vxvol(ADM)

Synopsis

Description

Each operation can be applied to only one disk group at a time, due to internal implementation constraints. Any volume operands will be used to determine a default disk group, according to the standard disk group selection rules described in vxintro(ADM). A specific disk group can be selected with -g diskgroup.

The recognized operation keywords are:

init

Perform an initialization action on a volume. This can be applied to volumes that were created by vxmake and that have not yet been initialized, or volumes that have been set to the uninitialized state with vxmend fix empty. The action to perform is specified by the init_type operand, which is usage-type-dependent. The volume operand determines which usage type to use for performing the operation.

rdpol

Set the read policy for a volume based on the policy operand. These are the recognized read policies:

round

Use a round-robin read order among the enabled, readable plexes associated with the volume. No plex operand should be specified for the round read-policy type.

prefer

Read preferentially from the plex named by the plex operand. If the plex is enabled, readable, and associated with the volume, then any read operation on the volume results in a read from that plex if all blocks requested in the read are contained in the plex. The plex operand is required for the prefer read-policy type.

select

Select a default policy based on plex associations to the volume. For a volume that contains one enabled, striped plex, the default is to prefer that plex. For any other set of plex associations, the default is to use a round-robin policy. No plex operand should be specified for the select read-policy type.

start

Enable disabled or detached volumes named by the volume operands. The process of enabling a volume is a highly usage-type-dependent operation and may result in transfers of data between plexes associated with the volume.

If the start operation is applied to an uninitialized volume (for example, a volume just created by vxmake), a default initialization will be used to initialize and enable the volume.

If the volume is not normally started because failures and disk removals have left all associated plexes with invalid data, the -f option can be used to try to start the volume, anyway. This can be used after replacing disks to enable the volume so that its contents can be restored from backup or reinitialized.

startall

Attempt to start all volumes that are disabled. If a -U usetype option is specified, then attempt to start all disabled volumes with the indicated usage type. This operation will not start uninitialized volumes. By default, start all volumes in the rootdg disk group. A different disk group can be specified with the -g option.

stop

Disable the enabled or detached volumes named by the volume operands.

The stop operation provides an interface to the usage type of a volume for shutting down operations on a volume in a clean manner. The specific method for cleanly stopping a volume, and the precise meaning of ``clean'' are both highly usage-type-dependent. By convention, -f can be used to force stopping of a volume that is in use, forcing I/O failures to be returned for any further volume device operations.

stopall

Attempt to stop all volumes that are enabled. If a -o usetype option is specified, then attempt to stop all disabled volumes with the indicated usage type. By default, stop all volumes in the rootdg disk group. A different disk group can be specified with the -g option.

resync

Examine all volumes named by the volume operands and perform any synchronization operations that are required. The exact procedure for this operation is usage-type specific.

recover

Some usage types support the concept of recovery of data for objects within a volume different from the mirror resynchronization model. The recover operation is used to specify this type of recovery. The exact procedure for this operation is usage type specific.

maint

Detach each volume named by the volume operands. When a volume is detached, normal read and write operations to the volume fail, although most volume ioctl operations can still be used.

set

Change specific volume characteristics. The changes to be made are given by arguments immediately after the set keyword of the form attribute=value. The set of volumes affected by the operation are given after these operands; thus the attribute list ends with an operand that does not contain an equal sign. To allow for volume names that contain an equal sign, an operand of -- can be used to terminate the attribute list. Each usage type represented by the list of volume operands is called once, with the set of all volumes with that usage type.

The set of attribute=value attribute arguments that are recognized depends upon the volume usage type (see the sections on usage types later in this manual page for information on available attributes). However, an attribute argument of the form

	len=number 

is expected to be interpreted (if at all) as requesting a change in the length of a volume, regardless of the volume's usage type. The number value is interpreted as a standard length number (see vxintro(ADM)).

Options

-g diskgroup

Specify the disk group for the operation. The disk group can be specified either by name or by disk group ID. See vxdg(ADM) for more information on disk groups.

-U usetype

Force the operation to be performed by the usage-type utility for this usage type.

-V

Display a list of utilities that would be called from vxvol, along with the arguments that would be passed. The -V option performs a ``mock run'' so the utilities are not actually called, and no changes are made to the volume configuration database.

-f

Force an operation in some situations where the operation has questionable semantics. For example, -f may be used to reduce the length of a volume with vxvol set, to stop a volume that is currently open or mounted as a file system, or to attempt to start a volume that has no plexes with valid data.

-o useopt

Pass in usage-type-specific options to the operation. By convention, the following usage-type- specific options should be implemented by all usage types:

bg

Perform any extended revive operations in background processes after the volume and one or more plexes have been enabled. A volume that is started or whose length is changed successfully with this option will be usable immediately after the operation completes, although recovery operations may affect performance of the volume for an extended period of time.

plexfork[=count]

Perform up to the specified number of plex revive operations simultaneously. If no count is specified, then a suitable small number is used (normally 10).

delayrecover

Do not perform any plex revive operations when starting a volume. Simply enable the volume and any plexes. This may leave some stale plexes, and may leave a mirrored volume in a special read-writeback (NEEDSYNC) recover state that performs limited plex recovery for each read to the volume.

slow[=iodelay]

Reduce the system performance impact of plex recovery operations and volume length changes. Startup recovery and length change consistency operations are usually a set of short operations on small regions of the volume (normally from 16K bytes to 128K bytes). This option inserts a delay between the recovery of each such region. A specific delay can be specified with iodelay as a number of milliseconds, or else a default is chosen (normally 250 milliseconds).

iosize=size

Perform recovery operations in regions with the length specified by size, which is a standard Volume Manager length number (see vxintro(ADM)). Specifying a larger number typically causes the operation to complete sooner, but with greater impact on other processes using the volume. The default I/O size is typically 32 kilobytes.

verbose

Print a message for each volume that is successfully started. Without this option, messages appear only for volumes that fail to start.

Fsgen and gen usage types

In addition to the standard -o options required for all usage types, the fsgen and gen usage types provide the following additional options:

force

Force an operation that is not normally performed as part of the operational model of the Volume Manager and may have adverse effects on data. This is the same as -f.

norecov

This can only be used with the gen usage type. Prevent the start operation from recovering plexes through the vxplex utility. Instead, all STALE and ACTIVE plexes are simply treated as equivalent to CLEAN plexes, and are thus enabled without being made consistent. This can be used for volumes whose contents are recreated for each use.

An example of a possible use for this attribute is a swap area and the /tmp file system. In the case of /tmp, the model assumes that mkfs is used to create an empty file system after the volume has been started.

init

These are the recognized uses of the vxvol init operation:

vxvol init clean volume [ plex ]

Set the state for the specified plex to CLEAN, and set all other plexes to STALE. The vxvol start operation can then be used to recover the volume from the CLEAN plex. This operation requires that the volume not be enabled.

If the specified volume has only one plex, then the plex argument is not required as it defaults to that plex. If specified, then the plex argument must represent a plex that is associated with the volume.

vxvol init active volume

Set the state for all plexes associated with volume to ACTIVE and enable the volume and its plexes. This is used to initialize a single or multiple-plex volume where all plexes are known to have identical contents.

vxvol init enable volume

Enable the volume and its plexes but leave the volume uninitialized. This operation can be used only for non-enabled volumes. It is used to temporarily enable a volume so that data can be loaded onto it to make it consistent. Once the data has been loaded, init active should be used to fully enable the volume. init active could be used, for example, if a complete image of the volume is to be loaded from a tape.

vxvol init zero volume

Write zero blocks to all plexes in the volume, up to the length of the volume. After the writes complete, the state of each plex is set to ACTIVE and the volume and its plexes are enabled. init zero volume could be used, for example, before running mkfs to put a file system on the volume.

If this operation is interrupted by a signal, then an attempt is made to restore all affected records to their original state, or to a state that is roughly equivalent to their original state. If this attempt is interrupted, such as through another signal, then the user many need to perform some cleanup. A set of commands to perform this cleanup are written to the standard error before the volume utility exits.

start

Starting an uninitialized gen or fsgen volume enables the volume and its plexes, sets the plexes to the ACTIVE state, and recovers the plexes to ensure that each plex has the same contents. If the volume has only one plex, then the volume is immediately set to the ACTIVE state; otherwise, the volume is set to the SYNC state and a special read/write-back mode is set to recover regions of the volume on every read operation. The volume is then read from beginning to end to make all plexes consistent, then the volume is set to the ACTIVE state.

Starting a volume with no active dirty region logging involves enabling all CLEAN and ACTIVE plexes and putting them in the ACTIVE state. If an I/O failure was logged against the plex, or if a disk replacement caused a plex to become stale, then the plex is considered STALE. If any of the subdisks for the plex reside on a removed or inaccessible disk, then the plex is ignored for the purposes of starting the volume.

If two or more plexes were enabled, and if the volume was active at the time the system went down, then the state for the volume is set to SYNC and a special read/write-back recovery mode is used to recover consistency of the volume, segment-by-segment, on every read. A process is then started (in the background with the -o bg option) to recover consistency for the entire length of the volume.

If any plexes were considered STALE, then those plexes are attached by calling vxplex att. The number of concurrent plex attach operations are limited based on the rules for -o plexfork.

Recovery of plexes with a dirty region log uses the same rules as for volumes without a valid dirty region log, except that recovery of non-stale plexes is done by scanning the contents of the dirty region log and recovering consistency for those regions listed in the log as requiring recovery.

In addition to enabling the volume and managing the recovery of plex consistency, starting a volume clears any transient operations that were being applied to a volume before the system was rebooted. Starting a volume dissociates and removes temporary plexes or subdisks, and dissociates plexes that were being attached if the attach operation did not complete. Snapshot plexes created by vxassist are also removed.

If the volume is unstartable because there are no valid, non-stale plexes and the -f flag is then specified, all STALE plexes that do not contain unusable subdisks (subdisks on failed or removed disks) will be changed to ACTIVE. The volume will then be started and synchronized from those plexes.

stop

Stopping an fsgen or gen volume disables the volume and its associated plexes. In addition, the utility state for each ACTIVE plex is changed as follows:

• If the plex is detached or disabled, set the state for the plex to STALE. If all plexes are set to STALE, then the volume cannot be started until vxmend is used to change the state of one or more plexes to CLEAN or ACTIVE. A plex normally becomes detached as a result of an I/O error on the plex, or a disk failure or replacement. I/O failures will not normally detach the last remaining enabled plex in a volume, so disk removal operations are the only normal operational method of making a volume unstartable.

• If the plex is volatile, i.e., one of the subdisks in the plex is defined on a disk with the volatile attribute (see vxdisk(ADM)), then set the plex state to STALE.

• If the volume is enabled and the plex is also enabled, then set the plex state to CLEAN.

• If the volume is detached and the plex is enabled, then the plex state is left as ACTIVE. A volume can be left detached, with remaining valid plexes, only as a result of calling vxvol maint to detach an enabled volume.

Normally, the stop operation fails if any extended operations are using the volume or any of its associated plexes. Such operations are detected as a nonempty value for the tutil0 field in a volume or plex record. If the -f option is specified, then the stop operation ignores volume and plex tutil0 fields.

The -f option must also be given to force the stopping of a volume that is open or mounted as a file system. In this case, a warning message is still written to the standard error, but the stop operation is not otherwise affected. Stopping an open or mounted volume is not normally advisable. Volumes that have possibly differing plex contents will be re-synchronized to contain consistent data. Any such volumes that are in the NEEDSYNC state will be recovered using a read/write-back recovery mode and then put into the ACTIVE state.

Plexes in the SYNC state may already be under recovery and the volume command will take no action to recover them unless the command was invoked with the -o force option.

maint

The -f option is required to detach an enabled volume. Also, a warning is written to the standard error for volumes that are open or mounted.

set

The attributes that can be changed are:

len=number

Change the length of each volume specified by the volume operands to number sectors. number is a standard Volume Manager length number (see vxintro(ADM)). Decreasing the length of a volume requires use of -f.

If the volume is enabled, then count the number of enabled, read-write plexes that would remain complete after the length change. The operation fails if this number would become zero, but the number of sparse plexes would become greater than 1. Changing the length of a volume with one enabled plex beyond the length of the plex requires use of the -f option.

If the volume is not enabled, count the number of CLEAN and ACTIVE plexes that would remain complete after the length change, then use the algorithm mentioned previously for determining whether the operation is allowed or requires use of -f.

In order to ensure that the new region of the volume is consistent across all plexes of the volume, the volume is put into a SYNC state and read/write-back mode, and a read loop is then performed against the volume. Once this loop has completed, the volume is put back into the ACTIVE state.

logtype=type

Set the type of logging to be used on the volume. This change can be applied only to volumes that are stopped and that have no ACTIVE plexes. Allowed log types are drl (logs the regions involved in all volume writes), none (never does logging), and undef (never does logging). If the logging type is set to undef, then a future vxsd aslog or vxplex att operation will change it to drl. See the fsgen and gen sections of vxsd(ADM) and vxplex(ADM) for more information.

loglen=size

Set the size for logs used with the volume. The size value is a standard Volume Manager length numbers (see vxintro(ADM)).

startopts=volume_options

Set options that are applied to the volume every time the volume is started, independently of options specified with the volume start command. This is a set of comma- separated options of the same form used with the -o option letter. At the present time, only the noattach and verbose options can be applied to volumes in this manner. Unrecognized or inappropriate options are ignored.

RAID-5 usage type

checkpt=size

Set the checkpoint size for a volume. A complete resynchronization of a volume via VOL_R5_RESYNC ioctls can take an extended amount of time. It is conceivable in some circumstances that the operation could be stopped before it complete (such as by a system crash). To avoid having to restart the synchronization from the beginning of the volume (after a certain amount of the volume has been synchronized), a transaction is issued to record the offset to which the resynchronization has completed. This size is called the checkpoint length and can be set using the checkpt option. The default checkpoint length is 64 megabytes.

force

Force an operation that is not normally performed as part of the operational model of the Volume Manager. It may have adverse effects on the data contained by the volume. This is the same as -f.

delayrecover

This prevents the start operation from undergoing some recovery operations. RAID-5 logs, if any are valid, will still be replayed; however, no parity resynchronizations or subdisk recoveries will be performed.

unsafe

This allows access to certain volumes earlier in the starting process than is normally allowed by the operating process of RAID-5 volumes. This can have adverse effects on the data, and can also result in the RAID-5 volume becoming unusable after a system crash or a power failure.

syncstartok

This allows the delayrecover option to be ignored if the volume must undergo parity resynchronizations or subdisk recoveries before the volume can be enabled.

init

The raid5 usage type recognizes the following uses of the init operation:

vxvol init active volume

Zero the RAID-5 log plexes, if any, and make the volume available for use. The parity in the volume is marked as stale, though no parity resynchronization is performed; the volume is left with a state of NEEDSYNC.

vxvol init zero volume

Write zeros to the RAID-5 log plexes, if any, and write zeros to the entire length of the volume. This is achieved by issuing the VOL_R5_ZERO ioctl for the entire altitude of the volume. The volume is left in the ACTIVE state.

start

Starting an uninitialized volume (one with a state of EMPTY) zeros any RAID-5 log plexes and then resynchronizes the parity of the volume by issuing VOL_R5_RESYNC ioctls. All subdisks are marked as non-stale and read-write. The volume and RAID-5 plex are then enabled and marked as ACTIVE, and all valid RAID-5 log plexes are marked as LOG. If any RAID-5 log plex proves to be invalid (such as having its NODAREC flag set) its state is set to BADLOG.

Starting a volume that has been shut down cleanly or is not marked as dirty enables the RAID-5 plex and RAID-5 log plexes, and sets the volume kernel state to detached to zero the RAID-5 log plexes for the volume, if any. Once this is completed, all valid RAID-5 log plexes are set to LOG and the volume is enabled and put in the ACTIVE state.

Starting a volume that was not shut down cleanly requires that the parity be resynchronized. If the volume has valid RAID-5 log plexes, the volume is first detached and has its state set to REPLAY, and all log plexes and the RAID-5 plex are enabled. If there are any valid RAID- 5 log plexes, their contents are read and their data is written to the appropriate regions of the RAID-5 plex. If reading the RAID-5 logs fails, the logs are marked as invalid and the parity is resynchronized as if there were no logs. Once the replay is complete, the RAID-5 logs are enabled and the volume is enabled and its state is set to ACTIVE.

If the volume needs resynchronization and no valid log plexes exist, the parity must be fully resynchronized. The volume is enabled and its state is set to RESYNC, and the RAID-5 plex is enabled. If usable RAID-5 plexes are available, but contain invalid data, they are zeroed. The parity is then resynchronized by issuing VOL_R5_RESYNC ioctls for the entire length of the volume. Once this is completed, the volume's state is set to ACTIVE. Any usable RAID-5 logs are enabled and set to the LOG state.

If a volume requires full resynchronization (i.e., has no usable logs) and the RAID-5 plex has stale or unusable subdisks, the volume is unusable and the start operation will fail. This can be overridden by using the -f flag or the -o force option . In this case, any stale subdisks are marked as non-stale and a full resynchronization is performed; however, this may result in some invalid data being introduced into the volume. If multiple subdisks at the same altitude in the RAID-5 plex are unusable (such as because they have their NODEVICE flag set), the volume is unusable and cannot be overridden.

Once any parity resynchronization has been completed, any subdisks still marked as stale are recovered. This is done by marking the subdisk as stale and write-only and issuing VOL_R5_RECOVER ioctls to regenerate the data on the stale subdisks. The subdisk is then marked as non-stale and read-write.

If the -o delayrecover option is specified, the only recoveries that will be performed are log replays. If the volume requires a parity resynchronization, it is enabled and left in the NEEDSYNC operation, and its parity is marked as stale. Any subdisk recoveries that are needed are not performed, and the stale subdisks are marked as stale.

Normally, if a volume has no RAID-5 logs, it will not be enabled with a stale subdisk or an unusable subdisk because were the system to crash or a power failure occurred while the volume was is in use, the parity could become stale and the volume would be unusable. This behavior can be overridden by specifying the -o unsafe option, which will cause the volume to be enabled during the above situations. As the name suggests, this is considered unsafe because doing so could cause data loss.

If only the -o delayrecover option is specified to start a volume with a stale subdisk or an unusable subdisk, the start operation will fail. In this case, the delayrecover option can be ignored by also specifying the -o syncstartok option.

stop

Stopping a raid5 volume disables the volume and its associated plexes. If the volume is in the SYNC state, it is changed to the NEEDSYNC state so that recovery will be performed at the next start. Any invalid or detached RAID-5 logs are set to the BADLOG state so that they will not be used during the next start.

Normally, the stop operation will fail if any extended operations are using the volume or any of its plexes. Such operations are detected as a non-empty value for the tutil0 field in a volume or plex record. If the -f option is specified, then the stop operation ignores volume and plex tutil0 fields.

resync

The resync operation examines the named volumes to see if they are enabled and if the parity in any part of a volume is stale; this is normally indicated by a volume state of NEEDSYNC. If so, the volume is placed in the SYNC state and VOL_R5_RESYNC ioctls are issued to resynchronize the parity in those regions. Upon completion, the volume is placed in the ACTIVE state.

recover

The raid5 usage type supports the following invocations of the recover operation:

recover [ subdisk ... ]

The recover operation can be used to initiate a recovery of subdisks containing invalid data. If subdisks are specified and are stale, they are recovered in the order specified. This is done by setting the stale and write-only flags on the subdisks and issuing VOL_R5_RECOVER ioctls to regenerate the data. After a successful recovery, the subdisk is marked as non-stale and read-write.

If no subdisk arguments are specified, the subdisks of the RAID-5 plex of the volume are checked to see if they are stale or have invalid contents. If any are found, they are recovered as described above.

set

The attributes that can be set for raid5 volumes are:

len=number

Change the length of the volumes specified to be number sectors. number is a standard Volume Manager length specification (see vxintro(ADM)). Decreasing the length of a volume requires -f.

The volume length cannot be increased such that the RAID-5 plex is sparse in respect to the new volume length; this would make the volume unusable.

In order to assure that the new region of the volume is consistent, the new region of the volume (from the old length to the new length) is filled with zeros by issuing VOL_R5_ZERO ioctls before the length is reset.

loglen=size

Set the size of the RAID-5 log for the volume. This cannot be set if the volume has no logs. If the length is being increased, the operation will not be allowed if it would cause any of the RAID-5 log plexes to become sparse in respect to the new length.

startopts=volume_options

Set options that are applied to the volume every time the volume is started, independently of options specified with the volume start command. This is a set of comma- separated options of the same form used with the -o option letter. Unrecognized or inappropriate options are ignored.

Files

/etc/vx/type/usetype/vxvol

The utility that performs volume operations for a particular volume usage type.

/dev/vx/dsk/group/volume

The device node that can be used for mounting a file system created on the volume named volume in the disk group named group. Volumes in group rootdg are also directly under the /dev/ vx/dsk directory.

/dev/vx/rdsk/group/volume

The device node that can be used for issuing raw I/O requests and also for issuing ioctl requests to the volume named volume in disk group named group. Volumes in group rootdg are also directly under the /dev/vx/rdsk directory.

Exit codes

See vxintro(ADM) for a list of standard exit codes.

References