Device Mapper Multipath（DM-Multipath）可以将服务器节点和存储阵列之间的多条I/O链路配置为一个单独的设备。这些I/O链路是由不同的线缆、交换机、控制器组成的SAN物理链路。Multipath将这些链路聚合在一起，生成一个单独的新的设备。
1.DM-Multipath概览：
（1）数据冗余
DM-Multipath可以实现在active/passive模式下的灾难转移。在active/passive模式下，只有一半的链路在工作，如果链路上的某一部分（线缆、交换机、控制器）出现故障，DM-Multipath就会切换到另一半链路上。
（2）提高性能
DM-Multipath也可以配置为active/active模式，从而I/O任务以round-robin的方式分布到所有的链路上去。通过配置，DM-Multipath还可以检测链路上的负载情况，动态地进行负载均衡。
DM-Multipath的存储阵列的支持情况可以查看multipath.conf.default。对于支持列表中没有的存储，可以在multipath的配置文件multipath.conf中自行添加。
2.DM-Multipath的组成：
（1）dm-multipath内核模块实现I/O重定向、支持failover以及链路的聚合
（2）mutilpath命令查看和配置multipath设备。它一般由/etc/rc.sysinit启动，也会在系统发现新的块设备时由udev启动，或者由initramfs在系统启动时执行。
（3）multipathd后台进程监控链路。当链路失效或者恢复时，它会发起链路组的切换。它实现了对multipath设备的动态修改。不过multipath.conf修改后，它需要重新启动。
（4）kpartx命令根据设备上分区表创建device mapper设备。当需要在DM-MP设备上使用基于DOS的分区时，这个命令是必须。
不使用DM-Multipath，每一条从服务器到存储的链路都被系统识别为一个单独的设备。DM-Multipath可以在这些底层的设备之上创建一个单一的multipath设备，实现对这些链路的组织和管理。
3.Multipath设备的标志符
每一个multipath设备都有一个WWID（World Wide Identifier），这个id是全球唯一而且不可更改的。默认情况下，multipath设备的名称被设置为它的WWID。不过也可以在配置文件中使用_friendly_names选项，为设备取一个别名，别名为mpath[n]。
例如，一台服务器节点有2块HBA卡，通过一台没有划分zone的交换机，连接到2个磁盘阵列控制器，服务器系统中会发现4个设备：/dev/sda、/dev/sdb、/dev/sdc、/dev/sdd。DM-Multipath会按照配置文件在这些底层设备上创建一个拥有一个唯一WWID号的multipath设备。如果配置文件中_friendly_names选项被设置为yes，则这个multipath设备会被命名为mpath[n]。
当新的设备被DM-Multipath接管之后，新的设备文件会在/dev/目录下3个不同的地方出现：/dev/mapper/mpath[n]、/dev/mpath/mpath[n]、/dev/dm-[n]。
（1）/dev/mapper/目录下的文件，是早在系统启动的过程中就创建了。访问multipath设备时就使用这些文件，例如创建lvm；
（2）/dev/mpath/目录下的文件，是为了可以方便的在同一个目录下查看所有的multipath设备。这个文件由udev创建。如果系统在启动过程中需要访问multipath设备，不要使用这些文件。不用在这些设备文件上建立lvm；
（3）/dev/dm-[n]只为了内部使用目的，永远不要对这些文件进行操作。
4.统一multipath设备的命名
当配置文件中_friendly_names被设置为yes，在该服务器节点上这个设备名是唯一而且确定的，但是不能保证在使用这些链路的其它服务器节点上的multipath设备的名称能够相互保持一致。如果只是建立lvm，那么这个问题不会有什么影响。但是如果希望不同服务器节点上的multipath设备的名称能够统一，必须使用下面其中一种方法：
（1）在配置文件中的multipaths段使用alias选项为设备设置别名，并在不同的服务器上保持一致；
（2）如果希望不同服务器上multipath设备的user-friendly名称保持一致，首先在一台服务器上建立所有multipath设备，然后把bindings文件拷贝到所有其它系统统一命名的服务器。binds文件的位置是/var/lib/multipath/bindings。在配置文件可以使用bindings_file参数修改bindings文件的位置。
5.Multipath设备上建立lv
创建multipath设备之后，可以像使用物理设备文件一样在multipath设备上建立pv。例如，假定multipath设备为/dev/mapper/mpath0，使用
pvscreate /dev/mapper/mpath0
即可将mpath0建立为物理卷。同样可以继续建立卷族和逻辑卷。
当在配置为active/passive模式的multipath设备上建立逻辑卷时，需要在lvm的配置文件lvm.conf中增加过滤器，将multipath设备下层的设备添加到过滤列表中。这时因为DM-Multipath会自动切换数据链路，当遇到failover和failback的情况时，如果下层的设备没有在配置文件中过滤，lvm会扫描这些passive状态下的数据链路。Passive状态的链路改变到active状态需要执行一些命令，所以lvm就会在这个时候报错。
为了过滤所有的SCSI设备，在lvm.conf中的devices段，添加下面的配置：
filter = [ "r/disk/", "r/sd.*/", "a/.*/"
6.部署DM-Multipath
6.1开始部署
（1）编辑/etc/multipath.conf，注释掉下面几行：
devnode_blacklist {
devnode "*"
}
（2）multipath的默认配置已经集成在系统之中，不需要在/etc/multipath.conf中重新配置。
path_grouping_policy的默认值为failover。在原始配置中default段设置了multipath设备的默认名称是mpath[n]的形式，如果没有这一段配置（即_friendly_names=yes），设备的默认名称是它的WWID号。
（3）保存配置文件并退出。
（4）执行下列命令：
modprobe dm-multipath
sevice multipathd start
multipath -v2
注：multipath -v2会打印出已经聚合的数据链路。
（5）使用 chkconfig multipathd on，让multipath服务开机自动启动。
6.2排除本地scsi磁盘
很多系统都安装有本地scsi磁盘，DM-Multipath是不建议在这些磁盘上使用的。可以按照下面的步骤取消对本地scsi磁盘的映射。
（1）使用 multipath -v2 确认本地磁盘的信息。如下面的示例（sda为本地scsi磁盘）：
[root@localhost ~]# multipath -v2
create: SIBM-ESXSST336732LC____F3ET0EP0Q000072428BX1
[size=33 GB][features="0"][hwhandler="0"]
\_ round-robin 0
\_ 0:0:0:0 sda 8:0
device-mapper ioctl cmd 9 failed: Invalid argument
device-mapper ioctl cmd 14 failed: No such device or address
create: 3600a0b80001327d80000006d43621677
[size=12 GB][features="0"][hwhandler="0"]
\_ round-robin 0
\_ 2:0:0:0 sdb 8:16
\_ 3:0:0:0 sdf 8:80
create: 3600 a0b80001327510000009a436215ec
[size=12 GB][features="0"][hwhandler="0"]
\_ round-robin 0
\_ 2:0:0:1 sdc 8:32
\_ 3:0:0:1 sdg 8:96
create: 3600a0b80001327d800000070436216b3
[size=12 GB][features="0"][hwhandler="0"]
\_ round-robin 0
\_ 2:0:0:2 sdd 8:48
\_ 3:0:0:2 sdh 8:112
create: 3600a0b80001327510000009b4362163e
[size=12 GB][features="0"][hwhandler="0"]
\_ round-robin 0
\_ 2:0:0:3 sde 8:64
\_ 3:0:0:3 sdi 8:128
（2）为了防止DM-Multipath对/dev/sda做映射，编辑/etc/multipath.conf中的devnode_blacklist段。可以使用devnode的方式过滤sda，不过系统中sda的命名不一定是固定的，所以最好使用wwid的方式。从上面的输出中可以看到/dev/sda的wwid为“SIBM-ESXSST336732LC____F3ET0EP0Q000072428BX1”，在配置文件添加：
devnode_blacklist{
wwid SIBM-ESXSST336732LC____F3ET0EP0Q000072428BX1
}
（3）执行下面的命令使配置生效，并重新打印multipath设备列表。
multipath -F
multipath -v2
6.3在DM-Multipath中增加新的设备类型
DM-Multipath支持大部分的存储阵列。默认的配置，可以查看multipath.conf.default文件。
如果希望添加默认不支持的存储设备，可以在/etc/multipath.conf中添加相应的信息。例如在配置为鉴中添加HP Open-V：
devices {
device {
vendor "HP"
product "OPEN-V"
getuid_callout "/sbin/scsi_id -g -u -p0x80 -s /block/%n"
}
}
7.DM-Multipath配置文件
7.1概览
DM-Multipath的配置文件分为以下几个部分：
devnode_blacklist
不使用DM-Multipath的设备列表。默认的情况下，所有的设备都在列表中。启用DM-Multipath的时候一般会将devnote_blacklist段注释掉；
defaults
DM-Multipath的默认通用配置；
multipaths
单独配置每一个multipath设备的属性。这些配置会覆盖在defaults段和devices段的配置；
devices
单独配置每一个存储控制器。这些配置会覆盖defaults段的配置。如果使用的存储控制器不被DM-Multipath支持，那么就需要为这种类型的控制器加一个devices subsection。
DM-Multipath确定multipath设备的属性时，会首先使用multipaths段的内容，然后是devices段，最好读取defaults段。
7.2配置blacklist
devnode_blacklist指定了系统在配置multipath设备时不使用的设备，默认情况所有的设备都在这个列表中。注释掉默认的一行之后，可以在列表中加入某一种类型的设备或者某一特定的设备。禁用设备有2种方法：
（1）使用wwid：
可以使用wwid指定特定的设备，如：
blacklist {
wwid 26353900f02796769
}
（2）使用设备名：
如：
devnode_blacklist {
devnode "^sd[a-z]"
}
这一段配置会禁用所有的SCSI磁盘设备。虽然可以使用这种方法禁用单一特定的设备，但是并不建议这样做。因为除非是使用了udev固定了设备的设备名，则设备的名称在每次重启之后是有可能发生变化的。
因为一些设备并不支持DM-Multipath，所以下面列出的设备是默认禁用的：
blacklist {
devnode "^(ram|raw|loop|fd|md|dm-|sr|scd|st)[0-9]*"
devnode "^hd[a-z]"
devnode "^cciss!c[0-9]d[0-9]*"
}
device-mapper-multipathing，需要完全的升级操作系统。早期的红帽企业Linux 4
不包含这个功能。编辑文件/etc/multipath.conf。在文件头部注释掉如下内容：
devnode_blacklist {
devnode "*"
}
如下例：
# devnode_blacklist {
# devnode "*"
# }
取消文件 /etc/multipath.conf中这段的注释，这段内容能让 device-mapper multipathing 不用扫描所有的设备。 etc/multipath.conf file:
defaults {
multipath_tool "/sbin/multipath -v0"
udev_dir /dev
polling_interval 10
default_selector "round-robin 0"
default_path_grouping_policy multibus
default_getuid_callout "/sbin/scsi_id -g -u -s /block/%n"
default_prio_callout "/bin/true"
default_features "0"
rr_wmin_io 100
failback immediate
}
devnode_blacklist {
wwid 26353900f02796769
devnode "^(ram|raw|loop|fd|md|dm-|sr|scd|st)[0-9]*"
devnode "^hd[a-z][[0-9]*]"
devnode "^cciss!c[0-9]d[0-9]*[p[0-9]*]"
}
这段设置了默认的 device-mapper 的动作，并且去出了通常不会有多路径的设备，如IDE硬盘和软盘。默认的hd*设备的黑名单存在这一个排印错误。需要修改。
devnode "^hd[a-z][[0-9]*]"
把上行修改为如下:
devnode "^hd[a-z][0-9]*"
为了实现简单的failover功能，下例中，defaults 组中的默认的 default_path_grouping_policy 选项被设置成为 failover。
defaults {
multipath_tool "/sbin/multipath -v0"
udev_dir /dev
polling_interval 10
default_selector "round-robin 0"
default_path_grouping_policy failover
default_getuid_callout "/sbin/scsi_id -g -u -s /block/%n"
default_prio_callout "/bin/true"
default_features "0"
rr_wmin_io 100
failback immediate
}
退出编辑并保存设置，执行如下命令
modprobe dm-multipath
modprobe dm-round-robin
service multipathd start
multipath -v2
命令 multipath -v2 能显示多路径，从而能知道那些设备工作在多路径下。如果没有所有输出，确认所有的SAN连接被正确的设置，系统有没有正确的开启多路径功能。执行如下命令确认多路径服务有无启动。
chkconfig multipathd on
设备的设备名会被生成，/dev/dm-#，#指代的是多路径组，如果/dev/sda是唯一的多路径设备，/dev/dm-0将会是
/dev/sda和/dev/sdb的多路径设备。注意：fdisk不能用于设备/dev/dm-#，使用fdisk只能操作基础磁盘，要在设备映射多路
径映射设备上创建/dev/dm-#分区的操作. 执行一下命令。
kpartx -a /dev/dm-#
注意： dmsetup ls ?target=multipath
是个协助侦测系统上多路径设备的命令。如果在多路径设备数据库中没有发现硬件，请查看文章“How can I add moreproducts
into the mutipathing database?”
device-mapper-multipath-0.4.7
=============================
RHEL4 U3 Device Mapper Multipath Usage
Maintainer
------------
Benjamin Marzinski
bmarzins@redhat.com
Overview
------------
Device Mapper Multipath (DM-MP) allows nodes to route I/O over multiple paths to
a storage controller. A path refers to the connection from an HBA port to a
storage controller port. As paths fail and new paths come up, DM-MP reroutes
the I/O over the available paths.
When there are multiple paths to a storage controller, each path
appears as a separate device. DM-MP creates a new device on top of
those devices. For example, a node with two HBAs attached to a storage
controller with two ports via a single unzoned FC switch sees four
devices: /dev/sda, /dev/sdb, /dev/sdc, and /dev/sdd. DM-MP creates a
single device, /dev/mpath/mpath1 that reroutes I/O to those four
underlying devices.
DM-MP consists of the
following components:
o dm-multipath kernel module -- This module reroutes I/O and fails
over paths and path groups.
o multipath command -- This command configures, lists, and removes multipath
devices. The command is run in rc.sysinit during startup, and by
udev, whenever a block device is added.
o multipathd daemon -- This daemon monitors paths, checking to see if faulty
paths have been fixed. As paths come back up, multipathd may also
initiate path group switches to ensure that the optimal path group
is being used. Also, it is possible to interactively modify a
multipath device.
o kpartx command -- This command creates Device Mapper devices for the
partitions on a device. It is necessary to use this command for DOS-
based partitions with DM-MP.
DM-MP works with a variety of storage arrays. It
auto-configures the following storage arrays:
o 3PARdata VV
o Compaq HSV110
o Compaq MSA1000
o DDN SAN DataDirector
o DEC HSG80
o EMC SYMMETRIX
o EMC CLARiiON
o FSC CentricStor
o GNBD
o HITACHI DF400
o HITACHI DF500
o HITACHI DF600
o HP HSV110
o HP HSV210
o HP A6189A
o HP Open-
o IBM 3542
o IBM ProFibre 4000R
o NETAPP
o SGI TP9100
o SGI TP9300
o SGI TP9400
o SGI TP9500
o STK OPENstroage D280
o SUN StorEdge 3510
o SUN T4
Storage arrays not included in the list may require entries in the
/etc/multipath.conf file.
NOTE: Some storage arrays require special handling of I/O errors and
path-group switching. Those require separate hardware handler
kernel modules.
Terms and Concepts
---------------------
Hardware Handler:
A kernel module that performs hardware-specific actions when switching
path groups and dealing with I/O errors.
Path:
The connection from an HBA port to a storage controller port for a LUN.
Each path appears as a separate device. Paths can be in
various states (refer to "Path States").
Path States:
ready - Path is able to handle I/O requests.
shaky - Path is up, but temporarily not available for normal
operations.
faulty - Path is unable to handle I/O requests.
ghost - Path is a passive path, on an active/passive controller.
NOTE: The shaky and ghost states only exist for certain
storage arrays.
Path Group:
A grouping of paths. With DM-MP, only one path group--the
active path group--receives I/O at any time. Within a path
group, DM-MP selects which ready path should receive I/O in a
round robin fashion. Path groups can be in various states (refer to
"Path Group States").
Path Group States:
active - Path group currently receiving I/O requests.
enabled - Path groups to try if the active path group has no paths
in the ready state.
disabled - Path groups to try if the active path group and all
enabled path groups have no paths in the active state.
NOTE: The disabled state only exists for certain storage arrays.
Path Priority:
Each path can have a priority assigned to it by a callout program.
Path priorities can be used to group paths by priority and change
their relative weights for the round robin path selector.
Path Group Priority:
Each path group has a priority that is equal to the sum of the
priorities of all the non-faulty paths in the group. By default, the
multipathd daemon tries to ensure that the path group with the
highest priority is always in the active state.
Failover:
When I/O to a path fails, the dm-multipath module tries to switch to
an enabled path group. If there are no enabled path groups with
any paths in the ready state, dm-multipath tries to switch to a disabled
path group. If necessary, dm-multipath runs the hardware handler for the
multipath device.
Failback:
At regular intervals, multipathd checks the current priority of
all path groups. If the current path group is not the highest priority
path group, multipathd reacts according to the failback mode.
By default, multipathd immediately switches to the highest priority
path group. Other options for multipathd are to (a) wait for a
user-defined length of time (for the path groups to stabilize)
and then switch or (b) for multipathd to do nothing and wait for
manual intervention. Failback can be forced at any time by
running the multipath command.
Multipath device:
The multipath device is the device mapper device created by
dm-multipath. A multipath device can be identified by either
its WWID or its alias. A multipath device has one or more path
groups. It also has numerous attributes defined in the
following file:
/usr/share/doc/device-mapper-multipathd-0.4.5/multipath.conf.annotated
alias:
The alias is the name of a multipath device. By default, the
alias is set to the WWID. However, by setting the
"user_friendly_names" configuration option, the alias is set to a
unique name of the form mpath<n>. The alias name can also be
explicitly set for each multipath device in the configuration file.
NOTE: While the alias in guaranteed to be unique on a node, it
is not guaranteed to be the same on all nodes using the
multipath device. Also, it may change.
WWID:
The WWID (World Wide Identifier) is an identifier for the
multipath device that is guaranteed to be globally unique and
unchanging. It is determined by the getuid callout program.
Using DM-MP
------------------------------------
Initial setup:
1. If it is not already installed. Install the device-mapper-multipath
package.
2. Edit /etc/multipath.conf. For new installations, all devices are blacklisted.
The default blacklist is listed in the commented out section of
/etc/multipath.conf. If you comment out or delete the following lines in
/etc/multipath.conf, the default blacklist takes effect:
devnode_blacklist {
devnode "*"
}
For some conditions, that may not be sufficient. If DM-MP is
multipathing devices that you do not want it to work on, you can
blacklist the devices by either device name or WWID.
NOTE: It is safest to blacklist individual devices by WWID, because
their device names may change.
Several other configuration options are detailed later in this
document. To check the effects of configuration changes, you can
do a dry run with the following command:
# multipath -v2 -d
3. Set the multipathd init script to run at boot time. by issuing the commands
# chkconfig --add multipathd
# chkconfig multipathd on
4. start dm-multipath (This is only necessary the first time. On
reboot, this should happen automatically).
# multipath
# /etc/init.d/multipathd start
After initial setup, all access to the multipathed storage should go through the
multipath device.
GNBD devices will not be automatically multipathed after they are imported.
The command
# multipath
must be run after every time the devices are imported. Otherwise the Multipath
devices will not be created.
Configuration File:
Many features of DM-MP are configurable using the configuration file,
/etc/multipath.conf.
For a complete list of all options with descriptions, refer to
/usr/share/doc/device-mapper-multipathd-0.4.5/multipath.conf.annotated
The configuration file is divided into four sections: system defaults,
blacklisted devices (devnode_blacklist), per storage array model settings
(devices), and per multipath device settings (multipaths). The per multipath
device settings are used for the multipath device with a matching "wwid"
value. The per storage array model settings are used for all multipath devices
with matching "vendor" and "product" values. To determine the attributes of a
multipath device, first the per multipath settings are checked, then the per
controller settings, then the system defaults. The blacklisted device section
is described setup step 2.
NOTE: There are compiled-in defaults for the "defaults", "devnode_blacklist",
and "devices" sections of the configuration file. To see what these
are, refer to the following file:
/usr/share/doc/device-mapper-multipathd-0.4.5/multipath.conf.defaults
If you are using one of the storage arrays listed in the preceding
text (in "Overview"), you probably do not need to modify the "devices"
subsection. If you are using a simple disk enclosure, the defaults
should work. If you are using a storage array that is not
listed, you may need to create a "devices" subsection for your array.
Reconfiguring a running system:
If any changes are make to the configuration file that will effect multipathd
(check /usr/share/doc/device-mapper-multipathd-0.4.5/multipath.conf.annotated
to see if mulipathd is in the options scope), you must restart multipathd
for these to take effect. To do that, run
# /etc/init.d/multipathd restart
Explanation of output
-----------------------
When you create, modify, or list a multipath device, you get a printout of
the current device setup. The format is as follows.
For each multipath device:
action_if_any: alias (wwid_if_different_from_alias)
[size][features][hardware_handler]
For each path group:
\_ scheduling_policy [path_group_priority_if_known][path_group_status_if_known]
For each path:
\_ host:channel:id:lun devnode major:minor [dm_status_if_known][path_status]
The dm status (dm_status_if_known) is like the path status
(path_status), but from the kernel‘s point of view. The dm status has two
states: "failed", which is analogous to "faulty", and "active" which
covers all other path states. Occasionally, the path state and the
dm state of a device will temporarily not agree.
NOTE: When a multipath device is being created or modified, the path group
status and the dm status are not known. Also, the features are not always
correct. When a multipath device is being listed, the path group priority is not
known.
Restrictions
---------------
DM-MP cannot be run on either the root or boot device.
Other Sources of information
----------------------------
Configuration file explanation:
/usr/share/doc/device-mapper-multipathd-0.4.5/multipath.conf.annotated
Upstream documentation:
http://christophe.varoqui.free.fr/wiki/wakka.php?wiki=Home
mailing list:
dm-devel@redhat.com
Subscribe to this from https://www.redhat.com/mailman/listinfo/dm-devel.
The list archives are at https://www.redhat.com/archives/dm-devel/
Man pages:
multipath.8, multipathd.8, kpartx.8 mpath_ctl.8
http://www.ibm.com/developerworks/cn/linux/l-devmapper/