bonding: vlans don't get deleted on enslave failure
[projects/modsched/linux.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/dma.h>
58 #include <linux/uaccess.h>
59 #include <linux/errno.h>
60 #include <linux/netdevice.h>
61 #include <linux/inetdevice.h>
62 #include <linux/igmp.h>
63 #include <linux/etherdevice.h>
64 #include <linux/skbuff.h>
65 #include <net/sock.h>
66 #include <linux/rtnetlink.h>
67 #include <linux/smp.h>
68 #include <linux/if_ether.h>
69 #include <net/arp.h>
70 #include <linux/mii.h>
71 #include <linux/ethtool.h>
72 #include <linux/if_vlan.h>
73 #include <linux/if_bonding.h>
74 #include <linux/jiffies.h>
75 #include <linux/preempt.h>
76 #include <net/route.h>
77 #include <net/net_namespace.h>
78 #include <net/netns/generic.h>
79 #include <net/pkt_sched.h>
80 #include "bonding.h"
81 #include "bond_3ad.h"
82 #include "bond_alb.h"
83
84 /*---------------------------- Module parameters ----------------------------*/
85
86 /* monitor all links that often (in milliseconds). <=0 disables monitoring */
87 #define BOND_LINK_MON_INTERV    0
88 #define BOND_LINK_ARP_INTERV    0
89
90 static int max_bonds    = BOND_DEFAULT_MAX_BONDS;
91 static int tx_queues    = BOND_DEFAULT_TX_QUEUES;
92 static int num_peer_notif = 1;
93 static int miimon       = BOND_LINK_MON_INTERV;
94 static int updelay;
95 static int downdelay;
96 static int use_carrier  = 1;
97 static char *mode;
98 static char *primary;
99 static char *primary_reselect;
100 static char *lacp_rate;
101 static int min_links;
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 int all_slaves_active = 0;
109 static struct bond_params bonding_defaults;
110 static int resend_igmp = BOND_DEFAULT_RESEND_IGMP;
111
112 module_param(max_bonds, int, 0);
113 MODULE_PARM_DESC(max_bonds, "Max number of bonded devices");
114 module_param(tx_queues, int, 0);
115 MODULE_PARM_DESC(tx_queues, "Max number of transmit queues (default = 16)");
116 module_param_named(num_grat_arp, num_peer_notif, int, 0644);
117 MODULE_PARM_DESC(num_grat_arp, "Number of peer notifications to send on "
118                                "failover event (alias of num_unsol_na)");
119 module_param_named(num_unsol_na, num_peer_notif, int, 0644);
120 MODULE_PARM_DESC(num_unsol_na, "Number of peer notifications to send on "
121                                "failover event (alias of num_grat_arp)");
122 module_param(miimon, int, 0);
123 MODULE_PARM_DESC(miimon, "Link check interval in milliseconds");
124 module_param(updelay, int, 0);
125 MODULE_PARM_DESC(updelay, "Delay before considering link up, in milliseconds");
126 module_param(downdelay, int, 0);
127 MODULE_PARM_DESC(downdelay, "Delay before considering link down, "
128                             "in milliseconds");
129 module_param(use_carrier, int, 0);
130 MODULE_PARM_DESC(use_carrier, "Use netif_carrier_ok (vs MII ioctls) in miimon; "
131                               "0 for off, 1 for on (default)");
132 module_param(mode, charp, 0);
133 MODULE_PARM_DESC(mode, "Mode of operation; 0 for balance-rr, "
134                        "1 for active-backup, 2 for balance-xor, "
135                        "3 for broadcast, 4 for 802.3ad, 5 for balance-tlb, "
136                        "6 for balance-alb");
137 module_param(primary, charp, 0);
138 MODULE_PARM_DESC(primary, "Primary network device to use");
139 module_param(primary_reselect, charp, 0);
140 MODULE_PARM_DESC(primary_reselect, "Reselect primary slave "
141                                    "once it comes up; "
142                                    "0 for always (default), "
143                                    "1 for only if speed of primary is "
144                                    "better, "
145                                    "2 for only on active slave "
146                                    "failure");
147 module_param(lacp_rate, charp, 0);
148 MODULE_PARM_DESC(lacp_rate, "LACPDU tx rate to request from 802.3ad partner; "
149                             "0 for slow, 1 for fast");
150 module_param(ad_select, charp, 0);
151 MODULE_PARM_DESC(ad_select, "803.ad aggregation selection logic; "
152                             "0 for stable (default), 1 for bandwidth, "
153                             "2 for count");
154 module_param(min_links, int, 0);
155 MODULE_PARM_DESC(min_links, "Minimum number of available links before turning on carrier");
156
157 module_param(xmit_hash_policy, charp, 0);
158 MODULE_PARM_DESC(xmit_hash_policy, "balance-xor and 802.3ad hashing method; "
159                                    "0 for layer 2 (default), 1 for layer 3+4, "
160                                    "2 for layer 2+3");
161 module_param(arp_interval, int, 0);
162 MODULE_PARM_DESC(arp_interval, "arp interval in milliseconds");
163 module_param_array(arp_ip_target, charp, NULL, 0);
164 MODULE_PARM_DESC(arp_ip_target, "arp targets in n.n.n.n form");
165 module_param(arp_validate, charp, 0);
166 MODULE_PARM_DESC(arp_validate, "validate src/dst of ARP probes; "
167                                "0 for none (default), 1 for active, "
168                                "2 for backup, 3 for all");
169 module_param(fail_over_mac, charp, 0);
170 MODULE_PARM_DESC(fail_over_mac, "For active-backup, do not set all slaves to "
171                                 "the same MAC; 0 for none (default), "
172                                 "1 for active, 2 for follow");
173 module_param(all_slaves_active, int, 0);
174 MODULE_PARM_DESC(all_slaves_active, "Keep all frames received on an interface"
175                                      "by setting active flag for all slaves; "
176                                      "0 for never (default), 1 for always.");
177 module_param(resend_igmp, int, 0);
178 MODULE_PARM_DESC(resend_igmp, "Number of IGMP membership reports to send on "
179                               "link failure");
180
181 /*----------------------------- Global variables ----------------------------*/
182
183 #ifdef CONFIG_NET_POLL_CONTROLLER
184 atomic_t netpoll_block_tx = ATOMIC_INIT(0);
185 #endif
186
187 int bond_net_id __read_mostly;
188
189 static __be32 arp_target[BOND_MAX_ARP_TARGETS];
190 static int arp_ip_count;
191 static int bond_mode    = BOND_MODE_ROUNDROBIN;
192 static int xmit_hashtype = BOND_XMIT_POLICY_LAYER2;
193 static int lacp_fast;
194
195 const struct bond_parm_tbl bond_lacp_tbl[] = {
196 {       "slow",         AD_LACP_SLOW},
197 {       "fast",         AD_LACP_FAST},
198 {       NULL,           -1},
199 };
200
201 const struct bond_parm_tbl bond_mode_tbl[] = {
202 {       "balance-rr",           BOND_MODE_ROUNDROBIN},
203 {       "active-backup",        BOND_MODE_ACTIVEBACKUP},
204 {       "balance-xor",          BOND_MODE_XOR},
205 {       "broadcast",            BOND_MODE_BROADCAST},
206 {       "802.3ad",              BOND_MODE_8023AD},
207 {       "balance-tlb",          BOND_MODE_TLB},
208 {       "balance-alb",          BOND_MODE_ALB},
209 {       NULL,                   -1},
210 };
211
212 const struct bond_parm_tbl xmit_hashtype_tbl[] = {
213 {       "layer2",               BOND_XMIT_POLICY_LAYER2},
214 {       "layer3+4",             BOND_XMIT_POLICY_LAYER34},
215 {       "layer2+3",             BOND_XMIT_POLICY_LAYER23},
216 {       NULL,                   -1},
217 };
218
219 const struct bond_parm_tbl arp_validate_tbl[] = {
220 {       "none",                 BOND_ARP_VALIDATE_NONE},
221 {       "active",               BOND_ARP_VALIDATE_ACTIVE},
222 {       "backup",               BOND_ARP_VALIDATE_BACKUP},
223 {       "all",                  BOND_ARP_VALIDATE_ALL},
224 {       NULL,                   -1},
225 };
226
227 const struct bond_parm_tbl fail_over_mac_tbl[] = {
228 {       "none",                 BOND_FOM_NONE},
229 {       "active",               BOND_FOM_ACTIVE},
230 {       "follow",               BOND_FOM_FOLLOW},
231 {       NULL,                   -1},
232 };
233
234 const struct bond_parm_tbl pri_reselect_tbl[] = {
235 {       "always",               BOND_PRI_RESELECT_ALWAYS},
236 {       "better",               BOND_PRI_RESELECT_BETTER},
237 {       "failure",              BOND_PRI_RESELECT_FAILURE},
238 {       NULL,                   -1},
239 };
240
241 struct bond_parm_tbl ad_select_tbl[] = {
242 {       "stable",       BOND_AD_STABLE},
243 {       "bandwidth",    BOND_AD_BANDWIDTH},
244 {       "count",        BOND_AD_COUNT},
245 {       NULL,           -1},
246 };
247
248 /*-------------------------- Forward declarations ---------------------------*/
249
250 static int bond_init(struct net_device *bond_dev);
251 static void bond_uninit(struct net_device *bond_dev);
252
253 /*---------------------------- General routines -----------------------------*/
254
255 const char *bond_mode_name(int mode)
256 {
257         static const char *names[] = {
258                 [BOND_MODE_ROUNDROBIN] = "load balancing (round-robin)",
259                 [BOND_MODE_ACTIVEBACKUP] = "fault-tolerance (active-backup)",
260                 [BOND_MODE_XOR] = "load balancing (xor)",
261                 [BOND_MODE_BROADCAST] = "fault-tolerance (broadcast)",
262                 [BOND_MODE_8023AD] = "IEEE 802.3ad Dynamic link aggregation",
263                 [BOND_MODE_TLB] = "transmit load balancing",
264                 [BOND_MODE_ALB] = "adaptive load balancing",
265         };
266
267         if (mode < 0 || mode > BOND_MODE_ALB)
268                 return "unknown";
269
270         return names[mode];
271 }
272
273 /*---------------------------------- VLAN -----------------------------------*/
274
275 /**
276  * bond_add_vlan - add a new vlan id on bond
277  * @bond: bond that got the notification
278  * @vlan_id: the vlan id to add
279  *
280  * Returns -ENOMEM if allocation failed.
281  */
282 static int bond_add_vlan(struct bonding *bond, unsigned short vlan_id)
283 {
284         struct vlan_entry *vlan;
285
286         pr_debug("bond: %s, vlan id %d\n",
287                  (bond ? bond->dev->name : "None"), vlan_id);
288
289         vlan = kzalloc(sizeof(struct vlan_entry), GFP_KERNEL);
290         if (!vlan)
291                 return -ENOMEM;
292
293         INIT_LIST_HEAD(&vlan->vlan_list);
294         vlan->vlan_id = vlan_id;
295
296         write_lock_bh(&bond->lock);
297
298         list_add_tail(&vlan->vlan_list, &bond->vlan_list);
299
300         write_unlock_bh(&bond->lock);
301
302         pr_debug("added VLAN ID %d on bond %s\n", vlan_id, bond->dev->name);
303
304         return 0;
305 }
306
307 /**
308  * bond_del_vlan - delete a vlan id from bond
309  * @bond: bond that got the notification
310  * @vlan_id: the vlan id to delete
311  *
312  * returns -ENODEV if @vlan_id was not found in @bond.
313  */
314 static int bond_del_vlan(struct bonding *bond, unsigned short vlan_id)
315 {
316         struct vlan_entry *vlan;
317         int res = -ENODEV;
318
319         pr_debug("bond: %s, vlan id %d\n", bond->dev->name, vlan_id);
320
321         block_netpoll_tx();
322         write_lock_bh(&bond->lock);
323
324         list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
325                 if (vlan->vlan_id == vlan_id) {
326                         list_del(&vlan->vlan_list);
327
328                         if (bond_is_lb(bond))
329                                 bond_alb_clear_vlan(bond, vlan_id);
330
331                         pr_debug("removed VLAN ID %d from bond %s\n",
332                                  vlan_id, bond->dev->name);
333
334                         kfree(vlan);
335
336                         res = 0;
337                         goto out;
338                 }
339         }
340
341         pr_debug("couldn't find VLAN ID %d in bond %s\n",
342                  vlan_id, bond->dev->name);
343
344 out:
345         write_unlock_bh(&bond->lock);
346         unblock_netpoll_tx();
347         return res;
348 }
349
350 /**
351  * bond_next_vlan - safely skip to the next item in the vlans list.
352  * @bond: the bond we're working on
353  * @curr: item we're advancing from
354  *
355  * Returns %NULL if list is empty, bond->next_vlan if @curr is %NULL,
356  * or @curr->next otherwise (even if it is @curr itself again).
357  *
358  * Caller must hold bond->lock
359  */
360 struct vlan_entry *bond_next_vlan(struct bonding *bond, struct vlan_entry *curr)
361 {
362         struct vlan_entry *next, *last;
363
364         if (list_empty(&bond->vlan_list))
365                 return NULL;
366
367         if (!curr) {
368                 next = list_entry(bond->vlan_list.next,
369                                   struct vlan_entry, vlan_list);
370         } else {
371                 last = list_entry(bond->vlan_list.prev,
372                                   struct vlan_entry, vlan_list);
373                 if (last == curr) {
374                         next = list_entry(bond->vlan_list.next,
375                                           struct vlan_entry, vlan_list);
376                 } else {
377                         next = list_entry(curr->vlan_list.next,
378                                           struct vlan_entry, vlan_list);
379                 }
380         }
381
382         return next;
383 }
384
385 /**
386  * bond_dev_queue_xmit - Prepare skb for xmit.
387  *
388  * @bond: bond device that got this skb for tx.
389  * @skb: hw accel VLAN tagged skb to transmit
390  * @slave_dev: slave that is supposed to xmit this skbuff
391  */
392 int bond_dev_queue_xmit(struct bonding *bond, struct sk_buff *skb,
393                         struct net_device *slave_dev)
394 {
395         skb->dev = slave_dev;
396
397         BUILD_BUG_ON(sizeof(skb->queue_mapping) !=
398                      sizeof(qdisc_skb_cb(skb)->slave_dev_queue_mapping));
399         skb->queue_mapping = qdisc_skb_cb(skb)->slave_dev_queue_mapping;
400
401         if (unlikely(netpoll_tx_running(bond->dev)))
402                 bond_netpoll_send_skb(bond_get_slave_by_dev(bond, slave_dev), skb);
403         else
404                 dev_queue_xmit(skb);
405
406         return 0;
407 }
408
409 /*
410  * In the following 2 functions, bond_vlan_rx_add_vid and bond_vlan_rx_kill_vid,
411  * We don't protect the slave list iteration with a lock because:
412  * a. This operation is performed in IOCTL context,
413  * b. The operation is protected by the RTNL semaphore in the 8021q code,
414  * c. Holding a lock with BH disabled while directly calling a base driver
415  *    entry point is generally a BAD idea.
416  *
417  * The design of synchronization/protection for this operation in the 8021q
418  * module is good for one or more VLAN devices over a single physical device
419  * and cannot be extended for a teaming solution like bonding, so there is a
420  * potential race condition here where a net device from the vlan group might
421  * be referenced (either by a base driver or the 8021q code) while it is being
422  * removed from the system. However, it turns out we're not making matters
423  * worse, and if it works for regular VLAN usage it will work here too.
424 */
425
426 /**
427  * bond_vlan_rx_add_vid - Propagates adding an id to slaves
428  * @bond_dev: bonding net device that got called
429  * @vid: vlan id being added
430  */
431 static int bond_vlan_rx_add_vid(struct net_device *bond_dev, uint16_t vid)
432 {
433         struct bonding *bond = netdev_priv(bond_dev);
434         struct slave *slave, *stop_at;
435         int i, res;
436
437         bond_for_each_slave(bond, slave, i) {
438                 res = vlan_vid_add(slave->dev, vid);
439                 if (res)
440                         goto unwind;
441         }
442
443         res = bond_add_vlan(bond, vid);
444         if (res) {
445                 pr_err("%s: Error: Failed to add vlan id %d\n",
446                        bond_dev->name, vid);
447                 return res;
448         }
449
450         return 0;
451
452 unwind:
453         /* unwind from head to the slave that failed */
454         stop_at = slave;
455         bond_for_each_slave_from_to(bond, slave, i, bond->first_slave, stop_at)
456                 vlan_vid_del(slave->dev, vid);
457
458         return res;
459 }
460
461 /**
462  * bond_vlan_rx_kill_vid - Propagates deleting an id to slaves
463  * @bond_dev: bonding net device that got called
464  * @vid: vlan id being removed
465  */
466 static int bond_vlan_rx_kill_vid(struct net_device *bond_dev, uint16_t vid)
467 {
468         struct bonding *bond = netdev_priv(bond_dev);
469         struct slave *slave;
470         int i, res;
471
472         bond_for_each_slave(bond, slave, i)
473                 vlan_vid_del(slave->dev, vid);
474
475         res = bond_del_vlan(bond, vid);
476         if (res) {
477                 pr_err("%s: Error: Failed to remove vlan id %d\n",
478                        bond_dev->name, vid);
479                 return res;
480         }
481
482         return 0;
483 }
484
485 static void bond_add_vlans_on_slave(struct bonding *bond, struct net_device *slave_dev)
486 {
487         struct vlan_entry *vlan;
488         int res;
489
490         list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
491                 res = vlan_vid_add(slave_dev, vlan->vlan_id);
492                 if (res)
493                         pr_warning("%s: Failed to add vlan id %d to device %s\n",
494                                    bond->dev->name, vlan->vlan_id,
495                                    slave_dev->name);
496         }
497 }
498
499 static void bond_del_vlans_from_slave(struct bonding *bond,
500                                       struct net_device *slave_dev)
501 {
502         struct vlan_entry *vlan;
503
504         list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
505                 if (!vlan->vlan_id)
506                         continue;
507                 vlan_vid_del(slave_dev, vlan->vlan_id);
508         }
509 }
510
511 /*------------------------------- Link status -------------------------------*/
512
513 /*
514  * Set the carrier state for the master according to the state of its
515  * slaves.  If any slaves are up, the master is up.  In 802.3ad mode,
516  * do special 802.3ad magic.
517  *
518  * Returns zero if carrier state does not change, nonzero if it does.
519  */
520 static int bond_set_carrier(struct bonding *bond)
521 {
522         struct slave *slave;
523         int i;
524
525         if (bond->slave_cnt == 0)
526                 goto down;
527
528         if (bond->params.mode == BOND_MODE_8023AD)
529                 return bond_3ad_set_carrier(bond);
530
531         bond_for_each_slave(bond, slave, i) {
532                 if (slave->link == BOND_LINK_UP) {
533                         if (!netif_carrier_ok(bond->dev)) {
534                                 netif_carrier_on(bond->dev);
535                                 return 1;
536                         }
537                         return 0;
538                 }
539         }
540
541 down:
542         if (netif_carrier_ok(bond->dev)) {
543                 netif_carrier_off(bond->dev);
544                 return 1;
545         }
546         return 0;
547 }
548
549 /*
550  * Get link speed and duplex from the slave's base driver
551  * using ethtool. If for some reason the call fails or the
552  * values are invalid, set speed and duplex to -1,
553  * and return.
554  */
555 static void bond_update_speed_duplex(struct slave *slave)
556 {
557         struct net_device *slave_dev = slave->dev;
558         struct ethtool_cmd ecmd;
559         u32 slave_speed;
560         int res;
561
562         slave->speed = SPEED_UNKNOWN;
563         slave->duplex = DUPLEX_UNKNOWN;
564
565         res = __ethtool_get_settings(slave_dev, &ecmd);
566         if (res < 0)
567                 return;
568
569         slave_speed = ethtool_cmd_speed(&ecmd);
570         if (slave_speed == 0 || slave_speed == ((__u32) -1))
571                 return;
572
573         switch (ecmd.duplex) {
574         case DUPLEX_FULL:
575         case DUPLEX_HALF:
576                 break;
577         default:
578                 return;
579         }
580
581         slave->speed = slave_speed;
582         slave->duplex = ecmd.duplex;
583
584         return;
585 }
586
587 /*
588  * if <dev> supports MII link status reporting, check its link status.
589  *
590  * We either do MII/ETHTOOL ioctls, or check netif_carrier_ok(),
591  * depending upon the setting of the use_carrier parameter.
592  *
593  * Return either BMSR_LSTATUS, meaning that the link is up (or we
594  * can't tell and just pretend it is), or 0, meaning that the link is
595  * down.
596  *
597  * If reporting is non-zero, instead of faking link up, return -1 if
598  * both ETHTOOL and MII ioctls fail (meaning the device does not
599  * support them).  If use_carrier is set, return whatever it says.
600  * It'd be nice if there was a good way to tell if a driver supports
601  * netif_carrier, but there really isn't.
602  */
603 static int bond_check_dev_link(struct bonding *bond,
604                                struct net_device *slave_dev, int reporting)
605 {
606         const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
607         int (*ioctl)(struct net_device *, struct ifreq *, int);
608         struct ifreq ifr;
609         struct mii_ioctl_data *mii;
610
611         if (!reporting && !netif_running(slave_dev))
612                 return 0;
613
614         if (bond->params.use_carrier)
615                 return netif_carrier_ok(slave_dev) ? BMSR_LSTATUS : 0;
616
617         /* Try to get link status using Ethtool first. */
618         if (slave_dev->ethtool_ops->get_link)
619                 return slave_dev->ethtool_ops->get_link(slave_dev) ?
620                         BMSR_LSTATUS : 0;
621
622         /* Ethtool can't be used, fallback to MII ioctls. */
623         ioctl = slave_ops->ndo_do_ioctl;
624         if (ioctl) {
625                 /* TODO: set pointer to correct ioctl on a per team member */
626                 /*       bases to make this more efficient. that is, once  */
627                 /*       we determine the correct ioctl, we will always    */
628                 /*       call it and not the others for that team          */
629                 /*       member.                                           */
630
631                 /*
632                  * We cannot assume that SIOCGMIIPHY will also read a
633                  * register; not all network drivers (e.g., e100)
634                  * support that.
635                  */
636
637                 /* Yes, the mii is overlaid on the ifreq.ifr_ifru */
638                 strncpy(ifr.ifr_name, slave_dev->name, IFNAMSIZ);
639                 mii = if_mii(&ifr);
640                 if (IOCTL(slave_dev, &ifr, SIOCGMIIPHY) == 0) {
641                         mii->reg_num = MII_BMSR;
642                         if (IOCTL(slave_dev, &ifr, SIOCGMIIREG) == 0)
643                                 return mii->val_out & BMSR_LSTATUS;
644                 }
645         }
646
647         /*
648          * If reporting, report that either there's no dev->do_ioctl,
649          * or both SIOCGMIIREG and get_link failed (meaning that we
650          * cannot report link status).  If not reporting, pretend
651          * we're ok.
652          */
653         return reporting ? -1 : BMSR_LSTATUS;
654 }
655
656 /*----------------------------- Multicast list ------------------------------*/
657
658 /*
659  * Push the promiscuity flag down to appropriate slaves
660  */
661 static int bond_set_promiscuity(struct bonding *bond, int inc)
662 {
663         int err = 0;
664         if (USES_PRIMARY(bond->params.mode)) {
665                 /* write lock already acquired */
666                 if (bond->curr_active_slave) {
667                         err = dev_set_promiscuity(bond->curr_active_slave->dev,
668                                                   inc);
669                 }
670         } else {
671                 struct slave *slave;
672                 int i;
673                 bond_for_each_slave(bond, slave, i) {
674                         err = dev_set_promiscuity(slave->dev, inc);
675                         if (err)
676                                 return err;
677                 }
678         }
679         return err;
680 }
681
682 /*
683  * Push the allmulti flag down to all slaves
684  */
685 static int bond_set_allmulti(struct bonding *bond, int inc)
686 {
687         int err = 0;
688         if (USES_PRIMARY(bond->params.mode)) {
689                 /* write lock already acquired */
690                 if (bond->curr_active_slave) {
691                         err = dev_set_allmulti(bond->curr_active_slave->dev,
692                                                inc);
693                 }
694         } else {
695                 struct slave *slave;
696                 int i;
697                 bond_for_each_slave(bond, slave, i) {
698                         err = dev_set_allmulti(slave->dev, inc);
699                         if (err)
700                                 return err;
701                 }
702         }
703         return err;
704 }
705
706 /*
707  * Add a Multicast address to slaves
708  * according to mode
709  */
710 static void bond_mc_add(struct bonding *bond, void *addr)
711 {
712         if (USES_PRIMARY(bond->params.mode)) {
713                 /* write lock already acquired */
714                 if (bond->curr_active_slave)
715                         dev_mc_add(bond->curr_active_slave->dev, addr);
716         } else {
717                 struct slave *slave;
718                 int i;
719
720                 bond_for_each_slave(bond, slave, i)
721                         dev_mc_add(slave->dev, addr);
722         }
723 }
724
725 /*
726  * Remove a multicast address from slave
727  * according to mode
728  */
729 static void bond_mc_del(struct bonding *bond, void *addr)
730 {
731         if (USES_PRIMARY(bond->params.mode)) {
732                 /* write lock already acquired */
733                 if (bond->curr_active_slave)
734                         dev_mc_del(bond->curr_active_slave->dev, addr);
735         } else {
736                 struct slave *slave;
737                 int i;
738                 bond_for_each_slave(bond, slave, i) {
739                         dev_mc_del(slave->dev, addr);
740                 }
741         }
742 }
743
744
745 static void __bond_resend_igmp_join_requests(struct net_device *dev)
746 {
747         struct in_device *in_dev;
748
749         in_dev = __in_dev_get_rcu(dev);
750         if (in_dev)
751                 ip_mc_rejoin_groups(in_dev);
752 }
753
754 /*
755  * Retrieve the list of registered multicast addresses for the bonding
756  * device and retransmit an IGMP JOIN request to the current active
757  * slave.
758  */
759 static void bond_resend_igmp_join_requests(struct bonding *bond)
760 {
761         struct net_device *bond_dev, *vlan_dev, *upper_dev;
762         struct vlan_entry *vlan;
763
764         rcu_read_lock();
765         read_lock(&bond->lock);
766
767         bond_dev = bond->dev;
768
769         /* rejoin all groups on bond device */
770         __bond_resend_igmp_join_requests(bond_dev);
771
772         /*
773          * if bond is enslaved to a bridge,
774          * then rejoin all groups on its master
775          */
776         upper_dev = netdev_master_upper_dev_get_rcu(bond_dev);
777         if (upper_dev && upper_dev->priv_flags & IFF_EBRIDGE)
778                 __bond_resend_igmp_join_requests(upper_dev);
779
780         /* rejoin all groups on vlan devices */
781         list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
782                 vlan_dev = __vlan_find_dev_deep(bond_dev,
783                                                 vlan->vlan_id);
784                 if (vlan_dev)
785                         __bond_resend_igmp_join_requests(vlan_dev);
786         }
787
788         if (--bond->igmp_retrans > 0)
789                 queue_delayed_work(bond->wq, &bond->mcast_work, HZ/5);
790
791         read_unlock(&bond->lock);
792         rcu_read_unlock();
793 }
794
795 static void bond_resend_igmp_join_requests_delayed(struct work_struct *work)
796 {
797         struct bonding *bond = container_of(work, struct bonding,
798                                             mcast_work.work);
799         rcu_read_lock();
800         bond_resend_igmp_join_requests(bond);
801         rcu_read_unlock();
802 }
803
804 /*
805  * flush all members of flush->mc_list from device dev->mc_list
806  */
807 static void bond_mc_list_flush(struct net_device *bond_dev,
808                                struct net_device *slave_dev)
809 {
810         struct bonding *bond = netdev_priv(bond_dev);
811         struct netdev_hw_addr *ha;
812
813         netdev_for_each_mc_addr(ha, bond_dev)
814                 dev_mc_del(slave_dev, ha->addr);
815
816         if (bond->params.mode == BOND_MODE_8023AD) {
817                 /* del lacpdu mc addr from mc list */
818                 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
819
820                 dev_mc_del(slave_dev, lacpdu_multicast);
821         }
822 }
823
824 /*--------------------------- Active slave change ---------------------------*/
825
826 /*
827  * Update the mc list and multicast-related flags for the new and
828  * old active slaves (if any) according to the multicast mode, and
829  * promiscuous flags unconditionally.
830  */
831 static void bond_mc_swap(struct bonding *bond, struct slave *new_active,
832                          struct slave *old_active)
833 {
834         struct netdev_hw_addr *ha;
835
836         if (!USES_PRIMARY(bond->params.mode))
837                 /* nothing to do -  mc list is already up-to-date on
838                  * all slaves
839                  */
840                 return;
841
842         if (old_active) {
843                 if (bond->dev->flags & IFF_PROMISC)
844                         dev_set_promiscuity(old_active->dev, -1);
845
846                 if (bond->dev->flags & IFF_ALLMULTI)
847                         dev_set_allmulti(old_active->dev, -1);
848
849                 netdev_for_each_mc_addr(ha, bond->dev)
850                         dev_mc_del(old_active->dev, ha->addr);
851         }
852
853         if (new_active) {
854                 /* FIXME: Signal errors upstream. */
855                 if (bond->dev->flags & IFF_PROMISC)
856                         dev_set_promiscuity(new_active->dev, 1);
857
858                 if (bond->dev->flags & IFF_ALLMULTI)
859                         dev_set_allmulti(new_active->dev, 1);
860
861                 netdev_for_each_mc_addr(ha, bond->dev)
862                         dev_mc_add(new_active->dev, ha->addr);
863         }
864 }
865
866 /*
867  * bond_do_fail_over_mac
868  *
869  * Perform special MAC address swapping for fail_over_mac settings
870  *
871  * Called with RTNL, bond->lock for read, curr_slave_lock for write_bh.
872  */
873 static void bond_do_fail_over_mac(struct bonding *bond,
874                                   struct slave *new_active,
875                                   struct slave *old_active)
876         __releases(&bond->curr_slave_lock)
877         __releases(&bond->lock)
878         __acquires(&bond->lock)
879         __acquires(&bond->curr_slave_lock)
880 {
881         u8 tmp_mac[ETH_ALEN];
882         struct sockaddr saddr;
883         int rv;
884
885         switch (bond->params.fail_over_mac) {
886         case BOND_FOM_ACTIVE:
887                 if (new_active) {
888                         memcpy(bond->dev->dev_addr,  new_active->dev->dev_addr,
889                                new_active->dev->addr_len);
890                         write_unlock_bh(&bond->curr_slave_lock);
891                         read_unlock(&bond->lock);
892                         call_netdevice_notifiers(NETDEV_CHANGEADDR, bond->dev);
893                         read_lock(&bond->lock);
894                         write_lock_bh(&bond->curr_slave_lock);
895                 }
896                 break;
897         case BOND_FOM_FOLLOW:
898                 /*
899                  * if new_active && old_active, swap them
900                  * if just old_active, do nothing (going to no active slave)
901                  * if just new_active, set new_active to bond's MAC
902                  */
903                 if (!new_active)
904                         return;
905
906                 write_unlock_bh(&bond->curr_slave_lock);
907                 read_unlock(&bond->lock);
908
909                 if (old_active) {
910                         memcpy(tmp_mac, new_active->dev->dev_addr, ETH_ALEN);
911                         memcpy(saddr.sa_data, old_active->dev->dev_addr,
912                                ETH_ALEN);
913                         saddr.sa_family = new_active->dev->type;
914                 } else {
915                         memcpy(saddr.sa_data, bond->dev->dev_addr, ETH_ALEN);
916                         saddr.sa_family = bond->dev->type;
917                 }
918
919                 rv = dev_set_mac_address(new_active->dev, &saddr);
920                 if (rv) {
921                         pr_err("%s: Error %d setting MAC of slave %s\n",
922                                bond->dev->name, -rv, new_active->dev->name);
923                         goto out;
924                 }
925
926                 if (!old_active)
927                         goto out;
928
929                 memcpy(saddr.sa_data, tmp_mac, ETH_ALEN);
930                 saddr.sa_family = old_active->dev->type;
931
932                 rv = dev_set_mac_address(old_active->dev, &saddr);
933                 if (rv)
934                         pr_err("%s: Error %d setting MAC of slave %s\n",
935                                bond->dev->name, -rv, new_active->dev->name);
936 out:
937                 read_lock(&bond->lock);
938                 write_lock_bh(&bond->curr_slave_lock);
939                 break;
940         default:
941                 pr_err("%s: bond_do_fail_over_mac impossible: bad policy %d\n",
942                        bond->dev->name, bond->params.fail_over_mac);
943                 break;
944         }
945
946 }
947
948 static bool bond_should_change_active(struct bonding *bond)
949 {
950         struct slave *prim = bond->primary_slave;
951         struct slave *curr = bond->curr_active_slave;
952
953         if (!prim || !curr || curr->link != BOND_LINK_UP)
954                 return true;
955         if (bond->force_primary) {
956                 bond->force_primary = false;
957                 return true;
958         }
959         if (bond->params.primary_reselect == BOND_PRI_RESELECT_BETTER &&
960             (prim->speed < curr->speed ||
961              (prim->speed == curr->speed && prim->duplex <= curr->duplex)))
962                 return false;
963         if (bond->params.primary_reselect == BOND_PRI_RESELECT_FAILURE)
964                 return false;
965         return true;
966 }
967
968 /**
969  * find_best_interface - select the best available slave to be the active one
970  * @bond: our bonding struct
971  *
972  * Warning: Caller must hold curr_slave_lock for writing.
973  */
974 static struct slave *bond_find_best_slave(struct bonding *bond)
975 {
976         struct slave *new_active, *old_active;
977         struct slave *bestslave = NULL;
978         int mintime = bond->params.updelay;
979         int i;
980
981         new_active = bond->curr_active_slave;
982
983         if (!new_active) { /* there were no active slaves left */
984                 if (bond->slave_cnt > 0)   /* found one slave */
985                         new_active = bond->first_slave;
986                 else
987                         return NULL; /* still no slave, return NULL */
988         }
989
990         if ((bond->primary_slave) &&
991             bond->primary_slave->link == BOND_LINK_UP &&
992             bond_should_change_active(bond)) {
993                 new_active = bond->primary_slave;
994         }
995
996         /* remember where to stop iterating over the slaves */
997         old_active = new_active;
998
999         bond_for_each_slave_from(bond, new_active, i, old_active) {
1000                 if (new_active->link == BOND_LINK_UP) {
1001                         return new_active;
1002                 } else if (new_active->link == BOND_LINK_BACK &&
1003                            IS_UP(new_active->dev)) {
1004                         /* link up, but waiting for stabilization */
1005                         if (new_active->delay < mintime) {
1006                                 mintime = new_active->delay;
1007                                 bestslave = new_active;
1008                         }
1009                 }
1010         }
1011
1012         return bestslave;
1013 }
1014
1015 static bool bond_should_notify_peers(struct bonding *bond)
1016 {
1017         struct slave *slave = bond->curr_active_slave;
1018
1019         pr_debug("bond_should_notify_peers: bond %s slave %s\n",
1020                  bond->dev->name, slave ? slave->dev->name : "NULL");
1021
1022         if (!slave || !bond->send_peer_notif ||
1023             test_bit(__LINK_STATE_LINKWATCH_PENDING, &slave->dev->state))
1024                 return false;
1025
1026         bond->send_peer_notif--;
1027         return true;
1028 }
1029
1030 /**
1031  * change_active_interface - change the active slave into the specified one
1032  * @bond: our bonding struct
1033  * @new: the new slave to make the active one
1034  *
1035  * Set the new slave to the bond's settings and unset them on the old
1036  * curr_active_slave.
1037  * Setting include flags, mc-list, promiscuity, allmulti, etc.
1038  *
1039  * If @new's link state is %BOND_LINK_BACK we'll set it to %BOND_LINK_UP,
1040  * because it is apparently the best available slave we have, even though its
1041  * updelay hasn't timed out yet.
1042  *
1043  * If new_active is not NULL, caller must hold bond->lock for read and
1044  * curr_slave_lock for write_bh.
1045  */
1046 void bond_change_active_slave(struct bonding *bond, struct slave *new_active)
1047 {
1048         struct slave *old_active = bond->curr_active_slave;
1049
1050         if (old_active == new_active)
1051                 return;
1052
1053         if (new_active) {
1054                 new_active->jiffies = jiffies;
1055
1056                 if (new_active->link == BOND_LINK_BACK) {
1057                         if (USES_PRIMARY(bond->params.mode)) {
1058                                 pr_info("%s: making interface %s the new active one %d ms earlier.\n",
1059                                         bond->dev->name, new_active->dev->name,
1060                                         (bond->params.updelay - new_active->delay) * bond->params.miimon);
1061                         }
1062
1063                         new_active->delay = 0;
1064                         new_active->link = BOND_LINK_UP;
1065
1066                         if (bond->params.mode == BOND_MODE_8023AD)
1067                                 bond_3ad_handle_link_change(new_active, BOND_LINK_UP);
1068
1069                         if (bond_is_lb(bond))
1070                                 bond_alb_handle_link_change(bond, new_active, BOND_LINK_UP);
1071                 } else {
1072                         if (USES_PRIMARY(bond->params.mode)) {
1073                                 pr_info("%s: making interface %s the new active one.\n",
1074                                         bond->dev->name, new_active->dev->name);
1075                         }
1076                 }
1077         }
1078
1079         if (USES_PRIMARY(bond->params.mode))
1080                 bond_mc_swap(bond, new_active, old_active);
1081
1082         if (bond_is_lb(bond)) {
1083                 bond_alb_handle_active_change(bond, new_active);
1084                 if (old_active)
1085                         bond_set_slave_inactive_flags(old_active);
1086                 if (new_active)
1087                         bond_set_slave_active_flags(new_active);
1088         } else {
1089                 bond->curr_active_slave = new_active;
1090         }
1091
1092         if (bond->params.mode == BOND_MODE_ACTIVEBACKUP) {
1093                 if (old_active)
1094                         bond_set_slave_inactive_flags(old_active);
1095
1096                 if (new_active) {
1097                         bool should_notify_peers = false;
1098
1099                         bond_set_slave_active_flags(new_active);
1100
1101                         if (bond->params.fail_over_mac)
1102                                 bond_do_fail_over_mac(bond, new_active,
1103                                                       old_active);
1104
1105                         if (netif_running(bond->dev)) {
1106                                 bond->send_peer_notif =
1107                                         bond->params.num_peer_notif;
1108                                 should_notify_peers =
1109                                         bond_should_notify_peers(bond);
1110                         }
1111
1112                         write_unlock_bh(&bond->curr_slave_lock);
1113                         read_unlock(&bond->lock);
1114
1115                         call_netdevice_notifiers(NETDEV_BONDING_FAILOVER, bond->dev);
1116                         if (should_notify_peers)
1117                                 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS,
1118                                                          bond->dev);
1119
1120                         read_lock(&bond->lock);
1121                         write_lock_bh(&bond->curr_slave_lock);
1122                 }
1123         }
1124
1125         /* resend IGMP joins since active slave has changed or
1126          * all were sent on curr_active_slave.
1127          * resend only if bond is brought up with the affected
1128          * bonding modes and the retransmission is enabled */
1129         if (netif_running(bond->dev) && (bond->params.resend_igmp > 0) &&
1130             ((USES_PRIMARY(bond->params.mode) && new_active) ||
1131              bond->params.mode == BOND_MODE_ROUNDROBIN)) {
1132                 bond->igmp_retrans = bond->params.resend_igmp;
1133                 queue_delayed_work(bond->wq, &bond->mcast_work, 0);
1134         }
1135 }
1136
1137 /**
1138  * bond_select_active_slave - select a new active slave, if needed
1139  * @bond: our bonding struct
1140  *
1141  * This functions should be called when one of the following occurs:
1142  * - The old curr_active_slave has been released or lost its link.
1143  * - The primary_slave has got its link back.
1144  * - A slave has got its link back and there's no old curr_active_slave.
1145  *
1146  * Caller must hold bond->lock for read and curr_slave_lock for write_bh.
1147  */
1148 void bond_select_active_slave(struct bonding *bond)
1149 {
1150         struct slave *best_slave;
1151         int rv;
1152
1153         best_slave = bond_find_best_slave(bond);
1154         if (best_slave != bond->curr_active_slave) {
1155                 bond_change_active_slave(bond, best_slave);
1156                 rv = bond_set_carrier(bond);
1157                 if (!rv)
1158                         return;
1159
1160                 if (netif_carrier_ok(bond->dev)) {
1161                         pr_info("%s: first active interface up!\n",
1162                                 bond->dev->name);
1163                 } else {
1164                         pr_info("%s: now running without any active interface !\n",
1165                                 bond->dev->name);
1166                 }
1167         }
1168 }
1169
1170 /*--------------------------- slave list handling ---------------------------*/
1171
1172 /*
1173  * This function attaches the slave to the end of list.
1174  *
1175  * bond->lock held for writing by caller.
1176  */
1177 static void bond_attach_slave(struct bonding *bond, struct slave *new_slave)
1178 {
1179         if (bond->first_slave == NULL) { /* attaching the first slave */
1180                 new_slave->next = new_slave;
1181                 new_slave->prev = new_slave;
1182                 bond->first_slave = new_slave;
1183         } else {
1184                 new_slave->next = bond->first_slave;
1185                 new_slave->prev = bond->first_slave->prev;
1186                 new_slave->next->prev = new_slave;
1187                 new_slave->prev->next = new_slave;
1188         }
1189
1190         bond->slave_cnt++;
1191 }
1192
1193 /*
1194  * This function detaches the slave from the list.
1195  * WARNING: no check is made to verify if the slave effectively
1196  * belongs to <bond>.
1197  * Nothing is freed on return, structures are just unchained.
1198  * If any slave pointer in bond was pointing to <slave>,
1199  * it should be changed by the calling function.
1200  *
1201  * bond->lock held for writing by caller.
1202  */
1203 static void bond_detach_slave(struct bonding *bond, struct slave *slave)
1204 {
1205         if (slave->next)
1206                 slave->next->prev = slave->prev;
1207
1208         if (slave->prev)
1209                 slave->prev->next = slave->next;
1210
1211         if (bond->first_slave == slave) { /* slave is the first slave */
1212                 if (bond->slave_cnt > 1) { /* there are more slave */
1213                         bond->first_slave = slave->next;
1214                 } else {
1215                         bond->first_slave = NULL; /* slave was the last one */
1216                 }
1217         }
1218
1219         slave->next = NULL;
1220         slave->prev = NULL;
1221         bond->slave_cnt--;
1222 }
1223
1224 #ifdef CONFIG_NET_POLL_CONTROLLER
1225 static inline int slave_enable_netpoll(struct slave *slave)
1226 {
1227         struct netpoll *np;
1228         int err = 0;
1229
1230         np = kzalloc(sizeof(*np), GFP_ATOMIC);
1231         err = -ENOMEM;
1232         if (!np)
1233                 goto out;
1234
1235         err = __netpoll_setup(np, slave->dev, GFP_ATOMIC);
1236         if (err) {
1237                 kfree(np);
1238                 goto out;
1239         }
1240         slave->np = np;
1241 out:
1242         return err;
1243 }
1244 static inline void slave_disable_netpoll(struct slave *slave)
1245 {
1246         struct netpoll *np = slave->np;
1247
1248         if (!np)
1249                 return;
1250
1251         slave->np = NULL;
1252         __netpoll_free_async(np);
1253 }
1254 static inline bool slave_dev_support_netpoll(struct net_device *slave_dev)
1255 {
1256         if (slave_dev->priv_flags & IFF_DISABLE_NETPOLL)
1257                 return false;
1258         if (!slave_dev->netdev_ops->ndo_poll_controller)
1259                 return false;
1260         return true;
1261 }
1262
1263 static void bond_poll_controller(struct net_device *bond_dev)
1264 {
1265 }
1266
1267 static void __bond_netpoll_cleanup(struct bonding *bond)
1268 {
1269         struct slave *slave;
1270         int i;
1271
1272         bond_for_each_slave(bond, slave, i)
1273                 if (IS_UP(slave->dev))
1274                         slave_disable_netpoll(slave);
1275 }
1276 static void bond_netpoll_cleanup(struct net_device *bond_dev)
1277 {
1278         struct bonding *bond = netdev_priv(bond_dev);
1279
1280         read_lock(&bond->lock);
1281         __bond_netpoll_cleanup(bond);
1282         read_unlock(&bond->lock);
1283 }
1284
1285 static int bond_netpoll_setup(struct net_device *dev, struct netpoll_info *ni, gfp_t gfp)
1286 {
1287         struct bonding *bond = netdev_priv(dev);
1288         struct slave *slave;
1289         int i, err = 0;
1290
1291         read_lock(&bond->lock);
1292         bond_for_each_slave(bond, slave, i) {
1293                 err = slave_enable_netpoll(slave);
1294                 if (err) {
1295                         __bond_netpoll_cleanup(bond);
1296                         break;
1297                 }
1298         }
1299         read_unlock(&bond->lock);
1300         return err;
1301 }
1302
1303 static struct netpoll_info *bond_netpoll_info(struct bonding *bond)
1304 {
1305         return bond->dev->npinfo;
1306 }
1307
1308 #else
1309 static inline int slave_enable_netpoll(struct slave *slave)
1310 {
1311         return 0;
1312 }
1313 static inline void slave_disable_netpoll(struct slave *slave)
1314 {
1315 }
1316 static void bond_netpoll_cleanup(struct net_device *bond_dev)
1317 {
1318 }
1319 #endif
1320
1321 /*---------------------------------- IOCTL ----------------------------------*/
1322
1323 static void bond_set_dev_addr(struct net_device *bond_dev,
1324                               struct net_device *slave_dev)
1325 {
1326         pr_debug("bond_dev=%p\n", bond_dev);
1327         pr_debug("slave_dev=%p\n", slave_dev);
1328         pr_debug("slave_dev->addr_len=%d\n", slave_dev->addr_len);
1329         memcpy(bond_dev->dev_addr, slave_dev->dev_addr, slave_dev->addr_len);
1330         bond_dev->addr_assign_type = NET_ADDR_SET;
1331         call_netdevice_notifiers(NETDEV_CHANGEADDR, bond_dev);
1332 }
1333
1334 static netdev_features_t bond_fix_features(struct net_device *dev,
1335         netdev_features_t features)
1336 {
1337         struct slave *slave;
1338         struct bonding *bond = netdev_priv(dev);
1339         netdev_features_t mask;
1340         int i;
1341
1342         read_lock(&bond->lock);
1343
1344         if (!bond->first_slave) {
1345                 /* Disable adding VLANs to empty bond. But why? --mq */
1346                 features |= NETIF_F_VLAN_CHALLENGED;
1347                 goto out;
1348         }
1349
1350         mask = features;
1351         features &= ~NETIF_F_ONE_FOR_ALL;
1352         features |= NETIF_F_ALL_FOR_ALL;
1353
1354         bond_for_each_slave(bond, slave, i) {
1355                 features = netdev_increment_features(features,
1356                                                      slave->dev->features,
1357                                                      mask);
1358         }
1359
1360 out:
1361         read_unlock(&bond->lock);
1362         return features;
1363 }
1364
1365 #define BOND_VLAN_FEATURES      (NETIF_F_ALL_CSUM | NETIF_F_SG | \
1366                                  NETIF_F_FRAGLIST | NETIF_F_ALL_TSO | \
1367                                  NETIF_F_HIGHDMA | NETIF_F_LRO)
1368
1369 static void bond_compute_features(struct bonding *bond)
1370 {
1371         struct slave *slave;
1372         struct net_device *bond_dev = bond->dev;
1373         netdev_features_t vlan_features = BOND_VLAN_FEATURES;
1374         unsigned short max_hard_header_len = ETH_HLEN;
1375         unsigned int gso_max_size = GSO_MAX_SIZE;
1376         u16 gso_max_segs = GSO_MAX_SEGS;
1377         int i;
1378         unsigned int flags, dst_release_flag = IFF_XMIT_DST_RELEASE;
1379
1380         read_lock(&bond->lock);
1381
1382         if (!bond->first_slave)
1383                 goto done;
1384
1385         bond_for_each_slave(bond, slave, i) {
1386                 vlan_features = netdev_increment_features(vlan_features,
1387                         slave->dev->vlan_features, BOND_VLAN_FEATURES);
1388
1389                 dst_release_flag &= slave->dev->priv_flags;
1390                 if (slave->dev->hard_header_len > max_hard_header_len)
1391                         max_hard_header_len = slave->dev->hard_header_len;
1392
1393                 gso_max_size = min(gso_max_size, slave->dev->gso_max_size);
1394                 gso_max_segs = min(gso_max_segs, slave->dev->gso_max_segs);
1395         }
1396
1397 done:
1398         bond_dev->vlan_features = vlan_features;
1399         bond_dev->hard_header_len = max_hard_header_len;
1400         bond_dev->gso_max_segs = gso_max_segs;
1401         netif_set_gso_max_size(bond_dev, gso_max_size);
1402
1403         flags = bond_dev->priv_flags & ~IFF_XMIT_DST_RELEASE;
1404         bond_dev->priv_flags = flags | dst_release_flag;
1405
1406         read_unlock(&bond->lock);
1407
1408         netdev_change_features(bond_dev);
1409 }
1410
1411 static void bond_setup_by_slave(struct net_device *bond_dev,
1412                                 struct net_device *slave_dev)
1413 {
1414         struct bonding *bond = netdev_priv(bond_dev);
1415
1416         bond_dev->header_ops        = slave_dev->header_ops;
1417
1418         bond_dev->type              = slave_dev->type;
1419         bond_dev->hard_header_len   = slave_dev->hard_header_len;
1420         bond_dev->addr_len          = slave_dev->addr_len;
1421
1422         memcpy(bond_dev->broadcast, slave_dev->broadcast,
1423                 slave_dev->addr_len);
1424         bond->setup_by_slave = 1;
1425 }
1426
1427 /* On bonding slaves other than the currently active slave, suppress
1428  * duplicates except for alb non-mcast/bcast.
1429  */
1430 static bool bond_should_deliver_exact_match(struct sk_buff *skb,
1431                                             struct slave *slave,
1432                                             struct bonding *bond)
1433 {
1434         if (bond_is_slave_inactive(slave)) {
1435                 if (bond->params.mode == BOND_MODE_ALB &&
1436                     skb->pkt_type != PACKET_BROADCAST &&
1437                     skb->pkt_type != PACKET_MULTICAST)
1438                         return false;
1439                 return true;
1440         }
1441         return false;
1442 }
1443
1444 static rx_handler_result_t bond_handle_frame(struct sk_buff **pskb)
1445 {
1446         struct sk_buff *skb = *pskb;
1447         struct slave *slave;
1448         struct bonding *bond;
1449         int (*recv_probe)(const struct sk_buff *, struct bonding *,
1450                           struct slave *);
1451         int ret = RX_HANDLER_ANOTHER;
1452
1453         skb = skb_share_check(skb, GFP_ATOMIC);
1454         if (unlikely(!skb))
1455                 return RX_HANDLER_CONSUMED;
1456
1457         *pskb = skb;
1458
1459         slave = bond_slave_get_rcu(skb->dev);
1460         bond = slave->bond;
1461
1462         if (bond->params.arp_interval)
1463                 slave->dev->last_rx = jiffies;
1464
1465         recv_probe = ACCESS_ONCE(bond->recv_probe);
1466         if (recv_probe) {
1467                 ret = recv_probe(skb, bond, slave);
1468                 if (ret == RX_HANDLER_CONSUMED) {
1469                         consume_skb(skb);
1470                         return ret;
1471                 }
1472         }
1473
1474         if (bond_should_deliver_exact_match(skb, slave, bond)) {
1475                 return RX_HANDLER_EXACT;
1476         }
1477
1478         skb->dev = bond->dev;
1479
1480         if (bond->params.mode == BOND_MODE_ALB &&
1481             bond->dev->priv_flags & IFF_BRIDGE_PORT &&
1482             skb->pkt_type == PACKET_HOST) {
1483
1484                 if (unlikely(skb_cow_head(skb,
1485                                           skb->data - skb_mac_header(skb)))) {
1486                         kfree_skb(skb);
1487                         return RX_HANDLER_CONSUMED;
1488                 }
1489                 memcpy(eth_hdr(skb)->h_dest, bond->dev->dev_addr, ETH_ALEN);
1490         }
1491
1492         return ret;
1493 }
1494
1495 static int bond_master_upper_dev_link(struct net_device *bond_dev,
1496                                       struct net_device *slave_dev)
1497 {
1498         int err;
1499
1500         err = netdev_master_upper_dev_link(slave_dev, bond_dev);
1501         if (err)
1502                 return err;
1503         slave_dev->flags |= IFF_SLAVE;
1504         rtmsg_ifinfo(RTM_NEWLINK, slave_dev, IFF_SLAVE);
1505         return 0;
1506 }
1507
1508 static void bond_upper_dev_unlink(struct net_device *bond_dev,
1509                                   struct net_device *slave_dev)
1510 {
1511         netdev_upper_dev_unlink(slave_dev, bond_dev);
1512         slave_dev->flags &= ~IFF_SLAVE;
1513         rtmsg_ifinfo(RTM_NEWLINK, slave_dev, IFF_SLAVE);
1514 }
1515
1516 /* enslave device <slave> to bond device <master> */
1517 int bond_enslave(struct net_device *bond_dev, struct net_device *slave_dev)
1518 {
1519         struct bonding *bond = netdev_priv(bond_dev);
1520         const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
1521         struct slave *new_slave = NULL;
1522         struct netdev_hw_addr *ha;
1523         struct sockaddr addr;
1524         int link_reporting;
1525         int res = 0;
1526
1527         if (!bond->params.use_carrier &&
1528             slave_dev->ethtool_ops->get_link == NULL &&
1529             slave_ops->ndo_do_ioctl == NULL) {
1530                 pr_warning("%s: Warning: no link monitoring support for %s\n",
1531                            bond_dev->name, slave_dev->name);
1532         }
1533
1534         /* already enslaved */
1535         if (slave_dev->flags & IFF_SLAVE) {
1536                 pr_debug("Error, Device was already enslaved\n");
1537                 return -EBUSY;
1538         }
1539
1540         /* vlan challenged mutual exclusion */
1541         /* no need to lock since we're protected by rtnl_lock */
1542         if (slave_dev->features & NETIF_F_VLAN_CHALLENGED) {
1543                 pr_debug("%s: NETIF_F_VLAN_CHALLENGED\n", slave_dev->name);
1544                 if (vlan_uses_dev(bond_dev)) {
1545                         pr_err("%s: Error: cannot enslave VLAN challenged slave %s on VLAN enabled bond %s\n",
1546                                bond_dev->name, slave_dev->name, bond_dev->name);
1547                         return -EPERM;
1548                 } else {
1549                         pr_warning("%s: Warning: enslaved VLAN challenged slave %s. Adding VLANs will be blocked as long as %s is part of bond %s\n",
1550                                    bond_dev->name, slave_dev->name,
1551                                    slave_dev->name, bond_dev->name);
1552                 }
1553         } else {
1554                 pr_debug("%s: ! NETIF_F_VLAN_CHALLENGED\n", slave_dev->name);
1555         }
1556
1557         /*
1558          * Old ifenslave binaries are no longer supported.  These can
1559          * be identified with moderate accuracy by the state of the slave:
1560          * the current ifenslave will set the interface down prior to
1561          * enslaving it; the old ifenslave will not.
1562          */
1563         if ((slave_dev->flags & IFF_UP)) {
1564                 pr_err("%s is up. This may be due to an out of date ifenslave.\n",
1565                        slave_dev->name);
1566                 res = -EPERM;
1567                 goto err_undo_flags;
1568         }
1569
1570         /* set bonding device ether type by slave - bonding netdevices are
1571          * created with ether_setup, so when the slave type is not ARPHRD_ETHER
1572          * there is a need to override some of the type dependent attribs/funcs.
1573          *
1574          * bond ether type mutual exclusion - don't allow slaves of dissimilar
1575          * ether type (eg ARPHRD_ETHER and ARPHRD_INFINIBAND) share the same bond
1576          */
1577         if (bond->slave_cnt == 0) {
1578                 if (bond_dev->type != slave_dev->type) {
1579                         pr_debug("%s: change device type from %d to %d\n",
1580                                  bond_dev->name,
1581                                  bond_dev->type, slave_dev->type);
1582
1583                         res = call_netdevice_notifiers(NETDEV_PRE_TYPE_CHANGE,
1584                                                        bond_dev);
1585                         res = notifier_to_errno(res);
1586                         if (res) {
1587                                 pr_err("%s: refused to change device type\n",
1588                                        bond_dev->name);
1589                                 res = -EBUSY;
1590                                 goto err_undo_flags;
1591                         }
1592
1593                         /* Flush unicast and multicast addresses */
1594                         dev_uc_flush(bond_dev);
1595                         dev_mc_flush(bond_dev);
1596
1597                         if (slave_dev->type != ARPHRD_ETHER)
1598                                 bond_setup_by_slave(bond_dev, slave_dev);
1599                         else {
1600                                 ether_setup(bond_dev);
1601                                 bond_dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1602                         }
1603
1604                         call_netdevice_notifiers(NETDEV_POST_TYPE_CHANGE,
1605                                                  bond_dev);
1606                 }
1607         } else if (bond_dev->type != slave_dev->type) {
1608                 pr_err("%s ether type (%d) is different from other slaves (%d), can not enslave it.\n",
1609                        slave_dev->name,
1610                        slave_dev->type, bond_dev->type);
1611                 res = -EINVAL;
1612                 goto err_undo_flags;
1613         }
1614
1615         if (slave_ops->ndo_set_mac_address == NULL) {
1616                 if (bond->slave_cnt == 0) {
1617                         pr_warning("%s: Warning: The first slave device specified does not support setting the MAC address. Setting fail_over_mac to active.",
1618                                    bond_dev->name);
1619                         bond->params.fail_over_mac = BOND_FOM_ACTIVE;
1620                 } else if (bond->params.fail_over_mac != BOND_FOM_ACTIVE) {
1621                         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",
1622                                bond_dev->name);
1623                         res = -EOPNOTSUPP;
1624                         goto err_undo_flags;
1625                 }
1626         }
1627
1628         call_netdevice_notifiers(NETDEV_JOIN, slave_dev);
1629
1630         /* If this is the first slave, then we need to set the master's hardware
1631          * address to be the same as the slave's. */
1632         if (bond->slave_cnt == 0 && bond->dev_addr_from_first)
1633                 bond_set_dev_addr(bond->dev, slave_dev);
1634
1635         new_slave = kzalloc(sizeof(struct slave), GFP_KERNEL);
1636         if (!new_slave) {
1637                 res = -ENOMEM;
1638                 goto err_undo_flags;
1639         }
1640
1641         /*
1642          * Set the new_slave's queue_id to be zero.  Queue ID mapping
1643          * is set via sysfs or module option if desired.
1644          */
1645         new_slave->queue_id = 0;
1646
1647         /* Save slave's original mtu and then set it to match the bond */
1648         new_slave->original_mtu = slave_dev->mtu;
1649         res = dev_set_mtu(slave_dev, bond->dev->mtu);
1650         if (res) {
1651                 pr_debug("Error %d calling dev_set_mtu\n", res);
1652                 goto err_free;
1653         }
1654
1655         /*
1656          * Save slave's original ("permanent") mac address for modes
1657          * that need it, and for restoring it upon release, and then
1658          * set it to the master's address
1659          */
1660         memcpy(new_slave->perm_hwaddr, slave_dev->dev_addr, ETH_ALEN);
1661
1662         if (!bond->params.fail_over_mac) {
1663                 /*
1664                  * Set slave to master's mac address.  The application already
1665                  * set the master's mac address to that of the first slave
1666                  */
1667                 memcpy(addr.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
1668                 addr.sa_family = slave_dev->type;
1669                 res = dev_set_mac_address(slave_dev, &addr);
1670                 if (res) {
1671                         pr_debug("Error %d calling set_mac_address\n", res);
1672                         goto err_restore_mtu;
1673                 }
1674         }
1675
1676         res = bond_master_upper_dev_link(bond_dev, slave_dev);
1677         if (res) {
1678                 pr_debug("Error %d calling bond_master_upper_dev_link\n", res);
1679                 goto err_restore_mac;
1680         }
1681
1682         /* open the slave since the application closed it */
1683         res = dev_open(slave_dev);
1684         if (res) {
1685                 pr_debug("Opening slave %s failed\n", slave_dev->name);
1686                 goto err_unset_master;
1687         }
1688
1689         new_slave->bond = bond;
1690         new_slave->dev = slave_dev;
1691         slave_dev->priv_flags |= IFF_BONDING;
1692
1693         if (bond_is_lb(bond)) {
1694                 /* bond_alb_init_slave() must be called before all other stages since
1695                  * it might fail and we do not want to have to undo everything
1696                  */
1697                 res = bond_alb_init_slave(bond, new_slave);
1698                 if (res)
1699                         goto err_close;
1700         }
1701
1702         /* If the mode USES_PRIMARY, then the new slave gets the
1703          * master's promisc (and mc) settings only if it becomes the
1704          * curr_active_slave, and that is taken care of later when calling
1705          * bond_change_active()
1706          */
1707         if (!USES_PRIMARY(bond->params.mode)) {
1708                 /* set promiscuity level to new slave */
1709                 if (bond_dev->flags & IFF_PROMISC) {
1710                         res = dev_set_promiscuity(slave_dev, 1);
1711                         if (res)
1712                                 goto err_close;
1713                 }
1714
1715                 /* set allmulti level to new slave */
1716                 if (bond_dev->flags & IFF_ALLMULTI) {
1717                         res = dev_set_allmulti(slave_dev, 1);
1718                         if (res)
1719                                 goto err_close;
1720                 }
1721
1722                 netif_addr_lock_bh(bond_dev);
1723                 /* upload master's mc_list to new slave */
1724                 netdev_for_each_mc_addr(ha, bond_dev)
1725                         dev_mc_add(slave_dev, ha->addr);
1726                 netif_addr_unlock_bh(bond_dev);
1727         }
1728
1729         if (bond->params.mode == BOND_MODE_8023AD) {
1730                 /* add lacpdu mc addr to mc list */
1731                 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
1732
1733                 dev_mc_add(slave_dev, lacpdu_multicast);
1734         }
1735
1736         bond_add_vlans_on_slave(bond, slave_dev);
1737
1738         write_lock_bh(&bond->lock);
1739
1740         bond_attach_slave(bond, new_slave);
1741
1742         new_slave->delay = 0;
1743         new_slave->link_failure_count = 0;
1744
1745         write_unlock_bh(&bond->lock);
1746
1747         bond_compute_features(bond);
1748
1749         bond_update_speed_duplex(new_slave);
1750
1751         read_lock(&bond->lock);
1752
1753         new_slave->last_arp_rx = jiffies -
1754                 (msecs_to_jiffies(bond->params.arp_interval) + 1);
1755
1756         if (bond->params.miimon && !bond->params.use_carrier) {
1757                 link_reporting = bond_check_dev_link(bond, slave_dev, 1);
1758
1759                 if ((link_reporting == -1) && !bond->params.arp_interval) {
1760                         /*
1761                          * miimon is set but a bonded network driver
1762                          * does not support ETHTOOL/MII and
1763                          * arp_interval is not set.  Note: if
1764                          * use_carrier is enabled, we will never go
1765                          * here (because netif_carrier is always
1766                          * supported); thus, we don't need to change
1767                          * the messages for netif_carrier.
1768                          */
1769                         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",
1770                                bond_dev->name, slave_dev->name);
1771                 } else if (link_reporting == -1) {
1772                         /* unable get link status using mii/ethtool */
1773                         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",
1774                                    bond_dev->name, slave_dev->name);
1775                 }
1776         }
1777
1778         /* check for initial state */
1779         if (bond->params.miimon) {
1780                 if (bond_check_dev_link(bond, slave_dev, 0) == BMSR_LSTATUS) {
1781                         if (bond->params.updelay) {
1782                                 new_slave->link = BOND_LINK_BACK;
1783                                 new_slave->delay = bond->params.updelay;
1784                         } else {
1785                                 new_slave->link = BOND_LINK_UP;
1786                         }
1787                 } else {
1788                         new_slave->link = BOND_LINK_DOWN;
1789                 }
1790         } else if (bond->params.arp_interval) {
1791                 new_slave->link = (netif_carrier_ok(slave_dev) ?
1792                         BOND_LINK_UP : BOND_LINK_DOWN);
1793         } else {
1794                 new_slave->link = BOND_LINK_UP;
1795         }
1796
1797         if (new_slave->link != BOND_LINK_DOWN)
1798                 new_slave->jiffies = jiffies;
1799         pr_debug("Initial state of slave_dev is BOND_LINK_%s\n",
1800                 new_slave->link == BOND_LINK_DOWN ? "DOWN" :
1801                         (new_slave->link == BOND_LINK_UP ? "UP" : "BACK"));
1802
1803         if (USES_PRIMARY(bond->params.mode) && bond->params.primary[0]) {
1804                 /* if there is a primary slave, remember it */
1805                 if (strcmp(bond->params.primary, new_slave->dev->name) == 0) {
1806                         bond->primary_slave = new_slave;
1807                         bond->force_primary = true;
1808                 }
1809         }
1810
1811         write_lock_bh(&bond->curr_slave_lock);
1812
1813         switch (bond->params.mode) {
1814         case BOND_MODE_ACTIVEBACKUP:
1815                 bond_set_slave_inactive_flags(new_slave);
1816                 bond_select_active_slave(bond);
1817                 break;
1818         case BOND_MODE_8023AD:
1819                 /* in 802.3ad mode, the internal mechanism
1820                  * will activate the slaves in the selected
1821                  * aggregator
1822                  */
1823                 bond_set_slave_inactive_flags(new_slave);
1824                 /* if this is the first slave */
1825                 if (bond->slave_cnt == 1) {
1826                         SLAVE_AD_INFO(new_slave).id = 1;
1827                         /* Initialize AD with the number of times that the AD timer is called in 1 second
1828                          * can be called only after the mac address of the bond is set
1829                          */
1830                         bond_3ad_initialize(bond, 1000/AD_TIMER_INTERVAL);
1831                 } else {
1832                         SLAVE_AD_INFO(new_slave).id =
1833                                 SLAVE_AD_INFO(new_slave->prev).id + 1;
1834                 }
1835
1836                 bond_3ad_bind_slave(new_slave);
1837                 break;
1838         case BOND_MODE_TLB:
1839         case BOND_MODE_ALB:
1840                 bond_set_active_slave(new_slave);
1841                 bond_set_slave_inactive_flags(new_slave);
1842                 bond_select_active_slave(bond);
1843                 break;
1844         default:
1845                 pr_debug("This slave is always active in trunk mode\n");
1846
1847                 /* always active in trunk mode */
1848                 bond_set_active_slave(new_slave);
1849
1850                 /* In trunking mode there is little meaning to curr_active_slave
1851                  * anyway (it holds no special properties of the bond device),
1852                  * so we can change it without calling change_active_interface()
1853                  */
1854                 if (!bond->curr_active_slave && new_slave->link == BOND_LINK_UP)
1855                         bond->curr_active_slave = new_slave;
1856
1857                 break;
1858         } /* switch(bond_mode) */
1859
1860         write_unlock_bh(&bond->curr_slave_lock);
1861
1862         bond_set_carrier(bond);
1863
1864 #ifdef CONFIG_NET_POLL_CONTROLLER
1865         slave_dev->npinfo = bond_netpoll_info(bond);
1866         if (slave_dev->npinfo) {
1867                 if (slave_enable_netpoll(new_slave)) {
1868                         read_unlock(&bond->lock);
1869                         pr_info("Error, %s: master_dev is using netpoll, "
1870                                  "but new slave device does not support netpoll.\n",
1871                                  bond_dev->name);
1872                         res = -EBUSY;
1873                         goto err_detach;
1874                 }
1875         }
1876 #endif
1877
1878         read_unlock(&bond->lock);
1879
1880         res = bond_create_slave_symlinks(bond_dev, slave_dev);
1881         if (res)
1882                 goto err_detach;
1883
1884         res = netdev_rx_handler_register(slave_dev, bond_handle_frame,
1885                                          new_slave);
1886         if (res) {
1887                 pr_debug("Error %d calling netdev_rx_handler_register\n", res);
1888                 goto err_dest_symlinks;
1889         }
1890
1891         pr_info("%s: enslaving %s as a%s interface with a%s link.\n",
1892                 bond_dev->name, slave_dev->name,
1893                 bond_is_active_slave(new_slave) ? "n active" : " backup",
1894                 new_slave->link != BOND_LINK_DOWN ? "n up" : " down");
1895
1896         /* enslave is successful */
1897         return 0;
1898
1899 /* Undo stages on error */
1900 err_dest_symlinks:
1901         bond_destroy_slave_symlinks(bond_dev, slave_dev);
1902
1903 err_detach:
1904         if (!USES_PRIMARY(bond->params.mode)) {
1905                 netif_addr_lock_bh(bond_dev);
1906                 bond_mc_list_flush(bond_dev, slave_dev);
1907                 netif_addr_unlock_bh(bond_dev);
1908         }
1909         bond_del_vlans_from_slave(bond, slave_dev);
1910         write_lock_bh(&bond->lock);
1911         bond_detach_slave(bond, new_slave);
1912         write_unlock_bh(&bond->lock);
1913
1914 err_close:
1915         slave_dev->priv_flags &= ~IFF_BONDING;
1916         dev_close(slave_dev);
1917
1918 err_unset_master:
1919         bond_upper_dev_unlink(bond_dev, slave_dev);
1920
1921 err_restore_mac:
1922         if (!bond->params.fail_over_mac) {
1923                 /* XXX TODO - fom follow mode needs to change master's
1924                  * MAC if this slave's MAC is in use by the bond, or at
1925                  * least print a warning.
1926                  */
1927                 memcpy(addr.sa_data, new_slave->perm_hwaddr, ETH_ALEN);
1928                 addr.sa_family = slave_dev->type;
1929                 dev_set_mac_address(slave_dev, &addr);
1930         }
1931
1932 err_restore_mtu:
1933         dev_set_mtu(slave_dev, new_slave->original_mtu);
1934
1935 err_free:
1936         kfree(new_slave);
1937
1938 err_undo_flags:
1939         bond_compute_features(bond);
1940
1941         return res;
1942 }
1943
1944 /*
1945  * Try to release the slave device <slave> from the bond device <master>
1946  * It is legal to access curr_active_slave without a lock because all the function
1947  * is write-locked. If "all" is true it means that the function is being called
1948  * while destroying a bond interface and all slaves are being released.
1949  *
1950  * The rules for slave state should be:
1951  *   for Active/Backup:
1952  *     Active stays on all backups go down
1953  *   for Bonded connections:
1954  *     The first up interface should be left on and all others downed.
1955  */
1956 static int __bond_release_one(struct net_device *bond_dev,
1957                               struct net_device *slave_dev,
1958                               bool all)
1959 {
1960         struct bonding *bond = netdev_priv(bond_dev);
1961         struct slave *slave, *oldcurrent;
1962         struct sockaddr addr;
1963         netdev_features_t old_features = bond_dev->features;
1964
1965         /* slave is not a slave or master is not master of this slave */
1966         if (!(slave_dev->flags & IFF_SLAVE) ||
1967             !netdev_has_upper_dev(slave_dev, bond_dev)) {
1968                 pr_err("%s: Error: cannot release %s.\n",
1969                        bond_dev->name, slave_dev->name);
1970                 return -EINVAL;
1971         }
1972
1973         block_netpoll_tx();
1974         write_lock_bh(&bond->lock);
1975
1976         slave = bond_get_slave_by_dev(bond, slave_dev);
1977         if (!slave) {
1978                 /* not a slave of this bond */
1979                 pr_info("%s: %s not enslaved\n",
1980                         bond_dev->name, slave_dev->name);
1981                 write_unlock_bh(&bond->lock);
1982                 unblock_netpoll_tx();
1983                 return -EINVAL;
1984         }
1985
1986         write_unlock_bh(&bond->lock);
1987         /* unregister rx_handler early so bond_handle_frame wouldn't be called
1988          * for this slave anymore.
1989          */
1990         netdev_rx_handler_unregister(slave_dev);
1991         write_lock_bh(&bond->lock);
1992
1993         if (!all && !bond->params.fail_over_mac) {
1994                 if (ether_addr_equal(bond_dev->dev_addr, slave->perm_hwaddr) &&
1995                     bond->slave_cnt > 1)
1996                         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",
1997                                    bond_dev->name, slave_dev->name,
1998                                    slave->perm_hwaddr,
1999                                    bond_dev->name, slave_dev->name);
2000         }
2001
2002         /* Inform AD package of unbinding of slave. */
2003         if (bond->params.mode == BOND_MODE_8023AD) {
2004                 /* must be called before the slave is
2005                  * detached from the list
2006                  */
2007                 bond_3ad_unbind_slave(slave);
2008         }
2009
2010         pr_info("%s: releasing %s interface %s\n",
2011                 bond_dev->name,
2012                 bond_is_active_slave(slave) ? "active" : "backup",
2013                 slave_dev->name);
2014
2015         oldcurrent = bond->curr_active_slave;
2016
2017         bond->current_arp_slave = NULL;
2018
2019         /* release the slave from its bond */
2020         bond_detach_slave(bond, slave);
2021
2022         if (bond->primary_slave == slave)
2023                 bond->primary_slave = NULL;
2024
2025         if (oldcurrent == slave)
2026                 bond_change_active_slave(bond, NULL);
2027
2028         if (bond_is_lb(bond)) {
2029                 /* Must be called only after the slave has been
2030                  * detached from the list and the curr_active_slave
2031                  * has been cleared (if our_slave == old_current),
2032                  * but before a new active slave is selected.
2033                  */
2034                 write_unlock_bh(&bond->lock);
2035                 bond_alb_deinit_slave(bond, slave);
2036                 write_lock_bh(&bond->lock);
2037         }
2038
2039         if (all) {
2040                 bond->curr_active_slave = NULL;
2041         } else if (oldcurrent == slave) {
2042                 /*
2043                  * Note that we hold RTNL over this sequence, so there
2044                  * is no concern that another slave add/remove event
2045                  * will interfere.
2046                  */
2047                 write_unlock_bh(&bond->lock);
2048                 read_lock(&bond->lock);
2049                 write_lock_bh(&bond->curr_slave_lock);
2050
2051                 bond_select_active_slave(bond);
2052
2053                 write_unlock_bh(&bond->curr_slave_lock);
2054                 read_unlock(&bond->lock);
2055                 write_lock_bh(&bond->lock);
2056         }
2057
2058         if (bond->slave_cnt == 0) {
2059                 bond_set_carrier(bond);
2060                 eth_hw_addr_random(bond_dev);
2061                 bond->dev_addr_from_first = true;
2062
2063                 if (bond_vlan_used(bond)) {
2064                         pr_warning("%s: Warning: clearing HW address of %s while it still has VLANs.\n",
2065                                    bond_dev->name, bond_dev->name);
2066                         pr_warning("%s: When re-adding slaves, make sure the bond's HW address matches its VLANs'.\n",
2067                                    bond_dev->name);
2068                 }
2069         }
2070
2071         write_unlock_bh(&bond->lock);
2072         unblock_netpoll_tx();
2073
2074         if (bond->slave_cnt == 0) {
2075                 call_netdevice_notifiers(NETDEV_CHANGEADDR, bond->dev);
2076                 call_netdevice_notifiers(NETDEV_RELEASE, bond->dev);
2077         }
2078
2079         bond_compute_features(bond);
2080         if (!(bond_dev->features & NETIF_F_VLAN_CHALLENGED) &&
2081             (old_features & NETIF_F_VLAN_CHALLENGED))
2082                 pr_info("%s: last VLAN challenged slave %s left bond %s. VLAN blocking is removed\n",
2083                         bond_dev->name, slave_dev->name, bond_dev->name);
2084
2085         /* must do this from outside any spinlocks */
2086         bond_destroy_slave_symlinks(bond_dev, slave_dev);
2087
2088         bond_del_vlans_from_slave(bond, slave_dev);
2089
2090         /* If the mode USES_PRIMARY, then we should only remove its
2091          * promisc and mc settings if it was the curr_active_slave, but that was
2092          * already taken care of above when we detached the slave
2093          */
2094         if (!USES_PRIMARY(bond->params.mode)) {
2095                 /* unset promiscuity level from slave */
2096                 if (bond_dev->flags & IFF_PROMISC)
2097                         dev_set_promiscuity(slave_dev, -1);
2098
2099                 /* unset allmulti level from slave */
2100                 if (bond_dev->flags & IFF_ALLMULTI)
2101                         dev_set_allmulti(slave_dev, -1);
2102
2103                 /* flush master's mc_list from slave */
2104                 netif_addr_lock_bh(bond_dev);
2105                 bond_mc_list_flush(bond_dev, slave_dev);
2106                 netif_addr_unlock_bh(bond_dev);
2107         }
2108
2109         bond_upper_dev_unlink(bond_dev, slave_dev);
2110
2111         slave_disable_netpoll(slave);
2112
2113         /* close slave before restoring its mac address */
2114         dev_close(slave_dev);
2115
2116         if (bond->params.fail_over_mac != BOND_FOM_ACTIVE) {
2117                 /* restore original ("permanent") mac address */
2118                 memcpy(addr.sa_data, slave->perm_hwaddr, ETH_ALEN);
2119                 addr.sa_family = slave_dev->type;
2120                 dev_set_mac_address(slave_dev, &addr);
2121         }
2122
2123         dev_set_mtu(slave_dev, slave->original_mtu);
2124
2125         slave_dev->priv_flags &= ~IFF_BONDING;
2126
2127         kfree(slave);
2128
2129         return 0;  /* deletion OK */
2130 }
2131
2132 /* A wrapper used because of ndo_del_link */
2133 int bond_release(struct net_device *bond_dev, struct net_device *slave_dev)
2134 {
2135         return __bond_release_one(bond_dev, slave_dev, false);
2136 }
2137
2138 /*
2139 * First release a slave and then destroy the bond if no more slaves are left.
2140 * Must be under rtnl_lock when this function is called.
2141 */
2142 static int  bond_release_and_destroy(struct net_device *bond_dev,
2143                                      struct net_device *slave_dev)
2144 {
2145         struct bonding *bond = netdev_priv(bond_dev);
2146         int ret;
2147
2148         ret = bond_release(bond_dev, slave_dev);
2149         if ((ret == 0) && (bond->slave_cnt == 0)) {
2150                 bond_dev->priv_flags |= IFF_DISABLE_NETPOLL;
2151                 pr_info("%s: destroying bond %s.\n",
2152                         bond_dev->name, bond_dev->name);
2153                 unregister_netdevice(bond_dev);
2154         }
2155         return ret;
2156 }
2157
2158 /*
2159  * This function changes the active slave to slave <slave_dev>.
2160  * It returns -EINVAL in the following cases.
2161  *  - <slave_dev> is not found in the list.
2162  *  - There is not active slave now.
2163  *  - <slave_dev> is already active.
2164  *  - The link state of <slave_dev> is not BOND_LINK_UP.
2165  *  - <slave_dev> is not running.
2166  * In these cases, this function does nothing.
2167  * In the other cases, current_slave pointer is changed and 0 is returned.
2168  */
2169 static int bond_ioctl_change_active(struct net_device *bond_dev, struct net_device *slave_dev)
2170 {
2171         struct bonding *bond = netdev_priv(bond_dev);
2172         struct slave *old_active = NULL;
2173         struct slave *new_active = NULL;
2174         int res = 0;
2175
2176         if (!USES_PRIMARY(bond->params.mode))
2177                 return -EINVAL;
2178
2179         /* Verify that bond_dev is indeed the master of slave_dev */
2180         if (!(slave_dev->flags & IFF_SLAVE) ||
2181             !netdev_has_upper_dev(slave_dev, bond_dev))
2182                 return -EINVAL;
2183
2184         read_lock(&bond->lock);
2185
2186         read_lock(&bond->curr_slave_lock);
2187         old_active = bond->curr_active_slave;
2188         read_unlock(&bond->curr_slave_lock);
2189
2190         new_active = bond_get_slave_by_dev(bond, slave_dev);
2191
2192         /*
2193          * Changing to the current active: do nothing; return success.
2194          */
2195         if (new_active && (new_active == old_active)) {
2196                 read_unlock(&bond->lock);
2197                 return 0;
2198         }
2199
2200         if ((new_active) &&
2201             (old_active) &&
2202             (new_active->link == BOND_LINK_UP) &&
2203             IS_UP(new_active->dev)) {
2204                 block_netpoll_tx();
2205                 write_lock_bh(&bond->curr_slave_lock);
2206                 bond_change_active_slave(bond, new_active);
2207                 write_unlock_bh(&bond->curr_slave_lock);
2208                 unblock_netpoll_tx();
2209         } else
2210                 res = -EINVAL;
2211
2212         read_unlock(&bond->lock);
2213
2214         return res;
2215 }
2216
2217 static int bond_info_query(struct net_device *bond_dev, struct ifbond *info)
2218 {
2219         struct bonding *bond = netdev_priv(bond_dev);
2220
2221         info->bond_mode = bond->params.mode;
2222         info->miimon = bond->params.miimon;
2223
2224         read_lock(&bond->lock);
2225         info->num_slaves = bond->slave_cnt;
2226         read_unlock(&bond->lock);
2227
2228         return 0;
2229 }
2230
2231 static int bond_slave_info_query(struct net_device *bond_dev, struct ifslave *info)
2232 {
2233         struct bonding *bond = netdev_priv(bond_dev);
2234         struct slave *slave;
2235         int i, res = -ENODEV;
2236
2237         read_lock(&bond->lock);
2238
2239         bond_for_each_slave(bond, slave, i) {
2240                 if (i == (int)info->slave_id) {
2241                         res = 0;
2242                         strcpy(info->slave_name, slave->dev->name);
2243                         info->link = slave->link;
2244                         info->state = bond_slave_state(slave);
2245                         info->link_failure_count = slave->link_failure_count;
2246                         break;
2247                 }
2248         }
2249
2250         read_unlock(&bond->lock);
2251
2252         return res;
2253 }
2254
2255 /*-------------------------------- Monitoring -------------------------------*/
2256
2257
2258 static int bond_miimon_inspect(struct bonding *bond)
2259 {
2260         struct slave *slave;
2261         int i, link_state, commit = 0;
2262         bool ignore_updelay;
2263
2264         ignore_updelay = !bond->curr_active_slave ? true : false;
2265
2266         bond_for_each_slave(bond, slave, i) {
2267                 slave->new_link = BOND_LINK_NOCHANGE;
2268
2269                 link_state = bond_check_dev_link(bond, slave->dev, 0);
2270
2271                 switch (slave->link) {
2272                 case BOND_LINK_UP:
2273                         if (link_state)
2274                                 continue;
2275
2276                         slave->link = BOND_LINK_FAIL;
2277                         slave->delay = bond->params.downdelay;
2278                         if (slave->delay) {
2279                                 pr_info("%s: link status down for %sinterface %s, disabling it in %d ms.\n",
2280                                         bond->dev->name,
2281                                         (bond->params.mode ==
2282                                          BOND_MODE_ACTIVEBACKUP) ?
2283                                         (bond_is_active_slave(slave) ?
2284                                          "active " : "backup ") : "",
2285                                         slave->dev->name,
2286                                         bond->params.downdelay * bond->params.miimon);
2287                         }
2288                         /*FALLTHRU*/
2289                 case BOND_LINK_FAIL:
2290                         if (link_state) {
2291                                 /*
2292                                  * recovered before downdelay expired
2293                                  */
2294                                 slave->link = BOND_LINK_UP;
2295                                 slave->jiffies = jiffies;
2296                                 pr_info("%s: link status up again after %d ms for interface %s.\n",
2297                                         bond->dev->name,
2298                                         (bond->params.downdelay - slave->delay) *
2299                                         bond->params.miimon,
2300                                         slave->dev->name);
2301                                 continue;
2302                         }
2303
2304                         if (slave->delay <= 0) {
2305                                 slave->new_link = BOND_LINK_DOWN;
2306                                 commit++;
2307                                 continue;
2308                         }
2309
2310                         slave->delay--;
2311                         break;
2312
2313                 case BOND_LINK_DOWN:
2314                         if (!link_state)
2315                                 continue;
2316
2317                         slave->link = BOND_LINK_BACK;
2318                         slave->delay = bond->params.updelay;
2319
2320                         if (slave->delay) {
2321                                 pr_info("%s: link status up for interface %s, enabling it in %d ms.\n",
2322                                         bond->dev->name, slave->dev->name,
2323                                         ignore_updelay ? 0 :
2324                                         bond->params.updelay *
2325                                         bond->params.miimon);
2326                         }
2327                         /*FALLTHRU*/
2328                 case BOND_LINK_BACK:
2329                         if (!link_state) {
2330                                 slave->link = BOND_LINK_DOWN;
2331                                 pr_info("%s: link status down again after %d ms for interface %s.\n",
2332                                         bond->dev->name,
2333                                         (bond->params.updelay - slave->delay) *
2334                                         bond->params.miimon,
2335                                         slave->dev->name);
2336
2337                                 continue;
2338                         }
2339
2340                         if (ignore_updelay)
2341                                 slave->delay = 0;
2342
2343                         if (slave->delay <= 0) {
2344                                 slave->new_link = BOND_LINK_UP;
2345                                 commit++;
2346                                 ignore_updelay = false;
2347                                 continue;
2348                         }
2349
2350                         slave->delay--;
2351                         break;
2352                 }
2353         }
2354
2355         return commit;
2356 }
2357
2358 static void bond_miimon_commit(struct bonding *bond)
2359 {
2360         struct slave *slave;
2361         int i;
2362
2363         bond_for_each_slave(bond, slave, i) {
2364                 switch (slave->new_link) {
2365                 case BOND_LINK_NOCHANGE:
2366                         continue;
2367
2368                 case BOND_LINK_UP:
2369                         slave->link = BOND_LINK_UP;
2370                         slave->jiffies = jiffies;
2371
2372                         if (bond->params.mode == BOND_MODE_8023AD) {
2373                                 /* prevent it from being the active one */
2374                                 bond_set_backup_slave(slave);
2375                         } else if (bond->params.mode != BOND_MODE_ACTIVEBACKUP) {
2376                                 /* make it immediately active */
2377                                 bond_set_active_slave(slave);
2378                         } else if (slave != bond->primary_slave) {
2379                                 /* prevent it from being the active one */
2380                                 bond_set_backup_slave(slave);
2381                         }
2382
2383                         pr_info("%s: link status definitely up for interface %s, %u Mbps %s duplex.\n",
2384                                 bond->dev->name, slave->dev->name,
2385                                 slave->speed, slave->duplex ? "full" : "half");
2386
2387                         /* notify ad that the link status has changed */
2388                         if (bond->params.mode == BOND_MODE_8023AD)
2389                                 bond_3ad_handle_link_change(slave, BOND_LINK_UP);
2390
2391                         if (bond_is_lb(bond))
2392                                 bond_alb_handle_link_change(bond, slave,
2393                                                             BOND_LINK_UP);
2394
2395                         if (!bond->curr_active_slave ||
2396                             (slave == bond->primary_slave))
2397                                 goto do_failover;
2398
2399                         continue;
2400
2401                 case BOND_LINK_DOWN:
2402                         if (slave->link_failure_count < UINT_MAX)
2403                                 slave->link_failure_count++;
2404
2405                         slave->link = BOND_LINK_DOWN;
2406
2407                         if (bond->params.mode == BOND_MODE_ACTIVEBACKUP ||
2408                             bond->params.mode == BOND_MODE_8023AD)
2409                                 bond_set_slave_inactive_flags(slave);
2410
2411                         pr_info("%s: link status definitely down for interface %s, disabling it\n",
2412                                 bond->dev->name, slave->dev->name);
2413
2414                         if (bond->params.mode == BOND_MODE_8023AD)
2415                                 bond_3ad_handle_link_change(slave,
2416                                                             BOND_LINK_DOWN);
2417
2418                         if (bond_is_lb(bond))
2419                                 bond_alb_handle_link_change(bond, slave,
2420                                                             BOND_LINK_DOWN);
2421
2422                         if (slave == bond->curr_active_slave)
2423                                 goto do_failover;
2424
2425                         continue;
2426
2427                 default:
2428                         pr_err("%s: invalid new link %d on slave %s\n",
2429                                bond->dev->name, slave->new_link,
2430                                slave->dev->name);
2431                         slave->new_link = BOND_LINK_NOCHANGE;
2432
2433                         continue;
2434                 }
2435
2436 do_failover:
2437                 ASSERT_RTNL();
2438                 block_netpoll_tx();
2439                 write_lock_bh(&bond->curr_slave_lock);
2440                 bond_select_active_slave(bond);
2441                 write_unlock_bh(&bond->curr_slave_lock);
2442                 unblock_netpoll_tx();
2443         }
2444
2445         bond_set_carrier(bond);
2446 }
2447
2448 /*
2449  * bond_mii_monitor
2450  *
2451  * Really a wrapper that splits the mii monitor into two phases: an
2452  * inspection, then (if inspection indicates something needs to be done)
2453  * an acquisition of appropriate locks followed by a commit phase to
2454  * implement whatever link state changes are indicated.
2455  */
2456 void bond_mii_monitor(struct work_struct *work)
2457 {
2458         struct bonding *bond = container_of(work, struct bonding,
2459                                             mii_work.work);
2460         bool should_notify_peers = false;
2461         unsigned long delay;
2462
2463         read_lock(&bond->lock);
2464
2465         delay = msecs_to_jiffies(bond->params.miimon);
2466
2467         if (bond->slave_cnt == 0)
2468                 goto re_arm;
2469
2470         should_notify_peers = bond_should_notify_peers(bond);
2471
2472         if (bond_miimon_inspect(bond)) {
2473                 read_unlock(&bond->lock);
2474
2475                 /* Race avoidance with bond_close cancel of workqueue */
2476                 if (!rtnl_trylock()) {
2477                         read_lock(&bond->lock);
2478                         delay = 1;
2479                         should_notify_peers = false;
2480                         goto re_arm;
2481                 }
2482
2483                 read_lock(&bond->lock);
2484
2485                 bond_miimon_commit(bond);
2486
2487                 read_unlock(&bond->lock);
2488                 rtnl_unlock();  /* might sleep, hold no other locks */
2489                 read_lock(&bond->lock);
2490         }
2491
2492 re_arm:
2493         if (bond->params.miimon)
2494                 queue_delayed_work(bond->wq, &bond->mii_work, delay);
2495
2496         read_unlock(&bond->lock);
2497
2498         if (should_notify_peers) {
2499                 if (!rtnl_trylock()) {
2500                         read_lock(&bond->lock);
2501                         bond->send_peer_notif++;
2502                         read_unlock(&bond->lock);
2503                         return;
2504                 }
2505                 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, bond->dev);
2506                 rtnl_unlock();
2507         }
2508 }
2509
2510 static int bond_has_this_ip(struct bonding *bond, __be32 ip)
2511 {
2512         struct vlan_entry *vlan;
2513         struct net_device *vlan_dev;
2514
2515         if (ip == bond_confirm_addr(bond->dev, 0, ip))
2516                 return 1;
2517
2518         list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
2519                 rcu_read_lock();
2520                 vlan_dev = __vlan_find_dev_deep(bond->dev, vlan->vlan_id);
2521                 rcu_read_unlock();
2522                 if (vlan_dev && ip == bond_confirm_addr(vlan_dev, 0, ip))
2523                         return 1;
2524         }
2525
2526         return 0;
2527 }
2528
2529 /*
2530  * We go to the (large) trouble of VLAN tagging ARP frames because
2531  * switches in VLAN mode (especially if ports are configured as
2532  * "native" to a VLAN) might not pass non-tagged frames.
2533  */
2534 static void bond_arp_send(struct net_device *slave_dev, int arp_op, __be32 dest_ip, __be32 src_ip, unsigned short vlan_id)
2535 {
2536         struct sk_buff *skb;
2537
2538         pr_debug("arp %d on slave %s: dst %x src %x vid %d\n", arp_op,
2539                  slave_dev->name, dest_ip, src_ip, vlan_id);
2540
2541         skb = arp_create(arp_op, ETH_P_ARP, dest_ip, slave_dev, src_ip,
2542                          NULL, slave_dev->dev_addr, NULL);
2543
2544         if (!skb) {
2545                 pr_err("ARP packet allocation failed\n");
2546                 return;
2547         }
2548         if (vlan_id) {
2549                 skb = vlan_put_tag(skb, vlan_id);
2550                 if (!skb) {
2551                         pr_err("failed to insert VLAN tag\n");
2552                         return;
2553                 }
2554         }
2555         arp_xmit(skb);
2556 }
2557
2558
2559 static void bond_arp_send_all(struct bonding *bond, struct slave *slave)
2560 {
2561         int i, vlan_id;
2562         __be32 *targets = bond->params.arp_targets;
2563         struct vlan_entry *vlan;
2564         struct net_device *vlan_dev = NULL;
2565         struct rtable *rt;
2566
2567         for (i = 0; (i < BOND_MAX_ARP_TARGETS); i++) {
2568                 __be32 addr;
2569                 if (!targets[i])
2570                         break;
2571                 pr_debug("basa: target %x\n", targets[i]);
2572                 if (!bond_vlan_used(bond)) {
2573                         pr_debug("basa: empty vlan: arp_send\n");
2574                         addr = bond_confirm_addr(bond->dev, targets[i], 0);
2575                         bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2576                                       addr, 0);
2577                         continue;
2578                 }
2579
2580                 /*
2581                  * If VLANs are configured, we do a route lookup to
2582                  * determine which VLAN interface would be used, so we
2583                  * can tag the ARP with the proper VLAN tag.
2584                  */
2585                 rt = ip_route_output(dev_net(bond->dev), targets[i], 0,
2586                                      RTO_ONLINK, 0);
2587                 if (IS_ERR(rt)) {
2588                         if (net_ratelimit()) {
2589                                 pr_warning("%s: no route to arp_ip_target %pI4\n",
2590                                            bond->dev->name, &targets[i]);
2591                         }
2592                         continue;
2593                 }
2594
2595                 /*
2596                  * This target is not on a VLAN
2597                  */
2598                 if (rt->dst.dev == bond->dev) {
2599                         ip_rt_put(rt);
2600                         pr_debug("basa: rtdev == bond->dev: arp_send\n");
2601                         addr = bond_confirm_addr(bond->dev, targets[i], 0);
2602                         bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2603                                       addr, 0);
2604                         continue;
2605                 }
2606
2607                 vlan_id = 0;
2608                 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
2609                         rcu_read_lock();
2610                         vlan_dev = __vlan_find_dev_deep(bond->dev,
2611                                                         vlan->vlan_id);
2612                         rcu_read_unlock();
2613                         if (vlan_dev == rt->dst.dev) {
2614                                 vlan_id = vlan->vlan_id;
2615                                 pr_debug("basa: vlan match on %s %d\n",
2616                                        vlan_dev->name, vlan_id);
2617                                 break;
2618                         }
2619                 }
2620
2621                 if (vlan_id && vlan_dev) {
2622                         ip_rt_put(rt);
2623                         addr = bond_confirm_addr(vlan_dev, targets[i], 0);
2624                         bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2625                                       addr, vlan_id);
2626                         continue;
2627                 }
2628
2629                 if (net_ratelimit()) {
2630                         pr_warning("%s: no path to arp_ip_target %pI4 via rt.dev %s\n",
2631                                    bond->dev->name, &targets[i],
2632                                    rt->dst.dev ? rt->dst.dev->name : "NULL");
2633                 }
2634                 ip_rt_put(rt);
2635         }
2636 }
2637
2638 static void bond_validate_arp(struct bonding *bond, struct slave *slave, __be32 sip, __be32 tip)
2639 {
2640         int i;
2641         __be32 *targets = bond->params.arp_targets;
2642
2643         for (i = 0; (i < BOND_MAX_ARP_TARGETS) && targets[i]; i++) {
2644                 pr_debug("bva: sip %pI4 tip %pI4 t[%d] %pI4 bhti(tip) %d\n",
2645                          &sip, &tip, i, &targets[i],
2646                          bond_has_this_ip(bond, tip));
2647                 if (sip == targets[i]) {
2648                         if (bond_has_this_ip(bond, tip))
2649                                 slave->last_arp_rx = jiffies;
2650                         return;
2651                 }
2652         }
2653 }
2654
2655 static int bond_arp_rcv(const struct sk_buff *skb, struct bonding *bond,
2656                         struct slave *slave)
2657 {
2658         struct arphdr *arp = (struct arphdr *)skb->data;
2659         unsigned char *arp_ptr;
2660         __be32 sip, tip;
2661         int alen;
2662
2663         if (skb->protocol != __cpu_to_be16(ETH_P_ARP))
2664                 return RX_HANDLER_ANOTHER;
2665
2666         read_lock(&bond->lock);
2667         alen = arp_hdr_len(bond->dev);
2668
2669         pr_debug("bond_arp_rcv: bond %s skb->dev %s\n",
2670                  bond->dev->name, skb->dev->name);
2671
2672         if (alen > skb_headlen(skb)) {
2673                 arp = kmalloc(alen, GFP_ATOMIC);
2674                 if (!arp)
2675                         goto out_unlock;
2676                 if (skb_copy_bits(skb, 0, arp, alen) < 0)
2677                         goto out_unlock;
2678         }
2679
2680         if (arp->ar_hln != bond->dev->addr_len ||
2681             skb->pkt_type == PACKET_OTHERHOST ||
2682             skb->pkt_type == PACKET_LOOPBACK ||
2683             arp->ar_hrd != htons(ARPHRD_ETHER) ||
2684             arp->ar_pro != htons(ETH_P_IP) ||
2685             arp->ar_pln != 4)
2686                 goto out_unlock;
2687
2688         arp_ptr = (unsigned char *)(arp + 1);
2689         arp_ptr += bond->dev->addr_len;
2690         memcpy(&sip, arp_ptr, 4);
2691         arp_ptr += 4 + bond->dev->addr_len;
2692         memcpy(&tip, arp_ptr, 4);
2693
2694         pr_debug("bond_arp_rcv: %s %s/%d av %d sv %d sip %pI4 tip %pI4\n",
2695                  bond->dev->name, slave->dev->name, bond_slave_state(slave),
2696                  bond->params.arp_validate, slave_do_arp_validate(bond, slave),
2697                  &sip, &tip);
2698
2699         /*
2700          * Backup slaves won't see the ARP reply, but do come through
2701          * here for each ARP probe (so we swap the sip/tip to validate
2702          * the probe).  In a "redundant switch, common router" type of
2703          * configuration, the ARP probe will (hopefully) travel from
2704          * the active, through one switch, the router, then the other
2705          * switch before reaching the backup.
2706          */
2707         if (bond_is_active_slave(slave))
2708                 bond_validate_arp(bond, slave, sip, tip);
2709         else
2710                 bond_validate_arp(bond, slave, tip, sip);
2711
2712 out_unlock:
2713         read_unlock(&bond->lock);
2714         if (arp != (struct arphdr *)skb->data)
2715                 kfree(arp);
2716         return RX_HANDLER_ANOTHER;
2717 }
2718
2719 /*
2720  * this function is called regularly to monitor each slave's link
2721  * ensuring that traffic is being sent and received when arp monitoring
2722  * is used in load-balancing mode. if the adapter has been dormant, then an
2723  * arp is transmitted to generate traffic. see activebackup_arp_monitor for
2724  * arp monitoring in active backup mode.
2725  */
2726 void bond_loadbalance_arp_mon(struct work_struct *work)
2727 {
2728         struct bonding *bond = container_of(work, struct bonding,
2729                                             arp_work.work);
2730         struct slave *slave, *oldcurrent;
2731         int do_failover = 0;
2732         int delta_in_ticks, extra_ticks;
2733         int i;
2734
2735         read_lock(&bond->lock);
2736
2737         delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
2738         extra_ticks = delta_in_ticks / 2;
2739
2740         if (bond->slave_cnt == 0)
2741                 goto re_arm;
2742
2743         read_lock(&bond->curr_slave_lock);
2744         oldcurrent = bond->curr_active_slave;
2745         read_unlock(&bond->curr_slave_lock);
2746
2747         /* see if any of the previous devices are up now (i.e. they have
2748          * xmt and rcv traffic). the curr_active_slave does not come into
2749          * the picture unless it is null. also, slave->jiffies is not needed
2750          * here because we send an arp on each slave and give a slave as
2751          * long as it needs to get the tx/rx within the delta.
2752          * TODO: what about up/down delay in arp mode? it wasn't here before
2753          *       so it can wait
2754          */
2755         bond_for_each_slave(bond, slave, i) {
2756                 unsigned long trans_start = dev_trans_start(slave->dev);
2757
2758                 if (slave->link != BOND_LINK_UP) {
2759                         if (time_in_range(jiffies,
2760                                 trans_start - delta_in_ticks,
2761                                 trans_start + delta_in_ticks + extra_ticks) &&
2762                             time_in_range(jiffies,
2763                                 slave->dev->last_rx - delta_in_ticks,
2764                                 slave->dev->last_rx + delta_in_ticks + extra_ticks)) {
2765
2766                                 slave->link  = BOND_LINK_UP;
2767                                 bond_set_active_slave(slave);
2768
2769                                 /* primary_slave has no meaning in round-robin
2770                                  * mode. the window of a slave being up and
2771                                  * curr_active_slave being null after enslaving
2772                                  * is closed.
2773                                  */
2774                                 if (!oldcurrent) {
2775                                         pr_info("%s: link status definitely up for interface %s, ",
2776                                                 bond->dev->name,
2777                                                 slave->dev->name);
2778                                         do_failover = 1;
2779                                 } else {
2780                                         pr_info("%s: interface %s is now up\n",
2781                                                 bond->dev->name,
2782                                                 slave->dev->name);
2783                                 }
2784                         }
2785                 } else {
2786                         /* slave->link == BOND_LINK_UP */
2787
2788                         /* not all switches will respond to an arp request
2789                          * when the source ip is 0, so don't take the link down
2790                          * if we don't know our ip yet
2791                          */
2792                         if (!time_in_range(jiffies,
2793                                 trans_start - delta_in_ticks,
2794                                 trans_start + 2 * delta_in_ticks + extra_ticks) ||
2795                             !time_in_range(jiffies,
2796                                 slave->dev->last_rx - delta_in_ticks,
2797                                 slave->dev->last_rx + 2 * delta_in_ticks + extra_ticks)) {
2798
2799                                 slave->link  = BOND_LINK_DOWN;
2800                                 bond_set_backup_slave(slave);
2801
2802                                 if (slave->link_failure_count < UINT_MAX)
2803                                         slave->link_failure_count++;
2804
2805                                 pr_info("%s: interface %s is now down.\n",
2806                                         bond->dev->name,
2807                                         slave->dev->name);
2808
2809                                 if (slave == oldcurrent)
2810                                         do_failover = 1;
2811                         }
2812                 }
2813
2814                 /* note: if switch is in round-robin mode, all links
2815                  * must tx arp to ensure all links rx an arp - otherwise
2816                  * links may oscillate or not come up at all; if switch is
2817                  * in something like xor mode, there is nothing we can
2818                  * do - all replies will be rx'ed on same link causing slaves
2819                  * to be unstable during low/no traffic periods
2820                  */
2821                 if (IS_UP(slave->dev))
2822                         bond_arp_send_all(bond, slave);
2823         }
2824
2825         if (do_failover) {
2826                 block_netpoll_tx();
2827                 write_lock_bh(&bond->curr_slave_lock);
2828
2829                 bond_select_active_slave(bond);
2830
2831                 write_unlock_bh(&bond->curr_slave_lock);
2832                 unblock_netpoll_tx();
2833         }
2834
2835 re_arm:
2836         if (bond->params.arp_interval)
2837                 queue_delayed_work(bond->wq, &bond->arp_work, delta_in_ticks);
2838
2839         read_unlock(&bond->lock);
2840 }
2841
2842 /*
2843  * Called to inspect slaves for active-backup mode ARP monitor link state
2844  * changes.  Sets new_link in slaves to specify what action should take
2845  * place for the slave.  Returns 0 if no changes are found, >0 if changes
2846  * to link states must be committed.
2847  *
2848  * Called with bond->lock held for read.
2849  */
2850 static int bond_ab_arp_inspect(struct bonding *bond, int delta_in_ticks)
2851 {
2852         struct slave *slave;
2853         int i, commit = 0;
2854         unsigned long trans_start;
2855         int extra_ticks;
2856
2857         /* All the time comparisons below need some extra time. Otherwise, on
2858          * fast networks the ARP probe/reply may arrive within the same jiffy
2859          * as it was sent.  Then, the next time the ARP monitor is run, one
2860          * arp_interval will already have passed in the comparisons.
2861          */
2862         extra_ticks = delta_in_ticks / 2;
2863
2864         bond_for_each_slave(bond, slave, i) {
2865                 slave->new_link = BOND_LINK_NOCHANGE;
2866
2867                 if (slave->link != BOND_LINK_UP) {
2868                         if (time_in_range(jiffies,
2869                                 slave_last_rx(bond, slave) - delta_in_ticks,
2870                                 slave_last_rx(bond, slave) + delta_in_ticks + extra_ticks)) {
2871
2872                                 slave->new_link = BOND_LINK_UP;
2873                                 commit++;
2874                         }
2875
2876                         continue;
2877                 }
2878
2879                 /*
2880                  * Give slaves 2*delta after being enslaved or made
2881                  * active.  This avoids bouncing, as the last receive
2882                  * times need a full ARP monitor cycle to be updated.
2883                  */
2884                 if (time_in_range(jiffies,
2885                                   slave->jiffies - delta_in_ticks,
2886                                   slave->jiffies + 2 * delta_in_ticks + extra_ticks))
2887                         continue;
2888
2889                 /*
2890                  * Backup slave is down if:
2891                  * - No current_arp_slave AND
2892                  * - more than 3*delta since last receive AND
2893                  * - the bond has an IP address
2894                  *
2895                  * Note: a non-null current_arp_slave indicates
2896                  * the curr_active_slave went down and we are
2897                  * searching for a new one; under this condition
2898                  * we only take the curr_active_slave down - this
2899                  * gives each slave a chance to tx/rx traffic
2900                  * before being taken out
2901                  */
2902                 if (!bond_is_active_slave(slave) &&
2903                     !bond->current_arp_slave &&
2904                     !time_in_range(jiffies,
2905                         slave_last_rx(bond, slave) - delta_in_ticks,
2906                         slave_last_rx(bond, slave) + 3 * delta_in_ticks + extra_ticks)) {
2907
2908                         slave->new_link = BOND_LINK_DOWN;
2909                         commit++;
2910                 }
2911
2912                 /*
2913                  * Active slave is down if:
2914                  * - more than 2*delta since transmitting OR
2915                  * - (more than 2*delta since receive AND
2916                  *    the bond has an IP address)
2917                  */
2918                 trans_start = dev_trans_start(slave->dev);
2919                 if (bond_is_active_slave(slave) &&
2920                     (!time_in_range(jiffies,
2921                         trans_start - delta_in_ticks,
2922                         trans_start + 2 * delta_in_ticks + extra_ticks) ||
2923                      !time_in_range(jiffies,
2924                         slave_last_rx(bond, slave) - delta_in_ticks,
2925                         slave_last_rx(bond, slave) + 2 * delta_in_ticks + extra_ticks))) {
2926
2927                         slave->new_link = BOND_LINK_DOWN;
2928                         commit++;
2929                 }
2930         }
2931
2932         return commit;
2933 }
2934
2935 /*
2936  * Called to commit link state changes noted by inspection step of
2937  * active-backup mode ARP monitor.
2938  *
2939  * Called with RTNL and bond->lock for read.
2940  */
2941 static void bond_ab_arp_commit(struct bonding *bond, int delta_in_ticks)
2942 {
2943         struct slave *slave;
2944         int i;
2945         unsigned long trans_start;
2946
2947         bond_for_each_slave(bond, slave, i) {
2948                 switch (slave->new_link) {
2949                 case BOND_LINK_NOCHANGE:
2950                         continue;
2951
2952                 case BOND_LINK_UP:
2953                         trans_start = dev_trans_start(slave->dev);
2954                         if ((!bond->curr_active_slave &&
2955                              time_in_range(jiffies,
2956                                            trans_start - delta_in_ticks,
2957                                            trans_start + delta_in_ticks + delta_in_ticks / 2)) ||
2958                             bond->curr_active_slave != slave) {
2959                                 slave->link = BOND_LINK_UP;
2960                                 if (bond->current_arp_slave) {
2961                                         bond_set_slave_inactive_flags(
2962                                                 bond->current_arp_slave);
2963                                         bond->current_arp_slave = NULL;
2964                                 }
2965
2966                                 pr_info("%s: link status definitely up for interface %s.\n",
2967                                         bond->dev->name, slave->dev->name);
2968
2969                                 if (!bond->curr_active_slave ||
2970                                     (slave == bond->primary_slave))
2971                                         goto do_failover;
2972
2973                         }
2974
2975                         continue;
2976
2977                 case BOND_LINK_DOWN:
2978                         if (slave->link_failure_count < UINT_MAX)
2979                                 slave->link_failure_count++;
2980
2981                         slave->link = BOND_LINK_DOWN;
2982                         bond_set_slave_inactive_flags(slave);
2983
2984                         pr_info("%s: link status definitely down for interface %s, disabling it\n",
2985                                 bond->dev->name, slave->dev->name);
2986
2987                         if (slave == bond->curr_active_slave) {
2988                                 bond->current_arp_slave = NULL;
2989                                 goto do_failover;
2990                         }
2991
2992                         continue;
2993
2994                 default:
2995                         pr_err("%s: impossible: new_link %d on slave %s\n",
2996                                bond->dev->name, slave->new_link,
2997                                slave->dev->name);
2998                         continue;
2999                 }
3000
3001 do_failover:
3002                 ASSERT_RTNL();
3003                 block_netpoll_tx();
3004                 write_lock_bh(&bond->curr_slave_lock);
3005                 bond_select_active_slave(bond);
3006                 write_unlock_bh(&bond->curr_slave_lock);
3007                 unblock_netpoll_tx();
3008         }
3009
3010         bond_set_carrier(bond);
3011 }
3012
3013 /*
3014  * Send ARP probes for active-backup mode ARP monitor.
3015  *
3016  * Called with bond->lock held for read.
3017  */
3018 static void bond_ab_arp_probe(struct bonding *bond)
3019 {
3020         struct slave *slave;
3021         int i;
3022
3023         read_lock(&bond->curr_slave_lock);
3024
3025         if (bond->current_arp_slave && bond->curr_active_slave)
3026                 pr_info("PROBE: c_arp %s && cas %s BAD\n",
3027                         bond->current_arp_slave->dev->name,
3028                         bond->curr_active_slave->dev->name);
3029
3030         if (bond->curr_active_slave) {
3031                 bond_arp_send_all(bond, bond->curr_active_slave);
3032                 read_unlock(&bond->curr_slave_lock);
3033                 return;
3034         }
3035
3036         read_unlock(&bond->curr_slave_lock);
3037
3038         /* if we don't have a curr_active_slave, search for the next available
3039          * backup slave from the current_arp_slave and make it the candidate
3040          * for becoming the curr_active_slave
3041          */
3042
3043         if (!bond->current_arp_slave) {
3044                 bond->current_arp_slave = bond->first_slave;
3045                 if (!bond->current_arp_slave)
3046                         return;
3047         }
3048
3049         bond_set_slave_inactive_flags(bond->current_arp_slave);
3050
3051         /* search for next candidate */
3052         bond_for_each_slave_from(bond, slave, i, bond->current_arp_slave->next) {
3053                 if (IS_UP(slave->dev)) {
3054                         slave->link = BOND_LINK_BACK;
3055                         bond_set_slave_active_flags(slave);
3056                         bond_arp_send_all(bond, slave);
3057                         slave->jiffies = jiffies;
3058                         bond->current_arp_slave = slave;
3059                         break;
3060                 }
3061
3062                 /* if the link state is up at this point, we
3063                  * mark it down - this can happen if we have
3064                  * simultaneous link failures and
3065                  * reselect_active_interface doesn't make this
3066                  * one the current slave so it is still marked
3067                  * up when it is actually down
3068                  */
3069                 if (slave->link == BOND_LINK_UP) {
3070                         slave->link = BOND_LINK_DOWN;
3071                         if (slave->link_failure_count < UINT_MAX)
3072                                 slave->link_failure_count++;
3073
3074                         bond_set_slave_inactive_flags(slave);
3075
3076                         pr_info("%s: backup interface %s is now down.\n",
3077                                 bond->dev->name, slave->dev->name);
3078                 }
3079         }
3080 }
3081
3082 void bond_activebackup_arp_mon(struct work_struct *work)
3083 {
3084         struct bonding *bond = container_of(work, struct bonding,
3085                                             arp_work.work);
3086         bool should_notify_peers = false;
3087         int delta_in_ticks;
3088
3089         read_lock(&bond->lock);
3090