bonding: fix potential deadlock in bond_uninit()
[linux-2.6.git] / drivers / net / bonding / bond_main.c
1 /*
2  * originally based on the dummy device.
3  *
4  * Copyright 1999, Thomas Davis, tadavis@lbl.gov.
5  * Licensed under the GPL. Based on dummy.c, and eql.c devices.
6  *
7  * bonding.c: an Ethernet Bonding driver
8  *
9  * This is useful to talk to a Cisco EtherChannel compatible equipment:
10  *      Cisco 5500
11  *      Sun Trunking (Solaris)
12  *      Alteon AceDirector Trunks
13  *      Linux Bonding
14  *      and probably many L2 switches ...
15  *
16  * How it works:
17  *    ifconfig bond0 ipaddress netmask up
18  *      will setup a network device, with an ip address.  No mac address
19  *      will be assigned at this time.  The hw mac address will come from
20  *      the first slave bonded to the channel.  All slaves will then use
21  *      this hw mac address.
22  *
23  *    ifconfig bond0 down
24  *         will release all slaves, marking them as down.
25  *
26  *    ifenslave bond0 eth0
27  *      will attach eth0 to bond0 as a slave.  eth0 hw mac address will either
28  *      a: be used as initial mac address
29  *      b: if a hw mac address already is there, eth0's hw mac address
30  *         will then be set from bond0.
31  *
32  */
33
34 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
35
36 #include <linux/kernel.h>
37 #include <linux/module.h>
38 #include <linux/types.h>
39 #include <linux/fcntl.h>
40 #include <linux/interrupt.h>
41 #include <linux/ptrace.h>
42 #include <linux/ioport.h>
43 #include <linux/in.h>
44 #include <net/ip.h>
45 #include <linux/ip.h>
46 #include <linux/tcp.h>
47 #include <linux/udp.h>
48 #include <linux/slab.h>
49 #include <linux/string.h>
50 #include <linux/init.h>
51 #include <linux/timer.h>
52 #include <linux/socket.h>
53 #include <linux/ctype.h>
54 #include <linux/inet.h>
55 #include <linux/bitops.h>
56 #include <linux/io.h>
57 #include <asm/system.h>
58 #include <asm/dma.h>
59 #include <linux/uaccess.h>
60 #include <linux/errno.h>
61 #include <linux/netdevice.h>
62 #include <linux/inetdevice.h>
63 #include <linux/igmp.h>
64 #include <linux/etherdevice.h>
65 #include <linux/skbuff.h>
66 #include <net/sock.h>
67 #include <linux/rtnetlink.h>
68 #include <linux/proc_fs.h>
69 #include <linux/seq_file.h>
70 #include <linux/smp.h>
71 #include <linux/if_ether.h>
72 #include <net/arp.h>
73 #include <linux/mii.h>
74 #include <linux/ethtool.h>
75 #include <linux/if_vlan.h>
76 #include <linux/if_bonding.h>
77 #include <linux/jiffies.h>
78 #include <net/route.h>
79 #include <net/net_namespace.h>
80 #include <net/netns/generic.h>
81 #include "bonding.h"
82 #include "bond_3ad.h"
83 #include "bond_alb.h"
84
85 /*---------------------------- Module parameters ----------------------------*/
86
87 /* monitor all links that often (in milliseconds). <=0 disables monitoring */
88 #define BOND_LINK_MON_INTERV    0
89 #define BOND_LINK_ARP_INTERV    0
90
91 static int max_bonds    = BOND_DEFAULT_MAX_BONDS;
92 static int num_grat_arp = 1;
93 static int num_unsol_na = 1;
94 static int miimon       = BOND_LINK_MON_INTERV;
95 static int updelay;
96 static int downdelay;
97 static int use_carrier  = 1;
98 static char *mode;
99 static char *primary;
100 static char *primary_reselect;
101 static char *lacp_rate;
102 static char *ad_select;
103 static char *xmit_hash_policy;
104 static int arp_interval = BOND_LINK_ARP_INTERV;
105 static char *arp_ip_target[BOND_MAX_ARP_TARGETS];
106 static char *arp_validate;
107 static char *fail_over_mac;
108 static struct bond_params bonding_defaults;
109
110 module_param(max_bonds, int, 0);
111 MODULE_PARM_DESC(max_bonds, "Max number of bonded devices");
112 module_param(num_grat_arp, int, 0644);
113 MODULE_PARM_DESC(num_grat_arp, "Number of gratuitous ARP packets to send on failover event");
114 module_param(num_unsol_na, int, 0644);
115 MODULE_PARM_DESC(num_unsol_na, "Number of unsolicited IPv6 Neighbor Advertisements packets to send on failover event");
116 module_param(miimon, int, 0);
117 MODULE_PARM_DESC(miimon, "Link check interval in milliseconds");
118 module_param(updelay, int, 0);
119 MODULE_PARM_DESC(updelay, "Delay before considering link up, in milliseconds");
120 module_param(downdelay, int, 0);
121 MODULE_PARM_DESC(downdelay, "Delay before considering link down, "
122                             "in milliseconds");
123 module_param(use_carrier, int, 0);
124 MODULE_PARM_DESC(use_carrier, "Use netif_carrier_ok (vs MII ioctls) in miimon; "
125                               "0 for off, 1 for on (default)");
126 module_param(mode, charp, 0);
127 MODULE_PARM_DESC(mode, "Mode of operation : 0 for balance-rr, "
128                        "1 for active-backup, 2 for balance-xor, "
129                        "3 for broadcast, 4 for 802.3ad, 5 for balance-tlb, "
130                        "6 for balance-alb");
131 module_param(primary, charp, 0);
132 MODULE_PARM_DESC(primary, "Primary network device to use");
133 module_param(primary_reselect, charp, 0);
134 MODULE_PARM_DESC(primary_reselect, "Reselect primary slave "
135                                    "once it comes up; "
136                                    "0 for always (default), "
137                                    "1 for only if speed of primary is "
138                                    "better, "
139                                    "2 for only on active slave "
140                                    "failure");
141 module_param(lacp_rate, charp, 0);
142 MODULE_PARM_DESC(lacp_rate, "LACPDU tx rate to request from 802.3ad partner "
143                             "(slow/fast)");
144 module_param(ad_select, charp, 0);
145 MODULE_PARM_DESC(ad_select, "803.ad aggregation selection logic: stable (0, default), bandwidth (1), count (2)");
146 module_param(xmit_hash_policy, charp, 0);
147 MODULE_PARM_DESC(xmit_hash_policy, "XOR hashing method: 0 for layer 2 (default)"
148                                    ", 1 for layer 3+4");
149 module_param(arp_interval, int, 0);
150 MODULE_PARM_DESC(arp_interval, "arp interval in milliseconds");
151 module_param_array(arp_ip_target, charp, NULL, 0);
152 MODULE_PARM_DESC(arp_ip_target, "arp targets in n.n.n.n form");
153 module_param(arp_validate, charp, 0);
154 MODULE_PARM_DESC(arp_validate, "validate src/dst of ARP probes: none (default), active, backup or all");
155 module_param(fail_over_mac, charp, 0);
156 MODULE_PARM_DESC(fail_over_mac, "For active-backup, do not set all slaves to the same MAC.  none (default), active or follow");
157
158 /*----------------------------- Global variables ----------------------------*/
159
160 static const char * const version =
161         DRV_DESCRIPTION ": v" DRV_VERSION " (" DRV_RELDATE ")\n";
162
163 int bond_net_id __read_mostly;
164
165 static __be32 arp_target[BOND_MAX_ARP_TARGETS];
166 static int arp_ip_count;
167 static int bond_mode    = BOND_MODE_ROUNDROBIN;
168 static int xmit_hashtype = BOND_XMIT_POLICY_LAYER2;
169 static int lacp_fast;
170
171
172 const struct bond_parm_tbl bond_lacp_tbl[] = {
173 {       "slow",         AD_LACP_SLOW},
174 {       "fast",         AD_LACP_FAST},
175 {       NULL,           -1},
176 };
177
178 const struct bond_parm_tbl bond_mode_tbl[] = {
179 {       "balance-rr",           BOND_MODE_ROUNDROBIN},
180 {       "active-backup",        BOND_MODE_ACTIVEBACKUP},
181 {       "balance-xor",          BOND_MODE_XOR},
182 {       "broadcast",            BOND_MODE_BROADCAST},
183 {       "802.3ad",              BOND_MODE_8023AD},
184 {       "balance-tlb",          BOND_MODE_TLB},
185 {       "balance-alb",          BOND_MODE_ALB},
186 {       NULL,                   -1},
187 };
188
189 const struct bond_parm_tbl xmit_hashtype_tbl[] = {
190 {       "layer2",               BOND_XMIT_POLICY_LAYER2},
191 {       "layer3+4",             BOND_XMIT_POLICY_LAYER34},
192 {       "layer2+3",             BOND_XMIT_POLICY_LAYER23},
193 {       NULL,                   -1},
194 };
195
196 const struct bond_parm_tbl arp_validate_tbl[] = {
197 {       "none",                 BOND_ARP_VALIDATE_NONE},
198 {       "active",               BOND_ARP_VALIDATE_ACTIVE},
199 {       "backup",               BOND_ARP_VALIDATE_BACKUP},
200 {       "all",                  BOND_ARP_VALIDATE_ALL},
201 {       NULL,                   -1},
202 };
203
204 const struct bond_parm_tbl fail_over_mac_tbl[] = {
205 {       "none",                 BOND_FOM_NONE},
206 {       "active",               BOND_FOM_ACTIVE},
207 {       "follow",               BOND_FOM_FOLLOW},
208 {       NULL,                   -1},
209 };
210
211 const struct bond_parm_tbl pri_reselect_tbl[] = {
212 {       "always",               BOND_PRI_RESELECT_ALWAYS},
213 {       "better",               BOND_PRI_RESELECT_BETTER},
214 {       "failure",              BOND_PRI_RESELECT_FAILURE},
215 {       NULL,                   -1},
216 };
217
218 struct bond_parm_tbl ad_select_tbl[] = {
219 {       "stable",       BOND_AD_STABLE},
220 {       "bandwidth",    BOND_AD_BANDWIDTH},
221 {       "count",        BOND_AD_COUNT},
222 {       NULL,           -1},
223 };
224
225 /*-------------------------- Forward declarations ---------------------------*/
226
227 static void bond_send_gratuitous_arp(struct bonding *bond);
228 static int bond_init(struct net_device *bond_dev);
229 static void bond_uninit(struct net_device *bond_dev);
230
231 /*---------------------------- General routines -----------------------------*/
232
233 static const char *bond_mode_name(int mode)
234 {
235         static const char *names[] = {
236                 [BOND_MODE_ROUNDROBIN] = "load balancing (round-robin)",
237                 [BOND_MODE_ACTIVEBACKUP] = "fault-tolerance (active-backup)",
238                 [BOND_MODE_XOR] = "load balancing (xor)",
239                 [BOND_MODE_BROADCAST] = "fault-tolerance (broadcast)",
240                 [BOND_MODE_8023AD] = "IEEE 802.3ad Dynamic link aggregation",
241                 [BOND_MODE_TLB] = "transmit load balancing",
242                 [BOND_MODE_ALB] = "adaptive load balancing",
243         };
244
245         if (mode < 0 || mode > BOND_MODE_ALB)
246                 return "unknown";
247
248         return names[mode];
249 }
250
251 /*---------------------------------- VLAN -----------------------------------*/
252
253 /**
254  * bond_add_vlan - add a new vlan id on bond
255  * @bond: bond that got the notification
256  * @vlan_id: the vlan id to add
257  *
258  * Returns -ENOMEM if allocation failed.
259  */
260 static int bond_add_vlan(struct bonding *bond, unsigned short vlan_id)
261 {
262         struct vlan_entry *vlan;
263
264         pr_debug("bond: %s, vlan id %d\n",
265                  (bond ? bond->dev->name : "None"), vlan_id);
266
267         vlan = kzalloc(sizeof(struct vlan_entry), GFP_KERNEL);
268         if (!vlan)
269                 return -ENOMEM;
270
271         INIT_LIST_HEAD(&vlan->vlan_list);
272         vlan->vlan_id = vlan_id;
273
274         write_lock_bh(&bond->lock);
275
276         list_add_tail(&vlan->vlan_list, &bond->vlan_list);
277
278         write_unlock_bh(&bond->lock);
279
280         pr_debug("added VLAN ID %d on bond %s\n", vlan_id, bond->dev->name);
281
282         return 0;
283 }
284
285 /**
286  * bond_del_vlan - delete a vlan id from bond
287  * @bond: bond that got the notification
288  * @vlan_id: the vlan id to delete
289  *
290  * returns -ENODEV if @vlan_id was not found in @bond.
291  */
292 static int bond_del_vlan(struct bonding *bond, unsigned short vlan_id)
293 {
294         struct vlan_entry *vlan;
295         int res = -ENODEV;
296
297         pr_debug("bond: %s, vlan id %d\n", bond->dev->name, vlan_id);
298
299         write_lock_bh(&bond->lock);
300
301         list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
302                 if (vlan->vlan_id == vlan_id) {
303                         list_del(&vlan->vlan_list);
304
305                         if (bond_is_lb(bond))
306                                 bond_alb_clear_vlan(bond, vlan_id);
307
308                         pr_debug("removed VLAN ID %d from bond %s\n",
309                                  vlan_id, bond->dev->name);
310
311                         kfree(vlan);
312
313                         if (list_empty(&bond->vlan_list) &&
314                             (bond->slave_cnt == 0)) {
315                                 /* Last VLAN removed and no slaves, so
316                                  * restore block on adding VLANs. This will
317                                  * be removed once new slaves that are not
318                                  * VLAN challenged will be added.
319                                  */
320                                 bond->dev->features |= NETIF_F_VLAN_CHALLENGED;
321                         }
322
323                         res = 0;
324                         goto out;
325                 }
326         }
327
328         pr_debug("couldn't find VLAN ID %d in bond %s\n",
329                  vlan_id, bond->dev->name);
330
331 out:
332         write_unlock_bh(&bond->lock);
333         return res;
334 }
335
336 /**
337  * bond_has_challenged_slaves
338  * @bond: the bond we're working on
339  *
340  * Searches the slave list. Returns 1 if a vlan challenged slave
341  * was found, 0 otherwise.
342  *
343  * Assumes bond->lock is held.
344  */
345 static int bond_has_challenged_slaves(struct bonding *bond)
346 {
347         struct slave *slave;
348         int i;
349
350         bond_for_each_slave(bond, slave, i) {
351                 if (slave->dev->features & NETIF_F_VLAN_CHALLENGED) {
352                         pr_debug("found VLAN challenged slave - %s\n",
353                                  slave->dev->name);
354                         return 1;
355                 }
356         }
357
358         pr_debug("no VLAN challenged slaves found\n");
359         return 0;
360 }
361
362 /**
363  * bond_next_vlan - safely skip to the next item in the vlans list.
364  * @bond: the bond we're working on
365  * @curr: item we're advancing from
366  *
367  * Returns %NULL if list is empty, bond->next_vlan if @curr is %NULL,
368  * or @curr->next otherwise (even if it is @curr itself again).
369  *
370  * Caller must hold bond->lock
371  */
372 struct vlan_entry *bond_next_vlan(struct bonding *bond, struct vlan_entry *curr)
373 {
374         struct vlan_entry *next, *last;
375
376         if (list_empty(&bond->vlan_list))
377                 return NULL;
378
379         if (!curr) {
380                 next = list_entry(bond->vlan_list.next,
381                                   struct vlan_entry, vlan_list);
382         } else {
383                 last = list_entry(bond->vlan_list.prev,
384                                   struct vlan_entry, vlan_list);
385                 if (last == curr) {
386                         next = list_entry(bond->vlan_list.next,
387                                           struct vlan_entry, vlan_list);
388                 } else {
389                         next = list_entry(curr->vlan_list.next,
390                                           struct vlan_entry, vlan_list);
391                 }
392         }
393
394         return next;
395 }
396
397 /**
398  * bond_dev_queue_xmit - Prepare skb for xmit.
399  *
400  * @bond: bond device that got this skb for tx.
401  * @skb: hw accel VLAN tagged skb to transmit
402  * @slave_dev: slave that is supposed to xmit this skbuff
403  *
404  * When the bond gets an skb to transmit that is
405  * already hardware accelerated VLAN tagged, and it
406  * needs to relay this skb to a slave that is not
407  * hw accel capable, the skb needs to be "unaccelerated",
408  * i.e. strip the hwaccel tag and re-insert it as part
409  * of the payload.
410  */
411 int bond_dev_queue_xmit(struct bonding *bond, struct sk_buff *skb,
412                         struct net_device *slave_dev)
413 {
414         unsigned short uninitialized_var(vlan_id);
415
416         if (!list_empty(&bond->vlan_list) &&
417             !(slave_dev->features & NETIF_F_HW_VLAN_TX) &&
418             vlan_get_tag(skb, &vlan_id) == 0) {
419                 skb->dev = slave_dev;
420                 skb = vlan_put_tag(skb, vlan_id);
421                 if (!skb) {
422                         /* vlan_put_tag() frees the skb in case of error,
423                          * so return success here so the calling functions
424                          * won't attempt to free is again.
425                          */
426                         return 0;
427                 }
428         } else {
429                 skb->dev = slave_dev;
430         }
431
432         skb->priority = 1;
433         dev_queue_xmit(skb);
434
435         return 0;
436 }
437
438 /*
439  * In the following 3 functions, bond_vlan_rx_register(), bond_vlan_rx_add_vid
440  * and bond_vlan_rx_kill_vid, We don't protect the slave list iteration with a
441  * lock because:
442  * a. This operation is performed in IOCTL context,
443  * b. The operation is protected by the RTNL semaphore in the 8021q code,
444  * c. Holding a lock with BH disabled while directly calling a base driver
445  *    entry point is generally a BAD idea.
446  *
447  * The design of synchronization/protection for this operation in the 8021q
448  * module is good for one or more VLAN devices over a single physical device
449  * and cannot be extended for a teaming solution like bonding, so there is a
450  * potential race condition here where a net device from the vlan group might
451  * be referenced (either by a base driver or the 8021q code) while it is being
452  * removed from the system. However, it turns out we're not making matters
453  * worse, and if it works for regular VLAN usage it will work here too.
454 */
455
456 /**
457  * bond_vlan_rx_register - Propagates registration to slaves
458  * @bond_dev: bonding net device that got called
459  * @grp: vlan group being registered
460  */
461 static void bond_vlan_rx_register(struct net_device *bond_dev,
462                                   struct vlan_group *grp)
463 {
464         struct bonding *bond = netdev_priv(bond_dev);
465         struct slave *slave;
466         int i;
467
468         bond->vlgrp = grp;
469
470         bond_for_each_slave(bond, slave, i) {
471                 struct net_device *slave_dev = slave->dev;
472                 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
473
474                 if ((slave_dev->features & NETIF_F_HW_VLAN_RX) &&
475                     slave_ops->ndo_vlan_rx_register) {
476                         slave_ops->ndo_vlan_rx_register(slave_dev, grp);
477                 }
478         }
479 }
480
481 /**
482  * bond_vlan_rx_add_vid - Propagates adding an id to slaves
483  * @bond_dev: bonding net device that got called
484  * @vid: vlan id being added
485  */
486 static void bond_vlan_rx_add_vid(struct net_device *bond_dev, uint16_t vid)
487 {
488         struct bonding *bond = netdev_priv(bond_dev);
489         struct slave *slave;
490         int i, res;
491
492         bond_for_each_slave(bond, slave, i) {
493                 struct net_device *slave_dev = slave->dev;
494                 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
495
496                 if ((slave_dev->features & NETIF_F_HW_VLAN_FILTER) &&
497                     slave_ops->ndo_vlan_rx_add_vid) {
498                         slave_ops->ndo_vlan_rx_add_vid(slave_dev, vid);
499                 }
500         }
501
502         res = bond_add_vlan(bond, vid);
503         if (res) {
504                 pr_err("%s: Error: Failed to add vlan id %d\n",
505                        bond_dev->name, vid);
506         }
507 }
508
509 /**
510  * bond_vlan_rx_kill_vid - Propagates deleting an id to slaves
511  * @bond_dev: bonding net device that got called
512  * @vid: vlan id being removed
513  */
514 static void bond_vlan_rx_kill_vid(struct net_device *bond_dev, uint16_t vid)
515 {
516         struct bonding *bond = netdev_priv(bond_dev);
517         struct slave *slave;
518         struct net_device *vlan_dev;
519         int i, res;
520
521         bond_for_each_slave(bond, slave, i) {
522                 struct net_device *slave_dev = slave->dev;
523                 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
524
525                 if ((slave_dev->features & NETIF_F_HW_VLAN_FILTER) &&
526                     slave_ops->ndo_vlan_rx_kill_vid) {
527                         /* Save and then restore vlan_dev in the grp array,
528                          * since the slave's driver might clear it.
529                          */
530                         vlan_dev = vlan_group_get_device(bond->vlgrp, vid);
531                         slave_ops->ndo_vlan_rx_kill_vid(slave_dev, vid);
532                         vlan_group_set_device(bond->vlgrp, vid, vlan_dev);
533                 }
534         }
535
536         res = bond_del_vlan(bond, vid);
537         if (res) {
538                 pr_err("%s: Error: Failed to remove vlan id %d\n",
539                        bond_dev->name, vid);
540         }
541 }
542
543 static void bond_add_vlans_on_slave(struct bonding *bond, struct net_device *slave_dev)
544 {
545         struct vlan_entry *vlan;
546         const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
547
548         write_lock_bh(&bond->lock);
549
550         if (list_empty(&bond->vlan_list))
551                 goto out;
552
553         if ((slave_dev->features & NETIF_F_HW_VLAN_RX) &&
554             slave_ops->ndo_vlan_rx_register)
555                 slave_ops->ndo_vlan_rx_register(slave_dev, bond->vlgrp);
556
557         if (!(slave_dev->features & NETIF_F_HW_VLAN_FILTER) ||
558             !(slave_ops->ndo_vlan_rx_add_vid))
559                 goto out;
560
561         list_for_each_entry(vlan, &bond->vlan_list, vlan_list)
562                 slave_ops->ndo_vlan_rx_add_vid(slave_dev, vlan->vlan_id);
563
564 out:
565         write_unlock_bh(&bond->lock);
566 }
567
568 static void bond_del_vlans_from_slave(struct bonding *bond,
569                                       struct net_device *slave_dev)
570 {
571         const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
572         struct vlan_entry *vlan;
573         struct net_device *vlan_dev;
574
575         write_lock_bh(&bond->lock);
576
577         if (list_empty(&bond->vlan_list))
578                 goto out;
579
580         if (!(slave_dev->features & NETIF_F_HW_VLAN_FILTER) ||
581             !(slave_ops->ndo_vlan_rx_kill_vid))
582                 goto unreg;
583
584         list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
585                 /* Save and then restore vlan_dev in the grp array,
586                  * since the slave's driver might clear it.
587                  */
588                 vlan_dev = vlan_group_get_device(bond->vlgrp, vlan->vlan_id);
589                 slave_ops->ndo_vlan_rx_kill_vid(slave_dev, vlan->vlan_id);
590                 vlan_group_set_device(bond->vlgrp, vlan->vlan_id, vlan_dev);
591         }
592
593 unreg:
594         if ((slave_dev->features & NETIF_F_HW_VLAN_RX) &&
595             slave_ops->ndo_vlan_rx_register)
596                 slave_ops->ndo_vlan_rx_register(slave_dev, NULL);
597
598 out:
599         write_unlock_bh(&bond->lock);
600 }
601
602 /*------------------------------- Link status -------------------------------*/
603
604 /*
605  * Set the carrier state for the master according to the state of its
606  * slaves.  If any slaves are up, the master is up.  In 802.3ad mode,
607  * do special 802.3ad magic.
608  *
609  * Returns zero if carrier state does not change, nonzero if it does.
610  */
611 static int bond_set_carrier(struct bonding *bond)
612 {
613         struct slave *slave;
614         int i;
615
616         if (bond->slave_cnt == 0)
617                 goto down;
618
619         if (bond->params.mode == BOND_MODE_8023AD)
620                 return bond_3ad_set_carrier(bond);
621
622         bond_for_each_slave(bond, slave, i) {
623                 if (slave->link == BOND_LINK_UP) {
624                         if (!netif_carrier_ok(bond->dev)) {
625                                 netif_carrier_on(bond->dev);
626                                 return 1;
627                         }
628                         return 0;
629                 }
630         }
631
632 down:
633         if (netif_carrier_ok(bond->dev)) {
634                 netif_carrier_off(bond->dev);
635                 return 1;
636         }
637         return 0;
638 }
639
640 /*
641  * Get link speed and duplex from the slave's base driver
642  * using ethtool. If for some reason the call fails or the
643  * values are invalid, fake speed and duplex to 100/Full
644  * and return error.
645  */
646 static int bond_update_speed_duplex(struct slave *slave)
647 {
648         struct net_device *slave_dev = slave->dev;
649         struct ethtool_cmd etool;
650         int res;
651
652         /* Fake speed and duplex */
653         slave->speed = SPEED_100;
654         slave->duplex = DUPLEX_FULL;
655
656         if (!slave_dev->ethtool_ops || !slave_dev->ethtool_ops->get_settings)
657                 return -1;
658
659         res = slave_dev->ethtool_ops->get_settings(slave_dev, &etool);
660         if (res < 0)
661                 return -1;
662
663         switch (etool.speed) {
664         case SPEED_10:
665         case SPEED_100:
666         case SPEED_1000:
667         case SPEED_10000:
668                 break;
669         default:
670                 return -1;
671         }
672
673         switch (etool.duplex) {
674         case DUPLEX_FULL:
675         case DUPLEX_HALF:
676                 break;
677         default:
678                 return -1;
679         }
680
681         slave->speed = etool.speed;
682         slave->duplex = etool.duplex;
683
684         return 0;
685 }
686
687 /*
688  * if <dev> supports MII link status reporting, check its link status.
689  *
690  * We either do MII/ETHTOOL ioctls, or check netif_carrier_ok(),
691  * depending upon the setting of the use_carrier parameter.
692  *
693  * Return either BMSR_LSTATUS, meaning that the link is up (or we
694  * can't tell and just pretend it is), or 0, meaning that the link is
695  * down.
696  *
697  * If reporting is non-zero, instead of faking link up, return -1 if
698  * both ETHTOOL and MII ioctls fail (meaning the device does not
699  * support them).  If use_carrier is set, return whatever it says.
700  * It'd be nice if there was a good way to tell if a driver supports
701  * netif_carrier, but there really isn't.
702  */
703 static int bond_check_dev_link(struct bonding *bond,
704                                struct net_device *slave_dev, int reporting)
705 {
706         const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
707         int (*ioctl)(struct net_device *, struct ifreq *, int);
708         struct ifreq ifr;
709         struct mii_ioctl_data *mii;
710
711         if (!reporting && !netif_running(slave_dev))
712                 return 0;
713
714         if (bond->params.use_carrier)
715                 return netif_carrier_ok(slave_dev) ? BMSR_LSTATUS : 0;
716
717         /* Try to get link status using Ethtool first. */
718         if (slave_dev->ethtool_ops) {
719                 if (slave_dev->ethtool_ops->get_link) {
720                         u32 link;
721
722                         link = slave_dev->ethtool_ops->get_link(slave_dev);
723
724                         return link ? BMSR_LSTATUS : 0;
725                 }
726         }
727
728         /* Ethtool can't be used, fallback to MII ioctls. */
729         ioctl = slave_ops->ndo_do_ioctl;
730         if (ioctl) {
731                 /* TODO: set pointer to correct ioctl on a per team member */
732                 /*       bases to make this more efficient. that is, once  */
733                 /*       we determine the correct ioctl, we will always    */
734                 /*       call it and not the others for that team          */
735                 /*       member.                                           */
736
737                 /*
738                  * We cannot assume that SIOCGMIIPHY will also read a
739                  * register; not all network drivers (e.g., e100)
740                  * support that.
741                  */
742
743                 /* Yes, the mii is overlaid on the ifreq.ifr_ifru */
744                 strncpy(ifr.ifr_name, slave_dev->name, IFNAMSIZ);
745                 mii = if_mii(&ifr);
746                 if (IOCTL(slave_dev, &ifr, SIOCGMIIPHY) == 0) {
747                         mii->reg_num = MII_BMSR;
748                         if (IOCTL(slave_dev, &ifr, SIOCGMIIREG) == 0)
749                                 return mii->val_out & BMSR_LSTATUS;
750                 }
751         }
752
753         /*
754          * If reporting, report that either there's no dev->do_ioctl,
755          * or both SIOCGMIIREG and get_link failed (meaning that we
756          * cannot report link status).  If not reporting, pretend
757          * we're ok.
758          */
759         return reporting ? -1 : BMSR_LSTATUS;
760 }
761
762 /*----------------------------- Multicast list ------------------------------*/
763
764 /*
765  * Returns 0 if dmi1 and dmi2 are the same, non-0 otherwise
766  */
767 static inline int bond_is_dmi_same(const struct dev_mc_list *dmi1,
768                                    const struct dev_mc_list *dmi2)
769 {
770         return memcmp(dmi1->dmi_addr, dmi2->dmi_addr, dmi1->dmi_addrlen) == 0 &&
771                         dmi1->dmi_addrlen == dmi2->dmi_addrlen;
772 }
773
774 /*
775  * returns dmi entry if found, NULL otherwise
776  */
777 static struct dev_mc_list *bond_mc_list_find_dmi(struct dev_mc_list *dmi,
778                                                  struct dev_mc_list *mc_list)
779 {
780         struct dev_mc_list *idmi;
781
782         for (idmi = mc_list; idmi; idmi = idmi->next) {
783                 if (bond_is_dmi_same(dmi, idmi))
784                         return idmi;
785         }
786
787         return NULL;
788 }
789
790 /*
791  * Push the promiscuity flag down to appropriate slaves
792  */
793 static int bond_set_promiscuity(struct bonding *bond, int inc)
794 {
795         int err = 0;
796         if (USES_PRIMARY(bond->params.mode)) {
797                 /* write lock already acquired */
798                 if (bond->curr_active_slave) {
799                         err = dev_set_promiscuity(bond->curr_active_slave->dev,
800                                                   inc);
801                 }
802         } else {
803                 struct slave *slave;
804                 int i;
805                 bond_for_each_slave(bond, slave, i) {
806                         err = dev_set_promiscuity(slave->dev, inc);
807                         if (err)
808                                 return err;
809                 }
810         }
811         return err;
812 }
813
814 /*
815  * Push the allmulti flag down to all slaves
816  */
817 static int bond_set_allmulti(struct bonding *bond, int inc)
818 {
819         int err = 0;
820         if (USES_PRIMARY(bond->params.mode)) {
821                 /* write lock already acquired */
822                 if (bond->curr_active_slave) {
823                         err = dev_set_allmulti(bond->curr_active_slave->dev,
824                                                inc);
825                 }
826         } else {
827                 struct slave *slave;
828                 int i;
829                 bond_for_each_slave(bond, slave, i) {
830                         err = dev_set_allmulti(slave->dev, inc);
831                         if (err)
832                                 return err;
833                 }
834         }
835         return err;
836 }
837
838 /*
839  * Add a Multicast address to slaves
840  * according to mode
841  */
842 static void bond_mc_add(struct bonding *bond, void *addr, int alen)
843 {
844         if (USES_PRIMARY(bond->params.mode)) {
845                 /* write lock already acquired */
846                 if (bond->curr_active_slave)
847                         dev_mc_add(bond->curr_active_slave->dev, addr, alen, 0);
848         } else {
849                 struct slave *slave;
850                 int i;
851
852                 bond_for_each_slave(bond, slave, i)
853                         dev_mc_add(slave->dev, addr, alen, 0);
854         }
855 }
856
857 /*
858  * Remove a multicast address from slave
859  * according to mode
860  */
861 static void bond_mc_delete(struct bonding *bond, void *addr, int alen)
862 {
863         if (USES_PRIMARY(bond->params.mode)) {
864                 /* write lock already acquired */
865                 if (bond->curr_active_slave)
866                         dev_mc_delete(bond->curr_active_slave->dev, addr,
867                                       alen, 0);
868         } else {
869                 struct slave *slave;
870                 int i;
871                 bond_for_each_slave(bond, slave, i) {
872                         dev_mc_delete(slave->dev, addr, alen, 0);
873                 }
874         }
875 }
876
877
878 /*
879  * Retrieve the list of registered multicast addresses for the bonding
880  * device and retransmit an IGMP JOIN request to the current active
881  * slave.
882  */
883 static void bond_resend_igmp_join_requests(struct bonding *bond)
884 {
885         struct in_device *in_dev;
886         struct ip_mc_list *im;
887
888         rcu_read_lock();
889         in_dev = __in_dev_get_rcu(bond->dev);
890         if (in_dev) {
891                 for (im = in_dev->mc_list; im; im = im->next)
892                         ip_mc_rejoin_group(im);
893         }
894
895         rcu_read_unlock();
896 }
897
898 /*
899  * Totally destroys the mc_list in bond
900  */
901 static void bond_mc_list_destroy(struct bonding *bond)
902 {
903         struct dev_mc_list *dmi;
904
905         dmi = bond->mc_list;
906         while (dmi) {
907                 bond->mc_list = dmi->next;
908                 kfree(dmi);
909                 dmi = bond->mc_list;
910         }
911
912         bond->mc_list = NULL;
913 }
914
915 /*
916  * Copy all the Multicast addresses from src to the bonding device dst
917  */
918 static int bond_mc_list_copy(struct dev_mc_list *mc_list, struct bonding *bond,
919                              gfp_t gfp_flag)
920 {
921         struct dev_mc_list *dmi, *new_dmi;
922
923         for (dmi = mc_list; dmi; dmi = dmi->next) {
924                 new_dmi = kmalloc(sizeof(struct dev_mc_list), gfp_flag);
925
926                 if (!new_dmi) {
927                         /* FIXME: Potential memory leak !!! */
928                         return -ENOMEM;
929                 }
930
931                 new_dmi->next = bond->mc_list;
932                 bond->mc_list = new_dmi;
933                 new_dmi->dmi_addrlen = dmi->dmi_addrlen;
934                 memcpy(new_dmi->dmi_addr, dmi->dmi_addr, dmi->dmi_addrlen);
935                 new_dmi->dmi_users = dmi->dmi_users;
936                 new_dmi->dmi_gusers = dmi->dmi_gusers;
937         }
938
939         return 0;
940 }
941
942 /*
943  * flush all members of flush->mc_list from device dev->mc_list
944  */
945 static void bond_mc_list_flush(struct net_device *bond_dev,
946                                struct net_device *slave_dev)
947 {
948         struct bonding *bond = netdev_priv(bond_dev);
949         struct dev_mc_list *dmi;
950
951         for (dmi = bond_dev->mc_list; dmi; dmi = dmi->next)
952                 dev_mc_delete(slave_dev, dmi->dmi_addr, dmi->dmi_addrlen, 0);
953
954         if (bond->params.mode == BOND_MODE_8023AD) {
955                 /* del lacpdu mc addr from mc list */
956                 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
957
958                 dev_mc_delete(slave_dev, lacpdu_multicast, ETH_ALEN, 0);
959         }
960 }
961
962 /*--------------------------- Active slave change ---------------------------*/
963
964 /*
965  * Update the mc list and multicast-related flags for the new and
966  * old active slaves (if any) according to the multicast mode, and
967  * promiscuous flags unconditionally.
968  */
969 static void bond_mc_swap(struct bonding *bond, struct slave *new_active,
970                          struct slave *old_active)
971 {
972         struct dev_mc_list *dmi;
973
974         if (!USES_PRIMARY(bond->params.mode))
975                 /* nothing to do -  mc list is already up-to-date on
976                  * all slaves
977                  */
978                 return;
979
980         if (old_active) {
981                 if (bond->dev->flags & IFF_PROMISC)
982                         dev_set_promiscuity(old_active->dev, -1);
983
984                 if (bond->dev->flags & IFF_ALLMULTI)
985                         dev_set_allmulti(old_active->dev, -1);
986
987                 for (dmi = bond->dev->mc_list; dmi; dmi = dmi->next)
988                         dev_mc_delete(old_active->dev, dmi->dmi_addr,
989                                       dmi->dmi_addrlen, 0);
990         }
991
992         if (new_active) {
993                 /* FIXME: Signal errors upstream. */
994                 if (bond->dev->flags & IFF_PROMISC)
995                         dev_set_promiscuity(new_active->dev, 1);
996
997                 if (bond->dev->flags & IFF_ALLMULTI)
998                         dev_set_allmulti(new_active->dev, 1);
999
1000                 for (dmi = bond->dev->mc_list; dmi; dmi = dmi->next)
1001                         dev_mc_add(new_active->dev, dmi->dmi_addr,
1002                                    dmi->dmi_addrlen, 0);
1003                 bond_resend_igmp_join_requests(bond);
1004         }
1005 }
1006
1007 /*
1008  * bond_do_fail_over_mac
1009  *
1010  * Perform special MAC address swapping for fail_over_mac settings
1011  *
1012  * Called with RTNL, bond->lock for read, curr_slave_lock for write_bh.
1013  */
1014 static void bond_do_fail_over_mac(struct bonding *bond,
1015                                   struct slave *new_active,
1016                                   struct slave *old_active)
1017         __releases(&bond->curr_slave_lock)
1018         __releases(&bond->lock)
1019         __acquires(&bond->lock)
1020         __acquires(&bond->curr_slave_lock)
1021 {
1022         u8 tmp_mac[ETH_ALEN];
1023         struct sockaddr saddr;
1024         int rv;
1025
1026         switch (bond->params.fail_over_mac) {
1027         case BOND_FOM_ACTIVE:
1028                 if (new_active)
1029                         memcpy(bond->dev->dev_addr,  new_active->dev->dev_addr,
1030                                new_active->dev->addr_len);
1031                 break;
1032         case BOND_FOM_FOLLOW:
1033                 /*
1034                  * if new_active && old_active, swap them
1035                  * if just old_active, do nothing (going to no active slave)
1036                  * if just new_active, set new_active to bond's MAC
1037                  */
1038                 if (!new_active)
1039                         return;
1040
1041                 write_unlock_bh(&bond->curr_slave_lock);
1042                 read_unlock(&bond->lock);
1043
1044                 if (old_active) {
1045                         memcpy(tmp_mac, new_active->dev->dev_addr, ETH_ALEN);
1046                         memcpy(saddr.sa_data, old_active->dev->dev_addr,
1047                                ETH_ALEN);
1048                         saddr.sa_family = new_active->dev->type;
1049                 } else {
1050                         memcpy(saddr.sa_data, bond->dev->dev_addr, ETH_ALEN);
1051                         saddr.sa_family = bond->dev->type;
1052                 }
1053
1054                 rv = dev_set_mac_address(new_active->dev, &saddr);
1055                 if (rv) {
1056                         pr_err("%s: Error %d setting MAC of slave %s\n",
1057                                bond->dev->name, -rv, new_active->dev->name);
1058                         goto out;
1059                 }
1060
1061                 if (!old_active)
1062                         goto out;
1063
1064                 memcpy(saddr.sa_data, tmp_mac, ETH_ALEN);
1065                 saddr.sa_family = old_active->dev->type;
1066
1067                 rv = dev_set_mac_address(old_active->dev, &saddr);
1068                 if (rv)
1069                         pr_err("%s: Error %d setting MAC of slave %s\n",
1070                                bond->dev->name, -rv, new_active->dev->name);
1071 out:
1072                 read_lock(&bond->lock);
1073                 write_lock_bh(&bond->curr_slave_lock);
1074                 break;
1075         default:
1076                 pr_err("%s: bond_do_fail_over_mac impossible: bad policy %d\n",
1077                        bond->dev->name, bond->params.fail_over_mac);
1078                 break;
1079         }
1080
1081 }
1082
1083 static bool bond_should_change_active(struct bonding *bond)
1084 {
1085         struct slave *prim = bond->primary_slave;
1086         struct slave *curr = bond->curr_active_slave;
1087
1088         if (!prim || !curr || curr->link != BOND_LINK_UP)
1089                 return true;
1090         if (bond->force_primary) {
1091                 bond->force_primary = false;
1092                 return true;
1093         }
1094         if (bond->params.primary_reselect == BOND_PRI_RESELECT_BETTER &&
1095             (prim->speed < curr->speed ||
1096              (prim->speed == curr->speed && prim->duplex <= curr->duplex)))
1097                 return false;
1098         if (bond->params.primary_reselect == BOND_PRI_RESELECT_FAILURE)
1099                 return false;
1100         return true;
1101 }
1102
1103 /**
1104  * find_best_interface - select the best available slave to be the active one
1105  * @bond: our bonding struct
1106  *
1107  * Warning: Caller must hold curr_slave_lock for writing.
1108  */
1109 static struct slave *bond_find_best_slave(struct bonding *bond)
1110 {
1111         struct slave *new_active, *old_active;
1112         struct slave *bestslave = NULL;
1113         int mintime = bond->params.updelay;
1114         int i;
1115
1116         new_active = bond->curr_active_slave;
1117
1118         if (!new_active) { /* there were no active slaves left */
1119                 if (bond->slave_cnt > 0)   /* found one slave */
1120                         new_active = bond->first_slave;
1121                 else
1122                         return NULL; /* still no slave, return NULL */
1123         }
1124
1125         if ((bond->primary_slave) &&
1126             bond->primary_slave->link == BOND_LINK_UP &&
1127             bond_should_change_active(bond)) {
1128                 new_active = bond->primary_slave;
1129         }
1130
1131         /* remember where to stop iterating over the slaves */
1132         old_active = new_active;
1133
1134         bond_for_each_slave_from(bond, new_active, i, old_active) {
1135                 if (new_active->link == BOND_LINK_UP) {
1136                         return new_active;
1137                 } else if (new_active->link == BOND_LINK_BACK &&
1138                            IS_UP(new_active->dev)) {
1139                         /* link up, but waiting for stabilization */
1140                         if (new_active->delay < mintime) {
1141                                 mintime = new_active->delay;
1142                                 bestslave = new_active;
1143                         }
1144                 }
1145         }
1146
1147         return bestslave;
1148 }
1149
1150 /**
1151  * change_active_interface - change the active slave into the specified one
1152  * @bond: our bonding struct
1153  * @new: the new slave to make the active one
1154  *
1155  * Set the new slave to the bond's settings and unset them on the old
1156  * curr_active_slave.
1157  * Setting include flags, mc-list, promiscuity, allmulti, etc.
1158  *
1159  * If @new's link state is %BOND_LINK_BACK we'll set it to %BOND_LINK_UP,
1160  * because it is apparently the best available slave we have, even though its
1161  * updelay hasn't timed out yet.
1162  *
1163  * If new_active is not NULL, caller must hold bond->lock for read and
1164  * curr_slave_lock for write_bh.
1165  */
1166 void bond_change_active_slave(struct bonding *bond, struct slave *new_active)
1167 {
1168         struct slave *old_active = bond->curr_active_slave;
1169
1170         if (old_active == new_active)
1171                 return;
1172
1173         if (new_active) {
1174                 new_active->jiffies = jiffies;
1175
1176                 if (new_active->link == BOND_LINK_BACK) {
1177                         if (USES_PRIMARY(bond->params.mode)) {
1178                                 pr_info("%s: making interface %s the new active one %d ms earlier.\n",
1179                                         bond->dev->name, new_active->dev->name,
1180                                         (bond->params.updelay - new_active->delay) * bond->params.miimon);
1181                         }
1182
1183                         new_active->delay = 0;
1184                         new_active->link = BOND_LINK_UP;
1185
1186                         if (bond->params.mode == BOND_MODE_8023AD)
1187                                 bond_3ad_handle_link_change(new_active, BOND_LINK_UP);
1188
1189                         if (bond_is_lb(bond))
1190                                 bond_alb_handle_link_change(bond, new_active, BOND_LINK_UP);
1191                 } else {
1192                         if (USES_PRIMARY(bond->params.mode)) {
1193                                 pr_info("%s: making interface %s the new active one.\n",
1194                                         bond->dev->name, new_active->dev->name);
1195                         }
1196                 }
1197         }
1198
1199         if (USES_PRIMARY(bond->params.mode))
1200                 bond_mc_swap(bond, new_active, old_active);
1201
1202         if (bond_is_lb(bond)) {
1203                 bond_alb_handle_active_change(bond, new_active);
1204                 if (old_active)
1205                         bond_set_slave_inactive_flags(old_active);
1206                 if (new_active)
1207                         bond_set_slave_active_flags(new_active);
1208         } else {
1209                 bond->curr_active_slave = new_active;
1210         }
1211
1212         if (bond->params.mode == BOND_MODE_ACTIVEBACKUP) {
1213                 if (old_active)
1214                         bond_set_slave_inactive_flags(old_active);
1215
1216                 if (new_active) {
1217                         bond_set_slave_active_flags(new_active);
1218
1219                         if (bond->params.fail_over_mac)
1220                                 bond_do_fail_over_mac(bond, new_active,
1221                                                       old_active);
1222
1223                         bond->send_grat_arp = bond->params.num_grat_arp;
1224                         bond_send_gratuitous_arp(bond);
1225
1226                         bond->send_unsol_na = bond->params.num_unsol_na;
1227                         bond_send_unsolicited_na(bond);
1228
1229                         write_unlock_bh(&bond->curr_slave_lock);
1230                         read_unlock(&bond->lock);
1231
1232                         netdev_bonding_change(bond->dev, NETDEV_BONDING_FAILOVER);
1233
1234                         read_lock(&bond->lock);
1235                         write_lock_bh(&bond->curr_slave_lock);
1236                 }
1237         }
1238
1239         /* resend IGMP joins since all were sent on curr_active_slave */
1240         if (bond->params.mode == BOND_MODE_ROUNDROBIN) {
1241                 bond_resend_igmp_join_requests(bond);
1242         }
1243 }
1244
1245 /**
1246  * bond_select_active_slave - select a new active slave, if needed
1247  * @bond: our bonding struct
1248  *
1249  * This functions should be called when one of the following occurs:
1250  * - The old curr_active_slave has been released or lost its link.
1251  * - The primary_slave has got its link back.
1252  * - A slave has got its link back and there's no old curr_active_slave.
1253  *
1254  * Caller must hold bond->lock for read and curr_slave_lock for write_bh.
1255  */
1256 void bond_select_active_slave(struct bonding *bond)
1257 {
1258         struct slave *best_slave;
1259         int rv;
1260
1261         best_slave = bond_find_best_slave(bond);
1262         if (best_slave != bond->curr_active_slave) {
1263                 bond_change_active_slave(bond, best_slave);
1264                 rv = bond_set_carrier(bond);
1265                 if (!rv)
1266                         return;
1267
1268                 if (netif_carrier_ok(bond->dev)) {
1269                         pr_info("%s: first active interface up!\n",
1270                                 bond->dev->name);
1271                 } else {
1272                         pr_info("%s: now running without any active interface !\n",
1273                                 bond->dev->name);
1274                 }
1275         }
1276 }
1277
1278 /*--------------------------- slave list handling ---------------------------*/
1279
1280 /*
1281  * This function attaches the slave to the end of list.
1282  *
1283  * bond->lock held for writing by caller.
1284  */
1285 static void bond_attach_slave(struct bonding *bond, struct slave *new_slave)
1286 {
1287         if (bond->first_slave == NULL) { /* attaching the first slave */
1288                 new_slave->next = new_slave;
1289                 new_slave->prev = new_slave;
1290                 bond->first_slave = new_slave;
1291         } else {
1292                 new_slave->next = bond->first_slave;
1293                 new_slave->prev = bond->first_slave->prev;
1294                 new_slave->next->prev = new_slave;
1295                 new_slave->prev->next = new_slave;
1296         }
1297
1298         bond->slave_cnt++;
1299 }
1300
1301 /*
1302  * This function detaches the slave from the list.
1303  * WARNING: no check is made to verify if the slave effectively
1304  * belongs to <bond>.
1305  * Nothing is freed on return, structures are just unchained.
1306  * If any slave pointer in bond was pointing to <slave>,
1307  * it should be changed by the calling function.
1308  *
1309  * bond->lock held for writing by caller.
1310  */
1311 static void bond_detach_slave(struct bonding *bond, struct slave *slave)
1312 {
1313         if (slave->next)
1314                 slave->next->prev = slave->prev;
1315
1316         if (slave->prev)
1317                 slave->prev->next = slave->next;
1318
1319         if (bond->first_slave == slave) { /* slave is the first slave */
1320                 if (bond->slave_cnt > 1) { /* there are more slave */
1321                         bond->first_slave = slave->next;
1322                 } else {
1323                         bond->first_slave = NULL; /* slave was the last one */
1324                 }
1325         }
1326
1327         slave->next = NULL;
1328         slave->prev = NULL;
1329         bond->slave_cnt--;
1330 }
1331
1332 /*---------------------------------- IOCTL ----------------------------------*/
1333
1334 static int bond_sethwaddr(struct net_device *bond_dev,
1335                           struct net_device *slave_dev)
1336 {
1337         pr_debug("bond_dev=%p\n", bond_dev);
1338         pr_debug("slave_dev=%p\n", slave_dev);
1339         pr_debug("slave_dev->addr_len=%d\n", slave_dev->addr_len);
1340         memcpy(bond_dev->dev_addr, slave_dev->dev_addr, slave_dev->addr_len);
1341         return 0;
1342 }
1343
1344 #define BOND_VLAN_FEATURES \
1345         (NETIF_F_VLAN_CHALLENGED | NETIF_F_HW_VLAN_RX | NETIF_F_HW_VLAN_TX | \
1346          NETIF_F_HW_VLAN_FILTER)
1347
1348 /*
1349  * Compute the common dev->feature set available to all slaves.  Some
1350  * feature bits are managed elsewhere, so preserve those feature bits
1351  * on the master device.
1352  */
1353 static int bond_compute_features(struct bonding *bond)
1354 {
1355         struct slave *slave;
1356         struct net_device *bond_dev = bond->dev;
1357         unsigned long features = bond_dev->features;
1358         unsigned long vlan_features = 0;
1359         unsigned short max_hard_header_len = max((u16)ETH_HLEN,
1360                                                 bond_dev->hard_header_len);
1361         int i;
1362
1363         features &= ~(NETIF_F_ALL_CSUM | BOND_VLAN_FEATURES);
1364         features |=  NETIF_F_GSO_MASK | NETIF_F_NO_CSUM;
1365
1366         if (!bond->first_slave)
1367                 goto done;
1368
1369         features &= ~NETIF_F_ONE_FOR_ALL;
1370
1371         vlan_features = bond->first_slave->dev->vlan_features;
1372         bond_for_each_slave(bond, slave, i) {
1373                 features = netdev_increment_features(features,
1374                                                      slave->dev->features,
1375                                                      NETIF_F_ONE_FOR_ALL);
1376                 vlan_features = netdev_increment_features(vlan_features,
1377                                                         slave->dev->vlan_features,
1378                                                         NETIF_F_ONE_FOR_ALL);
1379                 if (slave->dev->hard_header_len > max_hard_header_len)
1380                         max_hard_header_len = slave->dev->hard_header_len;
1381         }
1382
1383 done:
1384         features |= (bond_dev->features & BOND_VLAN_FEATURES);
1385         bond_dev->features = netdev_fix_features(features, NULL);
1386         bond_dev->vlan_features = netdev_fix_features(vlan_features, NULL);
1387         bond_dev->hard_header_len = max_hard_header_len;
1388
1389         return 0;
1390 }
1391
1392 static void bond_setup_by_slave(struct net_device *bond_dev,
1393                                 struct net_device *slave_dev)
1394 {
1395         struct bonding *bond = netdev_priv(bond_dev);
1396
1397         bond_dev->header_ops        = slave_dev->header_ops;
1398
1399         bond_dev->type              = slave_dev->type;
1400         bond_dev->hard_header_len   = slave_dev->hard_header_len;
1401         bond_dev->addr_len          = slave_dev->addr_len;
1402
1403         memcpy(bond_dev->broadcast, slave_dev->broadcast,
1404                 slave_dev->addr_len);
1405         bond->setup_by_slave = 1;
1406 }
1407
1408 /* enslave device <slave> to bond device <master> */
1409 int bond_enslave(struct net_device *bond_dev, struct net_device *slave_dev)
1410 {
1411         struct bonding *bond = netdev_priv(bond_dev);
1412         const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
1413         struct slave *new_slave = NULL;
1414         struct dev_mc_list *dmi;
1415         struct sockaddr addr;
1416         int link_reporting;
1417         int old_features = bond_dev->features;
1418         int res = 0;
1419
1420         if (!bond->params.use_carrier && slave_dev->ethtool_ops == NULL &&
1421                 slave_ops->ndo_do_ioctl == NULL) {
1422                 pr_warning("%s: Warning: no link monitoring support for %s\n",
1423                            bond_dev->name, slave_dev->name);
1424         }
1425
1426         /* bond must be initialized by bond_open() before enslaving */
1427         if (!(bond_dev->flags & IFF_UP)) {
1428                 pr_warning("%s: master_dev is not up in bond_enslave\n",
1429                            bond_dev->name);
1430         }
1431
1432         /* already enslaved */
1433         if (slave_dev->flags & IFF_SLAVE) {
1434                 pr_debug("Error, Device was already enslaved\n");
1435                 return -EBUSY;
1436         }
1437
1438         /* vlan challenged mutual exclusion */
1439         /* no need to lock since we're protected by rtnl_lock */
1440         if (slave_dev->features & NETIF_F_VLAN_CHALLENGED) {
1441                 pr_debug("%s: NETIF_F_VLAN_CHALLENGED\n", slave_dev->name);
1442                 if (!list_empty(&bond->vlan_list)) {
1443                         pr_err("%s: Error: cannot enslave VLAN challenged slave %s on VLAN enabled bond %s\n",
1444                                bond_dev->name, slave_dev->name, bond_dev->name);
1445                         return -EPERM;
1446                 } else {
1447                         pr_warning("%s: Warning: enslaved VLAN challenged slave %s. Adding VLANs will be blocked as long as %s is part of bond %s\n",
1448                                    bond_dev->name, slave_dev->name,
1449                                    slave_dev->name, bond_dev->name);
1450                         bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
1451                 }
1452         } else {
1453                 pr_debug("%s: ! NETIF_F_VLAN_CHALLENGED\n", slave_dev->name);
1454                 if (bond->slave_cnt == 0) {
1455                         /* First slave, and it is not VLAN challenged,
1456                          * so remove the block of adding VLANs over the bond.
1457                          */
1458                         bond_dev->features &= ~NETIF_F_VLAN_CHALLENGED;
1459                 }
1460         }
1461
1462         /*
1463          * Old ifenslave binaries are no longer supported.  These can
1464          * be identified with moderate accuracy by the state of the slave:
1465          * the current ifenslave will set the interface down prior to
1466          * enslaving it; the old ifenslave will not.
1467          */
1468         if ((slave_dev->flags & IFF_UP)) {
1469                 pr_err("%s is up. This may be due to an out of date ifenslave.\n",
1470                        slave_dev->name);
1471                 res = -EPERM;
1472                 goto err_undo_flags;
1473         }
1474
1475         /* set bonding device ether type by slave - bonding netdevices are
1476          * created with ether_setup, so when the slave type is not ARPHRD_ETHER
1477          * there is a need to override some of the type dependent attribs/funcs.
1478          *
1479          * bond ether type mutual exclusion - don't allow slaves of dissimilar
1480          * ether type (eg ARPHRD_ETHER and ARPHRD_INFINIBAND) share the same bond
1481          */
1482         if (bond->slave_cnt == 0) {
1483                 if (bond_dev->type != slave_dev->type) {
1484                         pr_debug("%s: change device type from %d to %d\n",
1485                                  bond_dev->name,
1486                                  bond_dev->type, slave_dev->type);
1487
1488                         netdev_bonding_change(bond_dev, NETDEV_BONDING_OLDTYPE);
1489
1490                         if (slave_dev->type != ARPHRD_ETHER)
1491                                 bond_setup_by_slave(bond_dev, slave_dev);
1492                         else
1493                                 ether_setup(bond_dev);
1494
1495                         netdev_bonding_change(bond_dev, NETDEV_BONDING_NEWTYPE);
1496                 }
1497         } else if (bond_dev->type != slave_dev->type) {
1498                 pr_err("%s ether type (%d) is different from other slaves (%d), can not enslave it.\n",
1499                        slave_dev->name,
1500                        slave_dev->type, bond_dev->type);
1501                 res = -EINVAL;
1502                 goto err_undo_flags;
1503         }
1504
1505         if (slave_ops->ndo_set_mac_address == NULL) {
1506                 if (bond->slave_cnt == 0) {
1507                         pr_warning("%s: Warning: The first slave device specified does not support setting the MAC address. Setting fail_over_mac to active.",
1508                                    bond_dev->name);
1509                         bond->params.fail_over_mac = BOND_FOM_ACTIVE;
1510                 } else if (bond->params.fail_over_mac != BOND_FOM_ACTIVE) {
1511                         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",
1512                                bond_dev->name);
1513                         res = -EOPNOTSUPP;
1514                         goto err_undo_flags;
1515                 }
1516         }
1517
1518         new_slave = kzalloc(sizeof(struct slave), GFP_KERNEL);
1519         if (!new_slave) {
1520                 res = -ENOMEM;
1521                 goto err_undo_flags;
1522         }
1523
1524         /* save slave's original flags before calling
1525          * netdev_set_master and dev_open
1526          */
1527         new_slave->original_flags = slave_dev->flags;
1528
1529         /*
1530          * Save slave's original ("permanent") mac address for modes
1531          * that need it, and for restoring it upon release, and then
1532          * set it to the master's address
1533          */
1534         memcpy(new_slave->perm_hwaddr, slave_dev->dev_addr, ETH_ALEN);
1535
1536         if (!bond->params.fail_over_mac) {
1537                 /*
1538                  * Set slave to master's mac address.  The application already
1539                  * set the master's mac address to that of the first slave
1540                  */
1541                 memcpy(addr.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
1542                 addr.sa_family = slave_dev->type;
1543                 res = dev_set_mac_address(slave_dev, &addr);
1544                 if (res) {
1545                         pr_debug("Error %d calling set_mac_address\n", res);
1546                         goto err_free;
1547                 }
1548         }
1549
1550         res = netdev_set_master(slave_dev, bond_dev);
1551         if (res) {
1552                 pr_debug("Error %d calling netdev_set_master\n", res);
1553                 goto err_restore_mac;
1554         }
1555         /* open the slave since the application closed it */
1556         res = dev_open(slave_dev);
1557         if (res) {
1558                 pr_debug("Opening slave %s failed\n", slave_dev->name);
1559                 goto err_unset_master;
1560         }
1561
1562         new_slave->dev = slave_dev;
1563         slave_dev->priv_flags |= IFF_BONDING;
1564
1565         if (bond_is_lb(bond)) {
1566                 /* bond_alb_init_slave() must be called before all other stages since
1567                  * it might fail and we do not want to have to undo everything
1568                  */
1569                 res = bond_alb_init_slave(bond, new_slave);
1570                 if (res)
1571                         goto err_close;
1572         }
1573
1574         /* If the mode USES_PRIMARY, then the new slave gets the
1575          * master's promisc (and mc) settings only if it becomes the
1576          * curr_active_slave, and that is taken care of later when calling
1577          * bond_change_active()
1578          */
1579         if (!USES_PRIMARY(bond->params.mode)) {
1580                 /* set promiscuity level to new slave */
1581                 if (bond_dev->flags & IFF_PROMISC) {
1582                         res = dev_set_promiscuity(slave_dev, 1);
1583                         if (res)
1584                                 goto err_close;
1585                 }
1586
1587                 /* set allmulti level to new slave */
1588                 if (bond_dev->flags & IFF_ALLMULTI) {
1589                         res = dev_set_allmulti(slave_dev, 1);
1590                         if (res)
1591                                 goto err_close;
1592                 }
1593
1594                 netif_addr_lock_bh(bond_dev);
1595                 /* upload master's mc_list to new slave */
1596                 for (dmi = bond_dev->mc_list; dmi; dmi = dmi->next)
1597                         dev_mc_add(slave_dev, dmi->dmi_addr,
1598                                    dmi->dmi_addrlen, 0);
1599                 netif_addr_unlock_bh(bond_dev);
1600         }
1601
1602         if (bond->params.mode == BOND_MODE_8023AD) {
1603                 /* add lacpdu mc addr to mc list */
1604                 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
1605
1606                 dev_mc_add(slave_dev, lacpdu_multicast, ETH_ALEN, 0);
1607         }
1608
1609         bond_add_vlans_on_slave(bond, slave_dev);
1610
1611         write_lock_bh(&bond->lock);
1612
1613         bond_attach_slave(bond, new_slave);
1614
1615         new_slave->delay = 0;
1616         new_slave->link_failure_count = 0;
1617
1618         bond_compute_features(bond);
1619
1620         write_unlock_bh(&bond->lock);
1621
1622         read_lock(&bond->lock);
1623
1624         new_slave->last_arp_rx = jiffies;
1625
1626         if (bond->params.miimon && !bond->params.use_carrier) {
1627                 link_reporting = bond_check_dev_link(bond, slave_dev, 1);
1628
1629                 if ((link_reporting == -1) && !bond->params.arp_interval) {
1630                         /*
1631                          * miimon is set but a bonded network driver
1632                          * does not support ETHTOOL/MII and
1633                          * arp_interval is not set.  Note: if
1634                          * use_carrier is enabled, we will never go
1635                          * here (because netif_carrier is always
1636                          * supported); thus, we don't need to change
1637                          * the messages for netif_carrier.
1638                          */
1639                         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",
1640                                bond_dev->name, slave_dev->name);
1641                 } else if (link_reporting == -1) {
1642                         /* unable get link status using mii/ethtool */
1643                         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",
1644                                    bond_dev->name, slave_dev->name);
1645                 }
1646         }
1647
1648         /* check for initial state */
1649         if (!bond->params.miimon ||
1650             (bond_check_dev_link(bond, slave_dev, 0) == BMSR_LSTATUS)) {
1651                 if (bond->params.updelay) {
1652                         pr_debug("Initial state of slave_dev is BOND_LINK_BACK\n");
1653                         new_slave->link  = BOND_LINK_BACK;
1654                         new_slave->delay = bond->params.updelay;
1655                 } else {
1656                         pr_debug("Initial state of slave_dev is BOND_LINK_UP\n");
1657                         new_slave->link  = BOND_LINK_UP;
1658                 }
1659                 new_slave->jiffies = jiffies;
1660         } else {
1661                 pr_debug("Initial state of slave_dev is BOND_LINK_DOWN\n");
1662                 new_slave->link  = BOND_LINK_DOWN;
1663         }
1664
1665         if (bond_update_speed_duplex(new_slave) &&
1666             (new_slave->link != BOND_LINK_DOWN)) {
1667                 pr_warning("%s: Warning: failed to get speed and duplex from %s, assumed to be 100Mb/sec and Full.\n",
1668                            bond_dev->name, new_slave->dev->name);
1669
1670                 if (bond->params.mode == BOND_MODE_8023AD) {
1671                         pr_warning("%s: Warning: Operation of 802.3ad mode requires ETHTOOL support in base driver for proper aggregator selection.\n",
1672                                    bond_dev->name);
1673                 }
1674         }
1675
1676         if (USES_PRIMARY(bond->params.mode) && bond->params.primary[0]) {
1677                 /* if there is a primary slave, remember it */
1678                 if (strcmp(bond->params.primary, new_slave->dev->name) == 0) {
1679                         bond->primary_slave = new_slave;
1680                         bond->force_primary = true;
1681                 }
1682         }
1683
1684         write_lock_bh(&bond->curr_slave_lock);
1685
1686         switch (bond->params.mode) {
1687         case BOND_MODE_ACTIVEBACKUP:
1688                 bond_set_slave_inactive_flags(new_slave);
1689                 bond_select_active_slave(bond);
1690                 break;
1691         case BOND_MODE_8023AD:
1692                 /* in 802.3ad mode, the internal mechanism
1693                  * will activate the slaves in the selected
1694                  * aggregator
1695                  */
1696                 bond_set_slave_inactive_flags(new_slave);
1697                 /* if this is the first slave */
1698                 if (bond->slave_cnt == 1) {
1699                         SLAVE_AD_INFO(new_slave).id = 1;
1700                         /* Initialize AD with the number of times that the AD timer is called in 1 second
1701                          * can be called only after the mac address of the bond is set
1702                          */
1703                         bond_3ad_initialize(bond, 1000/AD_TIMER_INTERVAL,
1704                                             bond->params.lacp_fast);
1705                 } else {
1706                         SLAVE_AD_INFO(new_slave).id =
1707                                 SLAVE_AD_INFO(new_slave->prev).id + 1;
1708                 }
1709
1710                 bond_3ad_bind_slave(new_slave);
1711                 break;
1712         case BOND_MODE_TLB:
1713         case BOND_MODE_ALB:
1714                 new_slave->state = BOND_STATE_ACTIVE;
1715                 bond_set_slave_inactive_flags(new_slave);
1716                 bond_select_active_slave(bond);
1717                 break;
1718         default:
1719                 pr_debug("This slave is always active in trunk mode\n");
1720
1721                 /* always active in trunk mode */
1722                 new_slave->state = BOND_STATE_ACTIVE;
1723
1724                 /* In trunking mode there is little meaning to curr_active_slave
1725                  * anyway (it holds no special properties of the bond device),
1726                  * so we can change it without calling change_active_interface()
1727                  */
1728                 if (!bond->curr_active_slave)
1729                         bond->curr_active_slave = new_slave;
1730
1731                 break;
1732         } /* switch(bond_mode) */
1733
1734         write_unlock_bh(&bond->curr_slave_lock);
1735
1736         bond_set_carrier(bond);
1737
1738         read_unlock(&bond->lock);
1739
1740         res = bond_create_slave_symlinks(bond_dev, slave_dev);
1741         if (res)
1742                 goto err_close;
1743
1744         pr_info("%s: enslaving %s as a%s interface with a%s link.\n",
1745                 bond_dev->name, slave_dev->name,
1746                 new_slave->state == BOND_STATE_ACTIVE ? "n active" : " backup",
1747                 new_slave->link != BOND_LINK_DOWN ? "n up" : " down");
1748
1749         /* enslave is successful */
1750         return 0;
1751
1752 /* Undo stages on error */
1753 err_close:
1754         dev_close(slave_dev);
1755
1756 err_unset_master:
1757         netdev_set_master(slave_dev, NULL);
1758
1759 err_restore_mac:
1760         if (!bond->params.fail_over_mac) {
1761                 /* XXX TODO - fom follow mode needs to change master's
1762                  * MAC if this slave's MAC is in use by the bond, or at
1763                  * least print a warning.
1764                  */
1765                 memcpy(addr.sa_data, new_slave->perm_hwaddr, ETH_ALEN);
1766                 addr.sa_family = slave_dev->type;
1767                 dev_set_mac_address(slave_dev, &addr);
1768         }
1769
1770 err_free:
1771         kfree(new_slave);
1772
1773 err_undo_flags:
1774         bond_dev->features = old_features;
1775
1776         return res;
1777 }
1778
1779 /*
1780  * Try to release the slave device <slave> from the bond device <master>
1781  * It is legal to access curr_active_slave without a lock because all the function
1782  * is write-locked.
1783  *
1784  * The rules for slave state should be:
1785  *   for Active/Backup:
1786  *     Active stays on all backups go down
1787  *   for Bonded connections:
1788  *     The first up interface should be left on and all others downed.
1789  */
1790 int bond_release(struct net_device *bond_dev, struct net_device *slave_dev)
1791 {
1792         struct bonding *bond = netdev_priv(bond_dev);
1793         struct slave *slave, *oldcurrent;
1794         struct sockaddr addr;
1795
1796         /* slave is not a slave or master is not master of this slave */
1797         if (!(slave_dev->flags & IFF_SLAVE) ||
1798             (slave_dev->master != bond_dev)) {
1799                 pr_err("%s: Error: cannot release %s.\n",
1800                        bond_dev->name, slave_dev->name);
1801                 return -EINVAL;
1802         }
1803
1804         write_lock_bh(&bond->lock);
1805
1806         slave = bond_get_slave_by_dev(bond, slave_dev);
1807         if (!slave) {
1808                 /* not a slave of this bond */
1809                 pr_info("%s: %s not enslaved\n",
1810                         bond_dev->name, slave_dev->name);
1811                 write_unlock_bh(&bond->lock);
1812                 return -EINVAL;
1813         }
1814
1815         if (!bond->params.fail_over_mac) {
1816                 if (!compare_ether_addr(bond_dev->dev_addr, slave->perm_hwaddr) &&
1817                     bond->slave_cnt > 1)
1818                         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",
1819                                    bond_dev->name, slave_dev->name,
1820                                    slave->perm_hwaddr,
1821                                    bond_dev->name, slave_dev->name);
1822         }
1823
1824         /* Inform AD package of unbinding of slave. */
1825         if (bond->params.mode == BOND_MODE_8023AD) {
1826                 /* must be called before the slave is
1827                  * detached from the list
1828                  */
1829                 bond_3ad_unbind_slave(slave);
1830         }
1831
1832         pr_info("%s: releasing %s interface %s\n",
1833                 bond_dev->name,
1834                 (slave->state == BOND_STATE_ACTIVE) ? "active" : "backup",
1835                 slave_dev->name);
1836
1837         oldcurrent = bond->curr_active_slave;
1838
1839         bond->current_arp_slave = NULL;
1840
1841         /* release the slave from its bond */
1842         bond_detach_slave(bond, slave);
1843
1844         bond_compute_features(bond);
1845
1846         if (bond->primary_slave == slave)
1847                 bond->primary_slave = NULL;
1848
1849         if (oldcurrent == slave)
1850                 bond_change_active_slave(bond, NULL);
1851
1852         if (bond_is_lb(bond)) {
1853                 /* Must be called only after the slave has been
1854                  * detached from the list and the curr_active_slave
1855                  * has been cleared (if our_slave == old_current),
1856                  * but before a new active slave is selected.
1857                  */
1858                 write_unlock_bh(&bond->lock);
1859                 bond_alb_deinit_slave(bond, slave);
1860                 write_lock_bh(&bond->lock);
1861         }
1862
1863         if (oldcurrent == slave) {
1864                 /*
1865                  * Note that we hold RTNL over this sequence, so there
1866                  * is no concern that another slave add/remove event
1867                  * will interfere.
1868                  */
1869                 write_unlock_bh(&bond->lock);
1870                 read_lock(&bond->lock);
1871                 write_lock_bh(&bond->curr_slave_lock);
1872
1873                 bond_select_active_slave(bond);
1874
1875                 write_unlock_bh(&bond->curr_slave_lock);
1876                 read_unlock(&bond->lock);
1877                 write_lock_bh(&bond->lock);
1878         }
1879
1880         if (bond->slave_cnt == 0) {
1881                 bond_set_carrier(bond);
1882
1883                 /* if the last slave was removed, zero the mac address
1884                  * of the master so it will be set by the application
1885                  * to the mac address of the first slave
1886                  */
1887                 memset(bond_dev->dev_addr, 0, bond_dev->addr_len);
1888
1889                 if (list_empty(&bond->vlan_list)) {
1890                         bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
1891                 } else {
1892                         pr_warning("%s: Warning: clearing HW address of %s while it still has VLANs.\n",
1893                                    bond_dev->name, bond_dev->name);
1894                         pr_warning("%s: When re-adding slaves, make sure the bond's HW address matches its VLANs'.\n",
1895                                    bond_dev->name);
1896                 }
1897         } else if ((bond_dev->features & NETIF_F_VLAN_CHALLENGED) &&
1898                    !bond_has_challenged_slaves(bond)) {
1899                 pr_info("%s: last VLAN challenged slave %s left bond %s. VLAN blocking is removed\n",
1900                         bond_dev->name, slave_dev->name, bond_dev->name);
1901                 bond_dev->features &= ~NETIF_F_VLAN_CHALLENGED;
1902         }
1903
1904         write_unlock_bh(&bond->lock);
1905
1906         /* must do this from outside any spinlocks */
1907         bond_destroy_slave_symlinks(bond_dev, slave_dev);
1908
1909         bond_del_vlans_from_slave(bond, slave_dev);
1910
1911         /* If the mode USES_PRIMARY, then we should only remove its
1912          * promisc and mc settings if it was the curr_active_slave, but that was
1913          * already taken care of above when we detached the slave
1914          */
1915         if (!USES_PRIMARY(bond->params.mode)) {
1916                 /* unset promiscuity level from slave */
1917                 if (bond_dev->flags & IFF_PROMISC)
1918                         dev_set_promiscuity(slave_dev, -1);
1919
1920                 /* unset allmulti level from slave */
1921                 if (bond_dev->flags & IFF_ALLMULTI)
1922                         dev_set_allmulti(slave_dev, -1);
1923
1924                 /* flush master's mc_list from slave */
1925                 netif_addr_lock_bh(bond_dev);
1926                 bond_mc_list_flush(bond_dev, slave_dev);
1927                 netif_addr_unlock_bh(bond_dev);
1928         }
1929
1930         netdev_set_master(slave_dev, NULL);
1931
1932         /* close slave before restoring its mac address */
1933         dev_close(slave_dev);
1934
1935         if (bond->params.fail_over_mac != BOND_FOM_ACTIVE) {
1936                 /* restore original ("permanent") mac address */
1937                 memcpy(addr.sa_data, slave->perm_hwaddr, ETH_ALEN);
1938                 addr.sa_family = slave_dev->type;
1939                 dev_set_mac_address(slave_dev, &addr);
1940         }
1941
1942         slave_dev->priv_flags &= ~(IFF_MASTER_8023AD | IFF_MASTER_ALB |
1943                                    IFF_SLAVE_INACTIVE | IFF_BONDING |
1944                                    IFF_SLAVE_NEEDARP);
1945
1946         kfree(slave);
1947
1948         return 0;  /* deletion OK */
1949 }
1950
1951 /*
1952 * First release a slave and than destroy the bond if no more slaves are left.
1953 * Must be under rtnl_lock when this function is called.
1954 */
1955 int  bond_release_and_destroy(struct net_device *bond_dev,
1956                               struct net_device *slave_dev)
1957 {
1958         struct bonding *bond = netdev_priv(bond_dev);
1959         int ret;
1960
1961         ret = bond_release(bond_dev, slave_dev);
1962         if ((ret == 0) && (bond->slave_cnt == 0)) {
1963                 pr_info("%s: destroying bond %s.\n",
1964                         bond_dev->name, bond_dev->name);
1965                 unregister_netdevice(bond_dev);
1966         }
1967         return ret;
1968 }
1969
1970 /*
1971  * This function releases all slaves.
1972  */
1973 static int bond_release_all(struct net_device *bond_dev)
1974 {
1975         struct bonding *bond = netdev_priv(bond_dev);
1976         struct slave *slave;
1977         struct net_device *slave_dev;
1978         struct sockaddr addr;
1979
1980         write_lock_bh(&bond->lock);
1981
1982         netif_carrier_off(bond_dev);
1983
1984         if (bond->slave_cnt == 0)
1985                 goto out;
1986
1987         bond->current_arp_slave = NULL;
1988         bond->primary_slave = NULL;
1989         bond_change_active_slave(bond, NULL);
1990
1991         while ((slave = bond->first_slave) != NULL) {
1992                 /* Inform AD package of unbinding of slave
1993                  * before slave is detached from the list.
1994                  */
1995                 if (bond->params.mode == BOND_MODE_8023AD)
1996                         bond_3ad_unbind_slave(slave);
1997
1998                 slave_dev = slave->dev;
1999                 bond_detach_slave(bond, slave);
2000
2001                 /* now that the slave is detached, unlock and perform
2002                  * all the undo steps that should not be called from
2003                  * within a lock.
2004                  */
2005                 write_unlock_bh(&bond->lock);
2006
2007                 if (bond_is_lb(bond)) {
2008                         /* must be called only after the slave
2009                          * has been detached from the list
2010                          */
2011                         bond_alb_deinit_slave(bond, slave);
2012                 }
2013
2014                 bond_compute_features(bond);
2015
2016                 bond_destroy_slave_symlinks(bond_dev, slave_dev);
2017                 bond_del_vlans_from_slave(bond, slave_dev);
2018
2019                 /* If the mode USES_PRIMARY, then we should only remove its
2020                  * promisc and mc settings if it was the curr_active_slave, but that was
2021                  * already taken care of above when we detached the slave
2022                  */
2023                 if (!USES_PRIMARY(bond->params.mode)) {
2024                         /* unset promiscuity level from slave */
2025                         if (bond_dev->flags & IFF_PROMISC)
2026                                 dev_set_promiscuity(slave_dev, -1);
2027
2028                         /* unset allmulti level from slave */
2029                         if (bond_dev->flags & IFF_ALLMULTI)
2030                                 dev_set_allmulti(slave_dev, -1);
2031
2032                         /* flush master's mc_list from slave */
2033                         netif_addr_lock_bh(bond_dev);
2034                         bond_mc_list_flush(bond_dev, slave_dev);
2035                         netif_addr_unlock_bh(bond_dev);
2036                 }
2037
2038                 netdev_set_master(slave_dev, NULL);
2039
2040                 /* close slave before restoring its mac address */
2041                 dev_close(slave_dev);
2042
2043                 if (!bond->params.fail_over_mac) {
2044                         /* restore original ("permanent") mac address*/
2045                         memcpy(addr.sa_data, slave->perm_hwaddr, ETH_ALEN);
2046                         addr.sa_family = slave_dev->type;
2047                         dev_set_mac_address(slave_dev, &addr);
2048                 }
2049
2050                 slave_dev->priv_flags &= ~(IFF_MASTER_8023AD | IFF_MASTER_ALB |
2051                                            IFF_SLAVE_INACTIVE);
2052
2053                 kfree(slave);
2054
2055                 /* re-acquire the lock before getting the next slave */
2056                 write_lock_bh(&bond->lock);
2057         }
2058
2059         /* zero the mac address of the master so it will be
2060          * set by the application to the mac address of the
2061          * first slave
2062          */
2063         memset(bond_dev->dev_addr, 0, bond_dev->addr_len);
2064
2065         if (list_empty(&bond->vlan_list))
2066                 bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
2067         else {
2068                 pr_warning("%s: Warning: clearing HW address of %s while it still has VLANs.\n",
2069                            bond_dev->name, bond_dev->name);
2070                 pr_warning("%s: When re-adding slaves, make sure the bond's HW address matches its VLANs'.\n",
2071                            bond_dev->name);
2072         }
2073
2074         pr_info("%s: released all slaves\n", bond_dev->name);
2075
2076 out:
2077         write_unlock_bh(&bond->lock);
2078
2079         return 0;
2080 }
2081
2082 /*
2083  * This function changes the active slave to slave <slave_dev>.
2084  * It returns -EINVAL in the following cases.
2085  *  - <slave_dev> is not found in the list.
2086  *  - There is not active slave now.
2087  *  - <slave_dev> is already active.
2088  *  - The link state of <slave_dev> is not BOND_LINK_UP.
2089  *  - <slave_dev> is not running.
2090  * In these cases, this function does nothing.
2091  * In the other cases, current_slave pointer is changed and 0 is returned.
2092  */
2093 static int bond_ioctl_change_active(struct net_device *bond_dev, struct net_device *slave_dev)
2094 {
2095         struct bonding *bond = netdev_priv(bond_dev);
2096         struct slave *old_active = NULL;
2097         struct slave *new_active = NULL;
2098         int res = 0;
2099
2100         if (!USES_PRIMARY(bond->params.mode))
2101                 return -EINVAL;
2102
2103         /* Verify that master_dev is indeed the master of slave_dev */
2104         if (!(slave_dev->flags & IFF_SLAVE) || (slave_dev->master != bond_dev))
2105                 return -EINVAL;
2106
2107         read_lock(&bond->lock);
2108
2109         read_lock(&bond->curr_slave_lock);
2110         old_active = bond->curr_active_slave;
2111         read_unlock(&bond->curr_slave_lock);
2112
2113         new_active = bond_get_slave_by_dev(bond, slave_dev);
2114
2115         /*
2116          * Changing to the current active: do nothing; return success.
2117          */
2118         if (new_active && (new_active == old_active)) {
2119                 read_unlock(&bond->lock);
2120                 return 0;
2121         }
2122
2123         if ((new_active) &&
2124             (old_active) &&
2125             (new_active->link == BOND_LINK_UP) &&
2126             IS_UP(new_active->dev)) {
2127                 write_lock_bh(&bond->curr_slave_lock);
2128                 bond_change_active_slave(bond, new_active);
2129                 write_unlock_bh(&bond->curr_slave_lock);
2130         } else
2131                 res = -EINVAL;
2132
2133         read_unlock(&bond->lock);
2134
2135         return res;
2136 }
2137
2138 static int bond_info_query(struct net_device *bond_dev, struct ifbond *info)
2139 {
2140         struct bonding *bond = netdev_priv(bond_dev);
2141
2142         info->bond_mode = bond->params.mode;
2143         info->miimon = bond->params.miimon;
2144
2145         read_lock(&bond->lock);
2146         info->num_slaves = bond->slave_cnt;
2147         read_unlock(&bond->lock);
2148
2149         return 0;
2150 }
2151
2152 static int bond_slave_info_query(struct net_device *bond_dev, struct ifslave *info)
2153 {
2154         struct bonding *bond = netdev_priv(bond_dev);
2155         struct slave *slave;
2156         int i, res = -ENODEV;
2157
2158         read_lock(&bond->lock);
2159
2160         bond_for_each_slave(bond, slave, i) {
2161                 if (i == (int)info->slave_id) {
2162                         res = 0;
2163                         strcpy(info->slave_name, slave->dev->name);
2164                         info->link = slave->link;
2165                         info->state = slave->state;
2166                         info->link_failure_count = slave->link_failure_count;
2167                         break;
2168                 }
2169         }
2170
2171         read_unlock(&bond->lock);
2172
2173         return res;
2174 }
2175
2176 /*-------------------------------- Monitoring -------------------------------*/
2177
2178
2179 static int bond_miimon_inspect(struct bonding *bond)
2180 {
2181         struct slave *slave;
2182         int i, link_state, commit = 0;
2183         bool ignore_updelay;
2184
2185         ignore_updelay = !bond->curr_active_slave ? true : false;
2186
2187         bond_for_each_slave(bond, slave, i) {
2188                 slave->new_link = BOND_LINK_NOCHANGE;
2189
2190                 link_state = bond_check_dev_link(bond, slave->dev, 0);
2191
2192                 switch (slave->link) {
2193                 case BOND_LINK_UP:
2194                         if (link_state)
2195                                 continue;
2196
2197                         slave->link = BOND_LINK_FAIL;
2198                         slave->delay = bond->params.downdelay;
2199                         if (slave->delay) {
2200                                 pr_info("%s: link status down for %sinterface %s, disabling it in %d ms.\n",
2201                                         bond->dev->name,
2202                                         (bond->params.mode ==
2203                                          BOND_MODE_ACTIVEBACKUP) ?
2204                                         ((slave->state == BOND_STATE_ACTIVE) ?
2205                                          "active " : "backup ") : "",
2206                                         slave->dev->name,
2207                                         bond->params.downdelay * bond->params.miimon);
2208                         }
2209                         /*FALLTHRU*/
2210                 case BOND_LINK_FAIL:
2211                         if (link_state) {
2212                                 /*
2213                                  * recovered before downdelay expired
2214                                  */
2215                                 slave->link = BOND_LINK_UP;
2216                                 slave->jiffies = jiffies;
2217                                 pr_info("%s: link status up again after %d ms for interface %s.\n",
2218                                         bond->dev->name,
2219                                         (bond->params.downdelay - slave->delay) *
2220                                         bond->params.miimon,
2221                                         slave->dev->name);
2222                                 continue;
2223                         }
2224
2225                         if (slave->delay <= 0) {
2226                                 slave->new_link = BOND_LINK_DOWN;
2227                                 commit++;
2228                                 continue;
2229                         }
2230
2231                         slave->delay--;
2232                         break;
2233
2234                 case BOND_LINK_DOWN:
2235                         if (!link_state)
2236                                 continue;
2237
2238                         slave->link = BOND_LINK_BACK;
2239                         slave->delay = bond->params.updelay;
2240
2241                         if (slave->delay) {
2242                                 pr_info("%s: link status up for interface %s, enabling it in %d ms.\n",
2243                                         bond->dev->name, slave->dev->name,
2244                                         ignore_updelay ? 0 :
2245                                         bond->params.updelay *
2246                                         bond->params.miimon);
2247                         }
2248                         /*FALLTHRU*/
2249                 case BOND_LINK_BACK:
2250                         if (!link_state) {
2251                                 slave->link = BOND_LINK_DOWN;
2252                                 pr_info("%s: link status down again after %d ms for interface %s.\n",
2253                                         bond->dev->name,
2254                                         (bond->params.updelay - slave->delay) *
2255                                         bond->params.miimon,
2256                                         slave->dev->name);
2257
2258                                 continue;
2259                         }
2260
2261                         if (ignore_updelay)
2262                                 slave->delay = 0;
2263
2264                         if (slave->delay <= 0) {
2265                                 slave->new_link = BOND_LINK_UP;
2266                                 commit++;
2267                                 ignore_updelay = false;
2268                                 continue;
2269                         }
2270
2271                         slave->delay--;
2272                         break;
2273                 }
2274         }
2275
2276         return commit;
2277 }
2278
2279 static void bond_miimon_commit(struct bonding *bond)
2280 {
2281         struct slave *slave;
2282         int i;
2283
2284         bond_for_each_slave(bond, slave, i) {
2285                 switch (slave->new_link) {
2286                 case BOND_LINK_NOCHANGE:
2287                         continue;
2288
2289                 case BOND_LINK_UP:
2290                         slave->link = BOND_LINK_UP;
2291                         slave->jiffies = jiffies;
2292
2293                         if (bond->params.mode == BOND_MODE_8023AD) {
2294                                 /* prevent it from being the active one */
2295                                 slave->state = BOND_STATE_BACKUP;
2296                         } else if (bond->params.mode != BOND_MODE_ACTIVEBACKUP) {
2297                                 /* make it immediately active */
2298                                 slave->state = BOND_STATE_ACTIVE;
2299                         } else if (slave != bond->primary_slave) {
2300                                 /* prevent it from being the active one */
2301                                 slave->state = BOND_STATE_BACKUP;
2302                         }
2303
2304                         pr_info("%s: link status definitely up for interface %s.\n",
2305                                 bond->dev->name, slave->dev->name);
2306
2307                         /* notify ad that the link status has changed */
2308                         if (bond->params.mode == BOND_MODE_8023AD)
2309                                 bond_3ad_handle_link_change(slave, BOND_LINK_UP);
2310
2311                         if (bond_is_lb(bond))
2312                                 bond_alb_handle_link_change(bond, slave,
2313                                                             BOND_LINK_UP);
2314
2315                         if (!bond->curr_active_slave ||
2316                             (slave == bond->primary_slave))
2317                                 goto do_failover;
2318
2319                         continue;
2320
2321                 case BOND_LINK_DOWN:
2322                         if (slave->link_failure_count < UINT_MAX)
2323                                 slave->link_failure_count++;
2324
2325                         slave->link = BOND_LINK_DOWN;
2326
2327                         if (bond->params.mode == BOND_MODE_ACTIVEBACKUP ||
2328                             bond->params.mode == BOND_MODE_8023AD)
2329                                 bond_set_slave_inactive_flags(slave);
2330
2331                         pr_info("%s: link status definitely down for interface %s, disabling it\n",
2332                                 bond->dev->name, slave->dev->name);
2333
2334                         if (bond->params.mode == BOND_MODE_8023AD)
2335                                 bond_3ad_handle_link_change(slave,
2336                                                             BOND_LINK_DOWN);
2337
2338                         if (bond_is_lb(bond))
2339                                 bond_alb_handle_link_change(bond, slave,
2340                                                             BOND_LINK_DOWN);
2341
2342                         if (slave == bond->curr_active_slave)
2343                                 goto do_failover;
2344
2345                         continue;
2346
2347                 default:
2348                         pr_err("%s: invalid new link %d on slave %s\n",
2349                                bond->dev->name, slave->new_link,
2350                                slave->dev->name);
2351                         slave->new_link = BOND_LINK_NOCHANGE;
2352
2353                         continue;
2354                 }
2355
2356 do_failover:
2357                 ASSERT_RTNL();
2358                 write_lock_bh(&bond->curr_slave_lock);
2359                 bond_select_active_slave(bond);
2360                 write_unlock_bh(&bond->curr_slave_lock);
2361         }
2362
2363         bond_set_carrier(bond);
2364 }
2365
2366 /*
2367  * bond_mii_monitor
2368  *
2369  * Really a wrapper that splits the mii monitor into two phases: an
2370  * inspection, then (if inspection indicates something needs to be done)
2371  * an acquisition of appropriate locks followed by a commit phase to
2372  * implement whatever link state changes are indicated.
2373  */
2374 void bond_mii_monitor(struct work_struct *work)
2375 {
2376         struct bonding *bond = container_of(work, struct bonding,
2377                                             mii_work.work);
2378
2379         read_lock(&bond->lock);
2380         if (bond->kill_timers)
2381                 goto out;
2382
2383         if (bond->slave_cnt == 0)
2384                 goto re_arm;
2385
2386         if (bond->send_grat_arp) {
2387                 read_lock(&bond->curr_slave_lock);
2388                 bond_send_gratuitous_arp(bond);
2389                 read_unlock(&bond->curr_slave_lock);
2390         }
2391
2392         if (bond->send_unsol_na) {
2393                 read_lock(&bond->curr_slave_lock);
2394                 bond_send_unsolicited_na(bond);
2395                 read_unlock(&bond->curr_slave_lock);
2396         }
2397
2398         if (bond_miimon_inspect(bond)) {
2399                 read_unlock(&bond->lock);
2400                 rtnl_lock();
2401                 read_lock(&bond->lock);
2402
2403                 bond_miimon_commit(bond);
2404
2405                 read_unlock(&bond->lock);
2406                 rtnl_unlock();  /* might sleep, hold no other locks */
2407                 read_lock(&bond->lock);
2408         }
2409
2410 re_arm:
2411         if (bond->params.miimon)
2412                 queue_delayed_work(bond->wq, &bond->mii_work,
2413                                    msecs_to_jiffies(bond->params.miimon));
2414 out:
2415         read_unlock(&bond->lock);
2416 }
2417
2418 static __be32 bond_glean_dev_ip(struct net_device *dev)
2419 {
2420         struct in_device *idev;
2421         struct in_ifaddr *ifa;
2422         __be32 addr = 0;
2423
2424         if (!dev)
2425                 return 0;
2426
2427         rcu_read_lock();
2428         idev = __in_dev_get_rcu(dev);
2429         if (!idev)
2430                 goto out;
2431
2432         ifa = idev->ifa_list;
2433         if (!ifa)
2434                 goto out;
2435
2436         addr = ifa->ifa_local;
2437 out:
2438         rcu_read_unlock();
2439         return addr;
2440 }
2441
2442 static int bond_has_this_ip(struct bonding *bond, __be32 ip)
2443 {
2444         struct vlan_entry *vlan;
2445
2446         if (ip == bond->master_ip)
2447                 return 1;
2448
2449         list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
2450                 if (ip == vlan->vlan_ip)
2451                         return 1;
2452         }
2453
2454         return 0;
2455 }
2456
2457 /*
2458  * We go to the (large) trouble of VLAN tagging ARP frames because
2459  * switches in VLAN mode (especially if ports are configured as
2460  * "native" to a VLAN) might not pass non-tagged frames.
2461  */
2462 static void bond_arp_send(struct net_device *slave_dev, int arp_op, __be32 dest_ip, __be32 src_ip, unsigned short vlan_id)
2463 {
2464         struct sk_buff *skb;
2465
2466         pr_debug("arp %d on slave %s: dst %x src %x vid %d\n", arp_op,
2467                  slave_dev->name, dest_ip, src_ip, vlan_id);
2468
2469         skb = arp_create(arp_op, ETH_P_ARP, dest_ip, slave_dev, src_ip,
2470                          NULL, slave_dev->dev_addr, NULL);
2471
2472         if (!skb) {
2473                 pr_err("ARP packet allocation failed\n");
2474                 return;
2475         }
2476         if (vlan_id) {
2477                 skb = vlan_put_tag(skb, vlan_id);
2478                 if (!skb) {
2479                         pr_err("failed to insert VLAN tag\n");
2480                         return;
2481                 }
2482         }
2483         arp_xmit(skb);
2484 }
2485
2486
2487 static void bond_arp_send_all(struct bonding *bond, struct slave *slave)
2488 {
2489         int i, vlan_id, rv;
2490         __be32 *targets = bond->params.arp_targets;
2491         struct vlan_entry *vlan;
2492         struct net_device *vlan_dev;
2493         struct flowi fl;
2494         struct rtable *rt;
2495
2496         for (i = 0; (i < BOND_MAX_ARP_TARGETS); i++) {
2497                 if (!targets[i])
2498                         break;
2499                 pr_debug("basa: target %x\n", targets[i]);
2500                 if (list_empty(&bond->vlan_list)) {
2501                         pr_debug("basa: empty vlan: arp_send\n");
2502                         bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2503                                       bond->master_ip, 0);
2504                         continue;
2505                 }
2506
2507                 /*
2508                  * If VLANs are configured, we do a route lookup to
2509                  * determine which VLAN interface would be used, so we
2510                  * can tag the ARP with the proper VLAN tag.
2511                  */
2512                 memset(&fl, 0, sizeof(fl));
2513                 fl.fl4_dst = targets[i];
2514                 fl.fl4_tos = RTO_ONLINK;
2515
2516                 rv = ip_route_output_key(dev_net(bond->dev), &rt, &fl);
2517                 if (rv) {
2518                         if (net_ratelimit()) {
2519                                 pr_warning("%s: no route to arp_ip_target %pI4\n",
2520                                            bond->dev->name, &fl.fl4_dst);
2521                         }
2522                         continue;
2523                 }
2524
2525                 /*
2526                  * This target is not on a VLAN
2527                  */
2528                 if (rt->u.dst.dev == bond->dev) {
2529                         ip_rt_put(rt);
2530                         pr_debug("basa: rtdev == bond->dev: arp_send\n");
2531                         bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2532                                       bond->master_ip, 0);
2533                         continue;
2534                 }
2535
2536                 vlan_id = 0;
2537                 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
2538                         vlan_dev = vlan_group_get_device(bond->vlgrp, vlan->vlan_id);
2539                         if (vlan_dev == rt->u.dst.dev) {
2540                                 vlan_id = vlan->vlan_id;
2541                                 pr_debug("basa: vlan match on %s %d\n",
2542                                        vlan_dev->name, vlan_id);
2543                                 break;
2544                         }
2545                 }
2546
2547                 if (vlan_id) {
2548                         ip_rt_put(rt);
2549                         bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2550                                       vlan->vlan_ip, vlan_id);
2551                         continue;
2552                 }
2553
2554                 if (net_ratelimit()) {
2555                         pr_warning("%s: no path to arp_ip_target %pI4 via rt.dev %s\n",
2556                                    bond->dev->name, &fl.fl4_dst,
2557                                    rt->u.dst.dev ? rt->u.dst.dev->name : "NULL");
2558                 }
2559                 ip_rt_put(rt);
2560         }
2561 }
2562
2563 /*
2564  * Kick out a gratuitous ARP for an IP on the bonding master plus one
2565  * for each VLAN above us.
2566  *
2567  * Caller must hold curr_slave_lock for read or better
2568  */
2569 static void bond_send_gratuitous_arp(struct bonding *bond)
2570 {
2571         struct slave *slave = bond->curr_active_slave;
2572         struct vlan_entry *vlan;
2573         struct net_device *vlan_dev;
2574
2575         pr_debug("bond_send_grat_arp: bond %s slave %s\n",
2576                  bond->dev->name, slave ? slave->dev->name : "NULL");
2577
2578         if (!slave || !bond->send_grat_arp ||
2579             test_bit(__LINK_STATE_LINKWATCH_PENDING, &slave->dev->state))
2580                 return;
2581
2582         bond->send_grat_arp--;
2583
2584         if (bond->master_ip) {
2585                 bond_arp_send(slave->dev, ARPOP_REPLY, bond->master_ip,
2586                                 bond->master_ip, 0);
2587         }
2588
2589         list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
2590                 vlan_dev = vlan_group_get_device(bond->vlgrp, vlan->vlan_id);
2591                 if (vlan->vlan_ip) {
2592                         bond_arp_send(slave->dev, ARPOP_REPLY, vlan->vlan_ip,
2593                                       vlan->vlan_ip, vlan->vlan_id);
2594                 }
2595         }
2596 }
2597
2598 static void bond_validate_arp(struct bonding *bond, struct slave *slave, __be32 sip, __be32 tip)
2599 {
2600         int i;
2601         __be32 *targets = bond->params.arp_targets;
2602
2603         for (i = 0; (i < BOND_MAX_ARP_TARGETS) && targets[i]; i++) {
2604                 pr_debug("bva: sip %pI4 tip %pI4 t[%d] %pI4 bhti(tip) %d\n",
2605                          &sip, &tip, i, &targets[i],
2606                          bond_has_this_ip(bond, tip));
2607                 if (sip == targets[i]) {
2608                         if (bond_has_this_ip(bond, tip))
2609                                 slave->last_arp_rx = jiffies;
2610                         return;
2611                 }
2612         }
2613 }
2614
2615 static int bond_arp_rcv(struct sk_buff *skb, struct net_device *dev, struct packet_type *pt, struct net_device *orig_dev)
2616 {
2617         struct arphdr *arp;
2618         struct slave *slave;
2619         struct bonding *bond;
2620         unsigned char *arp_ptr;
2621         __be32 sip, tip;
2622
2623         if (dev->priv_flags & IFF_802_1Q_VLAN) {
2624                 /*
2625                  * When using VLANS and bonding, dev and oriv_dev may be
2626                  * incorrect if the physical interface supports VLAN
2627                  * acceleration.  With this change ARP validation now
2628                  * works for hosts only reachable on the VLAN interface.
2629                  */
2630                 dev = vlan_dev_real_dev(dev);
2631                 orig_dev = dev_get_by_index_rcu(dev_net(skb->dev),skb->skb_iif);
2632         }
2633
2634         if (!(dev->priv_flags & IFF_BONDING) || !(dev->flags & IFF_MASTER))
2635                 goto out;
2636
2637         bond = netdev_priv(dev);
2638         read_lock(&bond->lock);
2639
2640         pr_debug("bond_arp_rcv: bond %s skb->dev %s orig_dev %s\n",
2641                  bond->dev->name, skb->dev ? skb->dev->name : "NULL",
2642                  orig_dev ? orig_dev->name : "NULL");
2643
2644         slave = bond_get_slave_by_dev(bond, orig_dev);
2645         if (!slave || !slave_do_arp_validate(bond, slave))
2646                 goto out_unlock;
2647
2648         if (!pskb_may_pull(skb, arp_hdr_len(dev)))
2649                 goto out_unlock;
2650
2651         arp = arp_hdr(skb);
2652         if (arp->ar_hln != dev->addr_len ||
2653             skb->pkt_type == PACKET_OTHERHOST ||
2654             skb->pkt_type == PACKET_LOOPBACK ||
2655             arp->ar_hrd != htons(ARPHRD_ETHER) ||
2656             arp->ar_pro != htons(ETH_P_IP) ||
2657             arp->ar_pln != 4)
2658                 goto out_unlock;
2659
2660         arp_ptr = (unsigned char *)(arp + 1);
2661         arp_ptr += dev->addr_len;
2662         memcpy(&sip, arp_ptr, 4);
2663         arp_ptr += 4 + dev->addr_len;
2664         memcpy(&tip, arp_ptr, 4);
2665
2666         pr_debug("bond_arp_rcv: %s %s/%d av %d sv %d sip %pI4 tip %pI4\n",
2667                  bond->dev->name, slave->dev->name, slave->state,
2668                  bond->params.arp_validate, slave_do_arp_validate(bond, slave),
2669                  &sip, &tip);
2670
2671         /*
2672          * Backup slaves won't see the ARP reply, but do come through
2673          * here for each ARP probe (so we swap the sip/tip to validate
2674          * the probe).  In a "redundant switch, common router" type of
2675          * configuration, the ARP probe will (hopefully) travel from
2676          * the active, through one switch, the router, then the other
2677          * switch before reaching the backup.
2678          */
2679         if (slave->state == BOND_STATE_ACTIVE)
2680                 bond_validate_arp(bond, slave, sip, tip);
2681         else
2682                 bond_validate_arp(bond, slave, tip, sip);
2683
2684 out_unlock:
2685         read_unlock(&bond->lock);
2686 out:
2687         dev_kfree_skb(skb);
2688         return NET_RX_SUCCESS;
2689 }
2690
2691 /*
2692  * this function is called regularly to monitor each slave's link
2693  * ensuring that traffic is being sent and received when arp monitoring
2694  * is used in load-balancing mode. if the adapter has been dormant, then an
2695  * arp is transmitted to generate traffic. see activebackup_arp_monitor for
2696  * arp monitoring in active backup mode.
2697  */
2698 void bond_loadbalance_arp_mon(struct work_struct *work)
2699 {
2700         struct bonding *bond = container_of(work, struct bonding,
2701                                             arp_work.work);
2702         struct slave *slave, *oldcurrent;
2703         int do_failover = 0;
2704         int delta_in_ticks;
2705         int i;
2706
2707         read_lock(&bond->lock);
2708
2709         delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
2710
2711         if (bond->kill_timers)
2712                 goto out;
2713
2714         if (bond->slave_cnt == 0)
2715                 goto re_arm;
2716
2717         read_lock(&bond->curr_slave_lock);
2718         oldcurrent = bond->curr_active_slave;
2719         read_unlock(&bond->curr_slave_lock);
2720
2721         /* see if any of the previous devices are up now (i.e. they have
2722          * xmt and rcv traffic). the curr_active_slave does not come into
2723          * the picture unless it is null. also, slave->jiffies is not needed
2724          * here because we send an arp on each slave and give a slave as
2725          * long as it needs to get the tx/rx within the delta.
2726          * TODO: what about up/down delay in arp mode? it wasn't here before
2727          *       so it can wait
2728          */
2729         bond_for_each_slave(bond, slave, i) {
2730                 if (slave->link != BOND_LINK_UP) {
2731                         if (time_before_eq(jiffies, dev_trans_start(slave->dev) + delta_in_ticks) &&
2732                             time_before_eq(jiffies, slave->dev->last_rx + delta_in_ticks)) {
2733
2734                                 slave->link  = BOND_LINK_UP;
2735                                 slave->state = BOND_STATE_ACTIVE;
2736
2737                                 /* primary_slave has no meaning in round-robin
2738                                  * mode. the window of a slave being up and
2739                                  * curr_active_slave being null after enslaving
2740                                  * is closed.
2741                                  */
2742                                 if (!oldcurrent) {
2743                                         pr_info("%s: link status definitely up for interface %s, ",
2744                                                 bond->dev->name,
2745                                                 slave->dev->name);
2746                                         do_failover = 1;
2747                                 } else {
2748                                         pr_info("%s: interface %s is now up\n",
2749                                                 bond->dev->name,
2750                                                 slave->dev->name);
2751                                 }
2752                         }
2753                 } else {
2754                         /* slave->link == BOND_LINK_UP */
2755
2756                         /* not all switches will respond to an arp request
2757                          * when the source ip is 0, so don't take the link down
2758                          * if we don't know our ip yet
2759                          */
2760                         if (time_after_eq(jiffies, dev_trans_start(slave->dev) + 2*delta_in_ticks) ||
2761                             (time_after_eq(jiffies, slave->dev->last_rx + 2*delta_in_ticks))) {
2762
2763                                 slave->link  = BOND_LINK_DOWN;
2764                                 slave->state = BOND_STATE_BACKUP;
2765
2766                                 if (slave->link_failure_count < UINT_MAX)
2767                                         slave->link_failure_count++;
2768
2769                                 pr_info("%s: interface %s is now down.\n",
2770                                         bond->dev->name,
2771                                         slave->dev->name);
2772
2773                                 if (slave == oldcurrent)
2774                                         do_failover = 1;
2775                         }
2776                 }
2777
2778                 /* note: if switch is in round-robin mode, all links
2779                  * must tx arp to ensure all links rx an arp - otherwise
2780                  * links may oscillate or not come up at all; if switch is
2781                  * in something like xor mode, there is nothing we can
2782                  * do - all replies will be rx'ed on same link causing slaves
2783                  * to be unstable during low/no traffic periods
2784                  */
2785                 if (IS_UP(slave->dev))
2786                         bond_arp_send_all(bond, slave);
2787         }
2788
2789         if (do_failover) {
2790                 write_lock_bh(&bond->curr_slave_lock);
2791
2792                 bond_select_active_slave(bond);
2793
2794                 write_unlock_bh(&bond->curr_slave_lock);
2795         }
2796
2797 re_arm:
2798         if (bond->params.arp_interval)
2799                 queue_delayed_work(bond->wq, &bond->arp_work, delta_in_ticks);
2800 out:
2801         read_unlock(&bond->lock);
2802 }
2803
2804 /*
2805  * Called to inspect slaves for active-backup mode ARP monitor link state
2806  * changes.  Sets new_link in slaves to specify what action should take
2807  * place for the slave.  Returns 0 if no changes are found, >0 if changes
2808  * to link states must be committed.
2809  *
2810  * Called with bond->lock held for read.
2811  */
2812 static int bond_ab_arp_inspect(struct bonding *bond, int delta_in_ticks)
2813 {
2814         struct slave *slave;
2815         int i, commit = 0;
2816
2817         bond_for_each_slave(bond, slave, i) {
2818                 slave->new_link = BOND_LINK_NOCHANGE;
2819
2820                 if (slave->link != BOND_LINK_UP) {
2821                         if (time_before_eq(jiffies, slave_last_rx(bond, slave) +
2822                                            delta_in_ticks)) {
2823                                 slave->new_link = BOND_LINK_UP;
2824                                 commit++;
2825                         }
2826
2827                         continue;
2828                 }
2829
2830                 /*
2831                  * Give slaves 2*delta after being enslaved or made
2832                  * active.  This avoids bouncing, as the last receive
2833                  * times need a full ARP monitor cycle to be updated.
2834                  */
2835                 if (!time_after_eq(jiffies, slave->jiffies +
2836                                    2 * delta_in_ticks))
2837                         continue;
2838
2839                 /*
2840                  * Backup slave is down if:
2841                  * - No current_arp_slave AND
2842                  * - more than 3*delta since last receive AND
2843                  * - the bond has an IP address
2844                  *
2845                  * Note: a non-null current_arp_slave indicates
2846                  * the curr_active_slave went down and we are
2847                  * searching for a new one; under this condition
2848                  * we only take the curr_active_slave down - this
2849                  * gives each slave a chance to tx/rx traffic
2850                  * before being taken out
2851                  */
2852                 if (slave->state == BOND_STATE_BACKUP &&
2853                     !bond->current_arp_slave &&
2854                     time_after(jiffies, slave_last_rx(bond, slave) +
2855                                3 * delta_in_ticks)) {
2856                         slave->new_link = BOND_LINK_DOWN;
2857                         commit++;
2858                 }
2859
2860                 /*
2861                  * Active slave is down if:
2862                  * - more than 2*delta since transmitting OR
2863                  * - (more than 2*delta since receive AND
2864                  *    the bond has an IP address)
2865                  */
2866                 if ((slave->state == BOND_STATE_ACTIVE) &&
2867                     (time_after_eq(jiffies, dev_trans_start(slave->dev) +
2868                                     2 * delta_in_ticks) ||
2869                       (time_after_eq(jiffies, slave_last_rx(bond, slave)
2870                                      + 2 * delta_in_ticks)))) {
2871                         slave->new_link = BOND_LINK_DOWN;
2872                         commit++;
2873                 }
2874         }
2875
2876         return commit;
2877 }
2878
2879 /*
2880  * Called to commit link state changes noted by inspection step of
2881  * active-backup mode ARP monitor.
2882  *
2883  * Called with RTNL and bond->lock for read.
2884  */
2885 static void bond_ab_arp_commit(struct bonding *bond, int delta_in_ticks)
2886 {
2887         struct slave *slave;
2888         int i;
2889
2890         bond_for_each_slave(bond, slave, i) {
2891                 switch (slave->new_link) {
2892                 case BOND_LINK_NOCHANGE:
2893                         continue;
2894
2895                 case BOND_LINK_UP:
2896                         if ((!bond->curr_active_slave &&
2897                              time_before_eq(jiffies,
2898                                             dev_trans_start(slave->dev) +
2899                                             delta_in_ticks)) ||
2900                             bond->curr_active_slave != slave) {
2901                                 slave->link = BOND_LINK_UP;
2902                                 bond->current_arp_slave = NULL;
2903
2904                                 pr_info("%s: link status definitely up for interface %s.\n",
2905                                         bond->dev->name, slave->dev->name);
2906
2907                                 if (!bond->curr_active_slave ||
2908                                     (slave == bond->primary_slave))
2909                                         goto do_failover;
2910
2911                         }
2912
2913                         continue;
2914
2915                 case BOND_LINK_DOWN:
2916                         if (slave->link_failure_count < UINT_MAX)
2917                                 slave->link_failure_count++;
2918
2919                         slave->link = BOND_LINK_DOWN;
2920                         bond_set_slave_inactive_flags(slave);
2921
2922                         pr_info("%s: link status definitely down for interface %s, disabling it\n",
2923                                 bond->dev->name, slave->dev->name);
2924
2925                         if (slave == bond->curr_active_slave) {
2926                                 bond->current_arp_slave = NULL;
2927                                 goto do_failover;
2928                         }
2929
2930                         continue;
2931
2932                 default:
2933                         pr_err("%s: impossible: new_link %d on slave %s\n",
2934                                bond->dev->name, slave->new_link,
2935                                slave->dev->name);
2936                         continue;
2937                 }
2938
2939 do_failover:
2940                 ASSERT_RTNL();
2941                 write_lock_bh(&bond->curr_slave_lock);
2942                 bond_select_active_slave(bond);
2943                 write_unlock_bh(&bond->curr_slave_lock);
2944         }
2945
2946         bond_set_carrier(bond);
2947 }
2948
2949 /*
2950  * Send ARP probes for active-backup mode ARP monitor.
2951  *
2952  * Called with bond->lock held for read.
2953  */
2954 static void bond_ab_arp_probe(struct bonding *bond)
2955 {
2956         struct slave *slave;
2957         int i;
2958
2959         read_lock(&bond->curr_slave_lock);
2960
2961         if (bond->current_arp_slave && bond->curr_active_slave)
2962                 pr_info("PROBE: c_arp %s && cas %s BAD\n",
2963                         bond->current_arp_slave->dev->name,
2964                         bond->curr_active_slave->dev->name);
2965
2966         if (bond->curr_active_slave) {
2967                 bond_arp_send_all(bond, bond->curr_active_slave);
2968                 read_unlock(&bond->curr_slave_lock);
2969                 return;
2970         }
2971
2972         read_unlock(&bond->curr_slave_lock);
2973
2974         /* if we don't have a curr_active_slave, search for the next available
2975          * backup slave from the current_arp_slave and make it the candidate
2976          * for becoming the curr_active_slave
2977          */
2978
2979         if (!bond->current_arp_slave) {
2980                 bond->current_arp_slave = bond->first_slave;
2981                 if (!bond->current_arp_slave)
2982                         return;
2983         }
2984
2985         bond_set_slave_inactive_flags(bond->current_arp_slave);
2986
2987         /* search for next candidate */
2988         bond_for_each_slave_from(bond, slave, i, bond->current_arp_slave->next) {
2989                 if (IS_UP(slave->dev)) {
2990                         slave->link = BOND_LINK_BACK;
2991                         bond_set_slave_active_flags(slave);
2992                         bond_arp_send_all(bond, slave);
2993                         slave->jiffies = jiffies;
2994                         bond->current_arp_slave = slave;
2995                         break;
2996                 }
2997
2998                 /* if the link state is up at this point, we
2999                  * mark it down - this can happen if we have
3000                  * simultaneous link failures and
3001                  * reselect_active_interface doesn't make this
3002                  * one the current slave so it is still marked
3003                  * up when it is actually down
3004                  */
3005                 if (slave->link == BOND_LINK_UP) {
3006                         slave->link = BOND_LINK_DOWN;
3007                         if (slave->link_failure_count < UINT_MAX)
3008                                 slave->link_failure_count++;
3009
3010                         bond_set_slave_inactive_flags(slave);
3011
3012                         pr_info("%s: backup interface %s is now down.\n",
3013                                 bond->dev->name, slave->dev->name);
3014                 }
3015         }
3016 }
3017
3018 void bond_activebackup_arp_mon(struct work_struct *work)
3019 {
3020         struct bonding *bond = container_of(work, struct bonding,
3021                                             arp_work.work);
3022         int delta_in_ticks;
3023
3024         read_lock(&bond->lock);
3025
3026         if (bond->kill_timers)
3027                 goto out;
3028
3029         delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
3030
3031         if (bond->slave_cnt == 0)
3032                 goto re_arm;
3033
3034         if (bond->send_grat_arp) {
3035                 read_lock(&bond->curr_slave_lock);
3036                 bond_send_gratuitous_arp(bond);
3037                 read_unlock(&bond->curr_slave_lock);
3038         }
3039
3040         if (bond->send_unsol_na) {
3041                 read_lock(&bond->curr_slave_lock);
3042                 bond_send_unsolicited_na(bond);
3043                 read_unlock(&bond->curr_slave_lock);
3044         }
3045
3046         if (bond_ab_arp_inspect(bond, delta_in_ticks)) {
3047                 read_unlock(&bond->lock);
3048                 rtnl_lock();
3049                 read_lock(&bond->lock);
3050
3051                 bond_ab_arp_commit(bond, delta_in_ticks);
3052
3053                 read_unlock(&bond->lock);
3054                 rtnl_unlock();
3055                 read_lock(&bond->lock);
3056         }
3057
3058         bond_ab_arp_probe(bond);
3059
3060 re_arm:
3061         if (bond->params.arp_interval)
3062                 queue_delayed_work(bond->wq, &bond->arp_work, delta_in_ticks);
3063 out:
3064         read_unlock(&bond->lock);
3065 }
3066
3067 /*------------------------------ proc/seq_file-------------------------------*/
3068
3069 #ifdef CONFIG_PROC_FS
3070
3071 static void *bond_info_seq_start(struct seq_file *seq, loff_t *pos)
3072         __acquires(&dev_base_lock)
3073         __acquires(&bond->lock)
3074 {
3075         struct bonding *bond = seq->private;
3076         loff_t off = 0;
3077         struct slave *slave;
3078         int i;
3079
3080         /* make sure the bond won't be taken away */
3081         read_lock(&dev_base_lock);
3082         read_lock(&bond->lock);
3083
3084         if (*pos == 0)
3085                 return SEQ_START_TOKEN;
3086
3087         bond_for_each_slave(bond, slave, i) {
3088                 if (++off == *pos)
3089                         return slave;
3090         }
3091
3092         return NULL;
3093 }
3094
3095 static void *bond_info_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3096 {
3097         struct bonding *bond = seq->private;
3098         struct slave *slave = v;
3099
3100         ++*pos;
3101         if (v == SEQ_START_TOKEN)
3102                 return bond->first_slave;
3103
3104         slave = slave->next;
3105
3106         return (slave == bond->first_slave) ? NULL : slave;
3107 }
3108
3109 static void bond_info_seq_stop(struct seq_file *seq, void *v)
3110         __releases(&bond->lock)
3111         __releases(&dev_base_lock)
3112 {
3113         struct bonding *bond = seq->private;
3114
3115         read_unlock(&bond->lock);
3116         read_unlock(&dev_base_lock);
3117 }
3118
3119 static void bond_info_show_master(struct seq_file *seq)
3120 {
3121         struct bonding *bond = seq->private;
3122         struct slave *curr;
3123         int i;
3124
3125         read_lock(&bond->curr_slave_lock);
3126         curr = bond->curr_active_slave;
3127         read_unlock(&bond->curr_slave_lock);
3128
3129         seq_printf(seq, "Bonding Mode: %s",
3130                    bond_mode_name(bond->params.mode));
3131
3132         if (bond->params.mode == BOND_MODE_ACTIVEBACKUP &&
3133             bond->params.fail_over_mac)
3134                 seq_printf(seq, " (fail_over_mac %s)",
3135                    fail_over_mac_tbl[bond->params.fail_over_mac].modename);
3136
3137         seq_printf(seq, "\n");
3138
3139         if (bond->params.mode == BOND_MODE_XOR ||
3140                 bond->params.mode == BOND_MODE_8023AD) {
3141                 seq_printf(seq, "Transmit Hash Policy: %s (%d)\n",
3142                         xmit_hashtype_tbl[bond->params.xmit_policy].modename,
3143                         bond->params.xmit_policy);
3144         }
3145
3146         if (USES_PRIMARY(bond->params.mode)) {
3147                 seq_printf(seq, "Primary Slave: %s",
3148                            (bond->primary_slave) ?
3149                            bond->primary_slave->dev->name : "None");
3150                 if (bond->primary_slave)
3151                         seq_printf(seq, " (primary_reselect %s)",
3152                    pri_reselect_tbl[bond->params.primary_reselect].modename);
3153
3154                 seq_printf(seq, "\nCurrently Active Slave: %s\n",
3155                            (curr) ? curr->dev->name : "None");
3156         }
3157
3158         seq_printf(seq, "MII Status: %s\n", netif_carrier_ok(bond->dev) ?
3159                    "up" : "down");
3160         seq_printf(seq, "MII Polling Interval (ms): %d\n", bond->params.miimon);
3161         seq_printf(seq, "Up Delay (ms): %d\n",
3162                    bond->params.updelay * bond->params.miimon);
3163         seq_printf(seq, "Down Delay (ms): %d\n",
3164                    bond->params.downdelay * bond->params.miimon);
3165
3166
3167         /* ARP information */
3168         if (bond->params.arp_interval > 0) {
3169                 int printed = 0;
3170                 seq_printf(seq, "ARP Polling Interval (ms): %d\n",
3171                                 bond->params.arp_interval);
3172
3173                 seq_printf(seq, "ARP IP target/s (n.n.n.n form):");
3174
3175                 for (i = 0; (i < BOND_MAX_ARP_TARGETS); i++) {
3176                         if (!bond->params.arp_targets[i])
3177                                 break;
3178                         if (printed)
3179                                 seq_printf(seq, ",");
3180                         seq_printf(seq, " %pI4", &bond->params.arp_targets[i]);
3181                         printed = 1;
3182                 }
3183                 seq_printf(seq, "\n");
3184         }
3185
3186         if (bond->params.mode == BOND_MODE_8023AD) {
3187                 struct ad_info ad_info;
3188
3189                 seq_puts(seq, "\n802.3ad info\n");
3190                 seq_printf(seq, "LACP rate: %s\n",
3191                            (bond->params.lacp_fast) ? "fast" : "slow");
3192                 seq_printf(seq, "Aggregator selection policy (ad_select): %s\n",
3193                            ad_select_tbl[bond->params.ad_select].modename);
3194
3195                 if (bond_3ad_get_active_agg_info(bond, &ad_info)) {
3196                         seq_printf(seq, "bond %s has no active aggregator\n",
3197                                    bond->dev->name);
3198                 } else {
3199                         seq_printf(seq, "Active Aggregator Info:\n");
3200
3201                         seq_printf(seq, "\tAggregator ID: %d\n",
3202                                    ad_info.aggregator_id);
3203                         seq_printf(seq, "\tNumber of ports: %d\n",
3204                                    ad_info.ports);
3205                         seq_printf(seq, "\tActor Key: %d\n",
3206                                    ad_info.actor_key);
3207                         seq_printf(seq, "\tPartner Key: %d\n",
3208                                    ad_info.partner_key);
3209                         seq_printf(seq, "\tPartner Mac Address: %pM\n",
3210                                    ad_info.partner_system);
3211                 }
3212         }
3213 }
3214
3215 static void bond_info_show_slave(struct seq_file *seq,
3216                                  const struct slave *slave)
3217 {
3218         struct bonding *bond = seq->private;
3219
3220         seq_printf(seq, "\nSlave Interface: %s\n", slave->dev->name);
3221         seq_printf(seq, "MII Status: %s\n",
3222                    (slave->link == BOND_LINK_UP) ?  "up" : "down");
3223         seq_printf(seq, "Link Failure Count: %u\n",
3224                    slave->link_failure_count);
3225
3226         seq_printf(seq, "Permanent HW addr: %pM\n", slave->perm_hwaddr);
3227
3228         if (bond->params.mode == BOND_MODE_8023AD) {
3229                 const struct aggregator *agg
3230                         = SLAVE_AD_INFO(slave).port.aggregator;
3231
3232                 if (agg)
3233                         seq_printf(seq, "Aggregator ID: %d\n",
3234                                    agg->aggregator_identifier);
3235                 else
3236                         seq_puts(seq, "Aggregator ID: N/A\n");
3237         }
3238 }
3239
3240 static int bond_info_seq_show(struct seq_file *seq, void *v)
3241 {
3242         if (v == SEQ_START_TOKEN) {
3243                 seq_printf(seq, "%s\n", version);
3244                 bond_info_show_master(seq);
3245         } else
3246                 bond_info_show_slave(seq, v);
3247
3248         return 0;
3249 }
3250
3251 static const struct seq_operations bond_info_seq_ops = {
3252         .start = bond_info_seq_start,
3253         .next  = bond_info_seq_next,
3254         .stop  = bond_info_seq_stop,
3255         .show  = bond_info_seq_show,
3256 };
3257
3258 static int bond_info_open(struct inode *inode, struct file *file)
3259 {
3260         struct seq_file *seq;
3261         struct proc_dir_entry *proc;
3262         int res;
3263
3264         res = seq_open(file, &bond_info_seq_ops);
3265         if (!res) {
3266                 /* recover the pointer buried in proc_dir_entry data */
3267                 seq = file->private_data;
3268                 proc = PDE(inode);
3269                 seq->private = proc->data;
3270         }
3271
3272         return res;
3273 }
3274
3275 static const struct file_operations bond_info_fops = {
3276         .owner   = THIS_MODULE,
3277         .open    = bond_info_open,
3278         .read    = seq_read,
3279         .llseek  = seq_lseek,
3280         .release = seq_release,
3281 };
3282
3283 static void bond_create_proc_entry(struct bonding *bond)
3284 {
3285         struct net_device *bond_dev = bond->dev;
3286         struct bond_net *bn = net_generic(dev_net(bond_dev), bond_net_id);
3287
3288         if (bn->proc_dir) {
3289                 bond->proc_entry = proc_create_data(bond_dev->name,
3290                                                     S_IRUGO, bn->proc_dir,
3291                                                     &bond_info_fops, bond);
3292                 if (bond->proc_entry == NULL)
3293                         pr_warning("Warning: Cannot create /proc/net/%s/%s\n",
3294                                    DRV_NAME, bond_dev->name);
3295                 else
3296                         memcpy(bond->proc_file_name, bond_dev->name, IFNAMSIZ);
3297         }
3298 }
3299
3300 static void bond_remove_proc_entry(struct bonding *bond)
3301 {
3302         struct net_device *bond_dev = bond->dev;
3303         struct bond_net *bn = net_generic(dev_net(bond_dev), bond_net_id);
3304
3305         if (bn->proc_dir && bond->proc_entry) {
3306                 remove_proc_entry(bond->proc_file_name, bn->proc_dir);
3307                 memset(bond->proc_file_name, 0, IFNAMSIZ);
3308                 bond->proc_entry = NULL;
3309         }
3310 }
3311
3312 /* Create the bonding directory under /proc/net, if doesn't exist yet.
3313  * Caller must hold rtnl_lock.
3314  */
3315 static void __net_init bond_create_proc_dir(struct bond_net *bn)
3316 {
3317         if (!bn->proc_dir) {
3318                 bn->proc_dir = proc_mkdir(DRV_NAME, bn->net->proc_net);
3319                 if (!bn->proc_dir)
3320                         pr_warning("Warning: cannot create /proc/net/%s\n",
3321                                    DRV_NAME);
3322         }
3323 }
3324
3325 /* Destroy the bonding directory under /proc/net, if empty.
3326  * Caller must hold rtnl_lock.
3327  */
3328 static void __net_exit bond_destroy_proc_dir(struct bond_net *bn)
3329 {
3330         if (bn->proc_dir) {
3331                 remove_proc_entry(DRV_NAME, bn->net->proc_net);
3332                 bn->proc_dir = NULL;
3333         }
3334 }
3335
3336 #else /* !CONFIG_PROC_FS */
3337
3338 static void bond_create_proc_entry(struct bonding *bond)
3339 {
3340 }
3341
3342 static void bond_remove_proc_entry(struct bonding *bond)
3343 {
3344 }
3345
3346 static inline void bond_create_proc_dir(struct bond_net *bn)
3347 {
3348 }
3349
3350 static inline void bond_destroy_proc_dir(struct bond_net *bn)
3351 {
3352 }
3353
3354 #endif /* CONFIG_PROC_FS */
3355
3356
3357 /*-------------------------- netdev event handling --------------------------*/
3358
3359 /*
3360  * Change device name
3361  */
3362 static int bond_event_changename(struct bonding *bond)
3363 {
3364         bond_remove_proc_entry(bond);
3365         bond_create_proc_entry(bond);
3366
3367         return NOTIFY_DONE;
3368 }
3369
3370 static int bond_master_netdev_event(unsigned long event,
3371                                     struct net_device *bond_dev)
3372 {
3373         struct bonding *event_bond = netdev_priv(bond_dev);
3374
3375         switch (event) {
3376         case NETDEV_CHANGENAME:
3377                 return bond_event_changename(event_bond);
3378         default:
3379                 break;
3380         }
3381
3382         return NOTIFY_DONE;
3383 }
3384
3385 static int bond_slave_netdev_event(unsigned long event,
3386                                    struct net_device *slave_dev)
3387 {
3388         struct net_device *bond_dev = slave_dev->master;
3389         struct bonding *bond = netdev_priv(bond_dev);
3390
3391         switch (event) {
3392         case NETDEV_UNREGISTER:
3393                 if (bond_dev) {
3394                         if (bond->setup_by_slave)
3395                                 bond_release_and_destroy(bond_dev, slave_dev);
3396                         else
3397                                 bond_release(bond_dev, slave_dev);
3398                 }
3399                 break;
3400         case NETDEV_CHANGE:
3401                 if (bond->params.mode == BOND_MODE_8023AD || bond_is_lb(bond)) {
3402                         struct slave *slave;
3403
3404                         slave = bond_get_slave_by_dev(bond, slave_dev);
3405                         if (slave) {
3406                                 u16 old_speed = slave->speed;
3407                                 u16 old_duplex = slave->duplex;
3408
3409                                 bond_update_speed_duplex(slave);
3410
3411                                 if (bond_is_lb(bond))
3412                                         break;
3413
3414                                 if (old_speed != slave->speed)
3415                                         bond_3ad_adapter_speed_changed(slave);
3416                                 if (old_duplex != slave->duplex)
3417                                         bond_3ad_adapter_duplex_changed(slave);
3418                         }
3419                 }
3420
3421                 break;
3422         case NETDEV_DOWN:
3423                 /*
3424                  * ... Or is it this?
3425                  */
3426                 break;
3427         case NETDEV_CHANGEMTU:
3428                 /*
3429                  * TODO: Should slaves be allowed to
3430                  * independently alter their MTU?  For
3431                  * an active-backup bond, slaves need
3432                  * not be the same type of device, so
3433                  * MTUs may vary.  For other modes,
3434                  * slaves arguably should have the
3435                  * same MTUs. To do this, we'd need to
3436                  * take over the slave's change_mtu
3437                  * function for the duration of their
3438                  * servitude.
3439                  */
3440                 break;
3441         case NETDEV_CHANGENAME:
3442                 /*
3443                  * TODO: handle changing the primary's name
3444                  */
3445                 break;
3446         case NETDEV_FEAT_CHANGE:
3447                 bond_compute_features(bond);
3448                 break;
3449         default:
3450                 break;
3451         }
3452
3453         return NOTIFY_DONE;
3454 }
3455
3456 /*
3457  * bond_netdev_event: handle netdev notifier chain events.
3458  *
3459  * This function receives events for the netdev chain.  The caller (an
3460  * ioctl handler calling blocking_notifier_call_chain) holds the necessary
3461  * locks for us to safely manipulate the slave devices (RTNL lock,
3462  * dev_probe_lock).
3463  */
3464 static int bond_netdev_event(struct notifier_block *this,
3465                              unsigned long event, void *ptr)
3466 {
3467         struct net_device *event_dev = (struct net_device *)ptr;
3468
3469         pr_debug("event_dev: %s, event: %lx\n",
3470                  event_dev ? event_dev->name : "None",
3471                  event);
3472
3473         if (!(event_dev->priv_flags & IFF_BONDING))
3474                 return NOTIFY_DONE;
3475
3476         if (event_dev->flags & IFF_MASTER) {
3477                 pr_debug("IFF_MASTER\n");
3478                 return bond_master_netdev_event(event, event_dev);
3479         }
3480
3481         if (event_dev->flags & IFF_SLAVE) {
3482                 pr_debug("IFF_SLAVE\n");
3483                 return bond_slave_netdev_event(event, event_dev);
3484         }
3485
3486         return NOTIFY_DONE;
3487 }
3488
3489 /*
3490  * bond_inetaddr_event: handle inetaddr notifier chain events.
3491  *
3492  * We keep track of device IPs primarily to use as source addresses in
3493  * ARP monitor probes (rather than spewing out broadcasts all the time).
3494  *
3495  * We track one IP for the main device (if it has one), plus one per VLAN.
3496  */
3497 static int bond_inetaddr_event(struct notifier_block *this, unsigned long event, void *ptr)
3498 {
3499         struct in_ifaddr *ifa = ptr;
3500         struct net_device *vlan_dev, *event_dev = ifa->ifa_dev->dev;
3501         struct bond_net *bn = net_generic(dev_net(event_dev), bond_net_id);
3502         struct bonding *bond;
3503         struct vlan_entry *vlan;
3504
3505         list_for_each_entry(bond, &bn->dev_list, bond_list) {
3506                 if (bond->dev == event_dev) {
3507                         switch (event) {
3508                         case NETDEV_UP:
3509                                 bond->master_ip = ifa->ifa_local;
3510                                 return NOTIFY_OK;
3511                         case NETDEV_DOWN:
3512                                 bond->master_ip = bond_glean_dev_ip(bond->dev);
3513                                 return NOTIFY_OK;
3514                         default:
3515                                 return NOTIFY_DONE;
3516                         }
3517                 }
3518
3519                 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
3520                         vlan_dev = vlan_group_get_device(bond->vlgrp, vlan->vlan_id);
3521                         if (vlan_dev == event_dev) {
3522                                 switch (event) {
3523                                 case NETDEV_UP:
3524                                         vlan->vlan_ip = ifa->ifa_local;
3525                                         return NOTIFY_OK;
3526                                 case NETDEV_DOWN:
3527                                         vlan->vlan_ip =
3528                                                 bond_glean_dev_ip(vlan_dev);
3529                                         return NOTIFY_OK;
3530                                 default:
3531                                         return NOTIFY_DONE;
3532                                 }
3533                         }
3534                 }
3535         }
3536         return NOTIFY_DONE;
3537 }
3538
3539 static struct notifier_block bond_netdev_notifier = {
3540         .notifier_call = bond_netdev_event,
3541 };
3542
3543 static struct notifier_block bond_inetaddr_notifier = {
3544         .notifier_call = bond_inetaddr_event,
3545 };
3546
3547 /*-------------------------- Packet type handling ---------------------------*/
3548
3549 /* register to receive lacpdus on a bond */
3550 static void bond_register_lacpdu(struct bonding *bond)
3551 {
3552         struct packet_type *pk_type = &(BOND_AD_INFO(bond).ad_pkt_type);
3553
3554         /* initialize packet type */
3555         pk_type->type = PKT_TYPE_LACPDU;
3556         pk_type->dev = bond->dev;
3557         pk_type->func = bond_3ad_lacpdu_recv;
3558
3559         dev_add_pack(pk_type);
3560 }
3561
3562 /* unregister to receive lacpdus on a bond */
3563 static void bond_unregister_lacpdu(struct bonding *bond)
3564 {
3565         dev_remove_pack(&(BOND_AD_INFO(bond).ad_pkt_type));
3566 }
3567
3568 void bond_register_arp(struct bonding *bond)
3569 {
3570         struct packet_type *pt = &bond->arp_mon_pt;
3571
3572         if (pt->type)
3573                 return;
3574
3575         pt->type = htons(ETH_P_ARP);
3576         pt->dev = bond->dev;
3577         pt->func = bond_arp_rcv;
3578         dev_add_pack(pt);
3579 }
3580
3581 void bond_unregister_arp(struct bonding *bond)
3582 {
3583         struct packet_type *pt = &bond->arp_mon_pt;
3584
3585         dev_remove_pack(pt);
3586         pt->type = 0;
3587 }
3588
3589 /*---------------------------- Hashing Policies -----------------------------*/
3590
3591 /*
3592  * Hash for the output device based upon layer 2 and layer 3 data. If
3593  * the packet is not IP mimic bond_xmit_hash_policy_l2()
3594  */
3595 static int bond_xmit_hash_policy_l23(struct sk_buff *skb, int count)
3596 {
3597         struct ethhdr *data = (struct ethhdr *)skb->data;
3598         struct iphdr *iph = ip_hdr(skb);
3599
3600         if (skb->protocol == htons(ETH_P_IP)) {
3601                 return ((ntohl(iph->saddr ^ iph->daddr) & 0xffff) ^
3602                         (data->h_dest[5] ^ data->h_source[5])) % count;
3603         }
3604
3605         return (data->h_dest[5] ^ data->h_source[5]) % count;
3606 }
3607
3608 /*
3609  * Hash for the output device based upon layer 3 and layer 4 data. If
3610  * the packet is a frag or not TCP or UDP, just use layer 3 data.  If it is
3611  * altogether not IP, mimic bond_xmit_hash_policy_l2()
3612  */
3613 static int bond_xmit_hash_policy_l34(struct sk_buff *skb, int count)
3614 {
3615         struct ethhdr *data = (struct ethhdr *)skb->data;
3616         struct iphdr *iph = ip_hdr(skb);
3617         __be16 *layer4hdr = (__be16 *)((u32 *)iph + iph->ihl);
3618         int layer4_xor = 0;
3619
3620         if (skb->protocol == htons(ETH_P_IP)) {
3621                 if (!(iph->frag_off & htons(IP_MF|IP_OFFSET)) &&
3622                     (iph->protocol == IPPROTO_TCP ||
3623                      iph->protocol == IPPROTO_UDP)) {
3624                         layer4_xor = ntohs((*layer4hdr ^ *(layer4hdr + 1)));
3625                 }
3626                 return (layer4_xor ^
3627                         ((ntohl(iph->saddr ^ iph->daddr)) & 0xffff)) % count;
3628
3629         }
3630
3631         return (data->h_dest[5] ^ data->h_source[5]) % count;
3632 }
3633
3634 /*
3635  * Hash for the output device based upon layer 2 data
3636  */
3637 static int bond_xmit_hash_policy_l2(struct sk_buff *skb, int count)
3638 {
3639         struct ethhdr *data = (struct ethhdr *)skb->data;
3640
3641         return (data->h_dest[5] ^ data->h_source[5]) % count;
3642 }
3643
3644 /*-------------------------- Device entry points ----------------------------*/
3645
3646 static int bond_open(struct net_device *bond_dev)
3647 {
3648         struct bonding *bond = netdev_priv(bond_dev);
3649
3650         bond->kill_timers = 0;
3651
3652         if (bond_is_lb(bond)) {
3653                 /* bond_alb_initialize must be called before the timer
3654                  * is started.
3655                  */
3656                 if (bond_alb_initialize(bond, (bond->params.mode == BOND_MODE_ALB))) {
3657                         /* something went wrong - fail the open operation */
3658                         return -ENOMEM;
3659                 }
3660
3661                 INIT_DELAYED_WORK(&bond->alb_work, bond_alb_monitor);
3662                 queue_delayed_work(bond->wq, &bond->alb_work, 0);
3663         }
3664
3665         if (bond->params.miimon) {  /* link check interval, in milliseconds. */
3666                 INIT_DELAYED_WORK(&bond->mii_work, bond_mii_monitor);
3667                 queue_delayed_work(bond->wq, &bond->mii_work, 0);
3668         }
3669
3670         if (bond->params.arp_interval) {  /* arp interval, in milliseconds. */
3671                 if (bond->params.mode == BOND_MODE_ACTIVEBACKUP)
3672                         INIT_DELAYED_WORK(&bond->arp_work,
3673                                           bond_activebackup_arp_mon);
3674                 else
3675                         INIT_DELAYED_WORK(&bond->arp_work,
3676                                           bond_loadbalance_arp_mon);
3677
3678                 queue_delayed_work(bond->wq, &bond->arp_work, 0);
3679                 if (bond->params.arp_validate)
3680                         bond_register_arp(bond);
3681         }
3682
3683         if (bond->params.mode == BOND_MODE_8023AD) {
3684                 INIT_DELAYED_WORK(&bond->ad_work, bond_3ad_state_machine_handler);
3685                 queue_delayed_work(bond->wq, &bond->ad_work, 0);
3686                 /* register to receive LACPDUs */
3687                 bond_register_lacpdu(bond);
3688                 bond_3ad_initiate_agg_selection(bond, 1);
3689         }
3690
3691         return 0;
3692 }
3693
3694 static int bond_close(struct net_device *bond_dev)
3695 {
3696         struct bonding *bond = netdev_priv(bond_dev);
3697
3698         if (bond->params.mode == BOND_MODE_8023AD) {
3699                 /* Unregister the receive of LACPDUs */
3700                 bond_unregister_lacpdu(bond);
3701         }
3702
3703         if (bond->params.arp_validate)
3704                 bond_unregister_arp(bond);
3705
3706         write_lock_bh(&bond->lock);
3707
3708         bond->send_grat_arp = 0;
3709         bond->send_unsol_na = 0;
3710
3711         /* signal timers not to re-arm */
3712         bond->kill_timers = 1;
3713
3714         write_unlock_bh(&bond->lock);
3715
3716         if (bond->params.miimon) {  /* link check interval, in milliseconds. */
3717                 cancel_delayed_work(&bond->mii_work);
3718         }
3719
3720         if (bond->params.arp_interval) {  /* arp interval, in milliseconds. */
3721                 cancel_delayed_work(&bond->arp_work);
3722         }
3723
3724         switch (bond->params.mode) {
3725         case BOND_MODE_8023AD:
3726                 cancel_delayed_work(&bond->ad_work);
3727                 break;
3728         case BOND_MODE_TLB:
3729         case BOND_MODE_ALB:
3730                 cancel_delayed_work(&bond->alb_work);
3731                 break;
3732         default:
3733                 break;
3734         }
3735
3736
3737         if (bond_is_lb(bond)) {
3738                 /* Must be called only after all
3739                  * slaves have been released
3740                  */
3741                 bond_alb_deinitialize(bond);
3742         }
3743
3744         return 0;
3745 }
3746
3747 static struct net_device_stats *bond_get_stats(struct net_device *bond_dev)
3748 {
3749         struct bonding *bond = netdev_priv(bond_dev);
3750         struct net_device_stats *stats = &bond_dev->stats;
3751         struct net_device_stats local_stats;
3752         struct slave *slave;
3753         int i;
3754
3755         memset(&local_stats, 0, sizeof(struct net_device_stats));
3756
3757         read_lock_bh(&bond->lock);
3758
3759         bond_for_each_slave(bond, slave, i) {
3760                 const struct net_device_stats *sstats = dev_get_stats(slave->dev);
3761
3762                 local_stats.rx_packets += sstats->rx_packets;
3763                 local_stats.rx_bytes += sstats->rx_bytes;
3764                 local_stats.rx_errors += sstats->rx_errors;
3765                 local_stats.rx_dropped += sstats->rx_dropped;
3766
3767                 local_stats.tx_packets += sstats->tx_packets;
3768                 local_stats.tx_bytes += sstats->tx_bytes;
3769                 local_stats.tx_errors += sstats->tx_errors;
3770                 local_stats.tx_dropped += sstats->tx_dropped;
3771
3772                 local_stats.multicast += sstats->multicast;
3773                 local_stats.collisions += sstats->collisions;
3774
3775                 local_stats.rx_length_errors += sstats->rx_length_errors;
3776                 local_stats.rx_over_errors += sstats->rx_over_errors;
3777                 local_stats.rx_crc_errors += sstats->rx_crc_errors;
3778                 local_stats.rx_frame_errors += sstats->rx_frame_errors;
3779                 local_stats.rx_fifo_errors += sstats->rx_fifo_errors;
3780                 local_stats.rx_missed_errors += sstats->rx_missed_errors;
3781
3782                 local_stats.tx_aborted_errors += sstats->tx_aborted_errors;
3783                 local_stats.tx_carrier_errors += sstats->tx_carrier_errors;
3784                 local_stats.tx_fifo_errors += sstats->tx_fifo_errors;
3785                 local_stats.tx_heartbeat_errors += sstats->tx_heartbeat_errors;
3786                 local_stats.tx_window_errors += sstats->tx_window_errors;
3787         }
3788
3789         memcpy(stats, &local_stats, sizeof(struct net_device_stats));
3790
3791         read_unlock_bh(&bond->lock);
3792
3793         return stats;
3794 }
3795
3796 static int bond_do_ioctl(struct net_device *bond_dev, struct ifreq *ifr, int cmd)
3797 {
3798         struct net_device *slave_dev = NULL;
3799         struct ifbond k_binfo;
3800         struct ifbond __user *u_binfo = NULL;
3801         struct ifslave k_sinfo;
3802         struct ifslave __user *u_sinfo = NULL;
3803         struct mii_ioctl_data *mii = NULL;
3804         int res = 0;
3805
3806         pr_debug("bond_ioctl: master=%s, cmd=%d\n", bond_dev->name, cmd);
3807
3808         switch (cmd) {
3809         case SIOCGMIIPHY:
3810                 mii = if_mii(ifr);
3811                 if (!mii)
3812                         return -EINVAL;
3813
3814                 mii->phy_id = 0;
3815                 /* Fall Through */
3816         case SIOCGMIIREG:
3817                 /*
3818                  * We do this again just in case we were called by SIOCGMIIREG
3819                  * instead of SIOCGMIIPHY.
3820                  */
3821                 mii = if_mii(ifr);
3822                 if (!mii)
3823                         return -EINVAL;
3824
3825
3826                 if (mii->reg_num == 1) {
3827                         struct bonding *bond = netdev_priv(bond_dev);
3828                         mii->val_out = 0;
3829                         read_lock(&bond->lock);
3830                         read_lock(&bond->curr_slave_lock);
3831                         if (netif_carrier_ok(bond->dev))
3832                                 mii->val_out = BMSR_LSTATUS;
3833
3834                         read_unlock(&bond->curr_slave_lock);
3835                         read_unlock(&bond->lock);
3836                 }
3837
3838                 return 0;
3839         case BOND_INFO_QUERY_OLD:
3840         case SIOCBONDINFOQUERY:
3841                 u_binfo = (struct ifbond __user *)ifr->ifr_data;
3842
3843                 if (copy_from_user(&k_binfo, u_binfo, sizeof(ifbond)))
3844                         return -EFAULT;
3845
3846                 res = bond_info_query(bond_dev, &k_binfo);
3847                 if (res == 0 &&
3848                     copy_to_user(u_binfo, &k_binfo, sizeof(ifbond)))
3849                         return -EFAULT;
3850
3851                 return res;
3852         case BOND_SLAVE_INFO_QUERY_OLD:
3853         case SIOCBONDSLAVEINFOQUERY:
3854                 u_sinfo = (struct ifslave __user *)ifr->ifr_data;
3855
3856                 if (copy_from_user(&k_sinfo, u_sinfo, sizeof(ifslave)))
3857                         return -EFAULT;
3858
3859                 res = bond_slave_info_query(bond_dev, &k_sinfo);
3860                 if (res == 0 &&
3861                     copy_to_user(u_sinfo, &k_sinfo, sizeof(ifslave)))
3862                         return -EFAULT;
3863
3864                 return res;
3865         default:
3866                 /* Go on */
3867                 break;
3868         }
3869
3870         if (!capable(CAP_NET_ADMIN))
3871                 return -EPERM;
3872
3873         slave_dev = dev_get_by_name(dev_net(bond_dev), ifr->ifr_slave);
3874
3875         pr_debug("slave_dev=%p:\n", slave_dev);
3876
3877         if (!slave_dev)
3878                 res = -ENODEV;
3879         else {
3880                 pr_debug("slave_dev->name=%s:\n", slave_dev->name);
3881                 switch (cmd) {
3882                 case BOND_ENSLAVE_OLD:
3883                 case SIOCBONDENSLAVE:
3884                         res = bond_enslave(bond_dev, slave_dev);
3885                         break;
3886                 case BOND_RELEASE_OLD:
3887                 case SIOCBONDRELEASE:
3888                         res = bond_release(bond_dev, slave_dev);
3889                         break;
3890                 case BOND_SETHWADDR_OLD:
3891                 case SIOCBONDSETHWADDR:
3892                         res = bond_sethwaddr(bond_dev, slave_dev);
3893                         break;
3894                 case BOND_CHANGE_ACTIVE_OLD:
3895                 case SIOCBONDCHANGEACTIVE:
3896                         res = bond_ioctl_change_active(bond_dev, slave_dev);
3897                         break;
3898                 default:
3899                         res = -EOPNOTSUPP;
3900                 }
3901
3902                 dev_put(slave_dev);
3903         }
3904
3905         return res;
3906 }
3907
3908 static void bond_set_multicast_list(struct net_device *bond_dev)
3909 {
3910         struct bonding *bond = netdev_priv(bond_dev);
3911         struct dev_mc_list *dmi;
3912
3913         /*
3914          * Do promisc before checking multicast_mode
3915          */
3916         if ((bond_dev->flags & IFF_PROMISC) && !(bond->flags & IFF_PROMISC))
3917                 /*
3918                  * FIXME: Need to handle the error when one of the multi-slaves
3919                  * encounters error.
3920                  */
3921                 bond_set_promiscuity(bond, 1);
3922
3923
3924         if (!(bond_dev->flags & IFF_PROMISC) && (bond->flags & IFF_PROMISC))
3925                 bond_set_promiscuity(bond, -1);
3926
3927
3928         /* set allmulti flag to slaves */
3929         if ((bond_dev->flags & IFF_ALLMULTI) && !(bond->flags & IFF_ALLMULTI))
3930                 /*
3931                  * FIXME: Need to handle the error when one of the multi-slaves
3932                  * encounters error.
3933                  */
3934                 bond_set_allmulti(bond, 1);
3935
3936
3937         if (!(bond_dev->flags & IFF_ALLMULTI) && (bond->flags & IFF_ALLMULTI))
3938                 bond_set_allmulti(bond, -1);
3939
3940
3941         read_lock(&bond->lock);
3942
3943         bond->flags = bond_dev->flags;
3944
3945         /* looking for addresses to add to slaves' mc list */
3946         for (dmi = bond_dev->mc_list; dmi; dmi = dmi->next) {
3947                 if (!bond_mc_list_find_dmi(dmi, bond->mc_list))
3948                         bond_mc_add(bond, dmi->dmi_addr, dmi->dmi_addrlen);
3949         }
3950
3951         /* looking for addresses to delete from slaves' list */
3952         for (dmi = bond->mc_list; dmi; dmi = dmi->next) {
3953                 if (!bond_mc_list_find_dmi(dmi, bond_dev->mc_list))
3954                         bond_mc_delete(bond, dmi->dmi_addr, dmi->dmi_addrlen);
3955         }
3956
3957         /* save master's multicast list */
3958         bond_mc_list_destroy(bond);
3959         bond_mc_list_copy(bond_dev->mc_list, bond, GFP_ATOMIC);
3960
3961         read_unlock(&bond->lock);
3962 }
3963
3964 static int bond_neigh_setup(struct net_device *dev, struct neigh_parms *parms)
3965 {
3966         struct bonding *bond = netdev_priv(dev);
3967         struct slave *slave = bond->first_slave;
3968
3969         if (slave) {
3970                 const struct net_device_ops *slave_ops
3971                         = slave->dev->netdev_ops;
3972                 if (slave_ops->ndo_neigh_setup)
3973                         return slave_ops->ndo_neigh_setup(slave->dev, parms);
3974         }
3975         return 0;
3976 }
3977
3978 /*
3979  * Change the MTU of all of a master's slaves to match the master
3980  */
3981 static int bond_change_mtu(struct net_device *bond_dev, int new_mtu)
3982 {
3983         struct bonding *bond = netdev_priv(bond_dev);
3984         struct slave *slave, *stop_at;
3985         int res = 0;
3986         int i;
3987
3988         pr_debug("bond=%p, name=%s, new_mtu=%d\n", bond,
3989                  (bond_dev ? bond_dev->name : "None"), new_mtu);
3990
3991         /* Can't hold bond->lock with bh disabled here since
3992          * some base drivers panic. On the other hand we can't
3993          * hold bond->lock without bh disabled because we'll
3994          * deadlock. The only solution is to rely on the fact
3995          * that we're under rtnl_lock here, and the slaves
3996          * list won't change. This doesn't solve the problem
3997          * of setting the slave's MTU while it is
3998          * transmitting, but the assumption is that the base
3999          * driver can handle that.
4000          *
4001          * TODO: figure out a way to safely iterate the slaves
4002          * list, but without holding a lock around the actual
4003          * call to the base driver.
4004          */
4005
4006         bond_for_each_slave(bond, slave, i) {
4007                 pr_debug("s %p s->p %p c_m %p\n",
4008                          slave,
4009                          slave->prev,
4010                          slave->dev->netdev_ops->ndo_change_mtu);
4011
4012                 res = dev_set_mtu(slave->dev, new_mtu);
4013
4014                 if (res) {
4015                         /* If we failed to set the slave's mtu to the new value
4016                          * we must abort the operation even in ACTIVE_BACKUP
4017                          * mode, because if we allow the backup slaves to have
4018                          * different mtu values than the active slave we'll
4019                          * need to change their mtu when doing a failover. That
4020                          * means changing their mtu from timer context, which
4021                          * is probably not a good idea.
4022                          */
4023                         pr_debug("err %d %s\n", res, slave->dev->name);
4024                         goto unwind;
4025                 }
4026         }
4027
4028         bond_dev->mtu = new_mtu;
4029
4030         return 0;
4031
4032 unwind:
4033         /* unwind from head to the slave that failed */
4034         stop_at = slave;
4035         bond_for_each_slave_from_to(bond, slave, i, bond->first_slave, stop_at) {
4036                 int tmp_res;
4037
4038                 tmp_res = dev_set_mtu(slave->dev, bond_dev->mtu);
4039                 if (tmp_res) {
4040                         pr_debug("unwind err %d dev %s\n",
4041                                  tmp_res, slave->dev->name);
4042                 }
4043         }
4044
4045         return res;
4046 }
4047
4048 /*
4049  * Change HW address
4050  *
4051  * Note that many devices must be down to change the HW address, and
4052  * downing the master releases all slaves.  We can make bonds full of
4053  * bonding devices to test this, however.
4054  */
4055 static int bond_set_mac_address(struct net_device *bond_dev, void *addr)
4056 {
4057         struct bonding *bond = netdev_priv(bond_dev);
4058         struct sockaddr *sa = addr, tmp_sa;
4059         struct slave *slave, *stop_at;
4060         int res = 0;
4061         int i;
4062
4063         if (bond->params.mode == BOND_MODE_ALB)
4064                 return bond_alb_set_mac_address(bond_dev, addr);
4065
4066
4067         pr_debug("bond=%p, name=%s\n",
4068                  bond, bond_dev ? bond_dev->name : "None");
4069
4070         /*
4071          * If fail_over_mac is set to active, do nothing and return
4072          * success.  Returning an error causes ifenslave to fail.
4073          */
4074         if (bond->params.fail_over_mac == BOND_FOM_ACTIVE)
4075                 return 0;
4076
4077         if (!is_valid_ether_addr(sa->sa_data))
4078                 return -EADDRNOTAVAIL;
4079
4080         /* Can't hold bond->lock with bh disabled here since
4081          * some base drivers panic. On the other hand we can't
4082          * hold bond->lock without bh disabled because we'll
4083          * deadlock. The only solution is to rely on the fact
4084          * that we're under rtnl_lock here, and the slaves
4085          * list won't change. This doesn't solve the problem
4086          * of setting the slave's hw address while it is
4087          * transmitting, but the assumption is that the base
4088          * driver can handle that.
4089          *
4090          * TODO: figure out a way to safely iterate the slaves
4091          * list, but without holding a lock around the actual
4092          * call to the base driver.
4093          */
4094
4095         bond_for_each_slave(bond, slave, i) {
4096                 const struct net_device_ops *slave_ops = slave->dev->netdev_ops;
4097                 pr_debug("slave %p %s\n", slave, slave->dev->name);
4098
4099                 if (slave_ops->ndo_set_mac_address == NULL) {
4100                         res = -EOPNOTSUPP;
4101                         pr_debug("EOPNOTSUPP %s\n", slave->dev->name);
4102                         goto unwind;
4103                 }
4104
4105                 res = dev_set_mac_address(slave->dev, addr);
4106                 if (res) {
4107                         /* TODO: consider downing the slave
4108                          * and retry ?
4109                          * User should expect communications
4110                          * breakage anyway until ARP finish
4111                          * updating, so...
4112                          */
4113                         pr_debug("err %d %s\n", res, slave->dev->name);
4114                         goto unwind;
4115                 }
4116         }
4117
4118         /* success */
4119         memcpy(bond_dev->dev_addr, sa->sa_data, bond_dev->addr_len);
4120         return 0;
4121
4122 unwind:
4123         memcpy(tmp_sa.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
4124         tmp_sa.sa_family = bond_dev->type;
4125
4126         /* unwind from head to the slave that failed */
4127         stop_at = slave;
4128         bond_for_each_slave_from_to(bond, slave, i, bond->first_slave, stop_at) {
4129                 int tmp_res;
4130
4131                 tmp_res = dev_set_mac_address(slave->dev, &tmp_sa);
4132                 if (tmp_res) {
4133                         pr_debug("unwind err %d dev %s\n",
4134                                  tmp_res, slave->dev->name);
4135                 }
4136         }
4137
4138         return res;
4139 }
4140
4141 static int bond_xmit_roundrobin(struct sk_buff *skb, struct net_device *bond_dev)
4142 {
4143         struct bonding *bond = netdev_priv(bond_dev);
4144         struct slave *slave, *start_at;
4145         int i, slave_no, res = 1;
4146         struct iphdr *iph = ip_hdr(skb);
4147
4148         read_lock(&bond->lock);
4149
4150         if (!BOND_IS_OK(bond))
4151                 goto out;
4152         /*
4153          * Start with the curr_active_slave that joined the bond as the
4154          * default for sending IGMP traffic.  For failover purposes one
4155          * needs to maintain some consistency for the interface that will
4156          * send the join/membership reports.  The curr_active_slave found
4157          * will send all of this type of traffic.
4158          */
4159         if ((iph->protocol == IPPROTO_IGMP) &&
4160             (skb->protocol == htons(ETH_P_IP))) {
4161
4162                 read_lock(&bond->curr_slave_lock);
4163                 slave = bond->curr_active_slave;
4164                 read_unlock(&bond->curr_slave_lock);
4165
4166                 if (!slave)
4167                         goto out;
4168         } else {
4169                 /*
4170                  * Concurrent TX may collide on rr_tx_counter; we accept
4171                  * that as being rare enough not to justify using an
4172                  * atomic op here.
4173                  */
4174                 slave_no = bond->rr_tx_counter++ % bond->slave_cnt;
4175
4176                 bond_for_each_slave(bond, slave, i) {
4177                         slave_no--;
4178                         if (slave_no < 0)
4179                                 break;
4180                 }
4181         }
4182
4183         start_at = slave;
4184         bond_for_each_slave_from(bond, slave, i, start_at) {
4185                 if (IS_UP(slave->dev) &&
4186                     (slave->link == BOND_LINK_UP) &&
4187                     (slave->state == BOND_STATE_ACTIVE)) {
4188                         res = bond_dev_queue_xmit(bond, skb, slave->dev);
4189                         break;
4190                 }
4191         }
4192
4193 out:
4194         if (res) {
4195                 /* no suitable interface, frame not sent */
4196                 dev_kfree_skb(skb);
4197         }
4198         read_unlock(&bond->lock);
4199         return NETDEV_TX_OK;
4200 }
4201
4202
4203 /*
4204  * in active-backup mode, we know that bond->curr_active_slave is always valid if
4205  * the bond has a usable interface.
4206  */
4207 static int bond_xmit_activebackup(struct sk_buff *skb, struct net_device *bond_dev)
4208 {
4209         struct bonding *bond = netdev_priv(bond_dev);
4210         int res = 1;
4211
4212         read_lock(&bond->lock);
4213         read_lock(&bond->curr_slave_lock);
4214
4215         if (!BOND_IS_OK(bond))
4216                 goto out;
4217
4218         if (!bond->curr_active_slave)
4219                 goto out;
4220
4221         res = bond_dev_queue_xmit(bond, skb, bond->curr_active_slave->dev);
4222
4223 out:
4224         if (res)
4225                 /* no suitable interface, frame not sent */
4226                 dev_kfree_skb(skb);
4227
4228         read_unlock(&bond->curr_slave_lock);
4229         read_unlock(&bond->lock);
4230         return NETDEV_TX_OK;
4231 }
4232
4233 /*
4234  * In bond_xmit_xor() , we determine the output device by using a pre-
4235  * determined xmit_hash_policy(), If the selected device is not enabled,
4236  * find the next active slave.
4237  */
4238 static int bond_xmit_xor(struct sk_buff *skb, struct net_device *bond_dev)
4239 {
4240         struct bonding *bond = netdev_priv(bond_dev);
4241         struct slave *slave, *start_at;
4242         int slave_no;
4243         int i;
4244         int res = 1;
4245
4246         read_lock(&bond->lock);
4247
4248         if (!BOND_IS_OK(bond))
4249                 goto out;
4250
4251         slave_no = bond->xmit_hash_policy(skb, bond->slave_cnt);
4252
4253         bond_for_each_slave(bond, slave, i) {
4254                 slave_no--;
4255                 if (slave_no < 0)
4256                         break;
4257         }
4258
4259         start_at = slave;
4260
4261         bond_for_each_slave_from(bond, slave, i, start_at) {
4262                 if (IS_UP(slave->dev) &&
4263                     (slave->link == BOND_LINK_UP) &&
4264                     (slave->state == BOND_STATE_ACTIVE)) {
4265                         res = bond_dev_queue_xmit(bond, skb, slave->dev);
4266                         break;
4267                 }
4268         }
4269
4270 out:
4271         if (res) {
4272                 /* no suitable interface, frame not sent */
4273                 dev_kfree_skb(skb);
4274         }
4275         read_unlock(&bond->lock);
4276         return NETDEV_TX_OK;
4277 }
4278
4279 /*
4280  * in broadcast mode, we send everything to all usable interfaces.
4281  */
4282 static int bond_xmit_broadcast(struct sk_buff *skb, struct net_device *bond_dev)
4283 {
4284         struct bonding *bond = netdev_priv(bond_dev);
4285         struct slave *slave, *start_at;
4286         struct net_device *tx_dev = NULL;
4287         int i;
4288         int res = 1;
4289
4290         read_lock(&bond->lock);
4291
4292         if (!BOND_IS_OK(bond))
4293                 goto out;
4294
4295         read_lock(&bond->curr_slave_lock);
4296         start_at = bond->curr_active_slave;
4297         read_unlock(&bond->curr_slave_lock);
4298
4299         if (!start_at)
4300                 goto out;
4301
4302         bond_for_each_slave_from(bond, slave, i, start_at) {
4303                 if (IS_UP(slave->dev) &&
4304                     (slave->link == BOND_LINK_UP) &&
4305                     (slave->state == BOND_STATE_ACTIVE)) {
4306                         if (tx_dev) {
4307                                 struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
4308                                 if (!skb2) {
4309                                         pr_err("%s: Error: bond_xmit_broadcast(): skb_clone() failed\n",
4310                                                bond_dev->name);
4311                                         continue;
4312                                 }
4313
4314                                 res = bond_dev_queue_xmit(bond, skb2, tx_dev);
4315                                 if (res) {
4316                                         dev_kfree_skb(skb2);
4317                                         continue;
4318                                 }
4319                         }
4320                         tx_dev = slave->dev;
4321                 }
4322         }
4323
4324         if (tx_dev)
4325                 res = bond_dev_queue_xmit(bond, skb, tx_dev);
4326
4327 out:
4328         if (res)
4329                 /* no suitable interface, frame not sent */
4330                 dev_kfree_skb(skb);
4331
4332         /* frame sent to all suitable interfaces */
4333         read_unlock(&bond->lock);
4334         return NETDEV_TX_OK;
4335 }
4336
4337 /*------------------------- Device initialization ---------------------------*/
4338
4339 static void bond_set_xmit_hash_policy(struct bonding *bond)
4340 {
4341         switch (bond->params.xmit_policy) {
4342         case BOND_XMIT_POLICY_LAYER23:
4343                 bond->xmit_hash_policy = bond_xmit_hash_policy_l23;
4344                 break;
4345         case BOND_XMIT_POLICY_LAYER34:
4346                 bond->xmit_hash_policy = bond_xmit_hash_policy_l34;
4347                 break;
4348         case BOND_XMIT_POLICY_LAYER2:
4349         default:
4350                 bond->xmit_hash_policy = bond_xmit_hash_policy_l2;
4351                 break;
4352         }
4353 }
4354
4355 static netdev_tx_t bond_start_xmit(struct sk_buff *skb, struct net_device *dev)
4356 {
4357         const struct bonding *bond = netdev_priv(dev);
4358
4359         switch (bond->params.mode) {
4360         case BOND_MODE_ROUNDROBIN:
4361                 return bond_xmit_roundrobin(skb, dev);
4362         case BOND_MODE_ACTIVEBACKUP:
4363                 return bond_xmit_activebackup(skb, dev);
4364         case BOND_MODE_XOR:
4365                 return bond_xmit_xor(skb, dev);
4366         case BOND_MODE_BROADCAST:
4367                 return bond_xmit_broadcast(skb, dev);
4368         case BOND_MODE_8023AD:
4369                 return bond_3ad_xmit_xor(skb, dev);
4370         case BOND_MODE_ALB:
4371         case BOND_MODE_TLB:
4372                 return bond_alb_xmit(skb, dev);
4373         default:
4374                 /* Should never happen, mode already checked */
4375                 pr_err("%s: Error: Unknown bonding mode %d\n",
4376                        dev->name, bond->params.mode);
4377                 WARN_ON_ONCE(1);
4378                 dev_kfree_skb(skb);
4379                 return NETDEV_TX_OK;
4380         }
4381 }
4382
4383
4384 /*
4385  * set bond mode specific net device operations
4386  */
4387 void bond_set_mode_ops(struct bonding *bond, int mode)
4388 {
4389         struct net_device *bond_dev = bond->dev;
4390
4391         switch (mode) {
4392         case BOND_MODE_ROUNDROBIN:
4393                 break;
4394         case BOND_MODE_ACTIVEBACKUP:
4395                 break;
4396         case BOND_MODE_XOR:
4397                 bond_set_xmit_hash_policy(bond);
4398                 break;
4399         case BOND_MODE_BROADCAST:
4400                 break;
4401         case BOND_MODE_8023AD:
4402                 bond_set_master_3ad_flags(bond);
4403                 bond_set_xmit_hash_policy(bond);
4404                 break;
4405         case BOND_MODE_ALB:
4406                 bond_set_master_alb_flags(bond);
4407                 /* FALLTHRU */
4408         case BOND_MODE_TLB:
4409                 break;
4410         default:
4411                 /* Should never happen, mode already checked */
4412                 pr_err("%s: Error: Unknown bonding mode %d\n",
4413                        bond_dev->name, mode);
4414                 break;
4415         }
4416 }
4417
4418 static void bond_ethtool_get_drvinfo(struct net_device *bond_dev,
4419                                     struct ethtool_drvinfo *drvinfo)
4420 {
4421         strncpy(drvinfo->driver, DRV_NAME, 32);
4422         strncpy(drvinfo->version, DRV_VERSION, 32);
4423         snprintf(drvinfo->fw_version, 32, "%d", BOND_ABI_VERSION);
4424 }
4425
4426 static const struct ethtool_ops bond_ethtool_ops = {
4427         .get_drvinfo            = bond_ethtool_get_drvinfo,
4428         .get_link               = ethtool_op_get_link,
4429         .get_tx_csum            = ethtool_op_get_tx_csum,
4430         .get_sg                 = ethtool_op_get_sg,
4431         .get_tso                = ethtool_op_get_tso,
4432         .get_ufo                = ethtool_op_get_ufo,
4433         .get_flags              = ethtool_op_get_flags,
4434 };
4435
4436 static const struct net_device_ops bond_netdev_ops = {
4437         .ndo_init               = bond_init,
4438         .ndo_uninit             = bond_uninit,
4439         .ndo_open               = bond_open,
4440         .ndo_stop               = bond_close,
4441         .ndo_start_xmit         = bond_start_xmit,
4442         .ndo_get_stats          = bond_get_stats,
4443         .ndo_do_ioctl           = bond_do_ioctl,
4444         .ndo_set_multicast_list = bond_set_multicast_list,
4445         .ndo_change_mtu         = bond_change_mtu,
4446         .ndo_set_mac_address    = bond_set_mac_address,
4447         .ndo_neigh_setup        = bond_neigh_setup,
4448         .ndo_vlan_rx_register   = bond_vlan_rx_register,
4449         .ndo_vlan_rx_add_vid    = bond_vlan_rx_add_vid,
4450         .ndo_vlan_rx_kill_vid   = bond_vlan_rx_kill_vid,
4451 };
4452
4453 static void bond_destructor(struct net_device *bond_dev)
4454 {
4455         struct bonding *bond = netdev_priv(bond_dev);
4456         if (bond->wq)
4457                 destroy_workqueue(bond->wq);
4458         free_netdev(bond_dev);
4459 }
4460
4461 static void bond_setup(struct net_device *bond_dev)
4462 {
4463         struct bonding *bond = netdev_priv(bond_dev);
4464
4465         /* initialize rwlocks */
4466         rwlock_init(&bond->lock);
4467         rwlock_init(&bond->curr_slave_lock);
4468
4469         bond->params = bonding_defaults;
4470
4471         /* Initialize pointers */
4472         bond->dev = bond_dev;
4473         INIT_LIST_HEAD(&bond->vlan_list);
4474
4475         /* Initialize the device entry points */
4476         ether_setup(bond_dev);
4477         bond_dev->netdev_ops = &bond_netdev_ops;
4478         bond_dev->ethtool_ops = &bond_ethtool_ops;
4479         bond_set_mode_ops(bond, bond->params.mode);
4480
4481         bond_dev->destructor = bond_destructor;
4482
4483         /* Initialize the device options */
4484         bond_dev->tx_queue_len = 0;
4485         bond_dev->flags |= IFF_MASTER|IFF_MULTICAST;
4486         bond_dev->priv_flags |= IFF_BONDING;
4487         bond_dev->priv_flags &= ~IFF_XMIT_DST_RELEASE;
4488
4489         if (bond->params.arp_interval)
4490                 bond_dev->priv_flags |= IFF_MASTER_ARPMON;
4491
4492         /* At first, we block adding VLANs. That's the only way to
4493          * prevent problems that occur when adding VLANs over an
4494          * empty bond. The block will be removed once non-challenged
4495          * slaves are enslaved.
4496          */
4497         bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
4498
4499         /* don't acquire bond device's netif_tx_lock when
4500          * transmitting */
4501         bond_dev->features |= NETIF_F_LLTX;
4502
4503         /* By default, we declare the bond to be fully
4504          * VLAN hardware accelerated capable. Special
4505          * care is taken in the various xmit functions
4506          * when there are slaves that are not hw accel
4507          * capable
4508          */
4509         bond_dev->features |= (NETIF_F_HW_VLAN_TX |
4510                                NETIF_F_HW_VLAN_RX |
4511                                NETIF_F_HW_VLAN_FILTER);
4512
4513 }
4514
4515 static void bond_work_cancel_all(struct bonding *bond)
4516 {
4517         write_lock_bh(&bond->lock);
4518         bond->kill_timers = 1;
4519         write_unlock_bh(&bond->lock);
4520
4521         if (bond->params.miimon && delayed_work_pending(&bond->mii_work))
4522                 cancel_delayed_work(&bond->mii_work);
4523
4524         if (bond->params.arp_interval && delayed_work_pending(&bond->arp_work))
4525                 cancel_delayed_work(&bond->arp_work);
4526
4527         if (bond->params.mode == BOND_MODE_ALB &&
4528             delayed_work_pending(&bond->alb_work))
4529                 cancel_delayed_work(&bond->alb_work);
4530
4531         if (bond->params.mode == BOND_MODE_8023AD &&
4532             delayed_work_pending(&bond->ad_work))
4533                 cancel_delayed_work(&bond->ad_work);
4534 }
4535
4536 /*
4537 * Destroy a bonding device.
4538 * Must be under rtnl_lock when this function is called.
4539 */
4540 static void bond_uninit(struct net_device *bond_dev)
4541 {
4542         struct bonding *bond = netdev_priv(bond_dev);
4543
4544         /* Release the bonded slaves */
4545         bond_release_all(bond_dev);
4546
4547         list_del(&bond->bond_list);
4548
4549         bond_work_cancel_all(bond);
4550
4551         bond_remove_proc_entry(bond);
4552
4553         netif_addr_lock_bh(bond_dev);
4554         bond_mc_list_destroy(bond);
4555         netif_addr_unlock_bh(bond_dev);
4556 }
4557
4558 /*------------------------- Module initialization ---------------------------*/
4559
4560 /*
4561  * Convert string input module parms.  Accept either the
4562  * number of the mode or its string name.  A bit complicated because
4563  * some mode names are substrings of other names, and calls from sysfs
4564  * may have whitespace in the name (trailing newlines, for example).
4565  */
4566 int bond_parse_parm(const char *buf, const struct bond_parm_tbl *tbl)
4567 {
4568         int modeint = -1, i, rv;
4569         char *p, modestr[BOND_MAX_MODENAME_LEN + 1] = { 0, };
4570
4571         for (p = (char *)buf; *p; p++)
4572                 if (!(isdigit(*p) || isspace(*p)))
4573                         break;
4574
4575         if (*p)
4576                 rv = sscanf(buf, "%20s", modestr);
4577         else
4578                 rv = sscanf(buf, "%d", &modeint);
4579
4580         if (!rv)
4581                 return -1;
4582
4583         for (i = 0; tbl[i].modename; i++) {
4584                 if (modeint == tbl[i].mode)
4585                         return tbl[i].mode;
4586                 if (strcmp(modestr, tbl[i].modename) == 0)
4587                         return tbl[i].mode;
4588         }
4589
4590         return -1;
4591 }
4592
4593 static int bond_check_params(struct bond_params *params)
4594 {
4595         int arp_validate_value, fail_over_mac_value, primary_reselect_value;
4596
4597         /*
4598          * Convert string parameters.
4599          */
4600         if (mode) {
4601                 bond_mode = bond_parse_parm(mode, bond_mode_tbl);
4602                 if (bond_mode == -1) {
4603                         pr_err("Error: Invalid bonding mode \"%s\"\n",
4604                                mode == NULL ? "NULL" : mode);
4605                         return -EINVAL;
4606                 }
4607         }
4608
4609         if (xmit_hash_policy) {
4610                 if ((bond_mode != BOND_MODE_XOR) &&
4611                     (bond_mode != BOND_MODE_8023AD)) {
4612                         pr_info("xmit_hash_policy param is irrelevant in mode %s\n",
4613                                bond_mode_name(bond_mode));
4614                 } else {
4615                         xmit_hashtype = bond_parse_parm(xmit_hash_policy,
4616                                                         xmit_hashtype_tbl);
4617                         if (xmit_hashtype == -1) {
4618                                 pr_err("Error: Invalid xmit_hash_policy \"%s\"\n",
4619                                        xmit_hash_policy == NULL ? "NULL" :
4620                                        xmit_hash_policy);
4621                                 return -EINVAL;
4622                         }
4623                 }
4624         }
4625
4626         if (lacp_rate) {
4627                 if (bond_mode != BOND_MODE_8023AD) {
4628                         pr_info("lacp_rate param is irrelevant in mode %s\n",
4629                                 bond_mode_name(bond_mode));
4630                 } else {
4631                         lacp_fast = bond_parse_parm(lacp_rate, bond_lacp_tbl);
4632                         if (lacp_fast == -1) {
4633                                 pr_err("Error: Invalid lacp rate \"%s\"\n",
4634                                        lacp_rate == NULL ? "NULL" : lacp_rate);
4635                                 return -EINVAL;
4636                         }
4637                 }
4638         }
4639
4640         if (ad_select) {
4641                 params->ad_select = bond_parse_parm(ad_select, ad_select_tbl);
4642                 if (params->ad_select == -1) {
4643                         pr_err("Error: Invalid ad_select \"%s\"\n",
4644                                ad_select == NULL ? "NULL" : ad_select);
4645                         return -EINVAL;
4646                 }
4647
4648                 if (bond_mode != BOND_MODE_8023AD) {
4649                         pr_warning("ad_select param only affects 802.3ad mode\n");
4650                 }
4651         } else {
4652                 params->ad_select = BOND_AD_STABLE;
4653         }
4654
4655         if (max_bonds < 0) {
4656                 pr_warning("Warning: max_bonds (%d) not in range %d-%d, so it was reset to BOND_DEFAULT_MAX_BONDS (%d)\n",
4657                            max_bonds, 0, INT_MAX, BOND_DEFAULT_MAX_BONDS);
4658                 max_bonds = BOND_DEFAULT_MAX_BONDS;
4659         }
4660
4661         if (miimon < 0) {
4662                 pr_warning("Warning: miimon module parameter (%d), not in range 0-%d, so it was reset to %d\n",
4663                            miimon, INT_MAX, BOND_LINK_MON_INTERV);
4664                 miimon = BOND_LINK_MON_INTERV;
4665         }
4666
4667         if (updelay < 0) {
4668                 pr_warning("Warning: updelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4669                            updelay, INT_MAX);
4670                 updelay = 0;
4671         }
4672
4673         if (downdelay < 0) {
4674                 pr_warning("Warning: downdelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4675                            downdelay, INT_MAX);
4676                 downdelay = 0;
4677         }
4678
4679         if ((use_carrier != 0) && (use_carrier != 1)) {
4680                 pr_warning("Warning: use_carrier module parameter (%d), not of valid value (0/1), so it was set to 1\n",
4681                            use_carrier);
4682                 use_carrier = 1;
4683         }
4684
4685         if (num_grat_arp < 0 || num_grat_arp > 255) {
4686                 pr_warning("Warning: num_grat_arp (%d) not in range 0-255 so it was reset to 1 \n",
4687                            num_grat_arp);
4688                 num_grat_arp = 1;
4689         }
4690
4691         if (num_unsol_na < 0 || num_unsol_na > 255) {
4692                 pr_warning("Warning: num_unsol_na (%d) not in range 0-255 so it was reset to 1 \n",
4693                            num_unsol_na);
4694                 num_unsol_na = 1;
4695         }
4696
4697         /* reset values for 802.3ad */
4698         if (bond_mode == BOND_MODE_8023AD) {
4699                 if (!miimon) {
4700                         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");
4701                         pr_warning("Forcing miimon to 100msec\n");
4702                         miimon = 100;
4703                 }
4704         }
4705
4706         /* reset values for TLB/ALB */
4707         if ((bond_mode == BOND_MODE_TLB) ||
4708             (bond_mode == BOND_MODE_ALB)) {
4709                 if (!miimon) {
4710                         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");
4711                         pr_warning("Forcing miimon to 100msec\n");
4712                         miimon = 100;
4713                 }
4714         }
4715
4716         if (bond_mode == BOND_MODE_ALB) {
4717                 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",
4718                           updelay);
4719         }
4720
4721         if (!miimon) {
4722                 if (updelay || downdelay) {
4723                         /* just warn the user the up/down delay will have
4724                          * no effect since miimon is zero...
4725                          */
4726                         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",
4727                                    updelay, downdelay);
4728                 }
4729         } else {
4730                 /* don't allow arp monitoring */
4731                 if (arp_interval) {
4732                         pr_warning("Warning: miimon (%d) and arp_interval (%d) can't be used simultaneously, disabling ARP monitoring\n",
4733                                    miimon, arp_interval);
4734                         arp_interval = 0;
4735                 }
4736
4737                 if ((updelay % miimon) != 0) {
4738                         pr_warning("Warning: updelay (%d) is not a multiple of miimon (%d), updelay rounded to %d ms\n",
4739                                    updelay, miimon,
4740                                    (updelay / miimon) * miimon);
4741                 }
4742
4743                 updelay /= miimon;
4744
4745                 if ((downdelay % miimon) != 0) {
4746                         pr_warning("Warning: downdelay (%d) is not a multiple of miimon (%d), downdelay rounded to %d ms\n",
4747                                    downdelay, miimon,
4748                                    (downdelay / miimon) * miimon);
4749                 }
4750
4751                 downdelay /= miimon;
4752         }
4753
4754         if (arp_interval < 0) {
4755                 pr_warning("Warning: arp_interval module parameter (%d) , not in range 0-%d, so it was reset to %d\n",
4756                            arp_interval, INT_MAX, BOND_LINK_ARP_INTERV);
4757                 arp_interval = BOND_LINK_ARP_INTERV;
4758         }
4759
4760         for (arp_ip_count = 0;
4761              (arp_ip_count < BOND_MAX_ARP_TARGETS) && arp_ip_target[arp_ip_count];
4762              arp_ip_count++) {
4763                 /* not complete check, but should be good enough to
4764                    catch mistakes */
4765                 if (!isdigit(arp_ip_target[arp_ip_count][0])) {
4766                         pr_warning("Warning: bad arp_ip_target module parameter (%s), ARP monitoring will not be performed\n",
4767                                    arp_ip_target[arp_ip_count]);
4768                         arp_interval = 0;
4769                 } else {
4770                         __be32 ip = in_aton(arp_ip_target[arp_ip_count]);
4771                         arp_target[arp_ip_count] = ip;
4772                 }
4773         }
4774
4775         if (arp_interval && !arp_ip_count) {
4776                 /* don't allow arping if no arp_ip_target given... */
4777                 pr_warning("Warning: arp_interval module parameter (%d) specified without providing an arp_ip_target parameter, arp_interval was reset to 0\n",
4778                            arp_interval);
4779                 arp_interval = 0;
4780         }
4781
4782         if (arp_validate) {
4783                 if (bond_mode != BOND_MODE_ACTIVEBACKUP) {
4784                         pr_err("arp_validate only supported in active-backup mode\n");
4785                         return -EINVAL;
4786                 }
4787                 if (!arp_interval) {
4788                         pr_err("arp_validate requires arp_interval\n");
4789                         return -EINVAL;
4790                 }
4791
4792                 arp_validate_value = bond_parse_parm(arp_validate,
4793                                                      arp_validate_tbl);
4794                 if (arp_validate_value == -1) {
4795                         pr_err("Error: invalid arp_validate \"%s\"\n",
4796                                arp_validate == NULL ? "NULL" : arp_validate);
4797                         return -EINVAL;
4798                 }
4799         } else
4800                 arp_validate_value = 0;
4801
4802         if (miimon) {
4803                 pr_info("MII link monitoring set to %d ms\n", miimon);
4804         } else if (arp_interval) {
4805                 int i;
4806
4807                 pr_info("ARP monitoring set to %d ms, validate %s, with %d target(s):",
4808                         arp_interval,
4809                         arp_validate_tbl[arp_validate_value].modename,
4810                         arp_ip_count);
4811
4812                 for (i = 0; i < arp_ip_count; i++)
4813                         pr_info(" %s", arp_ip_target[i]);
4814
4815                 pr_info("\n");
4816
4817         } else if (max_bonds) {
4818                 /* miimon and arp_interval not set, we need one so things
4819                  * work as expected, see bonding.txt for details
4820                  */
4821                 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");
4822         }
4823
4824         if (primary && !USES_PRIMARY(bond_mode)) {
4825                 /* currently, using a primary only makes sense
4826                  * in active backup, TLB or ALB modes
4827                  */
4828                 pr_warning("Warning: %s primary device specified but has no effect in %s mode\n",
4829                            primary, bond_mode_name(bond_mode));
4830                 primary = NULL;
4831         }
4832
4833         if (primary && primary_reselect) {
4834                 primary_reselect_value = bond_parse_parm(primary_reselect,
4835                                                          pri_reselect_tbl);
4836                 if (primary_reselect_value == -1) {
4837                         pr_err("Error: Invalid primary_reselect \"%s\"\n",
4838                                primary_reselect ==
4839                                         NULL ? "NULL" : primary_reselect);
4840                         return -EINVAL;
4841                 }
4842         } else {
4843                 primary_reselect_value = BOND_PRI_RESELECT_ALWAYS;
4844         }
4845
4846         if (fail_over_mac) {
4847                 fail_over_mac_value = bond_parse_parm(fail_over_mac,
4848                                                       fail_over_mac_tbl);
4849                 if (fail_over_mac_value == -1) {
4850                         pr_err("Error: invalid fail_over_mac \"%s\"\n",
4851                                arp_validate == NULL ? "NULL" : arp_validate);
4852                         return -EINVAL;
4853                 }
4854
4855                 if (bond_mode != BOND_MODE_ACTIVEBACKUP)
4856                         pr_warning("Warning: fail_over_mac only affects active-backup mode.\n");
4857         } else {
4858                 fail_over_mac_value = BOND_FOM_NONE;
4859         }
4860
4861         /* fill params struct with the proper values */
4862         params->mode = bond_mode;
4863         params->xmit_policy = xmit_hashtype;
4864         params->miimon = miimon;
4865         params->num_grat_arp = num_grat_arp;
4866         params->num_unsol_na = num_unsol_na;
4867         params->arp_interval = arp_interval;
4868         params->arp_validate = arp_validate_value;
4869         params->updelay = updelay;
4870         params->downdelay = downdelay;
4871         params->use_carrier = use_carrier;
4872         params->lacp_fast = lacp_fast;
4873         params->primary[0] = 0;
4874         params->primary_reselect = primary_reselect_value;
4875         params->fail_over_mac = fail_over_mac_value;
4876
4877         if (primary) {
4878                 strncpy(params->primary, primary, IFNAMSIZ);
4879                 params->primary[IFNAMSIZ - 1] = 0;
4880         }
4881
4882         memcpy(params->arp_targets, arp_target, sizeof(arp_target));
4883
4884         return 0;
4885 }
4886
4887 static struct lock_class_key bonding_netdev_xmit_lock_key;
4888 static struct lock_class_key bonding_netdev_addr_lock_key;
4889
4890 static void bond_set_lockdep_class_one(struct net_device *dev,
4891                                        struct netdev_queue *txq,
4892                                        void *_unused)
4893 {
4894         lockdep_set_class(&txq->_xmit_lock,
4895                           &bonding_netdev_xmit_lock_key);
4896 }
4897
4898 static void bond_set_lockdep_class(struct net_device *dev)
4899 {
4900         lockdep_set_class(&dev->addr_list_lock,
4901                           &bonding_netdev_addr_lock_key);
4902         netdev_for_each_tx_queue(dev, bond_set_lockdep_class_one, NULL);
4903 }
4904
4905 /*
4906  * Called from registration process
4907  */
4908 static int bond_init(struct net_device *bond_dev)
4909 {
4910         struct bonding *bond = netdev_priv(bond_dev);
4911         struct bond_net *bn = net_generic(dev_net(bond_dev), bond_net_id);
4912
4913         pr_debug("Begin bond_init for %s\n", bond_dev->name);
4914
4915         bond->wq = create_singlethread_workqueue(bond_dev->name);
4916         if (!bond->wq)
4917                 return -ENOMEM;
4918
4919         bond_set_lockdep_class(bond_dev);
4920
4921         netif_carrier_off(bond_dev);
4922
4923         bond_create_proc_entry(bond);
4924         list_add_tail(&bond->bond_list, &bn->dev_list);
4925
4926         bond_prepare_sysfs_group(bond);
4927         return 0;
4928 }
4929
4930 static int bond_validate(struct nlattr *tb[], struct nlattr *data[])
4931 {
4932         if (tb[IFLA_ADDRESS]) {
4933                 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN)
4934                         return -EINVAL;
4935                 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS])))
4936                         return -EADDRNOTAVAIL;
4937         }
4938         return 0;
4939 }
4940
4941 static struct rtnl_link_ops bond_link_ops __read_mostly = {
4942         .kind           = "bond",
4943         .priv_size      = sizeof(struct bonding),
4944         .setup          = bond_setup,
4945         .validate       = bond_validate,
4946 };
4947
4948 /* Create a new bond based on the specified name and bonding parameters.
4949  * If name is NULL, obtain a suitable "bond%d" name for us.
4950  * Caller must NOT hold rtnl_lock; we need to release it here before we
4951  * set up our sysfs entries.
4952  */
4953 int bond_create(struct net *net, const char *name)
4954 {
4955         struct net_device *bond_dev;
4956         int res;
4957
4958         rtnl_lock();
4959
4960         bond_dev = alloc_netdev(sizeof(struct bonding), name ? name : "",
4961                                 bond_setup);
4962         if (!bond_dev) {
4963                 pr_err("%s: eek! can't alloc netdev!\n", name);
4964                 rtnl_unlock();
4965                 return -ENOMEM;
4966         }
4967
4968         dev_net_set(bond_dev, net);
4969         bond_dev->rtnl_link_ops = &bond_link_ops;
4970
4971         if (!name) {
4972                 res = dev_alloc_name(bond_dev, "bond%d");
4973                 if (res < 0)
4974                         goto out;
4975         }
4976
4977         res = register_netdevice(bond_dev);
4978
4979 out:
4980         rtnl_unlock();
4981         if (res < 0)
4982                 bond_destructor(bond_dev);
4983         return res;
4984 }
4985
4986 static int __net_init bond_net_init(struct net *net)
4987 {
4988         struct bond_net *bn = net_generic(net, bond_net_id);
4989
4990         bn->net = net;
4991         INIT_LIST_HEAD(&bn->dev_list);
4992
4993         bond_create_proc_dir(bn);
4994         
4995         return 0;
4996 }
4997
4998 static void __net_exit bond_net_exit(struct net *net)
4999 {
5000         struct bond_net *bn = net_generic(net, bond_net_id);
5001
5002         bond_destroy_proc_dir(bn);
5003 }
5004
5005 static struct pernet_operations bond_net_ops = {
5006         .init = bond_net_init,
5007         .exit = bond_net_exit,
5008         .id   = &bond_net_id,
5009         .size = sizeof(struct bond_net),
5010 };
5011
5012 static int __init bonding_init(void)
5013 {
5014         int i;
5015         int res;
5016
5017         pr_info("%s", version);
5018
5019         res = bond_check_params(&bonding_defaults);
5020         if (res)
5021                 goto out;
5022
5023         res = register_pernet_subsys(&bond_net_ops);
5024         if (res)
5025                 goto out;
5026
5027         res = rtnl_link_register(&bond_link_ops);
5028         if (res)
5029                 goto err_link;
5030
5031         for (i = 0; i < max_bonds; i++) {
5032                 res = bond_create(&init_net, NULL);
5033                 if (res)
5034                         goto err;
5035         }
5036
5037         res = bond_create_sysfs();
5038         if (res)
5039                 goto err;
5040
5041         register_netdevice_notifier(&bond_netdev_notifier);
5042         register_inetaddr_notifier(&bond_inetaddr_notifier);
5043         bond_register_ipv6_notifier();
5044 out:
5045         return res;
5046 err:
5047         rtnl_link_unregister(&bond_link_ops);
5048 err_link:
5049         unregister_pernet_subsys(&bond_net_ops);
5050         goto out;
5051
5052 }
5053
5054 static void __exit bonding_exit(void)
5055 {
5056         unregister_netdevice_notifier(&bond_netdev_notifier);
5057         unregister_inetaddr_notifier(&bond_inetaddr_notifier);
5058         bond_unregister_ipv6_notifier();
5059
5060         bond_destroy_sysfs();
5061
5062         rtnl_link_unregister(&bond_link_ops);
5063         unregister_pernet_subsys(&bond_net_ops);
5064 }
5065
5066 module_init(bonding_init);
5067 module_exit(bonding_exit);
5068 MODULE_LICENSE("GPL");
5069 MODULE_VERSION(DRV_VERSION);
5070 MODULE_DESCRIPTION(DRV_DESCRIPTION ", v" DRV_VERSION);
5071 MODULE_AUTHOR("Thomas Davis, tadavis@lbl.gov and many others");
5072 MODULE_ALIAS_RTNL_LINK("bond");