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 current_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 (GPF_KERNEL to GPF_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 current_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 current_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 * - In conjunction with fix for ifenslave -c, in
386 * bond_change_active(), changing to the already active slave
387 * is no longer an error (it successfully does nothing).
388 *
389 * 2003/06/30 - Amir Noam <amir.noam at intel dot com>
390 * - Fixed bond_change_active() for ALB/TLB modes.
391 */
392
393 #include <linux/config.h>
394 #include <linux/kernel.h>
395 #include <linux/module.h>
396 #include <linux/sched.h>
397 #include <linux/types.h>
398 #include <linux/fcntl.h>
399 #include <linux/interrupt.h>
400 #include <linux/ptrace.h>
401 #include <linux/ioport.h>
402 #include <linux/in.h>
403 #include <linux/ip.h>
404 #include <linux/slab.h>
405 #include <linux/string.h>
406 #include <linux/init.h>
407 #include <linux/timer.h>
408 #include <linux/socket.h>
409 #include <linux/ctype.h>
410 #include <linux/inet.h>
411 #include <asm/system.h>
412 #include <asm/bitops.h>
413 #include <asm/io.h>
414 #include <asm/dma.h>
415 #include <asm/uaccess.h>
416 #include <linux/errno.h>
417
418 #include <linux/netdevice.h>
419 #include <linux/inetdevice.h>
420 #include <linux/etherdevice.h>
421 #include <linux/skbuff.h>
422 #include <net/sock.h>
423 #include <linux/rtnetlink.h>
424
425 #include <linux/if_bonding.h>
426 #include <linux/smp.h>
427 #include <linux/if_ether.h>
428 #include <net/arp.h>
429 #include <linux/mii.h>
430 #include <linux/ethtool.h>
431 #include "bonding.h"
432 #include "bond_3ad.h"
433 #include "bond_alb.h"
434
435 #define DRV_VERSION "2.2.14"
436 #define DRV_RELDATE "June 30, 2003"
437 #define DRV_NAME "bonding"
438 #define DRV_DESCRIPTION "Ethernet Channel Bonding Driver"
439
440 static const char *version =
441 DRV_NAME ".c:v" DRV_VERSION " (" DRV_RELDATE ")\n";
442
443 /* monitor all links that often (in milliseconds). <=0 disables monitoring */
444 #ifndef BOND_LINK_MON_INTERV
445 #define BOND_LINK_MON_INTERV 0
446 #endif
447
448 #ifndef BOND_LINK_ARP_INTERV
449 #define BOND_LINK_ARP_INTERV 0
450 #endif
451
452 #ifndef MAX_ARP_IP_TARGETS
453 #define MAX_ARP_IP_TARGETS 16
454 #endif
455
456 struct bond_parm_tbl {
457 char *modename;
458 int mode;
459 };
460
461 static int arp_interval = BOND_LINK_ARP_INTERV;
462 static char *arp_ip_target[MAX_ARP_IP_TARGETS] = { NULL, };
463 static u32 arp_target[MAX_ARP_IP_TARGETS] = { 0, } ;
464 static int arp_ip_count = 0;
465 static u32 my_ip = 0;
466 char *arp_target_hw_addr = NULL;
467
468 static char *primary= NULL;
469
470 static int app_abi_ver = 0;
471 static int orig_app_abi_ver = -1; /* This is used to save the first ABI version
472 * we receive from the application. Once set,
473 * it won't be changed, and the module will
474 * refuse to enslave/release interfaces if the
475 * command comes from an application using
476 * another ABI version.
477 */
478
479 static int max_bonds = BOND_DEFAULT_MAX_BONDS;
480 static int miimon = BOND_LINK_MON_INTERV;
481 static int use_carrier = 1;
482 static int bond_mode = BOND_MODE_ROUNDROBIN;
483 static int updelay = 0;
484 static int downdelay = 0;
485
486 static char *mode = NULL;
487
488 static struct bond_parm_tbl bond_mode_tbl[] = {
489 { "balance-rr", BOND_MODE_ROUNDROBIN},
490 { "active-backup", BOND_MODE_ACTIVEBACKUP},
491 { "balance-xor", BOND_MODE_XOR},
492 { "broadcast", BOND_MODE_BROADCAST},
493 { "802.3ad", BOND_MODE_8023AD},
494 { "balance-tlb", BOND_MODE_TLB},
495 { "balance-alb", BOND_MODE_ALB},
496 { NULL, -1},
497 };
498
499 static int multicast_mode = BOND_MULTICAST_ALL;
500 static char *multicast = NULL;
501
502 static struct bond_parm_tbl bond_mc_tbl[] = {
503 { "disabled", BOND_MULTICAST_DISABLED},
504 { "active", BOND_MULTICAST_ACTIVE},
505 { "all", BOND_MULTICAST_ALL},
506 { NULL, -1},
507 };
508
509 static int lacp_fast = 0;
510 static char *lacp_rate = NULL;
511
512 static struct bond_parm_tbl bond_lacp_tbl[] = {
513 { "slow", AD_LACP_SLOW},
514 { "fast", AD_LACP_FAST},
515 { NULL, -1},
516 };
517
518 static int first_pass = 1;
519 static struct bonding *these_bonds = NULL;
520 static struct net_device *dev_bonds = NULL;
521
522 MODULE_PARM(max_bonds, "i");
523 MODULE_PARM_DESC(max_bonds, "Max number of bonded devices");
524 MODULE_PARM(miimon, "i");
525 MODULE_PARM_DESC(miimon, "Link check interval in milliseconds");
526 MODULE_PARM(use_carrier, "i");
527 MODULE_PARM_DESC(use_carrier, "Use netif_carrier_ok (vs MII ioctls) in miimon; 0 for off, 1 for on (default)");
528 MODULE_PARM(mode, "s");
529 MODULE_PARM_DESC(mode, "Mode of operation : 0 for round robin, 1 for active-backup, 2 for xor");
530 MODULE_PARM(arp_interval, "i");
531 MODULE_PARM_DESC(arp_interval, "arp interval in milliseconds");
532 MODULE_PARM(arp_ip_target, "1-" __MODULE_STRING(MAX_ARP_IP_TARGETS) "s");
533 MODULE_PARM_DESC(arp_ip_target, "arp targets in n.n.n.n form");
534 MODULE_PARM(updelay, "i");
535 MODULE_PARM_DESC(updelay, "Delay before considering link up, in milliseconds");
536 MODULE_PARM(downdelay, "i");
537 MODULE_PARM_DESC(downdelay, "Delay before considering link down, in milliseconds");
538 MODULE_PARM(primary, "s");
539 MODULE_PARM_DESC(primary, "Primary network device to use");
540 MODULE_PARM(multicast, "s");
541 MODULE_PARM_DESC(multicast, "Mode for multicast support : 0 for none, 1 for active slave, 2 for all slaves (default)");
542 MODULE_PARM(lacp_rate, "s");
543 MODULE_PARM_DESC(lacp_rate, "LACPDU tx rate to request from 802.3ad partner (slow/fast)");
544
545 static int bond_xmit_roundrobin(struct sk_buff *skb, struct net_device *dev);
546 static int bond_xmit_xor(struct sk_buff *skb, struct net_device *dev);
547 static int bond_xmit_activebackup(struct sk_buff *skb, struct net_device *dev);
548 static struct net_device_stats *bond_get_stats(struct net_device *dev);
549 static void bond_mii_monitor(struct net_device *dev);
550 static void loadbalance_arp_monitor(struct net_device *dev);
551 static void activebackup_arp_monitor(struct net_device *dev);
552 static int bond_event(struct notifier_block *this, unsigned long event, void *ptr);
553 static void bond_mc_list_destroy(struct bonding *bond);
554 static void bond_mc_add(bonding_t *bond, void *addr, int alen);
555 static void bond_mc_delete(bonding_t *bond, void *addr, int alen);
556 static int bond_mc_list_copy (struct dev_mc_list *src, struct bonding *dst, int gpf_flag);
557 static inline int dmi_same(struct dev_mc_list *dmi1, struct dev_mc_list *dmi2);
558 static void bond_set_promiscuity(bonding_t *bond, int inc);
559 static void bond_set_allmulti(bonding_t *bond, int inc);
560 static struct dev_mc_list* bond_mc_list_find_dmi(struct dev_mc_list *dmi, struct dev_mc_list *mc_list);
561 static void bond_mc_update(bonding_t *bond, slave_t *new, slave_t *old);
562 static int bond_enslave(struct net_device *master, struct net_device *slave);
563 static int bond_release(struct net_device *master, struct net_device *slave);
564 static int bond_release_all(struct net_device *master);
565 static int bond_sethwaddr(struct net_device *master, struct net_device *slave);
566
567 /*
568 * bond_get_info is the interface into the /proc filesystem. This is
569 * a different interface than the BOND_INFO_QUERY ioctl. That is done
570 * through the generic networking ioctl interface, and bond_info_query
571 * is the internal function which provides that information.
572 */
573 static int bond_get_info(char *buf, char **start, off_t offset, int length);
574
575 /* Caller must hold bond->ptrlock for write */
576 static inline struct slave*
577 bond_assign_current_slave(struct bonding *bond,struct slave *newslave)
578 {
579 if ((bond_mode == BOND_MODE_TLB) ||
580 (bond_mode == BOND_MODE_ALB)) {
581 bond_alb_assign_current_slave(bond, newslave);
582 } else {
583 bond->current_slave = newslave;
584 }
585
586 return bond->current_slave;
587 }
588
589 /* #define BONDING_DEBUG 1 */
590
591 /* several macros */
592
593 static void arp_send_all(slave_t *slave)
594 {
595 int i;
596
597 for (i = 0; (i<MAX_ARP_IP_TARGETS) && arp_target[i]; i++) {
598 arp_send(ARPOP_REQUEST, ETH_P_ARP, arp_target[i], slave->dev,
599 my_ip, arp_target_hw_addr, slave->dev->dev_addr,
600 arp_target_hw_addr);
601 }
602 }
603
604
605 static const char *
606 bond_mode_name(void)
607 {
608 switch (bond_mode) {
609 case BOND_MODE_ROUNDROBIN :
610 return "load balancing (round-robin)";
611 case BOND_MODE_ACTIVEBACKUP :
612 return "fault-tolerance (active-backup)";
613 case BOND_MODE_XOR :
614 return "load balancing (xor)";
615 case BOND_MODE_BROADCAST :
616 return "fault-tolerance (broadcast)";
617 case BOND_MODE_8023AD:
618 return "IEEE 802.3ad Dynamic link aggregation";
619 case BOND_MODE_TLB:
620 return "transmit load balancing";
621 case BOND_MODE_ALB:
622 return "adaptive load balancing";
623 default:
624 return "unknown";
625 }
626 }
627
628 static const char *
629 multicast_mode_name(void)
630 {
631 switch(multicast_mode) {
632 case BOND_MULTICAST_DISABLED :
633 return "disabled";
634 case BOND_MULTICAST_ACTIVE :
635 return "active slave only";
636 case BOND_MULTICAST_ALL :
637 return "all slaves";
638 default :
639 return "unknown";
640 }
641 }
642
643 void bond_set_slave_inactive_flags(slave_t *slave)
644 {
645 slave->state = BOND_STATE_BACKUP;
646 slave->dev->flags |= IFF_NOARP;
647 }
648
649 void bond_set_slave_active_flags(slave_t *slave)
650 {
651 slave->state = BOND_STATE_ACTIVE;
652 slave->dev->flags &= ~IFF_NOARP;
653 }
654
655 /*
656 * This function counts and verifies the the number of attached
657 * slaves, checking the count against the expected value (given that incr
658 * is either 1 or -1, for add or removal of a slave). Only
659 * bond_xmit_xor() uses the slave_cnt value, but this is still a good
660 * consistency check.
661 */
662 static inline void
663 update_slave_cnt(bonding_t *bond, int incr)
664 {
665 slave_t *slave = NULL;
666 int expect = bond->slave_cnt + incr;
667
668 bond->slave_cnt = 0;
669 for (slave = bond->prev; slave != (slave_t*)bond;
670 slave = slave->prev) {
671 bond->slave_cnt++;
672 }
673
674 if (expect != bond->slave_cnt)
675 BUG();
676 }
677
678 /*
679 * Set MAC. Differs from eth_mac_addr in that we allow changes while
680 * netif_running().
681 */
682 static int
683 bond_set_mac_address(struct net_device *dev, void *p)
684 {
685 struct sockaddr *addr = p;
686
687 memcpy(dev->dev_addr, addr->sa_data, dev->addr_len);
688 return 0;
689 }
690
691 /*
692 * This function detaches the slave <slave> from the list <bond>.
693 * WARNING: no check is made to verify if the slave effectively
694 * belongs to <bond>. It returns <slave> in case it's needed.
695 * Nothing is freed on return, structures are just unchained.
696 * If the bond->current_slave pointer was pointing to <slave>,
697 * it's replaced with bond->next, or NULL if not applicable.
698 *
699 * bond->lock held for writing by caller.
700 */
701 static slave_t *
702 bond_detach_slave(bonding_t *bond, slave_t *slave)
703 {
704 if ((bond == NULL) || (slave == NULL) ||
705 ((void *)bond == (void *)slave)) {
706 printk(KERN_ERR
707 "bond_detach_slave(): trying to detach "
708 "slave %p from bond %p\n", bond, slave);
709 return slave;
710 }
711
712 if (bond->next == slave) { /* is the slave at the head ? */
713 if (bond->prev == slave) { /* is the slave alone ? */
714 bond->prev = bond->next = (slave_t *)bond;
715 } else { /* not alone */
716 bond->next = slave->next;
717 slave->next->prev = (slave_t *)bond;
718 bond->prev->next = slave->next;
719 }
720 } else {
721 slave->prev->next = slave->next;
722 if (bond->prev == slave) { /* is this slave the last one ? */
723 bond->prev = slave->prev;
724 } else {
725 slave->next->prev = slave->prev;
726 }
727 }
728
729 update_slave_cnt(bond, -1);
730
731 /* no need to hold ptrlock since bond lock is
732 * already held for writing
733 */
734 if (slave == bond->current_slave) {
735 if ( bond->next != (slave_t *)bond) { /* found one slave */
736 bond_assign_current_slave(bond, bond->next);
737 } else {
738 bond_assign_current_slave(bond, NULL);
739 }
740 }
741
742 return slave;
743 }
744
745 /*
746 * This function attaches the slave <slave> to the list <bond>.
747 *
748 * bond->lock held for writing by caller.
749 */
750 static void
751 bond_attach_slave(struct bonding *bond, struct slave *new_slave)
752 {
753 /*
754 * queue to the end of the slaves list, make the first element its
755 * successor, the last one its predecessor, and make it the bond's
756 * predecessor.
757 *
758 * Just to clarify, so future bonding driver hackers don't go through
759 * the same confusion stage I did trying to figure this out, the
760 * slaves are stored in a double linked circular list, sortof.
761 * In the ->next direction, the last slave points to the first slave,
762 * bypassing bond; only the slaves are in the ->next direction.
763 * In the ->prev direction, however, the first slave points to bond
764 * and bond points to the last slave.
765 *
766 * It looks like a circle with a little bubble hanging off one side
767 * in the ->prev direction only.
768 *
769 * When going through the list once, its best to start at bond->prev
770 * and go in the ->prev direction, testing for bond. Doing this
771 * in the ->next direction doesn't work. Trust me, I know this now.
772 * :) -mts 2002.03.14
773 */
774 new_slave->prev = bond->prev;
775 new_slave->prev->next = new_slave;
776 bond->prev = new_slave;
777 new_slave->next = bond->next;
778
779 update_slave_cnt(bond, 1);
780 }
781
782
783 /*
784 * Less bad way to call ioctl from within the kernel; this needs to be
785 * done some other way to get the call out of interrupt context.
786 * Needs "ioctl" variable to be supplied by calling context.
787 */
788 #define IOCTL(dev, arg, cmd) ({ \
789 int ret; \
790 mm_segment_t fs = get_fs(); \
791 set_fs(get_ds()); \
792 ret = ioctl(dev, arg, cmd); \
793 set_fs(fs); \
794 ret; })
795
796 /*
797 * Get link speed and duplex from the slave's base driver
798 * using ethtool. If for some reason the call fails or the
799 * values are invalid, fake speed and duplex to 100/Full
800 * and return error.
801 */
802 static int bond_update_speed_duplex(struct slave *slave)
803 {
804 struct net_device *dev = slave->dev;
805 static int (* ioctl)(struct net_device *, struct ifreq *, int);
806 struct ifreq ifr;
807 struct ethtool_cmd etool;
808
809 ioctl = dev->do_ioctl;
810 if (ioctl) {
811 etool.cmd = ETHTOOL_GSET;
812 ifr.ifr_data = (char*)&etool;
813 if (IOCTL(dev, &ifr, SIOCETHTOOL) == 0) {
814 slave->speed = etool.speed;
815 slave->duplex = etool.duplex;
816 } else {
817 goto err_out;
818 }
819 } else {
820 goto err_out;
821 }
822
823 switch (slave->speed) {
824 case SPEED_10:
825 case SPEED_100:
826 case SPEED_1000:
827 break;
828 default:
829 goto err_out;
830 }
831
832 switch (slave->duplex) {
833 case DUPLEX_FULL:
834 case DUPLEX_HALF:
835 break;
836 default:
837 goto err_out;
838 }
839
840 return 0;
841
842 err_out:
843 /* Fake speed and duplex */
844 slave->speed = SPEED_100;
845 slave->duplex = DUPLEX_FULL;
846 return -1;
847 }
848
849 /*
850 * if <dev> supports MII link status reporting, check its link status.
851 *
852 * We either do MII/ETHTOOL ioctls, or check netif_carrier_ok(),
853 * depening upon the setting of the use_carrier parameter.
854 *
855 * Return either BMSR_LSTATUS, meaning that the link is up (or we
856 * can't tell and just pretend it is), or 0, meaning that the link is
857 * down.
858 *
859 * If reporting is non-zero, instead of faking link up, return -1 if
860 * both ETHTOOL and MII ioctls fail (meaning the device does not
861 * support them). If use_carrier is set, return whatever it says.
862 * It'd be nice if there was a good way to tell if a driver supports
863 * netif_carrier, but there really isn't.
864 */
865 static int
866 bond_check_dev_link(struct net_device *dev, int reporting)
867 {
868 static int (* ioctl)(struct net_device *, struct ifreq *, int);
869 struct ifreq ifr;
870 struct mii_ioctl_data *mii;
871 struct ethtool_value etool;
872
873 if (use_carrier) {
874 return netif_carrier_ok(dev) ? BMSR_LSTATUS : 0;
875 }
876
877 ioctl = dev->do_ioctl;
878 if (ioctl) {
879 /* TODO: set pointer to correct ioctl on a per team member */
880 /* bases to make this more efficient. that is, once */
881 /* we determine the correct ioctl, we will always */
882 /* call it and not the others for that team */
883 /* member. */
884
885 /*
886 * We cannot assume that SIOCGMIIPHY will also read a
887 * register; not all network drivers (e.g., e100)
888 * support that.
889 */
890
891 /* Yes, the mii is overlaid on the ifreq.ifr_ifru */
892 mii = (struct mii_ioctl_data *)&ifr.ifr_data;
893 if (IOCTL(dev, &ifr, SIOCGMIIPHY) == 0) {
894 mii->reg_num = MII_BMSR;
895 if (IOCTL(dev, &ifr, SIOCGMIIREG) == 0) {
896 return mii->val_out & BMSR_LSTATUS;
897 }
898 }
899
900 /* try SIOCETHTOOL ioctl, some drivers cache ETHTOOL_GLINK */
901 /* for a period of time so we attempt to get link status */
902 /* from it last if the above MII ioctls fail... */
903 etool.cmd = ETHTOOL_GLINK;
904 ifr.ifr_data = (char*)&etool;
905 if (IOCTL(dev, &ifr, SIOCETHTOOL) == 0) {
906 if (etool.data == 1) {
907 return BMSR_LSTATUS;
908 } else {
909 #ifdef BONDING_DEBUG
910 printk(KERN_INFO
911 ":: SIOCETHTOOL shows link down \n");
912 #endif
913 return 0;
914 }
915 }
916
917 }
918
919 /*
920 * If reporting, report that either there's no dev->do_ioctl,
921 * or both SIOCGMIIREG and SIOCETHTOOL failed (meaning that we
922 * cannot report link status). If not reporting, pretend
923 * we're ok.
924 */
925 return reporting ? -1 : BMSR_LSTATUS;
926 }
927
928 static u16 bond_check_mii_link(bonding_t *bond)
929 {
930 int has_active_interface = 0;
931
932 read_lock_bh(&bond->lock);
933 read_lock(&bond->ptrlock);
934 has_active_interface = (bond->current_slave != NULL);
935 read_unlock(&bond->ptrlock);
936 read_unlock_bh(&bond->lock);
937
938 return (has_active_interface ? BMSR_LSTATUS : 0);
939 }
940
941 /* register to receive lacpdus on a bond */
942 static void bond_register_lacpdu(struct bonding *bond)
943 {
944 struct packet_type* pk_type = &(BOND_AD_INFO(bond).ad_pkt_type);
945
946 /* initialize packet type */
947 pk_type->type = PKT_TYPE_LACPDU;
948 pk_type->dev = bond->device;
949 pk_type->func = bond_3ad_lacpdu_recv;
950 pk_type->data = (void*)1; /* understand shared skbs */
951
952 dev_add_pack(pk_type);
953 }
954
955 /* unregister to receive lacpdus on a bond */
956 static void bond_unregister_lacpdu(struct bonding *bond)
957 {
958 dev_remove_pack(&(BOND_AD_INFO(bond).ad_pkt_type));
959 }
960
961 static int bond_open(struct net_device *dev)
962 {
963 struct bonding *bond = (struct bonding *)(dev->priv);
964 struct timer_list *timer = &((struct bonding *)(dev->priv))->mii_timer;
965 struct timer_list *arp_timer = &((struct bonding *)(dev->priv))->arp_timer;
966
967 if ((bond_mode == BOND_MODE_TLB) ||
968 (bond_mode == BOND_MODE_ALB)) {
969 struct timer_list *alb_timer = &(BOND_ALB_INFO(bond).alb_timer);
970
971 /* bond_alb_initialize must be called before the timer
972 * is started.
973 */
974 if (bond_alb_initialize(bond, (bond_mode == BOND_MODE_ALB))) {
975 /* something went wrong - fail the open operation */
976 return -1;
977 }
978
979 init_timer(alb_timer);
980 alb_timer->expires = jiffies + 1;
981 alb_timer->data = (unsigned long)bond;
982 alb_timer->function = (void *)&bond_alb_monitor;
983 add_timer(alb_timer);
984 }
985
986 MOD_INC_USE_COUNT;
987
988 if (miimon > 0) { /* link check interval, in milliseconds. */
989 init_timer(timer);
990 timer->expires = jiffies + (miimon * HZ / 1000);
991 timer->data = (unsigned long)dev;
992 timer->function = (void *)&bond_mii_monitor;
993 add_timer(timer);
994 }
995
996 if (arp_interval> 0) { /* arp interval, in milliseconds. */
997 init_timer(arp_timer);
998 arp_timer->expires = jiffies + (arp_interval * HZ / 1000);
999 arp_timer->data = (unsigned long)dev;
1000 if (bond_mode == BOND_MODE_ACTIVEBACKUP) {
1001 arp_timer->function = (void *)&activebackup_arp_monitor;
1002 } else {
1003 arp_timer->function = (void *)&loadbalance_arp_monitor;
1004 }
1005 add_timer(arp_timer);
1006 }
1007
1008 if (bond_mode == BOND_MODE_8023AD) {
1009 struct timer_list *ad_timer = &(BOND_AD_INFO(bond).ad_timer);
1010 init_timer(ad_timer);
1011 ad_timer->expires = jiffies + (AD_TIMER_INTERVAL * HZ / 1000);
1012 ad_timer->data = (unsigned long)bond;
1013 ad_timer->function = (void *)&bond_3ad_state_machine_handler;
1014 add_timer(ad_timer);
1015
1016 /* register to receive LACPDUs */
1017 bond_register_lacpdu(bond);
1018 }
1019
1020 return 0;
1021 }
1022
1023 static int bond_close(struct net_device *master)
1024 {
1025 bonding_t *bond = (struct bonding *) master->priv;
1026
1027 write_lock_bh(&bond->lock);
1028
1029 if (miimon > 0) { /* link check interval, in milliseconds. */
1030 del_timer(&bond->mii_timer);
1031 }
1032 if (arp_interval> 0) { /* arp interval, in milliseconds. */
1033 del_timer(&bond->arp_timer);
1034 if (arp_target_hw_addr != NULL) {
1035 kfree(arp_target_hw_addr);
1036 arp_target_hw_addr = NULL;
1037 }
1038 }
1039
1040 if (bond_mode == BOND_MODE_8023AD) {
1041 del_timer_sync(&(BOND_AD_INFO(bond).ad_timer));
1042
1043 /* Unregister the receive of LACPDUs */
1044 bond_unregister_lacpdu(bond);
1045 }
1046
1047 bond_mc_list_destroy (bond);
1048
1049 write_unlock_bh(&bond->lock);
1050
1051 /* Release the bonded slaves */
1052 bond_release_all(master);
1053
1054 if ((bond_mode == BOND_MODE_TLB) ||
1055 (bond_mode == BOND_MODE_ALB)) {
1056 del_timer_sync(&(BOND_ALB_INFO(bond).alb_timer));
1057
1058 bond_alb_deinitialize(bond);
1059 }
1060
1061 MOD_DEC_USE_COUNT;
1062 return 0;
1063 }
1064
1065 /*
1066 * flush all members of flush->mc_list from device dev->mc_list
1067 */
1068 static void bond_mc_list_flush(struct net_device *dev, struct net_device *flush)
1069 {
1070 struct dev_mc_list *dmi;
1071
1072 for (dmi = flush->mc_list; dmi != NULL; dmi = dmi->next)
1073 dev_mc_delete(dev, dmi->dmi_addr, dmi->dmi_addrlen, 0);
1074
1075 if (bond_mode == BOND_MODE_8023AD) {
1076 /* del lacpdu mc addr from mc list */
1077 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
1078
1079 dev_mc_delete(dev, lacpdu_multicast, ETH_ALEN, 0);
1080 }
1081 }
1082
1083 /*
1084 * Totally destroys the mc_list in bond
1085 */
1086 static void bond_mc_list_destroy(struct bonding *bond)
1087 {
1088 struct dev_mc_list *dmi;
1089
1090 dmi = bond->mc_list;
1091 while (dmi) {
1092 bond->mc_list = dmi->next;
1093 kfree(dmi);
1094 dmi = bond->mc_list;
1095 }
1096 }
1097
1098 /*
1099 * Add a Multicast address to every slave in the bonding group
1100 */
1101 static void bond_mc_add(bonding_t *bond, void *addr, int alen)
1102 {
1103 slave_t *slave;
1104 switch (multicast_mode) {
1105 case BOND_MULTICAST_ACTIVE :
1106 /* write lock already acquired */
1107 if (bond->current_slave != NULL)
1108 dev_mc_add(bond->current_slave->dev, addr, alen, 0);
1109 break;
1110 case BOND_MULTICAST_ALL :
1111 for (slave = bond->prev; slave != (slave_t*)bond; slave = slave->prev)
1112 dev_mc_add(slave->dev, addr, alen, 0);
1113 break;
1114 case BOND_MULTICAST_DISABLED :
1115 break;
1116 }
1117 }
1118
1119 /*
1120 * Remove a multicast address from every slave in the bonding group
1121 */
1122 static void bond_mc_delete(bonding_t *bond, void *addr, int alen)
1123 {
1124 slave_t *slave;
1125 switch (multicast_mode) {
1126 case BOND_MULTICAST_ACTIVE :
1127 /* write lock already acquired */
1128 if (bond->current_slave != NULL)
1129 dev_mc_delete(bond->current_slave->dev, addr, alen, 0);
1130 break;
1131 case BOND_MULTICAST_ALL :
1132 for (slave = bond->prev; slave != (slave_t*)bond; slave = slave->prev)
1133 dev_mc_delete(slave->dev, addr, alen, 0);
1134 break;
1135 case BOND_MULTICAST_DISABLED :
1136 break;
1137 }
1138 }
1139
1140 /*
1141 * Copy all the Multicast addresses from src to the bonding device dst
1142 */
1143 static int bond_mc_list_copy (struct dev_mc_list *src, struct bonding *dst,
1144 int gpf_flag)
1145 {
1146 struct dev_mc_list *dmi, *new_dmi;
1147
1148 for (dmi = src; dmi != NULL; dmi = dmi->next) {
1149 new_dmi = kmalloc(sizeof(struct dev_mc_list), gpf_flag);
1150
1151 if (new_dmi == NULL) {
1152 return -ENOMEM;
1153 }
1154
1155 new_dmi->next = dst->mc_list;
1156 dst->mc_list = new_dmi;
1157
1158 new_dmi->dmi_addrlen = dmi->dmi_addrlen;
1159 memcpy(new_dmi->dmi_addr, dmi->dmi_addr, dmi->dmi_addrlen);
1160 new_dmi->dmi_users = dmi->dmi_users;
1161 new_dmi->dmi_gusers = dmi->dmi_gusers;
1162 }
1163 return 0;
1164 }
1165
1166 /*
1167 * Returns 0 if dmi1 and dmi2 are the same, non-0 otherwise
1168 */
1169 static inline int dmi_same(struct dev_mc_list *dmi1, struct dev_mc_list *dmi2)
1170 {
1171 return memcmp(dmi1->dmi_addr, dmi2->dmi_addr, dmi1->dmi_addrlen) == 0 &&
1172 dmi1->dmi_addrlen == dmi2->dmi_addrlen;
1173 }
1174
1175 /*
1176 * Push the promiscuity flag down to all slaves
1177 */
1178 static void bond_set_promiscuity(bonding_t *bond, int inc)
1179 {
1180 slave_t *slave;
1181 switch (multicast_mode) {
1182 case BOND_MULTICAST_ACTIVE :
1183 /* write lock already acquired */
1184 if (bond->current_slave != NULL)
1185 dev_set_promiscuity(bond->current_slave->dev, inc);
1186 break;
1187 case BOND_MULTICAST_ALL :
1188 for (slave = bond->prev; slave != (slave_t*)bond; slave = slave->prev)
1189 dev_set_promiscuity(slave->dev, inc);
1190 break;
1191 case BOND_MULTICAST_DISABLED :
1192 break;
1193 }
1194 }
1195
1196 /*
1197 * Push the allmulti flag down to all slaves
1198 */
1199 static void bond_set_allmulti(bonding_t *bond, int inc)
1200 {
1201 slave_t *slave;
1202 switch (multicast_mode) {
1203 case BOND_MULTICAST_ACTIVE :
1204 /* write lock already acquired */
1205 if (bond->current_slave != NULL)
1206 dev_set_allmulti(bond->current_slave->dev, inc);
1207 break;
1208 case BOND_MULTICAST_ALL :
1209 for (slave = bond->prev; slave != (slave_t*)bond; slave = slave->prev)
1210 dev_set_allmulti(slave->dev, inc);
1211 break;
1212 case BOND_MULTICAST_DISABLED :
1213 break;
1214 }
1215 }
1216
1217 /*
1218 * returns dmi entry if found, NULL otherwise
1219 */
1220 static struct dev_mc_list* bond_mc_list_find_dmi(struct dev_mc_list *dmi,
1221 struct dev_mc_list *mc_list)
1222 {
1223 struct dev_mc_list *idmi;
1224
1225 for (idmi = mc_list; idmi != NULL; idmi = idmi->next) {
1226 if (dmi_same(dmi, idmi)) {
1227 return idmi;
1228 }
1229 }
1230 return NULL;
1231 }
1232
1233 static void set_multicast_list(struct net_device *master)
1234 {
1235 bonding_t *bond = master->priv;
1236 struct dev_mc_list *dmi;
1237
1238 if (multicast_mode == BOND_MULTICAST_DISABLED)
1239 return;
1240 /*
1241 * Lock the private data for the master
1242 */
1243 write_lock_bh(&bond->lock);
1244
1245 /* set promiscuity flag to slaves */
1246 if ( (master->flags & IFF_PROMISC) && !(bond->flags & IFF_PROMISC) )
1247 bond_set_promiscuity(bond, 1);
1248
1249 if ( !(master->flags & IFF_PROMISC) && (bond->flags & IFF_PROMISC) )
1250 bond_set_promiscuity(bond, -1);
1251
1252 /* set allmulti flag to slaves */
1253 if ( (master->flags & IFF_ALLMULTI) && !(bond->flags & IFF_ALLMULTI) )
1254 bond_set_allmulti(bond, 1);
1255
1256 if ( !(master->flags & IFF_ALLMULTI) && (bond->flags & IFF_ALLMULTI) )
1257 bond_set_allmulti(bond, -1);
1258
1259 bond->flags = master->flags;
1260
1261 /* looking for addresses to add to slaves' mc list */
1262 for (dmi = master->mc_list; dmi != NULL; dmi = dmi->next) {
1263 if (bond_mc_list_find_dmi(dmi, bond->mc_list) == NULL)
1264 bond_mc_add(bond, dmi->dmi_addr, dmi->dmi_addrlen);
1265 }
1266
1267 /* looking for addresses to delete from slaves' list */
1268 for (dmi = bond->mc_list; dmi != NULL; dmi = dmi->next) {
1269 if (bond_mc_list_find_dmi(dmi, master->mc_list) == NULL)
1270 bond_mc_delete(bond, dmi->dmi_addr, dmi->dmi_addrlen);
1271 }
1272
1273
1274 /* save master's multicast list */
1275 bond_mc_list_destroy (bond);
1276 bond_mc_list_copy (master->mc_list, bond, GFP_ATOMIC);
1277
1278 write_unlock_bh(&bond->lock);
1279 }
1280
1281 /*
1282 * Update the mc list and multicast-related flags for the new and
1283 * old active slaves (if any) according to the multicast mode
1284 */
1285 static void bond_mc_update(bonding_t *bond, slave_t *new, slave_t *old)
1286 {
1287 struct dev_mc_list *dmi;
1288
1289 switch(multicast_mode) {
1290 case BOND_MULTICAST_ACTIVE :
1291 if (bond->device->flags & IFF_PROMISC) {
1292 if (old != NULL && new != old)
1293 dev_set_promiscuity(old->dev, -1);
1294 dev_set_promiscuity(new->dev, 1);
1295 }
1296 if (bond->device->flags & IFF_ALLMULTI) {
1297 if (old != NULL && new != old)
1298 dev_set_allmulti(old->dev, -1);
1299 dev_set_allmulti(new->dev, 1);
1300 }
1301 /* first remove all mc addresses from old slave if any,
1302 and _then_ add them to new active slave */
1303 if (old != NULL && new != old) {
1304 for (dmi = bond->device->mc_list; dmi != NULL; dmi = dmi->next)
1305 dev_mc_delete(old->dev, dmi->dmi_addr, dmi->dmi_addrlen, 0);
1306 }
1307 for (dmi = bond->device->mc_list; dmi != NULL; dmi = dmi->next)
1308 dev_mc_add(new->dev, dmi->dmi_addr, dmi->dmi_addrlen, 0);
1309 break;
1310 case BOND_MULTICAST_ALL :
1311 /* nothing to do: mc list is already up-to-date on all slaves */
1312 break;
1313 case BOND_MULTICAST_DISABLED :
1314 break;
1315 }
1316 }
1317
1318 /* enslave device <slave> to bond device <master> */
1319 static int bond_enslave(struct net_device *master_dev,
1320 struct net_device *slave_dev)
1321 {
1322 bonding_t *bond = NULL;
1323 slave_t *new_slave = NULL;
1324 unsigned long rflags = 0;
1325 int err = 0;
1326 struct dev_mc_list *dmi;
1327 struct in_ifaddr **ifap;
1328 struct in_ifaddr *ifa;
1329 int link_reporting;
1330 struct sockaddr addr;
1331
1332 if (master_dev == NULL || slave_dev == NULL) {
1333 return -ENODEV;
1334 }
1335 bond = (struct bonding *) master_dev->priv;
1336
1337 if (slave_dev->do_ioctl == NULL) {
1338 printk(KERN_DEBUG
1339 "Warning : no link monitoring support for %s\n",
1340 slave_dev->name);
1341 }
1342
1343
1344 /* bond must be initialized by bond_open() before enslaving */
1345 if (!(master_dev->flags & IFF_UP)) {
1346 #ifdef BONDING_DEBUG
1347 printk(KERN_CRIT "Error, master_dev is not up\n");
1348 #endif
1349 return -EPERM;
1350 }
1351
1352 /* already enslaved */
1353 if (master_dev->flags & IFF_SLAVE || slave_dev->flags & IFF_SLAVE) {
1354 #ifdef BONDING_DEBUG
1355 printk(KERN_CRIT "Error, Device was already enslaved\n");
1356 #endif
1357 return -EBUSY;
1358 }
1359
1360 if (app_abi_ver >= 1) {
1361 /* The application is using an ABI, which requires the
1362 * slave interface to be closed.
1363 */
1364 if ((slave_dev->flags & IFF_UP)) {
1365 #ifdef BONDING_DEBUG
1366 printk(KERN_CRIT "Error, slave_dev is up\n");
1367 #endif
1368 return -EPERM;
1369 }
1370
1371 if (slave_dev->set_mac_address == NULL) {
1372 printk(KERN_CRIT
1373 "The slave device you specified does not support"
1374 " setting the MAC address.\n");
1375 printk(KERN_CRIT
1376 "Your kernel likely does not support slave"
1377 " devices.\n");
1378
1379 return -EOPNOTSUPP;
1380 }
1381 } else {
1382 /* The application is not using an ABI, which requires the
1383 * slave interface to be open.
1384 */
1385 if (!(slave_dev->flags & IFF_UP)) {
1386 #ifdef BONDING_DEBUG
1387 printk(KERN_CRIT "Error, slave_dev is not running\n");
1388 #endif
1389 return -EINVAL;
1390 }
1391
1392 if ((bond_mode == BOND_MODE_8023AD) ||
1393 (bond_mode == BOND_MODE_TLB) ||
1394 (bond_mode == BOND_MODE_ALB)) {
1395 printk(KERN_ERR
1396 "bonding: Error: to use %s mode, you must "
1397 "upgrade ifenslave.\n", bond_mode_name());
1398 return -EOPNOTSUPP;
1399 }
1400 }
1401
1402 if ((new_slave = kmalloc(sizeof(slave_t), GFP_KERNEL)) == NULL) {
1403 return -ENOMEM;
1404 }
1405 memset(new_slave, 0, sizeof(slave_t));
1406
1407 /* save slave's original flags before calling
1408 * netdev_set_master and dev_open
1409 */
1410 new_slave->original_flags = slave_dev->flags;
1411
1412 if (app_abi_ver >= 1) {
1413 /* save slave's original ("permanent") mac address for
1414 * modes that needs it, and for restoring it upon release,
1415 * and then set it to the master's address
1416 */
1417 memcpy(new_slave->perm_hwaddr, slave_dev->dev_addr, ETH_ALEN);
1418
1419 if (bond->slave_cnt > 0) {
1420 /* set slave to master's mac address
1421 * The application already set the master's
1422 * mac address to that of the first slave
1423 */
1424 memcpy(addr.sa_data, master_dev->dev_addr, ETH_ALEN);
1425 addr.sa_family = slave_dev->type;
1426 err = slave_dev->set_mac_address(slave_dev, &addr);
1427 if (err) {
1428 #ifdef BONDING_DEBUG
1429 printk(KERN_CRIT "Error %d calling set_mac_address\n", err);
1430 #endif
1431 goto err_free;
1432 }
1433 }
1434
1435 /* open the slave since the application closed it */
1436 err = dev_open(slave_dev);
1437 if (err) {
1438 #ifdef BONDING_DEBUG
1439 printk(KERN_CRIT "Openning slave %s failed\n", slave_dev->name);
1440 #endif
1441 goto err_restore_mac;
1442 }
1443 }
1444
1445 err = netdev_set_master(slave_dev, master_dev);
1446 if (err) {
1447 #ifdef BONDING_DEBUG
1448 printk(KERN_CRIT "Error %d calling netdev_set_master\n", err);
1449 #endif
1450 if (app_abi_ver < 1) {
1451 goto err_free;
1452 } else {
1453 goto err_close;
1454 }
1455 }
1456
1457 new_slave->dev = slave_dev;
1458
1459 if ((bond_mode == BOND_MODE_TLB) ||
1460 (bond_mode == BOND_MODE_ALB)) {
1461 /* bond_alb_init_slave() must be called before all other stages since
1462 * it might fail and we do not want to have to undo everything
1463 */
1464 err = bond_alb_init_slave(bond, new_slave);
1465 if (err) {
1466 goto err_unset_master;
1467 }
1468 }
1469
1470 if (multicast_mode == BOND_MULTICAST_ALL) {
1471 /* set promiscuity level to new slave */
1472 if (master_dev->flags & IFF_PROMISC)
1473 dev_set_promiscuity(slave_dev, 1);
1474
1475 /* set allmulti level to new slave */
1476 if (master_dev->flags & IFF_ALLMULTI)
1477 dev_set_allmulti(slave_dev, 1);
1478
1479 /* upload master's mc_list to new slave */
1480 for (dmi = master_dev->mc_list; dmi != NULL; dmi = dmi->next)
1481 dev_mc_add (slave_dev, dmi->dmi_addr, dmi->dmi_addrlen, 0);
1482 }
1483
1484 if (bond_mode == BOND_MODE_8023AD) {
1485 /* add lacpdu mc addr to mc list */
1486 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
1487
1488 dev_mc_add(slave_dev, lacpdu_multicast, ETH_ALEN, 0);
1489 }
1490
1491 write_lock_bh(&bond->lock);
1492
1493 bond_attach_slave(bond, new_slave);
1494 new_slave->delay = 0;
1495 new_slave->link_failure_count = 0;
1496
1497 if (miimon > 0 && !use_carrier) {
1498 link_reporting = bond_check_dev_link(slave_dev, 1);
1499
1500 if ((link_reporting == -1) && (arp_interval == 0)) {
1501 /*
1502 * miimon is set but a bonded network driver
1503 * does not support ETHTOOL/MII and
1504 * arp_interval is not set. Note: if
1505 * use_carrier is enabled, we will never go
1506 * here (because netif_carrier is always
1507 * supported); thus, we don't need to change
1508 * the messages for netif_carrier.
1509 */
1510 printk(KERN_ERR
1511 "bond_enslave(): MII and ETHTOOL support not "
1512 "available for interface %s, and "
1513 "arp_interval/arp_ip_target module parameters "
1514 "not specified, thus bonding will not detect "
1515 "link failures! see bonding.txt for details.\n",
1516 slave_dev->name);
1517 } else if (link_reporting == -1) {
1518 /* unable get link status using mii/ethtool */
1519 printk(KERN_WARNING
1520 "bond_enslave: can't get link status from "
1521 "interface %s; the network driver associated "
1522 "with this interface does not support "
1523 "MII or ETHTOOL link status reporting, thus "
1524 "miimon has no effect on this interface.\n",
1525 slave_dev->name);
1526 }
1527 }
1528
1529 /* check for initial state */
1530 if ((miimon <= 0) ||
1531 (bond_check_dev_link(slave_dev, 0) == BMSR_LSTATUS)) {
1532 if (updelay) {
1533 #ifdef BONDING_DEBUG
1534 printk(KERN_CRIT "Initial state of slave_dev is "
1535 "BOND_LINK_BACK\n");
1536 #endif
1537 new_slave->link = BOND_LINK_BACK;
1538 new_slave->delay = updelay;
1539 }
1540 else {
1541 #ifdef BONDING_DEBUG
1542 printk(KERN_DEBUG "Initial state of slave_dev is "
1543 "BOND_LINK_UP\n");
1544 #endif
1545 new_slave->link = BOND_LINK_UP;
1546 }
1547 new_slave->jiffies = jiffies;
1548 }
1549 else {
1550 #ifdef BONDING_DEBUG
1551 printk(KERN_CRIT "Initial state of slave_dev is "
1552 "BOND_LINK_DOWN\n");
1553 #endif
1554 new_slave->link = BOND_LINK_DOWN;
1555 }
1556
1557 if (bond_update_speed_duplex(new_slave) &&
1558 (new_slave->link != BOND_LINK_DOWN)) {
1559
1560 printk(KERN_WARNING
1561 "bond_enslave(): failed to get speed/duplex from %s, "
1562 "speed forced to 100Mbps, duplex forced to Full.\n",
1563 new_slave->dev->name);
1564 if (bond_mode == BOND_MODE_8023AD) {
1565 printk(KERN_WARNING
1566 "Operation of 802.3ad mode requires ETHTOOL support "
1567 "in base driver for proper aggregator selection.\n");
1568 }
1569 }
1570
1571 /* if we're in active-backup mode, we need one and only one active
1572 * interface. The backup interfaces will have their NOARP flag set
1573 * because we need them to be completely deaf and not to respond to
1574 * any ARP request on the network to avoid fooling a switch. Thus,
1575 * since we guarantee that current_slave always point to the last
1576 * usable interface, we just have to verify this interface's flag.
1577 */
1578 if (bond_mode == BOND_MODE_ACTIVEBACKUP) {
1579 if (((bond->current_slave == NULL)
1580 || (bond->current_slave->dev->flags & IFF_NOARP))
1581 && (new_slave->link != BOND_LINK_DOWN)) {
1582 #ifdef BONDING_DEBUG
1583 printk(KERN_CRIT "This is the first active slave\n");
1584 #endif
1585 /* first slave or no active slave yet, and this link
1586 is OK, so make this interface the active one */
1587 bond_assign_current_slave(bond, new_slave);
1588 bond_set_slave_active_flags(new_slave);
1589 bond_mc_update(bond, new_slave, NULL);
1590 }
1591 else {
1592 #ifdef BONDING_DEBUG
1593 printk(KERN_CRIT "This is just a backup slave\n");
1594 #endif
1595 bond_set_slave_inactive_flags(new_slave);
1596 }
1597 if (((struct in_device *)slave_dev->ip_ptr) != NULL) {
1598 read_lock_irqsave(&(((struct in_device *)slave_dev->ip_ptr)->lock), rflags);
1599 ifap= &(((struct in_device *)slave_dev->ip_ptr)->ifa_list);
1600 ifa = *ifap;
1601 if (ifa != NULL)
1602 my_ip = ifa->ifa_address;
1603 read_unlock_irqrestore(&(((struct in_device *)slave_dev->ip_ptr)->lock), rflags);
1604 }
1605
1606 /* if there is a primary slave, remember it */
1607 if (primary != NULL) {
1608 if (strcmp(primary, new_slave->dev->name) == 0) {
1609 bond->primary_slave = new_slave;
1610 }
1611 }
1612 } else if (bond_mode == BOND_MODE_8023AD) {
1613 /* in 802.3ad mode, the internal mechanism
1614 * will activate the slaves in the selected
1615 * aggregator
1616 */
1617 bond_set_slave_inactive_flags(new_slave);
1618 /* if this is the first slave */
1619 if (new_slave == bond->next) {
1620 SLAVE_AD_INFO(new_slave).id = 1;
1621 /* Initialize AD with the number of times that the AD timer is called in 1 second
1622 * can be called only after the mac address of the bond is set
1623 */
1624 bond_3ad_initialize(bond, 1000/AD_TIMER_INTERVAL,
1625 lacp_fast);
1626 } else {
1627 SLAVE_AD_INFO(new_slave).id =
1628 SLAVE_AD_INFO(new_slave->prev).id + 1;
1629 }
1630
1631 bond_3ad_bind_slave(new_slave);
1632 } else if ((bond_mode == BOND_MODE_TLB) ||
1633 (bond_mode == BOND_MODE_ALB)) {
1634 new_slave->state = BOND_STATE_ACTIVE;
1635 if ((bond->current_slave == NULL) && (new_slave->link != BOND_LINK_DOWN)) {
1636 /* first slave or no active slave yet, and this link
1637 * is OK, so make this interface the active one
1638 */
1639 bond_assign_current_slave(bond, new_slave);
1640 }
1641
1642 /* if there is a primary slave, remember it */
1643 if (primary != NULL) {
1644 if (strcmp(primary, new_slave->dev->name) == 0) {
1645 bond->primary_slave = new_slave;
1646 }
1647 }
1648 } else {
1649 #ifdef BONDING_DEBUG
1650 printk(KERN_CRIT "This slave is always active in trunk mode\n");
1651 #endif
1652 /* always active in trunk mode */
1653 new_slave->state = BOND_STATE_ACTIVE;
1654 if (bond->current_slave == NULL)
1655 bond_assign_current_slave(bond, new_slave);
1656 }
1657
1658 write_unlock_bh(&bond->lock);
1659
1660 if (app_abi_ver < 1) {
1661 /*
1662 * !!! This is to support old versions of ifenslave.
1663 * We can remove this in 2.5 because our ifenslave takes
1664 * care of this for us.
1665 * We check to see if the master has a mac address yet.
1666 * If not, we'll give it the mac address of our slave device.
1667 */
1668 int ndx = 0;
1669
1670 for (ndx = 0; ndx < slave_dev->addr_len; ndx++) {
1671 #ifdef BONDING_DEBUG
1672 printk(KERN_DEBUG
1673 "Checking ndx=%d of master_dev->dev_addr\n", ndx);
1674 #endif
1675 if (master_dev->dev_addr[ndx] != 0) {
1676 #ifdef BONDING_DEBUG
1677 printk(KERN_DEBUG
1678 "Found non-zero byte at ndx=%d\n", ndx);
1679 #endif
1680 break;
1681 }
1682 }
1683 if (ndx == slave_dev->addr_len) {
1684 /*
1685 * We got all the way through the address and it was
1686 * all 0's.
1687 */
1688 #ifdef BONDING_DEBUG
1689 printk(KERN_DEBUG "%s doesn't have a MAC address yet. ",
1690 master_dev->name);
1691 printk(KERN_DEBUG "Going to give assign it from %s.\n",
1692 slave_dev->name);
1693 #endif
1694 bond_sethwaddr(master_dev, slave_dev);
1695 }
1696 }
1697
1698 printk (KERN_INFO "%s: enslaving %s as a%s interface with a%s link.\n",
1699 master_dev->name, slave_dev->name,
1700 new_slave->state == BOND_STATE_ACTIVE ? "n active" : " backup",
1701 new_slave->link != BOND_LINK_DOWN ? "n up" : " down");
1702
1703 /* enslave is successful */
1704 return 0;
1705
1706 /* Undo stages on error */
1707 err_unset_master:
1708 netdev_set_master(slave_dev, NULL);
1709
1710 err_close:
1711 dev_close(slave_dev);
1712
1713 err_restore_mac:
1714 memcpy(addr.sa_data, new_slave->perm_hwaddr, ETH_ALEN);
1715 addr.sa_family = slave_dev->type;
1716 slave_dev->set_mac_address(slave_dev, &addr);
1717
1718 err_free:
1719 kfree(new_slave);
1720 return err;
1721 }
1722
1723 /*
1724 * This function changes the active slave to slave <slave_dev>.
1725 * It returns -EINVAL in the following cases.
1726 * - <slave_dev> is not found in the list.
1727 * - There is not active slave now.
1728 * - <slave_dev> is already active.
1729 * - The link state of <slave_dev> is not BOND_LINK_UP.
1730 * - <slave_dev> is not running.
1731 * In these cases, this fuction does nothing.
1732 * In the other cases, currnt_slave pointer is changed and 0 is returned.
1733 */
1734 static int bond_change_active(struct net_device *master_dev, struct net_device *slave_dev)
1735 {
1736 bonding_t *bond;
1737 slave_t *slave;
1738 slave_t *oldactive = NULL;
1739 slave_t *newactive = NULL;
1740 int ret = 0;
1741
1742 if (master_dev == NULL || slave_dev == NULL) {
1743 return -ENODEV;
1744 }
1745
1746 bond = (struct bonding *) master_dev->priv;
1747 write_lock_bh(&bond->lock);
1748 slave = (slave_t *)bond;
1749 oldactive = bond->current_slave;
1750
1751 while ((slave = slave->prev) != (slave_t *)bond) {
1752 if(slave_dev == slave->dev) {
1753 newactive = slave;
1754 break;
1755 }
1756 }
1757
1758 /*
1759 * Changing to the current active: do nothing; return success.
1760 */
1761 if (newactive && (newactive == oldactive)) {
1762 write_unlock_bh(&bond->lock);
1763 return 0;
1764 }
1765
1766 if ((newactive != NULL)&&
1767 (oldactive != NULL)&&
1768 (newactive->link == BOND_LINK_UP)&&
1769 IS_UP(newactive->dev)) {
1770 if (bond_mode == BOND_MODE_ACTIVEBACKUP) {
1771 bond_set_slave_inactive_flags(oldactive);
1772 bond_set_slave_active_flags(newactive);
1773 }
1774
1775 bond_mc_update(bond, newactive, oldactive);
1776 bond_assign_current_slave(bond, newactive);
1777 printk("%s : activate %s(old : %s)\n",
1778 master_dev->name, newactive->dev->name,
1779 oldactive->dev->name);
1780 } else {
1781 ret = -EINVAL;
1782 }
1783 write_unlock_bh(&bond->lock);
1784 return ret;
1785 }
1786
1787 /* Choose a new valid interface from the pool, set it active
1788 * and make it the current slave. If no valid interface is
1789 * found, the oldest slave in BACK state is choosen and
1790 * activated. If none is found, it's considered as no
1791 * interfaces left so the current slave is set to NULL.
1792 * The result is a pointer to the current slave.
1793 *
1794 * Since this function sends messages tails through printk, the caller
1795 * must have started something like `printk(KERN_INFO "xxxx ");'.
1796 *
1797 * Warning: Caller must hold ptrlock for writing.
1798 */
1799 slave_t *change_active_interface(bonding_t *bond)
1800 {
1801 slave_t *newslave, *oldslave;
1802 slave_t *bestslave = NULL;
1803 int mintime;
1804
1805 newslave = oldslave = bond->current_slave;
1806
1807 if (newslave == NULL) { /* there were no active slaves left */
1808 if (bond->next != (slave_t *)bond) { /* found one slave */
1809 newslave = bond_assign_current_slave(bond, bond->next);
1810 } else {
1811
1812 printk (" but could not find any %s interface.\n",
1813 (bond_mode == BOND_MODE_ACTIVEBACKUP) ? "backup":"other");
1814 bond_assign_current_slave(bond, NULL);
1815 return NULL; /* still no slave, return NULL */
1816 }
1817 } else if (bond_mode == BOND_MODE_ACTIVEBACKUP) {
1818 /* make sure oldslave doesn't send arps - this could
1819 * cause a ping-pong effect between interfaces since they
1820 * would be able to tx arps - in active backup only one
1821 * slave should be able to tx arps, and that should be
1822 * the current_slave; the only exception is when all
1823 * slaves have gone down, then only one non-current slave can
1824 * send arps at a time; clearing oldslaves' mc list is handled
1825 * later in this function.
1826 */
1827 bond_set_slave_inactive_flags(oldslave);
1828 }
1829
1830 mintime = updelay;
1831
1832 /* first try the primary link; if arping, a link must tx/rx traffic
1833 * before it can be considered the current_slave - also, we would skip
1834 * slaves between the current_slave and primary_slave that may be up
1835 * and able to arp
1836 */
1837 if ((bond->primary_slave != NULL) && (arp_interval == 0)) {
1838 if (IS_UP(bond->primary_slave->dev))
1839 newslave = bond->primary_slave;
1840 }
1841
1842 do {
1843 if (IS_UP(newslave->dev)) {
1844 if (newslave->link == BOND_LINK_UP) {
1845 /* this one is immediately usable */
1846 if (bond_mode == BOND_MODE_ACTIVEBACKUP) {
1847 bond_set_slave_active_flags(newslave);
1848 bond_mc_update(bond, newslave, oldslave);
1849 printk (" and making interface %s the active one.\n",
1850 newslave->dev->name);
1851 }
1852 else {
1853 printk (" and setting pointer to interface %s.\n",
1854 newslave->dev->name);
1855 }
1856
1857 bond_assign_current_slave(bond, newslave);
1858 return newslave;
1859 }
1860 else if (newslave->link == BOND_LINK_BACK) {
1861 /* link up, but waiting for stabilization */
1862 if (newslave->delay < mintime) {
1863 mintime = newslave->delay;
1864 bestslave = newslave;
1865 }
1866 }
1867 }
1868 } while ((newslave = newslave->next) != oldslave);
1869
1870 /* no usable backup found, we'll see if we at least got a link that was
1871 coming back for a long time, and could possibly already be usable.
1872 */
1873
1874 if (bestslave != NULL) {
1875 /* early take-over. */
1876 printk (" and making interface %s the active one %d ms earlier.\n",
1877 bestslave->dev->name,
1878 (updelay - bestslave->delay)*miimon);
1879
1880 bestslave->delay = 0;
1881 bestslave->link = BOND_LINK_UP;
1882 bestslave->jiffies = jiffies;
1883 bond_set_slave_active_flags(bestslave);
1884 bond_mc_update(bond, bestslave, oldslave);
1885 bond_assign_current_slave(bond, bestslave);
1886 return bestslave;
1887 }
1888
1889 if ((bond_mode == BOND_MODE_ACTIVEBACKUP) &&
1890 (multicast_mode == BOND_MULTICAST_ACTIVE) &&
1891 (oldslave != NULL)) {
1892 /* flush bonds (master's) mc_list from oldslave since it wasn't
1893 * updated (and deleted) above
1894 */
1895 bond_mc_list_flush(oldslave->dev, bond->device);
1896 if (bond->device->flags & IFF_PROMISC) {
1897 dev_set_promiscuity(oldslave->dev, -1);
1898 }
1899 if (bond->device->flags & IFF_ALLMULTI) {
1900 dev_set_allmulti(oldslave->dev, -1);
1901 }
1902 }
1903
1904 printk (" but could not find any %s interface.\n",
1905 (bond_mode == BOND_MODE_ACTIVEBACKUP) ? "backup":"other");
1906
1907 /* absolutely nothing found. let's return NULL */
1908 bond_assign_current_slave(bond, NULL);
1909 return NULL;
1910 }
1911
1912 /*
1913 * Try to release the slave device <slave> from the bond device <master>
1914 * It is legal to access current_slave without a lock because all the function
1915 * is write-locked.
1916 *
1917 * The rules for slave state should be:
1918 * for Active/Backup:
1919 * Active stays on all backups go down
1920 * for Bonded connections:
1921 * The first up interface should be left on and all others downed.
1922 */
1923 static int bond_release(struct net_device *master, struct net_device *slave)
1924 {
1925 bonding_t *bond;
1926 slave_t *our_slave, *old_current;
1927 struct sockaddr addr;
1928
1929 if (master == NULL || slave == NULL) {
1930 return -ENODEV;
1931 }
1932
1933 bond = (struct bonding *) master->priv;
1934
1935 /* master already enslaved, or slave not enslaved,
1936 or no slave for this master */
1937 if ((master->flags & IFF_SLAVE) || !(slave->flags & IFF_SLAVE)) {
1938 printk (KERN_DEBUG "%s: cannot release %s.\n", master->name, slave->name);
1939 return -EINVAL;
1940 }
1941
1942 write_lock_bh(&bond->lock);
1943 bond->current_arp_slave = NULL;
1944 our_slave = (slave_t *)bond;
1945 old_current = bond->current_slave;
1946 while ((our_slave = our_slave->prev) != (slave_t *)bond) {
1947 if (our_slave->dev == slave) {
1948 int mac_addr_differ = memcmp(bond->device->dev_addr,
1949 our_slave->perm_hwaddr,
1950 ETH_ALEN);
1951 if (!mac_addr_differ && (bond->slave_cnt > 1)) {
1952 printk(KERN_WARNING "WARNING: the permanent HWaddr of %s "
1953 "- %02X:%02X:%02X:%02X:%02X:%02X - "
1954 "is still in use by %s. Set the HWaddr "
1955 "of %s to a different address "
1956 "to avoid conflicts.\n",
1957 slave->name,
1958 slave->dev_addr[0],
1959 slave->dev_addr[1],
1960 slave->dev_addr[2],
1961 slave->dev_addr[3],
1962 slave->dev_addr[4],
1963 slave->dev_addr[5],
1964 bond->device->name,
1965 slave->name);
1966 }
1967
1968 /* Inform AD package of unbinding of slave. */
1969 if (bond_mode == BOND_MODE_8023AD) {
1970 /* must be called before the slave is
1971 * detached from the list
1972 */
1973 bond_3ad_unbind_slave(our_slave);
1974 }
1975
1976 /* release the slave from its bond */
1977 bond_detach_slave(bond, our_slave);
1978
1979 if (bond->primary_slave == our_slave) {
1980 bond->primary_slave = NULL;
1981 }
1982
1983 printk (KERN_INFO "%s: releasing %s interface %s",
1984 master->name,
1985 (our_slave->state == BOND_STATE_ACTIVE) ? "active" : "backup",
1986 slave->name);
1987
1988 if (our_slave == old_current) {
1989 /* find a new interface and be verbose */
1990 change_active_interface(bond);
1991 } else {
1992 printk(".\n");
1993 }
1994
1995 if (bond->current_slave == NULL) {
1996 printk(KERN_INFO
1997 "%s: now running without any active interface !\n",
1998 master->name);
1999 }
2000
2001 if ((bond_mode == BOND_MODE_TLB) ||
2002 (bond_mode == BOND_MODE_ALB)) {
2003 /* must be called only after the slave has been
2004 * detached from the list and the current_slave
2005 * has been replaced (if our_slave == old_current)
2006 */
2007 bond_alb_deinit_slave(bond, our_slave);
2008 }
2009
2010 break;
2011 }
2012
2013 }
2014 write_unlock_bh(&bond->lock);
2015
2016 if (our_slave == (slave_t *)bond) {
2017 /* if we get here, it's because the device was not found */
2018 printk (KERN_INFO "%s: %s not enslaved\n", master->name, slave->name);
2019 return -EINVAL;
2020 }
2021
2022 /* undo settings and restore original values */
2023
2024 if (multicast_mode == BOND_MULTICAST_ALL) {
2025 /* flush master's mc_list from slave */
2026 bond_mc_list_flush (slave, master);
2027
2028 /* unset promiscuity level from slave */
2029 if (master->flags & IFF_PROMISC)
2030 dev_set_promiscuity(slave, -1);
2031
2032 /* unset allmulti level from slave */
2033 if (master->flags & IFF_ALLMULTI)
2034 dev_set_allmulti(slave, -1);
2035 }
2036
2037 netdev_set_master(slave, NULL);
2038
2039 /* close slave before restoring its mac address */
2040 dev_close(slave);
2041
2042 if (app_abi_ver >= 1) {
2043 /* restore original ("permanent") mac address */
2044 memcpy(addr.sa_data, our_slave->perm_hwaddr, ETH_ALEN);
2045 addr.sa_family = slave->type;
2046 slave->set_mac_address(slave, &addr);
2047 }
2048
2049 /* restore the original state of the
2050 * IFF_NOARP flag that might have been
2051 * set by bond_set_slave_inactive_flags()
2052 */
2053 if ((our_slave->original_flags & IFF_NOARP) == 0) {
2054 slave->flags &= ~IFF_NOARP;
2055 }
2056
2057 kfree(our_slave);
2058
2059 /* if the last slave was removed, zero the mac address
2060 * of the master so it will be set by the application
2061 * to the mac address of the first slave
2062 */
2063 if (bond->next == (slave_t*)bond) {
2064 memset(master->dev_addr, 0, master->addr_len);
2065 }
2066
2067 return 0; /* deletion OK */
2068 }
2069
2070 /*
2071 * This function releases all slaves.
2072 */
2073 static int bond_release_all(struct net_device *master)
2074 {
2075 bonding_t *bond;
2076 slave_t *our_slave, *old_current;
2077 struct net_device *slave_dev;
2078 struct sockaddr addr;
2079 int err = 0;
2080
2081 if (master == NULL) {
2082 return -ENODEV;
2083 }
2084
2085 if (master->flags & IFF_SLAVE) {
2086 return -EINVAL;
2087 }
2088
2089 bond = (struct bonding *) master->priv;
2090
2091 write_lock_bh(&bond->lock);
2092 if (bond->next == (struct slave *) bond) {
2093 err = -EINVAL;
2094 goto out;
2095 }
2096
2097 old_current = bond->current_slave;
2098 bond_assign_current_slave(bond, NULL);
2099 bond->current_arp_slave = NULL;
2100 bond->primary_slave = NULL;
2101
2102 while ((our_slave = bond->prev) != (slave_t *)bond) {
2103 /* Inform AD package of unbinding of slave
2104 * before slave is detached from the list.
2105 */
2106 if (bond_mode == BOND_MODE_8023AD) {
2107 bond_3ad_unbind_slave(our_slave);
2108 }
2109
2110 slave_dev = our_slave->dev;
2111 bond_detach_slave(bond, our_slave);
2112
2113 if ((bond_mode == BOND_MODE_TLB) ||
2114 (bond_mode == BOND_MODE_ALB)) {
2115 /* must be called only after the slave
2116 * has been detached from the list
2117 */
2118 bond_alb_deinit_slave(bond, our_slave);
2119 }
2120
2121 /* now that the slave is detached, unlock and perform
2122 * all the undo steps that should not be called from
2123 * within a lock.
2124 */
2125 write_unlock_bh(&bond->lock);
2126
2127 if (multicast_mode == BOND_MULTICAST_ALL
2128 || (multicast_mode == BOND_MULTICAST_ACTIVE
2129 && old_current == our_slave)) {
2130
2131 /* flush master's mc_list from slave */
2132 bond_mc_list_flush (slave_dev, master);
2133
2134 /* unset promiscuity level from slave */
2135 if (master->flags & IFF_PROMISC)
2136 dev_set_promiscuity(slave_dev, -1);
2137
2138 /* unset allmulti level from slave */
2139 if (master->flags & IFF_ALLMULTI)
2140 dev_set_allmulti(slave_dev, -1);
2141 }
2142
2143 netdev_set_master(slave_dev, NULL);
2144
2145 /* close slave before restoring its mac address */
2146 dev_close(slave_dev);
2147
2148 if (app_abi_ver >= 1) {
2149 /* restore original ("permanent") mac address*/
2150 memcpy(addr.sa_data, our_slave->perm_hwaddr, ETH_ALEN);
2151 addr.sa_family = slave_dev->type;
2152 slave_dev->set_mac_address(slave_dev, &addr);
2153 }
2154
2155 /* restore the original state of the IFF_NOARP flag that might have
2156 * been set by bond_set_slave_inactive_flags()
2157 */
2158 if ((our_slave->original_flags & IFF_NOARP) == 0) {
2159 slave_dev->flags &= ~IFF_NOARP;
2160 }
2161
2162 kfree(our_slave);
2163
2164 /* re-acquire the lock before getting the next slave */
2165 write_lock_bh(&bond->lock);
2166 }
2167
2168 /* zero the mac address of the master so it will be
2169 * set by the application to the mac address of the
2170 * first slave
2171 */
2172 memset(master->dev_addr, 0, master->addr_len);
2173
2174 printk (KERN_INFO "%s: released all slaves\n", master->name);
2175
2176 out:
2177 write_unlock_bh(&bond->lock);
2178
2179 return err;
2180 }
2181
2182 /* this function is called regularly to monitor each slave's link. */
2183 static void bond_mii_monitor(struct net_device *master)
2184 {
2185 bonding_t *bond = (struct bonding *) master->priv;
2186 slave_t *slave, *bestslave, *oldcurrent;
2187 int slave_died = 0;
2188
2189 read_lock(&bond->lock);
2190
2191 /* we will try to read the link status of each of our slaves, and
2192 * set their IFF_RUNNING flag appropriately. For each slave not
2193 * supporting MII status, we won't do anything so that a user-space
2194 * program could monitor the link itself if needed.
2195 */
2196
2197 bestslave = NULL;
2198 slave = (slave_t *)bond;
2199
2200 read_lock(&bond->ptrlock);
2201 oldcurrent = bond->current_slave;
2202 read_unlock(&bond->ptrlock);
2203
2204 while ((slave = slave->prev) != (slave_t *)bond) {
2205 /* use updelay+1 to match an UP slave even when updelay is 0 */
2206 int mindelay = updelay + 1;
2207 struct net_device *dev = slave->dev;
2208 int link_state;
2209 u16 old_speed = slave->speed;
2210 u8 old_duplex = slave->duplex;
2211
2212 link_state = bond_check_dev_link(dev, 0);
2213
2214 switch (slave->link) {
2215 case BOND_LINK_UP: /* the link was up */
2216 if (link_state == BMSR_LSTATUS) {
2217 /* link stays up, tell that this one
2218 is immediately available */
2219 if (IS_UP(dev) && (mindelay > -2)) {
2220 /* -2 is the best case :
2221 this slave was already up */
2222 mindelay = -2;
2223 bestslave = slave;
2224 }
2225 break;
2226 }
2227 else { /* link going down */
2228 slave->link = BOND_LINK_FAIL;
2229 slave->delay = downdelay;
2230 if (slave->link_failure_count < UINT_MAX) {
2231 slave->link_failure_count++;
2232 }
2233 if (downdelay > 0) {
2234 printk (KERN_INFO
2235 "%s: link status down for %sinterface "
2236 "%s, disabling it in %d ms.\n",
2237 master->name,
2238 IS_UP(dev)
2239 ? ((bond_mode == BOND_MODE_ACTIVEBACKUP)
2240 ? ((slave == oldcurrent)
2241 ? "active " : "backup ")
2242 : "")
2243 : "idle ",
2244 dev->name,
2245 downdelay * miimon);
2246 }
2247 }
2248 /* no break ! fall through the BOND_LINK_FAIL test to
2249 ensure proper action to be taken
2250 */
2251 case BOND_LINK_FAIL: /* the link has just gone down */
2252 if (link_state != BMSR_LSTATUS) {
2253 /* link stays down */
2254 if (slave->delay <= 0) {
2255 /* link down for too long time */
2256 slave->link = BOND_LINK_DOWN;
2257 /* in active/backup mode, we must
2258 * completely disable this interface
2259 */
2260 if ((bond_mode == BOND_MODE_ACTIVEBACKUP) ||
2261 (bond_mode == BOND_MODE_8023AD)) {
2262 bond_set_slave_inactive_flags(slave);
2263 }
2264 printk(KERN_INFO
2265 "%s: link status definitely down "
2266 "for interface %s, disabling it",
2267 master->name,
2268 dev->name);
2269
2270 /* notify ad that the link status has changed */
2271 if (bond_mode == BOND_MODE_8023AD) {
2272 bond_3ad_handle_link_change(slave, BOND_LINK_DOWN);
2273 }
2274
2275 if ((bond_mode == BOND_MODE_TLB) ||
2276 (bond_mode == BOND_MODE_ALB)) {
2277 bond_alb_handle_link_change(bond, slave, BOND_LINK_DOWN);
2278 }
2279
2280 write_lock(&bond->ptrlock);
2281 if (slave == bond->current_slave) {
2282 /* find a new interface and be verbose */
2283 change_active_interface(bond);
2284 } else {
2285 printk(".\n");
2286 }
2287 write_unlock(&bond->ptrlock);
2288 slave_died = 1;
2289 } else {
2290 slave->delay--;
2291 }
2292 } else {
2293 /* link up again */
2294 slave->link = BOND_LINK_UP;
2295 slave->jiffies = jiffies;
2296 printk(KERN_INFO
2297 "%s: link status up again after %d ms "
2298 "for interface %s.\n",
2299 master->name,
2300 (downdelay - slave->delay) * miimon,
2301 dev->name);
2302
2303 if (IS_UP(dev) && (mindelay > -1)) {
2304 /* -1 is a good case : this slave went
2305 down only for a short time */
2306 mindelay = -1;
2307 bestslave = slave;
2308 }
2309 }
2310 break;
2311 case BOND_LINK_DOWN: /* the link was down */
2312 if (link_state != BMSR_LSTATUS) {
2313 /* the link stays down, nothing more to do */
2314 break;
2315 } else { /* link going up */
2316 slave->link = BOND_LINK_BACK;
2317 slave->delay = updelay;
2318
2319 if (updelay > 0) {
2320 /* if updelay == 0, no need to
2321 advertise about a 0 ms delay */
2322 printk (KERN_INFO
2323 "%s: link status up for interface"
2324 " %s, enabling it in %d ms.\n",
2325 master->name,
2326 dev->name,
2327 updelay * miimon);
2328 }
2329 }
2330 /* no break ! fall through the BOND_LINK_BACK state in
2331 case there's something to do.
2332 */
2333 case BOND_LINK_BACK: /* the link has just come back */
2334 if (link_state != BMSR_LSTATUS) {
2335 /* link down again */
2336 slave->link = BOND_LINK_DOWN;
2337 printk(KERN_INFO
2338 "%s: link status down again after %d ms "
2339 "for interface %s.\n",
2340 master->name,
2341 (updelay - slave->delay) * miimon,
2342 dev->name);
2343 } else {
2344 /* link stays up */
2345 if (slave->delay == 0) {
2346 /* now the link has been up for long time enough */
2347 slave->link = BOND_LINK_UP;
2348 slave->jiffies = jiffies;
2349
2350 if (bond_mode == BOND_MODE_8023AD) {
2351 /* prevent it from being the active one */
2352 slave->state = BOND_STATE_BACKUP;
2353 }
2354 else if (bond_mode != BOND_MODE_ACTIVEBACKUP) {
2355 /* make it immediately active */
2356 slave->state = BOND_STATE_ACTIVE;
2357 } else if (slave != bond->primary_slave) {
2358 /* prevent it from being the active one */
2359 slave->state = BOND_STATE_BACKUP;
2360 }
2361
2362 printk(KERN_INFO
2363 "%s: link status definitely up "
2364 "for interface %s.\n",
2365 master->name,
2366 dev->name);
2367
2368 /* notify ad that the link status has changed */
2369 if (bond_mode == BOND_MODE_8023AD) {
2370 bond_3ad_handle_link_change(slave, BOND_LINK_UP);
2371 }
2372
2373 if ((bond_mode == BOND_MODE_TLB) ||
2374 (bond_mode == BOND_MODE_ALB)) {
2375 bond_alb_handle_link_change(bond, slave, BOND_LINK_UP);
2376 }
2377
2378 write_lock(&bond->ptrlock);
2379 if ( (bond->primary_slave != NULL)
2380 && (slave == bond->primary_slave) )
2381 change_active_interface(bond);
2382 write_unlock(&bond->ptrlock);
2383 }
2384 else
2385 slave->delay--;
2386
2387 /* we'll also look for the mostly eligible slave */
2388 if (bond->primary_slave == NULL) {
2389 if (IS_UP(dev) && (slave->delay < mindelay)) {
2390 mindelay = slave->delay;
2391 bestslave = slave;
2392 }
2393 } else if ( (IS_UP(bond->primary_slave->dev)) ||
2394 ( (!IS_UP(bond->primary_slave->dev)) &&
2395 (IS_UP(dev) && (slave->delay < mindelay)) ) ) {
2396 mindelay = slave->delay;
2397 bestslave = slave;
2398 }
2399 }
2400 break;
2401 } /* end of switch */
2402
2403 bond_update_speed_duplex(slave);
2404
2405 if (bond_mode == BOND_MODE_8023AD) {
2406 if (old_speed != slave->speed) {
2407 bond_3ad_adapter_speed_changed(slave);
2408 }
2409 if (old_duplex != slave->duplex) {
2410 bond_3ad_adapter_duplex_changed(slave);
2411 }
2412 }
2413
2414 } /* end of while */
2415
2416 /*
2417 * if there's no active interface and we discovered that one
2418 * of the slaves could be activated earlier, so we do it.
2419 */
2420 read_lock(&bond->ptrlock);
2421 oldcurrent = bond->current_slave;
2422 read_unlock(&bond->ptrlock);
2423
2424 /* no active interface at the moment or need to bring up the primary */
2425 if (oldcurrent == NULL) { /* no active interface at the moment */
2426 if (bestslave != NULL) { /* last chance to find one ? */
2427 if (bestslave->link == BOND_LINK_UP) {
2428 printk (KERN_INFO
2429 "%s: making interface %s the new active one.\n",
2430 master->name, bestslave->dev->name);
2431 } else {
2432 printk (KERN_INFO
2433 "%s: making interface %s the new "
2434 "active one %d ms earlier.\n",
2435 master->name, bestslave->dev->name,
2436 (updelay - bestslave->delay) * miimon);
2437
2438 bestslave->delay = 0;
2439 bestslave->link = BOND_LINK_UP;
2440 bestslave->jiffies = jiffies;
2441
2442 /* notify ad that the link status has changed */
2443 if (bond_mode == BOND_MODE_8023AD) {
2444 bond_3ad_handle_link_change(bestslave, BOND_LINK_UP);
2445 }
2446
2447 if ((bond_mode == BOND_MODE_TLB) ||
2448 (bond_mode == BOND_MODE_ALB)) {
2449 bond_alb_handle_link_change(bond, bestslave, BOND_LINK_UP);
2450 }
2451 }
2452
2453 if (bond_mode == BOND_MODE_ACTIVEBACKUP) {
2454 bond_set_slave_active_flags(bestslave);
2455 bond_mc_update(bond, bestslave, NULL);
2456 } else if (bond_mode != BOND_MODE_8023AD) {
2457 bestslave->state = BOND_STATE_ACTIVE;
2458 }
2459 write_lock(&bond->ptrlock);
2460 bond_assign_current_slave(bond, bestslave);
2461 write_unlock(&bond->ptrlock);
2462 } else if (slave_died) {
2463 /* print this message only once a slave has just died */
2464 printk(KERN_INFO
2465 "%s: now running without any active interface !\n",
2466 master->name);
2467 }
2468 }
2469
2470 read_unlock(&bond->lock);
2471 /* re-arm the timer */
2472 mod_timer(&bond->mii_timer, jiffies + (miimon * HZ / 1000));
2473 }
2474
2475 /*
2476 * this function is called regularly to monitor each slave's link
2477 * ensuring that traffic is being sent and received when arp monitoring
2478 * is used in load-balancing mode. if the adapter has been dormant, then an
2479 * arp is transmitted to generate traffic. see activebackup_arp_monitor for
2480 * arp monitoring in active backup mode.
2481 */
2482 static void loadbalance_arp_monitor(struct net_device *master)
2483 {
2484 bonding_t *bond;
2485 slave_t *slave;
2486 int the_delta_in_ticks = arp_interval * HZ / 1000;
2487 int next_timer = jiffies + (arp_interval * HZ / 1000);
2488
2489 bond = (struct bonding *) master->priv;
2490 if (master->priv == NULL) {
2491 mod_timer(&bond->arp_timer, next_timer);
2492 return;
2493 }
2494
2495 /* TODO: investigate why rtnl_shlock_nowait and rtnl_exlock_nowait
2496 * are called below and add comment why they are required...
2497 */
2498 if ((!IS_UP(master)) || rtnl_shlock_nowait()) {
2499 mod_timer(&bond->arp_timer, next_timer);
2500 return;
2501 }
2502
2503 if (rtnl_exlock_nowait()) {
2504 rtnl_shunlock();
2505 mod_timer(&bond->arp_timer, next_timer);
2506 return;
2507 }
2508
2509 read_lock(&bond->lock);
2510
2511 /* see if any of the previous devices are up now (i.e. they have
2512 * xmt and rcv traffic). the current_slave does not come into
2513 * the picture unless it is null. also, slave->jiffies is not needed
2514 * here because we send an arp on each slave and give a slave as
2515 * long as it needs to get the tx/rx within the delta.
2516 * TODO: what about up/down delay in arp mode? it wasn't here before
2517 * so it can wait
2518 */
2519 slave = (slave_t *)bond;
2520 while ((slave = slave->prev) != (slave_t *)bond) {
2521
2522 if (slave->link != BOND_LINK_UP) {
2523
2524 if (((jiffies - slave->dev->trans_start) <=
2525 the_delta_in_ticks) &&
2526 ((jiffies - slave->dev->last_rx) <=
2527 the_delta_in_ticks)) {
2528
2529 slave->link = BOND_LINK_UP;
2530 slave->state = BOND_STATE_ACTIVE;
2531
2532 /* primary_slave has no meaning in round-robin
2533 * mode. the window of a slave being up and
2534 * current_slave being null after enslaving
2535 * is closed.
2536 */
2537 write_lock(&bond->ptrlock);
2538 if (bond->current_slave == NULL) {
2539 printk(KERN_INFO
2540 "%s: link status definitely up "
2541 "for interface %s, ",
2542 master->name,
2543 slave->dev->name);
2544 change_active_interface(bond);
2545 } else {
2546 printk(KERN_INFO
2547 "%s: interface %s is now up\n",
2548 master->name,
2549 slave->dev->name);
2550 }
2551 write_unlock(&bond->ptrlock);
2552 }
2553 } else {
2554 /* slave->link == BOND_LINK_UP */
2555
2556 /* not all switches will respond to an arp request
2557 * when the source ip is 0, so don't take the link down
2558 * if we don't know our ip yet
2559 */
2560 if (((jiffies - slave->dev->trans_start) >=
2561 (2*the_delta_in_ticks)) ||
2562 (((jiffies - slave->dev->last_rx) >=
2563 (2*the_delta_in_ticks)) && my_ip !=0)) {
2564 slave->link = BOND_LINK_DOWN;
2565 slave->state = BOND_STATE_BACKUP;
2566 if (slave->link_failure_count < UINT_MAX) {
2567 slave->link_failure_count++;
2568 }
2569 printk(KERN_INFO
2570 "%s: interface %s is now down.\n",
2571 master->name,
2572 slave->dev->name);
2573
2574 write_lock(&bond->ptrlock);
2575 if (slave == bond->current_slave) {
2576 change_active_interface(bond);
2577 }
2578 write_unlock(&bond->ptrlock);
2579 }
2580 }
2581
2582 /* note: if switch is in round-robin mode, all links
2583 * must tx arp to ensure all links rx an arp - otherwise
2584 * links may oscillate or not come up at all; if switch is
2585 * in something like xor mode, there is nothing we can
2586 * do - all replies will be rx'ed on same link causing slaves
2587 * to be unstable during low/no traffic periods
2588 */
2589 if (IS_UP(slave->dev)) {
2590 arp_send_all(slave);
2591 }
2592 }
2593
2594 read_unlock(&bond->lock);
2595 rtnl_exunlock();
2596 rtnl_shunlock();
2597
2598 /* re-arm the timer */
2599 mod_timer(&bond->arp_timer, next_timer);
2600 }
2601
2602 /*
2603 * When using arp monitoring in active-backup mode, this function is
2604 * called to determine if any backup slaves have went down or a new
2605 * current slave needs to be found.
2606 * The backup slaves never generate traffic, they are considered up by merely
2607 * receiving traffic. If the current slave goes down, each backup slave will
2608 * be given the opportunity to tx/rx an arp before being taken down - this
2609 * prevents all slaves from being taken down due to the current slave not
2610 * sending any traffic for the backups to receive. The arps are not necessarily
2611 * necessary, any tx and rx traffic will keep the current slave up. While any
2612 * rx traffic will keep the backup slaves up, the current slave is responsible
2613 * for generating traffic to keep them up regardless of any other traffic they
2614 * may have received.
2615 * see loadbalance_arp_monitor for arp monitoring in load balancing mode
2616 */
2617 static void activebackup_arp_monitor(struct net_device *master)
2618 {
2619 bonding_t *bond;
2620 slave_t *slave;
2621 int the_delta_in_ticks = arp_interval * HZ / 1000;
2622 int next_timer = jiffies + (arp_interval * HZ / 1000);
2623
2624 bond = (struct bonding *) master->priv;
2625 if (master->priv == NULL) {
2626 mod_timer(&bond->arp_timer, next_timer);
2627 return;
2628 }
2629
2630 if (!IS_UP(master)) {
2631 mod_timer(&bond->arp_timer, next_timer);
2632 return;
2633 }
2634
2635 read_lock(&bond->lock);
2636
2637 /* determine if any slave has come up or any backup slave has
2638 * gone down
2639 * TODO: what about up/down delay in arp mode? it wasn't here before
2640 * so it can wait
2641 */
2642 slave = (slave_t *)bond;
2643 while ((slave = slave->prev) != (slave_t *)bond) {
2644
2645 if (slave->link != BOND_LINK_UP) {
2646 if ((jiffies - slave->dev->last_rx) <=
2647 the_delta_in_ticks) {
2648
2649 slave->link = BOND_LINK_UP;
2650 write_lock(&bond->ptrlock);
2651 if ((bond->current_slave == NULL) &&
2652 ((jiffies - slave->dev->trans_start) <=
2653 the_delta_in_ticks)) {
2654 bond_assign_current_slave(bond, slave);
2655 bond_set_slave_active_flags(slave);
2656 bond_mc_update(bond, slave, NULL);
2657 bond->current_arp_slave = NULL;
2658 } else if (bond->current_slave != slave) {
2659 /* this slave has just come up but we
2660 * already have a current slave; this
2661 * can also happen if bond_enslave adds
2662 * a new slave that is up while we are
2663 * searching for a new slave
2664 */
2665 bond_set_slave_inactive_flags(slave);
2666 bond->current_arp_slave = NULL;
2667 }
2668
2669 if (slave == bond->current_slave) {
2670 printk(KERN_INFO
2671 "%s: %s is up and now the "
2672 "active interface\n",
2673 master->name,
2674 slave->dev->name);
2675 } else {
2676 printk(KERN_INFO
2677 "%s: backup interface %s is "
2678 "now up\n",
2679 master->name,
2680 slave->dev->name);
2681 }
2682
2683 write_unlock(&bond->ptrlock);
2684 }
2685 } else {
2686 read_lock(&bond->ptrlock);
2687 if ((slave != bond->current_slave) &&
2688 (bond->current_arp_slave == NULL) &&
2689 (((jiffies - slave->dev->last_rx) >=
2690 3*the_delta_in_ticks) && (my_ip != 0))) {
2691 /* a backup slave has gone down; three times
2692 * the delta allows the current slave to be
2693 * taken out before the backup slave.
2694 * note: a non-null current_arp_slave indicates
2695 * the current_slave went down and we are
2696 * searching for a new one; under this
2697 * condition we only take the current_slave
2698 * down - this gives each slave a chance to
2699 * tx/rx traffic before being taken out
2700 */
2701 read_unlock(&bond->ptrlock);
2702 slave->link = BOND_LINK_DOWN;
2703 if (slave->link_failure_count < UINT_MAX) {
2704 slave->link_failure_count++;
2705 }
2706 bond_set_slave_inactive_flags(slave);
2707 printk(KERN_INFO
2708 "%s: backup interface %s is now down\n",
2709 master->name,
2710 slave->dev->name);
2711 } else {
2712 read_unlock(&bond->ptrlock);
2713 }
2714 }
2715 }
2716
2717 read_lock(&bond->ptrlock);
2718 slave = bond->current_slave;
2719 read_unlock(&bond->ptrlock);
2720
2721 if (slave != NULL) {
2722
2723 /* if we have sent traffic in the past 2*arp_intervals but
2724 * haven't xmit and rx traffic in that time interval, select
2725 * a different slave. slave->jiffies is only updated when
2726 * a slave first becomes the current_slave - not necessarily
2727 * after every arp; this ensures the slave has a full 2*delta
2728 * before being taken out. if a primary is being used, check
2729 * if it is up and needs to take over as the current_slave
2730 */
2731 if ((((jiffies - slave->dev->trans_start) >=
2732 (2*the_delta_in_ticks)) ||
2733 (((jiffies - slave->dev->last_rx) >=
2734 (2*the_delta_in_ticks)) && (my_ip != 0))) &&
2735 ((jiffies - slave->jiffies) >= 2*the_delta_in_ticks)) {
2736
2737 slave->link = BOND_LINK_DOWN;
2738 if (slave->link_failure_count < UINT_MAX) {
2739 slave->link_failure_count++;
2740 }
2741 printk(KERN_INFO "%s: link status down for "
2742 "active interface %s, disabling it",
2743 master->name,
2744 slave->dev->name);
2745 write_lock(&bond->ptrlock);
2746 slave = change_active_interface(bond);
2747 write_unlock(&bond->ptrlock);
2748 bond->current_arp_slave = slave;
2749 if (slave != NULL) {
2750 slave->jiffies = jiffies;
2751 }
2752
2753 } else if ((bond->primary_slave != NULL) &&
2754 (bond->primary_slave != slave) &&
2755 (bond->primary_slave->link == BOND_LINK_UP)) {
2756 /* at this point, slave is the current_slave */
2757 printk(KERN_INFO
2758 "%s: changing from interface %s to primary "
2759 "interface %s\n",
2760 master->name,
2761 slave->dev->name,
2762 bond->primary_slave->dev->name);
2763
2764 /* primary is up so switch to it */
2765 bond_set_slave_inactive_flags(slave);
2766 bond_mc_update(bond, bond->primary_slave, slave);
2767 write_lock(&bond->ptrlock);
2768 bond_assign_current_slave(bond, bond->primary_slave);
2769 write_unlock(&bond->ptrlock);
2770 slave = bond->primary_slave;
2771 bond_set_slave_active_flags(slave);
2772 slave->jiffies = jiffies;
2773 } else {
2774 bond->current_arp_slave = NULL;
2775 }
2776
2777 /* the current slave must tx an arp to ensure backup slaves
2778 * rx traffic
2779 */
2780 if ((slave != NULL) && (my_ip != 0)) {
2781 arp_send_all(slave);
2782 }
2783 }
2784
2785 /* if we don't have a current_slave, search for the next available
2786 * backup slave from the current_arp_slave and make it the candidate
2787 * for becoming the current_slave
2788 */
2789 if (slave == NULL) {
2790
2791 if ((bond->current_arp_slave == NULL) ||
2792 (bond->current_arp_slave == (slave_t *)bond)) {
2793 bond->current_arp_slave = bond->prev;
2794 }
2795
2796 if (bond->current_arp_slave != (slave_t *)bond) {
2797 bond_set_slave_inactive_flags(bond->current_arp_slave);
2798 slave = bond->current_arp_slave->next;
2799
2800 /* search for next candidate */
2801 do {
2802 if (IS_UP(slave->dev)) {
2803 slave->link = BOND_LINK_BACK;
2804 bond_set_slave_active_flags(slave);
2805 arp_send_all(slave);
2806 slave->jiffies = jiffies;
2807 bond->current_arp_slave = slave;
2808 break;
2809 }
2810
2811 /* if the link state is up at this point, we
2812 * mark it down - this can happen if we have
2813 * simultaneous link failures and
2814 * change_active_interface doesn't make this
2815 * one the current slave so it is still marked
2816 * up when it is actually down
2817 */
2818 if (slave->link == BOND_LINK_UP) {
2819 slave->link = BOND_LINK_DOWN;
2820 if (slave->link_failure_count <
2821 UINT_MAX) {
2822 slave->link_failure_count++;
2823 }
2824
2825 bond_set_slave_inactive_flags(slave);
2826 printk(KERN_INFO
2827 "%s: backup interface "
2828 "%s is now down.\n",
2829 master->name,
2830 slave->dev->name);
2831 }
2832 } while ((slave = slave->next) !=
2833 bond->current_arp_slave->next);
2834 }
2835 }
2836
2837 read_unlock(&bond->lock);
2838 mod_timer(&bond->arp_timer, next_timer);
2839 }
2840
2841 static int bond_sethwaddr(struct net_device *master, struct net_device *slave)
2842 {
2843 #ifdef BONDING_DEBUG
2844 printk(KERN_CRIT "bond_sethwaddr: master=%x\n", (unsigned int)master);
2845 printk(KERN_CRIT "bond_sethwaddr: slave=%x\n", (unsigned int)slave);
2846 printk(KERN_CRIT "bond_sethwaddr: slave->addr_len=%d\n", slave->addr_len);
2847 #endif
2848 memcpy(master->dev_addr, slave->dev_addr, slave->addr_len);
2849 return 0;
2850 }
2851
2852 static int bond_info_query(struct net_device *master, struct ifbond *info)
2853 {
2854 bonding_t *bond = (struct bonding *) master->priv;
2855 slave_t *slave;
2856
2857 info->bond_mode = bond_mode;
2858 info->num_slaves = 0;
2859 info->miimon = miimon;
2860
2861 read_lock_bh(&bond->lock);
2862 for (slave = bond->prev; slave != (slave_t *)bond; slave = slave->prev) {
2863 info->num_slaves++;
2864 }
2865 read_unlock_bh(&bond->lock);
2866
2867 return 0;
2868 }
2869
2870 static int bond_slave_info_query(struct net_device *master,
2871 struct ifslave *info)
2872 {
2873 bonding_t *bond = (struct bonding *) master->priv;
2874 slave_t *slave;
2875 int cur_ndx = 0;
2876
2877 if (info->slave_id < 0) {
2878 return -ENODEV;
2879 }
2880
2881 read_lock_bh(&bond->lock);
2882 for (slave = bond->prev;
2883 slave != (slave_t *)bond && cur_ndx < info->slave_id;
2884 slave = slave->prev) {
2885 cur_ndx++;
2886 }
2887 read_unlock_bh(&bond->lock);
2888
2889 if (slave != (slave_t *)bond) {
2890 strcpy(info->slave_name, slave->dev->name);
2891 info->link = slave->link;
2892 info->state = slave->state;
2893 info->link_failure_count = slave->link_failure_count;
2894 } else {
2895 return -ENODEV;
2896 }
2897
2898 return 0;
2899 }
2900
2901 static int bond_ethtool_ioctl(struct net_device *master_dev, struct ifreq *ifr)
2902 {
2903 void *addr = ifr->ifr_data;
2904 uint32_t cmd;
2905
2906 if (get_user(cmd, (uint32_t *) addr))
2907 return -EFAULT;
2908
2909 switch (cmd) {
2910
2911 case ETHTOOL_GDRVINFO:
2912 {
2913 struct ethtool_drvinfo info;
2914 char *endptr;
2915
2916 if (copy_from_user(&info, addr, sizeof(info)))
2917 return -EFAULT;
2918
2919 if (strcmp(info.driver, "ifenslave") == 0) {
2920 int new_abi_ver;
2921
2922 new_abi_ver = simple_strtoul(info.fw_version,
2923 &endptr, 0);
2924 if (*endptr) {
2925 printk(KERN_ERR
2926 "bonding: Error: got invalid ABI"
2927 " version from application\n");
2928
2929 return -EINVAL;
2930 }
2931
2932 if (orig_app_abi_ver == -1) {
2933 orig_app_abi_ver = new_abi_ver;
2934 }
2935
2936 app_abi_ver = new_abi_ver;
2937 }
2938
2939 strncpy(info.driver, DRV_NAME, 32);
2940 strncpy(info.version, DRV_VERSION, 32);
2941 snprintf(info.fw_version, 32, "%d", BOND_ABI_VERSION);
2942
2943 if (copy_to_user(addr, &info, sizeof(info)))
2944 return -EFAULT;
2945
2946 return 0;
2947 }
2948 break;
2949 default:
2950 return -EOPNOTSUPP;
2951 }
2952 }
2953
2954 static int bond_ioctl(struct net_device *master_dev, struct ifreq *ifr, int cmd)
2955 {
2956 struct net_device *slave_dev = NULL;
2957 struct ifbond *u_binfo = NULL, k_binfo;
2958 struct ifslave *u_sinfo = NULL, k_sinfo;
2959 struct mii_ioctl_data *mii = NULL;
2960 int prev_abi_ver = orig_app_abi_ver;
2961 int ret = 0;
2962
2963 #ifdef BONDING_DEBUG
2964 printk(KERN_INFO "bond_ioctl: master=%s, cmd=%d\n",
2965 master_dev->name, cmd);
2966 #endif
2967
2968 switch (cmd) {
2969 case SIOCETHTOOL:
2970 return bond_ethtool_ioctl(master_dev, ifr);
2971
2972 case SIOCGMIIPHY:
2973 mii = (struct mii_ioctl_data *)&ifr->ifr_data;
2974 if (mii == NULL) {
2975 return -EINVAL;
2976 }
2977 mii->phy_id = 0;
2978 /* Fall Through */
2979 case SIOCGMIIREG:
2980 /*
2981 * We do this again just in case we were called by SIOCGMIIREG
2982 * instead of SIOCGMIIPHY.
2983 */
2984 mii = (struct mii_ioctl_data *)&ifr->ifr_data;
2985 if (mii == NULL) {
2986 return -EINVAL;
2987 }
2988 if (mii->reg_num == 1) {
2989 mii->val_out = bond_check_mii_link(
2990 (struct bonding *)master_dev->priv);
2991 }
2992 return 0;
2993 case BOND_INFO_QUERY_OLD:
2994 case SIOCBONDINFOQUERY:
2995 u_binfo = (struct ifbond *)ifr->ifr_data;
2996 if (copy_from_user(&k_binfo, u_binfo, sizeof(ifbond))) {
2997 return -EFAULT;
2998 }
2999 ret = bond_info_query(master_dev, &k_binfo);
3000 if (ret == 0) {
3001 if (copy_to_user(u_binfo, &k_binfo, sizeof(ifbond))) {
3002 return -EFAULT;
3003 }
3004 }
3005 return ret;
3006 case BOND_SLAVE_INFO_QUERY_OLD:
3007 case SIOCBONDSLAVEINFOQUERY:
3008 u_sinfo = (struct ifslave *)ifr->ifr_data;
3009 if (copy_from_user(&k_sinfo, u_sinfo, sizeof(ifslave))) {
3010 return -EFAULT;
3011 }
3012 ret = bond_slave_info_query(master_dev, &k_sinfo);
3013 if (ret == 0) {
3014 if (copy_to_user(u_sinfo, &k_sinfo, sizeof(ifslave))) {
3015 return -EFAULT;
3016 }
3017 }
3018 return ret;
3019 }
3020
3021 if (!capable(CAP_NET_ADMIN)) {
3022 return -EPERM;
3023 }
3024
3025 if (orig_app_abi_ver == -1) {
3026 /* no orig_app_abi_ver was provided yet, so we'll use the
3027 * current one from now on, even if it's 0
3028 */
3029 orig_app_abi_ver = app_abi_ver;
3030
3031 } else if (orig_app_abi_ver != app_abi_ver) {
3032 printk(KERN_ERR
3033 "bonding: Error: already using ifenslave ABI "
3034 "version %d; to upgrade ifenslave to version %d, "
3035 "you must first reload bonding.\n",
3036 orig_app_abi_ver, app_abi_ver);
3037 return -EINVAL;
3038 }
3039
3040 slave_dev = dev_get_by_name(ifr->ifr_slave);
3041
3042 #ifdef BONDING_DEBUG
3043 printk(KERN_INFO "slave_dev=%x: \n", (unsigned int)slave_dev);
3044 printk(KERN_INFO "slave_dev->name=%s: \n", slave_dev->name);
3045 #endif
3046
3047 if (slave_dev == NULL) {
3048 ret = -ENODEV;
3049 } else {
3050 switch (cmd) {
3051 case BOND_ENSLAVE_OLD:
3052 case SIOCBONDENSLAVE:
3053 ret = bond_enslave(master_dev, slave_dev);
3054 break;
3055 case BOND_RELEASE_OLD:
3056 case SIOCBONDRELEASE:
3057 ret = bond_release(master_dev, slave_dev);
3058 break;
3059 case BOND_SETHWADDR_OLD:
3060 case SIOCBONDSETHWADDR:
3061 ret = bond_sethwaddr(master_dev, slave_dev);
3062 break;
3063 case BOND_CHANGE_ACTIVE_OLD:
3064 case SIOCBONDCHANGEACTIVE:
3065 if ((bond_mode == BOND_MODE_ACTIVEBACKUP) ||
3066 (bond_mode == BOND_MODE_TLB) ||
3067 (bond_mode == BOND_MODE_ALB)) {
3068 ret = bond_change_active(master_dev, slave_dev);
3069 }
3070 else {
3071 ret = -EINVAL;
3072 }
3073 break;
3074 default:
3075 ret = -EOPNOTSUPP;
3076 }
3077 dev_put(slave_dev);
3078 }
3079
3080 if (ret < 0) {
3081 /* The ioctl failed, so there's no point in changing the
3082 * orig_app_abi_ver. We'll restore it's value just in case
3083 * we've changed it earlier in this function.
3084 */
3085 orig_app_abi_ver = prev_abi_ver;
3086 }
3087
3088 return ret;
3089 }
3090
3091 #ifdef CONFIG_NET_FASTROUTE
3092 static int bond_accept_fastpath(struct net_device *dev, struct dst_entry *dst)
3093 {
3094 return -1;
3095 }
3096 #endif
3097
3098 /*
3099 * in broadcast mode, we send everything to all usable interfaces.
3100 */
3101 static int bond_xmit_broadcast(struct sk_buff *skb, struct net_device *dev)
3102 {
3103 slave_t *slave, *start_at;
3104 struct bonding *bond = (struct bonding *) dev->priv;
3105 struct net_device *device_we_should_send_to = 0;
3106
3107 if (!IS_UP(dev)) { /* bond down */
3108 dev_kfree_skb(skb);
3109 return 0;
3110 }
3111
3112 read_lock(&bond->lock);
3113
3114 read_lock(&bond->ptrlock);
3115 slave = start_at = bond->current_slave;
3116 read_unlock(&bond->ptrlock);
3117
3118 if (slave == NULL) { /* we're at the root, get the first slave */
3119 /* no suitable interface, frame not sent */
3120 read_unlock(&bond->lock);
3121 dev_kfree_skb(skb);
3122 return 0;
3123 }
3124
3125 do {
3126 if (IS_UP(slave->dev)
3127 && (slave->link == BOND_LINK_UP)
3128 && (slave->state == BOND_STATE_ACTIVE)) {
3129 if (device_we_should_send_to) {
3130 struct sk_buff *skb2;
3131 if ((skb2 = skb_clone(skb, GFP_ATOMIC)) == NULL) {
3132 printk(KERN_ERR "bond_xmit_broadcast: skb_clone() failed\n");
3133 continue;
3134 }
3135
3136 skb2->dev = device_we_should_send_to;
3137 skb2->priority = 1;
3138 dev_queue_xmit(skb2);
3139 }
3140 device_we_should_send_to = slave->dev;
3141 }
3142 } while ((slave = slave->next) != start_at);
3143
3144 if (device_we_should_send_to) {
3145 skb->dev = device_we_should_send_to;
3146 skb->priority = 1;
3147 dev_queue_xmit(skb);
3148 } else
3149 dev_kfree_skb(skb);
3150
3151 /* frame sent to all suitable interfaces */
3152 read_unlock(&bond->lock);
3153 return 0;
3154 }
3155
3156 static int bond_xmit_roundrobin(struct sk_buff *skb, struct net_device *dev)
3157 {
3158 slave_t *slave, *start_at;
3159 struct bonding *bond = (struct bonding *) dev->priv;
3160
3161 if (!IS_UP(dev)) { /* bond down */
3162 dev_kfree_skb(skb);
3163 return 0;
3164 }
3165
3166 read_lock(&bond->lock);
3167
3168 read_lock(&bond->ptrlock);
3169 slave = start_at = bond->current_slave;
3170 read_unlock(&bond->ptrlock);
3171
3172 if (slave == NULL) { /* we're at the root, get the first slave */
3173 /* no suitable interface, frame not sent */
3174 dev_kfree_skb(skb);
3175 read_unlock(&bond->lock);
3176 return 0;
3177 }
3178
3179 do {
3180 if (IS_UP(slave->dev)
3181 && (slave->link == BOND_LINK_UP)
3182 && (slave->state == BOND_STATE_ACTIVE)) {
3183
3184 skb->dev = slave->dev;
3185 skb->priority = 1;
3186 dev_queue_xmit(skb);
3187
3188 write_lock(&bond->ptrlock);
3189 bond_assign_current_slave(bond, slave->next);
3190 write_unlock(&bond->ptrlock);
3191
3192 read_unlock(&bond->lock);
3193 return 0;
3194 }
3195 } while ((slave = slave->next) != start_at);
3196
3197 /* no suitable interface, frame not sent */
3198 dev_kfree_skb(skb);
3199 read_unlock(&bond->lock);
3200 return 0;
3201 }
3202
3203 /*
3204 * in XOR mode, we determine the output device by performing xor on
3205 * the source and destination hw adresses. If this device is not
3206 * enabled, find the next slave following this xor slave.
3207 */
3208 static int bond_xmit_xor(struct sk_buff *skb, struct net_device *dev)
3209 {
3210 slave_t *slave, *start_at;
3211 struct bonding *bond = (struct bonding *) dev->priv;
3212 struct ethhdr *data = (struct ethhdr *)skb->data;
3213 int slave_no;
3214
3215 if (!IS_UP(dev)) { /* bond down */
3216 dev_kfree_skb(skb);
3217 return 0;
3218 }
3219
3220 read_lock(&bond->lock);
3221 slave = bond->prev;
3222
3223 /* we're at the root, get the first slave */
3224 if (bond->slave_cnt == 0) {
3225 /* no suitable interface, frame not sent */
3226 dev_kfree_skb(skb);
3227 read_unlock(&bond->lock);
3228 return 0;
3229 }
3230
3231 slave_no = (data->h_dest[5]^slave->dev->dev_addr[5]) % bond->slave_cnt;
3232
3233 while ( (slave_no > 0) && (slave != (slave_t *)bond) ) {
3234 slave = slave->prev;
3235 slave_no--;
3236 }
3237 start_at = slave;
3238
3239 do {
3240 if (IS_UP(slave->dev)
3241 && (slave->link == BOND_LINK_UP)
3242 && (slave->state == BOND_STATE_ACTIVE)) {
3243
3244 skb->dev = slave->dev;
3245 skb->priority = 1;
3246 dev_queue_xmit(skb);
3247
3248 read_unlock(&bond->lock);
3249 return 0;
3250 }
3251 } while ((slave = slave->next) != start_at);
3252
3253 /* no suitable interface, frame not sent */
3254 dev_kfree_skb(skb);
3255 read_unlock(&bond->lock);
3256 return 0;
3257 }
3258
3259 /*
3260 * in active-backup mode, we know that bond->current_slave is always valid if
3261 * the bond has a usable interface.
3262 */
3263 static int bond_xmit_activebackup(struct sk_buff *skb, struct net_device *dev)
3264 {
3265 struct bonding *bond = (struct bonding *) dev->priv;
3266 int ret;
3267
3268 if (!IS_UP(dev)) { /* bond down */
3269 dev_kfree_skb(skb);
3270 return 0;
3271 }
3272
3273 /* if we are sending arp packets, try to at least
3274 identify our own ip address */
3275 if ( (arp_interval > 0) && (my_ip == 0) &&
3276 (skb->protocol == __constant_htons(ETH_P_ARP) ) ) {
3277 char *the_ip = (((char *)skb->data))
3278 + sizeof(struct ethhdr)
3279 + sizeof(struct arphdr) +
3280 ETH_ALEN;
3281 memcpy(&my_ip, the_ip, 4);
3282 }
3283
3284 /* if we are sending arp packets and don't know
3285 * the target hw address, save it so we don't need
3286 * to use a broadcast address.
3287 * don't do this if in active backup mode because the slaves must
3288 * receive packets to stay up, and the only ones they receive are
3289 * broadcasts.
3290 */
3291 if ( (bond_mode != BOND_MODE_ACTIVEBACKUP) &&
3292 (arp_ip_count == 1) &&
3293 (arp_interval > 0) && (arp_target_hw_addr == NULL) &&
3294 (skb->protocol == __constant_htons(ETH_P_IP) ) ) {
3295 struct ethhdr *eth_hdr =
3296 (struct ethhdr *) (((char *)skb->data));
3297 struct iphdr *ip_hdr = (struct iphdr *)(eth_hdr + 1);
3298
3299 if (arp_target[0] == ip_hdr->daddr) {
3300 arp_target_hw_addr = kmalloc(ETH_ALEN, GFP_KERNEL);
3301 if (arp_target_hw_addr != NULL)
3302 memcpy(arp_target_hw_addr, eth_hdr->h_dest, ETH_ALEN);
3303 }
3304 }
3305
3306 read_lock(&bond->lock);
3307
3308 read_lock(&bond->ptrlock);
3309 if (bond->current_slave != NULL) { /* one usable interface */
3310 skb->dev = bond->current_slave->dev;
3311 read_unlock(&bond->ptrlock);
3312 skb->priority = 1;
3313 ret = dev_queue_xmit(skb);
3314 read_unlock(&bond->lock);
3315 return 0;
3316 }
3317 else {
3318 read_unlock(&bond->ptrlock);
3319 }
3320
3321 /* no suitable interface, frame not sent */
3322 #ifdef BONDING_DEBUG
3323 printk(KERN_INFO "There was no suitable interface, so we don't transmit\n");
3324 #endif
3325 dev_kfree_skb(skb);
3326 read_unlock(&bond->lock);
3327 return 0;
3328 }
3329
3330 static struct net_device_stats *bond_get_stats(struct net_device *dev)
3331 {
3332 bonding_t *bond = dev->priv;
3333 struct net_device_stats *stats = bond->stats, *sstats;
3334 slave_t *slave;
3335
3336 memset(bond->stats, 0, sizeof(struct net_device_stats));
3337
3338 read_lock_bh(&bond->lock);
3339
3340 for (slave = bond->prev; slave != (slave_t *)bond; slave = slave->prev) {
3341 sstats = slave->dev->get_stats(slave->dev);
3342
3343 stats->rx_packets += sstats->rx_packets;
3344 stats->rx_bytes += sstats->rx_bytes;
3345 stats->rx_errors += sstats->rx_errors;
3346 stats->rx_dropped += sstats->rx_dropped;
3347
3348 stats->tx_packets += sstats->tx_packets;
3349 stats->tx_bytes += sstats->tx_bytes;
3350 stats->tx_errors += sstats->tx_errors;
3351 stats->tx_dropped += sstats->tx_dropped;
3352
3353 stats->multicast += sstats->multicast;
3354 stats->collisions += sstats->collisions;
3355
3356 stats->rx_length_errors += sstats->rx_length_errors;
3357 stats->rx_over_errors += sstats->rx_over_errors;
3358 stats->rx_crc_errors += sstats->rx_crc_errors;
3359 stats->rx_frame_errors += sstats->rx_frame_errors;
3360 stats->rx_fifo_errors += sstats->rx_fifo_errors;
3361 stats->rx_missed_errors += sstats->rx_missed_errors;
3362
3363 stats->tx_aborted_errors += sstats->tx_aborted_errors;
3364 stats->tx_carrier_errors += sstats->tx_carrier_errors;
3365 stats->tx_fifo_errors += sstats->tx_fifo_errors;
3366 stats->tx_heartbeat_errors += sstats->tx_heartbeat_errors;
3367 stats->tx_window_errors += sstats->tx_window_errors;
3368
3369 }
3370
3371 read_unlock_bh(&bond->lock);
3372 return stats;
3373 }
3374
3375 static int bond_get_info(char *buf, char **start, off_t offset, int length)
3376 {
3377 bonding_t *bond = these_bonds;
3378 int len = 0;
3379 off_t begin = 0;
3380 u16 link;
3381 slave_t *slave = NULL;
3382
3383 len += sprintf(buf + len, "%s\n", version);
3384
3385 while (bond != NULL) {
3386 /*
3387 * This function locks the mutex, so we can't lock it until
3388 * afterwards
3389 */
3390 link = bond_check_mii_link(bond);
3391
3392 len += sprintf(buf + len, "Bonding Mode: %s\n",
3393 bond_mode_name());
3394
3395 if ((bond_mode == BOND_MODE_ACTIVEBACKUP) ||
3396 (bond_mode == BOND_MODE_TLB) ||
3397 (bond_mode == BOND_MODE_ALB)) {
3398 read_lock_bh(&bond->lock);
3399 read_lock(&bond->ptrlock);
3400 if (bond->current_slave != NULL) {
3401 len += sprintf(buf + len,
3402 "Currently Active Slave: %s\n",
3403 bond->current_slave->dev->name);
3404 }
3405 read_unlock(&bond->ptrlock);
3406 read_unlock_bh(&bond->lock);
3407 }
3408
3409 len += sprintf(buf + len, "MII Status: ");
3410 len += sprintf(buf + len,
3411 link == BMSR_LSTATUS ? "up\n" : "down\n");
3412 len += sprintf(buf + len, "MII Polling Interval (ms): %d\n",
3413 miimon);
3414 len += sprintf(buf + len, "Up Delay (ms): %d\n",
3415 updelay * miimon);
3416 len += sprintf(buf + len, "Down Delay (ms): %d\n",
3417 downdelay * miimon);
3418 len += sprintf(buf + len, "Multicast Mode: %s\n",
3419 multicast_mode_name());
3420
3421 read_lock_bh(&bond->lock);
3422
3423 if (bond_mode == BOND_MODE_8023AD) {
3424 struct ad_info ad_info;
3425
3426 len += sprintf(buf + len, "\n802.3ad info\n");
3427
3428 if (bond_3ad_get_active_agg_info(bond, &ad_info)) {
3429 len += sprintf(buf + len, "bond %s has no active aggregator\n", bond->device->name);
3430 } else {
3431 len += sprintf(buf + len, "Active Aggregator Info:\n");
3432
3433 len += sprintf(buf + len, "\tAggregator ID: %d\n", ad_info.aggregator_id);
3434 len += sprintf(buf + len, "\tNumber of ports: %d\n", ad_info.ports);
3435 len += sprintf(buf + len, "\tActor Key: %d\n", ad_info.actor_key);
3436 len += sprintf(buf + len, "\tPartner Key: %d\n", ad_info.partner_key);
3437 len += sprintf(buf + len, "\tPartner Mac Address: %02x:%02x:%02x:%02x:%02x:%02x\n",
3438 ad_info.partner_system[0],
3439 ad_info.partner_system[1],
3440 ad_info.partner_system[2],
3441 ad_info.partner_system[3],
3442 ad_info.partner_system[4],
3443 ad_info.partner_system[5]);
3444 }
3445 }
3446
3447 for (slave = bond->prev; slave != (slave_t *)bond;
3448 slave = slave->prev) {
3449 len += sprintf(buf + len, "\nSlave Interface: %s\n", slave->dev->name);
3450
3451 len += sprintf(buf + len, "MII Status: ");
3452
3453 len += sprintf(buf + len,
3454 slave->link == BOND_LINK_UP ?
3455 "up\n" : "down\n");
3456 len += sprintf(buf + len, "Link Failure Count: %d\n",
3457 slave->link_failure_count);
3458
3459 if (app_abi_ver >= 1) {
3460 len += sprintf(buf + len,
3461 "Permanent HW addr: %02x:%02x:%02x:%02x:%02x:%02x\n",
3462 slave->perm_hwaddr[0],
3463 slave->perm_hwaddr[1],
3464 slave->perm_hwaddr[2],
3465 slave->perm_hwaddr[3],
3466 slave->perm_hwaddr[4],
3467 slave->perm_hwaddr[5]);
3468 }
3469
3470 if (bond_mode == BOND_MODE_8023AD) {
3471 struct aggregator *agg = SLAVE_AD_INFO(slave).port.aggregator;
3472
3473 if (agg) {
3474 len += sprintf(buf + len, "Aggregator ID: %d\n",
3475 agg->aggregator_identifier);
3476 } else {
3477 len += sprintf(buf + len, "Aggregator ID: N/A\n");
3478 }
3479 }
3480 }
3481 read_unlock_bh(&bond->lock);
3482
3483 /*
3484 * Figure out the calcs for the /proc/net interface
3485 */
3486 *start = buf + (offset - begin);
3487 len -= (offset - begin);
3488 if (len > length) {
3489 len = length;
3490 }
3491 if (len < 0) {
3492 len = 0;
3493 }
3494
3495
3496 bond = bond->next_bond;
3497 }
3498 return len;
3499 }
3500
3501 static int bond_event(struct notifier_block *this, unsigned long event,
3502 void *ptr)
3503 {
3504 struct bonding *this_bond = (struct bonding *)these_bonds;
3505 struct bonding *last_bond;
3506 struct net_device *event_dev = (struct net_device *)ptr;
3507
3508 /* while there are bonds configured */
3509 while (this_bond != NULL) {
3510 if (this_bond == event_dev->priv ) {
3511 switch (event) {
3512 case NETDEV_UNREGISTER:
3513 /*
3514 * remove this bond from a linked list of
3515 * bonds
3516 */
3517 if (this_bond == these_bonds) {
3518 these_bonds = this_bond->next_bond;
3519 } else {
3520 for (last_bond = these_bonds;
3521 last_bond != NULL;
3522 last_bond = last_bond->next_bond) {
3523 if (last_bond->next_bond ==
3524 this_bond) {
3525 last_bond->next_bond =
3526 this_bond->next_bond;
3527 }
3528 }
3529 }
3530 return NOTIFY_DONE;
3531
3532 default:
3533 return NOTIFY_DONE;
3534 }
3535 } else if (this_bond->device == event_dev->master) {
3536 switch (event) {
3537 case NETDEV_UNREGISTER:
3538 bond_release(this_bond->device, event_dev);
3539 break;
3540 }
3541 return NOTIFY_DONE;
3542 }
3543 this_bond = this_bond->next_bond;
3544 }
3545 return NOTIFY_DONE;
3546 }
3547
3548 static struct notifier_block bond_netdev_notifier = {
3549 notifier_call: bond_event,
3550 };
3551
3552 static int __init bond_init(struct net_device *dev)
3553 {
3554 bonding_t *bond, *this_bond, *last_bond;
3555 int count;
3556
3557 #ifdef BONDING_DEBUG
3558 printk (KERN_INFO "Begin bond_init for %s\n", dev->name);
3559 #endif
3560 bond = kmalloc(sizeof(struct bonding), GFP_KERNEL);
3561 if (bond == NULL) {
3562 return -ENOMEM;
3563 }
3564 memset(bond, 0, sizeof(struct bonding));
3565
3566 /* initialize rwlocks */
3567 rwlock_init(&bond->lock);
3568 rwlock_init(&bond->ptrlock);
3569
3570 bond->stats = kmalloc(sizeof(struct net_device_stats), GFP_KERNEL);
3571 if (bond->stats == NULL) {
3572 kfree(bond);
3573 return -ENOMEM;
3574 }
3575 memset(bond->stats, 0, sizeof(struct net_device_stats));
3576
3577 bond->next = bond->prev = (slave_t *)bond;
3578 bond->current_slave = NULL;
3579 bond->current_arp_slave = NULL;
3580 bond->device = dev;
3581 dev->priv = bond;
3582
3583 /* Initialize the device structure. */
3584 switch (bond_mode) {
3585 case BOND_MODE_ACTIVEBACKUP:
3586 dev->hard_start_xmit = bond_xmit_activebackup;
3587 break;
3588 case BOND_MODE_ROUNDROBIN:
3589 dev->hard_start_xmit = bond_xmit_roundrobin;
3590 break;
3591 case BOND_MODE_XOR:
3592 dev->hard_start_xmit = bond_xmit_xor;
3593 break;
3594 case BOND_MODE_BROADCAST:
3595 dev->hard_start_xmit = bond_xmit_broadcast;
3596 break;
3597 case BOND_MODE_8023AD:
3598 dev->hard_start_xmit = bond_3ad_xmit_xor;
3599 break;
3600 case BOND_MODE_TLB:
3601 case BOND_MODE_ALB:
3602 dev->hard_start_xmit = bond_alb_xmit;
3603 break;
3604 default:
3605 printk(KERN_ERR "Unknown bonding mode %d\n", bond_mode);
3606 kfree(bond->stats);
3607 kfree(bond);
3608 return -EINVAL;
3609 }
3610
3611 dev->get_stats = bond_get_stats;
3612 dev->open = bond_open;
3613 dev->stop = bond_close;
3614 dev->set_multicast_list = set_multicast_list;
3615 dev->do_ioctl = bond_ioctl;
3616
3617 /*
3618 * Fill in the fields of the device structure with ethernet-generic
3619 * values.
3620 */
3621
3622 ether_setup(dev);
3623
3624 dev->set_mac_address = bond_set_mac_address;
3625 dev->tx_queue_len = 0;
3626 dev->flags |= IFF_MASTER|IFF_MULTICAST;
3627 #ifdef CONFIG_NET_FASTROUTE
3628 dev->accept_fastpath = bond_accept_fastpath;
3629 #endif
3630
3631 printk(KERN_INFO "%s registered with", dev->name);
3632 if (miimon > 0) {
3633 printk(" MII link monitoring set to %d ms", miimon);
3634 updelay /= miimon;
3635 downdelay /= miimon;
3636 } else {
3637 printk("out MII link monitoring");
3638 }
3639 printk(", in %s mode.\n", bond_mode_name());
3640
3641 printk(KERN_INFO "%s registered with", dev->name);
3642 if (arp_interval > 0) {
3643 printk(" ARP monitoring set to %d ms with %d target(s):",
3644 arp_interval, arp_ip_count);
3645 for (count=0 ; count<arp_ip_count ; count++)
3646 printk (" %s", arp_ip_target[count]);
3647 printk("\n");
3648 } else {
3649 printk("out ARP monitoring\n");
3650 }
3651
3652 #ifdef CONFIG_PROC_FS
3653 bond->bond_proc_dir = proc_mkdir(dev->name, proc_net);
3654 if (bond->bond_proc_dir == NULL) {
3655 printk(KERN_ERR "%s: Cannot init /proc/net/%s/\n",
3656 dev->name, dev->name);
3657 kfree(bond->stats);
3658 kfree(bond);
3659 return -ENOMEM;
3660 }
3661 bond->bond_proc_info_file =
3662 create_proc_info_entry("info", 0, bond->bond_proc_dir,
3663 bond_get_info);
3664 if (bond->bond_proc_info_file == NULL) {
3665 printk(KERN_ERR "%s: Cannot init /proc/net/%s/info\n",
3666 dev->name, dev->name);
3667 remove_proc_entry(dev->name, proc_net);
3668 kfree(bond->stats);
3669 kfree(bond);
3670 return -ENOMEM;
3671 }
3672 #endif /* CONFIG_PROC_FS */
3673
3674 if (first_pass == 1) {
3675 these_bonds = bond;
3676 register_netdevice_notifier(&bond_netdev_notifier);
3677 first_pass = 0;
3678 } else {
3679 last_bond = these_bonds;
3680 this_bond = these_bonds->next_bond;
3681 while (this_bond != NULL) {
3682 last_bond = this_bond;
3683 this_bond = this_bond->next_bond;
3684 }
3685 last_bond->next_bond = bond;
3686 }
3687
3688 return 0;
3689 }
3690
3691 /*
3692 static int __init bond_probe(struct net_device *dev)
3693 {
3694 bond_init(dev);
3695 return 0;
3696 }
3697 */
3698
3699 /*
3700 * Convert string input module parms. Accept either the
3701 * number of the mode or its string name.
3702 */
3703 static inline int
3704 bond_parse_parm(char *mode_arg, struct bond_parm_tbl *tbl)
3705 {
3706 int i;
3707
3708 for (i = 0; tbl[i].modename != NULL; i++) {
3709 if ((isdigit(*mode_arg) &&
3710 tbl[i].mode == simple_strtol(mode_arg, NULL, 0)) ||
3711 (0 == strncmp(mode_arg, tbl[i].modename,
3712 strlen(tbl[i].modename)))) {
3713 return tbl[i].mode;
3714 }
3715 }
3716
3717 return -1;
3718 }
3719
3720
3721 static int __init bonding_init(void)
3722 {
3723 int no;
3724 int err;
3725
3726 /* Find a name for this unit */
3727 static struct net_device *dev_bond = NULL;
3728
3729 printk(KERN_INFO "%s", version);
3730
3731 /*
3732 * Convert string parameters.
3733 */
3734 if (mode) {
3735 bond_mode = bond_parse_parm(mode, bond_mode_tbl);
3736 if (bond_mode == -1) {
3737 printk(KERN_WARNING
3738 "bonding_init(): Invalid bonding mode \"%s\"\n",
3739 mode == NULL ? "NULL" : mode);
3740 return -EINVAL;
3741 }
3742 }
3743
3744 if (multicast) {
3745 multicast_mode = bond_parse_parm(multicast, bond_mc_tbl);
3746 if (multicast_mode == -1) {
3747 printk(KERN_WARNING
3748 "bonding_init(): Invalid multicast mode \"%s\"\n",
3749 multicast == NULL ? "NULL" : multicast);
3750 return -EINVAL;
3751 }
3752 }
3753
3754 if (lacp_rate) {
3755 if (bond_mode != BOND_MODE_8023AD) {
3756 printk(KERN_WARNING
3757 "lacp_rate param is irrelevant in mode %s\n",
3758 bond_mode_name());
3759 } else {
3760 lacp_fast = bond_parse_parm(lacp_rate, bond_lacp_tbl);
3761 if (lacp_fast == -1) {
3762 printk(KERN_WARNING
3763 "bonding_init(): Invalid lacp rate "
3764 "\"%s\"\n",
3765 lacp_rate == NULL ? "NULL" : lacp_rate);
3766
3767 return -EINVAL;
3768 }
3769 }
3770 }
3771
3772 if (max_bonds < 1 || max_bonds > INT_MAX) {
3773 printk(KERN_WARNING
3774 "bonding_init(): max_bonds (%d) not in range %d-%d, "
3775 "so it was reset to BOND_DEFAULT_MAX_BONDS (%d)",
3776 max_bonds, 1, INT_MAX, BOND_DEFAULT_MAX_BONDS);
3777 max_bonds = BOND_DEFAULT_MAX_BONDS;
3778 }
3779 dev_bond = dev_bonds = kmalloc(max_bonds*sizeof(struct net_device),
3780 GFP_KERNEL);
3781 if (dev_bond == NULL) {
3782 return -ENOMEM;
3783 }
3784 memset(dev_bonds, 0, max_bonds*sizeof(struct net_device));
3785
3786 if (miimon < 0) {
3787 printk(KERN_WARNING
3788 "bonding_init(): miimon module parameter (%d), "
3789 "not in range 0-%d, so it was reset to %d\n",
3790 miimon, INT_MAX, BOND_LINK_MON_INTERV);
3791 miimon = BOND_LINK_MON_INTERV;
3792 }
3793
3794 if (updelay < 0) {
3795 printk(KERN_WARNING
3796 "bonding_init(): updelay module parameter (%d), "
3797 "not in range 0-%d, so it was reset to 0\n",
3798 updelay, INT_MAX);
3799 updelay = 0;
3800 }
3801
3802 if (downdelay < 0) {
3803 printk(KERN_WARNING
3804 "bonding_init(): downdelay module parameter (%d), "
3805 "not in range 0-%d, so it was reset to 0\n",
3806 downdelay, INT_MAX);
3807 downdelay = 0;
3808 }
3809
3810 /* reset values for 802.3ad */
3811 if (bond_mode == BOND_MODE_8023AD) {
3812 if (arp_interval != 0) {
3813 printk(KERN_WARNING "bonding_init(): ARP monitoring"
3814 "can't be used simultaneously with 802.3ad, "
3815 "disabling ARP monitoring\n");
3816 arp_interval = 0;
3817 }
3818
3819 if (miimon == 0) {
3820 printk(KERN_ERR
3821 "bonding_init(): miimon must be specified, "
3822 "otherwise bonding will not detect link failure, "
3823 "speed and duplex which are essential "
3824 "for 802.3ad operation\n");
3825 printk(KERN_ERR "Forcing miimon to 100msec\n");
3826 miimon = 100;
3827 }
3828
3829 if (multicast_mode != BOND_MULTICAST_ALL) {
3830 printk(KERN_ERR
3831 "bonding_init(): Multicast mode must "
3832 "be set to ALL for 802.3ad\n");
3833 printk(KERN_ERR "Forcing Multicast mode to ALL\n");
3834 multicast_mode = BOND_MULTICAST_ALL;
3835 }
3836 }
3837
3838 /* reset values for TLB/ALB */
3839 if ((bond_mode == BOND_MODE_TLB) ||
3840 (bond_mode == BOND_MODE_ALB)) {
3841 if (miimon == 0) {
3842 printk(KERN_ERR
3843 "bonding_init(): miimon must be specified, "
3844 "otherwise bonding will not detect link failure "
3845 "and link speed which are essential "
3846 "for TLB/ALB load balancing\n");
3847 printk(KERN_ERR "Forcing miimon to 100msec\n");
3848 miimon = 100;
3849 }
3850
3851 if (multicast_mode != BOND_MULTICAST_ACTIVE) {
3852 printk(KERN_ERR
3853 "bonding_init(): Multicast mode must "
3854 "be set to ACTIVE for TLB/ALB\n");
3855 printk(KERN_ERR "Forcing Multicast mode to ACTIVE\n");
3856 multicast_mode = BOND_MULTICAST_ACTIVE;
3857 }
3858 }
3859
3860 if (bond_mode == BOND_MODE_ALB) {
3861 printk(KERN_INFO
3862 "In ALB mode you might experience client disconnections"
3863 " upon reconnection of a link if the bonding module"
3864 " updelay parameter (%d msec) is incompatible with the"
3865 " forwarding delay time of the switch\n", updelay);
3866 }
3867
3868 if (miimon == 0) {
3869 if ((updelay != 0) || (downdelay != 0)) {
3870 /* just warn the user the up/down delay will have
3871 * no effect since miimon is zero...
3872 */
3873 printk(KERN_WARNING
3874 "bonding_init(): miimon module parameter not "
3875 "set and updelay (%d) or downdelay (%d) module "
3876 "parameter is set; updelay and downdelay have "
3877 "no effect unless miimon is set\n",
3878 updelay, downdelay);
3879 }
3880 } else {
3881 /* don't allow arp monitoring */
3882 if (arp_interval != 0) {
3883 printk(KERN_WARNING
3884 "bonding_init(): miimon (%d) and arp_interval "
3885 "(%d) can't be used simultaneously, "
3886 "disabling ARP monitoring\n",
3887 miimon, arp_interval);
3888 arp_interval = 0;
3889 }
3890
3891 if ((updelay % miimon) != 0) {
3892 /* updelay will be rounded in bond_init() when it
3893 * is divided by miimon, we just inform user here
3894 */
3895 printk(KERN_WARNING
3896 "bonding_init(): updelay (%d) is not a multiple "
3897 "of miimon (%d), updelay rounded to %d ms\n",
3898 updelay, miimon, (updelay / miimon) * miimon);
3899 }
3900
3901 if ((downdelay % miimon) != 0) {
3902 /* downdelay will be rounded in bond_init() when it
3903 * is divided by miimon, we just inform user here
3904 */
3905 printk(KERN_WARNING
3906 "bonding_init(): downdelay (%d) is not a "
3907 "multiple of miimon (%d), downdelay rounded "
3908 "to %d ms\n",
3909 downdelay, miimon,
3910 (downdelay / miimon) * miimon);
3911 }
3912 }
3913
3914 if (arp_interval < 0) {
3915 printk(KERN_WARNING
3916 "bonding_init(): arp_interval module parameter (%d), "
3917 "not in range 0-%d, so it was reset to %d\n",
3918 arp_interval, INT_MAX, BOND_LINK_ARP_INTERV);
3919 arp_interval = BOND_LINK_ARP_INTERV;
3920 }
3921
3922 for (arp_ip_count=0 ;
3923 (arp_ip_count < MAX_ARP_IP_TARGETS) && arp_ip_target[arp_ip_count];
3924 arp_ip_count++ ) {
3925 /* not complete check, but should be good enough to
3926 catch mistakes */
3927 if (!isdigit(arp_ip_target[arp_ip_count][0])) {
3928 printk(KERN_WARNING
3929 "bonding_init(): bad arp_ip_target module "
3930 "parameter (%s), ARP monitoring will not be "
3931 "performed\n",
3932 arp_ip_target[arp_ip_count]);
3933 arp_interval = 0;
3934 } else {
3935 u32 ip = in_aton(arp_ip_target[arp_ip_count]);
3936 arp_target[arp_ip_count] = ip;
3937 }
3938 }
3939
3940
3941 if ( (arp_interval > 0) && (arp_ip_count==0)) {
3942 /* don't allow arping if no arp_ip_target given... */
3943 printk(KERN_WARNING
3944 "bonding_init(): arp_interval module parameter "
3945 "(%d) specified without providing an arp_ip_target "
3946 "parameter, arp_interval was reset to 0\n",
3947 arp_interval);
3948 arp_interval = 0;
3949 }
3950
3951 if ((miimon == 0) && (arp_interval == 0)) {
3952 /* miimon and arp_interval not set, we need one so things
3953 * work as expected, see bonding.txt for details
3954 */
3955 printk(KERN_ERR
3956 "bonding_init(): either miimon or "
3957 "arp_interval and arp_ip_target module parameters "
3958 "must be specified, otherwise bonding will not detect "
3959 "link failures! see bonding.txt for details.\n");
3960 }
3961
3962 if ((primary != NULL) && (bond_mode != BOND_MODE_ACTIVEBACKUP) &&
3963 (bond_mode != BOND_MODE_TLB) &&
3964 (bond_mode != BOND_MODE_ALB)){
3965 /* currently, using a primary only makes sense
3966 * in active backup, TLB or ALB modes
3967 */
3968 printk(KERN_WARNING
3969 "bonding_init(): %s primary device specified but has "
3970 "no effect in %s mode\n",
3971 primary, bond_mode_name());
3972 primary = NULL;
3973 }
3974
3975
3976 for (no = 0; no < max_bonds; no++) {
3977 dev_bond->init = bond_init;
3978
3979 err = dev_alloc_name(dev_bond,"bond%d");
3980 if (err < 0) {
3981 kfree(dev_bonds);
3982 return err;
3983 }
3984 SET_MODULE_OWNER(dev_bond);
3985 if (register_netdev(dev_bond) != 0) {
3986 kfree(dev_bonds);
3987 return -EIO;
3988 }
3989 dev_bond++;
3990 }
3991 return 0;
3992 }
3993
3994 static void __exit bonding_exit(void)
3995 {
3996 struct net_device *dev_bond = dev_bonds;
3997 struct bonding *bond;
3998 int no;
3999
4000 unregister_netdevice_notifier(&bond_netdev_notifier);
4001
4002 for (no = 0; no < max_bonds; no++) {
4003
4004 #ifdef CONFIG_PROC_FS
4005 bond = (struct bonding *) dev_bond->priv;
4006 remove_proc_entry("info", bond->bond_proc_dir);
4007 remove_proc_entry(dev_bond->name, proc_net);
4008 #endif
4009 unregister_netdev(dev_bond);
4010 kfree(bond->stats);
4011 kfree(dev_bond->priv);
4012
4013 dev_bond->priv = NULL;
4014 dev_bond++;
4015 }
4016 kfree(dev_bonds);
4017 }
4018
4019 module_init(bonding_init);
4020 module_exit(bonding_exit);
4021 MODULE_LICENSE("GPL");
4022 MODULE_DESCRIPTION(DRV_DESCRIPTION ", v" DRV_VERSION);
4023
4024 /*
4025 * Local variables:
4026 * c-indent-level: 8
4027 * c-basic-offset: 8
4028 * tab-width: 8
4029 * End:
4030 */
Cache object: 5efea603a3f77a9aa528e22822972c06
|