1// SPDX-License-Identifier: GPL-2.0-or-later
2/*
3 * Copyright(c) 1999 - 2004 Intel Corporation. All rights reserved.
4 */
5
6#include <linux/skbuff.h>
7#include <linux/netdevice.h>
8#include <linux/etherdevice.h>
9#include <linux/pkt_sched.h>
10#include <linux/spinlock.h>
11#include <linux/slab.h>
12#include <linux/timer.h>
13#include <linux/ip.h>
14#include <linux/ipv6.h>
15#include <linux/if_arp.h>
16#include <linux/if_ether.h>
17#include <linux/if_bonding.h>
18#include <linux/if_vlan.h>
19#include <linux/in.h>
20#include <net/arp.h>
21#include <net/ipv6.h>
22#include <net/ndisc.h>
23#include <asm/byteorder.h>
24#include <net/bonding.h>
25#include <net/bond_alb.h>
26
27static const u8 mac_v6_allmcast[ETH_ALEN + 2] __long_aligned = {
28 0x33, 0x33, 0x00, 0x00, 0x00, 0x01
29};
30static const int alb_delta_in_ticks = HZ / ALB_TIMER_TICKS_PER_SEC;
31
32#pragma pack(1)
33struct learning_pkt {
34 u8 mac_dst[ETH_ALEN];
35 u8 mac_src[ETH_ALEN];
36 __be16 type;
37 u8 padding[ETH_ZLEN - ETH_HLEN];
38};
39
40struct arp_pkt {
41 __be16 hw_addr_space;
42 __be16 prot_addr_space;
43 u8 hw_addr_len;
44 u8 prot_addr_len;
45 __be16 op_code;
46 u8 mac_src[ETH_ALEN]; /* sender hardware address */
47 __be32 ip_src; /* sender IP address */
48 u8 mac_dst[ETH_ALEN]; /* target hardware address */
49 __be32 ip_dst; /* target IP address */
50};
51#pragma pack()
52
53/* Forward declaration */
54static void alb_send_learning_packets(struct slave *slave, const u8 mac_addr[],
55 bool strict_match);
56static void rlb_purge_src_ip(struct bonding *bond, struct arp_pkt *arp);
57static void rlb_src_unlink(struct bonding *bond, u32 index);
58static void rlb_src_link(struct bonding *bond, u32 ip_src_hash,
59 u32 ip_dst_hash);
60
61static inline u8 _simple_hash(const u8 *hash_start, int hash_size)
62{
63 int i;
64 u8 hash = 0;
65
66 for (i = 0; i < hash_size; i++)
67 hash ^= hash_start[i];
68
69 return hash;
70}
71
72/*********************** tlb specific functions ***************************/
73
74static inline void tlb_init_table_entry(struct tlb_client_info *entry, int save_load)
75{
76 if (save_load) {
77 entry->load_history = 1 + entry->tx_bytes /
78 BOND_TLB_REBALANCE_INTERVAL;
79 entry->tx_bytes = 0;
80 }
81
82 entry->tx_slave = NULL;
83 entry->next = TLB_NULL_INDEX;
84 entry->prev = TLB_NULL_INDEX;
85}
86
87static inline void tlb_init_slave(struct slave *slave)
88{
89 SLAVE_TLB_INFO(slave).load = 0;
90 SLAVE_TLB_INFO(slave).head = TLB_NULL_INDEX;
91}
92
93static void __tlb_clear_slave(struct bonding *bond, struct slave *slave,
94 int save_load)
95{
96 struct tlb_client_info *tx_hash_table;
97 u32 index;
98
99 /* clear slave from tx_hashtbl */
100 tx_hash_table = BOND_ALB_INFO(bond).tx_hashtbl;
101
102 /* skip this if we've already freed the tx hash table */
103 if (tx_hash_table) {
104 index = SLAVE_TLB_INFO(slave).head;
105 while (index != TLB_NULL_INDEX) {
106 u32 next_index = tx_hash_table[index].next;
107
108 tlb_init_table_entry(entry: &tx_hash_table[index], save_load);
109 index = next_index;
110 }
111 }
112
113 tlb_init_slave(slave);
114}
115
116static void tlb_clear_slave(struct bonding *bond, struct slave *slave,
117 int save_load)
118{
119 spin_lock_bh(lock: &bond->mode_lock);
120 __tlb_clear_slave(bond, slave, save_load);
121 spin_unlock_bh(lock: &bond->mode_lock);
122}
123
124/* Must be called before starting the monitor timer */
125static int tlb_initialize(struct bonding *bond)
126{
127 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
128 int size = TLB_HASH_TABLE_SIZE * sizeof(struct tlb_client_info);
129 struct tlb_client_info *new_hashtbl;
130 int i;
131
132 new_hashtbl = kzalloc(size, GFP_KERNEL);
133 if (!new_hashtbl)
134 return -ENOMEM;
135
136 spin_lock_bh(lock: &bond->mode_lock);
137
138 bond_info->tx_hashtbl = new_hashtbl;
139
140 for (i = 0; i < TLB_HASH_TABLE_SIZE; i++)
141 tlb_init_table_entry(entry: &bond_info->tx_hashtbl[i], save_load: 0);
142
143 spin_unlock_bh(lock: &bond->mode_lock);
144
145 return 0;
146}
147
148/* Must be called only after all slaves have been released */
149static void tlb_deinitialize(struct bonding *bond)
150{
151 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
152
153 spin_lock_bh(lock: &bond->mode_lock);
154
155 kfree(objp: bond_info->tx_hashtbl);
156 bond_info->tx_hashtbl = NULL;
157
158 spin_unlock_bh(lock: &bond->mode_lock);
159}
160
161static long long compute_gap(struct slave *slave)
162{
163 return (s64) (slave->speed << 20) - /* Convert to Megabit per sec */
164 (s64) (SLAVE_TLB_INFO(slave).load << 3); /* Bytes to bits */
165}
166
167static struct slave *tlb_get_least_loaded_slave(struct bonding *bond)
168{
169 struct slave *slave, *least_loaded;
170 struct list_head *iter;
171 long long max_gap;
172
173 least_loaded = NULL;
174 max_gap = LLONG_MIN;
175
176 /* Find the slave with the largest gap */
177 bond_for_each_slave_rcu(bond, slave, iter) {
178 if (bond_slave_can_tx(slave)) {
179 long long gap = compute_gap(slave);
180
181 if (max_gap < gap) {
182 least_loaded = slave;
183 max_gap = gap;
184 }
185 }
186 }
187
188 return least_loaded;
189}
190
191static struct slave *__tlb_choose_channel(struct bonding *bond, u32 hash_index,
192 u32 skb_len)
193{
194 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
195 struct tlb_client_info *hash_table;
196 struct slave *assigned_slave;
197
198 hash_table = bond_info->tx_hashtbl;
199 assigned_slave = hash_table[hash_index].tx_slave;
200 if (!assigned_slave) {
201 assigned_slave = tlb_get_least_loaded_slave(bond);
202
203 if (assigned_slave) {
204 struct tlb_slave_info *slave_info =
205 &(SLAVE_TLB_INFO(assigned_slave));
206 u32 next_index = slave_info->head;
207
208 hash_table[hash_index].tx_slave = assigned_slave;
209 hash_table[hash_index].next = next_index;
210 hash_table[hash_index].prev = TLB_NULL_INDEX;
211
212 if (next_index != TLB_NULL_INDEX)
213 hash_table[next_index].prev = hash_index;
214
215 slave_info->head = hash_index;
216 slave_info->load +=
217 hash_table[hash_index].load_history;
218 }
219 }
220
221 if (assigned_slave)
222 hash_table[hash_index].tx_bytes += skb_len;
223
224 return assigned_slave;
225}
226
227static struct slave *tlb_choose_channel(struct bonding *bond, u32 hash_index,
228 u32 skb_len)
229{
230 struct slave *tx_slave;
231
232 /* We don't need to disable softirq here, because
233 * tlb_choose_channel() is only called by bond_alb_xmit()
234 * which already has softirq disabled.
235 */
236 spin_lock(lock: &bond->mode_lock);
237 tx_slave = __tlb_choose_channel(bond, hash_index, skb_len);
238 spin_unlock(lock: &bond->mode_lock);
239
240 return tx_slave;
241}
242
243/*********************** rlb specific functions ***************************/
244
245/* when an ARP REPLY is received from a client update its info
246 * in the rx_hashtbl
247 */
248static void rlb_update_entry_from_arp(struct bonding *bond, struct arp_pkt *arp)
249{
250 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
251 struct rlb_client_info *client_info;
252 u32 hash_index;
253
254 spin_lock_bh(lock: &bond->mode_lock);
255
256 hash_index = _simple_hash(hash_start: (u8 *)&(arp->ip_src), hash_size: sizeof(arp->ip_src));
257 client_info = &(bond_info->rx_hashtbl[hash_index]);
258
259 if ((client_info->assigned) &&
260 (client_info->ip_src == arp->ip_dst) &&
261 (client_info->ip_dst == arp->ip_src) &&
262 (!ether_addr_equal_64bits(addr1: client_info->mac_dst, addr2: arp->mac_src))) {
263 /* update the clients MAC address */
264 ether_addr_copy(dst: client_info->mac_dst, src: arp->mac_src);
265 client_info->ntt = 1;
266 bond_info->rx_ntt = 1;
267 }
268
269 spin_unlock_bh(lock: &bond->mode_lock);
270}
271
272static int rlb_arp_recv(const struct sk_buff *skb, struct bonding *bond,
273 struct slave *slave)
274{
275 struct arp_pkt *arp, _arp;
276
277 if (skb->protocol != cpu_to_be16(ETH_P_ARP))
278 goto out;
279
280 arp = skb_header_pointer(skb, offset: 0, len: sizeof(_arp), buffer: &_arp);
281 if (!arp)
282 goto out;
283
284 /* We received an ARP from arp->ip_src.
285 * We might have used this IP address previously (on the bonding host
286 * itself or on a system that is bridged together with the bond).
287 * However, if arp->mac_src is different than what is stored in
288 * rx_hashtbl, some other host is now using the IP and we must prevent
289 * sending out client updates with this IP address and the old MAC
290 * address.
291 * Clean up all hash table entries that have this address as ip_src but
292 * have a different mac_src.
293 */
294 rlb_purge_src_ip(bond, arp);
295
296 if (arp->op_code == htons(ARPOP_REPLY)) {
297 /* update rx hash table for this ARP */
298 rlb_update_entry_from_arp(bond, arp);
299 slave_dbg(bond->dev, slave->dev, "Server received an ARP Reply from client\n");
300 }
301out:
302 return RX_HANDLER_ANOTHER;
303}
304
305/* Caller must hold rcu_read_lock() */
306static struct slave *__rlb_next_rx_slave(struct bonding *bond)
307{
308 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
309 struct slave *before = NULL, *rx_slave = NULL, *slave;
310 struct list_head *iter;
311 bool found = false;
312
313 bond_for_each_slave_rcu(bond, slave, iter) {
314 if (!bond_slave_can_tx(slave))
315 continue;
316 if (!found) {
317 if (!before || before->speed < slave->speed)
318 before = slave;
319 } else {
320 if (!rx_slave || rx_slave->speed < slave->speed)
321 rx_slave = slave;
322 }
323 if (slave == bond_info->rx_slave)
324 found = true;
325 }
326 /* we didn't find anything after the current or we have something
327 * better before and up to the current slave
328 */
329 if (!rx_slave || (before && rx_slave->speed < before->speed))
330 rx_slave = before;
331
332 if (rx_slave)
333 bond_info->rx_slave = rx_slave;
334
335 return rx_slave;
336}
337
338/* Caller must hold RTNL, rcu_read_lock is obtained only to silence checkers */
339static struct slave *rlb_next_rx_slave(struct bonding *bond)
340{
341 struct slave *rx_slave;
342
343 ASSERT_RTNL();
344
345 rcu_read_lock();
346 rx_slave = __rlb_next_rx_slave(bond);
347 rcu_read_unlock();
348
349 return rx_slave;
350}
351
352/* teach the switch the mac of a disabled slave
353 * on the primary for fault tolerance
354 *
355 * Caller must hold RTNL
356 */
357static void rlb_teach_disabled_mac_on_primary(struct bonding *bond,
358 const u8 addr[])
359{
360 struct slave *curr_active = rtnl_dereference(bond->curr_active_slave);
361
362 if (!curr_active)
363 return;
364
365 if (!bond->alb_info.primary_is_promisc) {
366 if (!dev_set_promiscuity(dev: curr_active->dev, inc: 1))
367 bond->alb_info.primary_is_promisc = 1;
368 else
369 bond->alb_info.primary_is_promisc = 0;
370 }
371
372 bond->alb_info.rlb_promisc_timeout_counter = 0;
373
374 alb_send_learning_packets(slave: curr_active, mac_addr: addr, strict_match: true);
375}
376
377/* slave being removed should not be active at this point
378 *
379 * Caller must hold rtnl.
380 */
381static void rlb_clear_slave(struct bonding *bond, struct slave *slave)
382{
383 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
384 struct rlb_client_info *rx_hash_table;
385 u32 index, next_index;
386
387 /* clear slave from rx_hashtbl */
388 spin_lock_bh(lock: &bond->mode_lock);
389
390 rx_hash_table = bond_info->rx_hashtbl;
391 index = bond_info->rx_hashtbl_used_head;
392 for (; index != RLB_NULL_INDEX; index = next_index) {
393 next_index = rx_hash_table[index].used_next;
394 if (rx_hash_table[index].slave == slave) {
395 struct slave *assigned_slave = rlb_next_rx_slave(bond);
396
397 if (assigned_slave) {
398 rx_hash_table[index].slave = assigned_slave;
399 if (is_valid_ether_addr(addr: rx_hash_table[index].mac_dst)) {
400 bond_info->rx_hashtbl[index].ntt = 1;
401 bond_info->rx_ntt = 1;
402 /* A slave has been removed from the
403 * table because it is either disabled
404 * or being released. We must retry the
405 * update to avoid clients from not
406 * being updated & disconnecting when
407 * there is stress
408 */
409 bond_info->rlb_update_retry_counter =
410 RLB_UPDATE_RETRY;
411 }
412 } else { /* there is no active slave */
413 rx_hash_table[index].slave = NULL;
414 }
415 }
416 }
417
418 spin_unlock_bh(lock: &bond->mode_lock);
419
420 if (slave != rtnl_dereference(bond->curr_active_slave))
421 rlb_teach_disabled_mac_on_primary(bond, addr: slave->dev->dev_addr);
422}
423
424static void rlb_update_client(struct rlb_client_info *client_info)
425{
426 int i;
427
428 if (!client_info->slave || !is_valid_ether_addr(addr: client_info->mac_dst))
429 return;
430
431 for (i = 0; i < RLB_ARP_BURST_SIZE; i++) {
432 struct sk_buff *skb;
433
434 skb = arp_create(ARPOP_REPLY, ETH_P_ARP,
435 dest_ip: client_info->ip_dst,
436 dev: client_info->slave->dev,
437 src_ip: client_info->ip_src,
438 dest_hw: client_info->mac_dst,
439 src_hw: client_info->slave->dev->dev_addr,
440 target_hw: client_info->mac_dst);
441 if (!skb) {
442 slave_err(client_info->slave->bond->dev,
443 client_info->slave->dev,
444 "failed to create an ARP packet\n");
445 continue;
446 }
447
448 skb->dev = client_info->slave->dev;
449
450 if (client_info->vlan_id) {
451 __vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q),
452 vlan_tci: client_info->vlan_id);
453 }
454
455 arp_xmit(skb);
456 }
457}
458
459/* sends ARP REPLIES that update the clients that need updating */
460static void rlb_update_rx_clients(struct bonding *bond)
461{
462 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
463 struct rlb_client_info *client_info;
464 u32 hash_index;
465
466 spin_lock_bh(lock: &bond->mode_lock);
467
468 hash_index = bond_info->rx_hashtbl_used_head;
469 for (; hash_index != RLB_NULL_INDEX;
470 hash_index = client_info->used_next) {
471 client_info = &(bond_info->rx_hashtbl[hash_index]);
472 if (client_info->ntt) {
473 rlb_update_client(client_info);
474 if (bond_info->rlb_update_retry_counter == 0)
475 client_info->ntt = 0;
476 }
477 }
478
479 /* do not update the entries again until this counter is zero so that
480 * not to confuse the clients.
481 */
482 bond_info->rlb_update_delay_counter = RLB_UPDATE_DELAY;
483
484 spin_unlock_bh(lock: &bond->mode_lock);
485}
486
487/* The slave was assigned a new mac address - update the clients */
488static void rlb_req_update_slave_clients(struct bonding *bond, struct slave *slave)
489{
490 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
491 struct rlb_client_info *client_info;
492 int ntt = 0;
493 u32 hash_index;
494
495 spin_lock_bh(lock: &bond->mode_lock);
496
497 hash_index = bond_info->rx_hashtbl_used_head;
498 for (; hash_index != RLB_NULL_INDEX;
499 hash_index = client_info->used_next) {
500 client_info = &(bond_info->rx_hashtbl[hash_index]);
501
502 if ((client_info->slave == slave) &&
503 is_valid_ether_addr(addr: client_info->mac_dst)) {
504 client_info->ntt = 1;
505 ntt = 1;
506 }
507 }
508
509 /* update the team's flag only after the whole iteration */
510 if (ntt) {
511 bond_info->rx_ntt = 1;
512 /* fasten the change */
513 bond_info->rlb_update_retry_counter = RLB_UPDATE_RETRY;
514 }
515
516 spin_unlock_bh(lock: &bond->mode_lock);
517}
518
519/* mark all clients using src_ip to be updated */
520static void rlb_req_update_subnet_clients(struct bonding *bond, __be32 src_ip)
521{
522 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
523 struct rlb_client_info *client_info;
524 u32 hash_index;
525
526 spin_lock(lock: &bond->mode_lock);
527
528 hash_index = bond_info->rx_hashtbl_used_head;
529 for (; hash_index != RLB_NULL_INDEX;
530 hash_index = client_info->used_next) {
531 client_info = &(bond_info->rx_hashtbl[hash_index]);
532
533 if (!client_info->slave) {
534 netdev_err(dev: bond->dev, format: "found a client with no channel in the client's hash table\n");
535 continue;
536 }
537 /* update all clients using this src_ip, that are not assigned
538 * to the team's address (curr_active_slave) and have a known
539 * unicast mac address.
540 */
541 if ((client_info->ip_src == src_ip) &&
542 !ether_addr_equal_64bits(addr1: client_info->slave->dev->dev_addr,
543 addr2: bond->dev->dev_addr) &&
544 is_valid_ether_addr(addr: client_info->mac_dst)) {
545 client_info->ntt = 1;
546 bond_info->rx_ntt = 1;
547 }
548 }
549
550 spin_unlock(lock: &bond->mode_lock);
551}
552
553static struct slave *rlb_choose_channel(struct sk_buff *skb,
554 struct bonding *bond,
555 const struct arp_pkt *arp)
556{
557 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
558 struct slave *assigned_slave, *curr_active_slave;
559 struct rlb_client_info *client_info;
560 u32 hash_index = 0;
561
562 spin_lock(lock: &bond->mode_lock);
563
564 curr_active_slave = rcu_dereference(bond->curr_active_slave);
565
566 hash_index = _simple_hash(hash_start: (u8 *)&arp->ip_dst, hash_size: sizeof(arp->ip_dst));
567 client_info = &(bond_info->rx_hashtbl[hash_index]);
568
569 if (client_info->assigned) {
570 if ((client_info->ip_src == arp->ip_src) &&
571 (client_info->ip_dst == arp->ip_dst)) {
572 /* the entry is already assigned to this client */
573 if (!is_broadcast_ether_addr(addr: arp->mac_dst)) {
574 /* update mac address from arp */
575 ether_addr_copy(dst: client_info->mac_dst, src: arp->mac_dst);
576 }
577 ether_addr_copy(dst: client_info->mac_src, src: arp->mac_src);
578
579 assigned_slave = client_info->slave;
580 if (assigned_slave) {
581 spin_unlock(lock: &bond->mode_lock);
582 return assigned_slave;
583 }
584 } else {
585 /* the entry is already assigned to some other client,
586 * move the old client to primary (curr_active_slave) so
587 * that the new client can be assigned to this entry.
588 */
589 if (curr_active_slave &&
590 client_info->slave != curr_active_slave) {
591 client_info->slave = curr_active_slave;
592 rlb_update_client(client_info);
593 }
594 }
595 }
596 /* assign a new slave */
597 assigned_slave = __rlb_next_rx_slave(bond);
598
599 if (assigned_slave) {
600 if (!(client_info->assigned &&
601 client_info->ip_src == arp->ip_src)) {
602 /* ip_src is going to be updated,
603 * fix the src hash list
604 */
605 u32 hash_src = _simple_hash(hash_start: (u8 *)&arp->ip_src,
606 hash_size: sizeof(arp->ip_src));
607 rlb_src_unlink(bond, index: hash_index);
608 rlb_src_link(bond, ip_src_hash: hash_src, ip_dst_hash: hash_index);
609 }
610
611 client_info->ip_src = arp->ip_src;
612 client_info->ip_dst = arp->ip_dst;
613 /* arp->mac_dst is broadcast for arp requests.
614 * will be updated with clients actual unicast mac address
615 * upon receiving an arp reply.
616 */
617 ether_addr_copy(dst: client_info->mac_dst, src: arp->mac_dst);
618 ether_addr_copy(dst: client_info->mac_src, src: arp->mac_src);
619 client_info->slave = assigned_slave;
620
621 if (is_valid_ether_addr(addr: client_info->mac_dst)) {
622 client_info->ntt = 1;
623 bond->alb_info.rx_ntt = 1;
624 } else {
625 client_info->ntt = 0;
626 }
627
628 if (vlan_get_tag(skb, vlan_tci: &client_info->vlan_id))
629 client_info->vlan_id = 0;
630
631 if (!client_info->assigned) {
632 u32 prev_tbl_head = bond_info->rx_hashtbl_used_head;
633
634 bond_info->rx_hashtbl_used_head = hash_index;
635 client_info->used_next = prev_tbl_head;
636 if (prev_tbl_head != RLB_NULL_INDEX) {
637 bond_info->rx_hashtbl[prev_tbl_head].used_prev =
638 hash_index;
639 }
640 client_info->assigned = 1;
641 }
642 }
643
644 spin_unlock(lock: &bond->mode_lock);
645
646 return assigned_slave;
647}
648
649/* chooses (and returns) transmit channel for arp reply
650 * does not choose channel for other arp types since they are
651 * sent on the curr_active_slave
652 */
653static struct slave *rlb_arp_xmit(struct sk_buff *skb, struct bonding *bond)
654{
655 struct slave *tx_slave = NULL;
656 struct net_device *dev;
657 struct arp_pkt *arp;
658
659 if (!pskb_network_may_pull(skb, len: sizeof(*arp)))
660 return NULL;
661 arp = (struct arp_pkt *)skb_network_header(skb);
662
663 /* Don't modify or load balance ARPs that do not originate
664 * from the bond itself or a VLAN directly above the bond.
665 */
666 if (!bond_slave_has_mac_rcu(bond, mac: arp->mac_src))
667 return NULL;
668
669 dev = ip_dev_find(net: dev_net(dev: bond->dev), addr: arp->ip_src);
670 if (dev) {
671 if (netif_is_any_bridge_master(dev)) {
672 dev_put(dev);
673 return NULL;
674 }
675 dev_put(dev);
676 }
677
678 if (arp->op_code == htons(ARPOP_REPLY)) {
679 /* the arp must be sent on the selected rx channel */
680 tx_slave = rlb_choose_channel(skb, bond, arp);
681 if (tx_slave)
682 bond_hw_addr_copy(dst: arp->mac_src, src: tx_slave->dev->dev_addr,
683 len: tx_slave->dev->addr_len);
684 netdev_dbg(bond->dev, "(slave %s): Server sent ARP Reply packet\n",
685 tx_slave ? tx_slave->dev->name : "NULL");
686 } else if (arp->op_code == htons(ARPOP_REQUEST)) {
687 /* Create an entry in the rx_hashtbl for this client as a
688 * place holder.
689 * When the arp reply is received the entry will be updated
690 * with the correct unicast address of the client.
691 */
692 tx_slave = rlb_choose_channel(skb, bond, arp);
693
694 /* The ARP reply packets must be delayed so that
695 * they can cancel out the influence of the ARP request.
696 */
697 bond->alb_info.rlb_update_delay_counter = RLB_UPDATE_DELAY;
698
699 /* arp requests are broadcast and are sent on the primary
700 * the arp request will collapse all clients on the subnet to
701 * the primary slave. We must register these clients to be
702 * updated with their assigned mac.
703 */
704 rlb_req_update_subnet_clients(bond, src_ip: arp->ip_src);
705 netdev_dbg(bond->dev, "(slave %s): Server sent ARP Request packet\n",
706 tx_slave ? tx_slave->dev->name : "NULL");
707 }
708
709 return tx_slave;
710}
711
712static void rlb_rebalance(struct bonding *bond)
713{
714 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
715 struct slave *assigned_slave;
716 struct rlb_client_info *client_info;
717 int ntt;
718 u32 hash_index;
719
720 spin_lock_bh(lock: &bond->mode_lock);
721
722 ntt = 0;
723 hash_index = bond_info->rx_hashtbl_used_head;
724 for (; hash_index != RLB_NULL_INDEX;
725 hash_index = client_info->used_next) {
726 client_info = &(bond_info->rx_hashtbl[hash_index]);
727 assigned_slave = __rlb_next_rx_slave(bond);
728 if (assigned_slave && (client_info->slave != assigned_slave)) {
729 client_info->slave = assigned_slave;
730 if (!is_zero_ether_addr(addr: client_info->mac_dst)) {
731 client_info->ntt = 1;
732 ntt = 1;
733 }
734 }
735 }
736
737 /* update the team's flag only after the whole iteration */
738 if (ntt)
739 bond_info->rx_ntt = 1;
740 spin_unlock_bh(lock: &bond->mode_lock);
741}
742
743/* Caller must hold mode_lock */
744static void rlb_init_table_entry_dst(struct rlb_client_info *entry)
745{
746 entry->used_next = RLB_NULL_INDEX;
747 entry->used_prev = RLB_NULL_INDEX;
748 entry->assigned = 0;
749 entry->slave = NULL;
750 entry->vlan_id = 0;
751}
752static void rlb_init_table_entry_src(struct rlb_client_info *entry)
753{
754 entry->src_first = RLB_NULL_INDEX;
755 entry->src_prev = RLB_NULL_INDEX;
756 entry->src_next = RLB_NULL_INDEX;
757}
758
759static void rlb_init_table_entry(struct rlb_client_info *entry)
760{
761 memset(entry, 0, sizeof(struct rlb_client_info));
762 rlb_init_table_entry_dst(entry);
763 rlb_init_table_entry_src(entry);
764}
765
766static void rlb_delete_table_entry_dst(struct bonding *bond, u32 index)
767{
768 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
769 u32 next_index = bond_info->rx_hashtbl[index].used_next;
770 u32 prev_index = bond_info->rx_hashtbl[index].used_prev;
771
772 if (index == bond_info->rx_hashtbl_used_head)
773 bond_info->rx_hashtbl_used_head = next_index;
774 if (prev_index != RLB_NULL_INDEX)
775 bond_info->rx_hashtbl[prev_index].used_next = next_index;
776 if (next_index != RLB_NULL_INDEX)
777 bond_info->rx_hashtbl[next_index].used_prev = prev_index;
778}
779
780/* unlink a rlb hash table entry from the src list */
781static void rlb_src_unlink(struct bonding *bond, u32 index)
782{
783 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
784 u32 next_index = bond_info->rx_hashtbl[index].src_next;
785 u32 prev_index = bond_info->rx_hashtbl[index].src_prev;
786
787 bond_info->rx_hashtbl[index].src_next = RLB_NULL_INDEX;
788 bond_info->rx_hashtbl[index].src_prev = RLB_NULL_INDEX;
789
790 if (next_index != RLB_NULL_INDEX)
791 bond_info->rx_hashtbl[next_index].src_prev = prev_index;
792
793 if (prev_index == RLB_NULL_INDEX)
794 return;
795
796 /* is prev_index pointing to the head of this list? */
797 if (bond_info->rx_hashtbl[prev_index].src_first == index)
798 bond_info->rx_hashtbl[prev_index].src_first = next_index;
799 else
800 bond_info->rx_hashtbl[prev_index].src_next = next_index;
801
802}
803
804static void rlb_delete_table_entry(struct bonding *bond, u32 index)
805{
806 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
807 struct rlb_client_info *entry = &(bond_info->rx_hashtbl[index]);
808
809 rlb_delete_table_entry_dst(bond, index);
810 rlb_init_table_entry_dst(entry);
811
812 rlb_src_unlink(bond, index);
813}
814
815/* add the rx_hashtbl[ip_dst_hash] entry to the list
816 * of entries with identical ip_src_hash
817 */
818static void rlb_src_link(struct bonding *bond, u32 ip_src_hash, u32 ip_dst_hash)
819{
820 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
821 u32 next;
822
823 bond_info->rx_hashtbl[ip_dst_hash].src_prev = ip_src_hash;
824 next = bond_info->rx_hashtbl[ip_src_hash].src_first;
825 bond_info->rx_hashtbl[ip_dst_hash].src_next = next;
826 if (next != RLB_NULL_INDEX)
827 bond_info->rx_hashtbl[next].src_prev = ip_dst_hash;
828 bond_info->rx_hashtbl[ip_src_hash].src_first = ip_dst_hash;
829}
830
831/* deletes all rx_hashtbl entries with arp->ip_src if their mac_src does
832 * not match arp->mac_src
833 */
834static void rlb_purge_src_ip(struct bonding *bond, struct arp_pkt *arp)
835{
836 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
837 u32 ip_src_hash = _simple_hash(hash_start: (u8 *)&(arp->ip_src), hash_size: sizeof(arp->ip_src));
838 u32 index;
839
840 spin_lock_bh(lock: &bond->mode_lock);
841
842 index = bond_info->rx_hashtbl[ip_src_hash].src_first;
843 while (index != RLB_NULL_INDEX) {
844 struct rlb_client_info *entry = &(bond_info->rx_hashtbl[index]);
845 u32 next_index = entry->src_next;
846
847 if (entry->ip_src == arp->ip_src &&
848 !ether_addr_equal_64bits(addr1: arp->mac_src, addr2: entry->mac_src))
849 rlb_delete_table_entry(bond, index);
850 index = next_index;
851 }
852 spin_unlock_bh(lock: &bond->mode_lock);
853}
854
855static int rlb_initialize(struct bonding *bond)
856{
857 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
858 struct rlb_client_info *new_hashtbl;
859 int size = RLB_HASH_TABLE_SIZE * sizeof(struct rlb_client_info);
860 int i;
861
862 new_hashtbl = kmalloc(size, GFP_KERNEL);
863 if (!new_hashtbl)
864 return -1;
865
866 spin_lock_bh(lock: &bond->mode_lock);
867
868 bond_info->rx_hashtbl = new_hashtbl;
869
870 bond_info->rx_hashtbl_used_head = RLB_NULL_INDEX;
871
872 for (i = 0; i < RLB_HASH_TABLE_SIZE; i++)
873 rlb_init_table_entry(entry: bond_info->rx_hashtbl + i);
874
875 spin_unlock_bh(lock: &bond->mode_lock);
876
877 /* register to receive ARPs */
878 bond->recv_probe = rlb_arp_recv;
879
880 return 0;
881}
882
883static void rlb_deinitialize(struct bonding *bond)
884{
885 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
886
887 spin_lock_bh(lock: &bond->mode_lock);
888
889 kfree(objp: bond_info->rx_hashtbl);
890 bond_info->rx_hashtbl = NULL;
891 bond_info->rx_hashtbl_used_head = RLB_NULL_INDEX;
892
893 spin_unlock_bh(lock: &bond->mode_lock);
894}
895
896static void rlb_clear_vlan(struct bonding *bond, unsigned short vlan_id)
897{
898 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
899 u32 curr_index;
900
901 spin_lock_bh(lock: &bond->mode_lock);
902
903 curr_index = bond_info->rx_hashtbl_used_head;
904 while (curr_index != RLB_NULL_INDEX) {
905 struct rlb_client_info *curr = &(bond_info->rx_hashtbl[curr_index]);
906 u32 next_index = bond_info->rx_hashtbl[curr_index].used_next;
907
908 if (curr->vlan_id == vlan_id)
909 rlb_delete_table_entry(bond, index: curr_index);
910
911 curr_index = next_index;
912 }
913
914 spin_unlock_bh(lock: &bond->mode_lock);
915}
916
917/*********************** tlb/rlb shared functions *********************/
918
919static void alb_send_lp_vid(struct slave *slave, const u8 mac_addr[],
920 __be16 vlan_proto, u16 vid)
921{
922 struct learning_pkt pkt;
923 struct sk_buff *skb;
924 int size = sizeof(struct learning_pkt);
925
926 memset(&pkt, 0, size);
927 ether_addr_copy(dst: pkt.mac_dst, src: mac_addr);
928 ether_addr_copy(dst: pkt.mac_src, src: mac_addr);
929 pkt.type = cpu_to_be16(ETH_P_LOOPBACK);
930
931 skb = dev_alloc_skb(length: size);
932 if (!skb)
933 return;
934
935 skb_put_data(skb, data: &pkt, len: size);
936
937 skb_reset_mac_header(skb);
938 skb->network_header = skb->mac_header + ETH_HLEN;
939 skb->protocol = pkt.type;
940 skb->priority = TC_PRIO_CONTROL;
941 skb->dev = slave->dev;
942
943 slave_dbg(slave->bond->dev, slave->dev,
944 "Send learning packet: mac %pM vlan %d\n", mac_addr, vid);
945
946 if (vid)
947 __vlan_hwaccel_put_tag(skb, vlan_proto, vlan_tci: vid);
948
949 dev_queue_xmit(skb);
950}
951
952struct alb_walk_data {
953 struct bonding *bond;
954 struct slave *slave;
955 const u8 *mac_addr;
956 bool strict_match;
957};
958
959static int alb_upper_dev_walk(struct net_device *upper,
960 struct netdev_nested_priv *priv)
961{
962 struct alb_walk_data *data = (struct alb_walk_data *)priv->data;
963 bool strict_match = data->strict_match;
964 const u8 *mac_addr = data->mac_addr;
965 struct bonding *bond = data->bond;
966 struct slave *slave = data->slave;
967 struct bond_vlan_tag *tags;
968
969 if (is_vlan_dev(dev: upper) &&
970 bond->dev->lower_level == upper->lower_level - 1) {
971 if (upper->addr_assign_type == NET_ADDR_STOLEN) {
972 alb_send_lp_vid(slave, mac_addr,
973 vlan_proto: vlan_dev_vlan_proto(dev: upper),
974 vid: vlan_dev_vlan_id(dev: upper));
975 } else {
976 alb_send_lp_vid(slave, mac_addr: upper->dev_addr,
977 vlan_proto: vlan_dev_vlan_proto(dev: upper),
978 vid: vlan_dev_vlan_id(dev: upper));
979 }
980 }
981
982 /* If this is a macvlan device, then only send updates
983 * when strict_match is turned off.
984 */
985 if (netif_is_macvlan(dev: upper) && !strict_match) {
986 tags = bond_verify_device_path(start_dev: bond->dev, end_dev: upper, level: 0);
987 if (IS_ERR_OR_NULL(ptr: tags))
988 return -ENOMEM;
989
990 alb_send_lp_vid(slave, mac_addr: upper->dev_addr,
991 vlan_proto: tags[0].vlan_proto, vid: tags[0].vlan_id);
992 kfree(objp: tags);
993 }
994
995 return 0;
996}
997
998static void alb_send_learning_packets(struct slave *slave, const u8 mac_addr[],
999 bool strict_match)
1000{
1001 struct bonding *bond = bond_get_bond_by_slave(slave);
1002 struct netdev_nested_priv priv;
1003 struct alb_walk_data data = {
1004 .strict_match = strict_match,
1005 .mac_addr = mac_addr,
1006 .slave = slave,
1007 .bond = bond,
1008 };
1009
1010 priv.data = (void *)&data;
1011 /* send untagged */
1012 alb_send_lp_vid(slave, mac_addr, vlan_proto: 0, vid: 0);
1013
1014 /* loop through all devices and see if we need to send a packet
1015 * for that device.
1016 */
1017 rcu_read_lock();
1018 netdev_walk_all_upper_dev_rcu(dev: bond->dev, fn: alb_upper_dev_walk, priv: &priv);
1019 rcu_read_unlock();
1020}
1021
1022static int alb_set_slave_mac_addr(struct slave *slave, const u8 addr[],
1023 unsigned int len)
1024{
1025 struct net_device *dev = slave->dev;
1026 struct sockaddr_storage ss;
1027
1028 if (BOND_MODE(slave->bond) == BOND_MODE_TLB) {
1029 __dev_addr_set(dev, addr, len);
1030 return 0;
1031 }
1032
1033 /* for rlb each slave must have a unique hw mac addresses so that
1034 * each slave will receive packets destined to a different mac
1035 */
1036 memcpy(ss.__data, addr, len);
1037 ss.ss_family = dev->type;
1038 if (dev_set_mac_address(dev, sa: (struct sockaddr *)&ss, NULL)) {
1039 slave_err(slave->bond->dev, dev, "dev_set_mac_address on slave failed! ALB mode requires that the base driver support setting the hw address also when the network device's interface is open\n");
1040 return -EOPNOTSUPP;
1041 }
1042 return 0;
1043}
1044
1045/* Swap MAC addresses between two slaves.
1046 *
1047 * Called with RTNL held, and no other locks.
1048 */
1049static void alb_swap_mac_addr(struct slave *slave1, struct slave *slave2)
1050{
1051 u8 tmp_mac_addr[MAX_ADDR_LEN];
1052
1053 bond_hw_addr_copy(dst: tmp_mac_addr, src: slave1->dev->dev_addr,
1054 len: slave1->dev->addr_len);
1055 alb_set_slave_mac_addr(slave: slave1, addr: slave2->dev->dev_addr,
1056 len: slave2->dev->addr_len);
1057 alb_set_slave_mac_addr(slave: slave2, addr: tmp_mac_addr,
1058 len: slave1->dev->addr_len);
1059
1060}
1061
1062/* Send learning packets after MAC address swap.
1063 *
1064 * Called with RTNL and no other locks
1065 */
1066static void alb_fasten_mac_swap(struct bonding *bond, struct slave *slave1,
1067 struct slave *slave2)
1068{
1069 int slaves_state_differ = (bond_slave_can_tx(slave: slave1) != bond_slave_can_tx(slave: slave2));
1070 struct slave *disabled_slave = NULL;
1071
1072 ASSERT_RTNL();
1073
1074 /* fasten the change in the switch */
1075 if (bond_slave_can_tx(slave: slave1)) {
1076 alb_send_learning_packets(slave: slave1, mac_addr: slave1->dev->dev_addr, strict_match: false);
1077 if (bond->alb_info.rlb_enabled) {
1078 /* inform the clients that the mac address
1079 * has changed
1080 */
1081 rlb_req_update_slave_clients(bond, slave: slave1);
1082 }
1083 } else {
1084 disabled_slave = slave1;
1085 }
1086
1087 if (bond_slave_can_tx(slave: slave2)) {
1088 alb_send_learning_packets(slave: slave2, mac_addr: slave2->dev->dev_addr, strict_match: false);
1089 if (bond->alb_info.rlb_enabled) {
1090 /* inform the clients that the mac address
1091 * has changed
1092 */
1093 rlb_req_update_slave_clients(bond, slave: slave2);
1094 }
1095 } else {
1096 disabled_slave = slave2;
1097 }
1098
1099 if (bond->alb_info.rlb_enabled && slaves_state_differ) {
1100 /* A disabled slave was assigned an active mac addr */
1101 rlb_teach_disabled_mac_on_primary(bond,
1102 addr: disabled_slave->dev->dev_addr);
1103 }
1104}
1105
1106/**
1107 * alb_change_hw_addr_on_detach
1108 * @bond: bonding we're working on
1109 * @slave: the slave that was just detached
1110 *
1111 * We assume that @slave was already detached from the slave list.
1112 *
1113 * If @slave's permanent hw address is different both from its current
1114 * address and from @bond's address, then somewhere in the bond there's
1115 * a slave that has @slave's permanet address as its current address.
1116 * We'll make sure that slave no longer uses @slave's permanent address.
1117 *
1118 * Caller must hold RTNL and no other locks
1119 */
1120static void alb_change_hw_addr_on_detach(struct bonding *bond, struct slave *slave)
1121{
1122 int perm_curr_diff;
1123 int perm_bond_diff;
1124 struct slave *found_slave;
1125
1126 perm_curr_diff = !ether_addr_equal_64bits(addr1: slave->perm_hwaddr,
1127 addr2: slave->dev->dev_addr);
1128 perm_bond_diff = !ether_addr_equal_64bits(addr1: slave->perm_hwaddr,
1129 addr2: bond->dev->dev_addr);
1130
1131 if (perm_curr_diff && perm_bond_diff) {
1132 found_slave = bond_slave_has_mac(bond, mac: slave->perm_hwaddr);
1133
1134 if (found_slave) {
1135 alb_swap_mac_addr(slave1: slave, slave2: found_slave);
1136 alb_fasten_mac_swap(bond, slave1: slave, slave2: found_slave);
1137 }
1138 }
1139}
1140
1141/**
1142 * alb_handle_addr_collision_on_attach
1143 * @bond: bonding we're working on
1144 * @slave: the slave that was just attached
1145 *
1146 * checks uniqueness of slave's mac address and handles the case the
1147 * new slave uses the bonds mac address.
1148 *
1149 * If the permanent hw address of @slave is @bond's hw address, we need to
1150 * find a different hw address to give @slave, that isn't in use by any other
1151 * slave in the bond. This address must be, of course, one of the permanent
1152 * addresses of the other slaves.
1153 *
1154 * We go over the slave list, and for each slave there we compare its
1155 * permanent hw address with the current address of all the other slaves.
1156 * If no match was found, then we've found a slave with a permanent address
1157 * that isn't used by any other slave in the bond, so we can assign it to
1158 * @slave.
1159 *
1160 * assumption: this function is called before @slave is attached to the
1161 * bond slave list.
1162 */
1163static int alb_handle_addr_collision_on_attach(struct bonding *bond, struct slave *slave)
1164{
1165 struct slave *has_bond_addr = rcu_access_pointer(bond->curr_active_slave);
1166 struct slave *tmp_slave1, *free_mac_slave = NULL;
1167 struct list_head *iter;
1168
1169 if (!bond_has_slaves(bond)) {
1170 /* this is the first slave */
1171 return 0;
1172 }
1173
1174 /* if slave's mac address differs from bond's mac address
1175 * check uniqueness of slave's mac address against the other
1176 * slaves in the bond.
1177 */
1178 if (!ether_addr_equal_64bits(addr1: slave->perm_hwaddr, addr2: bond->dev->dev_addr)) {
1179 if (!bond_slave_has_mac(bond, mac: slave->dev->dev_addr))
1180 return 0;
1181
1182 /* Try setting slave mac to bond address and fall-through
1183 * to code handling that situation below...
1184 */
1185 alb_set_slave_mac_addr(slave, addr: bond->dev->dev_addr,
1186 len: bond->dev->addr_len);
1187 }
1188
1189 /* The slave's address is equal to the address of the bond.
1190 * Search for a spare address in the bond for this slave.
1191 */
1192 bond_for_each_slave(bond, tmp_slave1, iter) {
1193 if (!bond_slave_has_mac(bond, mac: tmp_slave1->perm_hwaddr)) {
1194 /* no slave has tmp_slave1's perm addr
1195 * as its curr addr
1196 */
1197 free_mac_slave = tmp_slave1;
1198 break;
1199 }
1200
1201 if (!has_bond_addr) {
1202 if (ether_addr_equal_64bits(addr1: tmp_slave1->dev->dev_addr,
1203 addr2: bond->dev->dev_addr)) {
1204
1205 has_bond_addr = tmp_slave1;
1206 }
1207 }
1208 }
1209
1210 if (free_mac_slave) {
1211 alb_set_slave_mac_addr(slave, addr: free_mac_slave->perm_hwaddr,
1212 len: free_mac_slave->dev->addr_len);
1213
1214 slave_warn(bond->dev, slave->dev, "the slave hw address is in use by the bond; giving it the hw address of %s\n",
1215 free_mac_slave->dev->name);
1216
1217 } else if (has_bond_addr) {
1218 slave_err(bond->dev, slave->dev, "the slave hw address is in use by the bond; couldn't find a slave with a free hw address to give it (this should not have happened)\n");
1219 return -EFAULT;
1220 }
1221
1222 return 0;
1223}
1224
1225/**
1226 * alb_set_mac_address
1227 * @bond: bonding we're working on
1228 * @addr: MAC address to set
1229 *
1230 * In TLB mode all slaves are configured to the bond's hw address, but set
1231 * their dev_addr field to different addresses (based on their permanent hw
1232 * addresses).
1233 *
1234 * For each slave, this function sets the interface to the new address and then
1235 * changes its dev_addr field to its previous value.
1236 *
1237 * Unwinding assumes bond's mac address has not yet changed.
1238 */
1239static int alb_set_mac_address(struct bonding *bond, void *addr)
1240{
1241 struct slave *slave, *rollback_slave;
1242 struct list_head *iter;
1243 struct sockaddr_storage ss;
1244 char tmp_addr[MAX_ADDR_LEN];
1245 int res;
1246
1247 if (bond->alb_info.rlb_enabled)
1248 return 0;
1249
1250 bond_for_each_slave(bond, slave, iter) {
1251 /* save net_device's current hw address */
1252 bond_hw_addr_copy(dst: tmp_addr, src: slave->dev->dev_addr,
1253 len: slave->dev->addr_len);
1254
1255 res = dev_set_mac_address(dev: slave->dev, sa: addr, NULL);
1256
1257 /* restore net_device's hw address */
1258 dev_addr_set(dev: slave->dev, addr: tmp_addr);
1259
1260 if (res)
1261 goto unwind;
1262 }
1263
1264 return 0;
1265
1266unwind:
1267 memcpy(ss.__data, bond->dev->dev_addr, bond->dev->addr_len);
1268 ss.ss_family = bond->dev->type;
1269
1270 /* unwind from head to the slave that failed */
1271 bond_for_each_slave(bond, rollback_slave, iter) {
1272 if (rollback_slave == slave)
1273 break;
1274 bond_hw_addr_copy(dst: tmp_addr, src: rollback_slave->dev->dev_addr,
1275 len: rollback_slave->dev->addr_len);
1276 dev_set_mac_address(dev: rollback_slave->dev,
1277 sa: (struct sockaddr *)&ss, NULL);
1278 dev_addr_set(dev: rollback_slave->dev, addr: tmp_addr);
1279 }
1280
1281 return res;
1282}
1283
1284/* determine if the packet is NA or NS */
1285static bool alb_determine_nd(struct sk_buff *skb, struct bonding *bond)
1286{
1287 struct ipv6hdr *ip6hdr;
1288 struct icmp6hdr *hdr;
1289
1290 if (!pskb_network_may_pull(skb, len: sizeof(*ip6hdr)))
1291 return true;
1292
1293 ip6hdr = ipv6_hdr(skb);
1294 if (ip6hdr->nexthdr != IPPROTO_ICMPV6)
1295 return false;
1296
1297 if (!pskb_network_may_pull(skb, len: sizeof(*ip6hdr) + sizeof(*hdr)))
1298 return true;
1299
1300 hdr = icmp6_hdr(skb);
1301 return hdr->icmp6_type == NDISC_NEIGHBOUR_ADVERTISEMENT ||
1302 hdr->icmp6_type == NDISC_NEIGHBOUR_SOLICITATION;
1303}
1304
1305/************************ exported alb functions ************************/
1306
1307int bond_alb_initialize(struct bonding *bond, int rlb_enabled)
1308{
1309 int res;
1310
1311 res = tlb_initialize(bond);
1312 if (res)
1313 return res;
1314
1315 if (rlb_enabled) {
1316 res = rlb_initialize(bond);
1317 if (res) {
1318 tlb_deinitialize(bond);
1319 return res;
1320 }
1321 bond->alb_info.rlb_enabled = 1;
1322 } else {
1323 bond->alb_info.rlb_enabled = 0;
1324 }
1325
1326 return 0;
1327}
1328
1329void bond_alb_deinitialize(struct bonding *bond)
1330{
1331 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
1332
1333 tlb_deinitialize(bond);
1334
1335 if (bond_info->rlb_enabled)
1336 rlb_deinitialize(bond);
1337}
1338
1339static netdev_tx_t bond_do_alb_xmit(struct sk_buff *skb, struct bonding *bond,
1340 struct slave *tx_slave)
1341{
1342 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
1343 struct ethhdr *eth_data = eth_hdr(skb);
1344
1345 if (!tx_slave) {
1346 /* unbalanced or unassigned, send through primary */
1347 tx_slave = rcu_dereference(bond->curr_active_slave);
1348 if (bond->params.tlb_dynamic_lb)
1349 bond_info->unbalanced_load += skb->len;
1350 }
1351
1352 if (tx_slave && bond_slave_can_tx(slave: tx_slave)) {
1353 if (tx_slave != rcu_access_pointer(bond->curr_active_slave)) {
1354 ether_addr_copy(dst: eth_data->h_source,
1355 src: tx_slave->dev->dev_addr);
1356 }
1357
1358 return bond_dev_queue_xmit(bond, skb, slave_dev: tx_slave->dev);
1359 }
1360
1361 if (tx_slave && bond->params.tlb_dynamic_lb) {
1362 spin_lock(lock: &bond->mode_lock);
1363 __tlb_clear_slave(bond, slave: tx_slave, save_load: 0);
1364 spin_unlock(lock: &bond->mode_lock);
1365 }
1366
1367 /* no suitable interface, frame not sent */
1368 return bond_tx_drop(dev: bond->dev, skb);
1369}
1370
1371struct slave *bond_xmit_tlb_slave_get(struct bonding *bond,
1372 struct sk_buff *skb)
1373{
1374 struct slave *tx_slave = NULL;
1375 struct ethhdr *eth_data;
1376 u32 hash_index;
1377
1378 skb_reset_mac_header(skb);
1379 eth_data = eth_hdr(skb);
1380
1381 /* Do not TX balance any multicast or broadcast */
1382 if (!is_multicast_ether_addr(addr: eth_data->h_dest)) {
1383 switch (skb->protocol) {
1384 case htons(ETH_P_IPV6):
1385 if (alb_determine_nd(skb, bond))
1386 break;
1387 fallthrough;
1388 case htons(ETH_P_IP):
1389 hash_index = bond_xmit_hash(bond, skb);
1390 if (bond->params.tlb_dynamic_lb) {
1391 tx_slave = tlb_choose_channel(bond,
1392 hash_index: hash_index & 0xFF,
1393 skb_len: skb->len);
1394 } else {
1395 struct bond_up_slave *slaves;
1396 unsigned int count;
1397
1398 slaves = rcu_dereference(bond->usable_slaves);
1399 count = slaves ? READ_ONCE(slaves->count) : 0;
1400 if (likely(count))
1401 tx_slave = slaves->arr[hash_index %
1402 count];
1403 }
1404 break;
1405 }
1406 }
1407 return tx_slave;
1408}
1409
1410netdev_tx_t bond_tlb_xmit(struct sk_buff *skb, struct net_device *bond_dev)
1411{
1412 struct bonding *bond = netdev_priv(dev: bond_dev);
1413 struct slave *tx_slave;
1414
1415 tx_slave = bond_xmit_tlb_slave_get(bond, skb);
1416 return bond_do_alb_xmit(skb, bond, tx_slave);
1417}
1418
1419struct slave *bond_xmit_alb_slave_get(struct bonding *bond,
1420 struct sk_buff *skb)
1421{
1422 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
1423 static const __be32 ip_bcast = htonl(0xffffffff);
1424 struct slave *tx_slave = NULL;
1425 const u8 *hash_start = NULL;
1426 bool do_tx_balance = true;
1427 struct ethhdr *eth_data;
1428 u32 hash_index = 0;
1429 int hash_size = 0;
1430
1431 skb_reset_mac_header(skb);
1432 eth_data = eth_hdr(skb);
1433
1434 switch (ntohs(skb->protocol)) {
1435 case ETH_P_IP: {
1436 const struct iphdr *iph;
1437
1438 if (is_broadcast_ether_addr(addr: eth_data->h_dest) ||
1439 !pskb_network_may_pull(skb, len: sizeof(*iph))) {
1440 do_tx_balance = false;
1441 break;
1442 }
1443 iph = ip_hdr(skb);
1444 if (iph->daddr == ip_bcast || iph->protocol == IPPROTO_IGMP) {
1445 do_tx_balance = false;
1446 break;
1447 }
1448 hash_start = (char *)&(iph->daddr);
1449 hash_size = sizeof(iph->daddr);
1450 break;
1451 }
1452 case ETH_P_IPV6: {
1453 const struct ipv6hdr *ip6hdr;
1454
1455 /* IPv6 doesn't really use broadcast mac address, but leave
1456 * that here just in case.
1457 */
1458 if (is_broadcast_ether_addr(addr: eth_data->h_dest)) {
1459 do_tx_balance = false;
1460 break;
1461 }
1462
1463 /* IPv6 uses all-nodes multicast as an equivalent to
1464 * broadcasts in IPv4.
1465 */
1466 if (ether_addr_equal_64bits(addr1: eth_data->h_dest, addr2: mac_v6_allmcast)) {
1467 do_tx_balance = false;
1468 break;
1469 }
1470
1471 if (alb_determine_nd(skb, bond)) {
1472 do_tx_balance = false;
1473 break;
1474 }
1475
1476 /* The IPv6 header is pulled by alb_determine_nd */
1477 /* Additionally, DAD probes should not be tx-balanced as that
1478 * will lead to false positives for duplicate addresses and
1479 * prevent address configuration from working.
1480 */
1481 ip6hdr = ipv6_hdr(skb);
1482 if (ipv6_addr_any(a: &ip6hdr->saddr)) {
1483 do_tx_balance = false;
1484 break;
1485 }
1486
1487 hash_start = (char *)&ip6hdr->daddr;
1488 hash_size = sizeof(ip6hdr->daddr);
1489 break;
1490 }
1491 case ETH_P_ARP:
1492 do_tx_balance = false;
1493 if (bond_info->rlb_enabled)
1494 tx_slave = rlb_arp_xmit(skb, bond);
1495 break;
1496 default:
1497 do_tx_balance = false;
1498 break;
1499 }
1500
1501 if (do_tx_balance) {
1502 if (bond->params.tlb_dynamic_lb) {
1503 hash_index = _simple_hash(hash_start, hash_size);
1504 tx_slave = tlb_choose_channel(bond, hash_index, skb_len: skb->len);
1505 } else {
1506 /*
1507 * do_tx_balance means we are free to select the tx_slave
1508 * So we do exactly what tlb would do for hash selection
1509 */
1510
1511 struct bond_up_slave *slaves;
1512 unsigned int count;
1513
1514 slaves = rcu_dereference(bond->usable_slaves);
1515 count = slaves ? READ_ONCE(slaves->count) : 0;
1516 if (likely(count))
1517 tx_slave = slaves->arr[bond_xmit_hash(bond, skb) %
1518 count];
1519 }
1520 }
1521 return tx_slave;
1522}
1523
1524netdev_tx_t bond_alb_xmit(struct sk_buff *skb, struct net_device *bond_dev)
1525{
1526 struct bonding *bond = netdev_priv(dev: bond_dev);
1527 struct slave *tx_slave = NULL;
1528
1529 tx_slave = bond_xmit_alb_slave_get(bond, skb);
1530 return bond_do_alb_xmit(skb, bond, tx_slave);
1531}
1532
1533void bond_alb_monitor(struct work_struct *work)
1534{
1535 struct bonding *bond = container_of(work, struct bonding,
1536 alb_work.work);
1537 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
1538 struct list_head *iter;
1539 struct slave *slave;
1540
1541 if (!bond_has_slaves(bond)) {
1542 atomic_set(v: &bond_info->tx_rebalance_counter, i: 0);
1543 bond_info->lp_counter = 0;
1544 goto re_arm;
1545 }
1546
1547 rcu_read_lock();
1548
1549 atomic_inc(v: &bond_info->tx_rebalance_counter);
1550 bond_info->lp_counter++;
1551
1552 /* send learning packets */
1553 if (bond_info->lp_counter >= BOND_ALB_LP_TICKS(bond)) {
1554 bool strict_match;
1555
1556 bond_for_each_slave_rcu(bond, slave, iter) {
1557 /* If updating current_active, use all currently
1558 * user mac addresses (!strict_match). Otherwise, only
1559 * use mac of the slave device.
1560 * In RLB mode, we always use strict matches.
1561 */
1562 strict_match = (slave != rcu_access_pointer(bond->curr_active_slave) ||
1563 bond_info->rlb_enabled);
1564 alb_send_learning_packets(slave, mac_addr: slave->dev->dev_addr,
1565 strict_match);
1566 }
1567 bond_info->lp_counter = 0;
1568 }
1569
1570 /* rebalance tx traffic */
1571 if (atomic_read(v: &bond_info->tx_rebalance_counter) >= BOND_TLB_REBALANCE_TICKS) {
1572 bond_for_each_slave_rcu(bond, slave, iter) {
1573 tlb_clear_slave(bond, slave, save_load: 1);
1574 if (slave == rcu_access_pointer(bond->curr_active_slave)) {
1575 SLAVE_TLB_INFO(slave).load =
1576 bond_info->unbalanced_load /
1577 BOND_TLB_REBALANCE_INTERVAL;
1578 bond_info->unbalanced_load = 0;
1579 }
1580 }
1581 atomic_set(v: &bond_info->tx_rebalance_counter, i: 0);
1582 }
1583
1584 if (bond_info->rlb_enabled) {
1585 if (bond_info->primary_is_promisc &&
1586 (++bond_info->rlb_promisc_timeout_counter >= RLB_PROMISC_TIMEOUT)) {
1587
1588 /* dev_set_promiscuity requires rtnl and
1589 * nothing else. Avoid race with bond_close.
1590 */
1591 rcu_read_unlock();
1592 if (!rtnl_trylock())
1593 goto re_arm;
1594
1595 bond_info->rlb_promisc_timeout_counter = 0;
1596
1597 /* If the primary was set to promiscuous mode
1598 * because a slave was disabled then
1599 * it can now leave promiscuous mode.
1600 */
1601 dev_set_promiscuity(rtnl_dereference(bond->curr_active_slave)->dev,
1602 inc: -1);
1603 bond_info->primary_is_promisc = 0;
1604
1605 rtnl_unlock();
1606 rcu_read_lock();
1607 }
1608
1609 if (bond_info->rlb_rebalance) {
1610 bond_info->rlb_rebalance = 0;
1611 rlb_rebalance(bond);
1612 }
1613
1614 /* check if clients need updating */
1615 if (bond_info->rx_ntt) {
1616 if (bond_info->rlb_update_delay_counter) {
1617 --bond_info->rlb_update_delay_counter;
1618 } else {
1619 rlb_update_rx_clients(bond);
1620 if (bond_info->rlb_update_retry_counter)
1621 --bond_info->rlb_update_retry_counter;
1622 else
1623 bond_info->rx_ntt = 0;
1624 }
1625 }
1626 }
1627 rcu_read_unlock();
1628re_arm:
1629 queue_delayed_work(wq: bond->wq, dwork: &bond->alb_work, delay: alb_delta_in_ticks);
1630}
1631
1632/* assumption: called before the slave is attached to the bond
1633 * and not locked by the bond lock
1634 */
1635int bond_alb_init_slave(struct bonding *bond, struct slave *slave)
1636{
1637 int res;
1638
1639 res = alb_set_slave_mac_addr(slave, addr: slave->perm_hwaddr,
1640 len: slave->dev->addr_len);
1641 if (res)
1642 return res;
1643
1644 res = alb_handle_addr_collision_on_attach(bond, slave);
1645 if (res)
1646 return res;
1647
1648 tlb_init_slave(slave);
1649
1650 /* order a rebalance ASAP */
1651 atomic_set(v: &bond->alb_info.tx_rebalance_counter,
1652 BOND_TLB_REBALANCE_TICKS);
1653
1654 if (bond->alb_info.rlb_enabled)
1655 bond->alb_info.rlb_rebalance = 1;
1656
1657 return 0;
1658}
1659
1660/* Remove slave from tlb and rlb hash tables, and fix up MAC addresses
1661 * if necessary.
1662 *
1663 * Caller must hold RTNL and no other locks
1664 */
1665void bond_alb_deinit_slave(struct bonding *bond, struct slave *slave)
1666{
1667 if (bond_has_slaves(bond))
1668 alb_change_hw_addr_on_detach(bond, slave);
1669
1670 tlb_clear_slave(bond, slave, save_load: 0);
1671
1672 if (bond->alb_info.rlb_enabled) {
1673 bond->alb_info.rx_slave = NULL;
1674 rlb_clear_slave(bond, slave);
1675 }
1676
1677}
1678
1679void bond_alb_handle_link_change(struct bonding *bond, struct slave *slave, char link)
1680{
1681 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
1682
1683 if (link == BOND_LINK_DOWN) {
1684 tlb_clear_slave(bond, slave, save_load: 0);
1685 if (bond->alb_info.rlb_enabled)
1686 rlb_clear_slave(bond, slave);
1687 } else if (link == BOND_LINK_UP) {
1688 /* order a rebalance ASAP */
1689 atomic_set(v: &bond_info->tx_rebalance_counter,
1690 BOND_TLB_REBALANCE_TICKS);
1691 if (bond->alb_info.rlb_enabled) {
1692 bond->alb_info.rlb_rebalance = 1;
1693 /* If the updelay module parameter is smaller than the
1694 * forwarding delay of the switch the rebalance will
1695 * not work because the rebalance arp replies will
1696 * not be forwarded to the clients..
1697 */
1698 }
1699 }
1700
1701 if (bond_is_nondyn_tlb(bond)) {
1702 if (bond_update_slave_arr(bond, NULL))
1703 pr_err("Failed to build slave-array for TLB mode.\n");
1704 }
1705}
1706
1707/**
1708 * bond_alb_handle_active_change - assign new curr_active_slave
1709 * @bond: our bonding struct
1710 * @new_slave: new slave to assign
1711 *
1712 * Set the bond->curr_active_slave to @new_slave and handle
1713 * mac address swapping and promiscuity changes as needed.
1714 *
1715 * Caller must hold RTNL
1716 */
1717void bond_alb_handle_active_change(struct bonding *bond, struct slave *new_slave)
1718{
1719 struct slave *swap_slave;
1720 struct slave *curr_active;
1721
1722 curr_active = rtnl_dereference(bond->curr_active_slave);
1723 if (curr_active == new_slave)
1724 return;
1725
1726 if (curr_active && bond->alb_info.primary_is_promisc) {
1727 dev_set_promiscuity(dev: curr_active->dev, inc: -1);
1728 bond->alb_info.primary_is_promisc = 0;
1729 bond->alb_info.rlb_promisc_timeout_counter = 0;
1730 }
1731
1732 swap_slave = curr_active;
1733 rcu_assign_pointer(bond->curr_active_slave, new_slave);
1734
1735 if (!new_slave || !bond_has_slaves(bond))
1736 return;
1737
1738 /* set the new curr_active_slave to the bonds mac address
1739 * i.e. swap mac addresses of old curr_active_slave and new curr_active_slave
1740 */
1741 if (!swap_slave)
1742 swap_slave = bond_slave_has_mac(bond, mac: bond->dev->dev_addr);
1743
1744 /* Arrange for swap_slave and new_slave to temporarily be
1745 * ignored so we can mess with their MAC addresses without
1746 * fear of interference from transmit activity.
1747 */
1748 if (swap_slave)
1749 tlb_clear_slave(bond, slave: swap_slave, save_load: 1);
1750 tlb_clear_slave(bond, slave: new_slave, save_load: 1);
1751
1752 /* in TLB mode, the slave might flip down/up with the old dev_addr,
1753 * and thus filter bond->dev_addr's packets, so force bond's mac
1754 */
1755 if (BOND_MODE(bond) == BOND_MODE_TLB) {
1756 struct sockaddr_storage ss;
1757 u8 tmp_addr[MAX_ADDR_LEN];
1758
1759 bond_hw_addr_copy(dst: tmp_addr, src: new_slave->dev->dev_addr,
1760 len: new_slave->dev->addr_len);
1761
1762 bond_hw_addr_copy(dst: ss.__data, src: bond->dev->dev_addr,
1763 len: bond->dev->addr_len);
1764 ss.ss_family = bond->dev->type;
1765 /* we don't care if it can't change its mac, best effort */
1766 dev_set_mac_address(dev: new_slave->dev, sa: (struct sockaddr *)&ss,
1767 NULL);
1768
1769 dev_addr_set(dev: new_slave->dev, addr: tmp_addr);
1770 }
1771
1772 /* curr_active_slave must be set before calling alb_swap_mac_addr */
1773 if (swap_slave) {
1774 /* swap mac address */
1775 alb_swap_mac_addr(slave1: swap_slave, slave2: new_slave);
1776 alb_fasten_mac_swap(bond, slave1: swap_slave, slave2: new_slave);
1777 } else {
1778 /* set the new_slave to the bond mac address */
1779 alb_set_slave_mac_addr(slave: new_slave, addr: bond->dev->dev_addr,
1780 len: bond->dev->addr_len);
1781 alb_send_learning_packets(slave: new_slave, mac_addr: bond->dev->dev_addr,
1782 strict_match: false);
1783 }
1784}
1785
1786/* Called with RTNL */
1787int bond_alb_set_mac_address(struct net_device *bond_dev, void *addr)
1788{
1789 struct bonding *bond = netdev_priv(dev: bond_dev);
1790 struct sockaddr_storage *ss = addr;
1791 struct slave *curr_active;
1792 struct slave *swap_slave;
1793 int res;
1794
1795 if (!is_valid_ether_addr(addr: ss->__data))
1796 return -EADDRNOTAVAIL;
1797
1798 res = alb_set_mac_address(bond, addr);
1799 if (res)
1800 return res;
1801
1802 dev_addr_set(dev: bond_dev, addr: ss->__data);
1803
1804 /* If there is no curr_active_slave there is nothing else to do.
1805 * Otherwise we'll need to pass the new address to it and handle
1806 * duplications.
1807 */
1808 curr_active = rtnl_dereference(bond->curr_active_slave);
1809 if (!curr_active)
1810 return 0;
1811
1812 swap_slave = bond_slave_has_mac(bond, mac: bond_dev->dev_addr);
1813
1814 if (swap_slave) {
1815 alb_swap_mac_addr(slave1: swap_slave, slave2: curr_active);
1816 alb_fasten_mac_swap(bond, slave1: swap_slave, slave2: curr_active);
1817 } else {
1818 alb_set_slave_mac_addr(slave: curr_active, addr: bond_dev->dev_addr,
1819 len: bond_dev->addr_len);
1820
1821 alb_send_learning_packets(slave: curr_active,
1822 mac_addr: bond_dev->dev_addr, strict_match: false);
1823 if (bond->alb_info.rlb_enabled) {
1824 /* inform clients mac address has changed */
1825 rlb_req_update_slave_clients(bond, slave: curr_active);
1826 }
1827 }
1828
1829 return 0;
1830}
1831
1832void bond_alb_clear_vlan(struct bonding *bond, unsigned short vlan_id)
1833{
1834 if (bond->alb_info.rlb_enabled)
1835 rlb_clear_vlan(bond, vlan_id);
1836}
1837
1838

source code of linux/drivers/net/bonding/bond_alb.c