aboutsummaryrefslogtreecommitdiffstats
path: root/drivers/net/bonding
diff options
context:
space:
mode:
authorLinus Torvalds <torvalds@ppc970.osdl.org>2005-04-16 18:20:36 -0400
committerLinus Torvalds <torvalds@ppc970.osdl.org>2005-04-16 18:20:36 -0400
commit1da177e4c3f41524e886b7f1b8a0c1fc7321cac2 (patch)
tree0bba044c4ce775e45a88a51686b5d9f90697ea9d /drivers/net/bonding
Linux-2.6.12-rc2v2.6.12-rc2
Initial git repository build. I'm not bothering with the full history, even though we have it. We can create a separate "historical" git archive of that later if we want to, and in the meantime it's about 3.2GB when imported into git - space that would just make the early git days unnecessarily complicated, when we don't have a lot of good infrastructure for it. Let it rip!
Diffstat (limited to 'drivers/net/bonding')
-rw-r--r--drivers/net/bonding/Makefile8
-rw-r--r--drivers/net/bonding/bond_3ad.c2451
-rw-r--r--drivers/net/bonding/bond_3ad.h300
-rw-r--r--drivers/net/bonding/bond_alb.c1696
-rw-r--r--drivers/net/bonding/bond_alb.h141
-rw-r--r--drivers/net/bonding/bond_main.c4708
-rw-r--r--drivers/net/bonding/bonding.h252
7 files changed, 9556 insertions, 0 deletions
diff --git a/drivers/net/bonding/Makefile b/drivers/net/bonding/Makefile
new file mode 100644
index 000000000000..cf50384b469e
--- /dev/null
+++ b/drivers/net/bonding/Makefile
@@ -0,0 +1,8 @@
1#
2# Makefile for the Ethernet Bonding driver
3#
4
5obj-$(CONFIG_BONDING) += bonding.o
6
7bonding-objs := bond_main.o bond_3ad.o bond_alb.o
8
diff --git a/drivers/net/bonding/bond_3ad.c b/drivers/net/bonding/bond_3ad.c
new file mode 100644
index 000000000000..6233c4ffb805
--- /dev/null
+++ b/drivers/net/bonding/bond_3ad.c
@@ -0,0 +1,2451 @@
1/*
2 * Copyright(c) 1999 - 2004 Intel Corporation. All rights reserved.
3 *
4 * This program is free software; you can redistribute it and/or modify it
5 * under the terms of the GNU General Public License as published by the Free
6 * Software Foundation; either version 2 of the License, or (at your option)
7 * any later version.
8 *
9 * This program is distributed in the hope that it will be useful, but WITHOUT
10 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
11 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for
12 * more details.
13 *
14 * You should have received a copy of the GNU General Public License along with
15 * this program; if not, write to the Free Software Foundation, Inc., 59
16 * Temple Place - Suite 330, Boston, MA 02111-1307, USA.
17 *
18 * The full GNU General Public License is included in this distribution in the
19 * file called LICENSE.
20 *
21 *
22 * Changes:
23 *
24 * 2003/05/01 - Tsippy Mendelson <tsippy.mendelson at intel dot com> and
25 * Amir Noam <amir.noam at intel dot com>
26 * - Added support for lacp_rate module param.
27 *
28 * 2003/05/01 - Shmulik Hen <shmulik.hen at intel dot com>
29 * - Based on discussion on mailing list, changed locking scheme
30 * to use lock/unlock or lock_bh/unlock_bh appropriately instead
31 * of lock_irqsave/unlock_irqrestore. The new scheme helps exposing
32 * hidden bugs and solves system hangs that occurred due to the fact
33 * that holding lock_irqsave doesn't prevent softirqs from running.
34 * This also increases total throughput since interrupts are not
35 * blocked on each transmitted packets or monitor timeout.
36 *
37 * 2003/05/01 - Shmulik Hen <shmulik.hen at intel dot com>
38 * - Renamed bond_3ad_link_status_changed() to
39 * bond_3ad_handle_link_change() for compatibility with TLB.
40 *
41 * 2003/05/20 - Amir Noam <amir.noam at intel dot com>
42 * - Fix long fail over time when releasing last slave of an active
43 * aggregator - send LACPDU on unbind of slave to tell partner this
44 * port is no longer aggregatable.
45 *
46 * 2003/06/25 - Tsippy Mendelson <tsippy.mendelson at intel dot com>
47 * - Send LACPDU as highest priority packet to further fix the above
48 * problem on very high Tx traffic load where packets may get dropped
49 * by the slave.
50 *
51 * 2003/12/01 - Shmulik Hen <shmulik.hen at intel dot com>
52 * - Code cleanup and style changes
53 */
54
55//#define BONDING_DEBUG 1
56
57#include <linux/skbuff.h>
58#include <linux/if_ether.h>
59#include <linux/netdevice.h>
60#include <linux/spinlock.h>
61#include <linux/ethtool.h>
62#include <linux/if_bonding.h>
63#include <linux/pkt_sched.h>
64#include "bonding.h"
65#include "bond_3ad.h"
66
67// General definitions
68#define AD_SHORT_TIMEOUT 1
69#define AD_LONG_TIMEOUT 0
70#define AD_STANDBY 0x2
71#define AD_MAX_TX_IN_SECOND 3
72#define AD_COLLECTOR_MAX_DELAY 0
73
74// Timer definitions(43.4.4 in the 802.3ad standard)
75#define AD_FAST_PERIODIC_TIME 1
76#define AD_SLOW_PERIODIC_TIME 30
77#define AD_SHORT_TIMEOUT_TIME (3*AD_FAST_PERIODIC_TIME)
78#define AD_LONG_TIMEOUT_TIME (3*AD_SLOW_PERIODIC_TIME)
79#define AD_CHURN_DETECTION_TIME 60
80#define AD_AGGREGATE_WAIT_TIME 2
81
82// Port state definitions(43.4.2.2 in the 802.3ad standard)
83#define AD_STATE_LACP_ACTIVITY 0x1
84#define AD_STATE_LACP_TIMEOUT 0x2
85#define AD_STATE_AGGREGATION 0x4
86#define AD_STATE_SYNCHRONIZATION 0x8
87#define AD_STATE_COLLECTING 0x10
88#define AD_STATE_DISTRIBUTING 0x20
89#define AD_STATE_DEFAULTED 0x40
90#define AD_STATE_EXPIRED 0x80
91
92// Port Variables definitions used by the State Machines(43.4.7 in the 802.3ad standard)
93#define AD_PORT_BEGIN 0x1
94#define AD_PORT_LACP_ENABLED 0x2
95#define AD_PORT_ACTOR_CHURN 0x4
96#define AD_PORT_PARTNER_CHURN 0x8
97#define AD_PORT_READY 0x10
98#define AD_PORT_READY_N 0x20
99#define AD_PORT_MATCHED 0x40
100#define AD_PORT_STANDBY 0x80
101#define AD_PORT_SELECTED 0x100
102#define AD_PORT_MOVED 0x200
103
104// Port Key definitions
105// key is determined according to the link speed, duplex and
106// user key(which is yet not supported)
107// ------------------------------------------------------------
108// Port key : | User key | Speed |Duplex|
109// ------------------------------------------------------------
110// 16 6 1 0
111#define AD_DUPLEX_KEY_BITS 0x1
112#define AD_SPEED_KEY_BITS 0x3E
113#define AD_USER_KEY_BITS 0xFFC0
114
115//dalloun
116#define AD_LINK_SPEED_BITMASK_1MBPS 0x1
117#define AD_LINK_SPEED_BITMASK_10MBPS 0x2
118#define AD_LINK_SPEED_BITMASK_100MBPS 0x4
119#define AD_LINK_SPEED_BITMASK_1000MBPS 0x8
120//endalloun
121
122// compare MAC addresses
123#define MAC_ADDRESS_COMPARE(A, B) memcmp(A, B, ETH_ALEN)
124
125static struct mac_addr null_mac_addr = {{0, 0, 0, 0, 0, 0}};
126static u16 ad_ticks_per_sec;
127static const int ad_delta_in_ticks = (AD_TIMER_INTERVAL * HZ) / 1000;
128
129// ================= 3AD api to bonding and kernel code ==================
130static u16 __get_link_speed(struct port *port);
131static u8 __get_duplex(struct port *port);
132static inline void __initialize_port_locks(struct port *port);
133//conversions
134static void __ntohs_lacpdu(struct lacpdu *lacpdu);
135static u16 __ad_timer_to_ticks(u16 timer_type, u16 Par);
136
137
138// ================= ad code helper functions ==================
139//needed by ad_rx_machine(...)
140static void __record_pdu(struct lacpdu *lacpdu, struct port *port);
141static void __record_default(struct port *port);
142static void __update_selected(struct lacpdu *lacpdu, struct port *port);
143static void __update_default_selected(struct port *port);
144static void __choose_matched(struct lacpdu *lacpdu, struct port *port);
145static void __update_ntt(struct lacpdu *lacpdu, struct port *port);
146
147//needed for ad_mux_machine(..)
148static void __attach_bond_to_agg(struct port *port);
149static void __detach_bond_from_agg(struct port *port);
150static int __agg_ports_are_ready(struct aggregator *aggregator);
151static void __set_agg_ports_ready(struct aggregator *aggregator, int val);
152
153//needed for ad_agg_selection_logic(...)
154static u32 __get_agg_bandwidth(struct aggregator *aggregator);
155static struct aggregator *__get_active_agg(struct aggregator *aggregator);
156
157
158// ================= main 802.3ad protocol functions ==================
159static int ad_lacpdu_send(struct port *port);
160static int ad_marker_send(struct port *port, struct marker *marker);
161static void ad_mux_machine(struct port *port);
162static void ad_rx_machine(struct lacpdu *lacpdu, struct port *port);
163static void ad_tx_machine(struct port *port);
164static void ad_periodic_machine(struct port *port);
165static void ad_port_selection_logic(struct port *port);
166static void ad_agg_selection_logic(struct aggregator *aggregator);
167static void ad_clear_agg(struct aggregator *aggregator);
168static void ad_initialize_agg(struct aggregator *aggregator);
169static void ad_initialize_port(struct port *port, int lacp_fast);
170static void ad_initialize_lacpdu(struct lacpdu *Lacpdu);
171static void ad_enable_collecting_distributing(struct port *port);
172static void ad_disable_collecting_distributing(struct port *port);
173static void ad_marker_info_received(struct marker *marker_info, struct port *port);
174static void ad_marker_response_received(struct marker *marker, struct port *port);
175
176
177/////////////////////////////////////////////////////////////////////////////////
178// ================= api to bonding and kernel code ==================
179/////////////////////////////////////////////////////////////////////////////////
180
181/**
182 * __get_bond_by_port - get the port's bonding struct
183 * @port: the port we're looking at
184 *
185 * Return @port's bonding struct, or %NULL if it can't be found.
186 */
187static inline struct bonding *__get_bond_by_port(struct port *port)
188{
189 if (port->slave == NULL) {
190 return NULL;
191 }
192
193 return bond_get_bond_by_slave(port->slave);
194}
195
196/**
197 * __get_first_port - get the first port in the bond
198 * @bond: the bond we're looking at
199 *
200 * Return the port of the first slave in @bond, or %NULL if it can't be found.
201 */
202static inline struct port *__get_first_port(struct bonding *bond)
203{
204 if (bond->slave_cnt == 0) {
205 return NULL;
206 }
207
208 return &(SLAVE_AD_INFO(bond->first_slave).port);
209}
210
211/**
212 * __get_next_port - get the next port in the bond
213 * @port: the port we're looking at
214 *
215 * Return the port of the slave that is next in line of @port's slave in the
216 * bond, or %NULL if it can't be found.
217 */
218static inline struct port *__get_next_port(struct port *port)
219{
220 struct bonding *bond = __get_bond_by_port(port);
221 struct slave *slave = port->slave;
222
223 // If there's no bond for this port, or this is the last slave
224 if ((bond == NULL) || (slave->next == bond->first_slave)) {
225 return NULL;
226 }
227
228 return &(SLAVE_AD_INFO(slave->next).port);
229}
230
231/**
232 * __get_first_agg - get the first aggregator in the bond
233 * @bond: the bond we're looking at
234 *
235 * Return the aggregator of the first slave in @bond, or %NULL if it can't be
236 * found.
237 */
238static inline struct aggregator *__get_first_agg(struct port *port)
239{
240 struct bonding *bond = __get_bond_by_port(port);
241
242 // If there's no bond for this port, or bond has no slaves
243 if ((bond == NULL) || (bond->slave_cnt == 0)) {
244 return NULL;
245 }
246
247 return &(SLAVE_AD_INFO(bond->first_slave).aggregator);
248}
249
250/**
251 * __get_next_agg - get the next aggregator in the bond
252 * @aggregator: the aggregator we're looking at
253 *
254 * Return the aggregator of the slave that is next in line of @aggregator's
255 * slave in the bond, or %NULL if it can't be found.
256 */
257static inline struct aggregator *__get_next_agg(struct aggregator *aggregator)
258{
259 struct slave *slave = aggregator->slave;
260 struct bonding *bond = bond_get_bond_by_slave(slave);
261
262 // If there's no bond for this aggregator, or this is the last slave
263 if ((bond == NULL) || (slave->next == bond->first_slave)) {
264 return NULL;
265 }
266
267 return &(SLAVE_AD_INFO(slave->next).aggregator);
268}
269
270/**
271 * __disable_port - disable the port's slave
272 * @port: the port we're looking at
273 *
274 */
275static inline void __disable_port(struct port *port)
276{
277 bond_set_slave_inactive_flags(port->slave);
278}
279
280/**
281 * __enable_port - enable the port's slave, if it's up
282 * @port: the port we're looking at
283 *
284 */
285static inline void __enable_port(struct port *port)
286{
287 struct slave *slave = port->slave;
288
289 if ((slave->link == BOND_LINK_UP) && IS_UP(slave->dev)) {
290 bond_set_slave_active_flags(slave);
291 }
292}
293
294/**
295 * __port_is_enabled - check if the port's slave is in active state
296 * @port: the port we're looking at
297 *
298 */
299static inline int __port_is_enabled(struct port *port)
300{
301 return(port->slave->state == BOND_STATE_ACTIVE);
302}
303
304/**
305 * __get_agg_selection_mode - get the aggregator selection mode
306 * @port: the port we're looking at
307 *
308 * Get the aggregator selection mode. Can be %BANDWIDTH or %COUNT.
309 */
310static inline u32 __get_agg_selection_mode(struct port *port)
311{
312 struct bonding *bond = __get_bond_by_port(port);
313
314 if (bond == NULL) {
315 return AD_BANDWIDTH;
316 }
317
318 return BOND_AD_INFO(bond).agg_select_mode;
319}
320
321/**
322 * __check_agg_selection_timer - check if the selection timer has expired
323 * @port: the port we're looking at
324 *
325 */
326static inline int __check_agg_selection_timer(struct port *port)
327{
328 struct bonding *bond = __get_bond_by_port(port);
329
330 if (bond == NULL) {
331 return 0;
332 }
333
334 return BOND_AD_INFO(bond).agg_select_timer ? 1 : 0;
335}
336
337/**
338 * __get_rx_machine_lock - lock the port's RX machine
339 * @port: the port we're looking at
340 *
341 */
342static inline void __get_rx_machine_lock(struct port *port)
343{
344 spin_lock(&(SLAVE_AD_INFO(port->slave).rx_machine_lock));
345}
346
347/**
348 * __release_rx_machine_lock - unlock the port's RX machine
349 * @port: the port we're looking at
350 *
351 */
352static inline void __release_rx_machine_lock(struct port *port)
353{
354 spin_unlock(&(SLAVE_AD_INFO(port->slave).rx_machine_lock));
355}
356
357/**
358 * __get_link_speed - get a port's speed
359 * @port: the port we're looking at
360 *
361 * Return @port's speed in 802.3ad bitmask format. i.e. one of:
362 * 0,
363 * %AD_LINK_SPEED_BITMASK_10MBPS,
364 * %AD_LINK_SPEED_BITMASK_100MBPS,
365 * %AD_LINK_SPEED_BITMASK_1000MBPS
366 */
367static u16 __get_link_speed(struct port *port)
368{
369 struct slave *slave = port->slave;
370 u16 speed;
371
372 /* this if covers only a special case: when the configuration starts with
373 * link down, it sets the speed to 0.
374 * This is done in spite of the fact that the e100 driver reports 0 to be
375 * compatible with MVT in the future.*/
376 if (slave->link != BOND_LINK_UP) {
377 speed=0;
378 } else {
379 switch (slave->speed) {
380 case SPEED_10:
381 speed = AD_LINK_SPEED_BITMASK_10MBPS;
382 break;
383
384 case SPEED_100:
385 speed = AD_LINK_SPEED_BITMASK_100MBPS;
386 break;
387
388 case SPEED_1000:
389 speed = AD_LINK_SPEED_BITMASK_1000MBPS;
390 break;
391
392 default:
393 speed = 0; // unknown speed value from ethtool. shouldn't happen
394 break;
395 }
396 }
397
398 dprintk("Port %d Received link speed %d update from adapter\n", port->actor_port_number, speed);
399 return speed;
400}
401
402/**
403 * __get_duplex - get a port's duplex
404 * @port: the port we're looking at
405 *
406 * Return @port's duplex in 802.3ad bitmask format. i.e.:
407 * 0x01 if in full duplex
408 * 0x00 otherwise
409 */
410static u8 __get_duplex(struct port *port)
411{
412 struct slave *slave = port->slave;
413
414 u8 retval;
415
416 // handling a special case: when the configuration starts with
417 // link down, it sets the duplex to 0.
418 if (slave->link != BOND_LINK_UP) {
419 retval=0x0;
420 } else {
421 switch (slave->duplex) {
422 case DUPLEX_FULL:
423 retval=0x1;
424 dprintk("Port %d Received status full duplex update from adapter\n", port->actor_port_number);
425 break;
426 case DUPLEX_HALF:
427 default:
428 retval=0x0;
429 dprintk("Port %d Received status NOT full duplex update from adapter\n", port->actor_port_number);
430 break;
431 }
432 }
433 return retval;
434}
435
436/**
437 * __initialize_port_locks - initialize a port's RX machine spinlock
438 * @port: the port we're looking at
439 *
440 */
441static inline void __initialize_port_locks(struct port *port)
442{
443 // make sure it isn't called twice
444 spin_lock_init(&(SLAVE_AD_INFO(port->slave).rx_machine_lock));
445}
446
447//conversions
448/**
449 * __ntohs_lacpdu - convert the contents of a LACPDU to host byte order
450 * @lacpdu: the speicifed lacpdu
451 *
452 * For each multi-byte field in the lacpdu, convert its content
453 */
454static void __ntohs_lacpdu(struct lacpdu *lacpdu)
455{
456 if (lacpdu) {
457 lacpdu->actor_system_priority = ntohs(lacpdu->actor_system_priority);
458 lacpdu->actor_key = ntohs(lacpdu->actor_key);
459 lacpdu->actor_port_priority = ntohs(lacpdu->actor_port_priority);
460 lacpdu->actor_port = ntohs(lacpdu->actor_port);
461 lacpdu->partner_system_priority = ntohs(lacpdu->partner_system_priority);
462 lacpdu->partner_key = ntohs(lacpdu->partner_key);
463 lacpdu->partner_port_priority = ntohs(lacpdu->partner_port_priority);
464 lacpdu->partner_port = ntohs(lacpdu->partner_port);
465 lacpdu->collector_max_delay = ntohs(lacpdu->collector_max_delay);
466 }
467}
468
469/**
470 * __ad_timer_to_ticks - convert a given timer type to AD module ticks
471 * @timer_type: which timer to operate
472 * @par: timer parameter. see below
473 *
474 * If @timer_type is %current_while_timer, @par indicates long/short timer.
475 * If @timer_type is %periodic_timer, @par is one of %FAST_PERIODIC_TIME,
476 * %SLOW_PERIODIC_TIME.
477 */
478static u16 __ad_timer_to_ticks(u16 timer_type, u16 par)
479{
480 u16 retval=0; //to silence the compiler
481
482 switch (timer_type) {
483 case AD_CURRENT_WHILE_TIMER: // for rx machine usage
484 if (par) { // for short or long timeout
485 retval = (AD_SHORT_TIMEOUT_TIME*ad_ticks_per_sec); // short timeout
486 } else {
487 retval = (AD_LONG_TIMEOUT_TIME*ad_ticks_per_sec); // long timeout
488 }
489 break;
490 case AD_ACTOR_CHURN_TIMER: // for local churn machine
491 retval = (AD_CHURN_DETECTION_TIME*ad_ticks_per_sec);
492 break;
493 case AD_PERIODIC_TIMER: // for periodic machine
494 retval = (par*ad_ticks_per_sec); // long timeout
495 break;
496 case AD_PARTNER_CHURN_TIMER: // for remote churn machine
497 retval = (AD_CHURN_DETECTION_TIME*ad_ticks_per_sec);
498 break;
499 case AD_WAIT_WHILE_TIMER: // for selection machine
500 retval = (AD_AGGREGATE_WAIT_TIME*ad_ticks_per_sec);
501 break;
502 }
503 return retval;
504}
505
506
507/////////////////////////////////////////////////////////////////////////////////
508// ================= ad_rx_machine helper functions ==================
509/////////////////////////////////////////////////////////////////////////////////
510
511/**
512 * __record_pdu - record parameters from a received lacpdu
513 * @lacpdu: the lacpdu we've received
514 * @port: the port we're looking at
515 *
516 * Record the parameter values for the Actor carried in a received lacpdu as
517 * the current partner operational parameter values and sets
518 * actor_oper_port_state.defaulted to FALSE.
519 */
520static void __record_pdu(struct lacpdu *lacpdu, struct port *port)
521{
522 // validate lacpdu and port
523 if (lacpdu && port) {
524 // record the new parameter values for the partner operational
525 port->partner_oper_port_number = lacpdu->actor_port;
526 port->partner_oper_port_priority = lacpdu->actor_port_priority;
527 port->partner_oper_system = lacpdu->actor_system;
528 port->partner_oper_system_priority = lacpdu->actor_system_priority;
529 port->partner_oper_key = lacpdu->actor_key;
530 // zero partener's lase states
531 port->partner_oper_port_state = 0;
532 port->partner_oper_port_state |= (lacpdu->actor_state & AD_STATE_LACP_ACTIVITY);
533 port->partner_oper_port_state |= (lacpdu->actor_state & AD_STATE_LACP_TIMEOUT);
534 port->partner_oper_port_state |= (lacpdu->actor_state & AD_STATE_AGGREGATION);
535 port->partner_oper_port_state |= (lacpdu->actor_state & AD_STATE_SYNCHRONIZATION);
536 port->partner_oper_port_state |= (lacpdu->actor_state & AD_STATE_COLLECTING);
537 port->partner_oper_port_state |= (lacpdu->actor_state & AD_STATE_DISTRIBUTING);
538 port->partner_oper_port_state |= (lacpdu->actor_state & AD_STATE_DEFAULTED);
539 port->partner_oper_port_state |= (lacpdu->actor_state & AD_STATE_EXPIRED);
540
541 // set actor_oper_port_state.defaulted to FALSE
542 port->actor_oper_port_state &= ~AD_STATE_DEFAULTED;
543
544 // set the partner sync. to on if the partner is sync. and the port is matched
545 if ((port->sm_vars & AD_PORT_MATCHED) && (lacpdu->actor_state & AD_STATE_SYNCHRONIZATION)) {
546 port->partner_oper_port_state |= AD_STATE_SYNCHRONIZATION;
547 } else {
548 port->partner_oper_port_state &= ~AD_STATE_SYNCHRONIZATION;
549 }
550 }
551}
552
553/**
554 * __record_default - record default parameters
555 * @port: the port we're looking at
556 *
557 * This function records the default parameter values for the partner carried
558 * in the Partner Admin parameters as the current partner operational parameter
559 * values and sets actor_oper_port_state.defaulted to TRUE.
560 */
561static void __record_default(struct port *port)
562{
563 // validate the port
564 if (port) {
565 // record the partner admin parameters
566 port->partner_oper_port_number = port->partner_admin_port_number;
567 port->partner_oper_port_priority = port->partner_admin_port_priority;
568 port->partner_oper_system = port->partner_admin_system;
569 port->partner_oper_system_priority = port->partner_admin_system_priority;
570 port->partner_oper_key = port->partner_admin_key;
571 port->partner_oper_port_state = port->partner_admin_port_state;
572
573 // set actor_oper_port_state.defaulted to true
574 port->actor_oper_port_state |= AD_STATE_DEFAULTED;
575 }
576}
577
578/**
579 * __update_selected - update a port's Selected variable from a received lacpdu
580 * @lacpdu: the lacpdu we've received
581 * @port: the port we're looking at
582 *
583 * Update the value of the selected variable, using parameter values from a
584 * newly received lacpdu. The parameter values for the Actor carried in the
585 * received PDU are compared with the corresponding operational parameter
586 * values for the ports partner. If one or more of the comparisons shows that
587 * the value(s) received in the PDU differ from the current operational values,
588 * then selected is set to FALSE and actor_oper_port_state.synchronization is
589 * set to out_of_sync. Otherwise, selected remains unchanged.
590 */
591static void __update_selected(struct lacpdu *lacpdu, struct port *port)
592{
593 // validate lacpdu and port
594 if (lacpdu && port) {
595 // check if any parameter is different
596 if ((lacpdu->actor_port != port->partner_oper_port_number) ||
597 (lacpdu->actor_port_priority != port->partner_oper_port_priority) ||
598 MAC_ADDRESS_COMPARE(&(lacpdu->actor_system), &(port->partner_oper_system)) ||
599 (lacpdu->actor_system_priority != port->partner_oper_system_priority) ||
600 (lacpdu->actor_key != port->partner_oper_key) ||
601 ((lacpdu->actor_state & AD_STATE_AGGREGATION) != (port->partner_oper_port_state & AD_STATE_AGGREGATION))
602 ) {
603 // update the state machine Selected variable
604 port->sm_vars &= ~AD_PORT_SELECTED;
605 }
606 }
607}
608
609/**
610 * __update_default_selected - update a port's Selected variable from Partner
611 * @port: the port we're looking at
612 *
613 * This function updates the value of the selected variable, using the partner
614 * administrative parameter values. The administrative values are compared with
615 * the corresponding operational parameter values for the partner. If one or
616 * more of the comparisons shows that the administrative value(s) differ from
617 * the current operational values, then Selected is set to FALSE and
618 * actor_oper_port_state.synchronization is set to OUT_OF_SYNC. Otherwise,
619 * Selected remains unchanged.
620 */
621static void __update_default_selected(struct port *port)
622{
623 // validate the port
624 if (port) {
625 // check if any parameter is different
626 if ((port->partner_admin_port_number != port->partner_oper_port_number) ||
627 (port->partner_admin_port_priority != port->partner_oper_port_priority) ||
628 MAC_ADDRESS_COMPARE(&(port->partner_admin_system), &(port->partner_oper_system)) ||
629 (port->partner_admin_system_priority != port->partner_oper_system_priority) ||
630 (port->partner_admin_key != port->partner_oper_key) ||
631 ((port->partner_admin_port_state & AD_STATE_AGGREGATION) != (port->partner_oper_port_state & AD_STATE_AGGREGATION))
632 ) {
633 // update the state machine Selected variable
634 port->sm_vars &= ~AD_PORT_SELECTED;
635 }
636 }
637}
638
639/**
640 * __choose_matched - update a port's matched variable from a received lacpdu
641 * @lacpdu: the lacpdu we've received
642 * @port: the port we're looking at
643 *
644 * Update the value of the matched variable, using parameter values from a
645 * newly received lacpdu. Parameter values for the partner carried in the
646 * received PDU are compared with the corresponding operational parameter
647 * values for the actor. Matched is set to TRUE if all of these parameters
648 * match and the PDU parameter partner_state.aggregation has the same value as
649 * actor_oper_port_state.aggregation and lacp will actively maintain the link
650 * in the aggregation. Matched is also set to TRUE if the value of
651 * actor_state.aggregation in the received PDU is set to FALSE, i.e., indicates
652 * an individual link and lacp will actively maintain the link. Otherwise,
653 * matched is set to FALSE. LACP is considered to be actively maintaining the
654 * link if either the PDU's actor_state.lacp_activity variable is TRUE or both
655 * the actor's actor_oper_port_state.lacp_activity and the PDU's
656 * partner_state.lacp_activity variables are TRUE.
657 */
658static void __choose_matched(struct lacpdu *lacpdu, struct port *port)
659{
660 // validate lacpdu and port
661 if (lacpdu && port) {
662 // check if all parameters are alike
663 if (((lacpdu->partner_port == port->actor_port_number) &&
664 (lacpdu->partner_port_priority == port->actor_port_priority) &&
665 !MAC_ADDRESS_COMPARE(&(lacpdu->partner_system), &(port->actor_system)) &&
666 (lacpdu->partner_system_priority == port->actor_system_priority) &&
667 (lacpdu->partner_key == port->actor_oper_port_key) &&
668 ((lacpdu->partner_state & AD_STATE_AGGREGATION) == (port->actor_oper_port_state & AD_STATE_AGGREGATION))) ||
669 // or this is individual link(aggregation == FALSE)
670 ((lacpdu->actor_state & AD_STATE_AGGREGATION) == 0)
671 ) {
672 // update the state machine Matched variable
673 port->sm_vars |= AD_PORT_MATCHED;
674 } else {
675 port->sm_vars &= ~AD_PORT_MATCHED;
676 }
677 }
678}
679
680/**
681 * __update_ntt - update a port's ntt variable from a received lacpdu
682 * @lacpdu: the lacpdu we've received
683 * @port: the port we're looking at
684 *
685 * Updates the value of the ntt variable, using parameter values from a newly
686 * received lacpdu. The parameter values for the partner carried in the
687 * received PDU are compared with the corresponding operational parameter
688 * values for the Actor. If one or more of the comparisons shows that the
689 * value(s) received in the PDU differ from the current operational values,
690 * then ntt is set to TRUE. Otherwise, ntt remains unchanged.
691 */
692static void __update_ntt(struct lacpdu *lacpdu, struct port *port)
693{
694 // validate lacpdu and port
695 if (lacpdu && port) {
696 // check if any parameter is different
697 if ((lacpdu->partner_port != port->actor_port_number) ||
698 (lacpdu->partner_port_priority != port->actor_port_priority) ||
699 MAC_ADDRESS_COMPARE(&(lacpdu->partner_system), &(port->actor_system)) ||
700 (lacpdu->partner_system_priority != port->actor_system_priority) ||
701 (lacpdu->partner_key != port->actor_oper_port_key) ||
702 ((lacpdu->partner_state & AD_STATE_LACP_ACTIVITY) != (port->actor_oper_port_state & AD_STATE_LACP_ACTIVITY)) ||
703 ((lacpdu->partner_state & AD_STATE_LACP_TIMEOUT) != (port->actor_oper_port_state & AD_STATE_LACP_TIMEOUT)) ||
704 ((lacpdu->partner_state & AD_STATE_SYNCHRONIZATION) != (port->actor_oper_port_state & AD_STATE_SYNCHRONIZATION)) ||
705 ((lacpdu->partner_state & AD_STATE_AGGREGATION) != (port->actor_oper_port_state & AD_STATE_AGGREGATION))
706 ) {
707 // set ntt to be TRUE
708 port->ntt = 1;
709 }
710 }
711}
712
713/**
714 * __attach_bond_to_agg
715 * @port: the port we're looking at
716 *
717 * Handle the attaching of the port's control parser/multiplexer and the
718 * aggregator. This function does nothing since the parser/multiplexer of the
719 * receive and the parser/multiplexer of the aggregator are already combined.
720 */
721static void __attach_bond_to_agg(struct port *port)
722{
723 port=NULL; // just to satisfy the compiler
724 // This function does nothing since the parser/multiplexer of the receive
725 // and the parser/multiplexer of the aggregator are already combined
726}
727
728/**
729 * __detach_bond_from_agg
730 * @port: the port we're looking at
731 *
732 * Handle the detaching of the port's control parser/multiplexer from the
733 * aggregator. This function does nothing since the parser/multiplexer of the
734 * receive and the parser/multiplexer of the aggregator are already combined.
735 */
736static void __detach_bond_from_agg(struct port *port)
737{
738 port=NULL; // just to satisfy the compiler
739 // This function does nothing sience the parser/multiplexer of the receive
740 // and the parser/multiplexer of the aggregator are already combined
741}
742
743/**
744 * __agg_ports_are_ready - check if all ports in an aggregator are ready
745 * @aggregator: the aggregator we're looking at
746 *
747 */
748static int __agg_ports_are_ready(struct aggregator *aggregator)
749{
750 struct port *port;
751 int retval = 1;
752
753 if (aggregator) {
754 // scan all ports in this aggregator to verfy if they are all ready
755 for (port=aggregator->lag_ports; port; port=port->next_port_in_aggregator) {
756 if (!(port->sm_vars & AD_PORT_READY_N)) {
757 retval = 0;
758 break;
759 }
760 }
761 }
762
763 return retval;
764}
765
766/**
767 * __set_agg_ports_ready - set value of Ready bit in all ports of an aggregator
768 * @aggregator: the aggregator we're looking at
769 * @val: Should the ports' ready bit be set on or off
770 *
771 */
772static void __set_agg_ports_ready(struct aggregator *aggregator, int val)
773{
774 struct port *port;
775
776 for (port=aggregator->lag_ports; port; port=port->next_port_in_aggregator) {
777 if (val) {
778 port->sm_vars |= AD_PORT_READY;
779 } else {
780 port->sm_vars &= ~AD_PORT_READY;
781 }
782 }
783}
784
785/**
786 * __get_agg_bandwidth - get the total bandwidth of an aggregator
787 * @aggregator: the aggregator we're looking at
788 *
789 */
790static u32 __get_agg_bandwidth(struct aggregator *aggregator)
791{
792 u32 bandwidth=0;
793 u32 basic_speed;
794
795 if (aggregator->num_of_ports) {
796 basic_speed = __get_link_speed(aggregator->lag_ports);
797 switch (basic_speed) {
798 case AD_LINK_SPEED_BITMASK_1MBPS:
799 bandwidth = aggregator->num_of_ports;
800 break;
801 case AD_LINK_SPEED_BITMASK_10MBPS:
802 bandwidth = aggregator->num_of_ports * 10;
803 break;
804 case AD_LINK_SPEED_BITMASK_100MBPS:
805 bandwidth = aggregator->num_of_ports * 100;
806 break;
807 case AD_LINK_SPEED_BITMASK_1000MBPS:
808 bandwidth = aggregator->num_of_ports * 1000;
809 break;
810 default:
811 bandwidth=0; // to silent the compilor ....
812 }
813 }
814 return bandwidth;
815}
816
817/**
818 * __get_active_agg - get the current active aggregator
819 * @aggregator: the aggregator we're looking at
820 *
821 */
822static struct aggregator *__get_active_agg(struct aggregator *aggregator)
823{
824 struct aggregator *retval = NULL;
825
826 for (; aggregator; aggregator = __get_next_agg(aggregator)) {
827 if (aggregator->is_active) {
828 retval = aggregator;
829 break;
830 }
831 }
832
833 return retval;
834}
835
836/**
837 * __update_lacpdu_from_port - update a port's lacpdu fields
838 * @port: the port we're looking at
839 *
840 */
841static inline void __update_lacpdu_from_port(struct port *port)
842{
843 struct lacpdu *lacpdu = &port->lacpdu;
844
845 /* update current actual Actor parameters */
846 /* lacpdu->subtype initialized
847 * lacpdu->version_number initialized
848 * lacpdu->tlv_type_actor_info initialized
849 * lacpdu->actor_information_length initialized
850 */
851
852 lacpdu->actor_system_priority = port->actor_system_priority;
853 lacpdu->actor_system = port->actor_system;
854 lacpdu->actor_key = port->actor_oper_port_key;
855 lacpdu->actor_port_priority = port->actor_port_priority;
856 lacpdu->actor_port = port->actor_port_number;
857 lacpdu->actor_state = port->actor_oper_port_state;
858
859 /* lacpdu->reserved_3_1 initialized
860 * lacpdu->tlv_type_partner_info initialized
861 * lacpdu->partner_information_length initialized
862 */
863
864 lacpdu->partner_system_priority = port->partner_oper_system_priority;
865 lacpdu->partner_system = port->partner_oper_system;
866 lacpdu->partner_key = port->partner_oper_key;
867 lacpdu->partner_port_priority = port->partner_oper_port_priority;
868 lacpdu->partner_port = port->partner_oper_port_number;
869 lacpdu->partner_state = port->partner_oper_port_state;
870
871 /* lacpdu->reserved_3_2 initialized
872 * lacpdu->tlv_type_collector_info initialized
873 * lacpdu->collector_information_length initialized
874 * collector_max_delay initialized
875 * reserved_12[12] initialized
876 * tlv_type_terminator initialized
877 * terminator_length initialized
878 * reserved_50[50] initialized
879 */
880
881 /* Convert all non u8 parameters to Big Endian for transmit */
882 __ntohs_lacpdu(lacpdu);
883}
884
885//////////////////////////////////////////////////////////////////////////////////////
886// ================= main 802.3ad protocol code ======================================
887//////////////////////////////////////////////////////////////////////////////////////
888
889/**
890 * ad_lacpdu_send - send out a lacpdu packet on a given port
891 * @port: the port we're looking at
892 *
893 * Returns: 0 on success
894 * < 0 on error
895 */
896static int ad_lacpdu_send(struct port *port)
897{
898 struct slave *slave = port->slave;
899 struct sk_buff *skb;
900 struct lacpdu_header *lacpdu_header;
901 int length = sizeof(struct lacpdu_header);
902 struct mac_addr lacpdu_multicast_address = AD_MULTICAST_LACPDU_ADDR;
903
904 skb = dev_alloc_skb(length);
905 if (!skb) {
906 return -ENOMEM;
907 }
908
909 skb->dev = slave->dev;
910 skb->mac.raw = skb->data;
911 skb->nh.raw = skb->data + ETH_HLEN;
912 skb->protocol = PKT_TYPE_LACPDU;
913 skb->priority = TC_PRIO_CONTROL;
914
915 lacpdu_header = (struct lacpdu_header *)skb_put(skb, length);
916
917 lacpdu_header->ad_header.destination_address = lacpdu_multicast_address;
918 /* Note: source addres is set to be the member's PERMANENT address, because we use it
919 to identify loopback lacpdus in receive. */
920 lacpdu_header->ad_header.source_address = *((struct mac_addr *)(slave->perm_hwaddr));
921 lacpdu_header->ad_header.length_type = PKT_TYPE_LACPDU;
922
923 lacpdu_header->lacpdu = port->lacpdu; // struct copy
924
925 dev_queue_xmit(skb);
926
927 return 0;
928}
929
930/**
931 * ad_marker_send - send marker information/response on a given port
932 * @port: the port we're looking at
933 * @marker: marker data to send
934 *
935 * Returns: 0 on success
936 * < 0 on error
937 */
938static int ad_marker_send(struct port *port, struct marker *marker)
939{
940 struct slave *slave = port->slave;
941 struct sk_buff *skb;
942 struct marker_header *marker_header;
943 int length = sizeof(struct marker_header);
944 struct mac_addr lacpdu_multicast_address = AD_MULTICAST_LACPDU_ADDR;
945
946 skb = dev_alloc_skb(length + 16);
947 if (!skb) {
948 return -ENOMEM;
949 }
950
951 skb_reserve(skb, 16);
952
953 skb->dev = slave->dev;
954 skb->mac.raw = skb->data;
955 skb->nh.raw = skb->data + ETH_HLEN;
956 skb->protocol = PKT_TYPE_LACPDU;
957
958 marker_header = (struct marker_header *)skb_put(skb, length);
959
960 marker_header->ad_header.destination_address = lacpdu_multicast_address;
961 /* Note: source addres is set to be the member's PERMANENT address, because we use it
962 to identify loopback MARKERs in receive. */
963 marker_header->ad_header.source_address = *((struct mac_addr *)(slave->perm_hwaddr));
964 marker_header->ad_header.length_type = PKT_TYPE_LACPDU;
965
966 marker_header->marker = *marker; // struct copy
967
968 dev_queue_xmit(skb);
969
970 return 0;
971}
972
973/**
974 * ad_mux_machine - handle a port's mux state machine
975 * @port: the port we're looking at
976 *
977 */
978static void ad_mux_machine(struct port *port)
979{
980 mux_states_t last_state;
981
982 // keep current State Machine state to compare later if it was changed
983 last_state = port->sm_mux_state;
984
985 if (port->sm_vars & AD_PORT_BEGIN) {
986 port->sm_mux_state = AD_MUX_DETACHED; // next state
987 } else {
988 switch (port->sm_mux_state) {
989 case AD_MUX_DETACHED:
990 if ((port->sm_vars & AD_PORT_SELECTED) || (port->sm_vars & AD_PORT_STANDBY)) { // if SELECTED or STANDBY
991 port->sm_mux_state = AD_MUX_WAITING; // next state
992 }
993 break;
994 case AD_MUX_WAITING:
995 // if SELECTED == FALSE return to DETACH state
996 if (!(port->sm_vars & AD_PORT_SELECTED)) { // if UNSELECTED
997 port->sm_vars &= ~AD_PORT_READY_N;
998 // in order to withhold the Selection Logic to check all ports READY_N value
999 // every callback cycle to update ready variable, we check READY_N and update READY here
1000 __set_agg_ports_ready(port->aggregator, __agg_ports_are_ready(port->aggregator));
1001 port->sm_mux_state = AD_MUX_DETACHED; // next state
1002 break;
1003 }
1004
1005 // check if the wait_while_timer expired
1006 if (port->sm_mux_timer_counter && !(--port->sm_mux_timer_counter)) {
1007 port->sm_vars |= AD_PORT_READY_N;
1008 }
1009
1010 // in order to withhold the selection logic to check all ports READY_N value
1011 // every callback cycle to update ready variable, we check READY_N and update READY here
1012 __set_agg_ports_ready(port->aggregator, __agg_ports_are_ready(port->aggregator));
1013
1014 // if the wait_while_timer expired, and the port is in READY state, move to ATTACHED state
1015 if ((port->sm_vars & AD_PORT_READY) && !port->sm_mux_timer_counter) {
1016 port->sm_mux_state = AD_MUX_ATTACHED; // next state
1017 }
1018 break;
1019 case AD_MUX_ATTACHED:
1020 // check also if agg_select_timer expired(so the edable port will take place only after this timer)
1021 if ((port->sm_vars & AD_PORT_SELECTED) && (port->partner_oper_port_state & AD_STATE_SYNCHRONIZATION) && !__check_agg_selection_timer(port)) {
1022 port->sm_mux_state = AD_MUX_COLLECTING_DISTRIBUTING;// next state
1023 } else if (!(port->sm_vars & AD_PORT_SELECTED) || (port->sm_vars & AD_PORT_STANDBY)) { // if UNSELECTED or STANDBY
1024 port->sm_vars &= ~AD_PORT_READY_N;
1025 // in order to withhold the selection logic to check all ports READY_N value
1026 // every callback cycle to update ready variable, we check READY_N and update READY here
1027 __set_agg_ports_ready(port->aggregator, __agg_ports_are_ready(port->aggregator));
1028 port->sm_mux_state = AD_MUX_DETACHED;// next state
1029 }
1030 break;
1031 case AD_MUX_COLLECTING_DISTRIBUTING:
1032 if (!(port->sm_vars & AD_PORT_SELECTED) || (port->sm_vars & AD_PORT_STANDBY) ||
1033 !(port->partner_oper_port_state & AD_STATE_SYNCHRONIZATION)
1034 ) {
1035 port->sm_mux_state = AD_MUX_ATTACHED;// next state
1036
1037 } else {
1038 // if port state hasn't changed make
1039 // sure that a collecting distributing
1040 // port in an active aggregator is enabled
1041 if (port->aggregator &&
1042 port->aggregator->is_active &&
1043 !__port_is_enabled(port)) {
1044
1045 __enable_port(port);
1046 }
1047 }
1048 break;
1049 default: //to silence the compiler
1050 break;
1051 }
1052 }
1053
1054 // check if the state machine was changed
1055 if (port->sm_mux_state != last_state) {
1056 dprintk("Mux Machine: Port=%d, Last State=%d, Curr State=%d\n", port->actor_port_number, last_state, port->sm_mux_state);
1057 switch (port->sm_mux_state) {
1058 case AD_MUX_DETACHED:
1059 __detach_bond_from_agg(port);
1060 port->actor_oper_port_state &= ~AD_STATE_SYNCHRONIZATION;
1061 ad_disable_collecting_distributing(port);
1062 port->actor_oper_port_state &= ~AD_STATE_COLLECTING;
1063 port->actor_oper_port_state &= ~AD_STATE_DISTRIBUTING;
1064 port->ntt = 1;
1065 break;
1066 case AD_MUX_WAITING:
1067 port->sm_mux_timer_counter = __ad_timer_to_ticks(AD_WAIT_WHILE_TIMER, 0);
1068 break;
1069 case AD_MUX_ATTACHED:
1070 __attach_bond_to_agg(port);
1071 port->actor_oper_port_state |= AD_STATE_SYNCHRONIZATION;
1072 port->actor_oper_port_state &= ~AD_STATE_COLLECTING;
1073 port->actor_oper_port_state &= ~AD_STATE_DISTRIBUTING;
1074 ad_disable_collecting_distributing(port);
1075 port->ntt = 1;
1076 break;
1077 case AD_MUX_COLLECTING_DISTRIBUTING:
1078 port->actor_oper_port_state |= AD_STATE_COLLECTING;
1079 port->actor_oper_port_state |= AD_STATE_DISTRIBUTING;
1080 ad_enable_collecting_distributing(port);
1081 port->ntt = 1;
1082 break;
1083 default: //to silence the compiler
1084 break;
1085 }
1086 }
1087}
1088
1089/**
1090 * ad_rx_machine - handle a port's rx State Machine
1091 * @lacpdu: the lacpdu we've received
1092 * @port: the port we're looking at
1093 *
1094 * If lacpdu arrived, stop previous timer (if exists) and set the next state as
1095 * CURRENT. If timer expired set the state machine in the proper state.
1096 * In other cases, this function checks if we need to switch to other state.
1097 */
1098static void ad_rx_machine(struct lacpdu *lacpdu, struct port *port)
1099{
1100 rx_states_t last_state;
1101
1102 // Lock to prevent 2 instances of this function to run simultaneously(rx interrupt and periodic machine callback)
1103 __get_rx_machine_lock(port);
1104
1105 // keep current State Machine state to compare later if it was changed
1106 last_state = port->sm_rx_state;
1107
1108 // check if state machine should change state
1109 // first, check if port was reinitialized
1110 if (port->sm_vars & AD_PORT_BEGIN) {
1111 port->sm_rx_state = AD_RX_INITIALIZE; // next state
1112 }
1113 // check if port is not enabled
1114 else if (!(port->sm_vars & AD_PORT_BEGIN) && !port->is_enabled && !(port->sm_vars & AD_PORT_MOVED)) {
1115 port->sm_rx_state = AD_RX_PORT_DISABLED; // next state
1116 }
1117 // check if new lacpdu arrived
1118 else if (lacpdu && ((port->sm_rx_state == AD_RX_EXPIRED) || (port->sm_rx_state == AD_RX_DEFAULTED) || (port->sm_rx_state == AD_RX_CURRENT))) {
1119 port->sm_rx_timer_counter = 0; // zero timer
1120 port->sm_rx_state = AD_RX_CURRENT;
1121 } else {
1122 // if timer is on, and if it is expired
1123 if (port->sm_rx_timer_counter && !(--port->sm_rx_timer_counter)) {
1124 switch (port->sm_rx_state) {
1125 case AD_RX_EXPIRED:
1126 port->sm_rx_state = AD_RX_DEFAULTED; // next state
1127 break;
1128 case AD_RX_CURRENT:
1129 port->sm_rx_state = AD_RX_EXPIRED; // next state
1130 break;
1131 default: //to silence the compiler
1132 break;
1133 }
1134 } else {
1135 // if no lacpdu arrived and no timer is on
1136 switch (port->sm_rx_state) {
1137 case AD_RX_PORT_DISABLED:
1138 if (port->sm_vars & AD_PORT_MOVED) {
1139 port->sm_rx_state = AD_RX_INITIALIZE; // next state
1140 } else if (port->is_enabled && (port->sm_vars & AD_PORT_LACP_ENABLED)) {
1141 port->sm_rx_state = AD_RX_EXPIRED; // next state
1142 } else if (port->is_enabled && ((port->sm_vars & AD_PORT_LACP_ENABLED) == 0)) {
1143 port->sm_rx_state = AD_RX_LACP_DISABLED; // next state
1144 }
1145 break;
1146 default: //to silence the compiler
1147 break;
1148
1149 }
1150 }
1151 }
1152
1153 // check if the State machine was changed or new lacpdu arrived
1154 if ((port->sm_rx_state != last_state) || (lacpdu)) {
1155 dprintk("Rx Machine: Port=%d, Last State=%d, Curr State=%d\n", port->actor_port_number, last_state, port->sm_rx_state);
1156 switch (port->sm_rx_state) {
1157 case AD_RX_INITIALIZE:
1158 if (!(port->actor_oper_port_key & AD_DUPLEX_KEY_BITS)) {
1159 port->sm_vars &= ~AD_PORT_LACP_ENABLED;
1160 } else {
1161 port->sm_vars |= AD_PORT_LACP_ENABLED;
1162 }
1163 port->sm_vars &= ~AD_PORT_SELECTED;
1164 __record_default(port);
1165 port->actor_oper_port_state &= ~AD_STATE_EXPIRED;
1166 port->sm_vars &= ~AD_PORT_MOVED;
1167 port->sm_rx_state = AD_RX_PORT_DISABLED; // next state
1168
1169 /*- Fall Through -*/
1170
1171 case AD_RX_PORT_DISABLED:
1172 port->sm_vars &= ~AD_PORT_MATCHED;
1173 break;
1174 case AD_RX_LACP_DISABLED:
1175 port->sm_vars &= ~AD_PORT_SELECTED;
1176 __record_default(port);
1177 port->partner_oper_port_state &= ~AD_STATE_AGGREGATION;
1178 port->sm_vars |= AD_PORT_MATCHED;
1179 port->actor_oper_port_state &= ~AD_STATE_EXPIRED;
1180 break;
1181 case AD_RX_EXPIRED:
1182 //Reset of the Synchronization flag. (Standard 43.4.12)
1183 //This reset cause to disable this port in the COLLECTING_DISTRIBUTING state of the
1184 //mux machine in case of EXPIRED even if LINK_DOWN didn't arrive for the port.
1185 port->partner_oper_port_state &= ~AD_STATE_SYNCHRONIZATION;
1186 port->sm_vars &= ~AD_PORT_MATCHED;
1187 port->partner_oper_port_state |= AD_SHORT_TIMEOUT;
1188 port->sm_rx_timer_counter = __ad_timer_to_ticks(AD_CURRENT_WHILE_TIMER, (u16)(AD_SHORT_TIMEOUT));
1189 port->actor_oper_port_state |= AD_STATE_EXPIRED;
1190 break;
1191 case AD_RX_DEFAULTED:
1192 __update_default_selected(port);
1193 __record_default(port);
1194 port->sm_vars |= AD_PORT_MATCHED;
1195 port->actor_oper_port_state &= ~AD_STATE_EXPIRED;
1196 break;
1197 case AD_RX_CURRENT:
1198 // detect loopback situation
1199 if (!MAC_ADDRESS_COMPARE(&(lacpdu->actor_system), &(port->actor_system))) {
1200 // INFO_RECEIVED_LOOPBACK_FRAMES
1201 printk(KERN_ERR DRV_NAME ": An illegal loopback occurred on adapter (%s)\n",
1202 port->slave->dev->name);
1203 printk(KERN_ERR "Check the configuration to verify that all Adapters "
1204 "are connected to 802.3ad compliant switch ports\n");
1205 __release_rx_machine_lock(port);
1206 return;
1207 }
1208 __update_selected(lacpdu, port);
1209 __update_ntt(lacpdu, port);
1210 __record_pdu(lacpdu, port);
1211 __choose_matched(lacpdu, port);
1212 port->sm_rx_timer_counter = __ad_timer_to_ticks(AD_CURRENT_WHILE_TIMER, (u16)(port->actor_oper_port_state & AD_STATE_LACP_TIMEOUT));
1213 port->actor_oper_port_state &= ~AD_STATE_EXPIRED;
1214 // verify that if the aggregator is enabled, the port is enabled too.
1215 //(because if the link goes down for a short time, the 802.3ad will not
1216 // catch it, and the port will continue to be disabled)
1217 if (port->aggregator && port->aggregator->is_active && !__port_is_enabled(port)) {
1218 __enable_port(port);
1219 }
1220 break;
1221 default: //to silence the compiler
1222 break;
1223 }
1224 }
1225 __release_rx_machine_lock(port);
1226}
1227
1228/**
1229 * ad_tx_machine - handle a port's tx state machine
1230 * @port: the port we're looking at
1231 *
1232 */
1233static void ad_tx_machine(struct port *port)
1234{
1235 // check if tx timer expired, to verify that we do not send more than 3 packets per second
1236 if (port->sm_tx_timer_counter && !(--port->sm_tx_timer_counter)) {
1237 // check if there is something to send
1238 if (port->ntt && (port->sm_vars & AD_PORT_LACP_ENABLED)) {
1239 __update_lacpdu_from_port(port);
1240 // send the lacpdu
1241 if (ad_lacpdu_send(port) >= 0) {
1242 dprintk("Sent LACPDU on port %d\n", port->actor_port_number);
1243 // mark ntt as false, so it will not be sent again until demanded
1244 port->ntt = 0;
1245 }
1246 }
1247 // restart tx timer(to verify that we will not exceed AD_MAX_TX_IN_SECOND
1248 port->sm_tx_timer_counter=ad_ticks_per_sec/AD_MAX_TX_IN_SECOND;
1249 }
1250}
1251
1252/**
1253 * ad_periodic_machine - handle a port's periodic state machine
1254 * @port: the port we're looking at
1255 *
1256 * Turn ntt flag on priodically to perform periodic transmission of lacpdu's.
1257 */
1258static void ad_periodic_machine(struct port *port)
1259{
1260 periodic_states_t last_state;
1261
1262 // keep current state machine state to compare later if it was changed
1263 last_state = port->sm_periodic_state;
1264
1265 // check if port was reinitialized
1266 if (((port->sm_vars & AD_PORT_BEGIN) || !(port->sm_vars & AD_PORT_LACP_ENABLED) || !port->is_enabled) ||
1267 (!(port->actor_oper_port_state & AD_STATE_LACP_ACTIVITY) && !(port->partner_oper_port_state & AD_STATE_LACP_ACTIVITY))
1268 ) {
1269 port->sm_periodic_state = AD_NO_PERIODIC; // next state
1270 }
1271 // check if state machine should change state
1272 else if (port->sm_periodic_timer_counter) {
1273 // check if periodic state machine expired
1274 if (!(--port->sm_periodic_timer_counter)) {
1275 // if expired then do tx
1276 port->sm_periodic_state = AD_PERIODIC_TX; // next state
1277 } else {
1278 // If not expired, check if there is some new timeout parameter from the partner state
1279 switch (port->sm_periodic_state) {
1280 case AD_FAST_PERIODIC:
1281 if (!(port->partner_oper_port_state & AD_STATE_LACP_TIMEOUT)) {
1282 port->sm_periodic_state = AD_SLOW_PERIODIC; // next state
1283 }
1284 break;
1285 case AD_SLOW_PERIODIC:
1286 if ((port->partner_oper_port_state & AD_STATE_LACP_TIMEOUT)) {
1287 // stop current timer
1288 port->sm_periodic_timer_counter = 0;
1289 port->sm_periodic_state = AD_PERIODIC_TX; // next state
1290 }
1291 break;
1292 default: //to silence the compiler
1293 break;
1294 }
1295 }
1296 } else {
1297 switch (port->sm_periodic_state) {
1298 case AD_NO_PERIODIC:
1299 port->sm_periodic_state = AD_FAST_PERIODIC; // next state
1300 break;
1301 case AD_PERIODIC_TX:
1302 if (!(port->partner_oper_port_state & AD_STATE_LACP_TIMEOUT)) {
1303 port->sm_periodic_state = AD_SLOW_PERIODIC; // next state
1304 } else {
1305 port->sm_periodic_state = AD_FAST_PERIODIC; // next state
1306 }
1307 break;
1308 default: //to silence the compiler
1309 break;
1310 }
1311 }
1312
1313 // check if the state machine was changed
1314 if (port->sm_periodic_state != last_state) {
1315 dprintk("Periodic Machine: Port=%d, Last State=%d, Curr State=%d\n", port->actor_port_number, last_state, port->sm_periodic_state);
1316 switch (port->sm_periodic_state) {
1317 case AD_NO_PERIODIC:
1318 port->sm_periodic_timer_counter = 0; // zero timer
1319 break;
1320 case AD_FAST_PERIODIC:
1321 port->sm_periodic_timer_counter = __ad_timer_to_ticks(AD_PERIODIC_TIMER, (u16)(AD_FAST_PERIODIC_TIME))-1; // decrement 1 tick we lost in the PERIODIC_TX cycle
1322 break;
1323 case AD_SLOW_PERIODIC:
1324 port->sm_periodic_timer_counter = __ad_timer_to_ticks(AD_PERIODIC_TIMER, (u16)(AD_SLOW_PERIODIC_TIME))-1; // decrement 1 tick we lost in the PERIODIC_TX cycle
1325 break;
1326 case AD_PERIODIC_TX:
1327 port->ntt = 1;
1328 break;
1329 default: //to silence the compiler
1330 break;
1331 }
1332 }
1333}
1334
1335/**
1336 * ad_port_selection_logic - select aggregation groups
1337 * @port: the port we're looking at
1338 *
1339 * Select aggregation groups, and assign each port for it's aggregetor. The
1340 * selection logic is called in the inititalization (after all the handshkes),
1341 * and after every lacpdu receive (if selected is off).
1342 */
1343static void ad_port_selection_logic(struct port *port)
1344{
1345 struct aggregator *aggregator, *free_aggregator = NULL, *temp_aggregator;
1346 struct port *last_port = NULL, *curr_port;
1347 int found = 0;
1348
1349 // if the port is already Selected, do nothing
1350 if (port->sm_vars & AD_PORT_SELECTED) {
1351 return;
1352 }
1353
1354 // if the port is connected to other aggregator, detach it
1355 if (port->aggregator) {
1356 // detach the port from its former aggregator
1357 temp_aggregator=port->aggregator;
1358 for (curr_port=temp_aggregator->lag_ports; curr_port; last_port=curr_port, curr_port=curr_port->next_port_in_aggregator) {
1359 if (curr_port == port) {
1360 temp_aggregator->num_of_ports--;
1361 if (!last_port) {// if it is the first port attached to the aggregator
1362 temp_aggregator->lag_ports=port->next_port_in_aggregator;
1363 } else {// not the first port attached to the aggregator
1364 last_port->next_port_in_aggregator=port->next_port_in_aggregator;
1365 }
1366
1367 // clear the port's relations to this aggregator
1368 port->aggregator = NULL;
1369 port->next_port_in_aggregator=NULL;
1370 port->actor_port_aggregator_identifier=0;
1371
1372 dprintk("Port %d left LAG %d\n", port->actor_port_number, temp_aggregator->aggregator_identifier);
1373 // if the aggregator is empty, clear its parameters, and set it ready to be attached
1374 if (!temp_aggregator->lag_ports) {
1375 ad_clear_agg(temp_aggregator);
1376 }
1377 break;
1378 }
1379 }
1380 if (!curr_port) { // meaning: the port was related to an aggregator but was not on the aggregator port list
1381 printk(KERN_WARNING DRV_NAME ": Warning: Port %d (on %s) was "
1382 "related to aggregator %d but was not on its port list\n",
1383 port->actor_port_number, port->slave->dev->name,
1384 port->aggregator->aggregator_identifier);
1385 }
1386 }
1387 // search on all aggregators for a suitable aggregator for this port
1388 for (aggregator = __get_first_agg(port); aggregator;
1389 aggregator = __get_next_agg(aggregator)) {
1390
1391 // keep a free aggregator for later use(if needed)
1392 if (!aggregator->lag_ports) {
1393 if (!free_aggregator) {
1394 free_aggregator=aggregator;
1395 }
1396 continue;
1397 }
1398 // check if current aggregator suits us
1399 if (((aggregator->actor_oper_aggregator_key == port->actor_oper_port_key) && // if all parameters match AND
1400 !MAC_ADDRESS_COMPARE(&(aggregator->partner_system), &(port->partner_oper_system)) &&
1401 (aggregator->partner_system_priority == port->partner_oper_system_priority) &&
1402 (aggregator->partner_oper_aggregator_key == port->partner_oper_key)
1403 ) &&
1404 ((MAC_ADDRESS_COMPARE(&(port->partner_oper_system), &(null_mac_addr)) && // partner answers
1405 !aggregator->is_individual) // but is not individual OR
1406 )
1407 ) {
1408 // attach to the founded aggregator
1409 port->aggregator = aggregator;
1410 port->actor_port_aggregator_identifier=port->aggregator->aggregator_identifier;
1411 port->next_port_in_aggregator=aggregator->lag_ports;
1412 port->aggregator->num_of_ports++;
1413 aggregator->lag_ports=port;
1414 dprintk("Port %d joined LAG %d(existing LAG)\n", port->actor_port_number, port->aggregator->aggregator_identifier);
1415
1416 // mark this port as selected
1417 port->sm_vars |= AD_PORT_SELECTED;
1418 found = 1;
1419 break;
1420 }
1421 }
1422
1423 // the port couldn't find an aggregator - attach it to a new aggregator
1424 if (!found) {
1425 if (free_aggregator) {
1426 // assign port a new aggregator
1427 port->aggregator = free_aggregator;
1428 port->actor_port_aggregator_identifier=port->aggregator->aggregator_identifier;
1429
1430 // update the new aggregator's parameters
1431 // if port was responsed from the end-user
1432 if (port->actor_oper_port_key & AD_DUPLEX_KEY_BITS) {// if port is full duplex
1433 port->aggregator->is_individual = 0;
1434 } else {
1435 port->aggregator->is_individual = 1;
1436 }
1437
1438 port->aggregator->actor_admin_aggregator_key = port->actor_admin_port_key;
1439 port->aggregator->actor_oper_aggregator_key = port->actor_oper_port_key;
1440 port->aggregator->partner_system=port->partner_oper_system;
1441 port->aggregator->partner_system_priority = port->partner_oper_system_priority;
1442 port->aggregator->partner_oper_aggregator_key = port->partner_oper_key;
1443 port->aggregator->receive_state = 1;
1444 port->aggregator->transmit_state = 1;
1445 port->aggregator->lag_ports = port;
1446 port->aggregator->num_of_ports++;
1447
1448 // mark this port as selected
1449 port->sm_vars |= AD_PORT_SELECTED;
1450
1451 dprintk("Port %d joined LAG %d(new LAG)\n", port->actor_port_number, port->aggregator->aggregator_identifier);
1452 } else {
1453 printk(KERN_ERR DRV_NAME ": Port %d (on %s) did not find a suitable aggregator\n",
1454 port->actor_port_number, port->slave->dev->name);
1455 }
1456 }
1457 // if all aggregator's ports are READY_N == TRUE, set ready=TRUE in all aggregator's ports
1458 // else set ready=FALSE in all aggregator's ports
1459 __set_agg_ports_ready(port->aggregator, __agg_ports_are_ready(port->aggregator));
1460
1461 if (!__check_agg_selection_timer(port) && (aggregator = __get_first_agg(port))) {
1462 ad_agg_selection_logic(aggregator);
1463 }
1464}
1465
1466/**
1467 * ad_agg_selection_logic - select an aggregation group for a team
1468 * @aggregator: the aggregator we're looking at
1469 *
1470 * It is assumed that only one aggregator may be selected for a team.
1471 * The logic of this function is to select (at first time) the aggregator with
1472 * the most ports attached to it, and to reselect the active aggregator only if
1473 * the previous aggregator has no more ports related to it.
1474 *
1475 * FIXME: this function MUST be called with the first agg in the bond, or
1476 * __get_active_agg() won't work correctly. This function should be better
1477 * called with the bond itself, and retrieve the first agg from it.
1478 */
1479static void ad_agg_selection_logic(struct aggregator *aggregator)
1480{
1481 struct aggregator *best_aggregator = NULL, *active_aggregator = NULL;
1482 struct aggregator *last_active_aggregator = NULL, *origin_aggregator;
1483 struct port *port;
1484 u16 num_of_aggs=0;
1485
1486 origin_aggregator = aggregator;
1487
1488 //get current active aggregator
1489 last_active_aggregator = __get_active_agg(aggregator);
1490
1491 // search for the aggregator with the most ports attached to it.
1492 do {
1493 // count how many candidate lag's we have
1494 if (aggregator->lag_ports) {
1495 num_of_aggs++;
1496 }
1497 if (aggregator->is_active && !aggregator->is_individual && // if current aggregator is the active aggregator
1498 MAC_ADDRESS_COMPARE(&(aggregator->partner_system), &(null_mac_addr))) { // and partner answers to 802.3ad PDUs
1499 if (aggregator->num_of_ports) { // if any ports attached to the current aggregator
1500 best_aggregator=NULL; // disregard the best aggregator that was chosen by now
1501 break; // stop the selection of other aggregator if there are any ports attached to this active aggregator
1502 } else { // no ports attached to this active aggregator
1503 aggregator->is_active = 0; // mark this aggregator as not active anymore
1504 }
1505 }
1506 if (aggregator->num_of_ports) { // if any ports attached
1507 if (best_aggregator) { // if there is a candidte aggregator
1508 //The reasons for choosing new best aggregator:
1509 // 1. if current agg is NOT individual and the best agg chosen so far is individual OR
1510 // current and best aggs are both individual or both not individual, AND
1511 // 2a. current agg partner reply but best agg partner do not reply OR
1512 // 2b. current agg partner reply OR current agg partner do not reply AND best agg partner also do not reply AND
1513 // current has more ports/bandwidth, or same amount of ports but current has faster ports, THEN
1514 // current agg become best agg so far
1515
1516 //if current agg is NOT individual and the best agg chosen so far is individual change best_aggregator
1517 if (!aggregator->is_individual && best_aggregator->is_individual) {
1518 best_aggregator=aggregator;
1519 }
1520 // current and best aggs are both individual or both not individual
1521 else if ((aggregator->is_individual && best_aggregator->is_individual) ||
1522 (!aggregator->is_individual && !best_aggregator->is_individual)) {
1523 // current and best aggs are both individual or both not individual AND
1524 // current agg partner reply but best agg partner do not reply
1525 if ((MAC_ADDRESS_COMPARE(&(aggregator->partner_system), &(null_mac_addr)) &&
1526 !MAC_ADDRESS_COMPARE(&(best_aggregator->partner_system), &(null_mac_addr)))) {
1527 best_aggregator=aggregator;
1528 }
1529 // current agg partner reply OR current agg partner do not reply AND best agg partner also do not reply
1530 else if (! (!MAC_ADDRESS_COMPARE(&(aggregator->partner_system), &(null_mac_addr)) &&
1531 MAC_ADDRESS_COMPARE(&(best_aggregator->partner_system), &(null_mac_addr)))) {
1532 if ((__get_agg_selection_mode(aggregator->lag_ports) == AD_BANDWIDTH)&&
1533 (__get_agg_bandwidth(aggregator) > __get_agg_bandwidth(best_aggregator))) {
1534 best_aggregator=aggregator;
1535 } else if (__get_agg_selection_mode(aggregator->lag_ports) == AD_COUNT) {
1536 if (((aggregator->num_of_ports > best_aggregator->num_of_ports) &&
1537 (aggregator->actor_oper_aggregator_key & AD_SPEED_KEY_BITS))||
1538 ((aggregator->num_of_ports == best_aggregator->num_of_ports) &&
1539 ((u16)(aggregator->actor_oper_aggregator_key & AD_SPEED_KEY_BITS) >
1540 (u16)(best_aggregator->actor_oper_aggregator_key & AD_SPEED_KEY_BITS)))) {
1541 best_aggregator=aggregator;
1542 }
1543 }
1544 }
1545 }
1546 } else {
1547 best_aggregator=aggregator;
1548 }
1549 }
1550 aggregator->is_active = 0; // mark all aggregators as not active anymore
1551 } while ((aggregator = __get_next_agg(aggregator)));
1552
1553 // if we have new aggregator selected, don't replace the old aggregator if it has an answering partner,
1554 // or if both old aggregator and new aggregator don't have answering partner
1555 if (best_aggregator) {
1556 if (last_active_aggregator && last_active_aggregator->lag_ports && last_active_aggregator->lag_ports->is_enabled &&
1557 (MAC_ADDRESS_COMPARE(&(last_active_aggregator->partner_system), &(null_mac_addr)) || // partner answers OR
1558 (!MAC_ADDRESS_COMPARE(&(last_active_aggregator->partner_system), &(null_mac_addr)) && // both old and new
1559 !MAC_ADDRESS_COMPARE(&(best_aggregator->partner_system), &(null_mac_addr)))) // partner do not answer
1560 ) {
1561 // if new aggregator has link, and old aggregator does not, replace old aggregator.(do nothing)
1562 // -> don't replace otherwise.
1563 if (!(!last_active_aggregator->actor_oper_aggregator_key && best_aggregator->actor_oper_aggregator_key)) {
1564 best_aggregator=NULL;
1565 last_active_aggregator->is_active = 1; // don't replace good old aggregator
1566
1567 }
1568 }
1569 }
1570
1571 // if there is new best aggregator, activate it
1572 if (best_aggregator) {
1573 for (aggregator = __get_first_agg(best_aggregator->lag_ports);
1574 aggregator;
1575 aggregator = __get_next_agg(aggregator)) {
1576
1577 dprintk("Agg=%d; Ports=%d; a key=%d; p key=%d; Indiv=%d; Active=%d\n",
1578 aggregator->aggregator_identifier, aggregator->num_of_ports,
1579 aggregator->actor_oper_aggregator_key, aggregator->partner_oper_aggregator_key,
1580 aggregator->is_individual, aggregator->is_active);
1581 }
1582
1583 // check if any partner replys
1584 if (best_aggregator->is_individual) {
1585 printk(KERN_WARNING DRV_NAME ": Warning: No 802.3ad response from the link partner "
1586 "for any adapters in the bond\n");
1587 }
1588
1589 // check if there are more than one aggregator
1590 if (num_of_aggs > 1) {
1591 dprintk("Warning: More than one Link Aggregation Group was "
1592 "found in the bond. Only one group will function in the bond\n");
1593 }
1594
1595 best_aggregator->is_active = 1;
1596 dprintk("LAG %d choosed as the active LAG\n", best_aggregator->aggregator_identifier);
1597 dprintk("Agg=%d; Ports=%d; a key=%d; p key=%d; Indiv=%d; Active=%d\n",
1598 best_aggregator->aggregator_identifier, best_aggregator->num_of_ports,
1599 best_aggregator->actor_oper_aggregator_key, best_aggregator->partner_oper_aggregator_key,
1600 best_aggregator->is_individual, best_aggregator->is_active);
1601
1602 // disable the ports that were related to the former active_aggregator
1603 if (last_active_aggregator) {
1604 for (port=last_active_aggregator->lag_ports; port; port=port->next_port_in_aggregator) {
1605 __disable_port(port);
1606 }
1607 }
1608 }
1609
1610 // if the selected aggregator is of join individuals(partner_system is NULL), enable their ports
1611 active_aggregator = __get_active_agg(origin_aggregator);
1612
1613 if (active_aggregator) {
1614 if (!MAC_ADDRESS_COMPARE(&(active_aggregator->partner_system), &(null_mac_addr))) {
1615 for (port=active_aggregator->lag_ports; port; port=port->next_port_in_aggregator) {
1616 __enable_port(port);
1617 }
1618 }
1619 }
1620}
1621
1622/**
1623 * ad_clear_agg - clear a given aggregator's parameters
1624 * @aggregator: the aggregator we're looking at
1625 *
1626 */
1627static void ad_clear_agg(struct aggregator *aggregator)
1628{
1629 if (aggregator) {
1630 aggregator->is_individual = 0;
1631 aggregator->actor_admin_aggregator_key = 0;
1632 aggregator->actor_oper_aggregator_key = 0;
1633 aggregator->partner_system = null_mac_addr;
1634 aggregator->partner_system_priority = 0;
1635 aggregator->partner_oper_aggregator_key = 0;
1636 aggregator->receive_state = 0;
1637 aggregator->transmit_state = 0;
1638 aggregator->lag_ports = NULL;
1639 aggregator->is_active = 0;
1640 aggregator->num_of_ports = 0;
1641 dprintk("LAG %d was cleared\n", aggregator->aggregator_identifier);
1642 }
1643}
1644
1645/**
1646 * ad_initialize_agg - initialize a given aggregator's parameters
1647 * @aggregator: the aggregator we're looking at
1648 *
1649 */
1650static void ad_initialize_agg(struct aggregator *aggregator)
1651{
1652 if (aggregator) {
1653 ad_clear_agg(aggregator);
1654
1655 aggregator->aggregator_mac_address = null_mac_addr;
1656 aggregator->aggregator_identifier = 0;
1657 aggregator->slave = NULL;
1658 }
1659}
1660
1661/**
1662 * ad_initialize_port - initialize a given port's parameters
1663 * @aggregator: the aggregator we're looking at
1664 * @lacp_fast: boolean. whether fast periodic should be used
1665 *
1666 */
1667static void ad_initialize_port(struct port *port, int lacp_fast)
1668{
1669 if (port) {
1670 port->actor_port_number = 1;
1671 port->actor_port_priority = 0xff;
1672 port->actor_system = null_mac_addr;
1673 port->actor_system_priority = 0xffff;
1674 port->actor_port_aggregator_identifier = 0;
1675 port->ntt = 0;
1676 port->actor_admin_port_key = 1;
1677 port->actor_oper_port_key = 1;
1678 port->actor_admin_port_state = AD_STATE_AGGREGATION | AD_STATE_LACP_ACTIVITY;
1679 port->actor_oper_port_state = AD_STATE_AGGREGATION | AD_STATE_LACP_ACTIVITY;
1680
1681 if (lacp_fast) {
1682 port->actor_oper_port_state |= AD_STATE_LACP_TIMEOUT;
1683 }
1684
1685 port->partner_admin_system = null_mac_addr;
1686 port->partner_oper_system = null_mac_addr;
1687 port->partner_admin_system_priority = 0xffff;
1688 port->partner_oper_system_priority = 0xffff;
1689 port->partner_admin_key = 1;
1690 port->partner_oper_key = 1;
1691 port->partner_admin_port_number = 1;
1692 port->partner_oper_port_number = 1;
1693 port->partner_admin_port_priority = 0xff;
1694 port->partner_oper_port_priority = 0xff;
1695 port->partner_admin_port_state = 1;
1696 port->partner_oper_port_state = 1;
1697 port->is_enabled = 1;
1698 // ****** private parameters ******
1699 port->sm_vars = 0x3;
1700 port->sm_rx_state = 0;
1701 port->sm_rx_timer_counter = 0;
1702 port->sm_periodic_state = 0;
1703 port->sm_periodic_timer_counter = 0;
1704 port->sm_mux_state = 0;
1705 port->sm_mux_timer_counter = 0;
1706 port->sm_tx_state = 0;
1707 port->sm_tx_timer_counter = 0;
1708 port->slave = NULL;
1709 port->aggregator = NULL;
1710 port->next_port_in_aggregator = NULL;
1711 port->transaction_id = 0;
1712
1713 ad_initialize_lacpdu(&(port->lacpdu));
1714 }
1715}
1716
1717/**
1718 * ad_enable_collecting_distributing - enable a port's transmit/receive
1719 * @port: the port we're looking at
1720 *
1721 * Enable @port if it's in an active aggregator
1722 */
1723static void ad_enable_collecting_distributing(struct port *port)
1724{
1725 if (port->aggregator->is_active) {
1726 dprintk("Enabling port %d(LAG %d)\n", port->actor_port_number, port->aggregator->aggregator_identifier);
1727 __enable_port(port);
1728 }
1729}
1730
1731/**
1732 * ad_disable_collecting_distributing - disable a port's transmit/receive
1733 * @port: the port we're looking at
1734 *
1735 */
1736static void ad_disable_collecting_distributing(struct port *port)
1737{
1738 if (port->aggregator && MAC_ADDRESS_COMPARE(&(port->aggregator->partner_system), &(null_mac_addr))) {
1739 dprintk("Disabling port %d(LAG %d)\n", port->actor_port_number, port->aggregator->aggregator_identifier);
1740 __disable_port(port);
1741 }
1742}
1743
1744#if 0
1745/**
1746 * ad_marker_info_send - send a marker information frame
1747 * @port: the port we're looking at
1748 *
1749 * This function does nothing since we decided not to implement send and handle
1750 * response for marker PDU's, in this stage, but only to respond to marker
1751 * information.
1752 */
1753static void ad_marker_info_send(struct port *port)
1754{
1755 struct marker marker;
1756 u16 index;
1757
1758 // fill the marker PDU with the appropriate values
1759 marker.subtype = 0x02;
1760 marker.version_number = 0x01;
1761 marker.tlv_type = AD_MARKER_INFORMATION_SUBTYPE;
1762 marker.marker_length = 0x16;
1763 // convert requester_port to Big Endian
1764 marker.requester_port = (((port->actor_port_number & 0xFF) << 8) |((u16)(port->actor_port_number & 0xFF00) >> 8));
1765 marker.requester_system = port->actor_system;
1766 // convert requester_port(u32) to Big Endian
1767 marker.requester_transaction_id = (((++port->transaction_id & 0xFF) << 24) |((port->transaction_id & 0xFF00) << 8) |((port->transaction_id & 0xFF0000) >> 8) |((port->transaction_id & 0xFF000000) >> 24));
1768 marker.pad = 0;
1769 marker.tlv_type_terminator = 0x00;
1770 marker.terminator_length = 0x00;
1771 for (index=0; index<90; index++) {
1772 marker.reserved_90[index]=0;
1773 }
1774
1775 // send the marker information
1776 if (ad_marker_send(port, &marker) >= 0) {
1777 dprintk("Sent Marker Information on port %d\n", port->actor_port_number);
1778 }
1779}
1780#endif
1781
1782/**
1783 * ad_marker_info_received - handle receive of a Marker information frame
1784 * @marker_info: Marker info received
1785 * @port: the port we're looking at
1786 *
1787 */
1788static void ad_marker_info_received(struct marker *marker_info,struct port *port)
1789{
1790 struct marker marker;
1791
1792 // copy the received marker data to the response marker
1793 //marker = *marker_info;
1794 memcpy(&marker, marker_info, sizeof(struct marker));
1795 // change the marker subtype to marker response
1796 marker.tlv_type=AD_MARKER_RESPONSE_SUBTYPE;
1797 // send the marker response
1798
1799 if (ad_marker_send(port, &marker) >= 0) {
1800 dprintk("Sent Marker Response on port %d\n", port->actor_port_number);
1801 }
1802}
1803
1804/**
1805 * ad_marker_response_received - handle receive of a marker response frame
1806 * @marker: marker PDU received
1807 * @port: the port we're looking at
1808 *
1809 * This function does nothing since we decided not to implement send and handle
1810 * response for marker PDU's, in this stage, but only to respond to marker
1811 * information.
1812 */
1813static void ad_marker_response_received(struct marker *marker, struct port *port)
1814{
1815 marker=NULL; // just to satisfy the compiler
1816 port=NULL; // just to satisfy the compiler
1817 // DO NOTHING, SINCE WE DECIDED NOT TO IMPLEMENT THIS FEATURE FOR NOW
1818}
1819
1820/**
1821 * ad_initialize_lacpdu - initialize a given lacpdu structure
1822 * @lacpdu: lacpdu structure to initialize
1823 *
1824 */
1825static void ad_initialize_lacpdu(struct lacpdu *lacpdu)
1826{
1827 u16 index;
1828
1829 // initialize lacpdu data
1830 lacpdu->subtype = 0x01;
1831 lacpdu->version_number = 0x01;
1832 lacpdu->tlv_type_actor_info = 0x01;
1833 lacpdu->actor_information_length = 0x14;
1834 // lacpdu->actor_system_priority updated on send
1835 // lacpdu->actor_system updated on send
1836 // lacpdu->actor_key updated on send
1837 // lacpdu->actor_port_priority updated on send
1838 // lacpdu->actor_port updated on send
1839 // lacpdu->actor_state updated on send
1840 lacpdu->tlv_type_partner_info = 0x02;
1841 lacpdu->partner_information_length = 0x14;
1842 for (index=0; index<=2; index++) {
1843 lacpdu->reserved_3_1[index]=0;
1844 }
1845 // lacpdu->partner_system_priority updated on send
1846 // lacpdu->partner_system updated on send
1847 // lacpdu->partner_key updated on send
1848 // lacpdu->partner_port_priority updated on send
1849 // lacpdu->partner_port updated on send
1850 // lacpdu->partner_state updated on send
1851 for (index=0; index<=2; index++) {
1852 lacpdu->reserved_3_2[index]=0;
1853 }
1854 lacpdu->tlv_type_collector_info = 0x03;
1855 lacpdu->collector_information_length= 0x10;
1856 lacpdu->collector_max_delay = AD_COLLECTOR_MAX_DELAY;
1857 for (index=0; index<=11; index++) {
1858 lacpdu->reserved_12[index]=0;
1859 }
1860 lacpdu->tlv_type_terminator = 0x00;
1861 lacpdu->terminator_length = 0;
1862 for (index=0; index<=49; index++) {
1863 lacpdu->reserved_50[index]=0;
1864 }
1865}
1866
1867//////////////////////////////////////////////////////////////////////////////////////
1868// ================= AD exported functions to the main bonding code ==================
1869//////////////////////////////////////////////////////////////////////////////////////
1870
1871// Check aggregators status in team every T seconds
1872#define AD_AGGREGATOR_SELECTION_TIMER 8
1873
1874static u16 aggregator_identifier;
1875
1876/**
1877 * bond_3ad_initialize - initialize a bond's 802.3ad parameters and structures
1878 * @bond: bonding struct to work on
1879 * @tick_resolution: tick duration (millisecond resolution)
1880 * @lacp_fast: boolean. whether fast periodic should be used
1881 *
1882 * Can be called only after the mac address of the bond is set.
1883 */
1884void bond_3ad_initialize(struct bonding *bond, u16 tick_resolution, int lacp_fast)
1885{
1886 // check that the bond is not initialized yet
1887 if (MAC_ADDRESS_COMPARE(&(BOND_AD_INFO(bond).system.sys_mac_addr), &(bond->dev->dev_addr))) {
1888
1889 aggregator_identifier = 0;
1890
1891 BOND_AD_INFO(bond).lacp_fast = lacp_fast;
1892 BOND_AD_INFO(bond).system.sys_priority = 0xFFFF;
1893 BOND_AD_INFO(bond).system.sys_mac_addr = *((struct mac_addr *)bond->dev->dev_addr);
1894
1895 // initialize how many times this module is called in one second(should be about every 100ms)
1896 ad_ticks_per_sec = tick_resolution;
1897
1898 // initialize the aggregator selection timer(to activate an aggregation selection after initialize)
1899 BOND_AD_INFO(bond).agg_select_timer = (AD_AGGREGATOR_SELECTION_TIMER * ad_ticks_per_sec);
1900 BOND_AD_INFO(bond).agg_select_mode = AD_BANDWIDTH;
1901 }
1902}
1903
1904/**
1905 * bond_3ad_bind_slave - initialize a slave's port
1906 * @slave: slave struct to work on
1907 *
1908 * Returns: 0 on success
1909 * < 0 on error
1910 */
1911int bond_3ad_bind_slave(struct slave *slave)
1912{
1913 struct bonding *bond = bond_get_bond_by_slave(slave);
1914 struct port *port;
1915 struct aggregator *aggregator;
1916
1917 if (bond == NULL) {
1918 printk(KERN_ERR "The slave %s is not attached to its bond\n", slave->dev->name);
1919 return -1;
1920 }
1921
1922 //check that the slave has not been intialized yet.
1923 if (SLAVE_AD_INFO(slave).port.slave != slave) {
1924
1925 // port initialization
1926 port = &(SLAVE_AD_INFO(slave).port);
1927
1928 ad_initialize_port(port, BOND_AD_INFO(bond).lacp_fast);
1929
1930 port->slave = slave;
1931 port->actor_port_number = SLAVE_AD_INFO(slave).id;
1932 // key is determined according to the link speed, duplex and user key(which is yet not supported)
1933 // ------------------------------------------------------------
1934 // Port key : | User key | Speed |Duplex|
1935 // ------------------------------------------------------------
1936 // 16 6 1 0
1937 port->actor_admin_port_key = 0; // initialize this parameter
1938 port->actor_admin_port_key |= __get_duplex(port);
1939 port->actor_admin_port_key |= (__get_link_speed(port) << 1);
1940 port->actor_oper_port_key = port->actor_admin_port_key;
1941 // if the port is not full duplex, then the port should be not lacp Enabled
1942 if (!(port->actor_oper_port_key & AD_DUPLEX_KEY_BITS)) {
1943 port->sm_vars &= ~AD_PORT_LACP_ENABLED;
1944 }
1945 // actor system is the bond's system
1946 port->actor_system = BOND_AD_INFO(bond).system.sys_mac_addr;
1947 // tx timer(to verify that no more than MAX_TX_IN_SECOND lacpdu's are sent in one second)
1948 port->sm_tx_timer_counter = ad_ticks_per_sec/AD_MAX_TX_IN_SECOND;
1949 port->aggregator = NULL;
1950 port->next_port_in_aggregator = NULL;
1951
1952 __disable_port(port);
1953 __initialize_port_locks(port);
1954
1955
1956 // aggregator initialization
1957 aggregator = &(SLAVE_AD_INFO(slave).aggregator);
1958
1959 ad_initialize_agg(aggregator);
1960
1961 aggregator->aggregator_mac_address = *((struct mac_addr *)bond->dev->dev_addr);
1962 aggregator->aggregator_identifier = (++aggregator_identifier);
1963 aggregator->slave = slave;
1964 aggregator->is_active = 0;
1965 aggregator->num_of_ports = 0;
1966 }
1967
1968 return 0;
1969}
1970
1971/**
1972 * bond_3ad_unbind_slave - deinitialize a slave's port
1973 * @slave: slave struct to work on
1974 *
1975 * Search for the aggregator that is related to this port, remove the
1976 * aggregator and assign another aggregator for other port related to it
1977 * (if any), and remove the port.
1978 */
1979void bond_3ad_unbind_slave(struct slave *slave)
1980{
1981 struct port *port, *prev_port, *temp_port;
1982 struct aggregator *aggregator, *new_aggregator, *temp_aggregator;
1983 int select_new_active_agg = 0;
1984
1985 // find the aggregator related to this slave
1986 aggregator = &(SLAVE_AD_INFO(slave).aggregator);
1987
1988 // find the port related to this slave
1989 port = &(SLAVE_AD_INFO(slave).port);
1990
1991 // if slave is null, the whole port is not initialized
1992 if (!port->slave) {
1993 printk(KERN_WARNING DRV_NAME ": Trying to unbind an uninitialized port on %s\n", slave->dev->name);
1994 return;
1995 }
1996
1997 dprintk("Unbinding Link Aggregation Group %d\n", aggregator->aggregator_identifier);
1998
1999 /* Tell the partner that this port is not suitable for aggregation */
2000 port->actor_oper_port_state &= ~AD_STATE_AGGREGATION;
2001 __update_lacpdu_from_port(port);
2002 ad_lacpdu_send(port);
2003
2004 // check if this aggregator is occupied
2005 if (aggregator->lag_ports) {
2006 // check if there are other ports related to this aggregator except
2007 // the port related to this slave(thats ensure us that there is a
2008 // reason to search for new aggregator, and that we will find one
2009 if ((aggregator->lag_ports != port) || (aggregator->lag_ports->next_port_in_aggregator)) {
2010 // find new aggregator for the related port(s)
2011 new_aggregator = __get_first_agg(port);
2012 for (; new_aggregator; new_aggregator = __get_next_agg(new_aggregator)) {
2013 // if the new aggregator is empty, or it connected to to our port only
2014 if (!new_aggregator->lag_ports || ((new_aggregator->lag_ports == port) && !new_aggregator->lag_ports->next_port_in_aggregator)) {
2015 break;
2016 }
2017 }
2018 // if new aggregator found, copy the aggregator's parameters
2019 // and connect the related lag_ports to the new aggregator
2020 if ((new_aggregator) && ((!new_aggregator->lag_ports) || ((new_aggregator->lag_ports == port) && !new_aggregator->lag_ports->next_port_in_aggregator))) {
2021 dprintk("Some port(s) related to LAG %d - replaceing with LAG %d\n", aggregator->aggregator_identifier, new_aggregator->aggregator_identifier);
2022
2023 if ((new_aggregator->lag_ports == port) && new_aggregator->is_active) {
2024 printk(KERN_INFO DRV_NAME ": Removing an active aggregator\n");
2025 // select new active aggregator
2026 select_new_active_agg = 1;
2027 }
2028
2029 new_aggregator->is_individual = aggregator->is_individual;
2030 new_aggregator->actor_admin_aggregator_key = aggregator->actor_admin_aggregator_key;
2031 new_aggregator->actor_oper_aggregator_key = aggregator->actor_oper_aggregator_key;
2032 new_aggregator->partner_system = aggregator->partner_system;
2033 new_aggregator->partner_system_priority = aggregator->partner_system_priority;
2034 new_aggregator->partner_oper_aggregator_key = aggregator->partner_oper_aggregator_key;
2035 new_aggregator->receive_state = aggregator->receive_state;
2036 new_aggregator->transmit_state = aggregator->transmit_state;
2037 new_aggregator->lag_ports = aggregator->lag_ports;
2038 new_aggregator->is_active = aggregator->is_active;
2039 new_aggregator->num_of_ports = aggregator->num_of_ports;
2040
2041 // update the information that is written on the ports about the aggregator
2042 for (temp_port=aggregator->lag_ports; temp_port; temp_port=temp_port->next_port_in_aggregator) {
2043 temp_port->aggregator=new_aggregator;
2044 temp_port->actor_port_aggregator_identifier = new_aggregator->aggregator_identifier;
2045 }
2046
2047 // clear the aggregator
2048 ad_clear_agg(aggregator);
2049
2050 if (select_new_active_agg) {
2051 ad_agg_selection_logic(__get_first_agg(port));
2052 }
2053 } else {
2054 printk(KERN_WARNING DRV_NAME ": Warning: unbinding aggregator, "
2055 "and could not find a new aggregator for its ports\n");
2056 }
2057 } else { // in case that the only port related to this aggregator is the one we want to remove
2058 select_new_active_agg = aggregator->is_active;
2059 // clear the aggregator
2060 ad_clear_agg(aggregator);
2061 if (select_new_active_agg) {
2062 printk(KERN_INFO "Removing an active aggregator\n");
2063 // select new active aggregator
2064 ad_agg_selection_logic(__get_first_agg(port));
2065 }
2066 }
2067 }
2068
2069 dprintk("Unbinding port %d\n", port->actor_port_number);
2070 // find the aggregator that this port is connected to
2071 temp_aggregator = __get_first_agg(port);
2072 for (; temp_aggregator; temp_aggregator = __get_next_agg(temp_aggregator)) {
2073 prev_port = NULL;
2074 // search the port in the aggregator's related ports
2075 for (temp_port=temp_aggregator->lag_ports; temp_port; prev_port=temp_port, temp_port=temp_port->next_port_in_aggregator) {
2076 if (temp_port == port) { // the aggregator found - detach the port from this aggregator
2077 if (prev_port) {
2078 prev_port->next_port_in_aggregator = temp_port->next_port_in_aggregator;
2079 } else {
2080 temp_aggregator->lag_ports = temp_port->next_port_in_aggregator;
2081 }
2082 temp_aggregator->num_of_ports--;
2083 if (temp_aggregator->num_of_ports==0) {
2084 select_new_active_agg = temp_aggregator->is_active;
2085 // clear the aggregator
2086 ad_clear_agg(temp_aggregator);
2087 if (select_new_active_agg) {
2088 printk(KERN_INFO "Removing an active aggregator\n");
2089 // select new active aggregator
2090 ad_agg_selection_logic(__get_first_agg(port));
2091 }
2092 }
2093 break;
2094 }
2095 }
2096 }
2097 port->slave=NULL;
2098}
2099
2100/**
2101 * bond_3ad_state_machine_handler - handle state machines timeout
2102 * @bond: bonding struct to work on
2103 *
2104 * The state machine handling concept in this module is to check every tick
2105 * which state machine should operate any function. The execution order is
2106 * round robin, so when we have an interaction between state machines, the
2107 * reply of one to each other might be delayed until next tick.
2108 *
2109 * This function also complete the initialization when the agg_select_timer
2110 * times out, and it selects an aggregator for the ports that are yet not
2111 * related to any aggregator, and selects the active aggregator for a bond.
2112 */
2113void bond_3ad_state_machine_handler(struct bonding *bond)
2114{
2115 struct port *port;
2116 struct aggregator *aggregator;
2117
2118 read_lock(&bond->lock);
2119
2120 if (bond->kill_timers) {
2121 goto out;
2122 }
2123
2124 //check if there are any slaves
2125 if (bond->slave_cnt == 0) {
2126 goto re_arm;
2127 }
2128
2129 // check if agg_select_timer timer after initialize is timed out
2130 if (BOND_AD_INFO(bond).agg_select_timer && !(--BOND_AD_INFO(bond).agg_select_timer)) {
2131 // select the active aggregator for the bond
2132 if ((port = __get_first_port(bond))) {
2133 if (!port->slave) {
2134 printk(KERN_WARNING DRV_NAME ": Warning: bond's first port is uninitialized\n");
2135 goto re_arm;
2136 }
2137
2138 aggregator = __get_first_agg(port);
2139 ad_agg_selection_logic(aggregator);
2140 }
2141 }
2142
2143 // for each port run the state machines
2144 for (port = __get_first_port(bond); port; port = __get_next_port(port)) {
2145 if (!port->slave) {
2146 printk(KERN_WARNING DRV_NAME ": Warning: Found an uninitialized port\n");
2147 goto re_arm;
2148 }
2149
2150 ad_rx_machine(NULL, port);
2151 ad_periodic_machine(port);
2152 ad_port_selection_logic(port);
2153 ad_mux_machine(port);
2154 ad_tx_machine(port);
2155
2156 // turn off the BEGIN bit, since we already handled it
2157 if (port->sm_vars & AD_PORT_BEGIN) {
2158 port->sm_vars &= ~AD_PORT_BEGIN;
2159 }
2160 }
2161
2162re_arm:
2163 mod_timer(&(BOND_AD_INFO(bond).ad_timer), jiffies + ad_delta_in_ticks);
2164out:
2165 read_unlock(&bond->lock);
2166}
2167
2168/**
2169 * bond_3ad_rx_indication - handle a received frame
2170 * @lacpdu: received lacpdu
2171 * @slave: slave struct to work on
2172 * @length: length of the data received
2173 *
2174 * It is assumed that frames that were sent on this NIC don't returned as new
2175 * received frames (loopback). Since only the payload is given to this
2176 * function, it check for loopback.
2177 */
2178static void bond_3ad_rx_indication(struct lacpdu *lacpdu, struct slave *slave, u16 length)
2179{
2180 struct port *port;
2181
2182 if (length >= sizeof(struct lacpdu)) {
2183
2184 port = &(SLAVE_AD_INFO(slave).port);
2185
2186 if (!port->slave) {
2187 printk(KERN_WARNING DRV_NAME ": Warning: port of slave %s is uninitialized\n", slave->dev->name);
2188 return;
2189 }
2190
2191 switch (lacpdu->subtype) {
2192 case AD_TYPE_LACPDU:
2193 __ntohs_lacpdu(lacpdu);
2194 dprintk("Received LACPDU on port %d\n", port->actor_port_number);
2195 ad_rx_machine(lacpdu, port);
2196 break;
2197
2198 case AD_TYPE_MARKER:
2199 // No need to convert fields to Little Endian since we don't use the marker's fields.
2200
2201 switch (((struct marker *)lacpdu)->tlv_type) {
2202 case AD_MARKER_INFORMATION_SUBTYPE:
2203 dprintk("Received Marker Information on port %d\n", port->actor_port_number);
2204 ad_marker_info_received((struct marker *)lacpdu, port);
2205 break;
2206
2207 case AD_MARKER_RESPONSE_SUBTYPE:
2208 dprintk("Received Marker Response on port %d\n", port->actor_port_number);
2209 ad_marker_response_received((struct marker *)lacpdu, port);
2210 break;
2211
2212 default:
2213 dprintk("Received an unknown Marker subtype on slot %d\n", port->actor_port_number);
2214 }
2215 }
2216 }
2217}
2218
2219/**
2220 * bond_3ad_adapter_speed_changed - handle a slave's speed change indication
2221 * @slave: slave struct to work on
2222 *
2223 * Handle reselection of aggregator (if needed) for this port.
2224 */
2225void bond_3ad_adapter_speed_changed(struct slave *slave)
2226{
2227 struct port *port;
2228
2229 port = &(SLAVE_AD_INFO(slave).port);
2230
2231 // if slave is null, the whole port is not initialized
2232 if (!port->slave) {
2233 printk(KERN_WARNING DRV_NAME ": Warning: speed changed for uninitialized port on %s\n",
2234 slave->dev->name);
2235 return;
2236 }
2237
2238 port->actor_admin_port_key &= ~AD_SPEED_KEY_BITS;
2239 port->actor_oper_port_key=port->actor_admin_port_key |= (__get_link_speed(port) << 1);
2240 dprintk("Port %d changed speed\n", port->actor_port_number);
2241 // there is no need to reselect a new aggregator, just signal the
2242 // state machines to reinitialize
2243 port->sm_vars |= AD_PORT_BEGIN;
2244}
2245
2246/**
2247 * bond_3ad_adapter_duplex_changed - handle a slave's duplex change indication
2248 * @slave: slave struct to work on
2249 *
2250 * Handle reselection of aggregator (if needed) for this port.
2251 */
2252void bond_3ad_adapter_duplex_changed(struct slave *slave)
2253{
2254 struct port *port;
2255
2256 port=&(SLAVE_AD_INFO(slave).port);
2257
2258 // if slave is null, the whole port is not initialized
2259 if (!port->slave) {
2260 printk(KERN_WARNING DRV_NAME ": Warning: duplex changed for uninitialized port on %s\n",
2261 slave->dev->name);
2262 return;
2263 }
2264
2265 port->actor_admin_port_key &= ~AD_DUPLEX_KEY_BITS;
2266 port->actor_oper_port_key=port->actor_admin_port_key |= __get_duplex(port);
2267 dprintk("Port %d changed duplex\n", port->actor_port_number);
2268 // there is no need to reselect a new aggregator, just signal the
2269 // state machines to reinitialize
2270 port->sm_vars |= AD_PORT_BEGIN;
2271}
2272
2273/**
2274 * bond_3ad_handle_link_change - handle a slave's link status change indication
2275 * @slave: slave struct to work on
2276 * @status: whether the link is now up or down
2277 *
2278 * Handle reselection of aggregator (if needed) for this port.
2279 */
2280void bond_3ad_handle_link_change(struct slave *slave, char link)
2281{
2282 struct port *port;
2283
2284 port = &(SLAVE_AD_INFO(slave).port);
2285
2286 // if slave is null, the whole port is not initialized
2287 if (!port->slave) {
2288 printk(KERN_WARNING DRV_NAME ": Warning: link status changed for uninitialized port on %s\n",
2289 slave->dev->name);
2290 return;
2291 }
2292
2293 // on link down we are zeroing duplex and speed since some of the adaptors(ce1000.lan) report full duplex/speed instead of N/A(duplex) / 0(speed)
2294 // on link up we are forcing recheck on the duplex and speed since some of he adaptors(ce1000.lan) report
2295 if (link == BOND_LINK_UP) {
2296 port->is_enabled = 1;
2297 port->actor_admin_port_key &= ~AD_DUPLEX_KEY_BITS;
2298 port->actor_oper_port_key=port->actor_admin_port_key |= __get_duplex(port);
2299 port->actor_admin_port_key &= ~AD_SPEED_KEY_BITS;
2300 port->actor_oper_port_key=port->actor_admin_port_key |= (__get_link_speed(port) << 1);
2301 } else {
2302 /* link has failed */
2303 port->is_enabled = 0;
2304 port->actor_admin_port_key &= ~AD_DUPLEX_KEY_BITS;
2305 port->actor_oper_port_key= (port->actor_admin_port_key &= ~AD_SPEED_KEY_BITS);
2306 }
2307 //BOND_PRINT_DBG(("Port %d changed link status to %s", port->actor_port_number, ((link == BOND_LINK_UP)?"UP":"DOWN")));
2308 // there is no need to reselect a new aggregator, just signal the
2309 // state machines to reinitialize
2310 port->sm_vars |= AD_PORT_BEGIN;
2311}
2312
2313/**
2314 * bond_3ad_get_active_agg_info - get information of the active aggregator
2315 * @bond: bonding struct to work on
2316 * @ad_info: ad_info struct to fill with the bond's info
2317 *
2318 * Returns: 0 on success
2319 * < 0 on error
2320 */
2321int bond_3ad_get_active_agg_info(struct bonding *bond, struct ad_info *ad_info)
2322{
2323 struct aggregator *aggregator = NULL;
2324 struct port *port;
2325
2326 for (port = __get_first_port(bond); port; port = __get_next_port(port)) {
2327 if (port->aggregator && port->aggregator->is_active) {
2328 aggregator = port->aggregator;
2329 break;
2330 }
2331 }
2332
2333 if (aggregator) {
2334 ad_info->aggregator_id = aggregator->aggregator_identifier;
2335 ad_info->ports = aggregator->num_of_ports;
2336 ad_info->actor_key = aggregator->actor_oper_aggregator_key;
2337 ad_info->partner_key = aggregator->partner_oper_aggregator_key;
2338 memcpy(ad_info->partner_system, aggregator->partner_system.mac_addr_value, ETH_ALEN);
2339 return 0;
2340 }
2341
2342 return -1;
2343}
2344
2345int bond_3ad_xmit_xor(struct sk_buff *skb, struct net_device *dev)
2346{
2347 struct slave *slave, *start_at;
2348 struct bonding *bond = dev->priv;
2349 struct ethhdr *data = (struct ethhdr *)skb->data;
2350 int slave_agg_no;
2351 int slaves_in_agg;
2352 int agg_id;
2353 int i;
2354 struct ad_info ad_info;
2355 int res = 1;
2356
2357 /* make sure that the slaves list will
2358 * not change during tx
2359 */
2360 read_lock(&bond->lock);
2361
2362 if (!BOND_IS_OK(bond)) {
2363 goto out;
2364 }
2365
2366 if (bond_3ad_get_active_agg_info(bond, &ad_info)) {
2367 printk(KERN_DEBUG "ERROR: bond_3ad_get_active_agg_info failed\n");
2368 goto out;
2369 }
2370
2371 slaves_in_agg = ad_info.ports;
2372 agg_id = ad_info.aggregator_id;
2373
2374 if (slaves_in_agg == 0) {
2375 /*the aggregator is empty*/
2376 printk(KERN_DEBUG "ERROR: active aggregator is empty\n");
2377 goto out;
2378 }
2379
2380 slave_agg_no = (data->h_dest[5]^bond->dev->dev_addr[5]) % slaves_in_agg;
2381
2382 bond_for_each_slave(bond, slave, i) {
2383 struct aggregator *agg = SLAVE_AD_INFO(slave).port.aggregator;
2384
2385 if (agg && (agg->aggregator_identifier == agg_id)) {
2386 slave_agg_no--;
2387 if (slave_agg_no < 0) {
2388 break;
2389 }
2390 }
2391 }
2392
2393 if (slave_agg_no >= 0) {
2394 printk(KERN_ERR DRV_NAME ": Error: Couldn't find a slave to tx on for aggregator ID %d\n", agg_id);
2395 goto out;
2396 }
2397
2398 start_at = slave;
2399
2400 bond_for_each_slave_from(bond, slave, i, start_at) {
2401 int slave_agg_id = 0;
2402 struct aggregator *agg = SLAVE_AD_INFO(slave).port.aggregator;
2403
2404 if (agg) {
2405 slave_agg_id = agg->aggregator_identifier;
2406 }
2407
2408 if (SLAVE_IS_OK(slave) && agg && (slave_agg_id == agg_id)) {
2409 res = bond_dev_queue_xmit(bond, skb, slave->dev);
2410 break;
2411 }
2412 }
2413
2414out:
2415 if (res) {
2416 /* no suitable interface, frame not sent */
2417 dev_kfree_skb(skb);
2418 }
2419 read_unlock(&bond->lock);
2420 return 0;
2421}
2422
2423int bond_3ad_lacpdu_recv(struct sk_buff *skb, struct net_device *dev, struct packet_type* ptype)
2424{
2425 struct bonding *bond = dev->priv;
2426 struct slave *slave = NULL;
2427 int ret = NET_RX_DROP;
2428
2429 if (!(dev->flags & IFF_MASTER)) {
2430 goto out;
2431 }
2432
2433 read_lock(&bond->lock);
2434 slave = bond_get_slave_by_dev((struct bonding *)dev->priv,
2435 skb->real_dev);
2436 if (slave == NULL) {
2437 goto out_unlock;
2438 }
2439
2440 bond_3ad_rx_indication((struct lacpdu *) skb->data, slave, skb->len);
2441
2442 ret = NET_RX_SUCCESS;
2443
2444out_unlock:
2445 read_unlock(&bond->lock);
2446out:
2447 dev_kfree_skb(skb);
2448
2449 return ret;
2450}
2451
diff --git a/drivers/net/bonding/bond_3ad.h b/drivers/net/bonding/bond_3ad.h
new file mode 100644
index 000000000000..f46823894187
--- /dev/null
+++ b/drivers/net/bonding/bond_3ad.h
@@ -0,0 +1,300 @@
1/*
2 * Copyright(c) 1999 - 2004 Intel Corporation. All rights reserved.
3 *
4 * This program is free software; you can redistribute it and/or modify it
5 * under the terms of the GNU General Public License as published by the Free
6 * Software Foundation; either version 2 of the License, or (at your option)
7 * any later version.
8 *
9 * This program is distributed in the hope that it will be useful, but WITHOUT
10 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
11 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for
12 * more details.
13 *
14 * You should have received a copy of the GNU General Public License along with
15 * this program; if not, write to the Free Software Foundation, Inc., 59
16 * Temple Place - Suite 330, Boston, MA 02111-1307, USA.
17 *
18 * The full GNU General Public License is included in this distribution in the
19 * file called LICENSE.
20 *
21 *
22 * Changes:
23 *
24 * 2003/05/01 - Tsippy Mendelson <tsippy.mendelson at intel dot com> and
25 * Amir Noam <amir.noam at intel dot com>
26 * - Added support for lacp_rate module param.
27 *
28 * 2003/05/01 - Shmulik Hen <shmulik.hen at intel dot com>
29 * - Renamed bond_3ad_link_status_changed() to
30 * bond_3ad_handle_link_change() for compatibility with TLB.
31 *
32 * 2003/12/01 - Shmulik Hen <shmulik.hen at intel dot com>
33 * - Code cleanup and style changes
34 */
35
36#ifndef __BOND_3AD_H__
37#define __BOND_3AD_H__
38
39#include <asm/byteorder.h>
40#include <linux/skbuff.h>
41#include <linux/netdevice.h>
42
43// General definitions
44#define BOND_ETH_P_LACPDU 0x8809
45#define PKT_TYPE_LACPDU __constant_htons(BOND_ETH_P_LACPDU)
46#define AD_TIMER_INTERVAL 100 /*msec*/
47
48#define MULTICAST_LACPDU_ADDR {0x01, 0x80, 0xC2, 0x00, 0x00, 0x02}
49#define AD_MULTICAST_LACPDU_ADDR {MULTICAST_LACPDU_ADDR}
50
51#define AD_LACP_SLOW 0
52#define AD_LACP_FAST 1
53
54typedef struct mac_addr {
55 u8 mac_addr_value[ETH_ALEN];
56} mac_addr_t;
57
58typedef enum {
59 AD_BANDWIDTH = 0,
60 AD_COUNT
61} agg_selection_t;
62
63// rx machine states(43.4.11 in the 802.3ad standard)
64typedef enum {
65 AD_RX_DUMMY,
66 AD_RX_INITIALIZE, // rx Machine
67 AD_RX_PORT_DISABLED, // rx Machine
68 AD_RX_LACP_DISABLED, // rx Machine
69 AD_RX_EXPIRED, // rx Machine
70 AD_RX_DEFAULTED, // rx Machine
71 AD_RX_CURRENT // rx Machine
72} rx_states_t;
73
74// periodic machine states(43.4.12 in the 802.3ad standard)
75typedef enum {
76 AD_PERIODIC_DUMMY,
77 AD_NO_PERIODIC, // periodic machine
78 AD_FAST_PERIODIC, // periodic machine
79 AD_SLOW_PERIODIC, // periodic machine
80 AD_PERIODIC_TX // periodic machine
81} periodic_states_t;
82
83// mux machine states(43.4.13 in the 802.3ad standard)
84typedef enum {
85 AD_MUX_DUMMY,
86 AD_MUX_DETACHED, // mux machine
87 AD_MUX_WAITING, // mux machine
88 AD_MUX_ATTACHED, // mux machine
89 AD_MUX_COLLECTING_DISTRIBUTING // mux machine
90} mux_states_t;
91
92// tx machine states(43.4.15 in the 802.3ad standard)
93typedef enum {
94 AD_TX_DUMMY,
95 AD_TRANSMIT // tx Machine
96} tx_states_t;
97
98// rx indication types
99typedef enum {
100 AD_TYPE_LACPDU = 1, // type lacpdu
101 AD_TYPE_MARKER // type marker
102} pdu_type_t;
103
104// rx marker indication types
105typedef enum {
106 AD_MARKER_INFORMATION_SUBTYPE = 1, // marker imformation subtype
107 AD_MARKER_RESPONSE_SUBTYPE // marker response subtype
108} marker_subtype_t;
109
110// timers types(43.4.9 in the 802.3ad standard)
111typedef enum {
112 AD_CURRENT_WHILE_TIMER,
113 AD_ACTOR_CHURN_TIMER,
114 AD_PERIODIC_TIMER,
115 AD_PARTNER_CHURN_TIMER,
116 AD_WAIT_WHILE_TIMER
117} ad_timers_t;
118
119#pragma pack(1)
120
121typedef struct ad_header {
122 struct mac_addr destination_address;
123 struct mac_addr source_address;
124 u16 length_type;
125} ad_header_t;
126
127// Link Aggregation Control Protocol(LACP) data unit structure(43.4.2.2 in the 802.3ad standard)
128typedef struct lacpdu {
129 u8 subtype; // = LACP(= 0x01)
130 u8 version_number;
131 u8 tlv_type_actor_info; // = actor information(type/length/value)
132 u8 actor_information_length; // = 20
133 u16 actor_system_priority;
134 struct mac_addr actor_system;
135 u16 actor_key;
136 u16 actor_port_priority;
137 u16 actor_port;
138 u8 actor_state;
139 u8 reserved_3_1[3]; // = 0
140 u8 tlv_type_partner_info; // = partner information
141 u8 partner_information_length; // = 20
142 u16 partner_system_priority;
143 struct mac_addr partner_system;
144 u16 partner_key;
145 u16 partner_port_priority;
146 u16 partner_port;
147 u8 partner_state;
148 u8 reserved_3_2[3]; // = 0
149 u8 tlv_type_collector_info; // = collector information
150 u8 collector_information_length; // = 16
151 u16 collector_max_delay;
152 u8 reserved_12[12];
153 u8 tlv_type_terminator; // = terminator
154 u8 terminator_length; // = 0
155 u8 reserved_50[50]; // = 0
156} lacpdu_t;
157
158typedef struct lacpdu_header {
159 struct ad_header ad_header;
160 struct lacpdu lacpdu;
161} lacpdu_header_t;
162
163// Marker Protocol Data Unit(PDU) structure(43.5.3.2 in the 802.3ad standard)
164typedef struct marker {
165 u8 subtype; // = 0x02 (marker PDU)
166 u8 version_number; // = 0x01
167 u8 tlv_type; // = 0x01 (marker information)
168 // = 0x02 (marker response information)
169 u8 marker_length; // = 0x16
170 u16 requester_port; // The number assigned to the port by the requester
171 struct mac_addr requester_system; // The requester's system id
172 u32 requester_transaction_id; // The transaction id allocated by the requester,
173 u16 pad; // = 0
174 u8 tlv_type_terminator; // = 0x00
175 u8 terminator_length; // = 0x00
176 u8 reserved_90[90]; // = 0
177} marker_t;
178
179typedef struct marker_header {
180 struct ad_header ad_header;
181 struct marker marker;
182} marker_header_t;
183
184#pragma pack()
185
186struct slave;
187struct bonding;
188struct ad_info;
189struct port;
190
191#ifdef __ia64__
192#pragma pack(8)
193#endif
194
195// aggregator structure(43.4.5 in the 802.3ad standard)
196typedef struct aggregator {
197 struct mac_addr aggregator_mac_address;
198 u16 aggregator_identifier;
199 u16 is_individual; // BOOLEAN
200 u16 actor_admin_aggregator_key;
201 u16 actor_oper_aggregator_key;
202 struct mac_addr partner_system;
203 u16 partner_system_priority;
204 u16 partner_oper_aggregator_key;
205 u16 receive_state; // BOOLEAN
206 u16 transmit_state; // BOOLEAN
207 struct port *lag_ports;
208 // ****** PRIVATE PARAMETERS ******
209 struct slave *slave; // pointer to the bond slave that this aggregator belongs to
210 u16 is_active; // BOOLEAN. Indicates if this aggregator is active
211 u16 num_of_ports;
212} aggregator_t;
213
214// port structure(43.4.6 in the 802.3ad standard)
215typedef struct port {
216 u16 actor_port_number;
217 u16 actor_port_priority;
218 struct mac_addr actor_system; // This parameter is added here although it is not specified in the standard, just for simplification
219 u16 actor_system_priority; // This parameter is added here although it is not specified in the standard, just for simplification
220 u16 actor_port_aggregator_identifier;
221 u16 ntt; // BOOLEAN
222 u16 actor_admin_port_key;
223 u16 actor_oper_port_key;
224 u8 actor_admin_port_state;
225 u8 actor_oper_port_state;
226 struct mac_addr partner_admin_system;
227 struct mac_addr partner_oper_system;
228 u16 partner_admin_system_priority;
229 u16 partner_oper_system_priority;
230 u16 partner_admin_key;
231 u16 partner_oper_key;
232 u16 partner_admin_port_number;
233 u16 partner_oper_port_number;
234 u16 partner_admin_port_priority;
235 u16 partner_oper_port_priority;
236 u8 partner_admin_port_state;
237 u8 partner_oper_port_state;
238 u16 is_enabled; // BOOLEAN
239 // ****** PRIVATE PARAMETERS ******
240 u16 sm_vars; // all state machines variables for this port
241 rx_states_t sm_rx_state; // state machine rx state
242 u16 sm_rx_timer_counter; // state machine rx timer counter
243 periodic_states_t sm_periodic_state;// state machine periodic state
244 u16 sm_periodic_timer_counter; // state machine periodic timer counter
245 mux_states_t sm_mux_state; // state machine mux state
246 u16 sm_mux_timer_counter; // state machine mux timer counter
247 tx_states_t sm_tx_state; // state machine tx state
248 u16 sm_tx_timer_counter; // state machine tx timer counter(allways on - enter to transmit state 3 time per second)
249 struct slave *slave; // pointer to the bond slave that this port belongs to
250 struct aggregator *aggregator; // pointer to an aggregator that this port related to
251 struct port *next_port_in_aggregator; // Next port on the linked list of the parent aggregator
252 u32 transaction_id; // continuous number for identification of Marker PDU's;
253 struct lacpdu lacpdu; // the lacpdu that will be sent for this port
254} port_t;
255
256// system structure
257typedef struct ad_system {
258 u16 sys_priority;
259 struct mac_addr sys_mac_addr;
260} ad_system_t;
261
262#ifdef __ia64__
263#pragma pack()
264#endif
265
266// ================= AD Exported structures to the main bonding code ==================
267#define BOND_AD_INFO(bond) ((bond)->ad_info)
268#define SLAVE_AD_INFO(slave) ((slave)->ad_info)
269
270struct ad_bond_info {
271 ad_system_t system; // 802.3ad system structure
272 u32 agg_select_timer; // Timer to select aggregator after all adapter's hand shakes
273 u32 agg_select_mode; // Mode of selection of active aggregator(bandwidth/count)
274 int lacp_fast; /* whether fast periodic tx should be
275 * requested
276 */
277 struct timer_list ad_timer;
278 struct packet_type ad_pkt_type;
279};
280
281struct ad_slave_info {
282 struct aggregator aggregator; // 802.3ad aggregator structure
283 struct port port; // 802.3ad port structure
284 spinlock_t rx_machine_lock; // To avoid race condition between callback and receive interrupt
285 u16 id;
286};
287
288// ================= AD Exported functions to the main bonding code ==================
289void bond_3ad_initialize(struct bonding *bond, u16 tick_resolution, int lacp_fast);
290int bond_3ad_bind_slave(struct slave *slave);
291void bond_3ad_unbind_slave(struct slave *slave);
292void bond_3ad_state_machine_handler(struct bonding *bond);
293void bond_3ad_adapter_speed_changed(struct slave *slave);
294void bond_3ad_adapter_duplex_changed(struct slave *slave);
295void bond_3ad_handle_link_change(struct slave *slave, char link);
296int bond_3ad_get_active_agg_info(struct bonding *bond, struct ad_info *ad_info);
297int bond_3ad_xmit_xor(struct sk_buff *skb, struct net_device *dev);
298int bond_3ad_lacpdu_recv(struct sk_buff *skb, struct net_device *dev, struct packet_type* ptype);
299#endif //__BOND_3AD_H__
300
diff --git a/drivers/net/bonding/bond_alb.c b/drivers/net/bonding/bond_alb.c
new file mode 100644
index 000000000000..5ce606d9dc03
--- /dev/null
+++ b/drivers/net/bonding/bond_alb.c
@@ -0,0 +1,1696 @@
1/*
2 * Copyright(c) 1999 - 2004 Intel Corporation. All rights reserved.
3 *
4 * This program is free software; you can redistribute it and/or modify it
5 * under the terms of the GNU General Public License as published by the
6 * Free Software Foundation; either version 2 of the License, or
7 * (at your option) any later version.
8 *
9 * This program is distributed in the hope that it will be useful, but
10 * WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY
11 * or FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
12 * for more details.
13 *
14 * You should have received a copy of the GNU General Public License along
15 * with this program; if not, write to the Free Software Foundation, Inc.,
16 * 59 Temple Place - Suite 330, Boston, MA 02111-1307, USA.
17 *
18 * The full GNU General Public License is included in this distribution in the
19 * file called LICENSE.
20 *
21 *
22 * Changes:
23 *
24 * 2003/06/25 - Shmulik Hen <shmulik.hen at intel dot com>
25 * - Fixed signed/unsigned calculation errors that caused load sharing
26 * to collapse to one slave under very heavy UDP Tx stress.
27 *
28 * 2003/08/06 - Amir Noam <amir.noam at intel dot com>
29 * - Add support for setting bond's MAC address with special
30 * handling required for ALB/TLB.
31 *
32 * 2003/12/01 - Shmulik Hen <shmulik.hen at intel dot com>
33 * - Code cleanup and style changes
34 *
35 * 2003/12/30 - Amir Noam <amir.noam at intel dot com>
36 * - Fixed: Cannot remove and re-enslave the original active slave.
37 *
38 * 2004/01/14 - Shmulik Hen <shmulik.hen at intel dot com>
39 * - Add capability to tag self generated packets in ALB/TLB modes.
40 */
41
42//#define BONDING_DEBUG 1
43
44#include <linux/skbuff.h>
45#include <linux/netdevice.h>
46#include <linux/etherdevice.h>
47#include <linux/pkt_sched.h>
48#include <linux/spinlock.h>
49#include <linux/slab.h>
50#include <linux/timer.h>
51#include <linux/ip.h>
52#include <linux/ipv6.h>
53#include <linux/if_arp.h>
54#include <linux/if_ether.h>
55#include <linux/if_bonding.h>
56#include <linux/if_vlan.h>
57#include <linux/in.h>
58#include <net/ipx.h>
59#include <net/arp.h>
60#include <asm/byteorder.h>
61#include "bonding.h"
62#include "bond_alb.h"
63
64
65#define ALB_TIMER_TICKS_PER_SEC 10 /* should be a divisor of HZ */
66#define BOND_TLB_REBALANCE_INTERVAL 10 /* In seconds, periodic re-balancing.
67 * Used for division - never set
68 * to zero !!!
69 */
70#define BOND_ALB_LP_INTERVAL 1 /* In seconds, periodic send of
71 * learning packets to the switch
72 */
73
74#define BOND_TLB_REBALANCE_TICKS (BOND_TLB_REBALANCE_INTERVAL \
75 * ALB_TIMER_TICKS_PER_SEC)
76
77#define BOND_ALB_LP_TICKS (BOND_ALB_LP_INTERVAL \
78 * ALB_TIMER_TICKS_PER_SEC)
79
80#define TLB_HASH_TABLE_SIZE 256 /* The size of the clients hash table.
81 * Note that this value MUST NOT be smaller
82 * because the key hash table is BYTE wide !
83 */
84
85
86#define TLB_NULL_INDEX 0xffffffff
87#define MAX_LP_BURST 3
88
89/* rlb defs */
90#define RLB_HASH_TABLE_SIZE 256
91#define RLB_NULL_INDEX 0xffffffff
92#define RLB_UPDATE_DELAY 2*ALB_TIMER_TICKS_PER_SEC /* 2 seconds */
93#define RLB_ARP_BURST_SIZE 2
94#define RLB_UPDATE_RETRY 3 /* 3-ticks - must be smaller than the rlb
95 * rebalance interval (5 min).
96 */
97/* RLB_PROMISC_TIMEOUT = 10 sec equals the time that the current slave is
98 * promiscuous after failover
99 */
100#define RLB_PROMISC_TIMEOUT 10*ALB_TIMER_TICKS_PER_SEC
101
102static const u8 mac_bcast[ETH_ALEN] = {0xff,0xff,0xff,0xff,0xff,0xff};
103static const int alb_delta_in_ticks = HZ / ALB_TIMER_TICKS_PER_SEC;
104
105#pragma pack(1)
106struct learning_pkt {
107 u8 mac_dst[ETH_ALEN];
108 u8 mac_src[ETH_ALEN];
109 u16 type;
110 u8 padding[ETH_ZLEN - ETH_HLEN];
111};
112
113struct arp_pkt {
114 u16 hw_addr_space;
115 u16 prot_addr_space;
116 u8 hw_addr_len;
117 u8 prot_addr_len;
118 u16 op_code;
119 u8 mac_src[ETH_ALEN]; /* sender hardware address */
120 u32 ip_src; /* sender IP address */
121 u8 mac_dst[ETH_ALEN]; /* target hardware address */
122 u32 ip_dst; /* target IP address */
123};
124#pragma pack()
125
126/* Forward declaration */
127static void alb_send_learning_packets(struct slave *slave, u8 mac_addr[]);
128
129static inline u8 _simple_hash(u8 *hash_start, int hash_size)
130{
131 int i;
132 u8 hash = 0;
133
134 for (i = 0; i < hash_size; i++) {
135 hash ^= hash_start[i];
136 }
137
138 return hash;
139}
140
141/*********************** tlb specific functions ***************************/
142
143static inline void _lock_tx_hashtbl(struct bonding *bond)
144{
145 spin_lock(&(BOND_ALB_INFO(bond).tx_hashtbl_lock));
146}
147
148static inline void _unlock_tx_hashtbl(struct bonding *bond)
149{
150 spin_unlock(&(BOND_ALB_INFO(bond).tx_hashtbl_lock));
151}
152
153/* Caller must hold tx_hashtbl lock */
154static inline void tlb_init_table_entry(struct tlb_client_info *entry, int save_load)
155{
156 if (save_load) {
157 entry->load_history = 1 + entry->tx_bytes /
158 BOND_TLB_REBALANCE_INTERVAL;
159 entry->tx_bytes = 0;
160 }
161
162 entry->tx_slave = NULL;
163 entry->next = TLB_NULL_INDEX;
164 entry->prev = TLB_NULL_INDEX;
165}
166
167static inline void tlb_init_slave(struct slave *slave)
168{
169 SLAVE_TLB_INFO(slave).load = 0;
170 SLAVE_TLB_INFO(slave).head = TLB_NULL_INDEX;
171}
172
173/* Caller must hold bond lock for read */
174static void tlb_clear_slave(struct bonding *bond, struct slave *slave, int save_load)
175{
176 struct tlb_client_info *tx_hash_table;
177 u32 index;
178
179 _lock_tx_hashtbl(bond);
180
181 /* clear slave from tx_hashtbl */
182 tx_hash_table = BOND_ALB_INFO(bond).tx_hashtbl;
183
184 index = SLAVE_TLB_INFO(slave).head;
185 while (index != TLB_NULL_INDEX) {
186 u32 next_index = tx_hash_table[index].next;
187 tlb_init_table_entry(&tx_hash_table[index], save_load);
188 index = next_index;
189 }
190
191 _unlock_tx_hashtbl(bond);
192
193 tlb_init_slave(slave);
194}
195
196/* Must be called before starting the monitor timer */
197static int tlb_initialize(struct bonding *bond)
198{
199 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
200 int size = TLB_HASH_TABLE_SIZE * sizeof(struct tlb_client_info);
201 int i;
202
203 spin_lock_init(&(bond_info->tx_hashtbl_lock));
204
205 _lock_tx_hashtbl(bond);
206
207 bond_info->tx_hashtbl = kmalloc(size, GFP_KERNEL);
208 if (!bond_info->tx_hashtbl) {
209 printk(KERN_ERR DRV_NAME
210 ": Error: %s: Failed to allocate TLB hash table\n",
211 bond->dev->name);
212 _unlock_tx_hashtbl(bond);
213 return -1;
214 }
215
216 memset(bond_info->tx_hashtbl, 0, size);
217
218 for (i = 0; i < TLB_HASH_TABLE_SIZE; i++) {
219 tlb_init_table_entry(&bond_info->tx_hashtbl[i], 1);
220 }
221
222 _unlock_tx_hashtbl(bond);
223
224 return 0;
225}
226
227/* Must be called only after all slaves have been released */
228static void tlb_deinitialize(struct bonding *bond)
229{
230 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
231
232 _lock_tx_hashtbl(bond);
233
234 kfree(bond_info->tx_hashtbl);
235 bond_info->tx_hashtbl = NULL;
236
237 _unlock_tx_hashtbl(bond);
238}
239
240/* Caller must hold bond lock for read */
241static struct slave *tlb_get_least_loaded_slave(struct bonding *bond)
242{
243 struct slave *slave, *least_loaded;
244 s64 max_gap;
245 int i, found = 0;
246
247 /* Find the first enabled slave */
248 bond_for_each_slave(bond, slave, i) {
249 if (SLAVE_IS_OK(slave)) {
250 found = 1;
251 break;
252 }
253 }
254
255 if (!found) {
256 return NULL;
257 }
258
259 least_loaded = slave;
260 max_gap = (s64)(slave->speed << 20) - /* Convert to Megabit per sec */
261 (s64)(SLAVE_TLB_INFO(slave).load << 3); /* Bytes to bits */
262
263 /* Find the slave with the largest gap */
264 bond_for_each_slave_from(bond, slave, i, least_loaded) {
265 if (SLAVE_IS_OK(slave)) {
266 s64 gap = (s64)(slave->speed << 20) -
267 (s64)(SLAVE_TLB_INFO(slave).load << 3);
268 if (max_gap < gap) {
269 least_loaded = slave;
270 max_gap = gap;
271 }
272 }
273 }
274
275 return least_loaded;
276}
277
278/* Caller must hold bond lock for read */
279static struct slave *tlb_choose_channel(struct bonding *bond, u32 hash_index, u32 skb_len)
280{
281 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
282 struct tlb_client_info *hash_table;
283 struct slave *assigned_slave;
284
285 _lock_tx_hashtbl(bond);
286
287 hash_table = bond_info->tx_hashtbl;
288 assigned_slave = hash_table[hash_index].tx_slave;
289 if (!assigned_slave) {
290 assigned_slave = tlb_get_least_loaded_slave(bond);
291
292 if (assigned_slave) {
293 struct tlb_slave_info *slave_info =
294 &(SLAVE_TLB_INFO(assigned_slave));
295 u32 next_index = slave_info->head;
296
297 hash_table[hash_index].tx_slave = assigned_slave;
298 hash_table[hash_index].next = next_index;
299 hash_table[hash_index].prev = TLB_NULL_INDEX;
300
301 if (next_index != TLB_NULL_INDEX) {
302 hash_table[next_index].prev = hash_index;
303 }
304
305 slave_info->head = hash_index;
306 slave_info->load +=
307 hash_table[hash_index].load_history;
308 }
309 }
310
311 if (assigned_slave) {
312 hash_table[hash_index].tx_bytes += skb_len;
313 }
314
315 _unlock_tx_hashtbl(bond);
316
317 return assigned_slave;
318}
319
320/*********************** rlb specific functions ***************************/
321static inline void _lock_rx_hashtbl(struct bonding *bond)
322{
323 spin_lock(&(BOND_ALB_INFO(bond).rx_hashtbl_lock));
324}
325
326static inline void _unlock_rx_hashtbl(struct bonding *bond)
327{
328 spin_unlock(&(BOND_ALB_INFO(bond).rx_hashtbl_lock));
329}
330
331/* when an ARP REPLY is received from a client update its info
332 * in the rx_hashtbl
333 */
334static void rlb_update_entry_from_arp(struct bonding *bond, struct arp_pkt *arp)
335{
336 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
337 struct rlb_client_info *client_info;
338 u32 hash_index;
339
340 _lock_rx_hashtbl(bond);
341
342 hash_index = _simple_hash((u8*)&(arp->ip_src), sizeof(arp->ip_src));
343 client_info = &(bond_info->rx_hashtbl[hash_index]);
344
345 if ((client_info->assigned) &&
346 (client_info->ip_src == arp->ip_dst) &&
347 (client_info->ip_dst == arp->ip_src)) {
348 /* update the clients MAC address */
349 memcpy(client_info->mac_dst, arp->mac_src, ETH_ALEN);
350 client_info->ntt = 1;
351 bond_info->rx_ntt = 1;
352 }
353
354 _unlock_rx_hashtbl(bond);
355}
356
357static int rlb_arp_recv(struct sk_buff *skb, struct net_device *bond_dev, struct packet_type *ptype)
358{
359 struct bonding *bond = bond_dev->priv;
360 struct arp_pkt *arp = (struct arp_pkt *)skb->data;
361 int res = NET_RX_DROP;
362
363 if (!(bond_dev->flags & IFF_MASTER)) {
364 goto out;
365 }
366
367 if (!arp) {
368 dprintk("Packet has no ARP data\n");
369 goto out;
370 }
371
372 if (skb->len < sizeof(struct arp_pkt)) {
373 dprintk("Packet is too small to be an ARP\n");
374 goto out;
375 }
376
377 if (arp->op_code == htons(ARPOP_REPLY)) {
378 /* update rx hash table for this ARP */
379 rlb_update_entry_from_arp(bond, arp);
380 dprintk("Server received an ARP Reply from client\n");
381 }
382
383 res = NET_RX_SUCCESS;
384
385out:
386 dev_kfree_skb(skb);
387
388 return res;
389}
390
391/* Caller must hold bond lock for read */
392static struct slave *rlb_next_rx_slave(struct bonding *bond)
393{
394 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
395 struct slave *rx_slave, *slave, *start_at;
396 int i = 0;
397
398 if (bond_info->next_rx_slave) {
399 start_at = bond_info->next_rx_slave;
400 } else {
401 start_at = bond->first_slave;
402 }
403
404 rx_slave = NULL;
405
406 bond_for_each_slave_from(bond, slave, i, start_at) {
407 if (SLAVE_IS_OK(slave)) {
408 if (!rx_slave) {
409 rx_slave = slave;
410 } else if (slave->speed > rx_slave->speed) {
411 rx_slave = slave;
412 }
413 }
414 }
415
416 if (rx_slave) {
417 bond_info->next_rx_slave = rx_slave->next;
418 }
419
420 return rx_slave;
421}
422
423/* teach the switch the mac of a disabled slave
424 * on the primary for fault tolerance
425 *
426 * Caller must hold bond->curr_slave_lock for write or bond lock for write
427 */
428static void rlb_teach_disabled_mac_on_primary(struct bonding *bond, u8 addr[])
429{
430 if (!bond->curr_active_slave) {
431 return;
432 }
433
434 if (!bond->alb_info.primary_is_promisc) {
435 bond->alb_info.primary_is_promisc = 1;
436 dev_set_promiscuity(bond->curr_active_slave->dev, 1);
437 }
438
439 bond->alb_info.rlb_promisc_timeout_counter = 0;
440
441 alb_send_learning_packets(bond->curr_active_slave, addr);
442}
443
444/* slave being removed should not be active at this point
445 *
446 * Caller must hold bond lock for read
447 */
448static void rlb_clear_slave(struct bonding *bond, struct slave *slave)
449{
450 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
451 struct rlb_client_info *rx_hash_table;
452 u32 index, next_index;
453
454 /* clear slave from rx_hashtbl */
455 _lock_rx_hashtbl(bond);
456
457 rx_hash_table = bond_info->rx_hashtbl;
458 index = bond_info->rx_hashtbl_head;
459 for (; index != RLB_NULL_INDEX; index = next_index) {
460 next_index = rx_hash_table[index].next;
461 if (rx_hash_table[index].slave == slave) {
462 struct slave *assigned_slave = rlb_next_rx_slave(bond);
463
464 if (assigned_slave) {
465 rx_hash_table[index].slave = assigned_slave;
466 if (memcmp(rx_hash_table[index].mac_dst,
467 mac_bcast, ETH_ALEN)) {
468 bond_info->rx_hashtbl[index].ntt = 1;
469 bond_info->rx_ntt = 1;
470 /* A slave has been removed from the
471 * table because it is either disabled
472 * or being released. We must retry the
473 * update to avoid clients from not
474 * being updated & disconnecting when
475 * there is stress
476 */
477 bond_info->rlb_update_retry_counter =
478 RLB_UPDATE_RETRY;
479 }
480 } else { /* there is no active slave */
481 rx_hash_table[index].slave = NULL;
482 }
483 }
484 }
485
486 _unlock_rx_hashtbl(bond);
487
488 write_lock(&bond->curr_slave_lock);
489
490 if (slave != bond->curr_active_slave) {
491 rlb_teach_disabled_mac_on_primary(bond, slave->dev->dev_addr);
492 }
493
494 write_unlock(&bond->curr_slave_lock);
495}
496
497static void rlb_update_client(struct rlb_client_info *client_info)
498{
499 int i;
500
501 if (!client_info->slave) {
502 return;
503 }
504
505 for (i = 0; i < RLB_ARP_BURST_SIZE; i++) {
506 struct sk_buff *skb;
507
508 skb = arp_create(ARPOP_REPLY, ETH_P_ARP,
509 client_info->ip_dst,
510 client_info->slave->dev,
511 client_info->ip_src,
512 client_info->mac_dst,
513 client_info->slave->dev->dev_addr,
514 client_info->mac_dst);
515 if (!skb) {
516 printk(KERN_ERR DRV_NAME
517 ": Error: failed to create an ARP packet\n");
518 continue;
519 }
520
521 skb->dev = client_info->slave->dev;
522
523 if (client_info->tag) {
524 skb = vlan_put_tag(skb, client_info->vlan_id);
525 if (!skb) {
526 printk(KERN_ERR DRV_NAME
527 ": Error: failed to insert VLAN tag\n");
528 continue;
529 }
530 }
531
532 arp_xmit(skb);
533 }
534}
535
536/* sends ARP REPLIES that update the clients that need updating */
537static void rlb_update_rx_clients(struct bonding *bond)
538{
539 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
540 struct rlb_client_info *client_info;
541 u32 hash_index;
542
543 _lock_rx_hashtbl(bond);
544
545 hash_index = bond_info->rx_hashtbl_head;
546 for (; hash_index != RLB_NULL_INDEX; hash_index = client_info->next) {
547 client_info = &(bond_info->rx_hashtbl[hash_index]);
548 if (client_info->ntt) {
549 rlb_update_client(client_info);
550 if (bond_info->rlb_update_retry_counter == 0) {
551 client_info->ntt = 0;
552 }
553 }
554 }
555
556 /* do not update the entries again untill this counter is zero so that
557 * not to confuse the clients.
558 */
559 bond_info->rlb_update_delay_counter = RLB_UPDATE_DELAY;
560
561 _unlock_rx_hashtbl(bond);
562}
563
564/* The slave was assigned a new mac address - update the clients */
565static void rlb_req_update_slave_clients(struct bonding *bond, struct slave *slave)
566{
567 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
568 struct rlb_client_info *client_info;
569 int ntt = 0;
570 u32 hash_index;
571
572 _lock_rx_hashtbl(bond);
573
574 hash_index = bond_info->rx_hashtbl_head;
575 for (; hash_index != RLB_NULL_INDEX; hash_index = client_info->next) {
576 client_info = &(bond_info->rx_hashtbl[hash_index]);
577
578 if ((client_info->slave == slave) &&
579 memcmp(client_info->mac_dst, mac_bcast, ETH_ALEN)) {
580 client_info->ntt = 1;
581 ntt = 1;
582 }
583 }
584
585 // update the team's flag only after the whole iteration
586 if (ntt) {
587 bond_info->rx_ntt = 1;
588 //fasten the change
589 bond_info->rlb_update_retry_counter = RLB_UPDATE_RETRY;
590 }
591
592 _unlock_rx_hashtbl(bond);
593}
594
595/* mark all clients using src_ip to be updated */
596static void rlb_req_update_subnet_clients(struct bonding *bond, u32 src_ip)
597{
598 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
599 struct rlb_client_info *client_info;
600 u32 hash_index;
601
602 _lock_rx_hashtbl(bond);
603
604 hash_index = bond_info->rx_hashtbl_head;
605 for (; hash_index != RLB_NULL_INDEX; hash_index = client_info->next) {
606 client_info = &(bond_info->rx_hashtbl[hash_index]);
607
608 if (!client_info->slave) {
609 printk(KERN_ERR DRV_NAME
610 ": Error: found a client with no channel in "
611 "the client's hash table\n");
612 continue;
613 }
614 /*update all clients using this src_ip, that are not assigned
615 * to the team's address (curr_active_slave) and have a known
616 * unicast mac address.
617 */
618 if ((client_info->ip_src == src_ip) &&
619 memcmp(client_info->slave->dev->dev_addr,
620 bond->dev->dev_addr, ETH_ALEN) &&
621 memcmp(client_info->mac_dst, mac_bcast, ETH_ALEN)) {
622 client_info->ntt = 1;
623 bond_info->rx_ntt = 1;
624 }
625 }
626
627 _unlock_rx_hashtbl(bond);
628}
629
630/* Caller must hold both bond and ptr locks for read */
631static struct slave *rlb_choose_channel(struct sk_buff *skb, struct bonding *bond)
632{
633 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
634 struct arp_pkt *arp = (struct arp_pkt *)skb->nh.raw;
635 struct slave *assigned_slave;
636 struct rlb_client_info *client_info;
637 u32 hash_index = 0;
638
639 _lock_rx_hashtbl(bond);
640
641 hash_index = _simple_hash((u8 *)&arp->ip_dst, sizeof(arp->ip_src));
642 client_info = &(bond_info->rx_hashtbl[hash_index]);
643
644 if (client_info->assigned) {
645 if ((client_info->ip_src == arp->ip_src) &&
646 (client_info->ip_dst == arp->ip_dst)) {
647 /* the entry is already assigned to this client */
648 if (memcmp(arp->mac_dst, mac_bcast, ETH_ALEN)) {
649 /* update mac address from arp */
650 memcpy(client_info->mac_dst, arp->mac_dst, ETH_ALEN);
651 }
652
653 assigned_slave = client_info->slave;
654 if (assigned_slave) {
655 _unlock_rx_hashtbl(bond);
656 return assigned_slave;
657 }
658 } else {
659 /* the entry is already assigned to some other client,
660 * move the old client to primary (curr_active_slave) so
661 * that the new client can be assigned to this entry.
662 */
663 if (bond->curr_active_slave &&
664 client_info->slave != bond->curr_active_slave) {
665 client_info->slave = bond->curr_active_slave;
666 rlb_update_client(client_info);
667 }
668 }
669 }
670 /* assign a new slave */
671 assigned_slave = rlb_next_rx_slave(bond);
672
673 if (assigned_slave) {
674 client_info->ip_src = arp->ip_src;
675 client_info->ip_dst = arp->ip_dst;
676 /* arp->mac_dst is broadcast for arp reqeusts.
677 * will be updated with clients actual unicast mac address
678 * upon receiving an arp reply.
679 */
680 memcpy(client_info->mac_dst, arp->mac_dst, ETH_ALEN);
681 client_info->slave = assigned_slave;
682
683 if (memcmp(client_info->mac_dst, mac_bcast, ETH_ALEN)) {
684 client_info->ntt = 1;
685 bond->alb_info.rx_ntt = 1;
686 } else {
687 client_info->ntt = 0;
688 }
689
690 if (!list_empty(&bond->vlan_list)) {
691 unsigned short vlan_id;
692 int res = vlan_get_tag(skb, &vlan_id);
693 if (!res) {
694 client_info->tag = 1;
695 client_info->vlan_id = vlan_id;
696 }
697 }
698
699 if (!client_info->assigned) {
700 u32 prev_tbl_head = bond_info->rx_hashtbl_head;
701 bond_info->rx_hashtbl_head = hash_index;
702 client_info->next = prev_tbl_head;
703 if (prev_tbl_head != RLB_NULL_INDEX) {
704 bond_info->rx_hashtbl[prev_tbl_head].prev =
705 hash_index;
706 }
707 client_info->assigned = 1;
708 }
709 }
710
711 _unlock_rx_hashtbl(bond);
712
713 return assigned_slave;
714}
715
716/* chooses (and returns) transmit channel for arp reply
717 * does not choose channel for other arp types since they are
718 * sent on the curr_active_slave
719 */
720static struct slave *rlb_arp_xmit(struct sk_buff *skb, struct bonding *bond)
721{
722 struct arp_pkt *arp = (struct arp_pkt *)skb->nh.raw;
723 struct slave *tx_slave = NULL;
724
725 if (arp->op_code == __constant_htons(ARPOP_REPLY)) {
726 /* the arp must be sent on the selected
727 * rx channel
728 */
729 tx_slave = rlb_choose_channel(skb, bond);
730 if (tx_slave) {
731 memcpy(arp->mac_src,tx_slave->dev->dev_addr, ETH_ALEN);
732 }
733 dprintk("Server sent ARP Reply packet\n");
734 } else if (arp->op_code == __constant_htons(ARPOP_REQUEST)) {
735 /* Create an entry in the rx_hashtbl for this client as a
736 * place holder.
737 * When the arp reply is received the entry will be updated
738 * with the correct unicast address of the client.
739 */
740 rlb_choose_channel(skb, bond);
741
742 /* The ARP relpy packets must be delayed so that
743 * they can cancel out the influence of the ARP request.
744 */
745 bond->alb_info.rlb_update_delay_counter = RLB_UPDATE_DELAY;
746
747 /* arp requests are broadcast and are sent on the primary
748 * the arp request will collapse all clients on the subnet to
749 * the primary slave. We must register these clients to be
750 * updated with their assigned mac.
751 */
752 rlb_req_update_subnet_clients(bond, arp->ip_src);
753 dprintk("Server sent ARP Request packet\n");
754 }
755
756 return tx_slave;
757}
758
759/* Caller must hold bond lock for read */
760static void rlb_rebalance(struct bonding *bond)
761{
762 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
763 struct slave *assigned_slave;
764 struct rlb_client_info *client_info;
765 int ntt;
766 u32 hash_index;
767
768 _lock_rx_hashtbl(bond);
769
770 ntt = 0;
771 hash_index = bond_info->rx_hashtbl_head;
772 for (; hash_index != RLB_NULL_INDEX; hash_index = client_info->next) {
773 client_info = &(bond_info->rx_hashtbl[hash_index]);
774 assigned_slave = rlb_next_rx_slave(bond);
775 if (assigned_slave && (client_info->slave != assigned_slave)) {
776 client_info->slave = assigned_slave;
777 client_info->ntt = 1;
778 ntt = 1;
779 }
780 }
781
782 /* update the team's flag only after the whole iteration */
783 if (ntt) {
784 bond_info->rx_ntt = 1;
785 }
786 _unlock_rx_hashtbl(bond);
787}
788
789/* Caller must hold rx_hashtbl lock */
790static void rlb_init_table_entry(struct rlb_client_info *entry)
791{
792 memset(entry, 0, sizeof(struct rlb_client_info));
793 entry->next = RLB_NULL_INDEX;
794 entry->prev = RLB_NULL_INDEX;
795}
796
797static int rlb_initialize(struct bonding *bond)
798{
799 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
800 struct packet_type *pk_type = &(BOND_ALB_INFO(bond).rlb_pkt_type);
801 int size = RLB_HASH_TABLE_SIZE * sizeof(struct rlb_client_info);
802 int i;
803
804 spin_lock_init(&(bond_info->rx_hashtbl_lock));
805
806 _lock_rx_hashtbl(bond);
807
808 bond_info->rx_hashtbl = kmalloc(size, GFP_KERNEL);
809 if (!bond_info->rx_hashtbl) {
810 printk(KERN_ERR DRV_NAME
811 ": Error: %s: Failed to allocate RLB hash table\n",
812 bond->dev->name);
813 _unlock_rx_hashtbl(bond);
814 return -1;
815 }
816
817 bond_info->rx_hashtbl_head = RLB_NULL_INDEX;
818
819 for (i = 0; i < RLB_HASH_TABLE_SIZE; i++) {
820 rlb_init_table_entry(bond_info->rx_hashtbl + i);
821 }
822
823 _unlock_rx_hashtbl(bond);
824
825 /*initialize packet type*/
826 pk_type->type = __constant_htons(ETH_P_ARP);
827 pk_type->dev = bond->dev;
828 pk_type->func = rlb_arp_recv;
829
830 /* register to receive ARPs */
831 dev_add_pack(pk_type);
832
833 return 0;
834}
835
836static void rlb_deinitialize(struct bonding *bond)
837{
838 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
839
840 dev_remove_pack(&(bond_info->rlb_pkt_type));
841
842 _lock_rx_hashtbl(bond);
843
844 kfree(bond_info->rx_hashtbl);
845 bond_info->rx_hashtbl = NULL;
846 bond_info->rx_hashtbl_head = RLB_NULL_INDEX;
847
848 _unlock_rx_hashtbl(bond);
849}
850
851static void rlb_clear_vlan(struct bonding *bond, unsigned short vlan_id)
852{
853 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
854 u32 curr_index;
855
856 _lock_rx_hashtbl(bond);
857
858 curr_index = bond_info->rx_hashtbl_head;
859 while (curr_index != RLB_NULL_INDEX) {
860 struct rlb_client_info *curr = &(bond_info->rx_hashtbl[curr_index]);
861 u32 next_index = bond_info->rx_hashtbl[curr_index].next;
862 u32 prev_index = bond_info->rx_hashtbl[curr_index].prev;
863
864 if (curr->tag && (curr->vlan_id == vlan_id)) {
865 if (curr_index == bond_info->rx_hashtbl_head) {
866 bond_info->rx_hashtbl_head = next_index;
867 }
868 if (prev_index != RLB_NULL_INDEX) {
869 bond_info->rx_hashtbl[prev_index].next = next_index;
870 }
871 if (next_index != RLB_NULL_INDEX) {
872 bond_info->rx_hashtbl[next_index].prev = prev_index;
873 }
874
875 rlb_init_table_entry(curr);
876 }
877
878 curr_index = next_index;
879 }
880
881 _unlock_rx_hashtbl(bond);
882}
883
884/*********************** tlb/rlb shared functions *********************/
885
886static void alb_send_learning_packets(struct slave *slave, u8 mac_addr[])
887{
888 struct bonding *bond = bond_get_bond_by_slave(slave);
889 struct learning_pkt pkt;
890 int size = sizeof(struct learning_pkt);
891 int i;
892
893 memset(&pkt, 0, size);
894 memcpy(pkt.mac_dst, mac_addr, ETH_ALEN);
895 memcpy(pkt.mac_src, mac_addr, ETH_ALEN);
896 pkt.type = __constant_htons(ETH_P_LOOP);
897
898 for (i = 0; i < MAX_LP_BURST; i++) {
899 struct sk_buff *skb;
900 char *data;
901
902 skb = dev_alloc_skb(size);
903 if (!skb) {
904 return;
905 }
906
907 data = skb_put(skb, size);
908 memcpy(data, &pkt, size);
909
910 skb->mac.raw = data;
911 skb->nh.raw = data + ETH_HLEN;
912 skb->protocol = pkt.type;
913 skb->priority = TC_PRIO_CONTROL;
914 skb->dev = slave->dev;
915
916 if (!list_empty(&bond->vlan_list)) {
917 struct vlan_entry *vlan;
918
919 vlan = bond_next_vlan(bond,
920 bond->alb_info.current_alb_vlan);
921
922 bond->alb_info.current_alb_vlan = vlan;
923 if (!vlan) {
924 kfree_skb(skb);
925 continue;
926 }
927
928 skb = vlan_put_tag(skb, vlan->vlan_id);
929 if (!skb) {
930 printk(KERN_ERR DRV_NAME
931 ": Error: failed to insert VLAN tag\n");
932 continue;
933 }
934 }
935
936 dev_queue_xmit(skb);
937 }
938}
939
940/* hw is a boolean parameter that determines whether we should try and
941 * set the hw address of the device as well as the hw address of the
942 * net_device
943 */
944static int alb_set_slave_mac_addr(struct slave *slave, u8 addr[], int hw)
945{
946 struct net_device *dev = slave->dev;
947 struct sockaddr s_addr;
948
949 if (!hw) {
950 memcpy(dev->dev_addr, addr, dev->addr_len);
951 return 0;
952 }
953
954 /* for rlb each slave must have a unique hw mac addresses so that */
955 /* each slave will receive packets destined to a different mac */
956 memcpy(s_addr.sa_data, addr, dev->addr_len);
957 s_addr.sa_family = dev->type;
958 if (dev_set_mac_address(dev, &s_addr)) {
959 printk(KERN_ERR DRV_NAME
960 ": Error: dev_set_mac_address of dev %s failed! ALB "
961 "mode requires that the base driver support setting "
962 "the hw address also when the network device's "
963 "interface is open\n",
964 dev->name);
965 return -EOPNOTSUPP;
966 }
967 return 0;
968}
969
970/* Caller must hold bond lock for write or curr_slave_lock for write*/
971static void alb_swap_mac_addr(struct bonding *bond, struct slave *slave1, struct slave *slave2)
972{
973 struct slave *disabled_slave = NULL;
974 u8 tmp_mac_addr[ETH_ALEN];
975 int slaves_state_differ;
976
977 slaves_state_differ = (SLAVE_IS_OK(slave1) != SLAVE_IS_OK(slave2));
978
979 memcpy(tmp_mac_addr, slave1->dev->dev_addr, ETH_ALEN);
980 alb_set_slave_mac_addr(slave1, slave2->dev->dev_addr, bond->alb_info.rlb_enabled);
981 alb_set_slave_mac_addr(slave2, tmp_mac_addr, bond->alb_info.rlb_enabled);
982
983 /* fasten the change in the switch */
984 if (SLAVE_IS_OK(slave1)) {
985 alb_send_learning_packets(slave1, slave1->dev->dev_addr);
986 if (bond->alb_info.rlb_enabled) {
987 /* inform the clients that the mac address
988 * has changed
989 */
990 rlb_req_update_slave_clients(bond, slave1);
991 }
992 } else {
993 disabled_slave = slave1;
994 }
995
996 if (SLAVE_IS_OK(slave2)) {
997 alb_send_learning_packets(slave2, slave2->dev->dev_addr);
998 if (bond->alb_info.rlb_enabled) {
999 /* inform the clients that the mac address
1000 * has changed
1001 */
1002 rlb_req_update_slave_clients(bond, slave2);
1003 }
1004 } else {
1005 disabled_slave = slave2;
1006 }
1007
1008 if (bond->alb_info.rlb_enabled && slaves_state_differ) {
1009 /* A disabled slave was assigned an active mac addr */
1010 rlb_teach_disabled_mac_on_primary(bond,
1011 disabled_slave->dev->dev_addr);
1012 }
1013}
1014
1015/**
1016 * alb_change_hw_addr_on_detach
1017 * @bond: bonding we're working on
1018 * @slave: the slave that was just detached
1019 *
1020 * We assume that @slave was already detached from the slave list.
1021 *
1022 * If @slave's permanent hw address is different both from its current
1023 * address and from @bond's address, then somewhere in the bond there's
1024 * a slave that has @slave's permanet address as its current address.
1025 * We'll make sure that that slave no longer uses @slave's permanent address.
1026 *
1027 * Caller must hold bond lock
1028 */
1029static void alb_change_hw_addr_on_detach(struct bonding *bond, struct slave *slave)
1030{
1031 int perm_curr_diff;
1032 int perm_bond_diff;
1033
1034 perm_curr_diff = memcmp(slave->perm_hwaddr,
1035 slave->dev->dev_addr,
1036 ETH_ALEN);
1037 perm_bond_diff = memcmp(slave->perm_hwaddr,
1038 bond->dev->dev_addr,
1039 ETH_ALEN);
1040
1041 if (perm_curr_diff && perm_bond_diff) {
1042 struct slave *tmp_slave;
1043 int i, found = 0;
1044
1045 bond_for_each_slave(bond, tmp_slave, i) {
1046 if (!memcmp(slave->perm_hwaddr,
1047 tmp_slave->dev->dev_addr,
1048 ETH_ALEN)) {
1049 found = 1;
1050 break;
1051 }
1052 }
1053
1054 if (found) {
1055 alb_swap_mac_addr(bond, slave, tmp_slave);
1056 }
1057 }
1058}
1059
1060/**
1061 * alb_handle_addr_collision_on_attach
1062 * @bond: bonding we're working on
1063 * @slave: the slave that was just attached
1064 *
1065 * checks uniqueness of slave's mac address and handles the case the
1066 * new slave uses the bonds mac address.
1067 *
1068 * If the permanent hw address of @slave is @bond's hw address, we need to
1069 * find a different hw address to give @slave, that isn't in use by any other
1070 * slave in the bond. This address must be, of course, one of the premanent
1071 * addresses of the other slaves.
1072 *
1073 * We go over the slave list, and for each slave there we compare its
1074 * permanent hw address with the current address of all the other slaves.
1075 * If no match was found, then we've found a slave with a permanent address
1076 * that isn't used by any other slave in the bond, so we can assign it to
1077 * @slave.
1078 *
1079 * assumption: this function is called before @slave is attached to the
1080 * bond slave list.
1081 *
1082 * caller must hold the bond lock for write since the mac addresses are compared
1083 * and may be swapped.
1084 */
1085static int alb_handle_addr_collision_on_attach(struct bonding *bond, struct slave *slave)
1086{
1087 struct slave *tmp_slave1, *tmp_slave2, *free_mac_slave;
1088 struct slave *has_bond_addr = bond->curr_active_slave;
1089 int i, j, found = 0;
1090
1091 if (bond->slave_cnt == 0) {
1092 /* this is the first slave */
1093 return 0;
1094 }
1095
1096 /* if slave's mac address differs from bond's mac address
1097 * check uniqueness of slave's mac address against the other
1098 * slaves in the bond.
1099 */
1100 if (memcmp(slave->perm_hwaddr, bond->dev->dev_addr, ETH_ALEN)) {
1101 bond_for_each_slave(bond, tmp_slave1, i) {
1102 if (!memcmp(tmp_slave1->dev->dev_addr, slave->dev->dev_addr,
1103 ETH_ALEN)) {
1104 found = 1;
1105 break;
1106 }
1107 }
1108
1109 if (found) {
1110 /* a slave was found that is using the mac address
1111 * of the new slave
1112 */
1113 printk(KERN_ERR DRV_NAME
1114 ": Error: the hw address of slave %s is not "
1115 "unique - cannot enslave it!",
1116 slave->dev->name);
1117 return -EINVAL;
1118 }
1119
1120 return 0;
1121 }
1122
1123 /* The slave's address is equal to the address of the bond.
1124 * Search for a spare address in the bond for this slave.
1125 */
1126 free_mac_slave = NULL;
1127
1128 bond_for_each_slave(bond, tmp_slave1, i) {
1129 found = 0;
1130 bond_for_each_slave(bond, tmp_slave2, j) {
1131 if (!memcmp(tmp_slave1->perm_hwaddr,
1132 tmp_slave2->dev->dev_addr,
1133 ETH_ALEN)) {
1134 found = 1;
1135 break;
1136 }
1137 }
1138
1139 if (!found) {
1140 /* no slave has tmp_slave1's perm addr
1141 * as its curr addr
1142 */
1143 free_mac_slave = tmp_slave1;
1144 break;
1145 }
1146
1147 if (!has_bond_addr) {
1148 if (!memcmp(tmp_slave1->dev->dev_addr,
1149 bond->dev->dev_addr,
1150 ETH_ALEN)) {
1151
1152 has_bond_addr = tmp_slave1;
1153 }
1154 }
1155 }
1156
1157 if (free_mac_slave) {
1158 alb_set_slave_mac_addr(slave, free_mac_slave->perm_hwaddr,
1159 bond->alb_info.rlb_enabled);
1160
1161 printk(KERN_WARNING DRV_NAME
1162 ": Warning: the hw address of slave %s is in use by "
1163 "the bond; giving it the hw address of %s\n",
1164 slave->dev->name, free_mac_slave->dev->name);
1165
1166 } else if (has_bond_addr) {
1167 printk(KERN_ERR DRV_NAME
1168 ": Error: the hw address of slave %s is in use by the "
1169 "bond; couldn't find a slave with a free hw address to "
1170 "give it (this should not have happened)\n",
1171 slave->dev->name);
1172 return -EFAULT;
1173 }
1174
1175 return 0;
1176}
1177
1178/**
1179 * alb_set_mac_address
1180 * @bond:
1181 * @addr:
1182 *
1183 * In TLB mode all slaves are configured to the bond's hw address, but set
1184 * their dev_addr field to different addresses (based on their permanent hw
1185 * addresses).
1186 *
1187 * For each slave, this function sets the interface to the new address and then
1188 * changes its dev_addr field to its previous value.
1189 *
1190 * Unwinding assumes bond's mac address has not yet changed.
1191 */
1192static int alb_set_mac_address(struct bonding *bond, void *addr)
1193{
1194 struct sockaddr sa;
1195 struct slave *slave, *stop_at;
1196 char tmp_addr[ETH_ALEN];
1197 int res;
1198 int i;
1199
1200 if (bond->alb_info.rlb_enabled) {
1201 return 0;
1202 }
1203
1204 bond_for_each_slave(bond, slave, i) {
1205 if (slave->dev->set_mac_address == NULL) {
1206 res = -EOPNOTSUPP;
1207 goto unwind;
1208 }
1209
1210 /* save net_device's current hw address */
1211 memcpy(tmp_addr, slave->dev->dev_addr, ETH_ALEN);
1212
1213 res = dev_set_mac_address(slave->dev, addr);
1214
1215 /* restore net_device's hw address */
1216 memcpy(slave->dev->dev_addr, tmp_addr, ETH_ALEN);
1217
1218 if (res) {
1219 goto unwind;
1220 }
1221 }
1222
1223 return 0;
1224
1225unwind:
1226 memcpy(sa.sa_data, bond->dev->dev_addr, bond->dev->addr_len);
1227 sa.sa_family = bond->dev->type;
1228
1229 /* unwind from head to the slave that failed */
1230 stop_at = slave;
1231 bond_for_each_slave_from_to(bond, slave, i, bond->first_slave, stop_at) {
1232 memcpy(tmp_addr, slave->dev->dev_addr, ETH_ALEN);
1233 dev_set_mac_address(slave->dev, &sa);
1234 memcpy(slave->dev->dev_addr, tmp_addr, ETH_ALEN);
1235 }
1236
1237 return res;
1238}
1239
1240/************************ exported alb funcions ************************/
1241
1242int bond_alb_initialize(struct bonding *bond, int rlb_enabled)
1243{
1244 int res;
1245
1246 res = tlb_initialize(bond);
1247 if (res) {
1248 return res;
1249 }
1250
1251 if (rlb_enabled) {
1252 bond->alb_info.rlb_enabled = 1;
1253 /* initialize rlb */
1254 res = rlb_initialize(bond);
1255 if (res) {
1256 tlb_deinitialize(bond);
1257 return res;
1258 }
1259 }
1260
1261 return 0;
1262}
1263
1264void bond_alb_deinitialize(struct bonding *bond)
1265{
1266 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
1267
1268 tlb_deinitialize(bond);
1269
1270 if (bond_info->rlb_enabled) {
1271 rlb_deinitialize(bond);
1272 }
1273}
1274
1275int bond_alb_xmit(struct sk_buff *skb, struct net_device *bond_dev)
1276{
1277 struct bonding *bond = bond_dev->priv;
1278 struct ethhdr *eth_data;
1279 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
1280 struct slave *tx_slave = NULL;
1281 static u32 ip_bcast = 0xffffffff;
1282 int hash_size = 0;
1283 int do_tx_balance = 1;
1284 u32 hash_index = 0;
1285 u8 *hash_start = NULL;
1286 int res = 1;
1287
1288 skb->mac.raw = (unsigned char *)skb->data;
1289 eth_data = eth_hdr(skb);
1290
1291 /* make sure that the curr_active_slave and the slaves list do
1292 * not change during tx
1293 */
1294 read_lock(&bond->lock);
1295 read_lock(&bond->curr_slave_lock);
1296
1297 if (!BOND_IS_OK(bond)) {
1298 goto out;
1299 }
1300
1301 switch (ntohs(skb->protocol)) {
1302 case ETH_P_IP:
1303 if ((memcmp(eth_data->h_dest, mac_bcast, ETH_ALEN) == 0) ||
1304 (skb->nh.iph->daddr == ip_bcast) ||
1305 (skb->nh.iph->protocol == IPPROTO_IGMP)) {
1306 do_tx_balance = 0;
1307 break;
1308 }
1309 hash_start = (char*)&(skb->nh.iph->daddr);
1310 hash_size = sizeof(skb->nh.iph->daddr);
1311 break;
1312 case ETH_P_IPV6:
1313 if (memcmp(eth_data->h_dest, mac_bcast, ETH_ALEN) == 0) {
1314 do_tx_balance = 0;
1315 break;
1316 }
1317
1318 hash_start = (char*)&(skb->nh.ipv6h->daddr);
1319 hash_size = sizeof(skb->nh.ipv6h->daddr);
1320 break;
1321 case ETH_P_IPX:
1322 if (ipx_hdr(skb)->ipx_checksum !=
1323 __constant_htons(IPX_NO_CHECKSUM)) {
1324 /* something is wrong with this packet */
1325 do_tx_balance = 0;
1326 break;
1327 }
1328
1329 if (ipx_hdr(skb)->ipx_type != IPX_TYPE_NCP) {
1330 /* The only protocol worth balancing in
1331 * this family since it has an "ARP" like
1332 * mechanism
1333 */
1334 do_tx_balance = 0;
1335 break;
1336 }
1337
1338 hash_start = (char*)eth_data->h_dest;
1339 hash_size = ETH_ALEN;
1340 break;
1341 case ETH_P_ARP:
1342 do_tx_balance = 0;
1343 if (bond_info->rlb_enabled) {
1344 tx_slave = rlb_arp_xmit(skb, bond);
1345 }
1346 break;
1347 default:
1348 do_tx_balance = 0;
1349 break;
1350 }
1351
1352 if (do_tx_balance) {
1353 hash_index = _simple_hash(hash_start, hash_size);
1354 tx_slave = tlb_choose_channel(bond, hash_index, skb->len);
1355 }
1356
1357 if (!tx_slave) {
1358 /* unbalanced or unassigned, send through primary */
1359 tx_slave = bond->curr_active_slave;
1360 bond_info->unbalanced_load += skb->len;
1361 }
1362
1363 if (tx_slave && SLAVE_IS_OK(tx_slave)) {
1364 if (tx_slave != bond->curr_active_slave) {
1365 memcpy(eth_data->h_source,
1366 tx_slave->dev->dev_addr,
1367 ETH_ALEN);
1368 }
1369
1370 res = bond_dev_queue_xmit(bond, skb, tx_slave->dev);
1371 } else {
1372 if (tx_slave) {
1373 tlb_clear_slave(bond, tx_slave, 0);
1374 }
1375 }
1376
1377out:
1378 if (res) {
1379 /* no suitable interface, frame not sent */
1380 dev_kfree_skb(skb);
1381 }
1382 read_unlock(&bond->curr_slave_lock);
1383 read_unlock(&bond->lock);
1384 return 0;
1385}
1386
1387void bond_alb_monitor(struct bonding *bond)
1388{
1389 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
1390 struct slave *slave;
1391 int i;
1392
1393 read_lock(&bond->lock);
1394
1395 if (bond->kill_timers) {
1396 goto out;
1397 }
1398
1399 if (bond->slave_cnt == 0) {
1400 bond_info->tx_rebalance_counter = 0;
1401 bond_info->lp_counter = 0;
1402 goto re_arm;
1403 }
1404
1405 bond_info->tx_rebalance_counter++;
1406 bond_info->lp_counter++;
1407
1408 /* send learning packets */
1409 if (bond_info->lp_counter >= BOND_ALB_LP_TICKS) {
1410 /* change of curr_active_slave involves swapping of mac addresses.
1411 * in order to avoid this swapping from happening while
1412 * sending the learning packets, the curr_slave_lock must be held for
1413 * read.
1414 */
1415 read_lock(&bond->curr_slave_lock);
1416
1417 bond_for_each_slave(bond, slave, i) {
1418 alb_send_learning_packets(slave,slave->dev->dev_addr);
1419 }
1420
1421 read_unlock(&bond->curr_slave_lock);
1422
1423 bond_info->lp_counter = 0;
1424 }
1425
1426 /* rebalance tx traffic */
1427 if (bond_info->tx_rebalance_counter >= BOND_TLB_REBALANCE_TICKS) {
1428
1429 read_lock(&bond->curr_slave_lock);
1430
1431 bond_for_each_slave(bond, slave, i) {
1432 tlb_clear_slave(bond, slave, 1);
1433 if (slave == bond->curr_active_slave) {
1434 SLAVE_TLB_INFO(slave).load =
1435 bond_info->unbalanced_load /
1436 BOND_TLB_REBALANCE_INTERVAL;
1437 bond_info->unbalanced_load = 0;
1438 }
1439 }
1440
1441 read_unlock(&bond->curr_slave_lock);
1442
1443 bond_info->tx_rebalance_counter = 0;
1444 }
1445
1446 /* handle rlb stuff */
1447 if (bond_info->rlb_enabled) {
1448 /* the following code changes the promiscuity of the
1449 * the curr_active_slave. It needs to be locked with a
1450 * write lock to protect from other code that also
1451 * sets the promiscuity.
1452 */
1453 write_lock(&bond->curr_slave_lock);
1454
1455 if (bond_info->primary_is_promisc &&
1456 (++bond_info->rlb_promisc_timeout_counter >= RLB_PROMISC_TIMEOUT)) {
1457
1458 bond_info->rlb_promisc_timeout_counter = 0;
1459
1460 /* If the primary was set to promiscuous mode
1461 * because a slave was disabled then
1462 * it can now leave promiscuous mode.
1463 */
1464 dev_set_promiscuity(bond->curr_active_slave->dev, -1);
1465 bond_info->primary_is_promisc = 0;
1466 }
1467
1468 write_unlock(&bond->curr_slave_lock);
1469
1470 if (bond_info->rlb_rebalance) {
1471 bond_info->rlb_rebalance = 0;
1472 rlb_rebalance(bond);
1473 }
1474
1475 /* check if clients need updating */
1476 if (bond_info->rx_ntt) {
1477 if (bond_info->rlb_update_delay_counter) {
1478 --bond_info->rlb_update_delay_counter;
1479 } else {
1480 rlb_update_rx_clients(bond);
1481 if (bond_info->rlb_update_retry_counter) {
1482 --bond_info->rlb_update_retry_counter;
1483 } else {
1484 bond_info->rx_ntt = 0;
1485 }
1486 }
1487 }
1488 }
1489
1490re_arm:
1491 mod_timer(&(bond_info->alb_timer), jiffies + alb_delta_in_ticks);
1492out:
1493 read_unlock(&bond->lock);
1494}
1495
1496/* assumption: called before the slave is attached to the bond
1497 * and not locked by the bond lock
1498 */
1499int bond_alb_init_slave(struct bonding *bond, struct slave *slave)
1500{
1501 int res;
1502
1503 res = alb_set_slave_mac_addr(slave, slave->perm_hwaddr,
1504 bond->alb_info.rlb_enabled);
1505 if (res) {
1506 return res;
1507 }
1508
1509 /* caller must hold the bond lock for write since the mac addresses
1510 * are compared and may be swapped.
1511 */
1512 write_lock_bh(&bond->lock);
1513
1514 res = alb_handle_addr_collision_on_attach(bond, slave);
1515
1516 write_unlock_bh(&bond->lock);
1517
1518 if (res) {
1519 return res;
1520 }
1521
1522 tlb_init_slave(slave);
1523
1524 /* order a rebalance ASAP */
1525 bond->alb_info.tx_rebalance_counter = BOND_TLB_REBALANCE_TICKS;
1526
1527 if (bond->alb_info.rlb_enabled) {
1528 bond->alb_info.rlb_rebalance = 1;
1529 }
1530
1531 return 0;
1532}
1533
1534/* Caller must hold bond lock for write */
1535void bond_alb_deinit_slave(struct bonding *bond, struct slave *slave)
1536{
1537 if (bond->slave_cnt > 1) {
1538 alb_change_hw_addr_on_detach(bond, slave);
1539 }
1540
1541 tlb_clear_slave(bond, slave, 0);
1542
1543 if (bond->alb_info.rlb_enabled) {
1544 bond->alb_info.next_rx_slave = NULL;
1545 rlb_clear_slave(bond, slave);
1546 }
1547}
1548
1549/* Caller must hold bond lock for read */
1550void bond_alb_handle_link_change(struct bonding *bond, struct slave *slave, char link)
1551{
1552 struct alb_bond_info *bond_info = &(BOND_ALB_INFO(bond));
1553
1554 if (link == BOND_LINK_DOWN) {
1555 tlb_clear_slave(bond, slave, 0);
1556 if (bond->alb_info.rlb_enabled) {
1557 rlb_clear_slave(bond, slave);
1558 }
1559 } else if (link == BOND_LINK_UP) {
1560 /* order a rebalance ASAP */
1561 bond_info->tx_rebalance_counter = BOND_TLB_REBALANCE_TICKS;
1562 if (bond->alb_info.rlb_enabled) {
1563 bond->alb_info.rlb_rebalance = 1;
1564 /* If the updelay module parameter is smaller than the
1565 * forwarding delay of the switch the rebalance will
1566 * not work because the rebalance arp replies will
1567 * not be forwarded to the clients..
1568 */
1569 }
1570 }
1571}
1572
1573/**
1574 * bond_alb_handle_active_change - assign new curr_active_slave
1575 * @bond: our bonding struct
1576 * @new_slave: new slave to assign
1577 *
1578 * Set the bond->curr_active_slave to @new_slave and handle
1579 * mac address swapping and promiscuity changes as needed.
1580 *
1581 * Caller must hold bond curr_slave_lock for write (or bond lock for write)
1582 */
1583void bond_alb_handle_active_change(struct bonding *bond, struct slave *new_slave)
1584{
1585 struct slave *swap_slave;
1586 int i;
1587
1588 if (bond->curr_active_slave == new_slave) {
1589 return;
1590 }
1591
1592 if (bond->curr_active_slave && bond->alb_info.primary_is_promisc) {
1593 dev_set_promiscuity(bond->curr_active_slave->dev, -1);
1594 bond->alb_info.primary_is_promisc = 0;
1595 bond->alb_info.rlb_promisc_timeout_counter = 0;
1596 }
1597
1598 swap_slave = bond->curr_active_slave;
1599 bond->curr_active_slave = new_slave;
1600
1601 if (!new_slave || (bond->slave_cnt == 0)) {
1602 return;
1603 }
1604
1605 /* set the new curr_active_slave to the bonds mac address
1606 * i.e. swap mac addresses of old curr_active_slave and new curr_active_slave
1607 */
1608 if (!swap_slave) {
1609 struct slave *tmp_slave;
1610 /* find slave that is holding the bond's mac address */
1611 bond_for_each_slave(bond, tmp_slave, i) {
1612 if (!memcmp(tmp_slave->dev->dev_addr,
1613 bond->dev->dev_addr, ETH_ALEN)) {
1614 swap_slave = tmp_slave;
1615 break;
1616 }
1617 }
1618 }
1619
1620 /* curr_active_slave must be set before calling alb_swap_mac_addr */
1621 if (swap_slave) {
1622 /* swap mac address */
1623 alb_swap_mac_addr(bond, swap_slave, new_slave);
1624 } else {
1625 /* set the new_slave to the bond mac address */
1626 alb_set_slave_mac_addr(new_slave, bond->dev->dev_addr,
1627 bond->alb_info.rlb_enabled);
1628 /* fasten bond mac on new current slave */
1629 alb_send_learning_packets(new_slave, bond->dev->dev_addr);
1630 }
1631}
1632
1633int bond_alb_set_mac_address(struct net_device *bond_dev, void *addr)
1634{
1635 struct bonding *bond = bond_dev->priv;
1636 struct sockaddr *sa = addr;
1637 struct slave *slave, *swap_slave;
1638 int res;
1639 int i;
1640
1641 if (!is_valid_ether_addr(sa->sa_data)) {
1642 return -EADDRNOTAVAIL;
1643 }
1644
1645 res = alb_set_mac_address(bond, addr);
1646 if (res) {
1647 return res;
1648 }
1649
1650 memcpy(bond_dev->dev_addr, sa->sa_data, bond_dev->addr_len);
1651
1652 /* If there is no curr_active_slave there is nothing else to do.
1653 * Otherwise we'll need to pass the new address to it and handle
1654 * duplications.
1655 */
1656 if (!bond->curr_active_slave) {
1657 return 0;
1658 }
1659
1660 swap_slave = NULL;
1661
1662 bond_for_each_slave(bond, slave, i) {
1663 if (!memcmp(slave->dev->dev_addr, bond_dev->dev_addr, ETH_ALEN)) {
1664 swap_slave = slave;
1665 break;
1666 }
1667 }
1668
1669 if (swap_slave) {
1670 alb_swap_mac_addr(bond, swap_slave, bond->curr_active_slave);
1671 } else {
1672 alb_set_slave_mac_addr(bond->curr_active_slave, bond_dev->dev_addr,
1673 bond->alb_info.rlb_enabled);
1674
1675 alb_send_learning_packets(bond->curr_active_slave, bond_dev->dev_addr);
1676 if (bond->alb_info.rlb_enabled) {
1677 /* inform clients mac address has changed */
1678 rlb_req_update_slave_clients(bond, bond->curr_active_slave);
1679 }
1680 }
1681
1682 return 0;
1683}
1684
1685void bond_alb_clear_vlan(struct bonding *bond, unsigned short vlan_id)
1686{
1687 if (bond->alb_info.current_alb_vlan &&
1688 (bond->alb_info.current_alb_vlan->vlan_id == vlan_id)) {
1689 bond->alb_info.current_alb_vlan = NULL;
1690 }
1691
1692 if (bond->alb_info.rlb_enabled) {
1693 rlb_clear_vlan(bond, vlan_id);
1694 }
1695}
1696
diff --git a/drivers/net/bonding/bond_alb.h b/drivers/net/bonding/bond_alb.h
new file mode 100644
index 000000000000..e4091cd8d654
--- /dev/null
+++ b/drivers/net/bonding/bond_alb.h
@@ -0,0 +1,141 @@
1/*
2 * Copyright(c) 1999 - 2004 Intel Corporation. All rights reserved.
3 *
4 * This program is free software; you can redistribute it and/or modify it
5 * under the terms of the GNU General Public License as published by the
6 * Free Software Foundation; either version 2 of the License, or
7 * (at your option) any later version.
8 *
9 * This program is distributed in the hope that it will be useful, but
10 * WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY
11 * or FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
12 * for more details.
13 *
14 * You should have received a copy of the GNU General Public License along
15 * with this program; if not, write to the Free Software Foundation, Inc.,
16 * 59 Temple Place - Suite 330, Boston, MA 02111-1307, USA.
17 *
18 * The full GNU General Public License is included in this distribution in the
19 * file called LICENSE.
20 *
21 *
22 * Changes:
23 *
24 * 2003/08/06 - Amir Noam <amir.noam at intel dot com>
25 * - Add support for setting bond's MAC address with special
26 * handling required for ALB/TLB.
27 *
28 * 2003/12/01 - Shmulik Hen <shmulik.hen at intel dot com>
29 * - Code cleanup and style changes
30 */
31
32#ifndef __BOND_ALB_H__
33#define __BOND_ALB_H__
34
35#include <linux/if_ether.h>
36
37struct bonding;
38struct slave;
39
40#define BOND_ALB_INFO(bond) ((bond)->alb_info)
41#define SLAVE_TLB_INFO(slave) ((slave)->tlb_info)
42
43struct tlb_client_info {
44 struct slave *tx_slave; /* A pointer to slave used for transmiting
45 * packets to a Client that the Hash function
46 * gave this entry index.
47 */
48 u32 tx_bytes; /* Each Client acumulates the BytesTx that
49 * were tranmitted to it, and after each
50 * CallBack the LoadHistory is devided
51 * by the balance interval
52 */
53 u32 load_history; /* This field contains the amount of Bytes
54 * that were transmitted to this client by
55 * the server on the previous balance
56 * interval in Bps.
57 */
58 u32 next; /* The next Hash table entry index, assigned
59 * to use the same adapter for transmit.
60 */
61 u32 prev; /* The previous Hash table entry index,
62 * assigned to use the same
63 */
64};
65
66/* -------------------------------------------------------------------------
67 * struct rlb_client_info contains all info related to a specific rx client
68 * connection. This is the Clients Hash Table entry struct
69 * -------------------------------------------------------------------------
70 */
71struct rlb_client_info {
72 u32 ip_src; /* the server IP address */
73 u32 ip_dst; /* the client IP address */
74 u8 mac_dst[ETH_ALEN]; /* the client MAC address */
75 u32 next; /* The next Hash table entry index */
76 u32 prev; /* The previous Hash table entry index */
77 u8 assigned; /* checking whether this entry is assigned */
78 u8 ntt; /* flag - need to transmit client info */
79 struct slave *slave; /* the slave assigned to this client */
80 u8 tag; /* flag - need to tag skb */
81 unsigned short vlan_id; /* VLAN tag associated with IP address */
82};
83
84struct tlb_slave_info {
85 u32 head; /* Index to the head of the bi-directional clients
86 * hash table entries list. The entries in the list
87 * are the entries that were assigned to use this
88 * slave for transmit.
89 */
90 u32 load; /* Each slave sums the loadHistory of all clients
91 * assigned to it
92 */
93};
94
95struct alb_bond_info {
96 struct timer_list alb_timer;
97 struct tlb_client_info *tx_hashtbl; /* Dynamically allocated */
98 spinlock_t tx_hashtbl_lock;
99 u32 unbalanced_load;
100 int tx_rebalance_counter;
101 int lp_counter;
102 /* -------- rlb parameters -------- */
103 int rlb_enabled;
104 struct packet_type rlb_pkt_type;
105 struct rlb_client_info *rx_hashtbl; /* Receive hash table */
106 spinlock_t rx_hashtbl_lock;
107 u32 rx_hashtbl_head;
108 u8 rx_ntt; /* flag - need to transmit
109 * to all rx clients
110 */
111 struct slave *next_rx_slave;/* next slave to be assigned
112 * to a new rx client for
113 */
114 u32 rlb_interval_counter;
115 u8 primary_is_promisc; /* boolean */
116 u32 rlb_promisc_timeout_counter;/* counts primary
117 * promiscuity time
118 */
119 u32 rlb_update_delay_counter;
120 u32 rlb_update_retry_counter;/* counter of retries
121 * of client update
122 */
123 u8 rlb_rebalance; /* flag - indicates that the
124 * rx traffic should be
125 * rebalanced
126 */
127 struct vlan_entry *current_alb_vlan;
128};
129
130int bond_alb_initialize(struct bonding *bond, int rlb_enabled);
131void bond_alb_deinitialize(struct bonding *bond);
132int bond_alb_init_slave(struct bonding *bond, struct slave *slave);
133void bond_alb_deinit_slave(struct bonding *bond, struct slave *slave);
134void bond_alb_handle_link_change(struct bonding *bond, struct slave *slave, char link);
135void bond_alb_handle_active_change(struct bonding *bond, struct slave *new_slave);
136int bond_alb_xmit(struct sk_buff *skb, struct net_device *bond_dev);
137void bond_alb_monitor(struct bonding *bond);
138int bond_alb_set_mac_address(struct net_device *bond_dev, void *addr);
139void bond_alb_clear_vlan(struct bonding *bond, unsigned short vlan_id);
140#endif /* __BOND_ALB_H__ */
141
diff --git a/drivers/net/bonding/bond_main.c b/drivers/net/bonding/bond_main.c
new file mode 100644
index 000000000000..770e28f98fd5
--- /dev/null
+++ b/drivers/net/bonding/bond_main.c
@@ -0,0 +1,4708 @@
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 * v0.1 - first working version.
33 * v0.2 - changed stats to be calculated by summing slaves stats.
34 *
35 * Changes:
36 * Arnaldo Carvalho de Melo <acme@conectiva.com.br>
37 * - fix leaks on failure at bond_init
38 *
39 * 2000/09/30 - Willy Tarreau <willy at meta-x.org>
40 * - added trivial code to release a slave device.
41 * - fixed security bug (CAP_NET_ADMIN not checked)
42 * - implemented MII link monitoring to disable dead links :
43 * All MII capable slaves are checked every <miimon> milliseconds
44 * (100 ms seems good). This value can be changed by passing it to
45 * insmod. A value of zero disables the monitoring (default).
46 * - fixed an infinite loop in bond_xmit_roundrobin() when there's no
47 * good slave.
48 * - made the code hopefully SMP safe
49 *
50 * 2000/10/03 - Willy Tarreau <willy at meta-x.org>
51 * - optimized slave lists based on relevant suggestions from Thomas Davis
52 * - implemented active-backup method to obtain HA with two switches:
53 * stay as long as possible on the same active interface, while we
54 * also monitor the backup one (MII link status) because we want to know
55 * if we are able to switch at any time. ( pass "mode=1" to insmod )
56 * - lots of stress testings because we need it to be more robust than the
57 * wires ! :->
58 *
59 * 2000/10/09 - Willy Tarreau <willy at meta-x.org>
60 * - added up and down delays after link state change.
61 * - optimized the slaves chaining so that when we run forward, we never
62 * repass through the bond itself, but we can find it by searching
63 * backwards. Renders the deletion more difficult, but accelerates the
64 * scan.
65 * - smarter enslaving and releasing.
66 * - finer and more robust SMP locking
67 *
68 * 2000/10/17 - Willy Tarreau <willy at meta-x.org>
69 * - fixed two potential SMP race conditions
70 *
71 * 2000/10/18 - Willy Tarreau <willy at meta-x.org>
72 * - small fixes to the monitoring FSM in case of zero delays
73 * 2000/11/01 - Willy Tarreau <willy at meta-x.org>
74 * - fixed first slave not automatically used in trunk mode.
75 * 2000/11/10 : spelling of "EtherChannel" corrected.
76 * 2000/11/13 : fixed a race condition in case of concurrent accesses to ioctl().
77 * 2000/12/16 : fixed improper usage of rtnl_exlock_nowait().
78 *
79 * 2001/1/3 - Chad N. Tindel <ctindel at ieee dot org>
80 * - The bonding driver now simulates MII status monitoring, just like
81 * a normal network device. It will show that the link is down iff
82 * every slave in the bond shows that their links are down. If at least
83 * one slave is up, the bond's MII status will appear as up.
84 *
85 * 2001/2/7 - Chad N. Tindel <ctindel at ieee dot org>
86 * - Applications can now query the bond from user space to get
87 * information which may be useful. They do this by calling
88 * the BOND_INFO_QUERY ioctl. Once the app knows how many slaves
89 * are in the bond, it can call the BOND_SLAVE_INFO_QUERY ioctl to
90 * get slave specific information (# link failures, etc). See
91 * <linux/if_bonding.h> for more details. The structs of interest
92 * are ifbond and ifslave.
93 *
94 * 2001/4/5 - Chad N. Tindel <ctindel at ieee dot org>
95 * - Ported to 2.4 Kernel
96 *
97 * 2001/5/2 - Jeffrey E. Mast <jeff at mastfamily dot com>
98 * - When a device is detached from a bond, the slave device is no longer
99 * left thinking that is has a master.
100 *
101 * 2001/5/16 - Jeffrey E. Mast <jeff at mastfamily dot com>
102 * - memset did not appropriately initialized the bond rw_locks. Used
103 * rwlock_init to initialize to unlocked state to prevent deadlock when
104 * first attempting a lock
105 * - Called SET_MODULE_OWNER for bond device
106 *
107 * 2001/5/17 - Tim Anderson <tsa at mvista.com>
108 * - 2 paths for releasing for slave release; 1 through ioctl
109 * and 2) through close. Both paths need to release the same way.
110 * - the free slave in bond release is changing slave status before
111 * the free. The netdev_set_master() is intended to change slave state
112 * so it should not be done as part of the release process.
113 * - Simple rule for slave state at release: only the active in A/B and
114 * only one in the trunked case.
115 *
116 * 2001/6/01 - Tim Anderson <tsa at mvista.com>
117 * - Now call dev_close when releasing a slave so it doesn't screw up
118 * out routing table.
119 *
120 * 2001/6/01 - Chad N. Tindel <ctindel at ieee dot org>
121 * - Added /proc support for getting bond and slave information.
122 * Information is in /proc/net/<bond device>/info.
123 * - Changed the locking when calling bond_close to prevent deadlock.
124 *
125 * 2001/8/05 - Janice Girouard <girouard at us.ibm.com>
126 * - correct problem where refcnt of slave is not incremented in bond_ioctl
127 * so the system hangs when halting.
128 * - correct locking problem when unable to malloc in bond_enslave.
129 * - adding bond_xmit_xor logic.
130 * - adding multiple bond device support.
131 *
132 * 2001/8/13 - Erik Habbinga <erik_habbinga at hp dot com>
133 * - correct locking problem with rtnl_exlock_nowait
134 *
135 * 2001/8/23 - Janice Girouard <girouard at us.ibm.com>
136 * - bzero initial dev_bonds, to correct oops
137 * - convert SIOCDEVPRIVATE to new MII ioctl calls
138 *
139 * 2001/9/13 - Takao Indoh <indou dot takao at jp dot fujitsu dot com>
140 * - Add the BOND_CHANGE_ACTIVE ioctl implementation
141 *
142 * 2001/9/14 - Mark Huth <mhuth at mvista dot com>
143 * - Change MII_LINK_READY to not check for end of auto-negotiation,
144 * but only for an up link.
145 *
146 * 2001/9/20 - Chad N. Tindel <ctindel at ieee dot org>
147 * - Add the device field to bonding_t. Previously the net_device
148 * corresponding to a bond wasn't available from the bonding_t
149 * structure.
150 *
151 * 2001/9/25 - Janice Girouard <girouard at us.ibm.com>
152 * - add arp_monitor for active backup mode
153 *
154 * 2001/10/23 - Takao Indoh <indou dot takao at jp dot fujitsu dot com>
155 * - Various memory leak fixes
156 *
157 * 2001/11/5 - Mark Huth <mark dot huth at mvista dot com>
158 * - Don't take rtnl lock in bond_mii_monitor as it deadlocks under
159 * certain hotswap conditions.
160 * Note: this same change may be required in bond_arp_monitor ???
161 * - Remove possibility of calling bond_sethwaddr with NULL slave_dev ptr
162 * - Handle hot swap ethernet interface deregistration events to remove
163 * kernel oops following hot swap of enslaved interface
164 *
165 * 2002/1/2 - Chad N. Tindel <ctindel at ieee dot org>
166 * - Restore original slave flags at release time.
167 *
168 * 2002/02/18 - Erik Habbinga <erik_habbinga at hp dot com>
169 * - bond_release(): calling kfree on our_slave after call to
170 * bond_restore_slave_flags, not before
171 * - bond_enslave(): saving slave flags into original_flags before
172 * call to netdev_set_master, so the IFF_SLAVE flag doesn't end
173 * up in original_flags
174 *
175 * 2002/04/05 - Mark Smith <mark.smith at comdev dot cc> and
176 * Steve Mead <steve.mead at comdev dot cc>
177 * - Port Gleb Natapov's multicast support patchs from 2.4.12
178 * to 2.4.18 adding support for multicast.
179 *
180 * 2002/06/10 - Tony Cureington <tony.cureington * hp_com>
181 * - corrected uninitialized pointer (ifr.ifr_data) in bond_check_dev_link;
182 * actually changed function to use MIIPHY, then MIIREG, and finally
183 * ETHTOOL to determine the link status
184 * - fixed bad ifr_data pointer assignments in bond_ioctl
185 * - corrected mode 1 being reported as active-backup in bond_get_info;
186 * also added text to distinguish type of load balancing (rr or xor)
187 * - change arp_ip_target module param from "1-12s" (array of 12 ptrs)
188 * to "s" (a single ptr)
189 *
190 * 2002/08/30 - Jay Vosburgh <fubar at us dot ibm dot com>
191 * - Removed acquisition of xmit_lock in set_multicast_list; caused
192 * deadlock on SMP (lock is held by caller).
193 * - Revamped SIOCGMIIPHY, SIOCGMIIREG portion of bond_check_dev_link().
194 *
195 * 2002/09/18 - Jay Vosburgh <fubar at us dot ibm dot com>
196 * - Fixed up bond_check_dev_link() (and callers): removed some magic
197 * numbers, banished local MII_ defines, wrapped ioctl calls to
198 * prevent EFAULT errors
199 *
200 * 2002/9/30 - Jay Vosburgh <fubar at us dot ibm dot com>
201 * - make sure the ip target matches the arp_target before saving the
202 * hw address.
203 *
204 * 2002/9/30 - Dan Eisner <eisner at 2robots dot com>
205 * - make sure my_ip is set before taking down the link, since
206 * not all switches respond if the source ip is not set.
207 *
208 * 2002/10/8 - Janice Girouard <girouard at us dot ibm dot com>
209 * - read in the local ip address when enslaving a device
210 * - add primary support
211 * - make sure 2*arp_interval has passed when a new device
212 * is brought on-line before taking it down.
213 *
214 * 2002/09/11 - Philippe De Muyter <phdm at macqel dot be>
215 * - Added bond_xmit_broadcast logic.
216 * - Added bond_mode() support function.
217 *
218 * 2002/10/26 - Laurent Deniel <laurent.deniel at free.fr>
219 * - allow to register multicast addresses only on active slave
220 * (useful in active-backup mode)
221 * - add multicast module parameter
222 * - fix deletion of multicast groups after unloading module
223 *
224 * 2002/11/06 - Kameshwara Rayaprolu <kameshwara.rao * wipro_com>
225 * - Changes to prevent panic from closing the device twice; if we close
226 * the device in bond_release, we must set the original_flags to down
227 * so it won't be closed again by the network layer.
228 *
229 * 2002/11/07 - Tony Cureington <tony.cureington * hp_com>
230 * - Fix arp_target_hw_addr memory leak
231 * - Created activebackup_arp_monitor function to handle arp monitoring
232 * in active backup mode - the bond_arp_monitor had several problems...
233 * such as allowing slaves to tx arps sequentially without any delay
234 * for a response
235 * - Renamed bond_arp_monitor to loadbalance_arp_monitor and re-wrote
236 * this function to just handle arp monitoring in load-balancing mode;
237 * it is a lot more compact now
238 * - Changes to ensure one and only one slave transmits in active-backup
239 * mode
240 * - Robustesize parameters; warn users about bad combinations of
241 * parameters; also if miimon is specified and a network driver does
242 * not support MII or ETHTOOL, inform the user of this
243 * - Changes to support link_failure_count when in arp monitoring mode
244 * - Fix up/down delay reported in /proc
245 * - Added version; log version; make version available from "modinfo -d"
246 * - Fixed problem in bond_check_dev_link - if the first IOCTL (SIOCGMIIPH)
247 * failed, the ETHTOOL ioctl never got a chance
248 *
249 * 2002/11/16 - Laurent Deniel <laurent.deniel at free.fr>
250 * - fix multicast handling in activebackup_arp_monitor
251 * - remove one unnecessary and confusing curr_active_slave == slave test
252 * in activebackup_arp_monitor
253 *
254 * 2002/11/17 - Laurent Deniel <laurent.deniel at free.fr>
255 * - fix bond_slave_info_query when slave_id = num_slaves
256 *
257 * 2002/11/19 - Janice Girouard <girouard at us dot ibm dot com>
258 * - correct ifr_data reference. Update ifr_data reference
259 * to mii_ioctl_data struct values to avoid confusion.
260 *
261 * 2002/11/22 - Bert Barbe <bert.barbe at oracle dot com>
262 * - Add support for multiple arp_ip_target
263 *
264 * 2002/12/13 - Jay Vosburgh <fubar at us dot ibm dot com>
265 * - Changed to allow text strings for mode and multicast, e.g.,
266 * insmod bonding mode=active-backup. The numbers still work.
267 * One change: an invalid choice will cause module load failure,
268 * rather than the previous behavior of just picking one.
269 * - Minor cleanups; got rid of dup ctype stuff, atoi function
270 *
271 * 2003/02/07 - Jay Vosburgh <fubar at us dot ibm dot com>
272 * - Added use_carrier module parameter that causes miimon to
273 * use netif_carrier_ok() test instead of MII/ETHTOOL ioctls.
274 * - Minor cleanups; consolidated ioctl calls to one function.
275 *
276 * 2003/02/07 - Tony Cureington <tony.cureington * hp_com>
277 * - Fix bond_mii_monitor() logic error that could result in
278 * bonding round-robin mode ignoring links after failover/recovery
279 *
280 * 2003/03/17 - Jay Vosburgh <fubar at us dot ibm dot com>
281 * - kmalloc fix (GFP_KERNEL to GFP_ATOMIC) reported by
282 * Shmulik dot Hen at intel.com.
283 * - Based on discussion on mailing list, changed use of
284 * update_slave_cnt(), created wrapper functions for adding/removing
285 * slaves, changed bond_xmit_xor() to check slave_cnt instead of
286 * checking slave and slave->dev (which only worked by accident).
287 * - Misc code cleanup: get arp_send() prototype from header file,
288 * add max_bonds to bonding.txt.
289 *
290 * 2003/03/18 - Tsippy Mendelson <tsippy.mendelson at intel dot com> and
291 * Shmulik Hen <shmulik.hen at intel dot com>
292 * - Make sure only bond_attach_slave() and bond_detach_slave() can
293 * manipulate the slave list, including slave_cnt, even when in
294 * bond_release_all().
295 * - Fixed hang in bond_release() with traffic running:
296 * netdev_set_master() must not be called from within the bond lock.
297 *
298 * 2003/03/18 - Tsippy Mendelson <tsippy.mendelson at intel dot com> and
299 * Shmulik Hen <shmulik.hen at intel dot com>
300 * - Fixed hang in bond_enslave() with traffic running:
301 * netdev_set_master() must not be called from within the bond lock.
302 *
303 * 2003/03/18 - Amir Noam <amir.noam at intel dot com>
304 * - Added support for getting slave's speed and duplex via ethtool.
305 * Needed for 802.3ad and other future modes.
306 *
307 * 2003/03/18 - Tsippy Mendelson <tsippy.mendelson at intel dot com> and
308 * Shmulik Hen <shmulik.hen at intel dot com>
309 * - Enable support of modes that need to use the unique mac address of
310 * each slave.
311 * * bond_enslave(): Moved setting the slave's mac address, and
312 * openning it, from the application to the driver. This breaks
313 * backward comaptibility with old versions of ifenslave that open
314 * the slave before enalsving it !!!.
315 * * bond_release(): The driver also takes care of closing the slave
316 * and restoring its original mac address.
317 * - Removed the code that restores all base driver's flags.
318 * Flags are automatically restored once all undo stages are done
319 * properly.
320 * - Block possibility of enslaving before the master is up. This
321 * prevents putting the system in an unstable state.
322 *
323 * 2003/03/18 - Amir Noam <amir.noam at intel dot com>,
324 * Tsippy Mendelson <tsippy.mendelson at intel dot com> and
325 * Shmulik Hen <shmulik.hen at intel dot com>
326 * - Added support for IEEE 802.3ad Dynamic link aggregation mode.
327 *
328 * 2003/05/01 - Amir Noam <amir.noam at intel dot com>
329 * - Added ABI version control to restore compatibility between
330 * new/old ifenslave and new/old bonding.
331 *
332 * 2003/05/01 - Shmulik Hen <shmulik.hen at intel dot com>
333 * - Fixed bug in bond_release_all(): save old value of curr_active_slave
334 * before setting it to NULL.
335 * - Changed driver versioning scheme to include version number instead
336 * of release date (that is already in another field). There are 3
337 * fields X.Y.Z where:
338 * X - Major version - big behavior changes
339 * Y - Minor version - addition of features
340 * Z - Extra version - minor changes and bug fixes
341 * The current version is 1.0.0 as a base line.
342 *
343 * 2003/05/01 - Tsippy Mendelson <tsippy.mendelson at intel dot com> and
344 * Amir Noam <amir.noam at intel dot com>
345 * - Added support for lacp_rate module param.
346 * - Code beautification and style changes (mainly in comments).
347 * new version - 1.0.1
348 *
349 * 2003/05/01 - Shmulik Hen <shmulik.hen at intel dot com>
350 * - Based on discussion on mailing list, changed locking scheme
351 * to use lock/unlock or lock_bh/unlock_bh appropriately instead
352 * of lock_irqsave/unlock_irqrestore. The new scheme helps exposing
353 * hidden bugs and solves system hangs that occurred due to the fact
354 * that holding lock_irqsave doesn't prevent softirqs from running.
355 * This also increases total throughput since interrupts are not
356 * blocked on each transmitted packets or monitor timeout.
357 * new version - 2.0.0
358 *
359 * 2003/05/01 - Shmulik Hen <shmulik.hen at intel dot com>
360 * - Added support for Transmit load balancing mode.
361 * - Concentrate all assignments of curr_active_slave to a single point
362 * so specific modes can take actions when the primary adapter is
363 * changed.
364 * - Take the updelay parameter into consideration during bond_enslave
365 * since some adapters loose their link during setting the device.
366 * - Renamed bond_3ad_link_status_changed() to
367 * bond_3ad_handle_link_change() for compatibility with TLB.
368 * new version - 2.1.0
369 *
370 * 2003/05/01 - Tsippy Mendelson <tsippy.mendelson at intel dot com>
371 * - Added support for Adaptive load balancing mode which is
372 * equivalent to Transmit load balancing + Receive load balancing.
373 * new version - 2.2.0
374 *
375 * 2003/05/15 - Jay Vosburgh <fubar at us dot ibm dot com>
376 * - Applied fix to activebackup_arp_monitor posted to bonding-devel
377 * by Tony Cureington <tony.cureington * hp_com>. Fixes ARP
378 * monitor endless failover bug. Version to 2.2.10
379 *
380 * 2003/05/20 - Amir Noam <amir.noam at intel dot com>
381 * - Fixed bug in ABI version control - Don't commit to a specific
382 * ABI version if receiving unsupported ioctl commands.
383 *
384 * 2003/05/22 - Jay Vosburgh <fubar at us dot ibm dot com>
385 * - Fix ifenslave -c causing bond to loose existing routes;
386 * added bond_set_mac_address() that doesn't require the
387 * bond to be down.
388 * - In conjunction with fix for ifenslave -c, in
389 * bond_change_active(), changing to the already active slave
390 * is no longer an error (it successfully does nothing).
391 *
392 * 2003/06/30 - Amir Noam <amir.noam at intel dot com>
393 * - Fixed bond_change_active() for ALB/TLB modes.
394 * Version to 2.2.14.
395 *
396 * 2003/07/29 - Amir Noam <amir.noam at intel dot com>
397 * - Fixed ARP monitoring bug.
398 * Version to 2.2.15.
399 *
400 * 2003/07/31 - Willy Tarreau <willy at ods dot org>
401 * - Fixed kernel panic when using ARP monitoring without
402 * setting bond's IP address.
403 * Version to 2.2.16.
404 *
405 * 2003/08/06 - Amir Noam <amir.noam at intel dot com>
406 * - Back port from 2.6: use alloc_netdev(); fix /proc handling;
407 * made stats a part of bond struct so no need to allocate
408 * and free it separately; use standard list operations instead
409 * of pre-allocated array of bonds.
410 * Version to 2.3.0.
411 *
412 * 2003/08/07 - Jay Vosburgh <fubar at us dot ibm dot com>,
413 * Amir Noam <amir.noam at intel dot com> and
414 * Shmulik Hen <shmulik.hen at intel dot com>
415 * - Propagating master's settings: Distinguish between modes that
416 * use a primary slave from those that don't, and propagate settings
417 * accordingly; Consolidate change_active opeartions and add
418 * reselect_active and find_best opeartions; Decouple promiscuous
419 * handling from the multicast mode setting; Add support for changing
420 * HW address and MTU with proper unwind; Consolidate procfs code,
421 * add CHANGENAME handler; Enhance netdev notification handling.
422 * Version to 2.4.0.
423 *
424 * 2003/09/15 - Stephen Hemminger <shemminger at osdl dot org>,
425 * Amir Noam <amir.noam at intel dot com>
426 * - Convert /proc to seq_file interface.
427 * Change /proc/net/bondX/info to /proc/net/bonding/bondX.
428 * Set version to 2.4.1.
429 *
430 * 2003/11/20 - Amir Noam <amir.noam at intel dot com>
431 * - Fix /proc creation/destruction.
432 *
433 * 2003/12/01 - Shmulik Hen <shmulik.hen at intel dot com>
434 * - Massive cleanup - Set version to 2.5.0
435 * Code changes:
436 * o Consolidate format of prints and debug prints.
437 * o Remove bonding_t/slave_t typedefs and consolidate all casts.
438 * o Remove dead code and unnecessary checks.
439 * o Consolidate starting/stopping timers.
440 * o Consolidate handling of primary module param throughout the code.
441 * o Removed multicast module param support - all settings are done
442 * according to mode.
443 * o Slave list iteration - bond is no longer part of the list,
444 * added cyclic list iteration macros.
445 * o Consolidate error handling in all xmit functions.
446 * Style changes:
447 * o Consolidate function naming and declarations.
448 * o Consolidate function params and local variables names.
449 * o Consolidate return values.
450 * o Consolidate curly braces.
451 * o Consolidate conditionals format.
452 * o Change struct member names and types.
453 * o Chomp trailing spaces, remove empty lines, fix indentations.
454 * o Re-organize code according to context.
455 *
456 * 2003/12/30 - Amir Noam <amir.noam at intel dot com>
457 * - Fixed: Cannot remove and re-enslave the original active slave.
458 * - Fixed: Releasing the original active slave causes mac address
459 * duplication.
460 * - Add support for slaves that use ethtool_ops.
461 * Set version to 2.5.3.
462 *
463 * 2004/01/05 - Amir Noam <amir.noam at intel dot com>
464 * - Save bonding parameters per bond instead of using the global values.
465 * Set version to 2.5.4.
466 *
467 * 2004/01/14 - Shmulik Hen <shmulik.hen at intel dot com>
468 * - Enhance VLAN support:
469 * * Add support for VLAN hardware acceleration capable slaves.
470 * * Add capability to tag self generated packets in ALB/TLB modes.
471 * Set version to 2.6.0.
472 * 2004/10/29 - Mitch Williams <mitch.a.williams at intel dot com>
473 * - Fixed bug when unloading module while using 802.3ad. If
474 * spinlock debugging is turned on, this causes a stack dump.
475 * Solution is to move call to dev_remove_pack outside of the
476 * spinlock.
477 * Set version to 2.6.1.
478 *
479 */
480
481//#define BONDING_DEBUG 1
482
483#include <linux/config.h>
484#include <linux/kernel.h>
485#include <linux/module.h>
486#include <linux/sched.h>
487#include <linux/types.h>
488#include <linux/fcntl.h>
489#include <linux/interrupt.h>
490#include <linux/ptrace.h>
491#include <linux/ioport.h>
492#include <linux/in.h>
493#include <linux/ip.h>
494#include <linux/slab.h>
495#include <linux/string.h>
496#include <linux/init.h>
497#include <linux/timer.h>
498#include <linux/socket.h>
499#include <linux/ctype.h>
500#include <linux/inet.h>
501#include <linux/bitops.h>
502#include <asm/system.h>
503#include <asm/io.h>
504#include <asm/dma.h>
505#include <asm/uaccess.h>
506#include <linux/errno.h>
507#include <linux/netdevice.h>
508#include <linux/inetdevice.h>
509#include <linux/etherdevice.h>
510#include <linux/skbuff.h>
511#include <net/sock.h>
512#include <linux/rtnetlink.h>
513#include <linux/proc_fs.h>
514#include <linux/seq_file.h>
515#include <linux/smp.h>
516#include <linux/if_ether.h>
517#include <net/arp.h>
518#include <linux/mii.h>
519#include <linux/ethtool.h>
520#include <linux/if_vlan.h>
521#include <linux/if_bonding.h>
522#include "bonding.h"
523#include "bond_3ad.h"
524#include "bond_alb.h"
525
526/*---------------------------- Module parameters ----------------------------*/
527
528/* monitor all links that often (in milliseconds). <=0 disables monitoring */
529#define BOND_LINK_MON_INTERV 0
530#define BOND_LINK_ARP_INTERV 0
531
532static int max_bonds = BOND_DEFAULT_MAX_BONDS;
533static int miimon = BOND_LINK_MON_INTERV;
534static int updelay = 0;
535static int downdelay = 0;
536static int use_carrier = 1;
537static char *mode = NULL;
538static char *primary = NULL;
539static char *lacp_rate = NULL;
540static int arp_interval = BOND_LINK_ARP_INTERV;
541static char *arp_ip_target[BOND_MAX_ARP_TARGETS] = { NULL, };
542
543module_param(max_bonds, int, 0);
544MODULE_PARM_DESC(max_bonds, "Max number of bonded devices");
545module_param(miimon, int, 0);
546MODULE_PARM_DESC(miimon, "Link check interval in milliseconds");
547module_param(updelay, int, 0);
548MODULE_PARM_DESC(updelay, "Delay before considering link up, in milliseconds");
549module_param(downdelay, int, 0);
550MODULE_PARM_DESC(downdelay, "Delay before considering link down, in milliseconds");
551module_param(use_carrier, int, 0);
552MODULE_PARM_DESC(use_carrier, "Use netif_carrier_ok (vs MII ioctls) in miimon; 0 for off, 1 for on (default)");
553module_param(mode, charp, 0);
554MODULE_PARM_DESC(mode, "Mode of operation : 0 for round robin, 1 for active-backup, 2 for xor");
555module_param(primary, charp, 0);
556MODULE_PARM_DESC(primary, "Primary network device to use");
557module_param(lacp_rate, charp, 0);
558MODULE_PARM_DESC(lacp_rate, "LACPDU tx rate to request from 802.3ad partner (slow/fast)");
559module_param(arp_interval, int, 0);
560MODULE_PARM_DESC(arp_interval, "arp interval in milliseconds");
561module_param_array(arp_ip_target, charp, NULL, 0);
562MODULE_PARM_DESC(arp_ip_target, "arp targets in n.n.n.n form");
563
564/*----------------------------- Global variables ----------------------------*/
565
566static const char *version =
567 DRV_DESCRIPTION ": v" DRV_VERSION " (" DRV_RELDATE ")\n";
568
569static LIST_HEAD(bond_dev_list);
570
571#ifdef CONFIG_PROC_FS
572static struct proc_dir_entry *bond_proc_dir = NULL;
573#endif
574
575static u32 arp_target[BOND_MAX_ARP_TARGETS] = { 0, } ;
576static int arp_ip_count = 0;
577static u32 my_ip = 0;
578static int bond_mode = BOND_MODE_ROUNDROBIN;
579static int lacp_fast = 0;
580static int app_abi_ver = 0;
581static int orig_app_abi_ver = -1; /* This is used to save the first ABI version
582 * we receive from the application. Once set,
583 * it won't be changed, and the module will
584 * refuse to enslave/release interfaces if the
585 * command comes from an application using
586 * another ABI version.
587 */
588
589struct bond_parm_tbl {
590 char *modename;
591 int mode;
592};
593
594static struct bond_parm_tbl bond_lacp_tbl[] = {
595{ "slow", AD_LACP_SLOW},
596{ "fast", AD_LACP_FAST},
597{ NULL, -1},
598};
599
600static struct bond_parm_tbl bond_mode_tbl[] = {
601{ "balance-rr", BOND_MODE_ROUNDROBIN},
602{ "active-backup", BOND_MODE_ACTIVEBACKUP},
603{ "balance-xor", BOND_MODE_XOR},
604{ "broadcast", BOND_MODE_BROADCAST},
605{ "802.3ad", BOND_MODE_8023AD},
606{ "balance-tlb", BOND_MODE_TLB},
607{ "balance-alb", BOND_MODE_ALB},
608{ NULL, -1},
609};
610
611/*-------------------------- Forward declarations ---------------------------*/
612
613static inline void bond_set_mode_ops(struct net_device *bond_dev, int mode);
614
615/*---------------------------- General routines -----------------------------*/
616
617static const char *bond_mode_name(int mode)
618{
619 switch (mode) {
620 case BOND_MODE_ROUNDROBIN :
621 return "load balancing (round-robin)";
622 case BOND_MODE_ACTIVEBACKUP :
623 return "fault-tolerance (active-backup)";
624 case BOND_MODE_XOR :
625 return "load balancing (xor)";
626 case BOND_MODE_BROADCAST :
627 return "fault-tolerance (broadcast)";
628 case BOND_MODE_8023AD:
629 return "IEEE 802.3ad Dynamic link aggregation";
630 case BOND_MODE_TLB:
631 return "transmit load balancing";
632 case BOND_MODE_ALB:
633 return "adaptive load balancing";
634 default:
635 return "unknown";
636 }
637}
638
639/*---------------------------------- VLAN -----------------------------------*/
640
641/**
642 * bond_add_vlan - add a new vlan id on bond
643 * @bond: bond that got the notification
644 * @vlan_id: the vlan id to add
645 *
646 * Returns -ENOMEM if allocation failed.
647 */
648static int bond_add_vlan(struct bonding *bond, unsigned short vlan_id)
649{
650 struct vlan_entry *vlan;
651
652 dprintk("bond: %s, vlan id %d\n",
653 (bond ? bond->dev->name: "None"), vlan_id);
654
655 vlan = kmalloc(sizeof(struct vlan_entry), GFP_KERNEL);
656 if (!vlan) {
657 return -ENOMEM;
658 }
659
660 INIT_LIST_HEAD(&vlan->vlan_list);
661 vlan->vlan_id = vlan_id;
662
663 write_lock_bh(&bond->lock);
664
665 list_add_tail(&vlan->vlan_list, &bond->vlan_list);
666
667 write_unlock_bh(&bond->lock);
668
669 dprintk("added VLAN ID %d on bond %s\n", vlan_id, bond->dev->name);
670
671 return 0;
672}
673
674/**
675 * bond_del_vlan - delete a vlan id from bond
676 * @bond: bond that got the notification
677 * @vlan_id: the vlan id to delete
678 *
679 * returns -ENODEV if @vlan_id was not found in @bond.
680 */
681static int bond_del_vlan(struct bonding *bond, unsigned short vlan_id)
682{
683 struct vlan_entry *vlan, *next;
684 int res = -ENODEV;
685
686 dprintk("bond: %s, vlan id %d\n", bond->dev->name, vlan_id);
687
688 write_lock_bh(&bond->lock);
689
690 list_for_each_entry_safe(vlan, next, &bond->vlan_list, vlan_list) {
691 if (vlan->vlan_id == vlan_id) {
692 list_del(&vlan->vlan_list);
693
694 if ((bond->params.mode == BOND_MODE_TLB) ||
695 (bond->params.mode == BOND_MODE_ALB)) {
696 bond_alb_clear_vlan(bond, vlan_id);
697 }
698
699 dprintk("removed VLAN ID %d from bond %s\n", vlan_id,
700 bond->dev->name);
701
702 kfree(vlan);
703
704 if (list_empty(&bond->vlan_list) &&
705 (bond->slave_cnt == 0)) {
706 /* Last VLAN removed and no slaves, so
707 * restore block on adding VLANs. This will
708 * be removed once new slaves that are not
709 * VLAN challenged will be added.
710 */
711 bond->dev->features |= NETIF_F_VLAN_CHALLENGED;
712 }
713
714 res = 0;
715 goto out;
716 }
717 }
718
719 dprintk("couldn't find VLAN ID %d in bond %s\n", vlan_id,
720 bond->dev->name);
721
722out:
723 write_unlock_bh(&bond->lock);
724 return res;
725}
726
727/**
728 * bond_has_challenged_slaves
729 * @bond: the bond we're working on
730 *
731 * Searches the slave list. Returns 1 if a vlan challenged slave
732 * was found, 0 otherwise.
733 *
734 * Assumes bond->lock is held.
735 */
736static int bond_has_challenged_slaves(struct bonding *bond)
737{
738 struct slave *slave;
739 int i;
740
741 bond_for_each_slave(bond, slave, i) {
742 if (slave->dev->features & NETIF_F_VLAN_CHALLENGED) {
743 dprintk("found VLAN challenged slave - %s\n",
744 slave->dev->name);
745 return 1;
746 }
747 }
748
749 dprintk("no VLAN challenged slaves found\n");
750 return 0;
751}
752
753/**
754 * bond_next_vlan - safely skip to the next item in the vlans list.
755 * @bond: the bond we're working on
756 * @curr: item we're advancing from
757 *
758 * Returns %NULL if list is empty, bond->next_vlan if @curr is %NULL,
759 * or @curr->next otherwise (even if it is @curr itself again).
760 *
761 * Caller must hold bond->lock
762 */
763struct vlan_entry *bond_next_vlan(struct bonding *bond, struct vlan_entry *curr)
764{
765 struct vlan_entry *next, *last;
766
767 if (list_empty(&bond->vlan_list)) {
768 return NULL;
769 }
770
771 if (!curr) {
772 next = list_entry(bond->vlan_list.next,
773 struct vlan_entry, vlan_list);
774 } else {
775 last = list_entry(bond->vlan_list.prev,
776 struct vlan_entry, vlan_list);
777 if (last == curr) {
778 next = list_entry(bond->vlan_list.next,
779 struct vlan_entry, vlan_list);
780 } else {
781 next = list_entry(curr->vlan_list.next,
782 struct vlan_entry, vlan_list);
783 }
784 }
785
786 return next;
787}
788
789/**
790 * bond_dev_queue_xmit - Prepare skb for xmit.
791 *
792 * @bond: bond device that got this skb for tx.
793 * @skb: hw accel VLAN tagged skb to transmit
794 * @slave_dev: slave that is supposed to xmit this skbuff
795 *
796 * When the bond gets an skb to transmit that is
797 * already hardware accelerated VLAN tagged, and it
798 * needs to relay this skb to a slave that is not
799 * hw accel capable, the skb needs to be "unaccelerated",
800 * i.e. strip the hwaccel tag and re-insert it as part
801 * of the payload.
802 */
803int bond_dev_queue_xmit(struct bonding *bond, struct sk_buff *skb, struct net_device *slave_dev)
804{
805 unsigned short vlan_id;
806
807 if (!list_empty(&bond->vlan_list) &&
808 !(slave_dev->features & NETIF_F_HW_VLAN_TX) &&
809 vlan_get_tag(skb, &vlan_id) == 0) {
810 skb->dev = slave_dev;
811 skb = vlan_put_tag(skb, vlan_id);
812 if (!skb) {
813 /* vlan_put_tag() frees the skb in case of error,
814 * so return success here so the calling functions
815 * won't attempt to free is again.
816 */
817 return 0;
818 }
819 } else {
820 skb->dev = slave_dev;
821 }
822
823 skb->priority = 1;
824 dev_queue_xmit(skb);
825
826 return 0;
827}
828
829/*
830 * In the following 3 functions, bond_vlan_rx_register(), bond_vlan_rx_add_vid
831 * and bond_vlan_rx_kill_vid, We don't protect the slave list iteration with a
832 * lock because:
833 * a. This operation is performed in IOCTL context,
834 * b. The operation is protected by the RTNL semaphore in the 8021q code,
835 * c. Holding a lock with BH disabled while directly calling a base driver
836 * entry point is generally a BAD idea.
837 *
838 * The design of synchronization/protection for this operation in the 8021q
839 * module is good for one or more VLAN devices over a single physical device
840 * and cannot be extended for a teaming solution like bonding, so there is a
841 * potential race condition here where a net device from the vlan group might
842 * be referenced (either by a base driver or the 8021q code) while it is being
843 * removed from the system. However, it turns out we're not making matters
844 * worse, and if it works for regular VLAN usage it will work here too.
845*/
846
847/**
848 * bond_vlan_rx_register - Propagates registration to slaves
849 * @bond_dev: bonding net device that got called
850 * @grp: vlan group being registered
851 */
852static void bond_vlan_rx_register(struct net_device *bond_dev, struct vlan_group *grp)
853{
854 struct bonding *bond = bond_dev->priv;
855 struct slave *slave;
856 int i;
857
858 bond->vlgrp = grp;
859
860 bond_for_each_slave(bond, slave, i) {
861 struct net_device *slave_dev = slave->dev;
862
863 if ((slave_dev->features & NETIF_F_HW_VLAN_RX) &&
864 slave_dev->vlan_rx_register) {
865 slave_dev->vlan_rx_register(slave_dev, grp);
866 }
867 }
868}
869
870/**
871 * bond_vlan_rx_add_vid - Propagates adding an id to slaves
872 * @bond_dev: bonding net device that got called
873 * @vid: vlan id being added
874 */
875static void bond_vlan_rx_add_vid(struct net_device *bond_dev, uint16_t vid)
876{
877 struct bonding *bond = bond_dev->priv;
878 struct slave *slave;
879 int i, res;
880
881 bond_for_each_slave(bond, slave, i) {
882 struct net_device *slave_dev = slave->dev;
883
884 if ((slave_dev->features & NETIF_F_HW_VLAN_FILTER) &&
885 slave_dev->vlan_rx_add_vid) {
886 slave_dev->vlan_rx_add_vid(slave_dev, vid);
887 }
888 }
889
890 res = bond_add_vlan(bond, vid);
891 if (res) {
892 printk(KERN_ERR DRV_NAME
893 ": %s: Failed to add vlan id %d\n",
894 bond_dev->name, vid);
895 }
896}
897
898/**
899 * bond_vlan_rx_kill_vid - Propagates deleting an id to slaves
900 * @bond_dev: bonding net device that got called
901 * @vid: vlan id being removed
902 */
903static void bond_vlan_rx_kill_vid(struct net_device *bond_dev, uint16_t vid)
904{
905 struct bonding *bond = bond_dev->priv;
906 struct slave *slave;
907 struct net_device *vlan_dev;
908 int i, res;
909
910 bond_for_each_slave(bond, slave, i) {
911 struct net_device *slave_dev = slave->dev;
912
913 if ((slave_dev->features & NETIF_F_HW_VLAN_FILTER) &&
914 slave_dev->vlan_rx_kill_vid) {
915 /* Save and then restore vlan_dev in the grp array,
916 * since the slave's driver might clear it.
917 */
918 vlan_dev = bond->vlgrp->vlan_devices[vid];
919 slave_dev->vlan_rx_kill_vid(slave_dev, vid);
920 bond->vlgrp->vlan_devices[vid] = vlan_dev;
921 }
922 }
923
924 res = bond_del_vlan(bond, vid);
925 if (res) {
926 printk(KERN_ERR DRV_NAME
927 ": %s: Failed to remove vlan id %d\n",
928 bond_dev->name, vid);
929 }
930}
931
932static void bond_add_vlans_on_slave(struct bonding *bond, struct net_device *slave_dev)
933{
934 struct vlan_entry *vlan;
935
936 write_lock_bh(&bond->lock);
937
938 if (list_empty(&bond->vlan_list)) {
939 goto out;
940 }
941
942 if ((slave_dev->features & NETIF_F_HW_VLAN_RX) &&
943 slave_dev->vlan_rx_register) {
944 slave_dev->vlan_rx_register(slave_dev, bond->vlgrp);
945 }
946
947 if (!(slave_dev->features & NETIF_F_HW_VLAN_FILTER) ||
948 !(slave_dev->vlan_rx_add_vid)) {
949 goto out;
950 }
951
952 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
953 slave_dev->vlan_rx_add_vid(slave_dev, vlan->vlan_id);
954 }
955
956out:
957 write_unlock_bh(&bond->lock);
958}
959
960static void bond_del_vlans_from_slave(struct bonding *bond, struct net_device *slave_dev)
961{
962 struct vlan_entry *vlan;
963 struct net_device *vlan_dev;
964
965 write_lock_bh(&bond->lock);
966
967 if (list_empty(&bond->vlan_list)) {
968 goto out;
969 }
970
971 if (!(slave_dev->features & NETIF_F_HW_VLAN_FILTER) ||
972 !(slave_dev->vlan_rx_kill_vid)) {
973 goto unreg;
974 }
975
976 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
977 /* Save and then restore vlan_dev in the grp array,
978 * since the slave's driver might clear it.
979 */
980 vlan_dev = bond->vlgrp->vlan_devices[vlan->vlan_id];
981 slave_dev->vlan_rx_kill_vid(slave_dev, vlan->vlan_id);
982 bond->vlgrp->vlan_devices[vlan->vlan_id] = vlan_dev;
983 }
984
985unreg:
986 if ((slave_dev->features & NETIF_F_HW_VLAN_RX) &&
987 slave_dev->vlan_rx_register) {
988 slave_dev->vlan_rx_register(slave_dev, NULL);
989 }
990
991out:
992 write_unlock_bh(&bond->lock);
993}
994
995/*------------------------------- Link status -------------------------------*/
996
997/*
998 * Get link speed and duplex from the slave's base driver
999 * using ethtool. If for some reason the call fails or the
1000 * values are invalid, fake speed and duplex to 100/Full
1001 * and return error.
1002 */
1003static int bond_update_speed_duplex(struct slave *slave)
1004{
1005 struct net_device *slave_dev = slave->dev;
1006 static int (* ioctl)(struct net_device *, struct ifreq *, int);
1007 struct ifreq ifr;
1008 struct ethtool_cmd etool;
1009
1010 /* Fake speed and duplex */
1011 slave->speed = SPEED_100;
1012 slave->duplex = DUPLEX_FULL;
1013
1014 if (slave_dev->ethtool_ops) {
1015 u32 res;
1016
1017 if (!slave_dev->ethtool_ops->get_settings) {
1018 return -1;
1019 }
1020
1021 res = slave_dev->ethtool_ops->get_settings(slave_dev, &etool);
1022 if (res < 0) {
1023 return -1;
1024 }
1025
1026 goto verify;
1027 }
1028
1029 ioctl = slave_dev->do_ioctl;
1030 strncpy(ifr.ifr_name, slave_dev->name, IFNAMSIZ);
1031 etool.cmd = ETHTOOL_GSET;
1032 ifr.ifr_data = (char*)&etool;
1033 if (!ioctl || (IOCTL(slave_dev, &ifr, SIOCETHTOOL) < 0)) {
1034 return -1;
1035 }
1036
1037verify:
1038 switch (etool.speed) {
1039 case SPEED_10:
1040 case SPEED_100:
1041 case SPEED_1000:
1042 break;
1043 default:
1044 return -1;
1045 }
1046
1047 switch (etool.duplex) {
1048 case DUPLEX_FULL:
1049 case DUPLEX_HALF:
1050 break;
1051 default:
1052 return -1;
1053 }
1054
1055 slave->speed = etool.speed;
1056 slave->duplex = etool.duplex;
1057
1058 return 0;
1059}
1060
1061/*
1062 * if <dev> supports MII link status reporting, check its link status.
1063 *
1064 * We either do MII/ETHTOOL ioctls, or check netif_carrier_ok(),
1065 * depening upon the setting of the use_carrier parameter.
1066 *
1067 * Return either BMSR_LSTATUS, meaning that the link is up (or we
1068 * can't tell and just pretend it is), or 0, meaning that the link is
1069 * down.
1070 *
1071 * If reporting is non-zero, instead of faking link up, return -1 if
1072 * both ETHTOOL and MII ioctls fail (meaning the device does not
1073 * support them). If use_carrier is set, return whatever it says.
1074 * It'd be nice if there was a good way to tell if a driver supports
1075 * netif_carrier, but there really isn't.
1076 */
1077static int bond_check_dev_link(struct bonding *bond, struct net_device *slave_dev, int reporting)
1078{
1079 static int (* ioctl)(struct net_device *, struct ifreq *, int);
1080 struct ifreq ifr;
1081 struct mii_ioctl_data *mii;
1082 struct ethtool_value etool;
1083
1084 if (bond->params.use_carrier) {
1085 return netif_carrier_ok(slave_dev) ? BMSR_LSTATUS : 0;
1086 }
1087
1088 ioctl = slave_dev->do_ioctl;
1089 if (ioctl) {
1090 /* TODO: set pointer to correct ioctl on a per team member */
1091 /* bases to make this more efficient. that is, once */
1092 /* we determine the correct ioctl, we will always */
1093 /* call it and not the others for that team */
1094 /* member. */
1095
1096 /*
1097 * We cannot assume that SIOCGMIIPHY will also read a
1098 * register; not all network drivers (e.g., e100)
1099 * support that.
1100 */
1101
1102 /* Yes, the mii is overlaid on the ifreq.ifr_ifru */
1103 strncpy(ifr.ifr_name, slave_dev->name, IFNAMSIZ);
1104 mii = if_mii(&ifr);
1105 if (IOCTL(slave_dev, &ifr, SIOCGMIIPHY) == 0) {
1106 mii->reg_num = MII_BMSR;
1107 if (IOCTL(slave_dev, &ifr, SIOCGMIIREG) == 0) {
1108 return (mii->val_out & BMSR_LSTATUS);
1109 }
1110 }
1111 }
1112
1113 /* try SIOCETHTOOL ioctl, some drivers cache ETHTOOL_GLINK */
1114 /* for a period of time so we attempt to get link status */
1115 /* from it last if the above MII ioctls fail... */
1116 if (slave_dev->ethtool_ops) {
1117 if (slave_dev->ethtool_ops->get_link) {
1118 u32 link;
1119
1120 link = slave_dev->ethtool_ops->get_link(slave_dev);
1121
1122 return link ? BMSR_LSTATUS : 0;
1123 }
1124 }
1125
1126 if (ioctl) {
1127 strncpy(ifr.ifr_name, slave_dev->name, IFNAMSIZ);
1128 etool.cmd = ETHTOOL_GLINK;
1129 ifr.ifr_data = (char*)&etool;
1130 if (IOCTL(slave_dev, &ifr, SIOCETHTOOL) == 0) {
1131 if (etool.data == 1) {
1132 return BMSR_LSTATUS;
1133 } else {
1134 dprintk("SIOCETHTOOL shows link down\n");
1135 return 0;
1136 }
1137 }
1138 }
1139
1140 /*
1141 * If reporting, report that either there's no dev->do_ioctl,
1142 * or both SIOCGMIIREG and SIOCETHTOOL failed (meaning that we
1143 * cannot report link status). If not reporting, pretend
1144 * we're ok.
1145 */
1146 return (reporting ? -1 : BMSR_LSTATUS);
1147}
1148
1149/*----------------------------- Multicast list ------------------------------*/
1150
1151/*
1152 * Returns 0 if dmi1 and dmi2 are the same, non-0 otherwise
1153 */
1154static inline int bond_is_dmi_same(struct dev_mc_list *dmi1, struct dev_mc_list *dmi2)
1155{
1156 return memcmp(dmi1->dmi_addr, dmi2->dmi_addr, dmi1->dmi_addrlen) == 0 &&
1157 dmi1->dmi_addrlen == dmi2->dmi_addrlen;
1158}
1159
1160/*
1161 * returns dmi entry if found, NULL otherwise
1162 */
1163static struct dev_mc_list *bond_mc_list_find_dmi(struct dev_mc_list *dmi, struct dev_mc_list *mc_list)
1164{
1165 struct dev_mc_list *idmi;
1166
1167 for (idmi = mc_list; idmi; idmi = idmi->next) {
1168 if (bond_is_dmi_same(dmi, idmi)) {
1169 return idmi;
1170 }
1171 }
1172
1173 return NULL;
1174}
1175
1176/*
1177 * Push the promiscuity flag down to appropriate slaves
1178 */
1179static void bond_set_promiscuity(struct bonding *bond, int inc)
1180{
1181 if (USES_PRIMARY(bond->params.mode)) {
1182 /* write lock already acquired */
1183 if (bond->curr_active_slave) {
1184 dev_set_promiscuity(bond->curr_active_slave->dev, inc);
1185 }
1186 } else {
1187 struct slave *slave;
1188 int i;
1189 bond_for_each_slave(bond, slave, i) {
1190 dev_set_promiscuity(slave->dev, inc);
1191 }
1192 }
1193}
1194
1195/*
1196 * Push the allmulti flag down to all slaves
1197 */
1198static void bond_set_allmulti(struct bonding *bond, int inc)
1199{
1200 if (USES_PRIMARY(bond->params.mode)) {
1201 /* write lock already acquired */
1202 if (bond->curr_active_slave) {
1203 dev_set_allmulti(bond->curr_active_slave->dev, inc);
1204 }
1205 } else {
1206 struct slave *slave;
1207 int i;
1208 bond_for_each_slave(bond, slave, i) {
1209 dev_set_allmulti(slave->dev, inc);
1210 }
1211 }
1212}
1213
1214/*
1215 * Add a Multicast address to slaves
1216 * according to mode
1217 */
1218static void bond_mc_add(struct bonding *bond, void *addr, int alen)
1219{
1220 if (USES_PRIMARY(bond->params.mode)) {
1221 /* write lock already acquired */
1222 if (bond->curr_active_slave) {
1223 dev_mc_add(bond->curr_active_slave->dev, addr, alen, 0);
1224 }
1225 } else {
1226 struct slave *slave;
1227 int i;
1228 bond_for_each_slave(bond, slave, i) {
1229 dev_mc_add(slave->dev, addr, alen, 0);
1230 }
1231 }
1232}
1233
1234/*
1235 * Remove a multicast address from slave
1236 * according to mode
1237 */
1238static void bond_mc_delete(struct bonding *bond, void *addr, int alen)
1239{
1240 if (USES_PRIMARY(bond->params.mode)) {
1241 /* write lock already acquired */
1242 if (bond->curr_active_slave) {
1243 dev_mc_delete(bond->curr_active_slave->dev, addr, alen, 0);
1244 }
1245 } else {
1246 struct slave *slave;
1247 int i;
1248 bond_for_each_slave(bond, slave, i) {
1249 dev_mc_delete(slave->dev, addr, alen, 0);
1250 }
1251 }
1252}
1253
1254/*
1255 * Totally destroys the mc_list in bond
1256 */
1257static void bond_mc_list_destroy(struct bonding *bond)
1258{
1259 struct dev_mc_list *dmi;
1260
1261 dmi = bond->mc_list;
1262 while (dmi) {
1263 bond->mc_list = dmi->next;
1264 kfree(dmi);
1265 dmi = bond->mc_list;
1266 }
1267}
1268
1269/*
1270 * Copy all the Multicast addresses from src to the bonding device dst
1271 */
1272static int bond_mc_list_copy(struct dev_mc_list *mc_list, struct bonding *bond, int gpf_flag)
1273{
1274 struct dev_mc_list *dmi, *new_dmi;
1275
1276 for (dmi = mc_list; dmi; dmi = dmi->next) {
1277 new_dmi = kmalloc(sizeof(struct dev_mc_list), gpf_flag);
1278
1279 if (!new_dmi) {
1280 /* FIXME: Potential memory leak !!! */
1281 return -ENOMEM;
1282 }
1283
1284 new_dmi->next = bond->mc_list;
1285 bond->mc_list = new_dmi;
1286 new_dmi->dmi_addrlen = dmi->dmi_addrlen;
1287 memcpy(new_dmi->dmi_addr, dmi->dmi_addr, dmi->dmi_addrlen);
1288 new_dmi->dmi_users = dmi->dmi_users;
1289 new_dmi->dmi_gusers = dmi->dmi_gusers;
1290 }
1291
1292 return 0;
1293}
1294
1295/*
1296 * flush all members of flush->mc_list from device dev->mc_list
1297 */
1298static void bond_mc_list_flush(struct net_device *bond_dev, struct net_device *slave_dev)
1299{
1300 struct bonding *bond = bond_dev->priv;
1301 struct dev_mc_list *dmi;
1302
1303 for (dmi = bond_dev->mc_list; dmi; dmi = dmi->next) {
1304 dev_mc_delete(slave_dev, dmi->dmi_addr, dmi->dmi_addrlen, 0);
1305 }
1306
1307 if (bond->params.mode == BOND_MODE_8023AD) {
1308 /* del lacpdu mc addr from mc list */
1309 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
1310
1311 dev_mc_delete(slave_dev, lacpdu_multicast, ETH_ALEN, 0);
1312 }
1313}
1314
1315/*--------------------------- Active slave change ---------------------------*/
1316
1317/*
1318 * Update the mc list and multicast-related flags for the new and
1319 * old active slaves (if any) according to the multicast mode, and
1320 * promiscuous flags unconditionally.
1321 */
1322static void bond_mc_swap(struct bonding *bond, struct slave *new_active, struct slave *old_active)
1323{
1324 struct dev_mc_list *dmi;
1325
1326 if (!USES_PRIMARY(bond->params.mode)) {
1327 /* nothing to do - mc list is already up-to-date on
1328 * all slaves
1329 */
1330 return;
1331 }
1332
1333 if (old_active) {
1334 if (bond->dev->flags & IFF_PROMISC) {
1335 dev_set_promiscuity(old_active->dev, -1);
1336 }
1337
1338 if (bond->dev->flags & IFF_ALLMULTI) {
1339 dev_set_allmulti(old_active->dev, -1);
1340 }
1341
1342 for (dmi = bond->dev->mc_list; dmi; dmi = dmi->next) {
1343 dev_mc_delete(old_active->dev, dmi->dmi_addr, dmi->dmi_addrlen, 0);
1344 }
1345 }
1346
1347 if (new_active) {
1348 if (bond->dev->flags & IFF_PROMISC) {
1349 dev_set_promiscuity(new_active->dev, 1);
1350 }
1351
1352 if (bond->dev->flags & IFF_ALLMULTI) {
1353 dev_set_allmulti(new_active->dev, 1);
1354 }
1355
1356 for (dmi = bond->dev->mc_list; dmi; dmi = dmi->next) {
1357 dev_mc_add(new_active->dev, dmi->dmi_addr, dmi->dmi_addrlen, 0);
1358 }
1359 }
1360}
1361
1362/**
1363 * find_best_interface - select the best available slave to be the active one
1364 * @bond: our bonding struct
1365 *
1366 * Warning: Caller must hold curr_slave_lock for writing.
1367 */
1368static struct slave *bond_find_best_slave(struct bonding *bond)
1369{
1370 struct slave *new_active, *old_active;
1371 struct slave *bestslave = NULL;
1372 int mintime = bond->params.updelay;
1373 int i;
1374
1375 new_active = old_active = bond->curr_active_slave;
1376
1377 if (!new_active) { /* there were no active slaves left */
1378 if (bond->slave_cnt > 0) { /* found one slave */
1379 new_active = bond->first_slave;
1380 } else {
1381 return NULL; /* still no slave, return NULL */
1382 }
1383 }
1384
1385 /* first try the primary link; if arping, a link must tx/rx traffic
1386 * before it can be considered the curr_active_slave - also, we would skip
1387 * slaves between the curr_active_slave and primary_slave that may be up
1388 * and able to arp
1389 */
1390 if ((bond->primary_slave) &&
1391 (!bond->params.arp_interval) &&
1392 (IS_UP(bond->primary_slave->dev))) {
1393 new_active = bond->primary_slave;
1394 }
1395
1396 /* remember where to stop iterating over the slaves */
1397 old_active = new_active;
1398
1399 bond_for_each_slave_from(bond, new_active, i, old_active) {
1400 if (IS_UP(new_active->dev)) {
1401 if (new_active->link == BOND_LINK_UP) {
1402 return new_active;
1403 } else if (new_active->link == BOND_LINK_BACK) {
1404 /* link up, but waiting for stabilization */
1405 if (new_active->delay < mintime) {
1406 mintime = new_active->delay;
1407 bestslave = new_active;
1408 }
1409 }
1410 }
1411 }
1412
1413 return bestslave;
1414}
1415
1416/**
1417 * change_active_interface - change the active slave into the specified one
1418 * @bond: our bonding struct
1419 * @new: the new slave to make the active one
1420 *
1421 * Set the new slave to the bond's settings and unset them on the old
1422 * curr_active_slave.
1423 * Setting include flags, mc-list, promiscuity, allmulti, etc.
1424 *
1425 * If @new's link state is %BOND_LINK_BACK we'll set it to %BOND_LINK_UP,
1426 * because it is apparently the best available slave we have, even though its
1427 * updelay hasn't timed out yet.
1428 *
1429 * Warning: Caller must hold curr_slave_lock for writing.
1430 */
1431static void bond_change_active_slave(struct bonding *bond, struct slave *new_active)
1432{
1433 struct slave *old_active = bond->curr_active_slave;
1434
1435 if (old_active == new_active) {
1436 return;
1437 }
1438
1439 if (new_active) {
1440 if (new_active->link == BOND_LINK_BACK) {
1441 if (USES_PRIMARY(bond->params.mode)) {
1442 printk(KERN_INFO DRV_NAME
1443 ": %s: making interface %s the new "
1444 "active one %d ms earlier.\n",
1445 bond->dev->name, new_active->dev->name,
1446 (bond->params.updelay - new_active->delay) * bond->params.miimon);
1447 }
1448
1449 new_active->delay = 0;
1450 new_active->link = BOND_LINK_UP;
1451 new_active->jiffies = jiffies;
1452
1453 if (bond->params.mode == BOND_MODE_8023AD) {
1454 bond_3ad_handle_link_change(new_active, BOND_LINK_UP);
1455 }
1456
1457 if ((bond->params.mode == BOND_MODE_TLB) ||
1458 (bond->params.mode == BOND_MODE_ALB)) {
1459 bond_alb_handle_link_change(bond, new_active, BOND_LINK_UP);
1460 }
1461 } else {
1462 if (USES_PRIMARY(bond->params.mode)) {
1463 printk(KERN_INFO DRV_NAME
1464 ": %s: making interface %s the new "
1465 "active one.\n",
1466 bond->dev->name, new_active->dev->name);
1467 }
1468 }
1469 }
1470
1471 if (bond->params.mode == BOND_MODE_ACTIVEBACKUP) {
1472 if (old_active) {
1473 bond_set_slave_inactive_flags(old_active);
1474 }
1475
1476 if (new_active) {
1477 bond_set_slave_active_flags(new_active);
1478 }
1479 }
1480
1481 if (USES_PRIMARY(bond->params.mode)) {
1482 bond_mc_swap(bond, new_active, old_active);
1483 }
1484
1485 if ((bond->params.mode == BOND_MODE_TLB) ||
1486 (bond->params.mode == BOND_MODE_ALB)) {
1487 bond_alb_handle_active_change(bond, new_active);
1488 } else {
1489 bond->curr_active_slave = new_active;
1490 }
1491}
1492
1493/**
1494 * bond_select_active_slave - select a new active slave, if needed
1495 * @bond: our bonding struct
1496 *
1497 * This functions shoud be called when one of the following occurs:
1498 * - The old curr_active_slave has been released or lost its link.
1499 * - The primary_slave has got its link back.
1500 * - A slave has got its link back and there's no old curr_active_slave.
1501 *
1502 * Warning: Caller must hold curr_slave_lock for writing.
1503 */
1504static void bond_select_active_slave(struct bonding *bond)
1505{
1506 struct slave *best_slave;
1507
1508 best_slave = bond_find_best_slave(bond);
1509 if (best_slave != bond->curr_active_slave) {
1510 bond_change_active_slave(bond, best_slave);
1511 }
1512}
1513
1514/*--------------------------- slave list handling ---------------------------*/
1515
1516/*
1517 * This function attaches the slave to the end of list.
1518 *
1519 * bond->lock held for writing by caller.
1520 */
1521static void bond_attach_slave(struct bonding *bond, struct slave *new_slave)
1522{
1523 if (bond->first_slave == NULL) { /* attaching the first slave */
1524 new_slave->next = new_slave;
1525 new_slave->prev = new_slave;
1526 bond->first_slave = new_slave;
1527 } else {
1528 new_slave->next = bond->first_slave;
1529 new_slave->prev = bond->first_slave->prev;
1530 new_slave->next->prev = new_slave;
1531 new_slave->prev->next = new_slave;
1532 }
1533
1534 bond->slave_cnt++;
1535}
1536
1537/*
1538 * This function detaches the slave from the list.
1539 * WARNING: no check is made to verify if the slave effectively
1540 * belongs to <bond>.
1541 * Nothing is freed on return, structures are just unchained.
1542 * If any slave pointer in bond was pointing to <slave>,
1543 * it should be changed by the calling function.
1544 *
1545 * bond->lock held for writing by caller.
1546 */
1547static void bond_detach_slave(struct bonding *bond, struct slave *slave)
1548{
1549 if (slave->next) {
1550 slave->next->prev = slave->prev;
1551 }
1552
1553 if (slave->prev) {
1554 slave->prev->next = slave->next;
1555 }
1556
1557 if (bond->first_slave == slave) { /* slave is the first slave */
1558 if (bond->slave_cnt > 1) { /* there are more slave */
1559 bond->first_slave = slave->next;
1560 } else {
1561 bond->first_slave = NULL; /* slave was the last one */
1562 }
1563 }
1564
1565 slave->next = NULL;
1566 slave->prev = NULL;
1567 bond->slave_cnt--;
1568}
1569
1570/*---------------------------------- IOCTL ----------------------------------*/
1571
1572static int bond_sethwaddr(struct net_device *bond_dev, struct net_device *slave_dev)
1573{
1574 dprintk("bond_dev=%p\n", bond_dev);
1575 dprintk("slave_dev=%p\n", slave_dev);
1576 dprintk("slave_dev->addr_len=%d\n", slave_dev->addr_len);
1577 memcpy(bond_dev->dev_addr, slave_dev->dev_addr, slave_dev->addr_len);
1578 return 0;
1579}
1580
1581/* enslave device <slave> to bond device <master> */
1582static int bond_enslave(struct net_device *bond_dev, struct net_device *slave_dev)
1583{
1584 struct bonding *bond = bond_dev->priv;
1585 struct slave *new_slave = NULL;
1586 struct dev_mc_list *dmi;
1587 struct sockaddr addr;
1588 int link_reporting;
1589 int old_features = bond_dev->features;
1590 int res = 0;
1591
1592 if (slave_dev->do_ioctl == NULL) {
1593 printk(KERN_WARNING DRV_NAME
1594 ": Warning : no link monitoring support for %s\n",
1595 slave_dev->name);
1596 }
1597
1598 /* bond must be initialized by bond_open() before enslaving */
1599 if (!(bond_dev->flags & IFF_UP)) {
1600 dprintk("Error, master_dev is not up\n");
1601 return -EPERM;
1602 }
1603
1604 /* already enslaved */
1605 if (slave_dev->flags & IFF_SLAVE) {
1606 dprintk("Error, Device was already enslaved\n");
1607 return -EBUSY;
1608 }
1609
1610 /* vlan challenged mutual exclusion */
1611 /* no need to lock since we're protected by rtnl_lock */
1612 if (slave_dev->features & NETIF_F_VLAN_CHALLENGED) {
1613 dprintk("%s: NETIF_F_VLAN_CHALLENGED\n", slave_dev->name);
1614 if (!list_empty(&bond->vlan_list)) {
1615 printk(KERN_ERR DRV_NAME
1616 ": Error: cannot enslave VLAN "
1617 "challenged slave %s on VLAN enabled "
1618 "bond %s\n", slave_dev->name,
1619 bond_dev->name);
1620 return -EPERM;
1621 } else {
1622 printk(KERN_WARNING DRV_NAME
1623 ": Warning: enslaved VLAN challenged "
1624 "slave %s. Adding VLANs will be blocked as "
1625 "long as %s is part of bond %s\n",
1626 slave_dev->name, slave_dev->name,
1627 bond_dev->name);
1628 bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
1629 }
1630 } else {
1631 dprintk("%s: ! NETIF_F_VLAN_CHALLENGED\n", slave_dev->name);
1632 if (bond->slave_cnt == 0) {
1633 /* First slave, and it is not VLAN challenged,
1634 * so remove the block of adding VLANs over the bond.
1635 */
1636 bond_dev->features &= ~NETIF_F_VLAN_CHALLENGED;
1637 }
1638 }
1639
1640 if (app_abi_ver >= 1) {
1641 /* The application is using an ABI, which requires the
1642 * slave interface to be closed.
1643 */
1644 if ((slave_dev->flags & IFF_UP)) {
1645 printk(KERN_ERR DRV_NAME
1646 ": Error: %s is up\n",
1647 slave_dev->name);
1648 res = -EPERM;
1649 goto err_undo_flags;
1650 }
1651
1652 if (slave_dev->set_mac_address == NULL) {
1653 printk(KERN_ERR DRV_NAME
1654 ": Error: The slave device you specified does "
1655 "not support setting the MAC address.\n");
1656 printk(KERN_ERR
1657 "Your kernel likely does not support slave "
1658 "devices.\n");
1659
1660 res = -EOPNOTSUPP;
1661 goto err_undo_flags;
1662 }
1663 } else {
1664 /* The application is not using an ABI, which requires the
1665 * slave interface to be open.
1666 */
1667 if (!(slave_dev->flags & IFF_UP)) {
1668 printk(KERN_ERR DRV_NAME
1669 ": Error: %s is not running\n",
1670 slave_dev->name);
1671 res = -EINVAL;
1672 goto err_undo_flags;
1673 }
1674
1675 if ((bond->params.mode == BOND_MODE_8023AD) ||
1676 (bond->params.mode == BOND_MODE_TLB) ||
1677 (bond->params.mode == BOND_MODE_ALB)) {
1678 printk(KERN_ERR DRV_NAME
1679 ": Error: to use %s mode, you must upgrade "
1680 "ifenslave.\n",
1681 bond_mode_name(bond->params.mode));
1682 res = -EOPNOTSUPP;
1683 goto err_undo_flags;
1684 }
1685 }
1686
1687 new_slave = kmalloc(sizeof(struct slave), GFP_KERNEL);
1688 if (!new_slave) {
1689 res = -ENOMEM;
1690 goto err_undo_flags;
1691 }
1692
1693 memset(new_slave, 0, sizeof(struct slave));
1694
1695 /* save slave's original flags before calling
1696 * netdev_set_master and dev_open
1697 */
1698 new_slave->original_flags = slave_dev->flags;
1699
1700 if (app_abi_ver >= 1) {
1701 /* save slave's original ("permanent") mac address for
1702 * modes that needs it, and for restoring it upon release,
1703 * and then set it to the master's address
1704 */
1705 memcpy(new_slave->perm_hwaddr, slave_dev->dev_addr, ETH_ALEN);
1706
1707 /* set slave to master's mac address
1708 * The application already set the master's
1709 * mac address to that of the first slave
1710 */
1711 memcpy(addr.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
1712 addr.sa_family = slave_dev->type;
1713 res = dev_set_mac_address(slave_dev, &addr);
1714 if (res) {
1715 dprintk("Error %d calling set_mac_address\n", res);
1716 goto err_free;
1717 }
1718
1719 /* open the slave since the application closed it */
1720 res = dev_open(slave_dev);
1721 if (res) {
1722 dprintk("Openning slave %s failed\n", slave_dev->name);
1723 goto err_restore_mac;
1724 }
1725 }
1726
1727 res = netdev_set_master(slave_dev, bond_dev);
1728 if (res) {
1729 dprintk("Error %d calling netdev_set_master\n", res);
1730 if (app_abi_ver < 1) {
1731 goto err_free;
1732 } else {
1733 goto err_close;
1734 }
1735 }
1736
1737 new_slave->dev = slave_dev;
1738
1739 if ((bond->params.mode == BOND_MODE_TLB) ||
1740 (bond->params.mode == BOND_MODE_ALB)) {
1741 /* bond_alb_init_slave() must be called before all other stages since
1742 * it might fail and we do not want to have to undo everything
1743 */
1744 res = bond_alb_init_slave(bond, new_slave);
1745 if (res) {
1746 goto err_unset_master;
1747 }
1748 }
1749
1750 /* If the mode USES_PRIMARY, then the new slave gets the
1751 * master's promisc (and mc) settings only if it becomes the
1752 * curr_active_slave, and that is taken care of later when calling
1753 * bond_change_active()
1754 */
1755 if (!USES_PRIMARY(bond->params.mode)) {
1756 /* set promiscuity level to new slave */
1757 if (bond_dev->flags & IFF_PROMISC) {
1758 dev_set_promiscuity(slave_dev, 1);
1759 }
1760
1761 /* set allmulti level to new slave */
1762 if (bond_dev->flags & IFF_ALLMULTI) {
1763 dev_set_allmulti(slave_dev, 1);
1764 }
1765
1766 /* upload master's mc_list to new slave */
1767 for (dmi = bond_dev->mc_list; dmi; dmi = dmi->next) {
1768 dev_mc_add (slave_dev, dmi->dmi_addr, dmi->dmi_addrlen, 0);
1769 }
1770 }
1771
1772 if (bond->params.mode == BOND_MODE_8023AD) {
1773 /* add lacpdu mc addr to mc list */
1774 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
1775
1776 dev_mc_add(slave_dev, lacpdu_multicast, ETH_ALEN, 0);
1777 }
1778
1779 bond_add_vlans_on_slave(bond, slave_dev);
1780
1781 write_lock_bh(&bond->lock);
1782
1783 bond_attach_slave(bond, new_slave);
1784
1785 new_slave->delay = 0;
1786 new_slave->link_failure_count = 0;
1787
1788 if (bond->params.miimon && !bond->params.use_carrier) {
1789 link_reporting = bond_check_dev_link(bond, slave_dev, 1);
1790
1791 if ((link_reporting == -1) && !bond->params.arp_interval) {
1792 /*
1793 * miimon is set but a bonded network driver
1794 * does not support ETHTOOL/MII and
1795 * arp_interval is not set. Note: if
1796 * use_carrier is enabled, we will never go
1797 * here (because netif_carrier is always
1798 * supported); thus, we don't need to change
1799 * the messages for netif_carrier.
1800 */
1801 printk(KERN_WARNING DRV_NAME
1802 ": Warning: MII and ETHTOOL support not "
1803 "available for interface %s, and "
1804 "arp_interval/arp_ip_target module parameters "
1805 "not specified, thus bonding will not detect "
1806 "link failures! see bonding.txt for details.\n",
1807 slave_dev->name);
1808 } else if (link_reporting == -1) {
1809 /* unable get link status using mii/ethtool */
1810 printk(KERN_WARNING DRV_NAME
1811 ": Warning: can't get link status from "
1812 "interface %s; the network driver associated "
1813 "with this interface does not support MII or "
1814 "ETHTOOL link status reporting, thus miimon "
1815 "has no effect on this interface.\n",
1816 slave_dev->name);
1817 }
1818 }
1819
1820 /* check for initial state */
1821 if (!bond->params.miimon ||
1822 (bond_check_dev_link(bond, slave_dev, 0) == BMSR_LSTATUS)) {
1823 if (bond->params.updelay) {
1824 dprintk("Initial state of slave_dev is "
1825 "BOND_LINK_BACK\n");
1826 new_slave->link = BOND_LINK_BACK;
1827 new_slave->delay = bond->params.updelay;
1828 } else {
1829 dprintk("Initial state of slave_dev is "
1830 "BOND_LINK_UP\n");
1831 new_slave->link = BOND_LINK_UP;
1832 }
1833 new_slave->jiffies = jiffies;
1834 } else {
1835 dprintk("Initial state of slave_dev is "
1836 "BOND_LINK_DOWN\n");
1837 new_slave->link = BOND_LINK_DOWN;
1838 }
1839
1840 if (bond_update_speed_duplex(new_slave) &&
1841 (new_slave->link != BOND_LINK_DOWN)) {
1842 printk(KERN_WARNING DRV_NAME
1843 ": Warning: failed to get speed and duplex from %s, "
1844 "assumed to be 100Mb/sec and Full.\n",
1845 new_slave->dev->name);
1846
1847 if (bond->params.mode == BOND_MODE_8023AD) {
1848 printk(KERN_WARNING
1849 "Operation of 802.3ad mode requires ETHTOOL "
1850 "support in base driver for proper aggregator "
1851 "selection.\n");
1852 }
1853 }
1854
1855 if (USES_PRIMARY(bond->params.mode) && bond->params.primary[0]) {
1856 /* if there is a primary slave, remember it */
1857 if (strcmp(bond->params.primary, new_slave->dev->name) == 0) {
1858 bond->primary_slave = new_slave;
1859 }
1860 }
1861
1862 switch (bond->params.mode) {
1863 case BOND_MODE_ACTIVEBACKUP:
1864 /* if we're in active-backup mode, we need one and only one active
1865 * interface. The backup interfaces will have their NOARP flag set
1866 * because we need them to be completely deaf and not to respond to
1867 * any ARP request on the network to avoid fooling a switch. Thus,
1868 * since we guarantee that curr_active_slave always point to the last
1869 * usable interface, we just have to verify this interface's flag.
1870 */
1871 if (((!bond->curr_active_slave) ||
1872 (bond->curr_active_slave->dev->flags & IFF_NOARP)) &&
1873 (new_slave->link != BOND_LINK_DOWN)) {
1874 dprintk("This is the first active slave\n");
1875 /* first slave or no active slave yet, and this link
1876 is OK, so make this interface the active one */
1877 bond_change_active_slave(bond, new_slave);
1878 } else {
1879 dprintk("This is just a backup slave\n");
1880 bond_set_slave_inactive_flags(new_slave);
1881 }
1882 break;
1883 case BOND_MODE_8023AD:
1884 /* in 802.3ad mode, the internal mechanism
1885 * will activate the slaves in the selected
1886 * aggregator
1887 */
1888 bond_set_slave_inactive_flags(new_slave);
1889 /* if this is the first slave */
1890 if (bond->slave_cnt == 1) {
1891 SLAVE_AD_INFO(new_slave).id = 1;
1892 /* Initialize AD with the number of times that the AD timer is called in 1 second
1893 * can be called only after the mac address of the bond is set
1894 */
1895 bond_3ad_initialize(bond, 1000/AD_TIMER_INTERVAL,
1896 bond->params.lacp_fast);
1897 } else {
1898 SLAVE_AD_INFO(new_slave).id =
1899 SLAVE_AD_INFO(new_slave->prev).id + 1;
1900 }
1901
1902 bond_3ad_bind_slave(new_slave);
1903 break;
1904 case BOND_MODE_TLB:
1905 case BOND_MODE_ALB:
1906 new_slave->state = BOND_STATE_ACTIVE;
1907 if ((!bond->curr_active_slave) &&
1908 (new_slave->link != BOND_LINK_DOWN)) {
1909 /* first slave or no active slave yet, and this link
1910 * is OK, so make this interface the active one
1911 */
1912 bond_change_active_slave(bond, new_slave);
1913 }
1914 break;
1915 default:
1916 dprintk("This slave is always active in trunk mode\n");
1917
1918 /* always active in trunk mode */
1919 new_slave->state = BOND_STATE_ACTIVE;
1920
1921 /* In trunking mode there is little meaning to curr_active_slave
1922 * anyway (it holds no special properties of the bond device),
1923 * so we can change it without calling change_active_interface()
1924 */
1925 if (!bond->curr_active_slave) {
1926 bond->curr_active_slave = new_slave;
1927 }
1928 break;
1929 } /* switch(bond_mode) */
1930
1931 write_unlock_bh(&bond->lock);
1932
1933 if (app_abi_ver < 1) {
1934 /*
1935 * !!! This is to support old versions of ifenslave.
1936 * We can remove this in 2.5 because our ifenslave takes
1937 * care of this for us.
1938 * We check to see if the master has a mac address yet.
1939 * If not, we'll give it the mac address of our slave device.
1940 */
1941 int ndx = 0;
1942
1943 for (ndx = 0; ndx < bond_dev->addr_len; ndx++) {
1944 dprintk("Checking ndx=%d of bond_dev->dev_addr\n",
1945 ndx);
1946 if (bond_dev->dev_addr[ndx] != 0) {
1947 dprintk("Found non-zero byte at ndx=%d\n",
1948 ndx);
1949 break;
1950 }
1951 }
1952
1953 if (ndx == bond_dev->addr_len) {
1954 /*
1955 * We got all the way through the address and it was
1956 * all 0's.
1957 */
1958 dprintk("%s doesn't have a MAC address yet. \n",
1959 bond_dev->name);
1960 dprintk("Going to give assign it from %s.\n",
1961 slave_dev->name);
1962 bond_sethwaddr(bond_dev, slave_dev);
1963 }
1964 }
1965
1966 printk(KERN_INFO DRV_NAME
1967 ": %s: enslaving %s as a%s interface with a%s link.\n",
1968 bond_dev->name, slave_dev->name,
1969 new_slave->state == BOND_STATE_ACTIVE ? "n active" : " backup",
1970 new_slave->link != BOND_LINK_DOWN ? "n up" : " down");
1971
1972 /* enslave is successful */
1973 return 0;
1974
1975/* Undo stages on error */
1976err_unset_master:
1977 netdev_set_master(slave_dev, NULL);
1978
1979err_close:
1980 dev_close(slave_dev);
1981
1982err_restore_mac:
1983 memcpy(addr.sa_data, new_slave->perm_hwaddr, ETH_ALEN);
1984 addr.sa_family = slave_dev->type;
1985 dev_set_mac_address(slave_dev, &addr);
1986
1987err_free:
1988 kfree(new_slave);
1989
1990err_undo_flags:
1991 bond_dev->features = old_features;
1992
1993 return res;
1994}
1995
1996/*
1997 * Try to release the slave device <slave> from the bond device <master>
1998 * It is legal to access curr_active_slave without a lock because all the function
1999 * is write-locked.
2000 *
2001 * The rules for slave state should be:
2002 * for Active/Backup:
2003 * Active stays on all backups go down
2004 * for Bonded connections:
2005 * The first up interface should be left on and all others downed.
2006 */
2007static int bond_release(struct net_device *bond_dev, struct net_device *slave_dev)
2008{
2009 struct bonding *bond = bond_dev->priv;
2010 struct slave *slave, *oldcurrent;
2011 struct sockaddr addr;
2012 int mac_addr_differ;
2013
2014 /* slave is not a slave or master is not master of this slave */
2015 if (!(slave_dev->flags & IFF_SLAVE) ||
2016 (slave_dev->master != bond_dev)) {
2017 printk(KERN_ERR DRV_NAME
2018 ": Error: %s: cannot release %s.\n",
2019 bond_dev->name, slave_dev->name);
2020 return -EINVAL;
2021 }
2022
2023 write_lock_bh(&bond->lock);
2024
2025 slave = bond_get_slave_by_dev(bond, slave_dev);
2026 if (!slave) {
2027 /* not a slave of this bond */
2028 printk(KERN_INFO DRV_NAME
2029 ": %s: %s not enslaved\n",
2030 bond_dev->name, slave_dev->name);
2031 return -EINVAL;
2032 }
2033
2034 mac_addr_differ = memcmp(bond_dev->dev_addr,
2035 slave->perm_hwaddr,
2036 ETH_ALEN);
2037 if (!mac_addr_differ && (bond->slave_cnt > 1)) {
2038 printk(KERN_WARNING DRV_NAME
2039 ": Warning: the permanent HWaddr of %s "
2040 "- %02X:%02X:%02X:%02X:%02X:%02X - is "
2041 "still in use by %s. Set the HWaddr of "
2042 "%s to a different address to avoid "
2043 "conflicts.\n",
2044 slave_dev->name,
2045 slave->perm_hwaddr[0],
2046 slave->perm_hwaddr[1],
2047 slave->perm_hwaddr[2],
2048 slave->perm_hwaddr[3],
2049 slave->perm_hwaddr[4],
2050 slave->perm_hwaddr[5],
2051 bond_dev->name,
2052 slave_dev->name);
2053 }
2054
2055 /* Inform AD package of unbinding of slave. */
2056 if (bond->params.mode == BOND_MODE_8023AD) {
2057 /* must be called before the slave is
2058 * detached from the list
2059 */
2060 bond_3ad_unbind_slave(slave);
2061 }
2062
2063 printk(KERN_INFO DRV_NAME
2064 ": %s: releasing %s interface %s\n",
2065 bond_dev->name,
2066 (slave->state == BOND_STATE_ACTIVE)
2067 ? "active" : "backup",
2068 slave_dev->name);
2069
2070 oldcurrent = bond->curr_active_slave;
2071
2072 bond->current_arp_slave = NULL;
2073
2074 /* release the slave from its bond */
2075 bond_detach_slave(bond, slave);
2076
2077 if (bond->primary_slave == slave) {
2078 bond->primary_slave = NULL;
2079 }
2080
2081 if (oldcurrent == slave) {
2082 bond_change_active_slave(bond, NULL);
2083 }
2084
2085 if ((bond->params.mode == BOND_MODE_TLB) ||
2086 (bond->params.mode == BOND_MODE_ALB)) {
2087 /* Must be called only after the slave has been
2088 * detached from the list and the curr_active_slave
2089 * has been cleared (if our_slave == old_current),
2090 * but before a new active slave is selected.
2091 */
2092 bond_alb_deinit_slave(bond, slave);
2093 }
2094
2095 if (oldcurrent == slave) {
2096 bond_select_active_slave(bond);
2097
2098 if (!bond->curr_active_slave) {
2099 printk(KERN_INFO DRV_NAME
2100 ": %s: now running without any active "
2101 "interface !\n",
2102 bond_dev->name);
2103 }
2104 }
2105
2106 if (bond->slave_cnt == 0) {
2107 /* if the last slave was removed, zero the mac address
2108 * of the master so it will be set by the application
2109 * to the mac address of the first slave
2110 */
2111 memset(bond_dev->dev_addr, 0, bond_dev->addr_len);
2112
2113 if (list_empty(&bond->vlan_list)) {
2114 bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
2115 } else {
2116 printk(KERN_WARNING DRV_NAME
2117 ": Warning: clearing HW address of %s while it "
2118 "still has VLANs.\n",
2119 bond_dev->name);
2120 printk(KERN_WARNING DRV_NAME
2121 ": When re-adding slaves, make sure the bond's "
2122 "HW address matches its VLANs'.\n");
2123 }
2124 } else if ((bond_dev->features & NETIF_F_VLAN_CHALLENGED) &&
2125 !bond_has_challenged_slaves(bond)) {
2126 printk(KERN_INFO DRV_NAME
2127 ": last VLAN challenged slave %s "
2128 "left bond %s. VLAN blocking is removed\n",
2129 slave_dev->name, bond_dev->name);
2130 bond_dev->features &= ~NETIF_F_VLAN_CHALLENGED;
2131 }
2132
2133 write_unlock_bh(&bond->lock);
2134
2135 bond_del_vlans_from_slave(bond, slave_dev);
2136
2137 /* If the mode USES_PRIMARY, then we should only remove its
2138 * promisc and mc settings if it was the curr_active_slave, but that was
2139 * already taken care of above when we detached the slave
2140 */
2141 if (!USES_PRIMARY(bond->params.mode)) {
2142 /* unset promiscuity level from slave */
2143 if (bond_dev->flags & IFF_PROMISC) {
2144 dev_set_promiscuity(slave_dev, -1);
2145 }
2146
2147 /* unset allmulti level from slave */
2148 if (bond_dev->flags & IFF_ALLMULTI) {
2149 dev_set_allmulti(slave_dev, -1);
2150 }
2151
2152 /* flush master's mc_list from slave */
2153 bond_mc_list_flush(bond_dev, slave_dev);
2154 }
2155
2156 netdev_set_master(slave_dev, NULL);
2157
2158 /* close slave before restoring its mac address */
2159 dev_close(slave_dev);
2160
2161 if (app_abi_ver >= 1) {
2162 /* restore original ("permanent") mac address */
2163 memcpy(addr.sa_data, slave->perm_hwaddr, ETH_ALEN);
2164 addr.sa_family = slave_dev->type;
2165 dev_set_mac_address(slave_dev, &addr);
2166 }
2167
2168 /* restore the original state of the
2169 * IFF_NOARP flag that might have been
2170 * set by bond_set_slave_inactive_flags()
2171 */
2172 if ((slave->original_flags & IFF_NOARP) == 0) {
2173 slave_dev->flags &= ~IFF_NOARP;
2174 }
2175
2176 kfree(slave);
2177
2178 return 0; /* deletion OK */
2179}
2180
2181/*
2182 * This function releases all slaves.
2183 */
2184static int bond_release_all(struct net_device *bond_dev)
2185{
2186 struct bonding *bond = bond_dev->priv;
2187 struct slave *slave;
2188 struct net_device *slave_dev;
2189 struct sockaddr addr;
2190
2191 write_lock_bh(&bond->lock);
2192
2193 if (bond->slave_cnt == 0) {
2194 goto out;
2195 }
2196
2197 bond->current_arp_slave = NULL;
2198 bond->primary_slave = NULL;
2199 bond_change_active_slave(bond, NULL);
2200
2201 while ((slave = bond->first_slave) != NULL) {
2202 /* Inform AD package of unbinding of slave
2203 * before slave is detached from the list.
2204 */
2205 if (bond->params.mode == BOND_MODE_8023AD) {
2206 bond_3ad_unbind_slave(slave);
2207 }
2208
2209 slave_dev = slave->dev;
2210 bond_detach_slave(bond, slave);
2211
2212 if ((bond->params.mode == BOND_MODE_TLB) ||
2213 (bond->params.mode == BOND_MODE_ALB)) {
2214 /* must be called only after the slave
2215 * has been detached from the list
2216 */
2217 bond_alb_deinit_slave(bond, slave);
2218 }
2219
2220 /* now that the slave is detached, unlock and perform
2221 * all the undo steps that should not be called from
2222 * within a lock.
2223 */
2224 write_unlock_bh(&bond->lock);
2225
2226 bond_del_vlans_from_slave(bond, slave_dev);
2227
2228 /* If the mode USES_PRIMARY, then we should only remove its
2229 * promisc and mc settings if it was the curr_active_slave, but that was
2230 * already taken care of above when we detached the slave
2231 */
2232 if (!USES_PRIMARY(bond->params.mode)) {
2233 /* unset promiscuity level from slave */
2234 if (bond_dev->flags & IFF_PROMISC) {
2235 dev_set_promiscuity(slave_dev, -1);
2236 }
2237
2238 /* unset allmulti level from slave */
2239 if (bond_dev->flags & IFF_ALLMULTI) {
2240 dev_set_allmulti(slave_dev, -1);
2241 }
2242
2243 /* flush master's mc_list from slave */
2244 bond_mc_list_flush(bond_dev, slave_dev);
2245 }
2246
2247 netdev_set_master(slave_dev, NULL);
2248
2249 /* close slave before restoring its mac address */
2250 dev_close(slave_dev);
2251
2252 if (app_abi_ver >= 1) {
2253 /* restore original ("permanent") mac address*/
2254 memcpy(addr.sa_data, slave->perm_hwaddr, ETH_ALEN);
2255 addr.sa_family = slave_dev->type;
2256 dev_set_mac_address(slave_dev, &addr);
2257 }
2258
2259 /* restore the original state of the IFF_NOARP flag that might have
2260 * been set by bond_set_slave_inactive_flags()
2261 */
2262 if ((slave->original_flags & IFF_NOARP) == 0) {
2263 slave_dev->flags &= ~IFF_NOARP;
2264 }
2265
2266 kfree(slave);
2267
2268 /* re-acquire the lock before getting the next slave */
2269 write_lock_bh(&bond->lock);
2270 }
2271
2272 /* zero the mac address of the master so it will be
2273 * set by the application to the mac address of the
2274 * first slave
2275 */
2276 memset(bond_dev->dev_addr, 0, bond_dev->addr_len);
2277
2278 if (list_empty(&bond->vlan_list)) {
2279 bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
2280 } else {
2281 printk(KERN_WARNING DRV_NAME
2282 ": Warning: clearing HW address of %s while it "
2283 "still has VLANs.\n",
2284 bond_dev->name);
2285 printk(KERN_WARNING DRV_NAME
2286 ": When re-adding slaves, make sure the bond's "
2287 "HW address matches its VLANs'.\n");
2288 }
2289
2290 printk(KERN_INFO DRV_NAME
2291 ": %s: released all slaves\n",
2292 bond_dev->name);
2293
2294out:
2295 write_unlock_bh(&bond->lock);
2296
2297 return 0;
2298}
2299
2300/*
2301 * This function changes the active slave to slave <slave_dev>.
2302 * It returns -EINVAL in the following cases.
2303 * - <slave_dev> is not found in the list.
2304 * - There is not active slave now.
2305 * - <slave_dev> is already active.
2306 * - The link state of <slave_dev> is not BOND_LINK_UP.
2307 * - <slave_dev> is not running.
2308 * In these cases, this fuction does nothing.
2309 * In the other cases, currnt_slave pointer is changed and 0 is returned.
2310 */
2311static int bond_ioctl_change_active(struct net_device *bond_dev, struct net_device *slave_dev)
2312{
2313 struct bonding *bond = bond_dev->priv;
2314 struct slave *old_active = NULL;
2315 struct slave *new_active = NULL;
2316 int res = 0;
2317
2318 if (!USES_PRIMARY(bond->params.mode)) {
2319 return -EINVAL;
2320 }
2321
2322 /* Verify that master_dev is indeed the master of slave_dev */
2323 if (!(slave_dev->flags & IFF_SLAVE) ||
2324 (slave_dev->master != bond_dev)) {
2325 return -EINVAL;
2326 }
2327
2328 write_lock_bh(&bond->lock);
2329
2330 old_active = bond->curr_active_slave;
2331 new_active = bond_get_slave_by_dev(bond, slave_dev);
2332
2333 /*
2334 * Changing to the current active: do nothing; return success.
2335 */
2336 if (new_active && (new_active == old_active)) {
2337 write_unlock_bh(&bond->lock);
2338 return 0;
2339 }
2340
2341 if ((new_active) &&
2342 (old_active) &&
2343 (new_active->link == BOND_LINK_UP) &&
2344 IS_UP(new_active->dev)) {
2345 bond_change_active_slave(bond, new_active);
2346 } else {
2347 res = -EINVAL;
2348 }
2349
2350 write_unlock_bh(&bond->lock);
2351
2352 return res;
2353}
2354
2355static int bond_ethtool_ioctl(struct net_device *bond_dev, struct ifreq *ifr)
2356{
2357 struct ethtool_drvinfo info;
2358 void __user *addr = ifr->ifr_data;
2359 uint32_t cmd;
2360
2361 if (get_user(cmd, (uint32_t __user *)addr)) {
2362 return -EFAULT;
2363 }
2364
2365 switch (cmd) {
2366 case ETHTOOL_GDRVINFO:
2367 if (copy_from_user(&info, addr, sizeof(info))) {
2368 return -EFAULT;
2369 }
2370
2371 if (strcmp(info.driver, "ifenslave") == 0) {
2372 int new_abi_ver;
2373 char *endptr;
2374
2375 new_abi_ver = simple_strtoul(info.fw_version,
2376 &endptr, 0);
2377 if (*endptr) {
2378 printk(KERN_ERR DRV_NAME
2379 ": Error: got invalid ABI "
2380 "version from application\n");
2381
2382 return -EINVAL;
2383 }
2384
2385 if (orig_app_abi_ver == -1) {
2386 orig_app_abi_ver = new_abi_ver;
2387 }
2388
2389 app_abi_ver = new_abi_ver;
2390 }
2391
2392 strncpy(info.driver, DRV_NAME, 32);
2393 strncpy(info.version, DRV_VERSION, 32);
2394 snprintf(info.fw_version, 32, "%d", BOND_ABI_VERSION);
2395
2396 if (copy_to_user(addr, &info, sizeof(info))) {
2397 return -EFAULT;
2398 }
2399
2400 return 0;
2401 default:
2402 return -EOPNOTSUPP;
2403 }
2404}
2405
2406static int bond_info_query(struct net_device *bond_dev, struct ifbond *info)
2407{
2408 struct bonding *bond = bond_dev->priv;
2409
2410 info->bond_mode = bond->params.mode;
2411 info->miimon = bond->params.miimon;
2412
2413 read_lock_bh(&bond->lock);
2414 info->num_slaves = bond->slave_cnt;
2415 read_unlock_bh(&bond->lock);
2416
2417 return 0;
2418}
2419
2420static int bond_slave_info_query(struct net_device *bond_dev, struct ifslave *info)
2421{
2422 struct bonding *bond = bond_dev->priv;
2423 struct slave *slave;
2424 int i, found = 0;
2425
2426 if (info->slave_id < 0) {
2427 return -ENODEV;
2428 }
2429
2430 read_lock_bh(&bond->lock);
2431
2432 bond_for_each_slave(bond, slave, i) {
2433 if (i == (int)info->slave_id) {
2434 found = 1;
2435 break;
2436 }
2437 }
2438
2439 read_unlock_bh(&bond->lock);
2440
2441 if (found) {
2442 strcpy(info->slave_name, slave->dev->name);
2443 info->link = slave->link;
2444 info->state = slave->state;
2445 info->link_failure_count = slave->link_failure_count;
2446 } else {
2447 return -ENODEV;
2448 }
2449
2450 return 0;
2451}
2452
2453/*-------------------------------- Monitoring -------------------------------*/
2454
2455/* this function is called regularly to monitor each slave's link. */
2456static void bond_mii_monitor(struct net_device *bond_dev)
2457{
2458 struct bonding *bond = bond_dev->priv;
2459 struct slave *slave, *oldcurrent;
2460 int do_failover = 0;
2461 int delta_in_ticks;
2462 int i;
2463
2464 read_lock(&bond->lock);
2465
2466 delta_in_ticks = (bond->params.miimon * HZ) / 1000;
2467
2468 if (bond->kill_timers) {
2469 goto out;
2470 }
2471
2472 if (bond->slave_cnt == 0) {
2473 goto re_arm;
2474 }
2475
2476 /* we will try to read the link status of each of our slaves, and
2477 * set their IFF_RUNNING flag appropriately. For each slave not
2478 * supporting MII status, we won't do anything so that a user-space
2479 * program could monitor the link itself if needed.
2480 */
2481
2482 read_lock(&bond->curr_slave_lock);
2483 oldcurrent = bond->curr_active_slave;
2484 read_unlock(&bond->curr_slave_lock);
2485
2486 bond_for_each_slave(bond, slave, i) {
2487 struct net_device *slave_dev = slave->dev;
2488 int link_state;
2489 u16 old_speed = slave->speed;
2490 u8 old_duplex = slave->duplex;
2491
2492 link_state = bond_check_dev_link(bond, slave_dev, 0);
2493
2494 switch (slave->link) {
2495 case BOND_LINK_UP: /* the link was up */
2496 if (link_state == BMSR_LSTATUS) {
2497 /* link stays up, nothing more to do */
2498 break;
2499 } else { /* link going down */
2500 slave->link = BOND_LINK_FAIL;
2501 slave->delay = bond->params.downdelay;
2502
2503 if (slave->link_failure_count < UINT_MAX) {
2504 slave->link_failure_count++;
2505 }
2506
2507 if (bond->params.downdelay) {
2508 printk(KERN_INFO DRV_NAME
2509 ": %s: link status down for %s "
2510 "interface %s, disabling it in "
2511 "%d ms.\n",
2512 bond_dev->name,
2513 IS_UP(slave_dev)
2514 ? ((bond->params.mode == BOND_MODE_ACTIVEBACKUP)
2515 ? ((slave == oldcurrent)
2516 ? "active " : "backup ")
2517 : "")
2518 : "idle ",
2519 slave_dev->name,
2520 bond->params.downdelay * bond->params.miimon);
2521 }
2522 }
2523 /* no break ! fall through the BOND_LINK_FAIL test to
2524 ensure proper action to be taken
2525 */
2526 case BOND_LINK_FAIL: /* the link has just gone down */
2527 if (link_state != BMSR_LSTATUS) {
2528 /* link stays down */
2529 if (slave->delay <= 0) {
2530 /* link down for too long time */
2531 slave->link = BOND_LINK_DOWN;
2532
2533 /* in active/backup mode, we must
2534 * completely disable this interface
2535 */
2536 if ((bond->params.mode == BOND_MODE_ACTIVEBACKUP) ||
2537 (bond->params.mode == BOND_MODE_8023AD)) {
2538 bond_set_slave_inactive_flags(slave);
2539 }
2540
2541 printk(KERN_INFO DRV_NAME
2542 ": %s: link status definitely "
2543 "down for interface %s, "
2544 "disabling it\n",
2545 bond_dev->name,
2546 slave_dev->name);
2547
2548 /* notify ad that the link status has changed */
2549 if (bond->params.mode == BOND_MODE_8023AD) {
2550 bond_3ad_handle_link_change(slave, BOND_LINK_DOWN);
2551 }
2552
2553 if ((bond->params.mode == BOND_MODE_TLB) ||
2554 (bond->params.mode == BOND_MODE_ALB)) {
2555 bond_alb_handle_link_change(bond, slave, BOND_LINK_DOWN);
2556 }
2557
2558 if (slave == oldcurrent) {
2559 do_failover = 1;
2560 }
2561 } else {
2562 slave->delay--;
2563 }
2564 } else {
2565 /* link up again */
2566 slave->link = BOND_LINK_UP;
2567 slave->jiffies = jiffies;
2568 printk(KERN_INFO DRV_NAME
2569 ": %s: link status up again after %d "
2570 "ms for interface %s.\n",
2571 bond_dev->name,
2572 (bond->params.downdelay - slave->delay) * bond->params.miimon,
2573 slave_dev->name);
2574 }
2575 break;
2576 case BOND_LINK_DOWN: /* the link was down */
2577 if (link_state != BMSR_LSTATUS) {
2578 /* the link stays down, nothing more to do */
2579 break;
2580 } else { /* link going up */
2581 slave->link = BOND_LINK_BACK;
2582 slave->delay = bond->params.updelay;
2583
2584 if (bond->params.updelay) {
2585 /* if updelay == 0, no need to
2586 advertise about a 0 ms delay */
2587 printk(KERN_INFO DRV_NAME
2588 ": %s: link status up for "
2589 "interface %s, enabling it "
2590 "in %d ms.\n",
2591 bond_dev->name,
2592 slave_dev->name,
2593 bond->params.updelay * bond->params.miimon);
2594 }
2595 }
2596 /* no break ! fall through the BOND_LINK_BACK state in
2597 case there's something to do.
2598 */
2599 case BOND_LINK_BACK: /* the link has just come back */
2600 if (link_state != BMSR_LSTATUS) {
2601 /* link down again */
2602 slave->link = BOND_LINK_DOWN;
2603
2604 printk(KERN_INFO DRV_NAME
2605 ": %s: link status down again after %d "
2606 "ms for interface %s.\n",
2607 bond_dev->name,
2608 (bond->params.updelay - slave->delay) * bond->params.miimon,
2609 slave_dev->name);
2610 } else {
2611 /* link stays up */
2612 if (slave->delay == 0) {
2613 /* now the link has been up for long time enough */
2614 slave->link = BOND_LINK_UP;
2615 slave->jiffies = jiffies;
2616
2617 if (bond->params.mode == BOND_MODE_8023AD) {
2618 /* prevent it from being the active one */
2619 slave->state = BOND_STATE_BACKUP;
2620 } else if (bond->params.mode != BOND_MODE_ACTIVEBACKUP) {
2621 /* make it immediately active */
2622 slave->state = BOND_STATE_ACTIVE;
2623 } else if (slave != bond->primary_slave) {
2624 /* prevent it from being the active one */
2625 slave->state = BOND_STATE_BACKUP;
2626 }
2627
2628 printk(KERN_INFO DRV_NAME
2629 ": %s: link status definitely "
2630 "up for interface %s.\n",
2631 bond_dev->name,
2632 slave_dev->name);
2633
2634 /* notify ad that the link status has changed */
2635 if (bond->params.mode == BOND_MODE_8023AD) {
2636 bond_3ad_handle_link_change(slave, BOND_LINK_UP);
2637 }
2638
2639 if ((bond->params.mode == BOND_MODE_TLB) ||
2640 (bond->params.mode == BOND_MODE_ALB)) {
2641 bond_alb_handle_link_change(bond, slave, BOND_LINK_UP);
2642 }
2643
2644 if ((!oldcurrent) ||
2645 (slave == bond->primary_slave)) {
2646 do_failover = 1;
2647 }
2648 } else {
2649 slave->delay--;
2650 }
2651 }
2652 break;
2653 default:
2654 /* Should not happen */
2655 printk(KERN_ERR "bonding: Error: %s Illegal value (link=%d)\n",
2656 slave->dev->name, slave->link);
2657 goto out;
2658 } /* end of switch (slave->link) */
2659
2660 bond_update_speed_duplex(slave);
2661
2662 if (bond->params.mode == BOND_MODE_8023AD) {
2663 if (old_speed != slave->speed) {
2664 bond_3ad_adapter_speed_changed(slave);
2665 }
2666
2667 if (old_duplex != slave->duplex) {
2668 bond_3ad_adapter_duplex_changed(slave);
2669 }
2670 }
2671
2672 } /* end of for */
2673
2674 if (do_failover) {
2675 write_lock(&bond->curr_slave_lock);
2676
2677 bond_select_active_slave(bond);
2678
2679 if (oldcurrent && !bond->curr_active_slave) {
2680 printk(KERN_INFO DRV_NAME
2681 ": %s: now running without any active "
2682 "interface !\n",
2683 bond_dev->name);
2684 }
2685
2686 write_unlock(&bond->curr_slave_lock);
2687 }
2688
2689re_arm:
2690 if (bond->params.miimon) {
2691 mod_timer(&bond->mii_timer, jiffies + delta_in_ticks);
2692 }
2693out:
2694 read_unlock(&bond->lock);
2695}
2696
2697static void bond_arp_send_all(struct bonding *bond, struct slave *slave)
2698{
2699 int i;
2700 u32 *targets = bond->params.arp_targets;
2701
2702 for (i = 0; (i < BOND_MAX_ARP_TARGETS) && targets[i]; i++) {
2703 arp_send(ARPOP_REQUEST, ETH_P_ARP, targets[i], slave->dev,
2704 my_ip, NULL, slave->dev->dev_addr,
2705 NULL);
2706 }
2707}
2708
2709/*
2710 * this function is called regularly to monitor each slave's link
2711 * ensuring that traffic is being sent and received when arp monitoring
2712 * is used in load-balancing mode. if the adapter has been dormant, then an
2713 * arp is transmitted to generate traffic. see activebackup_arp_monitor for
2714 * arp monitoring in active backup mode.
2715 */
2716static void bond_loadbalance_arp_mon(struct net_device *bond_dev)
2717{
2718 struct bonding *bond = bond_dev->priv;
2719 struct slave *slave, *oldcurrent;
2720 int do_failover = 0;
2721 int delta_in_ticks;
2722 int i;
2723
2724 read_lock(&bond->lock);
2725
2726 delta_in_ticks = (bond->params.arp_interval * HZ) / 1000;
2727
2728 if (bond->kill_timers) {
2729 goto out;
2730 }
2731
2732 if (bond->slave_cnt == 0) {
2733 goto re_arm;
2734 }
2735
2736 read_lock(&bond->curr_slave_lock);
2737 oldcurrent = bond->curr_active_slave;
2738 read_unlock(&bond->curr_slave_lock);
2739
2740 /* see if any of the previous devices are up now (i.e. they have
2741 * xmt and rcv traffic). the curr_active_slave does not come into
2742 * the picture unless it is null. also, slave->jiffies is not needed
2743 * here because we send an arp on each slave and give a slave as
2744 * long as it needs to get the tx/rx within the delta.
2745 * TODO: what about up/down delay in arp mode? it wasn't here before
2746 * so it can wait
2747 */
2748 bond_for_each_slave(bond, slave, i) {
2749 if (slave->link != BOND_LINK_UP) {
2750 if (((jiffies - slave->dev->trans_start) <= delta_in_ticks) &&
2751 ((jiffies - slave->dev->last_rx) <= delta_in_ticks)) {
2752
2753 slave->link = BOND_LINK_UP;
2754 slave->state = BOND_STATE_ACTIVE;
2755
2756 /* primary_slave has no meaning in round-robin
2757 * mode. the window of a slave being up and
2758 * curr_active_slave being null after enslaving
2759 * is closed.
2760 */
2761 if (!oldcurrent) {
2762 printk(KERN_INFO DRV_NAME
2763 ": %s: link status definitely "
2764 "up for interface %s, ",
2765 bond_dev->name,
2766 slave->dev->name);
2767 do_failover = 1;
2768 } else {
2769 printk(KERN_INFO DRV_NAME
2770 ": %s: interface %s is now up\n",
2771 bond_dev->name,
2772 slave->dev->name);
2773 }
2774 }
2775 } else {
2776 /* slave->link == BOND_LINK_UP */
2777
2778 /* not all switches will respond to an arp request
2779 * when the source ip is 0, so don't take the link down
2780 * if we don't know our ip yet
2781 */
2782 if (((jiffies - slave->dev->trans_start) >= (2*delta_in_ticks)) ||
2783 (((jiffies - slave->dev->last_rx) >= (2*delta_in_ticks)) &&
2784 my_ip)) {
2785
2786 slave->link = BOND_LINK_DOWN;
2787 slave->state = BOND_STATE_BACKUP;
2788
2789 if (slave->link_failure_count < UINT_MAX) {
2790 slave->link_failure_count++;
2791 }
2792
2793 printk(KERN_INFO DRV_NAME
2794 ": %s: interface %s is now down.\n",
2795 bond_dev->name,
2796 slave->dev->name);
2797
2798 if (slave == oldcurrent) {
2799 do_failover = 1;
2800 }
2801 }
2802 }
2803
2804 /* note: if switch is in round-robin mode, all links
2805 * must tx arp to ensure all links rx an arp - otherwise
2806 * links may oscillate or not come up at all; if switch is
2807 * in something like xor mode, there is nothing we can
2808 * do - all replies will be rx'ed on same link causing slaves
2809 * to be unstable during low/no traffic periods
2810 */
2811 if (IS_UP(slave->dev)) {
2812 bond_arp_send_all(bond, slave);
2813 }
2814 }
2815
2816 if (do_failover) {
2817 write_lock(&bond->curr_slave_lock);
2818
2819 bond_select_active_slave(bond);
2820
2821 if (oldcurrent && !bond->curr_active_slave) {
2822 printk(KERN_INFO DRV_NAME
2823 ": %s: now running without any active "
2824 "interface !\n",
2825 bond_dev->name);
2826 }
2827
2828 write_unlock(&bond->curr_slave_lock);
2829 }
2830
2831re_arm:
2832 if (bond->params.arp_interval) {
2833 mod_timer(&bond->arp_timer, jiffies + delta_in_ticks);
2834 }
2835out:
2836 read_unlock(&bond->lock);
2837}
2838
2839/*
2840 * When using arp monitoring in active-backup mode, this function is
2841 * called to determine if any backup slaves have went down or a new
2842 * current slave needs to be found.
2843 * The backup slaves never generate traffic, they are considered up by merely
2844 * receiving traffic. If the current slave goes down, each backup slave will
2845 * be given the opportunity to tx/rx an arp before being taken down - this
2846 * prevents all slaves from being taken down due to the current slave not
2847 * sending any traffic for the backups to receive. The arps are not necessarily
2848 * necessary, any tx and rx traffic will keep the current slave up. While any
2849 * rx traffic will keep the backup slaves up, the current slave is responsible
2850 * for generating traffic to keep them up regardless of any other traffic they
2851 * may have received.
2852 * see loadbalance_arp_monitor for arp monitoring in load balancing mode
2853 */
2854static void bond_activebackup_arp_mon(struct net_device *bond_dev)
2855{
2856 struct bonding *bond = bond_dev->priv;
2857 struct slave *slave;
2858 int delta_in_ticks;
2859 int i;
2860
2861 read_lock(&bond->lock);
2862
2863 delta_in_ticks = (bond->params.arp_interval * HZ) / 1000;
2864
2865 if (bond->kill_timers) {
2866 goto out;
2867 }
2868
2869 if (bond->slave_cnt == 0) {
2870 goto re_arm;
2871 }
2872
2873 /* determine if any slave has come up or any backup slave has
2874 * gone down
2875 * TODO: what about up/down delay in arp mode? it wasn't here before
2876 * so it can wait
2877 */
2878 bond_for_each_slave(bond, slave, i) {
2879 if (slave->link != BOND_LINK_UP) {
2880 if ((jiffies - slave->dev->last_rx) <= delta_in_ticks) {
2881
2882 slave->link = BOND_LINK_UP;
2883
2884 write_lock(&bond->curr_slave_lock);
2885
2886 if ((!bond->curr_active_slave) &&
2887 ((jiffies - slave->dev->trans_start) <= delta_in_ticks)) {
2888 bond_change_active_slave(bond, slave);
2889 bond->current_arp_slave = NULL;
2890 } else if (bond->curr_active_slave != slave) {
2891 /* this slave has just come up but we
2892 * already have a current slave; this
2893 * can also happen if bond_enslave adds
2894 * a new slave that is up while we are
2895 * searching for a new slave
2896 */
2897 bond_set_slave_inactive_flags(slave);
2898 bond->current_arp_slave = NULL;
2899 }
2900
2901 if (slave == bond->curr_active_slave) {
2902 printk(KERN_INFO DRV_NAME
2903 ": %s: %s is up and now the "
2904 "active interface\n",
2905 bond_dev->name,
2906 slave->dev->name);
2907 } else {
2908 printk(KERN_INFO DRV_NAME
2909 ": %s: backup interface %s is "
2910 "now up\n",
2911 bond_dev->name,
2912 slave->dev->name);
2913 }
2914
2915 write_unlock(&bond->curr_slave_lock);
2916 }
2917 } else {
2918 read_lock(&bond->curr_slave_lock);
2919
2920 if ((slave != bond->curr_active_slave) &&
2921 (!bond->current_arp_slave) &&
2922 (((jiffies - slave->dev->last_rx) >= 3*delta_in_ticks) &&
2923 my_ip)) {
2924 /* a backup slave has gone down; three times
2925 * the delta allows the current slave to be
2926 * taken out before the backup slave.
2927 * note: a non-null current_arp_slave indicates
2928 * the curr_active_slave went down and we are
2929 * searching for a new one; under this
2930 * condition we only take the curr_active_slave
2931 * down - this gives each slave a chance to
2932 * tx/rx traffic before being taken out
2933 */
2934
2935 read_unlock(&bond->curr_slave_lock);
2936
2937 slave->link = BOND_LINK_DOWN;
2938
2939 if (slave->link_failure_count < UINT_MAX) {
2940 slave->link_failure_count++;
2941 }
2942
2943 bond_set_slave_inactive_flags(slave);
2944
2945 printk(KERN_INFO DRV_NAME
2946 ": %s: backup interface %s is now down\n",
2947 bond_dev->name,
2948 slave->dev->name);
2949 } else {
2950 read_unlock(&bond->curr_slave_lock);
2951 }
2952 }
2953 }
2954
2955 read_lock(&bond->curr_slave_lock);
2956 slave = bond->curr_active_slave;
2957 read_unlock(&bond->curr_slave_lock);
2958
2959 if (slave) {
2960 /* if we have sent traffic in the past 2*arp_intervals but
2961 * haven't xmit and rx traffic in that time interval, select
2962 * a different slave. slave->jiffies is only updated when
2963 * a slave first becomes the curr_active_slave - not necessarily
2964 * after every arp; this ensures the slave has a full 2*delta
2965 * before being taken out. if a primary is being used, check
2966 * if it is up and needs to take over as the curr_active_slave
2967 */
2968 if ((((jiffies - slave->dev->trans_start) >= (2*delta_in_ticks)) ||
2969 (((jiffies - slave->dev->last_rx) >= (2*delta_in_ticks)) &&
2970 my_ip)) &&
2971 ((jiffies - slave->jiffies) >= 2*delta_in_ticks)) {
2972
2973 slave->link = BOND_LINK_DOWN;
2974
2975 if (slave->link_failure_count < UINT_MAX) {
2976 slave->link_failure_count++;
2977 }
2978
2979 printk(KERN_INFO DRV_NAME
2980 ": %s: link status down for active interface "
2981 "%s, disabling it\n",
2982 bond_dev->name,
2983 slave->dev->name);
2984
2985 write_lock(&bond->curr_slave_lock);
2986
2987 bond_select_active_slave(bond);
2988 slave = bond->curr_active_slave;
2989
2990 write_unlock(&bond->curr_slave_lock);
2991
2992 bond->current_arp_slave = slave;
2993
2994 if (slave) {
2995 slave->jiffies = jiffies;
2996 }
2997 } else if ((bond->primary_slave) &&
2998 (bond->primary_slave != slave) &&
2999 (bond->primary_slave->link == BOND_LINK_UP)) {
3000 /* at this point, slave is the curr_active_slave */
3001 printk(KERN_INFO DRV_NAME
3002 ": %s: changing from interface %s to primary "
3003 "interface %s\n",
3004 bond_dev->name,
3005 slave->dev->name,
3006 bond->primary_slave->dev->name);
3007
3008 /* primary is up so switch to it */
3009 write_lock(&bond->curr_slave_lock);
3010 bond_change_active_slave(bond, bond->primary_slave);
3011 write_unlock(&bond->curr_slave_lock);
3012
3013 slave = bond->primary_slave;
3014 slave->jiffies = jiffies;
3015 } else {
3016 bond->current_arp_slave = NULL;
3017 }
3018
3019 /* the current slave must tx an arp to ensure backup slaves
3020 * rx traffic
3021 */
3022 if (slave && my_ip) {
3023 bond_arp_send_all(bond, slave);
3024 }
3025 }
3026
3027 /* if we don't have a curr_active_slave, search for the next available
3028 * backup slave from the current_arp_slave and make it the candidate
3029 * for becoming the curr_active_slave
3030 */
3031 if (!slave) {
3032 if (!bond->current_arp_slave) {
3033 bond->current_arp_slave = bond->first_slave;
3034 }
3035
3036 if (bond->current_arp_slave) {
3037 bond_set_slave_inactive_flags(bond->current_arp_slave);
3038
3039 /* search for next candidate */
3040 bond_for_each_slave_from(bond, slave, i, bond->current_arp_slave) {
3041 if (IS_UP(slave->dev)) {
3042 slave->link = BOND_LINK_BACK;
3043 bond_set_slave_active_flags(slave);
3044 bond_arp_send_all(bond, slave);
3045 slave->jiffies = jiffies;
3046 bond->current_arp_slave = slave;
3047 break;
3048 }
3049
3050 /* if the link state is up at this point, we
3051 * mark it down - this can happen if we have
3052 * simultaneous link failures and
3053 * reselect_active_interface doesn't make this
3054 * one the current slave so it is still marked
3055 * up when it is actually down
3056 */
3057 if (slave->link == BOND_LINK_UP) {
3058 slave->link = BOND_LINK_DOWN;
3059 if (slave->link_failure_count < UINT_MAX) {
3060 slave->link_failure_count++;
3061 }
3062
3063 bond_set_slave_inactive_flags(slave);
3064
3065 printk(KERN_INFO DRV_NAME
3066 ": %s: backup interface %s is "
3067 "now down.\n",
3068 bond_dev->name,
3069 slave->dev->name);
3070 }
3071 }
3072 }
3073 }
3074
3075re_arm:
3076 if (bond->params.arp_interval) {
3077 mod_timer(&bond->arp_timer, jiffies + delta_in_ticks);
3078 }
3079out:
3080 read_unlock(&bond->lock);
3081}
3082
3083/*------------------------------ proc/seq_file-------------------------------*/
3084
3085#ifdef CONFIG_PROC_FS
3086
3087#define SEQ_START_TOKEN ((void *)1)
3088
3089static void *bond_info_seq_start(struct seq_file *seq, loff_t *pos)
3090{
3091 struct bonding *bond = seq->private;
3092 loff_t off = 0;
3093 struct slave *slave;
3094 int i;
3095
3096 /* make sure the bond won't be taken away */
3097 read_lock(&dev_base_lock);
3098 read_lock_bh(&bond->lock);
3099
3100 if (*pos == 0) {
3101 return SEQ_START_TOKEN;
3102 }
3103
3104 bond_for_each_slave(bond, slave, i) {
3105 if (++off == *pos) {
3106 return slave;
3107 }
3108 }
3109
3110 return NULL;
3111}
3112
3113static void *bond_info_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3114{
3115 struct bonding *bond = seq->private;
3116 struct slave *slave = v;
3117
3118 ++*pos;
3119 if (v == SEQ_START_TOKEN) {
3120 return bond->first_slave;
3121 }
3122
3123 slave = slave->next;
3124
3125 return (slave == bond->first_slave) ? NULL : slave;
3126}
3127
3128static void bond_info_seq_stop(struct seq_file *seq, void *v)
3129{
3130 struct bonding *bond = seq->private;
3131
3132 read_unlock_bh(&bond->lock);
3133 read_unlock(&dev_base_lock);
3134}
3135
3136static void bond_info_show_master(struct seq_file *seq)
3137{
3138 struct bonding *bond = seq->private;
3139 struct slave *curr;
3140
3141 read_lock(&bond->curr_slave_lock);
3142 curr = bond->curr_active_slave;
3143 read_unlock(&bond->curr_slave_lock);
3144
3145 seq_printf(seq, "Bonding Mode: %s\n",
3146 bond_mode_name(bond->params.mode));
3147
3148 if (USES_PRIMARY(bond->params.mode)) {
3149 seq_printf(seq, "Primary Slave: %s\n",
3150 (bond->params.primary[0]) ?
3151 bond->params.primary : "None");
3152
3153 seq_printf(seq, "Currently Active Slave: %s\n",
3154 (curr) ? curr->dev->name : "None");
3155 }
3156
3157 seq_printf(seq, "MII Status: %s\n", (curr) ? "up" : "down");
3158 seq_printf(seq, "MII Polling Interval (ms): %d\n", bond->params.miimon);
3159 seq_printf(seq, "Up Delay (ms): %d\n",
3160 bond->params.updelay * bond->params.miimon);
3161 seq_printf(seq, "Down Delay (ms): %d\n",
3162 bond->params.downdelay * bond->params.miimon);
3163
3164 if (bond->params.mode == BOND_MODE_8023AD) {
3165 struct ad_info ad_info;
3166
3167 seq_puts(seq, "\n802.3ad info\n");
3168 seq_printf(seq, "LACP rate: %s\n",
3169 (bond->params.lacp_fast) ? "fast" : "slow");
3170
3171 if (bond_3ad_get_active_agg_info(bond, &ad_info)) {
3172 seq_printf(seq, "bond %s has no active aggregator\n",
3173 bond->dev->name);
3174 } else {
3175 seq_printf(seq, "Active Aggregator Info:\n");
3176
3177 seq_printf(seq, "\tAggregator ID: %d\n",
3178 ad_info.aggregator_id);
3179 seq_printf(seq, "\tNumber of ports: %d\n",
3180 ad_info.ports);
3181 seq_printf(seq, "\tActor Key: %d\n",
3182 ad_info.actor_key);
3183 seq_printf(seq, "\tPartner Key: %d\n",
3184 ad_info.partner_key);
3185 seq_printf(seq, "\tPartner Mac Address: %02x:%02x:%02x:%02x:%02x:%02x\n",
3186 ad_info.partner_system[0],
3187 ad_info.partner_system[1],
3188 ad_info.partner_system[2],
3189 ad_info.partner_system[3],
3190 ad_info.partner_system[4],
3191 ad_info.partner_system[5]);
3192 }
3193 }
3194}
3195
3196static void bond_info_show_slave(struct seq_file *seq, const struct slave *slave)
3197{
3198 struct bonding *bond = seq->private;
3199
3200 seq_printf(seq, "\nSlave Interface: %s\n", slave->dev->name);
3201 seq_printf(seq, "MII Status: %s\n",
3202 (slave->link == BOND_LINK_UP) ? "up" : "down");
3203 seq_printf(seq, "Link Failure Count: %d\n",
3204 slave->link_failure_count);
3205
3206 if (app_abi_ver >= 1) {
3207 seq_printf(seq,
3208 "Permanent HW addr: %02x:%02x:%02x:%02x:%02x:%02x\n",
3209 slave->perm_hwaddr[0],
3210 slave->perm_hwaddr[1],
3211 slave->perm_hwaddr[2],
3212 slave->perm_hwaddr[3],
3213 slave->perm_hwaddr[4],
3214 slave->perm_hwaddr[5]);
3215 }
3216
3217 if (bond->params.mode == BOND_MODE_8023AD) {
3218 const struct aggregator *agg
3219 = SLAVE_AD_INFO(slave).port.aggregator;
3220
3221 if (agg) {
3222 seq_printf(seq, "Aggregator ID: %d\n",
3223 agg->aggregator_identifier);
3224 } else {
3225 seq_puts(seq, "Aggregator ID: N/A\n");
3226 }
3227 }
3228}
3229
3230static int bond_info_seq_show(struct seq_file *seq, void *v)
3231{
3232 if (v == SEQ_START_TOKEN) {
3233 seq_printf(seq, "%s\n", version);
3234 bond_info_show_master(seq);
3235 } else {
3236 bond_info_show_slave(seq, v);
3237 }
3238
3239 return 0;
3240}
3241
3242static struct seq_operations bond_info_seq_ops = {
3243 .start = bond_info_seq_start,
3244 .next = bond_info_seq_next,
3245 .stop = bond_info_seq_stop,
3246 .show = bond_info_seq_show,
3247};
3248
3249static int bond_info_open(struct inode *inode, struct file *file)
3250{
3251 struct seq_file *seq;
3252 struct proc_dir_entry *proc;
3253 int res;
3254
3255 res = seq_open(file, &bond_info_seq_ops);
3256 if (!res) {
3257 /* recover the pointer buried in proc_dir_entry data */
3258 seq = file->private_data;
3259 proc = PDE(inode);
3260 seq->private = proc->data;
3261 }
3262
3263 return res;
3264}
3265
3266static struct file_operations bond_info_fops = {
3267 .owner = THIS_MODULE,
3268 .open = bond_info_open,
3269 .read = seq_read,
3270 .llseek = seq_lseek,
3271 .release = seq_release,
3272};
3273
3274static int bond_create_proc_entry(struct bonding *bond)
3275{
3276 struct net_device *bond_dev = bond->dev;
3277
3278 if (bond_proc_dir) {
3279 bond->proc_entry = create_proc_entry(bond_dev->name,
3280 S_IRUGO,
3281 bond_proc_dir);
3282 if (bond->proc_entry == NULL) {
3283 printk(KERN_WARNING DRV_NAME
3284 ": Warning: Cannot create /proc/net/%s/%s\n",
3285 DRV_NAME, bond_dev->name);
3286 } else {
3287 bond->proc_entry->data = bond;
3288 bond->proc_entry->proc_fops = &bond_info_fops;
3289 bond->proc_entry->owner = THIS_MODULE;
3290 memcpy(bond->proc_file_name, bond_dev->name, IFNAMSIZ);
3291 }
3292 }
3293
3294 return 0;
3295}
3296
3297static void bond_remove_proc_entry(struct bonding *bond)
3298{
3299 if (bond_proc_dir && bond->proc_entry) {
3300 remove_proc_entry(bond->proc_file_name, bond_proc_dir);
3301 memset(bond->proc_file_name, 0, IFNAMSIZ);
3302 bond->proc_entry = NULL;
3303 }
3304}
3305
3306/* Create the bonding directory under /proc/net, if doesn't exist yet.
3307 * Caller must hold rtnl_lock.
3308 */
3309static void bond_create_proc_dir(void)
3310{
3311 int len = strlen(DRV_NAME);
3312
3313 for (bond_proc_dir = proc_net->subdir; bond_proc_dir;
3314 bond_proc_dir = bond_proc_dir->next) {
3315 if ((bond_proc_dir->namelen == len) &&
3316 !memcmp(bond_proc_dir->name, DRV_NAME, len)) {
3317 break;
3318 }
3319 }
3320
3321 if (!bond_proc_dir) {
3322 bond_proc_dir = proc_mkdir(DRV_NAME, proc_net);
3323 if (bond_proc_dir) {
3324 bond_proc_dir->owner = THIS_MODULE;
3325 } else {
3326 printk(KERN_WARNING DRV_NAME
3327 ": Warning: cannot create /proc/net/%s\n",
3328 DRV_NAME);
3329 }
3330 }
3331}
3332
3333/* Destroy the bonding directory under /proc/net, if empty.
3334 * Caller must hold rtnl_lock.
3335 */
3336static void bond_destroy_proc_dir(void)
3337{
3338 struct proc_dir_entry *de;
3339
3340 if (!bond_proc_dir) {
3341 return;
3342 }
3343
3344 /* verify that the /proc dir is empty */
3345 for (de = bond_proc_dir->subdir; de; de = de->next) {
3346 /* ignore . and .. */
3347 if (*(de->name) != '.') {
3348 break;
3349 }
3350 }
3351
3352 if (de) {
3353 if (bond_proc_dir->owner == THIS_MODULE) {
3354 bond_proc_dir->owner = NULL;
3355 }
3356 } else {
3357 remove_proc_entry(DRV_NAME, proc_net);
3358 bond_proc_dir = NULL;
3359 }
3360}
3361#endif /* CONFIG_PROC_FS */
3362
3363/*-------------------------- netdev event handling --------------------------*/
3364
3365/*
3366 * Change device name
3367 */
3368static int bond_event_changename(struct bonding *bond)
3369{
3370#ifdef CONFIG_PROC_FS
3371 bond_remove_proc_entry(bond);
3372 bond_create_proc_entry(bond);
3373#endif
3374
3375 return NOTIFY_DONE;
3376}
3377
3378static int bond_master_netdev_event(unsigned long event, struct net_device *bond_dev)
3379{
3380 struct bonding *event_bond = bond_dev->priv;
3381
3382 switch (event) {
3383 case NETDEV_CHANGENAME:
3384 return bond_event_changename(event_bond);
3385 case NETDEV_UNREGISTER:
3386 /*
3387 * TODO: remove a bond from the list?
3388 */
3389 break;
3390 default:
3391 break;
3392 }
3393
3394 return NOTIFY_DONE;
3395}
3396
3397static int bond_slave_netdev_event(unsigned long event, struct net_device *slave_dev)
3398{
3399 struct net_device *bond_dev = slave_dev->master;
3400
3401 switch (event) {
3402 case NETDEV_UNREGISTER:
3403 if (bond_dev) {
3404 bond_release(bond_dev, slave_dev);
3405 }
3406 break;
3407 case NETDEV_CHANGE:
3408 /*
3409 * TODO: is this what we get if somebody
3410 * sets up a hierarchical bond, then rmmod's
3411 * one of the slave bonding devices?
3412 */
3413 break;
3414 case NETDEV_DOWN:
3415 /*
3416 * ... Or is it this?
3417 */
3418 break;
3419 case NETDEV_CHANGEMTU:
3420 /*
3421 * TODO: Should slaves be allowed to
3422 * independently alter their MTU? For
3423 * an active-backup bond, slaves need
3424 * not be the same type of device, so
3425 * MTUs may vary. For other modes,
3426 * slaves arguably should have the
3427 * same MTUs. To do this, we'd need to
3428 * take over the slave's change_mtu
3429 * function for the duration of their
3430 * servitude.
3431 */
3432 break;
3433 case NETDEV_CHANGENAME:
3434 /*
3435 * TODO: handle changing the primary's name
3436 */
3437 break;
3438 default:
3439 break;
3440 }
3441
3442 return NOTIFY_DONE;
3443}
3444
3445/*
3446 * bond_netdev_event: handle netdev notifier chain events.
3447 *
3448 * This function receives events for the netdev chain. The caller (an
3449 * ioctl handler calling notifier_call_chain) holds the necessary
3450 * locks for us to safely manipulate the slave devices (RTNL lock,
3451 * dev_probe_lock).
3452 */
3453static int bond_netdev_event(struct notifier_block *this, unsigned long event, void *ptr)
3454{
3455 struct net_device *event_dev = (struct net_device *)ptr;
3456
3457 dprintk("event_dev: %s, event: %lx\n",
3458 (event_dev ? event_dev->name : "None"),
3459 event);
3460
3461 if (event_dev->flags & IFF_MASTER) {
3462 dprintk("IFF_MASTER\n");
3463 return bond_master_netdev_event(event, event_dev);
3464 }
3465
3466 if (event_dev->flags & IFF_SLAVE) {
3467 dprintk("IFF_SLAVE\n");
3468 return bond_slave_netdev_event(event, event_dev);
3469 }
3470
3471 return NOTIFY_DONE;
3472}
3473
3474static struct notifier_block bond_netdev_notifier = {
3475 .notifier_call = bond_netdev_event,
3476};
3477
3478/*-------------------------- Packet type handling ---------------------------*/
3479
3480/* register to receive lacpdus on a bond */
3481static void bond_register_lacpdu(struct bonding *bond)
3482{
3483 struct packet_type *pk_type = &(BOND_AD_INFO(bond).ad_pkt_type);
3484
3485 /* initialize packet type */
3486 pk_type->type = PKT_TYPE_LACPDU;
3487 pk_type->dev = bond->dev;
3488 pk_type->func = bond_3ad_lacpdu_recv;
3489
3490 dev_add_pack(pk_type);
3491}
3492
3493/* unregister to receive lacpdus on a bond */
3494static void bond_unregister_lacpdu(struct bonding *bond)
3495{
3496 dev_remove_pack(&(BOND_AD_INFO(bond).ad_pkt_type));
3497}
3498
3499/*-------------------------- Device entry points ----------------------------*/
3500
3501static int bond_open(struct net_device *bond_dev)
3502{
3503 struct bonding *bond = bond_dev->priv;
3504 struct timer_list *mii_timer = &bond->mii_timer;
3505 struct timer_list *arp_timer = &bond->arp_timer;
3506
3507 bond->kill_timers = 0;
3508
3509 if ((bond->params.mode == BOND_MODE_TLB) ||
3510 (bond->params.mode == BOND_MODE_ALB)) {
3511 struct timer_list *alb_timer = &(BOND_ALB_INFO(bond).alb_timer);
3512
3513 /* bond_alb_initialize must be called before the timer
3514 * is started.
3515 */
3516 if (bond_alb_initialize(bond, (bond->params.mode == BOND_MODE_ALB))) {
3517 /* something went wrong - fail the open operation */
3518 return -1;
3519 }
3520
3521 init_timer(alb_timer);
3522 alb_timer->expires = jiffies + 1;
3523 alb_timer->data = (unsigned long)bond;
3524 alb_timer->function = (void *)&bond_alb_monitor;
3525 add_timer(alb_timer);
3526 }
3527
3528 if (bond->params.miimon) { /* link check interval, in milliseconds. */
3529 init_timer(mii_timer);
3530 mii_timer->expires = jiffies + 1;
3531 mii_timer->data = (unsigned long)bond_dev;
3532 mii_timer->function = (void *)&bond_mii_monitor;
3533 add_timer(mii_timer);
3534 }
3535
3536 if (bond->params.arp_interval) { /* arp interval, in milliseconds. */
3537 init_timer(arp_timer);
3538 arp_timer->expires = jiffies + 1;
3539 arp_timer->data = (unsigned long)bond_dev;
3540 if (bond->params.mode == BOND_MODE_ACTIVEBACKUP) {
3541 arp_timer->function = (void *)&bond_activebackup_arp_mon;
3542 } else {
3543 arp_timer->function = (void *)&bond_loadbalance_arp_mon;
3544 }
3545 add_timer(arp_timer);
3546 }
3547
3548 if (bond->params.mode == BOND_MODE_8023AD) {
3549 struct timer_list *ad_timer = &(BOND_AD_INFO(bond).ad_timer);
3550 init_timer(ad_timer);
3551 ad_timer->expires = jiffies + 1;
3552 ad_timer->data = (unsigned long)bond;
3553 ad_timer->function = (void *)&bond_3ad_state_machine_handler;
3554 add_timer(ad_timer);
3555
3556 /* register to receive LACPDUs */
3557 bond_register_lacpdu(bond);
3558 }
3559
3560 return 0;
3561}
3562
3563static int bond_close(struct net_device *bond_dev)
3564{
3565 struct bonding *bond = bond_dev->priv;
3566
3567 if (bond->params.mode == BOND_MODE_8023AD) {
3568 /* Unregister the receive of LACPDUs */
3569 bond_unregister_lacpdu(bond);
3570 }
3571
3572 write_lock_bh(&bond->lock);
3573
3574 bond_mc_list_destroy(bond);
3575
3576 /* signal timers not to re-arm */
3577 bond->kill_timers = 1;
3578
3579 write_unlock_bh(&bond->lock);
3580
3581 /* del_timer_sync must run without holding the bond->lock
3582 * because a running timer might be trying to hold it too
3583 */
3584
3585 if (bond->params.miimon) { /* link check interval, in milliseconds. */
3586 del_timer_sync(&bond->mii_timer);
3587 }
3588
3589 if (bond->params.arp_interval) { /* arp interval, in milliseconds. */
3590 del_timer_sync(&bond->arp_timer);
3591 }
3592
3593 switch (bond->params.mode) {
3594 case BOND_MODE_8023AD:
3595 del_timer_sync(&(BOND_AD_INFO(bond).ad_timer));
3596 break;
3597 case BOND_MODE_TLB:
3598 case BOND_MODE_ALB:
3599 del_timer_sync(&(BOND_ALB_INFO(bond).alb_timer));
3600 break;
3601 default:
3602 break;
3603 }
3604
3605 /* Release the bonded slaves */
3606 bond_release_all(bond_dev);
3607
3608 if ((bond->params.mode == BOND_MODE_TLB) ||
3609 (bond->params.mode == BOND_MODE_ALB)) {
3610 /* Must be called only after all
3611 * slaves have been released
3612 */
3613 bond_alb_deinitialize(bond);
3614 }
3615
3616 return 0;
3617}
3618
3619static struct net_device_stats *bond_get_stats(struct net_device *bond_dev)
3620{
3621 struct bonding *bond = bond_dev->priv;
3622 struct net_device_stats *stats = &(bond->stats), *sstats;
3623 struct slave *slave;
3624 int i;
3625
3626 memset(stats, 0, sizeof(struct net_device_stats));
3627
3628 read_lock_bh(&bond->lock);
3629
3630 bond_for_each_slave(bond, slave, i) {
3631 sstats = slave->dev->get_stats(slave->dev);
3632
3633 stats->rx_packets += sstats->rx_packets;
3634 stats->rx_bytes += sstats->rx_bytes;
3635 stats->rx_errors += sstats->rx_errors;
3636 stats->rx_dropped += sstats->rx_dropped;
3637
3638 stats->tx_packets += sstats->tx_packets;
3639 stats->tx_bytes += sstats->tx_bytes;
3640 stats->tx_errors += sstats->tx_errors;
3641 stats->tx_dropped += sstats->tx_dropped;
3642
3643 stats->multicast += sstats->multicast;
3644 stats->collisions += sstats->collisions;
3645
3646 stats->rx_length_errors += sstats->rx_length_errors;
3647 stats->rx_over_errors += sstats->rx_over_errors;
3648 stats->rx_crc_errors += sstats->rx_crc_errors;
3649 stats->rx_frame_errors += sstats->rx_frame_errors;
3650 stats->rx_fifo_errors += sstats->rx_fifo_errors;
3651 stats->rx_missed_errors += sstats->rx_missed_errors;
3652
3653 stats->tx_aborted_errors += sstats->tx_aborted_errors;
3654 stats->tx_carrier_errors += sstats->tx_carrier_errors;
3655 stats->tx_fifo_errors += sstats->tx_fifo_errors;
3656 stats->tx_heartbeat_errors += sstats->tx_heartbeat_errors;
3657 stats->tx_window_errors += sstats->tx_window_errors;
3658 }
3659
3660 read_unlock_bh(&bond->lock);
3661
3662 return stats;
3663}
3664
3665static int bond_do_ioctl(struct net_device *bond_dev, struct ifreq *ifr, int cmd)
3666{
3667 struct net_device *slave_dev = NULL;
3668 struct ifbond k_binfo;
3669 struct ifbond __user *u_binfo = NULL;
3670 struct ifslave k_sinfo;
3671 struct ifslave __user *u_sinfo = NULL;
3672 struct mii_ioctl_data *mii = NULL;
3673 int prev_abi_ver = orig_app_abi_ver;
3674 int res = 0;
3675
3676 dprintk("bond_ioctl: master=%s, cmd=%d\n",
3677 bond_dev->name, cmd);
3678
3679 switch (cmd) {
3680 case SIOCETHTOOL:
3681 return bond_ethtool_ioctl(bond_dev, ifr);
3682 case SIOCGMIIPHY:
3683 mii = if_mii(ifr);
3684 if (!mii) {
3685 return -EINVAL;
3686 }
3687 mii->phy_id = 0;
3688 /* Fall Through */
3689 case SIOCGMIIREG:
3690 /*
3691 * We do this again just in case we were called by SIOCGMIIREG
3692 * instead of SIOCGMIIPHY.
3693 */
3694 mii = if_mii(ifr);
3695 if (!mii) {
3696 return -EINVAL;
3697 }
3698
3699 if (mii->reg_num == 1) {
3700 struct bonding *bond = bond_dev->priv;
3701 mii->val_out = 0;
3702 read_lock_bh(&bond->lock);
3703 read_lock(&bond->curr_slave_lock);
3704 if (bond->curr_active_slave) {
3705 mii->val_out = BMSR_LSTATUS;
3706 }
3707 read_unlock(&bond->curr_slave_lock);
3708 read_unlock_bh(&bond->lock);
3709 }
3710
3711 return 0;
3712 case BOND_INFO_QUERY_OLD:
3713 case SIOCBONDINFOQUERY:
3714 u_binfo = (struct ifbond __user *)ifr->ifr_data;
3715
3716 if (copy_from_user(&k_binfo, u_binfo, sizeof(ifbond))) {
3717 return -EFAULT;
3718 }
3719
3720 res = bond_info_query(bond_dev, &k_binfo);
3721 if (res == 0) {
3722 if (copy_to_user(u_binfo, &k_binfo, sizeof(ifbond))) {
3723 return -EFAULT;
3724 }
3725 }
3726
3727 return res;
3728 case BOND_SLAVE_INFO_QUERY_OLD:
3729 case SIOCBONDSLAVEINFOQUERY:
3730 u_sinfo = (struct ifslave __user *)ifr->ifr_data;
3731
3732 if (copy_from_user(&k_sinfo, u_sinfo, sizeof(ifslave))) {
3733 return -EFAULT;
3734 }
3735
3736 res = bond_slave_info_query(bond_dev, &k_sinfo);
3737 if (res == 0) {
3738 if (copy_to_user(u_sinfo, &k_sinfo, sizeof(ifslave))) {
3739 return -EFAULT;
3740 }
3741 }
3742
3743 return res;
3744 default:
3745 /* Go on */
3746 break;
3747 }
3748
3749 if (!capable(CAP_NET_ADMIN)) {
3750 return -EPERM;
3751 }
3752
3753 if (orig_app_abi_ver == -1) {
3754 /* no orig_app_abi_ver was provided yet, so we'll use the
3755 * current one from now on, even if it's 0
3756 */
3757 orig_app_abi_ver = app_abi_ver;
3758
3759 } else if (orig_app_abi_ver != app_abi_ver) {
3760 printk(KERN_ERR DRV_NAME
3761 ": Error: already using ifenslave ABI version %d; to "
3762 "upgrade ifenslave to version %d, you must first "
3763 "reload bonding.\n",
3764 orig_app_abi_ver, app_abi_ver);
3765 return -EINVAL;
3766 }
3767
3768 slave_dev = dev_get_by_name(ifr->ifr_slave);
3769
3770 dprintk("slave_dev=%p: \n", slave_dev);
3771
3772 if (!slave_dev) {
3773 res = -ENODEV;
3774 } else {
3775 dprintk("slave_dev->name=%s: \n", slave_dev->name);
3776 switch (cmd) {
3777 case BOND_ENSLAVE_OLD:
3778 case SIOCBONDENSLAVE:
3779 res = bond_enslave(bond_dev, slave_dev);
3780 break;
3781 case BOND_RELEASE_OLD:
3782 case SIOCBONDRELEASE:
3783 res = bond_release(bond_dev, slave_dev);
3784 break;
3785 case BOND_SETHWADDR_OLD:
3786 case SIOCBONDSETHWADDR:
3787 res = bond_sethwaddr(bond_dev, slave_dev);
3788 break;
3789 case BOND_CHANGE_ACTIVE_OLD:
3790 case SIOCBONDCHANGEACTIVE:
3791 res = bond_ioctl_change_active(bond_dev, slave_dev);
3792 break;
3793 default:
3794 res = -EOPNOTSUPP;
3795 }
3796
3797 dev_put(slave_dev);
3798 }
3799
3800 if (res < 0) {
3801 /* The ioctl failed, so there's no point in changing the
3802 * orig_app_abi_ver. We'll restore it's value just in case
3803 * we've changed it earlier in this function.
3804 */
3805 orig_app_abi_ver = prev_abi_ver;
3806 }
3807
3808 return res;
3809}
3810
3811static void bond_set_multicast_list(struct net_device *bond_dev)
3812{
3813 struct bonding *bond = bond_dev->priv;
3814 struct dev_mc_list *dmi;
3815
3816 write_lock_bh(&bond->lock);
3817
3818 /*
3819 * Do promisc before checking multicast_mode
3820 */
3821 if ((bond_dev->flags & IFF_PROMISC) && !(bond->flags & IFF_PROMISC)) {
3822 bond_set_promiscuity(bond, 1);
3823 }
3824
3825 if (!(bond_dev->flags & IFF_PROMISC) && (bond->flags & IFF_PROMISC)) {
3826 bond_set_promiscuity(bond, -1);
3827 }
3828
3829 /* set allmulti flag to slaves */
3830 if ((bond_dev->flags & IFF_ALLMULTI) && !(bond->flags & IFF_ALLMULTI)) {
3831 bond_set_allmulti(bond, 1);
3832 }
3833
3834 if (!(bond_dev->flags & IFF_ALLMULTI) && (bond->flags & IFF_ALLMULTI)) {
3835 bond_set_allmulti(bond, -1);
3836 }
3837
3838 bond->flags = bond_dev->flags;
3839
3840 /* looking for addresses to add to slaves' mc list */
3841 for (dmi = bond_dev->mc_list; dmi; dmi = dmi->next) {
3842 if (!bond_mc_list_find_dmi(dmi, bond->mc_list)) {
3843 bond_mc_add(bond, dmi->dmi_addr, dmi->dmi_addrlen);
3844 }
3845 }
3846
3847 /* looking for addresses to delete from slaves' list */
3848 for (dmi = bond->mc_list; dmi; dmi = dmi->next) {
3849 if (!bond_mc_list_find_dmi(dmi, bond_dev->mc_list)) {
3850 bond_mc_delete(bond, dmi->dmi_addr, dmi->dmi_addrlen);
3851 }
3852 }
3853
3854 /* save master's multicast list */
3855 bond_mc_list_destroy(bond);
3856 bond_mc_list_copy(bond_dev->mc_list, bond, GFP_ATOMIC);
3857
3858 write_unlock_bh(&bond->lock);
3859}
3860
3861/*
3862 * Change the MTU of all of a master's slaves to match the master
3863 */
3864static int bond_change_mtu(struct net_device *bond_dev, int new_mtu)
3865{
3866 struct bonding *bond = bond_dev->priv;
3867 struct slave *slave, *stop_at;
3868 int res = 0;
3869 int i;
3870
3871 dprintk("bond=%p, name=%s, new_mtu=%d\n", bond,
3872 (bond_dev ? bond_dev->name : "None"), new_mtu);
3873
3874 /* Can't hold bond->lock with bh disabled here since
3875 * some base drivers panic. On the other hand we can't
3876 * hold bond->lock without bh disabled because we'll
3877 * deadlock. The only solution is to rely on the fact
3878 * that we're under rtnl_lock here, and the slaves
3879 * list won't change. This doesn't solve the problem
3880 * of setting the slave's MTU while it is
3881 * transmitting, but the assumption is that the base
3882 * driver can handle that.
3883 *
3884 * TODO: figure out a way to safely iterate the slaves
3885 * list, but without holding a lock around the actual
3886 * call to the base driver.
3887 */
3888
3889 bond_for_each_slave(bond, slave, i) {
3890 dprintk("s %p s->p %p c_m %p\n", slave,
3891 slave->prev, slave->dev->change_mtu);
3892 res = dev_set_mtu(slave->dev, new_mtu);
3893
3894 if (res) {
3895 /* If we failed to set the slave's mtu to the new value
3896 * we must abort the operation even in ACTIVE_BACKUP
3897 * mode, because if we allow the backup slaves to have
3898 * different mtu values than the active slave we'll
3899 * need to change their mtu when doing a failover. That
3900 * means changing their mtu from timer context, which
3901 * is probably not a good idea.
3902 */
3903 dprintk("err %d %s\n", res, slave->dev->name);
3904 goto unwind;
3905 }
3906 }
3907
3908 bond_dev->mtu = new_mtu;
3909
3910 return 0;
3911
3912unwind:
3913 /* unwind from head to the slave that failed */
3914 stop_at = slave;
3915 bond_for_each_slave_from_to(bond, slave, i, bond->first_slave, stop_at) {
3916 int tmp_res;
3917
3918 tmp_res = dev_set_mtu(slave->dev, bond_dev->mtu);
3919 if (tmp_res) {
3920 dprintk("unwind err %d dev %s\n", tmp_res,
3921 slave->dev->name);
3922 }
3923 }
3924
3925 return res;
3926}
3927
3928/*
3929 * Change HW address
3930 *
3931 * Note that many devices must be down to change the HW address, and
3932 * downing the master releases all slaves. We can make bonds full of
3933 * bonding devices to test this, however.
3934 */
3935static int bond_set_mac_address(struct net_device *bond_dev, void *addr)
3936{
3937 struct bonding *bond = bond_dev->priv;
3938 struct sockaddr *sa = addr, tmp_sa;
3939 struct slave *slave, *stop_at;
3940 int res = 0;
3941 int i;
3942
3943 dprintk("bond=%p, name=%s\n", bond, (bond_dev ? bond_dev->name : "None"));
3944
3945 if (!is_valid_ether_addr(sa->sa_data)) {
3946 return -EADDRNOTAVAIL;
3947 }
3948
3949 /* Can't hold bond->lock with bh disabled here since
3950 * some base drivers panic. On the other hand we can't
3951 * hold bond->lock without bh disabled because we'll
3952 * deadlock. The only solution is to rely on the fact
3953 * that we're under rtnl_lock here, and the slaves
3954 * list won't change. This doesn't solve the problem
3955 * of setting the slave's hw address while it is
3956 * transmitting, but the assumption is that the base
3957 * driver can handle that.
3958 *
3959 * TODO: figure out a way to safely iterate the slaves
3960 * list, but without holding a lock around the actual
3961 * call to the base driver.
3962 */
3963
3964 bond_for_each_slave(bond, slave, i) {
3965 dprintk("slave %p %s\n", slave, slave->dev->name);
3966
3967 if (slave->dev->set_mac_address == NULL) {
3968 res = -EOPNOTSUPP;
3969 dprintk("EOPNOTSUPP %s\n", slave->dev->name);
3970 goto unwind;
3971 }
3972
3973 res = dev_set_mac_address(slave->dev, addr);
3974 if (res) {
3975 /* TODO: consider downing the slave
3976 * and retry ?
3977 * User should expect communications
3978 * breakage anyway until ARP finish
3979 * updating, so...
3980 */
3981 dprintk("err %d %s\n", res, slave->dev->name);
3982 goto unwind;
3983 }
3984 }
3985
3986 /* success */
3987 memcpy(bond_dev->dev_addr, sa->sa_data, bond_dev->addr_len);
3988 return 0;
3989
3990unwind:
3991 memcpy(tmp_sa.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
3992 tmp_sa.sa_family = bond_dev->type;
3993
3994 /* unwind from head to the slave that failed */
3995 stop_at = slave;
3996 bond_for_each_slave_from_to(bond, slave, i, bond->first_slave, stop_at) {
3997 int tmp_res;
3998
3999 tmp_res = dev_set_mac_address(slave->dev, &tmp_sa);
4000 if (tmp_res) {
4001 dprintk("unwind err %d dev %s\n", tmp_res,
4002 slave->dev->name);
4003 }
4004 }
4005
4006 return res;
4007}
4008
4009static int bond_xmit_roundrobin(struct sk_buff *skb, struct net_device *bond_dev)
4010{
4011 struct bonding *bond = bond_dev->priv;
4012 struct slave *slave, *start_at;
4013 int i;
4014 int res = 1;
4015
4016 read_lock(&bond->lock);
4017
4018 if (!BOND_IS_OK(bond)) {
4019 goto out;
4020 }
4021
4022 read_lock(&bond->curr_slave_lock);
4023 slave = start_at = bond->curr_active_slave;
4024 read_unlock(&bond->curr_slave_lock);
4025
4026 if (!slave) {
4027 goto out;
4028 }
4029
4030 bond_for_each_slave_from(bond, slave, i, start_at) {
4031 if (IS_UP(slave->dev) &&
4032 (slave->link == BOND_LINK_UP) &&
4033 (slave->state == BOND_STATE_ACTIVE)) {
4034 res = bond_dev_queue_xmit(bond, skb, slave->dev);
4035
4036 write_lock(&bond->curr_slave_lock);
4037 bond->curr_active_slave = slave->next;
4038 write_unlock(&bond->curr_slave_lock);
4039
4040 break;
4041 }
4042 }
4043
4044
4045out:
4046 if (res) {
4047 /* no suitable interface, frame not sent */
4048 dev_kfree_skb(skb);
4049 }
4050 read_unlock(&bond->lock);
4051 return 0;
4052}
4053
4054/*
4055 * in active-backup mode, we know that bond->curr_active_slave is always valid if
4056 * the bond has a usable interface.
4057 */
4058static int bond_xmit_activebackup(struct sk_buff *skb, struct net_device *bond_dev)
4059{
4060 struct bonding *bond = bond_dev->priv;
4061 int res = 1;
4062
4063 /* if we are sending arp packets, try to at least
4064 identify our own ip address */
4065 if (bond->params.arp_interval && !my_ip &&
4066 (skb->protocol == __constant_htons(ETH_P_ARP))) {
4067 char *the_ip = (char *)skb->data +
4068 sizeof(struct ethhdr) +
4069 sizeof(struct arphdr) +
4070 ETH_ALEN;
4071 memcpy(&my_ip, the_ip, 4);
4072 }
4073
4074 read_lock(&bond->lock);
4075 read_lock(&bond->curr_slave_lock);
4076
4077 if (!BOND_IS_OK(bond)) {
4078 goto out;
4079 }
4080
4081 if (bond->curr_active_slave) { /* one usable interface */
4082 res = bond_dev_queue_xmit(bond, skb, bond->curr_active_slave->dev);
4083 }
4084
4085out:
4086 if (res) {
4087 /* no suitable interface, frame not sent */
4088 dev_kfree_skb(skb);
4089 }
4090 read_unlock(&bond->curr_slave_lock);
4091 read_unlock(&bond->lock);
4092 return 0;
4093}
4094
4095/*
4096 * in XOR mode, we determine the output device by performing xor on
4097 * the source and destination hw adresses. If this device is not
4098 * enabled, find the next slave following this xor slave.
4099 */
4100static int bond_xmit_xor(struct sk_buff *skb, struct net_device *bond_dev)
4101{
4102 struct bonding *bond = bond_dev->priv;
4103 struct ethhdr *data = (struct ethhdr *)skb->data;
4104 struct slave *slave, *start_at;
4105 int slave_no;
4106 int i;
4107 int res = 1;
4108
4109 read_lock(&bond->lock);
4110
4111 if (!BOND_IS_OK(bond)) {
4112 goto out;
4113 }
4114
4115 slave_no = (data->h_dest[5]^bond_dev->dev_addr[5]) % bond->slave_cnt;
4116
4117 bond_for_each_slave(bond, slave, i) {
4118 slave_no--;
4119 if (slave_no < 0) {
4120 break;
4121 }
4122 }
4123
4124 start_at = slave;
4125
4126 bond_for_each_slave_from(bond, slave, i, start_at) {
4127 if (IS_UP(slave->dev) &&
4128 (slave->link == BOND_LINK_UP) &&
4129 (slave->state == BOND_STATE_ACTIVE)) {
4130 res = bond_dev_queue_xmit(bond, skb, slave->dev);
4131 break;
4132 }
4133 }
4134
4135out:
4136 if (res) {
4137 /* no suitable interface, frame not sent */
4138 dev_kfree_skb(skb);
4139 }
4140 read_unlock(&bond->lock);
4141 return 0;
4142}
4143
4144/*
4145 * in broadcast mode, we send everything to all usable interfaces.
4146 */
4147static int bond_xmit_broadcast(struct sk_buff *skb, struct net_device *bond_dev)
4148{
4149 struct bonding *bond = bond_dev->priv;
4150 struct slave *slave, *start_at;
4151 struct net_device *tx_dev = NULL;
4152 int i;
4153 int res = 1;
4154
4155 read_lock(&bond->lock);
4156
4157 if (!BOND_IS_OK(bond)) {
4158 goto out;
4159 }
4160
4161 read_lock(&bond->curr_slave_lock);
4162 start_at = bond->curr_active_slave;
4163 read_unlock(&bond->curr_slave_lock);
4164
4165 if (!start_at) {
4166 goto out;
4167 }
4168
4169 bond_for_each_slave_from(bond, slave, i, start_at) {
4170 if (IS_UP(slave->dev) &&
4171 (slave->link == BOND_LINK_UP) &&
4172 (slave->state == BOND_STATE_ACTIVE)) {
4173 if (tx_dev) {
4174 struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
4175 if (!skb2) {
4176 printk(KERN_ERR DRV_NAME
4177 ": Error: bond_xmit_broadcast(): "
4178 "skb_clone() failed\n");
4179 continue;
4180 }
4181
4182 res = bond_dev_queue_xmit(bond, skb2, tx_dev);
4183 if (res) {
4184 dev_kfree_skb(skb2);
4185 continue;
4186 }
4187 }
4188 tx_dev = slave->dev;
4189 }
4190 }
4191
4192 if (tx_dev) {
4193 res = bond_dev_queue_xmit(bond, skb, tx_dev);
4194 }
4195
4196out:
4197 if (res) {
4198 /* no suitable interface, frame not sent */
4199 dev_kfree_skb(skb);
4200 }
4201 /* frame sent to all suitable interfaces */
4202 read_unlock(&bond->lock);
4203 return 0;
4204}
4205
4206/*------------------------- Device initialization ---------------------------*/
4207
4208/*
4209 * set bond mode specific net device operations
4210 */
4211static inline void bond_set_mode_ops(struct net_device *bond_dev, int mode)
4212{
4213 switch (mode) {
4214 case BOND_MODE_ROUNDROBIN:
4215 bond_dev->hard_start_xmit = bond_xmit_roundrobin;
4216 break;
4217 case BOND_MODE_ACTIVEBACKUP:
4218 bond_dev->hard_start_xmit = bond_xmit_activebackup;
4219 break;
4220 case BOND_MODE_XOR:
4221 bond_dev->hard_start_xmit = bond_xmit_xor;
4222 break;
4223 case BOND_MODE_BROADCAST:
4224 bond_dev->hard_start_xmit = bond_xmit_broadcast;
4225 break;
4226 case BOND_MODE_8023AD:
4227 bond_dev->hard_start_xmit = bond_3ad_xmit_xor;
4228 break;
4229 case BOND_MODE_TLB:
4230 case BOND_MODE_ALB:
4231 bond_dev->hard_start_xmit = bond_alb_xmit;
4232 bond_dev->set_mac_address = bond_alb_set_mac_address;
4233 break;
4234 default:
4235 /* Should never happen, mode already checked */
4236 printk(KERN_ERR DRV_NAME
4237 ": Error: Unknown bonding mode %d\n",
4238 mode);
4239 break;
4240 }
4241}
4242
4243/*
4244 * Does not allocate but creates a /proc entry.
4245 * Allowed to fail.
4246 */
4247static int __init bond_init(struct net_device *bond_dev, struct bond_params *params)
4248{
4249 struct bonding *bond = bond_dev->priv;
4250
4251 dprintk("Begin bond_init for %s\n", bond_dev->name);
4252
4253 /* initialize rwlocks */
4254 rwlock_init(&bond->lock);
4255 rwlock_init(&bond->curr_slave_lock);
4256
4257 bond->params = *params; /* copy params struct */
4258
4259 /* Initialize pointers */
4260 bond->first_slave = NULL;
4261 bond->curr_active_slave = NULL;
4262 bond->current_arp_slave = NULL;
4263 bond->primary_slave = NULL;
4264 bond->dev = bond_dev;
4265 INIT_LIST_HEAD(&bond->vlan_list);
4266
4267 /* Initialize the device entry points */
4268 bond_dev->open = bond_open;
4269 bond_dev->stop = bond_close;
4270 bond_dev->get_stats = bond_get_stats;
4271 bond_dev->do_ioctl = bond_do_ioctl;
4272 bond_dev->set_multicast_list = bond_set_multicast_list;
4273 bond_dev->change_mtu = bond_change_mtu;
4274 bond_dev->set_mac_address = bond_set_mac_address;
4275
4276 bond_set_mode_ops(bond_dev, bond->params.mode);
4277
4278 bond_dev->destructor = free_netdev;
4279
4280 /* Initialize the device options */
4281 bond_dev->tx_queue_len = 0;
4282 bond_dev->flags |= IFF_MASTER|IFF_MULTICAST;
4283
4284 /* At first, we block adding VLANs. That's the only way to
4285 * prevent problems that occur when adding VLANs over an
4286 * empty bond. The block will be removed once non-challenged
4287 * slaves are enslaved.
4288 */
4289 bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
4290
4291 /* don't acquire bond device's xmit_lock when
4292 * transmitting */
4293 bond_dev->features |= NETIF_F_LLTX;
4294
4295 /* By default, we declare the bond to be fully
4296 * VLAN hardware accelerated capable. Special
4297 * care is taken in the various xmit functions
4298 * when there are slaves that are not hw accel
4299 * capable
4300 */
4301 bond_dev->vlan_rx_register = bond_vlan_rx_register;
4302 bond_dev->vlan_rx_add_vid = bond_vlan_rx_add_vid;
4303 bond_dev->vlan_rx_kill_vid = bond_vlan_rx_kill_vid;
4304 bond_dev->features |= (NETIF_F_HW_VLAN_TX |
4305 NETIF_F_HW_VLAN_RX |
4306 NETIF_F_HW_VLAN_FILTER);
4307
4308#ifdef CONFIG_PROC_FS
4309 bond_create_proc_entry(bond);
4310#endif
4311
4312 list_add_tail(&bond->bond_list, &bond_dev_list);
4313
4314 return 0;
4315}
4316
4317/* De-initialize device specific data.
4318 * Caller must hold rtnl_lock.
4319 */
4320static inline void bond_deinit(struct net_device *bond_dev)
4321{
4322 struct bonding *bond = bond_dev->priv;
4323
4324 list_del(&bond->bond_list);
4325
4326#ifdef CONFIG_PROC_FS
4327 bond_remove_proc_entry(bond);
4328#endif
4329}
4330
4331/* Unregister and free all bond devices.
4332 * Caller must hold rtnl_lock.
4333 */
4334static void bond_free_all(void)
4335{
4336 struct bonding *bond, *nxt;
4337
4338 list_for_each_entry_safe(bond, nxt, &bond_dev_list, bond_list) {
4339 struct net_device *bond_dev = bond->dev;
4340
4341 unregister_netdevice(bond_dev);
4342 bond_deinit(bond_dev);
4343 }
4344
4345#ifdef CONFIG_PROC_FS
4346 bond_destroy_proc_dir();
4347#endif
4348}
4349
4350/*------------------------- Module initialization ---------------------------*/
4351
4352/*
4353 * Convert string input module parms. Accept either the
4354 * number of the mode or its string name.
4355 */
4356static inline int bond_parse_parm(char *mode_arg, struct bond_parm_tbl *tbl)
4357{
4358 int i;
4359
4360 for (i = 0; tbl[i].modename; i++) {
4361 if ((isdigit(*mode_arg) &&
4362 tbl[i].mode == simple_strtol(mode_arg, NULL, 0)) ||
4363 (strncmp(mode_arg, tbl[i].modename,
4364 strlen(tbl[i].modename)) == 0)) {
4365 return tbl[i].mode;
4366 }
4367 }
4368
4369 return -1;
4370}
4371
4372static int bond_check_params(struct bond_params *params)
4373{
4374 /*
4375 * Convert string parameters.
4376 */
4377 if (mode) {
4378 bond_mode = bond_parse_parm(mode, bond_mode_tbl);
4379 if (bond_mode == -1) {
4380 printk(KERN_ERR DRV_NAME
4381 ": Error: Invalid bonding mode \"%s\"\n",
4382 mode == NULL ? "NULL" : mode);
4383 return -EINVAL;
4384 }
4385 }
4386
4387 if (lacp_rate) {
4388 if (bond_mode != BOND_MODE_8023AD) {
4389 printk(KERN_INFO DRV_NAME
4390 ": lacp_rate param is irrelevant in mode %s\n",
4391 bond_mode_name(bond_mode));
4392 } else {
4393 lacp_fast = bond_parse_parm(lacp_rate, bond_lacp_tbl);
4394 if (lacp_fast == -1) {
4395 printk(KERN_ERR DRV_NAME
4396 ": Error: Invalid lacp rate \"%s\"\n",
4397 lacp_rate == NULL ? "NULL" : lacp_rate);
4398 return -EINVAL;
4399 }
4400 }
4401 }
4402
4403 if (max_bonds < 1 || max_bonds > INT_MAX) {
4404 printk(KERN_WARNING DRV_NAME
4405 ": Warning: max_bonds (%d) not in range %d-%d, so it "
4406 "was reset to BOND_DEFAULT_MAX_BONDS (%d)",
4407 max_bonds, 1, INT_MAX, BOND_DEFAULT_MAX_BONDS);
4408 max_bonds = BOND_DEFAULT_MAX_BONDS;
4409 }
4410
4411 if (miimon < 0) {
4412 printk(KERN_WARNING DRV_NAME
4413 ": Warning: miimon module parameter (%d), "
4414 "not in range 0-%d, so it was reset to %d\n",
4415 miimon, INT_MAX, BOND_LINK_MON_INTERV);
4416 miimon = BOND_LINK_MON_INTERV;
4417 }
4418
4419 if (updelay < 0) {
4420 printk(KERN_WARNING DRV_NAME
4421 ": Warning: updelay module parameter (%d), "
4422 "not in range 0-%d, so it was reset to 0\n",
4423 updelay, INT_MAX);
4424 updelay = 0;
4425 }
4426
4427 if (downdelay < 0) {
4428 printk(KERN_WARNING DRV_NAME
4429 ": Warning: downdelay module parameter (%d), "
4430 "not in range 0-%d, so it was reset to 0\n",
4431 downdelay, INT_MAX);
4432 downdelay = 0;
4433 }
4434
4435 if ((use_carrier != 0) && (use_carrier != 1)) {
4436 printk(KERN_WARNING DRV_NAME
4437 ": Warning: use_carrier module parameter (%d), "
4438 "not of valid value (0/1), so it was set to 1\n",
4439 use_carrier);
4440 use_carrier = 1;
4441 }
4442
4443 /* reset values for 802.3ad */
4444 if (bond_mode == BOND_MODE_8023AD) {
4445 if (!miimon) {
4446 printk(KERN_WARNING DRV_NAME
4447 ": Warning: miimon must be specified, "
4448 "otherwise bonding will not detect link "
4449 "failure, speed and duplex which are "
4450 "essential for 802.3ad operation\n");
4451 printk(KERN_WARNING "Forcing miimon to 100msec\n");
4452 miimon = 100;
4453 }
4454 }
4455
4456 /* reset values for TLB/ALB */
4457 if ((bond_mode == BOND_MODE_TLB) ||
4458 (bond_mode == BOND_MODE_ALB)) {
4459 if (!miimon) {
4460 printk(KERN_WARNING DRV_NAME
4461 ": Warning: miimon must be specified, "
4462 "otherwise bonding will not detect link "
4463 "failure and link speed which are essential "
4464 "for TLB/ALB load balancing\n");
4465 printk(KERN_WARNING "Forcing miimon to 100msec\n");
4466 miimon = 100;
4467 }
4468 }
4469
4470 if (bond_mode == BOND_MODE_ALB) {
4471 printk(KERN_NOTICE DRV_NAME
4472 ": In ALB mode you might experience client "
4473 "disconnections upon reconnection of a link if the "
4474 "bonding module updelay parameter (%d msec) is "
4475 "incompatible with the forwarding delay time of the "
4476 "switch\n",
4477 updelay);
4478 }
4479
4480 if (!miimon) {
4481 if (updelay || downdelay) {
4482 /* just warn the user the up/down delay will have
4483 * no effect since miimon is zero...
4484 */
4485 printk(KERN_WARNING DRV_NAME
4486 ": Warning: miimon module parameter not set "
4487 "and updelay (%d) or downdelay (%d) module "
4488 "parameter is set; updelay and downdelay have "
4489 "no effect unless miimon is set\n",
4490 updelay, downdelay);
4491 }
4492 } else {
4493 /* don't allow arp monitoring */
4494 if (arp_interval) {
4495 printk(KERN_WARNING DRV_NAME
4496 ": Warning: miimon (%d) and arp_interval (%d) "
4497 "can't be used simultaneously, disabling ARP "
4498 "monitoring\n",
4499 miimon, arp_interval);
4500 arp_interval = 0;
4501 }
4502
4503 if ((updelay % miimon) != 0) {
4504 printk(KERN_WARNING DRV_NAME
4505 ": Warning: updelay (%d) is not a multiple "
4506 "of miimon (%d), updelay rounded to %d ms\n",
4507 updelay, miimon, (updelay / miimon) * miimon);
4508 }
4509
4510 updelay /= miimon;
4511
4512 if ((downdelay % miimon) != 0) {
4513 printk(KERN_WARNING DRV_NAME
4514 ": Warning: downdelay (%d) is not a multiple "
4515 "of miimon (%d), downdelay rounded to %d ms\n",
4516 downdelay, miimon,
4517 (downdelay / miimon) * miimon);
4518 }
4519
4520 downdelay /= miimon;
4521 }
4522
4523 if (arp_interval < 0) {
4524 printk(KERN_WARNING DRV_NAME
4525 ": Warning: arp_interval module parameter (%d) "
4526 ", not in range 0-%d, so it was reset to %d\n",
4527 arp_interval, INT_MAX, BOND_LINK_ARP_INTERV);
4528 arp_interval = BOND_LINK_ARP_INTERV;
4529 }
4530
4531 for (arp_ip_count = 0;
4532 (arp_ip_count < BOND_MAX_ARP_TARGETS) && arp_ip_target[arp_ip_count];
4533 arp_ip_count++) {
4534 /* not complete check, but should be good enough to
4535 catch mistakes */
4536 if (!isdigit(arp_ip_target[arp_ip_count][0])) {
4537 printk(KERN_WARNING DRV_NAME
4538 ": Warning: bad arp_ip_target module parameter "
4539 "(%s), ARP monitoring will not be performed\n",
4540 arp_ip_target[arp_ip_count]);
4541 arp_interval = 0;
4542 } else {
4543 u32 ip = in_aton(arp_ip_target[arp_ip_count]);
4544 arp_target[arp_ip_count] = ip;
4545 }
4546 }
4547
4548 if (arp_interval && !arp_ip_count) {
4549 /* don't allow arping if no arp_ip_target given... */
4550 printk(KERN_WARNING DRV_NAME
4551 ": Warning: arp_interval module parameter (%d) "
4552 "specified without providing an arp_ip_target "
4553 "parameter, arp_interval was reset to 0\n",
4554 arp_interval);
4555 arp_interval = 0;
4556 }
4557
4558 if (miimon) {
4559 printk(KERN_INFO DRV_NAME
4560 ": MII link monitoring set to %d ms\n",
4561 miimon);
4562 } else if (arp_interval) {
4563 int i;
4564
4565 printk(KERN_INFO DRV_NAME
4566 ": ARP monitoring set to %d ms with %d target(s):",
4567 arp_interval, arp_ip_count);
4568
4569 for (i = 0; i < arp_ip_count; i++)
4570 printk (" %s", arp_ip_target[i]);
4571
4572 printk("\n");
4573
4574 } else {
4575 /* miimon and arp_interval not set, we need one so things
4576 * work as expected, see bonding.txt for details
4577 */
4578 printk(KERN_WARNING DRV_NAME
4579 ": Warning: either miimon or arp_interval and "
4580 "arp_ip_target module parameters must be specified, "
4581 "otherwise bonding will not detect link failures! see "
4582 "bonding.txt for details.\n");
4583 }
4584
4585 if (primary && !USES_PRIMARY(bond_mode)) {
4586 /* currently, using a primary only makes sense
4587 * in active backup, TLB or ALB modes
4588 */
4589 printk(KERN_WARNING DRV_NAME
4590 ": Warning: %s primary device specified but has no "
4591 "effect in %s mode\n",
4592 primary, bond_mode_name(bond_mode));
4593 primary = NULL;
4594 }
4595
4596 /* fill params struct with the proper values */
4597 params->mode = bond_mode;
4598 params->miimon = miimon;
4599 params->arp_interval = arp_interval;
4600 params->updelay = updelay;
4601 params->downdelay = downdelay;
4602 params->use_carrier = use_carrier;
4603 params->lacp_fast = lacp_fast;
4604 params->primary[0] = 0;
4605
4606 if (primary) {
4607 strncpy(params->primary, primary, IFNAMSIZ);
4608 params->primary[IFNAMSIZ - 1] = 0;
4609 }
4610
4611 memcpy(params->arp_targets, arp_target, sizeof(arp_target));
4612
4613 return 0;
4614}
4615
4616static int __init bonding_init(void)
4617{
4618 struct bond_params params;
4619 int i;
4620 int res;
4621
4622 printk(KERN_INFO "%s", version);
4623
4624 res = bond_check_params(&params);
4625 if (res) {
4626 return res;
4627 }
4628
4629 rtnl_lock();
4630
4631#ifdef CONFIG_PROC_FS
4632 bond_create_proc_dir();
4633#endif
4634
4635 for (i = 0; i < max_bonds; i++) {
4636 struct net_device *bond_dev;
4637
4638 bond_dev = alloc_netdev(sizeof(struct bonding), "", ether_setup);
4639 if (!bond_dev) {
4640 res = -ENOMEM;
4641 goto out_err;
4642 }
4643
4644 res = dev_alloc_name(bond_dev, "bond%d");
4645 if (res < 0) {
4646 free_netdev(bond_dev);
4647 goto out_err;
4648 }
4649
4650 /* bond_init() must be called after dev_alloc_name() (for the
4651 * /proc files), but before register_netdevice(), because we
4652 * need to set function pointers.
4653 */
4654 res = bond_init(bond_dev, &params);
4655 if (res < 0) {
4656 free_netdev(bond_dev);
4657 goto out_err;
4658 }
4659
4660 SET_MODULE_OWNER(bond_dev);
4661
4662 res = register_netdevice(bond_dev);
4663 if (res < 0) {
4664 bond_deinit(bond_dev);
4665 free_netdev(bond_dev);
4666 goto out_err;
4667 }
4668 }
4669
4670 rtnl_unlock();
4671 register_netdevice_notifier(&bond_netdev_notifier);
4672
4673 return 0;
4674
4675out_err:
4676 /* free and unregister all bonds that were successfully added */
4677 bond_free_all();
4678
4679 rtnl_unlock();
4680
4681 return res;
4682}
4683
4684static void __exit bonding_exit(void)
4685{
4686 unregister_netdevice_notifier(&bond_netdev_notifier);
4687
4688 rtnl_lock();
4689 bond_free_all();
4690 rtnl_unlock();
4691}
4692
4693module_init(bonding_init);
4694module_exit(bonding_exit);
4695MODULE_LICENSE("GPL");
4696MODULE_VERSION(DRV_VERSION);
4697MODULE_DESCRIPTION(DRV_DESCRIPTION ", v" DRV_VERSION);
4698MODULE_AUTHOR("Thomas Davis, tadavis@lbl.gov and many others");
4699MODULE_SUPPORTED_DEVICE("most ethernet devices");
4700
4701/*
4702 * Local variables:
4703 * c-indent-level: 8
4704 * c-basic-offset: 8
4705 * tab-width: 8
4706 * End:
4707 */
4708
diff --git a/drivers/net/bonding/bonding.h b/drivers/net/bonding/bonding.h
new file mode 100644
index 000000000000..8c325308489d
--- /dev/null
+++ b/drivers/net/bonding/bonding.h
@@ -0,0 +1,252 @@
1/*
2 * Bond several ethernet interfaces into a Cisco, running 'Etherchannel'.
3 *
4 * Portions are (c) Copyright 1995 Simon "Guru Aleph-Null" Janes
5 * NCM: Network and Communications Management, Inc.
6 *
7 * BUT, I'm the one who modified it for ethernet, so:
8 * (c) Copyright 1999, Thomas Davis, tadavis@lbl.gov
9 *
10 * This software may be used and distributed according to the terms
11 * of the GNU Public License, incorporated herein by reference.
12 *
13 *
14 * 2003/03/18 - Amir Noam <amir.noam at intel dot com>,
15 * Tsippy Mendelson <tsippy.mendelson at intel dot com> and
16 * Shmulik Hen <shmulik.hen at intel dot com>
17 * - Added support for IEEE 802.3ad Dynamic link aggregation mode.
18 *
19 * 2003/05/01 - Tsippy Mendelson <tsippy.mendelson at intel dot com> and
20 * Amir Noam <amir.noam at intel dot com>
21 * - Code beautification and style changes (mainly in comments).
22 *
23 * 2003/05/01 - Shmulik Hen <shmulik.hen at intel dot com>
24 * - Added support for Transmit load balancing mode.
25 *
26 * 2003/12/01 - Shmulik Hen <shmulik.hen at intel dot com>
27 * - Code cleanup and style changes
28 */
29
30#ifndef _LINUX_BONDING_H
31#define _LINUX_BONDING_H
32
33#include <linux/timer.h>
34#include <linux/proc_fs.h>
35#include <linux/if_bonding.h>
36#include "bond_3ad.h"
37#include "bond_alb.h"
38
39#define DRV_VERSION "2.6.1"
40#define DRV_RELDATE "October 29, 2004"
41#define DRV_NAME "bonding"
42#define DRV_DESCRIPTION "Ethernet Channel Bonding Driver"
43
44#define BOND_MAX_ARP_TARGETS 16
45
46#ifdef BONDING_DEBUG
47#define dprintk(fmt, args...) \
48 printk(KERN_DEBUG \
49 DRV_NAME ": %s() %d: " fmt, __FUNCTION__, __LINE__ , ## args )
50#else
51#define dprintk(fmt, args...)
52#endif /* BONDING_DEBUG */
53
54#define IS_UP(dev) \
55 ((((dev)->flags & IFF_UP) == IFF_UP) && \
56 netif_running(dev) && \
57 netif_carrier_ok(dev))
58
59/*
60 * Checks whether bond is ready for transmit.
61 *
62 * Caller must hold bond->lock
63 */
64#define BOND_IS_OK(bond) \
65 (((bond)->dev->flags & IFF_UP) && \
66 netif_running((bond)->dev) && \
67 ((bond)->slave_cnt > 0))
68
69/*
70 * Checks whether slave is ready for transmit.
71 */
72#define SLAVE_IS_OK(slave) \
73 (((slave)->dev->flags & IFF_UP) && \
74 netif_running((slave)->dev) && \
75 ((slave)->link == BOND_LINK_UP) && \
76 ((slave)->state == BOND_STATE_ACTIVE))
77
78
79#define USES_PRIMARY(mode) \
80 (((mode) == BOND_MODE_ACTIVEBACKUP) || \
81 ((mode) == BOND_MODE_TLB) || \
82 ((mode) == BOND_MODE_ALB))
83
84/*
85 * Less bad way to call ioctl from within the kernel; this needs to be
86 * done some other way to get the call out of interrupt context.
87 * Needs "ioctl" variable to be supplied by calling context.
88 */
89#define IOCTL(dev, arg, cmd) ({ \
90 int res = 0; \
91 mm_segment_t fs = get_fs(); \
92 set_fs(get_ds()); \
93 res = ioctl(dev, arg, cmd); \
94 set_fs(fs); \
95 res; })
96
97/**
98 * bond_for_each_slave_from - iterate the slaves list from a starting point
99 * @bond: the bond holding this list.
100 * @pos: current slave.
101 * @cnt: counter for max number of moves
102 * @start: starting point.
103 *
104 * Caller must hold bond->lock
105 */
106#define bond_for_each_slave_from(bond, pos, cnt, start) \
107 for (cnt = 0, pos = start; \
108 cnt < (bond)->slave_cnt; \
109 cnt++, pos = (pos)->next)
110
111/**
112 * bond_for_each_slave_from_to - iterate the slaves list from start point to stop point
113 * @bond: the bond holding this list.
114 * @pos: current slave.
115 * @cnt: counter for number max of moves
116 * @start: start point.
117 * @stop: stop point.
118 *
119 * Caller must hold bond->lock
120 */
121#define bond_for_each_slave_from_to(bond, pos, cnt, start, stop) \
122 for (cnt = 0, pos = start; \
123 ((cnt < (bond)->slave_cnt) && (pos != (stop)->next)); \
124 cnt++, pos = (pos)->next)
125
126/**
127 * bond_for_each_slave - iterate the slaves list from head
128 * @bond: the bond holding this list.
129 * @pos: current slave.
130 * @cnt: counter for max number of moves
131 *
132 * Caller must hold bond->lock
133 */
134#define bond_for_each_slave(bond, pos, cnt) \
135 bond_for_each_slave_from(bond, pos, cnt, (bond)->first_slave)
136
137
138struct bond_params {
139 int mode;
140 int miimon;
141 int arp_interval;
142 int use_carrier;
143 int updelay;
144 int downdelay;
145 int lacp_fast;
146 char primary[IFNAMSIZ];
147 u32 arp_targets[BOND_MAX_ARP_TARGETS];
148};
149
150struct vlan_entry {
151 struct list_head vlan_list;
152 unsigned short vlan_id;
153};
154
155struct slave {
156 struct net_device *dev; /* first - usefull for panic debug */
157 struct slave *next;
158 struct slave *prev;
159 s16 delay;
160 u32 jiffies;
161 s8 link; /* one of BOND_LINK_XXXX */
162 s8 state; /* one of BOND_STATE_XXXX */
163 u32 original_flags;
164 u32 link_failure_count;
165 u16 speed;
166 u8 duplex;
167 u8 perm_hwaddr[ETH_ALEN];
168 struct ad_slave_info ad_info; /* HUGE - better to dynamically alloc */
169 struct tlb_slave_info tlb_info;
170};
171
172/*
173 * Here are the locking policies for the two bonding locks:
174 *
175 * 1) Get bond->lock when reading/writing slave list.
176 * 2) Get bond->curr_slave_lock when reading/writing bond->curr_active_slave.
177 * (It is unnecessary when the write-lock is put with bond->lock.)
178 * 3) When we lock with bond->curr_slave_lock, we must lock with bond->lock
179 * beforehand.
180 */
181struct bonding {
182 struct net_device *dev; /* first - usefull for panic debug */
183 struct slave *first_slave;
184 struct slave *curr_active_slave;
185 struct slave *current_arp_slave;
186 struct slave *primary_slave;
187 s32 slave_cnt; /* never change this value outside the attach/detach wrappers */
188 rwlock_t lock;
189 rwlock_t curr_slave_lock;
190 struct timer_list mii_timer;
191 struct timer_list arp_timer;
192 s8 kill_timers;
193 struct net_device_stats stats;
194#ifdef CONFIG_PROC_FS
195 struct proc_dir_entry *proc_entry;
196 char proc_file_name[IFNAMSIZ];
197#endif /* CONFIG_PROC_FS */
198 struct list_head bond_list;
199 struct dev_mc_list *mc_list;
200 u16 flags;
201 struct ad_bond_info ad_info;
202 struct alb_bond_info alb_info;
203 struct bond_params params;
204 struct list_head vlan_list;
205 struct vlan_group *vlgrp;
206};
207
208/**
209 * Returns NULL if the net_device does not belong to any of the bond's slaves
210 *
211 * Caller must hold bond lock for read
212 */
213extern inline struct slave *bond_get_slave_by_dev(struct bonding *bond, struct net_device *slave_dev)
214{
215 struct slave *slave = NULL;
216 int i;
217
218 bond_for_each_slave(bond, slave, i) {
219 if (slave->dev == slave_dev) {
220 break;
221 }
222 }
223
224 return slave;
225}
226
227extern inline struct bonding *bond_get_bond_by_slave(struct slave *slave)
228{
229 if (!slave || !slave->dev->master) {
230 return NULL;
231 }
232
233 return (struct bonding *)slave->dev->master->priv;
234}
235
236extern inline void bond_set_slave_inactive_flags(struct slave *slave)
237{
238 slave->state = BOND_STATE_BACKUP;
239 slave->dev->flags |= IFF_NOARP;
240}
241
242extern inline void bond_set_slave_active_flags(struct slave *slave)
243{
244 slave->state = BOND_STATE_ACTIVE;
245 slave->dev->flags &= ~IFF_NOARP;
246}
247
248struct vlan_entry *bond_next_vlan(struct bonding *bond, struct vlan_entry *curr);
249int bond_dev_queue_xmit(struct bonding *bond, struct sk_buff *skb, struct net_device *slave_dev);
250
251#endif /* _LINUX_BONDING_H */
252
generated by cgit v1.2.2 (git 2.25.0) at 2024-11-02 13:30:39 -0400