brintos / linux-shallow public Read only

1.. SPDX-License-Identifier: GPL-2.02 3===================================4Linux Ethernet Bonding Driver HOWTO5===================================6 7Latest update: 27 April 20118 9Initial release: Thomas Davis <tadavis at lbl.gov>10 11Corrections, HA extensions: 2000/10/03-15:12 13  - Willy Tarreau <willy at meta-x.org>14  - Constantine Gavrilov <const-g at xpert.com>15  - Chad N. Tindel <ctindel at ieee dot org>16  - Janice Girouard <girouard at us dot ibm dot com>17  - Jay Vosburgh <fubar at us dot ibm dot com>18 19Reorganized and updated Feb 2005 by Jay Vosburgh20Added Sysfs information: 2006/04/2421 22  - Mitch Williams <mitch.a.williams at intel.com>23 24Introduction25============26 27The Linux bonding driver provides a method for aggregating28multiple network interfaces into a single logical "bonded" interface.29The behavior of the bonded interfaces depends upon the mode; generally30speaking, modes provide either hot standby or load balancing services.31Additionally, link integrity monitoring may be performed.32 33The bonding driver originally came from Donald Becker's34beowulf patches for kernel 2.0. It has changed quite a bit since, and35the original tools from extreme-linux and beowulf sites will not work36with this version of the driver.37 38For new versions of the driver, updated userspace tools, and39who to ask for help, please follow the links at the end of this file.40 41.. Table of Contents42 43   1. Bonding Driver Installation44 45   2. Bonding Driver Options46 47   3. Configuring Bonding Devices48   3.1	Configuration with Sysconfig Support49   3.1.1		Using DHCP with Sysconfig50   3.1.2		Configuring Multiple Bonds with Sysconfig51   3.2	Configuration with Initscripts Support52   3.2.1		Using DHCP with Initscripts53   3.2.2		Configuring Multiple Bonds with Initscripts54   3.3	Configuring Bonding Manually with Ifenslave55   3.3.1		Configuring Multiple Bonds Manually56   3.4	Configuring Bonding Manually via Sysfs57   3.5	Configuration with Interfaces Support58   3.6	Overriding Configuration for Special Cases59   3.7 Configuring LACP for 802.3ad mode in a more secure way60 61   4. Querying Bonding Configuration62   4.1	Bonding Configuration63   4.2	Network Configuration64 65   5. Switch Configuration66 67   6. 802.1q VLAN Support68 69   7. Link Monitoring70   7.1	ARP Monitor Operation71   7.2	Configuring Multiple ARP Targets72   7.3	MII Monitor Operation73 74   8. Potential Trouble Sources75   8.1	Adventures in Routing76   8.2	Ethernet Device Renaming77   8.3	Painfully Slow Or No Failed Link Detection By Miimon78 79   9. SNMP agents80 81   10. Promiscuous mode82 83   11. Configuring Bonding for High Availability84   11.1	High Availability in a Single Switch Topology85   11.2	High Availability in a Multiple Switch Topology86   11.2.1		HA Bonding Mode Selection for Multiple Switch Topology87   11.2.2		HA Link Monitoring for Multiple Switch Topology88 89   12. Configuring Bonding for Maximum Throughput90   12.1	Maximum Throughput in a Single Switch Topology91   12.1.1		MT Bonding Mode Selection for Single Switch Topology92   12.1.2		MT Link Monitoring for Single Switch Topology93   12.2	Maximum Throughput in a Multiple Switch Topology94   12.2.1		MT Bonding Mode Selection for Multiple Switch Topology95   12.2.2		MT Link Monitoring for Multiple Switch Topology96 97   13. Switch Behavior Issues98   13.1	Link Establishment and Failover Delays99   13.2	Duplicated Incoming Packets100 101   14. Hardware Specific Considerations102   14.1	IBM BladeCenter103 104   15. Frequently Asked Questions105 106   16. Resources and Links107 108 1091. Bonding Driver Installation110==============================111 112Most popular distro kernels ship with the bonding driver113already available as a module. If your distro does not, or you114have need to compile bonding from source (e.g., configuring and115installing a mainline kernel from kernel.org), you'll need to perform116the following steps:117 1181.1 Configure and build the kernel with bonding119-----------------------------------------------120 121The current version of the bonding driver is available in the122drivers/net/bonding subdirectory of the most recent kernel source123(which is available on http://kernel.org).  Most users "rolling their124own" will want to use the most recent kernel from kernel.org.125 126Configure kernel with "make menuconfig" (or "make xconfig" or127"make config"), then select "Bonding driver support" in the "Network128device support" section.  It is recommended that you configure the129driver as module since it is currently the only way to pass parameters130to the driver or configure more than one bonding device.131 132Build and install the new kernel and modules.133 1341.2 Bonding Control Utility135---------------------------136 137It is recommended to configure bonding via iproute2 (netlink)138or sysfs, the old ifenslave control utility is obsolete.139 1402. Bonding Driver Options141=========================142 143Options for the bonding driver are supplied as parameters to the144bonding module at load time, or are specified via sysfs.145 146Module options may be given as command line arguments to the147insmod or modprobe command, but are usually specified in either the148``/etc/modprobe.d/*.conf`` configuration files, or in a distro-specific149configuration file (some of which are detailed in the next section).150 151Details on bonding support for sysfs is provided in the152"Configuring Bonding Manually via Sysfs" section, below.153 154The available bonding driver parameters are listed below. If a155parameter is not specified the default value is used.  When initially156configuring a bond, it is recommended "tail -f /var/log/messages" be157run in a separate window to watch for bonding driver error messages.158 159It is critical that either the miimon or arp_interval and160arp_ip_target parameters be specified, otherwise serious network161degradation will occur during link failures.  Very few devices do not162support at least miimon, so there is really no reason not to use it.163 164Options with textual values will accept either the text name165or, for backwards compatibility, the option value.  E.g.,166"mode=802.3ad" and "mode=4" set the same mode.167 168The parameters are as follows:169 170active_slave171 172	Specifies the new active slave for modes that support it173	(active-backup, balance-alb and balance-tlb).  Possible values174	are the name of any currently enslaved interface, or an empty175	string.  If a name is given, the slave and its link must be up in order176	to be selected as the new active slave.  If an empty string is177	specified, the current active slave is cleared, and a new active178	slave is selected automatically.179 180	Note that this is only available through the sysfs interface. No module181	parameter by this name exists.182 183	The normal value of this option is the name of the currently184	active slave, or the empty string if there is no active slave or185	the current mode does not use an active slave.186 187ad_actor_sys_prio188 189	In an AD system, this specifies the system priority. The allowed range190	is 1 - 65535. If the value is not specified, it takes 65535 as the191	default value.192 193	This parameter has effect only in 802.3ad mode and is available through194	SysFs interface.195 196ad_actor_system197 198	In an AD system, this specifies the mac-address for the actor in199	protocol packet exchanges (LACPDUs). The value cannot be a multicast200	address. If the all-zeroes MAC is specified, bonding will internally201	use the MAC of the bond itself. It is preferred to have the202	local-admin bit set for this mac but driver does not enforce it. If203	the value is not given then system defaults to using the masters'204	mac address as actors' system address.205 206	This parameter has effect only in 802.3ad mode and is available through207	SysFs interface.208 209ad_select210 211	Specifies the 802.3ad aggregation selection logic to use.  The212	possible values and their effects are:213 214	stable or 0215 216		The active aggregator is chosen by largest aggregate217		bandwidth.218 219		Reselection of the active aggregator occurs only when all220		slaves of the active aggregator are down or the active221		aggregator has no slaves.222 223		This is the default value.224 225	bandwidth or 1226 227		The active aggregator is chosen by largest aggregate228		bandwidth.  Reselection occurs if:229 230		- A slave is added to or removed from the bond231 232		- Any slave's link state changes233 234		- Any slave's 802.3ad association state changes235 236		- The bond's administrative state changes to up237 238	count or 2239 240		The active aggregator is chosen by the largest number of241		ports (slaves).  Reselection occurs as described under the242		"bandwidth" setting, above.243 244	The bandwidth and count selection policies permit failover of245	802.3ad aggregations when partial failure of the active aggregator246	occurs.  This keeps the aggregator with the highest availability247	(either in bandwidth or in number of ports) active at all times.248 249	This option was added in bonding version 3.4.0.250 251ad_user_port_key252 253	In an AD system, the port-key has three parts as shown below -254 255	   =====  ============256	   Bits   Use257	   =====  ============258	   00     Duplex259	   01-05  Speed260	   06-15  User-defined261	   =====  ============262 263	This defines the upper 10 bits of the port key. The values can be264	from 0 - 1023. If not given, the system defaults to 0.265 266	This parameter has effect only in 802.3ad mode and is available through267	SysFs interface.268 269all_slaves_active270 271	Specifies that duplicate frames (received on inactive ports) should be272	dropped (0) or delivered (1).273 274	Normally, bonding will drop duplicate frames (received on inactive275	ports), which is desirable for most users. But there are some times276	it is nice to allow duplicate frames to be delivered.277 278	The default value is 0 (drop duplicate frames received on inactive279	ports).280 281arp_interval282 283	Specifies the ARP link monitoring frequency in milliseconds.284 285	The ARP monitor works by periodically checking the slave286	devices to determine whether they have sent or received287	traffic recently (the precise criteria depends upon the288	bonding mode, and the state of the slave).  Regular traffic is289	generated via ARP probes issued for the addresses specified by290	the arp_ip_target option.291 292	This behavior can be modified by the arp_validate option,293	below.294 295	If ARP monitoring is used in an etherchannel compatible mode296	(modes 0 and 2), the switch should be configured in a mode297	that evenly distributes packets across all links. If the298	switch is configured to distribute the packets in an XOR299	fashion, all replies from the ARP targets will be received on300	the same link which could cause the other team members to301	fail.  ARP monitoring should not be used in conjunction with302	miimon.  A value of 0 disables ARP monitoring.  The default303	value is 0.304 305arp_ip_target306 307	Specifies the IP addresses to use as ARP monitoring peers when308	arp_interval is > 0.  These are the targets of the ARP request309	sent to determine the health of the link to the targets.310	Specify these values in ddd.ddd.ddd.ddd format.  Multiple IP311	addresses must be separated by a comma.  At least one IP312	address must be given for ARP monitoring to function.  The313	maximum number of targets that can be specified is 16.  The314	default value is no IP addresses.315 316ns_ip6_target317 318	Specifies the IPv6 addresses to use as IPv6 monitoring peers when319	arp_interval is > 0.  These are the targets of the NS request320	sent to determine the health of the link to the targets.321	Specify these values in ffff:ffff::ffff:ffff format.  Multiple IPv6322	addresses must be separated by a comma.  At least one IPv6323	address must be given for NS/NA monitoring to function.  The324	maximum number of targets that can be specified is 16.  The325	default value is no IPv6 addresses.326 327arp_validate328 329	Specifies whether or not ARP probes and replies should be330	validated in any mode that supports arp monitoring, or whether331	non-ARP traffic should be filtered (disregarded) for link332	monitoring purposes.333 334	Possible values are:335 336	none or 0337 338		No validation or filtering is performed.339 340	active or 1341 342		Validation is performed only for the active slave.343 344	backup or 2345 346		Validation is performed only for backup slaves.347 348	all or 3349 350		Validation is performed for all slaves.351 352	filter or 4353 354		Filtering is applied to all slaves. No validation is355		performed.356 357	filter_active or 5358 359		Filtering is applied to all slaves, validation is performed360		only for the active slave.361 362	filter_backup or 6363 364		Filtering is applied to all slaves, validation is performed365		only for backup slaves.366 367	Validation:368 369	Enabling validation causes the ARP monitor to examine the incoming370	ARP requests and replies, and only consider a slave to be up if it371	is receiving the appropriate ARP traffic.372 373	For an active slave, the validation checks ARP replies to confirm374	that they were generated by an arp_ip_target.  Since backup slaves375	do not typically receive these replies, the validation performed376	for backup slaves is on the broadcast ARP request sent out via the377	active slave.  It is possible that some switch or network378	configurations may result in situations wherein the backup slaves379	do not receive the ARP requests; in such a situation, validation380	of backup slaves must be disabled.381 382	The validation of ARP requests on backup slaves is mainly helping383	bonding to decide which slaves are more likely to work in case of384	the active slave failure, it doesn't really guarantee that the385	backup slave will work if it's selected as the next active slave.386 387	Validation is useful in network configurations in which multiple388	bonding hosts are concurrently issuing ARPs to one or more targets389	beyond a common switch.  Should the link between the switch and390	target fail (but not the switch itself), the probe traffic391	generated by the multiple bonding instances will fool the standard392	ARP monitor into considering the links as still up.  Use of393	validation can resolve this, as the ARP monitor will only consider394	ARP requests and replies associated with its own instance of395	bonding.396 397	Filtering:398 399	Enabling filtering causes the ARP monitor to only use incoming ARP400	packets for link availability purposes.  Arriving packets that are401	not ARPs are delivered normally, but do not count when determining402	if a slave is available.403 404	Filtering operates by only considering the reception of ARP405	packets (any ARP packet, regardless of source or destination) when406	determining if a slave has received traffic for link availability407	purposes.408 409	Filtering is useful in network configurations in which significant410	levels of third party broadcast traffic would fool the standard411	ARP monitor into considering the links as still up.  Use of412	filtering can resolve this, as only ARP traffic is considered for413	link availability purposes.414 415	This option was added in bonding version 3.1.0.416 417arp_all_targets418 419	Specifies the quantity of arp_ip_targets that must be reachable420	in order for the ARP monitor to consider a slave as being up.421	This option affects only active-backup mode for slaves with422	arp_validation enabled.423 424	Possible values are:425 426	any or 0427 428		consider the slave up only when any of the arp_ip_targets429		is reachable430 431	all or 1432 433		consider the slave up only when all of the arp_ip_targets434		are reachable435 436arp_missed_max437 438	Specifies the number of arp_interval monitor checks that must439	fail in order for an interface to be marked down by the ARP monitor.440 441	In order to provide orderly failover semantics, backup interfaces442	are permitted an extra monitor check (i.e., they must fail443	arp_missed_max + 1 times before being marked down).444 445	The default value is 2, and the allowable range is 1 - 255.446 447coupled_control448 449    Specifies whether the LACP state machine's MUX in the 802.3ad mode450    should have separate Collecting and Distributing states.451 452    This is by implementing the independent control state machine per453    IEEE 802.1AX-2008 5.4.15 in addition to the existing coupled control454    state machine.455 456    The default value is 1. This setting does not separate the Collecting457    and Distributing states, maintaining the bond in coupled control.458 459downdelay460 461	Specifies the time, in milliseconds, to wait before disabling462	a slave after a link failure has been detected.  This option463	is only valid for the miimon link monitor.  The downdelay464	value should be a multiple of the miimon value; if not, it465	will be rounded down to the nearest multiple.  The default466	value is 0.467 468fail_over_mac469 470	Specifies whether active-backup mode should set all slaves to471	the same MAC address at enslavement (the traditional472	behavior), or, when enabled, perform special handling of the473	bond's MAC address in accordance with the selected policy.474 475	Possible values are:476 477	none or 0478 479		This setting disables fail_over_mac, and causes480		bonding to set all slaves of an active-backup bond to481		the same MAC address at enslavement time.  This is the482		default.483 484	active or 1485 486		The "active" fail_over_mac policy indicates that the487		MAC address of the bond should always be the MAC488		address of the currently active slave.  The MAC489		address of the slaves is not changed; instead, the MAC490		address of the bond changes during a failover.491 492		This policy is useful for devices that cannot ever493		alter their MAC address, or for devices that refuse494		incoming broadcasts with their own source MAC (which495		interferes with the ARP monitor).496 497		The down side of this policy is that every device on498		the network must be updated via gratuitous ARP,499		vs. just updating a switch or set of switches (which500		often takes place for any traffic, not just ARP501		traffic, if the switch snoops incoming traffic to502		update its tables) for the traditional method.  If the503		gratuitous ARP is lost, communication may be504		disrupted.505 506		When this policy is used in conjunction with the mii507		monitor, devices which assert link up prior to being508		able to actually transmit and receive are particularly509		susceptible to loss of the gratuitous ARP, and an510		appropriate updelay setting may be required.511 512	follow or 2513 514		The "follow" fail_over_mac policy causes the MAC515		address of the bond to be selected normally (normally516		the MAC address of the first slave added to the bond).517		However, the second and subsequent slaves are not set518		to this MAC address while they are in a backup role; a519		slave is programmed with the bond's MAC address at520		failover time (and the formerly active slave receives521		the newly active slave's MAC address).522 523		This policy is useful for multiport devices that524		either become confused or incur a performance penalty525		when multiple ports are programmed with the same MAC526		address.527 528 529	The default policy is none, unless the first slave cannot530	change its MAC address, in which case the active policy is531	selected by default.532 533	This option may be modified via sysfs only when no slaves are534	present in the bond.535 536	This option was added in bonding version 3.2.0.  The "follow"537	policy was added in bonding version 3.3.0.538 539lacp_active540	Option specifying whether to send LACPDU frames periodically.541 542	off or 0543		LACPDU frames acts as "speak when spoken to".544 545	on or 1546		LACPDU frames are sent along the configured links547		periodically. See lacp_rate for more details.548 549	The default is on.550 551lacp_rate552 553	Option specifying the rate in which we'll ask our link partner554	to transmit LACPDU packets in 802.3ad mode.  Possible values555	are:556 557	slow or 0558		Request partner to transmit LACPDUs every 30 seconds559 560	fast or 1561		Request partner to transmit LACPDUs every 1 second562 563	The default is slow.564 565max_bonds566 567	Specifies the number of bonding devices to create for this568	instance of the bonding driver.  E.g., if max_bonds is 3, and569	the bonding driver is not already loaded, then bond0, bond1570	and bond2 will be created.  The default value is 1.  Specifying571	a value of 0 will load bonding, but will not create any devices.572 573miimon574 575	Specifies the MII link monitoring frequency in milliseconds.576	This determines how often the link state of each slave is577	inspected for link failures.  A value of zero disables MII578	link monitoring.  A value of 100 is a good starting point.579	The use_carrier option, below, affects how the link state is580	determined.  See the High Availability section for additional581	information.  The default value is 100 if arp_interval is not582	set.583 584min_links585 586	Specifies the minimum number of links that must be active before587	asserting carrier. It is similar to the Cisco EtherChannel min-links588	feature. This allows setting the minimum number of member ports that589	must be up (link-up state) before marking the bond device as up590	(carrier on). This is useful for situations where higher level services591	such as clustering want to ensure a minimum number of low bandwidth592	links are active before switchover. This option only affect 802.3ad593	mode.594 595	The default value is 0. This will cause carrier to be asserted (for596	802.3ad mode) whenever there is an active aggregator, regardless of the597	number of available links in that aggregator. Note that, because an598	aggregator cannot be active without at least one available link,599	setting this option to 0 or to 1 has the exact same effect.600 601mode602 603	Specifies one of the bonding policies. The default is604	balance-rr (round robin).  Possible values are:605 606	balance-rr or 0607 608		Round-robin policy: Transmit packets in sequential609		order from the first available slave through the610		last.  This mode provides load balancing and fault611		tolerance.612 613	active-backup or 1614 615		Active-backup policy: Only one slave in the bond is616		active.  A different slave becomes active if, and only617		if, the active slave fails.  The bond's MAC address is618		externally visible on only one port (network adapter)619		to avoid confusing the switch.620 621		In bonding version 2.6.2 or later, when a failover622		occurs in active-backup mode, bonding will issue one623		or more gratuitous ARPs on the newly active slave.624		One gratuitous ARP is issued for the bonding master625		interface and each VLAN interfaces configured above626		it, provided that the interface has at least one IP627		address configured.  Gratuitous ARPs issued for VLAN628		interfaces are tagged with the appropriate VLAN id.629 630		This mode provides fault tolerance.  The primary631		option, documented below, affects the behavior of this632		mode.633 634	balance-xor or 2635 636		XOR policy: Transmit based on the selected transmit637		hash policy.  The default policy is a simple [(source638		MAC address XOR'd with destination MAC address XOR639		packet type ID) modulo slave count].  Alternate transmit640		policies may be	selected via the xmit_hash_policy option,641		described below.642 643		This mode provides load balancing and fault tolerance.644 645	broadcast or 3646 647		Broadcast policy: transmits everything on all slave648		interfaces.  This mode provides fault tolerance.649 650	802.3ad or 4651 652		IEEE 802.3ad Dynamic link aggregation.  Creates653		aggregation groups that share the same speed and654		duplex settings.  Utilizes all slaves in the active655		aggregator according to the 802.3ad specification.656 657		Slave selection for outgoing traffic is done according658		to the transmit hash policy, which may be changed from659		the default simple XOR policy via the xmit_hash_policy660		option, documented below.  Note that not all transmit661		policies may be 802.3ad compliant, particularly in662		regards to the packet mis-ordering requirements of663		section 43.2.4 of the 802.3ad standard.  Differing664		peer implementations will have varying tolerances for665		noncompliance.666 667		Prerequisites:668 669		1. Ethtool support in the base drivers for retrieving670		the speed and duplex of each slave.671 672		2. A switch that supports IEEE 802.3ad Dynamic link673		aggregation.674 675		Most switches will require some type of configuration676		to enable 802.3ad mode.677 678	balance-tlb or 5679 680		Adaptive transmit load balancing: channel bonding that681		does not require any special switch support.682 683		In tlb_dynamic_lb=1 mode; the outgoing traffic is684		distributed according to the current load (computed685		relative to the speed) on each slave.686 687		In tlb_dynamic_lb=0 mode; the load balancing based on688		current load is disabled and the load is distributed689		only using the hash distribution.690 691		Incoming traffic is received by the current slave.692		If the receiving slave fails, another slave takes over693		the MAC address of the failed receiving slave.694 695		Prerequisite:696 697		Ethtool support in the base drivers for retrieving the698		speed of each slave.699 700	balance-alb or 6701 702		Adaptive load balancing: includes balance-tlb plus703		receive load balancing (rlb) for IPV4 traffic, and704		does not require any special switch support.  The705		receive load balancing is achieved by ARP negotiation.706		The bonding driver intercepts the ARP Replies sent by707		the local system on their way out and overwrites the708		source hardware address with the unique hardware709		address of one of the slaves in the bond such that710		different peers use different hardware addresses for711		the server.712 713		Receive traffic from connections created by the server714		is also balanced.  When the local system sends an ARP715		Request the bonding driver copies and saves the peer's716		IP information from the ARP packet.  When the ARP717		Reply arrives from the peer, its hardware address is718		retrieved and the bonding driver initiates an ARP719		reply to this peer assigning it to one of the slaves720		in the bond.  A problematic outcome of using ARP721		negotiation for balancing is that each time that an722		ARP request is broadcast it uses the hardware address723		of the bond.  Hence, peers learn the hardware address724		of the bond and the balancing of receive traffic725		collapses to the current slave.  This is handled by726		sending updates (ARP Replies) to all the peers with727		their individually assigned hardware address such that728		the traffic is redistributed.  Receive traffic is also729		redistributed when a new slave is added to the bond730		and when an inactive slave is re-activated.  The731		receive load is distributed sequentially (round robin)732		among the group of highest speed slaves in the bond.733 734		When a link is reconnected or a new slave joins the735		bond the receive traffic is redistributed among all736		active slaves in the bond by initiating ARP Replies737		with the selected MAC address to each of the738		clients. The updelay parameter (detailed below) must739		be set to a value equal or greater than the switch's740		forwarding delay so that the ARP Replies sent to the741		peers will not be blocked by the switch.742 743		Prerequisites:744 745		1. Ethtool support in the base drivers for retrieving746		the speed of each slave.747 748		2. Base driver support for setting the hardware749		address of a device while it is open.  This is750		required so that there will always be one slave in the751		team using the bond hardware address (the752		curr_active_slave) while having a unique hardware753		address for each slave in the bond.  If the754		curr_active_slave fails its hardware address is755		swapped with the new curr_active_slave that was756		chosen.757 758num_grat_arp,759num_unsol_na760 761	Specify the number of peer notifications (gratuitous ARPs and762	unsolicited IPv6 Neighbor Advertisements) to be issued after a763	failover event.  As soon as the link is up on the new slave764	(possibly immediately) a peer notification is sent on the765	bonding device and each VLAN sub-device. This is repeated at766	the rate specified by peer_notif_delay if the number is767	greater than 1.768 769	The valid range is 0 - 255; the default value is 1.  These options770	affect only the active-backup mode.  These options were added for771	bonding versions 3.3.0 and 3.4.0 respectively.772 773	From Linux 3.0 and bonding version 3.7.1, these notifications774	are generated by the ipv4 and ipv6 code and the numbers of775	repetitions cannot be set independently.776 777packets_per_slave778 779	Specify the number of packets to transmit through a slave before780	moving to the next one. When set to 0 then a slave is chosen at781	random.782 783	The valid range is 0 - 65535; the default value is 1. This option784	has effect only in balance-rr mode.785 786peer_notif_delay787 788	Specify the delay, in milliseconds, between each peer789	notification (gratuitous ARP and unsolicited IPv6 Neighbor790	Advertisement) when they are issued after a failover event.791	This delay should be a multiple of the MII link monitor interval792	(miimon).793 794	The valid range is 0 - 300000. The default value is 0, which means795	to match the value of the MII link monitor interval.796 797prio798	Slave priority. A higher number means higher priority.799	The primary slave has the highest priority. This option also800	follows the primary_reselect rules.801 802	This option could only be configured via netlink, and is only valid803	for active-backup(1), balance-tlb (5) and balance-alb (6) mode.804	The valid value range is a signed 32 bit integer.805 806	The default value is 0.807 808primary809 810	A string (eth0, eth2, etc) specifying which slave is the811	primary device.  The specified device will always be the812	active slave while it is available.  Only when the primary is813	off-line will alternate devices be used.  This is useful when814	one slave is preferred over another, e.g., when one slave has815	higher throughput than another.816 817	The primary option is only valid for active-backup(1),818	balance-tlb (5) and balance-alb (6) mode.819 820primary_reselect821 822	Specifies the reselection policy for the primary slave.  This823	affects how the primary slave is chosen to become the active slave824	when failure of the active slave or recovery of the primary slave825	occurs.  This option is designed to prevent flip-flopping between826	the primary slave and other slaves.  Possible values are:827 828	always or 0 (default)829 830		The primary slave becomes the active slave whenever it831		comes back up.832 833	better or 1834 835		The primary slave becomes the active slave when it comes836		back up, if the speed and duplex of the primary slave is837		better than the speed and duplex of the current active838		slave.839 840	failure or 2841 842		The primary slave becomes the active slave only if the843		current active slave fails and the primary slave is up.844 845	The primary_reselect setting is ignored in two cases:846 847		If no slaves are active, the first slave to recover is848		made the active slave.849 850		When initially enslaved, the primary slave is always made851		the active slave.852 853	Changing the primary_reselect policy via sysfs will cause an854	immediate selection of the best active slave according to the new855	policy.  This may or may not result in a change of the active856	slave, depending upon the circumstances.857 858	This option was added for bonding version 3.6.0.859 860tlb_dynamic_lb861 862	Specifies if dynamic shuffling of flows is enabled in tlb863	or alb mode. The value has no effect on any other modes.864 865	The default behavior of tlb mode is to shuffle active flows across866	slaves based on the load in that interval. This gives nice lb867	characteristics but can cause packet reordering. If re-ordering is868	a concern use this variable to disable flow shuffling and rely on869	load balancing provided solely by the hash distribution.870	xmit-hash-policy can be used to select the appropriate hashing for871	the setup.872 873	The sysfs entry can be used to change the setting per bond device874	and the initial value is derived from the module parameter. The875	sysfs entry is allowed to be changed only if the bond device is876	down.877 878	The default value is "1" that enables flow shuffling while value "0"879	disables it. This option was added in bonding driver 3.7.1880 881 882updelay883 884	Specifies the time, in milliseconds, to wait before enabling a885	slave after a link recovery has been detected.  This option is886	only valid for the miimon link monitor.  The updelay value887	should be a multiple of the miimon value; if not, it will be888	rounded down to the nearest multiple.  The default value is 0.889 890use_carrier891 892	Specifies whether or not miimon should use MII or ETHTOOL893	ioctls vs. netif_carrier_ok() to determine the link894	status. The MII or ETHTOOL ioctls are less efficient and895	utilize a deprecated calling sequence within the kernel.  The896	netif_carrier_ok() relies on the device driver to maintain its897	state with netif_carrier_on/off; at this writing, most, but898	not all, device drivers support this facility.899 900	If bonding insists that the link is up when it should not be,901	it may be that your network device driver does not support902	netif_carrier_on/off.  The default state for netif_carrier is903	"carrier on," so if a driver does not support netif_carrier,904	it will appear as if the link is always up.  In this case,905	setting use_carrier to 0 will cause bonding to revert to the906	MII / ETHTOOL ioctl method to determine the link state.907 908	A value of 1 enables the use of netif_carrier_ok(), a value of909	0 will use the deprecated MII / ETHTOOL ioctls.  The default910	value is 1.911 912xmit_hash_policy913 914	Selects the transmit hash policy to use for slave selection in915	balance-xor, 802.3ad, and tlb modes.  Possible values are:916 917	layer2918 919		Uses XOR of hardware MAC addresses and packet type ID920		field to generate the hash. The formula is921 922		hash = source MAC[5] XOR destination MAC[5] XOR packet type ID923		slave number = hash modulo slave count924 925		This algorithm will place all traffic to a particular926		network peer on the same slave.927 928		This algorithm is 802.3ad compliant.929 930	layer2+3931 932		This policy uses a combination of layer2 and layer3933		protocol information to generate the hash.934 935		Uses XOR of hardware MAC addresses and IP addresses to936		generate the hash.  The formula is937 938		hash = source MAC[5] XOR destination MAC[5] XOR packet type ID939		hash = hash XOR source IP XOR destination IP940		hash = hash XOR (hash RSHIFT 16)941		hash = hash XOR (hash RSHIFT 8)942		And then hash is reduced modulo slave count.943 944		If the protocol is IPv6 then the source and destination945		addresses are first hashed using ipv6_addr_hash.946 947		This algorithm will place all traffic to a particular948		network peer on the same slave.  For non-IP traffic,949		the formula is the same as for the layer2 transmit950		hash policy.951 952		This policy is intended to provide a more balanced953		distribution of traffic than layer2 alone, especially954		in environments where a layer3 gateway device is955		required to reach most destinations.956 957		This algorithm is 802.3ad compliant.958 959	layer3+4960 961		This policy uses upper layer protocol information,962		when available, to generate the hash.  This allows for963		traffic to a particular network peer to span multiple964		slaves, although a single connection will not span965		multiple slaves.966 967		The formula for unfragmented TCP and UDP packets is968 969		hash = source port, destination port (as in the header)970		hash = hash XOR source IP XOR destination IP971		hash = hash XOR (hash RSHIFT 16)972		hash = hash XOR (hash RSHIFT 8)973		hash = hash RSHIFT 1974		And then hash is reduced modulo slave count.975 976		If the protocol is IPv6 then the source and destination977		addresses are first hashed using ipv6_addr_hash.978 979		For fragmented TCP or UDP packets and all other IPv4 and980		IPv6 protocol traffic, the source and destination port981		information is omitted.  For non-IP traffic, the982		formula is the same as for the layer2 transmit hash983		policy.984 985		This algorithm is not fully 802.3ad compliant.  A986		single TCP or UDP conversation containing both987		fragmented and unfragmented packets will see packets988		striped across two interfaces.  This may result in out989		of order delivery.  Most traffic types will not meet990		this criteria, as TCP rarely fragments traffic, and991		most UDP traffic is not involved in extended992		conversations.  Other implementations of 802.3ad may993		or may not tolerate this noncompliance.994 995	encap2+3996 997		This policy uses the same formula as layer2+3 but it998		relies on skb_flow_dissect to obtain the header fields999		which might result in the use of inner headers if an1000		encapsulation protocol is used. For example this will1001		improve the performance for tunnel users because the1002		packets will be distributed according to the encapsulated1003		flows.1004 1005	encap3+41006 1007		This policy uses the same formula as layer3+4 but it1008		relies on skb_flow_dissect to obtain the header fields1009		which might result in the use of inner headers if an1010		encapsulation protocol is used. For example this will1011		improve the performance for tunnel users because the1012		packets will be distributed according to the encapsulated1013		flows.1014 1015	vlan+srcmac1016 1017		This policy uses a very rudimentary vlan ID and source mac1018		hash to load-balance traffic per-vlan, with failover1019		should one leg fail. The intended use case is for a bond1020		shared by multiple virtual machines, all configured to1021		use their own vlan, to give lacp-like functionality1022		without requiring lacp-capable switching hardware.1023 1024		The formula for the hash is simply1025 1026		hash = (vlan ID) XOR (source MAC vendor) XOR (source MAC dev)1027 1028	The default value is layer2.  This option was added in bonding1029	version 2.6.3.  In earlier versions of bonding, this parameter1030	does not exist, and the layer2 policy is the only policy.  The1031	layer2+3 value was added for bonding version 3.2.2.1032 1033resend_igmp1034 1035	Specifies the number of IGMP membership reports to be issued after1036	a failover event. One membership report is issued immediately after1037	the failover, subsequent packets are sent in each 200ms interval.1038 1039	The valid range is 0 - 255; the default value is 1. A value of 01040	prevents the IGMP membership report from being issued in response1041	to the failover event.1042 1043	This option is useful for bonding modes balance-rr (0), active-backup1044	(1), balance-tlb (5) and balance-alb (6), in which a failover can1045	switch the IGMP traffic from one slave to another.  Therefore a fresh1046	IGMP report must be issued to cause the switch to forward the incoming1047	IGMP traffic over the newly selected slave.1048 1049	This option was added for bonding version 3.7.0.1050 1051lp_interval1052 1053	Specifies the number of seconds between instances where the bonding1054	driver sends learning packets to each slaves peer switch.1055 1056	The valid range is 1 - 0x7fffffff; the default value is 1. This Option1057	has effect only in balance-tlb and balance-alb modes.1058 10593. Configuring Bonding Devices1060==============================1061 1062You can configure bonding using either your distro's network1063initialization scripts, or manually using either iproute2 or the1064sysfs interface.  Distros generally use one of three packages for the1065network initialization scripts: initscripts, sysconfig or interfaces.1066Recent versions of these packages have support for bonding, while older1067versions do not.1068 1069We will first describe the options for configuring bonding for1070distros using versions of initscripts, sysconfig and interfaces with full1071or partial support for bonding, then provide information on enabling1072bonding without support from the network initialization scripts (i.e.,1073older versions of initscripts or sysconfig).1074 1075If you're unsure whether your distro uses sysconfig,1076initscripts or interfaces, or don't know if it's new enough, have no fear.1077Determining this is fairly straightforward.1078 1079First, look for a file called interfaces in /etc/network directory.1080If this file is present in your system, then your system use interfaces. See1081Configuration with Interfaces Support.1082 1083Else, issue the command::1084 1085	$ rpm -qf /sbin/ifup1086 1087It will respond with a line of text starting with either1088"initscripts" or "sysconfig," followed by some numbers.  This is the1089package that provides your network initialization scripts.1090 1091Next, to determine if your installation supports bonding,1092issue the command::1093 1094    $ grep ifenslave /sbin/ifup1095 1096If this returns any matches, then your initscripts or1097sysconfig has support for bonding.1098 10993.1 Configuration with Sysconfig Support1100----------------------------------------1101 1102This section applies to distros using a version of sysconfig1103with bonding support, for example, SuSE Linux Enterprise Server 9.1104 1105SuSE SLES 9's networking configuration system does support1106bonding, however, at this writing, the YaST system configuration1107front end does not provide any means to work with bonding devices.1108Bonding devices can be managed by hand, however, as follows.1109 1110First, if they have not already been configured, configure the1111slave devices.  On SLES 9, this is most easily done by running the1112yast2 sysconfig configuration utility.  The goal is for to create an1113ifcfg-id file for each slave device.  The simplest way to accomplish1114this is to configure the devices for DHCP (this is only to get the1115file ifcfg-id file created; see below for some issues with DHCP).  The1116name of the configuration file for each device will be of the form::1117 1118    ifcfg-id-xx:xx:xx:xx:xx:xx1119 1120Where the "xx" portion will be replaced with the digits from1121the device's permanent MAC address.1122 1123Once the set of ifcfg-id-xx:xx:xx:xx:xx:xx files has been1124created, it is necessary to edit the configuration files for the slave1125devices (the MAC addresses correspond to those of the slave devices).1126Before editing, the file will contain multiple lines, and will look1127something like this::1128 1129	BOOTPROTO='dhcp'1130	STARTMODE='on'1131	USERCTL='no'1132	UNIQUE='XNzu.WeZGOGF+4wE'1133	_nm_name='bus-pci-0001:61:01.0'1134 1135Change the BOOTPROTO and STARTMODE lines to the following::1136 1137	BOOTPROTO='none'1138	STARTMODE='off'1139 1140Do not alter the UNIQUE or _nm_name lines.  Remove any other1141lines (USERCTL, etc).1142 1143Once the ifcfg-id-xx:xx:xx:xx:xx:xx files have been modified,1144it's time to create the configuration file for the bonding device1145itself.  This file is named ifcfg-bondX, where X is the number of the1146bonding device to create, starting at 0.  The first such file is1147ifcfg-bond0, the second is ifcfg-bond1, and so on.  The sysconfig1148network configuration system will correctly start multiple instances1149of bonding.1150 1151The contents of the ifcfg-bondX file is as follows::1152 1153	BOOTPROTO="static"1154	BROADCAST="10.0.2.255"1155	IPADDR="10.0.2.10"1156	NETMASK="255.255.0.0"1157	NETWORK="10.0.2.0"1158	REMOTE_IPADDR=""1159	STARTMODE="onboot"1160	BONDING_MASTER="yes"1161	BONDING_MODULE_OPTS="mode=active-backup miimon=100"1162	BONDING_SLAVE0="eth0"1163	BONDING_SLAVE1="bus-pci-0000:06:08.1"1164 1165Replace the sample BROADCAST, IPADDR, NETMASK and NETWORK1166values with the appropriate values for your network.1167 1168The STARTMODE specifies when the device is brought online.1169The possible values are:1170 1171	======== ======================================================1172	onboot	 The device is started at boot time.  If you're not1173		 sure, this is probably what you want.1174 1175	manual	 The device is started only when ifup is called1176		 manually.  Bonding devices may be configured this1177		 way if you do not wish them to start automatically1178		 at boot for some reason.1179 1180	hotplug  The device is started by a hotplug event.  This is not1181		 a valid choice for a bonding device.1182 1183	off or   The device configuration is ignored.1184	ignore1185	======== ======================================================1186 1187The line BONDING_MASTER='yes' indicates that the device is a1188bonding master device.  The only useful value is "yes."1189 1190The contents of BONDING_MODULE_OPTS are supplied to the1191instance of the bonding module for this device.  Specify the options1192for the bonding mode, link monitoring, and so on here.  Do not include1193the max_bonds bonding parameter; this will confuse the configuration1194system if you have multiple bonding devices.1195 1196Finally, supply one BONDING_SLAVEn="slave device" for each1197slave.  where "n" is an increasing value, one for each slave.  The1198"slave device" is either an interface name, e.g., "eth0", or a device1199specifier for the network device.  The interface name is easier to1200find, but the ethN names are subject to change at boot time if, e.g.,1201a device early in the sequence has failed.  The device specifiers1202(bus-pci-0000:06:08.1 in the example above) specify the physical1203network device, and will not change unless the device's bus location1204changes (for example, it is moved from one PCI slot to another).  The1205example above uses one of each type for demonstration purposes; most1206configurations will choose one or the other for all slave devices.1207 1208When all configuration files have been modified or created,1209networking must be restarted for the configuration changes to take1210effect.  This can be accomplished via the following::1211 1212	# /etc/init.d/network restart1213 1214Note that the network control script (/sbin/ifdown) will1215remove the bonding module as part of the network shutdown processing,1216so it is not necessary to remove the module by hand if, e.g., the1217module parameters have changed.1218 1219Also, at this writing, YaST/YaST2 will not manage bonding1220devices (they do not show bonding interfaces on its list of network1221devices).  It is necessary to edit the configuration file by hand to1222change the bonding configuration.1223 1224Additional general options and details of the ifcfg file1225format can be found in an example ifcfg template file::1226 1227	/etc/sysconfig/network/ifcfg.template1228 1229Note that the template does not document the various ``BONDING_*``1230settings described above, but does describe many of the other options.1231 12323.1.1 Using DHCP with Sysconfig1233-------------------------------1234 1235Under sysconfig, configuring a device with BOOTPROTO='dhcp'1236will cause it to query DHCP for its IP address information.  At this1237writing, this does not function for bonding devices; the scripts1238attempt to obtain the device address from DHCP prior to adding any of1239the slave devices.  Without active slaves, the DHCP requests are not1240sent to the network.1241 12423.1.2 Configuring Multiple Bonds with Sysconfig1243-----------------------------------------------1244 1245The sysconfig network initialization system is capable of1246handling multiple bonding devices.  All that is necessary is for each1247bonding instance to have an appropriately configured ifcfg-bondX file1248(as described above).  Do not specify the "max_bonds" parameter to any1249instance of bonding, as this will confuse sysconfig.  If you require1250multiple bonding devices with identical parameters, create multiple1251ifcfg-bondX files.1252 1253Because the sysconfig scripts supply the bonding module1254options in the ifcfg-bondX file, it is not necessary to add them to1255the system ``/etc/modules.d/*.conf`` configuration files.1256 12573.2 Configuration with Initscripts Support1258------------------------------------------1259 1260This section applies to distros using a recent version of1261initscripts with bonding support, for example, Red Hat Enterprise Linux1262version 3 or later, Fedora, etc.  On these systems, the network1263initialization scripts have knowledge of bonding, and can be configured to1264control bonding devices.  Note that older versions of the initscripts1265package have lower levels of support for bonding; this will be noted where1266applicable.1267 1268These distros will not automatically load the network adapter1269driver unless the ethX device is configured with an IP address.1270Because of this constraint, users must manually configure a1271network-script file for all physical adapters that will be members of1272a bondX link.  Network script files are located in the directory:1273 1274/etc/sysconfig/network-scripts1275 1276The file name must be prefixed with "ifcfg-eth" and suffixed1277with the adapter's physical adapter number.  For example, the script1278for eth0 would be named /etc/sysconfig/network-scripts/ifcfg-eth0.1279Place the following text in the file::1280 1281	DEVICE=eth01282	USERCTL=no1283	ONBOOT=yes1284	MASTER=bond01285	SLAVE=yes1286	BOOTPROTO=none1287 1288The DEVICE= line will be different for every ethX device and1289must correspond with the name of the file, i.e., ifcfg-eth1 must have1290a device line of DEVICE=eth1.  The setting of the MASTER= line will1291also depend on the final bonding interface name chosen for your bond.1292As with other network devices, these typically start at 0, and go up1293one for each device, i.e., the first bonding instance is bond0, the1294second is bond1, and so on.1295 1296Next, create a bond network script.  The file name for this1297script will be /etc/sysconfig/network-scripts/ifcfg-bondX where X is1298the number of the bond.  For bond0 the file is named "ifcfg-bond0",1299for bond1 it is named "ifcfg-bond1", and so on.  Within that file,1300place the following text::1301 1302	DEVICE=bond01303	IPADDR=192.168.1.11304	NETMASK=255.255.255.01305	NETWORK=192.168.1.01306	BROADCAST=192.168.1.2551307	ONBOOT=yes1308	BOOTPROTO=none1309	USERCTL=no1310 1311Be sure to change the networking specific lines (IPADDR,1312NETMASK, NETWORK and BROADCAST) to match your network configuration.1313 1314For later versions of initscripts, such as that found with Fedora13157 (or later) and Red Hat Enterprise Linux version 5 (or later), it is possible,1316and, indeed, preferable, to specify the bonding options in the ifcfg-bond01317file, e.g. a line of the format::1318 1319  BONDING_OPTS="mode=active-backup arp_interval=60 arp_ip_target=192.168.1.254"1320 1321will configure the bond with the specified options.  The options1322specified in BONDING_OPTS are identical to the bonding module parameters1323except for the arp_ip_target field when using versions of initscripts older1324than and 8.57 (Fedora 8) and 8.45.19 (Red Hat Enterprise Linux 5.2).  When1325using older versions each target should be included as a separate option and1326should be preceded by a '+' to indicate it should be added to the list of1327queried targets, e.g.,::1328 1329    arp_ip_target=+192.168.1.1 arp_ip_target=+192.168.1.21330 1331is the proper syntax to specify multiple targets.  When specifying1332options via BONDING_OPTS, it is not necessary to edit1333``/etc/modprobe.d/*.conf``.1334 1335For even older versions of initscripts that do not support1336BONDING_OPTS, it is necessary to edit /etc/modprobe.d/*.conf, depending upon1337your distro) to load the bonding module with your desired options when the1338bond0 interface is brought up.  The following lines in /etc/modprobe.d/*.conf1339will load the bonding module, and select its options:1340 1341	alias bond0 bonding1342	options bond0 mode=balance-alb miimon=1001343 1344Replace the sample parameters with the appropriate set of1345options for your configuration.1346 1347Finally run "/etc/rc.d/init.d/network restart" as root.  This1348will restart the networking subsystem and your bond link should be now1349up and running.1350 13513.2.1 Using DHCP with Initscripts1352---------------------------------1353 1354Recent versions of initscripts (the versions supplied with Fedora1355Core 3 and Red Hat Enterprise Linux 4, or later versions, are reported to1356work) have support for assigning IP information to bonding devices via1357DHCP.1358 1359To configure bonding for DHCP, configure it as described1360above, except replace the line "BOOTPROTO=none" with "BOOTPROTO=dhcp"1361and add a line consisting of "TYPE=Bonding".  Note that the TYPE value1362is case sensitive.1363 13643.2.2 Configuring Multiple Bonds with Initscripts1365-------------------------------------------------1366 1367Initscripts packages that are included with Fedora 7 and Red Hat1368Enterprise Linux 5 support multiple bonding interfaces by simply1369specifying the appropriate BONDING_OPTS= in ifcfg-bondX where X is the1370number of the bond.  This support requires sysfs support in the kernel,1371and a bonding driver of version 3.0.0 or later.  Other configurations may1372not support this method for specifying multiple bonding interfaces; for1373those instances, see the "Configuring Multiple Bonds Manually" section,1374below.1375 13763.3 Configuring Bonding Manually with iproute21377-----------------------------------------------1378 1379This section applies to distros whose network initialization1380scripts (the sysconfig or initscripts package) do not have specific1381knowledge of bonding.  One such distro is SuSE Linux Enterprise Server1382version 8.1383 1384The general method for these systems is to place the bonding1385module parameters into a config file in /etc/modprobe.d/ (as1386appropriate for the installed distro), then add modprobe and/or1387`ip link` commands to the system's global init script.  The name of1388the global init script differs; for sysconfig, it is1389/etc/init.d/boot.local and for initscripts it is /etc/rc.d/rc.local.1390 1391For example, if you wanted to make a simple bond of two e1001392devices (presumed to be eth0 and eth1), and have it persist across1393reboots, edit the appropriate file (/etc/init.d/boot.local or1394/etc/rc.d/rc.local), and add the following::1395 1396	modprobe bonding mode=balance-alb miimon=1001397	modprobe e1001398	ifconfig bond0 192.168.1.1 netmask 255.255.255.0 up1399	ip link set eth0 master bond01400	ip link set eth1 master bond01401 1402Replace the example bonding module parameters and bond01403network configuration (IP address, netmask, etc) with the appropriate1404values for your configuration.1405 1406Unfortunately, this method will not provide support for the1407ifup and ifdown scripts on the bond devices.  To reload the bonding1408configuration, it is necessary to run the initialization script, e.g.,::1409 1410	# /etc/init.d/boot.local1411 1412or::1413 1414	# /etc/rc.d/rc.local1415 1416It may be desirable in such a case to create a separate script1417which only initializes the bonding configuration, then call that1418separate script from within boot.local.  This allows for bonding to be1419enabled without re-running the entire global init script.1420 1421To shut down the bonding devices, it is necessary to first1422mark the bonding device itself as being down, then remove the1423appropriate device driver modules.  For our example above, you can do1424the following::1425 1426	# ifconfig bond0 down1427	# rmmod bonding1428	# rmmod e1001429 1430Again, for convenience, it may be desirable to create a script1431with these commands.1432 1433 14343.3.1 Configuring Multiple Bonds Manually1435-----------------------------------------1436 1437This section contains information on configuring multiple1438bonding devices with differing options for those systems whose network1439initialization scripts lack support for configuring multiple bonds.1440 1441If you require multiple bonding devices, but all with the same1442options, you may wish to use the "max_bonds" module parameter,1443documented above.1444 1445To create multiple bonding devices with differing options, it is1446preferable to use bonding parameters exported by sysfs, documented in the1447section below.1448 1449For versions of bonding without sysfs support, the only means to1450provide multiple instances of bonding with differing options is to load1451the bonding driver multiple times.  Note that current versions of the1452sysconfig network initialization scripts handle this automatically; if1453your distro uses these scripts, no special action is needed.  See the1454section Configuring Bonding Devices, above, if you're not sure about your1455network initialization scripts.1456 1457To load multiple instances of the module, it is necessary to1458specify a different name for each instance (the module loading system1459requires that every loaded module, even multiple instances of the same1460module, have a unique name).  This is accomplished by supplying multiple1461sets of bonding options in ``/etc/modprobe.d/*.conf``, for example::1462 1463	alias bond0 bonding1464	options bond0 -o bond0 mode=balance-rr miimon=1001465 1466	alias bond1 bonding1467	options bond1 -o bond1 mode=balance-alb miimon=501468 1469will load the bonding module two times.  The first instance is1470named "bond0" and creates the bond0 device in balance-rr mode with an1471miimon of 100.  The second instance is named "bond1" and creates the1472bond1 device in balance-alb mode with an miimon of 50.1473 1474In some circumstances (typically with older distributions),1475the above does not work, and the second bonding instance never sees1476its options.  In that case, the second options line can be substituted1477as follows::1478 1479	install bond1 /sbin/modprobe --ignore-install bonding -o bond1 \1480				     mode=balance-alb miimon=501481 1482This may be repeated any number of times, specifying a new and1483unique name in place of bond1 for each subsequent instance.1484 1485It has been observed that some Red Hat supplied kernels are unable1486to rename modules at load time (the "-o bond1" part).  Attempts to pass1487that option to modprobe will produce an "Operation not permitted" error.1488This has been reported on some Fedora Core kernels, and has been seen on1489RHEL 4 as well.  On kernels exhibiting this problem, it will be impossible1490to configure multiple bonds with differing parameters (as they are older1491kernels, and also lack sysfs support).1492 14933.4 Configuring Bonding Manually via Sysfs1494------------------------------------------1495 1496Starting with version 3.0.0, Channel Bonding may be configured1497via the sysfs interface.  This interface allows dynamic configuration1498of all bonds in the system without unloading the module.  It also1499allows for adding and removing bonds at runtime.  Ifenslave is no1500longer required, though it is still supported.1501 1502Use of the sysfs interface allows you to use multiple bonds1503with different configurations without having to reload the module.1504It also allows you to use multiple, differently configured bonds when1505bonding is compiled into the kernel.1506 1507You must have the sysfs filesystem mounted to configure1508bonding this way.  The examples in this document assume that you1509are using the standard mount point for sysfs, e.g. /sys.  If your1510sysfs filesystem is mounted elsewhere, you will need to adjust the1511example paths accordingly.1512 1513Creating and Destroying Bonds1514-----------------------------1515To add a new bond foo::1516 1517	# echo +foo > /sys/class/net/bonding_masters1518 1519To remove an existing bond bar::1520 1521	# echo -bar > /sys/class/net/bonding_masters1522 1523To show all existing bonds::1524 1525	# cat /sys/class/net/bonding_masters1526 1527.. note::1528 1529   due to 4K size limitation of sysfs files, this list may be1530   truncated if you have more than a few hundred bonds.  This is unlikely1531   to occur under normal operating conditions.1532 1533Adding and Removing Slaves1534--------------------------1535Interfaces may be enslaved to a bond using the file1536/sys/class/net/<bond>/bonding/slaves.  The semantics for this file1537are the same as for the bonding_masters file.1538 1539To enslave interface eth0 to bond bond0::1540 1541	# ifconfig bond0 up1542	# echo +eth0 > /sys/class/net/bond0/bonding/slaves1543 1544To free slave eth0 from bond bond0::1545 1546	# echo -eth0 > /sys/class/net/bond0/bonding/slaves1547 1548When an interface is enslaved to a bond, symlinks between the1549two are created in the sysfs filesystem.  In this case, you would get1550/sys/class/net/bond0/slave_eth0 pointing to /sys/class/net/eth0, and1551/sys/class/net/eth0/master pointing to /sys/class/net/bond0.1552 1553This means that you can tell quickly whether or not an1554interface is enslaved by looking for the master symlink.  Thus:1555# echo -eth0 > /sys/class/net/eth0/master/bonding/slaves1556will free eth0 from whatever bond it is enslaved to, regardless of1557the name of the bond interface.1558 1559Changing a Bond's Configuration1560-------------------------------1561Each bond may be configured individually by manipulating the1562files located in /sys/class/net/<bond name>/bonding1563 1564The names of these files correspond directly with the command-1565line parameters described elsewhere in this file, and, with the1566exception of arp_ip_target, they accept the same values.  To see the1567current setting, simply cat the appropriate file.1568 1569A few examples will be given here; for specific usage1570guidelines for each parameter, see the appropriate section in this1571document.1572 1573To configure bond0 for balance-alb mode::1574 1575	# ifconfig bond0 down1576	# echo 6 > /sys/class/net/bond0/bonding/mode1577	- or -1578	# echo balance-alb > /sys/class/net/bond0/bonding/mode1579 1580.. note::1581 1582   The bond interface must be down before the mode can be changed.1583 1584To enable MII monitoring on bond0 with a 1 second interval::1585 1586	# echo 1000 > /sys/class/net/bond0/bonding/miimon1587 1588.. note::1589 1590   If ARP monitoring is enabled, it will disabled when MII1591   monitoring is enabled, and vice-versa.1592 1593To add ARP targets::1594 1595	# echo +192.168.0.100 > /sys/class/net/bond0/bonding/arp_ip_target1596	# echo +192.168.0.101 > /sys/class/net/bond0/bonding/arp_ip_target1597 1598.. note::1599 1600   up to 16 target addresses may be specified.1601 1602To remove an ARP target::1603 1604	# echo -192.168.0.100 > /sys/class/net/bond0/bonding/arp_ip_target1605 1606To configure the interval between learning packet transmits::1607 1608	# echo 12 > /sys/class/net/bond0/bonding/lp_interval1609 1610.. note::1611 1612   the lp_interval is the number of seconds between instances where1613   the bonding driver sends learning packets to each slaves peer switch.  The1614   default interval is 1 second.1615 1616Example Configuration1617---------------------1618We begin with the same example that is shown in section 3.3,1619executed with sysfs, and without using ifenslave.1620 1621To make a simple bond of two e100 devices (presumed to be eth01622and eth1), and have it persist across reboots, edit the appropriate1623file (/etc/init.d/boot.local or /etc/rc.d/rc.local), and add the1624following::1625 1626	modprobe bonding1627	modprobe e1001628	echo balance-alb > /sys/class/net/bond0/bonding/mode1629	ifconfig bond0 192.168.1.1 netmask 255.255.255.0 up1630	echo 100 > /sys/class/net/bond0/bonding/miimon1631	echo +eth0 > /sys/class/net/bond0/bonding/slaves1632	echo +eth1 > /sys/class/net/bond0/bonding/slaves1633 1634To add a second bond, with two e1000 interfaces in1635active-backup mode, using ARP monitoring, add the following lines to1636your init script::1637 1638	modprobe e10001639	echo +bond1 > /sys/class/net/bonding_masters1640	echo active-backup > /sys/class/net/bond1/bonding/mode1641	ifconfig bond1 192.168.2.1 netmask 255.255.255.0 up1642	echo +192.168.2.100 /sys/class/net/bond1/bonding/arp_ip_target1643	echo 2000 > /sys/class/net/bond1/bonding/arp_interval1644	echo +eth2 > /sys/class/net/bond1/bonding/slaves1645	echo +eth3 > /sys/class/net/bond1/bonding/slaves1646 16473.5 Configuration with Interfaces Support1648-----------------------------------------1649 1650This section applies to distros which use /etc/network/interfaces file1651to describe network interface configuration, most notably Debian and its1652derivatives.1653 1654The ifup and ifdown commands on Debian don't support bonding out of1655the box. The ifenslave-2.6 package should be installed to provide bonding1656support.  Once installed, this package will provide ``bond-*`` options1657to be used into /etc/network/interfaces.1658 1659Note that ifenslave-2.6 package will load the bonding module and use1660the ifenslave command when appropriate.1661 1662Example Configurations1663----------------------1664 1665In /etc/network/interfaces, the following stanza will configure bond0, in1666active-backup mode, with eth0 and eth1 as slaves::1667 1668	auto bond01669	iface bond0 inet dhcp1670		bond-slaves eth0 eth11671		bond-mode active-backup1672		bond-miimon 1001673		bond-primary eth0 eth11674 1675If the above configuration doesn't work, you might have a system using1676upstart for system startup. This is most notably true for recent1677Ubuntu versions. The following stanza in /etc/network/interfaces will1678produce the same result on those systems::1679 1680	auto bond01681	iface bond0 inet dhcp1682		bond-slaves none1683		bond-mode active-backup1684		bond-miimon 1001685 1686	auto eth01687	iface eth0 inet manual1688		bond-master bond01689		bond-primary eth0 eth11690 1691	auto eth11692	iface eth1 inet manual1693		bond-master bond01694		bond-primary eth0 eth11695 1696For a full list of ``bond-*`` supported options in /etc/network/interfaces and1697some more advanced examples tailored to you particular distros, see the files in1698/usr/share/doc/ifenslave-2.6.1699 17003.6 Overriding Configuration for Special Cases1701----------------------------------------------1702 1703When using the bonding driver, the physical port which transmits a frame is1704typically selected by the bonding driver, and is not relevant to the user or1705system administrator.  The output port is simply selected using the policies of1706the selected bonding mode.  On occasion however, it is helpful to direct certain1707classes of traffic to certain physical interfaces on output to implement1708slightly more complex policies.  For example, to reach a web server over a1709bonded interface in which eth0 connects to a private network, while eth11710connects via a public network, it may be desirous to bias the bond to send said1711traffic over eth0 first, using eth1 only as a fall back, while all other traffic1712can safely be sent over either interface.  Such configurations may be achieved1713using the traffic control utilities inherent in linux.1714 1715By default the bonding driver is multiqueue aware and 16 queues are created1716when the driver initializes (see Documentation/networking/multiqueue.rst1717for details).  If more or less queues are desired the module parameter1718tx_queues can be used to change this value.  There is no sysfs parameter1719available as the allocation is done at module init time.1720 1721The output of the file /proc/net/bonding/bondX has changed so the output Queue1722ID is now printed for each slave::1723 1724	Bonding Mode: fault-tolerance (active-backup)1725	Primary Slave: None1726	Currently Active Slave: eth01727	MII Status: up1728	MII Polling Interval (ms): 01729	Up Delay (ms): 01730	Down Delay (ms): 01731 1732	Slave Interface: eth01733	MII Status: up1734	Link Failure Count: 01735	Permanent HW addr: 00:1a:a0:12:8f:cb1736	Slave queue ID: 01737 1738	Slave Interface: eth11739	MII Status: up1740	Link Failure Count: 01741	Permanent HW addr: 00:1a:a0:12:8f:cc1742	Slave queue ID: 21743 1744The queue_id for a slave can be set using the command::1745 1746	# echo "eth1:2" > /sys/class/net/bond0/bonding/queue_id1747 1748Any interface that needs a queue_id set should set it with multiple calls1749like the one above until proper priorities are set for all interfaces.  On1750distributions that allow configuration via initscripts, multiple 'queue_id'1751arguments can be added to BONDING_OPTS to set all needed slave queues.1752 1753These queue id's can be used in conjunction with the tc utility to configure1754a multiqueue qdisc and filters to bias certain traffic to transmit on certain1755slave devices.  For instance, say we wanted, in the above configuration to1756force all traffic bound to 192.168.1.100 to use eth1 in the bond as its output1757device. The following commands would accomplish this::1758 1759	# tc qdisc add dev bond0 handle 1 root multiq1760 1761	# tc filter add dev bond0 protocol ip parent 1: prio 1 u32 match ip \1762		dst 192.168.1.100 action skbedit queue_mapping 21763 1764These commands tell the kernel to attach a multiqueue queue discipline to the1765bond0 interface and filter traffic enqueued to it, such that packets with a dst1766ip of 192.168.1.100 have their output queue mapping value overwritten to 2.1767This value is then passed into the driver, causing the normal output path1768selection policy to be overridden, selecting instead qid 2, which maps to eth1.1769 1770Note that qid values begin at 1.  Qid 0 is reserved to initiate to the driver1771that normal output policy selection should take place.  One benefit to simply1772leaving the qid for a slave to 0 is the multiqueue awareness in the bonding1773driver that is now present.  This awareness allows tc filters to be placed on1774slave devices as well as bond devices and the bonding driver will simply act as1775a pass-through for selecting output queues on the slave device rather than1776output port selection.1777 1778This feature first appeared in bonding driver version 3.7.0 and support for1779output slave selection was limited to round-robin and active-backup modes.1780 17813.7 Configuring LACP for 802.3ad mode in a more secure way1782----------------------------------------------------------1783 1784When using 802.3ad bonding mode, the Actor (host) and Partner (switch)1785exchange LACPDUs.  These LACPDUs cannot be sniffed, because they are1786destined to link local mac addresses (which switches/bridges are not1787supposed to forward).  However, most of the values are easily predictable1788or are simply the machine's MAC address (which is trivially known to all1789other hosts in the same L2).  This implies that other machines in the L21790domain can spoof LACPDU packets from other hosts to the switch and potentially1791cause mayhem by joining (from the point of view of the switch) another1792machine's aggregate, thus receiving a portion of that hosts incoming1793traffic and / or spoofing traffic from that machine themselves (potentially1794even successfully terminating some portion of flows). Though this is not1795a likely scenario, one could avoid this possibility by simply configuring1796few bonding parameters:1797 1798   (a) ad_actor_system : You can set a random mac-address that can be used for1799       these LACPDU exchanges. The value can not be either NULL or Multicast.1800       Also it's preferable to set the local-admin bit. Following shell code1801       generates a random mac-address as described above::1802 1803	      # sys_mac_addr=$(printf '%02x:%02x:%02x:%02x:%02x:%02x' \1804				       $(( (RANDOM & 0xFE) | 0x02 )) \1805				       $(( RANDOM & 0xFF )) \1806				       $(( RANDOM & 0xFF )) \1807				       $(( RANDOM & 0xFF )) \1808				       $(( RANDOM & 0xFF )) \1809				       $(( RANDOM & 0xFF )))1810	      # echo $sys_mac_addr > /sys/class/net/bond0/bonding/ad_actor_system1811 1812   (b) ad_actor_sys_prio : Randomize the system priority. The default value1813       is 65535, but system can take the value from 1 - 65535. Following shell1814       code generates random priority and sets it::1815 1816	    # sys_prio=$(( 1 + RANDOM + RANDOM ))1817	    # echo $sys_prio > /sys/class/net/bond0/bonding/ad_actor_sys_prio1818 1819   (c) ad_user_port_key : Use the user portion of the port-key. The default1820       keeps this empty. These are the upper 10 bits of the port-key and value1821       ranges from 0 - 1023. Following shell code generates these 10 bits and1822       sets it::1823 1824	    # usr_port_key=$(( RANDOM & 0x3FF ))1825	    # echo $usr_port_key > /sys/class/net/bond0/bonding/ad_user_port_key1826 1827 18284 Querying Bonding Configuration1829=================================1830 18314.1 Bonding Configuration1832-------------------------1833 1834Each bonding device has a read-only file residing in the1835/proc/net/bonding directory.  The file contents include information1836about the bonding configuration, options and state of each slave.1837 1838For example, the contents of /proc/net/bonding/bond0 after the1839driver is loaded with parameters of mode=0 and miimon=1000 is1840generally as follows::1841 1842	Ethernet Channel Bonding Driver: 2.6.1 (October 29, 2004)1843	Bonding Mode: load balancing (round-robin)1844	Currently Active Slave: eth01845	MII Status: up1846	MII Polling Interval (ms): 10001847	Up Delay (ms): 01848	Down Delay (ms): 01849 1850	Slave Interface: eth11851	MII Status: up1852	Link Failure Count: 11853 1854	Slave Interface: eth01855	MII Status: up1856	Link Failure Count: 11857 1858The precise format and contents will change depending upon the1859bonding configuration, state, and version of the bonding driver.1860 18614.2 Network configuration1862-------------------------1863 1864The network configuration can be inspected using the ifconfig1865command.  Bonding devices will have the MASTER flag set; Bonding slave1866devices will have the SLAVE flag set.  The ifconfig output does not1867contain information on which slaves are associated with which masters.1868 1869In the example below, the bond0 interface is the master1870(MASTER) while eth0 and eth1 are slaves (SLAVE). Notice all slaves of1871bond0 have the same MAC address (HWaddr) as bond0 for all modes except1872TLB and ALB that require a unique MAC address for each slave::1873 1874  # /sbin/ifconfig1875  bond0     Link encap:Ethernet  HWaddr 00:C0:F0:1F:37:B41876	    inet addr:XXX.XXX.XXX.YYY  Bcast:XXX.XXX.XXX.255  Mask:255.255.252.01877	    UP BROADCAST RUNNING MASTER MULTICAST  MTU:1500  Metric:11878	    RX packets:7224794 errors:0 dropped:0 overruns:0 frame:01879	    TX packets:3286647 errors:1 dropped:0 overruns:1 carrier:01880	    collisions:0 txqueuelen:01881 1882  eth0      Link encap:Ethernet  HWaddr 00:C0:F0:1F:37:B41883	    UP BROADCAST RUNNING SLAVE MULTICAST  MTU:1500  Metric:11884	    RX packets:3573025 errors:0 dropped:0 overruns:0 frame:01885	    TX packets:1643167 errors:1 dropped:0 overruns:1 carrier:01886	    collisions:0 txqueuelen:1001887	    Interrupt:10 Base address:0x10801888 1889  eth1      Link encap:Ethernet  HWaddr 00:C0:F0:1F:37:B41890	    UP BROADCAST RUNNING SLAVE MULTICAST  MTU:1500  Metric:11891	    RX packets:3651769 errors:0 dropped:0 overruns:0 frame:01892	    TX packets:1643480 errors:0 dropped:0 overruns:0 carrier:01893	    collisions:0 txqueuelen:1001894	    Interrupt:9 Base address:0x14001895 18965. Switch Configuration1897=======================1898 1899For this section, "switch" refers to whatever system the1900bonded devices are directly connected to (i.e., where the other end of1901the cable plugs into).  This may be an actual dedicated switch device,1902or it may be another regular system (e.g., another computer running1903Linux),1904 1905The active-backup, balance-tlb and balance-alb modes do not1906require any specific configuration of the switch.1907 1908The 802.3ad mode requires that the switch have the appropriate1909ports configured as an 802.3ad aggregation.  The precise method used1910to configure this varies from switch to switch, but, for example, a1911Cisco 3550 series switch requires that the appropriate ports first be1912grouped together in a single etherchannel instance, then that1913etherchannel is set to mode "lacp" to enable 802.3ad (instead of1914standard EtherChannel).1915 1916The balance-rr, balance-xor and broadcast modes generally1917require that the switch have the appropriate ports grouped together.1918The nomenclature for such a group differs between switches, it may be1919called an "etherchannel" (as in the Cisco example, above), a "trunk1920group" or some other similar variation.  For these modes, each switch1921will also have its own configuration options for the switch's transmit1922policy to the bond.  Typical choices include XOR of either the MAC or1923IP addresses.  The transmit policy of the two peers does not need to1924match.  For these three modes, the bonding mode really selects a1925transmit policy for an EtherChannel group; all three will interoperate1926with another EtherChannel group.1927 1928 19296. 802.1q VLAN Support1930======================1931 1932It is possible to configure VLAN devices over a bond interface1933using the 8021q driver.  However, only packets coming from the 8021q1934driver and passing through bonding will be tagged by default.  Self1935generated packets, for example, bonding's learning packets or ARP1936packets generated by either ALB mode or the ARP monitor mechanism, are1937tagged internally by bonding itself.  As a result, bonding must1938"learn" the VLAN IDs configured above it, and use those IDs to tag1939self generated packets.1940 1941For reasons of simplicity, and to support the use of adapters1942that can do VLAN hardware acceleration offloading, the bonding1943interface declares itself as fully hardware offloading capable, it gets1944the add_vid/kill_vid notifications to gather the necessary1945information, and it propagates those actions to the slaves.  In case1946of mixed adapter types, hardware accelerated tagged packets that1947should go through an adapter that is not offloading capable are1948"un-accelerated" by the bonding driver so the VLAN tag sits in the1949regular location.1950 1951VLAN interfaces *must* be added on top of a bonding interface1952only after enslaving at least one slave.  The bonding interface has a1953hardware address of 00:00:00:00:00:00 until the first slave is added.1954If the VLAN interface is created prior to the first enslavement, it1955would pick up the all-zeroes hardware address.  Once the first slave1956is attached to the bond, the bond device itself will pick up the1957slave's hardware address, which is then available for the VLAN device.1958 1959Also, be aware that a similar problem can occur if all slaves1960are released from a bond that still has one or more VLAN interfaces on1961top of it.  When a new slave is added, the bonding interface will1962obtain its hardware address from the first slave, which might not1963match the hardware address of the VLAN interfaces (which was1964ultimately copied from an earlier slave).1965 1966There are two methods to insure that the VLAN device operates1967with the correct hardware address if all slaves are removed from a1968bond interface:1969 19701. Remove all VLAN interfaces then recreate them1971 19722. Set the bonding interface's hardware address so that it1973matches the hardware address of the VLAN interfaces.1974 1975Note that changing a VLAN interface's HW address would set the1976underlying device -- i.e. the bonding interface -- to promiscuous1977mode, which might not be what you want.1978 1979 19807. Link Monitoring1981==================1982 1983The bonding driver at present supports two schemes for1984monitoring a slave device's link state: the ARP monitor and the MII1985monitor.1986 1987At the present time, due to implementation restrictions in the1988bonding driver itself, it is not possible to enable both ARP and MII1989monitoring simultaneously.1990 19917.1 ARP Monitor Operation1992-------------------------1993 1994The ARP monitor operates as its name suggests: it sends ARP1995queries to one or more designated peer systems on the network, and1996uses the response as an indication that the link is operating.  This1997gives some assurance that traffic is actually flowing to and from one1998or more peers on the local network.1999 20007.2 Configuring Multiple ARP Targets2001------------------------------------2002 2003While ARP monitoring can be done with just one target, it can2004be useful in a High Availability setup to have several targets to2005monitor.  In the case of just one target, the target itself may go2006down or have a problem making it unresponsive to ARP requests.  Having2007an additional target (or several) increases the reliability of the ARP2008monitoring.2009 2010Multiple ARP targets must be separated by commas as follows::2011 2012 # example options for ARP monitoring with three targets2013 alias bond0 bonding2014 options bond0 arp_interval=60 arp_ip_target=192.168.0.1,192.168.0.3,192.168.0.92015 2016For just a single target the options would resemble::2017 2018    # example options for ARP monitoring with one target2019    alias bond0 bonding2020    options bond0 arp_interval=60 arp_ip_target=192.168.0.1002021 2022 20237.3 MII Monitor Operation2024-------------------------2025 2026The MII monitor monitors only the carrier state of the local2027network interface.  It accomplishes this in one of three ways: by2028depending upon the device driver to maintain its carrier state, by2029querying the device's MII registers, or by making an ethtool query to2030the device.2031 2032If the use_carrier module parameter is 1 (the default value),2033then the MII monitor will rely on the driver for carrier state2034information (via the netif_carrier subsystem).  As explained in the2035use_carrier parameter information, above, if the MII monitor fails to2036detect carrier loss on the device (e.g., when the cable is physically2037disconnected), it may be that the driver does not support2038netif_carrier.2039 2040If use_carrier is 0, then the MII monitor will first query the2041device's (via ioctl) MII registers and check the link state.  If that2042request fails (not just that it returns carrier down), then the MII2043monitor will make an ethtool ETHTOOL_GLINK request to attempt to obtain2044the same information.  If both methods fail (i.e., the driver either2045does not support or had some error in processing both the MII register2046and ethtool requests), then the MII monitor will assume the link is2047up.2048 20498. Potential Sources of Trouble2050===============================2051 20528.1 Adventures in Routing2053-------------------------2054 2055When bonding is configured, it is important that the slave2056devices not have routes that supersede routes of the master (or,2057generally, not have routes at all).  For example, suppose the bonding2058device bond0 has two slaves, eth0 and eth1, and the routing table is2059as follows::2060 2061  Kernel IP routing table2062  Destination     Gateway         Genmask         Flags   MSS Window  irtt Iface2063  10.0.0.0        0.0.0.0         255.255.0.0     U        40 0          0 eth02064  10.0.0.0        0.0.0.0         255.255.0.0     U        40 0          0 eth12065  10.0.0.0        0.0.0.0         255.255.0.0     U        40 0          0 bond02066  127.0.0.0       0.0.0.0         255.0.0.0       U        40 0          0 lo2067 2068This routing configuration will likely still update the2069receive/transmit times in the driver (needed by the ARP monitor), but2070may bypass the bonding driver (because outgoing traffic to, in this2071case, another host on network 10 would use eth0 or eth1 before bond0).2072 2073The ARP monitor (and ARP itself) may become confused by this2074configuration, because ARP requests (generated by the ARP monitor)2075will be sent on one interface (bond0), but the corresponding reply2076will arrive on a different interface (eth0).  This reply looks to ARP2077as an unsolicited ARP reply (because ARP matches replies on an2078interface basis), and is discarded.  The MII monitor is not affected2079by the state of the routing table.2080 2081The solution here is simply to insure that slaves do not have2082routes of their own, and if for some reason they must, those routes do2083not supersede routes of their master.  This should generally be the2084case, but unusual configurations or errant manual or automatic static2085route additions may cause trouble.2086 20878.2 Ethernet Device Renaming2088----------------------------2089 2090On systems with network configuration scripts that do not2091associate physical devices directly with network interface names (so2092that the same physical device always has the same "ethX" name), it may2093be necessary to add some special logic to config files in2094/etc/modprobe.d/.2095 2096For example, given a modules.conf containing the following::2097 2098	alias bond0 bonding2099	options bond0 mode=some-mode miimon=502100	alias eth0 tg32101	alias eth1 tg32102	alias eth2 e10002103	alias eth3 e10002104 2105If neither eth0 and eth1 are slaves to bond0, then when the2106bond0 interface comes up, the devices may end up reordered.  This2107happens because bonding is loaded first, then its slave device's2108drivers are loaded next.  Since no other drivers have been loaded,2109when the e1000 driver loads, it will receive eth0 and eth1 for its2110devices, but the bonding configuration tries to enslave eth2 and eth32111(which may later be assigned to the tg3 devices).2112 2113Adding the following::2114 2115	add above bonding e1000 tg32116 2117causes modprobe to load e1000 then tg3, in that order, when2118bonding is loaded.  This command is fully documented in the2119modules.conf manual page.2120 2121On systems utilizing modprobe an equivalent problem can occur.2122In this case, the following can be added to config files in2123/etc/modprobe.d/ as::2124 2125	softdep bonding pre: tg3 e10002126 2127This will load tg3 and e1000 modules before loading the bonding one.2128Full documentation on this can be found in the modprobe.d and modprobe2129manual pages.2130 21318.3. Painfully Slow Or No Failed Link Detection By Miimon2132---------------------------------------------------------2133 2134By default, bonding enables the use_carrier option, which2135instructs bonding to trust the driver to maintain carrier state.2136 2137As discussed in the options section, above, some drivers do2138not support the netif_carrier_on/_off link state tracking system.2139With use_carrier enabled, bonding will always see these links as up,2140regardless of their actual state.2141 2142Additionally, other drivers do support netif_carrier, but do2143not maintain it in real time, e.g., only polling the link state at2144some fixed interval.  In this case, miimon will detect failures, but2145only after some long period of time has expired.  If it appears that2146miimon is very slow in detecting link failures, try specifying2147use_carrier=0 to see if that improves the failure detection time.  If2148it does, then it may be that the driver checks the carrier state at a2149fixed interval, but does not cache the MII register values (so the2150use_carrier=0 method of querying the registers directly works).  If2151use_carrier=0 does not improve the failover, then the driver may cache2152the registers, or the problem may be elsewhere.2153 2154Also, remember that miimon only checks for the device's2155carrier state.  It has no way to determine the state of devices on or2156beyond other ports of a switch, or if a switch is refusing to pass2157traffic while still maintaining carrier on.2158 21599. SNMP agents2160===============2161 2162If running SNMP agents, the bonding driver should be loaded2163before any network drivers participating in a bond.  This requirement2164is due to the interface index (ipAdEntIfIndex) being associated to2165the first interface found with a given IP address.  That is, there is2166only one ipAdEntIfIndex for each IP address.  For example, if eth0 and2167eth1 are slaves of bond0 and the driver for eth0 is loaded before the2168bonding driver, the interface for the IP address will be associated2169with the eth0 interface.  This configuration is shown below, the IP2170address 192.168.1.1 has an interface index of 2 which indexes to eth02171in the ifDescr table (ifDescr.2).2172 2173::2174 2175     interfaces.ifTable.ifEntry.ifDescr.1 = lo2176     interfaces.ifTable.ifEntry.ifDescr.2 = eth02177     interfaces.ifTable.ifEntry.ifDescr.3 = eth12178     interfaces.ifTable.ifEntry.ifDescr.4 = eth22179     interfaces.ifTable.ifEntry.ifDescr.5 = eth32180     interfaces.ifTable.ifEntry.ifDescr.6 = bond02181     ip.ipAddrTable.ipAddrEntry.ipAdEntIfIndex.10.10.10.10 = 52182     ip.ipAddrTable.ipAddrEntry.ipAdEntIfIndex.192.168.1.1 = 22183     ip.ipAddrTable.ipAddrEntry.ipAdEntIfIndex.10.74.20.94 = 42184     ip.ipAddrTable.ipAddrEntry.ipAdEntIfIndex.127.0.0.1 = 12185 2186This problem is avoided by loading the bonding driver before2187any network drivers participating in a bond.  Below is an example of2188loading the bonding driver first, the IP address 192.168.1.1 is2189correctly associated with ifDescr.2.2190 2191     interfaces.ifTable.ifEntry.ifDescr.1 = lo2192     interfaces.ifTable.ifEntry.ifDescr.2 = bond02193     interfaces.ifTable.ifEntry.ifDescr.3 = eth02194     interfaces.ifTable.ifEntry.ifDescr.4 = eth12195     interfaces.ifTable.ifEntry.ifDescr.5 = eth22196     interfaces.ifTable.ifEntry.ifDescr.6 = eth32197     ip.ipAddrTable.ipAddrEntry.ipAdEntIfIndex.10.10.10.10 = 62198     ip.ipAddrTable.ipAddrEntry.ipAdEntIfIndex.192.168.1.1 = 22199     ip.ipAddrTable.ipAddrEntry.ipAdEntIfIndex.10.74.20.94 = 52200     ip.ipAddrTable.ipAddrEntry.ipAdEntIfIndex.127.0.0.1 = 12201 2202While some distributions may not report the interface name in2203ifDescr, the association between the IP address and IfIndex remains2204and SNMP functions such as Interface_Scan_Next will report that2205association.2206 220710. Promiscuous mode2208====================2209 2210When running network monitoring tools, e.g., tcpdump, it is2211common to enable promiscuous mode on the device, so that all traffic2212is seen (instead of seeing only traffic destined for the local host).2213The bonding driver handles promiscuous mode changes to the bonding2214master device (e.g., bond0), and propagates the setting to the slave2215devices.2216 2217For the balance-rr, balance-xor, broadcast, and 802.3ad modes,2218the promiscuous mode setting is propagated to all slaves.2219 2220For the active-backup, balance-tlb and balance-alb modes, the2221promiscuous mode setting is propagated only to the active slave.2222 2223For balance-tlb mode, the active slave is the slave currently2224receiving inbound traffic.2225 2226For balance-alb mode, the active slave is the slave used as a2227"primary."  This slave is used for mode-specific control traffic, for2228sending to peers that are unassigned or if the load is unbalanced.2229 2230For the active-backup, balance-tlb and balance-alb modes, when2231the active slave changes (e.g., due to a link failure), the2232promiscuous setting will be propagated to the new active slave.2233 223411. Configuring Bonding for High Availability2235=============================================2236 2237High Availability refers to configurations that provide2238maximum network availability by having redundant or backup devices,2239links or switches between the host and the rest of the world.  The2240goal is to provide the maximum availability of network connectivity2241(i.e., the network always works), even though other configurations2242could provide higher throughput.2243 224411.1 High Availability in a Single Switch Topology2245--------------------------------------------------2246 2247If two hosts (or a host and a single switch) are directly2248connected via multiple physical links, then there is no availability2249penalty to optimizing for maximum bandwidth.  In this case, there is2250only one switch (or peer), so if it fails, there is no alternative2251access to fail over to.  Additionally, the bonding load balance modes2252support link monitoring of their members, so if individual links fail,2253the load will be rebalanced across the remaining devices.2254 2255See Section 12, "Configuring Bonding for Maximum Throughput"2256for information on configuring bonding with one peer device.2257 225811.2 High Availability in a Multiple Switch Topology2259----------------------------------------------------2260 2261With multiple switches, the configuration of bonding and the2262network changes dramatically.  In multiple switch topologies, there is2263a trade off between network availability and usable bandwidth.2264 2265Below is a sample network, configured to maximize the2266availability of the network::2267 2268		|                                     |2269		|port3                           port3|2270	  +-----+----+                          +-----+----+2271	  |          |port2       ISL      port2|          |2272	  | switch A +--------------------------+ switch B |2273	  |          |                          |          |2274	  +-----+----+                          +-----++---+2275		|port1                           port1|2276		|             +-------+               |2277		+-------------+ host1 +---------------+2278			 eth0 +-------+ eth12279 2280In this configuration, there is a link between the two2281switches (ISL, or inter switch link), and multiple ports connecting to2282the outside world ("port3" on each switch).  There is no technical2283reason that this could not be extended to a third switch.2284 228511.2.1 HA Bonding Mode Selection for Multiple Switch Topology2286-------------------------------------------------------------2287 2288In a topology such as the example above, the active-backup and2289broadcast modes are the only useful bonding modes when optimizing for2290availability; the other modes require all links to terminate on the2291same peer for them to behave rationally.2292 2293active-backup:2294	This is generally the preferred mode, particularly if2295	the switches have an ISL and play together well.  If the2296	network configuration is such that one switch is specifically2297	a backup switch (e.g., has lower capacity, higher cost, etc),2298	then the primary option can be used to insure that the2299	preferred link is always used when it is available.2300 2301broadcast:2302	This mode is really a special purpose mode, and is suitable2303	only for very specific needs.  For example, if the two2304	switches are not connected (no ISL), and the networks beyond2305	them are totally independent.  In this case, if it is2306	necessary for some specific one-way traffic to reach both2307	independent networks, then the broadcast mode may be suitable.2308 230911.2.2 HA Link Monitoring Selection for Multiple Switch Topology2310----------------------------------------------------------------2311 2312The choice of link monitoring ultimately depends upon your2313switch.  If the switch can reliably fail ports in response to other2314failures, then either the MII or ARP monitors should work.  For2315example, in the above example, if the "port3" link fails at the remote2316end, the MII monitor has no direct means to detect this.  The ARP2317monitor could be configured with a target at the remote end of port3,2318thus detecting that failure without switch support.2319 2320In general, however, in a multiple switch topology, the ARP2321monitor can provide a higher level of reliability in detecting end to2322end connectivity failures (which may be caused by the failure of any2323individual component to pass traffic for any reason).  Additionally,2324the ARP monitor should be configured with multiple targets (at least2325one for each switch in the network).  This will insure that,2326regardless of which switch is active, the ARP monitor has a suitable2327target to query.2328 2329Note, also, that of late many switches now support a functionality2330generally referred to as "trunk failover."  This is a feature of the2331switch that causes the link state of a particular switch port to be set2332down (or up) when the state of another switch port goes down (or up).2333Its purpose is to propagate link failures from logically "exterior" ports2334to the logically "interior" ports that bonding is able to monitor via2335miimon.  Availability and configuration for trunk failover varies by2336switch, but this can be a viable alternative to the ARP monitor when using2337suitable switches.2338 233912. Configuring Bonding for Maximum Throughput2340==============================================2341 234212.1 Maximizing Throughput in a Single Switch Topology2343------------------------------------------------------2344 2345In a single switch configuration, the best method to maximize2346throughput depends upon the application and network environment.  The2347various load balancing modes each have strengths and weaknesses in2348different environments, as detailed below.2349 2350For this discussion, we will break down the topologies into2351two categories.  Depending upon the destination of most traffic, we2352categorize them into either "gatewayed" or "local" configurations.2353 2354In a gatewayed configuration, the "switch" is acting primarily2355as a router, and the majority of traffic passes through this router to2356other networks.  An example would be the following::2357 2358 2359     +----------+                     +----------+2360     |          |eth0            port1|          | to other networks2361     | Host A   +---------------------+ router   +------------------->2362     |          +---------------------+          | Hosts B and C are out2363     |          |eth1            port2|          | here somewhere2364     +----------+                     +----------+2365 2366The router may be a dedicated router device, or another host2367acting as a gateway.  For our discussion, the important point is that2368the majority of traffic from Host A will pass through the router to2369some other network before reaching its final destination.2370 2371In a gatewayed network configuration, although Host A may2372communicate with many other systems, all of its traffic will be sent2373and received via one other peer on the local network, the router.2374 2375Note that the case of two systems connected directly via2376multiple physical links is, for purposes of configuring bonding, the2377same as a gatewayed configuration.  In that case, it happens that all2378traffic is destined for the "gateway" itself, not some other network2379beyond the gateway.2380 2381In a local configuration, the "switch" is acting primarily as2382a switch, and the majority of traffic passes through this switch to2383reach other stations on the same network.  An example would be the2384following::2385 2386    +----------+            +----------+       +--------+2387    |          |eth0   port1|          +-------+ Host B |2388    |  Host A  +------------+  switch  |port3  +--------+2389    |          +------------+          |                  +--------+2390    |          |eth1   port2|          +------------------+ Host C |2391    +----------+            +----------+port4             +--------+2392 2393 2394Again, the switch may be a dedicated switch device, or another2395host acting as a gateway.  For our discussion, the important point is2396that the majority of traffic from Host A is destined for other hosts2397on the same local network (Hosts B and C in the above example).2398 2399In summary, in a gatewayed configuration, traffic to and from2400the bonded device will be to the same MAC level peer on the network2401(the gateway itself, i.e., the router), regardless of its final2402destination.  In a local configuration, traffic flows directly to and2403from the final destinations, thus, each destination (Host B, Host C)2404will be addressed directly by their individual MAC addresses.2405 2406This distinction between a gatewayed and a local network2407configuration is important because many of the load balancing modes2408available use the MAC addresses of the local network source and2409destination to make load balancing decisions.  The behavior of each2410mode is described below.2411 2412 241312.1.1 MT Bonding Mode Selection for Single Switch Topology2414-----------------------------------------------------------2415 2416This configuration is the easiest to set up and to understand,2417although you will have to decide which bonding mode best suits your2418needs.  The trade offs for each mode are detailed below:2419 2420balance-rr:2421	This mode is the only mode that will permit a single2422	TCP/IP connection to stripe traffic across multiple2423	interfaces. It is therefore the only mode that will allow a2424	single TCP/IP stream to utilize more than one interface's2425	worth of throughput.  This comes at a cost, however: the2426	striping generally results in peer systems receiving packets out2427	of order, causing TCP/IP's congestion control system to kick2428	in, often by retransmitting segments.2429 2430	It is possible to adjust TCP/IP's congestion limits by2431	altering the net.ipv4.tcp_reordering sysctl parameter.  The2432	usual default value is 3. But keep in mind TCP stack is able2433	to automatically increase this when it detects reorders.2434 2435	Note that the fraction of packets that will be delivered out of2436	order is highly variable, and is unlikely to be zero.  The level2437	of reordering depends upon a variety of factors, including the2438	networking interfaces, the switch, and the topology of the2439	configuration.  Speaking in general terms, higher speed network2440	cards produce more reordering (due to factors such as packet2441	coalescing), and a "many to many" topology will reorder at a2442	higher rate than a "many slow to one fast" configuration.2443 2444	Many switches do not support any modes that stripe traffic2445	(instead choosing a port based upon IP or MAC level addresses);2446	for those devices, traffic for a particular connection flowing2447	through the switch to a balance-rr bond will not utilize greater2448	than one interface's worth of bandwidth.2449 2450	If you are utilizing protocols other than TCP/IP, UDP for2451	example, and your application can tolerate out of order2452	delivery, then this mode can allow for single stream datagram2453	performance that scales near linearly as interfaces are added2454	to the bond.2455 2456	This mode requires the switch to have the appropriate ports2457	configured for "etherchannel" or "trunking."2458 2459active-backup:2460	There is not much advantage in this network topology to2461	the active-backup mode, as the inactive backup devices are all2462	connected to the same peer as the primary.  In this case, a2463	load balancing mode (with link monitoring) will provide the2464	same level of network availability, but with increased2465	available bandwidth.  On the plus side, active-backup mode2466	does not require any configuration of the switch, so it may2467	have value if the hardware available does not support any of2468	the load balance modes.2469 2470balance-xor:2471	This mode will limit traffic such that packets destined2472	for specific peers will always be sent over the same2473	interface.  Since the destination is determined by the MAC2474	addresses involved, this mode works best in a "local" network2475	configuration (as described above), with destinations all on2476	the same local network.  This mode is likely to be suboptimal2477	if all your traffic is passed through a single router (i.e., a2478	"gatewayed" network configuration, as described above).2479 2480	As with balance-rr, the switch ports need to be configured for2481	"etherchannel" or "trunking."2482 2483broadcast:2484	Like active-backup, there is not much advantage to this2485	mode in this type of network topology.2486 2487802.3ad:2488	This mode can be a good choice for this type of network2489	topology.  The 802.3ad mode is an IEEE standard, so all peers2490	that implement 802.3ad should interoperate well.  The 802.3ad2491	protocol includes automatic configuration of the aggregates,2492	so minimal manual configuration of the switch is needed2493	(typically only to designate that some set of devices is2494	available for 802.3ad).  The 802.3ad standard also mandates2495	that frames be delivered in order (within certain limits), so2496	in general single connections will not see misordering of2497	packets.  The 802.3ad mode does have some drawbacks: the2498	standard mandates that all devices in the aggregate operate at2499	the same speed and duplex.  Also, as with all bonding load2500	balance modes other than balance-rr, no single connection will2501	be able to utilize more than a single interface's worth of2502	bandwidth.2503 2504	Additionally, the linux bonding 802.3ad implementation2505	distributes traffic by peer (using an XOR of MAC addresses2506	and packet type ID), so in a "gatewayed" configuration, all2507	outgoing traffic will generally use the same device.  Incoming2508	traffic may also end up on a single device, but that is2509	dependent upon the balancing policy of the peer's 802.3ad2510	implementation.  In a "local" configuration, traffic will be2511	distributed across the devices in the bond.2512 2513	Finally, the 802.3ad mode mandates the use of the MII monitor,2514	therefore, the ARP monitor is not available in this mode.2515 2516balance-tlb:2517	The balance-tlb mode balances outgoing traffic by peer.2518	Since the balancing is done according to MAC address, in a2519	"gatewayed" configuration (as described above), this mode will2520	send all traffic across a single device.  However, in a2521	"local" network configuration, this mode balances multiple2522	local network peers across devices in a vaguely intelligent2523	manner (not a simple XOR as in balance-xor or 802.3ad mode),2524	so that mathematically unlucky MAC addresses (i.e., ones that2525	XOR to the same value) will not all "bunch up" on a single2526	interface.2527 2528	Unlike 802.3ad, interfaces may be of differing speeds, and no2529	special switch configuration is required.  On the down side,2530	in this mode all incoming traffic arrives over a single2531	interface, this mode requires certain ethtool support in the2532	network device driver of the slave interfaces, and the ARP2533	monitor is not available.2534 2535balance-alb:2536	This mode is everything that balance-tlb is, and more.2537	It has all of the features (and restrictions) of balance-tlb,2538	and will also balance incoming traffic from local network2539	peers (as described in the Bonding Module Options section,2540	above).2541 2542	The only additional down side to this mode is that the network2543	device driver must support changing the hardware address while2544	the device is open.2545 254612.1.2 MT Link Monitoring for Single Switch Topology2547----------------------------------------------------2548 2549The choice of link monitoring may largely depend upon which2550mode you choose to use.  The more advanced load balancing modes do not2551support the use of the ARP monitor, and are thus restricted to using2552the MII monitor (which does not provide as high a level of end to end2553assurance as the ARP monitor).2554 255512.2 Maximum Throughput in a Multiple Switch Topology2556-----------------------------------------------------2557 2558Multiple switches may be utilized to optimize for throughput2559when they are configured in parallel as part of an isolated network2560between two or more systems, for example::2561 2562		       +-----------+2563		       |  Host A   |2564		       +-+---+---+-+2565			 |   |   |2566		+--------+   |   +---------+2567		|            |             |2568	 +------+---+  +-----+----+  +-----+----+2569	 | Switch A |  | Switch B |  | Switch C |2570	 +------+---+  +-----+----+  +-----+----+2571		|            |             |2572		+--------+   |   +---------+2573			 |   |   |2574		       +-+---+---+-+2575		       |  Host B   |2576		       +-----------+2577 2578In this configuration, the switches are isolated from one2579another.  One reason to employ a topology such as this is for an2580isolated network with many hosts (a cluster configured for high2581performance, for example), using multiple smaller switches can be more2582cost effective than a single larger switch, e.g., on a network with 242583hosts, three 24 port switches can be significantly less expensive than2584a single 72 port switch.2585 2586If access beyond the network is required, an individual host2587can be equipped with an additional network device connected to an2588external network; this host then additionally acts as a gateway.2589 259012.2.1 MT Bonding Mode Selection for Multiple Switch Topology2591-------------------------------------------------------------2592 2593In actual practice, the bonding mode typically employed in2594configurations of this type is balance-rr.  Historically, in this2595network configuration, the usual caveats about out of order packet2596delivery are mitigated by the use of network adapters that do not do2597any kind of packet coalescing (via the use of NAPI, or because the2598device itself does not generate interrupts until some number of2599packets has arrived).  When employed in this fashion, the balance-rr2600mode allows individual connections between two hosts to effectively2601utilize greater than one interface's bandwidth.2602 260312.2.2 MT Link Monitoring for Multiple Switch Topology2604------------------------------------------------------2605 2606Again, in actual practice, the MII monitor is most often used2607in this configuration, as performance is given preference over2608availability.  The ARP monitor will function in this topology, but its2609advantages over the MII monitor are mitigated by the volume of probes2610needed as the number of systems involved grows (remember that each2611host in the network is configured with bonding).2612 261313. Switch Behavior Issues2614==========================2615 261613.1 Link Establishment and Failover Delays2617-------------------------------------------2618 2619Some switches exhibit undesirable behavior with regard to the2620timing of link up and down reporting by the switch.2621 2622First, when a link comes up, some switches may indicate that2623the link is up (carrier available), but not pass traffic over the2624interface for some period of time.  This delay is typically due to2625some type of autonegotiation or routing protocol, but may also occur2626during switch initialization (e.g., during recovery after a switch2627failure).  If you find this to be a problem, specify an appropriate2628value to the updelay bonding module option to delay the use of the2629relevant interface(s).2630 2631Second, some switches may "bounce" the link state one or more2632times while a link is changing state.  This occurs most commonly while2633the switch is initializing.  Again, an appropriate updelay value may2634help.2635 2636Note that when a bonding interface has no active links, the2637driver will immediately reuse the first link that goes up, even if the2638updelay parameter has been specified (the updelay is ignored in this2639case).  If there are slave interfaces waiting for the updelay timeout2640to expire, the interface that first went into that state will be2641immediately reused.  This reduces down time of the network if the2642value of updelay has been overestimated, and since this occurs only in2643cases with no connectivity, there is no additional penalty for2644ignoring the updelay.2645 2646In addition to the concerns about switch timings, if your2647switches take a long time to go into backup mode, it may be desirable2648to not activate a backup interface immediately after a link goes down.2649Failover may be delayed via the downdelay bonding module option.2650 265113.2 Duplicated Incoming Packets2652--------------------------------2653 2654NOTE: Starting with version 3.0.2, the bonding driver has logic to2655suppress duplicate packets, which should largely eliminate this problem.2656The following description is kept for reference.2657 2658It is not uncommon to observe a short burst of duplicated2659traffic when the bonding device is first used, or after it has been2660idle for some period of time.  This is most easily observed by issuing2661a "ping" to some other host on the network, and noticing that the2662output from ping flags duplicates (typically one per slave).2663 2664For example, on a bond in active-backup mode with five slaves2665all connected to one switch, the output may appear as follows::2666 2667	# ping -n 10.0.4.22668	PING 10.0.4.2 (10.0.4.2) from 10.0.3.10 : 56(84) bytes of data.2669	64 bytes from 10.0.4.2: icmp_seq=1 ttl=64 time=13.7 ms2670	64 bytes from 10.0.4.2: icmp_seq=1 ttl=64 time=13.8 ms (DUP!)2671	64 bytes from 10.0.4.2: icmp_seq=1 ttl=64 time=13.8 ms (DUP!)2672	64 bytes from 10.0.4.2: icmp_seq=1 ttl=64 time=13.8 ms (DUP!)2673	64 bytes from 10.0.4.2: icmp_seq=1 ttl=64 time=13.8 ms (DUP!)2674	64 bytes from 10.0.4.2: icmp_seq=2 ttl=64 time=0.216 ms2675	64 bytes from 10.0.4.2: icmp_seq=3 ttl=64 time=0.267 ms2676	64 bytes from 10.0.4.2: icmp_seq=4 ttl=64 time=0.222 ms2677 2678This is not due to an error in the bonding driver, rather, it2679is a side effect of how many switches update their MAC forwarding2680tables.  Initially, the switch does not associate the MAC address in2681the packet with a particular switch port, and so it may send the2682traffic to all ports until its MAC forwarding table is updated.  Since2683the interfaces attached to the bond may occupy multiple ports on a2684single switch, when the switch (temporarily) floods the traffic to all2685ports, the bond device receives multiple copies of the same packet2686(one per slave device).2687 2688The duplicated packet behavior is switch dependent, some2689switches exhibit this, and some do not.  On switches that display this2690behavior, it can be induced by clearing the MAC forwarding table (on2691most Cisco switches, the privileged command "clear mac address-table2692dynamic" will accomplish this).2693 269414. Hardware Specific Considerations2695====================================2696 2697This section contains additional information for configuring2698bonding on specific hardware platforms, or for interfacing bonding2699with particular switches or other devices.2700 270114.1 IBM BladeCenter2702--------------------2703 2704This applies to the JS20 and similar systems.2705 2706On the JS20 blades, the bonding driver supports only2707balance-rr, active-backup, balance-tlb and balance-alb modes.  This is2708largely due to the network topology inside the BladeCenter, detailed2709below.2710 2711JS20 network adapter information2712--------------------------------2713 2714All JS20s come with two Broadcom Gigabit Ethernet ports2715integrated on the planar (that's "motherboard" in IBM-speak).  In the2716BladeCenter chassis, the eth0 port of all JS20 blades is hard wired to2717I/O Module #1; similarly, all eth1 ports are wired to I/O Module #2.2718An add-on Broadcom daughter card can be installed on a JS20 to provide2719two more Gigabit Ethernet ports.  These ports, eth2 and eth3, are2720wired to I/O Modules 3 and 4, respectively.2721 2722Each I/O Module may contain either a switch or a passthrough2723module (which allows ports to be directly connected to an external2724switch).  Some bonding modes require a specific BladeCenter internal2725network topology in order to function; these are detailed below.2726 2727Additional BladeCenter-specific networking information can be2728found in two IBM Redbooks (www.ibm.com/redbooks):2729 2730- "IBM eServer BladeCenter Networking Options"2731- "IBM eServer BladeCenter Layer 2-7 Network Switching"2732 2733BladeCenter networking configuration2734------------------------------------2735 2736Because a BladeCenter can be configured in a very large number2737of ways, this discussion will be confined to describing basic2738configurations.2739 2740Normally, Ethernet Switch Modules (ESMs) are used in I/O2741modules 1 and 2.  In this configuration, the eth0 and eth1 ports of a2742JS20 will be connected to different internal switches (in the2743respective I/O modules).2744 2745A passthrough module (OPM or CPM, optical or copper,2746passthrough module) connects the I/O module directly to an external2747switch.  By using PMs in I/O module #1 and #2, the eth0 and eth12748interfaces of a JS20 can be redirected to the outside world and2749connected to a common external switch.2750 2751Depending upon the mix of ESMs and PMs, the network will2752appear to bonding as either a single switch topology (all PMs) or as a2753multiple switch topology (one or more ESMs, zero or more PMs).  It is2754also possible to connect ESMs together, resulting in a configuration2755much like the example in "High Availability in a Multiple Switch2756Topology," above.2757 2758Requirements for specific modes2759-------------------------------2760 2761The balance-rr mode requires the use of passthrough modules2762for devices in the bond, all connected to an common external switch.2763That switch must be configured for "etherchannel" or "trunking" on the2764appropriate ports, as is usual for balance-rr.2765 2766The balance-alb and balance-tlb modes will function with2767either switch modules or passthrough modules (or a mix).  The only2768specific requirement for these modes is that all network interfaces2769must be able to reach all destinations for traffic sent over the2770bonding device (i.e., the network must converge at some point outside2771the BladeCenter).2772 2773The active-backup mode has no additional requirements.2774 2775Link monitoring issues2776----------------------2777 2778When an Ethernet Switch Module is in place, only the ARP2779monitor will reliably detect link loss to an external switch.  This is2780nothing unusual, but examination of the BladeCenter cabinet would2781suggest that the "external" network ports are the ethernet ports for2782the system, when it fact there is a switch between these "external"2783ports and the devices on the JS20 system itself.  The MII monitor is2784only able to detect link failures between the ESM and the JS20 system.2785 2786When a passthrough module is in place, the MII monitor does2787detect failures to the "external" port, which is then directly2788connected to the JS20 system.2789 2790Other concerns2791--------------2792 2793The Serial Over LAN (SoL) link is established over the primary2794ethernet (eth0) only, therefore, any loss of link to eth0 will result2795in losing your SoL connection.  It will not fail over with other2796network traffic, as the SoL system is beyond the control of the2797bonding driver.2798 2799It may be desirable to disable spanning tree on the switch2800(either the internal Ethernet Switch Module, or an external switch) to2801avoid fail-over delay issues when using bonding.2802 2803 280415. Frequently Asked Questions2805==============================2806 28071.  Is it SMP safe?2808-------------------2809 2810Yes. The old 2.0.xx channel bonding patch was not SMP safe.2811The new driver was designed to be SMP safe from the start.2812 28132.  What type of cards will work with it?2814-----------------------------------------2815 2816Any Ethernet type cards (you can even mix cards - a Intel2817EtherExpress PRO/100 and a 3com 3c905b, for example).  For most modes,2818devices need not be of the same speed.2819 2820Starting with version 3.2.1, bonding also supports Infiniband2821slaves in active-backup mode.2822 28233.  How many bonding devices can I have?2824----------------------------------------2825 2826There is no limit.2827 28284.  How many slaves can a bonding device have?2829----------------------------------------------2830 2831This is limited only by the number of network interfaces Linux2832supports and/or the number of network cards you can place in your2833system.2834 28355.  What happens when a slave link dies?2836----------------------------------------2837 2838If link monitoring is enabled, then the failing device will be2839disabled.  The active-backup mode will fail over to a backup link, and2840other modes will ignore the failed link.  The link will continue to be2841monitored, and should it recover, it will rejoin the bond (in whatever2842manner is appropriate for the mode). See the sections on High2843Availability and the documentation for each mode for additional2844information.2845 2846Link monitoring can be enabled via either the miimon or2847arp_interval parameters (described in the module parameters section,2848above).  In general, miimon monitors the carrier state as sensed by2849the underlying network device, and the arp monitor (arp_interval)2850monitors connectivity to another host on the local network.2851 2852If no link monitoring is configured, the bonding driver will2853be unable to detect link failures, and will assume that all links are2854always available.  This will likely result in lost packets, and a2855resulting degradation of performance.  The precise performance loss2856depends upon the bonding mode and network configuration.2857 28586.  Can bonding be used for High Availability?2859----------------------------------------------2860 2861Yes.  See the section on High Availability for details.2862 28637.  Which switches/systems does it work with?2864---------------------------------------------2865 2866The full answer to this depends upon the desired mode.2867 2868In the basic balance modes (balance-rr and balance-xor), it2869works with any system that supports etherchannel (also called2870trunking).  Most managed switches currently available have such2871support, and many unmanaged switches as well.2872 2873The advanced balance modes (balance-tlb and balance-alb) do2874not have special switch requirements, but do need device drivers that2875support specific features (described in the appropriate section under2876module parameters, above).2877 2878In 802.3ad mode, it works with systems that support IEEE2879802.3ad Dynamic Link Aggregation.  Most managed and many unmanaged2880switches currently available support 802.3ad.2881 2882The active-backup mode should work with any Layer-II switch.2883 28848.  Where does a bonding device get its MAC address from?2885---------------------------------------------------------2886 2887When using slave devices that have fixed MAC addresses, or when2888the fail_over_mac option is enabled, the bonding device's MAC address is2889the MAC address of the active slave.2890 2891For other configurations, if not explicitly configured (with2892ifconfig or ip link), the MAC address of the bonding device is taken from2893its first slave device.  This MAC address is then passed to all following2894slaves and remains persistent (even if the first slave is removed) until2895the bonding device is brought down or reconfigured.2896 2897If you wish to change the MAC address, you can set it with2898ifconfig or ip link::2899 2900	# ifconfig bond0 hw ether 00:11:22:33:44:552901 2902	# ip link set bond0 address 66:77:88:99:aa:bb2903 2904The MAC address can be also changed by bringing down/up the2905device and then changing its slaves (or their order)::2906 2907	# ifconfig bond0 down ; modprobe -r bonding2908	# ifconfig bond0 .... up2909	# ifenslave bond0 eth...2910 2911This method will automatically take the address from the next2912slave that is added.2913 2914To restore your slaves' MAC addresses, you need to detach them2915from the bond (``ifenslave -d bond0 eth0``). The bonding driver will2916then restore the MAC addresses that the slaves had before they were2917enslaved.2918 291916. Resources and Links2920=======================2921 2922The latest version of the bonding driver can be found in the latest2923version of the linux kernel, found on http://kernel.org2924 2925The latest version of this document can be found in the latest kernel2926source (named Documentation/networking/bonding.rst).2927 2928Discussions regarding the development of the bonding driver take place2929on the main Linux network mailing list, hosted at vger.kernel.org. The list2930address is:2931 2932netdev@vger.kernel.org2933 2934The administrative interface (to subscribe or unsubscribe) can2935be found at:2936 2937http://vger.kernel.org/vger-lists.html#netdev2938