bonding: fix a race in IGMP handling
[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 <linux/preempt.h>
80 #include <net/route.h>
81 #include <net/net_namespace.h>
82 #include <net/netns/generic.h>
83 #include "bonding.h"
84 #include "bond_3ad.h"
85 #include "bond_alb.h"
86
87 /*---------------------------- Module parameters ----------------------------*/
88
89 /* monitor all links that often (in milliseconds). <=0 disables monitoring */
90 #define BOND_LINK_MON_INTERV    0
91 #define BOND_LINK_ARP_INTERV    0
92
93 static int max_bonds    = BOND_DEFAULT_MAX_BONDS;
94 static int tx_queues    = BOND_DEFAULT_TX_QUEUES;
95 static int num_grat_arp = 1;
96 static int num_unsol_na = 1;
97 static int miimon       = BOND_LINK_MON_INTERV;
98 static int updelay;
99 static int downdelay;
100 static int use_carrier  = 1;
101 static char *mode;
102 static char *primary;
103 static char *primary_reselect;
104 static char *lacp_rate;
105 static char *ad_select;
106 static char *xmit_hash_policy;
107 static int arp_interval = BOND_LINK_ARP_INTERV;
108 static char *arp_ip_target[BOND_MAX_ARP_TARGETS];
109 static char *arp_validate;
110 static char *fail_over_mac;
111 static int all_slaves_active = 0;
112 static struct bond_params bonding_defaults;
113 static int resend_igmp = BOND_DEFAULT_RESEND_IGMP;
114
115 module_param(max_bonds, int, 0);
116 MODULE_PARM_DESC(max_bonds, "Max number of bonded devices");
117 module_param(tx_queues, int, 0);
118 MODULE_PARM_DESC(tx_queues, "Max number of transmit queues (default = 16)");
119 module_param(num_grat_arp, int, 0644);
120 MODULE_PARM_DESC(num_grat_arp, "Number of gratuitous ARP packets to send on failover event");
121 module_param(num_unsol_na, int, 0644);
122 MODULE_PARM_DESC(num_unsol_na, "Number of unsolicited IPv6 Neighbor Advertisements packets to send on failover event");
123 module_param(miimon, int, 0);
124 MODULE_PARM_DESC(miimon, "Link check interval in milliseconds");
125 module_param(updelay, int, 0);
126 MODULE_PARM_DESC(updelay, "Delay before considering link up, in milliseconds");
127 module_param(downdelay, int, 0);
128 MODULE_PARM_DESC(downdelay, "Delay before considering link down, "
129                             "in milliseconds");
130 module_param(use_carrier, int, 0);
131 MODULE_PARM_DESC(use_carrier, "Use netif_carrier_ok (vs MII ioctls) in miimon; "
132                               "0 for off, 1 for on (default)");
133 module_param(mode, charp, 0);
134 MODULE_PARM_DESC(mode, "Mode of operation : 0 for balance-rr, "
135                        "1 for active-backup, 2 for balance-xor, "
136                        "3 for broadcast, 4 for 802.3ad, 5 for balance-tlb, "
137                        "6 for balance-alb");
138 module_param(primary, charp, 0);
139 MODULE_PARM_DESC(primary, "Primary network device to use");
140 module_param(primary_reselect, charp, 0);
141 MODULE_PARM_DESC(primary_reselect, "Reselect primary slave "
142                                    "once it comes up; "
143                                    "0 for always (default), "
144                                    "1 for only if speed of primary is "
145                                    "better, "
146                                    "2 for only on active slave "
147                                    "failure");
148 module_param(lacp_rate, charp, 0);
149 MODULE_PARM_DESC(lacp_rate, "LACPDU tx rate to request from 802.3ad partner "
150                             "(slow/fast)");
151 module_param(ad_select, charp, 0);
152 MODULE_PARM_DESC(ad_select, "803.ad aggregation selection logic: stable (0, default), bandwidth (1), count (2)");
153 module_param(xmit_hash_policy, charp, 0);
154 MODULE_PARM_DESC(xmit_hash_policy, "XOR hashing method: 0 for layer 2 (default)"
155                                    ", 1 for layer 3+4");
156 module_param(arp_interval, int, 0);
157 MODULE_PARM_DESC(arp_interval, "arp interval in milliseconds");
158 module_param_array(arp_ip_target, charp, NULL, 0);
159 MODULE_PARM_DESC(arp_ip_target, "arp targets in n.n.n.n form");
160 module_param(arp_validate, charp, 0);
161 MODULE_PARM_DESC(arp_validate, "validate src/dst of ARP probes: none (default), active, backup or all");
162 module_param(fail_over_mac, charp, 0);
163 MODULE_PARM_DESC(fail_over_mac, "For active-backup, do not set all slaves to the same MAC.  none (default), active or follow");
164 module_param(all_slaves_active, int, 0);
165 MODULE_PARM_DESC(all_slaves_active, "Keep all frames received on an interface"
166                                      "by setting active flag for all slaves.  "
167                                      "0 for never (default), 1 for always.");
168 module_param(resend_igmp, int, 0);
169 MODULE_PARM_DESC(resend_igmp, "Number of IGMP membership reports to send on link failure");
170
171 /*----------------------------- Global variables ----------------------------*/
172
173 #ifdef CONFIG_NET_POLL_CONTROLLER
174 cpumask_var_t netpoll_block_tx;
175 #endif
176
177 static const char * const version =
178         DRV_DESCRIPTION ": v" DRV_VERSION " (" DRV_RELDATE ")\n";
179
180 int bond_net_id __read_mostly;
181
182 static __be32 arp_target[BOND_MAX_ARP_TARGETS];
183 static int arp_ip_count;
184 static int bond_mode    = BOND_MODE_ROUNDROBIN;
185 static int xmit_hashtype = BOND_XMIT_POLICY_LAYER2;
186 static int lacp_fast;
187
188 const struct bond_parm_tbl bond_lacp_tbl[] = {
189 {       "slow",         AD_LACP_SLOW},
190 {       "fast",         AD_LACP_FAST},
191 {       NULL,           -1},
192 };
193
194 const struct bond_parm_tbl bond_mode_tbl[] = {
195 {       "balance-rr",           BOND_MODE_ROUNDROBIN},
196 {       "active-backup",        BOND_MODE_ACTIVEBACKUP},
197 {       "balance-xor",          BOND_MODE_XOR},
198 {       "broadcast",            BOND_MODE_BROADCAST},
199 {       "802.3ad",              BOND_MODE_8023AD},
200 {       "balance-tlb",          BOND_MODE_TLB},
201 {       "balance-alb",          BOND_MODE_ALB},
202 {       NULL,                   -1},
203 };
204
205 const struct bond_parm_tbl xmit_hashtype_tbl[] = {
206 {       "layer2",               BOND_XMIT_POLICY_LAYER2},
207 {       "layer3+4",             BOND_XMIT_POLICY_LAYER34},
208 {       "layer2+3",             BOND_XMIT_POLICY_LAYER23},
209 {       NULL,                   -1},
210 };
211
212 const struct bond_parm_tbl arp_validate_tbl[] = {
213 {       "none",                 BOND_ARP_VALIDATE_NONE},
214 {       "active",               BOND_ARP_VALIDATE_ACTIVE},
215 {       "backup",               BOND_ARP_VALIDATE_BACKUP},
216 {       "all",                  BOND_ARP_VALIDATE_ALL},
217 {       NULL,                   -1},
218 };
219
220 const struct bond_parm_tbl fail_over_mac_tbl[] = {
221 {       "none",                 BOND_FOM_NONE},
222 {       "active",               BOND_FOM_ACTIVE},
223 {       "follow",               BOND_FOM_FOLLOW},
224 {       NULL,                   -1},
225 };
226
227 const struct bond_parm_tbl pri_reselect_tbl[] = {
228 {       "always",               BOND_PRI_RESELECT_ALWAYS},
229 {       "better",               BOND_PRI_RESELECT_BETTER},
230 {       "failure",              BOND_PRI_RESELECT_FAILURE},
231 {       NULL,                   -1},
232 };
233
234 struct bond_parm_tbl ad_select_tbl[] = {
235 {       "stable",       BOND_AD_STABLE},
236 {       "bandwidth",    BOND_AD_BANDWIDTH},
237 {       "count",        BOND_AD_COUNT},
238 {       NULL,           -1},
239 };
240
241 /*-------------------------- Forward declarations ---------------------------*/
242
243 static void bond_send_gratuitous_arp(struct bonding *bond);
244 static int bond_init(struct net_device *bond_dev);
245 static void bond_uninit(struct net_device *bond_dev);
246
247 /*---------------------------- General routines -----------------------------*/
248
249 static const char *bond_mode_name(int mode)
250 {
251         static const char *names[] = {
252                 [BOND_MODE_ROUNDROBIN] = "load balancing (round-robin)",
253                 [BOND_MODE_ACTIVEBACKUP] = "fault-tolerance (active-backup)",
254                 [BOND_MODE_XOR] = "load balancing (xor)",
255                 [BOND_MODE_BROADCAST] = "fault-tolerance (broadcast)",
256                 [BOND_MODE_8023AD] = "IEEE 802.3ad Dynamic link aggregation",
257                 [BOND_MODE_TLB] = "transmit load balancing",
258                 [BOND_MODE_ALB] = "adaptive load balancing",
259         };
260
261         if (mode < 0 || mode > BOND_MODE_ALB)
262                 return "unknown";
263
264         return names[mode];
265 }
266
267 /*---------------------------------- VLAN -----------------------------------*/
268
269 /**
270  * bond_add_vlan - add a new vlan id on bond
271  * @bond: bond that got the notification
272  * @vlan_id: the vlan id to add
273  *
274  * Returns -ENOMEM if allocation failed.
275  */
276 static int bond_add_vlan(struct bonding *bond, unsigned short vlan_id)
277 {
278         struct vlan_entry *vlan;
279
280         pr_debug("bond: %s, vlan id %d\n",
281                  (bond ? bond->dev->name : "None"), vlan_id);
282
283         vlan = kzalloc(sizeof(struct vlan_entry), GFP_KERNEL);
284         if (!vlan)
285                 return -ENOMEM;
286
287         INIT_LIST_HEAD(&vlan->vlan_list);
288         vlan->vlan_id = vlan_id;
289
290         write_lock_bh(&bond->lock);
291
292         list_add_tail(&vlan->vlan_list, &bond->vlan_list);
293
294         write_unlock_bh(&bond->lock);
295
296         pr_debug("added VLAN ID %d on bond %s\n", vlan_id, bond->dev->name);
297
298         return 0;
299 }
300
301 /**
302  * bond_del_vlan - delete a vlan id from bond
303  * @bond: bond that got the notification
304  * @vlan_id: the vlan id to delete
305  *
306  * returns -ENODEV if @vlan_id was not found in @bond.
307  */
308 static int bond_del_vlan(struct bonding *bond, unsigned short vlan_id)
309 {
310         struct vlan_entry *vlan;
311         int res = -ENODEV;
312
313         pr_debug("bond: %s, vlan id %d\n", bond->dev->name, vlan_id);
314
315         block_netpoll_tx();
316         write_lock_bh(&bond->lock);
317
318         list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
319                 if (vlan->vlan_id == vlan_id) {
320                         list_del(&vlan->vlan_list);
321
322                         if (bond_is_lb(bond))
323                                 bond_alb_clear_vlan(bond, vlan_id);
324
325                         pr_debug("removed VLAN ID %d from bond %s\n",
326                                  vlan_id, bond->dev->name);
327
328                         kfree(vlan);
329
330                         if (list_empty(&bond->vlan_list) &&
331                             (bond->slave_cnt == 0)) {
332                                 /* Last VLAN removed and no slaves, so
333                                  * restore block on adding VLANs. This will
334                                  * be removed once new slaves that are not
335                                  * VLAN challenged will be added.
336                                  */
337                                 bond->dev->features |= NETIF_F_VLAN_CHALLENGED;
338                         }
339
340                         res = 0;
341                         goto out;
342                 }
343         }
344
345         pr_debug("couldn't find VLAN ID %d in bond %s\n",
346                  vlan_id, bond->dev->name);
347
348 out:
349         write_unlock_bh(&bond->lock);
350         unblock_netpoll_tx();
351         return res;
352 }
353
354 /**
355  * bond_has_challenged_slaves
356  * @bond: the bond we're working on
357  *
358  * Searches the slave list. Returns 1 if a vlan challenged slave
359  * was found, 0 otherwise.
360  *
361  * Assumes bond->lock is held.
362  */
363 static int bond_has_challenged_slaves(struct bonding *bond)
364 {
365         struct slave *slave;
366         int i;
367
368         bond_for_each_slave(bond, slave, i) {
369                 if (slave->dev->features & NETIF_F_VLAN_CHALLENGED) {
370                         pr_debug("found VLAN challenged slave - %s\n",
371                                  slave->dev->name);
372                         return 1;
373                 }
374         }
375
376         pr_debug("no VLAN challenged slaves found\n");
377         return 0;
378 }
379
380 /**
381  * bond_next_vlan - safely skip to the next item in the vlans list.
382  * @bond: the bond we're working on
383  * @curr: item we're advancing from
384  *
385  * Returns %NULL if list is empty, bond->next_vlan if @curr is %NULL,
386  * or @curr->next otherwise (even if it is @curr itself again).
387  *
388  * Caller must hold bond->lock
389  */
390 struct vlan_entry *bond_next_vlan(struct bonding *bond, struct vlan_entry *curr)
391 {
392         struct vlan_entry *next, *last;
393
394         if (list_empty(&bond->vlan_list))
395                 return NULL;
396
397         if (!curr) {
398                 next = list_entry(bond->vlan_list.next,
399                                   struct vlan_entry, vlan_list);
400         } else {
401                 last = list_entry(bond->vlan_list.prev,
402                                   struct vlan_entry, vlan_list);
403                 if (last == curr) {
404                         next = list_entry(bond->vlan_list.next,
405                                           struct vlan_entry, vlan_list);
406                 } else {
407                         next = list_entry(curr->vlan_list.next,
408                                           struct vlan_entry, vlan_list);
409                 }
410         }
411
412         return next;
413 }
414
415 /**
416  * bond_dev_queue_xmit - Prepare skb for xmit.
417  *
418  * @bond: bond device that got this skb for tx.
419  * @skb: hw accel VLAN tagged skb to transmit
420  * @slave_dev: slave that is supposed to xmit this skbuff
421  *
422  * When the bond gets an skb to transmit that is
423  * already hardware accelerated VLAN tagged, and it
424  * needs to relay this skb to a slave that is not
425  * hw accel capable, the skb needs to be "unaccelerated",
426  * i.e. strip the hwaccel tag and re-insert it as part
427  * of the payload.
428  */
429 int bond_dev_queue_xmit(struct bonding *bond, struct sk_buff *skb,
430                         struct net_device *slave_dev)
431 {
432         unsigned short uninitialized_var(vlan_id);
433
434         /* Test vlan_list not vlgrp to catch and handle 802.1p tags */
435         if (!list_empty(&bond->vlan_list) &&
436             !(slave_dev->features & NETIF_F_HW_VLAN_TX) &&
437             vlan_get_tag(skb, &vlan_id) == 0) {
438                 skb->dev = slave_dev;
439                 skb = vlan_put_tag(skb, vlan_id);
440                 if (!skb) {
441                         /* vlan_put_tag() frees the skb in case of error,
442                          * so return success here so the calling functions
443                          * won't attempt to free is again.
444                          */
445                         return 0;
446                 }
447         } else {
448                 skb->dev = slave_dev;
449         }
450
451         skb->priority = 1;
452 #ifdef CONFIG_NET_POLL_CONTROLLER
453         if (unlikely(bond->dev->priv_flags & IFF_IN_NETPOLL)) {
454                 struct netpoll *np = bond->dev->npinfo->netpoll;
455                 slave_dev->npinfo = bond->dev->npinfo;
456                 slave_dev->priv_flags |= IFF_IN_NETPOLL;
457                 netpoll_send_skb_on_dev(np, skb, slave_dev);
458                 slave_dev->priv_flags &= ~IFF_IN_NETPOLL;
459         } else
460 #endif
461                 dev_queue_xmit(skb);
462
463         return 0;
464 }
465
466 /*
467  * In the following 3 functions, bond_vlan_rx_register(), bond_vlan_rx_add_vid
468  * and bond_vlan_rx_kill_vid, We don't protect the slave list iteration with a
469  * lock because:
470  * a. This operation is performed in IOCTL context,
471  * b. The operation is protected by the RTNL semaphore in the 8021q code,
472  * c. Holding a lock with BH disabled while directly calling a base driver
473  *    entry point is generally a BAD idea.
474  *
475  * The design of synchronization/protection for this operation in the 8021q
476  * module is good for one or more VLAN devices over a single physical device
477  * and cannot be extended for a teaming solution like bonding, so there is a
478  * potential race condition here where a net device from the vlan group might
479  * be referenced (either by a base driver or the 8021q code) while it is being
480  * removed from the system. However, it turns out we're not making matters
481  * worse, and if it works for regular VLAN usage it will work here too.
482 */
483
484 /**
485  * bond_vlan_rx_register - Propagates registration to slaves
486  * @bond_dev: bonding net device that got called
487  * @grp: vlan group being registered
488  */
489 static void bond_vlan_rx_register(struct net_device *bond_dev,
490                                   struct vlan_group *grp)
491 {
492         struct bonding *bond = netdev_priv(bond_dev);
493         struct slave *slave;
494         int i;
495
496         write_lock_bh(&bond->lock);
497         bond->vlgrp = grp;
498         write_unlock_bh(&bond->lock);
499
500         bond_for_each_slave(bond, slave, i) {
501                 struct net_device *slave_dev = slave->dev;
502                 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
503
504                 if ((slave_dev->features & NETIF_F_HW_VLAN_RX) &&
505                     slave_ops->ndo_vlan_rx_register) {
506                         slave_ops->ndo_vlan_rx_register(slave_dev, grp);
507                 }
508         }
509 }
510
511 /**
512  * bond_vlan_rx_add_vid - Propagates adding an id to slaves
513  * @bond_dev: bonding net device that got called
514  * @vid: vlan id being added
515  */
516 static void bond_vlan_rx_add_vid(struct net_device *bond_dev, uint16_t vid)
517 {
518         struct bonding *bond = netdev_priv(bond_dev);
519         struct slave *slave;
520         int i, res;
521
522         bond_for_each_slave(bond, slave, i) {
523                 struct net_device *slave_dev = slave->dev;
524                 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
525
526                 if ((slave_dev->features & NETIF_F_HW_VLAN_FILTER) &&
527                     slave_ops->ndo_vlan_rx_add_vid) {
528                         slave_ops->ndo_vlan_rx_add_vid(slave_dev, vid);
529                 }
530         }
531
532         res = bond_add_vlan(bond, vid);
533         if (res) {
534                 pr_err("%s: Error: Failed to add vlan id %d\n",
535                        bond_dev->name, vid);
536         }
537 }
538
539 /**
540  * bond_vlan_rx_kill_vid - Propagates deleting an id to slaves
541  * @bond_dev: bonding net device that got called
542  * @vid: vlan id being removed
543  */
544 static void bond_vlan_rx_kill_vid(struct net_device *bond_dev, uint16_t vid)
545 {
546         struct bonding *bond = netdev_priv(bond_dev);
547         struct slave *slave;
548         struct net_device *vlan_dev;
549         int i, res;
550
551         bond_for_each_slave(bond, slave, i) {
552                 struct net_device *slave_dev = slave->dev;
553                 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
554
555                 if ((slave_dev->features & NETIF_F_HW_VLAN_FILTER) &&
556                     slave_ops->ndo_vlan_rx_kill_vid) {
557                         /* Save and then restore vlan_dev in the grp array,
558                          * since the slave's driver might clear it.
559                          */
560                         vlan_dev = vlan_group_get_device(bond->vlgrp, vid);
561                         slave_ops->ndo_vlan_rx_kill_vid(slave_dev, vid);
562                         vlan_group_set_device(bond->vlgrp, vid, vlan_dev);
563                 }
564         }
565
566         res = bond_del_vlan(bond, vid);
567         if (res) {
568                 pr_err("%s: Error: Failed to remove vlan id %d\n",
569                        bond_dev->name, vid);
570         }
571 }
572
573 static void bond_add_vlans_on_slave(struct bonding *bond, struct net_device *slave_dev)
574 {
575         struct vlan_entry *vlan;
576         const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
577
578         if (!bond->vlgrp)
579                 return;
580
581         if ((slave_dev->features & NETIF_F_HW_VLAN_RX) &&
582             slave_ops->ndo_vlan_rx_register)
583                 slave_ops->ndo_vlan_rx_register(slave_dev, bond->vlgrp);
584
585         if (!(slave_dev->features & NETIF_F_HW_VLAN_FILTER) ||
586             !(slave_ops->ndo_vlan_rx_add_vid))
587                 return;
588
589         list_for_each_entry(vlan, &bond->vlan_list, vlan_list)
590                 slave_ops->ndo_vlan_rx_add_vid(slave_dev, vlan->vlan_id);
591 }
592
593 static void bond_del_vlans_from_slave(struct bonding *bond,
594                                       struct net_device *slave_dev)
595 {
596         const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
597         struct vlan_entry *vlan;
598         struct net_device *vlan_dev;
599
600         if (!bond->vlgrp)
601                 return;
602
603         if (!(slave_dev->features & NETIF_F_HW_VLAN_FILTER) ||
604             !(slave_ops->ndo_vlan_rx_kill_vid))
605                 goto unreg;
606
607         list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
608                 if (!vlan->vlan_id)
609                         continue;
610                 /* Save and then restore vlan_dev in the grp array,
611                  * since the slave's driver might clear it.
612                  */
613                 vlan_dev = vlan_group_get_device(bond->vlgrp, vlan->vlan_id);
614                 slave_ops->ndo_vlan_rx_kill_vid(slave_dev, vlan->vlan_id);
615                 vlan_group_set_device(bond->vlgrp, vlan->vlan_id, vlan_dev);
616         }
617
618 unreg:
619         if ((slave_dev->features & NETIF_F_HW_VLAN_RX) &&
620             slave_ops->ndo_vlan_rx_register)
621                 slave_ops->ndo_vlan_rx_register(slave_dev, NULL);
622 }
623
624 /*------------------------------- Link status -------------------------------*/
625
626 /*
627  * Set the carrier state for the master according to the state of its
628  * slaves.  If any slaves are up, the master is up.  In 802.3ad mode,
629  * do special 802.3ad magic.
630  *
631  * Returns zero if carrier state does not change, nonzero if it does.
632  */
633 static int bond_set_carrier(struct bonding *bond)
634 {
635         struct slave *slave;
636         int i;
637
638         if (bond->slave_cnt == 0)
639                 goto down;
640
641         if (bond->params.mode == BOND_MODE_8023AD)
642                 return bond_3ad_set_carrier(bond);
643
644         bond_for_each_slave(bond, slave, i) {
645                 if (slave->link == BOND_LINK_UP) {
646                         if (!netif_carrier_ok(bond->dev)) {
647                                 netif_carrier_on(bond->dev);
648                                 return 1;
649                         }
650                         return 0;
651                 }
652         }
653
654 down:
655         if (netif_carrier_ok(bond->dev)) {
656                 netif_carrier_off(bond->dev);
657                 return 1;
658         }
659         return 0;
660 }
661
662 /*
663  * Get link speed and duplex from the slave's base driver
664  * using ethtool. If for some reason the call fails or the
665  * values are invalid, fake speed and duplex to 100/Full
666  * and return error.
667  */
668 static int bond_update_speed_duplex(struct slave *slave)
669 {
670         struct net_device *slave_dev = slave->dev;
671         struct ethtool_cmd etool;
672         int res;
673
674         /* Fake speed and duplex */
675         slave->speed = SPEED_100;
676         slave->duplex = DUPLEX_FULL;
677
678         if (!slave_dev->ethtool_ops || !slave_dev->ethtool_ops->get_settings)
679                 return -1;
680
681         res = slave_dev->ethtool_ops->get_settings(slave_dev, &etool);
682         if (res < 0)
683                 return -1;
684
685         switch (etool.speed) {
686         case SPEED_10:
687         case SPEED_100:
688         case SPEED_1000:
689         case SPEED_10000:
690                 break;
691         default:
692                 return -1;
693         }
694
695         switch (etool.duplex) {
696         case DUPLEX_FULL:
697         case DUPLEX_HALF:
698                 break;
699         default:
700                 return -1;
701         }
702
703         slave->speed = etool.speed;
704         slave->duplex = etool.duplex;
705
706         return 0;
707 }
708
709 /*
710  * if <dev> supports MII link status reporting, check its link status.
711  *
712  * We either do MII/ETHTOOL ioctls, or check netif_carrier_ok(),
713  * depending upon the setting of the use_carrier parameter.
714  *
715  * Return either BMSR_LSTATUS, meaning that the link is up (or we
716  * can't tell and just pretend it is), or 0, meaning that the link is
717  * down.
718  *
719  * If reporting is non-zero, instead of faking link up, return -1 if
720  * both ETHTOOL and MII ioctls fail (meaning the device does not
721  * support them).  If use_carrier is set, return whatever it says.
722  * It'd be nice if there was a good way to tell if a driver supports
723  * netif_carrier, but there really isn't.
724  */
725 static int bond_check_dev_link(struct bonding *bond,
726                                struct net_device *slave_dev, int reporting)
727 {
728         const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
729         int (*ioctl)(struct net_device *, struct ifreq *, int);
730         struct ifreq ifr;
731         struct mii_ioctl_data *mii;
732
733         if (!reporting && !netif_running(slave_dev))
734                 return 0;
735
736         if (bond->params.use_carrier)
737                 return netif_carrier_ok(slave_dev) ? BMSR_LSTATUS : 0;
738
739         /* Try to get link status using Ethtool first. */
740         if (slave_dev->ethtool_ops) {
741                 if (slave_dev->ethtool_ops->get_link) {
742                         u32 link;
743
744                         link = slave_dev->ethtool_ops->get_link(slave_dev);
745
746                         return link ? BMSR_LSTATUS : 0;
747                 }
748         }
749
750         /* Ethtool can't be used, fallback to MII ioctls. */
751         ioctl = slave_ops->ndo_do_ioctl;
752         if (ioctl) {
753                 /* TODO: set pointer to correct ioctl on a per team member */
754                 /*       bases to make this more efficient. that is, once  */
755                 /*       we determine the correct ioctl, we will always    */
756                 /*       call it and not the others for that team          */
757                 /*       member.                                           */
758
759                 /*
760                  * We cannot assume that SIOCGMIIPHY will also read a
761                  * register; not all network drivers (e.g., e100)
762                  * support that.
763                  */
764
765                 /* Yes, the mii is overlaid on the ifreq.ifr_ifru */
766                 strncpy(ifr.ifr_name, slave_dev->name, IFNAMSIZ);
767                 mii = if_mii(&ifr);
768                 if (IOCTL(slave_dev, &ifr, SIOCGMIIPHY) == 0) {
769                         mii->reg_num = MII_BMSR;
770                         if (IOCTL(slave_dev, &ifr, SIOCGMIIREG) == 0)
771                                 return mii->val_out & BMSR_LSTATUS;
772                 }
773         }
774
775         /*
776          * If reporting, report that either there's no dev->do_ioctl,
777          * or both SIOCGMIIREG and get_link failed (meaning that we
778          * cannot report link status).  If not reporting, pretend
779          * we're ok.
780          */
781         return reporting ? -1 : BMSR_LSTATUS;
782 }
783
784 /*----------------------------- Multicast list ------------------------------*/
785
786 /*
787  * Push the promiscuity flag down to appropriate slaves
788  */
789 static int bond_set_promiscuity(struct bonding *bond, int inc)
790 {
791         int err = 0;
792         if (USES_PRIMARY(bond->params.mode)) {
793                 /* write lock already acquired */
794                 if (bond->curr_active_slave) {
795                         err = dev_set_promiscuity(bond->curr_active_slave->dev,
796                                                   inc);
797                 }
798         } else {
799                 struct slave *slave;
800                 int i;
801                 bond_for_each_slave(bond, slave, i) {
802                         err = dev_set_promiscuity(slave->dev, inc);
803                         if (err)
804                                 return err;
805                 }
806         }
807         return err;
808 }
809
810 /*
811  * Push the allmulti flag down to all slaves
812  */
813 static int bond_set_allmulti(struct bonding *bond, int inc)
814 {
815         int err = 0;
816         if (USES_PRIMARY(bond->params.mode)) {
817                 /* write lock already acquired */
818                 if (bond->curr_active_slave) {
819                         err = dev_set_allmulti(bond->curr_active_slave->dev,
820                                                inc);
821                 }
822         } else {
823                 struct slave *slave;
824                 int i;
825                 bond_for_each_slave(bond, slave, i) {
826                         err = dev_set_allmulti(slave->dev, inc);
827                         if (err)
828                                 return err;
829                 }
830         }
831         return err;
832 }
833
834 /*
835  * Add a Multicast address to slaves
836  * according to mode
837  */
838 static void bond_mc_add(struct bonding *bond, void *addr)
839 {
840         if (USES_PRIMARY(bond->params.mode)) {
841                 /* write lock already acquired */
842                 if (bond->curr_active_slave)
843                         dev_mc_add(bond->curr_active_slave->dev, addr);
844         } else {
845                 struct slave *slave;
846                 int i;
847
848                 bond_for_each_slave(bond, slave, i)
849                         dev_mc_add(slave->dev, addr);
850         }
851 }
852
853 /*
854  * Remove a multicast address from slave
855  * according to mode
856  */
857 static void bond_mc_del(struct bonding *bond, void *addr)
858 {
859         if (USES_PRIMARY(bond->params.mode)) {
860                 /* write lock already acquired */
861                 if (bond->curr_active_slave)
862                         dev_mc_del(bond->curr_active_slave->dev, addr);
863         } else {
864                 struct slave *slave;
865                 int i;
866                 bond_for_each_slave(bond, slave, i) {
867                         dev_mc_del(slave->dev, addr);
868                 }
869         }
870 }
871
872
873 static void __bond_resend_igmp_join_requests(struct net_device *dev)
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(dev);
880         if (in_dev) {
881                 read_lock(&in_dev->mc_list_lock);
882                 for (im = in_dev->mc_list; im; im = im->next)
883                         ip_mc_rejoin_group(im);
884                 read_unlock(&in_dev->mc_list_lock);
885         }
886
887         rcu_read_unlock();
888 }
889
890 /*
891  * Retrieve the list of registered multicast addresses for the bonding
892  * device and retransmit an IGMP JOIN request to the current active
893  * slave.
894  */
895 static void bond_resend_igmp_join_requests(struct bonding *bond)
896 {
897         struct net_device *vlan_dev;
898         struct vlan_entry *vlan;
899
900         read_lock(&bond->lock);
901
902         /* rejoin all groups on bond device */
903         __bond_resend_igmp_join_requests(bond->dev);
904
905         /* rejoin all groups on vlan devices */
906         if (bond->vlgrp) {
907                 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
908                         vlan_dev = vlan_group_get_device(bond->vlgrp,
909                                                          vlan->vlan_id);
910                         if (vlan_dev)
911                                 __bond_resend_igmp_join_requests(vlan_dev);
912                 }
913         }
914
915         if (--bond->igmp_retrans > 0)
916                 queue_delayed_work(bond->wq, &bond->mcast_work, HZ/5);
917
918         read_unlock(&bond->lock);
919 }
920
921 static void bond_resend_igmp_join_requests_delayed(struct work_struct *work)
922 {
923         struct bonding *bond = container_of(work, struct bonding,
924                                                         mcast_work.work);
925         bond_resend_igmp_join_requests(bond);
926 }
927
928 /*
929  * flush all members of flush->mc_list from device dev->mc_list
930  */
931 static void bond_mc_list_flush(struct net_device *bond_dev,
932                                struct net_device *slave_dev)
933 {
934         struct bonding *bond = netdev_priv(bond_dev);
935         struct netdev_hw_addr *ha;
936
937         netdev_for_each_mc_addr(ha, bond_dev)
938                 dev_mc_del(slave_dev, ha->addr);
939
940         if (bond->params.mode == BOND_MODE_8023AD) {
941                 /* del lacpdu mc addr from mc list */
942                 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
943
944                 dev_mc_del(slave_dev, lacpdu_multicast);
945         }
946 }
947
948 /*--------------------------- Active slave change ---------------------------*/
949
950 /*
951  * Update the mc list and multicast-related flags for the new and
952  * old active slaves (if any) according to the multicast mode, and
953  * promiscuous flags unconditionally.
954  */
955 static void bond_mc_swap(struct bonding *bond, struct slave *new_active,
956                          struct slave *old_active)
957 {
958         struct netdev_hw_addr *ha;
959
960         if (!USES_PRIMARY(bond->params.mode))
961                 /* nothing to do -  mc list is already up-to-date on
962                  * all slaves
963                  */
964                 return;
965
966         if (old_active) {
967                 if (bond->dev->flags & IFF_PROMISC)
968                         dev_set_promiscuity(old_active->dev, -1);
969
970                 if (bond->dev->flags & IFF_ALLMULTI)
971                         dev_set_allmulti(old_active->dev, -1);
972
973                 netdev_for_each_mc_addr(ha, bond->dev)
974                         dev_mc_del(old_active->dev, ha->addr);
975         }
976
977         if (new_active) {
978                 /* FIXME: Signal errors upstream. */
979                 if (bond->dev->flags & IFF_PROMISC)
980                         dev_set_promiscuity(new_active->dev, 1);
981
982                 if (bond->dev->flags & IFF_ALLMULTI)
983                         dev_set_allmulti(new_active->dev, 1);
984
985                 netdev_for_each_mc_addr(ha, bond->dev)
986                         dev_mc_add(new_active->dev, ha->addr);
987         }
988 }
989
990 /*
991  * bond_do_fail_over_mac
992  *
993  * Perform special MAC address swapping for fail_over_mac settings
994  *
995  * Called with RTNL, bond->lock for read, curr_slave_lock for write_bh.
996  */
997 static void bond_do_fail_over_mac(struct bonding *bond,
998                                   struct slave *new_active,
999                                   struct slave *old_active)
1000         __releases(&bond->curr_slave_lock)
1001         __releases(&bond->lock)
1002         __acquires(&bond->lock)
1003         __acquires(&bond->curr_slave_lock)
1004 {
1005         u8 tmp_mac[ETH_ALEN];
1006         struct sockaddr saddr;
1007         int rv;
1008
1009         switch (bond->params.fail_over_mac) {
1010         case BOND_FOM_ACTIVE:
1011                 if (new_active)
1012                         memcpy(bond->dev->dev_addr,  new_active->dev->dev_addr,
1013                                new_active->dev->addr_len);
1014                 break;
1015         case BOND_FOM_FOLLOW:
1016                 /*
1017                  * if new_active && old_active, swap them
1018                  * if just old_active, do nothing (going to no active slave)
1019                  * if just new_active, set new_active to bond's MAC
1020                  */
1021                 if (!new_active)
1022                         return;
1023
1024                 write_unlock_bh(&bond->curr_slave_lock);
1025                 read_unlock(&bond->lock);
1026
1027                 if (old_active) {
1028                         memcpy(tmp_mac, new_active->dev->dev_addr, ETH_ALEN);
1029                         memcpy(saddr.sa_data, old_active->dev->dev_addr,
1030                                ETH_ALEN);
1031                         saddr.sa_family = new_active->dev->type;
1032                 } else {
1033                         memcpy(saddr.sa_data, bond->dev->dev_addr, ETH_ALEN);
1034                         saddr.sa_family = bond->dev->type;
1035                 }
1036
1037                 rv = dev_set_mac_address(new_active->dev, &saddr);
1038                 if (rv) {
1039                         pr_err("%s: Error %d setting MAC of slave %s\n",
1040                                bond->dev->name, -rv, new_active->dev->name);
1041                         goto out;
1042                 }
1043
1044                 if (!old_active)
1045                         goto out;
1046
1047                 memcpy(saddr.sa_data, tmp_mac, ETH_ALEN);
1048                 saddr.sa_family = old_active->dev->type;
1049
1050                 rv = dev_set_mac_address(old_active->dev, &saddr);
1051                 if (rv)
1052                         pr_err("%s: Error %d setting MAC of slave %s\n",
1053                                bond->dev->name, -rv, new_active->dev->name);
1054 out:
1055                 read_lock(&bond->lock);
1056                 write_lock_bh(&bond->curr_slave_lock);
1057                 break;
1058         default:
1059                 pr_err("%s: bond_do_fail_over_mac impossible: bad policy %d\n",
1060                        bond->dev->name, bond->params.fail_over_mac);
1061                 break;
1062         }
1063
1064 }
1065
1066 static bool bond_should_change_active(struct bonding *bond)
1067 {
1068         struct slave *prim = bond->primary_slave;
1069         struct slave *curr = bond->curr_active_slave;
1070
1071         if (!prim || !curr || curr->link != BOND_LINK_UP)
1072                 return true;
1073         if (bond->force_primary) {
1074                 bond->force_primary = false;
1075                 return true;
1076         }
1077         if (bond->params.primary_reselect == BOND_PRI_RESELECT_BETTER &&
1078             (prim->speed < curr->speed ||
1079              (prim->speed == curr->speed && prim->duplex <= curr->duplex)))
1080                 return false;
1081         if (bond->params.primary_reselect == BOND_PRI_RESELECT_FAILURE)
1082                 return false;
1083         return true;
1084 }
1085
1086 /**
1087  * find_best_interface - select the best available slave to be the active one
1088  * @bond: our bonding struct
1089  *
1090  * Warning: Caller must hold curr_slave_lock for writing.
1091  */
1092 static struct slave *bond_find_best_slave(struct bonding *bond)
1093 {
1094         struct slave *new_active, *old_active;
1095         struct slave *bestslave = NULL;
1096         int mintime = bond->params.updelay;
1097         int i;
1098
1099         new_active = bond->curr_active_slave;
1100
1101         if (!new_active) { /* there were no active slaves left */
1102                 if (bond->slave_cnt > 0)   /* found one slave */
1103                         new_active = bond->first_slave;
1104                 else
1105                         return NULL; /* still no slave, return NULL */
1106         }
1107
1108         if ((bond->primary_slave) &&
1109             bond->primary_slave->link == BOND_LINK_UP &&
1110             bond_should_change_active(bond)) {
1111                 new_active = bond->primary_slave;
1112         }
1113
1114         /* remember where to stop iterating over the slaves */
1115         old_active = new_active;
1116
1117         bond_for_each_slave_from(bond, new_active, i, old_active) {
1118                 if (new_active->link == BOND_LINK_UP) {
1119                         return new_active;
1120                 } else if (new_active->link == BOND_LINK_BACK &&
1121                            IS_UP(new_active->dev)) {
1122                         /* link up, but waiting for stabilization */
1123                         if (new_active->delay < mintime) {
1124                                 mintime = new_active->delay;
1125                                 bestslave = new_active;
1126                         }
1127                 }
1128         }
1129
1130         return bestslave;
1131 }
1132
1133 /**
1134  * change_active_interface - change the active slave into the specified one
1135  * @bond: our bonding struct
1136  * @new: the new slave to make the active one
1137  *
1138  * Set the new slave to the bond's settings and unset them on the old
1139  * curr_active_slave.
1140  * Setting include flags, mc-list, promiscuity, allmulti, etc.
1141  *
1142  * If @new's link state is %BOND_LINK_BACK we'll set it to %BOND_LINK_UP,
1143  * because it is apparently the best available slave we have, even though its
1144  * updelay hasn't timed out yet.
1145  *
1146  * If new_active is not NULL, caller must hold bond->lock for read and
1147  * curr_slave_lock for write_bh.
1148  */
1149 void bond_change_active_slave(struct bonding *bond, struct slave *new_active)
1150 {
1151         struct slave *old_active = bond->curr_active_slave;
1152
1153         if (old_active == new_active)
1154                 return;
1155
1156         if (new_active) {
1157                 new_active->jiffies = jiffies;
1158
1159                 if (new_active->link == BOND_LINK_BACK) {
1160                         if (USES_PRIMARY(bond->params.mode)) {
1161                                 pr_info("%s: making interface %s the new active one %d ms earlier.\n",
1162                                         bond->dev->name, new_active->dev->name,
1163                                         (bond->params.updelay - new_active->delay) * bond->params.miimon);
1164                         }
1165
1166                         new_active->delay = 0;
1167                         new_active->link = BOND_LINK_UP;
1168
1169                         if (bond->params.mode == BOND_MODE_8023AD)
1170                                 bond_3ad_handle_link_change(new_active, BOND_LINK_UP);
1171
1172                         if (bond_is_lb(bond))
1173                                 bond_alb_handle_link_change(bond, new_active, BOND_LINK_UP);
1174                 } else {
1175                         if (USES_PRIMARY(bond->params.mode)) {
1176                                 pr_info("%s: making interface %s the new active one.\n",
1177                                         bond->dev->name, new_active->dev->name);
1178                         }
1179                 }
1180         }
1181
1182         if (USES_PRIMARY(bond->params.mode))
1183                 bond_mc_swap(bond, new_active, old_active);
1184
1185         if (bond_is_lb(bond)) {
1186                 bond_alb_handle_active_change(bond, new_active);
1187                 if (old_active)
1188                         bond_set_slave_inactive_flags(old_active);
1189                 if (new_active)
1190                         bond_set_slave_active_flags(new_active);
1191         } else {
1192                 bond->curr_active_slave = new_active;
1193         }
1194
1195         if (bond->params.mode == BOND_MODE_ACTIVEBACKUP) {
1196                 if (old_active)
1197                         bond_set_slave_inactive_flags(old_active);
1198
1199                 if (new_active) {
1200                         bond_set_slave_active_flags(new_active);
1201
1202                         if (bond->params.fail_over_mac)
1203                                 bond_do_fail_over_mac(bond, new_active,
1204                                                       old_active);
1205
1206                         bond->send_grat_arp = bond->params.num_grat_arp;
1207                         bond_send_gratuitous_arp(bond);
1208
1209                         bond->send_unsol_na = bond->params.num_unsol_na;
1210                         bond_send_unsolicited_na(bond);
1211
1212                         write_unlock_bh(&bond->curr_slave_lock);
1213                         read_unlock(&bond->lock);
1214
1215                         netdev_bonding_change(bond->dev, NETDEV_BONDING_FAILOVER);
1216
1217                         read_lock(&bond->lock);
1218                         write_lock_bh(&bond->curr_slave_lock);
1219                 }
1220         }
1221
1222         /* resend IGMP joins since active slave has changed or
1223          * all were sent on curr_active_slave */
1224         if ((USES_PRIMARY(bond->params.mode) && new_active) ||
1225             bond->params.mode == BOND_MODE_ROUNDROBIN) {
1226                 bond->igmp_retrans = bond->params.resend_igmp;
1227                 queue_delayed_work(bond->wq, &bond->mcast_work, 0);
1228         }
1229 }
1230
1231 /**
1232  * bond_select_active_slave - select a new active slave, if needed
1233  * @bond: our bonding struct
1234  *
1235  * This functions should be called when one of the following occurs:
1236  * - The old curr_active_slave has been released or lost its link.
1237  * - The primary_slave has got its link back.
1238  * - A slave has got its link back and there's no old curr_active_slave.
1239  *
1240  * Caller must hold bond->lock for read and curr_slave_lock for write_bh.
1241  */
1242 void bond_select_active_slave(struct bonding *bond)
1243 {
1244         struct slave *best_slave;
1245         int rv;
1246
1247         best_slave = bond_find_best_slave(bond);
1248         if (best_slave != bond->curr_active_slave) {
1249                 bond_change_active_slave(bond, best_slave);
1250                 rv = bond_set_carrier(bond);
1251                 if (!rv)
1252                         return;
1253
1254                 if (netif_carrier_ok(bond->dev)) {
1255                         pr_info("%s: first active interface up!\n",
1256                                 bond->dev->name);
1257                 } else {
1258                         pr_info("%s: now running without any active interface !\n",
1259                                 bond->dev->name);
1260                 }
1261         }
1262 }
1263
1264 /*--------------------------- slave list handling ---------------------------*/
1265
1266 /*
1267  * This function attaches the slave to the end of list.
1268  *
1269  * bond->lock held for writing by caller.
1270  */
1271 static void bond_attach_slave(struct bonding *bond, struct slave *new_slave)
1272 {
1273         if (bond->first_slave == NULL) { /* attaching the first slave */
1274                 new_slave->next = new_slave;
1275                 new_slave->prev = new_slave;
1276                 bond->first_slave = new_slave;
1277         } else {
1278                 new_slave->next = bond->first_slave;
1279                 new_slave->prev = bond->first_slave->prev;
1280                 new_slave->next->prev = new_slave;
1281                 new_slave->prev->next = new_slave;
1282         }
1283
1284         bond->slave_cnt++;
1285 }
1286
1287 /*
1288  * This function detaches the slave from the list.
1289  * WARNING: no check is made to verify if the slave effectively
1290  * belongs to <bond>.
1291  * Nothing is freed on return, structures are just unchained.
1292  * If any slave pointer in bond was pointing to <slave>,
1293  * it should be changed by the calling function.
1294  *
1295  * bond->lock held for writing by caller.
1296  */
1297 static void bond_detach_slave(struct bonding *bond, struct slave *slave)
1298 {
1299         if (slave->next)
1300                 slave->next->prev = slave->prev;
1301
1302         if (slave->prev)
1303                 slave->prev->next = slave->next;
1304
1305         if (bond->first_slave == slave) { /* slave is the first slave */
1306                 if (bond->slave_cnt > 1) { /* there are more slave */
1307                         bond->first_slave = slave->next;
1308                 } else {
1309                         bond->first_slave = NULL; /* slave was the last one */
1310                 }
1311         }
1312
1313         slave->next = NULL;
1314         slave->prev = NULL;
1315         bond->slave_cnt--;
1316 }
1317
1318 #ifdef CONFIG_NET_POLL_CONTROLLER
1319 /*
1320  * You must hold read lock on bond->lock before calling this.
1321  */
1322 static bool slaves_support_netpoll(struct net_device *bond_dev)
1323 {
1324         struct bonding *bond = netdev_priv(bond_dev);
1325         struct slave *slave;
1326         int i = 0;
1327         bool ret = true;
1328
1329         bond_for_each_slave(bond, slave, i) {
1330                 if ((slave->dev->priv_flags & IFF_DISABLE_NETPOLL) ||
1331                     !slave->dev->netdev_ops->ndo_poll_controller)
1332                         ret = false;
1333         }
1334         return i != 0 && ret;
1335 }
1336
1337 static void bond_poll_controller(struct net_device *bond_dev)
1338 {
1339         struct bonding *bond = netdev_priv(bond_dev);
1340         struct slave *slave;
1341         int i;
1342
1343         bond_for_each_slave(bond, slave, i) {
1344                 if (slave->dev && IS_UP(slave->dev))
1345                         netpoll_poll_dev(slave->dev);
1346         }
1347 }
1348
1349 static void bond_netpoll_cleanup(struct net_device *bond_dev)
1350 {
1351         struct bonding *bond = netdev_priv(bond_dev);
1352         struct slave *slave;
1353         const struct net_device_ops *ops;
1354         int i;
1355
1356         read_lock(&bond->lock);
1357         bond_dev->npinfo = NULL;
1358         bond_for_each_slave(bond, slave, i) {
1359                 if (slave->dev) {
1360                         ops = slave->dev->netdev_ops;
1361                         if (ops->ndo_netpoll_cleanup)
1362                                 ops->ndo_netpoll_cleanup(slave->dev);
1363                         else
1364                                 slave->dev->npinfo = NULL;
1365                 }
1366         }
1367         read_unlock(&bond->lock);
1368 }
1369
1370 #else
1371
1372 static void bond_netpoll_cleanup(struct net_device *bond_dev)
1373 {
1374 }
1375
1376 #endif
1377
1378 /*---------------------------------- IOCTL ----------------------------------*/
1379
1380 static int bond_sethwaddr(struct net_device *bond_dev,
1381                           struct net_device *slave_dev)
1382 {
1383         pr_debug("bond_dev=%p\n", bond_dev);
1384         pr_debug("slave_dev=%p\n", slave_dev);
1385         pr_debug("slave_dev->addr_len=%d\n", slave_dev->addr_len);
1386         memcpy(bond_dev->dev_addr, slave_dev->dev_addr, slave_dev->addr_len);
1387         return 0;
1388 }
1389
1390 #define BOND_VLAN_FEATURES \
1391         (NETIF_F_VLAN_CHALLENGED | NETIF_F_HW_VLAN_RX | NETIF_F_HW_VLAN_TX | \
1392          NETIF_F_HW_VLAN_FILTER)
1393
1394 /*
1395  * Compute the common dev->feature set available to all slaves.  Some
1396  * feature bits are managed elsewhere, so preserve those feature bits
1397  * on the master device.
1398  */
1399 static int bond_compute_features(struct bonding *bond)
1400 {
1401         struct slave *slave;
1402         struct net_device *bond_dev = bond->dev;
1403         unsigned long features = bond_dev->features;
1404         unsigned long vlan_features = 0;
1405         unsigned short max_hard_header_len = max((u16)ETH_HLEN,
1406                                                 bond_dev->hard_header_len);
1407         int i;
1408
1409         features &= ~(NETIF_F_ALL_CSUM | BOND_VLAN_FEATURES);
1410         features |=  NETIF_F_GSO_MASK | NETIF_F_NO_CSUM;
1411
1412         if (!bond->first_slave)
1413                 goto done;
1414
1415         features &= ~NETIF_F_ONE_FOR_ALL;
1416
1417         vlan_features = bond->first_slave->dev->vlan_features;
1418         bond_for_each_slave(bond, slave, i) {
1419                 features = netdev_increment_features(features,
1420                                                      slave->dev->features,
1421                                                      NETIF_F_ONE_FOR_ALL);
1422                 vlan_features = netdev_increment_features(vlan_features,
1423                                                         slave->dev->vlan_features,
1424                                                         NETIF_F_ONE_FOR_ALL);
1425                 if (slave->dev->hard_header_len > max_hard_header_len)
1426                         max_hard_header_len = slave->dev->hard_header_len;
1427         }
1428
1429 done:
1430         features |= (bond_dev->features & BOND_VLAN_FEATURES);
1431         bond_dev->features = netdev_fix_features(features, NULL);
1432         bond_dev->vlan_features = netdev_fix_features(vlan_features, NULL);
1433         bond_dev->hard_header_len = max_hard_header_len;
1434
1435         return 0;
1436 }
1437
1438 static void bond_setup_by_slave(struct net_device *bond_dev,
1439                                 struct net_device *slave_dev)
1440 {
1441         struct bonding *bond = netdev_priv(bond_dev);
1442
1443         bond_dev->header_ops        = slave_dev->header_ops;
1444
1445         bond_dev->type              = slave_dev->type;
1446         bond_dev->hard_header_len   = slave_dev->hard_header_len;
1447         bond_dev->addr_len          = slave_dev->addr_len;
1448
1449         memcpy(bond_dev->broadcast, slave_dev->broadcast,
1450                 slave_dev->addr_len);
1451         bond->setup_by_slave = 1;
1452 }
1453
1454 /* enslave device <slave> to bond device <master> */
1455 int bond_enslave(struct net_device *bond_dev, struct net_device *slave_dev)
1456 {
1457         struct bonding *bond = netdev_priv(bond_dev);
1458         const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
1459         struct slave *new_slave = NULL;
1460         struct netdev_hw_addr *ha;
1461         struct sockaddr addr;
1462         int link_reporting;
1463         int old_features = bond_dev->features;
1464         int res = 0;
1465
1466         if (!bond->params.use_carrier && slave_dev->ethtool_ops == NULL &&
1467                 slave_ops->ndo_do_ioctl == NULL) {
1468                 pr_warning("%s: Warning: no link monitoring support for %s\n",
1469                            bond_dev->name, slave_dev->name);
1470         }
1471
1472         /* bond must be initialized by bond_open() before enslaving */
1473         if (!(bond_dev->flags & IFF_UP)) {
1474                 pr_warning("%s: master_dev is not up in bond_enslave\n",
1475                            bond_dev->name);
1476         }
1477
1478         /* already enslaved */
1479         if (slave_dev->flags & IFF_SLAVE) {
1480                 pr_debug("Error, Device was already enslaved\n");
1481                 return -EBUSY;
1482         }
1483
1484         /* vlan challenged mutual exclusion */
1485         /* no need to lock since we're protected by rtnl_lock */
1486         if (slave_dev->features & NETIF_F_VLAN_CHALLENGED) {
1487                 pr_debug("%s: NETIF_F_VLAN_CHALLENGED\n", slave_dev->name);
1488                 if (bond->vlgrp) {
1489                         pr_err("%s: Error: cannot enslave VLAN challenged slave %s on VLAN enabled bond %s\n",
1490                                bond_dev->name, slave_dev->name, bond_dev->name);
1491                         return -EPERM;
1492                 } else {
1493                         pr_warning("%s: Warning: enslaved VLAN challenged slave %s. Adding VLANs will be blocked as long as %s is part of bond %s\n",
1494                                    bond_dev->name, slave_dev->name,
1495                                    slave_dev->name, bond_dev->name);
1496                         bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
1497                 }
1498         } else {
1499                 pr_debug("%s: ! NETIF_F_VLAN_CHALLENGED\n", slave_dev->name);
1500                 if (bond->slave_cnt == 0) {
1501                         /* First slave, and it is not VLAN challenged,
1502                          * so remove the block of adding VLANs over the bond.
1503                          */
1504                         bond_dev->features &= ~NETIF_F_VLAN_CHALLENGED;
1505                 }
1506         }
1507
1508         /*
1509          * Old ifenslave binaries are no longer supported.  These can
1510          * be identified with moderate accuracy by the state of the slave:
1511          * the current ifenslave will set the interface down prior to
1512          * enslaving it; the old ifenslave will not.
1513          */
1514         if ((slave_dev->flags & IFF_UP)) {
1515                 pr_err("%s is up. This may be due to an out of date ifenslave.\n",
1516                        slave_dev->name);
1517                 res = -EPERM;
1518                 goto err_undo_flags;
1519         }
1520
1521         /* set bonding device ether type by slave - bonding netdevices are
1522          * created with ether_setup, so when the slave type is not ARPHRD_ETHER
1523          * there is a need to override some of the type dependent attribs/funcs.
1524          *
1525          * bond ether type mutual exclusion - don't allow slaves of dissimilar
1526          * ether type (eg ARPHRD_ETHER and ARPHRD_INFINIBAND) share the same bond
1527          */
1528         if (bond->slave_cnt == 0) {
1529                 if (bond_dev->type != slave_dev->type) {
1530                         pr_debug("%s: change device type from %d to %d\n",
1531                                  bond_dev->name,
1532                                  bond_dev->type, slave_dev->type);
1533
1534                         res = netdev_bonding_change(bond_dev,
1535                                                     NETDEV_PRE_TYPE_CHANGE);
1536                         res = notifier_to_errno(res);
1537                         if (res) {
1538                                 pr_err("%s: refused to change device type\n",
1539                                        bond_dev->name);
1540                                 res = -EBUSY;
1541                                 goto err_undo_flags;
1542                         }
1543
1544                         /* Flush unicast and multicast addresses */
1545                         dev_uc_flush(bond_dev);
1546                         dev_mc_flush(bond_dev);
1547
1548                         if (slave_dev->type != ARPHRD_ETHER)
1549                                 bond_setup_by_slave(bond_dev, slave_dev);
1550                         else
1551                                 ether_setup(bond_dev);
1552
1553                         netdev_bonding_change(bond_dev,
1554                                               NETDEV_POST_TYPE_CHANGE);
1555                 }
1556         } else if (bond_dev->type != slave_dev->type) {
1557                 pr_err("%s ether type (%d) is different from other slaves (%d), can not enslave it.\n",
1558                        slave_dev->name,
1559                        slave_dev->type, bond_dev->type);
1560                 res = -EINVAL;
1561                 goto err_undo_flags;
1562         }
1563
1564         if (slave_ops->ndo_set_mac_address == NULL) {
1565                 if (bond->slave_cnt == 0) {
1566                         pr_warning("%s: Warning: The first slave device specified does not support setting the MAC address. Setting fail_over_mac to active.",
1567                                    bond_dev->name);
1568                         bond->params.fail_over_mac = BOND_FOM_ACTIVE;
1569                 } else if (bond->params.fail_over_mac != BOND_FOM_ACTIVE) {
1570                         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",
1571                                bond_dev->name);
1572                         res = -EOPNOTSUPP;
1573                         goto err_undo_flags;
1574                 }
1575         }
1576
1577         /* If this is the first slave, then we need to set the master's hardware
1578          * address to be the same as the slave's. */
1579         if (bond->slave_cnt == 0)
1580                 memcpy(bond->dev->dev_addr, slave_dev->dev_addr,
1581                        slave_dev->addr_len);
1582
1583
1584         new_slave = kzalloc(sizeof(struct slave), GFP_KERNEL);
1585         if (!new_slave) {
1586                 res = -ENOMEM;
1587                 goto err_undo_flags;
1588         }
1589
1590         /*
1591          * Set the new_slave's queue_id to be zero.  Queue ID mapping
1592          * is set via sysfs or module option if desired.
1593          */
1594         new_slave->queue_id = 0;
1595
1596         /* Save slave's original mtu and then set it to match the bond */
1597         new_slave->original_mtu = slave_dev->mtu;
1598         res = dev_set_mtu(slave_dev, bond->dev->mtu);
1599         if (res) {
1600                 pr_debug("Error %d calling dev_set_mtu\n", res);
1601                 goto err_free;
1602         }
1603
1604         /*
1605          * Save slave's original ("permanent") mac address for modes
1606          * that need it, and for restoring it upon release, and then
1607          * set it to the master's address
1608          */
1609         memcpy(new_slave->perm_hwaddr, slave_dev->dev_addr, ETH_ALEN);
1610
1611         if (!bond->params.fail_over_mac) {
1612                 /*
1613                  * Set slave to master's mac address.  The application already
1614                  * set the master's mac address to that of the first slave
1615                  */
1616                 memcpy(addr.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
1617                 addr.sa_family = slave_dev->type;
1618                 res = dev_set_mac_address(slave_dev, &addr);
1619                 if (res) {
1620                         pr_debug("Error %d calling set_mac_address\n", res);
1621                         goto err_restore_mtu;
1622                 }
1623         }
1624
1625         res = netdev_set_master(slave_dev, bond_dev);
1626         if (res) {
1627                 pr_debug("Error %d calling netdev_set_master\n", res);
1628                 goto err_restore_mac;
1629         }
1630         /* open the slave since the application closed it */
1631         res = dev_open(slave_dev);
1632         if (res) {
1633                 pr_debug("Opening slave %s failed\n", slave_dev->name);
1634                 goto err_unset_master;
1635         }
1636
1637         new_slave->dev = slave_dev;
1638         slave_dev->priv_flags |= IFF_BONDING;
1639
1640         if (bond_is_lb(bond)) {
1641                 /* bond_alb_init_slave() must be called before all other stages since
1642                  * it might fail and we do not want to have to undo everything
1643                  */
1644                 res = bond_alb_init_slave(bond, new_slave);
1645                 if (res)
1646                         goto err_close;
1647         }
1648
1649         /* If the mode USES_PRIMARY, then the new slave gets the
1650          * master's promisc (and mc) settings only if it becomes the
1651          * curr_active_slave, and that is taken care of later when calling
1652          * bond_change_active()
1653          */
1654         if (!USES_PRIMARY(bond->params.mode)) {
1655                 /* set promiscuity level to new slave */
1656                 if (bond_dev->flags & IFF_PROMISC) {
1657                         res = dev_set_promiscuity(slave_dev, 1);
1658                         if (res)
1659                                 goto err_close;
1660                 }
1661
1662                 /* set allmulti level to new slave */
1663                 if (bond_dev->flags & IFF_ALLMULTI) {
1664                         res = dev_set_allmulti(slave_dev, 1);
1665                         if (res)
1666                                 goto err_close;
1667                 }
1668
1669                 netif_addr_lock_bh(bond_dev);
1670                 /* upload master's mc_list to new slave */
1671                 netdev_for_each_mc_addr(ha, bond_dev)
1672                         dev_mc_add(slave_dev, ha->addr);
1673                 netif_addr_unlock_bh(bond_dev);
1674         }
1675
1676         if (bond->params.mode == BOND_MODE_8023AD) {
1677                 /* add lacpdu mc addr to mc list */
1678                 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
1679
1680                 dev_mc_add(slave_dev, lacpdu_multicast);
1681         }
1682
1683         bond_add_vlans_on_slave(bond, slave_dev);
1684
1685         write_lock_bh(&bond->lock);
1686
1687         bond_attach_slave(bond, new_slave);
1688
1689         new_slave->delay = 0;
1690         new_slave->link_failure_count = 0;
1691
1692         bond_compute_features(bond);
1693
1694         write_unlock_bh(&bond->lock);
1695
1696         read_lock(&bond->lock);
1697
1698         new_slave->last_arp_rx = jiffies;
1699
1700         if (bond->params.miimon && !bond->params.use_carrier) {
1701                 link_reporting = bond_check_dev_link(bond, slave_dev, 1);
1702
1703                 if ((link_reporting == -1) && !bond->params.arp_interval) {
1704                         /*
1705                          * miimon is set but a bonded network driver
1706                          * does not support ETHTOOL/MII and
1707                          * arp_interval is not set.  Note: if
1708                          * use_carrier is enabled, we will never go
1709                          * here (because netif_carrier is always
1710                          * supported); thus, we don't need to change
1711                          * the messages for netif_carrier.
1712                          */
1713                         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",
1714                                bond_dev->name, slave_dev->name);
1715                 } else if (link_reporting == -1) {
1716                         /* unable get link status using mii/ethtool */
1717                         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",
1718                                    bond_dev->name, slave_dev->name);
1719                 }
1720         }
1721
1722         /* check for initial state */
1723         if (!bond->params.miimon ||
1724             (bond_check_dev_link(bond, slave_dev, 0) == BMSR_LSTATUS)) {
1725                 if (bond->params.updelay) {
1726                         pr_debug("Initial state of slave_dev is BOND_LINK_BACK\n");
1727                         new_slave->link  = BOND_LINK_BACK;
1728                         new_slave->delay = bond->params.updelay;
1729                 } else {
1730                         pr_debug("Initial state of slave_dev is BOND_LINK_UP\n");
1731                         new_slave->link  = BOND_LINK_UP;
1732                 }
1733                 new_slave->jiffies = jiffies;
1734         } else {
1735                 pr_debug("Initial state of slave_dev is BOND_LINK_DOWN\n");
1736                 new_slave->link  = BOND_LINK_DOWN;
1737         }
1738
1739         if (bond_update_speed_duplex(new_slave) &&
1740             (new_slave->link != BOND_LINK_DOWN)) {
1741                 pr_warning("%s: Warning: failed to get speed and duplex from %s, assumed to be 100Mb/sec and Full.\n",
1742                            bond_dev->name, new_slave->dev->name);
1743
1744                 if (bond->params.mode == BOND_MODE_8023AD) {
1745                         pr_warning("%s: Warning: Operation of 802.3ad mode requires ETHTOOL support in base driver for proper aggregator selection.\n",
1746                                    bond_dev->name);
1747                 }
1748         }
1749
1750         if (USES_PRIMARY(bond->params.mode) && bond->params.primary[0]) {
1751                 /* if there is a primary slave, remember it */
1752                 if (strcmp(bond->params.primary, new_slave->dev->name) == 0) {
1753                         bond->primary_slave = new_slave;
1754                         bond->force_primary = true;
1755                 }
1756         }
1757
1758         write_lock_bh(&bond->curr_slave_lock);
1759
1760         switch (bond->params.mode) {
1761         case BOND_MODE_ACTIVEBACKUP:
1762                 bond_set_slave_inactive_flags(new_slave);
1763                 bond_select_active_slave(bond);
1764                 break;
1765         case BOND_MODE_8023AD:
1766                 /* in 802.3ad mode, the internal mechanism
1767                  * will activate the slaves in the selected
1768                  * aggregator
1769                  */
1770                 bond_set_slave_inactive_flags(new_slave);
1771                 /* if this is the first slave */
1772                 if (bond->slave_cnt == 1) {
1773                         SLAVE_AD_INFO(new_slave).id = 1;
1774                         /* Initialize AD with the number of times that the AD timer is called in 1 second
1775                          * can be called only after the mac address of the bond is set
1776                          */
1777                         bond_3ad_initialize(bond, 1000/AD_TIMER_INTERVAL,
1778                                             bond->params.lacp_fast);
1779                 } else {
1780                         SLAVE_AD_INFO(new_slave).id =
1781                                 SLAVE_AD_INFO(new_slave->prev).id + 1;
1782                 }
1783
1784                 bond_3ad_bind_slave(new_slave);
1785                 break;
1786         case BOND_MODE_TLB:
1787         case BOND_MODE_ALB:
1788                 new_slave->state = BOND_STATE_ACTIVE;
1789                 bond_set_slave_inactive_flags(new_slave);
1790                 bond_select_active_slave(bond);
1791                 break;
1792         default:
1793                 pr_debug("This slave is always active in trunk mode\n");
1794
1795                 /* always active in trunk mode */
1796                 new_slave->state = BOND_STATE_ACTIVE;
1797
1798                 /* In trunking mode there is little meaning to curr_active_slave
1799                  * anyway (it holds no special properties of the bond device),
1800                  * so we can change it without calling change_active_interface()
1801                  */
1802                 if (!bond->curr_active_slave)
1803                         bond->curr_active_slave = new_slave;
1804
1805                 break;
1806         } /* switch(bond_mode) */
1807
1808         write_unlock_bh(&bond->curr_slave_lock);
1809
1810         bond_set_carrier(bond);
1811
1812 #ifdef CONFIG_NET_POLL_CONTROLLER
1813         if (slaves_support_netpoll(bond_dev)) {
1814                 bond_dev->priv_flags &= ~IFF_DISABLE_NETPOLL;
1815                 if (bond_dev->npinfo)
1816                         slave_dev->npinfo = bond_dev->npinfo;
1817         } else if (!(bond_dev->priv_flags & IFF_DISABLE_NETPOLL)) {
1818                 bond_dev->priv_flags |= IFF_DISABLE_NETPOLL;
1819                 pr_info("New slave device %s does not support netpoll\n",
1820                         slave_dev->name);
1821                 pr_info("Disabling netpoll support for %s\n", bond_dev->name);
1822         }
1823 #endif
1824         read_unlock(&bond->lock);
1825
1826         res = bond_create_slave_symlinks(bond_dev, slave_dev);
1827         if (res)
1828                 goto err_close;
1829
1830         pr_info("%s: enslaving %s as a%s interface with a%s link.\n",
1831                 bond_dev->name, slave_dev->name,
1832                 new_slave->state == BOND_STATE_ACTIVE ? "n active" : " backup",
1833                 new_slave->link != BOND_LINK_DOWN ? "n up" : " down");
1834
1835         /* enslave is successful */
1836         return 0;
1837
1838 /* Undo stages on error */
1839 err_close:
1840         dev_close(slave_dev);
1841
1842 err_unset_master:
1843         netdev_set_master(slave_dev, NULL);
1844
1845 err_restore_mac:
1846         if (!bond->params.fail_over_mac) {
1847                 /* XXX TODO - fom follow mode needs to change master's
1848                  * MAC if this slave's MAC is in use by the bond, or at
1849                  * least print a warning.
1850                  */
1851                 memcpy(addr.sa_data, new_slave->perm_hwaddr, ETH_ALEN);
1852                 addr.sa_family = slave_dev->type;
1853                 dev_set_mac_address(slave_dev, &addr);
1854         }
1855
1856 err_restore_mtu:
1857         dev_set_mtu(slave_dev, new_slave->original_mtu);
1858
1859 err_free:
1860         kfree(new_slave);
1861
1862 err_undo_flags:
1863         bond_dev->features = old_features;
1864
1865         return res;
1866 }
1867
1868 /*
1869  * Try to release the slave device <slave> from the bond device <master>
1870  * It is legal to access curr_active_slave without a lock because all the function
1871  * is write-locked.
1872  *
1873  * The rules for slave state should be:
1874  *   for Active/Backup:
1875  *     Active stays on all backups go down
1876  *   for Bonded connections:
1877  *     The first up interface should be left on and all others downed.
1878  */
1879 int bond_release(struct net_device *bond_dev, struct net_device *slave_dev)
1880 {
1881         struct bonding *bond = netdev_priv(bond_dev);
1882         struct slave *slave, *oldcurrent;
1883         struct sockaddr addr;
1884
1885         /* slave is not a slave or master is not master of this slave */
1886         if (!(slave_dev->flags & IFF_SLAVE) ||
1887             (slave_dev->master != bond_dev)) {
1888                 pr_err("%s: Error: cannot release %s.\n",
1889                        bond_dev->name, slave_dev->name);
1890                 return -EINVAL;
1891         }
1892
1893         block_netpoll_tx();
1894         netdev_bonding_change(bond_dev, NETDEV_BONDING_DESLAVE);
1895         write_lock_bh(&bond->lock);
1896
1897         slave = bond_get_slave_by_dev(bond, slave_dev);
1898         if (!slave) {
1899                 /* not a slave of this bond */
1900                 pr_info("%s: %s not enslaved\n",
1901                         bond_dev->name, slave_dev->name);
1902                 write_unlock_bh(&bond->lock);
1903                 unblock_netpoll_tx();
1904                 return -EINVAL;
1905         }
1906
1907         if (!bond->params.fail_over_mac) {
1908                 if (!compare_ether_addr(bond_dev->dev_addr, slave->perm_hwaddr) &&
1909                     bond->slave_cnt > 1)
1910                         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",
1911                                    bond_dev->name, slave_dev->name,
1912                                    slave->perm_hwaddr,
1913                                    bond_dev->name, slave_dev->name);
1914         }
1915
1916         /* Inform AD package of unbinding of slave. */
1917         if (bond->params.mode == BOND_MODE_8023AD) {
1918                 /* must be called before the slave is
1919                  * detached from the list
1920                  */
1921                 bond_3ad_unbind_slave(slave);
1922         }
1923
1924         pr_info("%s: releasing %s interface %s\n",
1925                 bond_dev->name,
1926                 (slave->state == BOND_STATE_ACTIVE) ? "active" : "backup",
1927                 slave_dev->name);
1928
1929         oldcurrent = bond->curr_active_slave;
1930
1931         bond->current_arp_slave = NULL;
1932
1933         /* release the slave from its bond */
1934         bond_detach_slave(bond, slave);
1935
1936         bond_compute_features(bond);
1937
1938         if (bond->primary_slave == slave)
1939                 bond->primary_slave = NULL;
1940
1941         if (oldcurrent == slave)
1942                 bond_change_active_slave(bond, NULL);
1943
1944         if (bond_is_lb(bond)) {
1945                 /* Must be called only after the slave has been
1946                  * detached from the list and the curr_active_slave
1947                  * has been cleared (if our_slave == old_current),
1948                  * but before a new active slave is selected.
1949                  */
1950                 write_unlock_bh(&bond->lock);
1951                 bond_alb_deinit_slave(bond, slave);
1952                 write_lock_bh(&bond->lock);
1953         }
1954
1955         if (oldcurrent == slave) {
1956                 /*
1957                  * Note that we hold RTNL over this sequence, so there
1958                  * is no concern that another slave add/remove event
1959                  * will interfere.
1960                  */
1961                 write_unlock_bh(&bond->lock);
1962                 read_lock(&bond->lock);
1963                 write_lock_bh(&bond->curr_slave_lock);
1964
1965                 bond_select_active_slave(bond);
1966
1967                 write_unlock_bh(&bond->curr_slave_lock);
1968                 read_unlock(&bond->lock);
1969                 write_lock_bh(&bond->lock);
1970         }
1971
1972         if (bond->slave_cnt == 0) {
1973                 bond_set_carrier(bond);
1974
1975                 /* if the last slave was removed, zero the mac address
1976                  * of the master so it will be set by the application
1977                  * to the mac address of the first slave
1978                  */
1979                 memset(bond_dev->dev_addr, 0, bond_dev->addr_len);
1980
1981                 if (!bond->vlgrp) {
1982                         bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
1983                 } else {
1984                         pr_warning("%s: Warning: clearing HW address of %s while it still has VLANs.\n",
1985                                    bond_dev->name, bond_dev->name);
1986                         pr_warning("%s: When re-adding slaves, make sure the bond's HW address matches its VLANs'.\n",
1987                                    bond_dev->name);
1988                 }
1989         } else if ((bond_dev->features & NETIF_F_VLAN_CHALLENGED) &&
1990                    !bond_has_challenged_slaves(bond)) {
1991                 pr_info("%s: last VLAN challenged slave %s left bond %s. VLAN blocking is removed\n",
1992                         bond_dev->name, slave_dev->name, bond_dev->name);
1993                 bond_dev->features &= ~NETIF_F_VLAN_CHALLENGED;
1994         }
1995
1996         write_unlock_bh(&bond->lock);
1997         unblock_netpoll_tx();
1998
1999         /* must do this from outside any spinlocks */
2000         bond_destroy_slave_symlinks(bond_dev, slave_dev);
2001
2002         bond_del_vlans_from_slave(bond, slave_dev);
2003
2004         /* If the mode USES_PRIMARY, then we should only remove its
2005          * promisc and mc settings if it was the curr_active_slave, but that was
2006          * already taken care of above when we detached the slave
2007          */
2008         if (!USES_PRIMARY(bond->params.mode)) {
2009                 /* unset promiscuity level from slave */
2010                 if (bond_dev->flags & IFF_PROMISC)
2011                         dev_set_promiscuity(slave_dev, -1);
2012
2013                 /* unset allmulti level from slave */
2014                 if (bond_dev->flags & IFF_ALLMULTI)
2015                         dev_set_allmulti(slave_dev, -1);
2016
2017                 /* flush master's mc_list from slave */
2018                 netif_addr_lock_bh(bond_dev);
2019                 bond_mc_list_flush(bond_dev, slave_dev);
2020                 netif_addr_unlock_bh(bond_dev);
2021         }
2022
2023         netdev_set_master(slave_dev, NULL);
2024
2025 #ifdef CONFIG_NET_POLL_CONTROLLER
2026         read_lock_bh(&bond->lock);
2027
2028         if (slaves_support_netpoll(bond_dev))
2029                 bond_dev->priv_flags &= ~IFF_DISABLE_NETPOLL;
2030         read_unlock_bh(&bond->lock);
2031         if (slave_dev->netdev_ops->ndo_netpoll_cleanup)
2032                 slave_dev->netdev_ops->ndo_netpoll_cleanup(slave_dev);
2033         else
2034                 slave_dev->npinfo = NULL;
2035 #endif
2036
2037         /* close slave before restoring its mac address */
2038         dev_close(slave_dev);
2039
2040         if (bond->params.fail_over_mac != BOND_FOM_ACTIVE) {
2041                 /* restore original ("permanent") mac address */
2042                 memcpy(addr.sa_data, slave->perm_hwaddr, ETH_ALEN);
2043                 addr.sa_family = slave_dev->type;
2044                 dev_set_mac_address(slave_dev, &addr);
2045         }
2046
2047         dev_set_mtu(slave_dev, slave->original_mtu);
2048
2049         slave_dev->priv_flags &= ~(IFF_MASTER_8023AD | IFF_MASTER_ALB |
2050                                    IFF_SLAVE_INACTIVE | IFF_BONDING |
2051                                    IFF_SLAVE_NEEDARP);
2052
2053         kfree(slave);
2054
2055         return 0;  /* deletion OK */
2056 }
2057
2058 /*
2059 * First release a slave and than destroy the bond if no more slaves are left.
2060 * Must be under rtnl_lock when this function is called.
2061 */
2062 static int  bond_release_and_destroy(struct net_device *bond_dev,
2063                                      struct net_device *slave_dev)
2064 {
2065         struct bonding *bond = netdev_priv(bond_dev);
2066         int ret;
2067
2068         ret = bond_release(bond_dev, slave_dev);
2069         if ((ret == 0) && (bond->slave_cnt == 0)) {
2070                 pr_info("%s: destroying bond %s.\n",
2071                         bond_dev->name, bond_dev->name);
2072                 unregister_netdevice(bond_dev);
2073         }
2074         return ret;
2075 }
2076
2077 /*
2078  * This function releases all slaves.
2079  */
2080 static int bond_release_all(struct net_device *bond_dev)
2081 {
2082         struct bonding *bond = netdev_priv(bond_dev);
2083         struct slave *slave;
2084         struct net_device *slave_dev;
2085         struct sockaddr addr;
2086
2087         write_lock_bh(&bond->lock);
2088
2089         netif_carrier_off(bond_dev);
2090
2091         if (bond->slave_cnt == 0)
2092                 goto out;
2093
2094         bond->current_arp_slave = NULL;
2095         bond->primary_slave = NULL;
2096         bond_change_active_slave(bond, NULL);
2097
2098         while ((slave = bond->first_slave) != NULL) {
2099                 /* Inform AD package of unbinding of slave
2100                  * before slave is detached from the list.
2101                  */
2102                 if (bond->params.mode == BOND_MODE_8023AD)
2103                         bond_3ad_unbind_slave(slave);
2104
2105                 slave_dev = slave->dev;
2106                 bond_detach_slave(bond, slave);
2107
2108                 /* now that the slave is detached, unlock and perform
2109                  * all the undo steps that should not be called from
2110                  * within a lock.
2111                  */
2112                 write_unlock_bh(&bond->lock);
2113
2114                 if (bond_is_lb(bond)) {
2115                         /* must be called only after the slave
2116                          * has been detached from the list
2117                          */
2118                         bond_alb_deinit_slave(bond, slave);
2119                 }
2120
2121                 bond_compute_features(bond);
2122
2123                 bond_destroy_slave_symlinks(bond_dev, slave_dev);
2124                 bond_del_vlans_from_slave(bond, slave_dev);
2125
2126                 /* If the mode USES_PRIMARY, then we should only remove its
2127                  * promisc and mc settings if it was the curr_active_slave, but that was
2128                  * already taken care of above when we detached the slave
2129                  */
2130                 if (!USES_PRIMARY(bond->params.mode)) {
2131                         /* unset promiscuity level from slave */
2132                         if (bond_dev->flags & IFF_PROMISC)
2133                                 dev_set_promiscuity(slave_dev, -1);
2134
2135                         /* unset allmulti level from slave */
2136                         if (bond_dev->flags & IFF_ALLMULTI)
2137                                 dev_set_allmulti(slave_dev, -1);
2138
2139                         /* flush master's mc_list from slave */
2140                         netif_addr_lock_bh(bond_dev);
2141                         bond_mc_list_flush(bond_dev, slave_dev);
2142                         netif_addr_unlock_bh(bond_dev);
2143                 }
2144
2145                 netdev_set_master(slave_dev, NULL);
2146
2147                 /* close slave before restoring its mac address */
2148                 dev_close(slave_dev);
2149
2150                 if (!bond->params.fail_over_mac) {
2151                         /* restore original ("permanent") mac address*/
2152                         memcpy(addr.sa_data, slave->perm_hwaddr, ETH_ALEN);
2153                         addr.sa_family = slave_dev->type;
2154                         dev_set_mac_address(slave_dev, &addr);
2155                 }
2156
2157                 slave_dev->priv_flags &= ~(IFF_MASTER_8023AD | IFF_MASTER_ALB |
2158                                            IFF_SLAVE_INACTIVE);
2159
2160                 kfree(slave);
2161
2162                 /* re-acquire the lock before getting the next slave */
2163                 write_lock_bh(&bond->lock);
2164         }
2165
2166         /* zero the mac address of the master so it will be
2167          * set by the application to the mac address of the
2168          * first slave
2169          */
2170         memset(bond_dev->dev_addr, 0, bond_dev->addr_len);
2171
2172         if (!bond->vlgrp) {
2173                 bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
2174         } else {
2175                 pr_warning("%s: Warning: clearing HW address of %s while it still has VLANs.\n",
2176                            bond_dev->name, bond_dev->name);
2177                 pr_warning("%s: When re-adding slaves, make sure the bond's HW address matches its VLANs'.\n",
2178                            bond_dev->name);
2179         }
2180
2181         pr_info("%s: released all slaves\n", bond_dev->name);
2182
2183 out:
2184         write_unlock_bh(&bond->lock);
2185         return 0;
2186 }
2187
2188 /*
2189  * This function changes the active slave to slave <slave_dev>.
2190  * It returns -EINVAL in the following cases.
2191  *  - <slave_dev> is not found in the list.
2192  *  - There is not active slave now.
2193  *  - <slave_dev> is already active.
2194  *  - The link state of <slave_dev> is not BOND_LINK_UP.
2195  *  - <slave_dev> is not running.
2196  * In these cases, this function does nothing.
2197  * In the other cases, current_slave pointer is changed and 0 is returned.
2198  */
2199 static int bond_ioctl_change_active(struct net_device *bond_dev, struct net_device *slave_dev)
2200 {
2201         struct bonding *bond = netdev_priv(bond_dev);
2202         struct slave *old_active = NULL;
2203         struct slave *new_active = NULL;
2204         int res = 0;
2205
2206         if (!USES_PRIMARY(bond->params.mode))
2207                 return -EINVAL;
2208
2209         /* Verify that master_dev is indeed the master of slave_dev */
2210         if (!(slave_dev->flags & IFF_SLAVE) || (slave_dev->master != bond_dev))
2211                 return -EINVAL;
2212
2213         read_lock(&bond->lock);
2214
2215         read_lock(&bond->curr_slave_lock);
2216         old_active = bond->curr_active_slave;
2217         read_unlock(&bond->curr_slave_lock);
2218
2219         new_active = bond_get_slave_by_dev(bond, slave_dev);
2220
2221         /*
2222          * Changing to the current active: do nothing; return success.
2223          */
2224         if (new_active && (new_active == old_active)) {
2225                 read_unlock(&bond->lock);
2226                 return 0;
2227         }
2228
2229         if ((new_active) &&
2230             (old_active) &&
2231             (new_active->link == BOND_LINK_UP) &&
2232             IS_UP(new_active->dev)) {
2233                 block_netpoll_tx();
2234                 write_lock_bh(&bond->curr_slave_lock);
2235                 bond_change_active_slave(bond, new_active);
2236                 write_unlock_bh(&bond->curr_slave_lock);
2237                 unblock_netpoll_tx();
2238         } else
2239                 res = -EINVAL;
2240
2241         read_unlock(&bond->lock);
2242
2243         return res;
2244 }
2245
2246 static int bond_info_query(struct net_device *bond_dev, struct ifbond *info)
2247 {
2248         struct bonding *bond = netdev_priv(bond_dev);
2249
2250         info->bond_mode = bond->params.mode;
2251         info->miimon = bond->params.miimon;
2252
2253         read_lock(&bond->lock);
2254         info->num_slaves = bond->slave_cnt;
2255         read_unlock(&bond->lock);
2256
2257         return 0;
2258 }
2259
2260 static int bond_slave_info_query(struct net_device *bond_dev, struct ifslave *info)
2261 {
2262         struct bonding *bond = netdev_priv(bond_dev);
2263         struct slave *slave;
2264         int i, res = -ENODEV;
2265
2266         read_lock(&bond->lock);
2267
2268         bond_for_each_slave(bond, slave, i) {
2269                 if (i == (int)info->slave_id) {
2270                         res = 0;
2271                         strcpy(info->slave_name, slave->dev->name);
2272                         info->link = slave->link;
2273                         info->state = slave->state;
2274                         info->link_failure_count = slave->link_failure_count;
2275                         break;
2276                 }
2277         }
2278
2279         read_unlock(&bond->lock);
2280
2281         return res;
2282 }
2283
2284 /*-------------------------------- Monitoring -------------------------------*/
2285
2286
2287 static int bond_miimon_inspect(struct bonding *bond)
2288 {
2289         struct slave *slave;
2290         int i, link_state, commit = 0;
2291         bool ignore_updelay;
2292
2293         ignore_updelay = !bond->curr_active_slave ? true : false;
2294
2295         bond_for_each_slave(bond, slave, i) {
2296                 slave->new_link = BOND_LINK_NOCHANGE;
2297
2298                 link_state = bond_check_dev_link(bond, slave->dev, 0);
2299
2300                 switch (slave->link) {
2301                 case BOND_LINK_UP:
2302                         if (link_state)
2303                                 continue;
2304
2305                         slave->link = BOND_LINK_FAIL;
2306                         slave->delay = bond->params.downdelay;
2307                         if (slave->delay) {
2308                                 pr_info("%s: link status down for %sinterface %s, disabling it in %d ms.\n",
2309                                         bond->dev->name,
2310                                         (bond->params.mode ==
2311                                          BOND_MODE_ACTIVEBACKUP) ?
2312                                         ((slave->state == BOND_STATE_ACTIVE) ?
2313                                          "active " : "backup ") : "",
2314                                         slave->dev->name,
2315                                         bond->params.downdelay * bond->params.miimon);
2316                         }
2317                         /*FALLTHRU*/
2318                 case BOND_LINK_FAIL:
2319                         if (link_state) {
2320                                 /*
2321                                  * recovered before downdelay expired
2322                                  */
2323                                 slave->link = BOND_LINK_UP;
2324                                 slave->jiffies = jiffies;
2325                                 pr_info("%s: link status up again after %d ms for interface %s.\n",
2326                                         bond->dev->name,
2327                                         (bond->params.downdelay - slave->delay) *
2328                                         bond->params.miimon,
2329                                         slave->dev->name);
2330                                 continue;
2331                         }
2332
2333                         if (slave->delay <= 0) {
2334                                 slave->new_link = BOND_LINK_DOWN;
2335                                 commit++;
2336                                 continue;
2337                         }
2338
2339                         slave->delay--;
2340                         break;
2341
2342                 case BOND_LINK_DOWN:
2343                         if (!link_state)
2344                                 continue;
2345
2346                         slave->link = BOND_LINK_BACK;
2347                         slave->delay = bond->params.updelay;
2348
2349                         if (slave->delay) {
2350                                 pr_info("%s: link status up for interface %s, enabling it in %d ms.\n",
2351                                         bond->dev->name, slave->dev->name,
2352                                         ignore_updelay ? 0 :
2353                                         bond->params.updelay *
2354                                         bond->params.miimon);
2355                         }
2356                         /*FALLTHRU*/
2357                 case BOND_LINK_BACK:
2358                         if (!link_state) {
2359                                 slave->link = BOND_LINK_DOWN;
2360                                 pr_info("%s: link status down again after %d ms for interface %s.\n",
2361                                         bond->dev->name,
2362                                         (bond->params.updelay - slave->delay) *
2363                                         bond->params.miimon,
2364                                         slave->dev->name);
2365
2366                                 continue;
2367                         }
2368
2369                         if (ignore_updelay)
2370                                 slave->delay = 0;
2371
2372                         if (slave->delay <= 0) {
2373                                 slave->new_link = BOND_LINK_UP;
2374                                 commit++;
2375                                 ignore_updelay = false;
2376                                 continue;
2377                         }
2378
2379                         slave->delay--;
2380                         break;
2381                 }
2382         }
2383
2384         return commit;
2385 }
2386
2387 static void bond_miimon_commit(struct bonding *bond)
2388 {
2389         struct slave *slave;
2390         int i;
2391
2392         bond_for_each_slave(bond, slave, i) {
2393                 switch (slave->new_link) {
2394                 case BOND_LINK_NOCHANGE:
2395                         continue;
2396
2397                 case BOND_LINK_UP:
2398                         slave->link = BOND_LINK_UP;
2399                         slave->jiffies = jiffies;
2400
2401                         if (bond->params.mode == BOND_MODE_8023AD) {
2402                                 /* prevent it from being the active one */
2403                                 slave->state = BOND_STATE_BACKUP;
2404                         } else if (bond->params.mode != BOND_MODE_ACTIVEBACKUP) {
2405                                 /* make it immediately active */
2406                                 slave->state = BOND_STATE_ACTIVE;
2407                         } else if (slave != bond->primary_slave) {
2408                                 /* prevent it from being the active one */
2409                                 slave->state = BOND_STATE_BACKUP;
2410                         }
2411
2412                         bond_update_speed_duplex(slave);
2413
2414                         pr_info("%s: link status definitely up for interface %s, %d Mbps %s duplex.\n",
2415                                 bond->dev->name, slave->dev->name,
2416                                 slave->speed, slave->duplex ? "full" : "half");
2417
2418                         /* notify ad that the link status has changed */
2419                         if (bond->params.mode == BOND_MODE_8023AD)
2420                                 bond_3ad_handle_link_change(slave, BOND_LINK_UP);
2421
2422                         if (bond_is_lb(bond))
2423                                 bond_alb_handle_link_change(bond, slave,
2424                                                             BOND_LINK_UP);
2425
2426                         if (!bond->curr_active_slave ||
2427                             (slave == bond->primary_slave))
2428                                 goto do_failover;
2429
2430                         continue;
2431
2432                 case BOND_LINK_DOWN:
2433                         if (slave->link_failure_count < UINT_MAX)
2434                                 slave->link_failure_count++;
2435
2436                         slave->link = BOND_LINK_DOWN;
2437
2438                         if (bond->params.mode == BOND_MODE_ACTIVEBACKUP ||
2439                             bond->params.mode == BOND_MODE_8023AD)
2440                                 bond_set_slave_inactive_flags(slave);
2441
2442                         pr_info("%s: link status definitely down for interface %s, disabling it\n",
2443                                 bond->dev->name, slave->dev->name);
2444
2445                         if (bond->params.mode == BOND_MODE_8023AD)
2446                                 bond_3ad_handle_link_change(slave,
2447                                                             BOND_LINK_DOWN);
2448
2449                         if (bond_is_lb(bond))
2450                                 bond_alb_handle_link_change(bond, slave,
2451                                                             BOND_LINK_DOWN);
2452
2453                         if (slave == bond->curr_active_slave)
2454                                 goto do_failover;
2455
2456                         continue;
2457
2458                 default:
2459                         pr_err("%s: invalid new link %d on slave %s\n",
2460                                bond->dev->name, slave->new_link,
2461                                slave->dev->name);
2462                         slave->new_link = BOND_LINK_NOCHANGE;
2463
2464                         continue;
2465                 }
2466
2467 do_failover:
2468                 ASSERT_RTNL();
2469                 block_netpoll_tx();
2470                 write_lock_bh(&bond->curr_slave_lock);
2471                 bond_select_active_slave(bond);
2472                 write_unlock_bh(&bond->curr_slave_lock);
2473                 unblock_netpoll_tx();
2474         }
2475
2476         bond_set_carrier(bond);
2477 }
2478
2479 /*
2480  * bond_mii_monitor
2481  *
2482  * Really a wrapper that splits the mii monitor into two phases: an
2483  * inspection, then (if inspection indicates something needs to be done)
2484  * an acquisition of appropriate locks followed by a commit phase to
2485  * implement whatever link state changes are indicated.
2486  */
2487 void bond_mii_monitor(struct work_struct *work)
2488 {
2489         struct bonding *bond = container_of(work, struct bonding,
2490                                             mii_work.work);
2491
2492         read_lock(&bond->lock);
2493         if (bond->kill_timers)
2494                 goto out;
2495
2496         if (bond->slave_cnt == 0)
2497                 goto re_arm;
2498
2499         if (bond->send_grat_arp) {
2500                 read_lock(&bond->curr_slave_lock);
2501                 bond_send_gratuitous_arp(bond);
2502                 read_unlock(&bond->curr_slave_lock);
2503         }
2504
2505         if (bond->send_unsol_na) {
2506                 read_lock(&bond->curr_slave_lock);
2507                 bond_send_unsolicited_na(bond);
2508                 read_unlock(&bond->curr_slave_lock);
2509         }
2510
2511         if (bond_miimon_inspect(bond)) {
2512                 read_unlock(&bond->lock);
2513                 rtnl_lock();
2514                 read_lock(&bond->lock);
2515
2516                 bond_miimon_commit(bond);
2517
2518                 read_unlock(&bond->lock);
2519                 rtnl_unlock();  /* might sleep, hold no other locks */
2520                 read_lock(&bond->lock);
2521         }
2522
2523 re_arm:
2524         if (bond->params.miimon)
2525                 queue_delayed_work(bond->wq, &bond->mii_work,
2526                                    msecs_to_jiffies(bond->params.miimon));
2527 out:
2528         read_unlock(&bond->lock);
2529 }
2530
2531 static __be32 bond_glean_dev_ip(struct net_device *dev)
2532 {
2533         struct in_device *idev;
2534         struct in_ifaddr *ifa;
2535         __be32 addr = 0;
2536
2537         if (!dev)
2538                 return 0;
2539
2540         rcu_read_lock();
2541         idev = __in_dev_get_rcu(dev);
2542         if (!idev)
2543                 goto out;
2544
2545         ifa = idev->ifa_list;
2546         if (!ifa)
2547                 goto out;
2548
2549         addr = ifa->ifa_local;
2550 out:
2551         rcu_read_unlock();
2552         return addr;
2553 }
2554
2555 static int bond_has_this_ip(struct bonding *bond, __be32 ip)
2556 {
2557         struct vlan_entry *vlan;
2558
2559         if (ip == bond->master_ip)
2560                 return 1;
2561
2562         list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
2563                 if (ip == vlan->vlan_ip)
2564                         return 1;
2565         }
2566
2567         return 0;
2568 }
2569
2570 /*
2571  * We go to the (large) trouble of VLAN tagging ARP frames because
2572  * switches in VLAN mode (especially if ports are configured as
2573  * "native" to a VLAN) might not pass non-tagged frames.
2574  */
2575 static void bond_arp_send(struct net_device *slave_dev, int arp_op, __be32 dest_ip, __be32 src_ip, unsigned short vlan_id)
2576 {
2577         struct sk_buff *skb;
2578
2579         pr_debug("arp %d on slave %s: dst %x src %x vid %d\n", arp_op,
2580                  slave_dev->name, dest_ip, src_ip, vlan_id);
2581
2582         skb = arp_create(arp_op, ETH_P_ARP, dest_ip, slave_dev, src_ip,
2583                          NULL, slave_dev->dev_addr, NULL);
2584
2585         if (!skb) {
2586                 pr_err("ARP packet allocation failed\n");
2587                 return;
2588         }
2589         if (vlan_id) {
2590                 skb = vlan_put_tag(skb, vlan_id);
2591                 if (!skb) {
2592                         pr_err("failed to insert VLAN tag\n");
2593                         return;
2594                 }
2595         }
2596         arp_xmit(skb);
2597 }
2598
2599
2600 static void bond_arp_send_all(struct bonding *bond, struct slave *slave)
2601 {
2602         int i, vlan_id, rv;
2603         __be32 *targets = bond->params.arp_targets;
2604         struct vlan_entry *vlan;
2605         struct net_device *vlan_dev;
2606         struct flowi fl;
2607         struct rtable *rt;
2608
2609         for (i = 0; (i < BOND_MAX_ARP_TARGETS); i++) {
2610                 if (!targets[i])
2611                         break;
2612                 pr_debug("basa: target %x\n", targets[i]);
2613                 if (!bond->vlgrp) {
2614                         pr_debug("basa: empty vlan: arp_send\n");
2615                         bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2616                                       bond->master_ip, 0);
2617                         continue;
2618                 }
2619
2620                 /*
2621                  * If VLANs are configured, we do a route lookup to
2622                  * determine which VLAN interface would be used, so we
2623                  * can tag the ARP with the proper VLAN tag.
2624                  */
2625                 memset(&fl, 0, sizeof(fl));
2626                 fl.fl4_dst = targets[i];
2627                 fl.fl4_tos = RTO_ONLINK;
2628
2629                 rv = ip_route_output_key(dev_net(bond->dev), &rt, &fl);
2630                 if (rv) {
2631                         if (net_ratelimit()) {
2632                                 pr_warning("%s: no route to arp_ip_target %pI4\n",
2633                                            bond->dev->name, &fl.fl4_dst);
2634                         }
2635                         continue;
2636                 }
2637
2638                 /*
2639                  * This target is not on a VLAN
2640                  */
2641                 if (rt->dst.dev == bond->dev) {
2642                         ip_rt_put(rt);
2643                         pr_debug("basa: rtdev == bond->dev: arp_send\n");
2644                         bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2645                                       bond->master_ip, 0);
2646                         continue;
2647                 }
2648
2649                 vlan_id = 0;
2650                 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
2651                         vlan_dev = vlan_group_get_device(bond->vlgrp, vlan->vlan_id);
2652                         if (vlan_dev == rt->dst.dev) {
2653                                 vlan_id = vlan->vlan_id;
2654                                 pr_debug("basa: vlan match on %s %d\n",
2655                                        vlan_dev->name, vlan_id);
2656                                 break;
2657                         }
2658                 }
2659
2660                 if (vlan_id) {
2661                         ip_rt_put(rt);
2662                         bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2663                                       vlan->vlan_ip, vlan_id);
2664                         continue;
2665                 }
2666
2667                 if (net_ratelimit()) {
2668                         pr_warning("%s: no path to arp_ip_target %pI4 via rt.dev %s\n",
2669                                    bond->dev->name, &fl.fl4_dst,
2670                                    rt->dst.dev ? rt->dst.dev->name : "NULL");
2671                 }
2672                 ip_rt_put(rt);
2673         }
2674 }
2675
2676 /*
2677  * Kick out a gratuitous ARP for an IP on the bonding master plus one
2678  * for each VLAN above us.
2679  *
2680  * Caller must hold curr_slave_lock for read or better
2681  */
2682 static void bond_send_gratuitous_arp(struct bonding *bond)
2683 {
2684         struct slave *slave = bond->curr_active_slave;
2685         struct vlan_entry *vlan;
2686         struct net_device *vlan_dev;
2687
2688         pr_debug("bond_send_grat_arp: bond %s slave %s\n",
2689                  bond->dev->name, slave ? slave->dev->name : "NULL");
2690
2691         if (!slave || !bond->send_grat_arp ||
2692             test_bit(__LINK_STATE_LINKWATCH_PENDING, &slave->dev->state))
2693                 return;
2694
2695         bond->send_grat_arp--;
2696
2697         if (bond->master_ip) {
2698                 bond_arp_send(slave->dev, ARPOP_REPLY, bond->master_ip,
2699                                 bond->master_ip, 0);
2700         }
2701
2702         if (!bond->vlgrp)
2703                 return;
2704
2705         list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
2706                 vlan_dev = vlan_group_get_device(bond->vlgrp, vlan->vlan_id);
2707                 if (vlan->vlan_ip) {
2708                         bond_arp_send(slave->dev, ARPOP_REPLY, vlan->vlan_ip,
2709                                       vlan->vlan_ip, vlan->vlan_id);
2710                 }
2711         }
2712 }
2713
2714 static void bond_validate_arp(struct bonding *bond, struct slave *slave, __be32 sip, __be32 tip)
2715 {
2716         int i;
2717         __be32 *targets = bond->params.arp_targets;
2718
2719         for (i = 0; (i < BOND_MAX_ARP_TARGETS) && targets[i]; i++) {
2720                 pr_debug("bva: sip %pI4 tip %pI4 t[%d] %pI4 bhti(tip) %d\n",
2721                          &sip, &tip, i, &targets[i],
2722                          bond_has_this_ip(bond, tip));
2723                 if (sip == targets[i]) {
2724                         if (bond_has_this_ip(bond, tip))
2725                                 slave->last_arp_rx = jiffies;
2726                         return;
2727                 }
2728         }
2729 }
2730
2731 static int bond_arp_rcv(struct sk_buff *skb, struct net_device *dev, struct packet_type *pt, struct net_device *orig_dev)
2732 {
2733         struct arphdr *arp;
2734         struct slave *slave;
2735         struct bonding *bond;
2736         unsigned char *arp_ptr;
2737         __be32 sip, tip;
2738
2739         if (dev->priv_flags & IFF_802_1Q_VLAN) {
2740                 /*
2741                  * When using VLANS and bonding, dev and oriv_dev may be
2742                  * incorrect if the physical interface supports VLAN
2743                  * acceleration.  With this change ARP validation now
2744                  * works for hosts only reachable on the VLAN interface.
2745                  */
2746                 dev = vlan_dev_real_dev(dev);
2747                 orig_dev = dev_get_by_index_rcu(dev_net(skb->dev),skb->skb_iif);
2748         }
2749
2750         if (!(dev->priv_flags & IFF_BONDING) || !(dev->flags & IFF_MASTER))
2751                 goto out;
2752
2753         bond = netdev_priv(dev);
2754         read_lock(&bond->lock);
2755
2756         pr_debug("bond_arp_rcv: bond %s skb->dev %s orig_dev %s\n",
2757                  bond->dev->name, skb->dev ? skb->dev->name : "NULL",
2758                  orig_dev ? orig_dev->name : "NULL");
2759
2760         slave = bond_get_slave_by_dev(bond, orig_dev);
2761         if (!slave || !slave_do_arp_validate(bond, slave))
2762                 goto out_unlock;
2763
2764         if (!pskb_may_pull(skb, arp_hdr_len(dev)))
2765                 goto out_unlock;
2766
2767         arp = arp_hdr(skb);
2768         if (arp->ar_hln != dev->addr_len ||
2769             skb->pkt_type == PACKET_OTHERHOST ||
2770             skb->pkt_type == PACKET_LOOPBACK ||
2771             arp->ar_hrd != htons(ARPHRD_ETHER) ||
2772             arp->ar_pro != htons(ETH_P_IP) ||
2773             arp->ar_pln != 4)
2774                 goto out_unlock;
2775
2776         arp_ptr = (unsigned char *)(arp + 1);
2777         arp_ptr += dev->addr_len;
2778         memcpy(&sip, arp_ptr, 4);
2779         arp_ptr += 4 + dev->addr_len;
2780         memcpy(&tip, arp_ptr, 4);
2781
2782         pr_debug("bond_arp_rcv: %s %s/%d av %d sv %d sip %pI4 tip %pI4\n",
2783                  bond->dev->name, slave->dev->name, slave->state,
2784                  bond->params.arp_validate, slave_do_arp_validate(bond, slave),
2785                  &sip, &tip);
2786
2787         /*
2788          * Backup slaves won't see the ARP reply, but do come through
2789          * here for each ARP probe (so we swap the sip/tip to validate
2790          * the probe).  In a "redundant switch, common router" type of
2791          * configuration, the ARP probe will (hopefully) travel from
2792          * the active, through one switch, the router, then the other
2793          * switch before reaching the backup.
2794          */
2795         if (slave->state == BOND_STATE_ACTIVE)
2796                 bond_validate_arp(bond, slave, sip, tip);
2797         else
2798                 bond_validate_arp(bond, slave, tip, sip);
2799
2800 out_unlock:
2801         read_unlock(&bond->lock);
2802 out:
2803         dev_kfree_skb(skb);
2804         return NET_RX_SUCCESS;
2805 }
2806
2807 /*
2808  * this function is called regularly to monitor each slave's link
2809  * ensuring that traffic is being sent and received when arp monitoring
2810  * is used in load-balancing mode. if the adapter has been dormant, then an
2811  * arp is transmitted to generate traffic. see activebackup_arp_monitor for
2812  * arp monitoring in active backup mode.
2813  */
2814 void bond_loadbalance_arp_mon(struct work_struct *work)
2815 {
2816         struct bonding *bond = container_of(work, struct bonding,
2817                                             arp_work.work);
2818         struct slave *slave, *oldcurrent;
2819         int do_failover = 0;
2820         int delta_in_ticks;
2821         int i;
2822
2823         read_lock(&bond->lock);
2824
2825         delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
2826
2827         if (bond->kill_timers)
2828                 goto out;
2829
2830         if (bond->slave_cnt == 0)
2831                 goto re_arm;
2832
2833         read_lock(&bond->curr_slave_lock);
2834         oldcurrent = bond->curr_active_slave;
2835         read_unlock(&bond->curr_slave_lock);
2836
2837         /* see if any of the previous devices are up now (i.e. they have
2838          * xmt and rcv traffic). the curr_active_slave does not come into
2839          * the picture unless it is null. also, slave->jiffies is not needed
2840          * here because we send an arp on each slave and give a slave as
2841          * long as it needs to get the tx/rx within the delta.
2842          * TODO: what about up/down delay in arp mode? it wasn't here before
2843          *       so it can wait
2844          */
2845         bond_for_each_slave(bond, slave, i) {
2846                 unsigned long trans_start = dev_trans_start(slave->dev);
2847
2848                 if (slave->link != BOND_LINK_UP) {
2849                         if (time_in_range(jiffies,
2850                                 trans_start - delta_in_ticks,
2851                                 trans_start + delta_in_ticks) &&
2852                             time_in_range(jiffies,
2853                                 slave->dev->last_rx - delta_in_ticks,
2854                                 slave->dev->last_rx + delta_in_ticks)) {
2855
2856                                 slave->link  = BOND_LINK_UP;
2857                                 slave->state = BOND_STATE_ACTIVE;
2858
2859                                 /* primary_slave has no meaning in round-robin
2860                                  * mode. the window of a slave being up and
2861                                  * curr_active_slave being null after enslaving
2862                                  * is closed.
2863                                  */
2864                                 if (!oldcurrent) {
2865                                         pr_info("%s: link status definitely up for interface %s, ",
2866                                                 bond->dev->name,
2867                                                 slave->dev->name);
2868                                         do_failover = 1;
2869                                 } else {
2870                                         pr_info("%s: interface %s is now up\n",
2871                                                 bond->dev->name,
2872                                                 slave->dev->name);
2873                                 }
2874                         }
2875                 } else {
2876                         /* slave->link == BOND_LINK_UP */
2877
2878                         /* not all switches will respond to an arp request
2879                          * when the source ip is 0, so don't take the link down
2880                          * if we don't know our ip yet
2881                          */
2882                         if (!time_in_range(jiffies,
2883                                 trans_start - delta_in_ticks,
2884                                 trans_start + 2 * delta_in_ticks) ||
2885                             !time_in_range(jiffies,
2886                                 slave->dev->last_rx - delta_in_ticks,
2887                                 slave->dev->last_rx + 2 * delta_in_ticks)) {
2888
2889                                 slave->link  = BOND_LINK_DOWN;
2890                                 slave->state = BOND_STATE_BACKUP;
2891
2892                                 if (slave->link_failure_count < UINT_MAX)
2893                                         slave->link_failure_count++;
2894
2895                                 pr_info("%s: interface %s is now down.\n",
2896                                         bond->dev->name,
2897                                         slave->dev->name);
2898
2899                                 if (slave == oldcurrent)
2900                                         do_failover = 1;
2901                         }
2902                 }
2903
2904                 /* note: if switch is in round-robin mode, all links
2905                  * must tx arp to ensure all links rx an arp - otherwise
2906                  * links may oscillate or not come up at all; if switch is
2907                  * in something like xor mode, there is nothing we can
2908                  * do - all replies will be rx'ed on same link causing slaves
2909                  * to be unstable during low/no traffic periods
2910                  */
2911                 if (IS_UP(slave->dev))
2912                         bond_arp_send_all(bond, slave);
2913         }
2914
2915         if (do_failover) {
2916                 block_netpoll_tx();
2917                 write_lock_bh(&bond->curr_slave_lock);
2918
2919                 bond_select_active_slave(bond);
2920
2921                 write_unlock_bh(&bond->curr_slave_lock);
2922                 unblock_netpoll_tx();
2923         }
2924
2925 re_arm:
2926         if (bond->params.arp_interval)
2927                 queue_delayed_work(bond->wq, &bond->arp_work, delta_in_ticks);
2928 out:
2929         read_unlock(&bond->lock);
2930 }
2931
2932 /*
2933  * Called to inspect slaves for active-backup mode ARP monitor link state
2934  * changes.  Sets new_link in slaves to specify what action should take
2935  * place for the slave.  Returns 0 if no changes are found, >0 if changes
2936  * to link states must be committed.
2937  *
2938  * Called with bond->lock held for read.
2939  */
2940 static int bond_ab_arp_inspect(struct bonding *bond, int delta_in_ticks)
2941 {
2942         struct slave *slave;
2943         int i, commit = 0;
2944         unsigned long trans_start;
2945
2946         bond_for_each_slave(bond, slave, i) {
2947                 slave->new_link = BOND_LINK_NOCHANGE;
2948
2949                 if (slave->link != BOND_LINK_UP) {
2950                         if (time_in_range(jiffies,
2951                                 slave_last_rx(bond, slave) - delta_in_ticks,
2952                                 slave_last_rx(bond, slave) + delta_in_ticks)) {
2953
2954                                 slave->new_link = BOND_LINK_UP;
2955                                 commit++;
2956                         }
2957
2958                         continue;
2959                 }
2960
2961                 /*
2962                  * Give slaves 2*delta after being enslaved or made
2963                  * active.  This avoids bouncing, as the last receive
2964                  * times need a full ARP monitor cycle to be updated.
2965                  */
2966                 if (time_in_range(jiffies,
2967                                   slave->jiffies - delta_in_ticks,
2968                                   slave->jiffies + 2 * delta_in_ticks))
2969                         continue;
2970
2971                 /*
2972                  * Backup slave is down if:
2973                  * - No current_arp_slave AND
2974                  * - more than 3*delta since last receive AND
2975                  * - the bond has an IP address
2976                  *
2977                  * Note: a non-null current_arp_slave indicates
2978                  * the curr_active_slave went down and we are
2979                  * searching for a new one; under this condition
2980                  * we only take the curr_active_slave down - this
2981                  * gives each slave a chance to tx/rx traffic
2982                  * before being taken out
2983                  */
2984                 if (slave->state == BOND_STATE_BACKUP &&
2985                     !bond->current_arp_slave &&
2986                     !time_in_range(jiffies,
2987                         slave_last_rx(bond, slave) - delta_in_ticks,
2988                         slave_last_rx(bond, slave) + 3 * delta_in_ticks)) {
2989
2990                         slave->new_link = BOND_LINK_DOWN;
2991                         commit++;
2992                 }
2993
2994                 /*
2995                  * Active slave is down if:
2996                  * - more than 2*delta since transmitting OR
2997                  * - (more than 2*delta since receive AND
2998                  *    the bond has an IP address)
2999                  */
3000                 trans_start = dev_trans_start(slave->dev);
3001                 if ((slave->state == BOND_STATE_ACTIVE) &&
3002                     (!time_in_range(jiffies,
3003                         trans_start - delta_in_ticks,
3004                         trans_start + 2 * delta_in_ticks) ||
3005                      !time_in_range(jiffies,
3006                         slave_last_rx(bond, slave) - delta_in_ticks,
3007                         slave_last_rx(bond, slave) + 2 * delta_in_ticks))) {
3008
3009                         slave->new_link = BOND_LINK_DOWN;
3010                         commit++;
3011                 }
3012         }
3013
3014         return commit;
3015 }
3016
3017 /*
3018  * Called to commit link state changes noted by inspection step of
3019  * active-backup mode ARP monitor.
3020  *
3021  * Called with RTNL and bond->lock for read.
3022  */
3023 static void bond_ab_arp_commit(struct bonding *bond, int delta_in_ticks)
3024 {
3025         struct slave *slave;
3026         int i;
3027         unsigned long trans_start;
3028
3029         bond_for_each_slave(bond, slave, i) {
3030                 switch (slave->new_link) {
3031                 case BOND_LINK_NOCHANGE:
3032                         continue;
3033
3034                 case BOND_LINK_UP:
3035                         trans_start = dev_trans_start(slave->dev);
3036                         if ((!bond->curr_active_slave &&
3037                              time_in_range(jiffies,
3038                                            trans_start - delta_in_ticks,
3039                                            trans_start + delta_in_ticks)) ||
3040                             bond->curr_active_slave != slave) {
3041                                 slave->link = BOND_LINK_UP;
3042                                 bond->current_arp_slave = NULL;
3043
3044                                 pr_info("%s: link status definitely up for interface %s.\n",
3045                                         bond->dev->name, slave->dev->name);
3046
3047                                 if (!bond->curr_active_slave ||
3048                                     (slave == bond->primary_slave))
3049                                         goto do_failover;
3050
3051                         }
3052
3053                         continue;
3054
3055                 case BOND_LINK_DOWN:
3056                         if (slave->link_failure_count < UINT_MAX)
3057                                 slave->link_failure_count++;
3058
3059                         slave->link = BOND_LINK_DOWN;
3060                         bond_set_slave_inactive_flags(slave);
3061
3062                         pr_info("%s: link status definitely down for interface %s, disabling it\n",
3063                                 bond->dev->name, slave->dev->name);
3064
3065                         if (slave == bond->curr_active_slave) {
3066                                 bond->current_arp_slave = NULL;
3067                                 goto do_failover;
3068                         }
3069
3070                         continue;
3071
3072                 default:
3073                         pr_err("%s: impossible: new_link %d on slave %s\n",
3074                                bond->dev->name, slave->new_link,
3075                                slave->dev->name);
3076                         continue;
3077                 }
3078
3079 do_failover:
3080                 ASSERT_RTNL();
3081                 block_netpoll_tx();
3082                 write_lock_bh(&bond->curr_slave_lock);
3083                 bond_select_active_slave(bond);
3084                 write_unlock_bh(&bond->curr_slave_lock);
3085                 unblock_netpoll_tx();
3086         }
3087
3088         bond_set_carrier(bond);
3089 }
3090
3091 /*
3092  * Send ARP probes for active-backup mode ARP monitor.
3093  *
3094  * Called with bond->lock held for read.
3095  */
3096 static void bond_ab_arp_probe(struct bonding *bond)
3097 {
3098         struct slave *slave;
3099         int i;
3100
3101         read_lock(&bond->curr_slave_lock);
3102
3103         if (bond->current_arp_slave && bond->curr_active_slave)
3104                 pr_info("PROBE: c_arp %s && cas %s BAD\n",
3105                         bond->current_arp_slave->dev->name,
3106                         bond->curr_active_slave->dev->name);
3107
3108         if (bond->curr_active_slave) {
3109                 bond_arp_send_all(bond, bond->curr_active_slave);
3110                 read_unlock(&bond->curr_slave_lock);
3111                 return;
3112         }
3113
3114         read_unlock(&bond->curr_slave_lock);
3115
3116         /* if we don't have a curr_active_slave, search for the next available
3117          * backup slave from the current_arp_slave and make it the candidate
3118          * for becoming the curr_active_slave
3119          */
3120
3121         if (!bond->current_arp_slave) {
3122                 bond->current_arp_slave = bond->first_slave;
3123                 if (!bond->current_arp_slave)
3124                         return;
3125         }
3126
3127         bond_set_slave_inactive_flags(bond->current_arp_slave);
3128
3129         /* search for next candidate */
3130         bond_for_each_slave_from(bond, slave, i, bond->current_arp_slave->next) {
3131                 if (IS_UP(slave->dev)) {
3132                         slave->link = BOND_LINK_BACK;
3133                         bond_set_slave_active_flags(slave);
3134                         bond_arp_send_all(bond, slave);
3135                         slave->jiffies = jiffies;
3136                         bond->current_arp_slave = slave;
3137                         break;
3138                 }
3139
3140                 /* if the link state is up at this point, we
3141                  * mark it down - this can happen if we have
3142                  * simultaneous link failures and
3143                  * reselect_active_interface doesn't make this
3144                  * one the current slave so it is still marked
3145                  * up when it is actually down
3146                  */
3147                 if (slave->link == BOND_LINK_UP) {
3148                         slave->link = BOND_LINK_DOWN;
3149                         if (slave->link_failure_count < UINT_MAX)
3150                                 slave->link_failure_count++;
3151
3152                         bond_set_slave_inactive_flags(slave);
3153
3154                         pr_info("%s: backup interface %s is now down.\n",
3155                                 bond->dev->name, slave->dev->name);
3156                 }
3157         }
3158 }
3159
3160 void bond_activebackup_arp_mon(struct work_struct *work)
3161 {
3162         struct bonding *bond = container_of(work, struct bonding,
3163                                             arp_work.work);
3164         int delta_in_ticks;
3165
3166         read_lock(&bond->lock);
3167
3168         if (bond->kill_timers)
3169                 goto out;
3170
3171         delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
3172
3173         if (bond->slave_cnt == 0)
3174                 goto re_arm;
3175
3176         if (bond->send_grat_arp) {
3177                 read_lock(&bond->curr_slave_lock);
3178                 bond_send_gratuitous_arp(bond);
3179                 read_unlock(&bond->curr_slave_lock);
3180         }
3181
3182         if (bond->send_unsol_na) {
3183                 read_lock(&bond->curr_slave_lock);
3184                 bond_send_unsolicited_na(bond);
3185                 read_unlock(&bond->curr_slave_lock);
3186         }
3187
3188         if (bond_ab_arp_inspect(bond, delta_in_ticks)) {
3189                 read_unlock(&bond->lock);
3190                 rtnl_lock();
3191                 read_lock(&bond->lock);
3192
3193                 bond_ab_arp_commit(bond, delta_in_ticks);
3194
3195                 read_unlock(&bond->lock);
3196                 rtnl_unlock();
3197                 read_lock(&bond->lock);
3198         }
3199
3200         bond_ab_arp_probe(bond);
3201
3202 re_arm:
3203         if (bond->params.arp_interval)
3204                 queue_delayed_work(bond->wq, &bond->arp_work, delta_in_ticks);
3205 out:
3206         read_unlock(&bond->lock);
3207 }
3208
3209 /*------------------------------ proc/seq_file-------------------------------*/
3210
3211 #ifdef CONFIG_PROC_FS
3212
3213 static void *bond_info_seq_start(struct seq_file *seq, loff_t *pos)
3214         __acquires(&dev_base_lock)
3215         __acquires(&bond->lock)
3216 {
3217         struct bonding *bond = seq->private;
3218         loff_t off = 0;
3219         struct slave *slave;
3220         int i;
3221
3222         /* make sure the bond won't be taken away */
3223         read_lock(&dev_base_lock);
3224         read_lock(&bond->lock);
3225
3226         if (*pos == 0)
3227                 return SEQ_START_TOKEN;
3228
3229         bond_for_each_slave(bond, slave, i) {
3230                 if (++off == *pos)
3231                         return slave;
3232         }
3233
3234         return NULL;
3235 }
3236
3237 static void *bond_info_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3238 {
3239         struct bonding *bond = seq->private;
3240         struct slave *slave = v;
3241
3242         ++*pos;
3243         if (v == SEQ_START_TOKEN)
3244                 return bond->first_slave;
3245
3246         slave = slave->next;
3247
3248         return (slave == bond->first_slave) ? NULL : slave;
3249 }
3250
3251 static void bond_info_seq_stop(struct seq_file *seq, void *v)
3252         __releases(&bond->lock)
3253         __releases(&dev_base_lock)
3254 {
3255         struct bonding *bond = seq->private;
3256
3257         read_unlock(&bond->lock);
3258         read_unlock(&dev_base_lock);
3259 }
3260
3261 static void bond_info_show_master(struct seq_file *seq)
3262 {
3263         struct bonding *bond = seq->private;
3264         struct slave *curr;
3265         int i;
3266
3267         read_lock(&bond->curr_slave_lock);
3268         curr = bond->curr_active_slave;
3269         read_unlock(&bond->curr_slave_lock);
3270
3271         seq_printf(seq, "Bonding Mode: %s",
3272                    bond_mode_name(bond->params.mode));
3273
3274         if (bond->params.mode == BOND_MODE_ACTIVEBACKUP &&
3275             bond->params.fail_over_mac)
3276                 seq_printf(seq, " (fail_over_mac %s)",
3277                    fail_over_mac_tbl[bond->params.fail_over_mac].modename);
3278
3279         seq_printf(seq, "\n");
3280
3281         if (bond->params.mode == BOND_MODE_XOR ||
3282                 bond->params.mode == BOND_MODE_8023AD) {
3283                 seq_printf(seq, "Transmit Hash Policy: %s (%d)\n",
3284                         xmit_hashtype_tbl[bond->params.xmit_policy].modename,
3285                         bond->params.xmit_policy);
3286         }
3287
3288         if (USES_PRIMARY(bond->params.mode)) {
3289                 seq_printf(seq, "Primary Slave: %s",
3290                            (bond->primary_slave) ?
3291                            bond->primary_slave->dev->name : "None");
3292                 if (bond->primary_slave)
3293                         seq_printf(seq, " (primary_reselect %s)",
3294                    pri_reselect_tbl[bond->params.primary_reselect].modename);
3295
3296                 seq_printf(seq, "\nCurrently Active Slave: %s\n",
3297                            (curr) ? curr->dev->name : "None");
3298         }
3299
3300         seq_printf(seq, "MII Status: %s\n", netif_carrier_ok(bond->dev) ?
3301                    "up" : "down");
3302         seq_printf(seq, "MII Polling Interval (ms): %d\n", bond->params.miimon);
3303         seq_printf(seq, "Up Delay (ms): %d\n",
3304                    bond->params.updelay * bond->params.miimon);
3305         seq_printf(seq, "Down Delay (ms): %d\n",
3306                    bond->params.downdelay * bond->params.miimon);
3307
3308
3309         /* ARP information */
3310         if (bond->params.arp_interval > 0) {
3311                 int printed = 0;
3312                 seq_printf(seq, "ARP Polling Interval (ms): %d\n",
3313                                 bond->params.arp_interval);
3314
3315                 seq_printf(seq, "ARP IP target/s (n.n.n.n form):");
3316
3317                 for (i = 0; (i < BOND_MAX_ARP_TARGETS); i++) {
3318                         if (!bond->params.arp_targets[i])
3319                                 break;
3320                         if (printed)
3321                                 seq_printf(seq, ",");
3322                         seq_printf(seq, " %pI4", &bond->params.arp_targets[i]);
3323                         printed = 1;
3324                 }
3325                 seq_printf(seq, "\n");
3326         }
3327
3328         if (bond->params.mode == BOND_MODE_8023AD) {
3329                 struct ad_info ad_info;
3330
3331                 seq_puts(seq, "\n802.3ad info\n");
3332                 seq_printf(seq, "LACP rate: %s\n",
3333                            (bond->params.lacp_fast) ? "fast" : "slow");
3334                 seq_printf(seq, "Aggregator selection policy (ad_select): %s\n",
3335                            ad_select_tbl[bond->params.ad_select].modename);
3336
3337                 if (bond_3ad_get_active_agg_info(bond, &ad_info)) {
3338                         seq_printf(seq, "bond %s has no active aggregator\n",
3339                                    bond->dev->name);
3340                 } else {
3341                         seq_printf(seq, "Active Aggregator Info:\n");
3342
3343                         seq_printf(seq, "\tAggregator ID: %d\n",
3344                                    ad_info.aggregator_id);
3345                         seq_printf(seq, "\tNumber of ports: %d\n",
3346                                    ad_info.ports);
3347                         seq_printf(seq, "\tActor Key: %d\n",
3348                                    ad_info.actor_key);
3349                         seq_printf(seq, "\tPartner Key: %d\n",
3350                                    ad_info.partner_key);
3351                         seq_printf(seq, "\tPartner Mac Address: %pM\n",
3352                                    ad_info.partner_system);
3353                 }
3354         }
3355 }
3356
3357 static void bond_info_show_slave(struct seq_file *seq,
3358                                  const struct slave *slave)
3359 {
3360         struct bonding *bond = seq->private;
3361
3362         seq_printf(seq, "\nSlave Interface: %s\n", slave->dev->name);
3363         seq_printf(seq, "MII Status: %s\n",
3364                    (slave->link == BOND_LINK_UP) ?  "up" : "down");
3365         seq_printf(seq, "Speed: %d Mbps\n", slave->speed);
3366         seq_printf(seq, "Duplex: %s\n", slave->duplex ? "full" : "half");
3367         seq_printf(seq, "Link Failure Count: %u\n",
3368                    slave->link_failure_count);
3369
3370         seq_printf(seq, "Permanent HW addr: %pM\n", slave->perm_hwaddr);
3371
3372         if (bond->params.mode == BOND_MODE_8023AD) {
3373                 const struct aggregator *agg
3374                         = SLAVE_AD_INFO(slave).port.aggregator;
3375
3376                 if (agg)
3377                         seq_printf(seq, "Aggregator ID: %d\n",
3378                                    agg->aggregator_identifier);
3379                 else
3380                         seq_puts(seq, "Aggregator ID: N/A\n");
3381         }
3382         seq_printf(seq, "Slave queue ID: %d\n", slave->queue_id);
3383 }
3384
3385 static int bond_info_seq_show(struct seq_file *seq, void *v)
3386 {
3387         if (v == SEQ_START_TOKEN) {
3388                 seq_printf(seq, "%s\n", version);
3389                 bond_info_show_master(seq);
3390         } else
3391                 bond_info_show_slave(seq, v);
3392
3393         return 0;
3394 }
3395
3396 static const struct seq_operations bond_info_seq_ops = {
3397         .start = bond_info_seq_start,
3398         .next  = bond_info_seq_next,
3399         .stop  = bond_info_seq_stop,
3400         .show  = bond_info_seq_show,
3401 };
3402
3403 static int bond_info_open(struct inode *inode, struct file *file)
3404 {
3405         struct seq_file *seq;
3406         struct proc_dir_entry *proc;
3407         int res;
3408
3409         res = seq_open(file, &bond_info_seq_ops);
3410         if (!res) {
3411                 /* recover the pointer buried in proc_dir_entry data */
3412                 seq = file->private_data;
3413                 proc = PDE(inode);
3414                 seq->private = proc->data;
3415         }
3416
3417         return res;
3418 }
3419
3420 static const struct file_operations bond_info_fops = {
3421         .owner   = THIS_MODULE,
3422         .open    = bond_info_open,
3423         .read    = seq_read,
3424         .llseek  = seq_lseek,
3425         .release = seq_release,
3426 };
3427
3428 static void bond_create_proc_entry(struct bonding *bond)
3429 {
3430         struct net_device *bond_dev = bond->dev;
3431         struct bond_net *bn = net_generic(dev_net(bond_dev), bond_net_id);
3432
3433         if (bn->proc_dir) {
3434                 bond->proc_entry = proc_create_data(bond_dev->name,
3435                                                     S_IRUGO, bn->proc_dir,
3436                                                     &bond_info_fops, bond);
3437                 if (bond->proc_entry == NULL)
3438                         pr_warning("Warning: Cannot create /proc/net/%s/%s\n",
3439                                    DRV_NAME, bond_dev->name);
3440                 else
3441                         memcpy(bond->proc_file_name, bond_dev->name, IFNAMSIZ);
3442         }
3443 }
3444
3445 static void bond_remove_proc_entry(struct bonding *bond)
3446 {
3447         struct net_device *bond_dev = bond->dev;
3448         struct bond_net *bn = net_generic(dev_net(bond_dev), bond_net_id);
3449
3450         if (bn->proc_dir && bond->proc_entry) {
3451                 remove_proc_entry(bond->proc_file_name, bn->proc_dir);
3452                 memset(bond->proc_file_name, 0, IFNAMSIZ);
3453                 bond->proc_entry = NULL;
3454         }
3455 }
3456
3457 /* Create the bonding directory under /proc/net, if doesn't exist yet.
3458  * Caller must hold rtnl_lock.
3459  */
3460 static void __net_init bond_create_proc_dir(struct bond_net *bn)
3461 {
3462         if (!bn->proc_dir) {
3463                 bn->proc_dir = proc_mkdir(DRV_NAME, bn->net->proc_net);
3464                 if (!bn->proc_dir)
3465                         pr_warning("Warning: cannot create /proc/net/%s\n",
3466                                    DRV_NAME);
3467         }
3468 }
3469
3470 /* Destroy the bonding directory under /proc/net, if empty.
3471  * Caller must hold rtnl_lock.
3472  */
3473 static void __net_exit bond_destroy_proc_dir(struct bond_net *bn)
3474 {
3475         if (bn->proc_dir) {
3476                 remove_proc_entry(DRV_NAME, bn->net->proc_net);
3477                 bn->proc_dir = NULL;
3478         }
3479 }
3480
3481 #else /* !CONFIG_PROC_FS */
3482
3483 static void bond_create_proc_entry(struct bonding *bond)
3484 {
3485 }
3486
3487 static void bond_remove_proc_entry(struct bonding *bond)
3488 {
3489 }
3490
3491 static inline void bond_create_proc_dir(struct bond_net *bn)
3492 {
3493 }
3494
3495 static inline void bond_destroy_proc_dir(struct bond_net *bn)
3496 {
3497 }
3498
3499 #endif /* CONFIG_PROC_FS */
3500
3501
3502 /*-------------------------- netdev event handling --------------------------*/
3503
3504 /*
3505  * Change device name
3506  */
3507 static int bond_event_changename(struct bonding *bond)
3508 {
3509         bond_remove_proc_entry(bond);
3510         bond_create_proc_entry(bond);
3511
3512         return NOTIFY_DONE;
3513 }
3514
3515 static int bond_master_netdev_event(unsigned long event,
3516                                     struct net_device *bond_dev)
3517 {
3518         struct bonding *event_bond = netdev_priv(bond_dev);
3519
3520         switch (event) {
3521         case NETDEV_CHANGENAME:
3522                 return bond_event_changename(event_bond);
3523         default:
3524                 break;
3525         }
3526
3527         return NOTIFY_DONE;
3528 }
3529
3530 static int bond_slave_netdev_event(unsigned long event,
3531                                    struct net_device *slave_dev)
3532 {
3533         struct net_device *bond_dev = slave_dev->master;
3534         struct bonding *bond = netdev_priv(bond_dev);
3535
3536         switch (event) {
3537         case NETDEV_UNREGISTER:
3538                 if (bond_dev) {
3539                         if (bond->setup_by_slave)
3540                                 bond_release_and_destroy(bond_dev, slave_dev);
3541                         else
3542                                 bond_release(bond_dev, slave_dev);
3543                 }
3544                 break;
3545         case NETDEV_CHANGE:
3546                 if (bond->params.mode == BOND_MODE_8023AD || bond_is_lb(bond)) {
3547                         struct slave *slave;
3548
3549                         slave = bond_get_slave_by_dev(bond, slave_dev);
3550                         if (slave) {
3551                                 u16 old_speed = slave->speed;
3552                                 u16 old_duplex = slave->duplex;
3553
3554                                 bond_update_speed_duplex(slave);
3555
3556                                 if (bond_is_lb(bond))
3557                                         break;
3558
3559                                 if (old_speed != slave->speed)
3560                                         bond_3ad_adapter_speed_changed(slave);
3561                                 if (old_duplex != slave->duplex)
3562                                         bond_3ad_adapter_duplex_changed(slave);
3563                         }
3564                 }
3565
3566                 break;
3567         case NETDEV_DOWN:
3568                 /*
3569                  * ... Or is it this?
3570                  */
3571                 break;
3572         case NETDEV_CHANGEMTU:
3573                 /*
3574                  * TODO: Should slaves be allowed to
3575                  * independently alter their MTU?  For
3576                  * an active-backup bond, slaves need
3577                  * not be the same type of device, so
3578                  * MTUs may vary.  For other modes,
3579                  * slaves arguably should have the
3580                  * same MTUs. To do this, we'd need to
3581                  * take over the slave's change_mtu
3582                  * function for the duration of their
3583                  * servitude.
3584                  */
3585                 break;
3586         case NETDEV_CHANGENAME:
3587                 /*
3588                  * TODO: handle changing the primary's name
3589                  */
3590                 break;
3591         case NETDEV_FEAT_CHANGE:
3592                 bond_compute_features(bond);
3593                 break;
3594         default:
3595                 break;
3596         }
3597
3598         return NOTIFY_DONE;
3599 }
3600
3601 /*
3602  * bond_netdev_event: handle netdev notifier chain events.
3603  *
3604  * This function receives events for the netdev chain.  The caller (an
3605  * ioctl handler calling blocking_notifier_call_chain) holds the necessary
3606  * locks for us to safely manipulate the slave devices (RTNL lock,
3607  * dev_probe_lock).
3608  */
3609 static int bond_netdev_event(struct notifier_block *this,
3610                              unsigned long event, void *ptr)
3611 {
3612         struct net_device *event_dev = (struct net_device *)ptr;
3613
3614         pr_debug("event_dev: %s, event: %lx\n",
3615                  event_dev ? event_dev->name : "None",
3616                  event);
3617
3618         if (!(event_dev->priv_flags & IFF_BONDING))
3619                 return NOTIFY_DONE;
3620
3621         if (event_dev->flags & IFF_MASTER) {
3622                 pr_debug("IFF_MASTER\n");
3623                 return bond_master_netdev_event(event, event_dev);
3624         }
3625
3626         if (event_dev->flags & IFF_SLAVE) {
3627                 pr_debug("IFF_SLAVE\n");
3628                 return bond_slave_netdev_event(event, event_dev);
3629         }
3630
3631         return NOTIFY_DONE;
3632 }
3633
3634 /*
3635  * bond_inetaddr_event: handle inetaddr notifier chain events.
3636  *
3637  * We keep track of device IPs primarily to use as source addresses in
3638  * ARP monitor probes (rather than spewing out broadcasts all the time).
3639  *
3640  * We track one IP for the main device (if it has one), plus one per VLAN.
3641  */
3642 static int bond_inetaddr_event(struct notifier_block *this, unsigned long event, void *ptr)
3643 {
3644         struct in_ifaddr *ifa = ptr;
3645         struct net_device *vlan_dev, *event_dev = ifa->ifa_dev->dev;
3646         struct bond_net *bn = net_generic(dev_net(event_dev), bond_net_id);
3647         struct bonding *bond;
3648         struct vlan_entry *vlan;
3649
3650         list_for_each_entry(bond, &bn->dev_list, bond_list) {
3651                 if (bond->dev == event_dev) {
3652                         switch (event) {
3653                         case NETDEV_UP:
3654                                 bond->master_ip = ifa->ifa_local;
3655                                 return NOTIFY_OK;
3656                         case NETDEV_DOWN:
3657                                 bond->master_ip = bond_glean_dev_ip(bond->dev);
3658                                 return NOTIFY_OK;
3659                         default:
3660                                 return NOTIFY_DONE;
3661                         }
3662                 }
3663
3664                 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
3665                         if (!bond->vlgrp)
3666                                 continue;
3667                         vlan_dev = vlan_group_get_device(bond->vlgrp, vlan->vlan_id);
3668                         if (vlan_dev == event_dev) {
3669                                 switch (event) {
3670                                 case NETDEV_UP:
3671                                         vlan->vlan_ip = ifa->ifa_local;
3672                                         return NOTIFY_OK;
3673                                 case NETDEV_DOWN:
3674                                         vlan->vlan_ip =
3675                                                 bond_glean_dev_ip(vlan_dev);
3676                                         return NOTIFY_OK;
3677                                 default:
3678                                         return NOTIFY_DONE;
3679                                 }
3680                         }
3681                 }
3682         }
3683         return NOTIFY_DONE;
3684 }
3685
3686 static struct notifier_block bond_netdev_notifier = {
3687         .notifier_call = bond_netdev_event,
3688 };
3689
3690 static struct notifier_block bond_inetaddr_notifier = {
3691         .notifier_call = bond_inetaddr_event,
3692 };
3693
3694 /*-------------------------- Packet type handling ---------------------------*/
3695
3696 /* register to receive lacpdus on a bond */
3697 static void bond_register_lacpdu(struct bonding *bond)
3698 {
3699         struct packet_type *pk_type = &(BOND_AD_INFO(bond).ad_pkt_type);
3700
3701         /* initialize packet type */
3702         pk_type->type = PKT_TYPE_LACPDU;
3703         pk_type->dev = bond->dev;
3704         pk_type->func = bond_3ad_lacpdu_recv;
3705
3706         dev_add_pack(pk_type);
3707 }
3708
3709 /* unregister to receive lacpdus on a bond */
3710 static void bond_unregister_lacpdu(struct bonding *bond)
3711 {
3712         dev_remove_pack(&(BOND_AD_INFO(bond).ad_pkt_type));
3713 }
3714
3715 void bond_register_arp(struct bonding *bond)
3716 {
3717         struct packet_type *pt = &bond->arp_mon_pt;
3718
3719         if (pt->type)
3720                 return;
3721
3722         pt->type = htons(ETH_P_ARP);
3723         pt->dev = bond->dev;
3724         pt->func = bond_arp_rcv;
3725         dev_add_pack(pt);
3726 }
3727
3728 void bond_unregister_arp(struct bonding *bond)
3729 {
3730         struct packet_type *pt = &bond->arp_mon_pt;
3731
3732         dev_remove_pack(pt);
3733         pt->type = 0;
3734 }
3735
3736 /*---------------------------- Hashing Policies -----------------------------*/
3737
3738 /*
3739  * Hash for the output device based upon layer 2 and layer 3 data. If
3740  * the packet is not IP mimic bond_xmit_hash_policy_l2()
3741  */
3742 static int bond_xmit_hash_policy_l23(struct sk_buff *skb, int count)
3743 {
3744         struct ethhdr *data = (struct ethhdr *)skb->data;
3745         struct iphdr *iph = ip_hdr(skb);
3746
3747         if (skb->protocol == htons(ETH_P_IP)) {
3748                 return ((ntohl(iph->saddr ^ iph->daddr) & 0xffff) ^
3749                         (data->h_dest[5] ^ data->h_source[5])) % count;
3750         }
3751
3752         return (data->h_dest[5] ^ data->h_source[5]) % count;
3753 }
3754
3755 /*
3756  * Hash for the output device based upon layer 3 and layer 4 data. If
3757  * the packet is a frag or not TCP or UDP, just use layer 3 data.  If it is
3758  * altogether not IP, mimic bond_xmit_hash_policy_l2()
3759  */
3760 static int bond_xmit_hash_policy_l34(struct sk_buff *skb, int count)
3761 {
3762         struct ethhdr *data = (struct ethhdr *)skb->data;
3763         struct iphdr *iph = ip_hdr(skb);
3764         __be16 *layer4hdr = (__be16 *)((u32 *)iph + iph->ihl);
3765         int layer4_xor = 0;
3766
3767         if (skb->protocol == htons(ETH_P_IP)) {
3768                 if (!(iph->frag_off & htons(IP_MF|IP_OFFSET)) &&
3769                     (iph->protocol == IPPROTO_TCP ||
3770                      iph->protocol == IPPROTO_UDP)) {
3771                         layer4_xor = ntohs((*layer4hdr ^ *(layer4hdr + 1)));
3772                 }
3773                 return (layer4_xor ^
3774                         ((ntohl(iph->saddr ^ iph->daddr)) & 0xffff)) % count;
3775
3776         }
3777
3778         return (data->h_dest[5] ^ data->h_source[5]) % count;
3779 }
3780
3781 /*
3782  * Hash for the output device based upon layer 2 data
3783  */
3784 static int bond_xmit_hash_policy_l2(struct sk_buff *skb, int count)
3785 {
3786         struct ethhdr *data = (struct ethhdr *)skb->data;
3787
3788         return (data->h_dest[5] ^ data->h_source[5]) % count;
3789 }
3790
3791 /*-------------------------- Device entry points ----------------------------*/
3792
3793 static int bond_open(struct net_device *bond_dev)
3794 {
3795         struct bonding *bond = netdev_priv(bond_dev);
3796
3797         bond->kill_timers = 0;
3798
3799         INIT_DELAYED_WORK(&bond->mcast_work, bond_resend_igmp_join_requests_delayed);
3800
3801         if (bond_is_lb(bond)) {
3802                 /* bond_alb_initialize must be called before the timer
3803                  * is started.
3804                  */
3805                 if (bond_alb_initialize(bond, (bond->params.mode == BOND_MODE_ALB))) {
3806                         /* something went wrong - fail the open operation */
3807                         return -ENOMEM;
3808                 }
3809
3810                 INIT_DELAYED_WORK(&bond->alb_work, bond_alb_monitor);
3811                 queue_delayed_work(bond->wq, &bond->alb_work, 0);
3812         }
3813
3814         if (bond->params.miimon) {  /* link check interval, in milliseconds. */
3815                 INIT_DELAYED_WORK(&bond->mii_work, bond_mii_monitor);
3816                 queue_delayed_work(bond->wq, &bond->mii_work, 0);
3817         }
3818
3819         if (bond->params.arp_interval) {  /* arp interval, in milliseconds. */
3820                 if (bond->params.mode == BOND_MODE_ACTIVEBACKUP)
3821                         INIT_DELAYED_WORK(&bond->arp_work,
3822                                           bond_activebackup_arp_mon);
3823                 else
3824                         INIT_DELAYED_WORK(&bond->arp_work,
3825                                           bond_loadbalance_arp_mon);
3826
3827                 queue_delayed_work(bond->wq, &bond->arp_work, 0);
3828                 if (bond->params.arp_validate)
3829                         bond_register_arp(bond);
3830         }
3831
3832         if (bond->params.mode == BOND_MODE_8023AD) {
3833                 INIT_DELAYED_WORK(&bond->ad_work, bond_3ad_state_machine_handler);
3834                 queue_delayed_work(bond->wq, &bond->ad_work, 0);
3835                 /* register to receive LACPDUs */
3836                 bond_register_lacpdu(bond);
3837                 bond_3ad_initiate_agg_selection(bond, 1);
3838         }
3839
3840         return 0;
3841 }
3842
3843 static int bond_close(struct net_device *bond_dev)
3844 {
3845         struct bonding *bond = netdev_priv(bond_dev);
3846
3847         if (bond->params.mode == BOND_MODE_8023AD) {
3848                 /* Unregister the receive of LACPDUs */
3849                 bond_unregister_lacpdu(bond);
3850         }
3851
3852         if (bond->params.arp_validate)
3853                 bond_unregister_arp(bond);
3854
3855         write_lock_bh(&bond->lock);
3856
3857         bond->send_grat_arp = 0;
3858         bond->send_unsol_na = 0;
3859
3860         /* signal timers not to re-arm */
3861         bond->kill_timers = 1;
3862
3863         write_unlock_bh(&bond->lock);
3864
3865         if (bond->params.miimon) {  /* link check interval, in milliseconds. */
3866                 cancel_delayed_work(&bond->mii_work);
3867         }
3868
3869         if (bond->params.arp_interval) {  /* arp interval, in milliseconds. */
3870                 cancel_delayed_work(&bond->arp_work);
3871         }
3872
3873         switch (bond->params.mode) {
3874         case BOND_MODE_8023AD:
3875                 cancel_delayed_work(&bond->ad_work);
3876                 break;
3877         case BOND_MODE_TLB:
3878         case BOND_MODE_ALB:
3879                 cancel_delayed_work(&bond->alb_work);
3880                 break;
3881         default:
3882                 break;
3883         }
3884
3885         if (delayed_work_pending(&bond->mcast_work))
3886                 cancel_delayed_work(&bond->mcast_work);
3887
3888         if (bond_is_lb(bond)) {
3889                 /* Must be called only after all
3890                  * slaves have been released
3891                  */
3892                 bond_alb_deinitialize(bond);
3893         }
3894
3895         return 0;
3896 }
3897
3898 static struct rtnl_link_stats64 *bond_get_stats(struct net_device *bond_dev,
3899                                                 struct rtnl_link_stats64 *stats)
3900 {
3901         struct bonding *bond = netdev_priv(bond_dev);
3902         struct rtnl_link_stats64 temp;
3903         struct slave *slave;
3904         int i;
3905
3906         memset(stats, 0, sizeof(*stats));
3907
3908         read_lock_bh(&bond->lock);
3909
3910         bond_for_each_slave(bond, slave, i) {
3911                 const struct rtnl_link_stats64 *sstats =
3912                         dev_get_stats(slave->dev, &temp);
3913
3914                 stats->rx_packets += sstats->rx_packets;
3915                 stats->rx_bytes += sstats->rx_bytes;
3916                 stats->rx_errors += sstats->rx_errors;
3917                 stats->rx_dropped += sstats->rx_dropped;
3918
3919                 stats->tx_packets += sstats->tx_packets;
3920                 stats->tx_bytes += sstats->tx_bytes;
3921                 stats->tx_errors += sstats->tx_errors;
3922                 stats->tx_dropped += sstats->tx_dropped;
3923
3924                 stats->multicast += sstats->multicast;
3925                 stats->collisions += sstats->collisions;
3926
3927                 stats->rx_length_errors += sstats->rx_length_errors;
3928                 stats->rx_over_errors += sstats->rx_over_errors;
3929                 stats->rx_crc_errors += sstats->rx_crc_errors;
3930                 stats->rx_frame_errors += sstats->rx_frame_errors;
3931                 stats->rx_fifo_errors += sstats->rx_fifo_errors;
3932                 stats->rx_missed_errors += sstats->rx_missed_errors;
3933
3934                 stats->tx_aborted_errors += sstats->tx_aborted_errors;
3935                 stats->tx_carrier_errors += sstats->tx_carrier_errors;
3936                 stats->tx_fifo_errors += sstats->tx_fifo_errors;
3937                 stats->tx_heartbeat_errors += sstats->tx_heartbeat_errors;
3938                 stats->tx_window_errors += sstats->tx_window_errors;
3939         }
3940
3941         read_unlock_bh(&bond->lock);
3942
3943         return stats;
3944 }
3945
3946 static int bond_do_ioctl(struct net_device *bond_dev, struct ifreq *ifr, int cmd)
3947 {
3948         struct net_device *slave_dev = NULL;
3949         struct ifbond k_binfo;
3950         struct ifbond __user *u_binfo = NULL;
3951         struct ifslave k_sinfo;
3952         struct ifslave __user *u_sinfo = NULL;
3953         struct mii_ioctl_data *mii = NULL;
3954         int res = 0;
3955
3956         pr_debug("bond_ioctl: master=%s, cmd=%d\n", bond_dev->name, cmd);
3957
3958         switch (cmd) {
3959         case SIOCGMIIPHY:
3960                 mii = if_mii(ifr);
3961                 if (!mii)
3962                         return -EINVAL;
3963
3964                 mii->phy_id = 0;
3965                 /* Fall Through */
3966         case SIOCGMIIREG:
3967                 /*
3968                  * We do this again just in case we were called by SIOCGMIIREG
3969                  * instead of SIOCGMIIPHY.
3970                  */
3971                 mii = if_mii(ifr);
3972                 if (!mii)
3973                         return -EINVAL;
3974
3975
3976                 if (mii->reg_num == 1) {
3977                         struct bonding *bond = netdev_priv(bond_dev);
3978                         mii->val_out = 0;
3979                         read_lock(&bond->lock);
3980                         read_lock(&bond->curr_slave_lock);
3981                         if (netif_carrier_ok(bond->dev))
3982                                 mii->val_out = BMSR_LSTATUS;
3983
3984                         read_unlock(&bond->curr_slave_lock);
3985                         read_unlock(&bond->lock);
3986                 }
3987
3988                 return 0;
3989         case BOND_INFO_QUERY_OLD:
3990         case SIOCBONDINFOQUERY:
3991                 u_binfo = (struct ifbond __user *)ifr->ifr_data;
3992
3993                 if (copy_from_user(&k_binfo, u_binfo, sizeof(ifbond)))
3994                         return -EFAULT;
3995
3996                 res = bond_info_query(bond_dev, &k_binfo);
3997                 if (res == 0 &&
3998                     copy_to_user(u_binfo, &k_binfo, sizeof(ifbond)))
3999                         return -EFAULT;
4000
4001                 return res;
4002         case BOND_SLAVE_INFO_QUERY_OLD:
4003         case SIOCBONDSLAVEINFOQUERY:
4004                 u_sinfo = (struct ifslave __user *)ifr->ifr_data;
4005
4006                 if (copy_from_user(&k_sinfo, u_sinfo, sizeof(ifslave)))
4007                         return -EFAULT;
4008
4009                 res = bond_slave_info_query(bond_dev, &k_sinfo);
4010                 if (res == 0 &&
4011                     copy_to_user(u_sinfo, &k_sinfo, sizeof(ifslave)))
4012                         return -EFAULT;
4013
4014                 return res;
4015         default:
4016                 /* Go on */
4017                 break;
4018         }
4019
4020         if (!capable(CAP_NET_ADMIN))
4021                 return -EPERM;
4022
4023         slave_dev = dev_get_by_name(dev_net(bond_dev), ifr->ifr_slave);
4024
4025         pr_debug("slave_dev=%p:\n", slave_dev);
4026
4027         if (!slave_dev)
4028                 res = -ENODEV;
4029         else {
4030                 pr_debug("slave_dev->name=%s:\n", slave_dev->name);
4031                 switch (cmd) {
4032                 case BOND_ENSLAVE_OLD:
4033                 case SIOCBONDENSLAVE:
4034                         res = bond_enslave(bond_dev, slave_dev);
4035                         break;
4036                 case BOND_RELEASE_OLD:
4037                 case SIOCBONDRELEASE:
4038                         res = bond_release(bond_dev, slave_dev);
4039                         break;
4040                 case BOND_SETHWADDR_OLD:
4041                 case SIOCBONDSETHWADDR:
4042                         res = bond_sethwaddr(bond_dev, slave_dev);
4043                         break;
4044                 case BOND_CHANGE_ACTIVE_OLD:
4045                 case SIOCBONDCHANGEACTIVE:
4046                         res = bond_ioctl_change_active(bond_dev, slave_dev);
4047                         break;
4048                 default:
4049                         res = -EOPNOTSUPP;
4050                 }
4051
4052                 dev_put(slave_dev);
4053         }
4054
4055         return res;
4056 }
4057
4058 static bool bond_addr_in_mc_list(unsigned char *addr,
4059                                  struct netdev_hw_addr_list *list,
4060                                  int addrlen)
4061 {
4062         struct netdev_hw_addr *ha;
4063
4064         netdev_hw_addr_list_for_each(ha, list)
4065                 if (!memcmp(ha->addr, addr, addrlen))
4066                         return true;
4067
4068         return false;
4069 }
4070
4071 static void bond_set_multicast_list(struct net_device *bond_dev)
4072 {
4073         struct bonding *bond = netdev_priv(bond_dev);
4074         struct netdev_hw_addr *ha;
4075         bool found;
4076
4077         /*
4078          * Do promisc before checking multicast_mode
4079          */
4080         if ((bond_dev->flags & IFF_PROMISC) && !(bond->flags & IFF_PROMISC))
4081                 /*
4082                  * FIXME: Need to handle the error when one of the multi-slaves
4083                  * encounters error.
4084                  */
4085                 bond_set_promiscuity(bond, 1);
4086
4087
4088         if (!(bond_dev->flags & IFF_PROMISC) && (bond->flags & IFF_PROMISC))
4089                 bond_set_promiscuity(bond, -1);
4090
4091
4092         /* set allmulti flag to slaves */
4093         if ((bond_dev->flags & IFF_ALLMULTI) && !(bond->flags & IFF_ALLMULTI))
4094                 /*
4095                  * FIXME: Need to handle the error when one of the multi-slaves
4096                  * encounters error.
4097                  */
4098                 bond_set_allmulti(bond, 1);
4099
4100
4101         if (!(bond_dev->flags & IFF_ALLMULTI) && (bond->flags & IFF_ALLMULTI))
4102                 bond_set_allmulti(bond, -1);
4103
4104
4105         read_lock(&bond->lock);
4106
4107         bond->flags = bond_dev->flags;
4108
4109         /* looking for addresses to add to slaves' mc list */
4110         netdev_for_each_mc_addr(ha, bond_dev) {
4111                 found = bond_addr_in_mc_list(ha->addr, &bond->mc_list,
4112                                              bond_dev->addr_len);
4113                 if (!found)
4114                         bond_mc_add(bond, ha->addr);
4115         }
4116
4117         /* looking for addresses to delete from slaves' list */
4118         netdev_hw_addr_list_for_each(ha, &bond->mc_list) {
4119                 found = bond_addr_in_mc_list(ha->addr, &bond_dev->mc,
4120                                              bond_dev->addr_len);
4121                 if (!found)
4122                         bond_mc_del(bond, ha->addr);
4123         }
4124
4125         /* save master's multicast list */
4126         __hw_addr_flush(&bond->mc_list);
4127         __hw_addr_add_multiple(&bond->mc_list, &bond_dev->mc,
4128                                bond_dev->addr_len, NETDEV_HW_ADDR_T_MULTICAST);
4129
4130         read_unlock(&bond->lock);
4131 }
4132
4133 static int bond_neigh_setup(struct net_device *dev, struct neigh_parms *parms)
4134 {
4135         struct bonding *bond = netdev_priv(dev);
4136         struct slave *slave = bond->first_slave;
4137
4138         if (slave) {
4139                 const struct net_device_ops *slave_ops
4140                         = slave->dev->netdev_ops;
4141                 if (slave_ops->ndo_neigh_setup)
4142                         return slave_ops->ndo_neigh_setup(slave->dev, parms);
4143         }
4144         return 0;
4145 }
4146
4147 /*
4148  * Change the MTU of all of a master's slaves to match the master
4149  */
4150 static int bond_change_mtu(struct net_device *bond_dev, int new_mtu)
4151 {
4152         struct bonding *bond = netdev_priv(bond_dev);
4153         struct slave *slave, *stop_at;
4154         int res = 0;
4155         int i;
4156
4157         pr_debug("bond=%p, name=%s, new_mtu=%d\n", bond,
4158                  (bond_dev ? bond_dev->name : "None"), new_mtu);
4159
4160         /* Can't hold bond->lock with bh disabled here since
4161          * some base drivers panic. On the other hand we can't
4162          * hold bond->lock without bh disabled because we'll
4163          * deadlock. The only solution is to rely on the fact
4164          * that we're under rtnl_lock here, and the slaves
4165          * list won't change. This doesn't solve the problem
4166          * of setting the slave's MTU while it is
4167          * transmitting, but the assumption is that the base
4168          * driver can handle that.
4169          *
4170          * TODO: figure out a way to safely iterate the slaves
4171          * list, but without holding a lock around the actual
4172          * call to the base driver.
4173          */
4174
4175         bond_for_each_slave(bond, slave, i) {
4176                 pr_debug("s %p s->p %p c_m %p\n",
4177                          slave,
4178                          slave->prev,
4179                          slave->dev->netdev_ops->ndo_change_mtu);
4180
4181                 res = dev_set_mtu(slave->dev, new_mtu);
4182
4183                 if (res) {
4184                         /* If we failed to set the slave's mtu to the new value
4185                          * we must abort the operation even in ACTIVE_BACKUP
4186                          * mode, because if we allow the backup slaves to have
4187                          * different mtu values than the active slave we'll
4188                          * need to change their mtu when doing a failover. That
4189                          * means changing their mtu from timer context, which
4190                          * is probably not a good idea.
4191                          */
4192                         pr_debug("err %d %s\n", res, slave->dev->name);
4193                         goto unwind;
4194                 }
4195         }
4196
4197         bond_dev->mtu = new_mtu;
4198
4199         return 0;
4200
4201 unwind:
4202         /* unwind from head to the slave that failed */
4203         stop_at = slave;
4204         bond_for_each_slave_from_to(bond, slave, i, bond->first_slave, stop_at) {
4205                 int tmp_res;
4206
4207                 tmp_res = dev_set_mtu(slave->dev, bond_dev->mtu);
4208                 if (tmp_res) {
4209                         pr_debug("unwind err %d dev %s\n",
4210                                  tmp_res, slave->dev->name);
4211                 }
4212         }
4213
4214         return res;
4215 }
4216
4217 /*
4218  * Change HW address
4219  *
4220  * Note that many devices must be down to change the HW address, and
4221  * downing the master releases all slaves.  We can make bonds full of
4222  * bonding devices to test this, however.
4223  */
4224 static int bond_set_mac_address(struct net_device *bond_dev, void *addr)
4225 {
4226         struct bonding *bond = netdev_priv(bond_dev);
4227         struct sockaddr *sa = addr, tmp_sa;
4228         struct slave *slave, *stop_at;
4229         int res = 0;
4230         int i;
4231
4232         if (bond->params.mode == BOND_MODE_ALB)
4233                 return bond_alb_set_mac_address(bond_dev, addr);
4234
4235
4236         pr_debug("bond=%p, name=%s\n",
4237                  bond, bond_dev ? bond_dev->name : "None");
4238
4239         /*
4240          * If fail_over_mac is set to active, do nothing and return
4241          * success.  Returning an error causes ifenslave to fail.
4242          */
4243         if (bond->params.fail_over_mac == BOND_FOM_ACTIVE)
4244                 return 0;
4245
4246         if (!is_valid_ether_addr(sa->sa_data))
4247                 return -EADDRNOTAVAIL;
4248
4249         /* Can't hold bond->lock with bh disabled here since
4250          * some base drivers panic. On the other hand we can't
4251          * hold bond->lock without bh disabled because we'll
4252          * deadlock. The only solution is to rely on the fact
4253          * that we're under rtnl_lock here, and the slaves
4254          * list won't change. This doesn't solve the problem
4255          * of setting the slave's hw address while it is
4256          * transmitting, but the assumption is that the base
4257          * driver can handle that.
4258          *
4259          * TODO: figure out a way to safely iterate the slaves
4260          * list, but without holding a lock around the actual
4261          * call to the base driver.
4262          */
4263
4264         bond_for_each_slave(bond, slave, i) {
4265                 const struct net_device_ops *slave_ops = slave->dev->netdev_ops;
4266                 pr_debug("slave %p %s\n", slave, slave->dev->name);
4267
4268                 if (slave_ops->ndo_set_mac_address == NULL) {
4269                         res = -EOPNOTSUPP;
4270                         pr_debug("EOPNOTSUPP %s\n", slave->dev->name);
4271                         goto unwind;
4272                 }
4273
4274                 res = dev_set_mac_address(slave->dev, addr);
4275                 if (res) {
4276                         /* TODO: consider downing the slave
4277                          * and retry ?
4278                          * User should expect communications
4279                          * breakage anyway until ARP finish
4280                          * updating, so...
4281                          */
4282                         pr_debug("err %d %s\n", res, slave->dev->name);
4283                         goto unwind;
4284                 }
4285         }
4286
4287         /* success */
4288         memcpy(bond_dev->dev_addr, sa->sa_data, bond_dev->addr_len);
4289         return 0;
4290
4291 unwind:
4292         memcpy(tmp_sa.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
4293         tmp_sa.sa_family = bond_dev->type;
4294
4295         /* unwind from head to the slave that failed */
4296         stop_at = slave;
4297         bond_for_each_slave_from_to(bond, slave, i, bond->first_slave, stop_at) {
4298                 int tmp_res;
4299
4300                 tmp_res = dev_set_mac_address(slave->dev, &tmp_sa);
4301                 if (tmp_res) {
4302                         pr_debug("unwind err %d dev %s\n",
4303                                  tmp_res, slave->dev->name);
4304                 }
4305         }
4306
4307         return res;
4308 }
4309
4310 static int bond_xmit_roundrobin(struct sk_buff *skb, struct net_device *bond_dev)
4311 {
4312         struct bonding *bond = netdev_priv(bond_dev);
4313         struct slave *slave, *start_at;
4314         int i, slave_no, res = 1;
4315         struct iphdr *iph = ip_hdr(skb);
4316
4317         read_lock(&bond->lock);
4318
4319         if (!BOND_IS_OK(bond))
4320                 goto out;
4321         /*
4322          * Start with the curr_active_slave that joined the bond as the
4323          * default for sending IGMP traffic.  For failover purposes one
4324          * needs to maintain some consistency for the interface that will
4325          * send the join/membership reports.  The curr_active_slave found
4326          * will send all of this type of traffic.
4327          */
4328         if ((iph->protocol == IPPROTO_IGMP) &&
4329             (skb->protocol == htons(ETH_P_IP))) {
4330
4331                 read_lock(&bond->curr_slave_lock);
4332                 slave = bond->curr_active_slave;
4333                 read_unlock(&bond->curr_slave_lock);
4334
4335                 if (!slave)
4336                         goto out;
4337         } else {
4338                 /*
4339                  * Concurrent TX may collide on rr_tx_counter; we accept
4340                  * that as being rare enough not to justify using an
4341                  * atomic op here.
4342                  */
4343                 slave_no = bond->rr_tx_counter++ % bond->slave_cnt;
4344
4345                 bond_for_each_slave(bond, slave, i) {
4346                         slave_no--;
4347                         if (slave_no < 0)
4348                                 break;
4349                 }
4350         }
4351
4352         start_at = slave;
4353         bond_for_each_slave_from(bond, slave, i, start_at) {
4354                 if (IS_UP(slave->dev) &&
4355                     (slave->link == BOND_LINK_UP) &&
4356                     (slave->state == BOND_STATE_ACTIVE)) {
4357                         res = bond_dev_queue_xmit(bond, skb, slave->dev);
4358                         break;
4359                 }
4360         }
4361
4362 out:
4363         if (res) {
4364                 /* no suitable interface, frame not sent */
4365                 dev_kfree_skb(skb);
4366         }
4367         read_unlock(&bond->lock);
4368         return NETDEV_TX_OK;
4369 }
4370
4371
4372 /*
4373  * in active-backup mode, we know that bond->curr_active_slave is always valid if
4374  * the bond has a usable interface.
4375  */
4376 static int bond_xmit_activebackup(struct sk_buff *skb, struct net_device *bond_dev)
4377 {
4378         struct bonding *bond = netdev_priv(bond_dev);
4379         int res = 1;
4380
4381         read_lock(&bond->lock);
4382         read_lock(&bond->curr_slave_lock);
4383
4384         if (!BOND_IS_OK(bond))
4385                 goto out;
4386
4387         if (!bond->curr_active_slave)
4388                 goto out;
4389
4390         res = bond_dev_queue_xmit(bond, skb, bond->curr_active_slave->dev);
4391
4392 out:
4393         if (res)
4394                 /* no suitable interface, frame not sent */
4395                 dev_kfree_skb(skb);
4396
4397         read_unlock(&bond->curr_slave_lock);
4398         read_unlock(&bond->lock);
4399         return NETDEV_TX_OK;
4400 }
4401
4402 /*
4403  * In bond_xmit_xor() , we determine the output device by using a pre-
4404  * determined xmit_hash_policy(), If the selected device is not enabled,
4405  * find the next active slave.
4406  */
4407 static int bond_xmit_xor(struct sk_buff *skb, struct net_device *bond_dev)
4408 {
4409         struct bonding *bond = netdev_priv(bond_dev);
4410         struct slave *slave, *start_at;
4411         int slave_no;
4412         int i;
4413         int res = 1;
4414
4415         read_lock(&bond->lock);
4416
4417         if (!BOND_IS_OK(bond))
4418                 goto out;
4419
4420         slave_no = bond->xmit_hash_policy(skb, bond->slave_cnt);
4421
4422         bond_for_each_slave(bond, slave, i) {
4423                 slave_no--;
4424                 if (slave_no < 0)
4425                         break;
4426         }
4427
4428         start_at = slave;
4429
4430         bond_for_each_slave_from(bond, slave, i, start_at) {
4431                 if (IS_UP(slave->dev) &&
4432                     (slave->link == BOND_LINK_UP) &&
4433                     (slave->state == BOND_STATE_ACTIVE)) {
4434                         res = bond_dev_queue_xmit(bond, skb, slave->dev);
4435                         break;
4436                 }
4437         }
4438
4439 out:
4440         if (res) {
4441                 /* no suitable interface, frame not sent */
4442                 dev_kfree_skb(skb);
4443         }
4444         read_unlock(&bond->lock);
4445         return NETDEV_TX_OK;
4446 }
4447
4448 /*
4449  * in broadcast mode, we send everything to all usable interfaces.
4450  */
4451 static int bond_xmit_broadcast(struct sk_buff *skb, struct net_device *bond_dev)
4452 {
4453         struct bonding *bond = netdev_priv(bond_dev);
4454         struct slave *slave, *start_at;
4455         struct net_device *tx_dev = NULL;
4456         int i;
4457         int res = 1;
4458
4459         read_lock(&bond->lock);
4460
4461         if (!BOND_IS_OK(bond))
4462                 goto out;
4463
4464         read_lock(&bond->curr_slave_lock);
4465         start_at = bond->curr_active_slave;
4466         read_unlock(&bond->curr_slave_lock);
4467
4468         if (!start_at)
4469                 goto out;
4470
4471         bond_for_each_slave_from(bond, slave, i, start_at) {
4472                 if (IS_UP(slave->dev) &&
4473                     (slave->link == BOND_LINK_UP) &&
4474                     (slave->state == BOND_STATE_ACTIVE)) {
4475                         if (tx_dev) {
4476                                 struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
4477                                 if (!skb2) {
4478                                         pr_err("%s: Error: bond_xmit_broadcast(): skb_clone() failed\n",
4479                                                bond_dev->name);
4480                                         continue;
4481                                 }
4482
4483                                 res = bond_dev_queue_xmit(bond, skb2, tx_dev);
4484                                 if (res) {
4485                                         dev_kfree_skb(skb2);
4486                                         continue;
4487                                 }
4488                         }
4489                         tx_dev = slave->dev;
4490                 }
4491         }
4492
4493         if (tx_dev)
4494                 res = bond_dev_queue_xmit(bond, skb, tx_dev);
4495
4496 out:
4497         if (res)
4498                 /* no suitable interface, frame not sent */
4499                 dev_kfree_skb(skb);
4500
4501         /* frame sent to all suitable interfaces */
4502         read_unlock(&bond->lock);
4503         return NETDEV_TX_OK;
4504 }
4505
4506 /*------------------------- Device initialization ---------------------------*/
4507
4508 static void bond_set_xmit_hash_policy(struct bonding *bond)
4509 {
4510         switch (bond->params.xmit_policy) {
4511         case BOND_XMIT_POLICY_LAYER23:
4512                 bond->xmit_hash_policy = bond_xmit_hash_policy_l23;
4513                 break;
4514         case BOND_XMIT_POLICY_LAYER34:
4515                 bond->xmit_hash_policy = bond_xmit_hash_policy_l34;
4516                 break;
4517         case BOND_XMIT_POLICY_LAYER2:
4518         default:
4519                 bond->xmit_hash_policy = bond_xmit_hash_policy_l2;
4520                 break;
4521         }
4522 }
4523
4524 /*
4525  * Lookup the slave that corresponds to a qid
4526  */
4527 static inline int bond_slave_override(struct bonding *bond,
4528                                       struct sk_buff *skb)
4529 {
4530         int i, res = 1;
4531         struct slave *slave = NULL;
4532         struct slave *check_slave;
4533
4534         read_lock(&bond->lock);
4535
4536         if (!BOND_IS_OK(bond) || !skb->queue_mapping)
4537                 goto out;
4538
4539         /* Find out if any slaves have the same mapping as this skb. */
4540         bond_for_each_slave(bond, check_slave, i) {
4541                 if (check_slave->queue_id == skb->queue_mapping) {
4542                         slave = check_slave;
4543                         break;
4544                 }
4545         }
4546
4547         /* If the slave isn't UP, use default transmit policy. */
4548         if (slave && slave->queue_id && IS_UP(slave->dev) &&
4549             (slave->link == BOND_LINK_UP)) {
4550                 res = bond_dev_queue_xmit(bond, skb, slave->dev);
4551         }
4552
4553 out:
4554         read_unlock(&bond->lock);
4555         return res;
4556 }
4557
4558 static u16 bond_select_queue(struct net_device *dev, struct sk_buff *skb)
4559 {
4560         /*
4561          * This helper function exists to help dev_pick_tx get the correct
4562          * destination queue.  Using a helper function skips the a call to
4563          * skb_tx_hash and will put the skbs in the queue we expect on their
4564          * way down to the bonding driver.
4565          */
4566         return skb->queue_mapping;
4567 }
4568
4569 static netdev_tx_t bond_start_xmit(struct sk_buff *skb, struct net_device *dev)
4570 {
4571         struct bonding *bond = netdev_priv(dev);
4572
4573         /*
4574          * If we risk deadlock from transmitting this in the
4575          * netpoll path, tell netpoll to queue the frame for later tx
4576          */
4577         if (is_netpoll_tx_blocked(dev))
4578                 return NETDEV_TX_BUSY;
4579
4580         if (TX_QUEUE_OVERRIDE(bond->params.mode)) {
4581                 if (!bond_slave_override(bond, skb))
4582                         return NETDEV_TX_OK;
4583         }
4584
4585         switch (bond->params.mode) {
4586         case BOND_MODE_ROUNDROBIN:
4587                 return bond_xmit_roundrobin(skb, dev);
4588         case BOND_MODE_ACTIVEBACKUP:
4589                 return bond_xmit_activebackup(skb, dev);
4590         case BOND_MODE_XOR:
4591                 return bond_xmit_xor(skb, dev);
4592         case BOND_MODE_BROADCAST:
4593                 return bond_xmit_broadcast(skb, dev);
4594         case BOND_MODE_8023AD:
4595                 return bond_3ad_xmit_xor(skb, dev);
4596         case BOND_MODE_ALB:
4597         case BOND_MODE_TLB:
4598                 return bond_alb_xmit(skb, dev);
4599         default:
4600                 /* Should never happen, mode already checked */
4601                 pr_err("%s: Error: Unknown bonding mode %d\n",
4602                        dev->name, bond->params.mode);
4603                 WARN_ON_ONCE(1);
4604                 dev_kfree_skb(skb);
4605                 return NETDEV_TX_OK;
4606         }
4607 }
4608
4609
4610 /*
4611  * set bond mode specific net device operations
4612  */
4613 void bond_set_mode_ops(struct bonding *bond, int mode)
4614 {
4615         struct net_device *bond_dev = bond->dev;
4616
4617         switch (mode) {
4618         case BOND_MODE_ROUNDROBIN:
4619                 break;
4620         case BOND_MODE_ACTIVEBACKUP:
4621                 break;
4622         case BOND_MODE_XOR:
4623                 bond_set_xmit_hash_policy(bond);
4624                 break;
4625         case BOND_MODE_BROADCAST:
4626                 break;
4627         case BOND_MODE_8023AD:
4628                 bond_set_master_3ad_flags(bond);
4629                 bond_set_xmit_hash_policy(bond);
4630                 break;
4631         case BOND_MODE_ALB:
4632                 bond_set_master_alb_flags(bond);
4633                 /* FALLTHRU */
4634         case BOND_MODE_TLB:
4635                 break;
4636         default:
4637                 /* Should never happen, mode already checked */
4638                 pr_err("%s: Error: Unknown bonding mode %d\n",
4639                        bond_dev->name, mode);
4640                 break;
4641         }
4642 }
4643
4644 static void bond_ethtool_get_drvinfo(struct net_device *bond_dev,
4645                                     struct ethtool_drvinfo *drvinfo)
4646 {
4647         strncpy(drvinfo->driver, DRV_NAME, 32);
4648         strncpy(drvinfo->version, DRV_VERSION, 32);
4649         snprintf(drvinfo->fw_version, 32, "%d", BOND_ABI_VERSION);
4650 }
4651
4652 static const struct ethtool_ops bond_ethtool_ops = {
4653         .get_drvinfo            = bond_ethtool_get_drvinfo,
4654         .get_link               = ethtool_op_get_link,
4655         .get_tx_csum            = ethtool_op_get_tx_csum,
4656         .get_sg                 = ethtool_op_get_sg,
4657         .get_tso                = ethtool_op_get_tso,
4658         .get_ufo                = ethtool_op_get_ufo,
4659         .get_flags              = ethtool_op_get_flags,
4660 };
4661
4662 static const struct net_device_ops bond_netdev_ops = {
4663         .ndo_init               = bond_init,
4664         .ndo_uninit             = bond_uninit,
4665         .ndo_open               = bond_open,
4666         .ndo_stop               = bond_close,
4667         .ndo_start_xmit         = bond_start_xmit,
4668         .ndo_select_queue       = bond_select_queue,
4669         .ndo_get_stats64        = bond_get_stats,
4670         .ndo_do_ioctl           = bond_do_ioctl,
4671         .ndo_set_multicast_list = bond_set_multicast_list,
4672         .ndo_change_mtu         = bond_change_mtu,
4673         .ndo_set_mac_address    = bond_set_mac_address,
4674         .ndo_neigh_setup        = bond_neigh_setup,
4675         .ndo_vlan_rx_register   = bond_vlan_rx_register,
4676         .ndo_vlan_rx_add_vid    = bond_vlan_rx_add_vid,
4677         .ndo_vlan_rx_kill_vid   = bond_vlan_rx_kill_vid,
4678 #ifdef CONFIG_NET_POLL_CONTROLLER
4679         .ndo_netpoll_cleanup    = bond_netpoll_cleanup,
4680         .ndo_poll_controller    = bond_poll_controller,
4681 #endif
4682 };
4683
4684 static void bond_destructor(struct net_device *bond_dev)
4685 {
4686         struct bonding *bond = netdev_priv(bond_dev);
4687         if (bond->wq)
4688                 destroy_workqueue(bond->wq);
4689         free_netdev(bond_dev);
4690 }
4691
4692 static void bond_setup(struct net_device *bond_dev)
4693 {
4694         struct bonding *bond = netdev_priv(bond_dev);
4695
4696         /* initialize rwlocks */
4697         rwlock_init(&bond->lock);
4698         rwlock_init(&bond->curr_slave_lock);
4699
4700         bond->params = bonding_defaults;
4701
4702         /* Initialize pointers */
4703         bond->dev = bond_dev;
4704         INIT_LIST_HEAD(&bond->vlan_list);
4705
4706         /* Initialize the device entry points */
4707         ether_setup(bond_dev);
4708         bond_dev->netdev_ops = &bond_netdev_ops;
4709         bond_dev->ethtool_ops = &bond_ethtool_ops;
4710         bond_set_mode_ops(bond, bond->params.mode);
4711
4712         bond_dev->destructor = bond_destructor;
4713
4714         /* Initialize the device options */
4715         bond_dev->tx_queue_len = 0;
4716         bond_dev->flags |= IFF_MASTER|IFF_MULTICAST;
4717         bond_dev->priv_flags |= IFF_BONDING;
4718         bond_dev->priv_flags &= ~IFF_XMIT_DST_RELEASE;
4719
4720         if (bond->params.arp_interval)
4721                 bond_dev->priv_flags |= IFF_MASTER_ARPMON;
4722
4723         /* At first, we block adding VLANs. That's the only way to
4724          * prevent problems that occur when adding VLANs over an
4725          * empty bond. The block will be removed once non-challenged
4726          * slaves are enslaved.
4727          */
4728         bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
4729
4730         /* don't acquire bond device's netif_tx_lock when
4731          * transmitting */
4732         bond_dev->features |= NETIF_F_LLTX;
4733
4734         /* By default, we declare the bond to be fully
4735          * VLAN hardware accelerated capable. Special
4736          * care is taken in the various xmit functions
4737          * when there are slaves that are not hw accel
4738          * capable
4739          */
4740         bond_dev->features |= (NETIF_F_HW_VLAN_TX |
4741                                NETIF_F_HW_VLAN_RX |
4742                                NETIF_F_HW_VLAN_FILTER);
4743
4744         /* By default, we enable GRO on bonding devices.
4745          * Actual support requires lowlevel drivers are GRO ready.
4746          */
4747         bond_dev->features |= NETIF_F_GRO;
4748 }
4749
4750 static void bond_work_cancel_all(struct bonding *bond)
4751 {
4752         write_lock_bh(&bond->lock);
4753         bond->kill_timers = 1;
4754         write_unlock_bh(&bond->lock);
4755
4756         if (bond->params.miimon && delayed_work_pending(&bond->mii_work))
4757                 cancel_delayed_work(&bond->mii_work);
4758
4759         if (bond->params.arp_interval && delayed_work_pending(&bond->arp_work))
4760                 cancel_delayed_work(&bond->arp_work);
4761
4762         if (bond->params.mode == BOND_MODE_ALB &&
4763             delayed_work_pending(&bond->alb_work))
4764                 cancel_delayed_work(&bond->alb_work);
4765
4766         if (bond->params.mode == BOND_MODE_8023AD &&
4767             delayed_work_pending(&bond->ad_work))
4768                 cancel_delayed_work(&bond->ad_work);
4769
4770         if (delayed_work_pending(&bond->mcast_work))
4771                 cancel_delayed_work(&bond->mcast_work);
4772 }
4773
4774 /*
4775 * Destroy a bonding device.
4776 * Must be under rtnl_lock when this function is called.
4777 */
4778 static void bond_uninit(struct net_device *bond_dev)
4779 {
4780         struct bonding *bond = netdev_priv(bond_dev);
4781         struct vlan_entry *vlan, *tmp;
4782
4783         bond_netpoll_cleanup(bond_dev);
4784
4785         /* Release the bonded slaves */
4786         bond_release_all(bond_dev);
4787
4788         list_del(&bond->bond_list);
4789
4790         bond_work_cancel_all(bond);
4791
4792         bond_remove_proc_entry(bond);
4793
4794         __hw_addr_flush(&bond->mc_list);
4795
4796         list_for_each_entry_safe(vlan, tmp, &bond->vlan_list, vlan_list) {
4797                 list_del(&vlan->vlan_list);
4798                 kfree(vlan);
4799         }
4800 }
4801
4802 /*------------------------- Module initialization ---------------------------*/
4803
4804 /*
4805  * Convert string input module parms.  Accept either the
4806  * number of the mode or its string name.  A bit complicated because
4807  * some mode names are substrings of other names, and calls from sysfs
4808  * may have whitespace in the name (trailing newlines, for example).
4809  */
4810 int bond_parse_parm(const char *buf, const struct bond_parm_tbl *tbl)
4811 {
4812         int modeint = -1, i, rv;
4813         char *p, modestr[BOND_MAX_MODENAME_LEN + 1] = { 0, };
4814
4815         for (p = (char *)buf; *p; p++)
4816                 if (!(isdigit(*p) || isspace(*p)))
4817                         break;
4818
4819         if (*p)
4820                 rv = sscanf(buf, "%20s", modestr);
4821         else
4822                 rv = sscanf(buf, "%d", &modeint);
4823
4824         if (!rv)
4825                 return -1;
4826
4827         for (i = 0; tbl[i].modename; i++) {
4828                 if (modeint == tbl[i].mode)
4829                         return tbl[i].mode;
4830                 if (strcmp(modestr, tbl[i].modename) == 0)
4831                         return tbl[i].mode;
4832         }
4833
4834         return -1;
4835 }
4836
4837 static int bond_check_params(struct bond_params *params)
4838 {
4839         int arp_validate_value, fail_over_mac_value, primary_reselect_value;
4840
4841         /*
4842          * Convert string parameters.
4843          */
4844         if (mode) {
4845                 bond_mode = bond_parse_parm(mode, bond_mode_tbl);
4846                 if (bond_mode == -1) {
4847                         pr_err("Error: Invalid bonding mode \"%s\"\n",
4848                                mode == NULL ? "NULL" : mode);
4849                         return -EINVAL;
4850                 }
4851         }
4852
4853         if (xmit_hash_policy) {
4854                 if ((bond_mode != BOND_MODE_XOR) &&
4855                     (bond_mode != BOND_MODE_8023AD)) {
4856                         pr_info("xmit_hash_policy param is irrelevant in mode %s\n",
4857                                bond_mode_name(bond_mode));
4858                 } else {
4859                         xmit_hashtype = bond_parse_parm(xmit_hash_policy,
4860                                                         xmit_hashtype_tbl);
4861                         if (xmit_hashtype == -1) {
4862                                 pr_err("Error: Invalid xmit_hash_policy \"%s\"\n",
4863                                        xmit_hash_policy == NULL ? "NULL" :
4864                                        xmit_hash_policy);
4865                                 return -EINVAL;
4866                         }
4867                 }
4868         }
4869
4870         if (lacp_rate) {
4871                 if (bond_mode != BOND_MODE_8023AD) {
4872                         pr_info("lacp_rate param is irrelevant in mode %s\n",
4873                                 bond_mode_name(bond_mode));
4874                 } else {
4875                         lacp_fast = bond_parse_parm(lacp_rate, bond_lacp_tbl);
4876                         if (lacp_fast == -1) {
4877                                 pr_err("Error: Invalid lacp rate \"%s\"\n",
4878                                        lacp_rate == NULL ? "NULL" : lacp_rate);
4879                                 return -EINVAL;
4880                         }
4881                 }
4882         }
4883
4884         if (ad_select) {
4885                 params->ad_select = bond_parse_parm(ad_select, ad_select_tbl);
4886                 if (params->ad_select == -1) {
4887                         pr_err("Error: Invalid ad_select \"%s\"\n",
4888                                ad_select == NULL ? "NULL" : ad_select);
4889                         return -EINVAL;
4890                 }
4891
4892                 if (bond_mode != BOND_MODE_8023AD) {
4893                         pr_warning("ad_select param only affects 802.3ad mode\n");
4894                 }
4895         } else {
4896                 params->ad_select = BOND_AD_STABLE;
4897         }
4898
4899         if (max_bonds < 0) {
4900                 pr_warning("Warning: max_bonds (%d) not in range %d-%d, so it was reset to BOND_DEFAULT_MAX_BONDS (%d)\n",
4901                            max_bonds, 0, INT_MAX, BOND_DEFAULT_MAX_BONDS);
4902                 max_bonds = BOND_DEFAULT_MAX_BONDS;
4903         }
4904
4905         if (miimon < 0) {
4906                 pr_warning("Warning: miimon module parameter (%d), not in range 0-%d, so it was reset to %d\n",
4907                            miimon, INT_MAX, BOND_LINK_MON_INTERV);
4908                 miimon = BOND_LINK_MON_INTERV;
4909         }
4910
4911         if (updelay < 0) {
4912                 pr_warning("Warning: updelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4913                            updelay, INT_MAX);
4914                 updelay = 0;
4915         }
4916
4917         if (downdelay < 0) {
4918                 pr_warning("Warning: downdelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4919                            downdelay, INT_MAX);
4920                 downdelay = 0;
4921         }
4922
4923         if ((use_carrier != 0) && (use_carrier != 1)) {
4924                 pr_warning("Warning: use_carrier module parameter (%d), not of valid value (0/1), so it was set to 1\n",
4925                            use_carrier);
4926                 use_carrier = 1;
4927         }
4928
4929         if (num_grat_arp < 0 || num_grat_arp > 255) {
4930                 pr_warning("Warning: num_grat_arp (%d) not in range 0-255 so it was reset to 1\n",
4931                            num_grat_arp);
4932                 num_grat_arp = 1;
4933         }
4934
4935         if (num_unsol_na < 0 || num_unsol_na > 255) {
4936                 pr_warning("Warning: num_unsol_na (%d) not in range 0-255 so it was reset to 1\n",
4937                            num_unsol_na);
4938                 num_unsol_na = 1;
4939         }
4940
4941         /* reset values for 802.3ad */
4942         if (bond_mode == BOND_MODE_8023AD) {
4943                 if (!miimon) {
4944                         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");
4945                         pr_warning("Forcing miimon to 100msec\n");
4946                         miimon = 100;
4947                 }
4948         }
4949
4950         if (tx_queues < 1 || tx_queues > 255) {
4951                 pr_warning("Warning: tx_queues (%d) should be between "
4952                            "1 and 255, resetting to %d\n",
4953                            tx_queues, BOND_DEFAULT_TX_QUEUES);
4954                 tx_queues = BOND_DEFAULT_TX_QUEUES;
4955         }
4956
4957         if ((all_slaves_active != 0) && (all_slaves_active != 1)) {
4958                 pr_warning("Warning: all_slaves_active module parameter (%d), "
4959                            "not of valid value (0/1), so it was set to "
4960                            "0\n", all_slaves_active);
4961                 all_slaves_active = 0;
4962         }
4963
4964         if (resend_igmp < 0 || resend_igmp > 255) {
4965                 pr_warning("Warning: resend_igmp (%d) should be between "
4966                            "0 and 255, resetting to %d\n",
4967                            resend_igmp, BOND_DEFAULT_RESEND_IGMP);
4968                 resend_igmp = BOND_DEFAULT_RESEND_IGMP;
4969         }
4970
4971         /* reset values for TLB/ALB */
4972         if ((bond_mode == BOND_MODE_TLB) ||
4973             (bond_mode == BOND_MODE_ALB)) {
4974                 if (!miimon) {
4975                         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");
4976                         pr_warning("Forcing miimon to 100msec\n");
4977                         miimon = 100;
4978                 }
4979         }
4980
4981         if (bond_mode == BOND_MODE_ALB) {
4982                 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",
4983                           updelay);
4984         }
4985
4986         if (!miimon) {
4987                 if (updelay || downdelay) {
4988                         /* just warn the user the up/down delay will have
4989                          * no effect since miimon is zero...
4990                          */
4991                         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",
4992                                    updelay, downdelay);
4993                 }
4994         } else {
4995                 /* don't allow arp monitoring */
4996                 if (arp_interval) {
4997                         pr_warning("Warning: miimon (%d) and arp_interval (%d) can't be used simultaneously, disabling ARP monitoring\n",
4998                                    miimon, arp_interval);
4999                         arp_interval = 0;
5000                 }
5001
5002                 if ((updelay % miimon) != 0) {
5003                         pr_warning("Warning: updelay (%d) is not a multiple of miimon (%d), updelay rounded to %d ms\n",
5004                                    updelay, miimon,
5005                                    (updelay / miimon) * miimon);
5006                 }
5007
5008                 updelay /= miimon;
5009
5010                 if ((downdelay % miimon) != 0) {
5011                         pr_warning("Warning: downdelay (%d) is not a multiple of miimon (%d), downdelay rounded to %d ms\n",
5012                                    downdelay, miimon,
5013                                    (downdelay / miimon) * miimon);
5014                 }
5015
5016                 downdelay /= miimon;
5017         }
5018
5019         if (arp_interval < 0) {
5020                 pr_warning("Warning: arp_interval module parameter (%d) , not in range 0-%d, so it was reset to %d\n",
5021                            arp_interval, INT_MAX, BOND_LINK_ARP_INTERV);
5022                 arp_interval = BOND_LINK_ARP_INTERV;
5023         }
5024
5025         for (arp_ip_count = 0;
5026              (arp_ip_count < BOND_MAX_ARP_TARGETS) && arp_ip_target[arp_ip_count];
5027              arp_ip_count++) {
5028                 /* not complete check, but should be good enough to
5029                    catch mistakes */
5030                 if (!isdigit(arp_ip_target[arp_ip_count][0])) {
5031                         pr_warning("Warning: bad arp_ip_target module parameter (%s), ARP monitoring will not be performed\n",
5032                                    arp_ip_target[arp_ip_count]);
5033                         arp_interval = 0;
5034                 } else {
5035                         __be32 ip = in_aton(arp_ip_target[arp_ip_count]);
5036                         arp_target[arp_ip_count] = ip;
5037                 }
5038         }
5039
5040         if (arp_interval && !arp_ip_count) {
5041                 /* don't allow arping if no arp_ip_target given... */
5042                 pr_warning("Warning: arp_interval module parameter (%d) specified without providing an arp_ip_target parameter, arp_interval was reset to 0\n",
5043                            arp_interval);
5044                 arp_interval = 0;
5045         }
5046
5047         if (arp_validate) {
5048                 if (bond_mode != BOND_MODE_ACTIVEBACKUP) {
5049                         pr_err("arp_validate only supported in active-backup mode\n");
5050                         return -EINVAL;
5051                 }
5052                 if (!arp_interval) {
5053                         pr_err("arp_validate requires arp_interval\n");
5054                         return -EINVAL;
5055                 }
5056
5057                 arp_validate_value = bond_parse_parm(arp_validate,
5058                                                      arp_validate_tbl);
5059                 if (arp_validate_value == -1) {
5060                         pr_err("Error: invalid arp_validate \"%s\"\n",
5061                                arp_validate == NULL ? "NULL" : arp_validate);
5062                         return -EINVAL;
5063                 }
5064         } else
5065                 arp_validate_value = 0;
5066
5067         if (miimon) {
5068                 pr_info("MII link monitoring set to %d ms\n", miimon);
5069         } else if (arp_interval) {
5070                 int i;
5071
5072                 pr_info("ARP monitoring set to %d ms, validate %s, with %d target(s):",
5073                         arp_interval,
5074                         arp_validate_tbl[arp_validate_value].modename,
5075                         arp_ip_count);
5076
5077                 for (i = 0; i < arp_ip_count; i++)
5078                         pr_info(" %s", arp_ip_target[i]);
5079
5080                 pr_info("\n");
5081
5082         } else if (max_bonds) {
5083                 /* miimon and arp_interval not set, we need one so things
5084                  * work as expected, see bonding.txt for details
5085                  */
5086                 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");
5087         }
5088
5089         if (primary && !USES_PRIMARY(bond_mode)) {
5090                 /* currently, using a primary only makes sense
5091                  * in active backup, TLB or ALB modes
5092                  */
5093                 pr_warning("Warning: %s primary device specified but has no effect in %s mode\n",
5094                            primary, bond_mode_name(bond_mode));
5095                 primary = NULL;
5096         }
5097
5098         if (primary && primary_reselect) {
5099                 primary_reselect_value = bond_parse_parm(primary_reselect,
5100                                                          pri_reselect_tbl);
5101                 if (primary_reselect_value == -1) {
5102                         pr_err("Error: Invalid primary_reselect \"%s\"\n",
5103                                primary_reselect ==
5104                                         NULL ? "NULL" : primary_reselect);
5105                         return -EINVAL;
5106                 }
5107         } else {
5108                 primary_reselect_value = BOND_PRI_RESELECT_ALWAYS;
5109         }
5110
5111         if (fail_over_mac) {
5112                 fail_over_mac_value = bond_parse_parm(fail_over_mac,
5113                                                       fail_over_mac_tbl);
5114                 if (fail_over_mac_value == -1) {
5115                         pr_err("Error: invalid fail_over_mac \"%s\"\n",
5116                                arp_validate == NULL ? "NULL" : arp_validate);
5117                         return -EINVAL;
5118                 }
5119
5120                 if (bond_mode != BOND_MODE_ACTIVEBACKUP)
5121                         pr_warning("Warning: fail_over_mac only affects active-backup mode.\n");
5122         } else {
5123                 fail_over_mac_value = BOND_FOM_NONE;
5124         }
5125
5126         /* fill params struct with the proper values */
5127         params->mode = bond_mode;
5128         params->xmit_policy = xmit_hashtype;
5129         params->miimon = miimon;
5130         params->num_grat_arp = num_grat_arp;
5131         params->num_unsol_na = num_unsol_na;
5132         params->arp_interval = arp_interval;
5133         params->arp_validate = arp_validate_value;
5134         params->updelay = updelay;
5135         params->downdelay = downdelay;
5136         params->use_carrier = use_carrier;
5137         params->lacp_fast = lacp_fast;
5138         params->primary[0] = 0;
5139         params->primary_reselect = primary_reselect_value;
5140         params->fail_over_mac = fail_over_mac_value;
5141         params->tx_queues = tx_queues;
5142         params->all_slaves_active = all_slaves_active;
5143         params->resend_igmp = resend_igmp;
5144
5145         if (primary) {
5146                 strncpy(params->primary, primary, IFNAMSIZ);
5147                 params->primary[IFNAMSIZ - 1] = 0;
5148         }
5149
5150         memcpy(params->arp_targets, arp_target, sizeof(arp_target));
5151
5152         return 0;
5153 }
5154
5155 static struct lock_class_key bonding_netdev_xmit_lock_key;
5156 static struct lock_class_key bonding_netdev_addr_lock_key;
5157
5158 static void bond_set_lockdep_class_one(struct net_device *dev,
5159                                        struct netdev_queue *txq,
5160                                        void *_unused)
5161 {
5162         lockdep_set_class(&txq->_xmit_lock,
5163                           &bonding_netdev_xmit_lock_key);
5164 }
5165
5166 static void bond_set_lockdep_class(struct net_device *dev)
5167 {
5168         lockdep_set_class(&dev->addr_list_lock,
5169                           &bonding_netdev_addr_lock_key);
5170         netdev_for_each_tx_queue(dev, bond_set_lockdep_class_one, NULL);
5171 }
5172
5173 /*
5174  * Called from registration process
5175  */
5176 static int bond_init(struct net_device *bond_dev)
5177 {
5178         struct bonding *bond = netdev_priv(bond_dev);
5179         struct bond_net *bn = net_generic(dev_net(bond_dev), bond_net_id);
5180
5181         pr_debug("Begin bond_init for %s\n", bond_dev->name);
5182
5183         bond->wq = create_singlethread_workqueue(bond_dev->name);
5184         if (!bond->wq)
5185                 return -ENOMEM;
5186
5187         bond_set_lockdep_class(bond_dev);
5188
5189         netif_carrier_off(bond_dev);
5190
5191         bond_create_proc_entry(bond);
5192         list_add_tail(&bond->bond_list, &bn->dev_list);
5193
5194         bond_prepare_sysfs_group(bond);
5195
5196         __hw_addr_init(&bond->mc_list);
5197         return 0;
5198 }
5199
5200 static int bond_validate(struct nlattr *tb[], struct nlattr *data[])
5201 {
5202         if (tb[IFLA_ADDRESS]) {
5203                 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN)
5204                         return -EINVAL;
5205                 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS])))
5206                         return -EADDRNOTAVAIL;
5207         }
5208         return 0;
5209 }
5210
5211 static struct rtnl_link_ops bond_link_ops __read_mostly = {
5212         .kind           = "bond",
5213         .priv_size      = sizeof(struct bonding),
5214         .setup          = bond_setup,
5215         .validate       = bond_validate,
5216 };
5217
5218 /* Create a new bond based on the specified name and bonding parameters.
5219  * If name is NULL, obtain a suitable "bond%d" name for us.
5220  * Caller must NOT hold rtnl_lock; we need to release it here before we
5221  * set up our sysfs entries.
5222  */
5223 int bond_create(struct net *net, const char *name)
5224 {
5225         struct net_device *bond_dev;
5226         int res;
5227
5228         rtnl_lock();
5229
5230         bond_dev = alloc_netdev_mq(sizeof(struct bonding), name ? name : "",
5231                                 bond_setup, tx_queues);
5232         if (!bond_dev) {
5233                 pr_err("%s: eek! can't alloc netdev!\n", name);
5234                 rtnl_unlock();
5235                 return -ENOMEM;
5236         }
5237
5238         dev_net_set(bond_dev, net);
5239         bond_dev->rtnl_link_ops = &bond_link_ops;
5240
5241         if (!name) {
5242                 res = dev_alloc_name(bond_dev, "bond%d");
5243                 if (res < 0)
5244                         goto out;
5245         } else {
5246                 /*
5247                  * If we're given a name to register
5248                  * we need to ensure that its not already
5249                  * registered
5250                  */
5251                 res = -EEXIST;
5252                 if (__dev_get_by_name(net, name) != NULL)
5253                         goto out;
5254         }
5255
5256         res = register_netdevice(bond_dev);
5257
5258 out:
5259         rtnl_unlock();
5260         if (res < 0)
5261                 bond_destructor(bond_dev);
5262         return res;
5263 }
5264
5265 static int __net_init bond_net_init(struct net *net)
5266 {
5267         struct bond_net *bn = net_generic(net, bond_net_id);
5268
5269         bn->net = net;
5270         INIT_LIST_HEAD(&bn->dev_list);
5271
5272         bond_create_proc_dir(bn);
5273         
5274         return 0;
5275 }
5276
5277 static void __net_exit bond_net_exit(struct net *net)
5278 {
5279         struct bond_net *bn = net_generic(net, bond_net_id);
5280
5281         bond_destroy_proc_dir(bn);
5282 }
5283
5284 static struct pernet_operations bond_net_ops = {
5285         .init = bond_net_init,
5286         .exit = bond_net_exit,
5287         .id   = &bond_net_id,
5288         .size = sizeof(struct bond_net),
5289 };
5290
5291 static int __init bonding_init(void)
5292 {
5293         int i;
5294         int res;
5295
5296         pr_info("%s", version);
5297
5298         res = bond_check_params(&bonding_defaults);
5299         if (res)
5300                 goto out;
5301
5302 #ifdef CONFIG_NET_POLL_CONTROLLER
5303         if (!alloc_cpumask_var(&netpoll_block_tx, GFP_KERNEL)) {
5304                 res = -ENOMEM;
5305                 goto out;
5306         }
5307 #endif
5308
5309         res = register_pernet_subsys(&bond_net_ops);
5310         if (res)
5311                 goto out;
5312
5313         res = rtnl_link_register(&bond_link_ops);
5314         if (res)
5315                 goto err_link;
5316
5317         for (i = 0; i < max_bonds; i++) {
5318                 res = bond_create(&init_net, NULL);
5319                 if (res)
5320                         goto err;
5321         }
5322
5323         res = bond_create_sysfs();
5324         if (res)
5325                 goto err;
5326
5327
5328         register_netdevice_notifier(&bond_netdev_notifier);
5329         register_inetaddr_notifier(&bond_inetaddr_notifier);
5330         bond_register_ipv6_notifier();
5331 out:
5332         return res;
5333 err:
5334         rtnl_link_unregister(&bond_link_ops);
5335 err_link:
5336         unregister_pernet_subsys(&bond_net_ops);
5337 #ifdef CONFIG_NET_POLL_CONTROLLER
5338         free_cpumask_var(netpoll_block_tx);
5339 #endif
5340         goto out;
5341
5342 }
5343
5344 static void __exit bonding_exit(void)
5345 {
5346         unregister_netdevice_notifier(&bond_netdev_notifier);
5347         unregister_inetaddr_notifier(&bond_inetaddr_notifier);
5348         bond_unregister_ipv6_notifier();
5349
5350         bond_destroy_sysfs();
5351
5352         rtnl_link_unregister(&bond_link_ops);
5353         unregister_pernet_subsys(&bond_net_ops);
5354
5355 #ifdef CONFIG_NET_POLL_CONTROLLER
5356         free_cpumask_var(netpoll_block_tx);
5357 #endif
5358 }
5359
5360 module_init(bonding_init);
5361 module_exit(bonding_exit);
5362 MODULE_LICENSE("GPL");
5363 MODULE_VERSION(DRV_VERSION);
5364 MODULE_DESCRIPTION(DRV_DESCRIPTION ", v" DRV_VERSION);
5365 MODULE_AUTHOR("Thomas Davis, tadavis@lbl.gov and many others");
5366 MODULE_ALIAS_RTNL_LINK("bond");