  * Returns zero if carrier state does not change, nonzero if it does.
627  */
628 static int bond_set_carrier(struct bonding *bond)
629 {
630         struct slave *slave;
631         int i;
632
633         if (bond->slave_cnt == 0)
634                 goto down;
635
636         if (bond->params.mode == BOND_MODE_8023AD)
637                 return bond_3ad_set_carrier(bond);
638
639         bond_for_each_slave(bond, slave, i) {
640                 if (slave->link == BOND_LINK_UP) {
641                         if (!netif_carrier_ok(bond->dev)) {
642                                 netif_carrier_on(bond->dev);
643                                 return 1;
644                         }
645                         return 0;
646                 }
647         }
648
649 down:
650         if (netif_carrier_ok(bond->dev)) {
651                 netif_carrier_off(bond->dev);
652                 return 1;
653         }
654         return 0;
655 }
656
657 /*
658  * Get link speed and duplex from the slave's base driver
659  * using ethtool. If for some reason the call fails or the
660  * values are invalid, fake speed and duplex to 100/Full
661  * and return error.
662  */
663 static int bond_update_speed_duplex(struct slave *slave)
664 {
665         struct net_device *slave_dev = slave->dev;
666         struct ethtool_cmd etool;
667         int res;
668
669         /* Fake speed and duplex */
670         slave->speed = SPEED_100;
671         slave->duplex = DUPLEX_FULL;
672
673         if (!slave_dev->ethtool_ops || !slave_dev->ethtool_ops->get_settings)
674                 return -1;
675
676         res = slave_dev->ethtool_ops->get_settings(slave_dev, &etool);
677         if (res < 0)
678                 return -1;
679
680         switch (etool.speed) {
681         case SPEED_10:
682         case SPEED_100:
683         case SPEED_1000:
684         case SPEED_10000:
685                 break;
686         default:
687                 return -1;
688         }
689
690         switch (etool.duplex) {
691         case DUPLEX_FULL:
692         case DUPLEX_HALF:
693                 break;
694         default:
695                 return -1;
696         }
697
698         slave->speed = etool.speed;
699         slave->duplex = etool.duplex;
700
701         return 0;
702 }
703
704 /*
705  * if <dev> supports MII link status reporting, check its link status.
706  *
707  * We either do MII/ETHTOOL ioctls, or check netif_carrier_ok(),
708  * depending upon the setting of the use_carrier parameter.
709  *
710  * Return either BMSR_LSTATUS, meaning that the link is up (or we
711  * can't tell and just pretend it is), or 0, meaning that the link is
712  * down.
713  *
714  * If reporting is non-zero, instead of faking link up, return -1 if
715  * both ETHTOOL and MII ioctls fail (meaning the device does not
716  * support them).  If use_carrier is set, return whatever it says.
717  * It'd be nice if there was a good way to tell if a driver supports
718  * netif_carrier, but there really isn't.
719  */
720 static int bond_check_dev_link(struct bonding *bond,
721                                struct net_device *slave_dev, int reporting)
722 {
723         const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
724         int (*ioctl)(struct net_device *, struct ifreq *, int);
725         struct ifreq ifr;
726         struct mii_ioctl_data *mii;
727
728         if (!reporting && !netif_running(slave_dev))
729                 return 0;
730
731         if (bond->params.use_carrier)
732                 return netif_carrier_ok(slave_dev) ? BMSR_LSTATUS : 0;
733
734         /* Try to get link status using Ethtool first. */
735         if (slave_dev->ethtool_ops) {
736                 if (slave_dev->ethtool_ops->get_link) {
737                         u32 link;
738
739                         link = slave_dev->ethtool_ops->get_link(slave_dev);
740
741                         return link ? BMSR_LSTATUS : 0;
742                 }
743         }
744
745         /* Ethtool can't be used, fallback to MII ioctls. */
746         ioctl = slave_ops->ndo_do_ioctl;
747         if (ioctl) {
748                 /* TODO: set pointer to correct ioctl on a per team member */
749                 /*       bases to make this more efficient. that is, once  */
750                 /*       we determine the correct ioctl, we will always    */
751                 /*       call it and not the others for that team          */
752                 /*       member.                                           */
753
754                 /*
755                  * We cannot assume that SIOCGMIIPHY will also read a
756                  * register; not all network drivers (e.g., e100)
757                  * support that.
758                  */
759
760                 /* Yes, the mii is overlaid on the ifreq.ifr_ifru */
761                 strncpy(ifr.ifr_name, slave_dev->name, IFNAMSIZ);
762                 mii = if_mii(&ifr);
763                 if (IOCTL(slave_dev, &ifr, SIOCGMIIPHY) == 0) {
764                         mii->reg_num = MII_BMSR;
765                         if (IOCTL(slave_dev, &ifr, SIOCGMIIREG) == 0)
766                                 return mii->val_out & BMSR_LSTATUS;
767                 }
768         }
769
770         /*
771          * If reporting, report that either there's no dev->do_ioctl,
772          * or both SIOCGMIIREG and get_link failed (meaning that we
773          * cannot report link status).  If not reporting, pretend
774          * we're ok.
775          */
776         return reporting ? -1 : BMSR_LSTATUS;
777 }
778
779 /*----------------------------- Multicast list ------------------------------*/
780
781 /*
782  * Push the promiscuity flag down to appropriate slaves
783  */
784 static int bond_set_promiscuity(struct bonding *bond, int inc)
785 {
786         int err = 0;
787         if (USES_PRIMARY(bond->params.mode)) {
788                 /* write lock already acquired */
789                 if (bond->curr_active_slave) {
790                         err = dev_set_promiscuity(bond->curr_active_slave->dev,
791                                                   inc);
792                 }
793         } else {
794                 struct slave *slave;
795                 int i;
796                 bond_for_each_slave(bond, slave, i) {
797                         err = dev_set_promiscuity(slave->dev, inc);
798                         if (err)
799                                 return err;
800                 }
801         }
802         return err;
803 }
804
805 /*
806  * Push the allmulti flag down to all slaves
807  */
808 static int bond_set_allmulti(struct bonding *bond, int inc)
809 {
810         int err = 0;
811         if (USES_PRIMARY(bond->params.mode)) {
812                 /* write lock already acquired */
813                 if (bond->curr_active_slave) {
814                         err = dev_set_allmulti(bond->curr_active_slave->dev,
815                                                inc);
816                 }
817         } else {
818                 struct slave *slave;
819                 int i;
820                 bond_for_each_slave(bond, slave, i) {
821                         err = dev_set_allmulti(slave->dev, inc);
822                         if (err)
823                                 return err;
824                 }
825         }
826         return err;
827 }
828
829 /*
830  * Add a Multicast address to slaves
831  * according to mode
832  */
833 static void bond_mc_add(struct bonding *bond, void *addr)
834 {
835         if (USES_PRIMARY(bond->params.mode)) {
836                 /* write lock already acquired */
837                 if (bond->curr_active_slave)
838                         dev_mc_add(bond->curr_active_slave->dev, addr);
839         } else {
840                 struct slave *slave;
841                 int i;
842
843                 bond_for_each_slave(bond, slave, i)
844                         dev_mc_add(slave->dev, addr);
845         }
846 }
847
848 /*
849  * Remove a multicast address from slave
850  * according to mode
851  */
852 static void bond_mc_del(struct bonding *bond, void *addr)
853 {
854         if (USES_PRIMARY(bond->params.mode)) {
855                 /* write lock already acquired */
856                 if (bond->curr_active_slave)
857                         dev_mc_del(bond->curr_active_slave->dev, addr);
858         } else {
859                 struct slave *slave;
860                 int i;
861                 bond_for_each_slave(bond, slave, i) {
862                         dev_mc_del(slave->dev, addr);
863                 }
864         }
865 }
866
867
868 /*
869  * Retrieve the list of registered multicast addresses for the bonding
870  * device and retransmit an IGMP JOIN request to the current active
871  * slave.
872  */
873 static void bond_resend_igmp_join_requests(struct bonding *bond)
874 {
875         struct in_device *in_dev;
876         struct ip_mc_list *im;
877
878         rcu_read_lock();
879         in_dev = __in_dev_get_rcu(bond->dev);
880         if (in_dev) {
881                 for (im = in_dev->mc_list; im; im = im->next)
882                         ip_mc_rejoin_group(im);
883         }
884
885         rcu_read_unlock();
886 }
887
888 /*
889  * flush all members of flush->mc_list from device dev->mc_list
890  */
891 static void bond_mc_list_flush(struct net_device *bond_dev,
892                                struct net_device *slave_dev)
893 {
894         struct bonding *bond = netdev_priv(bond_dev);
895         struct netdev_hw_addr *ha;
896
897         netdev_for_each_mc_addr(ha, bond_dev)
898                 dev_mc_del(slave_dev, ha->addr);
899
900         if (bond->params.mode == BOND_MODE_8023AD) {
901                 /* del lacpdu mc addr from mc list */
902                 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
903
904                 dev_mc_del(slave_dev, lacpdu_multicast);
905         }
906 }
907
908 /*--------------------------- Active slave change ---------------------------*/
909
910 /*
911  * Update the mc list and multicast-related flags for the new and
912  * old active slaves (if any) according to the multicast mode, and
913  * promiscuous flags unconditionally.
914  */
915 static void bond_mc_swap(struct bonding *bond, struct slave *new_active,
916                          struct slave *old_active)
917 {
918         struct netdev_hw_addr *ha;
919
920         if (!USES_PRIMARY(bond->params.mode))
921                 /* nothing to do -  mc list is already up-to-date on
922                  * all slaves
923                  */
924                 return;
925
926         if (old_active) {
927                 if (bond->dev->flags & IFF_PROMISC)
928                         dev_set_promiscuity(old_active->dev, -1);
929
930                 if (bond->dev->flags & IFF_ALLMULTI)
931                         dev_set_allmulti(old_active->dev, -1);
932
933                 netdev_for_each_mc_addr(ha, bond->dev)
934                         dev_mc_del(old_active->dev, ha->addr);
935         }
936
937         if (new_active) {
938                 /* FIXME: Signal errors upstream. */
939                 if (bond->dev->flags & IFF_PROMISC)
940                         dev_set_promiscuity(new_active->dev, 1);
941
942                 if (bond->dev->flags & IFF_ALLMULTI)
943                         dev_set_allmulti(new_active->dev, 1);
944
945                 netdev_for_each_mc_addr(ha, bond->dev)
946                         dev_mc_add(new_active->dev, ha->addr);
947                 bond_resend_igmp_join_requests(bond);
948         }
949 }
950
951 /*
952  * bond_do_fail_over_mac
953  *
954  * Perform special MAC address swapping for fail_over_mac settings
955  *
956  * Called with RTNL, bond->lock for read, curr_slave_lock for write_bh.
957  */
958 static void bond_do_fail_over_mac(struct bonding *bond,
959                                   struct slave *new_active,
960                                   struct slave *old_active)
961         __releases(&bond->curr_slave_lock)
962         __releases(&bond->lock)
963         __acquires(&bond->lock)
964         __acquires(&bond->curr_slave_lock)
965 {
966         u8 tmp_mac[ETH_ALEN];
967         struct sockaddr saddr;
968         int rv;
969
970         switch (bond->params.fail_over_mac) {
971         case BOND_FOM_ACTIVE:
972                 if (new_active)
973                         memcpy(bond->dev->dev_addr,  new_active->dev->dev_addr,
974                                new_active->dev->addr_len);
975                 break;
976         case BOND_FOM_FOLLOW:
977                 /*
978                  * if new_active && old_active, swap them
979                  * if just old_active, do nothing (going to no active slave)
980                  * if just new_active, set new_active to bond's MAC
981                  */
982                 if (!new_active)
983                         return;
984
985                 write_unlock_bh(&bond->curr_slave_lock);
986                 read_unlock(&bond->lock);
987
988                 if (old_active) {
989                         memcpy(tmp_mac, new_active->dev->dev_addr, ETH_ALEN);
990                         memcpy(saddr.sa_data, old_active->dev->dev_addr,
991                                ETH_ALEN);
992                         saddr.sa_family = new_active->dev->type;
993                 } else {
994                         memcpy(saddr.sa_data, bond->dev->dev_addr, ETH_ALEN);
995                         saddr.sa_family = bond->dev->type;
996                 }
997
998                 rv = dev_set_mac_address(new_active->dev, &saddr);
999                 if (rv) {
1000                         pr_err("%s: Error %d setting MAC of slave %s\n",
1001                                bond->dev->name, -rv, new_active->dev->name);
1002                         goto out;
1003                 }
1004
1005                 if (!old_active)
1006                         goto out;
1007
1008                 memcpy(saddr.sa_data, tmp_mac, ETH_ALEN);
1009                 saddr.sa_family = old_active->dev->type;
1010
1011                 rv = dev_set_mac_address(old_active->dev, &saddr);
1012                 if (rv)
1013                         pr_err("%s: Error %d setting MAC of slave %s\n",
1014                                bond->dev->name, -rv, new_active->dev->name);
1015 out:
1016                 read_lock(&bond->lock);
1017                 write_lock_bh(&bond->curr_slave_lock);
1018                 break;
1019         default:
1020                 pr_err("%s: bond_do_fail_over_mac impossible: bad policy %d\n",
1021                        bond->dev->name, bond->params.fail_over_mac);
1022                 break;
1023         }
1024
1025 }
1026
1027 static bool bond_should_change_active(struct bonding *bond)
1028 {
1029         struct slave *prim = bond->primary_slave;
1030         struct slave *curr = bond->curr_active_slave;
1031
1032         if (!prim || !curr || curr->link != BOND_LINK_UP)
1033                 return true;
1034         if (bond->force_primary) {
1035                 bond->force_primary = false;
1036                 return true;
1037         }
1038         if (bond->params.primary_reselect == BOND_PRI_RESELECT_BETTER &&
1039             (prim->speed < curr->speed ||
1040              (prim->speed == curr->speed && prim->duplex <= curr->duplex)))
1041                 return false;
1042         if (bond->params.primary_reselect == BOND_PRI_RESELECT_FAILURE)
1043                 return false;
1044         return true;
1045 }
1046
1047 /**
1048  * find_best_interface - select the best available slave to be the active one
1049  * @bond: our bonding struct
1050  *
1051  * Warning: Caller must hold curr_slave_lock for writing.
1052  */
1053 static struct slave *bond_find_best_slave(struct bonding *bond)
1054 {
1055         struct slave *new_active, *old_active;
1056         struct slave *bestslave = NULL;
1057         int mintime = bond->params.updelay;
1058         int i;
1059
1060         new_active = bond->curr_active_slave;
1061
1062         if (!new_active) { /* there were no active slaves left */
1063                 if (bond->slave_cnt > 0)   /* found one slave */
1064                         new_active = bond->first_slave;
1065                 else
1066                         return NULL; /* still no slave, return NULL */
1067         }
1068
1069         if ((bond->primary_slave) &&
1070             bond->primary_slave->link == BOND_LINK_UP &&
1071             bond_should_change_active(bond)) {
1072                 new_active = bond->primary_slave;
1073         }
1074
1075         /* remember where to stop iterating over the slaves */
1076         old_active = new_active;
1077
1078         bond_for_each_slave_from(bond, new_active, i, old_active) {
1079                 if (new_active->link == BOND_LINK_UP) {
1080                         return new_active;
1081                 } else if (new_active->link == BOND_LINK_BACK &&
1082                            IS_UP(new_active->dev)) {
1083                         /* link up, but waiting for stabilization */
1084                         if (new_active->delay < mintime) {
1085                                 mintime = new_active->delay;
1086                                 bestslave = new_active;
1087                         }
1088                 }
1089         }
1090
1091         return bestslave;
1092 }
1093
1094 /**
1095  * change_active_interface - change the active slave into the specified one
1096  * @bond: our bonding struct
1097  * @new: the new slave to make the active one
1098  *
1099  * Set the new slave to the bond's settings and unset them on the old
1100  * curr_active_slave.
1101  * Setting include flags, mc-list, promiscuity, allmulti, etc.
1102  *
1103  * If @new's link state is %BOND_LINK_BACK we'll set it to %BOND_LINK_UP,
1104  * because it is apparently the best available slave we have, even though its
1105  * updelay hasn't timed out yet.
1106  *
1107  * If new_active is not NULL, caller must hold bond->lock for read and
1108  * curr_slave_lock for write_bh.
1109  */
1110 void bond_change_active_slave(struct bonding *bond, struct slave *new_active)
1111 {
1112         struct slave *old_active = bond->curr_active_slave;
1113
1114         if (old_active == new_active)
1115                 return;
1116
1117         if (new_active) {
1118                 new_active->jiffies = jiffies;
1119
1120                 if (new_active->link == BOND_LINK_BACK) {
1121                         if (USES_PRIMARY(bond->params.mode)) {
1122                                 pr_info("%s: making interface %s the new active one %d ms earlier.\n",
1123                                         bond->dev->name, new_active->dev->name,
1124                                         (bond->params.updelay - new_active->delay) * bond->params.miimon);
1125                         }
1126
1127                         new_active->delay = 0;
1128                         new_active->link = BOND_LINK_UP;
1129
1130                         if (bond->params.mode == BOND_MODE_8023AD)
1131                                 bond_3ad_handle_link_change(new_active, BOND_LINK_UP);
1132
1133                         if (bond_is_lb(bond))
1134                                 bond_alb_handle_link_change(bond, new_active, BOND_LINK_UP);
1135                 } else {
1136                         if (USES_PRIMARY(bond->params.mode)) {
1137                                 pr_info("%s: making interface %s the new active one.\n",
1138                                         bond->dev->name, new_active->dev->name);
1139                         }
1140                 }
1141         }
1142
1143         if (USES_PRIMARY(bond->params.mode))
1144                 bond_mc_swap(bond, new_active, old_active);
1145
1146         if (bond_is_lb(bond)) {
1147                 bond_alb_handle_active_change(bond, new_active);
1148                 if (old_active)
1149                         bond_set_slave_inactive_flags(old_active);
1150                 if (new_active)
1151                         bond_set_slave_active_flags(new_active);
1152         } else {
1153                 bond->curr_active_slave = new_active;
1154         }
1155
1156         if (bond->params.mode == BOND_MODE_ACTIVEBACKUP) {
1157                 if (old_active)
1158                         bond_set_slave_inactive_flags(old_active);
1159
1160                 if (new_active) {
1161                         bond_set_slave_active_flags(new_active);
1162
1163                         if (bond->params.fail_over_mac)
1164                                 bond_do_fail_over_mac(bond, new_active,
1165                                                       old_active);
1166
1167                         bond->send_grat_arp = bond->params.num_grat_arp;
1168                         bond_send_gratuitous_arp(bond);
1169
1170                         bond->send_unsol_na = bond->params.num_unsol_na;
1171                         bond_send_unsolicited_na(bond);
1172
1173                         write_unlock_bh(&bond->curr_slave_lock);
1174                         read_unlock(&bond->lock);
1175
1176                         netdev_bonding_change(bond->dev, NETDEV_BONDING_FAILOVER);
1177
1178                         read_lock(&bond->lock);
1179                         write_lock_bh(&bond->curr_slave_lock);
1180                 }
1181         }
1182
1183         /* resend IGMP joins since all were sent on curr_active_slave */
1184         if (bond->params.mode == BOND_MODE_ROUNDROBIN) {
1185                 bond_resend_igmp_join_requests(bond);
1186         }
1187 }
1188
1189 /**
1190  * bond_select_active_slave - select a new active slave, if needed
1191  * @bond: our bonding struct
1192  *
1193  * This functions should be called when one of the following occurs:
1194  * - The old curr_active_slave has been released or lost its link.
1195  * - The primary_slave has got its link back.
1196  * - A slave has got its link back and there's no old curr_active_slave.
1197  *
1198  * Caller must hold bond->lock for read and curr_slave_lock for write_bh.
1199  */
1200 void bond_select_active_slave(struct bonding *bond)
1201 {
1202         struct slave *best_slave;
1203         int rv;
1204
1205         best_slave = bond_find_best_slave(bond);
1206         if (best_slave != bond->curr_active_slave) {
1207                 bond_change_active_slave(bond, best_slave);
1208                 rv = bond_set_carrier(bond);
1209                 if (!rv)
1210                         return;
1211
1212                 if (netif_carrier_ok(bond->dev)) {
1213                         pr_info("%s: first active interface up!\n",
1214                                 bond->dev->name);
1215                 } else {
1216                         pr_info("%s: now running without any active interface !\n",
1217                                 bond->dev->name);
1218                 }
1219         }
1220 }
1221
1222 /*--------------------------- slave list handling ---------------------------*/
1223
1224 /*
1225  * This function attaches the slave to the end of list.
1226  *
1227  * bond->lock held for writing by caller.
1228  */
1229 static void bond_attach_slave(struct bonding *bond, struct slave *new_slave)
1230 {
1231         if (bond->first_slave == NULL) { /* attaching the first slave */
1232                 new_slave->next = new_slave;
1233                 new_slave->prev = new_slave;
1234                 bond->first_slave = new_slave;
1235         } else {
1236                 new_slave->next = bond->first_slave;
1237                 new_slave->prev = bond->first_slave->prev;
1238                 new_slave->next->prev = new_slave;
1239                 new_slave->prev->next = new_slave;
1240         }
1241
1242         bond->slave_cnt++;
1243 }
1244
1245 /*
1246  * This function detaches the slave from the list.
1247  * WARNING: no check is made to verify if the slave effectively
1248  * belongs to <bond>.
1249  * Nothing is freed on return, structures are just unchained.
1250  * If any slave pointer in bond was pointing to <slave>,
1251  * it should be changed by the calling function.
1252  *
1253  * bond->lock held for writing by caller.
1254  */
1255 static void bond_detach_slave(struct bonding *bond, struct slave *slave)
1256 {
1257         if (slave->next)
1258                 slave->next->prev = slave->prev;
1259
1260         if (slave->prev)
1261                 slave->prev->next = slave->next;
1262
1263         if (bond->first_slave == slave) { /* slave is the first slave */
1264                 if (bond->slave_cnt > 1) { /* there are more slave */
1265                         bond->first_slave = slave->next;
1266                 } else {
1267                         bond->first_slave = NULL; /* slave was the last one */
1268                 }
1269         }
1270
1271         slave->next = NULL;
1272         slave->prev = NULL;
1273         bond->slave_cnt--;
1274 }
1275
1276 #ifdef CONFIG_NET_POLL_CONTROLLER
1277 /*
1278  * You must hold read lock on bond->lock before calling this.
1279  */
1280 static bool slaves_support_netpoll(struct net_device *bond_dev)
1281 {
1282         struct bonding *bond = netdev_priv(bond_dev);
1283         struct slave *slave;
1284         int i = 0;
1285         bool ret = true;
1286
1287         bond_for_each_slave(bond, slave, i) {
1288                 if ((slave->dev->priv_flags & IFF_DISABLE_NETPOLL) ||
1289                     !slave->dev->netdev_ops->ndo_poll_controller)
1290                         ret = false;
1291         }
1292         return i != 0 && ret;
1293 }
1294
1295 static void bond_poll_controller(struct net_device *bond_dev)
1296 {
1297         struct net_device *dev = bond_dev->npinfo->netpoll->real_dev;
1298         if (dev != bond_dev)
1299                 netpoll_poll_dev(dev);
1300 }
1301
1302 static void bond_netpoll_cleanup(struct net_device *bond_dev)
1303 {
1304         struct bonding *bond = netdev_priv(bond_dev);
1305         struct slave *slave;
1306         const struct net_device_ops *ops;
1307         int i;
1308
1309         read_lock(&bond->lock);
1310         bond_dev->npinfo = NULL;
1311         bond_for_each_slave(bond, slave, i) {
1312                 if (slave->dev) {
1313                         ops = slave->dev->netdev_ops;
1314                         if (ops->ndo_netpoll_cleanup)
1315                                 ops->ndo_netpoll_cleanup(slave->dev);
1316                         else
1317                                 slave->dev->npinfo = NULL;
1318                 }
1319         }
1320         read_unlock(&bond->lock);
1321 }
1322
1323 #else
1324
1325 static void bond_netpoll_cleanup(struct net_device *bond_dev)
1326 {
1327 }
1328
1329 #endif
1330
1331 /*---------------------------------- IOCTL ----------------------------------*/
1332
1333 static int bond_sethwaddr(struct net_device *bond_dev,
1334                           struct net_device *slave_dev)
1335 {
1336         pr_debug("bond_dev=%p\n", bond_dev);
1337         pr_debug("slave_dev=%p\n", slave_dev);
1338         pr_debug("slave_dev->addr_len=%d\n", slave_dev->addr_len);
1339         memcpy(bond_dev->dev_addr, slave_dev->dev_addr, slave_dev->addr_len);
1340         return 0;
1341 }
1342
1343 #define BOND_VLAN_FEATURES \
1344         (NETIF_F_VLAN_CHALLENGED | NETIF_F_HW_VLAN_RX | NETIF_F_HW_VLAN_TX | \
1345          NETIF_F_HW_VLAN_FILTER)
1346
1347 /*
1348  * Compute the common dev->feature set available to all slaves.  Some
1349  * feature bits are managed elsewhere, so preserve those feature bits
1350  * on the master device.
1351  */
1352 static int bond_compute_features(struct bonding *bond)
1353 {
1354         struct slave *slave;
1355         struct net_device *bond_dev = bond->dev;
1356         unsigned long features = bond_dev->features;
1357         unsigned long vlan_features = 0;
1358         unsigned short max_hard_header_len = max((u16)ETH_HLEN,
1359                                                 bond_dev->hard_header_len);
1360         int i;
1361
1362         features &= ~(NETIF_F_ALL_CSUM | BOND_VLAN_FEATURES);
1363         features |=  NETIF_F_GSO_MASK | NETIF_F_NO_CSUM;
1364
1365         if (!bond->first_slave)
1366                 goto done;
1367
1368         features &= ~NETIF_F_ONE_FOR_ALL;
1369
1370         vlan_features = bond->first_slave->dev->vlan_features;
1371         bond_for_each_slave(bond, slave, i) {
1372                 features = netdev_increment_features(features,
1373                                                      slave->dev->features,
1374                                                      NETIF_F_ONE_FOR_ALL);
1375                 vlan_features = netdev_increment_features(vlan_features,
1376                                                         slave->dev->vlan_features,
1377                                                         NETIF_F_ONE_FOR_ALL);
1378                 if (slave->dev->hard_header_len > max_hard_header_len)
1379                         max_hard_header_len = slave->dev->hard_header_len;
1380         }
1381
1382 done:
1383         features |= (bond_dev->features & BOND_VLAN_FEATURES);
1384         bond_dev->features = netdev_fix_features(features, NULL);
1385         bond_dev->vlan_features = netdev_fix_features(vlan_features, NULL);
1386         bond_dev->hard_header_len = max_hard_header_len;
1387
1388         return 0;
1389 }
1390
1391 static void bond_setup_by_slave(struct net_device *bond_dev,
1392                                 struct net_device *slave_dev)
1393 {
1394         struct bonding *bond = netdev_priv(bond_dev);
1395
1396         bond_dev->header_ops        = slave_dev->header_ops;
1397
1398         bond_dev->type              = slave_dev->type;
1399         bond_dev->hard_header_len   = slave_dev->hard_header_len;
1400         bond_dev->addr_len          = slave_dev->addr_len;
1401
1402         memcpy(bond_dev->broadcast, slave_dev->broadcast,
1403                 slave_dev->addr_len);
1404         bond->setup_by_slave = 1;
1405 }
1406
1407 /* enslave device <slave> to bond device <master> */
1408 int bond_enslave(struct net_device *bond_dev, struct net_device *slave_dev)
1409 {
1410         struct bonding *bond = netdev_priv(bond_dev);
1411         const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
1412         struct slave *new_slave = NULL;
1413         struct netdev_hw_addr *ha;
1414         struct sockaddr addr;
1415         int link_reporting;
1416         int old_features = bond_dev->features;
1417         int res = 0;
1418
1419         if (!bond->params.use_carrier && slave_dev->ethtool_ops == NULL &&
1420                 slave_ops->ndo_do_ioctl == NULL) {
1421                 pr_warning("%s: Warning: no link monitoring support for %s\n",
1422                            bond_dev->name, slave_dev->name);
1423         }
1424
1425         /* bond must be initialized by bond_open() before enslaving */
1426         if (!(bond_dev->flags & IFF_UP)) {
1427                 pr_warning("%s: master_dev is not up in bond_enslave\n",
1428                            bond_dev->name);
1429         }
1430
1431         /* already enslaved */
1432         if (slave_dev->flags & IFF_SLAVE) {
1433                 pr_debug("Error, Device was already enslaved\n");
1434                 return -EBUSY;
1435         }
1436
1437         /* vlan challenged mutual exclusion */
1438         /* no need to lock since we're protected by rtnl_lock */
1439         if (slave_dev->features & NETIF_F_VLAN_CHALLENGED) {
1440                 pr_debug("%s: NETIF_F_VLAN_CHALLENGED\n", slave_dev->name);
1441                 if (bond->vlgrp) {
1442                         pr_err("%s: Error: cannot enslave VLAN challenged slave %s on VLAN enabled bond %s\n",
1443                                bond_dev->name, slave_dev->name, bond_dev->name);
1444                         return -EPERM;
1445                 } else {
1446                         pr_warning("%s: Warning: enslaved VLAN challenged slave %s. Adding VLANs will be blocked as long as %s is part of bond %s\n",
1447                                    bond_dev->name, slave_dev->name,
1448                                    slave_dev->name, bond_dev->name);
1449                         bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
1450                 }
1451         } else {
1452                 pr_debug("%s: ! NETIF_F_VLAN_CHALLENGED\n", slave_dev->name);
1453                 if (bond->slave_cnt == 0) {
1454                         /* First slave, and it is not VLAN challenged,
1455                          * so remove the block of adding VLANs over the bond.
1456                          */
1457                         bond_dev->features &= ~NETIF_F_VLAN_CHALLENGED;
1458                 }
1459         }
1460
1461         /*
1462          * Old ifenslave binaries are no longer supported.  These can
1463          * be identified with moderate accuracy by the state of the slave:
1464          * the current ifenslave will set the interface down prior to
1465          * enslaving it; the old ifenslave will not.
1466          */
1467         if ((slave_dev->flags & IFF_UP)) {
1468                 pr_err("%s is up. This may be due to an out of date ifenslave.\n",
1469                        slave_dev->name);
1470                 res = -EPERM;
1471                 goto err_undo_flags;
1472         }
1473
1474         /* set bonding device ether type by slave - bonding netdevices are
1475          * created with ether_setup, so when the slave type is not ARPHRD_ETHER
1476          * there is a need to override some of the type dependent attribs/funcs.
1477          *
1478          * bond ether type mutual exclusion - don't allow slaves of dissimilar
1479          * ether type (eg ARPHRD_ETHER and ARPHRD_INFINIBAND) share the same bond
1480          */
1481         if (bond->slave_cnt == 0) {
1482                 if (bond_dev->type != slave_dev->type) {
1483                         pr_debug("%s: change device type from %d to %d\n",
1484                                  bond_dev->name,
1485                                  bond_dev->type, slave_dev->type);
1486
1487                         res = netdev_bonding_change(bond_dev,
1488                                                     NETDEV_PRE_TYPE_CHANGE);
1489                         res = notifier_to_errno(res);
1490                         if (res) {
1491                                 pr_err("%s: refused to change device type\n",
1492                                        bond_dev->name);
1493                                 res = -EBUSY;
1494                                 goto err_undo_flags;
1495                         }
1496
1497                         /* Flush unicast and multicast addresses */
1498                         dev_uc_flush(bond_dev);
1499                         dev_mc_flush(bond_dev);
1500
1501                         if (slave_dev->type != ARPHRD_ETHER)
1502                                 bond_setup_by_slave(bond_dev, slave_dev);
1503                         else
1504                                 ether_setup(bond_dev);
1505
1506                         netdev_bonding_change(bond_dev,
1507                                               NETDEV_POST_TYPE_CHANGE);
1508                 }
1509         } else if (bond_dev->type != slave_dev->type) {
1510                 pr_err("%s ether type (%d) is different from other slaves (%d), can not enslave it.\n",
1511                        slave_dev->name,
1512                        slave_dev->type, bond_dev->type);
1513                 res = -EINVAL;
1514                 goto err_undo_flags;
1515         }
1516
1517         if (slave_ops->ndo_set_mac_address == NULL) {
1518                 if (bond->slave_cnt == 0) {
1519                         pr_warning("%s: Warning: The first slave device specified does not support setting the MAC address. Setting fail_over_mac to active.",
1520                                    bond_dev->name);
1521                         bond->params.fail_over_mac = BOND_FOM_ACTIVE;
1522                 } else if (bond->params.fail_over_mac != BOND_FOM_ACTIVE) {
1523                         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",
1524                                bond_dev->name);
1525                         res = -EOPNOTSUPP;
1526                         goto err_undo_flags;
1527                 }
1528         }
1529
1530         /* If this is the first slave, then we need to set the master's hardware
1531          * address to be the same as the slave's. */
1532         if (bond->slave_cnt == 0)
1533                 memcpy(bond->dev->dev_addr, slave_dev->dev_addr,
1534                        slave_dev->addr_len);
1535
1536
1537         new_slave = kzalloc(sizeof(struct slave), GFP_KERNEL);
1538         if (!new_slave) {
1539                 res = -ENOMEM;
1540                 goto err_undo_flags;
1541         }
1542
1543         /*
1544          * Set the new_slave's queue_id to be zero.  Queue ID mapping
1545          * is set via sysfs or module option if desired.
1546          */
1547         new_slave->queue_id = 0;
1548
1549         /* Save slave's original mtu and then set it to match the bond */
1550         new_slave->original_mtu = slave_dev->mtu;
1551         res = dev_set_mtu(slave_dev, bond->dev->mtu);
1552         if (res) {
1553                 pr_debug("Error %d calling dev_set_mtu\n", res);
1554                 goto err_free;
1555         }
1556
1557         /*
1558          * Save slave's original ("permanent") mac address for modes
1559          * that need it, and for restoring it upon release, and then
1560          * set it to the master's address
1561          */
1562         memcpy(new_slave->perm_hwaddr, slave_dev->dev_addr, ETH_ALEN);
1563
1564         if (!bond->params.fail_over_mac) {
1565                 /*
1566                  * Set slave to master's mac address.  The application already
1567                  * set the master's mac address to that of the first slave
1568                  */
1569                 memcpy(addr.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
1570                 addr.sa_family = slave_dev->type;
1571                 res = dev_set_mac_address(slave_dev, &addr);
1572                 if (res) {
1573                         pr_debug("Error %d calling set_mac_address\n", res);
1574                         goto err_restore_mtu;
1575                 }
1576         }
1577
1578         res = netdev_set_master(slave_dev, bond_dev);
1579         if (res) {
1580                 pr_debug("Error %d calling netdev_set_master\n", res);
1581                 goto err_restore_mac;
1582         }
1583         /* open the slave since the application closed it */
1584         res = dev_open(slave_dev);
1585         if (res) {
1586                 pr_debug("Opening slave %s failed\n", slave_dev->name);
1587                 goto err_unset_master;
1588         }
1589
1590         new_slave->dev = slave_dev;
1591         slave_dev->priv_flags |= IFF_BONDING;
1592
1593         if (bond_is_lb(bond)) {
1594                 /* bond_alb_init_slave() must be called before all other stages since
1595                  * it might fail and we do not want to have to undo everything
1596                  */
1597                 res = bond_alb_init_slave(bond, new_slave);
1598                 if (res)
1599                         goto err_close;
1600         }
1601
1602         /* If the mode USES_PRIMARY, then the new slave gets the
1603          * master's promisc (and mc) settings only if it becomes the
1604          * curr_active_slave, and that is taken care of later when calling
1605          * bond_change_active()
1606          */
1607         if (!USES_PRIMARY(bond->params.mode)) {
1608                 /* set promiscuity level to new slave */
1609                 if (bond_dev->flags & IFF_PROMISC) {
1610                         res = dev_set_promiscuity(slave_dev, 1);
1611                         if (res)
1612                                 goto err_close;
1613                 }
1614
1615                 /* set allmulti level to new slave */
1616                 if (bond_dev->flags & IFF_ALLMULTI) {
1617                         res = dev_set_allmulti(slave_dev, 1);
1618                         if (res)
1619                                 goto err_close;
1620                 }
1621
1622                 netif_addr_lock_bh(bond_dev);
1623                 /* upload master's mc_list to new slave */
1624                 netdev_for_each_mc_addr(ha, bond_dev)
1625                         dev_mc_add(slave_dev, ha->addr);
1626                 netif_addr_unlock_bh(bond_dev);
1627         }
1628
1629         if (bond->params.mode == BOND_MODE_8023AD) {
1630                 /* add lacpdu mc addr to mc list */
1631                 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
1632
1633                 dev_mc_add(slave_dev, lacpdu_multicast);
1634         }
1635
1636         bond_add_vlans_on_slave(bond, slave_dev);
1637
1638         write_lock_bh(&bond->lock);
1639
1640         bond_attach_slave(bond, new_slave);
1641
1642         new_slave->delay = 0;
1643         new_slave->link_failure_count = 0;
1644
1645         bond_compute_features(bond);
1646
1647         write_unlock_bh(&bond->lock);
1648
1649         read_lock(&bond->lock);
1650
1651         new_slave->last_arp_rx = jiffies;
1652
1653         if (bond->params.miimon && !bond->params.use_carrier) {
1654                 link_reporting = bond_check_dev_link(bond, slave_dev, 1);
1655
1656                 if ((link_reporting == -1) && !bond->params.arp_interval) {
1657                         /*
1658                          * miimon is set but a bonded network driver
1659                          * does not support ETHTOOL/MII and
1660                          * arp_interval is not set.  Note: if
1661                          * use_carrier is enabled, we will never go
1662                          * here (because netif_carrier is always
1663                          * supported); thus, we don't need to change
1664                          * the messages for netif_carrier.
1665                          */
1666                         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",
1667                                bond_dev->name, slave_dev->name);
1668                 } else if (link_reporting == -1) {
1669                         /* unable get link status using mii/ethtool */
1670                         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",
1671                                    bond_dev->name, slave_dev->name);
1672                 }
1673         }
1674
1675         /* check for initial state */
1676         if (!bond->params.miimon ||
1677             (bond_check_dev_link(bond, slave_dev, 0) == BMSR_LSTATUS)) {
1678                 if (bond->params.updelay) {
1679                         pr_debug("Initial state of slave_dev is BOND_LINK_BACK\n");
1680                         new_slave->link  = BOND_LINK_BACK;
1681                         new_slave->delay = bond->params.updelay;
1682                 } else {
1683                         pr_debug("Initial state of slave_dev is BOND_LINK_UP\n");
1684                         new_slave->link  = BOND_LINK_UP;
1685                 }
1686                 new_slave->jiffies = jiffies;
1687         } else {
1688                 pr_debug("Initial state of slave_dev is BOND_LINK_DOWN\n");
1689                 new_slave->link  = BOND_LINK_DOWN;
1690         }
1691
1692         if (bond_update_speed_duplex(new_slave) &&
1693             (new_slave->link != BOND_LINK_DOWN)) {
1694                 pr_warning("%s: Warning: failed to get speed and duplex from %s, assumed to be 100Mb/sec and Full.\n",
1695                            bond_dev->name, new_slave->dev->name);
1696
1697                 if (bond->params.mode == BOND_MODE_8023AD) {
1698                         pr_warning("%s: Warning: Operation of 802.3ad mode requires ETHTOOL support in base driver for proper aggregator selection.\n",
1699                                    bond_dev->name);
1700                 }
1701         }
1702
1703         if (USES_PRIMARY(bond->params.mode) && bond->params.primary[0]) {
1704                 /* if there is a primary slave, remember it */
1705                 if (strcmp(bond->params.primary, new_slave->dev->name) == 0) {
1706                         bond->primary_slave = new_slave;
1707                         bond->force_primary = true;
1708                 }
1709         }
1710
1711         write_lock_bh(&bond->curr_slave_lock);
1712
1713         switch (bond->params.mode) {
1714         case BOND_MODE_ACTIVEBACKUP:
1715                 bond_set_slave_inactive_flags(new_slave);
1716                 bond_select_active_slave(bond);
1717                 break;
1718         case BOND_MODE_8023AD:
1719                 /* in 802.3ad mode, the internal mechanism
1720                  * will activate the slaves in the selected
1721                  * aggregator
1722                  */
1723                 bond_set_slave_inactive_flags(new_slave);
1724                 /* if this is the first slave */
1725                 if (bond->slave_cnt == 1) {
1726                         SLAVE_AD_INFO(new_slave).id = 1;
1727                         /* Initialize AD with the number of times that the AD timer is called in 1 second
1728                          * can be called only after the mac address of the bond is set
1729                          */
1730                         bond_3ad_initialize(bond, 1000/AD_TIMER_INTERVAL,
1731                                             bond->params.lacp_fast);
1732                 } else {
1733                         SLAVE_AD_INFO(new_slave).id =
1734                                 SLAVE_AD_INFO(new_slave->prev).id + 1;
1735                 }
1736
1737                 bond_3ad_bind_slave(new_slave);
1738                 break;
1739         case BOND_MODE_TLB:
1740         case BOND_MODE_ALB:
1741                 new_slave->state = BOND_STATE_ACTIVE;
1742                 bond_set_slave_inactive_flags(new_slave);
1743                 bond_select_active_slave(bond);
1744                 break;
1745         default:
1746                 pr_debug("This slave is always active in trunk mode\n");
1747
1748                 /* always active in trunk mode */
1749                 new_slave->state = BOND_STATE_ACTIVE;
1750
1751                 /* In trunking mode there is little meaning to curr_active_slave
1752                  * anyway (it holds no special properties of the bond device),
1753                  * so we can change it without calling change_active_interface()
1754                  */
1755                 if (!bond->curr_active_slave)
1756                         bond->curr_active_slave = new_slave;
1757
1758                 break;
1759         } /* switch(bond_mode) */
1760
1761         write_unlock_bh(&bond->curr_slave_lock);
1762
1763         bond_set_carrier(bond);
1764
1765 #ifdef CONFIG_NET_POLL_CONTROLLER
1766         /*
1767          * Netpoll and bonding is broken, make sure it is not initialized
1768          * until it is fixed.
1769          */
1770         if (disable_netpoll) {
1771                 bond_dev->priv_flags |= IFF_DISABLE_NETPOLL;
1772         } else {
1773                 if (slaves_support_netpoll(bond_dev)) {
1774                         bond_dev->priv_flags &= ~IFF_DISABLE_NETPOLL;
1775                         if (bond_dev->npinfo)
1776                                 slave_dev->npinfo = bond_dev->npinfo;
1777                 } else if (!(bond_dev->priv_flags & IFF_DISABLE_NETPOLL)) {
1778                         bond_dev->priv_flags |= IFF_DISABLE_NETPOLL;
1779                         pr_info("New slave device %s does not support netpoll\n",
1780                                 slave_dev->name);
1781                         pr_info("Disabling netpoll support for %s\n", bond_dev->name);
1782                 }
1783         }
1784 #endif
1785         read_unlock(&bond->lock);
1786
1787         res = bond_create_slave_symlinks(bond_dev, slave_dev);
1788         if (res)
1789                 goto err_close;
1790
1791         pr_info("%s: enslaving %s as a%s interface with a%s link.\n",
1792                 bond_dev->name, slave_dev->name,
1793                 new_slave->state == BOND_STATE_ACTIVE ? "n active" : " backup",
1794                 new_slave->link != BOND_LINK_DOWN ? "n up" : " down");
1795
1796         /* enslave is successful */
1797         return 0;
1798
1799 /* Undo stages on error */
1800 err_close:
1801         dev_close(slave_dev);
1802
1803 err_unset_master:
1804         netdev_set_master(slave_dev, NULL);
1805
1806 err_restore_mac:
1807         if (!bond->params.fail_over_mac) {
1808                 /* XXX TODO - fom follow mode needs to change master's
1809                  * MAC if this slave's MAC is in use by the bond, or at
1810                  * least print a warning.
1811                  */
1812                 memcpy(addr.sa_data, new_slave->perm_hwaddr, ETH_ALEN);
1813                 addr.sa_family = slave_dev->type;
1814                 dev_set_mac_address(slave_dev, &addr);
1815         }
1816
1817 err_restore_mtu:
1818         dev_set_mtu(slave_dev, new_slave->original_mtu);
1819
1820 err_free:
1821         kfree(new_slave);
1822
1823 err_undo_flags:
1824         bond_dev->features = old_features;
1825
1826         return res;
1827 }
1828
1829 /*
1830  * Try to release the slave device <slave> from the bond device <master>
1831  * It is legal to access curr_active_slave without a lock because all the function
1832  * is write-locked.
1833  *
1834  * The rules for slave state should be:
1835  *   for Active/Backup:
1836  *     Active stays on all backups go down
1837  *   for Bonded connections:
1838  *     The first up interface should be left on and all others downed.
1839  */
1840 int bond_release(struct net_device *bond_dev, struct net_device *slave_dev)
1841 {
1842         struct bonding *bond = netdev_priv(bond_dev);
1843         struct slave *slave, *oldcurrent;
1844         struct sockaddr addr;
1845
1846         /* slave is not a slave or master is not master of this slave */
1847         if (!(slave_dev->flags & IFF_SLAVE) ||
1848             (slave_dev->master != bond_dev)) {
1849                 pr_err("%s: Error: cannot release %s.\n",
1850                        bond_dev->name, slave_dev->name);
1851                 return -EINVAL;
1852         }
1853
1854         netdev_bonding_change(bond_dev, NETDEV_BONDING_DESLAVE);
1855         write_lock_bh(&bond->lock);
1856
1857         slave = bond_get_slave_by_dev(bond, slave_dev);
1858         if (!slave) {
1859                 /* not a slave of this bond */
1860                 pr_info("%s: %s not enslaved\n",
1861                         bond_dev->name, slave_dev->name);
1862                 write_unlock_bh(&bond->lock);
1863                 return -EINVAL;
1864         }
1865
1866         if (!bond->params.fail_over_mac) {
1867                 if (!compare_ether_addr(bond_dev->dev_addr, slave->perm_hwaddr) &&
1868                     bond->slave_cnt > 1)
1869                         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",
1870                                    bond_dev->name, slave_dev->name,
1871                                    slave->perm_hwaddr,
1872                                    bond_dev->name, slave_dev->name);
1873         }
1874
1875         /* Inform AD package of unbinding of slave. */
1876         if (bond->params.mode == BOND_MODE_8023AD) {
1877                 /* must be called before the slave is
1878                  * detached from the list
1879                  */
1880                 bond_3ad_unbind_slave(slave);
1881         }
1882
1883         pr_info("%s: releasing %s interface %s\n",
1884                 bond_dev->name,
1885                 (slave->state == BOND_STATE_ACTIVE) ? "active" : "backup",
1886                 slave_dev->name);
1887
1888         oldcurrent = bond->curr_active_slave;
1889
1890         bond->current_arp_slave = NULL;
1891
1892         /* release the slave from its bond */
1893         bond_detach_slave(bond, slave);
1894
1895         bond_compute_features(bond);
1896
1897         if (bond->primary_slave == slave)
1898                 bond->primary_slave = NULL;
1899
1900         if (oldcurrent == slave)
1901                 bond_change_active_slave(bond, NULL);
1902
1903         if (bond_is_lb(bond)) {
1904                 /* Must be called only after the slave has been
1905                  * detached from the list and the curr_active_slave
1906                  * has been cleared (if our_slave == old_current),
1907                  * but before a new active slave is selected.
1908                  */
1909                 write_unlock_bh(&bond->lock);
1910                 bond_alb_deinit_slave(bond, slave);
1911                 write_lock_bh(&bond->lock);
1912         }
1913
1914         if (oldcurrent == slave) {
1915                 /*
1916                  * Note that we hold RTNL over this sequence, so there
1917                  * is no concern that another slave add/remove event
1918                  * will interfere.
1919                  */
1920                 write_unlock_bh(&bond->lock);
1921                 read_lock(&bond->lock);
1922                 write_lock_bh(&bond->curr_slave_lock);
1923
1924                 bond_select_active_slave(bond);
1925
1926                 write_unlock_bh(&bond->curr_slave_lock);
1927                 read_unlock(&bond->lock);
1928                 write_lock_bh(&bond->lock);
1929         }
1930
1931         if (bond->slave_cnt == 0) {
1932                 bond_set_carrier(bond);
1933
1934                 /* if the last slave was removed, zero the mac address
1935                  * of the master so it will be set by the application
1936                  * to the mac address of the first slave
1937                  */
1938                 memset(bond_dev->dev_addr, 0, bond_dev->addr_len);
1939
1940                 if (!bond->vlgrp) {
1941                         bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
1942                 } else {
1943                         pr_warning("%s: Warning: clearing HW address of %s while it still has VLANs.\n",
1944                                    bond_dev->name, bond_dev->name);
1945                         pr_warning("%s: When re-adding slaves, make sure the bond's HW address matches its VLANs'.\n",
1946                                    bond_dev->name);
1947                 }
1948         } else if ((bond_dev->features & NETIF_F_VLAN_CHALLENGED) &&
1949                    !bond_has_challenged_slaves(bond)) {
1950                 pr_info("%s: last VLAN challenged slave %s left bond %s. VLAN blocking is removed\n",
1951                         bond_dev->name, slave_dev->name, bond_dev->name);
1952                 bond_dev->features &= ~NETIF_F_VLAN_CHALLENGED;
1953         }
1954
1955         write_unlock_bh(&bond->lock);
1956
1957         /* must do this from outside any spinlocks */
1958         bond_destroy_slave_symlinks(bond_dev, slave_dev);
1959
1960         bond_del_vlans_from_slave(bond, slave_dev);
1961
1962         /* If the mode USES_PRIMARY, then we should only remove its
1963          * promisc and mc settings if it was the curr_active_slave, but that was
1964          * already taken care of above when we detached the slave
1965          */
1966         if (!USES_PRIMARY(bond->params.mode)) {
1967                 /* unset promiscuity level from slave */
1968                 if (bond_dev->flags & IFF_PROMISC)
1969                         dev_set_promiscuity(slave_dev, -1);
1970
1971                 /* unset allmulti level from slave */
1972                 if (bond_dev->flags & IFF_ALLMULTI)
1973                         dev_set_allmulti(slave_dev, -1);
1974
1975                 /* flush master's mc_list from slave */
1976                 netif_addr_lock_bh(bond_dev);
1977                 bond_mc_list_flush(bond_dev, slave_dev);
1978                 netif_addr_unlock_bh(bond_dev);
1979         }
1980
1981         netdev_set_master(slave_dev, NULL);
1982
1983 #ifdef CONFIG_NET_POLL_CONTROLLER
1984         read_lock_bh(&bond->lock);
1985
1986          /* Make sure netpoll over stays disabled until fixed. */
1987         if (!disable_netpoll)
1988                 if (slaves_support_netpoll(bond_dev))
1989                                 bond_dev->priv_flags &= ~IFF_DISABLE_NETPOLL;
1990         read_unlock_bh(&bond->lock);
1991         if (slave_dev->netdev_ops->ndo_netpoll_cleanup)
1992                 slave_dev->netdev_ops->ndo_netpoll_cleanup(slave_dev);
1993         else
1994                 slave_dev->npinfo = NULL;
1995 #endif
1996
1997         /* close slave before restoring its mac address */
1998         dev_close(slave_dev);
1999
2000         if (bond->params.fail_over_mac != BOND_FOM_ACTIVE) {
2001                 /* restore original ("permanent") mac address */
2002                 memcpy(addr.sa_data, slave->perm_hwaddr, ETH_ALEN);
2003                 addr.sa_family = slave_dev->type;
2004                 dev_set_mac_address(slave_dev, &addr);
2005         }
2006
2007         dev_set_mtu(slave_dev, slave->original_mtu);
2008
2009         slave_dev->priv_flags &= ~(IFF_MASTER_8023AD | IFF_MASTER_ALB |
2010                                    IFF_SLAVE_INACTIVE | IFF_BONDING |
2011                                    IFF_SLAVE_NEEDARP);
2012
2013         kfree(slave);
2014
2015         return 0;  /* deletion OK */
2016 }
2017
2018 /*
2019 * First release a slave and than destroy the bond if no more slaves are left.
2020 * Must be under rtnl_lock when this function is called.
2021 */
2022 int  bond_release_and_destroy(struct net_device *bond_dev,
2023                               struct net_device *slave_dev)
2024 {
2025         struct bonding *bond = netdev_priv(bond_dev);
2026         int ret;
2027
2028         ret = bond_release(bond_dev, slave_dev);
2029         if ((ret == 0) && (bond->slave_cnt == 0)) {
2030                 pr_info("%s: destroying bond %s.\n",
2031                         bond_dev->name, bond_dev->name);
2032                 unregister_netdevice(bond_dev);
2033         }
2034         return ret;
2035 }
2036
2037 /*
2038  * This function releases all slaves.
2039  */
2040 static int bond_release_all(struct net_device *bond_dev)
2041 {
2042         struct bonding *bond = netdev_priv(bond_dev);
2043         struct slave *slave;
2044         struct net_device *slave_dev;
2045         struct sockaddr addr;
2046
2047         write_lock_bh(&bond->lock);
2048
2049         netif_carrier_off(bond_dev);
2050
2051         if (bond->slave_cnt == 0)
2052                 goto out;
2053
2054         bond->current_arp_slave = NULL;
2055         bond->primary_slave = NULL;
2056         bond_change_active_slave(bond, NULL);
2057
2058         while ((slave = bond->first_slave) != NULL) {
2059                 /* Inform AD package of unbinding of slave
2060                  * before slave is detached from the list.
2061                  */
2062                 if (bond->params.mode == BOND_MODE_8023AD)
2063                         bond_3ad_unbind_slave(slave);
2064
2065                 slave_dev = slave->dev;
2066                 bond_detach_slave(bond, slave);
2067
2068                 /* now that the slave is detached, unlock and perform
2069                  * all the undo steps that should not be called from
2070                  * within a lock.
2071                  */
2072                 write_unlock_bh(&bond->lock);
2073
2074                 if (bond_is_lb(bond)) {
2075                         /* must be called only after the slave
2076                          * has been detached from the list
2077                          */
2078                         bond_alb_deinit_slave(bond, slave);
2079                 }
2080
2081                 bond_compute_features(bond);
2082
2083                 bond_destroy_slave_symlinks(bond_dev, slave_dev);
2084                 bond_del_vlans_from_slave(bond, slave_dev);
2085
2086                 /* If the mode USES_PRIMARY, then we should only remove its
2087                  * promisc and mc settings if it was the curr_active_slave, but that was
2088                  * already taken care of above when we detached the slave
2089                  */
2090                 if (!USES_PRIMARY(bond->params.mode)) {
2091                         /* unset promiscuity level from slave */
2092                         if (bond_dev->flags & IFF_PROMISC)
2093                                 dev_set_promiscuity(slave_dev, -1);
2094
2095                         /* unset allmulti level from slave */
2096                         if (bond_dev->flags & IFF_ALLMULTI)
2097                                 dev_set_allmulti(slave_dev, -1);
2098
2099                         /* flush master's mc_list from slave */
2100                         netif_addr_lock_bh(bond_dev);
2101                         bond_mc_list_flush(bond_dev, slave_dev);
2102                         netif_addr_unlock_bh(bond_dev);
2103                 }
2104
2105                 netdev_set_master(slave_dev, NULL);
2106
2107                 /* close slave before restoring its mac address */
2108                 dev_close(slave_dev);
2109
2110                 if (!bond->params.fail_over_mac) {
2111                         /* restore original ("permanent") mac address*/
2112                         memcpy(addr.sa_data, slave->perm_hwaddr, ETH_ALEN);
2113                         addr.sa_family = slave_dev->type;
2114                         dev_set_mac_address(slave_dev, &addr);
2115                 }
2116
2117                 slave_dev->priv_flags &= ~(IFF_MASTER_8023AD | IFF_MASTER_ALB |
2118                                            IFF_SLAVE_INACTIVE);
2119
2120                 kfree(slave);
2121
2122                 /* re-acquire the lock before getting the next slave */
2123                 write_lock_bh(&bond->lock);
2124         }
2125
2126         /* zero the mac address of the master so it will be
2127          * set by the application to the mac address of the
2128          * first slave
2129          */
2130         memset(bond_dev->dev_addr, 0, bond_dev->addr_len);
2131
2132         if (!bond->vlgrp) {
2133                 bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
2134         } else {
2135                 pr_warning("%s: Warning: clearing HW address of %s while it still has VLANs.\n",
2136                            bond_dev->name, bond_dev->name);
2137                 pr_warning("%s: When re-adding slaves, make sure the bond's HW address matches its VLANs'.\n",
2138                            bond_dev->name);
2139         }
2140
2141         pr_info("%s: released all slaves\n", bond_dev->name);
2142
2143 out:
2144         write_unlock_bh(&bond->lock);
2145
2146         return 0;
2147 }
2148
2149 /*
2150  * This function changes the active slave to slave <slave_dev>.
2151  * It returns -EINVAL in the following cases.
2152  *  - <slave_dev> is not found in the list.
2153  *  - There is not active slave now.
2154  *  - <slave_dev> is already active.
2155  *  - The link state of <slave_dev> is not BOND_LINK_UP.
2156  *  - <slave_dev> is not running.
2157  * In these cases, this function does nothing.
2158  * In the other cases, current_slave pointer is changed and 0 is returned.
2159  */
2160 static int bond_ioctl_change_active(struct net_device *bond_dev, struct net_device *slave_dev)
2161 {
2162         struct bonding *bond = netdev_priv(bond_dev);
2163         struct slave *old_active = NULL;
2164         struct slave *new_active = NULL;
2165         int res = 0;
2166
2167         if (!USES_PRIMARY(bond->params.mode))
2168                 return -EINVAL;
2169
2170         /* Verify that master_dev is indeed the master of slave_dev */
2171         if (!(slave_dev->flags & IFF_SLAVE) || (slave_dev->master != bond_dev))
2172                 return -EINVAL;
2173
2174         read_lock(&bond->lock);
2175
2176         read_lock(&bond->curr_slave_lock);
2177         old_active = bond->curr_active_slave;
2178         read_unlock(&bond->curr_slave_lock);
2179
2180         new_active = bond_get_slave_by_dev(bond, slave_dev);
2181
2182         /*
2183          * Changing to the current active: do nothing; return success.
2184          */
2185         if (new_active && (new_active == old_active)) {
2186                 read_unlock(&bond->lock);
2187                 return 0;
2188         }
2189
2190         if ((new_active) &&
2191             (old_active) &&
2192             (new_active->link == BOND_LINK_UP) &&
2193             IS_UP(new_active->dev)) {
2194                 write_lock_bh(&bond->curr_slave_lock);
2195                 bond_change_active_slave(bond, new_active);
2196                 write_unlock_bh(&bond->curr_slave_lock);
2197         } else
2198                 res = -EINVAL;
2199
2200         read_unlock(&bond->lock);
2201
2202         return res;
2203 }
2204
2205 static int bond_info_query(struct net_device *bond_dev, struct ifbond *info)
2206 {
2207         struct bonding *bond = netdev_priv(bond_dev);
2208
2209         info->bond_mode = bond->params.mode;
2210         info->miimon = bond->params.miimon;
2211
2212         read_lock(&bond->lock);
2213         info->num_slaves = bond->slave_cnt;
2214         read_unlock(&bond->lock);
2215
2216         return 0;
2217 }
2218
2219 static int bond_slave_info_query(struct net_device *bond_dev, struct ifslave *info)
2220 {
2221         struct bonding *bond = netdev_priv(bond_dev);
2222         struct slave *slave;
2223         int i, res = -ENODEV;
2224
2225         read_lock(&bond->lock);
2226
2227         bond_for_each_slave(bond, slave, i) {
2228                 if (i == (int)info->slave_id) {
2229                         res = 0;
2230                         strcpy(info->slave_name, slave->dev->name);
2231                         info->link = slave->link;
2232                         info->state = slave->state;
2233                         info->link_failure_count = slave->link_failure_count;
2234                         break;
2235                 }
2236         }
2237
2238         read_unlock(&bond->lock);
2239
2240         return res;
2241 }
2242
2243 /*-------------------------------- Monitoring -------------------------------*/
2244
2245
2246 static int bond_miimon_inspect(struct bonding *bond)
2247 {
2248         struct slave *slave;
2249         int i, link_state, commit = 0;
2250         bool ignore_updelay;
2251
2252         ignore_updelay = !bond->curr_active_slave ? true : false;
2253
2254         bond_for_each_slave(bond, slave, i) {
2255                 slave->new_link = BOND_LINK_NOCHANGE;
2256
2257                 link_state = bond_check_dev_link(bond, slave->dev, 0);
2258
2259                 switch (slave->link) {
2260                 case BOND_LINK_UP:
2261                         if (link_state)
2262                                 continue;
2263
2264                         slave->link = BOND_LINK_FAIL;
2265                         slave->delay = bond->params.downdelay;
2266                         if (slave->delay) {
2267                                 pr_info("%s: link status down for %sinterface %s, disabling it in %d ms.\n",
2268                                         bond->dev->name,
2269                                         (bond->params.mode ==
2270                                          BOND_MODE_ACTIVEBACKUP) ?
2271                                         ((slave->state == BOND_STATE_ACTIVE) ?
2272                                          "active " : "backup ") : "",
2273                                         slave->dev->name,
2274                                         bond->params.downdelay * bond->params.miimon);
2275                         }
2276                         /*FALLTHRU*/
2277                 case BOND_LINK_FAIL:
2278                         if (link_state) {
2279                                 /*
2280                                  * recovered before downdelay expired
2281                                  */
2282                                 slave->link = BOND_LINK_UP;
2283                                 slave->jiffies = jiffies;
2284                                 pr_info("%s: link status up again after %d ms for interface %s.\n",
2285                                         bond->dev->name,
2286                                         (bond->params.downdelay - slave->delay) *
2287                                         bond->params.miimon,
2288                                         slave->dev->name);
2289                                 continue;
2290                         }
2291
2292                         if (slave->delay <= 0) {
2293                                 slave->new_link = BOND_LINK_DOWN;
2294                                 commit++;
2295                                 continue;
2296                         }
2297
2298                         slave->delay--;
2299                         break;
2300
2301                 case BOND_LINK_DOWN:
2302                         if (!link_state)
2303                                 continue;
2304
2305                         slave->link = BOND_LINK_BACK;
2306                         slave->delay = bond->params.updelay;
2307
2308                         if (slave->delay) {
2309                                 pr_info("%s: link status up for interface %s, enabling it in %d ms.\n",
2310                                         bond->dev->name, slave->dev->name,
2311                                         ignore_updelay ? 0 :
2312                                         bond->params.updelay *
2313                                         bond->params.miimon);
2314                         }
2315                         /*FALLTHRU*/
2316                 case BOND_LINK_BACK:
2317                         if (!link_state) {
2318                                 slave->link = BOND_LINK_DOWN;
2319                                 pr_info("%s: link status down again after %d ms for interface %s.\n",
2320                                         bond->dev->name,
2321                                         (bond->params.updelay - slave->delay) *
2322                                         bond->params.miimon,
2323                                         slave->dev->name);
2324
2325                                 continue;
2326                         }
2327
2328                         if (ignore_updelay)
2329                                 slave->delay = 0;
2330
2331                         if (slave->delay <= 0) {
2332                                 slave->new_link = BOND_LINK_UP;
2333                                 commit++;
2334                                 ignore_updelay = false;
2335                                 continue;
2336                         }
2337
2338                         slave->delay--;
2339                         break;
2340                 }
2341         }
2342
2343         return commit;
2344 }
2345
2346 static void bond_miimon_commit(struct bonding *bond)
2347 {
2348         struct slave *slave;
2349         int i;
2350
2351         bond_for_each_slave(bond, slave, i) {
2352                 switch (slave->new_link) {
2353                 case BOND_LINK_NOCHANGE:
2354                         continue;
2355
2356                 case BOND_LINK_UP:
2357                         slave->link = BOND_LINK_UP;
2358                         slave->jiffies = jiffies;
2359
2360                         if (bond->params.mode == BOND_MODE_8023AD) {
2361                                 /* prevent it from being the active one */
2362                                 slave->state = BOND_STATE_BACKUP;
2363                         } else if (bond->params.mode != BOND_MODE_ACTIVEBACKUP) {
2364                                 /* make it immediately active */
2365                                 slave->state = BOND_STATE_ACTIVE;
2366                         } else if (slave != bond->primary_slave) {
2367                                 /* prevent it from being the active one */
2368                                 slave->state = BOND_STATE_BACKUP;
2369                         }
2370
2371                         pr_info("%s: link status definitely up for interface %s.\n",
2372                                 bond->dev->name, slave->dev->name);
2373
2374                         /* notify ad that the link status has changed */
2375                         if (bond->params.mode == BOND_MODE_8023AD)
2376                                 bond_3ad_handle_link_change(slave, BOND_LINK_UP);
2377
2378                         if (bond_is_lb(bond))
2379                                 bond_alb_handle_link_change(bond, slave,
2380                                                             BOND_LINK_UP);
2381
2382                         if (!bond->curr_active_slave ||
2383                             (slave == bond->primary_slave))
2384                                 goto do_failover;
2385
2386                         continue;
2387
2388                 case BOND_LINK_DOWN:
2389                         if (slave->link_failure_count < UINT_MAX)
2390                                 slave->link_failure_count++;
2391
2392                         slave->link = BOND_LINK_DOWN;
2393
2394                         if (bond->params.mode == BOND_MODE_ACTIVEBACKUP ||
2395                             bond->params.mode == BOND_MODE_8023AD)
2396                                 bond_set_slave_inactive_flags(slave);
2397
2398                         pr_info("%s: link status definitely down for interface %s, disabling it\n",
2399                                 bond->dev->name, slave->dev->name);
2400
2401                         if (bond->params.mode == BOND_MODE_8023AD)
2402                                 bond_3ad_handle_link_change(slave,
2403                                                             BOND_LINK_DOWN);
2404
2405                         if (bond_is_lb(bond))
2406                                 bond_alb_handle_link_change(bond, slave,
2407                                                             BOND_LINK_DOWN);
2408
2409                         if (slave == bond->curr_active_slave)
2410                                 goto do_failover;
2411
2412                         continue;
2413
2414                 default:
2415                         pr_err("%s: invalid new link %d on slave %s\n",
2416                                bond->dev->name, slave->new_link,
2417                                slave->dev->name);
2418                         slave->new_link = BOND_LINK_NOCHANGE;
2419
2420                         continue;
2421                 }
2422
2423 do_failover:
2424                 ASSERT_RTNL();
2425                 write_lock_bh(&bond->curr_slave_lock);
2426                 bond_select_active_slave(bond);
2427                 write_unlock_bh(&bond->curr_slave_lock);
2428         }
2429
2430         bond_set_carrier(bond);
2431 }
2432
2433 /*
2434  * bond_mii_monitor
2435  *
2436  * Really a wrapper that splits the mii monitor into two phases: an
2437  * inspection, then (if inspection indicates something needs to be done)
2438  * an acquisition of appropriate locks followed by a commit phase to
2439  * implement whatever link state changes are indicated.
2440  */
2441 void bond_mii_monitor(struct work_struct *work)
2442 {
2443         struct bonding *bond = container_of(work, struct bonding,
2444                                             mii_work.work);
2445
2446         read_lock(&bond->lock);
2447         if (bond->kill_timers)
2448                 goto out;
2449
2450         if (bond->slave_cnt == 0)
2451                 goto re_arm;
2452
2453         if (bond->send_grat_arp) {
2454                 read_lock(&bond->curr_slave_lock);
2455                 bond_send_gratuitous_arp(bond);
2456                 read_unlock(&bond->curr_slave_lock);
2457         }
2458
2459         if (bond->send_unsol_na) {
2460                 read_lock(&bond->curr_slave_lock);
2461                 bond_send_unsolicited_na(bond);
2462                 read_unlock(&bond->curr_slave_lock);
2463         }
2464
2465         if (bond_miimon_inspect(bond)) {
2466                 read_unlock(&bond->lock);
2467                 rtnl_lock();
2468                 read_lock(&bond->lock);
2469
2470                 bond_miimon_commit(bond);
2471
2472                 read_unlock(&bond->lock);
2473                 rtnl_unlock();  /* might sleep, hold no other locks */
2474                 read_lock(&bond->lock);
2475         }
2476
2477 re_arm:
2478         if (bond->params.miimon)
2479                 queue_delayed_work(bond->wq, &bond->mii_work,
2480                                    msecs_to_jiffies(bond->params.miimon));
2481 out:
2482         read_unlock(&bond->lock);
2483 }
2484
2485 static __be32 bond_glean_dev_ip(struct net_device *dev)
2486 {
2487         struct in_device *idev;
2488         struct in_ifaddr *ifa;
2489         __be32 addr = 0;
2490
2491         if (!dev)
2492                 return 0;
2493
2494         rcu_read_lock();
2495         idev = __in_dev_get_rcu(dev);
2496         if (!idev)
2497                 goto out;
2498
2499         ifa = idev->ifa_list;
2500         if (!ifa)
2501                 goto out;
2502
2503         addr = ifa->ifa_local;
2504 out:
2505         rcu_read_unlock();
2506         return addr;
2507 }
2508
2509 static int bond_has_this_ip(struct bonding *bond, __be32 ip)
2510 {
2511         struct vlan_entry *vlan;
2512
2513         if (ip == bond->master_ip)
2514                 return 1;
2515
2516         list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
2517                 if (ip == vlan->vlan_ip)
2518                         return 1;
2519         }
2520
2521         return 0;
2522 }
2523
2524 /*
2525  * We go to the (large) trouble of VLAN tagging ARP frames because
2526  * switches in VLAN mode (especially if ports are configured as
2527  * "native" to a VLAN) might not pass non-tagged frames.
2528  */
2529 static void bond_arp_send(struct net_device *slave_dev, int arp_op, __be32 dest_ip, __be32 src_ip, unsigned short vlan_id)
2530 {
2531         struct sk_buff *skb;
2532
2533         pr_debug("arp %d on slave %s: dst %x src %x vid %d\n", arp_op,
2534                  slave_dev->name, dest_ip, src_ip, vlan_id);
2535
2536         skb = arp_create(arp_op, ETH_P_ARP, dest_ip, slave_dev, src_ip,
2537                          NULL, slave_dev->dev_addr, NULL);
2538
2539         if (!skb) {
2540                 pr_err("ARP packet allocation failed\n");
2541                 return;
2542         }
2543         if (vlan_id) {
2544                 skb = vlan_put_tag(skb, vlan_id);
2545                 if (!skb) {
2546                         pr_err("failed to insert VLAN tag\n");
2547                         return;
2548                 }
2549         }
2550         arp_xmit(skb);
2551 }
2552
2553
2554 static void bond_arp_send_all(struct bonding *bond, struct slave *slave)
2555 {
2556         int i, vlan_id, rv;
2557         __be32 *targets = bond->params.arp_targets;
2558         struct vlan_entry *vlan;
2559         struct net_device *vlan_dev;
2560         struct flowi fl;
2561         struct rtable *rt;
2562
2563         for (i = 0; (i < BOND_MAX_ARP_TARGETS); i++) {
2564                 if (!targets[i])
2565                         break;
2566                 pr_debug("basa: target %x\n", targets[i]);
2567                 if (!bond->vlgrp) {
2568                         pr_debug("basa: empty vlan: arp_send\n");
2569                         bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2570                                       bond->master_ip, 0);
2571                         continue;
2572                 }
2573
2574                 /*
2575                  * If VLANs are configured, we do a route lookup to
2576                  * determine which VLAN interface would be used, so we
2577                  * can tag the ARP with the proper VLAN tag.
2578                  */
2579                 memset(&fl, 0, sizeof(fl));
2580                 fl.fl4_dst = targets[i];
2581                 fl.fl4_tos = RTO_ONLINK;
2582
2583                 rv = ip_route_output_key(dev_net(bond->dev), &rt, &fl);
2584                 if (rv) {
2585                         if (net_ratelimit()) {
2586                                 pr_warning("%s: no route to arp_ip_target %pI4\n",
2587                                            bond->dev->name, &fl.fl4_dst);
2588                         }
2589                         continue;
2590                 }
2591
2592                 /*
2593                  * This target is not on a VLAN
2594                  */
2595                 if (rt->dst.dev == bond->dev) {
2596                         ip_rt_put(rt);
2597                         pr_debug("basa: rtdev == bond->dev: arp_send\n");
2598                         bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2599                                       bond->master_ip, 0);
2600                         continue;
2601                 }
2602
2603                 vlan_id = 0;
2604                 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
2605                         vlan_dev = vlan_group_get_device(bond->vlgrp, vlan->vlan_id);
2606                         if (vlan_dev == rt->dst.dev) {
2607                                 vlan_id = vlan->vlan_id;
2608                                 pr_debug("basa: vlan match on %s %d\n",
2609                                        vlan_dev->name, vlan_id);
2610                                 break;
2611                         }
2612                 }
2613
2614                 if (vlan_id) {
2615                         ip_rt_put(rt);
2616                         bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2617                                       vlan->vlan_ip, vlan_id);
2618                         continue;
2619                 }
2620
2621                 if (net_ratelimit()) {
2622                         pr_warning("%s: no path to arp_ip_target %pI4 via rt.dev %s\n",
2623                                    bond->dev->name, &fl.fl4_dst,
2624                                    rt->dst.dev ? rt->dst.dev->name : "NULL");
2625                 }
2626                 ip_rt_put(rt);
2627         }
2628 }
2629
2630 /*
2631  * Kick out a gratuitous ARP for an IP on the bonding master plus one
2632  * for each VLAN above us.
2633  *
2634  * Caller must hold curr_slave_lock for read or better
2635  */
2636 static void bond_send_gratuitous_arp(struct bonding *bond)
2637 {
2638         struct slave *slave = bond->curr_active_slave;
2639         struct vlan_entry *vlan;
2640         struct net_device *vlan_dev;
2641
2642         pr_debug("bond_send_grat_arp: bond %s slave %s\n",
2643                  bond->dev->name, slave ? slave->dev->name : "NULL");
2644
2645         if (!slave || !bond->send_grat_arp ||
2646             test_bit(__LINK_STATE_LINKWATCH_PENDING, &slave->dev->state))
2647                 return;
2648
2649         bond->send_grat_arp--;
2650
2651         if (bond->master_ip) {
2652                 bond_arp_send(slave->dev, ARPOP_REPLY, bond->master_ip,
2653                                 bond->master_ip, 0);
2654         }
2655
2656         if (!bond->vlgrp)
2657                 return;
2658
2659         list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
2660                 vlan_dev = vlan_group_get_device(bond->vlgrp, vlan->vlan_id);
2661                 if (vlan->vlan_ip) {
2662                         bond_arp_send(slave->dev, ARPOP_REPLY, vlan->vlan_ip,
2663                                       vlan->vlan_ip, vlan->vlan_id);
2664                 }
2665         }
2666 }
2667
2668 static void bond_validate_arp(struct bonding *bond, struct slave *slave, __be32 sip, __be32 tip)
2669 {
2670         int i;
2671         __be32 *targets = bond->params.arp_targets;
2672
2673         for (i = 0; (i < BOND_MAX_ARP_TARGETS) && targets[i]; i++) {
2674                 pr_debug("bva: sip %pI4 tip %pI4 t[%d] %pI4 bhti(tip) %d\n",
2675                          &sip, &tip, i, &targets[i],
2676                          bond_has_this_ip(bond, tip));
2677                 if (sip == targets[i]) {
2678                         if (bond_has_this_ip(bond, tip))
2679                                 slave->last_arp_rx = jiffies;
2680                         return;
2681                 }
2682         }
2683 }
2684
2685 static int bond_arp_rcv(struct sk_buff *skb, struct net_device *dev, struct packet_type *pt, struct net_device *orig_dev)
2686 {
2687         struct arphdr *arp;
2688         struct slave *slave;
2689         struct bonding *bond;
2690         unsigned char *arp_ptr;
2691         __be32 sip, tip;
2692
2693         if (dev->priv_flags & IFF_802_1Q_VLAN) {
2694                 /*
2695                  * When using VLANS and bonding, dev and oriv_dev may be
2696                  * incorrect if the physical interface supports VLAN
2697                  * acceleration.  With this change ARP validation now
2698                  * works for hosts only reachable on the VLAN interface.
2699                  */
2700                 dev = vlan_dev_real_dev(dev);
2701                 orig_dev = dev_get_by_index_rcu(dev_net(skb->dev),skb->skb_iif);
2702         }
2703
2704         if (!(dev->priv_flags & IFF_BONDING) || !(dev->flags & IFF_MASTER))
2705                 goto out;
2706
2707         bond = netdev_priv(dev);
2708         read_lock(&bond->lock);
2709
2710         pr_debug("bond_arp_rcv: bond %s skb->dev %s orig_dev %s\n",
2711                  bond->dev->name, skb->dev ? skb->dev->name : "NULL",
2712                  orig_dev ? orig_dev->name : "NULL");
2713
2714         slave = bond_get_slave_by_dev(bond, orig_dev);
2715         if (!slave || !slave_do_arp_validate(bond, slave))
2716                 goto out_unlock;
2717
2718         if (!pskb_may_pull(skb, arp_hdr_len(dev)))
2719                 goto out_unlock;
2720
2721         arp = arp_hdr(skb);
2722         if (arp->ar_hln != dev->addr_len ||
2723             skb->pkt_type == PACKET_OTHERHOST ||
2724             skb->pkt_type == PACKET_LOOPBACK ||
2725             arp->ar_hrd != htons(ARPHRD_ETHER) ||
2726             arp->ar_pro != htons(ETH_P_IP) ||
2727             arp->ar_pln != 4)
2728                 goto out_unlock;
2729
2730         arp_ptr = (unsigned char *)(arp + 1);
2731         arp_ptr += dev->addr_len;
2732         memcpy(&sip, arp_ptr, 4);
2733         arp_ptr += 4 + dev->addr_len;
2734         memcpy(&tip, arp_ptr, 4);
2735
2736         pr_debug("bond_arp_rcv: %s %s/%d av %d sv %d sip %pI4 tip %pI4\n",
2737                  bond->dev->name, slave->dev->name, slave->state,
2738                  bond->params.arp_validate, slave_do_arp_validate(bond, slave),
2739                  &sip, &tip);
2740
2741         /*
2742          * Backup slaves won't see the ARP reply, but do come through
2743          * here for each ARP probe (so we swap the sip/tip to validate
2744          * the probe).  In a "redundant switch, common router" type of
2745          * configuration, the ARP probe will (hopefully) travel from
2746          * the active, through one switch, the router, then the other
2747          * switch before reaching the backup.
2748          */
2749         if (slave->state == BOND_STATE_ACTIVE)
2750                 bond_validate_arp(bond, slave, sip, tip);
2751         else
2752                 bond_validate_arp(bond, slave, tip, sip);
2753
2754 out_unlock:
2755         read_unlock(&bond->lock);
2756 out:
2757         dev_kfree_skb(skb);
2758         return NET_RX_SUCCESS;
2759 }
2760
2761 /*
2762  * this function is called regularly to monitor each slave's link
2763  * ensuring that traffic is being sent and received when arp monitoring
2764  * is used in load-balancing mode. if the adapter has been dormant, then an
2765  * arp is transmitted to generate traffic. see activebackup_arp_monitor for
2766  * arp monitoring in active backup mode.
2767  */
2768 void bond_loadbalance_arp_mon(struct work_struct *work)
2769 {
2770         struct bonding *bond = container_of(work, struct bonding,
2771                                             arp_work.work);
2772         struct slave *slave, *oldcurrent;
2773         int do_failover = 0;
2774         int delta_in_ticks;
2775         int i;
2776
2777         read_lock(&bond->lock);
2778
2779         delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
2780
2781         if (bond->kill_timers)
2782                 goto out;
2783
2784         if (bond->slave_cnt == 0)
2785                 goto re_arm;
2786
2787         read_lock(&bond->curr_slave_lock);
2788         oldcurrent = bond->curr_active_slave;
2789         read_unlock(&bond->curr_slave_lock);
2790
2791         /* see if any of the previous devices are up now (i.e. they have
2792          * xmt and rcv traffic). the curr_active_slave does not come into
2793          * the picture unless it is null. also, slave->jiffies is not needed
2794          * here because we send an arp on each slave and give a slave as
2795          * long as it needs to get the tx/rx within the delta.
2796          * TODO: what about up/down delay in arp mode? it wasn't here before
2797          *       so it can wait
2798          */
2799         bond_for_each_slave(bond, slave, i) {
2800                 unsigned long trans_start = dev_trans_start(slave->dev);
2801
2802                 if (slave->link != BOND_LINK_UP) {
2803                         if (time_in_range(jiffies,
2804                                 trans_start - delta_in_ticks,
2805                                 trans_start + delta_in_ticks) &&
2806                             time_in_range(jiffies,
2807                                 slave->dev->last_rx - delta_in_ticks,
2808                                 slave->dev->last_rx + delta_in_ticks)) {
2809
2810                                 slave->link  = BOND_LINK_UP;
2811                                 slave->state = BOND_STATE_ACTIVE;
2812
2813                                 /* primary_slave has no meaning in round-robin
2814                                  * mode. the window of a slave being up and
2815                                  * curr_active_slave being null after enslaving
2816                                  * is closed.
2817                                  */
2818                                 if (!oldcurrent) {
2819                                         pr_info("%s: link status definitely up for interface %s, ",
2820                                                 bond->dev->name,
2821                                                 slave->dev->name);
2822                                         do_failover = 1;
2823                                 } else {
2824                                         pr_info("%s: interface %s is now up\n",
2825                                                 bond->dev->name,
2826                                                 slave->dev->name);
2827                                 }
2828                         }
2829                 } else {
2830                         /* slave->link == BOND_LINK_UP */
2831
2832                         /* not all switches will respond to an arp request
2833                          * when the source ip is 0, so don't take the link down
2834                          * if we don't know our ip yet
2835                          */
2836                         if (!time_in_range(jiffies,
2837                                 trans_start - delta_in_ticks,
2838                                 trans_start + 2 * delta_in_ticks) ||
2839                             !time_in_range(jiffies,
2840                                 slave->dev->last_rx - delta_in_ticks,
2841                                 slave->dev->last_rx + 2 * delta_in_ticks)) {
2842
2843                                 slave->link  = BOND_LINK_DOWN;
2844                                 slave->state = BOND_STATE_BACKUP;
2845
2846                                 if (slave->link_failure_count < UINT_MAX)
2847                                         slave->link_failure_count++;
2848
2849                                 pr_info("%s: interface %s is now down.\n",
2850                                         bond->dev->name,
2851                                         slave->dev->name);
2852
2853                                 if (slave == oldcurrent)
2854                                         do_failover = 1;
2855                         }
2856                 }
2857
2858                 /* note: if switch is in round-robin mode, all links
2859                  * must tx arp to ensure all links rx an arp - otherwise
2860                  * links may oscillate or not come up at all; if switch is
2861                  * in something like xor mode, there is nothing we can
2862                  * do - all replies will be rx'ed on same link causing slaves
2863                  * to be unstable during low/no traffic periods
2864                  */
2865                 if (IS_UP(slave->dev))
2866                         bond_arp_send_all(bond, slave);
2867         }
2868
2869         if (do_failover) {
2870                 write_lock_bh(&bond->curr_slave_lock);
2871
2872                 bond_select_active_slave(bond);
2873
2874                 write_unlock_bh(&bond->curr_slave_lock);
2875         }
2876
2877 re_arm:
2878         if (bond->params.arp_interval)
2879                 queue_delayed_work(bond->wq, &bond->arp_work, delta_in_ticks);
2880 out:
2881         read_unlock(&bond->lock);
2882 }
2883
2884 /*
2885  * Called to inspect slaves for active-backup mode ARP monitor link state
2886  * changes.  Sets new_link in slaves to specify what action should take
2887  * place for the slave.  Returns 0 if no changes are found, >0 if changes
2888  * to link states must be committed.
2889  *
2890  * Called with bond->lock held for read.
2891  */
2892 static int bond_ab_arp_inspect(struct bonding *bond, int delta_in_ticks)
2893 {
2894         struct slave *slave;
2895         int i, commit = 0;
2896         unsigned long trans_start;
2897
2898         bond_for_each_slave(bond, slave, i) {
2899                 slave->new_link = BOND_LINK_NOCHANGE;
2900
2901                 if (slave->link != BOND_LINK_UP) {
2902                         if (time_in_range(jiffies,
2903                                 slave_last_rx(bond, slave) - delta_in_ticks,
2904                                 slave_last_rx(bond, slave) + delta_in_ticks)) {
2905
2906                                 slave->new_link = BOND_LINK_UP;
2907                                 commit++;
2908                         }
2909
2910                         continue;
2911                 }
2912
2913                 /*
2914                  * Give slaves 2*delta after being enslaved or made
2915                  * active.  This avoids bouncing, as the last receive
2916                  * times need a full ARP monitor cycle to be updated.
2917                  */
2918                 if (time_in_range(jiffies,
2919                                   slave->jiffies - delta_in_ticks,
2920                                   slave->jiffies + 2 * delta_in_ticks))
2921                         continue;
2922
2923                 /*
2924                  * Backup slave is down if:
2925                  * - No current_arp_slave AND
2926                  * - more than 3*delta since last receive AND
2927                  * - the bond has an IP address
2928                  *
2929                  * Note: a non-null current_arp_slave indicates
2930                  * the curr_active_slave went down and we are
2931                  * searching for a new one; under this condition
2932                  * we only take the curr_active_slave down - this
2933                  * gives each slave a chance to tx/rx traffic
2934                  * before being taken out
2935                  */
2936                 if (slave->state == BOND_STATE_BACKUP &&
2937                     !bond->current_arp_slave &&
2938                     !time_in_range(jiffies,
2939                         slave_last_rx(bond, slave) - delta_in_ticks,
2940                         slave_last_rx(bond, slave) + 3 * delta_in_ticks)) {
2941
2942                         slave->new_link = BOND_LINK_DOWN;
2943                         commit++;
2944                 }
2945
2946                 /*
2947                  * Active slave is down if:
2948                  * - more than 2*delta since transmitting OR
2949                  * - (more than 2*delta since receive AND
2950                  *    the bond has an IP address)
2951                  */
2952                 trans_start = dev_trans_start(slave->dev);
2953                 if ((slave->state == BOND_STATE_ACTIVE) &&
2954                     (!time_in_range(jiffies,
2955                         trans_start - delta_in_ticks,
2956                         trans_start + 2 * delta_in_ticks) ||
2957                      !time_in_range(jiffies,
2958                         slave_last_rx(bond, slave) - delta_in_ticks,
2959                         slave_last_rx(bond, slave) + 2 * delta_in_ticks))) {
2960
2961                         slave->new_link = BOND_LINK_DOWN;
2962                         commit++;
2963                 }
2964         }
2965
2966         return commit;
2967 }
2968
2969 /*
2970  * Called to commit link state changes noted by inspection step of
2971  * active-backup mode ARP monitor.
2972  *
2973  * Called with RTNL and bond->lock for read.
2974  */
2975 static void bond_ab_arp_commit(struct bonding *bond, int delta_in_ticks)
2976 {
2977         struct slave *slave;
2978         int i;
2979         unsigned long trans_start;
2980
2981         bond_for_each_slave(bond, slave, i) {
2982                 switch (slave->new_link) {
2983                 case BOND_LINK_NOCHANGE:
2984                         continue;
2985
2986                 case BOND_LINK_UP:
2987                         trans_start = dev_trans_start(slave->dev);
2988                         if ((!bond->curr_active_slave &&
2989                              time_in_range(jiffies,
2990                                            trans_start - delta_in_ticks,
2991                                            trans_start + delta_in_ticks)) ||
2992                             bond->curr_active_slave != slave) {
2993                                 slave->link = BOND_LINK_UP;
2994                                 bond->current_arp_slave = NULL;
2995
2996                                 pr_info("%s: link status definitely up for interface %s.\n",
2997                                         bond->dev->name, slave->dev->name);
2998
2999                                 if (!bond->curr_active_slave ||
3000                                     (slave == bond->primary_slave))
3001                                         goto do_failover;
3002
3003                         }
3004
3005                         continue;
3006
3007                 case BOND_LINK_DOWN:
3008                         if (slave->link_failure_count < UINT_MAX)
3009                                 slave->link_failure_count++;
3010
3011                         slave->link = BOND_LINK_DOWN;
3012                         bond_set_slave_inactive_flags(slave);
3013
3014                         pr_info("%s: link status definitely down for interface %s, disabling it\n",
3015                                 bond->dev->name, slave->dev->name);
3016
3017                         if (slave == bond->curr_active_slave) {
3018                                 bond->current_arp_slave = NULL;
3019                                 goto do_failover;
3020                         }
3021
3022                         continue;
3023
3024                 default:
3025                         pr_err("%s: impossible: new_link %d on slave %s\n",
3026                                bond->dev->name, slave->new_link,
3027                                slave->dev->name);
3028                         continue;
3029                 }
3030
3031 do_failover:
3032                 ASSERT_RTNL();
3033                 write_lock_bh(&bond->curr_slave_lock);
3034                 bond_select_active_slave(bond);
3035                 write_unlock_bh(&bond->curr_slave_lock);
3036         }
3037
3038         bond_set_carrier(bond);
3039 }
3040
3041 /*
3042  * Send ARP probes for active-backup mode ARP monitor.
3043  *
3044  * Called with bond->lock held for read.
3045  */
3046 static void bond_ab_arp_probe(struct bonding *bond)
3047 {
3048         struct slave *slave;
3049         int i;
3050
3051         read_lock(&bond->curr_slave_lock);
3052
3053         if (bond->current_arp_slave && bond->curr_active_slave)
3054                 pr_info("PROBE: c_arp %s && cas %s BAD\n",
3055                         bond->current_arp_slave->dev->name,
3056                         bond->curr_active_slave->dev->name);
3057
3058         if (bond->curr_active_slave) {
3059                 bond_arp_send_all(bond, bond->curr_active_slave);
3060                 read_unlock(&bond->curr_slave_lock);
3061                 return;
3062         }
3063
3064         read_unlock(&bond->curr_slave_lock);
3065
3066         /* if we don't have a curr_active_slave, search for the next available
3067          * backup slave from the current_arp_slave and make it the candidate
3068          * for becoming the curr_active_slave
3069          */
3070
3071         if (!bond->current_arp_slave) {
3072                 bond->current_arp_slave = bond->first_slave;
3073                 if (!bond->current_arp_slave)
3074                         return;
3075         }
3076
3077         bond_set_slave_inactive_flags(bond->current_arp_slave);
3078
3079         /* search for next candidate */
3080         bond_for_each_slave_from(bond, slave, i, bond->current_arp_slave->next) {
3081                 if (IS_UP(slave->dev)) {
3082                         slave->link = BOND_LINK_BACK;
3083                         bond_set_slave_active_flags(slave);
3084                         bond_arp_send_all(bond, slave);
3085                         slave->jiffies = jiffies;
3086                         bond->current_arp_slave = slave;
3087                         break;
3088                 }
3089
3090                 /* if the link state is up at this point, we
3091                  * mark it down - this can happen if we have
3092                  * simultaneous link failures and
3093                  * reselect_active_interface doesn't make this
3094                  * one the current slave so it is still marked
3095                  * up when it is actually down
3096                  */
3097                 if (slave->link == BOND_LINK_UP) {
3098                         slave->link = BOND_LINK_DOWN;
3099                         if (slave->link_failure_count < UINT_MAX)
3100                                 slave->link_failure_count++;
3101
3102                         bond_set_slave_inactive_flags(slave);
3103
3104                         pr_info("%s: backup interface %s is now down.\n",
3105                                 bond->dev->name, slave->dev->name);
3106                 }
3107         }
3108 }
3109
3110 void bond_activebackup_arp_mon(struct work_struct *work)
3111 {
3112         struct bonding *bond = container_of(work, struct bonding,
3113                                             arp_work.work);
3114         int delta_in_ticks;
3115
3116         read_lock(&bond->lock);
3117
3118         if (bond->kill_timers)
3119                 goto out;
3120
3121         delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
3122
3123         if (bond->slave_cnt == 0)
3124                 goto re_arm;
3125
3126         if (bond->send_grat_arp) {
3127                 read_lock(&bond->curr_slave_lock);
3128                 bond_send_gratuitous_arp(bond);
3129                 read_unlock(&bond->curr_slave_lock);
3130         }
3131
3132         if (bond->send_unsol_na) {
3133                 read_lock(&bond->curr_slave_lock);
3134                 bond_send_unsolicited_na(bond);
3135                 read_unlock(&bond->curr_slave_lock);
3136         }
3137
3138         if (bond_ab_arp_inspect(bond, delta_in_ticks)) {
3139                 read_unlock(&bond->lock);
3140                 rtnl_lock();
3141                 read_lock(&bond->lock);
3142
3143                 bond_ab_arp_commit(bond, delta_in_ticks);
3144
3145                 read_unlock(&bond->lock);
3146                 rtnl_unlock();
3147                 read_lock(&bond->lock);
3148         }
3149
3150         bond_ab_arp_probe(bond);
3151
3152 re_arm:
3153         if (bond->params.arp_interval)
3154                 queue_delayed_work(bond->wq, &bond->arp_work, delta_in_ticks);
3155 out:
3156         read_unlock(&bond->lock);
3157 }
3158
3159 /*------------------------------ proc/seq_file-------------------------------*/
3160
3161 #ifdef CONFIG_PROC_FS
3162
3163 static void *bond_info_seq_start(struct seq_file *seq, loff_t *pos)
3164         __acquires(&dev_base_lock)
3165         __acquires(&bond->lock)
3166 {
3167         struct bonding *bond = seq->private;
3168         loff_t off = 0;
3169         struct slave *slave;
3170         int i;
3171
3172         /* make sure the bond won't be taken away */
3173         read_lock(&dev_base_lock);
3174         read_lock(&bond->lock);
3175
3176         if (*pos == 0)
3177                 return SEQ_START_TOKEN;
3178
3179         bond_for_each_slave(bond, slave, i) {
3180                 if (++off == *pos)
3181                         return slave;
3182         }
3183
3184         return NULL;
3185 }
3186
3187 static void *bond_info_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3188 {
3189         struct bonding *bond = seq->private;
3190         struct slave *slave = v;
3191
3192         ++*pos;
3193         if (v == SEQ_START_TOKEN)
3194                 return bond->first_slave;
3195
3196         slave = slave->next;
3197
3198         return (slave == bond->first_slave) ? NULL : slave;
3199 }
3200
3201 static void bond_info_seq_stop(struct seq_file *seq, void *v)
3202         __releases(&bond->lock)
3203         __releases(&dev_base_lock)
3204 {
3205         struct bonding *bond = seq->private;
3206
3207         read_unlock(&bond->lock);
3208         read_unlock(&dev_base_lock);
3209 }
3210
3211 static void bond_info_show_master(struct seq_file *seq)
3212 {
3213         struct bonding *bond = seq->private;
3214         struct slave *curr;
3215         int i;
3216
3217         read_lock(&bond->curr_slave_lock);
3218         curr = bond->curr_active_slave;
3219         read_unlock(&bond->curr_slave_lock);
3220
3221         seq_printf(seq, "Bonding Mode: %s",
3222                    bond_mode_name(bond->params.mode));
3223
3224         if (bond->params.mode == BOND_MODE_ACTIVEBACKUP &&
3225             bond->params.fail_over_mac)
3226                 seq_printf(seq, " (fail_over_mac %s)",
3227                    fail_over_mac_tbl[bond->params.fail_over_mac].modename);
3228
3229         seq_printf(seq, "\n");
3230
3231         if (bond->params.mode == BOND_MODE_XOR ||
3232                 bond->params.mode == BOND_MODE_8023AD) {
3233                 seq_printf(seq, "Transmit Hash Policy: %s (%d)\n",
3234                         xmit_hashtype_tbl[bond->params.xmit_policy].modename,
3235                         bond->params.xmit_policy);
3236         }
3237
3238         if (USES_PRIMARY(bond->params.mode)) {
3239                 seq_printf(seq, "Primary Slave: %s",
3240                            (bond->primary_slave) ?
3241                            bond->primary_slave->dev->name : "None");
3242                 if (bond->primary_slave)
3243                         seq_printf(seq, " (primary_reselect %s)",
3244                    pri_reselect_tbl[bond->params.primary_reselect].modename);
3245
3246                 seq_printf(seq, "\nCurrently Active Slave: %s\n",
3247                            (curr) ? curr->dev->name : "None");
3248         }
3249
3250         seq_printf(seq, "MII Status: %s\n", netif_carrier_ok(bond->dev) ?
3251                    "up" : "down");
3252         seq_printf(seq, "MII Polling Interval (ms): %d\n", bond->params.miimon);
3253         seq_printf(seq, "Up Delay (ms): %d\n",
3254                    bond->params.updelay * bond->params.miimon);
3255         seq_printf(seq, "Down Delay (ms): %d\n",
3256                    bond->params.downdelay * bond->params.miimon);
3257
3258
3259         /* ARP information */
3260         if (bond->params.arp_interval > 0) {
3261                 int printed = 0;
3262                 seq_printf(seq, "ARP Polling Interval (ms): %d\n",
3263                                 bond->params.arp_interval);
3264
3265                 seq_printf(seq, "ARP IP target/s (n.n.n.n form):");
3266
3267                 for (i = 0; (i < BOND_MAX_ARP_TARGETS); i++) {
3268                         if (!bond->params.arp_targets[i])
3269                                 break;
3270                         if (printed)
3271                                 seq_printf(seq, ",");
3272                         seq_printf(seq, " %pI4", &bond->params.arp_targets[i]);
3273                         printed = 1;
3274                 }
3275                 seq_printf(seq, "\n");
3276         }
3277
3278         if (bond->params.mode == BOND_MODE_8023AD) {
3279                 struct ad_info ad_info;
3280
3281                 seq_puts(seq, "\n802.3ad info\n");
3282                 seq_printf(seq, "LACP rate: %s\n",
3283                            (bond->params.lacp_fast) ? "fast" : "slow");
3284                 seq_printf(seq, "Aggregator selection policy (ad_select): %s\n",
3285                            ad_select_tbl[bond->params.ad_select].modename);
3286
3287                 if (bond_3ad_get_active_agg_info(bond, &ad_info)) {
3288                         seq_printf(seq, "bond %s has no active aggregator\n",
3289                                    bond->dev->name);
3290                 } else {
3291                         seq_printf(seq, "Active Aggregator Info:\n");
3292
3293                         seq_printf(seq, "\tAggregator ID: %d\n",
3294                                    ad_info.aggregator_id);
3295                         seq_printf(seq, "\tNumber of ports: %d\n",
3296                                    ad_info.ports);
3297                         seq_printf(seq, "\tActor Key: %d\n",
3298                                    ad_info.actor_key);
3299                         seq_printf(seq, "\tPartner Key: %d\n",
3300                                    ad_info.partner_key);
3301                         seq_printf(seq, "\tPartner Mac Address: %pM\n",
3302                                    ad_info.partner_system);
3303                 }
3304         }
3305 }
3306
3307 static void bond_info_show_slave(struct seq_file *seq,
3308                                  const struct slave *slave)
3309 {
3310         struct bonding *bond = seq->private;
3311
3312         seq_printf(seq, "\nSlave Interface: %s\n", slave->dev->name);
3313         seq_printf(seq, "MII Status: %s\n",
3314                    (slave->link == BOND_LINK_UP) ?  "up" : "down");
3315         seq_printf(seq, "Link Failure Count: %u\n",
3316                    slave->link_failure_count);
3317
3318         seq_printf(seq, "Permanent HW addr: %pM\n", slave->perm_hwaddr);
3319
3320         if (bond->params.mode == BOND_MODE_8023AD) {
3321                 const struct aggregator *agg
3322                         = SLAVE_AD_INFO(slave).port.aggregator;
3323
3324                 if (agg)
3325                         seq_printf(seq, "Aggregator ID: %d\n",
3326                                    agg->aggregator_identifier);
3327                 else
3328                         seq_puts(seq, "Aggregator ID: N/A\n");
3329         }
3330         seq_printf(seq, "Slave queue ID: %d\n", slave->queue_id);
3331 }
3332
3333 static int bond_info_seq_show(struct seq_file *seq, void *v)
3334 {
3335         if (v == SEQ_START_TOKEN) {
3336                 seq_printf(seq, "%s\n", version);
3337                 bond_info_show_master(seq);
3338         } else
3339                 bond_info_show_slave(seq, v);
3340
3341         return 0;
3342 }
3343
3344 static const struct seq_operations bond_info_seq_ops = {
3345         .start = bond_info_seq_start,
3346         .next  = bond_info_seq_next,
3347         .stop  = bond_info_seq_stop,
3348         .show  = bond_info_seq_show,
3349 };
3350
3351 static int bond_info_open(struct inode *inode, struct file *file)
3352 {
3353         struct seq_file *seq;
3354         struct proc_dir_entry *proc;
3355         int res;
3356
3357         res = seq_open(file, &bond_info_seq_ops);
3358         if (!res) {
3359                 /* recover the pointer buried in proc_dir_entry data */
3360                 seq = file->private_data;
3361                 proc = PDE(inode);
3362                 seq->private = proc->data;
3363         }
3364
3365         return res;
3366 }
3367
3368 static const struct file_operations bond_info_fops = {
3369         .owner   = THIS_MODULE,
3370         .open    = bond_info_open,
3371         .read    = seq_read,
3372         .llseek  = seq_lseek,
3373         .release = seq_release,
3374 };
3375
3376 static void bond_create_proc_entry(struct bonding *bond)
3377 {
3378         struct net_device *bond_dev = bond->dev;
3379         struct bond_net *bn = net_generic(dev_net(bond_dev), bond_net_id);
3380
3381         if (bn->proc_dir) {
3382                 bond->proc_entry = proc_create_data(bond_dev->name,
3383                                                     S_IRUGO, bn->proc_dir,
3384                                                     &bond_info_fops, bond);
3385                 if (bond->proc_entry == NULL)
3386                         pr_warning("Warning: Cannot create /proc/net/%s/%s\n",
3387                                    DRV_NAME, bond_dev->name);
3388                 else
3389                         memcpy(bond->proc_file_name, bond_dev->name, IFNAMSIZ);
3390         }
3391 }
3392
3393 static void bond_remove_proc_entry(struct bonding *bond)
3394 {
3395         struct net_device *bond_dev = bond->dev;
3396         struct bond_net *bn = net_generic(dev_net(bond_dev), bond_net_id);
3397
3398         if (bn->proc_dir && bond->proc_entry) {
3399                 remove_proc_entry(bond->proc_file_name, bn->proc_dir);
3400                 memset(bond->proc_file_name, 0, IFNAMSIZ);
3401                 bond->proc_entry = NULL;
3402         }
3403 }
3404
3405 /* Create the bonding directory under /proc/net, if doesn't exist yet.
3406  * Caller must hold rtnl_lock.
3407  */
3408 static void __net_init bond_create_proc_dir(struct bond_net *bn)
3409 {
3410         if (!bn->proc_dir) {
3411                 bn->proc_dir = proc_mkdir(DRV_NAME, bn->net->proc_net);
3412                 if (!bn->proc_dir)
3413                         pr_warning("Warning: cannot create /proc/net/%s\n",
3414                                    DRV_NAME);
3415         }
3416 }
3417
3418 /* Destroy the bonding directory under /proc/net, if empty.
3419  * Caller must hold rtnl_lock.
3420  */
3421 static void __net_exit bond_destroy_proc_dir(struct bond_net *bn)
3422 {
3423         if (bn->proc_dir) {
3424                 remove_proc_entry(DRV_NAME, bn->net->proc_net);
3425                 bn->proc_dir = NULL;
3426         }
3427 }
3428
3429 #else /* !CONFIG_PROC_FS */
3430
3431 static void bond_create_proc_entry(struct bonding *bond)
3432 {
3433 }
3434
3435 static void bond_remove_proc_entry(struct bonding *bond)
3436 {
3437 }
3438
3439 static inline void bond_create_proc_dir(struct bond_net *bn)
3440 {
3441 }
3442
3443 static inline void bond_destroy_proc_dir(struct bond_net *bn)
3444 {
3445 }
3446
3447 #endif /* CONFIG_PROC_FS */
3448
3449
3450 /*-------------------------- netdev event handling --------------------------*/
3451
3452 /*
3453  * Change device name
3454  */
3455 static int bond_event_changename(struct bonding *bond)
3456 {
3457         bond_remove_proc_entry(bond);
3458         bond_create_proc_entry(bond);
3459
3460         return NOTIFY_DONE;
3461 }
3462
3463 static int bond_master_netdev_event(unsigned long event,
3464                                     struct net_device *bond_dev)
3465 {
3466         struct bonding *event_bond = netdev_priv(bond_dev);
3467
3468         switch (event) {
3469         case NETDEV_CHANGENAME:
3470                 return bond_event_changename(event_bond);
3471         default:
3472                 break;
3473         }
3474
3475         return NOTIFY_DONE;
3476 }
3477
3478 static int bond_slave_netdev_event(unsigned long event,
3479                                    struct net_device *slave_dev)
3480 {
3481         struct net_device *bond_dev = slave_dev->master;
3482         struct bonding *bond = netdev_priv(bond_dev);
3483
3484         switch (event) {
3485         case NETDEV_UNREGISTER:
3486                 if (bond_dev) {
3487                         if (bond->setup_by_slave)
3488                                 bond_release_and_destroy(bond_dev, slave_dev);
3489                         else
3490                                 bond_release(bond_dev, slave_dev);
3491                 }
3492                 break;
3493         case NETDEV_CHANGE:
3494                 if (bond->params.mode == BOND_MODE_8023AD || bond_is_lb(bond)) {
3495                         struct slave *slave;
3496
3497                         slave = bond_get_slave_by_dev(bond, slave_dev);
3498                         if (slave) {
3499                                 u16 old_speed = slave->speed;
3500                                 u16 old_duplex = slave->duplex;
3501
3502                                 bond_update_speed_duplex(slave);
3503
3504                                 if (bond_is_lb(bond))
3505                                         break;
3506
3507                                 if (old_speed != slave->speed)
3508                                         bond_3ad_adapter_speed_changed(slave);
3509                                 if (old_duplex != slave->duplex)
3510                                         bond_3ad_adapter_duplex_changed(slave);
3511                         }
3512                 }
3513
3514                 break;
3515         case NETDEV_DOWN:
3516                 /*
3517                  * ... Or is it this?
3518                  */
3519                 break;
3520         case NETDEV_CHANGEMTU:
3521                 /*
3522                  * TODO: Should slaves be allowed to
3523                  * independently alter their MTU?  For
3524                  * an active-backup bond, slaves need
3525                  * not be the same type of device, so
3526                  * MTUs may vary.  For other modes,
3527                  * slaves arguably should have the
3528                  * same MTUs. To do this, we'd need to
3529                  * take over the slave's change_mtu
3530                  * function for the duration of their
3531                  * servitude.
3532                  */
3533                 break;
3534         case NETDEV_CHANGENAME:
3535                 /*
3536                  * TODO: handle changing the primary's name
3537                  */
3538                 break;
3539         case NETDEV_FEAT_CHANGE:
3540                 bond_compute_features(bond);
3541                 break;
3542         default:
3543                 break;
3544         }
3545
3546         return NOTIFY_DONE;
3547 }
3548
3549 /*
3550  * bond_netdev_event: handle netdev notifier chain events.
3551  *
3552  * This function receives events for the netdev chain.  The caller (an
3553  * ioctl handler calling blocking_notifier_call_chain) holds the necessary
3554  * locks for us to safely manipulate the slave devices (RTNL lock,
3555  * dev_probe_lock).
3556  */
3557 static int bond_netdev_event(struct notifier_block *this,
3558                              unsigned long event, void *ptr)
3559 {
3560         struct net_device *event_dev = (struct net_device *)ptr;
3561
3562         pr_debug("event_dev: %s, event: %lx\n",
3563                  event_dev ? event_dev->name : "None",
3564                  event);
3565
3566         if (!(event_dev->priv_flags & IFF_BONDING))
3567                 return NOTIFY_DONE;
3568
3569         if (event_dev->flags & IFF_MASTER) {
3570                 pr_debug("IFF_MASTER\n");
3571                 return bond_master_netdev_event(event, event_dev);
3572         }
3573
3574         if (event_dev->flags & IFF_SLAVE) {
3575                 pr_debug("IFF_SLAVE\n");
3576                 return bond_slave_netdev_event(event, event_dev);
3577         }
3578
3579         return NOTIFY_DONE;
3580 }
3581
3582 /*
3583  * bond_inetaddr_event: handle inetaddr notifier chain events.
3584  *
3585  * We keep track of device IPs primarily to use as source addresses in
3586  * ARP monitor probes (rather than spewing out broadcasts all the time).
3587  *
3588  * We track one IP for the main device (if it has one), plus one per VLAN.
3589  */
3590 static int bond_inetaddr_event(struct notifier_block *this, unsigned long event, void *ptr)
3591 {
3592         struct in_ifaddr *ifa = ptr;
3593         struct net_device *vlan_dev, *event_dev = ifa->ifa_dev->dev;
3594         struct bond_net *bn = net_generic(dev_net(event_dev), bond_net_id);
3595         struct bonding *bond;
3596         struct vlan_entry *vlan;
3597
3598         list_for_each_entry(bond, &bn->dev_list, bond_list) {
3599                 if (bond->dev == event_dev) {
3600                         switch (event) {
3601                         case NETDEV_UP:
3602                                 bond->master_ip = ifa->ifa_local;
3603                                 return NOTIFY_OK;
3604                         case NETDEV_DOWN:
3605                                 bond->master_ip = bond_glean_dev_ip(bond->dev);
3606                                 return NOTIFY_OK;
3607                         default:
3608                                 return NOTIFY_DONE;
3609                         }
3610                 }
3611
3612                 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
3613                         if (!bond->vlgrp)
3614                                 continue;
3615                         vlan_dev = vlan_group_get_device(bond->vlgrp, vlan->vlan_id);
3616                         if (vlan_dev == event_dev) {
3617                                 switch (event) {
3618                                 case NETDEV_UP:
3619                                         vlan->vlan_ip = ifa->ifa_local;
3620                                         return NOTIFY_OK;
3621                                 case NETDEV_DOWN:
3622                                         vlan->vlan_ip =
3623                                                 bond_glean_dev_ip(vlan_dev);
3624                                         return NOTIFY_OK;
3625                                 default:
3626                                         return NOTIFY_DONE;
3627                                 }
3628                         }
3629                 }
3630         }
3631         return NOTIFY_DONE;
3632 }
3633
3634 static struct notifier_block bond_netdev_notifier = {
3635         .notifier_call = bond_netdev_event,
3636 };
3637
3638 static struct notifier_block bond_inetaddr_notifier = {
3639         .notifier_call = bond_inetaddr_event,
3640 };
3641
3642 /*-------------------------- Packet type handling ---------------------------*/
3643
3644 /* register to receive lacpdus on a bond */
3645 static void bond_register_lacpdu(struct bonding *bond)
3646 {
3647         struct packet_type *pk_type = &(BOND_AD_INFO(bond).ad_pkt_type);
3648
3649         /* initialize packet type */
3650         pk_type->type = PKT_TYPE_LACPDU;
3651         pk_type->dev = bond->dev;
3652         pk_type->func = bond_3ad_lacpdu_recv;
3653
3654         dev_add_pack(pk_type);
3655 }
3656
3657 /* unregister to receive lacpdus on a bond */
3658 static void bond_unregister_lacpdu(struct bonding *bond)
3659 {
3660         dev_remove_pack(&(BOND_AD_INFO(bond).ad_pkt_type));
3661 }
3662
3663 void bond_register_arp(struct bonding *bond)
3664 {
3665         struct packet_type *pt = &bond->arp_mon_pt;
3666
3667         if (pt->type)
3668                 return;
3669
3670         pt->type = htons(ETH_P_ARP);
3671         pt->dev = bond->dev;
3672         pt->func = bond_arp_rcv;
3673         dev_add_pack(pt);
3674 }
3675
3676 void bond_unregister_arp(struct bonding *bond)
3677 {
3678         struct packet_type *pt = &bond->arp_mon_pt;
3679
3680         dev_remove_pack(pt);
3681         pt->type = 0;
3682 }
3683
3684 /*---------------------------- Hashing Policies -----------------------------*/
3685
3686 /*
3687  * Hash for the output device based upon layer 2 and layer 3 data. If
3688  * the packet is not IP mimic bond_xmit_hash_policy_l2()
3689  */
3690 static int bond_xmit_hash_policy_l23(struct sk_buff *skb, int count)
3691 {
3692         struct ethhdr *data = (struct ethhdr *)skb->data;
3693         struct iphdr *iph = ip_hdr(skb);
3694
3695         if (skb->protocol == htons(ETH_P_IP)) {
3696                 return ((ntohl(iph->saddr ^ iph->daddr) & 0xffff) ^
3697                         (data->h_dest[5] ^ data->h_source[5])) % count;
3698         }
3699
3700         return (data->h_dest[5] ^ data->h_source[5]) % count;
3701 }
3702
3703 /*
3704  * Hash for the output device based upon layer 3 and layer 4 data. If
3705  * the packet is a frag or not TCP or UDP, just use layer 3 data.  If it is
3706  * altogether not IP, mimic bond_xmit_hash_policy_l2()
3707  */
3708 static int bond_xmit_hash_policy_l34(struct sk_buff *skb, int count)
3709 {
3710         struct ethhdr *data = (struct ethhdr *)skb->data;
3711         struct iphdr *iph = ip_hdr(skb);
3712         __be16 *layer4hdr = (__be16 *)((u32 *)iph + iph->ihl);
3713         int layer4_xor = 0;
3714
3715         if (skb->protocol == htons(ETH_P_IP)) {
3716                 if (!(iph->frag_off & htons(IP_MF|IP_OFFSET)) &&
3717                     (iph->protocol == IPPROTO_TCP ||
3718                      iph->protocol == IPPROTO_UDP)) {
3719                         layer4_xor = ntohs((*layer4hdr ^ *(layer4hdr + 1)));
3720                 }
3721                 return (layer4_xor ^
3722                         ((ntohl(iph->saddr ^ iph->daddr)) & 0xffff)) % count;
3723
3724         }
3725
3726         return (data->h_dest[5] ^ data->h_source[5]) % count;
3727 }
3728
3729 /*
3730  * Hash for the output device based upon layer 2 data
3731  */
3732 static int bond_xmit_hash_policy_l2(struct sk_buff *skb, int count)
3733 {
3734         struct ethhdr *data = (struct ethhdr *)skb->data;
3735
3736         return (data->h_dest[5] ^ data->h_source[5]) % count;
3737 }
3738
3739 /*-------------------------- Device entry points ----------------------------*/
3740
3741 static int bond_open(struct net_device *bond_dev)
3742 {
3743         struct bonding *bond = netdev_priv(bond_dev);
3744
3745         bond->kill_timers = 0;
3746
3747         if (bond_is_lb(bond)) {
3748                 /* bond_alb_initialize must be called before the timer
3749                  * is started.
3750                  */
3751                 if (bond_alb_initialize(bond, (bond->params.mode == BOND_MODE_ALB))) {
3752                         /* something went wrong - fail the open operation */
3753                         return -ENOMEM;
3754                 }
3755
3756                 INIT_DELAYED_WORK(&bond->alb_work, bond_alb_monitor);
3757                 queue_delayed_work(bond->wq, &bond->alb_work, 0);
3758         }
3759
3760         if (bond->params.miimon) {  /* link check interval, in milliseconds. */
3761                 INIT_DELAYED_WORK(&bond->mii_work, bond_mii_monitor);
3762                 queue_delayed_work(bond->wq, &bond->mii_work, 0);
3763         }
3764
3765         if (bond->params.arp_interval) {  /* arp interval, in milliseconds. */
3766                 if (bond->params.mode == BOND_MODE_ACTIVEBACKUP)
3767                         INIT_DELAYED_WORK(&bond->arp_work,
3768                                           bond_activebackup_arp_mon);
3769                 else
3770                         INIT_DELAYED_WORK(&bond->arp_work,
3771                                           bond_loadbalance_arp_mon);
3772
3773                 queue_delayed_work(bond->wq, &bond->arp_work, 0);
3774                 if (bond->params.arp_validate)
3775                         bond_register_arp(bond);
3776         }
3777
3778         if (bond->params.mode == BOND_MODE_8023AD) {
3779                 INIT_DELAYED_WORK(&bond->ad_work, bond_3ad_state_machine_handler);
3780                 queue_delayed_work(bond->wq, &bond->ad_work, 0);
3781                 /* register to receive LACPDUs */
3782                 bond_register_lacpdu(bond);
3783                 bond_3ad_initiate_agg_selection(bond, 1);
3784         }
3785
3786         return 0;
3787 }
3788
3789 static int bond_close(struct net_device *bond_dev)
3790 {
3791         struct bonding *bond = netdev_priv(bond_dev);
3792
3793         if (bond->params.mode == BOND_MODE_8023AD) {
3794                 /* Unregister the receive of LACPDUs */
3795                 bond_unregister_lacpdu(bond);
3796         }
3797
3798         if (bond->params.arp_validate)
3799                 bond_unregister_arp(bond);
3800
3801         write_lock_bh(&bond->lock);
3802
3803         bond->send_grat_arp = 0;
3804         bond->send_unsol_na = 0;
3805
3806         /* signal timers not to re-arm */
3807         bond->kill_timers = 1;
3808
3809         write_unlock_bh(&bond->lock);
3810
3811         if (bond->params.miimon) {  /* link check interval, in milliseconds. */
3812                 cancel_delayed_work(&bond->mii_work);
3813         }
3814
3815         if (bond->params.arp_interval) {  /* arp interval, in milliseconds. */
3816                 cancel_delayed_work(&bond->arp_work);
3817         }
3818
3819         switch (bond->params.mode) {
3820         case BOND_MODE_8023AD:
3821                 cancel_delayed_work(&bond->ad_work);
3822                 break;
3823         case BOND_MODE_TLB:
3824         case BOND_MODE_ALB:
3825                 cancel_delayed_work(&bond->alb_work);
3826                 break;
3827         default:
3828                 break;
3829         }
3830
3831
3832         if (bond_is_lb(bond)) {
3833                 /* Must be called only after all
3834                  * slaves have been released
3835                  */
3836                 bond_alb_deinitialize(bond);
3837         }
3838
3839         return 0;
3840 }
3841
3842 static struct rtnl_link_stats64 *bond_get_stats(struct net_device *bond_dev,
3843                                                 struct rtnl_link_stats64 *stats)
3844 {
3845         struct bonding *bond = netdev_priv(bond_dev);
3846         struct rtnl_link_stats64 temp;
3847         struct slave *slave;
3848         int i;
3849
3850         memset(stats, 0, sizeof(*stats));
3851
3852         read_lock_bh(&bond->lock);
3853
3854         bond_for_each_slave(bond, slave, i) {
3855                 const struct rtnl_link_stats64 *sstats =
3856                         dev_get_stats(slave->dev, &temp);
3857
3858                 stats->rx_packets += sstats->rx_packets;
3859                 stats->rx_bytes += sstats->rx_bytes;
3860                 stats->rx_errors += sstats->rx_errors;
3861                 stats->rx_dropped += sstats->rx_dropped;
3862
3863                 stats->tx_packets += sstats->tx_packets;
3864                 stats->tx_bytes += sstats->tx_bytes;
3865                 stats->tx_errors += sstats->tx_errors;
3866                 stats->tx_dropped += sstats->tx_dropped;
3867
3868                 stats->multicast += sstats->multicast;
3869                 stats->collisions += sstats->collisions;
3870
3871                 stats->rx_length_errors += sstats->rx_length_errors;
3872                 stats->rx_over_errors += sstats->rx_over_errors;
3873                 stats->rx_crc_errors += sstats->rx_crc_errors;
3874                 stats->rx_frame_errors += sstats->rx_frame_errors;
3875                 stats->rx_fifo_errors += sstats->rx_fifo_errors;
3876                 stats->rx_missed_errors += sstats->rx_missed_errors;
3877
3878                 stats->tx_aborted_errors += sstats->tx_aborted_errors;
3879                 stats->tx_carrier_errors += sstats->tx_carrier_errors;
3880                 stats->tx_fifo_errors += sstats->tx_fifo_errors;
3881                 stats->tx_heartbeat_errors += sstats->tx_heartbeat_errors;
3882                 stats->tx_window_errors += sstats->tx_window_errors;
3883         }
3884
3885         read_unlock_bh(&bond->lock);
3886
3887         return stats;
3888 }
3889
3890 static int bond_do_ioctl(struct net_device *bond_dev, struct ifreq *ifr, int cmd)
3891 {
3892         struct net_device *slave_dev = NULL;
3893         struct ifbond k_binfo;
3894         struct ifbond __user *u_binfo = NULL;
3895         struct ifslave k_sinfo;
3896         struct ifslave __user *u_sinfo = NULL;
3897         struct mii_ioctl_data *mii = NULL;
3898         int res = 0;
3899
3900         pr_debug("bond_ioctl: master=%s, cmd=%d\n", bond_dev->name, cmd);
3901
3902         switch (cmd) {
3903         case SIOCGMIIPHY:
3904                 mii = if_mii(ifr);
3905                 if (!mii)
3906                         return -EINVAL;
3907
3908                 mii->phy_id = 0;
3909                 /* Fall Through */
3910         case SIOCGMIIREG:
3911                 /*
3912                  * We do this again just in case we were called by SIOCGMIIREG
3913                  * instead of SIOCGMIIPHY.
3914                  */
3915                 mii = if_mii(ifr);
3916                 if (!mii)
3917                         return -EINVAL;
3918
3919
3920                 if (mii->reg_num == 1) {
3921                         struct bonding *bond = netdev_priv(bond_dev);
3922                         mii->val_out = 0;
3923                         read_lock(&bond->lock);
3924                         read_lock(&bond->curr_slave_lock);
3925                         if (netif_carrier_ok(bond->dev))
3926                                 mii->val_out = BMSR_LSTATUS;
3927
3928                         read_unlock(&bond->curr_slave_lock);
3929                         read_unlock(&bond->lock);
3930                 }
3931
3932                 return 0;
3933         case BOND_INFO_QUERY_OLD:
3934         case SIOCBONDINFOQUERY:
3935                 u_binfo = (struct ifbond __user *)ifr->ifr_data;
3936
3937                 if (copy_from_user(&k_binfo, u_binfo, sizeof(ifbond)))
3938                         return -EFAULT;
3939
3940                 res = bond_info_query(bond_dev, &k_binfo);
3941                 if (res == 0 &&
3942                     copy_to_user(u_binfo, &k_binfo, sizeof(ifbond)))
3943                         return -EFAULT;
3944
3945                 return res;
3946         case BOND_SLAVE_INFO_QUERY_OLD:
3947         case SIOCBONDSLAVEINFOQUERY:
3948                 u_sinfo = (struct ifslave __user *)ifr->ifr_data;
3949
3950                 if (copy_from_user(&k_sinfo, u_sinfo, sizeof(ifslave)))
3951                         return -EFAULT;
3952
3953                 res = bond_slave_info_query(bond_dev, &k_sinfo);
3954                 if (res == 0 &&
3955                     copy_to_user(u_sinfo, &k_sinfo, sizeof(ifslave)))
3956                         return -EFAULT;
3957
3958                 return res;
3959         default:
3960                 /* Go on */
3961                 break;
3962         }
3963
3964         if (!capable(CAP_NET_ADMIN))
3965                 return -EPERM;
3966
3967         slave_dev = dev_get_by_name(dev_net(bond_dev), ifr->ifr_slave);
3968
3969         pr_debug("slave_dev=%p:\n", slave_dev);
3970
3971         if (!slave_dev)
3972                 res = -ENODEV;
3973         else {
3974                 pr_debug("slave_dev->name=%s:\n", slave_dev->name);
3975                 switch (cmd) {
3976                 case BOND_ENSLAVE_OLD:
3977                 case SIOCBONDENSLAVE:
3978                         res = bond_enslave(bond_dev, slave_dev);
3979                         break;
3980                 case BOND_RELEASE_OLD:
3981                 case SIOCBONDRELEASE:
3982                         res = bond_release(bond_dev, slave_dev);
3983                         break;
3984                 case BOND_SETHWADDR_OLD:
3985                 case SIOCBONDSETHWADDR:
3986                         res = bond_sethwaddr(bond_dev, slave_dev);
3987                         break;
3988                 case BOND_CHANGE_ACTIVE_OLD:
3989                 case SIOCBONDCHANGEACTIVE:
3990                         res = bond_ioctl_change_active(bond_dev, slave_dev);
3991                         break;
3992                 default:
3993                         res = -EOPNOTSUPP;
3994                 }
3995
3996                 dev_put(slave_dev);
3997         }
3998
3999         return res;
4000 }
4001
4002 static bool bond_addr_in_mc_list(unsigned char *addr,
4003                                  struct netdev_hw_addr_list *list,
4004                                  int addrlen)
4005 {
4006         struct netdev_hw_addr *ha;
4007
4008         netdev_hw_addr_list_for_each(ha, list)
4009                 if (!memcmp(ha->addr, addr, addrlen))
4010                         return true;
4011
4012         return false;
4013 }
4014
4015 static void bond_set_multicast_list(struct net_device *bond_dev)
4016 {
4017         struct bonding *bond = netdev_priv(bond_dev);
4018         struct netdev_hw_addr *ha;
4019         bool found;
4020
4021         /*
4022          * Do promisc before checking multicast_mode
4023          */
4024         if ((bond_dev->flags & IFF_PROMISC) && !(bond->flags & IFF_PROMISC))
4025                 /*
4026                  * FIXME: Need to handle the error when one of the multi-slaves
4027                  * encounters error.
4028                  */
4029                 bond_set_promiscuity(bond, 1);
4030
4031
4032         if (!(bond_dev->flags & IFF_PROMISC) && (bond->flags & IFF_PROMISC))
4033                 bond_set_promiscuity(bond, -1);
4034
4035
4036         /* set allmulti flag to slaves */
4037         if ((bond_dev->flags & IFF_ALLMULTI) && !(bond->flags & IFF_ALLMULTI))
4038                 /*
4039                  * FIXME: Need to handle the error when one of the multi-slaves
4040                  * encounters error.
4041                  */
4042                 bond_set_allmulti(bond, 1);
4043
4044
4045         if (!(bond_dev->flags & IFF_ALLMULTI) && (bond->flags & IFF_ALLMULTI))
4046                 bond_set_allmulti(bond, -1);
4047
4048
4049         read_lock(&bond->lock);
4050
4051         bond->flags = bond_dev->flags;
4052
4053         /* looking for addresses to add to slaves' mc list */
4054         netdev_for_each_mc_addr(ha, bond_dev) {
4055                 found = bond_addr_in_mc_list(ha->addr, &bond->mc_list,
4056                                              bond_dev->addr_len);
4057                 if (!found)
4058                         bond_mc_add(bond, ha->addr);
4059         }
4060
4061         /* looking for addresses to delete from slaves' list */
4062         netdev_hw_addr_list_for_each(ha, &bond->mc_list) {
4063                 found = bond_addr_in_mc_list(ha->addr, &bond_dev->mc,
4064                                              bond_dev->addr_len);
4065                 if (!found)
4066                         bond_mc_del(bond, ha->addr);
4067         }
4068
4069         /* save master's multicast list */
4070         __hw_addr_flush(&bond->mc_list);
4071         __hw_addr_add_multiple(&bond->mc_list, &bond_dev->mc,
4072                                bond_dev->addr_len, NETDEV_HW_ADDR_T_MULTICAST);
4073
4074         read_unlock(&bond->lock);
4075 }
4076
4077 static int bond_neigh_setup(struct net_device *dev, struct neigh_parms *parms)
4078 {
4079         struct bonding *bond = netdev_priv(dev);
4080         struct slave *slave = bond->first_slave;
4081
4082         if (slave) {
4083                 const struct net_device_ops *slave_ops
4084                         = slave->dev->netdev_ops;
4085                 if (slave_ops->ndo_neigh_setup)
4086                         return slave_ops->ndo_neigh_setup(slave->dev, parms);
4087         }
4088         return 0;
4089 }
4090
4091 /*
4092  * Change the MTU of all of a master's slaves to match the master
4093  */
4094 static int bond_change_mtu(struct net_device *bond_dev, int new_mtu)
4095 {
4096         struct bonding *bond = netdev_priv(bond_dev);
4097         struct slave *slave, *stop_at;
4098         int res = 0;
4099         int i;
4100
4101         pr_debug("bond=%p, name=%s, new_mtu=%d\n", bond,
4102                  (bond_dev ? bond_dev->name : "None"), new_mtu);
4103
4104         /* Can't hold bond->lock with bh disabled here since
4105          * some base drivers panic. On the other hand we can't
4106          * hold bond->lock without bh disabled because we'll
4107          * deadlock. The only solution is to rely on the fact
4108          * that we're under rtnl_lock here, and the slaves
4109          * list won't change. This doesn't solve the problem
4110          * of setting the slave's MTU while it is
4111          * transmitting, but the assumption is that the base
4112          * driver can handle that.
4113          *
4114          * TODO: figure out a way to safely iterate the slaves
4115          * list, but without holding a lock around the actual
4116          * call to the base driver.
4117          */
4118
4119         bond_for_each_slave(bond, slave, i) {
4120                 pr_debug("s %p s->p %p c_m %p\n",
4121                          slave,
4122                          slave->prev,
4123                          slave->dev->netdev_ops->ndo_change_mtu);
4124
4125                 res = dev_set_mtu(slave->dev, new_mtu);
4126
4127                 if (res) {
4128                         /* If we failed to set the slave's mtu to the new value
4129                          * we must abort the operation even in ACTIVE_BACKUP
4130                          * mode, because if we allow the backup slaves to have
4131                          * different mtu values than the active slave we'll
4132                          * need to change their mtu when doing a failover. That
4133                          * means changing their mtu from timer context, which
4134                          * is probably not a good idea.
4135                          */
4136                         pr_debug("err %d %s\n", res, slave->dev->name);
4137                         goto unwind;
4138                 }
4139         }
4140
4141         bond_dev->mtu = new_mtu;
4142
4143         return 0;
4144
4145 unwind:
4146         /* unwind from head to the slave that failed */
4147         stop_at = slave;
4148         bond_for_each_slave_from_to(bond, slave, i, bond->first_slave, stop_at) {
4149                 int tmp_res;
4150
4151                 tmp_res = dev_set_mtu(slave->dev, bond_dev->mtu);
4152                 if (tmp_res) {
4153                         pr_debug("unwind err %d dev %s\n",
4154                                  tmp_res, slave->dev->name);
4155                 }
4156         }
4157
4158         return res;
4159 }
4160
4161 /*
4162  * Change HW address
4163  *
4164  * Note that many devices must be down to change the HW address, and
4165  * downing the master releases all slaves.  We can make bonds full of
4166  * bonding devices to test this, however.
4167  */
4168 static int bond_set_mac_address(struct net_device *bond_dev, void *addr)
4169 {
4170         struct bonding *bond = netdev_priv(bond_dev);
4171         struct sockaddr *sa = addr, tmp_sa;
4172         struct slave *slave, *stop_at;
4173         int res = 0;
4174         int i;
4175
4176         if (bond->params.mode == BOND_MODE_ALB)
4177                 return bond_alb_set_mac_address(bond_dev, addr);
4178
4179
4180         pr_debug("bond=%p, name=%s\n",
4181                  bond, bond_dev ? bond_dev->name : "None");
4182
4183         /*
4184          * If fail_over_mac is set to active, do nothing and return
4185          * success.  Returning an error causes ifenslave to fail.
4186          */
4187         if (bond->params.fail_over_mac == BOND_FOM_ACTIVE)
4188                 return 0;
4189
4190         if (!is_valid_ether_addr(sa->sa_data))
4191                 return -EADDRNOTAVAIL;
4192
4193         /* Can't hold bond->lock with bh disabled here since
4194          * some base drivers panic. On the other hand we can't
4195          * hold bond->lock without bh disabled because we'll
4196          * deadlock. The only solution is to rely on the fact
4197          * that we're under rtnl_lock here, and the slaves
4198          * list won't change. This doesn't solve the problem
4199          * of setting the slave's hw address while it is
4200          * transmitting, but the assumption is that the base
4201          * driver can handle that.
4202          *
4203          * TODO: figure out a way to safely iterate the slaves
4204          * list, but without holding a lock around the actual
4205          * call to the base driver.
4206          */
4207
4208         bond_for_each_slave(bond, slave, i) {
4209                 const struct net_device_ops *slave_ops = slave->dev->netdev_ops;
4210                 pr_debug("slave %p %s\n", slave, slave->dev->name);
4211
4212                 if (slave_ops->ndo_set_mac_address == NULL) {
4213                         res = -EOPNOTSUPP;
4214                         pr_debug("EOPNOTSUPP %s\n", slave->dev->name);
4215                         goto unwind;
4216                 }
4217
4218                 res = dev_set_mac_address(slave->dev, addr);
4219                 if (res) {
4220                         /* TODO: consider downing the slave
4221                          * and retry ?
4222                          * User should expect communications
4223                          * breakage anyway until ARP finish
4224                          * updating, so...
4225                          */
4226                         pr_debug("err %d %s\n", res, slave->dev->name);
4227                         goto unwind;
4228                 }
4229         }
4230
4231         /* success */
4232         memcpy(bond_dev->dev_addr, sa->sa_data, bond_dev->addr_len);
4233         return 0;
4234
4235 unwind:
4236         memcpy(tmp_sa.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
4237         tmp_sa.sa_family = bond_dev->type;
4238
4239         /* unwind from head to the slave that failed */
4240         stop_at = slave;
4241         bond_for_each_slave_from_to(bond, slave, i, bond->first_slave, stop_at) {
4242                 int tmp_res;
4243
4244                 tmp_res = dev_set_mac_address(slave->dev, &tmp_sa);
4245                 if (tmp_res) {
4246                         pr_debug("unwind err %d dev %s\n",
4247                                  tmp_res, slave->dev->name);
4248                 }
4249         }
4250
4251         return res;
4252 }
4253
4254 static int bond_xmit_roundrobin(struct sk_buff *skb, struct net_device *bond_dev)
4255 {
4256         struct bonding *bond = netdev_priv(bond_dev);
4257         struct slave *slave, *start_at;
4258         int i, slave_no, res = 1;
4259         struct iphdr *iph = ip_hdr(skb);
4260
4261         read_lock(&bond->lock);
4262
4263         if (!BOND_IS_OK(bond))
4264                 goto out;
4265         /*
4266          * Start with the curr_active_slave that joined the bond as the
4267          * default for sending IGMP traffic.  For failover purposes one
4268          * needs to maintain some consistency for the interface that will
4269          * send the join/membership reports.  The curr_active_slave found
4270          * will send all of this type of traffic.
4271          */
4272         if ((iph->protocol == IPPROTO_IGMP) &&
4273             (skb->protocol == htons(ETH_P_IP))) {
4274
4275                 read_lock(&bond->curr_slave_lock);
4276                 slave = bond->curr_active_slave;
4277                 read_unlock(&bond->curr_slave_lock);
4278
4279                 if (!slave)
4280                         goto out;
4281         } else {
4282                 /*
4283                  * Concurrent TX may collide on rr_tx_counter; we accept
4284                  * that as being rare enough not to justify using an
4285                  * atomic op here.
4286                  */
4287                 slave_no = bond->rr_tx_counter++ % bond->slave_cnt;
4288
4289                 bond_for_each_slave(bond, slave, i) {
4290                         slave_no--;
4291                         if (slave_no < 0)
4292                                 break;
4293                 }
4294         }
4295
4296         start_at = slave;
4297         bond_for_each_slave_from(bond, slave, i, start_at) {
4298                 if (IS_UP(slave->dev) &&
4299                     (slave->link == BOND_LINK_UP) &&
4300                     (slave->state == BOND_STATE_ACTIVE)) {
4301                         res = bond_dev_queue_xmit(bond, skb, slave->dev);
4302                         break;
4303                 }
4304         }
4305
4306 out:
4307         if (res) {
4308                 /* no suitable interface, frame not sent */
4309                 dev_kfree_skb(skb);
4310         }
4311         read_unlock(&bond->lock);
4312         return NETDEV_TX_OK;
4313 }
4314
4315
4316 /*
4317  * in active-backup mode, we know that bond->curr_active_slave is always valid if
4318  * the bond has a usable interface.
4319  */
4320 static int bond_xmit_activebackup(struct sk_buff *skb, struct net_device *bond_dev)
4321 {
4322         struct bonding *bond = netdev_priv(bond_dev);
4323         int res = 1;
4324
4325         read_lock(&bond->lock);
4326         read_lock(&bond->curr_slave_lock);
4327
4328         if (!BOND_IS_OK(bond))
4329                 goto out;
4330
4331         if (!bond->curr_active_slave)
4332                 goto out;
4333
4334         res = bond_dev_queue_xmit(bond, skb, bond->curr_active_slave->dev);
4335
4336 out:
4337         if (res)
4338                 /* no suitable interface, frame not sent */
4339                 dev_kfree_skb(skb);
4340
4341         read_unlock(&bond->curr_slave_lock);
4342         read_unlock(&bond->lock);
4343         return NETDEV_TX_OK;
4344 }
4345
4346 /*
4347  * In bond_xmit_xor() , we determine the output device by using a pre-
4348  * determined xmit_hash_policy(), If the selected device is not enabled,
4349  * find the next active slave.
4350  */
4351 static int bond_xmit_xor(struct sk_buff *skb, struct net_device *bond_dev)
4352 {
4353         struct bonding *bond = netdev_priv(bond_dev);
4354         struct slave *slave, *start_at;
4355         int slave_no;
4356         int i;
4357         int res = 1;
4358
4359         read_lock(&bond->lock);
4360
4361         if (!BOND_IS_OK(bond))
4362                 goto out;
4363
4364         slave_no = bond->xmit_hash_policy(skb, bond->slave_cnt);
4365
4366         bond_for_each_slave(bond, slave, i) {
4367                 slave_no--;
4368                 if (slave_no < 0)
4369                         break;
4370         }
4371
4372         start_at = slave;
4373
4374         bond_for_each_slave_from(bond, slave, i, start_at) {
4375                 if (IS_UP(slave->dev) &&
4376                     (slave->link == BOND_LINK_UP) &&
4377                     (slave->state == BOND_STATE_ACTIVE)) {
4378                         res = bond_dev_queue_xmit(bond, skb, slave->dev);
4379                         break;
4380                 }
4381         }
4382
4383 out:
4384         if (res) {
4385                 /* no suitable interface, frame not sent */
4386                 dev_kfree_skb(skb);
4387         }
4388         read_unlock(&bond->lock);
4389         return NETDEV_TX_OK;
4390 }
4391
4392 /*
4393  * in broadcast mode, we send everything to all usable interfaces.
4394  */
4395 static int bond_xmit_broadcast(struct sk_buff *skb, struct net_device *bond_dev)
4396 {
4397         struct bonding *bond = netdev_priv(bond_dev);
4398         struct slave *slave, *start_at;
4399         struct net_device *tx_dev = NULL;
4400         int i;
4401         int res = 1;
4402
4403         read_lock(&bond->lock);
4404
4405         if (!BOND_IS_OK(bond))
4406                 goto out;
4407
4408         read_lock(&bond->curr_slave_lock);
4409         start_at = bond->curr_active_slave;
4410         read_unlock(&bond->curr_slave_lock);
4411
4412         if (!start_at)
4413                 goto out;
4414
4415         bond_for_each_slave_from(bond, slave, i, start_at) {
4416                 if (IS_UP(slave->dev) &&
4417                     (slave->link == BOND_LINK_UP) &&
4418                     (slave->state == BOND_STATE_ACTIVE)) {
4419                         if (tx_dev) {
4420                                 struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
4421                                 if (!skb2) {
4422                                         pr_err("%s: Error: bond_xmit_broadcast(): skb_clone() failed\n",
4423                                                bond_dev->name);
4424                                         continue;
4425                                 }
4426
4427                                 res = bond_dev_queue_xmit(bond, skb2, tx_dev);
4428                                 if (res) {
4429                                         dev_kfree_skb(skb2);
4430                                         continue;
4431                                 }
4432                         }
4433                         tx_dev = slave->dev;
4434                 }
4435         }
4436
4437         if (tx_dev)
4438                 res = bond_dev_queue_xmit(bond, skb, tx_dev);
4439
4440 out:
4441         if (res)
4442                 /* no suitable interface, frame not sent */
4443                 dev_kfree_skb(skb);
4444
4445         /* frame sent to all suitable interfaces */
4446         read_unlock(&bond->lock);
4447         return NETDEV_TX_OK;
4448 }
4449
4450 /*------------------------- Device initialization ---------------------------*/
4451
4452 static void bond_set_xmit_hash_policy(struct bonding *bond)
4453 {
4454         switch (bond->params.xmit_policy) {
4455         case BOND_XMIT_POLICY_LAYER23:
4456                 bond->xmit_hash_policy = bond_xmit_hash_policy_l23;
4457                 break;
4458         case BOND_XMIT_POLICY_LAYER34:
4459                 bond->xmit_hash_policy = bond_xmit_hash_policy_l34;
4460                 break;
4461         case BOND_XMIT_POLICY_LAYER2:
4462         default:
4463                 bond->xmit_hash_policy = bond_xmit_hash_policy_l2;
4464                 break;
4465         }
4466 }
4467
4468 /*
4469  * Lookup the slave that corresponds to a qid
4470  */
4471 static inline int bond_slave_override(struct bonding *bond,
4472                                       struct sk_buff *skb)
4473 {
4474         int i, res = 1;
4475         struct slave *slave = NULL;
4476         struct slave *check_slave;
4477
4478         read_lock(&bond->lock);
4479
4480         if (!BOND_IS_OK(bond) || !skb->queue_mapping)
4481                 goto out;
4482
4483         /* Find out if any slaves have the same mapping as this skb. */
4484         bond_for_each_slave(bond, check_slave, i) {
4485                 if (check_slave->queue_id == skb->queue_mapping) {
4486                         slave = check_slave;
4487                         break;
4488                 }
4489         }
4490
4491         /* If the slave isn't UP, use default transmit policy. */
4492         if (slave && slave->queue_id && IS_UP(slave->dev) &&
4493             (slave->link == BOND_LINK_UP)) {
4494                 res = bond_dev_queue_xmit(bond, skb, slave->dev);
4495         }
4496
4497 out:
4498         read_unlock(&bond->lock);
4499         return res;
4500 }
4501
4502 static u16 bond_select_queue(struct net_device *dev, struct sk_buff *skb)
4503 {
4504         /*
4505          * This helper function exists to help dev_pick_tx get the correct
4506          * destination queue.  Using a helper function skips the a call to
4507          * skb_tx_hash and will put the skbs in the queue we expect on their
4508          * way down to the bonding driver.
4509          */
4510         return skb->queue_mapping;
4511 }
4512
4513 static netdev_tx_t bond_start_xmit(struct sk_buff *skb, struct net_device *dev)
4514 {
4515         struct bonding *bond = netdev_priv(dev);
4516
4517         if (TX_QUEUE_OVERRIDE(bond->params.mode)) {
4518                 if (!bond_slave_override(bond, skb))
4519                         return NETDEV_TX_OK;
4520         }
4521
4522         switch (bond->params.mode) {
4523         case BOND_MODE_ROUNDROBIN:
4524                 return bond_xmit_roundrobin(skb, dev);
4525         case BOND_MODE_ACTIVEBACKUP:
4526                 return bond_xmit_activebackup(skb, dev);
4527         case BOND_MODE_XOR:
4528                 return bond_xmit_xor(skb, dev);
4529         case BOND_MODE_BROADCAST:
4530                 return bond_xmit_broadcast(skb, dev);
4531         case BOND_MODE_8023AD:
4532                 return bond_3ad_xmit_xor(skb, dev);
4533         case BOND_MODE_ALB:
4534         case BOND_MODE_TLB:
4535                 return bond_alb_xmit(skb, dev);
4536         default:
4537                 /* Should never happen, mode already checked */
4538                 pr_err("%s: Error: Unknown bonding mode %d\n",
4539                        dev->name, bond->params.mode);
4540                 WARN_ON_ONCE(1);
4541                 dev_kfree_skb(skb);
4542                 return NETDEV_TX_OK;
4543         }
4544 }
4545
4546
4547 /*
4548  * set bond mode specific net device operations
4549  */
4550 void bond_set_mode_ops(struct bonding *bond, int mode)
4551 {
4552         struct net_device *bond_dev = bond->dev;
4553
4554         switch (mode) {
4555         case BOND_MODE_ROUNDROBIN:
4556                 break;
4557         case BOND_MODE_ACTIVEBACKUP:
4558                 break;
4559         case BOND_MODE_XOR:
4560                 bond_set_xmit_hash_policy(bond);
4561                 break;
4562         case BOND_MODE_BROADCAST:
4563                 break;
4564         case BOND_MODE_8023AD:
4565                 bond_set_master_3ad_flags(bond);
4566                 bond_set_xmit_hash_policy(bond);
4567                 break;
4568         case BOND_MODE_ALB:
4569                 bond_set_master_alb_flags(bond);
4570                 /* FALLTHRU */
4571         case BOND_MODE_TLB:
4572                 break;
4573         default:
4574                 /* Should never happen, mode already checked */
4575                 pr_err("%s: Error: Unknown bonding mode %d\n",
4576                        bond_dev->name, mode);
4577                 break;
4578         }
4579 }
4580
4581 static void bond_ethtool_get_drvinfo(struct net_device *bond_dev,
4582                                     struct ethtool_drvinfo *drvinfo)
4583 {
4584         strncpy(drvinfo->driver, DRV_NAME, 32);
4585         strncpy(drvinfo->version, DRV_VERSION, 32);
4586         snprintf(drvinfo->fw_version, 32, "%d", BOND_ABI_VERSION);
4587 }
4588
4589 static const struct ethtool_ops bond_ethtool_ops = {
4590         .get_drvinfo            = bond_ethtool_get_drvinfo,
4591         .get_link               = ethtool_op_get_link,
4592         .get_tx_csum            = ethtool_op_get_tx_csum,
4593         .get_sg                 = ethtool_op_get_sg,
4594         .get_tso                = ethtool_op_get_tso,
4595         .get_ufo                = ethtool_op_get_ufo,
4596         .get_flags              = ethtool_op_get_flags,
4597 };
4598
4599 static const struct net_device_ops bond_netdev_ops = {
4600         .ndo_init               = bond_init,
4601         .ndo_uninit             = bond_uninit,
4602         .ndo_open               = bond_open,
4603         .ndo_stop               = bond_close,
4604         .ndo_start_xmit         = bond_start_xmit,
4605         .ndo_select_queue       = bond_select_queue,
4606         .ndo_get_stats64        = bond_get_stats,
4607         .ndo_do_ioctl           = bond_do_ioctl,
4608         .ndo_set_multicast_list = bond_set_multicast_list,
4609         .ndo_change_mtu         = bond_change_mtu,
4610         .ndo_set_mac_address    = bond_set_mac_address,
4611         .ndo_neigh_setup        = bond_neigh_setup,
4612         .ndo_vlan_rx_register   = bond_vlan_rx_register,
4613         .ndo_vlan_rx_add_vid    = bond_vlan_rx_add_vid,
4614         .ndo_vlan_rx_kill_vid   = bond_vlan_rx_kill_vid,
4615 #ifdef CONFIG_NET_POLL_CONTROLLER
4616         .ndo_netpoll_cleanup    = bond_netpoll_cleanup,
4617         .ndo_poll_controller    = bond_poll_controller,
4618 #endif
4619 };
4620
4621 static void bond_destructor(struct net_device *bond_dev)
4622 {
4623         struct bonding *bond = netdev_priv(bond_dev);
4624         if (bond->wq)
4625                 destroy_workqueue(bond->wq);
4626         free_netdev(bond_dev);
4627 }
4628
4629 static void bond_setup(struct net_device *bond_dev)
4630 {
4631         struct bonding *bond = netdev_priv(bond_dev);
4632
4633         /* initialize rwlocks */
4634         rwlock_init(&bond->lock);
4635         rwlock_init(&bond->curr_slave_lock);
4636
4637         bond->params = bonding_defaults;
4638
4639         /* Initialize pointers */
4640         bond->dev = bond_dev;
4641         INIT_LIST_HEAD(&bond->vlan_list);
4642
4643         /* Initialize the device entry points */
4644         ether_setup(bond_dev);
4645         bond_dev->netdev_ops = &bond_netdev_ops;
4646         bond_dev->ethtool_ops = &bond_ethtool_ops;
4647         bond_set_mode_ops(bond, bond->params.mode);
4648
4649         bond_dev->destructor = bond_destructor;
4650
4651         /* Initialize the device options */
4652         bond_dev->tx_queue_len = 0;
4653         bond_dev->flags |= IFF_MASTER|IFF_MULTICAST;
4654         bond_dev->priv_flags |= IFF_BONDING;
4655         bond_dev->priv_flags &= ~IFF_XMIT_DST_RELEASE;
4656
4657         if (bond->params.arp_interval)
4658                 bond_dev->priv_flags |= IFF_MASTER_ARPMON;
4659
4660         /* At first, we block adding VLANs. That's the only way to
4661          * prevent problems that occur when adding VLANs over an
4662          * empty bond. The block will be removed once non-challenged
4663          * slaves are enslaved.
4664          */
4665         bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
4666
4667         /* don't acquire bond device's netif_tx_lock when
4668          * transmitting */
4669         bond_dev->features |= NETIF_F_LLTX;
4670
4671         /* By default, we declare the bond to be fully
4672          * VLAN hardware accelerated capable. Special
4673          * care is taken in the various xmit functions
4674          * when there are slaves that are not hw accel
4675          * capable
4676          */
4677         bond_dev->features |= (NETIF_F_HW_VLAN_TX |
4678                                NETIF_F_HW_VLAN_RX |
4679                                NETIF_F_HW_VLAN_FILTER);
4680
4681         /* By default, we enable GRO on bonding devices.
4682          * Actual support requires lowlevel drivers are GRO ready.
4683          */
4684         bond_dev->features |= NETIF_F_GRO;
4685 }
4686
4687 static void bond_work_cancel_all(struct bonding *bond)
4688 {
4689         write_lock_bh(&bond->lock);
4690         bond->kill_timers = 1;
4691         write_unlock_bh(&bond->lock);
4692
4693         if (bond->params.miimon && delayed_work_pending(&bond->mii_work))
4694                 cancel_delayed_work(&bond->mii_work);
4695
4696         if (bond->params.arp_interval && delayed_work_pending(&bond->arp_work))
4697                 cancel_delayed_work(&bond->arp_work);
4698
4699         if (bond->params.mode == BOND_MODE_ALB &&
4700             delayed_work_pending(&bond->alb_work))
4701                 cancel_delayed_work(&bond->alb_work);
4702
4703         if (bond->params.mode == BOND_MODE_8023AD &&
4704             delayed_work_pending(&bond->ad_work))
4705                 cancel_delayed_work(&bond->ad_work);
4706 }
4707
4708 /*
4709 * Destroy a bonding device.
4710 * Must be under rtnl_lock when this function is called.
4711 */
4712 static void bond_uninit(struct net_device *bond_dev)
4713 {
4714         struct bonding *bond = netdev_priv(bond_dev);
4715         struct vlan_entry *vlan, *tmp;
4716
4717         bond_netpoll_cleanup(bond_dev);
4718
4719         /* Release the bonded slaves */
4720         bond_release_all(bond_dev);
4721
4722         list_del(&bond->bond_list);
4723
4724         bond_work_cancel_all(bond);
4725
4726         bond_remove_proc_entry(bond);
4727
4728         __hw_addr_flush(&bond->mc_list);
4729
4730         list_for_each_entry_safe(vlan, tmp, &bond->vlan_list, vlan_list) {
4731                 list_del(&vlan->vlan_list);
4732                 kfree(vlan);
4733         }
4734 }
4735
4736 /*------------------------- Module initialization ---------------------------*/
4737
4738 /*
4739  * Convert string input module parms.  Accept either the
4740  * number of the mode or its string name.  A bit complicated because
4741  * some mode names are substrings of other names, and calls from sysfs
4742  * may have whitespace in the name (trailing newlines, for example).
4743  */
4744 int bond_parse_parm(const char *buf, const struct bond_parm_tbl *tbl)
4745 {
4746         int modeint = -1, i, rv;
4747         char *p, modestr[BOND_MAX_MODENAME_LEN + 1] = { 0, };
4748
4749         for (p = (char *)buf; *p; p++)
4750                 if (!(isdigit(*p) || isspace(*p)))
4751                         break;
4752
4753         if (*p)
4754                 rv = sscanf(buf, "%20s", modestr);
4755         else
4756                 rv = sscanf(buf, "%d", &modeint);
4757
4758         if (!rv)
4759                 return -1;
4760
4761         for (i = 0; tbl[i].modename; i++) {
4762                 if (modeint == tbl[i].mode)
4763                         return tbl[i].mode;
4764                 if (strcmp(modestr, tbl[i].modename) == 0)
4765                         return tbl[i].mode;
4766         }
4767
4768         return -1;
4769 }
4770
4771 static int bond_check_params(struct bond_params *params)
4772 {
4773         int arp_validate_value, fail_over_mac_value, primary_reselect_value;
4774
4775         /*
4776          * Convert string parameters.
4777          */
4778         if (mode) {
4779                 bond_mode = bond_parse_parm(mode, bond_mode_tbl);
4780                 if (bond_mode == -1) {
4781                         pr_err("Error: Invalid bonding mode \"%s\"\n",
4782                                mode == NULL ? "NULL" : mode);
4783                         return -EINVAL;
4784                 }
4785         }
4786
4787         if (xmit_hash_policy) {
4788                 if ((bond_mode != BOND_MODE_XOR) &&
4789                     (bond_mode != BOND_MODE_8023AD)) {
4790                         pr_info("xmit_hash_policy param is irrelevant in mode %s\n",
4791                                bond_mode_name(bond_mode));
4792                 } else {
4793                         xmit_hashtype = bond_parse_parm(xmit_hash_policy,
4794                                                         xmit_hashtype_tbl);
4795                         if (xmit_hashtype == -1) {
4796                                 pr_err("Error: Invalid xmit_hash_policy \"%s\"\n",
4797                                        xmit_hash_policy == NULL ? "NULL" :
4798                                        xmit_hash_policy);
4799                                 return -EINVAL;
4800                         }
4801                 }
4802         }
4803
4804         if (lacp_rate) {
4805                 if (bond_mode != BOND_MODE_8023AD) {
4806                         pr_info("lacp_rate param is irrelevant in mode %s\n",
4807                                 bond_mode_name(bond_mode));
4808                 } else {
4809                         lacp_fast = bond_parse_parm(lacp_rate, bond_lacp_tbl);
4810                         if (lacp_fast == -1) {
4811                                 pr_err("Error: Invalid lacp rate \"%s\"\n",
4812                                        lacp_rate == NULL ? "NULL" : lacp_rate);
4813                                 return -EINVAL;
4814                         }
4815                 }
4816         }
4817
4818         if (ad_select) {
4819                 params->ad_select = bond_parse_parm(ad_select, ad_select_tbl);
4820                 if (params->ad_select == -1) {
4821                         pr_err("Error: Invalid ad_select \"%s\"\n",
4822                                ad_select == NULL ? "NULL" : ad_select);
4823                         return -EINVAL;
4824                 }
4825
4826                 if (bond_mode != BOND_MODE_8023AD) {
4827                         pr_warning("ad_select param only affects 802.3ad mode\n");
4828                 }
4829         } else {
4830                 params->ad_select = BOND_AD_STABLE;
4831         }
4832
4833         if (max_bonds < 0) {
4834                 pr_warning("Warning: max_bonds (%d) not in range %d-%d, so it was reset to BOND_DEFAULT_MAX_BONDS (%d)\n",
4835                            max_bonds, 0, INT_MAX, BOND_DEFAULT_MAX_BONDS);
4836                 max_bonds = BOND_DEFAULT_MAX_BONDS;
4837         }
4838
4839         if (miimon < 0) {
4840                 pr_warning("Warning: miimon module parameter (%d), not in range 0-%d, so it was reset to %d\n",
4841                            miimon, INT_MAX, BOND_LINK_MON_INTERV);
4842                 miimon = BOND_LINK_MON_INTERV;
4843         }
4844
4845         if (updelay < 0) {
4846                 pr_warning("Warning: updelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4847                            updelay, INT_MAX);
4848                 updelay = 0;
4849         }
4850
4851         if (downdelay < 0) {
4852                 pr_warning("Warning: downdelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4853                            downdelay, INT_MAX);
4854                 downdelay = 0;
4855         }
4856
4857         if ((use_carrier != 0) && (use_carrier != 1)) {
4858                 pr_warning("Warning: use_carrier module parameter (%d), not of valid value (0/1), so it was set to 1\n",
4859                            use_carrier);
4860                 use_carrier = 1;
4861         }
4862
4863         if (num_grat_arp < 0 || num_grat_arp > 255) {
4864                 pr_warning("Warning: num_grat_arp (%d) not in range 0-255 so it was reset to 1\n",
4865                            num_grat_arp);
4866                 num_grat_arp = 1;
4867         }
4868
4869         if (num_unsol_na < 0 || num_unsol_na > 255) {
4870                 pr_warning("Warning: num_unsol_na (%d) not in range 0-255 so it was reset to 1\n",
4871                            num_unsol_na);
4872                 num_unsol_na = 1;
4873         }
4874
4875         /* reset values for 802.3ad */
4876         if (bond_mode == BOND_MODE_8023AD) {
4877                 if (!miimon) {
4878                         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");
4879                         pr_warning("Forcing miimon to 100msec\n");
4880                         miimon = 100;
4881                 }
4882         }
4883
4884         if (tx_queues < 1 || tx_queues > 255) {
4885                 pr_warning("Warning: tx_queues (%d) should be between "
4886                            "1 and 255, resetting to %d\n",
4887                            tx_queues, BOND_DEFAULT_TX_QUEUES);
4888                 tx_queues = BOND_DEFAULT_TX_QUEUES;
4889         }
4890
4891         if ((all_slaves_active != 0) && (all_slaves_active != 1)) {
4892                 pr_warning("Warning: all_slaves_active module parameter (%d), "
4893                            "not of valid value (0/1), so it was set to "
4894                            "0\n", all_slaves_active);
4895                 all_slaves_active = 0;
4896         }
4897
4898         /* reset values for TLB/ALB */
4899         if ((bond_mode == BOND_MODE_TLB) ||
4900             (bond_mode == BOND_MODE_ALB)) {
4901                 if (!miimon) {
4902                         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");
4903                         pr_warning("Forcing miimon to 100msec\n");
4904                         miimon = 100;
4905                 }
4906         }
4907
4908         if (bond_mode == BOND_MODE_ALB) {
4909                 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",
4910                           updelay);
4911         }
4912
4913         if (!miimon) {
4914                 if (updelay || downdelay) {
4915                         /* just warn the user the up/down delay will have
4916                          * no effect since miimon is zero...
4917                          */
4918                         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",
4919                                    updelay, downdelay);
4920                 }
4921         } else {
4922                 /* don't allow arp monitoring */
4923                 if (arp_interval) {
4924                         pr_warning("Warning: miimon (%d) and arp_interval (%d) can't be used simultaneously, disabling ARP monitoring\n",
4925                                    miimon, arp_interval);
4926                         arp_interval = 0;
4927                 }
4928
4929                 if ((updelay % miimon) != 0) {
4930                         pr_warning("Warning: updelay (%d) is not a multiple of miimon (%d), updelay rounded to %d ms\n",
4931                                    updelay, miimon,
4932                                    (updelay / miimon) * miimon);
4933                 }
4934
4935                 updelay /= miimon;
4936
4937                 if ((downdelay % miimon) != 0) {
4938                         pr_warning("Warning: downdelay (%d) is not a multiple of miimon (%d), downdelay rounded to %d ms\n",
4939                                    downdelay, miimon,
4940                                    (downdelay / miimon) * miimon);
4941                 }
4942
4943                 downdelay /= miimon;
4944         }
4945
4946         if (arp_interval < 0) {
4947                 pr_warning("Warning: arp_interval module parameter (%d) , not in range 0-%d, so it was reset to %d\n",
4948                            arp_interval, INT_MAX, BOND_LINK_ARP_INTERV);
4949                 arp_interval = BOND_LINK_ARP_INTERV;
4950         }
4951
4952         for (arp_ip_count = 0;
4953              (arp_ip_count < BOND_MAX_ARP_TARGETS) && arp_ip_target[arp_ip_count];
4954              arp_ip_count++) {
4955                 /* not complete check, but should be good enough to
4956                    catch mistakes */
4957                 if (!isdigit(arp_ip_target[arp_ip_count][0])) {
4958                         pr_warning("Warning: bad arp_ip_target module parameter (%s), ARP monitoring will not be performed\n",
4959                                    arp_ip_target[arp_ip_count]);
4960                         arp_interval = 0;
4961                 } else {
4962                         __be32 ip = in_aton(arp_ip_target[arp_ip_count]);
4963                         arp_target[arp_ip_count] = ip;
4964                 }
4965         }
4966
4967         if (arp_interval && !arp_ip_count) {
4968                 /* don't allow arping if no arp_ip_target given... */
4969                 pr_warning("Warning: arp_interval module parameter (%d) specified without providing an arp_ip_target parameter, arp_interval was reset to 0\n",
4970                            arp_interval);
4971                 arp_interval = 0;
4972         }
4973
4974         if (arp_validate) {
4975                 if (bond_mode != BOND_MODE_ACTIVEBACKUP) {
4976                         pr_err("arp_validate only supported in active-backup mode\n");
4977                         return -EINVAL;
4978                 }
4979                 if (!arp_interval) {
4980                         pr_err("arp_validate requires arp_interval\n");
4981                         return -EINVAL;
4982                 }
4983
4984                 arp_validate_value = bond_parse_parm(arp_validate,
4985                                                      arp_validate_tbl);
4986                 if (arp_validate_value == -1) {
4987                         pr_err("Error: invalid arp_validate \"%s\"\n",
4988                                arp_validate == NULL ? "NULL" : arp_validate);
4989                         return -EINVAL;
4990                 }
4991         } else
4992                 arp_validate_value = 0;
4993
4994         if (miimon) {
4995                 pr_info("MII link monitoring set to %d ms\n", miimon);
4996         } else if (arp_interval) {
4997                 int i;
4998
4999                 pr_info("ARP monitoring set to %d ms, validate %s, with %d target(s):",
5000                         arp_interval,
5001                         arp_validate_tbl[arp_validate_value].modename,
5002                         arp_ip_count);
5003
5004                 for (i = 0; i < arp_ip_count; i++)
5005                         pr_info(" %s", arp_ip_target[i]);
5006
5007                 pr_info("\n");
5008
5009         } else if (max_bonds) {
5010                 /* miimon and arp_interval not set, we need one so things
5011                  * work as expected, see bonding.txt for details
5012                  */
5013                 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");
5014         }
5015
5016         if (primary && !USES_PRIMARY(bond_mode)) {
5017                 /* currently, using a primary only makes sense
5018                  * in active backup, TLB or ALB modes
5019                  */
5020                 pr_warning("Warning: %s primary device specified but has no effect in %s mode\n",
5021                            primary, bond_mode_name(bond_mode));
5022                 primary = NULL;
5023         }
5024
5025         if (primary && primary_reselect) {
5026                 primary_reselect_value = bond_parse_parm(primary_reselect,
5027                                                          pri_reselect_tbl);
5028                 if (primary_reselect_value == -1) {
5029                         pr_err("Error: Invalid primary_reselect \"%s\"\n",
5030                                primary_reselect ==
5031                                         NULL ? "NULL" : primary_reselect);
5032                         return -EINVAL;
5033                 }
5034         } else {
5035                 primary_reselect_value = BOND_PRI_RESELECT_ALWAYS;
5036         }
5037
5038         if (fail_over_mac) {
5039                 fail_over_mac_value = bond_parse_parm(fail_over_mac,
5040                                                       fail_over_mac_tbl);
5041                 if (fail_over_mac_value == -1) {
5042                         pr_err("Error: invalid fail_over_mac \"%s\"\n",
5043                                arp_validate == NULL ? "NULL" : arp_validate);
5044                         return -EINVAL;
5045                 }
5046
5047                 if (bond_mode != BOND_MODE_ACTIVEBACKUP)
5048                         pr_warning("Warning: fail_over_mac only affects active-backup mode.\n");
5049         } else {
5050                 fail_over_mac_value = BOND_FOM_NONE;
5051         }
5052
5053         /* fill params struct with the proper values */
5054         params->mode = bond_mode;
5055         params->xmit_policy = xmit_hashtype;
5056         params->miimon = miimon;
5057         params->num_grat_arp = num_grat_arp;
5058         params->num_unsol_na = num_unsol_na;
5059         params->arp_interval = arp_interval;
5060         params->arp_validate = arp_validate_value;
5061         params->updelay = updelay;
5062         params->downdelay = downdelay;
5063         params->use_carrier = use_carrier;
5064         params->lacp_fast = lacp_fast;
5065         params->primary[0] = 0;
5066         params->primary_reselect = primary_reselect_value;
5067         params->fail_over_mac = fail_over_mac_value;
5068         params->tx_queues = tx_queues;
5069         params->all_slaves_active = all_slaves_active;
5070
5071         if (primary) {
5072                 strncpy(params->primary, primary, IFNAMSIZ);
5073                 params->primary[IFNAMSIZ - 1] = 0;
5074         }
5075
5076         memcpy(params->arp_targets, arp_target, sizeof(arp_target));
5077
5078         return 0;
5079 }
5080
5081 static struct lock_class_key bonding_netdev_xmit_lock_key;
5082 static struct lock_class_key bonding_netdev_addr_lock_key;
5083
5084 static void bond_set_lockdep_class_one(struct net_device *dev,
5085                                        struct netdev_queue *txq,
5086                                        void *_unused)
5087 {
5088         lockdep_set_class(&txq->_xmit_lock,
5089                           &bonding_netdev_xmit_lock_key);
5090 }
5091
5092 static void bond_set_lockdep_class(struct net_device *dev)
5093 {
5094         lockdep_set_class(&dev->addr_list_lock,
5095                           &bonding_netdev_addr_lock_key);
5096         netdev_for_each_tx_queue(dev, bond_set_lockdep_class_one, NULL);
5097 }
5098
5099 /*
5100  * Called from registration process
5101  */
5102 static int bond_init(struct net_device *bond_dev)
5103 {
5104         struct bonding *bond = netdev_priv(bond_dev);
5105         struct bond_net *bn = net_generic(dev_net(bond_dev), bond_net_id);
5106
5107         pr_debug("Begin bond_init for %s\n", bond_dev->name);
5108
5109         bond->wq = create_singlethread_workqueue(bond_dev->name);
5110         if (!bond->wq)
5111                 return -ENOMEM;
5112
5113         bond_set_lockdep_class(bond_dev);
5114
5115         netif_carrier_off(bond_dev);
5116
5117         bond_create_proc_entry(bond);
5118         list_add_tail(&bond->bond_list, &bn->dev_list);
5119
5120         bond_prepare_sysfs_group(bond);
5121
5122         __hw_addr_init(&bond->mc_list);
5123         return 0;
5124 }
5125
5126 static int bond_validate(struct nlattr *tb[], struct nlattr *data[])
5127 {
5128         if (tb[IFLA_ADDRESS]) {
5129                 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN)
5130                         return -EINVAL;
5131                 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS])))
5132                         return -EADDRNOTAVAIL;
5133         }
5134         return 0;
5135 }
5136
5137 static struct rtnl_link_ops bond_link_ops __read_mostly = {
5138         .kind           = "bond",
5139         .priv_size      = sizeof(struct bonding),
5140         .setup          = bond_setup,
5141         .validate       = bond_validate,
5142 };
5143
5144 /* Create a new bond based on the specified name and bonding parameters.
5145  * If name is NULL, obtain a suitable "bond%d" name for us.
5146  * Caller must NOT hold rtnl_lock; we need to release it here before we
5147  * set up our sysfs entries.
5148  */
5149 int bond_create(struct net *net, const char *name)
5150 {
5151         struct net_device *bond_dev;
5152         int res;
5153
5154         rtnl_lock();
5155
5156         bond_dev = alloc_netdev_mq(sizeof(struct bonding), name ? name : "",
5157                                 bond_setup, tx_queues);
5158         if (!bond_dev) {
5159                 pr_err("%s: eek! can't alloc netdev!\n", name);
5160                 rtnl_unlock();
5161                 return -ENOMEM;
5162         }
5163
5164         dev_net_set(bond_dev, net);
5165         bond_dev->rtnl_link_ops = &bond_link_ops;
5166
5167         if (!name) {
5168                 res = dev_alloc_name(bond_dev, "bond%d");
5169                 if (res < 0)
5170                         goto out;
5171         }
5172
5173         res = register_netdevice(bond_dev);
5174
5175 out:
5176         rtnl_unlock();
5177         if (res < 0)
5178                 bond_destructor(bond_dev);
5179         return res;
5180 }
5181
5182 static int __net_init bond_net_init(struct net *net)
5183 {
5184         struct bond_net *bn = net_generic(net, bond_net_id);
5185
5186         bn->net = net;
5187         INIT_LIST_HEAD(&bn->dev_list);
5188
5189         bond_create_proc_dir(bn);
5190         
5191         return 0;
5192 }
5193
5194 static void __net_exit bond_net_exit(struct net *net)
5195 {
5196         struct bond_net *bn = net_generic(net, bond_net_id);
5197
5198         bond_destroy_proc_dir(bn);
5199 }
5200
5201 static struct pernet_operations bond_net_ops = {
5202         .init = bond_net_init,
5203         .exit = bond_net_exit,
5204         .id   = &bond_net_id,
5205         .size = sizeof(struct bond_net),
5206 };
5207
5208 static int __init bonding_init(void)
5209 {
5210         int i;
5211         int res;
5212
5213         pr_info("%s", version);
5214
5215         res = bond_check_params(&bonding_defaults);
5216         if (res)
5217                 goto out;
5218
5219         res = register_pernet_subsys(&bond_net_ops);
5220         if (res)
5221                 goto out;
5222
5223         res = rtnl_link_register(&bond_link_ops);
5224         if (res)
5225                 goto err_link;
5226
5227         for (i = 0; i < max_bonds; i++) {
5228                 res = bond_create(&init_net, NULL);
5229                 if (res)
5230                         goto err;
5231         }
5232
5233         res = bond_create_sysfs();
5234         if (res)
5235                 goto err;
5236
5237         register_netdevice_notifier(&bond_netdev_notifier);
5238         register_inetaddr_notifier(&bond_inetaddr_notifier);
5239         bond_register_ipv6_notifier();
5240 out:
5241         return res;
5242 err:
5243         rtnl_link_unregister(&bond_link_ops);
5244 err_link:
5245         unregister_pernet_subsys(&bond_net_ops);
5246         goto out;
5247
5248 }
5249
5250 static void __exit bonding_exit(void)
5251 {
5252         unregister_netdevice_notifier(&bond_netdev_notifier);
5253         unregister_inetaddr_notifier(&bond_inetaddr_notifier);
5254         bond_unregister_ipv6_notifier();
5255
5256         bond_destroy_sysfs();
5257
5258         rtnl_link_unregister(&bond_link_ops);
5259         unregister_pernet_subsys(&bond_net_ops);
5260 }
5261
5262 module_init(bonding_init);
5263 module_exit(bonding_exit);
5264 MODULE_LICENSE("GPL");
5265 MODULE_VERSION(DRV_VERSION);
5266 MODULE_DESCRIPTION(DRV_DESCRIPTION ", v" DRV_VERSION);
5267 MODULE_AUTHOR("Thomas Davis, tadavis@lbl.gov and many others");
5268 MODULE_ALIAS_RTNL_LINK("bond");