blob: 545f6fe025a831cc2262b933025a80c81d7854f8 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
2 * originally based on the dummy device.
3 *
4 * Copyright 1999, Thomas Davis, tadavis@lbl.gov.
5 * Licensed under the GPL. Based on dummy.c, and eql.c devices.
6 *
7 * bonding.c: an Ethernet Bonding driver
8 *
9 * This is useful to talk to a Cisco EtherChannel compatible equipment:
10 * Cisco 5500
11 * Sun Trunking (Solaris)
12 * Alteon AceDirector Trunks
13 * Linux Bonding
14 * and probably many L2 switches ...
15 *
16 * How it works:
17 * ifconfig bond0 ipaddress netmask up
18 * will setup a network device, with an ip address. No mac address
19 * will be assigned at this time. The hw mac address will come from
20 * the first slave bonded to the channel. All slaves will then use
21 * this hw mac address.
22 *
23 * ifconfig bond0 down
24 * will release all slaves, marking them as down.
25 *
26 * ifenslave bond0 eth0
27 * will attach eth0 to bond0 as a slave. eth0 hw mac address will either
28 * a: be used as initial mac address
29 * b: if a hw mac address already is there, eth0's hw mac address
30 * will then be set from bond0.
31 *
32 * v0.1 - first working version.
33 * v0.2 - changed stats to be calculated by summing slaves stats.
34 *
35 * Changes:
36 * Arnaldo Carvalho de Melo <acme@conectiva.com.br>
37 * - fix leaks on failure at bond_init
38 *
39 * 2000/09/30 - Willy Tarreau <willy at meta-x.org>
40 * - added trivial code to release a slave device.
41 * - fixed security bug (CAP_NET_ADMIN not checked)
42 * - implemented MII link monitoring to disable dead links :
43 * All MII capable slaves are checked every <miimon> milliseconds
44 * (100 ms seems good). This value can be changed by passing it to
45 * insmod. A value of zero disables the monitoring (default).
46 * - fixed an infinite loop in bond_xmit_roundrobin() when there's no
47 * good slave.
48 * - made the code hopefully SMP safe
49 *
50 * 2000/10/03 - Willy Tarreau <willy at meta-x.org>
51 * - optimized slave lists based on relevant suggestions from Thomas Davis
52 * - implemented active-backup method to obtain HA with two switches:
53 * stay as long as possible on the same active interface, while we
54 * also monitor the backup one (MII link status) because we want to know
55 * if we are able to switch at any time. ( pass "mode=1" to insmod )
56 * - lots of stress testings because we need it to be more robust than the
57 * wires ! :->
58 *
59 * 2000/10/09 - Willy Tarreau <willy at meta-x.org>
60 * - added up and down delays after link state change.
61 * - optimized the slaves chaining so that when we run forward, we never
62 * repass through the bond itself, but we can find it by searching
63 * backwards. Renders the deletion more difficult, but accelerates the
64 * scan.
65 * - smarter enslaving and releasing.
66 * - finer and more robust SMP locking
67 *
68 * 2000/10/17 - Willy Tarreau <willy at meta-x.org>
69 * - fixed two potential SMP race conditions
70 *
71 * 2000/10/18 - Willy Tarreau <willy at meta-x.org>
72 * - small fixes to the monitoring FSM in case of zero delays
73 * 2000/11/01 - Willy Tarreau <willy at meta-x.org>
74 * - fixed first slave not automatically used in trunk mode.
75 * 2000/11/10 : spelling of "EtherChannel" corrected.
76 * 2000/11/13 : fixed a race condition in case of concurrent accesses to ioctl().
77 * 2000/12/16 : fixed improper usage of rtnl_exlock_nowait().
78 *
79 * 2001/1/3 - Chad N. Tindel <ctindel at ieee dot org>
80 * - The bonding driver now simulates MII status monitoring, just like
81 * a normal network device. It will show that the link is down iff
82 * every slave in the bond shows that their links are down. If at least
83 * one slave is up, the bond's MII status will appear as up.
84 *
85 * 2001/2/7 - Chad N. Tindel <ctindel at ieee dot org>
86 * - Applications can now query the bond from user space to get
87 * information which may be useful. They do this by calling
88 * the BOND_INFO_QUERY ioctl. Once the app knows how many slaves
89 * are in the bond, it can call the BOND_SLAVE_INFO_QUERY ioctl to
90 * get slave specific information (# link failures, etc). See
91 * <linux/if_bonding.h> for more details. The structs of interest
92 * are ifbond and ifslave.
93 *
94 * 2001/4/5 - Chad N. Tindel <ctindel at ieee dot org>
95 * - Ported to 2.4 Kernel
96 *
97 * 2001/5/2 - Jeffrey E. Mast <jeff at mastfamily dot com>
98 * - When a device is detached from a bond, the slave device is no longer
99 * left thinking that is has a master.
100 *
101 * 2001/5/16 - Jeffrey E. Mast <jeff at mastfamily dot com>
102 * - memset did not appropriately initialized the bond rw_locks. Used
103 * rwlock_init to initialize to unlocked state to prevent deadlock when
104 * first attempting a lock
105 * - Called SET_MODULE_OWNER for bond device
106 *
107 * 2001/5/17 - Tim Anderson <tsa at mvista.com>
108 * - 2 paths for releasing for slave release; 1 through ioctl
109 * and 2) through close. Both paths need to release the same way.
110 * - the free slave in bond release is changing slave status before
111 * the free. The netdev_set_master() is intended to change slave state
112 * so it should not be done as part of the release process.
113 * - Simple rule for slave state at release: only the active in A/B and
114 * only one in the trunked case.
115 *
116 * 2001/6/01 - Tim Anderson <tsa at mvista.com>
117 * - Now call dev_close when releasing a slave so it doesn't screw up
118 * out routing table.
119 *
120 * 2001/6/01 - Chad N. Tindel <ctindel at ieee dot org>
121 * - Added /proc support for getting bond and slave information.
122 * Information is in /proc/net/<bond device>/info.
123 * - Changed the locking when calling bond_close to prevent deadlock.
124 *
125 * 2001/8/05 - Janice Girouard <girouard at us.ibm.com>
126 * - correct problem where refcnt of slave is not incremented in bond_ioctl
127 * so the system hangs when halting.
128 * - correct locking problem when unable to malloc in bond_enslave.
129 * - adding bond_xmit_xor logic.
130 * - adding multiple bond device support.
131 *
132 * 2001/8/13 - Erik Habbinga <erik_habbinga at hp dot com>
133 * - correct locking problem with rtnl_exlock_nowait
134 *
135 * 2001/8/23 - Janice Girouard <girouard at us.ibm.com>
136 * - bzero initial dev_bonds, to correct oops
137 * - convert SIOCDEVPRIVATE to new MII ioctl calls
138 *
139 * 2001/9/13 - Takao Indoh <indou dot takao at jp dot fujitsu dot com>
140 * - Add the BOND_CHANGE_ACTIVE ioctl implementation
141 *
142 * 2001/9/14 - Mark Huth <mhuth at mvista dot com>
143 * - Change MII_LINK_READY to not check for end of auto-negotiation,
144 * but only for an up link.
145 *
146 * 2001/9/20 - Chad N. Tindel <ctindel at ieee dot org>
147 * - Add the device field to bonding_t. Previously the net_device
148 * corresponding to a bond wasn't available from the bonding_t
149 * structure.
150 *
151 * 2001/9/25 - Janice Girouard <girouard at us.ibm.com>
152 * - add arp_monitor for active backup mode
153 *
154 * 2001/10/23 - Takao Indoh <indou dot takao at jp dot fujitsu dot com>
155 * - Various memory leak fixes
156 *
157 * 2001/11/5 - Mark Huth <mark dot huth at mvista dot com>
158 * - Don't take rtnl lock in bond_mii_monitor as it deadlocks under
159 * certain hotswap conditions.
160 * Note: this same change may be required in bond_arp_monitor ???
161 * - Remove possibility of calling bond_sethwaddr with NULL slave_dev ptr
162 * - Handle hot swap ethernet interface deregistration events to remove
163 * kernel oops following hot swap of enslaved interface
164 *
165 * 2002/1/2 - Chad N. Tindel <ctindel at ieee dot org>
166 * - Restore original slave flags at release time.
167 *
168 * 2002/02/18 - Erik Habbinga <erik_habbinga at hp dot com>
169 * - bond_release(): calling kfree on our_slave after call to
170 * bond_restore_slave_flags, not before
171 * - bond_enslave(): saving slave flags into original_flags before
172 * call to netdev_set_master, so the IFF_SLAVE flag doesn't end
173 * up in original_flags
174 *
175 * 2002/04/05 - Mark Smith <mark.smith at comdev dot cc> and
176 * Steve Mead <steve.mead at comdev dot cc>
177 * - Port Gleb Natapov's multicast support patchs from 2.4.12
178 * to 2.4.18 adding support for multicast.
179 *
180 * 2002/06/10 - Tony Cureington <tony.cureington * hp_com>
181 * - corrected uninitialized pointer (ifr.ifr_data) in bond_check_dev_link;
182 * actually changed function to use MIIPHY, then MIIREG, and finally
183 * ETHTOOL to determine the link status
184 * - fixed bad ifr_data pointer assignments in bond_ioctl
185 * - corrected mode 1 being reported as active-backup in bond_get_info;
186 * also added text to distinguish type of load balancing (rr or xor)
187 * - change arp_ip_target module param from "1-12s" (array of 12 ptrs)
188 * to "s" (a single ptr)
189 *
190 * 2002/08/30 - Jay Vosburgh <fubar at us dot ibm dot com>
191 * - Removed acquisition of xmit_lock in set_multicast_list; caused
192 * deadlock on SMP (lock is held by caller).
193 * - Revamped SIOCGMIIPHY, SIOCGMIIREG portion of bond_check_dev_link().
194 *
195 * 2002/09/18 - Jay Vosburgh <fubar at us dot ibm dot com>
196 * - Fixed up bond_check_dev_link() (and callers): removed some magic
197 * numbers, banished local MII_ defines, wrapped ioctl calls to
198 * prevent EFAULT errors
199 *
200 * 2002/9/30 - Jay Vosburgh <fubar at us dot ibm dot com>
201 * - make sure the ip target matches the arp_target before saving the
202 * hw address.
203 *
204 * 2002/9/30 - Dan Eisner <eisner at 2robots dot com>
205 * - make sure my_ip is set before taking down the link, since
206 * not all switches respond if the source ip is not set.
207 *
208 * 2002/10/8 - Janice Girouard <girouard at us dot ibm dot com>
209 * - read in the local ip address when enslaving a device
210 * - add primary support
211 * - make sure 2*arp_interval has passed when a new device
212 * is brought on-line before taking it down.
213 *
214 * 2002/09/11 - Philippe De Muyter <phdm at macqel dot be>
215 * - Added bond_xmit_broadcast logic.
216 * - Added bond_mode() support function.
217 *
218 * 2002/10/26 - Laurent Deniel <laurent.deniel at free.fr>
219 * - allow to register multicast addresses only on active slave
220 * (useful in active-backup mode)
221 * - add multicast module parameter
222 * - fix deletion of multicast groups after unloading module
223 *
224 * 2002/11/06 - Kameshwara Rayaprolu <kameshwara.rao * wipro_com>
225 * - Changes to prevent panic from closing the device twice; if we close
226 * the device in bond_release, we must set the original_flags to down
227 * so it won't be closed again by the network layer.
228 *
229 * 2002/11/07 - Tony Cureington <tony.cureington * hp_com>
230 * - Fix arp_target_hw_addr memory leak
231 * - Created activebackup_arp_monitor function to handle arp monitoring
232 * in active backup mode - the bond_arp_monitor had several problems...
233 * such as allowing slaves to tx arps sequentially without any delay
234 * for a response
235 * - Renamed bond_arp_monitor to loadbalance_arp_monitor and re-wrote
236 * this function to just handle arp monitoring in load-balancing mode;
237 * it is a lot more compact now
238 * - Changes to ensure one and only one slave transmits in active-backup
239 * mode
240 * - Robustesize parameters; warn users about bad combinations of
241 * parameters; also if miimon is specified and a network driver does
242 * not support MII or ETHTOOL, inform the user of this
243 * - Changes to support link_failure_count when in arp monitoring mode
244 * - Fix up/down delay reported in /proc
245 * - Added version; log version; make version available from "modinfo -d"
246 * - Fixed problem in bond_check_dev_link - if the first IOCTL (SIOCGMIIPH)
247 * failed, the ETHTOOL ioctl never got a chance
248 *
249 * 2002/11/16 - Laurent Deniel <laurent.deniel at free.fr>
250 * - fix multicast handling in activebackup_arp_monitor
251 * - remove one unnecessary and confusing curr_active_slave == slave test
252 * in activebackup_arp_monitor
253 *
254 * 2002/11/17 - Laurent Deniel <laurent.deniel at free.fr>
255 * - fix bond_slave_info_query when slave_id = num_slaves
256 *
257 * 2002/11/19 - Janice Girouard <girouard at us dot ibm dot com>
258 * - correct ifr_data reference. Update ifr_data reference
259 * to mii_ioctl_data struct values to avoid confusion.
260 *
261 * 2002/11/22 - Bert Barbe <bert.barbe at oracle dot com>
262 * - Add support for multiple arp_ip_target
263 *
264 * 2002/12/13 - Jay Vosburgh <fubar at us dot ibm dot com>
265 * - Changed to allow text strings for mode and multicast, e.g.,
266 * insmod bonding mode=active-backup. The numbers still work.
267 * One change: an invalid choice will cause module load failure,
268 * rather than the previous behavior of just picking one.
269 * - Minor cleanups; got rid of dup ctype stuff, atoi function
270 *
271 * 2003/02/07 - Jay Vosburgh <fubar at us dot ibm dot com>
272 * - Added use_carrier module parameter that causes miimon to
273 * use netif_carrier_ok() test instead of MII/ETHTOOL ioctls.
274 * - Minor cleanups; consolidated ioctl calls to one function.
275 *
276 * 2003/02/07 - Tony Cureington <tony.cureington * hp_com>
277 * - Fix bond_mii_monitor() logic error that could result in
278 * bonding round-robin mode ignoring links after failover/recovery
279 *
280 * 2003/03/17 - Jay Vosburgh <fubar at us dot ibm dot com>
281 * - kmalloc fix (GFP_KERNEL to GFP_ATOMIC) reported by
282 * Shmulik dot Hen at intel.com.
283 * - Based on discussion on mailing list, changed use of
284 * update_slave_cnt(), created wrapper functions for adding/removing
285 * slaves, changed bond_xmit_xor() to check slave_cnt instead of
286 * checking slave and slave->dev (which only worked by accident).
287 * - Misc code cleanup: get arp_send() prototype from header file,
288 * add max_bonds to bonding.txt.
289 *
290 * 2003/03/18 - Tsippy Mendelson <tsippy.mendelson at intel dot com> and
291 * Shmulik Hen <shmulik.hen at intel dot com>
292 * - Make sure only bond_attach_slave() and bond_detach_slave() can
293 * manipulate the slave list, including slave_cnt, even when in
294 * bond_release_all().
295 * - Fixed hang in bond_release() with traffic running:
296 * netdev_set_master() must not be called from within the bond lock.
297 *
298 * 2003/03/18 - Tsippy Mendelson <tsippy.mendelson at intel dot com> and
299 * Shmulik Hen <shmulik.hen at intel dot com>
300 * - Fixed hang in bond_enslave() with traffic running:
301 * netdev_set_master() must not be called from within the bond lock.
302 *
303 * 2003/03/18 - Amir Noam <amir.noam at intel dot com>
304 * - Added support for getting slave's speed and duplex via ethtool.
305 * Needed for 802.3ad and other future modes.
306 *
307 * 2003/03/18 - Tsippy Mendelson <tsippy.mendelson at intel dot com> and
308 * Shmulik Hen <shmulik.hen at intel dot com>
309 * - Enable support of modes that need to use the unique mac address of
310 * each slave.
311 * * bond_enslave(): Moved setting the slave's mac address, and
312 * openning it, from the application to the driver. This breaks
313 * backward comaptibility with old versions of ifenslave that open
314 * the slave before enalsving it !!!.
315 * * bond_release(): The driver also takes care of closing the slave
316 * and restoring its original mac address.
317 * - Removed the code that restores all base driver's flags.
318 * Flags are automatically restored once all undo stages are done
319 * properly.
320 * - Block possibility of enslaving before the master is up. This
321 * prevents putting the system in an unstable state.
322 *
323 * 2003/03/18 - Amir Noam <amir.noam at intel dot com>,
324 * Tsippy Mendelson <tsippy.mendelson at intel dot com> and
325 * Shmulik Hen <shmulik.hen at intel dot com>
326 * - Added support for IEEE 802.3ad Dynamic link aggregation mode.
327 *
328 * 2003/05/01 - Amir Noam <amir.noam at intel dot com>
329 * - Added ABI version control to restore compatibility between
330 * new/old ifenslave and new/old bonding.
331 *
332 * 2003/05/01 - Shmulik Hen <shmulik.hen at intel dot com>
333 * - Fixed bug in bond_release_all(): save old value of curr_active_slave
334 * before setting it to NULL.
335 * - Changed driver versioning scheme to include version number instead
336 * of release date (that is already in another field). There are 3
337 * fields X.Y.Z where:
338 * X - Major version - big behavior changes
339 * Y - Minor version - addition of features
340 * Z - Extra version - minor changes and bug fixes
341 * The current version is 1.0.0 as a base line.
342 *
343 * 2003/05/01 - Tsippy Mendelson <tsippy.mendelson at intel dot com> and
344 * Amir Noam <amir.noam at intel dot com>
345 * - Added support for lacp_rate module param.
346 * - Code beautification and style changes (mainly in comments).
347 * new version - 1.0.1
348 *
349 * 2003/05/01 - Shmulik Hen <shmulik.hen at intel dot com>
350 * - Based on discussion on mailing list, changed locking scheme
351 * to use lock/unlock or lock_bh/unlock_bh appropriately instead
352 * of lock_irqsave/unlock_irqrestore. The new scheme helps exposing
353 * hidden bugs and solves system hangs that occurred due to the fact
354 * that holding lock_irqsave doesn't prevent softirqs from running.
355 * This also increases total throughput since interrupts are not
356 * blocked on each transmitted packets or monitor timeout.
357 * new version - 2.0.0
358 *
359 * 2003/05/01 - Shmulik Hen <shmulik.hen at intel dot com>
360 * - Added support for Transmit load balancing mode.
361 * - Concentrate all assignments of curr_active_slave to a single point
362 * so specific modes can take actions when the primary adapter is
363 * changed.
364 * - Take the updelay parameter into consideration during bond_enslave
365 * since some adapters loose their link during setting the device.
366 * - Renamed bond_3ad_link_status_changed() to
367 * bond_3ad_handle_link_change() for compatibility with TLB.
368 * new version - 2.1.0
369 *
370 * 2003/05/01 - Tsippy Mendelson <tsippy.mendelson at intel dot com>
371 * - Added support for Adaptive load balancing mode which is
372 * equivalent to Transmit load balancing + Receive load balancing.
373 * new version - 2.2.0
374 *
375 * 2003/05/15 - Jay Vosburgh <fubar at us dot ibm dot com>
376 * - Applied fix to activebackup_arp_monitor posted to bonding-devel
377 * by Tony Cureington <tony.cureington * hp_com>. Fixes ARP
378 * monitor endless failover bug. Version to 2.2.10
379 *
380 * 2003/05/20 - Amir Noam <amir.noam at intel dot com>
381 * - Fixed bug in ABI version control - Don't commit to a specific
382 * ABI version if receiving unsupported ioctl commands.
383 *
384 * 2003/05/22 - Jay Vosburgh <fubar at us dot ibm dot com>
385 * - Fix ifenslave -c causing bond to loose existing routes;
386 * added bond_set_mac_address() that doesn't require the
387 * bond to be down.
388 * - In conjunction with fix for ifenslave -c, in
389 * bond_change_active(), changing to the already active slave
390 * is no longer an error (it successfully does nothing).
391 *
392 * 2003/06/30 - Amir Noam <amir.noam at intel dot com>
393 * - Fixed bond_change_active() for ALB/TLB modes.
394 * Version to 2.2.14.
395 *
396 * 2003/07/29 - Amir Noam <amir.noam at intel dot com>
397 * - Fixed ARP monitoring bug.
398 * Version to 2.2.15.
399 *
400 * 2003/07/31 - Willy Tarreau <willy at ods dot org>
401 * - Fixed kernel panic when using ARP monitoring without
402 * setting bond's IP address.
403 * Version to 2.2.16.
404 *
405 * 2003/08/06 - Amir Noam <amir.noam at intel dot com>
406 * - Back port from 2.6: use alloc_netdev(); fix /proc handling;
407 * made stats a part of bond struct so no need to allocate
408 * and free it separately; use standard list operations instead
409 * of pre-allocated array of bonds.
410 * Version to 2.3.0.
411 *
412 * 2003/08/07 - Jay Vosburgh <fubar at us dot ibm dot com>,
413 * Amir Noam <amir.noam at intel dot com> and
414 * Shmulik Hen <shmulik.hen at intel dot com>
415 * - Propagating master's settings: Distinguish between modes that
416 * use a primary slave from those that don't, and propagate settings
417 * accordingly; Consolidate change_active opeartions and add
418 * reselect_active and find_best opeartions; Decouple promiscuous
419 * handling from the multicast mode setting; Add support for changing
420 * HW address and MTU with proper unwind; Consolidate procfs code,
421 * add CHANGENAME handler; Enhance netdev notification handling.
422 * Version to 2.4.0.
423 *
424 * 2003/09/15 - Stephen Hemminger <shemminger at osdl dot org>,
425 * Amir Noam <amir.noam at intel dot com>
426 * - Convert /proc to seq_file interface.
427 * Change /proc/net/bondX/info to /proc/net/bonding/bondX.
428 * Set version to 2.4.1.
429 *
430 * 2003/11/20 - Amir Noam <amir.noam at intel dot com>
431 * - Fix /proc creation/destruction.
432 *
433 * 2003/12/01 - Shmulik Hen <shmulik.hen at intel dot com>
434 * - Massive cleanup - Set version to 2.5.0
435 * Code changes:
436 * o Consolidate format of prints and debug prints.
437 * o Remove bonding_t/slave_t typedefs and consolidate all casts.
438 * o Remove dead code and unnecessary checks.
439 * o Consolidate starting/stopping timers.
440 * o Consolidate handling of primary module param throughout the code.
441 * o Removed multicast module param support - all settings are done
442 * according to mode.
443 * o Slave list iteration - bond is no longer part of the list,
444 * added cyclic list iteration macros.
445 * o Consolidate error handling in all xmit functions.
446 * Style changes:
447 * o Consolidate function naming and declarations.
448 * o Consolidate function params and local variables names.
449 * o Consolidate return values.
450 * o Consolidate curly braces.
451 * o Consolidate conditionals format.
452 * o Change struct member names and types.
453 * o Chomp trailing spaces, remove empty lines, fix indentations.
454 * o Re-organize code according to context.
455 *
456 * 2003/12/30 - Amir Noam <amir.noam at intel dot com>
457 * - Fixed: Cannot remove and re-enslave the original active slave.
458 * - Fixed: Releasing the original active slave causes mac address
459 * duplication.
460 * - Add support for slaves that use ethtool_ops.
461 * Set version to 2.5.3.
462 *
463 * 2004/01/05 - Amir Noam <amir.noam at intel dot com>
464 * - Save bonding parameters per bond instead of using the global values.
465 * Set version to 2.5.4.
466 *
467 * 2004/01/14 - Shmulik Hen <shmulik.hen at intel dot com>
468 * - Enhance VLAN support:
469 * * Add support for VLAN hardware acceleration capable slaves.
470 * * Add capability to tag self generated packets in ALB/TLB modes.
471 * Set version to 2.6.0.
472 * 2004/10/29 - Mitch Williams <mitch.a.williams at intel dot com>
473 * - Fixed bug when unloading module while using 802.3ad. If
474 * spinlock debugging is turned on, this causes a stack dump.
475 * Solution is to move call to dev_remove_pack outside of the
476 * spinlock.
477 * Set version to 2.6.1.
Jay Vosburghc3ade5c2005-06-26 17:52:20 -0400478 * 2005/06/05 - Jay Vosburgh <fubar@us.ibm.com>
479 * - Support for generating gratuitous ARPs in active-backup mode.
480 * Includes support for VLAN tagging all bonding-generated ARPs
481 * as needed. Set version to 2.6.2.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700482 */
483
484//#define BONDING_DEBUG 1
485
486#include <linux/config.h>
487#include <linux/kernel.h>
488#include <linux/module.h>
489#include <linux/sched.h>
490#include <linux/types.h>
491#include <linux/fcntl.h>
492#include <linux/interrupt.h>
493#include <linux/ptrace.h>
494#include <linux/ioport.h>
495#include <linux/in.h>
496#include <linux/ip.h>
497#include <linux/slab.h>
498#include <linux/string.h>
499#include <linux/init.h>
500#include <linux/timer.h>
501#include <linux/socket.h>
502#include <linux/ctype.h>
503#include <linux/inet.h>
504#include <linux/bitops.h>
505#include <asm/system.h>
506#include <asm/io.h>
507#include <asm/dma.h>
508#include <asm/uaccess.h>
509#include <linux/errno.h>
510#include <linux/netdevice.h>
511#include <linux/inetdevice.h>
512#include <linux/etherdevice.h>
513#include <linux/skbuff.h>
514#include <net/sock.h>
515#include <linux/rtnetlink.h>
516#include <linux/proc_fs.h>
517#include <linux/seq_file.h>
518#include <linux/smp.h>
519#include <linux/if_ether.h>
520#include <net/arp.h>
521#include <linux/mii.h>
522#include <linux/ethtool.h>
523#include <linux/if_vlan.h>
524#include <linux/if_bonding.h>
Jay Vosburghc3ade5c2005-06-26 17:52:20 -0400525#include <net/route.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -0700526#include "bonding.h"
527#include "bond_3ad.h"
528#include "bond_alb.h"
529
530/*---------------------------- Module parameters ----------------------------*/
531
532/* monitor all links that often (in milliseconds). <=0 disables monitoring */
533#define BOND_LINK_MON_INTERV 0
534#define BOND_LINK_ARP_INTERV 0
535
536static int max_bonds = BOND_DEFAULT_MAX_BONDS;
537static int miimon = BOND_LINK_MON_INTERV;
538static int updelay = 0;
539static int downdelay = 0;
540static int use_carrier = 1;
541static char *mode = NULL;
542static char *primary = NULL;
543static char *lacp_rate = NULL;
544static int arp_interval = BOND_LINK_ARP_INTERV;
545static char *arp_ip_target[BOND_MAX_ARP_TARGETS] = { NULL, };
546
547module_param(max_bonds, int, 0);
548MODULE_PARM_DESC(max_bonds, "Max number of bonded devices");
549module_param(miimon, int, 0);
550MODULE_PARM_DESC(miimon, "Link check interval in milliseconds");
551module_param(updelay, int, 0);
552MODULE_PARM_DESC(updelay, "Delay before considering link up, in milliseconds");
553module_param(downdelay, int, 0);
554MODULE_PARM_DESC(downdelay, "Delay before considering link down, in milliseconds");
555module_param(use_carrier, int, 0);
556MODULE_PARM_DESC(use_carrier, "Use netif_carrier_ok (vs MII ioctls) in miimon; 0 for off, 1 for on (default)");
557module_param(mode, charp, 0);
558MODULE_PARM_DESC(mode, "Mode of operation : 0 for round robin, 1 for active-backup, 2 for xor");
559module_param(primary, charp, 0);
560MODULE_PARM_DESC(primary, "Primary network device to use");
561module_param(lacp_rate, charp, 0);
562MODULE_PARM_DESC(lacp_rate, "LACPDU tx rate to request from 802.3ad partner (slow/fast)");
563module_param(arp_interval, int, 0);
564MODULE_PARM_DESC(arp_interval, "arp interval in milliseconds");
565module_param_array(arp_ip_target, charp, NULL, 0);
566MODULE_PARM_DESC(arp_ip_target, "arp targets in n.n.n.n form");
567
568/*----------------------------- Global variables ----------------------------*/
569
570static const char *version =
571 DRV_DESCRIPTION ": v" DRV_VERSION " (" DRV_RELDATE ")\n";
572
573static LIST_HEAD(bond_dev_list);
574
575#ifdef CONFIG_PROC_FS
576static struct proc_dir_entry *bond_proc_dir = NULL;
577#endif
578
579static u32 arp_target[BOND_MAX_ARP_TARGETS] = { 0, } ;
580static int arp_ip_count = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700581static int bond_mode = BOND_MODE_ROUNDROBIN;
582static int lacp_fast = 0;
583static int app_abi_ver = 0;
584static int orig_app_abi_ver = -1; /* This is used to save the first ABI version
585 * we receive from the application. Once set,
586 * it won't be changed, and the module will
587 * refuse to enslave/release interfaces if the
588 * command comes from an application using
589 * another ABI version.
590 */
591
592struct bond_parm_tbl {
593 char *modename;
594 int mode;
595};
596
597static struct bond_parm_tbl bond_lacp_tbl[] = {
598{ "slow", AD_LACP_SLOW},
599{ "fast", AD_LACP_FAST},
600{ NULL, -1},
601};
602
603static struct bond_parm_tbl bond_mode_tbl[] = {
604{ "balance-rr", BOND_MODE_ROUNDROBIN},
605{ "active-backup", BOND_MODE_ACTIVEBACKUP},
606{ "balance-xor", BOND_MODE_XOR},
607{ "broadcast", BOND_MODE_BROADCAST},
608{ "802.3ad", BOND_MODE_8023AD},
609{ "balance-tlb", BOND_MODE_TLB},
610{ "balance-alb", BOND_MODE_ALB},
611{ NULL, -1},
612};
613
614/*-------------------------- Forward declarations ---------------------------*/
615
616static inline void bond_set_mode_ops(struct net_device *bond_dev, int mode);
Jay Vosburghc3ade5c2005-06-26 17:52:20 -0400617static void bond_send_gratuitous_arp(struct bonding *bond);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700618
619/*---------------------------- General routines -----------------------------*/
620
621static const char *bond_mode_name(int mode)
622{
623 switch (mode) {
624 case BOND_MODE_ROUNDROBIN :
625 return "load balancing (round-robin)";
626 case BOND_MODE_ACTIVEBACKUP :
627 return "fault-tolerance (active-backup)";
628 case BOND_MODE_XOR :
629 return "load balancing (xor)";
630 case BOND_MODE_BROADCAST :
631 return "fault-tolerance (broadcast)";
632 case BOND_MODE_8023AD:
633 return "IEEE 802.3ad Dynamic link aggregation";
634 case BOND_MODE_TLB:
635 return "transmit load balancing";
636 case BOND_MODE_ALB:
637 return "adaptive load balancing";
638 default:
639 return "unknown";
640 }
641}
642
643/*---------------------------------- VLAN -----------------------------------*/
644
645/**
646 * bond_add_vlan - add a new vlan id on bond
647 * @bond: bond that got the notification
648 * @vlan_id: the vlan id to add
649 *
650 * Returns -ENOMEM if allocation failed.
651 */
652static int bond_add_vlan(struct bonding *bond, unsigned short vlan_id)
653{
654 struct vlan_entry *vlan;
655
656 dprintk("bond: %s, vlan id %d\n",
657 (bond ? bond->dev->name: "None"), vlan_id);
658
659 vlan = kmalloc(sizeof(struct vlan_entry), GFP_KERNEL);
660 if (!vlan) {
661 return -ENOMEM;
662 }
663
664 INIT_LIST_HEAD(&vlan->vlan_list);
665 vlan->vlan_id = vlan_id;
Jay Vosburghc3ade5c2005-06-26 17:52:20 -0400666 vlan->vlan_ip = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700667
668 write_lock_bh(&bond->lock);
669
670 list_add_tail(&vlan->vlan_list, &bond->vlan_list);
671
672 write_unlock_bh(&bond->lock);
673
674 dprintk("added VLAN ID %d on bond %s\n", vlan_id, bond->dev->name);
675
676 return 0;
677}
678
679/**
680 * bond_del_vlan - delete a vlan id from bond
681 * @bond: bond that got the notification
682 * @vlan_id: the vlan id to delete
683 *
684 * returns -ENODEV if @vlan_id was not found in @bond.
685 */
686static int bond_del_vlan(struct bonding *bond, unsigned short vlan_id)
687{
688 struct vlan_entry *vlan, *next;
689 int res = -ENODEV;
690
691 dprintk("bond: %s, vlan id %d\n", bond->dev->name, vlan_id);
692
693 write_lock_bh(&bond->lock);
694
695 list_for_each_entry_safe(vlan, next, &bond->vlan_list, vlan_list) {
696 if (vlan->vlan_id == vlan_id) {
697 list_del(&vlan->vlan_list);
698
699 if ((bond->params.mode == BOND_MODE_TLB) ||
700 (bond->params.mode == BOND_MODE_ALB)) {
701 bond_alb_clear_vlan(bond, vlan_id);
702 }
703
704 dprintk("removed VLAN ID %d from bond %s\n", vlan_id,
705 bond->dev->name);
706
707 kfree(vlan);
708
709 if (list_empty(&bond->vlan_list) &&
710 (bond->slave_cnt == 0)) {
711 /* Last VLAN removed and no slaves, so
712 * restore block on adding VLANs. This will
713 * be removed once new slaves that are not
714 * VLAN challenged will be added.
715 */
716 bond->dev->features |= NETIF_F_VLAN_CHALLENGED;
717 }
718
719 res = 0;
720 goto out;
721 }
722 }
723
724 dprintk("couldn't find VLAN ID %d in bond %s\n", vlan_id,
725 bond->dev->name);
726
727out:
728 write_unlock_bh(&bond->lock);
729 return res;
730}
731
732/**
733 * bond_has_challenged_slaves
734 * @bond: the bond we're working on
735 *
736 * Searches the slave list. Returns 1 if a vlan challenged slave
737 * was found, 0 otherwise.
738 *
739 * Assumes bond->lock is held.
740 */
741static int bond_has_challenged_slaves(struct bonding *bond)
742{
743 struct slave *slave;
744 int i;
745
746 bond_for_each_slave(bond, slave, i) {
747 if (slave->dev->features & NETIF_F_VLAN_CHALLENGED) {
748 dprintk("found VLAN challenged slave - %s\n",
749 slave->dev->name);
750 return 1;
751 }
752 }
753
754 dprintk("no VLAN challenged slaves found\n");
755 return 0;
756}
757
758/**
759 * bond_next_vlan - safely skip to the next item in the vlans list.
760 * @bond: the bond we're working on
761 * @curr: item we're advancing from
762 *
763 * Returns %NULL if list is empty, bond->next_vlan if @curr is %NULL,
764 * or @curr->next otherwise (even if it is @curr itself again).
765 *
766 * Caller must hold bond->lock
767 */
768struct vlan_entry *bond_next_vlan(struct bonding *bond, struct vlan_entry *curr)
769{
770 struct vlan_entry *next, *last;
771
772 if (list_empty(&bond->vlan_list)) {
773 return NULL;
774 }
775
776 if (!curr) {
777 next = list_entry(bond->vlan_list.next,
778 struct vlan_entry, vlan_list);
779 } else {
780 last = list_entry(bond->vlan_list.prev,
781 struct vlan_entry, vlan_list);
782 if (last == curr) {
783 next = list_entry(bond->vlan_list.next,
784 struct vlan_entry, vlan_list);
785 } else {
786 next = list_entry(curr->vlan_list.next,
787 struct vlan_entry, vlan_list);
788 }
789 }
790
791 return next;
792}
793
794/**
795 * bond_dev_queue_xmit - Prepare skb for xmit.
796 *
797 * @bond: bond device that got this skb for tx.
798 * @skb: hw accel VLAN tagged skb to transmit
799 * @slave_dev: slave that is supposed to xmit this skbuff
800 *
801 * When the bond gets an skb to transmit that is
802 * already hardware accelerated VLAN tagged, and it
803 * needs to relay this skb to a slave that is not
804 * hw accel capable, the skb needs to be "unaccelerated",
805 * i.e. strip the hwaccel tag and re-insert it as part
806 * of the payload.
807 */
808int bond_dev_queue_xmit(struct bonding *bond, struct sk_buff *skb, struct net_device *slave_dev)
809{
810 unsigned short vlan_id;
811
812 if (!list_empty(&bond->vlan_list) &&
813 !(slave_dev->features & NETIF_F_HW_VLAN_TX) &&
814 vlan_get_tag(skb, &vlan_id) == 0) {
815 skb->dev = slave_dev;
816 skb = vlan_put_tag(skb, vlan_id);
817 if (!skb) {
818 /* vlan_put_tag() frees the skb in case of error,
819 * so return success here so the calling functions
820 * won't attempt to free is again.
821 */
822 return 0;
823 }
824 } else {
825 skb->dev = slave_dev;
826 }
827
828 skb->priority = 1;
829 dev_queue_xmit(skb);
830
831 return 0;
832}
833
834/*
835 * In the following 3 functions, bond_vlan_rx_register(), bond_vlan_rx_add_vid
836 * and bond_vlan_rx_kill_vid, We don't protect the slave list iteration with a
837 * lock because:
838 * a. This operation is performed in IOCTL context,
839 * b. The operation is protected by the RTNL semaphore in the 8021q code,
840 * c. Holding a lock with BH disabled while directly calling a base driver
841 * entry point is generally a BAD idea.
842 *
843 * The design of synchronization/protection for this operation in the 8021q
844 * module is good for one or more VLAN devices over a single physical device
845 * and cannot be extended for a teaming solution like bonding, so there is a
846 * potential race condition here where a net device from the vlan group might
847 * be referenced (either by a base driver or the 8021q code) while it is being
848 * removed from the system. However, it turns out we're not making matters
849 * worse, and if it works for regular VLAN usage it will work here too.
850*/
851
852/**
853 * bond_vlan_rx_register - Propagates registration to slaves
854 * @bond_dev: bonding net device that got called
855 * @grp: vlan group being registered
856 */
857static void bond_vlan_rx_register(struct net_device *bond_dev, struct vlan_group *grp)
858{
859 struct bonding *bond = bond_dev->priv;
860 struct slave *slave;
861 int i;
862
863 bond->vlgrp = grp;
864
865 bond_for_each_slave(bond, slave, i) {
866 struct net_device *slave_dev = slave->dev;
867
868 if ((slave_dev->features & NETIF_F_HW_VLAN_RX) &&
869 slave_dev->vlan_rx_register) {
870 slave_dev->vlan_rx_register(slave_dev, grp);
871 }
872 }
873}
874
875/**
876 * bond_vlan_rx_add_vid - Propagates adding an id to slaves
877 * @bond_dev: bonding net device that got called
878 * @vid: vlan id being added
879 */
880static void bond_vlan_rx_add_vid(struct net_device *bond_dev, uint16_t vid)
881{
882 struct bonding *bond = bond_dev->priv;
883 struct slave *slave;
884 int i, res;
885
886 bond_for_each_slave(bond, slave, i) {
887 struct net_device *slave_dev = slave->dev;
888
889 if ((slave_dev->features & NETIF_F_HW_VLAN_FILTER) &&
890 slave_dev->vlan_rx_add_vid) {
891 slave_dev->vlan_rx_add_vid(slave_dev, vid);
892 }
893 }
894
895 res = bond_add_vlan(bond, vid);
896 if (res) {
897 printk(KERN_ERR DRV_NAME
898 ": %s: Failed to add vlan id %d\n",
899 bond_dev->name, vid);
900 }
901}
902
903/**
904 * bond_vlan_rx_kill_vid - Propagates deleting an id to slaves
905 * @bond_dev: bonding net device that got called
906 * @vid: vlan id being removed
907 */
908static void bond_vlan_rx_kill_vid(struct net_device *bond_dev, uint16_t vid)
909{
910 struct bonding *bond = bond_dev->priv;
911 struct slave *slave;
912 struct net_device *vlan_dev;
913 int i, res;
914
915 bond_for_each_slave(bond, slave, i) {
916 struct net_device *slave_dev = slave->dev;
917
918 if ((slave_dev->features & NETIF_F_HW_VLAN_FILTER) &&
919 slave_dev->vlan_rx_kill_vid) {
920 /* Save and then restore vlan_dev in the grp array,
921 * since the slave's driver might clear it.
922 */
923 vlan_dev = bond->vlgrp->vlan_devices[vid];
924 slave_dev->vlan_rx_kill_vid(slave_dev, vid);
925 bond->vlgrp->vlan_devices[vid] = vlan_dev;
926 }
927 }
928
929 res = bond_del_vlan(bond, vid);
930 if (res) {
931 printk(KERN_ERR DRV_NAME
932 ": %s: Failed to remove vlan id %d\n",
933 bond_dev->name, vid);
934 }
935}
936
937static void bond_add_vlans_on_slave(struct bonding *bond, struct net_device *slave_dev)
938{
939 struct vlan_entry *vlan;
940
941 write_lock_bh(&bond->lock);
942
943 if (list_empty(&bond->vlan_list)) {
944 goto out;
945 }
946
947 if ((slave_dev->features & NETIF_F_HW_VLAN_RX) &&
948 slave_dev->vlan_rx_register) {
949 slave_dev->vlan_rx_register(slave_dev, bond->vlgrp);
950 }
951
952 if (!(slave_dev->features & NETIF_F_HW_VLAN_FILTER) ||
953 !(slave_dev->vlan_rx_add_vid)) {
954 goto out;
955 }
956
957 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
958 slave_dev->vlan_rx_add_vid(slave_dev, vlan->vlan_id);
959 }
960
961out:
962 write_unlock_bh(&bond->lock);
963}
964
965static void bond_del_vlans_from_slave(struct bonding *bond, struct net_device *slave_dev)
966{
967 struct vlan_entry *vlan;
968 struct net_device *vlan_dev;
969
970 write_lock_bh(&bond->lock);
971
972 if (list_empty(&bond->vlan_list)) {
973 goto out;
974 }
975
976 if (!(slave_dev->features & NETIF_F_HW_VLAN_FILTER) ||
977 !(slave_dev->vlan_rx_kill_vid)) {
978 goto unreg;
979 }
980
981 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
982 /* Save and then restore vlan_dev in the grp array,
983 * since the slave's driver might clear it.
984 */
985 vlan_dev = bond->vlgrp->vlan_devices[vlan->vlan_id];
986 slave_dev->vlan_rx_kill_vid(slave_dev, vlan->vlan_id);
987 bond->vlgrp->vlan_devices[vlan->vlan_id] = vlan_dev;
988 }
989
990unreg:
991 if ((slave_dev->features & NETIF_F_HW_VLAN_RX) &&
992 slave_dev->vlan_rx_register) {
993 slave_dev->vlan_rx_register(slave_dev, NULL);
994 }
995
996out:
997 write_unlock_bh(&bond->lock);
998}
999
1000/*------------------------------- Link status -------------------------------*/
1001
1002/*
1003 * Get link speed and duplex from the slave's base driver
1004 * using ethtool. If for some reason the call fails or the
1005 * values are invalid, fake speed and duplex to 100/Full
1006 * and return error.
1007 */
1008static int bond_update_speed_duplex(struct slave *slave)
1009{
1010 struct net_device *slave_dev = slave->dev;
1011 static int (* ioctl)(struct net_device *, struct ifreq *, int);
1012 struct ifreq ifr;
1013 struct ethtool_cmd etool;
1014
1015 /* Fake speed and duplex */
1016 slave->speed = SPEED_100;
1017 slave->duplex = DUPLEX_FULL;
1018
1019 if (slave_dev->ethtool_ops) {
1020 u32 res;
1021
1022 if (!slave_dev->ethtool_ops->get_settings) {
1023 return -1;
1024 }
1025
1026 res = slave_dev->ethtool_ops->get_settings(slave_dev, &etool);
1027 if (res < 0) {
1028 return -1;
1029 }
1030
1031 goto verify;
1032 }
1033
1034 ioctl = slave_dev->do_ioctl;
1035 strncpy(ifr.ifr_name, slave_dev->name, IFNAMSIZ);
1036 etool.cmd = ETHTOOL_GSET;
1037 ifr.ifr_data = (char*)&etool;
1038 if (!ioctl || (IOCTL(slave_dev, &ifr, SIOCETHTOOL) < 0)) {
1039 return -1;
1040 }
1041
1042verify:
1043 switch (etool.speed) {
1044 case SPEED_10:
1045 case SPEED_100:
1046 case SPEED_1000:
1047 break;
1048 default:
1049 return -1;
1050 }
1051
1052 switch (etool.duplex) {
1053 case DUPLEX_FULL:
1054 case DUPLEX_HALF:
1055 break;
1056 default:
1057 return -1;
1058 }
1059
1060 slave->speed = etool.speed;
1061 slave->duplex = etool.duplex;
1062
1063 return 0;
1064}
1065
1066/*
1067 * if <dev> supports MII link status reporting, check its link status.
1068 *
1069 * We either do MII/ETHTOOL ioctls, or check netif_carrier_ok(),
1070 * depening upon the setting of the use_carrier parameter.
1071 *
1072 * Return either BMSR_LSTATUS, meaning that the link is up (or we
1073 * can't tell and just pretend it is), or 0, meaning that the link is
1074 * down.
1075 *
1076 * If reporting is non-zero, instead of faking link up, return -1 if
1077 * both ETHTOOL and MII ioctls fail (meaning the device does not
1078 * support them). If use_carrier is set, return whatever it says.
1079 * It'd be nice if there was a good way to tell if a driver supports
1080 * netif_carrier, but there really isn't.
1081 */
1082static int bond_check_dev_link(struct bonding *bond, struct net_device *slave_dev, int reporting)
1083{
1084 static int (* ioctl)(struct net_device *, struct ifreq *, int);
1085 struct ifreq ifr;
1086 struct mii_ioctl_data *mii;
1087 struct ethtool_value etool;
1088
1089 if (bond->params.use_carrier) {
1090 return netif_carrier_ok(slave_dev) ? BMSR_LSTATUS : 0;
1091 }
1092
1093 ioctl = slave_dev->do_ioctl;
1094 if (ioctl) {
1095 /* TODO: set pointer to correct ioctl on a per team member */
1096 /* bases to make this more efficient. that is, once */
1097 /* we determine the correct ioctl, we will always */
1098 /* call it and not the others for that team */
1099 /* member. */
1100
1101 /*
1102 * We cannot assume that SIOCGMIIPHY will also read a
1103 * register; not all network drivers (e.g., e100)
1104 * support that.
1105 */
1106
1107 /* Yes, the mii is overlaid on the ifreq.ifr_ifru */
1108 strncpy(ifr.ifr_name, slave_dev->name, IFNAMSIZ);
1109 mii = if_mii(&ifr);
1110 if (IOCTL(slave_dev, &ifr, SIOCGMIIPHY) == 0) {
1111 mii->reg_num = MII_BMSR;
1112 if (IOCTL(slave_dev, &ifr, SIOCGMIIREG) == 0) {
1113 return (mii->val_out & BMSR_LSTATUS);
1114 }
1115 }
1116 }
1117
1118 /* try SIOCETHTOOL ioctl, some drivers cache ETHTOOL_GLINK */
1119 /* for a period of time so we attempt to get link status */
1120 /* from it last if the above MII ioctls fail... */
1121 if (slave_dev->ethtool_ops) {
1122 if (slave_dev->ethtool_ops->get_link) {
1123 u32 link;
1124
1125 link = slave_dev->ethtool_ops->get_link(slave_dev);
1126
1127 return link ? BMSR_LSTATUS : 0;
1128 }
1129 }
1130
1131 if (ioctl) {
1132 strncpy(ifr.ifr_name, slave_dev->name, IFNAMSIZ);
1133 etool.cmd = ETHTOOL_GLINK;
1134 ifr.ifr_data = (char*)&etool;
1135 if (IOCTL(slave_dev, &ifr, SIOCETHTOOL) == 0) {
1136 if (etool.data == 1) {
1137 return BMSR_LSTATUS;
1138 } else {
1139 dprintk("SIOCETHTOOL shows link down\n");
1140 return 0;
1141 }
1142 }
1143 }
1144
1145 /*
1146 * If reporting, report that either there's no dev->do_ioctl,
1147 * or both SIOCGMIIREG and SIOCETHTOOL failed (meaning that we
1148 * cannot report link status). If not reporting, pretend
1149 * we're ok.
1150 */
1151 return (reporting ? -1 : BMSR_LSTATUS);
1152}
1153
1154/*----------------------------- Multicast list ------------------------------*/
1155
1156/*
1157 * Returns 0 if dmi1 and dmi2 are the same, non-0 otherwise
1158 */
1159static inline int bond_is_dmi_same(struct dev_mc_list *dmi1, struct dev_mc_list *dmi2)
1160{
1161 return memcmp(dmi1->dmi_addr, dmi2->dmi_addr, dmi1->dmi_addrlen) == 0 &&
1162 dmi1->dmi_addrlen == dmi2->dmi_addrlen;
1163}
1164
1165/*
1166 * returns dmi entry if found, NULL otherwise
1167 */
1168static struct dev_mc_list *bond_mc_list_find_dmi(struct dev_mc_list *dmi, struct dev_mc_list *mc_list)
1169{
1170 struct dev_mc_list *idmi;
1171
1172 for (idmi = mc_list; idmi; idmi = idmi->next) {
1173 if (bond_is_dmi_same(dmi, idmi)) {
1174 return idmi;
1175 }
1176 }
1177
1178 return NULL;
1179}
1180
1181/*
1182 * Push the promiscuity flag down to appropriate slaves
1183 */
1184static void bond_set_promiscuity(struct bonding *bond, int inc)
1185{
1186 if (USES_PRIMARY(bond->params.mode)) {
1187 /* write lock already acquired */
1188 if (bond->curr_active_slave) {
1189 dev_set_promiscuity(bond->curr_active_slave->dev, inc);
1190 }
1191 } else {
1192 struct slave *slave;
1193 int i;
1194 bond_for_each_slave(bond, slave, i) {
1195 dev_set_promiscuity(slave->dev, inc);
1196 }
1197 }
1198}
1199
1200/*
1201 * Push the allmulti flag down to all slaves
1202 */
1203static void bond_set_allmulti(struct bonding *bond, int inc)
1204{
1205 if (USES_PRIMARY(bond->params.mode)) {
1206 /* write lock already acquired */
1207 if (bond->curr_active_slave) {
1208 dev_set_allmulti(bond->curr_active_slave->dev, inc);
1209 }
1210 } else {
1211 struct slave *slave;
1212 int i;
1213 bond_for_each_slave(bond, slave, i) {
1214 dev_set_allmulti(slave->dev, inc);
1215 }
1216 }
1217}
1218
1219/*
1220 * Add a Multicast address to slaves
1221 * according to mode
1222 */
1223static void bond_mc_add(struct bonding *bond, void *addr, int alen)
1224{
1225 if (USES_PRIMARY(bond->params.mode)) {
1226 /* write lock already acquired */
1227 if (bond->curr_active_slave) {
1228 dev_mc_add(bond->curr_active_slave->dev, addr, alen, 0);
1229 }
1230 } else {
1231 struct slave *slave;
1232 int i;
1233 bond_for_each_slave(bond, slave, i) {
1234 dev_mc_add(slave->dev, addr, alen, 0);
1235 }
1236 }
1237}
1238
1239/*
1240 * Remove a multicast address from slave
1241 * according to mode
1242 */
1243static void bond_mc_delete(struct bonding *bond, void *addr, int alen)
1244{
1245 if (USES_PRIMARY(bond->params.mode)) {
1246 /* write lock already acquired */
1247 if (bond->curr_active_slave) {
1248 dev_mc_delete(bond->curr_active_slave->dev, addr, alen, 0);
1249 }
1250 } else {
1251 struct slave *slave;
1252 int i;
1253 bond_for_each_slave(bond, slave, i) {
1254 dev_mc_delete(slave->dev, addr, alen, 0);
1255 }
1256 }
1257}
1258
1259/*
1260 * Totally destroys the mc_list in bond
1261 */
1262static void bond_mc_list_destroy(struct bonding *bond)
1263{
1264 struct dev_mc_list *dmi;
1265
1266 dmi = bond->mc_list;
1267 while (dmi) {
1268 bond->mc_list = dmi->next;
1269 kfree(dmi);
1270 dmi = bond->mc_list;
1271 }
1272}
1273
1274/*
1275 * Copy all the Multicast addresses from src to the bonding device dst
1276 */
1277static int bond_mc_list_copy(struct dev_mc_list *mc_list, struct bonding *bond, int gpf_flag)
1278{
1279 struct dev_mc_list *dmi, *new_dmi;
1280
1281 for (dmi = mc_list; dmi; dmi = dmi->next) {
1282 new_dmi = kmalloc(sizeof(struct dev_mc_list), gpf_flag);
1283
1284 if (!new_dmi) {
1285 /* FIXME: Potential memory leak !!! */
1286 return -ENOMEM;
1287 }
1288
1289 new_dmi->next = bond->mc_list;
1290 bond->mc_list = new_dmi;
1291 new_dmi->dmi_addrlen = dmi->dmi_addrlen;
1292 memcpy(new_dmi->dmi_addr, dmi->dmi_addr, dmi->dmi_addrlen);
1293 new_dmi->dmi_users = dmi->dmi_users;
1294 new_dmi->dmi_gusers = dmi->dmi_gusers;
1295 }
1296
1297 return 0;
1298}
1299
1300/*
1301 * flush all members of flush->mc_list from device dev->mc_list
1302 */
1303static void bond_mc_list_flush(struct net_device *bond_dev, struct net_device *slave_dev)
1304{
1305 struct bonding *bond = bond_dev->priv;
1306 struct dev_mc_list *dmi;
1307
1308 for (dmi = bond_dev->mc_list; dmi; dmi = dmi->next) {
1309 dev_mc_delete(slave_dev, dmi->dmi_addr, dmi->dmi_addrlen, 0);
1310 }
1311
1312 if (bond->params.mode == BOND_MODE_8023AD) {
1313 /* del lacpdu mc addr from mc list */
1314 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
1315
1316 dev_mc_delete(slave_dev, lacpdu_multicast, ETH_ALEN, 0);
1317 }
1318}
1319
1320/*--------------------------- Active slave change ---------------------------*/
1321
1322/*
1323 * Update the mc list and multicast-related flags for the new and
1324 * old active slaves (if any) according to the multicast mode, and
1325 * promiscuous flags unconditionally.
1326 */
1327static void bond_mc_swap(struct bonding *bond, struct slave *new_active, struct slave *old_active)
1328{
1329 struct dev_mc_list *dmi;
1330
1331 if (!USES_PRIMARY(bond->params.mode)) {
1332 /* nothing to do - mc list is already up-to-date on
1333 * all slaves
1334 */
1335 return;
1336 }
1337
1338 if (old_active) {
1339 if (bond->dev->flags & IFF_PROMISC) {
1340 dev_set_promiscuity(old_active->dev, -1);
1341 }
1342
1343 if (bond->dev->flags & IFF_ALLMULTI) {
1344 dev_set_allmulti(old_active->dev, -1);
1345 }
1346
1347 for (dmi = bond->dev->mc_list; dmi; dmi = dmi->next) {
1348 dev_mc_delete(old_active->dev, dmi->dmi_addr, dmi->dmi_addrlen, 0);
1349 }
1350 }
1351
1352 if (new_active) {
1353 if (bond->dev->flags & IFF_PROMISC) {
1354 dev_set_promiscuity(new_active->dev, 1);
1355 }
1356
1357 if (bond->dev->flags & IFF_ALLMULTI) {
1358 dev_set_allmulti(new_active->dev, 1);
1359 }
1360
1361 for (dmi = bond->dev->mc_list; dmi; dmi = dmi->next) {
1362 dev_mc_add(new_active->dev, dmi->dmi_addr, dmi->dmi_addrlen, 0);
1363 }
1364 }
1365}
1366
1367/**
1368 * find_best_interface - select the best available slave to be the active one
1369 * @bond: our bonding struct
1370 *
1371 * Warning: Caller must hold curr_slave_lock for writing.
1372 */
1373static struct slave *bond_find_best_slave(struct bonding *bond)
1374{
1375 struct slave *new_active, *old_active;
1376 struct slave *bestslave = NULL;
1377 int mintime = bond->params.updelay;
1378 int i;
1379
1380 new_active = old_active = bond->curr_active_slave;
1381
1382 if (!new_active) { /* there were no active slaves left */
1383 if (bond->slave_cnt > 0) { /* found one slave */
1384 new_active = bond->first_slave;
1385 } else {
1386 return NULL; /* still no slave, return NULL */
1387 }
1388 }
1389
1390 /* first try the primary link; if arping, a link must tx/rx traffic
1391 * before it can be considered the curr_active_slave - also, we would skip
1392 * slaves between the curr_active_slave and primary_slave that may be up
1393 * and able to arp
1394 */
1395 if ((bond->primary_slave) &&
1396 (!bond->params.arp_interval) &&
1397 (IS_UP(bond->primary_slave->dev))) {
1398 new_active = bond->primary_slave;
1399 }
1400
1401 /* remember where to stop iterating over the slaves */
1402 old_active = new_active;
1403
1404 bond_for_each_slave_from(bond, new_active, i, old_active) {
1405 if (IS_UP(new_active->dev)) {
1406 if (new_active->link == BOND_LINK_UP) {
1407 return new_active;
1408 } else if (new_active->link == BOND_LINK_BACK) {
1409 /* link up, but waiting for stabilization */
1410 if (new_active->delay < mintime) {
1411 mintime = new_active->delay;
1412 bestslave = new_active;
1413 }
1414 }
1415 }
1416 }
1417
1418 return bestslave;
1419}
1420
1421/**
1422 * change_active_interface - change the active slave into the specified one
1423 * @bond: our bonding struct
1424 * @new: the new slave to make the active one
1425 *
1426 * Set the new slave to the bond's settings and unset them on the old
1427 * curr_active_slave.
1428 * Setting include flags, mc-list, promiscuity, allmulti, etc.
1429 *
1430 * If @new's link state is %BOND_LINK_BACK we'll set it to %BOND_LINK_UP,
1431 * because it is apparently the best available slave we have, even though its
1432 * updelay hasn't timed out yet.
1433 *
1434 * Warning: Caller must hold curr_slave_lock for writing.
1435 */
1436static void bond_change_active_slave(struct bonding *bond, struct slave *new_active)
1437{
1438 struct slave *old_active = bond->curr_active_slave;
1439
1440 if (old_active == new_active) {
1441 return;
1442 }
1443
1444 if (new_active) {
1445 if (new_active->link == BOND_LINK_BACK) {
1446 if (USES_PRIMARY(bond->params.mode)) {
1447 printk(KERN_INFO DRV_NAME
1448 ": %s: making interface %s the new "
1449 "active one %d ms earlier.\n",
1450 bond->dev->name, new_active->dev->name,
1451 (bond->params.updelay - new_active->delay) * bond->params.miimon);
1452 }
1453
1454 new_active->delay = 0;
1455 new_active->link = BOND_LINK_UP;
1456 new_active->jiffies = jiffies;
1457
1458 if (bond->params.mode == BOND_MODE_8023AD) {
1459 bond_3ad_handle_link_change(new_active, BOND_LINK_UP);
1460 }
1461
1462 if ((bond->params.mode == BOND_MODE_TLB) ||
1463 (bond->params.mode == BOND_MODE_ALB)) {
1464 bond_alb_handle_link_change(bond, new_active, BOND_LINK_UP);
1465 }
1466 } else {
1467 if (USES_PRIMARY(bond->params.mode)) {
1468 printk(KERN_INFO DRV_NAME
1469 ": %s: making interface %s the new "
1470 "active one.\n",
1471 bond->dev->name, new_active->dev->name);
1472 }
1473 }
1474 }
1475
Linus Torvalds1da177e2005-04-16 15:20:36 -07001476 if (USES_PRIMARY(bond->params.mode)) {
1477 bond_mc_swap(bond, new_active, old_active);
1478 }
1479
1480 if ((bond->params.mode == BOND_MODE_TLB) ||
1481 (bond->params.mode == BOND_MODE_ALB)) {
1482 bond_alb_handle_active_change(bond, new_active);
1483 } else {
1484 bond->curr_active_slave = new_active;
1485 }
Jay Vosburghc3ade5c2005-06-26 17:52:20 -04001486
1487 if (bond->params.mode == BOND_MODE_ACTIVEBACKUP) {
1488 if (old_active) {
1489 bond_set_slave_inactive_flags(old_active);
1490 }
1491
1492 if (new_active) {
1493 bond_set_slave_active_flags(new_active);
1494 }
1495 bond_send_gratuitous_arp(bond);
1496 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001497}
1498
1499/**
1500 * bond_select_active_slave - select a new active slave, if needed
1501 * @bond: our bonding struct
1502 *
1503 * This functions shoud be called when one of the following occurs:
1504 * - The old curr_active_slave has been released or lost its link.
1505 * - The primary_slave has got its link back.
1506 * - A slave has got its link back and there's no old curr_active_slave.
1507 *
1508 * Warning: Caller must hold curr_slave_lock for writing.
1509 */
1510static void bond_select_active_slave(struct bonding *bond)
1511{
1512 struct slave *best_slave;
1513
1514 best_slave = bond_find_best_slave(bond);
1515 if (best_slave != bond->curr_active_slave) {
1516 bond_change_active_slave(bond, best_slave);
1517 }
1518}
1519
1520/*--------------------------- slave list handling ---------------------------*/
1521
1522/*
1523 * This function attaches the slave to the end of list.
1524 *
1525 * bond->lock held for writing by caller.
1526 */
1527static void bond_attach_slave(struct bonding *bond, struct slave *new_slave)
1528{
1529 if (bond->first_slave == NULL) { /* attaching the first slave */
1530 new_slave->next = new_slave;
1531 new_slave->prev = new_slave;
1532 bond->first_slave = new_slave;
1533 } else {
1534 new_slave->next = bond->first_slave;
1535 new_slave->prev = bond->first_slave->prev;
1536 new_slave->next->prev = new_slave;
1537 new_slave->prev->next = new_slave;
1538 }
1539
1540 bond->slave_cnt++;
1541}
1542
1543/*
1544 * This function detaches the slave from the list.
1545 * WARNING: no check is made to verify if the slave effectively
1546 * belongs to <bond>.
1547 * Nothing is freed on return, structures are just unchained.
1548 * If any slave pointer in bond was pointing to <slave>,
1549 * it should be changed by the calling function.
1550 *
1551 * bond->lock held for writing by caller.
1552 */
1553static void bond_detach_slave(struct bonding *bond, struct slave *slave)
1554{
1555 if (slave->next) {
1556 slave->next->prev = slave->prev;
1557 }
1558
1559 if (slave->prev) {
1560 slave->prev->next = slave->next;
1561 }
1562
1563 if (bond->first_slave == slave) { /* slave is the first slave */
1564 if (bond->slave_cnt > 1) { /* there are more slave */
1565 bond->first_slave = slave->next;
1566 } else {
1567 bond->first_slave = NULL; /* slave was the last one */
1568 }
1569 }
1570
1571 slave->next = NULL;
1572 slave->prev = NULL;
1573 bond->slave_cnt--;
1574}
1575
1576/*---------------------------------- IOCTL ----------------------------------*/
1577
1578static int bond_sethwaddr(struct net_device *bond_dev, struct net_device *slave_dev)
1579{
1580 dprintk("bond_dev=%p\n", bond_dev);
1581 dprintk("slave_dev=%p\n", slave_dev);
1582 dprintk("slave_dev->addr_len=%d\n", slave_dev->addr_len);
1583 memcpy(bond_dev->dev_addr, slave_dev->dev_addr, slave_dev->addr_len);
1584 return 0;
1585}
1586
1587/* enslave device <slave> to bond device <master> */
1588static int bond_enslave(struct net_device *bond_dev, struct net_device *slave_dev)
1589{
1590 struct bonding *bond = bond_dev->priv;
1591 struct slave *new_slave = NULL;
1592 struct dev_mc_list *dmi;
1593 struct sockaddr addr;
1594 int link_reporting;
1595 int old_features = bond_dev->features;
1596 int res = 0;
1597
1598 if (slave_dev->do_ioctl == NULL) {
1599 printk(KERN_WARNING DRV_NAME
1600 ": Warning : no link monitoring support for %s\n",
1601 slave_dev->name);
1602 }
1603
1604 /* bond must be initialized by bond_open() before enslaving */
1605 if (!(bond_dev->flags & IFF_UP)) {
1606 dprintk("Error, master_dev is not up\n");
1607 return -EPERM;
1608 }
1609
1610 /* already enslaved */
1611 if (slave_dev->flags & IFF_SLAVE) {
1612 dprintk("Error, Device was already enslaved\n");
1613 return -EBUSY;
1614 }
1615
1616 /* vlan challenged mutual exclusion */
1617 /* no need to lock since we're protected by rtnl_lock */
1618 if (slave_dev->features & NETIF_F_VLAN_CHALLENGED) {
1619 dprintk("%s: NETIF_F_VLAN_CHALLENGED\n", slave_dev->name);
1620 if (!list_empty(&bond->vlan_list)) {
1621 printk(KERN_ERR DRV_NAME
1622 ": Error: cannot enslave VLAN "
1623 "challenged slave %s on VLAN enabled "
1624 "bond %s\n", slave_dev->name,
1625 bond_dev->name);
1626 return -EPERM;
1627 } else {
1628 printk(KERN_WARNING DRV_NAME
1629 ": Warning: enslaved VLAN challenged "
1630 "slave %s. Adding VLANs will be blocked as "
1631 "long as %s is part of bond %s\n",
1632 slave_dev->name, slave_dev->name,
1633 bond_dev->name);
1634 bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
1635 }
1636 } else {
1637 dprintk("%s: ! NETIF_F_VLAN_CHALLENGED\n", slave_dev->name);
1638 if (bond->slave_cnt == 0) {
1639 /* First slave, and it is not VLAN challenged,
1640 * so remove the block of adding VLANs over the bond.
1641 */
1642 bond_dev->features &= ~NETIF_F_VLAN_CHALLENGED;
1643 }
1644 }
1645
1646 if (app_abi_ver >= 1) {
1647 /* The application is using an ABI, which requires the
1648 * slave interface to be closed.
1649 */
1650 if ((slave_dev->flags & IFF_UP)) {
1651 printk(KERN_ERR DRV_NAME
1652 ": Error: %s is up\n",
1653 slave_dev->name);
1654 res = -EPERM;
1655 goto err_undo_flags;
1656 }
1657
1658 if (slave_dev->set_mac_address == NULL) {
1659 printk(KERN_ERR DRV_NAME
1660 ": Error: The slave device you specified does "
1661 "not support setting the MAC address.\n");
1662 printk(KERN_ERR
1663 "Your kernel likely does not support slave "
1664 "devices.\n");
1665
1666 res = -EOPNOTSUPP;
1667 goto err_undo_flags;
1668 }
1669 } else {
1670 /* The application is not using an ABI, which requires the
1671 * slave interface to be open.
1672 */
1673 if (!(slave_dev->flags & IFF_UP)) {
1674 printk(KERN_ERR DRV_NAME
1675 ": Error: %s is not running\n",
1676 slave_dev->name);
1677 res = -EINVAL;
1678 goto err_undo_flags;
1679 }
1680
1681 if ((bond->params.mode == BOND_MODE_8023AD) ||
1682 (bond->params.mode == BOND_MODE_TLB) ||
1683 (bond->params.mode == BOND_MODE_ALB)) {
1684 printk(KERN_ERR DRV_NAME
1685 ": Error: to use %s mode, you must upgrade "
1686 "ifenslave.\n",
1687 bond_mode_name(bond->params.mode));
1688 res = -EOPNOTSUPP;
1689 goto err_undo_flags;
1690 }
1691 }
1692
1693 new_slave = kmalloc(sizeof(struct slave), GFP_KERNEL);
1694 if (!new_slave) {
1695 res = -ENOMEM;
1696 goto err_undo_flags;
1697 }
1698
1699 memset(new_slave, 0, sizeof(struct slave));
1700
1701 /* save slave's original flags before calling
1702 * netdev_set_master and dev_open
1703 */
1704 new_slave->original_flags = slave_dev->flags;
1705
1706 if (app_abi_ver >= 1) {
1707 /* save slave's original ("permanent") mac address for
1708 * modes that needs it, and for restoring it upon release,
1709 * and then set it to the master's address
1710 */
1711 memcpy(new_slave->perm_hwaddr, slave_dev->dev_addr, ETH_ALEN);
1712
1713 /* set slave to master's mac address
1714 * The application already set the master's
1715 * mac address to that of the first slave
1716 */
1717 memcpy(addr.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
1718 addr.sa_family = slave_dev->type;
1719 res = dev_set_mac_address(slave_dev, &addr);
1720 if (res) {
1721 dprintk("Error %d calling set_mac_address\n", res);
1722 goto err_free;
1723 }
1724
1725 /* open the slave since the application closed it */
1726 res = dev_open(slave_dev);
1727 if (res) {
1728 dprintk("Openning slave %s failed\n", slave_dev->name);
1729 goto err_restore_mac;
1730 }
1731 }
1732
1733 res = netdev_set_master(slave_dev, bond_dev);
1734 if (res) {
1735 dprintk("Error %d calling netdev_set_master\n", res);
1736 if (app_abi_ver < 1) {
1737 goto err_free;
1738 } else {
1739 goto err_close;
1740 }
1741 }
1742
1743 new_slave->dev = slave_dev;
1744
1745 if ((bond->params.mode == BOND_MODE_TLB) ||
1746 (bond->params.mode == BOND_MODE_ALB)) {
1747 /* bond_alb_init_slave() must be called before all other stages since
1748 * it might fail and we do not want to have to undo everything
1749 */
1750 res = bond_alb_init_slave(bond, new_slave);
1751 if (res) {
1752 goto err_unset_master;
1753 }
1754 }
1755
1756 /* If the mode USES_PRIMARY, then the new slave gets the
1757 * master's promisc (and mc) settings only if it becomes the
1758 * curr_active_slave, and that is taken care of later when calling
1759 * bond_change_active()
1760 */
1761 if (!USES_PRIMARY(bond->params.mode)) {
1762 /* set promiscuity level to new slave */
1763 if (bond_dev->flags & IFF_PROMISC) {
1764 dev_set_promiscuity(slave_dev, 1);
1765 }
1766
1767 /* set allmulti level to new slave */
1768 if (bond_dev->flags & IFF_ALLMULTI) {
1769 dev_set_allmulti(slave_dev, 1);
1770 }
1771
1772 /* upload master's mc_list to new slave */
1773 for (dmi = bond_dev->mc_list; dmi; dmi = dmi->next) {
1774 dev_mc_add (slave_dev, dmi->dmi_addr, dmi->dmi_addrlen, 0);
1775 }
1776 }
1777
1778 if (bond->params.mode == BOND_MODE_8023AD) {
1779 /* add lacpdu mc addr to mc list */
1780 u8 lacpdu_multicast[ETH_ALEN] = MULTICAST_LACPDU_ADDR;
1781
1782 dev_mc_add(slave_dev, lacpdu_multicast, ETH_ALEN, 0);
1783 }
1784
1785 bond_add_vlans_on_slave(bond, slave_dev);
1786
1787 write_lock_bh(&bond->lock);
1788
1789 bond_attach_slave(bond, new_slave);
1790
1791 new_slave->delay = 0;
1792 new_slave->link_failure_count = 0;
1793
1794 if (bond->params.miimon && !bond->params.use_carrier) {
1795 link_reporting = bond_check_dev_link(bond, slave_dev, 1);
1796
1797 if ((link_reporting == -1) && !bond->params.arp_interval) {
1798 /*
1799 * miimon is set but a bonded network driver
1800 * does not support ETHTOOL/MII and
1801 * arp_interval is not set. Note: if
1802 * use_carrier is enabled, we will never go
1803 * here (because netif_carrier is always
1804 * supported); thus, we don't need to change
1805 * the messages for netif_carrier.
1806 */
1807 printk(KERN_WARNING DRV_NAME
1808 ": Warning: MII and ETHTOOL support not "
1809 "available for interface %s, and "
1810 "arp_interval/arp_ip_target module parameters "
1811 "not specified, thus bonding will not detect "
1812 "link failures! see bonding.txt for details.\n",
1813 slave_dev->name);
1814 } else if (link_reporting == -1) {
1815 /* unable get link status using mii/ethtool */
1816 printk(KERN_WARNING DRV_NAME
1817 ": Warning: can't get link status from "
1818 "interface %s; the network driver associated "
1819 "with this interface does not support MII or "
1820 "ETHTOOL link status reporting, thus miimon "
1821 "has no effect on this interface.\n",
1822 slave_dev->name);
1823 }
1824 }
1825
1826 /* check for initial state */
1827 if (!bond->params.miimon ||
1828 (bond_check_dev_link(bond, slave_dev, 0) == BMSR_LSTATUS)) {
1829 if (bond->params.updelay) {
1830 dprintk("Initial state of slave_dev is "
1831 "BOND_LINK_BACK\n");
1832 new_slave->link = BOND_LINK_BACK;
1833 new_slave->delay = bond->params.updelay;
1834 } else {
1835 dprintk("Initial state of slave_dev is "
1836 "BOND_LINK_UP\n");
1837 new_slave->link = BOND_LINK_UP;
1838 }
1839 new_slave->jiffies = jiffies;
1840 } else {
1841 dprintk("Initial state of slave_dev is "
1842 "BOND_LINK_DOWN\n");
1843 new_slave->link = BOND_LINK_DOWN;
1844 }
1845
1846 if (bond_update_speed_duplex(new_slave) &&
1847 (new_slave->link != BOND_LINK_DOWN)) {
1848 printk(KERN_WARNING DRV_NAME
1849 ": Warning: failed to get speed and duplex from %s, "
1850 "assumed to be 100Mb/sec and Full.\n",
1851 new_slave->dev->name);
1852
1853 if (bond->params.mode == BOND_MODE_8023AD) {
1854 printk(KERN_WARNING
1855 "Operation of 802.3ad mode requires ETHTOOL "
1856 "support in base driver for proper aggregator "
1857 "selection.\n");
1858 }
1859 }
1860
1861 if (USES_PRIMARY(bond->params.mode) && bond->params.primary[0]) {
1862 /* if there is a primary slave, remember it */
1863 if (strcmp(bond->params.primary, new_slave->dev->name) == 0) {
1864 bond->primary_slave = new_slave;
1865 }
1866 }
1867
1868 switch (bond->params.mode) {
1869 case BOND_MODE_ACTIVEBACKUP:
1870 /* if we're in active-backup mode, we need one and only one active
1871 * interface. The backup interfaces will have their NOARP flag set
1872 * because we need them to be completely deaf and not to respond to
1873 * any ARP request on the network to avoid fooling a switch. Thus,
1874 * since we guarantee that curr_active_slave always point to the last
1875 * usable interface, we just have to verify this interface's flag.
1876 */
1877 if (((!bond->curr_active_slave) ||
1878 (bond->curr_active_slave->dev->flags & IFF_NOARP)) &&
1879 (new_slave->link != BOND_LINK_DOWN)) {
1880 dprintk("This is the first active slave\n");
1881 /* first slave or no active slave yet, and this link
1882 is OK, so make this interface the active one */
1883 bond_change_active_slave(bond, new_slave);
1884 } else {
1885 dprintk("This is just a backup slave\n");
1886 bond_set_slave_inactive_flags(new_slave);
1887 }
1888 break;
1889 case BOND_MODE_8023AD:
1890 /* in 802.3ad mode, the internal mechanism
1891 * will activate the slaves in the selected
1892 * aggregator
1893 */
1894 bond_set_slave_inactive_flags(new_slave);
1895 /* if this is the first slave */
1896 if (bond->slave_cnt == 1) {
1897 SLAVE_AD_INFO(new_slave).id = 1;
1898 /* Initialize AD with the number of times that the AD timer is called in 1 second
1899 * can be called only after the mac address of the bond is set
1900 */
1901 bond_3ad_initialize(bond, 1000/AD_TIMER_INTERVAL,
1902 bond->params.lacp_fast);
1903 } else {
1904 SLAVE_AD_INFO(new_slave).id =
1905 SLAVE_AD_INFO(new_slave->prev).id + 1;
1906 }
1907
1908 bond_3ad_bind_slave(new_slave);
1909 break;
1910 case BOND_MODE_TLB:
1911 case BOND_MODE_ALB:
1912 new_slave->state = BOND_STATE_ACTIVE;
1913 if ((!bond->curr_active_slave) &&
1914 (new_slave->link != BOND_LINK_DOWN)) {
1915 /* first slave or no active slave yet, and this link
1916 * is OK, so make this interface the active one
1917 */
1918 bond_change_active_slave(bond, new_slave);
1919 }
1920 break;
1921 default:
1922 dprintk("This slave is always active in trunk mode\n");
1923
1924 /* always active in trunk mode */
1925 new_slave->state = BOND_STATE_ACTIVE;
1926
1927 /* In trunking mode there is little meaning to curr_active_slave
1928 * anyway (it holds no special properties of the bond device),
1929 * so we can change it without calling change_active_interface()
1930 */
1931 if (!bond->curr_active_slave) {
1932 bond->curr_active_slave = new_slave;
1933 }
1934 break;
1935 } /* switch(bond_mode) */
1936
1937 write_unlock_bh(&bond->lock);
1938
1939 if (app_abi_ver < 1) {
1940 /*
1941 * !!! This is to support old versions of ifenslave.
1942 * We can remove this in 2.5 because our ifenslave takes
1943 * care of this for us.
1944 * We check to see if the master has a mac address yet.
1945 * If not, we'll give it the mac address of our slave device.
1946 */
1947 int ndx = 0;
1948
1949 for (ndx = 0; ndx < bond_dev->addr_len; ndx++) {
1950 dprintk("Checking ndx=%d of bond_dev->dev_addr\n",
1951 ndx);
1952 if (bond_dev->dev_addr[ndx] != 0) {
1953 dprintk("Found non-zero byte at ndx=%d\n",
1954 ndx);
1955 break;
1956 }
1957 }
1958
1959 if (ndx == bond_dev->addr_len) {
1960 /*
1961 * We got all the way through the address and it was
1962 * all 0's.
1963 */
1964 dprintk("%s doesn't have a MAC address yet. \n",
1965 bond_dev->name);
1966 dprintk("Going to give assign it from %s.\n",
1967 slave_dev->name);
1968 bond_sethwaddr(bond_dev, slave_dev);
1969 }
1970 }
1971
1972 printk(KERN_INFO DRV_NAME
1973 ": %s: enslaving %s as a%s interface with a%s link.\n",
1974 bond_dev->name, slave_dev->name,
1975 new_slave->state == BOND_STATE_ACTIVE ? "n active" : " backup",
1976 new_slave->link != BOND_LINK_DOWN ? "n up" : " down");
1977
1978 /* enslave is successful */
1979 return 0;
1980
1981/* Undo stages on error */
1982err_unset_master:
1983 netdev_set_master(slave_dev, NULL);
1984
1985err_close:
1986 dev_close(slave_dev);
1987
1988err_restore_mac:
1989 memcpy(addr.sa_data, new_slave->perm_hwaddr, ETH_ALEN);
1990 addr.sa_family = slave_dev->type;
1991 dev_set_mac_address(slave_dev, &addr);
1992
1993err_free:
1994 kfree(new_slave);
1995
1996err_undo_flags:
1997 bond_dev->features = old_features;
1998
1999 return res;
2000}
2001
2002/*
2003 * Try to release the slave device <slave> from the bond device <master>
2004 * It is legal to access curr_active_slave without a lock because all the function
2005 * is write-locked.
2006 *
2007 * The rules for slave state should be:
2008 * for Active/Backup:
2009 * Active stays on all backups go down
2010 * for Bonded connections:
2011 * The first up interface should be left on and all others downed.
2012 */
2013static int bond_release(struct net_device *bond_dev, struct net_device *slave_dev)
2014{
2015 struct bonding *bond = bond_dev->priv;
2016 struct slave *slave, *oldcurrent;
2017 struct sockaddr addr;
2018 int mac_addr_differ;
2019
2020 /* slave is not a slave or master is not master of this slave */
2021 if (!(slave_dev->flags & IFF_SLAVE) ||
2022 (slave_dev->master != bond_dev)) {
2023 printk(KERN_ERR DRV_NAME
2024 ": Error: %s: cannot release %s.\n",
2025 bond_dev->name, slave_dev->name);
2026 return -EINVAL;
2027 }
2028
2029 write_lock_bh(&bond->lock);
2030
2031 slave = bond_get_slave_by_dev(bond, slave_dev);
2032 if (!slave) {
2033 /* not a slave of this bond */
2034 printk(KERN_INFO DRV_NAME
2035 ": %s: %s not enslaved\n",
2036 bond_dev->name, slave_dev->name);
2037 return -EINVAL;
2038 }
2039
2040 mac_addr_differ = memcmp(bond_dev->dev_addr,
2041 slave->perm_hwaddr,
2042 ETH_ALEN);
2043 if (!mac_addr_differ && (bond->slave_cnt > 1)) {
2044 printk(KERN_WARNING DRV_NAME
2045 ": Warning: the permanent HWaddr of %s "
2046 "- %02X:%02X:%02X:%02X:%02X:%02X - is "
2047 "still in use by %s. Set the HWaddr of "
2048 "%s to a different address to avoid "
2049 "conflicts.\n",
2050 slave_dev->name,
2051 slave->perm_hwaddr[0],
2052 slave->perm_hwaddr[1],
2053 slave->perm_hwaddr[2],
2054 slave->perm_hwaddr[3],
2055 slave->perm_hwaddr[4],
2056 slave->perm_hwaddr[5],
2057 bond_dev->name,
2058 slave_dev->name);
2059 }
2060
2061 /* Inform AD package of unbinding of slave. */
2062 if (bond->params.mode == BOND_MODE_8023AD) {
2063 /* must be called before the slave is
2064 * detached from the list
2065 */
2066 bond_3ad_unbind_slave(slave);
2067 }
2068
2069 printk(KERN_INFO DRV_NAME
2070 ": %s: releasing %s interface %s\n",
2071 bond_dev->name,
2072 (slave->state == BOND_STATE_ACTIVE)
2073 ? "active" : "backup",
2074 slave_dev->name);
2075
2076 oldcurrent = bond->curr_active_slave;
2077
2078 bond->current_arp_slave = NULL;
2079
2080 /* release the slave from its bond */
2081 bond_detach_slave(bond, slave);
2082
2083 if (bond->primary_slave == slave) {
2084 bond->primary_slave = NULL;
2085 }
2086
2087 if (oldcurrent == slave) {
2088 bond_change_active_slave(bond, NULL);
2089 }
2090
2091 if ((bond->params.mode == BOND_MODE_TLB) ||
2092 (bond->params.mode == BOND_MODE_ALB)) {
2093 /* Must be called only after the slave has been
2094 * detached from the list and the curr_active_slave
2095 * has been cleared (if our_slave == old_current),
2096 * but before a new active slave is selected.
2097 */
2098 bond_alb_deinit_slave(bond, slave);
2099 }
2100
2101 if (oldcurrent == slave) {
2102 bond_select_active_slave(bond);
2103
2104 if (!bond->curr_active_slave) {
2105 printk(KERN_INFO DRV_NAME
2106 ": %s: now running without any active "
2107 "interface !\n",
2108 bond_dev->name);
2109 }
2110 }
2111
2112 if (bond->slave_cnt == 0) {
2113 /* if the last slave was removed, zero the mac address
2114 * of the master so it will be set by the application
2115 * to the mac address of the first slave
2116 */
2117 memset(bond_dev->dev_addr, 0, bond_dev->addr_len);
2118
2119 if (list_empty(&bond->vlan_list)) {
2120 bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
2121 } else {
2122 printk(KERN_WARNING DRV_NAME
2123 ": Warning: clearing HW address of %s while it "
2124 "still has VLANs.\n",
2125 bond_dev->name);
2126 printk(KERN_WARNING DRV_NAME
2127 ": When re-adding slaves, make sure the bond's "
2128 "HW address matches its VLANs'.\n");
2129 }
2130 } else if ((bond_dev->features & NETIF_F_VLAN_CHALLENGED) &&
2131 !bond_has_challenged_slaves(bond)) {
2132 printk(KERN_INFO DRV_NAME
2133 ": last VLAN challenged slave %s "
2134 "left bond %s. VLAN blocking is removed\n",
2135 slave_dev->name, bond_dev->name);
2136 bond_dev->features &= ~NETIF_F_VLAN_CHALLENGED;
2137 }
2138
2139 write_unlock_bh(&bond->lock);
2140
2141 bond_del_vlans_from_slave(bond, slave_dev);
2142
2143 /* If the mode USES_PRIMARY, then we should only remove its
2144 * promisc and mc settings if it was the curr_active_slave, but that was
2145 * already taken care of above when we detached the slave
2146 */
2147 if (!USES_PRIMARY(bond->params.mode)) {
2148 /* unset promiscuity level from slave */
2149 if (bond_dev->flags & IFF_PROMISC) {
2150 dev_set_promiscuity(slave_dev, -1);
2151 }
2152
2153 /* unset allmulti level from slave */
2154 if (bond_dev->flags & IFF_ALLMULTI) {
2155 dev_set_allmulti(slave_dev, -1);
2156 }
2157
2158 /* flush master's mc_list from slave */
2159 bond_mc_list_flush(bond_dev, slave_dev);
2160 }
2161
2162 netdev_set_master(slave_dev, NULL);
2163
2164 /* close slave before restoring its mac address */
2165 dev_close(slave_dev);
2166
2167 if (app_abi_ver >= 1) {
2168 /* restore original ("permanent") mac address */
2169 memcpy(addr.sa_data, slave->perm_hwaddr, ETH_ALEN);
2170 addr.sa_family = slave_dev->type;
2171 dev_set_mac_address(slave_dev, &addr);
2172 }
2173
2174 /* restore the original state of the
2175 * IFF_NOARP flag that might have been
2176 * set by bond_set_slave_inactive_flags()
2177 */
2178 if ((slave->original_flags & IFF_NOARP) == 0) {
2179 slave_dev->flags &= ~IFF_NOARP;
2180 }
2181
2182 kfree(slave);
2183
2184 return 0; /* deletion OK */
2185}
2186
2187/*
2188 * This function releases all slaves.
2189 */
2190static int bond_release_all(struct net_device *bond_dev)
2191{
2192 struct bonding *bond = bond_dev->priv;
2193 struct slave *slave;
2194 struct net_device *slave_dev;
2195 struct sockaddr addr;
2196
2197 write_lock_bh(&bond->lock);
2198
2199 if (bond->slave_cnt == 0) {
2200 goto out;
2201 }
2202
2203 bond->current_arp_slave = NULL;
2204 bond->primary_slave = NULL;
2205 bond_change_active_slave(bond, NULL);
2206
2207 while ((slave = bond->first_slave) != NULL) {
2208 /* Inform AD package of unbinding of slave
2209 * before slave is detached from the list.
2210 */
2211 if (bond->params.mode == BOND_MODE_8023AD) {
2212 bond_3ad_unbind_slave(slave);
2213 }
2214
2215 slave_dev = slave->dev;
2216 bond_detach_slave(bond, slave);
2217
2218 if ((bond->params.mode == BOND_MODE_TLB) ||
2219 (bond->params.mode == BOND_MODE_ALB)) {
2220 /* must be called only after the slave
2221 * has been detached from the list
2222 */
2223 bond_alb_deinit_slave(bond, slave);
2224 }
2225
2226 /* now that the slave is detached, unlock and perform
2227 * all the undo steps that should not be called from
2228 * within a lock.
2229 */
2230 write_unlock_bh(&bond->lock);
2231
2232 bond_del_vlans_from_slave(bond, slave_dev);
2233
2234 /* If the mode USES_PRIMARY, then we should only remove its
2235 * promisc and mc settings if it was the curr_active_slave, but that was
2236 * already taken care of above when we detached the slave
2237 */
2238 if (!USES_PRIMARY(bond->params.mode)) {
2239 /* unset promiscuity level from slave */
2240 if (bond_dev->flags & IFF_PROMISC) {
2241 dev_set_promiscuity(slave_dev, -1);
2242 }
2243
2244 /* unset allmulti level from slave */
2245 if (bond_dev->flags & IFF_ALLMULTI) {
2246 dev_set_allmulti(slave_dev, -1);
2247 }
2248
2249 /* flush master's mc_list from slave */
2250 bond_mc_list_flush(bond_dev, slave_dev);
2251 }
2252
2253 netdev_set_master(slave_dev, NULL);
2254
2255 /* close slave before restoring its mac address */
2256 dev_close(slave_dev);
2257
2258 if (app_abi_ver >= 1) {
2259 /* restore original ("permanent") mac address*/
2260 memcpy(addr.sa_data, slave->perm_hwaddr, ETH_ALEN);
2261 addr.sa_family = slave_dev->type;
2262 dev_set_mac_address(slave_dev, &addr);
2263 }
2264
2265 /* restore the original state of the IFF_NOARP flag that might have
2266 * been set by bond_set_slave_inactive_flags()
2267 */
2268 if ((slave->original_flags & IFF_NOARP) == 0) {
2269 slave_dev->flags &= ~IFF_NOARP;
2270 }
2271
2272 kfree(slave);
2273
2274 /* re-acquire the lock before getting the next slave */
2275 write_lock_bh(&bond->lock);
2276 }
2277
2278 /* zero the mac address of the master so it will be
2279 * set by the application to the mac address of the
2280 * first slave
2281 */
2282 memset(bond_dev->dev_addr, 0, bond_dev->addr_len);
2283
2284 if (list_empty(&bond->vlan_list)) {
2285 bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
2286 } else {
2287 printk(KERN_WARNING DRV_NAME
2288 ": Warning: clearing HW address of %s while it "
2289 "still has VLANs.\n",
2290 bond_dev->name);
2291 printk(KERN_WARNING DRV_NAME
2292 ": When re-adding slaves, make sure the bond's "
2293 "HW address matches its VLANs'.\n");
2294 }
2295
2296 printk(KERN_INFO DRV_NAME
2297 ": %s: released all slaves\n",
2298 bond_dev->name);
2299
2300out:
2301 write_unlock_bh(&bond->lock);
2302
2303 return 0;
2304}
2305
2306/*
2307 * This function changes the active slave to slave <slave_dev>.
2308 * It returns -EINVAL in the following cases.
2309 * - <slave_dev> is not found in the list.
2310 * - There is not active slave now.
2311 * - <slave_dev> is already active.
2312 * - The link state of <slave_dev> is not BOND_LINK_UP.
2313 * - <slave_dev> is not running.
2314 * In these cases, this fuction does nothing.
2315 * In the other cases, currnt_slave pointer is changed and 0 is returned.
2316 */
2317static int bond_ioctl_change_active(struct net_device *bond_dev, struct net_device *slave_dev)
2318{
2319 struct bonding *bond = bond_dev->priv;
2320 struct slave *old_active = NULL;
2321 struct slave *new_active = NULL;
2322 int res = 0;
2323
2324 if (!USES_PRIMARY(bond->params.mode)) {
2325 return -EINVAL;
2326 }
2327
2328 /* Verify that master_dev is indeed the master of slave_dev */
2329 if (!(slave_dev->flags & IFF_SLAVE) ||
2330 (slave_dev->master != bond_dev)) {
2331 return -EINVAL;
2332 }
2333
2334 write_lock_bh(&bond->lock);
2335
2336 old_active = bond->curr_active_slave;
2337 new_active = bond_get_slave_by_dev(bond, slave_dev);
2338
2339 /*
2340 * Changing to the current active: do nothing; return success.
2341 */
2342 if (new_active && (new_active == old_active)) {
2343 write_unlock_bh(&bond->lock);
2344 return 0;
2345 }
2346
2347 if ((new_active) &&
2348 (old_active) &&
2349 (new_active->link == BOND_LINK_UP) &&
2350 IS_UP(new_active->dev)) {
2351 bond_change_active_slave(bond, new_active);
2352 } else {
2353 res = -EINVAL;
2354 }
2355
2356 write_unlock_bh(&bond->lock);
2357
2358 return res;
2359}
2360
2361static int bond_ethtool_ioctl(struct net_device *bond_dev, struct ifreq *ifr)
2362{
2363 struct ethtool_drvinfo info;
2364 void __user *addr = ifr->ifr_data;
2365 uint32_t cmd;
2366
2367 if (get_user(cmd, (uint32_t __user *)addr)) {
2368 return -EFAULT;
2369 }
2370
2371 switch (cmd) {
2372 case ETHTOOL_GDRVINFO:
2373 if (copy_from_user(&info, addr, sizeof(info))) {
2374 return -EFAULT;
2375 }
2376
2377 if (strcmp(info.driver, "ifenslave") == 0) {
2378 int new_abi_ver;
2379 char *endptr;
2380
2381 new_abi_ver = simple_strtoul(info.fw_version,
2382 &endptr, 0);
2383 if (*endptr) {
2384 printk(KERN_ERR DRV_NAME
2385 ": Error: got invalid ABI "
2386 "version from application\n");
2387
2388 return -EINVAL;
2389 }
2390
2391 if (orig_app_abi_ver == -1) {
2392 orig_app_abi_ver = new_abi_ver;
2393 }
2394
2395 app_abi_ver = new_abi_ver;
2396 }
2397
2398 strncpy(info.driver, DRV_NAME, 32);
2399 strncpy(info.version, DRV_VERSION, 32);
2400 snprintf(info.fw_version, 32, "%d", BOND_ABI_VERSION);
2401
2402 if (copy_to_user(addr, &info, sizeof(info))) {
2403 return -EFAULT;
2404 }
2405
2406 return 0;
2407 default:
2408 return -EOPNOTSUPP;
2409 }
2410}
2411
2412static int bond_info_query(struct net_device *bond_dev, struct ifbond *info)
2413{
2414 struct bonding *bond = bond_dev->priv;
2415
2416 info->bond_mode = bond->params.mode;
2417 info->miimon = bond->params.miimon;
2418
2419 read_lock_bh(&bond->lock);
2420 info->num_slaves = bond->slave_cnt;
2421 read_unlock_bh(&bond->lock);
2422
2423 return 0;
2424}
2425
2426static int bond_slave_info_query(struct net_device *bond_dev, struct ifslave *info)
2427{
2428 struct bonding *bond = bond_dev->priv;
2429 struct slave *slave;
2430 int i, found = 0;
2431
2432 if (info->slave_id < 0) {
2433 return -ENODEV;
2434 }
2435
2436 read_lock_bh(&bond->lock);
2437
2438 bond_for_each_slave(bond, slave, i) {
2439 if (i == (int)info->slave_id) {
2440 found = 1;
2441 break;
2442 }
2443 }
2444
2445 read_unlock_bh(&bond->lock);
2446
2447 if (found) {
2448 strcpy(info->slave_name, slave->dev->name);
2449 info->link = slave->link;
2450 info->state = slave->state;
2451 info->link_failure_count = slave->link_failure_count;
2452 } else {
2453 return -ENODEV;
2454 }
2455
2456 return 0;
2457}
2458
2459/*-------------------------------- Monitoring -------------------------------*/
2460
2461/* this function is called regularly to monitor each slave's link. */
2462static void bond_mii_monitor(struct net_device *bond_dev)
2463{
2464 struct bonding *bond = bond_dev->priv;
2465 struct slave *slave, *oldcurrent;
2466 int do_failover = 0;
2467 int delta_in_ticks;
2468 int i;
2469
2470 read_lock(&bond->lock);
2471
2472 delta_in_ticks = (bond->params.miimon * HZ) / 1000;
2473
2474 if (bond->kill_timers) {
2475 goto out;
2476 }
2477
2478 if (bond->slave_cnt == 0) {
2479 goto re_arm;
2480 }
2481
2482 /* we will try to read the link status of each of our slaves, and
2483 * set their IFF_RUNNING flag appropriately. For each slave not
2484 * supporting MII status, we won't do anything so that a user-space
2485 * program could monitor the link itself if needed.
2486 */
2487
2488 read_lock(&bond->curr_slave_lock);
2489 oldcurrent = bond->curr_active_slave;
2490 read_unlock(&bond->curr_slave_lock);
2491
2492 bond_for_each_slave(bond, slave, i) {
2493 struct net_device *slave_dev = slave->dev;
2494 int link_state;
2495 u16 old_speed = slave->speed;
2496 u8 old_duplex = slave->duplex;
2497
2498 link_state = bond_check_dev_link(bond, slave_dev, 0);
2499
2500 switch (slave->link) {
2501 case BOND_LINK_UP: /* the link was up */
2502 if (link_state == BMSR_LSTATUS) {
2503 /* link stays up, nothing more to do */
2504 break;
2505 } else { /* link going down */
2506 slave->link = BOND_LINK_FAIL;
2507 slave->delay = bond->params.downdelay;
2508
2509 if (slave->link_failure_count < UINT_MAX) {
2510 slave->link_failure_count++;
2511 }
2512
2513 if (bond->params.downdelay) {
2514 printk(KERN_INFO DRV_NAME
2515 ": %s: link status down for %s "
2516 "interface %s, disabling it in "
2517 "%d ms.\n",
2518 bond_dev->name,
2519 IS_UP(slave_dev)
2520 ? ((bond->params.mode == BOND_MODE_ACTIVEBACKUP)
2521 ? ((slave == oldcurrent)
2522 ? "active " : "backup ")
2523 : "")
2524 : "idle ",
2525 slave_dev->name,
2526 bond->params.downdelay * bond->params.miimon);
2527 }
2528 }
2529 /* no break ! fall through the BOND_LINK_FAIL test to
2530 ensure proper action to be taken
2531 */
2532 case BOND_LINK_FAIL: /* the link has just gone down */
2533 if (link_state != BMSR_LSTATUS) {
2534 /* link stays down */
2535 if (slave->delay <= 0) {
2536 /* link down for too long time */
2537 slave->link = BOND_LINK_DOWN;
2538
2539 /* in active/backup mode, we must
2540 * completely disable this interface
2541 */
2542 if ((bond->params.mode == BOND_MODE_ACTIVEBACKUP) ||
2543 (bond->params.mode == BOND_MODE_8023AD)) {
2544 bond_set_slave_inactive_flags(slave);
2545 }
2546
2547 printk(KERN_INFO DRV_NAME
2548 ": %s: link status definitely "
2549 "down for interface %s, "
2550 "disabling it\n",
2551 bond_dev->name,
2552 slave_dev->name);
2553
2554 /* notify ad that the link status has changed */
2555 if (bond->params.mode == BOND_MODE_8023AD) {
2556 bond_3ad_handle_link_change(slave, BOND_LINK_DOWN);
2557 }
2558
2559 if ((bond->params.mode == BOND_MODE_TLB) ||
2560 (bond->params.mode == BOND_MODE_ALB)) {
2561 bond_alb_handle_link_change(bond, slave, BOND_LINK_DOWN);
2562 }
2563
2564 if (slave == oldcurrent) {
2565 do_failover = 1;
2566 }
2567 } else {
2568 slave->delay--;
2569 }
2570 } else {
2571 /* link up again */
2572 slave->link = BOND_LINK_UP;
2573 slave->jiffies = jiffies;
2574 printk(KERN_INFO DRV_NAME
2575 ": %s: link status up again after %d "
2576 "ms for interface %s.\n",
2577 bond_dev->name,
2578 (bond->params.downdelay - slave->delay) * bond->params.miimon,
2579 slave_dev->name);
2580 }
2581 break;
2582 case BOND_LINK_DOWN: /* the link was down */
2583 if (link_state != BMSR_LSTATUS) {
2584 /* the link stays down, nothing more to do */
2585 break;
2586 } else { /* link going up */
2587 slave->link = BOND_LINK_BACK;
2588 slave->delay = bond->params.updelay;
2589
2590 if (bond->params.updelay) {
2591 /* if updelay == 0, no need to
2592 advertise about a 0 ms delay */
2593 printk(KERN_INFO DRV_NAME
2594 ": %s: link status up for "
2595 "interface %s, enabling it "
2596 "in %d ms.\n",
2597 bond_dev->name,
2598 slave_dev->name,
2599 bond->params.updelay * bond->params.miimon);
2600 }
2601 }
2602 /* no break ! fall through the BOND_LINK_BACK state in
2603 case there's something to do.
2604 */
2605 case BOND_LINK_BACK: /* the link has just come back */
2606 if (link_state != BMSR_LSTATUS) {
2607 /* link down again */
2608 slave->link = BOND_LINK_DOWN;
2609
2610 printk(KERN_INFO DRV_NAME
2611 ": %s: link status down again after %d "
2612 "ms for interface %s.\n",
2613 bond_dev->name,
2614 (bond->params.updelay - slave->delay) * bond->params.miimon,
2615 slave_dev->name);
2616 } else {
2617 /* link stays up */
2618 if (slave->delay == 0) {
2619 /* now the link has been up for long time enough */
2620 slave->link = BOND_LINK_UP;
2621 slave->jiffies = jiffies;
2622
2623 if (bond->params.mode == BOND_MODE_8023AD) {
2624 /* prevent it from being the active one */
2625 slave->state = BOND_STATE_BACKUP;
2626 } else if (bond->params.mode != BOND_MODE_ACTIVEBACKUP) {
2627 /* make it immediately active */
2628 slave->state = BOND_STATE_ACTIVE;
2629 } else if (slave != bond->primary_slave) {
2630 /* prevent it from being the active one */
2631 slave->state = BOND_STATE_BACKUP;
2632 }
2633
2634 printk(KERN_INFO DRV_NAME
2635 ": %s: link status definitely "
2636 "up for interface %s.\n",
2637 bond_dev->name,
2638 slave_dev->name);
2639
2640 /* notify ad that the link status has changed */
2641 if (bond->params.mode == BOND_MODE_8023AD) {
2642 bond_3ad_handle_link_change(slave, BOND_LINK_UP);
2643 }
2644
2645 if ((bond->params.mode == BOND_MODE_TLB) ||
2646 (bond->params.mode == BOND_MODE_ALB)) {
2647 bond_alb_handle_link_change(bond, slave, BOND_LINK_UP);
2648 }
2649
2650 if ((!oldcurrent) ||
2651 (slave == bond->primary_slave)) {
2652 do_failover = 1;
2653 }
2654 } else {
2655 slave->delay--;
2656 }
2657 }
2658 break;
2659 default:
2660 /* Should not happen */
2661 printk(KERN_ERR "bonding: Error: %s Illegal value (link=%d)\n",
2662 slave->dev->name, slave->link);
2663 goto out;
2664 } /* end of switch (slave->link) */
2665
2666 bond_update_speed_duplex(slave);
2667
2668 if (bond->params.mode == BOND_MODE_8023AD) {
2669 if (old_speed != slave->speed) {
2670 bond_3ad_adapter_speed_changed(slave);
2671 }
2672
2673 if (old_duplex != slave->duplex) {
2674 bond_3ad_adapter_duplex_changed(slave);
2675 }
2676 }
2677
2678 } /* end of for */
2679
2680 if (do_failover) {
2681 write_lock(&bond->curr_slave_lock);
2682
2683 bond_select_active_slave(bond);
2684
2685 if (oldcurrent && !bond->curr_active_slave) {
2686 printk(KERN_INFO DRV_NAME
2687 ": %s: now running without any active "
2688 "interface !\n",
2689 bond_dev->name);
2690 }
2691
2692 write_unlock(&bond->curr_slave_lock);
2693 }
2694
2695re_arm:
2696 if (bond->params.miimon) {
2697 mod_timer(&bond->mii_timer, jiffies + delta_in_ticks);
2698 }
2699out:
2700 read_unlock(&bond->lock);
2701}
2702
Jay Vosburghc3ade5c2005-06-26 17:52:20 -04002703
2704static u32 bond_glean_dev_ip(struct net_device *dev)
2705{
2706 struct in_device *idev;
2707 struct in_ifaddr *ifa;
2708 u32 addr = 0;
2709
2710 if (!dev)
2711 return 0;
2712
2713 rcu_read_lock();
2714 idev = __in_dev_get(dev);
2715 if (!idev)
2716 goto out;
2717
2718 ifa = idev->ifa_list;
2719 if (!ifa)
2720 goto out;
2721
2722 addr = ifa->ifa_local;
2723out:
2724 rcu_read_unlock();
2725 return addr;
2726}
2727
2728static int bond_has_ip(struct bonding *bond)
2729{
2730 struct vlan_entry *vlan, *vlan_next;
2731
2732 if (bond->master_ip)
2733 return 1;
2734
2735 if (list_empty(&bond->vlan_list))
2736 return 0;
2737
2738 list_for_each_entry_safe(vlan, vlan_next, &bond->vlan_list,
2739 vlan_list) {
2740 if (vlan->vlan_ip)
2741 return 1;
2742 }
2743
2744 return 0;
2745}
2746
2747/*
2748 * We go to the (large) trouble of VLAN tagging ARP frames because
2749 * switches in VLAN mode (especially if ports are configured as
2750 * "native" to a VLAN) might not pass non-tagged frames.
2751 */
2752static void bond_arp_send(struct net_device *slave_dev, int arp_op, u32 dest_ip, u32 src_ip, unsigned short vlan_id)
2753{
2754 struct sk_buff *skb;
2755
2756 dprintk("arp %d on slave %s: dst %x src %x vid %d\n", arp_op,
2757 slave_dev->name, dest_ip, src_ip, vlan_id);
2758
2759 skb = arp_create(arp_op, ETH_P_ARP, dest_ip, slave_dev, src_ip,
2760 NULL, slave_dev->dev_addr, NULL);
2761
2762 if (!skb) {
2763 printk(KERN_ERR DRV_NAME ": ARP packet allocation failed\n");
2764 return;
2765 }
2766 if (vlan_id) {
2767 skb = vlan_put_tag(skb, vlan_id);
2768 if (!skb) {
2769 printk(KERN_ERR DRV_NAME ": failed to insert VLAN tag\n");
2770 return;
2771 }
2772 }
2773 arp_xmit(skb);
2774}
2775
2776
Linus Torvalds1da177e2005-04-16 15:20:36 -07002777static void bond_arp_send_all(struct bonding *bond, struct slave *slave)
2778{
Jay Vosburghc3ade5c2005-06-26 17:52:20 -04002779 int i, vlan_id, rv;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002780 u32 *targets = bond->params.arp_targets;
Jay Vosburghc3ade5c2005-06-26 17:52:20 -04002781 struct vlan_entry *vlan, *vlan_next;
2782 struct net_device *vlan_dev;
2783 struct flowi fl;
2784 struct rtable *rt;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002785
2786 for (i = 0; (i < BOND_MAX_ARP_TARGETS) && targets[i]; i++) {
Jay Vosburghc3ade5c2005-06-26 17:52:20 -04002787 dprintk("basa: target %x\n", targets[i]);
2788 if (list_empty(&bond->vlan_list)) {
2789 dprintk("basa: empty vlan: arp_send\n");
2790 bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2791 bond->master_ip, 0);
2792 continue;
2793 }
2794
2795 /*
2796 * If VLANs are configured, we do a route lookup to
2797 * determine which VLAN interface would be used, so we
2798 * can tag the ARP with the proper VLAN tag.
2799 */
2800 memset(&fl, 0, sizeof(fl));
2801 fl.fl4_dst = targets[i];
2802 fl.fl4_tos = RTO_ONLINK;
2803
2804 rv = ip_route_output_key(&rt, &fl);
2805 if (rv) {
2806 if (net_ratelimit()) {
2807 printk(KERN_WARNING DRV_NAME
2808 ": %s: no route to arp_ip_target %u.%u.%u.%u\n",
2809 bond->dev->name, NIPQUAD(fl.fl4_dst));
2810 }
2811 continue;
2812 }
2813
2814 /*
2815 * This target is not on a VLAN
2816 */
2817 if (rt->u.dst.dev == bond->dev) {
2818 dprintk("basa: rtdev == bond->dev: arp_send\n");
2819 bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2820 bond->master_ip, 0);
2821 continue;
2822 }
2823
2824 vlan_id = 0;
2825 list_for_each_entry_safe(vlan, vlan_next, &bond->vlan_list,
2826 vlan_list) {
2827 vlan_dev = bond->vlgrp->vlan_devices[vlan->vlan_id];
2828 if (vlan_dev == rt->u.dst.dev) {
2829 vlan_id = vlan->vlan_id;
2830 dprintk("basa: vlan match on %s %d\n",
2831 vlan_dev->name, vlan_id);
2832 break;
2833 }
2834 }
2835
2836 if (vlan_id) {
2837 bond_arp_send(slave->dev, ARPOP_REQUEST, targets[i],
2838 vlan->vlan_ip, vlan_id);
2839 continue;
2840 }
2841
2842 if (net_ratelimit()) {
2843 printk(KERN_WARNING DRV_NAME
2844 ": %s: no path to arp_ip_target %u.%u.%u.%u via rt.dev %s\n",
2845 bond->dev->name, NIPQUAD(fl.fl4_dst),
2846 rt->u.dst.dev ? rt->u.dst.dev->name : "NULL");
2847 }
2848 }
2849}
2850
2851/*
2852 * Kick out a gratuitous ARP for an IP on the bonding master plus one
2853 * for each VLAN above us.
2854 */
2855static void bond_send_gratuitous_arp(struct bonding *bond)
2856{
2857 struct slave *slave = bond->curr_active_slave;
2858 struct vlan_entry *vlan;
2859 struct net_device *vlan_dev;
2860
2861 dprintk("bond_send_grat_arp: bond %s slave %s\n", bond->dev->name,
2862 slave ? slave->dev->name : "NULL");
2863 if (!slave)
2864 return;
2865
2866 if (bond->master_ip) {
2867 bond_arp_send(slave->dev, ARPOP_REPLY, bond->master_ip,
2868 bond->master_ip, 0);
2869 }
2870
2871 list_for_each_entry(vlan, &bond->vlan_list, vlan_list) {
2872 vlan_dev = bond->vlgrp->vlan_devices[vlan->vlan_id];
2873 if (vlan->vlan_ip) {
2874 bond_arp_send(slave->dev, ARPOP_REPLY, vlan->vlan_ip,
2875 vlan->vlan_ip, vlan->vlan_id);
2876 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002877 }
2878}
2879
2880/*
2881 * this function is called regularly to monitor each slave's link
2882 * ensuring that traffic is being sent and received when arp monitoring
2883 * is used in load-balancing mode. if the adapter has been dormant, then an
2884 * arp is transmitted to generate traffic. see activebackup_arp_monitor for
2885 * arp monitoring in active backup mode.
2886 */
2887static void bond_loadbalance_arp_mon(struct net_device *bond_dev)
2888{
2889 struct bonding *bond = bond_dev->priv;
2890 struct slave *slave, *oldcurrent;
2891 int do_failover = 0;
2892 int delta_in_ticks;
2893 int i;
2894
2895 read_lock(&bond->lock);
2896
2897 delta_in_ticks = (bond->params.arp_interval * HZ) / 1000;
2898
2899 if (bond->kill_timers) {
2900 goto out;
2901 }
2902
2903 if (bond->slave_cnt == 0) {
2904 goto re_arm;
2905 }
2906
2907 read_lock(&bond->curr_slave_lock);
2908 oldcurrent = bond->curr_active_slave;
2909 read_unlock(&bond->curr_slave_lock);
2910
2911 /* see if any of the previous devices are up now (i.e. they have
2912 * xmt and rcv traffic). the curr_active_slave does not come into
2913 * the picture unless it is null. also, slave->jiffies is not needed
2914 * here because we send an arp on each slave and give a slave as
2915 * long as it needs to get the tx/rx within the delta.
2916 * TODO: what about up/down delay in arp mode? it wasn't here before
2917 * so it can wait
2918 */
2919 bond_for_each_slave(bond, slave, i) {
2920 if (slave->link != BOND_LINK_UP) {
2921 if (((jiffies - slave->dev->trans_start) <= delta_in_ticks) &&
2922 ((jiffies - slave->dev->last_rx) <= delta_in_ticks)) {
2923
2924 slave->link = BOND_LINK_UP;
2925 slave->state = BOND_STATE_ACTIVE;
2926
2927 /* primary_slave has no meaning in round-robin
2928 * mode. the window of a slave being up and
2929 * curr_active_slave being null after enslaving
2930 * is closed.
2931 */
2932 if (!oldcurrent) {
2933 printk(KERN_INFO DRV_NAME
2934 ": %s: link status definitely "
2935 "up for interface %s, ",
2936 bond_dev->name,
2937 slave->dev->name);
2938 do_failover = 1;
2939 } else {
2940 printk(KERN_INFO DRV_NAME
2941 ": %s: interface %s is now up\n",
2942 bond_dev->name,
2943 slave->dev->name);
2944 }
2945 }
2946 } else {
2947 /* slave->link == BOND_LINK_UP */
2948
2949 /* not all switches will respond to an arp request
2950 * when the source ip is 0, so don't take the link down
2951 * if we don't know our ip yet
2952 */
2953 if (((jiffies - slave->dev->trans_start) >= (2*delta_in_ticks)) ||
2954 (((jiffies - slave->dev->last_rx) >= (2*delta_in_ticks)) &&
Jay Vosburghc3ade5c2005-06-26 17:52:20 -04002955 bond_has_ip(bond))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002956
2957 slave->link = BOND_LINK_DOWN;
2958 slave->state = BOND_STATE_BACKUP;
2959
2960 if (slave->link_failure_count < UINT_MAX) {
2961 slave->link_failure_count++;
2962 }
2963
2964 printk(KERN_INFO DRV_NAME
2965 ": %s: interface %s is now down.\n",
2966 bond_dev->name,
2967 slave->dev->name);
2968
2969 if (slave == oldcurrent) {
2970 do_failover = 1;
2971 }
2972 }
2973 }
2974
2975 /* note: if switch is in round-robin mode, all links
2976 * must tx arp to ensure all links rx an arp - otherwise
2977 * links may oscillate or not come up at all; if switch is
2978 * in something like xor mode, there is nothing we can
2979 * do - all replies will be rx'ed on same link causing slaves
2980 * to be unstable during low/no traffic periods
2981 */
2982 if (IS_UP(slave->dev)) {
2983 bond_arp_send_all(bond, slave);
2984 }
2985 }
2986
2987 if (do_failover) {
2988 write_lock(&bond->curr_slave_lock);
2989
2990 bond_select_active_slave(bond);
2991
2992 if (oldcurrent && !bond->curr_active_slave) {
2993 printk(KERN_INFO DRV_NAME
2994 ": %s: now running without any active "
2995 "interface !\n",
2996 bond_dev->name);
2997 }
2998
2999 write_unlock(&bond->curr_slave_lock);
3000 }
3001
3002re_arm:
3003 if (bond->params.arp_interval) {
3004 mod_timer(&bond->arp_timer, jiffies + delta_in_ticks);
3005 }
3006out:
3007 read_unlock(&bond->lock);
3008}
3009
3010/*
3011 * When using arp monitoring in active-backup mode, this function is
3012 * called to determine if any backup slaves have went down or a new
3013 * current slave needs to be found.
3014 * The backup slaves never generate traffic, they are considered up by merely
3015 * receiving traffic. If the current slave goes down, each backup slave will
3016 * be given the opportunity to tx/rx an arp before being taken down - this
3017 * prevents all slaves from being taken down due to the current slave not
3018 * sending any traffic for the backups to receive. The arps are not necessarily
3019 * necessary, any tx and rx traffic will keep the current slave up. While any
3020 * rx traffic will keep the backup slaves up, the current slave is responsible
3021 * for generating traffic to keep them up regardless of any other traffic they
3022 * may have received.
3023 * see loadbalance_arp_monitor for arp monitoring in load balancing mode
3024 */
3025static void bond_activebackup_arp_mon(struct net_device *bond_dev)
3026{
3027 struct bonding *bond = bond_dev->priv;
3028 struct slave *slave;
3029 int delta_in_ticks;
3030 int i;
3031
3032 read_lock(&bond->lock);
3033
3034 delta_in_ticks = (bond->params.arp_interval * HZ) / 1000;
3035
3036 if (bond->kill_timers) {
3037 goto out;
3038 }
3039
3040 if (bond->slave_cnt == 0) {
3041 goto re_arm;
3042 }
3043
3044 /* determine if any slave has come up or any backup slave has
3045 * gone down
3046 * TODO: what about up/down delay in arp mode? it wasn't here before
3047 * so it can wait
3048 */
3049 bond_for_each_slave(bond, slave, i) {
3050 if (slave->link != BOND_LINK_UP) {
3051 if ((jiffies - slave->dev->last_rx) <= delta_in_ticks) {
3052
3053 slave->link = BOND_LINK_UP;
3054
3055 write_lock(&bond->curr_slave_lock);
3056
3057 if ((!bond->curr_active_slave) &&
3058 ((jiffies - slave->dev->trans_start) <= delta_in_ticks)) {
3059 bond_change_active_slave(bond, slave);
3060 bond->current_arp_slave = NULL;
3061 } else if (bond->curr_active_slave != slave) {
3062 /* this slave has just come up but we
3063 * already have a current slave; this
3064 * can also happen if bond_enslave adds
3065 * a new slave that is up while we are
3066 * searching for a new slave
3067 */
3068 bond_set_slave_inactive_flags(slave);
3069 bond->current_arp_slave = NULL;
3070 }
3071
3072 if (slave == bond->curr_active_slave) {
3073 printk(KERN_INFO DRV_NAME
3074 ": %s: %s is up and now the "
3075 "active interface\n",
3076 bond_dev->name,
3077 slave->dev->name);
3078 } else {
3079 printk(KERN_INFO DRV_NAME
3080 ": %s: backup interface %s is "
3081 "now up\n",
3082 bond_dev->name,
3083 slave->dev->name);
3084 }
3085
3086 write_unlock(&bond->curr_slave_lock);
3087 }
3088 } else {
3089 read_lock(&bond->curr_slave_lock);
3090
3091 if ((slave != bond->curr_active_slave) &&
3092 (!bond->current_arp_slave) &&
3093 (((jiffies - slave->dev->last_rx) >= 3*delta_in_ticks) &&
Jay Vosburghc3ade5c2005-06-26 17:52:20 -04003094 bond_has_ip(bond))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003095 /* a backup slave has gone down; three times
3096 * the delta allows the current slave to be
3097 * taken out before the backup slave.
3098 * note: a non-null current_arp_slave indicates
3099 * the curr_active_slave went down and we are
3100 * searching for a new one; under this
3101 * condition we only take the curr_active_slave
3102 * down - this gives each slave a chance to
3103 * tx/rx traffic before being taken out
3104 */
3105
3106 read_unlock(&bond->curr_slave_lock);
3107
3108 slave->link = BOND_LINK_DOWN;
3109
3110 if (slave->link_failure_count < UINT_MAX) {
3111 slave->link_failure_count++;
3112 }
3113
3114 bond_set_slave_inactive_flags(slave);
3115
3116 printk(KERN_INFO DRV_NAME
3117 ": %s: backup interface %s is now down\n",
3118 bond_dev->name,
3119 slave->dev->name);
3120 } else {
3121 read_unlock(&bond->curr_slave_lock);
3122 }
3123 }
3124 }
3125
3126 read_lock(&bond->curr_slave_lock);
3127 slave = bond->curr_active_slave;
3128 read_unlock(&bond->curr_slave_lock);
3129
3130 if (slave) {
3131 /* if we have sent traffic in the past 2*arp_intervals but
3132 * haven't xmit and rx traffic in that time interval, select
3133 * a different slave. slave->jiffies is only updated when
3134 * a slave first becomes the curr_active_slave - not necessarily
3135 * after every arp; this ensures the slave has a full 2*delta
3136 * before being taken out. if a primary is being used, check
3137 * if it is up and needs to take over as the curr_active_slave
3138 */
3139 if ((((jiffies - slave->dev->trans_start) >= (2*delta_in_ticks)) ||
Jay Vosburghc3ade5c2005-06-26 17:52:20 -04003140 (((jiffies - slave->dev->last_rx) >= (2*delta_in_ticks)) &&
3141 bond_has_ip(bond))) &&
Linus Torvalds1da177e2005-04-16 15:20:36 -07003142 ((jiffies - slave->jiffies) >= 2*delta_in_ticks)) {
3143
3144 slave->link = BOND_LINK_DOWN;
3145
3146 if (slave->link_failure_count < UINT_MAX) {
3147 slave->link_failure_count++;
3148 }
3149
3150 printk(KERN_INFO DRV_NAME
3151 ": %s: link status down for active interface "
3152 "%s, disabling it\n",
3153 bond_dev->name,
3154 slave->dev->name);
3155
3156 write_lock(&bond->curr_slave_lock);
3157
3158 bond_select_active_slave(bond);
3159 slave = bond->curr_active_slave;
3160
3161 write_unlock(&bond->curr_slave_lock);
3162
3163 bond->current_arp_slave = slave;
3164
3165 if (slave) {
3166 slave->jiffies = jiffies;
3167 }
3168 } else if ((bond->primary_slave) &&
3169 (bond->primary_slave != slave) &&
3170 (bond->primary_slave->link == BOND_LINK_UP)) {
3171 /* at this point, slave is the curr_active_slave */
3172 printk(KERN_INFO DRV_NAME
3173 ": %s: changing from interface %s to primary "
3174 "interface %s\n",
3175 bond_dev->name,
3176 slave->dev->name,
3177 bond->primary_slave->dev->name);
3178
3179 /* primary is up so switch to it */
3180 write_lock(&bond->curr_slave_lock);
3181 bond_change_active_slave(bond, bond->primary_slave);
3182 write_unlock(&bond->curr_slave_lock);
3183
3184 slave = bond->primary_slave;
3185 slave->jiffies = jiffies;
3186 } else {
3187 bond->current_arp_slave = NULL;
3188 }
3189
3190 /* the current slave must tx an arp to ensure backup slaves
3191 * rx traffic
3192 */
Jay Vosburghc3ade5c2005-06-26 17:52:20 -04003193 if (slave && bond_has_ip(bond)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003194 bond_arp_send_all(bond, slave);
3195 }
3196 }
3197
3198 /* if we don't have a curr_active_slave, search for the next available
3199 * backup slave from the current_arp_slave and make it the candidate
3200 * for becoming the curr_active_slave
3201 */
3202 if (!slave) {
3203 if (!bond->current_arp_slave) {
3204 bond->current_arp_slave = bond->first_slave;
3205 }
3206
3207 if (bond->current_arp_slave) {
3208 bond_set_slave_inactive_flags(bond->current_arp_slave);
3209
3210 /* search for next candidate */
Jay Vosburgh2f872f02005-05-26 12:56:59 -07003211 bond_for_each_slave_from(bond, slave, i, bond->current_arp_slave->next) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003212 if (IS_UP(slave->dev)) {
3213 slave->link = BOND_LINK_BACK;
3214 bond_set_slave_active_flags(slave);
3215 bond_arp_send_all(bond, slave);
3216 slave->jiffies = jiffies;
3217 bond->current_arp_slave = slave;
3218 break;
3219 }
3220
3221 /* if the link state is up at this point, we
3222 * mark it down - this can happen if we have
3223 * simultaneous link failures and
3224 * reselect_active_interface doesn't make this
3225 * one the current slave so it is still marked
3226 * up when it is actually down
3227 */
3228 if (slave->link == BOND_LINK_UP) {
3229 slave->link = BOND_LINK_DOWN;
3230 if (slave->link_failure_count < UINT_MAX) {
3231 slave->link_failure_count++;
3232 }
3233
3234 bond_set_slave_inactive_flags(slave);
3235
3236 printk(KERN_INFO DRV_NAME
3237 ": %s: backup interface %s is "
3238 "now down.\n",
3239 bond_dev->name,
3240 slave->dev->name);
3241 }
3242 }
3243 }
3244 }
3245
3246re_arm:
3247 if (bond->params.arp_interval) {
3248 mod_timer(&bond->arp_timer, jiffies + delta_in_ticks);
3249 }
3250out:
3251 read_unlock(&bond->lock);
3252}
3253
3254/*------------------------------ proc/seq_file-------------------------------*/
3255
3256#ifdef CONFIG_PROC_FS
3257
3258#define SEQ_START_TOKEN ((void *)1)
3259
3260static void *bond_info_seq_start(struct seq_file *seq, loff_t *pos)
3261{
3262 struct bonding *bond = seq->private;
3263 loff_t off = 0;
3264 struct slave *slave;
3265 int i;
3266
3267 /* make sure the bond won't be taken away */
3268 read_lock(&dev_base_lock);
3269 read_lock_bh(&bond->lock);
3270
3271 if (*pos == 0) {
3272 return SEQ_START_TOKEN;
3273 }
3274
3275 bond_for_each_slave(bond, slave, i) {
3276 if (++off == *pos) {
3277 return slave;
3278 }
3279 }
3280
3281 return NULL;
3282}
3283
3284static void *bond_info_seq_next(struct seq_file *seq, void *v, loff_t *pos)
3285{
3286 struct bonding *bond = seq->private;
3287 struct slave *slave = v;
3288
3289 ++*pos;
3290 if (v == SEQ_START_TOKEN) {
3291 return bond->first_slave;
3292 }
3293
3294 slave = slave->next;
3295
3296 return (slave == bond->first_slave) ? NULL : slave;
3297}
3298
3299static void bond_info_seq_stop(struct seq_file *seq, void *v)
3300{
3301 struct bonding *bond = seq->private;
3302
3303 read_unlock_bh(&bond->lock);
3304 read_unlock(&dev_base_lock);
3305}
3306
3307static void bond_info_show_master(struct seq_file *seq)
3308{
3309 struct bonding *bond = seq->private;
3310 struct slave *curr;
3311
3312 read_lock(&bond->curr_slave_lock);
3313 curr = bond->curr_active_slave;
3314 read_unlock(&bond->curr_slave_lock);
3315
3316 seq_printf(seq, "Bonding Mode: %s\n",
3317 bond_mode_name(bond->params.mode));
3318
3319 if (USES_PRIMARY(bond->params.mode)) {
3320 seq_printf(seq, "Primary Slave: %s\n",
3321 (bond->params.primary[0]) ?
3322 bond->params.primary : "None");
3323
3324 seq_printf(seq, "Currently Active Slave: %s\n",
3325 (curr) ? curr->dev->name : "None");
3326 }
3327
3328 seq_printf(seq, "MII Status: %s\n", (curr) ? "up" : "down");
3329 seq_printf(seq, "MII Polling Interval (ms): %d\n", bond->params.miimon);
3330 seq_printf(seq, "Up Delay (ms): %d\n",
3331 bond->params.updelay * bond->params.miimon);
3332 seq_printf(seq, "Down Delay (ms): %d\n",
3333 bond->params.downdelay * bond->params.miimon);
3334
3335 if (bond->params.mode == BOND_MODE_8023AD) {
3336 struct ad_info ad_info;
3337
3338 seq_puts(seq, "\n802.3ad info\n");
3339 seq_printf(seq, "LACP rate: %s\n",
3340 (bond->params.lacp_fast) ? "fast" : "slow");
3341
3342 if (bond_3ad_get_active_agg_info(bond, &ad_info)) {
3343 seq_printf(seq, "bond %s has no active aggregator\n",
3344 bond->dev->name);
3345 } else {
3346 seq_printf(seq, "Active Aggregator Info:\n");
3347
3348 seq_printf(seq, "\tAggregator ID: %d\n",
3349 ad_info.aggregator_id);
3350 seq_printf(seq, "\tNumber of ports: %d\n",
3351 ad_info.ports);
3352 seq_printf(seq, "\tActor Key: %d\n",
3353 ad_info.actor_key);
3354 seq_printf(seq, "\tPartner Key: %d\n",
3355 ad_info.partner_key);
3356 seq_printf(seq, "\tPartner Mac Address: %02x:%02x:%02x:%02x:%02x:%02x\n",
3357 ad_info.partner_system[0],
3358 ad_info.partner_system[1],
3359 ad_info.partner_system[2],
3360 ad_info.partner_system[3],
3361 ad_info.partner_system[4],
3362 ad_info.partner_system[5]);
3363 }
3364 }
3365}
3366
3367static void bond_info_show_slave(struct seq_file *seq, const struct slave *slave)
3368{
3369 struct bonding *bond = seq->private;
3370
3371 seq_printf(seq, "\nSlave Interface: %s\n", slave->dev->name);
3372 seq_printf(seq, "MII Status: %s\n",
3373 (slave->link == BOND_LINK_UP) ? "up" : "down");
3374 seq_printf(seq, "Link Failure Count: %d\n",
3375 slave->link_failure_count);
3376
3377 if (app_abi_ver >= 1) {
3378 seq_printf(seq,
3379 "Permanent HW addr: %02x:%02x:%02x:%02x:%02x:%02x\n",
3380 slave->perm_hwaddr[0],
3381 slave->perm_hwaddr[1],
3382 slave->perm_hwaddr[2],
3383 slave->perm_hwaddr[3],
3384 slave->perm_hwaddr[4],
3385 slave->perm_hwaddr[5]);
3386 }
3387
3388 if (bond->params.mode == BOND_MODE_8023AD) {
3389 const struct aggregator *agg
3390 = SLAVE_AD_INFO(slave).port.aggregator;
3391
3392 if (agg) {
3393 seq_printf(seq, "Aggregator ID: %d\n",
3394 agg->aggregator_identifier);
3395 } else {
3396 seq_puts(seq, "Aggregator ID: N/A\n");
3397 }
3398 }
3399}
3400
3401static int bond_info_seq_show(struct seq_file *seq, void *v)
3402{
3403 if (v == SEQ_START_TOKEN) {
3404 seq_printf(seq, "%s\n", version);
3405 bond_info_show_master(seq);
3406 } else {
3407 bond_info_show_slave(seq, v);
3408 }
3409
3410 return 0;
3411}
3412
3413static struct seq_operations bond_info_seq_ops = {
3414 .start = bond_info_seq_start,
3415 .next = bond_info_seq_next,
3416 .stop = bond_info_seq_stop,
3417 .show = bond_info_seq_show,
3418};
3419
3420static int bond_info_open(struct inode *inode, struct file *file)
3421{
3422 struct seq_file *seq;
3423 struct proc_dir_entry *proc;
3424 int res;
3425
3426 res = seq_open(file, &bond_info_seq_ops);
3427 if (!res) {
3428 /* recover the pointer buried in proc_dir_entry data */
3429 seq = file->private_data;
3430 proc = PDE(inode);
3431 seq->private = proc->data;
3432 }
3433
3434 return res;
3435}
3436
3437static struct file_operations bond_info_fops = {
3438 .owner = THIS_MODULE,
3439 .open = bond_info_open,
3440 .read = seq_read,
3441 .llseek = seq_lseek,
3442 .release = seq_release,
3443};
3444
3445static int bond_create_proc_entry(struct bonding *bond)
3446{
3447 struct net_device *bond_dev = bond->dev;
3448
3449 if (bond_proc_dir) {
3450 bond->proc_entry = create_proc_entry(bond_dev->name,
3451 S_IRUGO,
3452 bond_proc_dir);
3453 if (bond->proc_entry == NULL) {
3454 printk(KERN_WARNING DRV_NAME
3455 ": Warning: Cannot create /proc/net/%s/%s\n",
3456 DRV_NAME, bond_dev->name);
3457 } else {
3458 bond->proc_entry->data = bond;
3459 bond->proc_entry->proc_fops = &bond_info_fops;
3460 bond->proc_entry->owner = THIS_MODULE;
3461 memcpy(bond->proc_file_name, bond_dev->name, IFNAMSIZ);
3462 }
3463 }
3464
3465 return 0;
3466}
3467
3468static void bond_remove_proc_entry(struct bonding *bond)
3469{
3470 if (bond_proc_dir && bond->proc_entry) {
3471 remove_proc_entry(bond->proc_file_name, bond_proc_dir);
3472 memset(bond->proc_file_name, 0, IFNAMSIZ);
3473 bond->proc_entry = NULL;
3474 }
3475}
3476
3477/* Create the bonding directory under /proc/net, if doesn't exist yet.
3478 * Caller must hold rtnl_lock.
3479 */
3480static void bond_create_proc_dir(void)
3481{
3482 int len = strlen(DRV_NAME);
3483
3484 for (bond_proc_dir = proc_net->subdir; bond_proc_dir;
3485 bond_proc_dir = bond_proc_dir->next) {
3486 if ((bond_proc_dir->namelen == len) &&
3487 !memcmp(bond_proc_dir->name, DRV_NAME, len)) {
3488 break;
3489 }
3490 }
3491
3492 if (!bond_proc_dir) {
3493 bond_proc_dir = proc_mkdir(DRV_NAME, proc_net);
3494 if (bond_proc_dir) {
3495 bond_proc_dir->owner = THIS_MODULE;
3496 } else {
3497 printk(KERN_WARNING DRV_NAME
3498 ": Warning: cannot create /proc/net/%s\n",
3499 DRV_NAME);
3500 }
3501 }
3502}
3503
3504/* Destroy the bonding directory under /proc/net, if empty.
3505 * Caller must hold rtnl_lock.
3506 */
3507static void bond_destroy_proc_dir(void)
3508{
3509 struct proc_dir_entry *de;
3510
3511 if (!bond_proc_dir) {
3512 return;
3513 }
3514
3515 /* verify that the /proc dir is empty */
3516 for (de = bond_proc_dir->subdir; de; de = de->next) {
3517 /* ignore . and .. */
3518 if (*(de->name) != '.') {
3519 break;
3520 }
3521 }
3522
3523 if (de) {
3524 if (bond_proc_dir->owner == THIS_MODULE) {
3525 bond_proc_dir->owner = NULL;
3526 }
3527 } else {
3528 remove_proc_entry(DRV_NAME, proc_net);
3529 bond_proc_dir = NULL;
3530 }
3531}
3532#endif /* CONFIG_PROC_FS */
3533
3534/*-------------------------- netdev event handling --------------------------*/
3535
3536/*
3537 * Change device name
3538 */
3539static int bond_event_changename(struct bonding *bond)
3540{
3541#ifdef CONFIG_PROC_FS
3542 bond_remove_proc_entry(bond);
3543 bond_create_proc_entry(bond);
3544#endif
3545
3546 return NOTIFY_DONE;
3547}
3548
3549static int bond_master_netdev_event(unsigned long event, struct net_device *bond_dev)
3550{
3551 struct bonding *event_bond = bond_dev->priv;
3552
3553 switch (event) {
3554 case NETDEV_CHANGENAME:
3555 return bond_event_changename(event_bond);
3556 case NETDEV_UNREGISTER:
3557 /*
3558 * TODO: remove a bond from the list?
3559 */
3560 break;
3561 default:
3562 break;
3563 }
3564
3565 return NOTIFY_DONE;
3566}
3567
3568static int bond_slave_netdev_event(unsigned long event, struct net_device *slave_dev)
3569{
3570 struct net_device *bond_dev = slave_dev->master;
3571
3572 switch (event) {
3573 case NETDEV_UNREGISTER:
3574 if (bond_dev) {
3575 bond_release(bond_dev, slave_dev);
3576 }
3577 break;
3578 case NETDEV_CHANGE:
3579 /*
3580 * TODO: is this what we get if somebody
3581 * sets up a hierarchical bond, then rmmod's
3582 * one of the slave bonding devices?
3583 */
3584 break;
3585 case NETDEV_DOWN:
3586 /*
3587 * ... Or is it this?
3588 */
3589 break;
3590 case NETDEV_CHANGEMTU:
3591 /*
3592 * TODO: Should slaves be allowed to
3593 * independently alter their MTU? For
3594 * an active-backup bond, slaves need
3595 * not be the same type of device, so
3596 * MTUs may vary. For other modes,
3597 * slaves arguably should have the
3598 * same MTUs. To do this, we'd need to
3599 * take over the slave's change_mtu
3600 * function for the duration of their
3601 * servitude.
3602 */
3603 break;
3604 case NETDEV_CHANGENAME:
3605 /*
3606 * TODO: handle changing the primary's name
3607 */
3608 break;
3609 default:
3610 break;
3611 }
3612
3613 return NOTIFY_DONE;
3614}
3615
3616/*
3617 * bond_netdev_event: handle netdev notifier chain events.
3618 *
3619 * This function receives events for the netdev chain. The caller (an
3620 * ioctl handler calling notifier_call_chain) holds the necessary
3621 * locks for us to safely manipulate the slave devices (RTNL lock,
3622 * dev_probe_lock).
3623 */
3624static int bond_netdev_event(struct notifier_block *this, unsigned long event, void *ptr)
3625{
3626 struct net_device *event_dev = (struct net_device *)ptr;
3627
3628 dprintk("event_dev: %s, event: %lx\n",
3629 (event_dev ? event_dev->name : "None"),
3630 event);
3631
3632 if (event_dev->flags & IFF_MASTER) {
3633 dprintk("IFF_MASTER\n");
3634 return bond_master_netdev_event(event, event_dev);
3635 }
3636
3637 if (event_dev->flags & IFF_SLAVE) {
3638 dprintk("IFF_SLAVE\n");
3639 return bond_slave_netdev_event(event, event_dev);
3640 }
3641
3642 return NOTIFY_DONE;
3643}
3644
Jay Vosburghc3ade5c2005-06-26 17:52:20 -04003645/*
3646 * bond_inetaddr_event: handle inetaddr notifier chain events.
3647 *
3648 * We keep track of device IPs primarily to use as source addresses in
3649 * ARP monitor probes (rather than spewing out broadcasts all the time).
3650 *
3651 * We track one IP for the main device (if it has one), plus one per VLAN.
3652 */
3653static int bond_inetaddr_event(struct notifier_block *this, unsigned long event, void *ptr)
3654{
3655 struct in_ifaddr *ifa = ptr;
3656 struct net_device *vlan_dev, *event_dev = ifa->ifa_dev->dev;
3657 struct bonding *bond, *bond_next;
3658 struct vlan_entry *vlan, *vlan_next;
3659
3660 list_for_each_entry_safe(bond, bond_next, &bond_dev_list, bond_list) {
3661 if (bond->dev == event_dev) {
3662 switch (event) {
3663 case NETDEV_UP:
3664 bond->master_ip = ifa->ifa_local;
3665 return NOTIFY_OK;
3666 case NETDEV_DOWN:
3667 bond->master_ip = bond_glean_dev_ip(bond->dev);
3668 return NOTIFY_OK;
3669 default:
3670 return NOTIFY_DONE;
3671 }
3672 }
3673
3674 if (list_empty(&bond->vlan_list))
3675 continue;
3676
3677 list_for_each_entry_safe(vlan, vlan_next, &bond->vlan_list,
3678 vlan_list) {
3679 vlan_dev = bond->vlgrp->vlan_devices[vlan->vlan_id];
3680 if (vlan_dev == event_dev) {
3681 switch (event) {
3682 case NETDEV_UP:
3683 vlan->vlan_ip = ifa->ifa_local;
3684 return NOTIFY_OK;
3685 case NETDEV_DOWN:
3686 vlan->vlan_ip =
3687 bond_glean_dev_ip(vlan_dev);
3688 return NOTIFY_OK;
3689 default:
3690 return NOTIFY_DONE;
3691 }
3692 }
3693 }
3694 }
3695 return NOTIFY_DONE;
3696}
3697
Linus Torvalds1da177e2005-04-16 15:20:36 -07003698static struct notifier_block bond_netdev_notifier = {
3699 .notifier_call = bond_netdev_event,
3700};
3701
Jay Vosburghc3ade5c2005-06-26 17:52:20 -04003702static struct notifier_block bond_inetaddr_notifier = {
3703 .notifier_call = bond_inetaddr_event,
3704};
3705
Linus Torvalds1da177e2005-04-16 15:20:36 -07003706/*-------------------------- Packet type handling ---------------------------*/
3707
3708/* register to receive lacpdus on a bond */
3709static void bond_register_lacpdu(struct bonding *bond)
3710{
3711 struct packet_type *pk_type = &(BOND_AD_INFO(bond).ad_pkt_type);
3712
3713 /* initialize packet type */
3714 pk_type->type = PKT_TYPE_LACPDU;
3715 pk_type->dev = bond->dev;
3716 pk_type->func = bond_3ad_lacpdu_recv;
3717
3718 dev_add_pack(pk_type);
3719}
3720
3721/* unregister to receive lacpdus on a bond */
3722static void bond_unregister_lacpdu(struct bonding *bond)
3723{
3724 dev_remove_pack(&(BOND_AD_INFO(bond).ad_pkt_type));
3725}
3726
3727/*-------------------------- Device entry points ----------------------------*/
3728
3729static int bond_open(struct net_device *bond_dev)
3730{
3731 struct bonding *bond = bond_dev->priv;
3732 struct timer_list *mii_timer = &bond->mii_timer;
3733 struct timer_list *arp_timer = &bond->arp_timer;
3734
3735 bond->kill_timers = 0;
3736
3737 if ((bond->params.mode == BOND_MODE_TLB) ||
3738 (bond->params.mode == BOND_MODE_ALB)) {
3739 struct timer_list *alb_timer = &(BOND_ALB_INFO(bond).alb_timer);
3740
3741 /* bond_alb_initialize must be called before the timer
3742 * is started.
3743 */
3744 if (bond_alb_initialize(bond, (bond->params.mode == BOND_MODE_ALB))) {
3745 /* something went wrong - fail the open operation */
3746 return -1;
3747 }
3748
3749 init_timer(alb_timer);
3750 alb_timer->expires = jiffies + 1;
3751 alb_timer->data = (unsigned long)bond;
3752 alb_timer->function = (void *)&bond_alb_monitor;
3753 add_timer(alb_timer);
3754 }
3755
3756 if (bond->params.miimon) { /* link check interval, in milliseconds. */
3757 init_timer(mii_timer);
3758 mii_timer->expires = jiffies + 1;
3759 mii_timer->data = (unsigned long)bond_dev;
3760 mii_timer->function = (void *)&bond_mii_monitor;
3761 add_timer(mii_timer);
3762 }
3763
3764 if (bond->params.arp_interval) { /* arp interval, in milliseconds. */
3765 init_timer(arp_timer);
3766 arp_timer->expires = jiffies + 1;
3767 arp_timer->data = (unsigned long)bond_dev;
3768 if (bond->params.mode == BOND_MODE_ACTIVEBACKUP) {
3769 arp_timer->function = (void *)&bond_activebackup_arp_mon;
3770 } else {
3771 arp_timer->function = (void *)&bond_loadbalance_arp_mon;
3772 }
3773 add_timer(arp_timer);
3774 }
3775
3776 if (bond->params.mode == BOND_MODE_8023AD) {
3777 struct timer_list *ad_timer = &(BOND_AD_INFO(bond).ad_timer);
3778 init_timer(ad_timer);
3779 ad_timer->expires = jiffies + 1;
3780 ad_timer->data = (unsigned long)bond;
3781 ad_timer->function = (void *)&bond_3ad_state_machine_handler;
3782 add_timer(ad_timer);
3783
3784 /* register to receive LACPDUs */
3785 bond_register_lacpdu(bond);
3786 }
3787
3788 return 0;
3789}
3790
3791static int bond_close(struct net_device *bond_dev)
3792{
3793 struct bonding *bond = bond_dev->priv;
3794
3795 if (bond->params.mode == BOND_MODE_8023AD) {
3796 /* Unregister the receive of LACPDUs */
3797 bond_unregister_lacpdu(bond);
3798 }
3799
3800 write_lock_bh(&bond->lock);
3801
3802 bond_mc_list_destroy(bond);
3803
3804 /* signal timers not to re-arm */
3805 bond->kill_timers = 1;
3806
3807 write_unlock_bh(&bond->lock);
3808
3809 /* del_timer_sync must run without holding the bond->lock
3810 * because a running timer might be trying to hold it too
3811 */
3812
3813 if (bond->params.miimon) { /* link check interval, in milliseconds. */
3814 del_timer_sync(&bond->mii_timer);
3815 }
3816
3817 if (bond->params.arp_interval) { /* arp interval, in milliseconds. */
3818 del_timer_sync(&bond->arp_timer);
3819 }
3820
3821 switch (bond->params.mode) {
3822 case BOND_MODE_8023AD:
3823 del_timer_sync(&(BOND_AD_INFO(bond).ad_timer));
3824 break;
3825 case BOND_MODE_TLB:
3826 case BOND_MODE_ALB:
3827 del_timer_sync(&(BOND_ALB_INFO(bond).alb_timer));
3828 break;
3829 default:
3830 break;
3831 }
3832
3833 /* Release the bonded slaves */
3834 bond_release_all(bond_dev);
3835
3836 if ((bond->params.mode == BOND_MODE_TLB) ||
3837 (bond->params.mode == BOND_MODE_ALB)) {
3838 /* Must be called only after all
3839 * slaves have been released
3840 */
3841 bond_alb_deinitialize(bond);
3842 }
3843
3844 return 0;
3845}
3846
3847static struct net_device_stats *bond_get_stats(struct net_device *bond_dev)
3848{
3849 struct bonding *bond = bond_dev->priv;
3850 struct net_device_stats *stats = &(bond->stats), *sstats;
3851 struct slave *slave;
3852 int i;
3853
3854 memset(stats, 0, sizeof(struct net_device_stats));
3855
3856 read_lock_bh(&bond->lock);
3857
3858 bond_for_each_slave(bond, slave, i) {
3859 sstats = slave->dev->get_stats(slave->dev);
3860
3861 stats->rx_packets += sstats->rx_packets;
3862 stats->rx_bytes += sstats->rx_bytes;
3863 stats->rx_errors += sstats->rx_errors;
3864 stats->rx_dropped += sstats->rx_dropped;
3865
3866 stats->tx_packets += sstats->tx_packets;
3867 stats->tx_bytes += sstats->tx_bytes;
3868 stats->tx_errors += sstats->tx_errors;
3869 stats->tx_dropped += sstats->tx_dropped;
3870
3871 stats->multicast += sstats->multicast;
3872 stats->collisions += sstats->collisions;
3873
3874 stats->rx_length_errors += sstats->rx_length_errors;
3875 stats->rx_over_errors += sstats->rx_over_errors;
3876 stats->rx_crc_errors += sstats->rx_crc_errors;
3877 stats->rx_frame_errors += sstats->rx_frame_errors;
3878 stats->rx_fifo_errors += sstats->rx_fifo_errors;
3879 stats->rx_missed_errors += sstats->rx_missed_errors;
3880
3881 stats->tx_aborted_errors += sstats->tx_aborted_errors;
3882 stats->tx_carrier_errors += sstats->tx_carrier_errors;
3883 stats->tx_fifo_errors += sstats->tx_fifo_errors;
3884 stats->tx_heartbeat_errors += sstats->tx_heartbeat_errors;
3885 stats->tx_window_errors += sstats->tx_window_errors;
3886 }
3887
3888 read_unlock_bh(&bond->lock);
3889
3890 return stats;
3891}
3892
3893static int bond_do_ioctl(struct net_device *bond_dev, struct ifreq *ifr, int cmd)
3894{
3895 struct net_device *slave_dev = NULL;
3896 struct ifbond k_binfo;
3897 struct ifbond __user *u_binfo = NULL;
3898 struct ifslave k_sinfo;
3899 struct ifslave __user *u_sinfo = NULL;
3900 struct mii_ioctl_data *mii = NULL;
3901 int prev_abi_ver = orig_app_abi_ver;
3902 int res = 0;
3903
3904 dprintk("bond_ioctl: master=%s, cmd=%d\n",
3905 bond_dev->name, cmd);
3906
3907 switch (cmd) {
3908 case SIOCETHTOOL:
3909 return bond_ethtool_ioctl(bond_dev, ifr);
3910 case SIOCGMIIPHY:
3911 mii = if_mii(ifr);
3912 if (!mii) {
3913 return -EINVAL;
3914 }
3915 mii->phy_id = 0;
3916 /* Fall Through */
3917 case SIOCGMIIREG:
3918 /*
3919 * We do this again just in case we were called by SIOCGMIIREG
3920 * instead of SIOCGMIIPHY.
3921 */
3922 mii = if_mii(ifr);
3923 if (!mii) {
3924 return -EINVAL;
3925 }
3926
3927 if (mii->reg_num == 1) {
3928 struct bonding *bond = bond_dev->priv;
3929 mii->val_out = 0;
3930 read_lock_bh(&bond->lock);
3931 read_lock(&bond->curr_slave_lock);
3932 if (bond->curr_active_slave) {
3933 mii->val_out = BMSR_LSTATUS;
3934 }
3935 read_unlock(&bond->curr_slave_lock);
3936 read_unlock_bh(&bond->lock);
3937 }
3938
3939 return 0;
3940 case BOND_INFO_QUERY_OLD:
3941 case SIOCBONDINFOQUERY:
3942 u_binfo = (struct ifbond __user *)ifr->ifr_data;
3943
3944 if (copy_from_user(&k_binfo, u_binfo, sizeof(ifbond))) {
3945 return -EFAULT;
3946 }
3947
3948 res = bond_info_query(bond_dev, &k_binfo);
3949 if (res == 0) {
3950 if (copy_to_user(u_binfo, &k_binfo, sizeof(ifbond))) {
3951 return -EFAULT;
3952 }
3953 }
3954
3955 return res;
3956 case BOND_SLAVE_INFO_QUERY_OLD:
3957 case SIOCBONDSLAVEINFOQUERY:
3958 u_sinfo = (struct ifslave __user *)ifr->ifr_data;
3959
3960 if (copy_from_user(&k_sinfo, u_sinfo, sizeof(ifslave))) {
3961 return -EFAULT;
3962 }
3963
3964 res = bond_slave_info_query(bond_dev, &k_sinfo);
3965 if (res == 0) {
3966 if (copy_to_user(u_sinfo, &k_sinfo, sizeof(ifslave))) {
3967 return -EFAULT;
3968 }
3969 }
3970
3971 return res;
3972 default:
3973 /* Go on */
3974 break;
3975 }
3976
3977 if (!capable(CAP_NET_ADMIN)) {
3978 return -EPERM;
3979 }
3980
3981 if (orig_app_abi_ver == -1) {
3982 /* no orig_app_abi_ver was provided yet, so we'll use the
3983 * current one from now on, even if it's 0
3984 */
3985 orig_app_abi_ver = app_abi_ver;
3986
3987 } else if (orig_app_abi_ver != app_abi_ver) {
3988 printk(KERN_ERR DRV_NAME
3989 ": Error: already using ifenslave ABI version %d; to "
3990 "upgrade ifenslave to version %d, you must first "
3991 "reload bonding.\n",
3992 orig_app_abi_ver, app_abi_ver);
3993 return -EINVAL;
3994 }
3995
3996 slave_dev = dev_get_by_name(ifr->ifr_slave);
3997
3998 dprintk("slave_dev=%p: \n", slave_dev);
3999
4000 if (!slave_dev) {
4001 res = -ENODEV;
4002 } else {
4003 dprintk("slave_dev->name=%s: \n", slave_dev->name);
4004 switch (cmd) {
4005 case BOND_ENSLAVE_OLD:
4006 case SIOCBONDENSLAVE:
4007 res = bond_enslave(bond_dev, slave_dev);
4008 break;
4009 case BOND_RELEASE_OLD:
4010 case SIOCBONDRELEASE:
4011 res = bond_release(bond_dev, slave_dev);
4012 break;
4013 case BOND_SETHWADDR_OLD:
4014 case SIOCBONDSETHWADDR:
4015 res = bond_sethwaddr(bond_dev, slave_dev);
4016 break;
4017 case BOND_CHANGE_ACTIVE_OLD:
4018 case SIOCBONDCHANGEACTIVE:
4019 res = bond_ioctl_change_active(bond_dev, slave_dev);
4020 break;
4021 default:
4022 res = -EOPNOTSUPP;
4023 }
4024
4025 dev_put(slave_dev);
4026 }
4027
4028 if (res < 0) {
4029 /* The ioctl failed, so there's no point in changing the
4030 * orig_app_abi_ver. We'll restore it's value just in case
4031 * we've changed it earlier in this function.
4032 */
4033 orig_app_abi_ver = prev_abi_ver;
4034 }
4035
4036 return res;
4037}
4038
4039static void bond_set_multicast_list(struct net_device *bond_dev)
4040{
4041 struct bonding *bond = bond_dev->priv;
4042 struct dev_mc_list *dmi;
4043
4044 write_lock_bh(&bond->lock);
4045
4046 /*
4047 * Do promisc before checking multicast_mode
4048 */
4049 if ((bond_dev->flags & IFF_PROMISC) && !(bond->flags & IFF_PROMISC)) {
4050 bond_set_promiscuity(bond, 1);
4051 }
4052
4053 if (!(bond_dev->flags & IFF_PROMISC) && (bond->flags & IFF_PROMISC)) {
4054 bond_set_promiscuity(bond, -1);
4055 }
4056
4057 /* set allmulti flag to slaves */
4058 if ((bond_dev->flags & IFF_ALLMULTI) && !(bond->flags & IFF_ALLMULTI)) {
4059 bond_set_allmulti(bond, 1);
4060 }
4061
4062 if (!(bond_dev->flags & IFF_ALLMULTI) && (bond->flags & IFF_ALLMULTI)) {
4063 bond_set_allmulti(bond, -1);
4064 }
4065
4066 bond->flags = bond_dev->flags;
4067
4068 /* looking for addresses to add to slaves' mc list */
4069 for (dmi = bond_dev->mc_list; dmi; dmi = dmi->next) {
4070 if (!bond_mc_list_find_dmi(dmi, bond->mc_list)) {
4071 bond_mc_add(bond, dmi->dmi_addr, dmi->dmi_addrlen);
4072 }
4073 }
4074
4075 /* looking for addresses to delete from slaves' list */
4076 for (dmi = bond->mc_list; dmi; dmi = dmi->next) {
4077 if (!bond_mc_list_find_dmi(dmi, bond_dev->mc_list)) {
4078 bond_mc_delete(bond, dmi->dmi_addr, dmi->dmi_addrlen);
4079 }
4080 }
4081
4082 /* save master's multicast list */
4083 bond_mc_list_destroy(bond);
4084 bond_mc_list_copy(bond_dev->mc_list, bond, GFP_ATOMIC);
4085
4086 write_unlock_bh(&bond->lock);
4087}
4088
4089/*
4090 * Change the MTU of all of a master's slaves to match the master
4091 */
4092static int bond_change_mtu(struct net_device *bond_dev, int new_mtu)
4093{
4094 struct bonding *bond = bond_dev->priv;
4095 struct slave *slave, *stop_at;
4096 int res = 0;
4097 int i;
4098
4099 dprintk("bond=%p, name=%s, new_mtu=%d\n", bond,
4100 (bond_dev ? bond_dev->name : "None"), new_mtu);
4101
4102 /* Can't hold bond->lock with bh disabled here since
4103 * some base drivers panic. On the other hand we can't
4104 * hold bond->lock without bh disabled because we'll
4105 * deadlock. The only solution is to rely on the fact
4106 * that we're under rtnl_lock here, and the slaves
4107 * list won't change. This doesn't solve the problem
4108 * of setting the slave's MTU while it is
4109 * transmitting, but the assumption is that the base
4110 * driver can handle that.
4111 *
4112 * TODO: figure out a way to safely iterate the slaves
4113 * list, but without holding a lock around the actual
4114 * call to the base driver.
4115 */
4116
4117 bond_for_each_slave(bond, slave, i) {
4118 dprintk("s %p s->p %p c_m %p\n", slave,
4119 slave->prev, slave->dev->change_mtu);
4120 res = dev_set_mtu(slave->dev, new_mtu);
4121
4122 if (res) {
4123 /* If we failed to set the slave's mtu to the new value
4124 * we must abort the operation even in ACTIVE_BACKUP
4125 * mode, because if we allow the backup slaves to have
4126 * different mtu values than the active slave we'll
4127 * need to change their mtu when doing a failover. That
4128 * means changing their mtu from timer context, which
4129 * is probably not a good idea.
4130 */
4131 dprintk("err %d %s\n", res, slave->dev->name);
4132 goto unwind;
4133 }
4134 }
4135
4136 bond_dev->mtu = new_mtu;
4137
4138 return 0;
4139
4140unwind:
4141 /* unwind from head to the slave that failed */
4142 stop_at = slave;
4143 bond_for_each_slave_from_to(bond, slave, i, bond->first_slave, stop_at) {
4144 int tmp_res;
4145
4146 tmp_res = dev_set_mtu(slave->dev, bond_dev->mtu);
4147 if (tmp_res) {
4148 dprintk("unwind err %d dev %s\n", tmp_res,
4149 slave->dev->name);
4150 }
4151 }
4152
4153 return res;
4154}
4155
4156/*
4157 * Change HW address
4158 *
4159 * Note that many devices must be down to change the HW address, and
4160 * downing the master releases all slaves. We can make bonds full of
4161 * bonding devices to test this, however.
4162 */
4163static int bond_set_mac_address(struct net_device *bond_dev, void *addr)
4164{
4165 struct bonding *bond = bond_dev->priv;
4166 struct sockaddr *sa = addr, tmp_sa;
4167 struct slave *slave, *stop_at;
4168 int res = 0;
4169 int i;
4170
4171 dprintk("bond=%p, name=%s\n", bond, (bond_dev ? bond_dev->name : "None"));
4172
4173 if (!is_valid_ether_addr(sa->sa_data)) {
4174 return -EADDRNOTAVAIL;
4175 }
4176
4177 /* Can't hold bond->lock with bh disabled here since
4178 * some base drivers panic. On the other hand we can't
4179 * hold bond->lock without bh disabled because we'll
4180 * deadlock. The only solution is to rely on the fact
4181 * that we're under rtnl_lock here, and the slaves
4182 * list won't change. This doesn't solve the problem
4183 * of setting the slave's hw address while it is
4184 * transmitting, but the assumption is that the base
4185 * driver can handle that.
4186 *
4187 * TODO: figure out a way to safely iterate the slaves
4188 * list, but without holding a lock around the actual
4189 * call to the base driver.
4190 */
4191
4192 bond_for_each_slave(bond, slave, i) {
4193 dprintk("slave %p %s\n", slave, slave->dev->name);
4194
4195 if (slave->dev->set_mac_address == NULL) {
4196 res = -EOPNOTSUPP;
4197 dprintk("EOPNOTSUPP %s\n", slave->dev->name);
4198 goto unwind;
4199 }
4200
4201 res = dev_set_mac_address(slave->dev, addr);
4202 if (res) {
4203 /* TODO: consider downing the slave
4204 * and retry ?
4205 * User should expect communications
4206 * breakage anyway until ARP finish
4207 * updating, so...
4208 */
4209 dprintk("err %d %s\n", res, slave->dev->name);
4210 goto unwind;
4211 }
4212 }
4213
4214 /* success */
4215 memcpy(bond_dev->dev_addr, sa->sa_data, bond_dev->addr_len);
4216 return 0;
4217
4218unwind:
4219 memcpy(tmp_sa.sa_data, bond_dev->dev_addr, bond_dev->addr_len);
4220 tmp_sa.sa_family = bond_dev->type;
4221
4222 /* unwind from head to the slave that failed */
4223 stop_at = slave;
4224 bond_for_each_slave_from_to(bond, slave, i, bond->first_slave, stop_at) {
4225 int tmp_res;
4226
4227 tmp_res = dev_set_mac_address(slave->dev, &tmp_sa);
4228 if (tmp_res) {
4229 dprintk("unwind err %d dev %s\n", tmp_res,
4230 slave->dev->name);
4231 }
4232 }
4233
4234 return res;
4235}
4236
4237static int bond_xmit_roundrobin(struct sk_buff *skb, struct net_device *bond_dev)
4238{
4239 struct bonding *bond = bond_dev->priv;
4240 struct slave *slave, *start_at;
4241 int i;
4242 int res = 1;
4243
4244 read_lock(&bond->lock);
4245
4246 if (!BOND_IS_OK(bond)) {
4247 goto out;
4248 }
4249
4250 read_lock(&bond->curr_slave_lock);
4251 slave = start_at = bond->curr_active_slave;
4252 read_unlock(&bond->curr_slave_lock);
4253
4254 if (!slave) {
4255 goto out;
4256 }
4257
4258 bond_for_each_slave_from(bond, slave, i, start_at) {
4259 if (IS_UP(slave->dev) &&
4260 (slave->link == BOND_LINK_UP) &&
4261 (slave->state == BOND_STATE_ACTIVE)) {
4262 res = bond_dev_queue_xmit(bond, skb, slave->dev);
4263
4264 write_lock(&bond->curr_slave_lock);
4265 bond->curr_active_slave = slave->next;
4266 write_unlock(&bond->curr_slave_lock);
4267
4268 break;
4269 }
4270 }
4271
4272
4273out:
4274 if (res) {
4275 /* no suitable interface, frame not sent */
4276 dev_kfree_skb(skb);
4277 }
4278 read_unlock(&bond->lock);
4279 return 0;
4280}
4281
4282/*
4283 * in active-backup mode, we know that bond->curr_active_slave is always valid if
4284 * the bond has a usable interface.
4285 */
4286static int bond_xmit_activebackup(struct sk_buff *skb, struct net_device *bond_dev)
4287{
4288 struct bonding *bond = bond_dev->priv;
4289 int res = 1;
4290
Linus Torvalds1da177e2005-04-16 15:20:36 -07004291 read_lock(&bond->lock);
4292 read_lock(&bond->curr_slave_lock);
4293
4294 if (!BOND_IS_OK(bond)) {
4295 goto out;
4296 }
4297
4298 if (bond->curr_active_slave) { /* one usable interface */
4299 res = bond_dev_queue_xmit(bond, skb, bond->curr_active_slave->dev);
4300 }
4301
4302out:
4303 if (res) {
4304 /* no suitable interface, frame not sent */
4305 dev_kfree_skb(skb);
4306 }
4307 read_unlock(&bond->curr_slave_lock);
4308 read_unlock(&bond->lock);
4309 return 0;
4310}
4311
4312/*
4313 * in XOR mode, we determine the output device by performing xor on
4314 * the source and destination hw adresses. If this device is not
4315 * enabled, find the next slave following this xor slave.
4316 */
4317static int bond_xmit_xor(struct sk_buff *skb, struct net_device *bond_dev)
4318{
4319 struct bonding *bond = bond_dev->priv;
4320 struct ethhdr *data = (struct ethhdr *)skb->data;
4321 struct slave *slave, *start_at;
4322 int slave_no;
4323 int i;
4324 int res = 1;
4325
4326 read_lock(&bond->lock);
4327
4328 if (!BOND_IS_OK(bond)) {
4329 goto out;
4330 }
4331
4332 slave_no = (data->h_dest[5]^bond_dev->dev_addr[5]) % bond->slave_cnt;
4333
4334 bond_for_each_slave(bond, slave, i) {
4335 slave_no--;
4336 if (slave_no < 0) {
4337 break;
4338 }
4339 }
4340
4341 start_at = slave;
4342
4343 bond_for_each_slave_from(bond, slave, i, start_at) {
4344 if (IS_UP(slave->dev) &&
4345 (slave->link == BOND_LINK_UP) &&
4346 (slave->state == BOND_STATE_ACTIVE)) {
4347 res = bond_dev_queue_xmit(bond, skb, slave->dev);
4348 break;
4349 }
4350 }
4351
4352out:
4353 if (res) {
4354 /* no suitable interface, frame not sent */
4355 dev_kfree_skb(skb);
4356 }
4357 read_unlock(&bond->lock);
4358 return 0;
4359}
4360
4361/*
4362 * in broadcast mode, we send everything to all usable interfaces.
4363 */
4364static int bond_xmit_broadcast(struct sk_buff *skb, struct net_device *bond_dev)
4365{
4366 struct bonding *bond = bond_dev->priv;
4367 struct slave *slave, *start_at;
4368 struct net_device *tx_dev = NULL;
4369 int i;
4370 int res = 1;
4371
4372 read_lock(&bond->lock);
4373
4374 if (!BOND_IS_OK(bond)) {
4375 goto out;
4376 }
4377
4378 read_lock(&bond->curr_slave_lock);
4379 start_at = bond->curr_active_slave;
4380 read_unlock(&bond->curr_slave_lock);
4381
4382 if (!start_at) {
4383 goto out;
4384 }
4385
4386 bond_for_each_slave_from(bond, slave, i, start_at) {
4387 if (IS_UP(slave->dev) &&
4388 (slave->link == BOND_LINK_UP) &&
4389 (slave->state == BOND_STATE_ACTIVE)) {
4390 if (tx_dev) {
4391 struct sk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
4392 if (!skb2) {
4393 printk(KERN_ERR DRV_NAME
4394 ": Error: bond_xmit_broadcast(): "
4395 "skb_clone() failed\n");
4396 continue;
4397 }
4398
4399 res = bond_dev_queue_xmit(bond, skb2, tx_dev);
4400 if (res) {
4401 dev_kfree_skb(skb2);
4402 continue;
4403 }
4404 }
4405 tx_dev = slave->dev;
4406 }
4407 }
4408
4409 if (tx_dev) {
4410 res = bond_dev_queue_xmit(bond, skb, tx_dev);
4411 }
4412
4413out:
4414 if (res) {
4415 /* no suitable interface, frame not sent */
4416 dev_kfree_skb(skb);
4417 }
4418 /* frame sent to all suitable interfaces */
4419 read_unlock(&bond->lock);
4420 return 0;
4421}
4422
4423/*------------------------- Device initialization ---------------------------*/
4424
4425/*
4426 * set bond mode specific net device operations
4427 */
4428static inline void bond_set_mode_ops(struct net_device *bond_dev, int mode)
4429{
4430 switch (mode) {
4431 case BOND_MODE_ROUNDROBIN:
4432 bond_dev->hard_start_xmit = bond_xmit_roundrobin;
4433 break;
4434 case BOND_MODE_ACTIVEBACKUP:
4435 bond_dev->hard_start_xmit = bond_xmit_activebackup;
4436 break;
4437 case BOND_MODE_XOR:
4438 bond_dev->hard_start_xmit = bond_xmit_xor;
4439 break;
4440 case BOND_MODE_BROADCAST:
4441 bond_dev->hard_start_xmit = bond_xmit_broadcast;
4442 break;
4443 case BOND_MODE_8023AD:
4444 bond_dev->hard_start_xmit = bond_3ad_xmit_xor;
4445 break;
4446 case BOND_MODE_TLB:
4447 case BOND_MODE_ALB:
4448 bond_dev->hard_start_xmit = bond_alb_xmit;
4449 bond_dev->set_mac_address = bond_alb_set_mac_address;
4450 break;
4451 default:
4452 /* Should never happen, mode already checked */
4453 printk(KERN_ERR DRV_NAME
4454 ": Error: Unknown bonding mode %d\n",
4455 mode);
4456 break;
4457 }
4458}
4459
4460/*
4461 * Does not allocate but creates a /proc entry.
4462 * Allowed to fail.
4463 */
4464static int __init bond_init(struct net_device *bond_dev, struct bond_params *params)
4465{
4466 struct bonding *bond = bond_dev->priv;
4467
4468 dprintk("Begin bond_init for %s\n", bond_dev->name);
4469
4470 /* initialize rwlocks */
4471 rwlock_init(&bond->lock);
4472 rwlock_init(&bond->curr_slave_lock);
4473
4474 bond->params = *params; /* copy params struct */
4475
4476 /* Initialize pointers */
4477 bond->first_slave = NULL;
4478 bond->curr_active_slave = NULL;
4479 bond->current_arp_slave = NULL;
4480 bond->primary_slave = NULL;
4481 bond->dev = bond_dev;
4482 INIT_LIST_HEAD(&bond->vlan_list);
4483
4484 /* Initialize the device entry points */
4485 bond_dev->open = bond_open;
4486 bond_dev->stop = bond_close;
4487 bond_dev->get_stats = bond_get_stats;
4488 bond_dev->do_ioctl = bond_do_ioctl;
4489 bond_dev->set_multicast_list = bond_set_multicast_list;
4490 bond_dev->change_mtu = bond_change_mtu;
4491 bond_dev->set_mac_address = bond_set_mac_address;
4492
4493 bond_set_mode_ops(bond_dev, bond->params.mode);
4494
4495 bond_dev->destructor = free_netdev;
4496
4497 /* Initialize the device options */
4498 bond_dev->tx_queue_len = 0;
4499 bond_dev->flags |= IFF_MASTER|IFF_MULTICAST;
4500
4501 /* At first, we block adding VLANs. That's the only way to
4502 * prevent problems that occur when adding VLANs over an
4503 * empty bond. The block will be removed once non-challenged
4504 * slaves are enslaved.
4505 */
4506 bond_dev->features |= NETIF_F_VLAN_CHALLENGED;
4507
4508 /* don't acquire bond device's xmit_lock when
4509 * transmitting */
4510 bond_dev->features |= NETIF_F_LLTX;
4511
4512 /* By default, we declare the bond to be fully
4513 * VLAN hardware accelerated capable. Special
4514 * care is taken in the various xmit functions
4515 * when there are slaves that are not hw accel
4516 * capable
4517 */
4518 bond_dev->vlan_rx_register = bond_vlan_rx_register;
4519 bond_dev->vlan_rx_add_vid = bond_vlan_rx_add_vid;
4520 bond_dev->vlan_rx_kill_vid = bond_vlan_rx_kill_vid;
4521 bond_dev->features |= (NETIF_F_HW_VLAN_TX |
4522 NETIF_F_HW_VLAN_RX |
4523 NETIF_F_HW_VLAN_FILTER);
4524
4525#ifdef CONFIG_PROC_FS
4526 bond_create_proc_entry(bond);
4527#endif
4528
4529 list_add_tail(&bond->bond_list, &bond_dev_list);
4530
4531 return 0;
4532}
4533
4534/* De-initialize device specific data.
4535 * Caller must hold rtnl_lock.
4536 */
4537static inline void bond_deinit(struct net_device *bond_dev)
4538{
4539 struct bonding *bond = bond_dev->priv;
4540
4541 list_del(&bond->bond_list);
4542
4543#ifdef CONFIG_PROC_FS
4544 bond_remove_proc_entry(bond);
4545#endif
4546}
4547
4548/* Unregister and free all bond devices.
4549 * Caller must hold rtnl_lock.
4550 */
4551static void bond_free_all(void)
4552{
4553 struct bonding *bond, *nxt;
4554
4555 list_for_each_entry_safe(bond, nxt, &bond_dev_list, bond_list) {
4556 struct net_device *bond_dev = bond->dev;
4557
4558 unregister_netdevice(bond_dev);
4559 bond_deinit(bond_dev);
4560 }
4561
4562#ifdef CONFIG_PROC_FS
4563 bond_destroy_proc_dir();
4564#endif
4565}
4566
4567/*------------------------- Module initialization ---------------------------*/
4568
4569/*
4570 * Convert string input module parms. Accept either the
4571 * number of the mode or its string name.
4572 */
4573static inline int bond_parse_parm(char *mode_arg, struct bond_parm_tbl *tbl)
4574{
4575 int i;
4576
4577 for (i = 0; tbl[i].modename; i++) {
4578 if ((isdigit(*mode_arg) &&
4579 tbl[i].mode == simple_strtol(mode_arg, NULL, 0)) ||
4580 (strncmp(mode_arg, tbl[i].modename,
4581 strlen(tbl[i].modename)) == 0)) {
4582 return tbl[i].mode;
4583 }
4584 }
4585
4586 return -1;
4587}
4588
4589static int bond_check_params(struct bond_params *params)
4590{
4591 /*
4592 * Convert string parameters.
4593 */
4594 if (mode) {
4595 bond_mode = bond_parse_parm(mode, bond_mode_tbl);
4596 if (bond_mode == -1) {
4597 printk(KERN_ERR DRV_NAME
4598 ": Error: Invalid bonding mode \"%s\"\n",
4599 mode == NULL ? "NULL" : mode);
4600 return -EINVAL;
4601 }
4602 }
4603
4604 if (lacp_rate) {
4605 if (bond_mode != BOND_MODE_8023AD) {
4606 printk(KERN_INFO DRV_NAME
4607 ": lacp_rate param is irrelevant in mode %s\n",
4608 bond_mode_name(bond_mode));
4609 } else {
4610 lacp_fast = bond_parse_parm(lacp_rate, bond_lacp_tbl);
4611 if (lacp_fast == -1) {
4612 printk(KERN_ERR DRV_NAME
4613 ": Error: Invalid lacp rate \"%s\"\n",
4614 lacp_rate == NULL ? "NULL" : lacp_rate);
4615 return -EINVAL;
4616 }
4617 }
4618 }
4619
4620 if (max_bonds < 1 || max_bonds > INT_MAX) {
4621 printk(KERN_WARNING DRV_NAME
4622 ": Warning: max_bonds (%d) not in range %d-%d, so it "
4623 "was reset to BOND_DEFAULT_MAX_BONDS (%d)",
4624 max_bonds, 1, INT_MAX, BOND_DEFAULT_MAX_BONDS);
4625 max_bonds = BOND_DEFAULT_MAX_BONDS;
4626 }
4627
4628 if (miimon < 0) {
4629 printk(KERN_WARNING DRV_NAME
4630 ": Warning: miimon module parameter (%d), "
4631 "not in range 0-%d, so it was reset to %d\n",
4632 miimon, INT_MAX, BOND_LINK_MON_INTERV);
4633 miimon = BOND_LINK_MON_INTERV;
4634 }
4635
4636 if (updelay < 0) {
4637 printk(KERN_WARNING DRV_NAME
4638 ": Warning: updelay module parameter (%d), "
4639 "not in range 0-%d, so it was reset to 0\n",
4640 updelay, INT_MAX);
4641 updelay = 0;
4642 }
4643
4644 if (downdelay < 0) {
4645 printk(KERN_WARNING DRV_NAME
4646 ": Warning: downdelay module parameter (%d), "
4647 "not in range 0-%d, so it was reset to 0\n",
4648 downdelay, INT_MAX);
4649 downdelay = 0;
4650 }
4651
4652 if ((use_carrier != 0) && (use_carrier != 1)) {
4653 printk(KERN_WARNING DRV_NAME
4654 ": Warning: use_carrier module parameter (%d), "
4655 "not of valid value (0/1), so it was set to 1\n",
4656 use_carrier);
4657 use_carrier = 1;
4658 }
4659
4660 /* reset values for 802.3ad */
4661 if (bond_mode == BOND_MODE_8023AD) {
4662 if (!miimon) {
4663 printk(KERN_WARNING DRV_NAME
4664 ": Warning: miimon must be specified, "
4665 "otherwise bonding will not detect link "
4666 "failure, speed and duplex which are "
4667 "essential for 802.3ad operation\n");
4668 printk(KERN_WARNING "Forcing miimon to 100msec\n");
4669 miimon = 100;
4670 }
4671 }
4672
4673 /* reset values for TLB/ALB */
4674 if ((bond_mode == BOND_MODE_TLB) ||
4675 (bond_mode == BOND_MODE_ALB)) {
4676 if (!miimon) {
4677 printk(KERN_WARNING DRV_NAME
4678 ": Warning: miimon must be specified, "
4679 "otherwise bonding will not detect link "
4680 "failure and link speed which are essential "
4681 "for TLB/ALB load balancing\n");
4682 printk(KERN_WARNING "Forcing miimon to 100msec\n");
4683 miimon = 100;
4684 }
4685 }
4686
4687 if (bond_mode == BOND_MODE_ALB) {
4688 printk(KERN_NOTICE DRV_NAME
4689 ": In ALB mode you might experience client "
4690 "disconnections upon reconnection of a link if the "
4691 "bonding module updelay parameter (%d msec) is "
4692 "incompatible with the forwarding delay time of the "
4693 "switch\n",
4694 updelay);
4695 }
4696
4697 if (!miimon) {
4698 if (updelay || downdelay) {
4699 /* just warn the user the up/down delay will have
4700 * no effect since miimon is zero...
4701 */
4702 printk(KERN_WARNING DRV_NAME
4703 ": Warning: miimon module parameter not set "
4704 "and updelay (%d) or downdelay (%d) module "
4705 "parameter is set; updelay and downdelay have "
4706 "no effect unless miimon is set\n",
4707 updelay, downdelay);
4708 }
4709 } else {
4710 /* don't allow arp monitoring */
4711 if (arp_interval) {
4712 printk(KERN_WARNING DRV_NAME
4713 ": Warning: miimon (%d) and arp_interval (%d) "
4714 "can't be used simultaneously, disabling ARP "
4715 "monitoring\n",
4716 miimon, arp_interval);
4717 arp_interval = 0;
4718 }
4719
4720 if ((updelay % miimon) != 0) {
4721 printk(KERN_WARNING DRV_NAME
4722 ": Warning: updelay (%d) is not a multiple "
4723 "of miimon (%d), updelay rounded to %d ms\n",
4724 updelay, miimon, (updelay / miimon) * miimon);
4725 }
4726
4727 updelay /= miimon;
4728
4729 if ((downdelay % miimon) != 0) {
4730 printk(KERN_WARNING DRV_NAME
4731 ": Warning: downdelay (%d) is not a multiple "
4732 "of miimon (%d), downdelay rounded to %d ms\n",
4733 downdelay, miimon,
4734 (downdelay / miimon) * miimon);
4735 }
4736
4737 downdelay /= miimon;
4738 }
4739
4740 if (arp_interval < 0) {
4741 printk(KERN_WARNING DRV_NAME
4742 ": Warning: arp_interval module parameter (%d) "
4743 ", not in range 0-%d, so it was reset to %d\n",
4744 arp_interval, INT_MAX, BOND_LINK_ARP_INTERV);
4745 arp_interval = BOND_LINK_ARP_INTERV;
4746 }
4747
4748 for (arp_ip_count = 0;
4749 (arp_ip_count < BOND_MAX_ARP_TARGETS) && arp_ip_target[arp_ip_count];
4750 arp_ip_count++) {
4751 /* not complete check, but should be good enough to
4752 catch mistakes */
4753 if (!isdigit(arp_ip_target[arp_ip_count][0])) {
4754 printk(KERN_WARNING DRV_NAME
4755 ": Warning: bad arp_ip_target module parameter "
4756 "(%s), ARP monitoring will not be performed\n",
4757 arp_ip_target[arp_ip_count]);
4758 arp_interval = 0;
4759 } else {
4760 u32 ip = in_aton(arp_ip_target[arp_ip_count]);
4761 arp_target[arp_ip_count] = ip;
4762 }
4763 }
4764
4765 if (arp_interval && !arp_ip_count) {
4766 /* don't allow arping if no arp_ip_target given... */
4767 printk(KERN_WARNING DRV_NAME
4768 ": Warning: arp_interval module parameter (%d) "
4769 "specified without providing an arp_ip_target "
4770 "parameter, arp_interval was reset to 0\n",
4771 arp_interval);
4772 arp_interval = 0;
4773 }
4774
4775 if (miimon) {
4776 printk(KERN_INFO DRV_NAME
4777 ": MII link monitoring set to %d ms\n",
4778 miimon);
4779 } else if (arp_interval) {
4780 int i;
4781
4782 printk(KERN_INFO DRV_NAME
4783 ": ARP monitoring set to %d ms with %d target(s):",
4784 arp_interval, arp_ip_count);
4785
4786 for (i = 0; i < arp_ip_count; i++)
4787 printk (" %s", arp_ip_target[i]);
4788
4789 printk("\n");
4790
4791 } else {
4792 /* miimon and arp_interval not set, we need one so things
4793 * work as expected, see bonding.txt for details
4794 */
4795 printk(KERN_WARNING DRV_NAME
4796 ": Warning: either miimon or arp_interval and "
4797 "arp_ip_target module parameters must be specified, "
4798 "otherwise bonding will not detect link failures! see "
4799 "bonding.txt for details.\n");
4800 }
4801
4802 if (primary && !USES_PRIMARY(bond_mode)) {
4803 /* currently, using a primary only makes sense
4804 * in active backup, TLB or ALB modes
4805 */
4806 printk(KERN_WARNING DRV_NAME
4807 ": Warning: %s primary device specified but has no "
4808 "effect in %s mode\n",
4809 primary, bond_mode_name(bond_mode));
4810 primary = NULL;
4811 }
4812
4813 /* fill params struct with the proper values */
4814 params->mode = bond_mode;
4815 params->miimon = miimon;
4816 params->arp_interval = arp_interval;
4817 params->updelay = updelay;
4818 params->downdelay = downdelay;
4819 params->use_carrier = use_carrier;
4820 params->lacp_fast = lacp_fast;
4821 params->primary[0] = 0;
4822
4823 if (primary) {
4824 strncpy(params->primary, primary, IFNAMSIZ);
4825 params->primary[IFNAMSIZ - 1] = 0;
4826 }
4827
4828 memcpy(params->arp_targets, arp_target, sizeof(arp_target));
4829
4830 return 0;
4831}
4832
4833static int __init bonding_init(void)
4834{
4835 struct bond_params params;
4836 int i;
4837 int res;
4838
4839 printk(KERN_INFO "%s", version);
4840
4841 res = bond_check_params(&params);
4842 if (res) {
4843 return res;
4844 }
4845
4846 rtnl_lock();
4847
4848#ifdef CONFIG_PROC_FS
4849 bond_create_proc_dir();
4850#endif
4851
4852 for (i = 0; i < max_bonds; i++) {
4853 struct net_device *bond_dev;
4854
4855 bond_dev = alloc_netdev(sizeof(struct bonding), "", ether_setup);
4856 if (!bond_dev) {
4857 res = -ENOMEM;
4858 goto out_err;
4859 }
4860
4861 res = dev_alloc_name(bond_dev, "bond%d");
4862 if (res < 0) {
4863 free_netdev(bond_dev);
4864 goto out_err;
4865 }
4866
4867 /* bond_init() must be called after dev_alloc_name() (for the
4868 * /proc files), but before register_netdevice(), because we
4869 * need to set function pointers.
4870 */
4871 res = bond_init(bond_dev, &params);
4872 if (res < 0) {
4873 free_netdev(bond_dev);
4874 goto out_err;
4875 }
4876
4877 SET_MODULE_OWNER(bond_dev);
4878
4879 res = register_netdevice(bond_dev);
4880 if (res < 0) {
4881 bond_deinit(bond_dev);
4882 free_netdev(bond_dev);
4883 goto out_err;
4884 }
4885 }
4886
4887 rtnl_unlock();
4888 register_netdevice_notifier(&bond_netdev_notifier);
Jay Vosburghc3ade5c2005-06-26 17:52:20 -04004889 register_inetaddr_notifier(&bond_inetaddr_notifier);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004890
4891 return 0;
4892
4893out_err:
4894 /* free and unregister all bonds that were successfully added */
4895 bond_free_all();
4896
4897 rtnl_unlock();
4898
4899 return res;
4900}
4901
4902static void __exit bonding_exit(void)
4903{
4904 unregister_netdevice_notifier(&bond_netdev_notifier);
Jay Vosburghc3ade5c2005-06-26 17:52:20 -04004905 unregister_inetaddr_notifier(&bond_inetaddr_notifier);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004906
4907 rtnl_lock();
4908 bond_free_all();
4909 rtnl_unlock();
4910}
4911
4912module_init(bonding_init);
4913module_exit(bonding_exit);
4914MODULE_LICENSE("GPL");
4915MODULE_VERSION(DRV_VERSION);
4916MODULE_DESCRIPTION(DRV_DESCRIPTION ", v" DRV_VERSION);
4917MODULE_AUTHOR("Thomas Davis, tadavis@lbl.gov and many others");
4918MODULE_SUPPORTED_DEVICE("most ethernet devices");
4919
4920/*
4921 * Local variables:
4922 * c-indent-level: 8
4923 * c-basic-offset: 8
4924 * tab-width: 8
4925 * End:
4926 */
4927