projects / GitHub / mt8127 / android_kernel_alcatel_ttab.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
Merge branches 'devel-stable', 'entry', 'fixes', 'mach-types', 'misc' and 'smp-hotplu...
[GitHub/mt8127/android_kernel_alcatel_ttab.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 netif_addr_lock_bh(bond->dev);
850 netdev_for_each_mc_addr(ha, bond->dev)
851 dev_mc_del(old_active->dev, ha->addr);
852 netif_addr_unlock_bh(bond->dev);
853 }
854
855 if (new_active) {
856 /* FIXME: Signal errors upstream. */
857 if (bond->dev->flags & IFF_PROMISC)
858 dev_set_promiscuity(new_active->dev, 1);
859
860 if (bond->dev->flags & IFF_ALLMULTI)
861 dev_set_allmulti(new_active->dev, 1);
862
863 netif_addr_lock_bh(bond->dev);
864 netdev_for_each_mc_addr(ha, bond->dev)
865 dev_mc_add(new_active->dev, ha->addr);
866 netif_addr_unlock_bh(bond->dev);
867 }
868 }
869
870 /*
871 * bond_do_fail_over_mac
872 *
873 * Perform special MAC address swapping for fail_over_mac settings
874 *
875 * Called with RTNL, bond->lock for read, curr_slave_lock for write_bh.
876 */
877 static void bond_do_fail_over_mac(struct bonding *bond,
878 struct slave *new_active,
879 struct slave *old_active)
880 __releases(&bond->curr_slave_lock)
881 __releases(&bond->lock)
882 __acquires(&bond->lock)
883 __acquires(&bond->curr_slave_lock)
884 {
885 u8 tmp_mac[ETH_ALEN];
886 struct sockaddr saddr;
887 int rv;
888
889 switch (bond->params.fail_over_mac) {
890 case BOND_FOM_ACTIVE:
891 if (new_active) {
892 memcpy(bond->dev->dev_addr, new_active->dev->dev_addr,
893 new_active->dev->addr_len);
894 write_unlock_bh(&bond->curr_slave_lock);
895 read_unlock(&bond->lock);
896 call_netdevice_notifiers(NETDEV_CHANGEADDR, bond->dev);
897 read_lock(&bond->lock);
898 write_lock_bh(&bond->curr_slave_lock);
899 }
900 break;
901 case BOND_FOM_FOLLOW:
902 /*
903 * if new_active && old_active, swap them
904 * if just old_active, do nothing (going to no active slave)
905 * if just new_active, set new_active to bond's MAC
906 */
907 if (!new_active)
908 return;
909
910 write_unlock_bh(&bond->curr_slave_lock);
911 read_unlock(&bond->lock);
912
913 if (old_active) {
914 memcpy(tmp_mac, new_active->dev->dev_addr, ETH_ALEN);
915 memcpy(saddr.sa_data, old_active->dev->dev_addr,
916 ETH_ALEN);
917 saddr.sa_family = new_active->dev->type;
918 } else {
919 memcpy(saddr.sa_data, bond->dev->dev_addr, ETH_ALEN);
920 saddr.sa_family = bond->dev->type;
921 }
922
923 rv = dev_set_mac_address(new_active->dev, &saddr);
924 if (rv) {
925 pr_err("%s: Error %d setting MAC of slave %s\n",
926 bond->dev->name, -rv, new_active->dev->name);
927 goto out;
928 }
929
930 if (!old_active)
931 goto out;
932
933 memcpy(saddr.sa_data, tmp_mac, ETH_ALEN);
934 saddr.sa_family = old_active->dev->type;
935
936 rv = dev_set_mac_address(old_active->dev, &saddr);
937 if (rv)
938 pr_err("%s: Error %d setting MAC of slave %s\n",
939 bond->dev->name, -rv, new_active->dev->name);
940 out:
941 read_lock(&bond->lock);
942 write_lock_bh(&bond->curr_slave_lock);
943 break;
944 default:
945 pr_err("%s: bond_do_fail_over_mac impossible: bad policy %d\n",
946 bond->dev->name, bond->params.fail_over_mac);
947 break;
948 }
949
950 }
951
952 static bool bond_should_change_active(struct bonding *bond)
953 {
954 struct slave *prim = bond->primary_slave;
955 struct slave *curr = bond->curr_active_slave;
956
957 if (!prim || !curr || curr->link != BOND_LINK_UP)
958 return true;
959 if (bond->force_primary) {
960 bond->force_primary = false;
961 return true;
962 }
963 if (bond->params.primary_reselect == BOND_PRI_RESELECT_BETTER &&
964 (prim->speed < curr->speed ||
965 (prim->speed == curr->speed && prim->duplex <= curr->duplex)))
966 return false;
967 if (bond->params.primary_reselect == BOND_PRI_RESELECT_FAILURE)
968 return false;
969 return true;
970 }
971
972 /**
973 * find_best_interface - select the best available slave to be the active one
974 * @bond: our bonding struct
975 *
976 * Warning: Caller must hold curr_slave_lock for writing.
977 */
978 static struct slave *bond_find_best_slave(struct bonding *bond)
979 {
980 struct slave *new_active, *old_active;
981 struct slave *bestslave = NULL;
982 int mintime = bond->params.updelay;
983 int i;
984
985 new_active = bond->curr_active_slave;
986
987 if (!new_active) { /* there were no active slaves left */
988 if (bond->slave_cnt > 0) /* found one slave */
989 new_active = bond->first_slave;
990 else
991 return NULL; /* still no slave, return NULL */
992 }
993
994 if ((bond->primary_slave) &&
995 bond->primary_slave->link == BOND_LINK_UP &&
996 bond_should_change_active(bond)) {
997 new_active = bond->primary_slave;
998 }
999
1000 /* remember where to stop iterating over the slaves */
1001 old_active = new_active;
1002
1003 bond_for_each_slave_from(bond, new_active, i, old_active) {
1004 if (new_active->link == BOND_LINK_UP) {
1005 return new_active;
1006 } else if (new_active->link == BOND_LINK_BACK &&
1007 IS_UP(new_active->dev)) {
1008 /* link up, but waiting for stabilization */
1009 if (new_active->delay < mintime) {
1010 mintime = new_active->delay;
1011 bestslave = new_active;
1012 }
1013 }
1014 }
1015
1016 return bestslave;
1017 }
1018
1019 static bool bond_should_notify_peers(struct bonding *bond)
1020 {
1021 struct slave *slave = bond->curr_active_slave;
1022
1023 pr_debug("bond_should_notify_peers: bond %s slave %s\n",
1024 bond->dev->name, slave ? slave->dev->name : "NULL");
1025
1026 if (!slave || !bond->send_peer_notif ||
1027 test_bit(__LINK_STATE_LINKWATCH_PENDING, &slave->dev->state))
1028 return false;
1029
1030 bond->send_peer_notif--;
1031 return true;
1032 }
1033
1034 /**
1035 * change_active_interface - change the active slave into the specified one
1036 * @bond: our bonding struct
1037 * @new: the new slave to make the active one
1038 *
1039 * Set the new slave to the bond's settings and unset them on the old
1040 * curr_active_slave.
1041 * Setting include flags, mc-list, promiscuity, allmulti, etc.
1042 *
1043 * If @new's link state is %BOND_LINK_BACK we'll set it to %BOND_LINK_UP,
1044 * because it is apparently the best available slave we have, even though its
1045 * updelay hasn't timed out yet.
1046 *
1047 * If new_active is not NULL, caller must hold bond->lock for read and
1048 * curr_slave_lock for write_bh.
1049 */
1050 void bond_change_active_slave(struct bonding *bond, struct slave *new_active)
1051 {
1052 struct slave *old_active = bond->curr_active_slave;
1053
1054 if (old_active == new_active)
1055 return;
1056
1057 if (new_active) {
1058 new_active->jiffies = jiffies;
1059
1060 if (new_active->link == BOND_LINK_BACK) {
1061 if (USES_PRIMARY(bond->params.mode)) {
1062 pr_info("%s: making interface %s the new active one %d ms earlier.\n",
1063 bond->dev->name, new_active->dev->name,
1064 (bond->params.updelay - new_active->delay) * bond->params.miimon);
1065 }
1066
1067 new_active->delay = 0;
1068 new_active->link = BOND_LINK_UP;
1069
1070 if (bond->params.mode == BOND_MODE_8023AD)
1071 bond_3ad_handle_link_change(new_active, BOND_LINK_UP);
1072
1073 if (bond_is_lb(bond))
1074 bond_alb_handle_link_change(bond, new_active, BOND_LINK_UP);
1075 } else {
1076 if (USES_PRIMARY(bond->params.mode)) {
1077 pr_info("%s: making interface %s the new active one.\n",
1078 bond->dev->name, new_active->dev->name);
1079 }
1080 }
1081 }
1082
1083 if (USES_PRIMARY(bond->params.mode))
1084 bond_mc_swap(bond, new_active, old_active);
1085
1086 if (bond_is_lb(bond)) {
1087 bond_alb_handle_active_change(bond, new_active);
1088 if (old_active)
1089 bond_set_slave_inactive_flags(old_active);
1090 if (new_active)
1091 bond_set_slave_active_flags(new_active);
1092 } else {
1093 bond->curr_active_slave = new_active;
1094 }
1095
1096 if (bond->params.mode == BOND_MODE_ACTIVEBACKUP) {
1097 if (old_active)
1098 bond_set_slave_inactive_flags(old_active);
1099
1100 if (new_active) {
1101 bool should_notify_peers = false;
1102
1103 bond_set_slave_active_flags(new_active);
1104
1105 if (bond->params.fail_over_mac)
1106 bond_do_fail_over_mac(bond, new_active,
1107 old_active);
1108
1109 if (netif_running(bond->dev)) {
1110 bond->send_peer_notif =
1111 bond->params.num_peer_notif;
1112 should_notify_peers =
1113 bond_should_notify_peers(bond);
1114 }
1115
1116 write_unlock_bh(&bond->curr_slave_lock);
1117 read_unlock(&bond->lock);
1118
1119 call_netdevice_notifiers(NETDEV_BONDING_FAILOVER, bond->dev);
1120 if (should_notify_peers)
1121 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS,
1122 bond->dev);
1123
1124 read_lock(&bond->lock);
1125 write_lock_bh(&bond->curr_slave_lock);
1126 }
1127 }
1128
1129 /* resend IGMP joins since active slave has changed or
1130 * all were sent on curr_active_slave.
1131 * resend only if bond is brought up with the affected
1132 * bonding modes and the retransmission is enabled */
1133 if (netif_running(bond->dev) && (bond->params.resend_igmp > 0) &&
1134 ((USES_PRIMARY(bond->params.mode) && new_active) ||
1135 bond->params.mode == BOND_MODE_ROUNDROBIN)) {
1136 bond->igmp_retrans = bond->params.resend_igmp;
1137 queue_delayed_work(bond->wq, &bond->mcast_work, 0);
1138 }
1139 }
1140
1141 /**
1142 * bond_select_active_slave - select a new active slave, if needed
1143 * @bond: our bonding struct
1144 *
1145 * This functions should be called when one of the following occurs:
1146 * - The old curr_active_slave has been released or lost its link.
1147 * - The primary_slave has got its link back.
1148 * - A slave has got its link back and there's no old curr_active_slave.
1149 *
1150 * Caller must hold bond->lock for read and curr_slave_lock for write_bh.
1151 */
1152 void bond_select_active_slave(struct bonding *bond)
1153 {
1154 struct slave *best_slave;
1155 int rv;
1156
1157 best_slave = bond_find_best_slave(bond);
1158 if (best_slave != bond->curr_active_slave) {
1159 bond_change_active_slave(bond, best_slave);
1160 rv = bond_set_carrier(bond);
1161 if (!rv)
1162 return;
1163
1164 if (netif_carrier_ok(bond->dev)) {
1165 pr_info("%s: first active interface up!\n",
1166 bond->dev->name);
1167 } else {
1168 pr_info("%s: now running without any active interface !\n",
1169 bond->dev->name);
1170 }
1171 }
1172 }
1173
1174 /*--------------------------- slave list handling ---------------------------*/
1175
1176 /*
1177 * This function attaches the slave to the end of list.
1178 *
1179 * bond->lock held for writing by caller.
1180 */
1181 static void bond_attach_slave(struct bonding *bond, struct slave *new_slave)
1182 {
1183 if (bond->first_slave == NULL) { /* attaching the first slave */
1184 new_slave->next = new_slave;
1185 new_slave->prev = new_slave;
1186 bond->first_slave = new_slave;
1187 } else {
1188 new_slave->next = bond->first_slave;
1189 new_slave->prev = bond->first_slave->prev;
1190 new_slave->next->prev = new_slave;
1191 new_slave->prev->next = new_slave;
1192 }
1193
1194 bond->slave_cnt++;
1195 }
1196
1197 /*
1198 * This function detaches the slave from the list.
1199 * WARNING: no check is made to verify if the slave effectively
1200 * belongs to <bond>.
1201 * Nothing is freed on return, structures are just unchained.
1202 * If any slave pointer in bond was pointing to <slave>,
1203 * it should be changed by the calling function.
1204 *
1205 * bond->lock held for writing by caller.
1206 */
1207 static void bond_detach_slave(struct bonding *bond, struct slave *slave)
1208 {
1209 if (slave->next)
1210 slave->next->prev = slave->prev;
1211
1212 if (slave->prev)
1213 slave->prev->next = slave->next;
1214
1215 if (bond->first_slave == slave) { /* slave is the first slave */
1216 if (bond->slave_cnt > 1) { /* there are more slave */
1217 bond->first_slave = slave->next;
1218 } else {
1219 bond->first_slave = NULL; /* slave was the last one */
1220 }
1221 }
1222
1223 slave->next = NULL;
1224 slave->prev = NULL;
1225 bond->slave_cnt--;
1226 }
1227
1228 #ifdef CONFIG_NET_POLL_CONTROLLER
1229 static inline int slave_enable_netpoll(struct slave *slave)
1230 {
1231 struct netpoll *np;
1232 int err = 0;
1233
1234 np = kzalloc(sizeof(*np), GFP_ATOMIC);
1235 err = -ENOMEM;
1236 if (!np)
1237 goto out;
1238
1239 err = __netpoll_setup(np, slave->dev, GFP_ATOMIC);
1240 if (err) {
1241 kfree(np);
1242 goto out;
1243 }
1244 slave->np = np;
1245 out:
1246 return err;
1247 }
1248 static inline void slave_disable_netpoll(struct slave *slave)
1249 {
1250 struct netpoll *np = slave->np;
1251
1252 if (!np)
1253 return;
1254
1255 slave->np = NULL;
1256 __netpoll_free_async(np);
1257 }
1258 static inline bool slave_dev_support_netpoll(struct net_device *slave_dev)
1259 {
1260 if (slave_dev->priv_flags & IFF_DISABLE_NETPOLL)
1261 return false;
1262 if (!slave_dev->netdev_ops->ndo_poll_controller)
1263 return false;
1264 return true;
1265 }
1266
1267 static void bond_poll_controller(struct net_device *bond_dev)
1268 {
1269 }
1270
1271 static void __bond_netpoll_cleanup(struct bonding *bond)
1272 {
1273 struct slave *slave;
1274 int i;
1275
1276 bond_for_each_slave(bond, slave, i)
1277 if (IS_UP(slave->dev))
1278 slave_disable_netpoll(slave);
1279 }
1280 static void bond_netpoll_cleanup(struct net_device *bond_dev)
1281 {
1282 struct bonding *bond = netdev_priv(bond_dev);
1283
1284 read_lock(&bond->lock);
1285 __bond_netpoll_cleanup(bond);
1286 read_unlock(&bond->lock);
1287 }
1288
1289 static int bond_netpoll_setup(struct net_device *dev, struct netpoll_info *ni, gfp_t gfp)
1290 {
1291 struct bonding *bond = netdev_priv(dev);
1292 struct slave *slave;
1293 int i, err = 0;
1294
1295 read_lock(&bond->lock);
1296 bond_for_each_slave(bond, slave, i) {
1297 err = slave_enable_netpoll(slave);
1298 if (err) {
1299 __bond_netpoll_cleanup(bond);
1300 break;
1301 }
1302 }
1303 read_unlock(&bond->lock);
1304 return err;
1305 }
1306
1307 static struct netpoll_info *bond_netpoll_info(struct bonding *bond)
1308 {
1309 return bond->dev->npinfo;
1310 }
1311
1312 #else
1313 static inline int slave_enable_netpoll(struct slave *slave)
1314 {
1315 return 0;
1316 }
1317 static inline void slave_disable_netpoll(struct slave *slave)
1318 {
1319 }
1320 static void bond_netpoll_cleanup(struct net_device *bond_dev)
1321 {
1322 }
1323 #endif
1324
1325 /*---------------------------------- IOCTL ----------------------------------*/
1326
1327 static void bond_set_dev_addr(struct net_device *bond_dev,
1328 struct net_device *slave_dev)
1329 {
1330 pr_debug("bond_dev=%p\n", bond_dev);
1331 pr_debug("slave_dev=%p\n", slave_dev);
1332 pr_debug("slave_dev->addr_len=%d\n", slave_dev->addr_len);
1333 memcpy(bond_dev->dev_addr, slave_dev->dev_addr, slave_dev->addr_len);
1334 bond_dev->addr_assign_type = NET_ADDR_SET;
1335 call_netdevice_notifiers(NETDEV_CHANGEADDR, bond_dev);
1336 }
1337
1338 static netdev_features_t bond_fix_features(struct net_device *dev,
1339 netdev_features_t features)
1340 {
1341 struct slave *slave;
1342 struct bonding *bond = netdev_priv(dev);
1343 netdev_features_t mask;
1344 int i;
1345
1346 read_lock(&bond->lock);
1347
1348 if (!bond->first_slave) {
1349 /* Disable adding VLANs to empty bond. But why? --mq */
1350 features |= NETIF_F_VLAN_CHALLENGED;
1351 goto out;
1352 }
1353
1354 mask = features;
1355 features &= ~NETIF_F_ONE_FOR_ALL;
1356 features |= NETIF_F_ALL_FOR_ALL;
1357
1358 bond_for_each_slave(bond, slave, i) {
1359 features = netdev_increment_features(features,
1360 slave->dev->features,
1361 mask);
1362 }
1363
1364 out:
1365 read_unlock(&bond->lock);
1366 return features;
1367 }
1368
1369 #define BOND_VLAN_FEATURES (NETIF_F_ALL_CSUM | NETIF_F_SG | \
1370 NETIF_F_FRAGLIST | NETIF_F_ALL_TSO | \
1371 NETIF_F_HIGHDMA | NETIF_F_LRO)
1372
1373 static void bond_compute_features(struct bonding *bond)
1374 {
1375 struct slave *slave;
1376 struct net_device *bond_dev = bond->dev;
1377 netdev_features_t vlan_features = BOND_VLAN_FEATURES;
1378 unsigned short max_hard_header_len = ETH_HLEN;
1379 unsigned int gso_max_size = GSO_MAX_SIZE;
1380 u16 gso_max_segs = GSO_MAX_SEGS;
1381 int i;
1382 unsigned int flags, dst_release_flag = IFF_XMIT_DST_RELEASE;
1383
1384 read_lock(&bond->lock);
1385
1386 if (!bond->first_slave)
1387 goto done;
1388
1389 bond_for_each_slave(bond, slave, i) {
1390 vlan_features = netdev_increment_features(vlan_features,
1391 slave->dev->vlan_features, BOND_VLAN_FEATURES);
1392
1393 dst_release_flag &= slave->dev->priv_flags;
1394 if (slave->dev->hard_header_len > max_hard_header_len)
1395 max_hard_header_len = slave->dev->hard_header_len;
1396
1397 gso_max_size = min(gso_max_size, slave->dev->gso_max_size);
1398 gso_max_segs = min(gso_max_segs, slave->dev->gso_max_segs);
1399 }
1400
1401 done:
1402 bond_dev->vlan_features = vlan_features;
1403 bond_dev->hard_header_len = max_hard_header_len;
1404 bond_dev->gso_max_segs = gso_max_segs;
1405 netif_set_gso_max_size(bond_dev, gso_max_size);
1406
1407 flags = bond_dev->priv_flags & ~IFF_XMIT_DST_RELEASE;
1408 bond_dev->priv_flags = flags | dst_release_flag;
1409
1410 read_unlock(&bond->lock);
1411
1412 netdev_change_features(bond_dev);
1413 }
1414
1415 static void bond_setup_by_slave(struct net_device *bond_dev,
1416 struct net_device *slave_dev)
1417 {
1418 struct bonding *bond = netdev_priv(bond_dev);
1419
1420 bond_dev->header_ops = slave_dev->header_ops;
1421
1422 bond_dev->type = slave_dev->type;
1423 bond_dev->hard_header_len = slave_dev->hard_header_len;
1424 bond_dev->addr_len = slave_dev->addr_len;
1425
1426 memcpy(bond_dev->broadcast, slave_dev->broadcast,
1427 slave_dev->addr_len);
1428 bond->setup_by_slave = 1;
1429 }
1430
1431 /* On bonding slaves other than the currently active slave, suppress
1432 * duplicates except for alb non-mcast/bcast.
1433 */
1434 static bool bond_should_deliver_exact_match(struct sk_buff *skb,
1435 struct slave *slave,
1436 struct bonding *bond)
1437 {
1438 if (bond_is_slave_inactive(slave)) {
1439 if (bond->params.mode == BOND_MODE_ALB &&
1440 skb->pkt_type != PACKET_BROADCAST &&
1441 skb->pkt_type != PACKET_MULTICAST)
1442 return false;
1443 return true;
1444 }
1445 return false;
1446 }
1447
1448 static rx_handler_result_t bond_handle_frame(struct sk_buff **pskb)
1449 {
1450 struct sk_buff *skb = *pskb;
1451 struct slave *slave;
1452 struct bonding *bond;
1453 int (*recv_probe)(const struct sk_buff *, struct bonding *,
1454 struct slave *);
1455 int ret = RX_HANDLER_ANOTHER;
1456
1457 skb = skb_share_check(skb, GFP_ATOMIC);
1458 if (unlikely(!skb))
1459 return RX_HANDLER_CONSUMED;
1460
1461 *pskb = skb;
1462
1463 slave = bond_slave_get_rcu(skb->dev);
1464 bond = slave->bond;
1465
1466 if (bond->params.arp_interval)
1467 slave->dev->last_rx = jiffies;
1468
1469 recv_probe = ACCESS_ONCE(bond->recv_probe);
1470 if (recv_probe) {
1471 ret = recv_probe(skb, bond, slave);
1472 if (ret == RX_HANDLER_CONSUMED) {
1473 consume_skb(skb);
1474 return ret;
1475 }
1476 }
1477
1478 if (bond_should_deliver_exact_match(skb, slave, bond)) {
1479 return RX_HANDLER_EXACT;
1480 }
1481
1482 skb->dev = bond->dev;
1483
1484 if (bond->params.mode == BOND_MODE_ALB &&
1485 bond->dev->priv_flags & IFF_BRIDGE_PORT &&
1486 skb->pkt_type == PACKET_HOST) {
1487
1488 if (unlikely(skb_cow_head(skb,
1489 skb->data - skb_mac_header(skb)))) {
1490 kfree_skb(skb);
1491 return RX_HANDLER_CONSUMED;
1492 }
1493 memcpy(eth_hdr(skb)->h_dest, bond->dev->dev_addr, ETH_ALEN);
1494 }
1495
1496 return ret;
1497 }
1498
1499 static int bond_master_upper_dev_link(struct net_device *bond_dev,
1500 struct net_device *slave_dev)
1501 {
1502 int err;
1503
1504 err = netdev_master_upper_dev_link(slave_dev, bond_dev);
1505 if (err)
1506 return err;
1507 slave_dev->flags |= IFF_SLAVE;
1508 rtmsg_ifinfo(RTM_NEWLINK, slave_dev, IFF_SLAVE);
1509 return 0;
1510 }
1511
1512 static void bond_upper_dev_unlink(struct net_device *bond_dev,
1513 struct net_device *slave_dev)
1514 {
1515 netdev_upper_dev_unlink(slave_dev, bond_dev);
1516 slave_dev->flags &= ~IFF_SLAVE;
1517 rtmsg_ifinfo(RTM_NEWLINK, slave_dev, IFF_SLAVE);
1518 }
1519
1520 /* enslave device <slave> to bond device <master> */
1521 int bond_enslave(struct net_device *bond_dev, struct net_device *slave_dev)
1522 {
1523 struct bonding *bond = netdev_priv(bond_dev);
1524 const struct net_device_ops *slave_ops = slave_dev->netdev_ops;
1525 struct slave *new_slave = NULL;
1526 struct netdev_hw_addr *ha;
1527 struct sockaddr addr;
1528 int link_reporting;
1529 int res = 0;
1530
1531 if (!bond->params.use_carrier &&
1532 slave_dev->ethtool_ops->get_link == NULL &&
1533 slave_ops->ndo_do_ioctl == NULL) {
1534 pr_warning("%s: Warning: no link monitoring support for %s\n",
1535 bond_dev->name, slave_dev->name);
1536 }
1537
1538 /* already enslaved */
1539 if (slave_dev->flags & IFF_SLAVE) {
1540 pr_debug("Error, Device was already enslaved\n");
1541 return -EBUSY;
1542 }
1543
1544 /* vlan challenged mutual exclusion */
1545 /* no need to lock since we're protected by rtnl_lock */
1546 if (slave_dev->features & NETIF_F_VLAN_CHALLENGED) {
1547 pr_debug("%s: NETIF_F_VLAN_CHALLENGED\n", slave_dev->name);
1548 if (vlan_uses_dev(bond_dev)) {
1549 pr_err("%s: Error: cannot enslave VLAN challenged slave %s on VLAN enabled bond %s\n",
1550 bond_dev->name, slave_dev->name, bond_dev->name);
1551 return -EPERM;
1552 } else {
1553 pr_warning("%s: Warning: enslaved VLAN challenged slave %s. Adding VLANs will be blocked as long as %s is part of bond %s\n",
1554 bond_dev->name, slave_dev->name,
1555 slave_dev->name, bond_dev->name);
1556 }
1557 } else {
1558 pr_debug("%s: ! NETIF_F_VLAN_CHALLENGED\n", slave_dev->name);
1559 }
1560
1561 /*
1562 * Old ifenslave binaries are no longer supported. These can
1563 * be identified with moderate accuracy by the state of the slave:
1564 * the current ifenslave will set the interface down prior to
1565 * enslaving it; the old ifenslave will not.
1566 */
1567 if ((slave_dev->flags & IFF_UP)) {
1568 pr_err("%s is up. This may be due to an out of date ifenslave.\n",
1569 slave_dev->name);
1570 res = -EPERM;
1571 goto err_undo_flags;
1572 }
1573
1574 /* set bonding device ether type by slave - bonding netdevices are
1575 * created with ether_setup, so when the slave type is not ARPHRD_ETHER
1576 * there is a need to override some of the type dependent attribs/funcs.
1577 *
1578 * bond ether type mutual exclusion - don't allow slaves of dissimilar
1579 * ether type (eg ARPHRD_ETHER and ARPHRD_INFINIBAND) share the same bond
1580 */
1581 if (bond->slave_cnt == 0) {
1582 if (bond_dev->type != slave_dev->type) {
1583 pr_debug("%s: change device type from %d to %d\n",
1584 bond_dev->name,
1585 bond_dev->type, slave_dev->type);
1586
1587 res = call_netdevice_notifiers(NETDEV_PRE_TYPE_CHANGE,
1588 bond_dev);
1589 res = notifier_to_errno(res);
1590 if (res) {
1591 pr_err("%s: refused to change device type\n",
1592 bond_dev->name);
1593 res = -EBUSY;
1594 goto err_undo_flags;
1595 }
1596
1597 /* Flush unicast and multicast addresses */
1598 dev_uc_flush(bond_dev);
1599 dev_mc_flush(bond_dev);
1600
1601 if (slave_dev->type != ARPHRD_ETHER)
1602 bond_setup_by_slave(bond_dev, slave_dev);
1603 else {
1604 ether_setup(bond_dev);
1605 bond_dev->priv_flags &= ~IFF_TX_SKB_SHARING;
1606 }
1607
1608 call_netdevice_notifiers(NETDEV_POST_TYPE_CHANGE,
1609 bond_dev);
1610 }
1611 } else if (bond_dev->type != slave_dev->type) {
1612 pr_err("%s ether type (%d) is different from other slaves (%d), can not enslave it.\n",
1613 slave_dev->name,
1614 slave_dev->type, bond_dev->type);
1615 res = -EINVAL;
1616 goto err_undo_flags;
1617 }
1618
1619 if (slave_ops->ndo_set_mac_address == NULL) {
1620 if (bond->slave_cnt == 0) {
1621 pr_warning("%s: Warning: The first slave device specified does not support setting the MAC address. Setting fail_over_mac to active.",
1622 bond_dev->name);
1623 bond->params.fail_over_mac = BOND_FOM_ACTIVE;
1624 } else if (bond->params.fail_over_mac != BOND_FOM_ACTIVE) {
1625 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",
1626 bond_dev->name);
1627 res = -EOPNOTSUPP;
1628 goto err_undo_flags;
1629 }
1630 }
1631
1632 call_netdevice_notifiers(NETDEV_JOIN, slave_dev);
1633
1634 /* If this is the first slave, then we need to set the master's hardware
1635 * address to be the same as the slave's. */
1636 if (bond->slave_cnt == 0 && bond->dev_addr_from_first)
1637 bond_set_dev_addr(bond->dev, slave_dev);
1638
1639 new_slave = kzalloc(sizeof(struct slave), GFP_KERNEL);
1640 if (!new_slave) {
1641 res = -ENOMEM;
1642 goto err_undo_flags;
1643 }
1644
1645 /*
1646 * Set the new_slave's queue_id to be zero. Queue ID mapping
1647 * is set via sysfs or module option if desired.
1648 */
1649 new_slave->queue_id = 0;
1650
1651 /* Save slave's original mtu and then set it to match the bond */
1652 new_slave->original_mtu = slave_dev->mtu;
1653 res = dev_set_mtu(slave_dev, bond->dev->mtu);
1654 if (res) {
1655 pr_debug("Error %d calling dev_set_mtu\n", res);
1656 goto err_free;
1657 }
1658
1659 /*
1660 * Save slave's original ("permanent") mac address for modes
1661 * that need it, and for restoring it upon release, and then
1662 * set it to the master's address
1663 */
1664 memcpy(new_slave->perm_hwaddr, slave_dev->dev_addr, ETH_ALEN);
1665
1666 if (!bond->params.fail_over_mac) {
1667 /*
1668 * Set slave to master's mac address. The application already
1669 * set the master's mac address to that of the first slave
1670 */
1671 memcpy(addr.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
1672 addr.sa_family = slave_dev->type;
1673 res = dev_set_mac_address(slave_dev, &addr);
1674 if (res) {
1675 pr_debug("Error %d calling set_mac_address\n", res);
1676 goto err_restore_mtu;
1677 }
1678 }
1679
1680 res = bond_master_upper_dev_link(bond_dev, slave_dev);
1681 if (res) {
1682 pr_debug("Error %d calling bond_master_upper_dev_link\n", res);
1683 goto err_restore_mac;
1684 }
1685
1686 /* open the slave since the application closed it */
1687 res = dev_open(slave_dev);
1688 if (res) {
1689 pr_debug("Opening slave %s failed\n", slave_dev->name);
1690 goto err_unset_master;
1691 }
1692
1693 new_slave->bond = bond;
1694 new_slave->dev = slave_dev;
1695 slave_dev->priv_flags |= IFF_BONDING;
1696
1697 if (bond_is_lb(bond)) {
1698 /* bond_alb_init_slave() must be called before all other stages since
1699 * it might fail and we do not want to have to undo everything
1700 */
1701 res = bond_alb_init_slave(bond, new_slave);
1702 if (res)
1703 goto err_close;
1704 }
1705
1706 /* If the mode USES_PRIMARY, then the new slave gets the
1707 * master's promisc (and mc) settings only if it becomes the
1708 * curr_active_slave, and that is taken care of later when calling
1709 * bond_change_active()
1710 */
1711 if (!USES_PRIMARY(bond->params.mode)) {
1712 /* set promiscuity level to new slave */
1713 if (bond_dev->flags & IFF_PROMISC) {
1714 res = dev_set_promiscuity(slave_dev, 1);
1715 if (res)
1716 goto err_close;
1717 }
1718
1719 /* set allmulti level to new slave */
1720 if (bond_dev->flags & IFF_ALLMULTI) {
1721 res = dev_set_allmulti(slave_dev, 1);
1722 if (res)
1723 goto err_close;
1724 }
1725
1726 netif_addr_lock_bh(bond_dev);
1727 /* upload master's mc_list to new slave */
1728 netdev_for_each_mc_addr(ha, bond_dev)
1729 dev_mc_add(slave_dev, ha->addr);
1730 netif_addr_unlock_bh(bond_dev);
1731 }
1732
1733 if (bond->params.mode == BOND_MODE_8023AD) {
1734 /* add lacpdu mc addr to mc list */
1735 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
1736
1737 dev_mc_add(slave_dev, lacpdu_multicast);
1738 }
1739
1740 bond_add_vlans_on_slave(bond, slave_dev);
1741
1742 write_lock_bh(&bond->lock);
1743
1744 bond_attach_slave(bond, new_slave);
1745
1746 new_slave->delay = 0;
1747 new_slave->link_failure_count = 0;
1748
1749 write_unlock_bh(&bond->lock);
1750
1751 bond_compute_features(bond);
1752
1753 bond_update_speed_duplex(new_slave);
1754
1755 read_lock(&bond->lock);
1756
1757 new_slave->last_arp_rx = jiffies -
1758 (msecs_to_jiffies(bond->params.arp_interval) + 1);
1759
1760 if (bond->params.miimon && !bond->params.use_carrier) {
1761 link_reporting = bond_check_dev_link(bond, slave_dev, 1);
1762
1763 if ((link_reporting == -1) && !bond->params.arp_interval) {
1764 /*
1765 * miimon is set but a bonded network driver
1766 * does not support ETHTOOL/MII and
1767 * arp_interval is not set. Note: if
1768 * use_carrier is enabled, we will never go
1769 * here (because netif_carrier is always
1770 * supported); thus, we don't need to change
1771 * the messages for netif_carrier.
1772 */
1773 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",
1774 bond_dev->name, slave_dev->name);
1775 } else if (link_reporting == -1) {
1776 /* unable get link status using mii/ethtool */
1777 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",
1778 bond_dev->name, slave_dev->name);
1779 }
1780 }
1781
1782 /* check for initial state */
1783 if (bond->params.miimon) {
1784 if (bond_check_dev_link(bond, slave_dev, 0) == BMSR_LSTATUS) {
1785 if (bond->params.updelay) {
1786 new_slave->link = BOND_LINK_BACK;
1787 new_slave->delay = bond->params.updelay;
1788 } else {
1789 new_slave->link = BOND_LINK_UP;
1790 }
1791 } else {
1792 new_slave->link = BOND_LINK_DOWN;
1793 }
1794 } else if (bond->params.arp_interval) {
1795 new_slave->link = (netif_carrier_ok(slave_dev) ?
1796 BOND_LINK_UP : BOND_LINK_DOWN);
1797 } else {
1798 new_slave->link = BOND_LINK_UP;
1799 }
1800
1801 if (new_slave->link != BOND_LINK_DOWN)
1802 new_slave->jiffies = jiffies;
1803 pr_debug("Initial state of slave_dev is BOND_LINK_%s\n",
1804 new_slave->link == BOND_LINK_DOWN ? "DOWN" :
1805 (new_slave->link == BOND_LINK_UP ? "UP" : "BACK"));
1806
1807 if (USES_PRIMARY(bond->params.mode) && bond->params.primary[0]) {
1808 /* if there is a primary slave, remember it */
1809 if (strcmp(bond->params.primary, new_slave->dev->name) == 0) {
1810 bond->primary_slave = new_slave;
1811 bond->force_primary = true;
1812 }
1813 }
1814
1815 write_lock_bh(&bond->curr_slave_lock);
1816
1817 switch (bond->params.mode) {
1818 case BOND_MODE_ACTIVEBACKUP:
1819 bond_set_slave_inactive_flags(new_slave);
1820 bond_select_active_slave(bond);
1821 break;
1822 case BOND_MODE_8023AD:
1823 /* in 802.3ad mode, the internal mechanism
1824 * will activate the slaves in the selected
1825 * aggregator
1826 */
1827 bond_set_slave_inactive_flags(new_slave);
1828 /* if this is the first slave */
1829 if (bond->slave_cnt == 1) {
1830 SLAVE_AD_INFO(new_slave).id = 1;
1831 /* Initialize AD with the number of times that the AD timer is called in 1 second
1832 * can be called only after the mac address of the bond is set
1833 */
1834 bond_3ad_initialize(bond, 1000/AD_TIMER_INTERVAL);
1835 } else {
1836 SLAVE_AD_INFO(new_slave).id =
1837 SLAVE_AD_INFO(new_slave->prev).id + 1;
1838 }
1839
1840 bond_3ad_bind_slave(new_slave);
1841 break;
1842 case BOND_MODE_TLB:
1843 case BOND_MODE_ALB:
1844 bond_set_active_slave(new_slave);
1845 bond_set_slave_inactive_flags(new_slave);
1846 bond_select_active_slave(bond);
1847 break;
1848 default:
1849 pr_debug("This slave is always active in trunk mode\n");
1850
1851 /* always active in trunk mode */
1852 bond_set_active_slave(new_slave);
1853
1854 /* In trunking mode there is little meaning to curr_active_slave
1855 * anyway (it holds no special properties of the bond device),
1856 * so we can change it without calling change_active_interface()
1857 */
1858 if (!bond->curr_active_slave && new_slave->link == BOND_LINK_UP)
1859 bond->curr_active_slave = new_slave;
1860
1861 break;
1862 } /* switch(bond_mode) */
1863
1864 write_unlock_bh(&bond->curr_slave_lock);
1865
1866 bond_set_carrier(bond);
1867
1868 #ifdef CONFIG_NET_POLL_CONTROLLER
1869 slave_dev->npinfo = bond_netpoll_info(bond);
1870 if (slave_dev->npinfo) {
1871 if (slave_enable_netpoll(new_slave)) {
1872 read_unlock(&bond->lock);
1873 pr_info("Error, %s: master_dev is using netpoll, "
1874 "but new slave device does not support netpoll.\n",
1875 bond_dev->name);
1876 res = -EBUSY;
1877 goto err_detach;
1878 }
1879 }
1880 #endif
1881
1882 read_unlock(&bond->lock);
1883
1884 res = bond_create_slave_symlinks(bond_dev, slave_dev);
1885 if (res)
1886 goto err_detach;
1887
1888 res = netdev_rx_handler_register(slave_dev, bond_handle_frame,
1889 new_slave);
1890 if (res) {
1891 pr_debug("Error %d calling netdev_rx_handler_register\n", res);
1892 goto err_dest_symlinks;
1893 }
1894
1895 pr_info("%s: enslaving %s as a%s interface with a%s link.\n",
1896 bond_dev->name, slave_dev->name,
1897 bond_is_active_slave(new_slave) ? "n active" : " backup",
1898 new_slave->link != BOND_LINK_DOWN ? "n up" : " down");
1899
1900 /* enslave is successful */
1901 return 0;
1902
1903 /* Undo stages on error */
1904 err_dest_symlinks:
1905 bond_destroy_slave_symlinks(bond_dev, slave_dev);
1906
1907 err_detach:
1908 if (!USES_PRIMARY(bond->params.mode)) {
1909 netif_addr_lock_bh(bond_dev);
1910 bond_mc_list_flush(bond_dev, slave_dev);
1911 netif_addr_unlock_bh(bond_dev);
1912 }
1913 bond_del_vlans_from_slave(bond, slave_dev);
1914 write_lock_bh(&bond->lock);
1915 bond_detach_slave(bond, new_slave);
1916 if (bond->primary_slave == new_slave)
1917 bond->primary_slave = NULL;
1918 write_unlock_bh(&bond->lock);
1919 if (bond->curr_active_slave == new_slave) {
1920 read_lock(&bond->lock);
1921 write_lock_bh(&bond->curr_slave_lock);
1922 bond_change_active_slave(bond, NULL);
1923 bond_select_active_slave(bond);
1924 write_unlock_bh(&bond->curr_slave_lock);
1925 read_unlock(&bond->lock);
1926 }
1927 slave_disable_netpoll(new_slave);
1928
1929 err_close:
1930 slave_dev->priv_flags &= ~IFF_BONDING;
1931 dev_close(slave_dev);
1932
1933 err_unset_master:
1934 bond_upper_dev_unlink(bond_dev, slave_dev);
1935
1936 err_restore_mac:
1937 if (!bond->params.fail_over_mac) {
1938 /* XXX TODO - fom follow mode needs to change master's
1939 * MAC if this slave's MAC is in use by the bond, or at
1940 * least print a warning.
1941 */
1942 memcpy(addr.sa_data, new_slave->perm_hwaddr, ETH_ALEN);
1943 addr.sa_family = slave_dev->type;
1944 dev_set_mac_address(slave_dev, &addr);
1945 }
1946
1947 err_restore_mtu:
1948 dev_set_mtu(slave_dev, new_slave->original_mtu);
1949
1950 err_free:
1951 kfree(new_slave);
1952
1953 err_undo_flags:
1954 bond_compute_features(bond);
1955
1956 return res;
1957 }
1958
1959 /*
1960 * Try to release the slave device <slave> from the bond device <master>
1961 * It is legal to access curr_active_slave without a lock because all the function
1962 * is write-locked. If "all" is true it means that the function is being called
1963 * while destroying a bond interface and all slaves are being released.
1964 *
1965 * The rules for slave state should be:
1966 * for Active/Backup:
1967 * Active stays on all backups go down
1968 * for Bonded connections:
1969 * The first up interface should be left on and all others downed.
1970 */
1971 static int __bond_release_one(struct net_device *bond_dev,
1972 struct net_device *slave_dev,
1973 bool all)
1974 {
1975 struct bonding *bond = netdev_priv(bond_dev);
1976 struct slave *slave, *oldcurrent;
1977 struct sockaddr addr;
1978 netdev_features_t old_features = bond_dev->features;
1979
1980 /* slave is not a slave or master is not master of this slave */
1981 if (!(slave_dev->flags & IFF_SLAVE) ||
1982 !netdev_has_upper_dev(slave_dev, bond_dev)) {
1983 pr_err("%s: Error: cannot release %s.\n",
1984 bond_dev->name, slave_dev->name);
1985 return -EINVAL;
1986 }
1987
1988 block_netpoll_tx();
1989 write_lock_bh(&bond->lock);
1990
1991 slave = bond_get_slave_by_dev(bond, slave_dev);
1992 if (!slave) {
1993 /* not a slave of this bond */
1994 pr_info("%s: %s not enslaved\n",
1995 bond_dev->name, slave_dev->name);
1996 write_unlock_bh(&bond->lock);
1997 unblock_netpoll_tx();
1998 return -EINVAL;
1999 }
2000
2001 write_unlock_bh(&bond->lock);
2002 /* unregister rx_handler early so bond_handle_frame wouldn't be called
2003 * for this slave anymore.
2004 */
2005 netdev_rx_handler_unregister(slave_dev);
2006 write_lock_bh(&bond->lock);
2007
2008 if (!all && !bond->params.fail_over_mac) {
2009 if (ether_addr_equal(bond_dev->dev_addr, slave->perm_hwaddr) &&
2010 bond->slave_cnt > 1)
2011 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",
2012 bond_dev->name, slave_dev->name,
2013 slave->perm_hwaddr,
2014 bond_dev->name, slave_dev->name);
2015 }
2016
2017 /* Inform AD package of unbinding of slave. */
2018 if (bond->params.mode == BOND_MODE_8023AD) {
2019 /* must be called before the slave is
2020 * detached from the list
2021 */
2022 bond_3ad_unbind_slave(slave);
2023 }
2024
2025 pr_info("%s: releasing %s interface %s\n",
2026 bond_dev->name,
2027 bond_is_active_slave(slave) ? "active" : "backup",
2028 slave_dev->name);
2029
2030 oldcurrent = bond->curr_active_slave;
2031
2032 bond->current_arp_slave = NULL;
2033
2034 /* release the slave from its bond */
2035 bond_detach_slave(bond, slave);
2036
2037 if (bond->primary_slave == slave)
2038 bond->primary_slave = NULL;
2039
2040 if (oldcurrent == slave)
2041 bond_change_active_slave(bond, NULL);
2042
2043 if (bond_is_lb(bond)) {
2044 /* Must be called only after the slave has been
2045 * detached from the list and the curr_active_slave
2046 * has been cleared (if our_slave == old_current),
2047 * but before a new active slave is selected.
2048 */
2049 write_unlock_bh(&bond->lock);
2050 bond_alb_deinit_slave(bond, slave);
2051 write_lock_bh(&bond->lock);
2052 }
2053
2054 if (all) {
2055 bond->curr_active_slave = NULL;
2056 } else if (oldcurrent == slave) {
2057 /*
2058 * Note that we hold RTNL over this sequence, so there
2059 * is no concern that another slave add/remove event
2060 * will interfere.
2061 */
2062 write_unlock_bh(&bond->lock);
2063 read_lock(&bond->lock);
2064 write_lock_bh(&bond->curr_slave_lock);
2065
2066 bond_select_active_slave(bond);
2067
2068 write_unlock_bh(&bond->curr_slave_lock);
2069 read_unlock(&bond->lock);
2070 write_lock_bh(&bond->lock);
2071 }
2072
2073 if (bond->slave_cnt == 0) {
2074 bond_set_carrier(bond);
2075 eth_hw_addr_random(bond_dev);
2076 bond->dev_addr_from_first = true;
2077
2078 if (bond_vlan_used(bond)) {
2079 pr_warning("%s: Warning: clearing HW address of %s while it still has VLANs.\n",
2080 bond_dev->name, bond_dev->name);
2081 pr_warning("%s: When re-adding slaves, make sure the bond's HW address matches its VLANs'.\n",
2082 bond_dev->name);
2083 }
2084 }
2085
2086 write_unlock_bh(&bond->lock);
2087 unblock_netpoll_tx();
2088
2089 if (bond->slave_cnt == 0) {
2090 call_netdevice_notifiers(NETDEV_CHANGEADDR, bond->dev);
2091 call_netdevice_notifiers(NETDEV_RELEASE, bond->dev);
2092 }
2093
2094 bond_compute_features(bond);
2095 if (!(bond_dev->features & NETIF_F_VLAN_CHALLENGED) &&
2096 (old_features & NETIF_F_VLAN_CHALLENGED))
2097 pr_info("%s: last VLAN challenged slave %s left bond %s. VLAN blocking is removed\n",
2098 bond_dev->name, slave_dev->name, bond_dev->name);
2099
2100 /* must do this from outside any spinlocks */
2101 bond_destroy_slave_symlinks(bond_dev, slave_dev);
2102
2103 bond_del_vlans_from_slave(bond, slave_dev);
2104
2105 /* If the mode USES_PRIMARY, then we should only remove its
2106 * promisc and mc settings if it was the curr_active_slave, but that was
2107 * already taken care of above when we detached the slave
2108 */
2109 if (!USES_PRIMARY(bond->params.mode)) {
2110 /* unset promiscuity level from slave */
2111 if (bond_dev->flags & IFF_PROMISC)
2112 dev_set_promiscuity(slave_dev, -1);
2113
2114 /* unset allmulti level from slave */
2115 if (bond_dev->flags & IFF_ALLMULTI)
2116 dev_set_allmulti(slave_dev, -1);
2117
2118 /* flush master's mc_list from slave */
2119 netif_addr_lock_bh(bond_dev);
2120 bond_mc_list_flush(bond_dev, slave_dev);
2121 netif_addr_unlock_bh(bond_dev);
2122 }
2123
2124 bond_upper_dev_unlink(bond_dev, slave_dev);
2125
2126 slave_disable_netpoll(slave);
2127
2128 /* close slave before restoring its mac address */
2129 dev_close(slave_dev);
2130
2131 if (bond->params.fail_over_mac != BOND_FOM_ACTIVE) {
2132 /* restore original ("permanent") mac address */
2133 memcpy(addr.sa_data, slave->perm_hwaddr, ETH_ALEN);
2134 addr.sa_family = slave_dev->type;
2135 dev_set_mac_address(slave_dev, &addr);
2136 }
2137
2138 dev_set_mtu(slave_dev, slave->original_mtu);
2139
2140 slave_dev->priv_flags &= ~IFF_BONDING;
2141
2142 kfree(slave);
2143
2144 return 0; /* deletion OK */
2145 }
2146
2147 /* A wrapper used because of ndo_del_link */
2148 int bond_release(struct net_device *bond_dev, struct net_device *slave_dev)
2149 {
2150 return __bond_release_one(bond_dev, slave_dev, false);
2151 }
2152
2153 /*
2154 * First release a slave and then destroy the bond if no more slaves are left.
2155 * Must be under rtnl_lock when this function is called.
2156 */
2157 static int bond_release_and_destroy(struct net_device *bond_dev,
2158 struct net_device *slave_dev)
2159 {
2160 struct bonding *bond = netdev_priv(bond_dev);
2161 int ret;
2162
2163 ret = bond_release(bond_dev, slave_dev);
2164 if ((ret == 0) && (bond->slave_cnt == 0)) {
2165 bond_dev->priv_flags |= IFF_DISABLE_NETPOLL;
2166 pr_info("%s: destroying bond %s.\n",
2167 bond_dev->name, bond_dev->name);
2168 unregister_netdevice(bond_dev);
2169 }
2170 return ret;
2171 }
2172
2173 /*
2174 * This function changes the active slave to slave <slave_dev>.
2175 * It returns -EINVAL in the following cases.
2176 * - <slave_dev> is not found in the list.
2177 * - There is not active slave now.
2178 * - <slave_dev> is already active.
2179 * - The link state of <slave_dev> is not BOND_LINK_UP.
2180 * - <slave_dev> is not running.
2181 * In these cases, this function does nothing.
2182 * In the other cases, current_slave pointer is changed and 0 is returned.
2183 */
2184 static int bond_ioctl_change_active(struct net_device *bond_dev, struct net_device *slave_dev)
2185 {
2186 struct bonding *bond = netdev_priv(bond_dev);
2187 struct slave *old_active = NULL;
2188 struct slave *new_active = NULL;
2189 int res = 0;
2190
2191 if (!USES_PRIMARY(bond->params.mode))
2192 return -EINVAL;
2193
2194 /* Verify that bond_dev is indeed the master of slave_dev */
2195 if (!(slave_dev->flags & IFF_SLAVE) ||
2196 !netdev_has_upper_dev(slave_dev, bond_dev))
2197 return -EINVAL;
2198
2199 read_lock(&bond->lock);
2200
2201 read_lock(&bond->curr_slave_lock);
2202 old_active = bond->curr_active_slave;
2203 read_unlock(&bond->curr_slave_lock);
2204
2205 new_active = bond_get_slave_by_dev(bond, slave_dev);
2206
2207 /*
2208 * Changing to the current active: do nothing; return success.
2209 */
2210 if (new_active && (new_active == old_active)) {
2211 read_unlock(&bond->lock);
2212 return 0;
2213 }
2214
2215 if ((new_active) &&
2216 (old_active) &&
2217 (new_active->link == BOND_LINK_UP) &&
2218 IS_UP(new_active->dev)) {
2219 block_netpoll_tx();
2220 write_lock_bh(&bond->curr_slave_lock);
2221 bond_change_active_slave(bond, new_active);
2222 write_unlock_bh(&bond->curr_slave_lock);
2223 unblock_netpoll_tx();
2224 } else
2225 res = -EINVAL;
2226
2227 read_unlock(&bond->lock);
2228
2229 return res;
2230 }
2231
2232 static int bond_info_query(struct net_device *bond_dev, struct ifbond *info)
2233 {
2234 struct bonding *bond = netdev_priv(bond_dev);
2235
2236 info->bond_mode = bond->params.mode;
2237 info->miimon = bond->params.miimon;
2238
2239 read_lock(&bond->lock);
2240 info->num_slaves = bond->slave_cnt;
2241 read_unlock(&bond->lock);
2242
2243 return 0;
2244 }
2245
2246 static int bond_slave_info_query(struct net_device *bond_dev, struct ifslave *info)
2247 {
2248 struct bonding *bond = netdev_priv(bond_dev);
2249 struct slave *slave;
2250 int i, res = -ENODEV;
2251
2252 read_lock(&bond->lock);
2253
2254 bond_for_each_slave(bond, slave, i) {
2255 if (i == (int)info->slave_id) {
2256 res = 0;
2257 strcpy(info->slave_name, slave->dev->name);
2258 info->link = slave->link;
2259 info->state = bond_slave_state(slave);
2260 info->link_failure_count = slave->link_failure_count;
2261 break;
2262 }
2263 }
2264
2265 read_unlock(&bond->lock);
2266
2267 return res;
2268 }
2269
2270 /*-------------------------------- Monitoring -------------------------------*/
2271
2272
2273 static int bond_miimon_inspect(struct bonding *bond)
2274 {
2275 struct slave *slave;
2276 int i, link_state, commit = 0;
2277 bool ignore_updelay;
2278
2279 ignore_updelay = !bond->curr_active_slave ? true : false;
2280
2281 bond_for_each_slave(bond, slave, i) {
2282 slave->new_link = BOND_LINK_NOCHANGE;
2283
2284 link_state = bond_check_dev_link(bond, slave->dev, 0);
2285
2286 switch (slave->link) {
2287 case BOND_LINK_UP:
2288 if (link_state)
2289 continue;
2290
2291 slave->link = BOND_LINK_FAIL;
2292 slave->delay = bond->params.downdelay;
2293 if (slave->delay) {
2294 pr_info("%s: link status down for %sinterface %s, disabling it in %d ms.\n",
2295 bond->dev->name,
2296 (bond->params.mode ==
2297 BOND_MODE_ACTIVEBACKUP) ?
2298 (bond_is_active_slave(slave) ?
2299 "active " : "backup ") : "",
2300 slave->dev->name,
2301 bond->params.downdelay * bond->params.miimon);
2302 }
2303 /*FALLTHRU*/
2304 case BOND_LINK_FAIL:
2305 if (link_state) {
2306 /*
2307 * recovered before downdelay expired
2308 */
2309 slave->link = BOND_LINK_UP;
2310 slave->jiffies = jiffies;
2311 pr_info("%s: link status up again after %d ms for interface %s.\n",
2312 bond->dev->name,
2313 (bond->params.downdelay - slave->delay) *
2314 bond->params.miimon,
2315 slave->dev->name);
2316 continue;
2317 }
2318
2319 if (slave->delay <= 0) {
2320 slave->new_link = BOND_LINK_DOWN;
2321 commit++;
2322 continue;
2323 }
2324
2325 slave->delay--;
2326 break;
2327
2328 case BOND_LINK_DOWN:
2329 if (!link_state)
2330 continue;
2331
2332 slave->link = BOND_LINK_BACK;
2333 slave->delay = bond->params.updelay;
2334
2335 if (slave->delay) {
2336 pr_info("%s: link status up for interface %s, enabling it in %d ms.\n",
2337 bond->dev->name, slave->dev->name,
2338 ignore_updelay ? 0 :
2339 bond->params.updelay *
2340 bond->params.miimon);
2341 }
2342 /*FALLTHRU*/
2343 case BOND_LINK_BACK:
2344 if (!link_state) {
2345 slave->link = BOND_LINK_DOWN;
2346 pr_info("%s: link status down again after %d ms for interface %s.\n",
2347 bond->dev->name,
2348 (bond->params.updelay - slave->delay) *
2349 bond->params.miimon,
2350 slave->dev->name);
2351
2352 continue;
2353 }
2354
2355 if (ignore_updelay)
2356 slave->delay = 0;
2357
2358 if (slave->delay <= 0) {
2359 slave->new_link = BOND_LINK_UP;
2360 commit++;
2361 ignore_updelay = false;
2362 continue;
2363 }
2364
2365 slave->delay--;
2366 break;
2367 }
2368 }
2369
2370 return commit;
2371 }
2372
2373 static void bond_miimon_commit(struct bonding *bond)
2374 {
2375 struct slave *slave;
2376 int i;
2377
2378 bond_for_each_slave(bond, slave, i) {
2379 switch (slave->new_link) {
2380 case BOND_LINK_NOCHANGE:
2381 continue;
2382
2383 case BOND_LINK_UP:
2384 slave->link = BOND_LINK_UP;
2385 slave->jiffies = jiffies;
2386
2387 if (bond->params.mode == BOND_MODE_8023AD) {
2388 /* prevent it from being the active one */
2389 bond_set_backup_slave(slave);
2390 } else if (bond->params.mode != BOND_MODE_ACTIVEBACKUP) {
2391 /* make it immediately active */
2392 bond_set_active_slave(slave);
2393 } else if (slave != bond->primary_slave) {
2394 /* prevent it from being the active one */
2395 bond_set_backup_slave(slave);
2396 }
2397
2398 pr_info("%s: link status definitely up for interface %s, %u Mbps %s duplex.\n",
2399 bond->dev->name, slave->dev->name,
2400 slave->speed, slave->duplex ? "full" : "half");
2401
2402 /* notify ad that the link status has changed */
2403 if (bond->params.mode == BOND_MODE_8023AD)
2404 bond_3ad_handle_link_change(slave, BOND_LINK_UP);
2405
2406 if (bond_is_lb(bond))
2407 bond_alb_handle_link_change(bond, slave,
2408 BOND_LINK_UP);
2409
2410 if (!bond->curr_active_slave ||
2411 (slave == bond->primary_slave))
2412 goto do_failover;
2413
2414 continue;
2415
2416 case BOND_LINK_DOWN:
2417 if (slave->link_failure_count < UINT_MAX)
2418 slave->link_failure_count++;
2419
2420 slave->link = BOND_LINK_DOWN;
2421
2422 if (bond->params.mode == BOND_MODE_ACTIVEBACKUP ||
2423 bond->params.mode == BOND_MODE_8023AD)
2424 bond_set_slave_inactive_flags(slave);
2425
2426 pr_info("%s: link status definitely down for interface %s, disabling it\n",
2427 bond->dev->name, slave->dev->name);
2428
2429 if (bond->params.mode == BOND_MODE_8023AD)
2430 bond_3ad_handle_link_change(slave,
2431 BOND_LINK_DOWN);
2432
2433 if (bond_is_lb(bond))
2434 bond_alb_handle_link_change(bond, slave,
2435 BOND_LINK_DOWN);
2436
2437 if (slave == bond->curr_active_slave)
2438 goto do_failover;
2439
2440 continue;
2441
2442 default:
2443 pr_err("%s: invalid new link %d on slave %s\n",
2444 bond->dev->name, slave->new_link,
2445 slave->dev->name);
2446 slave->new_link = BOND_LINK_NOCHANGE;
2447
2448 continue;
2449 }
2450
2451 do_failover:
2452 ASSERT_RTNL();
2453 block_netpoll_tx();
2454 write_lock_bh(&bond->curr_slave_lock);
2455 bond_select_active_slave(bond);
2456 write_unlock_bh(&bond->curr_slave_lock);
2457 unblock_netpoll_tx();
2458 }
2459
2460 bond_set_carrier(bond);
2461 }
2462
2463 /*
2464 * bond_mii_monitor
2465 *
2466 * Really a wrapper that splits the mii monitor into two phases: an
2467 * inspection, then (if inspection indicates something needs to be done)
2468 * an acquisition of appropriate locks followed by a commit phase to
2469 * implement whatever link state changes are indicated.
2470 */
2471 void bond_mii_monitor(struct work_struct *work)
2472 {
2473 struct bonding *bond = container_of(work, struct bonding,
2474 mii_work.work);
2475 bool should_notify_peers = false;
2476 unsigned long delay;
2477
2478 read_lock(&bond->lock);
2479
2480 delay = msecs_to_jiffies(bond->params.miimon);
2481
2482 if (bond->slave_cnt == 0)
2483 goto re_arm;
2484
2485 should_notify_peers = bond_should_notify_peers(bond);
2486
2487 if (bond_miimon_inspect(bond)) {
2488 read_unlock(&bond->lock);
2489
2490 /* Race avoidance with bond_close cancel of workqueue */
2491 if (!rtnl_trylock()) {
2492 read_lock(&bond->lock);
2493 delay = 1;
2494 should_notify_peers = false;
2495 goto re_arm;
2496 }
2497
2498 read_lock(&bond->lock);
2499
2500 bond_miimon_commit(bond);
2501
2502 read_unlock(&bond->lock);
2503 rtnl_unlock(); /* might sleep, hold no other locks */
2504 read_lock(&bond->lock);
2505 }
2506
2507 re_arm:
2508 if (bond->params.miimon)
2509 queue_delayed_work(bond->wq, &bond->mii_work, delay);
2510
2511 read_unlock(&bond->lock);
2512
2513 if (should_notify_peers) {
2514 if (!rtnl_trylock()) {
2515 read_lock(&bond->lock);
2516 bond->send_peer_notif++;
2517 read_unlock(&bond->lock);
2518 return;
2519 }
2520 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, bond->dev);
2521 rtnl_unlock();
2522 }
2523 }
2524
2525 static int bond_has_this_ip(struct bonding *bond, __be32 ip)
2526 {
2527 struct vlan_entry *vlan;
2528 struct net_device *vlan_dev;
2529
2530 if (ip == bond_confirm_addr(bond->dev, 0, ip))
2531 return 1;
2532
2533 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
2534 rcu_read_lock();
2535 vlan_dev = __vlan_find_dev_deep(bond->dev, vlan->vlan_id);
2536 rcu_read_unlock();
2537 if (vlan_dev && ip == bond_confirm_addr(vlan_dev, 0, ip))
2538 return 1;
2539 }
2540
2541 return 0;
2542 }
2543
2544 /*
2545 * We go to the (large) trouble of VLAN tagging ARP frames because
2546 * switches in VLAN mode (especially if ports are configured as
2547 * "native" to a VLAN) might not pass non-tagged frames.
2548 */
2549 static void bond_arp_send(struct net_device *slave_dev, int arp_op, __be32 dest_ip, __be32 src_ip, unsigned short vlan_id)
2550 {
2551 struct sk_buff *skb;
2552
2553 pr_debug("arp %d on slave %s: dst %x src %x vid %d\n", arp_op,
2554 slave_dev->name, dest_ip, src_ip, vlan_id);
2555
2556 skb = arp_create(arp_op, ETH_P_ARP, dest_ip, slave_dev, src_ip,
2557 NULL, slave_dev->dev_addr, NULL);
2558
2559 if (!skb) {
2560 pr_err("ARP packet allocation failed\n");
2561 return;
2562 }
2563 if (vlan_id) {
2564 skb = vlan_put_tag(skb, vlan_id);
2565 if (!skb) {
2566 pr_err("failed to insert VLAN tag\n");
2567 return;
2568 }
2569 }
2570 arp_xmit(skb);
2571 }
2572
2573
2574 static void bond_arp_send_all(struct bonding *bond, struct slave *slave)
2575 {
2576 int i, vlan_id;
2577 __be32 *targets = bond->params.arp_targets;
2578 struct vlan_entry *vlan;
2579 struct net_device *vlan_dev = NULL;
2580 struct rtable *rt;
2581
2582 for (i = 0; (i < BOND_MAX_ARP_TARGETS); i++) {
2583 __be32 addr;
2584 if (!targets[i])
2585 break;
2586 pr_debug("basa: target %x\n", targets[i]);
2587 if (!bond_vlan_used(bond)) {
2588 pr_debug("basa: empty vlan: arp_send\n");
2589 addr = bond_confirm_addr(bond->dev, targets[i], 0);
2590 bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2591 addr, 0);
2592 continue;
2593 }
2594
2595 /*
2596 * If VLANs are configured, we do a route lookup to
2597 * determine which VLAN interface would be used, so we
2598 * can tag the ARP with the proper VLAN tag.
2599 */
2600 rt = ip_route_output(dev_net(bond->dev), targets[i], 0,
2601 RTO_ONLINK, 0);
2602 if (IS_ERR(rt)) {
2603 if (net_ratelimit()) {
2604 pr_warning("%s: no route to arp_ip_target %pI4\n",
2605 bond->dev->name, &targets[i]);
2606 }
2607 continue;
2608 }
2609
2610 /*
2611 * This target is not on a VLAN
2612 */
2613 if (rt->dst.dev == bond->dev) {
2614 ip_rt_put(rt);
2615 pr_debug("basa: rtdev == bond->dev: arp_send\n");
2616 addr = bond_confirm_addr(bond->dev, targets[i], 0);
2617 bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2618 addr, 0);
2619 continue;
2620 }
2621
2622 vlan_id = 0;
2623 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
2624 rcu_read_lock();
2625 vlan_dev = __vlan_find_dev_deep(bond->dev,
2626 vlan->vlan_id);
2627 rcu_read_unlock();
2628 if (vlan_dev == rt->dst.dev) {
2629 vlan_id = vlan->vlan_id;
2630 pr_debug("basa: vlan match on %s %d\n",
2631 vlan_dev->name, vlan_id);
2632 break;
2633 }
2634 }
2635
2636 if (vlan_id && vlan_dev) {
2637 ip_rt_put(rt);
2638 addr = bond_confirm_addr(vlan_dev, targets[i], 0);
2639 bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2640 addr, vlan_id);
2641 continue;
2642 }
2643
2644 if (net_ratelimit()) {
2645 pr_warning("%s: no path to arp_ip_target %pI4 via rt.dev %s\n",
2646 bond->dev->name, &targets[i],
2647 rt->dst.dev ? rt->dst.dev->name : "NULL");
2648 }
2649 ip_rt_put(rt);
2650 }
2651 }
2652
2653 static void bond_validate_arp(struct bonding *bond, struct slave *slave, __be32 sip, __be32 tip)
2654 {
2655 int i;
2656 __be32 *targets = bond->params.arp_targets;
2657
2658 for (i = 0; (i < BOND_MAX_ARP_TARGETS) && targets[i]; i++) {
2659 pr_debug("bva: sip %pI4 tip %pI4 t[%d] %pI4 bhti(tip) %d\n",
2660 &sip, &tip, i, &targets[i],
2661 bond_has_this_ip(bond, tip));
2662 if (sip == targets[i]) {
2663 if (bond_has_this_ip(bond, tip))
2664 slave->last_arp_rx = jiffies;
2665 return;
2666 }
2667 }
2668 }
2669
2670 static int bond_arp_rcv(const struct sk_buff *skb, struct bonding *bond,
2671 struct slave *slave)
2672 {
2673 struct arphdr *arp = (struct arphdr *)skb->data;
2674 unsigned char *arp_ptr;
2675 __be32 sip, tip;
2676 int alen;
2677
2678 if (skb->protocol != __cpu_to_be16(ETH_P_ARP))
2679 return RX_HANDLER_ANOTHER;
2680
2681 read_lock(&bond->lock);
2682 alen = arp_hdr_len(bond->dev);
2683
2684 pr_debug("bond_arp_rcv: bond %s skb->dev %s\n",
2685 bond->dev->name, skb->dev->name);
2686
2687 if (alen > skb_headlen(skb)) {
2688 arp = kmalloc(alen, GFP_ATOMIC);
2689 if (!arp)
2690 goto out_unlock;
2691 if (skb_copy_bits(skb, 0, arp, alen) < 0)
2692 goto out_unlock;
2693 }
2694
2695 if (arp->ar_hln != bond->dev->addr_len ||
2696 skb->pkt_type == PACKET_OTHERHOST ||
2697 skb->pkt_type == PACKET_LOOPBACK ||
2698 arp->ar_hrd != htons(ARPHRD_ETHER) ||
2699 arp->ar_pro != htons(ETH_P_IP) ||
2700 arp->ar_pln != 4)
2701 goto out_unlock;
2702
2703 arp_ptr = (unsigned char *)(arp + 1);
2704 arp_ptr += bond->dev->addr_len;
2705 memcpy(&sip, arp_ptr, 4);
2706 arp_ptr += 4 + bond->dev->addr_len;
2707 memcpy(&tip, arp_ptr, 4);
2708
2709 pr_debug("bond_arp_rcv: %s %s/%d av %d sv %d sip %pI4 tip %pI4\n",
2710 bond->dev->name, slave->dev->name, bond_slave_state(slave),
2711 bond->params.arp_validate, slave_do_arp_validate(bond, slave),
2712 &sip, &tip);
2713
2714 /*
2715 * Backup slaves won't see the ARP reply, but do come through
2716 * here for each ARP probe (so we swap the sip/tip to validate
2717 * the probe). In a "redundant switch, common router" type of
2718 * configuration, the ARP probe will (hopefully) travel from
2719 * the active, through one switch, the router, then the other
2720 * switch before reaching the backup.
2721 */
2722 if (bond_is_active_slave(slave))
2723 bond_validate_arp(bond, slave, sip, tip);
2724 else
2725 bond_validate_arp(bond, slave, tip, sip);
2726
2727 out_unlock:
2728 read_unlock(&bond->lock);
2729 if (arp != (struct arphdr *)skb->data)
2730 kfree(arp);
2731 return RX_HANDLER_ANOTHER;
2732 }
2733
2734 /*
2735 * this function is called regularly to monitor each slave's link
2736 * ensuring that traffic is being sent and received when arp monitoring
2737 * is used in load-balancing mode. if the adapter has been dormant, then an
2738 * arp is transmitted to generate traffic. see activebackup_arp_monitor for
2739 * arp monitoring in active backup mode.
2740 */
2741 void bond_loadbalance_arp_mon(struct work_struct *work)
2742 {
2743 struct bonding *bond = container_of(work, struct bonding,
2744 arp_work.work);
2745 struct slave *slave, *oldcurrent;
2746 int do_failover = 0;
2747 int delta_in_ticks, extra_ticks;
2748 int i;
2749
2750 read_lock(&bond->lock);
2751
2752 delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
2753 extra_ticks = delta_in_ticks / 2;
2754
2755 if (bond->slave_cnt == 0)
2756 goto re_arm;
2757
2758 read_lock(&bond->curr_slave_lock);
2759 oldcurrent = bond->curr_active_slave;
2760 read_unlock(&bond->curr_slave_lock);
2761
2762 /* see if any of the previous devices are up now (i.e. they have
2763 * xmt and rcv traffic). the curr_active_slave does not come into
2764 * the picture unless it is null. also, slave->jiffies is not needed
2765 * here because we send an arp on each slave and give a slave as
2766 * long as it needs to get the tx/rx within the delta.
2767 * TODO: what about up/down delay in arp mode? it wasn't here before
2768 * so it can wait
2769 */
2770 bond_for_each_slave(bond, slave, i) {
2771 unsigned long trans_start = dev_trans_start(slave->dev);
2772
2773 if (slave->link != BOND_LINK_UP) {
2774 if (time_in_range(jiffies,
2775 trans_start - delta_in_ticks,
2776 trans_start + delta_in_ticks + extra_ticks) &&
2777 time_in_range(jiffies,
2778 slave->dev->last_rx - delta_in_ticks,
2779 slave->dev->last_rx + delta_in_ticks + extra_ticks)) {
2780
2781 slave->link = BOND_LINK_UP;
2782 bond_set_active_slave(slave);
2783
2784 /* primary_slave has no meaning in round-robin
2785 * mode. the window of a slave being up and
2786 * curr_active_slave being null after enslaving
2787 * is closed.
2788 */
2789 if (!oldcurrent) {
2790 pr_info("%s: link status definitely up for interface %s, ",
2791 bond->dev->name,
2792 slave->dev->name);
2793 do_failover = 1;
2794 } else {
2795 pr_info("%s: interface %s is now up\n",
2796 bond->dev->name,
2797 slave->dev->name);
2798 }
2799 }
2800 } else {
2801 /* slave->link == BOND_LINK_UP */
2802
2803 /* not all switches will respond to an arp request
2804 * when the source ip is 0, so don't take the link down
2805 * if we don't know our ip yet
2806 */
2807 if (!time_in_range(jiffies,
2808 trans_start - delta_in_ticks,
2809 trans_start + 2 * delta_in_ticks + extra_ticks) ||
2810 !time_in_range(jiffies,
2811 slave->dev->last_rx - delta_in_ticks,
2812 slave->dev->last_rx + 2 * delta_in_ticks + extra_ticks)) {
2813
2814 slave->link = BOND_LINK_DOWN;
2815 bond_set_backup_slave(slave);
2816
2817 if (slave->link_failure_count < UINT_MAX)
2818 slave->link_failure_count++;
2819
2820 pr_info("%s: interface %s is now down.\n",
2821 bond->dev->name,
2822 slave->dev->name);
2823
2824 if (slave == oldcurrent)
2825 do_failover = 1;
2826 }
2827 }
2828
2829 /* note: if switch is in round-robin mode, all links
2830 * must tx arp to ensure all links rx an arp - otherwise
2831 * links may oscillate or not come up at all; if switch is
2832 * in something like xor mode, there is nothing we can
2833 * do - all replies will be rx'ed on same link causing slaves
2834 * to be unstable during low/no traffic periods
2835 */
2836 if (IS_UP(slave->dev))
2837 bond_arp_send_all(bond, slave);
2838 }
2839
2840 if (do_failover) {
2841 block_netpoll_tx();
2842 write_lock_bh(&bond->curr_slave_lock);
2843
2844 bond_select_active_slave(bond);
2845
2846 write_unlock_bh(&bond->curr_slave_lock);
2847 unblock_netpoll_tx();
2848 }
2849
2850 re_arm:
2851 if (bond->params.arp_interval)
2852 queue_delayed_work(bond->wq, &bond->arp_work, delta_in_ticks);
2853
2854 read_unlock(&bond->lock);
2855 }
2856
2857 /*
2858 * Called to inspect slaves for active-backup mode ARP monitor link state
2859 * changes. Sets new_link in slaves to specify what action should take
2860 * place for the slave. Returns 0 if no changes are found, >0 if changes
2861 * to link states must be committed.
2862 *
2863 * Called with bond->lock held for read.
2864 */
2865 static int bond_ab_arp_inspect(struct bonding *bond, int delta_in_ticks)
2866 {
2867 struct slave *slave;
2868 int i, commit = 0;
2869 unsigned long trans_start;
2870 int extra_ticks;
2871
2872 /* All the time comparisons below need some extra time. Otherwise, on
2873 * fast networks the ARP probe/reply may arrive within the same jiffy
2874 * as it was sent. Then, the next time the ARP monitor is run, one
2875 * arp_interval will already have passed in the comparisons.
2876 */
2877 extra_ticks = delta_in_ticks / 2;
2878
2879 bond_for_each_slave(bond, slave, i) {
2880 slave->new_link = BOND_LINK_NOCHANGE;
2881
2882 if (slave->link != BOND_LINK_UP) {
2883 if (time_in_range(jiffies,
2884 slave_last_rx(bond, slave) - delta_in_ticks,
2885 slave_last_rx(bond, slave) + delta_in_ticks + extra_ticks)) {
2886
2887 slave->new_link = BOND_LINK_UP;
2888 commit++;
2889 }
2890
2891 continue;
2892 }
2893
2894 /*
2895 * Give slaves 2*delta after being enslaved or made
2896 * active. This avoids bouncing, as the last receive
2897 * times need a full ARP monitor cycle to be updated.
2898 */
2899 if (time_in_range(jiffies,
2900 slave->jiffies - delta_in_ticks,
2901 slave->jiffies + 2 * delta_in_ticks + extra_ticks))
2902 continue;
2903
2904 /*
2905 * Backup slave is down if:
2906 * - No current_arp_slave AND
2907 * - more than 3*delta since last receive AND
2908 * - the bond has an IP address
2909 *
2910 * Note: a non-null current_arp_slave indicates
2911 * the curr_active_slave went down and we are
2912 * searching for a new one; under this condition
2913 * we only take the curr_active_slave down - this
2914 * gives each slave a chance to tx/rx traffic
2915 * before being taken out
2916 */
2917 if (!bond_is_active_slave(slave) &&
2918 !bond->current_arp_slave &&
2919 !time_in_range(jiffies,
2920 slave_last_rx(bond, slave) - delta_in_ticks,
2921 slave_last_rx(bond, slave) + 3 * delta_in_ticks + extra_ticks)) {
2922
2923 slave->new_link = BOND_LINK_DOWN;
2924 commit++;
2925 }
2926
2927 /*
2928 * Active slave is down if:
2929 * - more than 2*delta since transmitting OR
2930 * - (more than 2*delta since receive AND
2931 * the bond has an IP address)
2932 */
2933 trans_start = dev_trans_start(slave->dev);
2934 if (bond_is_active_slave(slave) &&
2935 (!time_in_range(jiffies,
2936 trans_start - delta_in_ticks,
2937 trans_start + 2 * delta_in_ticks + extra_ticks) ||
2938 !time_in_range(jiffies,
2939 slave_last_rx(bond, slave) - delta_in_ticks,
2940 slave_last_rx(bond, slave) + 2 * delta_in_ticks + extra_ticks))) {
2941
2942 slave->new_link = BOND_LINK_DOWN;
2943 commit++;
2944 }
2945 }
2946
2947 return commit;
2948 }
2949
2950 /*
2951 * Called to commit link state changes noted by inspection step of
2952 * active-backup mode ARP monitor.
2953 *
2954 * Called with RTNL and bond->lock for read.
2955 */
2956 static void bond_ab_arp_commit(struct bonding *bond, int delta_in_ticks)
2957 {
2958 struct slave *slave;
2959 int i;
2960 unsigned long trans_start;
2961
2962 bond_for_each_slave(bond, slave, i) {
2963 switch (slave->new_link) {
2964 case BOND_LINK_NOCHANGE:
2965 continue;
2966
2967 case BOND_LINK_UP:
2968 trans_start = dev_trans_start(slave->dev);
2969 if ((!bond->curr_active_slave &&
2970 time_in_range(jiffies,
2971 trans_start - delta_in_ticks,
2972 trans_start + delta_in_ticks + delta_in_ticks / 2)) ||
2973 bond->curr_active_slave != slave) {
2974 slave->link = BOND_LINK_UP;
2975 if (bond->current_arp_slave) {
2976 bond_set_slave_inactive_flags(
2977 bond->current_arp_slave);
2978 bond->current_arp_slave = NULL;
2979 }
2980
2981 pr_info("%s: link status definitely up for interface %s.\n",
2982 bond->dev->name, slave->dev->name);
2983
2984 if (!bond->curr_active_slave ||
2985 (slave == bond->primary_slave))
2986 goto do_failover;
2987
2988 }
2989
2990 continue;
2991
2992 case BOND_LINK_DOWN:
2993 if (slave->link_failure_count < UINT_MAX)
2994 slave->link_failure_count++;
2995
2996 slave->link = BOND_LINK_DOWN;
2997 bond_set_slave_inactive_flags(slave);
2998
2999 pr_info("%s: link status definitely down for interface %s, disabling it\n",
3000 bond->dev->name, slave->dev->name);
3001
3002 if (slave == bond->curr_active_slave) {
3003 bond->current_arp_slave = NULL;
3004 goto do_failover;
3005 }
3006
3007 continue;
3008
3009 default:
3010 pr_err("%s: impossible: new_link %d on slave %s\n",
3011 bond->dev->name, slave->new_link,
3012 slave->dev->name);
3013 continue;
3014 }
3015
3016 do_failover:
3017 ASSERT_RTNL();
3018 block_netpoll_tx();
3019 write_lock_bh(&bond->curr_slave_lock);
3020 bond_select_active_slave(bond);
3021 write_unlock_bh(&bond->curr_slave_lock);
3022 unblock_netpoll_tx();
3023 }
3024
3025 bond_set_carrier(bond);
3026 }
3027
3028 /*
3029 * Send ARP probes for active-backup mode ARP monitor.
3030 *
3031 * Called with bond->lock held for read.
3032 */
3033 static void bond_ab_arp_probe(struct bonding *bond)
3034 {
3035 struct slave *slave;
3036 int i;
3037
3038 read_lock(&bond->curr_slave_lock);
3039
3040 if (bond->current_arp_slave && bond->curr_active_slave)
3041 pr_info("PROBE: c_arp %s && cas %s BAD\n",
3042 bond->current_arp_slave->dev->name,
3043 bond->curr_active_slave->dev->name);
3044
3045 if (bond->curr_active_slave) {
3046 bond_arp_send_all(bond, bond->curr_active_slave);
3047 read_unlock(&bond->curr_slave_lock);
3048 return;
3049 }
3050
3051 read_unlock(&bond->curr_slave_lock);
3052
3053 /* if we don't have a curr_active_slave, search for the next available
3054 * backup slave from the current_arp_slave and make it the candidate
3055 * for becoming the curr_active_slave
3056 */
3057
3058 if (!bond->current_arp_slave) {
3059 bond->current_arp_slave = bond->first_slave;
3060 if (!bond->current_arp_slave)
3061 return;
3062 }
3063
3064 bond_set_slave_inactive_flags(bond->current_arp_slave);
3065
3066 /* search for next candidate */
3067 bond_for_each_slave_from(bond, slave, i, bond->current_arp_slave->next) {
3068 if (IS_UP(slave->dev)) {
3069 slave->link = BOND_LINK_BACK;
3070 bond_set_slave_active_flags(slave);
3071 bond_arp_send_all(bond, slave);
3072 slave->jiffies = jiffies;
3073 bond->current_arp_slave = slave;
3074 break;
3075 }
3076
3077 /* if the link state is up at this point, we
3078 * mark it down - this can happen if we have
3079 * simultaneous link failures and
3080 * reselect_active_interface doesn't make this
3081 * one the current slave so it is still marked
3082 * up when it is actually down
3083 */
3084 if (slave->link == BOND_LINK_UP) {
3085 slave->link = BOND_LINK_DOWN;
3086 if (slave->link_failure_count < UINT_MAX)
3087 slave->link_failure_count++;
3088
3089 bond_set_slave_inactive_flags(slave);
3090
3091 pr_info("%s: backup interface %s is now down.\n",
3092 bond->dev->name, slave->dev->name);
3093 }
3094 }
3095 }
3096
3097 void bond_activebackup_arp_mon(struct work_struct *work)
3098 {
3099 struct bonding *bond = container_of(work, struct bonding,
3100 arp_work.work);
3101 bool should_notify_peers = false;
3102 int delta_in_ticks;
3103
3104 read_lock(&bond->lock);
3105
3106 delta_in_ticks = msecs_to_jiffies(bond->params.arp_interval);
3107
3108 if (bond->slave_cnt == 0)
3109 goto re_arm;
3110
3111 should_notify_peers = bond_should_notify_peers(bond);
3112
3113 if (bond_ab_arp_inspect(bond, delta_in_ticks)) {
3114 read_unlock(&bond->lock);
3115
3116 /* Race avoidance with bond_close flush of workqueue */
3117 if (!rtnl_trylock()) {
3118 read_lock(&bond->lock);
3119 delta_in_ticks = 1;
3120 should_notify_peers = false;
3121 goto re_arm;
3122 }
3123
3124 read_lock(&bond->lock);
3125
3126 bond_ab_arp_commit(bond, delta_in_ticks);
3127
3128 read_unlock(&bond->lock);
3129 rtnl_unlock();
3130 read_lock(&bond->lock);
3131 }
3132
3133 bond_ab_arp_probe(bond);
3134
3135 re_arm:
3136 if (bond->params.arp_interval)
3137 queue_delayed_work(bond->wq, &bond->arp_work, delta_in_ticks);
3138
3139 read_unlock(&bond->lock);
3140
3141 if (should_notify_peers) {
3142 if (!rtnl_trylock()) {
3143 read_lock(&bond->lock);
3144 bond->send_peer_notif++;
3145 read_unlock(&bond->lock);
3146 return;
3147 }
3148 call_netdevice_notifiers(NETDEV_NOTIFY_PEERS, bond->dev);
3149 rtnl_unlock();
3150 }
3151 }
3152
3153 /*-------------------------- netdev event handling --------------------------*/
3154
3155 /*
3156 * Change device name
3157 */
3158 static int bond_event_changename(struct bonding *bond)
3159 {
3160 bond_remove_proc_entry(bond);
3161 bond_create_proc_entry(bond);
3162
3163 bond_debug_reregister(bond);
3164
3165 return NOTIFY_DONE;
3166 }
3167
3168 static int bond_master_netdev_event(unsigned long event,
3169 struct net_device *bond_dev)
3170 {
3171 struct bonding *event_bond = netdev_priv(bond_dev);
3172
3173 switch (event) {
3174 case NETDEV_CHANGENAME:
3175 return bond_event_changename(event_bond);
3176 case NETDEV_UNREGISTER:
3177 bond_remove_proc_entry(event_bond);
3178 break;
3179 case NETDEV_REGISTER:
3180 bond_create_proc_entry(event_bond);
3181 break;
3182 default:
3183 break;
3184 }
3185
3186 return NOTIFY_DONE;
3187 }
3188
3189 static int bond_slave_netdev_event(unsigned long event,
3190 struct net_device *slave_dev)
3191 {
3192 struct slave *slave = bond_slave_get_rtnl(slave_dev);
3193 struct bonding *bond;
3194 struct net_device *bond_dev;
3195 u32 old_speed;
3196 u8 old_duplex;
3197
3198 /* A netdev event can be generated while enslaving a device
3199 * before netdev_rx_handler_register is called in which case
3200 * slave will be NULL
3201 */
3202 if (!slave)
3203 return NOTIFY_DONE;
3204 bond_dev = slave->bond->dev;
3205 bond = slave->bond;
3206
3207 switch (event) {
3208 case NETDEV_UNREGISTER:
3209 if (bond->setup_by_slave)
3210 bond_release_and_destroy(bond_dev, slave_dev);
3211 else
3212 bond_release(bond_dev, slave_dev);
3213 break;
3214 case NETDEV_UP:
3215 case NETDEV_CHANGE:
3216 old_speed = slave->speed;
3217 old_duplex = slave->duplex;
3218
3219 bond_update_speed_duplex(slave);
3220
3221 if (bond->params.mode == BOND_MODE_8023AD) {
3222 if (old_speed != slave->speed)
3223 bond_3ad_adapter_speed_changed(slave);
3224 if (old_duplex != slave->duplex)
3225 bond_3ad_adapter_duplex_changed(slave);
3226 }
3227 break;
3228 case NETDEV_DOWN:
3229 /*
3230 * ... Or is it this?
3231 */
3232 break;
3233 case NETDEV_CHANGEMTU:
3234 /*
3235 * TODO: Should slaves be allowed to
3236 * independently alter their MTU? For
3237 * an active-backup bond, slaves need
3238 * not be the same type of device, so
3239 * MTUs may vary. For other modes,
3240 * slaves arguably should have the
3241 * same MTUs. To do this, we'd need to
3242 * take over the slave's change_mtu
3243 * function for the duration of their
3244 * servitude.
3245 */
3246 break;
3247 case NETDEV_CHANGENAME:
3248 /*
3249 * TODO: handle changing the primary's name
3250 */
3251 break;
3252 case NETDEV_FEAT_CHANGE:
3253 bond_compute_features(bond);
3254 break;
3255 default:
3256 break;
3257 }
3258
3259 return NOTIFY_DONE;
3260 }
3261
3262 /*
3263 * bond_netdev_event: handle netdev notifier chain events.
3264 *
3265 * This function receives events for the netdev chain. The caller (an
3266 * ioctl handler calling blocking_notifier_call_chain) holds the necessary
3267 * locks for us to safely manipulate the slave devices (RTNL lock,
3268 * dev_probe_lock).
3269 */
3270 static int bond_netdev_event(struct notifier_block *this,
3271 unsigned long event, void *ptr)
3272 {
3273 struct net_device *event_dev = (struct net_device *)ptr;
3274
3275 pr_debug("event_dev: %s, event: %lx\n",
3276 event_dev ? event_dev->name : "None",
3277 event);
3278
3279 if (!(event_dev->priv_flags & IFF_BONDING))
3280 return NOTIFY_DONE;
3281
3282 if (event_dev->flags & IFF_MASTER) {
3283 pr_debug("IFF_MASTER\n");
3284 return bond_master_netdev_event(event, event_dev);
3285 }
3286
3287 if (event_dev->flags & IFF_SLAVE) {
3288 pr_debug("IFF_SLAVE\n");
3289 return bond_slave_netdev_event(event, event_dev);
3290 }
3291
3292 return NOTIFY_DONE;
3293 }
3294
3295 static struct notifier_block bond_netdev_notifier = {
3296 .notifier_call = bond_netdev_event,
3297 };
3298
3299 /*---------------------------- Hashing Policies -----------------------------*/
3300
3301 /*
3302 * Hash for the output device based upon layer 2 data
3303 */
3304 static int bond_xmit_hash_policy_l2(struct sk_buff *skb, int count)
3305 {
3306 struct ethhdr *data = (struct ethhdr *)skb->data;
3307
3308 if (skb_headlen(skb) >= offsetof(struct ethhdr, h_proto))
3309 return (data->h_dest[5] ^ data->h_source[5]) % count;
3310
3311 return 0;
3312 }
3313
3314 /*
3315 * Hash for the output device based upon layer 2 and layer 3 data. If
3316 * the packet is not IP, fall back on bond_xmit_hash_policy_l2()
3317 */
3318 static int bond_xmit_hash_policy_l23(struct sk_buff *skb, int count)
3319 {
3320 const struct ethhdr *data;
3321 const struct iphdr *iph;
3322 const struct ipv6hdr *ipv6h;
3323 u32 v6hash;
3324 const __be32 *s, *d;
3325
3326 if (skb->protocol == htons(ETH_P_IP) &&
3327 pskb_network_may_pull(skb, sizeof(*iph))) {
3328 iph = ip_hdr(skb);
3329 data = (struct ethhdr *)skb->data;
3330 return ((ntohl(iph->saddr ^ iph->daddr) & 0xffff) ^
3331 (data->h_dest[5] ^ data->h_source[5])) % count;
3332 } else if (skb->protocol == htons(ETH_P_IPV6) &&
3333 pskb_network_may_pull(skb, sizeof(*ipv6h))) {
3334 ipv6h = ipv6_hdr(skb);
3335 data = (struct ethhdr *)skb->data;
3336 s = &ipv6h->saddr.s6_addr32[0];
3337 d = &ipv6h->daddr.s6_addr32[0];
3338 v6hash = (s[1] ^ d[1]) ^ (s[2] ^ d[2]) ^ (s[3] ^ d[3]);
3339 v6hash ^= (v6hash >> 24) ^ (v6hash >> 16) ^ (v6hash >> 8);
3340 return (v6hash ^ data->h_dest[5] ^ data->h_source[5]) % count;
3341 }
3342
3343 return bond_xmit_hash_policy_l2(skb, count);
3344 }
3345
3346 /*
3347 * Hash for the output device based upon layer 3 and layer 4 data. If
3348 * the packet is a frag or not TCP or UDP, just use layer 3 data. If it is
3349 * altogether not IP, fall back on bond_xmit_hash_policy_l2()
3350 */
3351 static int bond_xmit_hash_policy_l34(struct sk_buff *skb, int count)
3352 {
3353 u32 layer4_xor = 0;
3354 const struct iphdr *iph;
3355 const struct ipv6hdr *ipv6h;
3356 const __be32 *s, *d;
3357 const __be16 *l4 = NULL;
3358 __be16 _l4[2];
3359 int noff = skb_network_offset(skb);
3360 int poff;
3361
3362 if (skb->protocol == htons(ETH_P_IP) &&
3363 pskb_may_pull(skb, noff + sizeof(*iph))) {
3364 iph = ip_hdr(skb);
3365 poff = proto_ports_offset(iph->protocol);
3366
3367 if (!ip_is_fragment(iph) && poff >= 0) {
3368 l4 = skb_header_pointer(skb, noff + (iph->ihl << 2) + poff,
3369 sizeof(_l4), &_l4);
3370 if (l4)
3371 layer4_xor = ntohs(l4[0] ^ l4[1]);
3372 }
3373 return (layer4_xor ^
3374 ((ntohl(iph->saddr ^ iph->daddr)) & 0xffff)) % count;
3375 } else if (skb->protocol == htons(ETH_P_IPV6) &&
3376 pskb_may_pull(skb, noff + sizeof(*ipv6h))) {
3377 ipv6h = ipv6_hdr(skb);
3378 poff = proto_ports_offset(ipv6h->nexthdr);
3379 if (poff >= 0) {
3380 l4 = skb_header_pointer(skb, noff + sizeof(*ipv6h) + poff,
3381 sizeof(_l4), &_l4);
3382 if (l4)
3383 layer4_xor = ntohs(l4[0] ^ l4[1]);
3384 }
3385 s = &ipv6h->saddr.s6_addr32[0];
3386 d = &ipv6h->daddr.s6_addr32[0];
3387 layer4_xor ^= (s[1] ^ d[1]) ^ (s[2] ^ d[2]) ^ (s[3] ^ d[3]);
3388 layer4_xor ^= (layer4_xor >> 24) ^ (layer4_xor >> 16) ^
3389 (layer4_xor >> 8);
3390 return layer4_xor % count;
3391 }
3392
3393 return bond_xmit_hash_policy_l2(skb, count);
3394 }
3395
3396 /*-------------------------- Device entry points ----------------------------*/
3397
3398 static void bond_work_init_all(struct bonding *bond)
3399 {
3400 INIT_DELAYED_WORK(&bond->mcast_work,
3401 bond_resend_igmp_join_requests_delayed);
3402 INIT_DELAYED_WORK(&bond->alb_work, bond_alb_monitor);
3403 INIT_DELAYED_WORK(&bond->mii_work, bond_mii_monitor);
3404 if (bond->params.mode == BOND_MODE_ACTIVEBACKUP)
3405 INIT_DELAYED_WORK(&bond->arp_work, bond_activebackup_arp_mon);
3406 else
3407 INIT_DELAYED_WORK(&bond->arp_work, bond_loadbalance_arp_mon);
3408 INIT_DELAYED_WORK(&bond->ad_work, bond_3ad_state_machine_handler);
3409 }
3410
3411 static void bond_work_cancel_all(struct bonding *bond)
3412 {
3413 cancel_delayed_work_sync(&bond->mii_work);
3414 cancel_delayed_work_sync(&bond->arp_work);
3415 cancel_delayed_work_sync(&bond->alb_work);
3416 cancel_delayed_work_sync(&bond->ad_work);
3417 cancel_delayed_work_sync(&bond->mcast_work);
3418 }
3419
3420 static int bond_open(struct net_device *bond_dev)
3421 {
3422 struct bonding *bond = netdev_priv(bond_dev);
3423 struct slave *slave;
3424 int i;
3425
3426 /* reset slave->backup and slave->inactive */
3427 read_lock(&bond->lock);
3428 if (bond->slave_cnt > 0) {
3429 read_lock(&bond->curr_slave_lock);
3430 bond_for_each_slave(bond, slave, i) {
3431 if ((bond->params.mode == BOND_MODE_ACTIVEBACKUP)
3432 && (slave != bond->curr_active_slave)) {
3433 bond_set_slave_inactive_flags(slave);
3434 } else {
3435 bond_set_slave_active_flags(slave);
3436 }
3437 }
3438 read_unlock(&bond->curr_slave_lock);
3439 }
3440 read_unlock(&bond->lock);
3441
3442 bond_work_init_all(bond);
3443
3444 if (bond_is_lb(bond)) {
3445 /* bond_alb_initialize must be called before the timer
3446 * is started.
3447 */
3448 if (bond_alb_initialize(bond, (bond->params.mode == BOND_MODE_ALB)))
3449 return -ENOMEM;
3450 queue_delayed_work(bond->wq, &bond->alb_work, 0);
3451 }
3452
3453 if (bond->params.miimon) /* link check interval, in milliseconds. */
3454 queue_delayed_work(bond->wq, &bond->mii_work, 0);
3455
3456 if (bond->params.arp_interval) { /* arp interval, in milliseconds. */
3457 queue_delayed_work(bond->wq, &bond->arp_work, 0);
3458 if (bond->params.arp_validate)
3459 bond->recv_probe = bond_arp_rcv;
3460 }
3461
3462 if (bond->params.mode == BOND_MODE_8023AD) {
3463 queue_delayed_work(bond->wq, &bond->ad_work, 0);
3464 /* register to receive LACPDUs */
3465 bond->recv_probe = bond_3ad_lacpdu_recv;
3466 bond_3ad_initiate_agg_selection(bond, 1);
3467 }
3468
3469 return 0;
3470 }
3471
3472 static int bond_close(struct net_device *bond_dev)
3473 {
3474 struct bonding *bond = netdev_priv(bond_dev);
3475
3476 write_lock_bh(&bond->lock);
3477 bond->send_peer_notif = 0;
3478 write_unlock_bh(&bond->lock);
3479
3480 bond_work_cancel_all(bond);
3481 if (bond_is_lb(bond)) {
3482 /* Must be called only after all
3483 * slaves have been released
3484 */
3485 bond_alb_deinitialize(bond);
3486 }
3487 bond->recv_probe = NULL;
3488
3489 return 0;
3490 }
3491
3492 static struct rtnl_link_stats64 *bond_get_stats(struct net_device *bond_dev,
3493 struct rtnl_link_stats64 *stats)
3494 {
3495 struct bonding *bond = netdev_priv(bond_dev);
3496 struct rtnl_link_stats64 temp;
3497 struct slave *slave;
3498 int i;
3499
3500 memset(stats, 0, sizeof(*stats));
3501
3502 read_lock_bh(&bond->lock);
3503
3504 bond_for_each_slave(bond, slave, i) {
3505 const struct rtnl_link_stats64 *sstats =
3506 dev_get_stats(slave->dev, &temp);
3507
3508 stats->rx_packets += sstats->rx_packets;
3509 stats->rx_bytes += sstats->rx_bytes;
3510 stats->rx_errors += sstats->rx_errors;
3511 stats->rx_dropped += sstats->rx_dropped;
3512
3513 stats->tx_packets += sstats->tx_packets;
3514 stats->tx_bytes += sstats->tx_bytes;
3515 stats->tx_errors += sstats->tx_errors;
3516 stats->tx_dropped += sstats->tx_dropped;
3517
3518 stats->multicast += sstats->multicast;
3519 stats->collisions += sstats->collisions;
3520
3521 stats->rx_length_errors += sstats->rx_length_errors;
3522 stats->rx_over_errors += sstats->rx_over_errors;
3523 stats->rx_crc_errors += sstats->rx_crc_errors;
3524 stats->rx_frame_errors += sstats->rx_frame_errors;
3525 stats->rx_fifo_errors += sstats->rx_fifo_errors;
3526 stats->rx_missed_errors += sstats->rx_missed_errors;
3527
3528 stats->tx_aborted_errors += sstats->tx_aborted_errors;
3529 stats->tx_carrier_errors += sstats->tx_carrier_errors;
3530 stats->tx_fifo_errors += sstats->tx_fifo_errors;
3531 stats->tx_heartbeat_errors += sstats->tx_heartbeat_errors;
3532 stats->tx_window_errors += sstats->tx_window_errors;
3533 }
3534
3535 read_unlock_bh(&bond->lock);
3536
3537 return stats;
3538 }
3539
3540 static int bond_do_ioctl(struct net_device *bond_dev, struct ifreq *ifr, int cmd)
3541 {
3542 struct net_device *slave_dev = NULL;
3543 struct ifbond k_binfo;
3544 struct ifbond __user *u_binfo = NULL;
3545 struct ifslave k_sinfo;
3546 struct ifslave __user *u_sinfo = NULL;
3547 struct mii_ioctl_data *mii = NULL;
3548 struct net *net;
3549 int res = 0;
3550
3551 pr_debug("bond_ioctl: master=%s, cmd=%d\n", bond_dev->name, cmd);
3552
3553 switch (cmd) {
3554 case SIOCGMIIPHY:
3555 mii = if_mii(ifr);
3556 if (!mii)
3557 return -EINVAL;
3558
3559 mii->phy_id = 0;
3560 /* Fall Through */
3561 case SIOCGMIIREG:
3562 /*
3563 * We do this again just in case we were called by SIOCGMIIREG
3564 * instead of SIOCGMIIPHY.
3565 */
3566 mii = if_mii(ifr);
3567 if (!mii)
3568 return -EINVAL;
3569
3570
3571 if (mii->reg_num == 1) {
3572 struct bonding *bond = netdev_priv(bond_dev);
3573 mii->val_out = 0;
3574 read_lock(&bond->lock);
3575 read_lock(&bond->curr_slave_lock);
3576 if (netif_carrier_ok(bond->dev))
3577 mii->val_out = BMSR_LSTATUS;
3578
3579 read_unlock(&bond->curr_slave_lock);
3580 read_unlock(&bond->lock);
3581 }
3582
3583 return 0;
3584 case BOND_INFO_QUERY_OLD:
3585 case SIOCBONDINFOQUERY:
3586 u_binfo = (struct ifbond __user *)ifr->ifr_data;
3587
3588 if (copy_from_user(&k_binfo, u_binfo, sizeof(ifbond)))
3589 return -EFAULT;
3590
3591 res = bond_info_query(bond_dev, &k_binfo);
3592 if (res == 0 &&
3593 copy_to_user(u_binfo, &k_binfo, sizeof(ifbond)))
3594 return -EFAULT;
3595
3596 return res;
3597 case BOND_SLAVE_INFO_QUERY_OLD:
3598 case SIOCBONDSLAVEINFOQUERY:
3599 u_sinfo = (struct ifslave __user *)ifr->ifr_data;
3600
3601 if (copy_from_user(&k_sinfo, u_sinfo, sizeof(ifslave)))
3602 return -EFAULT;
3603
3604 res = bond_slave_info_query(bond_dev, &k_sinfo);
3605 if (res == 0 &&
3606 copy_to_user(u_sinfo, &k_sinfo, sizeof(ifslave)))
3607 return -EFAULT;
3608
3609 return res;
3610 default:
3611 /* Go on */
3612 break;
3613 }
3614
3615 net = dev_net(bond_dev);
3616
3617 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
3618 return -EPERM;
3619
3620 slave_dev = dev_get_by_name(net, ifr->ifr_slave);
3621
3622 pr_debug("slave_dev=%p:\n", slave_dev);
3623
3624 if (!slave_dev)
3625 res = -ENODEV;
3626 else {
3627 pr_debug("slave_dev->name=%s:\n", slave_dev->name);
3628 switch (cmd) {
3629 case BOND_ENSLAVE_OLD:
3630 case SIOCBONDENSLAVE:
3631 res = bond_enslave(bond_dev, slave_dev);
3632 break;
3633 case BOND_RELEASE_OLD:
3634 case SIOCBONDRELEASE:
3635 res = bond_release(bond_dev, slave_dev);
3636 break;
3637 case BOND_SETHWADDR_OLD:
3638 case SIOCBONDSETHWADDR:
3639 bond_set_dev_addr(bond_dev, slave_dev);
3640 res = 0;
3641 break;
3642 case BOND_CHANGE_ACTIVE_OLD:
3643 case SIOCBONDCHANGEACTIVE:
3644 res = bond_ioctl_change_active(bond_dev, slave_dev);
3645 break;
3646 default:
3647 res = -EOPNOTSUPP;
3648 }
3649
3650 dev_put(slave_dev);
3651 }
3652
3653 return res;
3654 }
3655
3656 static bool bond_addr_in_mc_list(unsigned char *addr,
3657 struct netdev_hw_addr_list *list,
3658 int addrlen)
3659 {
3660 struct netdev_hw_addr *ha;
3661
3662 netdev_hw_addr_list_for_each(ha, list)
3663 if (!memcmp(ha->addr, addr, addrlen))
3664 return true;
3665
3666 return false;
3667 }
3668
3669 static void bond_change_rx_flags(struct net_device *bond_dev, int change)
3670 {
3671 struct bonding *bond = netdev_priv(bond_dev);
3672
3673 if (change & IFF_PROMISC)
3674 bond_set_promiscuity(bond,
3675 bond_dev->flags & IFF_PROMISC ? 1 : -1);
3676
3677 if (change & IFF_ALLMULTI)
3678 bond_set_allmulti(bond,
3679 bond_dev->flags & IFF_ALLMULTI ? 1 : -1);
3680 }
3681
3682 static void bond_set_multicast_list(struct net_device *bond_dev)
3683 {
3684 struct bonding *bond = netdev_priv(bond_dev);
3685 struct netdev_hw_addr *ha;
3686 bool found;
3687
3688 read_lock(&bond->lock);
3689
3690 /* looking for addresses to add to slaves' mc list */
3691 netdev_for_each_mc_addr(ha, bond_dev) {
3692 found = bond_addr_in_mc_list(ha->addr, &bond->mc_list,
3693 bond_dev->addr_len);
3694 if (!found)
3695 bond_mc_add(bond, ha->addr);
3696 }
3697
3698 /* looking for addresses to delete from slaves' list */
3699 netdev_hw_addr_list_for_each(ha, &bond->mc_list) {
3700 found = bond_addr_in_mc_list(ha->addr, &bond_dev->mc,
3701 bond_dev->addr_len);
3702 if (!found)
3703 bond_mc_del(bond, ha->addr);
3704 }
3705
3706 /* save master's multicast list */
3707 __hw_addr_flush(&bond->mc_list);
3708 __hw_addr_add_multiple(&bond->mc_list, &bond_dev->mc,
3709 bond_dev->addr_len, NETDEV_HW_ADDR_T_MULTICAST);
3710
3711 read_unlock(&bond->lock);
3712 }
3713
3714 static int bond_neigh_init(struct neighbour *n)
3715 {
3716 struct bonding *bond = netdev_priv(n->dev);
3717 struct slave *slave = bond->first_slave;
3718 const struct net_device_ops *slave_ops;
3719 struct neigh_parms parms;
3720 int ret;
3721
3722 if (!slave)
3723 return 0;
3724
3725 slave_ops = slave->dev->netdev_ops;
3726
3727 if (!slave_ops->ndo_neigh_setup)
3728 return 0;
3729
3730 parms.neigh_setup = NULL;
3731 parms.neigh_cleanup = NULL;
3732 ret = slave_ops->ndo_neigh_setup(slave->dev, &parms);
3733 if (ret)
3734 return ret;
3735
3736 /*
3737 * Assign slave's neigh_cleanup to neighbour in case cleanup is called
3738 * after the last slave has been detached. Assumes that all slaves
3739 * utilize the same neigh_cleanup (true at this writing as only user
3740 * is ipoib).
3741 */
3742 n->parms->neigh_cleanup = parms.neigh_cleanup;
3743
3744 if (!parms.neigh_setup)
3745 return 0;
3746
3747 return parms.neigh_setup(n);
3748 }
3749
3750 /*
3751 * The bonding ndo_neigh_setup is called at init time beofre any
3752 * slave exists. So we must declare proxy setup function which will
3753 * be used at run time to resolve the actual slave neigh param setup.
3754 */
3755 static int bond_neigh_setup(struct net_device *dev,
3756 struct neigh_parms *parms)
3757 {
3758 parms->neigh_setup = bond_neigh_init;
3759
3760 return 0;
3761 }
3762
3763 /*
3764 * Change the MTU of all of a master's slaves to match the master
3765 */
3766 static int bond_change_mtu(struct net_device *bond_dev, int new_mtu)
3767 {
3768 struct bonding *bond = netdev_priv(bond_dev);
3769 struct slave *slave, *stop_at;
3770 int res = 0;
3771 int i;
3772
3773 pr_debug("bond=%p, name=%s, new_mtu=%d\n", bond,
3774 (bond_dev ? bond_dev->name : "None"), new_mtu);
3775
3776 /* Can't hold bond->lock with bh disabled here since
3777 * some base drivers panic. On the other hand we can't
3778 * hold bond->lock without bh disabled because we'll
3779 * deadlock. The only solution is to rely on the fact
3780 * that we're under rtnl_lock here, and the slaves
3781 * list won't change. This doesn't solve the problem
3782 * of setting the slave's MTU while it is
3783 * transmitting, but the assumption is that the base
3784 * driver can handle that.
3785 *
3786 * TODO: figure out a way to safely iterate the slaves
3787 * list, but without holding a lock around the actual
3788 * call to the base driver.
3789 */
3790
3791 bond_for_each_slave(bond, slave, i) {
3792 pr_debug("s %p s->p %p c_m %p\n",
3793 slave,
3794 slave->prev,
3795 slave->dev->netdev_ops->ndo_change_mtu);
3796
3797 res = dev_set_mtu(slave->dev, new_mtu);
3798
3799 if (res) {
3800 /* If we failed to set the slave's mtu to the new value
3801 * we must abort the operation even in ACTIVE_BACKUP
3802 * mode, because if we allow the backup slaves to have
3803 * different mtu values than the active slave we'll
3804 * need to change their mtu when doing a failover. That
3805 * means changing their mtu from timer context, which
3806 * is probably not a good idea.
3807 */
3808 pr_debug("err %d %s\n", res, slave->dev->name);
3809 goto unwind;
3810 }
3811 }
3812
3813 bond_dev->mtu = new_mtu;
3814
3815 return 0;
3816
3817 unwind:
3818 /* unwind from head to the slave that failed */
3819 stop_at = slave;
3820 bond_for_each_slave_from_to(bond, slave, i, bond->first_slave, stop_at) {
3821 int tmp_res;
3822
3823 tmp_res = dev_set_mtu(slave->dev, bond_dev->mtu);
3824 if (tmp_res) {
3825 pr_debug("unwind err %d dev %s\n",
3826 tmp_res, slave->dev->name);
3827 }
3828 }
3829
3830 return res;
3831 }
3832
3833 /*
3834 * Change HW address
3835 *
3836 * Note that many devices must be down to change the HW address, and
3837 * downing the master releases all slaves. We can make bonds full of
3838 * bonding devices to test this, however.
3839 */
3840 static int bond_set_mac_address(struct net_device *bond_dev, void *addr)
3841 {
3842 struct bonding *bond = netdev_priv(bond_dev);
3843 struct sockaddr *sa = addr, tmp_sa;
3844 struct slave *slave, *stop_at;
3845 int res = 0;
3846 int i;
3847
3848 if (bond->params.mode == BOND_MODE_ALB)
3849 return bond_alb_set_mac_address(bond_dev, addr);
3850
3851
3852 pr_debug("bond=%p, name=%s\n",
3853 bond, bond_dev ? bond_dev->name : "None");
3854
3855 /*
3856 * If fail_over_mac is set to active, do nothing and return
3857 * success. Returning an error causes ifenslave to fail.
3858 */
3859 if (bond->params.fail_over_mac == BOND_FOM_ACTIVE)
3860 return 0;
3861
3862 if (!is_valid_ether_addr(sa->sa_data))
3863 return -EADDRNOTAVAIL;
3864
3865 /* Can't hold bond->lock with bh disabled here since
3866 * some base drivers panic. On the other hand we can't
3867 * hold bond->lock without bh disabled because we'll
3868 * deadlock. The only solution is to rely on the fact
3869 * that we're under rtnl_lock here, and the slaves
3870 * list won't change. This doesn't solve the problem
3871 * of setting the slave's hw address while it is
3872 * transmitting, but the assumption is that the base
3873 * driver can handle that.
3874 *
3875 * TODO: figure out a way to safely iterate the slaves
3876 * list, but without holding a lock around the actual
3877 * call to the base driver.
3878 */
3879
3880 bond_for_each_slave(bond, slave, i) {
3881 const struct net_device_ops *slave_ops = slave->dev->netdev_ops;
3882 pr_debug("slave %p %s\n", slave, slave->dev->name);
3883
3884 if (slave_ops->ndo_set_mac_address == NULL) {
3885 res = -EOPNOTSUPP;
3886 pr_debug("EOPNOTSUPP %s\n", slave->dev->name);
3887 goto unwind;
3888 }
3889
3890 res = dev_set_mac_address(slave->dev, addr);
3891 if (res) {
3892 /* TODO: consider downing the slave
3893 * and retry ?
3894 * User should expect communications
3895 * breakage anyway until ARP finish
3896 * updating, so...
3897 */
3898 pr_debug("err %d %s\n", res, slave->dev->name);
3899 goto unwind;
3900 }
3901 }
3902
3903 /* success */
3904 memcpy(bond_dev->dev_addr, sa->sa_data, bond_dev->addr_len);
3905 return 0;
3906
3907 unwind:
3908 memcpy(tmp_sa.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
3909 tmp_sa.sa_family = bond_dev->type;
3910
3911 /* unwind from head to the slave that failed */
3912 stop_at = slave;
3913 bond_for_each_slave_from_to(bond, slave, i, bond->first_slave, stop_at) {
3914 int tmp_res;
3915
3916 tmp_res = dev_set_mac_address(slave->dev, &tmp_sa);
3917 if (tmp_res) {
3918 pr_debug("unwind err %d dev %s\n",
3919 tmp_res, slave->dev->name);
3920 }
3921 }
3922
3923 return res;
3924 }
3925
3926 static int bond_xmit_roundrobin(struct sk_buff *skb, struct net_device *bond_dev)
3927 {
3928 struct bonding *bond = netdev_priv(bond_dev);
3929 struct slave *slave, *start_at;
3930 int i, slave_no, res = 1;
3931 struct iphdr *iph = ip_hdr(skb);
3932
3933 /*
3934 * Start with the curr_active_slave that joined the bond as the
3935 * default for sending IGMP traffic. For failover purposes one
3936 * needs to maintain some consistency for the interface that will
3937 * send the join/membership reports. The curr_active_slave found
3938 * will send all of this type of traffic.
3939 */
3940 if ((iph->protocol == IPPROTO_IGMP) &&
3941 (skb->protocol == htons(ETH_P_IP))) {
3942
3943 read_lock(&bond->curr_slave_lock);
3944 slave = bond->curr_active_slave;
3945 read_unlock(&bond->curr_slave_lock);
3946
3947 if (!slave)
3948 goto out;
3949 } else {
3950 /*
3951 * Concurrent TX may collide on rr_tx_counter; we accept
3952 * that as being rare enough not to justify using an
3953 * atomic op here.
3954 */
3955 slave_no = bond->rr_tx_counter++ % bond->slave_cnt;
3956
3957 bond_for_each_slave(bond, slave, i) {
3958 slave_no--;
3959 if (slave_no < 0)
3960 break;
3961 }
3962 }
3963
3964 start_at = slave;
3965 bond_for_each_slave_from(bond, slave, i, start_at) {
3966 if (IS_UP(slave->dev) &&
3967 (slave->link == BOND_LINK_UP) &&
3968 bond_is_active_slave(slave)) {
3969 res = bond_dev_queue_xmit(bond, skb, slave->dev);
3970 break;
3971 }
3972 }
3973
3974 out:
3975 if (res) {
3976 /* no suitable interface, frame not sent */
3977 kfree_skb(skb);
3978 }
3979
3980 return NETDEV_TX_OK;
3981 }
3982
3983
3984 /*
3985 * in active-backup mode, we know that bond->curr_active_slave is always valid if
3986 * the bond has a usable interface.
3987 */
3988 static int bond_xmit_activebackup(struct sk_buff *skb, struct net_device *bond_dev)
3989 {
3990 struct bonding *bond = netdev_priv(bond_dev);
3991 int res = 1;
3992
3993 read_lock(&bond->curr_slave_lock);
3994
3995 if (bond->curr_active_slave)
3996 res = bond_dev_queue_xmit(bond, skb,
3997 bond->curr_active_slave->dev);
3998
3999 read_unlock(&bond->curr_slave_lock);
4000
4001 if (res)
4002 /* no suitable interface, frame not sent */
4003 kfree_skb(skb);
4004
4005 return NETDEV_TX_OK;
4006 }
4007
4008 /*
4009 * In bond_xmit_xor() , we determine the output device by using a pre-
4010 * determined xmit_hash_policy(), If the selected device is not enabled,
4011 * find the next active slave.
4012 */
4013 static int bond_xmit_xor(struct sk_buff *skb, struct net_device *bond_dev)
4014 {
4015 struct bonding *bond = netdev_priv(bond_dev);
4016 struct slave *slave, *start_at;
4017 int slave_no;
4018 int i;
4019 int res = 1;
4020
4021 slave_no = bond->xmit_hash_policy(skb, bond->slave_cnt);
4022
4023 bond_for_each_slave(bond, slave, i) {
4024 slave_no--;
4025 if (slave_no < 0)
4026 break;
4027 }
4028
4029 start_at = slave;
4030
4031 bond_for_each_slave_from(bond, slave, i, start_at) {
4032 if (IS_UP(slave->dev) &&
4033 (slave->link == BOND_LINK_UP) &&
4034 bond_is_active_slave(slave)) {
4035 res = bond_dev_queue_xmit(bond, skb, slave->dev);
4036 break;
4037 }
4038 }
4039
4040 if (res) {
4041 /* no suitable interface, frame not sent */
4042 kfree_skb(skb);
4043 }
4044
4045 return NETDEV_TX_OK;
4046 }
4047
4048 /*
4049 * in broadcast mode, we send everything to all usable interfaces.
4050 */
4051 static int bond_xmit_broadcast(struct sk_buff *skb, struct net_device *bond_dev)
4052 {
4053 struct bonding *bond = netdev_priv(bond_dev);
4054 struct slave *slave, *start_at;
4055 struct net_device *tx_dev = NULL;
4056 int i;
4057 int res = 1;
4058
4059 read_lock(&bond->curr_slave_lock);
4060 start_at = bond->curr_active_slave;
4061 read_unlock(&bond->curr_slave_lock);
4062
4063 if (!start_at)
4064 goto out;
4065
4066 bond_for_each_slave_from(bond, slave, i, start_at) {
4067 if (IS_UP(slave->dev) &&
4068 (slave->link == BOND_LINK_UP) &&
4069 bond_is_active_slave(slave)) {
4070 if (tx_dev) {
4071 struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
4072 if (!skb2) {
4073 pr_err("%s: Error: bond_xmit_broadcast(): skb_clone() failed\n",
4074 bond_dev->name);
4075 continue;
4076 }
4077
4078 res = bond_dev_queue_xmit(bond, skb2, tx_dev);
4079 if (res) {
4080 kfree_skb(skb2);
4081 continue;
4082 }
4083 }
4084 tx_dev = slave->dev;
4085 }
4086 }
4087
4088 if (tx_dev)
4089 res = bond_dev_queue_xmit(bond, skb, tx_dev);
4090
4091 out:
4092 if (res)
4093 /* no suitable interface, frame not sent */
4094 kfree_skb(skb);
4095
4096 /* frame sent to all suitable interfaces */
4097 return NETDEV_TX_OK;
4098 }
4099
4100 /*------------------------- Device initialization ---------------------------*/
4101
4102 static void bond_set_xmit_hash_policy(struct bonding *bond)
4103 {
4104 switch (bond->params.xmit_policy) {
4105 case BOND_XMIT_POLICY_LAYER23:
4106 bond->xmit_hash_policy = bond_xmit_hash_policy_l23;
4107 break;
4108 case BOND_XMIT_POLICY_LAYER34:
4109 bond->xmit_hash_policy = bond_xmit_hash_policy_l34;
4110 break;
4111 case BOND_XMIT_POLICY_LAYER2:
4112 default:
4113 bond->xmit_hash_policy = bond_xmit_hash_policy_l2;
4114 break;
4115 }
4116 }
4117
4118 /*
4119 * Lookup the slave that corresponds to a qid
4120 */
4121 static inline int bond_slave_override(struct bonding *bond,
4122 struct sk_buff *skb)
4123 {
4124 int i, res = 1;
4125 struct slave *slave = NULL;
4126 struct slave *check_slave;
4127
4128 if (!skb->queue_mapping)
4129 return 1;
4130
4131 /* Find out if any slaves have the same mapping as this skb. */
4132 bond_for_each_slave(bond, check_slave, i) {
4133 if (check_slave->queue_id == skb->queue_mapping) {
4134 slave = check_slave;
4135 break;
4136 }
4137 }
4138
4139 /* If the slave isn't UP, use default transmit policy. */
4140 if (slave && slave->queue_id && IS_UP(slave->dev) &&
4141 (slave->link == BOND_LINK_UP)) {
4142 res = bond_dev_queue_xmit(bond, skb, slave->dev);
4143 }
4144
4145 return res;
4146 }
4147
4148
4149 static u16 bond_select_queue(struct net_device *dev, struct sk_buff *skb)
4150 {
4151 /*
4152 * This helper function exists to help dev_pick_tx get the correct
4153 * destination queue. Using a helper function skips a call to
4154 * skb_tx_hash and will put the skbs in the queue we expect on their
4155 * way down to the bonding driver.
4156 */
4157 u16 txq = skb_rx_queue_recorded(skb) ? skb_get_rx_queue(skb) : 0;
4158
4159 /*
4160 * Save the original txq to restore before passing to the driver
4161 */
4162 qdisc_skb_cb(skb)->slave_dev_queue_mapping = skb->queue_mapping;
4163
4164 if (unlikely(txq >= dev->real_num_tx_queues)) {
4165 do {
4166 txq -= dev->real_num_tx_queues;
4167 } while (txq >= dev->real_num_tx_queues);
4168 }
4169 return txq;
4170 }
4171
4172 static netdev_tx_t __bond_start_xmit(struct sk_buff *skb, struct net_device *dev)
4173 {
4174 struct bonding *bond = netdev_priv(dev);
4175
4176 if (TX_QUEUE_OVERRIDE(bond->params.mode)) {
4177 if (!bond_slave_override(bond, skb))
4178 return NETDEV_TX_OK;
4179 }
4180
4181 switch (bond->params.mode) {
4182 case BOND_MODE_ROUNDROBIN:
4183 return bond_xmit_roundrobin(skb, dev);
4184 case BOND_MODE_ACTIVEBACKUP:
4185 return bond_xmit_activebackup(skb, dev);
4186 case BOND_MODE_XOR:
4187 return bond_xmit_xor(skb, dev);
4188 case BOND_MODE_BROADCAST:
4189 return bond_xmit_broadcast(skb, dev);
4190 case BOND_MODE_8023AD:
4191 return bond_3ad_xmit_xor(skb, dev);
4192 case BOND_MODE_ALB:
4193 case BOND_MODE_TLB:
4194 return bond_alb_xmit(skb, dev);
4195 default:
4196 /* Should never happen, mode already checked */
4197 pr_err("%s: Error: Unknown bonding mode %d\n",
4198 dev->name, bond->params.mode);
4199 WARN_ON_ONCE(1);
4200 kfree_skb(skb);
4201 return NETDEV_TX_OK;
4202 }
4203 }
4204
4205 static netdev_tx_t bond_start_xmit(struct sk_buff *skb, struct net_device *dev)
4206 {
4207 struct bonding *bond = netdev_priv(dev);
4208 netdev_tx_t ret = NETDEV_TX_OK;
4209
4210 /*
4211 * If we risk deadlock from transmitting this in the
4212 * netpoll path, tell netpoll to queue the frame for later tx
4213 */
4214 if (is_netpoll_tx_blocked(dev))
4215 return NETDEV_TX_BUSY;
4216
4217 read_lock(&bond->lock);
4218
4219 if (bond->slave_cnt)
4220 ret = __bond_start_xmit(skb, dev);
4221 else
4222 kfree_skb(skb);
4223
4224 read_unlock(&bond->lock);
4225
4226 return ret;
4227 }
4228
4229 /*
4230 * set bond mode specific net device operations
4231 */
4232 void bond_set_mode_ops(struct bonding *bond, int mode)
4233 {
4234 struct net_device *bond_dev = bond->dev;
4235
4236 switch (mode) {
4237 case BOND_MODE_ROUNDROBIN:
4238 break;
4239 case BOND_MODE_ACTIVEBACKUP:
4240 break;
4241 case BOND_MODE_XOR:
4242 bond_set_xmit_hash_policy(bond);
4243 break;
4244 case BOND_MODE_BROADCAST:
4245 break;
4246 case BOND_MODE_8023AD:
4247 bond_set_xmit_hash_policy(bond);
4248 break;
4249 case BOND_MODE_ALB:
4250 /* FALLTHRU */
4251 case BOND_MODE_TLB:
4252 break;
4253 default:
4254 /* Should never happen, mode already checked */
4255 pr_err("%s: Error: Unknown bonding mode %d\n",
4256 bond_dev->name, mode);
4257 break;
4258 }
4259 }
4260
4261 static void bond_ethtool_get_drvinfo(struct net_device *bond_dev,
4262 struct ethtool_drvinfo *drvinfo)
4263 {
4264 strlcpy(drvinfo->driver, DRV_NAME, sizeof(drvinfo->driver));
4265 strlcpy(drvinfo->version, DRV_VERSION, sizeof(drvinfo->version));
4266 snprintf(drvinfo->fw_version, sizeof(drvinfo->fw_version), "%d",
4267 BOND_ABI_VERSION);
4268 }
4269
4270 static const struct ethtool_ops bond_ethtool_ops = {
4271 .get_drvinfo = bond_ethtool_get_drvinfo,
4272 .get_link = ethtool_op_get_link,
4273 };
4274
4275 static const struct net_device_ops bond_netdev_ops = {
4276 .ndo_init = bond_init,
4277 .ndo_uninit = bond_uninit,
4278 .ndo_open = bond_open,
4279 .ndo_stop = bond_close,
4280 .ndo_start_xmit = bond_start_xmit,
4281 .ndo_select_queue = bond_select_queue,
4282 .ndo_get_stats64 = bond_get_stats,
4283 .ndo_do_ioctl = bond_do_ioctl,
4284 .ndo_change_rx_flags = bond_change_rx_flags,
4285 .ndo_set_rx_mode = bond_set_multicast_list,
4286 .ndo_change_mtu = bond_change_mtu,
4287 .ndo_set_mac_address = bond_set_mac_address,
4288 .ndo_neigh_setup = bond_neigh_setup,
4289 .ndo_vlan_rx_add_vid = bond_vlan_rx_add_vid,
4290 .ndo_vlan_rx_kill_vid = bond_vlan_rx_kill_vid,
4291 #ifdef CONFIG_NET_POLL_CONTROLLER
4292 .ndo_netpoll_setup = bond_netpoll_setup,
4293 .ndo_netpoll_cleanup = bond_netpoll_cleanup,
4294 .ndo_poll_controller = bond_poll_controller,
4295 #endif
4296 .ndo_add_slave = bond_enslave,
4297 .ndo_del_slave = bond_release,
4298 .ndo_fix_features = bond_fix_features,
4299 };
4300
4301 static const struct device_type bond_type = {
4302 .name = "bond",
4303 };
4304
4305 static void bond_destructor(struct net_device *bond_dev)
4306 {
4307 struct bonding *bond = netdev_priv(bond_dev);
4308 if (bond->wq)
4309 destroy_workqueue(bond->wq);
4310 free_netdev(bond_dev);
4311 }
4312
4313 static void bond_setup(struct net_device *bond_dev)
4314 {
4315 struct bonding *bond = netdev_priv(bond_dev);
4316
4317 /* initialize rwlocks */
4318 rwlock_init(&bond->lock);
4319 rwlock_init(&bond->curr_slave_lock);
4320
4321 bond->params = bonding_defaults;
4322
4323 /* Initialize pointers */
4324 bond->dev = bond_dev;
4325 INIT_LIST_HEAD(&bond->vlan_list);
4326
4327 /* Initialize the device entry points */
4328 ether_setup(bond_dev);
4329 bond_dev->netdev_ops = &bond_netdev_ops;
4330 bond_dev->ethtool_ops = &bond_ethtool_ops;
4331 bond_set_mode_ops(bond, bond->params.mode);
4332
4333 bond_dev->destructor = bond_destructor;
4334
4335 SET_NETDEV_DEVTYPE(bond_dev, &bond_type);
4336
4337 /* Initialize the device options */
4338 bond_dev->tx_queue_len = 0;
4339 bond_dev->flags |= IFF_MASTER|IFF_MULTICAST;
4340 bond_dev->priv_flags |= IFF_BONDING;
4341 bond_dev->priv_flags &= ~(IFF_XMIT_DST_RELEASE | IFF_TX_SKB_SHARING);
4342
4343 /* At first, we block adding VLANs. That's the only way to
4344 * prevent problems that occur when adding VLANs over an
4345 * empty bond. The block will be removed once non-challenged
4346 * slaves are enslaved.
4347 */
4348 bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
4349
4350 /* don't acquire bond device's netif_tx_lock when
4351 * transmitting */
4352 bond_dev->features |= NETIF_F_LLTX;
4353
4354 /* By default, we declare the bond to be fully
4355 * VLAN hardware accelerated capable. Special
4356 * care is taken in the various xmit functions
4357 * when there are slaves that are not hw accel
4358 * capable
4359 */
4360
4361 bond_dev->hw_features = BOND_VLAN_FEATURES |
4362 NETIF_F_HW_VLAN_TX |
4363 NETIF_F_HW_VLAN_RX |
4364 NETIF_F_HW_VLAN_FILTER;
4365
4366 bond_dev->hw_features &= ~(NETIF_F_ALL_CSUM & ~NETIF_F_HW_CSUM);
4367 bond_dev->features |= bond_dev->hw_features;
4368 }
4369
4370 /*
4371 * Destroy a bonding device.
4372 * Must be under rtnl_lock when this function is called.
4373 */
4374 static void bond_uninit(struct net_device *bond_dev)
4375 {
4376 struct bonding *bond = netdev_priv(bond_dev);
4377 struct vlan_entry *vlan, *tmp;
4378
4379 bond_netpoll_cleanup(bond_dev);
4380
4381 /* Release the bonded slaves */
4382 while (bond->first_slave != NULL)
4383 __bond_release_one(bond_dev, bond->first_slave->dev, true);
4384 pr_info("%s: released all slaves\n", bond_dev->name);
4385
4386 list_del(&bond->bond_list);
4387
4388 bond_debug_unregister(bond);
4389
4390 __hw_addr_flush(&bond->mc_list);
4391
4392 list_for_each_entry_safe(vlan, tmp, &bond->vlan_list, vlan_list) {
4393 list_del(&vlan->vlan_list);
4394 kfree(vlan);
4395 }
4396 }
4397
4398 /*------------------------- Module initialization ---------------------------*/
4399
4400 /*
4401 * Convert string input module parms. Accept either the
4402 * number of the mode or its string name. A bit complicated because
4403 * some mode names are substrings of other names, and calls from sysfs
4404 * may have whitespace in the name (trailing newlines, for example).
4405 */
4406 int bond_parse_parm(const char *buf, const struct bond_parm_tbl *tbl)
4407 {
4408 int modeint = -1, i, rv;
4409 char *p, modestr[BOND_MAX_MODENAME_LEN + 1] = { 0, };
4410
4411 for (p = (char *)buf; *p; p++)
4412 if (!(isdigit(*p) || isspace(*p)))
4413 break;
4414
4415 if (*p)
4416 rv = sscanf(buf, "%20s", modestr);
4417 else
4418 rv = sscanf(buf, "%d", &modeint);
4419
4420 if (!rv)
4421 return -1;
4422
4423 for (i = 0; tbl[i].modename; i++) {
4424 if (modeint == tbl[i].mode)
4425 return tbl[i].mode;
4426 if (strcmp(modestr, tbl[i].modename) == 0)
4427 return tbl[i].mode;
4428 }
4429
4430 return -1;
4431 }
4432
4433 static int bond_check_params(struct bond_params *params)
4434 {
4435 int arp_validate_value, fail_over_mac_value, primary_reselect_value;
4436
4437 /*
4438 * Convert string parameters.
4439 */
4440 if (mode) {
4441 bond_mode = bond_parse_parm(mode, bond_mode_tbl);
4442 if (bond_mode == -1) {
4443 pr_err("Error: Invalid bonding mode \"%s\"\n",
4444 mode == NULL ? "NULL" : mode);
4445 return -EINVAL;
4446 }
4447 }
4448
4449 if (xmit_hash_policy) {
4450 if ((bond_mode != BOND_MODE_XOR) &&
4451 (bond_mode != BOND_MODE_8023AD)) {
4452 pr_info("xmit_hash_policy param is irrelevant in mode %s\n",
4453 bond_mode_name(bond_mode));
4454 } else {
4455 xmit_hashtype = bond_parse_parm(xmit_hash_policy,
4456 xmit_hashtype_tbl);
4457 if (xmit_hashtype == -1) {
4458 pr_err("Error: Invalid xmit_hash_policy \"%s\"\n",
4459 xmit_hash_policy == NULL ? "NULL" :
4460 xmit_hash_policy);
4461 return -EINVAL;
4462 }
4463 }
4464 }
4465
4466 if (lacp_rate) {
4467 if (bond_mode != BOND_MODE_8023AD) {
4468 pr_info("lacp_rate param is irrelevant in mode %s\n",
4469 bond_mode_name(bond_mode));
4470 } else {
4471 lacp_fast = bond_parse_parm(lacp_rate, bond_lacp_tbl);
4472 if (lacp_fast == -1) {
4473 pr_err("Error: Invalid lacp rate \"%s\"\n",
4474 lacp_rate == NULL ? "NULL" : lacp_rate);
4475 return -EINVAL;
4476 }
4477 }
4478 }
4479
4480 if (ad_select) {
4481 params->ad_select = bond_parse_parm(ad_select, ad_select_tbl);
4482 if (params->ad_select == -1) {
4483 pr_err("Error: Invalid ad_select \"%s\"\n",
4484 ad_select == NULL ? "NULL" : ad_select);
4485 return -EINVAL;
4486 }
4487
4488 if (bond_mode != BOND_MODE_8023AD) {
4489 pr_warning("ad_select param only affects 802.3ad mode\n");
4490 }
4491 } else {
4492 params->ad_select = BOND_AD_STABLE;
4493 }
4494
4495 if (max_bonds < 0) {
4496 pr_warning("Warning: max_bonds (%d) not in range %d-%d, so it was reset to BOND_DEFAULT_MAX_BONDS (%d)\n",
4497 max_bonds, 0, INT_MAX, BOND_DEFAULT_MAX_BONDS);
4498 max_bonds = BOND_DEFAULT_MAX_BONDS;
4499 }
4500
4501 if (miimon < 0) {
4502 pr_warning("Warning: miimon module parameter (%d), not in range 0-%d, so it was reset to %d\n",
4503 miimon, INT_MAX, BOND_LINK_MON_INTERV);
4504 miimon = BOND_LINK_MON_INTERV;
4505 }
4506
4507 if (updelay < 0) {
4508 pr_warning("Warning: updelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4509 updelay, INT_MAX);
4510 updelay = 0;
4511 }
4512
4513 if (downdelay < 0) {
4514 pr_warning("Warning: downdelay module parameter (%d), not in range 0-%d, so it was reset to 0\n",
4515 downdelay, INT_MAX);
4516 downdelay = 0;
4517 }
4518
4519 if ((use_carrier != 0) && (use_carrier != 1)) {
4520 pr_warning("Warning: use_carrier module parameter (%d), not of valid value (0/1), so it was set to 1\n",
4521 use_carrier);
4522 use_carrier = 1;
4523 }
4524
4525 if (num_peer_notif < 0 || num_peer_notif > 255) {
4526 pr_warning("Warning: num_grat_arp/num_unsol_na (%d) not in range 0-255 so it was reset to 1\n",
4527 num_peer_notif);
4528 num_peer_notif = 1;
4529 }
4530
4531 /* reset values for 802.3ad */
4532 if (bond_mode == BOND_MODE_8023AD) {
4533 if (!miimon) {
4534 pr_warning("Warning: miimon must be specified, otherwise bonding will not detect link failure, speed and duplex which are essential for 802.3ad operation\n");
4535 pr_warning("Forcing miimon to 100msec\n");
4536 miimon = 100;
4537 }
4538 }
4539
4540 if (tx_queues < 1 || tx_queues > 255) {
4541 pr_warning("Warning: tx_queues (%d) should be between "
4542 "1 and 255, resetting to %d\n",
4543 tx_queues, BOND_DEFAULT_TX_QUEUES);
4544 tx_queues = BOND_DEFAULT_TX_QUEUES;
4545 }
4546
4547 if ((all_slaves_active != 0) && (all_slaves_active != 1)) {
4548 pr_warning("Warning: all_slaves_active module parameter (%d), "
4549 "not of valid value (0/1), so it was set to "
4550 "0\n", all_slaves_active);
4551 all_slaves_active = 0;
4552 }
4553
4554 if (resend_igmp < 0 || resend_igmp > 255) {
4555 pr_warning("Warning: resend_igmp (%d) should be between "
4556 "0 and 255, resetting to %d\n",
4557 resend_igmp, BOND_DEFAULT_RESEND_IGMP);
4558 resend_igmp = BOND_DEFAULT_RESEND_IGMP;
4559 }
4560
4561 /* reset values for TLB/ALB */
4562 if ((bond_mode == BOND_MODE_TLB) ||
4563 (bond_mode == BOND_MODE_ALB)) {
4564 if (!miimon) {
4565 pr_warning("Warning: miimon must be specified, otherwise bonding will not detect link failure and link speed which are essential for TLB/ALB load balancing\n");
4566 pr_warning("Forcing miimon to 100msec\n");
4567 miimon = 100;
4568 }
4569 }
4570
4571 if (bond_mode == BOND_MODE_ALB) {
4572 pr_notice("In ALB mode you might experience client disconnections upon reconnection of a link if the bonding module updelay parameter (%d msec) is incompatible with the forwarding delay time of the switch\n",
4573 updelay);
4574 }
4575
4576 if (!miimon) {
4577 if (updelay || downdelay) {
4578 /* just warn the user the up/down delay will have
4579 * no effect since miimon is zero...
4580 */
4581 pr_warning("Warning: miimon module parameter not set and updelay (%d) or downdelay (%d) module parameter is set; updelay and downdelay have no effect unless miimon is set\n",
4582 updelay, downdelay);
4583 }
4584 } else {
4585 /* don't allow arp monitoring */
4586 if (arp_interval) {
4587 pr_warning("Warning: miimon (%d) and arp_interval (%d) can't be used simultaneously, disabling ARP monitoring\n",
4588 miimon, arp_interval);
4589 arp_interval = 0;
4590 }
4591
4592 if ((updelay % miimon) != 0) {
4593 pr_warning("Warning: updelay (%d) is not a multiple of miimon (%d), updelay rounded to %d ms\n",
4594 updelay, miimon,
4595 (updelay / miimon) * miimon);
4596 }
4597
4598 updelay /= miimon;
4599
4600 if ((downdelay % miimon) != 0) {
4601 pr_warning("Warning: downdelay (%d) is not a multiple of miimon (%d), downdelay rounded to %d ms\n",
4602 downdelay, miimon,
4603 (downdelay / miimon) * miimon);
4604 }
4605
4606 downdelay /= miimon;
4607 }
4608
4609 if (arp_interval < 0) {
4610 pr_warning("Warning: arp_interval module parameter (%d) , not in range 0-%d, so it was reset to %d\n",
4611 arp_interval, INT_MAX, BOND_LINK_ARP_INTERV);
4612 arp_interval = BOND_LINK_ARP_INTERV;
4613 }
4614
4615 for (arp_ip_count = 0;
4616 (arp_ip_count < BOND_MAX_ARP_TARGETS) && arp_ip_target[arp_ip_count];
4617 arp_ip_count++) {
4618 /* not complete check, but should be good enough to
4619 catch mistakes */
4620 __be32 ip = in_aton(arp_ip_target[arp_ip_count]);
4621 if (!isdigit(arp_ip_target[arp_ip_count][0]) ||
4622 ip == 0 || ip == htonl(INADDR_BROADCAST)) {
4623 pr_warning("Warning: bad arp_ip_target module parameter (%s), ARP monitoring will not be performed\n",
4624 arp_ip_target[arp_ip_count]);
4625 arp_interval = 0;
4626 } else {
4627 arp_target[arp_ip_count] = ip;
4628 }
4629 }
4630
4631 if (arp_interval && !arp_ip_count) {
4632 /* don't allow arping if no arp_ip_target given... */
4633 pr_warning("Warning: arp_interval module parameter (%d) specified without providing an arp_ip_target parameter, arp_interval was reset to 0\n",
4634 arp_interval);
4635 arp_interval = 0;
4636 }
4637
4638 if (arp_validate) {
4639 if (bond_mode != BOND_MODE_ACTIVEBACKUP) {
4640 pr_err("arp_validate only supported in active-backup mode\n");
4641 return -EINVAL;
4642 }
4643 if (!arp_interval) {
4644 pr_err("arp_validate requires arp_interval\n");
4645 return -EINVAL;
4646 }
4647
4648 arp_validate_value = bond_parse_parm(arp_validate,
4649 arp_validate_tbl);
4650 if (arp_validate_value == -1) {
4651 pr_err("Error: invalid arp_validate \"%s\"\n",
4652 arp_validate == NULL ? "NULL" : arp_validate);
4653 return -EINVAL;
4654 }
4655 } else
4656 arp_validate_value = 0;
4657
4658 if (miimon) {
4659 pr_info("MII link monitoring set to %d ms\n", miimon);
4660 } else if (arp_interval) {
4661 int i;
4662
4663 pr_info("ARP monitoring set to %d ms, validate %s, with %d target(s):",
4664 arp_interval,
4665 arp_validate_tbl[arp_validate_value].modename,
4666 arp_ip_count);
4667
4668 for (i = 0; i < arp_ip_count; i++)
4669 pr_info(" %s", arp_ip_target[i]);
4670
4671 pr_info("\n");
4672
4673 } else if (max_bonds) {
4674 /* miimon and arp_interval not set, we need one so things
4675 * work as expected, see bonding.txt for details
4676 */
4677 pr_debug("Warning: either miimon or arp_interval and arp_ip_target module parameters must be specified, otherwise bonding will not detect link failures! see bonding.txt for details.\n");
4678 }
4679
4680 if (primary && !USES_PRIMARY(bond_mode)) {
4681 /* currently, using a primary only makes sense
4682 * in active backup, TLB or ALB modes
4683 */
4684 pr_warning("Warning: %s primary device specified but has no effect in %s mode\n",
4685 primary, bond_mode_name(bond_mode));
4686 primary = NULL;
4687 }
4688
4689 if (primary && primary_reselect) {
4690 primary_reselect_value = bond_parse_parm(primary_reselect,
4691 pri_reselect_tbl);
4692 if (primary_reselect_value == -1) {
4693 pr_err("Error: Invalid primary_reselect \"%s\"\n",
4694 primary_reselect ==
4695 NULL ? "NULL" : primary_reselect);
4696 return -EINVAL;
4697 }
4698 } else {
4699 primary_reselect_value = BOND_PRI_RESELECT_ALWAYS;
4700 }
4701
4702 if (fail_over_mac) {
4703 fail_over_mac_value = bond_parse_parm(fail_over_mac,
4704 fail_over_mac_tbl);
4705 if (fail_over_mac_value == -1) {
4706 pr_err("Error: invalid fail_over_mac \"%s\"\n",
4707 arp_validate == NULL ? "NULL" : arp_validate);
4708 return -EINVAL;
4709 }
4710
4711 if (bond_mode != BOND_MODE_ACTIVEBACKUP)
4712 pr_warning("Warning: fail_over_mac only affects active-backup mode.\n");
4713 } else {
4714 fail_over_mac_value = BOND_FOM_NONE;
4715 }
4716
4717 /* fill params struct with the proper values */
4718 params->mode = bond_mode;
4719 params->xmit_policy = xmit_hashtype;
4720 params->miimon = miimon;
4721 params->num_peer_notif = num_peer_notif;
4722 params->arp_interval = arp_interval;
4723 params->arp_validate = arp_validate_value;
4724 params->updelay = updelay;
4725 params->downdelay = downdelay;
4726 params->use_carrier = use_carrier;
4727 params->lacp_fast = lacp_fast;
4728 params->primary[0] = 0;
4729 params->primary_reselect = primary_reselect_value;
4730 params->fail_over_mac = fail_over_mac_value;
4731 params->tx_queues = tx_queues;
4732 params->all_slaves_active = all_slaves_active;
4733 params->resend_igmp = resend_igmp;
4734 params->min_links = min_links;
4735
4736 if (primary) {
4737 strncpy(params->primary, primary, IFNAMSIZ);
4738 params->primary[IFNAMSIZ - 1] = 0;
4739 }
4740
4741 memcpy(params->arp_targets, arp_target, sizeof(arp_target));
4742
4743 return 0;
4744 }
4745
4746 static struct lock_class_key bonding_netdev_xmit_lock_key;
4747 static struct lock_class_key bonding_netdev_addr_lock_key;
4748 static struct lock_class_key bonding_tx_busylock_key;
4749
4750 static void bond_set_lockdep_class_one(struct net_device *dev,
4751 struct netdev_queue *txq,
4752 void *_unused)
4753 {
4754 lockdep_set_class(&txq->_xmit_lock,
4755 &bonding_netdev_xmit_lock_key);
4756 }
4757
4758 static void bond_set_lockdep_class(struct net_device *dev)
4759 {
4760 lockdep_set_class(&dev->addr_list_lock,
4761 &bonding_netdev_addr_lock_key);
4762 netdev_for_each_tx_queue(dev, bond_set_lockdep_class_one, NULL);
4763 dev->qdisc_tx_busylock = &bonding_tx_busylock_key;
4764 }
4765
4766 /*
4767 * Called from registration process
4768 */
4769 static int bond_init(struct net_device *bond_dev)
4770 {
4771 struct bonding *bond = netdev_priv(bond_dev);
4772 struct bond_net *bn = net_generic(dev_net(bond_dev), bond_net_id);
4773 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
4774
4775 pr_debug("Begin bond_init for %s\n", bond_dev->name);
4776
4777 /*
4778 * Initialize locks that may be required during
4779 * en/deslave operations. All of the bond_open work
4780 * (of which this is part) should really be moved to
4781 * a phase prior to dev_open
4782 */
4783 spin_lock_init(&(bond_info->tx_hashtbl_lock));
4784 spin_lock_init(&(bond_info->rx_hashtbl_lock));
4785
4786 bond->wq = create_singlethread_workqueue(bond_dev->name);
4787 if (!bond->wq)
4788 return -ENOMEM;
4789
4790 bond_set_lockdep_class(bond_dev);
4791
4792 list_add_tail(&bond->bond_list, &bn->dev_list);
4793
4794 bond_prepare_sysfs_group(bond);
4795
4796 bond_debug_register(bond);
4797
4798 /* Ensure valid dev_addr */
4799 if (is_zero_ether_addr(bond_dev->dev_addr) &&
4800 bond_dev->addr_assign_type == NET_ADDR_PERM) {
4801 eth_hw_addr_random(bond_dev);
4802 bond->dev_addr_from_first = true;
4803 }
4804
4805 __hw_addr_init(&bond->mc_list);
4806 return 0;
4807 }
4808
4809 static int bond_validate(struct nlattr *tb[], struct nlattr *data[])
4810 {
4811 if (tb[IFLA_ADDRESS]) {
4812 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN)
4813 return -EINVAL;
4814 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS])))
4815 return -EADDRNOTAVAIL;
4816 }
4817 return 0;
4818 }
4819
4820 static unsigned int bond_get_num_tx_queues(void)
4821 {
4822 return tx_queues;
4823 }
4824
4825 static struct rtnl_link_ops bond_link_ops __read_mostly = {
4826 .kind = "bond",
4827 .priv_size = sizeof(struct bonding),
4828 .setup = bond_setup,
4829 .validate = bond_validate,
4830 .get_num_tx_queues = bond_get_num_tx_queues,
4831 .get_num_rx_queues = bond_get_num_tx_queues, /* Use the same number
4832 as for TX queues */
4833 };
4834
4835 /* Create a new bond based on the specified name and bonding parameters.
4836 * If name is NULL, obtain a suitable "bond%d" name for us.
4837 * Caller must NOT hold rtnl_lock; we need to release it here before we
4838 * set up our sysfs entries.
4839 */
4840 int bond_create(struct net *net, const char *name)
4841 {
4842 struct net_device *bond_dev;
4843 int res;
4844
4845 rtnl_lock();
4846
4847 bond_dev = alloc_netdev_mq(sizeof(struct bonding),
4848 name ? name : "bond%d",
4849 bond_setup, tx_queues);
4850 if (!bond_dev) {
4851 pr_err("%s: eek! can't alloc netdev!\n", name);
4852 rtnl_unlock();
4853 return -ENOMEM;
4854 }
4855
4856 dev_net_set(bond_dev, net);
4857 bond_dev->rtnl_link_ops = &bond_link_ops;
4858
4859 res = register_netdevice(bond_dev);
4860
4861 netif_carrier_off(bond_dev);
4862
4863 rtnl_unlock();
4864 if (res < 0)
4865 bond_destructor(bond_dev);
4866 return res;
4867 }
4868
4869 static int __net_init bond_net_init(struct net *net)
4870 {
4871 struct bond_net *bn = net_generic(net, bond_net_id);
4872
4873 bn->net = net;
4874 INIT_LIST_HEAD(&bn->dev_list);
4875
4876 bond_create_proc_dir(bn);
4877 bond_create_sysfs(bn);
4878
4879 return 0;
4880 }
4881
4882 static void __net_exit bond_net_exit(struct net *net)
4883 {
4884 struct bond_net *bn = net_generic(net, bond_net_id);
4885 struct bonding *bond, *tmp_bond;
4886 LIST_HEAD(list);
4887
4888 bond_destroy_sysfs(bn);
4889 bond_destroy_proc_dir(bn);
4890
4891 /* Kill off any bonds created after unregistering bond rtnl ops */
4892 rtnl_lock();
4893 list_for_each_entry_safe(bond, tmp_bond, &bn->dev_list, bond_list)
4894 unregister_netdevice_queue(bond->dev, &list);
4895 unregister_netdevice_many(&list);
4896 rtnl_unlock();
4897 }
4898
4899 static struct pernet_operations bond_net_ops = {
4900 .init = bond_net_init,
4901 .exit = bond_net_exit,
4902 .id = &bond_net_id,
4903 .size = sizeof(struct bond_net),
4904 };
4905
4906 static int __init bonding_init(void)
4907 {
4908 int i;
4909 int res;
4910
4911 pr_info("%s", bond_version);
4912
4913 res = bond_check_params(&bonding_defaults);
4914 if (res)
4915 goto out;
4916
4917 res = register_pernet_subsys(&bond_net_ops);
4918 if (res)
4919 goto out;
4920
4921 res = rtnl_link_register(&bond_link_ops);
4922 if (res)
4923 goto err_link;
4924
4925 bond_create_debugfs();
4926
4927 for (i = 0; i < max_bonds; i++) {
4928 res = bond_create(&init_net, NULL);
4929 if (res)
4930 goto err;
4931 }
4932
4933 register_netdevice_notifier(&bond_netdev_notifier);
4934 out:
4935 return res;
4936 err:
4937 rtnl_link_unregister(&bond_link_ops);
4938 err_link:
4939 unregister_pernet_subsys(&bond_net_ops);
4940 goto out;
4941
4942 }
4943
4944 static void __exit bonding_exit(void)
4945 {
4946 unregister_netdevice_notifier(&bond_netdev_notifier);
4947
4948 bond_destroy_debugfs();
4949
4950 rtnl_link_unregister(&bond_link_ops);
4951 unregister_pernet_subsys(&bond_net_ops);
4952
4953 #ifdef CONFIG_NET_POLL_CONTROLLER
4954 /*
4955 * Make sure we don't have an imbalance on our netpoll blocking
4956 */
4957 WARN_ON(atomic_read(&netpoll_block_tx));
4958 #endif
4959 }
4960
4961 module_init(bonding_init);
4962 module_exit(bonding_exit);
4963 MODULE_LICENSE("GPL");
4964 MODULE_VERSION(DRV_VERSION);
4965 MODULE_DESCRIPTION(DRV_DESCRIPTION ", v" DRV_VERSION);
4966 MODULE_AUTHOR("Thomas Davis, tadavis@lbl.gov and many others");
4967 MODULE_ALIAS_RTNL_LINK("bond");
kernel source for telekom puls (alcatel/mediatek)