Several significant updates:

* Better wording of sections dealing with physical storage
* A new section on assumptions gvirstor has on its consumer devices
  (components) and its interaction  with file systems
* Improved markup (by hrs@)

Reviewed by:	hrs
Approved by:	gnn (mentor)
This commit is contained in:
Ivan Voras 2009-01-04 11:31:03 +00:00
parent 90886ebcc2
commit 2b67c28a09
Notes: svn2git 2020-12-20 02:59:44 +00:00
svn path=/head/; revision=186735

View File

@ -1,5 +1,4 @@
.\" Copyright (c) 2005 Pawel Jakub Dawidek <pjd@FreeBSD.org>
.\" Copyright (c) 2005 Ivan Voras <ivoras@FreeBSD.org>
.\" Copyright (c) 2006-2008 Ivan Voras <ivoras@FreeBSD.org>
.\" All rights reserved.
.\"
.\" Redistribution and use in source and binary forms, with or without
@ -25,7 +24,7 @@
.\"
.\" $FreeBSD$
.\"
.Dd July 8, 2006
.Dd December 17, 2008
.Dt GVIRSTOR 8
.Os
.Sh NAME
@ -69,10 +68,17 @@
.Sh DESCRIPTION
The
.Nm
utility is used for setting up a storage device of arbitrary large size (for example,
several TB), consisting of an arbitrary number of physical storage devices with
total size <= the virtual size. Data for the virtual devices will be allocated from
physical devices on demand. In short, this is the virtual storage functionality.
utility is used for setting up a virtual storage device of arbitrary
large size
.Pq for example, several TB ,
consisting of an arbitrary number of physical storage devices with the
total size which is equal to or smaller than the virtual size. Data
for the virtual devices will be allocated from physical devices on
demand. The idea behind
.Nm
is similar to the concept of Virtual Memory in operating systems,
effectively allowing users to overcommit on storage
.Pq free file system space .
The first argument to
.Nm
indicates an action to be performed:
@ -82,12 +88,15 @@ Set up a virtual device from the given components with the specified
.Ar name .
Metadata are stored in the last sector of every component.
Argument
.Ar virsize
is the size of new virtual device, with default being 2 TiB (2097152 MiB).
.Fl s Ar virsize
is the size of new virtual device, with default being 2 TiB
.Pq 2097152 MiB .
Argument
.Ar chunksize
is the chunk size, with default being 4 MiB (4096 KiB).
The default is thus "-s 2097152 -m 4096".
.Fl m Ar chunksize
is the chunk size, with default being 4 MiB
.Pq 4096 KiB .
The default is thus
.Qq Fl s Ar 2097152 Fl m Ar 4096 .
.It Cm stop
Turn off an existing virtual device by its
.Ar name .
@ -96,8 +105,10 @@ As with other GEOM classes, stopped geoms cannot be started manually.
.It Cm add
Adds new components to existing virtual device by its
.Ar name .
The specified virstor device must exist and be active (i.e.
module loaded, device present in /dev).
The specified virstor device must exist and be active
.Pq i.e. module loaded, device present in Pa /dev .
This action can be safely performed while the virstor device is in use
.Pq Qo hot Qc operation
.It Cm remove
Removes components from existing virtual device by its
.Ar name .
@ -130,82 +141,97 @@ Hardcode providers' names in metadata.
Be more verbose.
.El
.Sh EXIT STATUS
Exit status is 0 on success, and 1 if the command fails.
The
.Nm
utility exits 0 on success, and 1 if an error occurs.
.Sh EXAMPLES
The following example shows how to create a virtual device of default size
(2 TiB), of default chunk (extent) size (4 MiB), with two physical devices for
backing storage.
.Pq 2 TiB ,
of default chunk
.Pq extent
size
.Pq 4 MiB ,
with two physical devices for backing storage.
.Bd -literal -offset indent
gvirstor label -v mydata /dev/ad4 /dev/ad6
newfs /dev/virstor/mydata
.No gvirstor label -v Ar mydata Ar /dev/ad4 Ar /dev/ad6
.No newfs Ar /dev/virstor/mydata
.Ed
.Pp
From now on, the virtual device will be available via the
.Pa /dev/virstor/mydata
device entry.
To add a new physical device / provider to an active virstor device:
To add a new physical device / component to an active virstor device:
.Bd -literal -offset indent
gvirstor add mydata ad8
.No gvirstor add Ar mydata Ar ad8
.Ed
.Pp
This will add physical storage (from ad8) to
This will add physical storage
.Ar ad8
to
.Pa /dev/virstor/mydata
device.
To see device status information (including how much physical storage
is still available for the virtual device), use:
To see device status information
.Pq including how much physical storage is still available for the virtual device ,
use:
.Bd -literal -offset indent
gvirstor list
.Ed
.Pp
All standard
.Xr geom 8
subcommands (e.g. "status", "help") are also supported.
.Sh SYSCTLs
subcommands
.Pq e.g. Cm status , Cm help
are also supported.
.Sh SYSCTL VARIABLES
.Nm
has several
has several
.Xr sysctl 8
tunable variables.
.Bd -literal -offset indent
.Pa int kern.geom.virstor.debug
.Va int kern.geom.virstor.debug
.Ed
.Pp
This sysctl controls verbosity of the kernel module, in the range
1 to 15. Messages that are marked with higher verbosity levels than
this are supressed. Default value is 5 and it's not
recommented to set this tunable to less than 2, because level 1 messages
1 to 15. Messages that are marked with higher verbosity levels than
this are suppressed. Default value is 5 and it's not
recommended to set this tunable to less than 2, because level 1 messages
are error events, and level 2 messages are system warnings.
.Bd -literal -offset indent
.Pa int kern.geom.virstor.chunk_watermark
.Va int kern.geom.virstor.chunk_watermark
.Ed
.Pp
Value in this sysctl sets warning watermark level for physical chunk usage
on a single component. The warning is issued when a virstor component
has less than this many free chunks (default 100).
has less than this many free chunks
.Pq default 100 .
.Bd -literal -offset indent
.Pa int kern.geom.virstor.component_watermark
.Va int kern.geom.virstor.component_watermark
.Ed
.Pp
Value in this sysctl sets warning watermark level for component usage.
The warning is issed when there are less than this many unallocated
components (default is 1).
The warning is issued when there are less than this many unallocated
components
.Pq default is 1 .
.Pp
All these sysctls are also available as
.Xr loader 8
tunables.
.Sh LOG MESSAGES
.Sh DIAGNOSTICS
.Nm
kernel module issues log messages with prefixes in standardised format,
kernel module issues log messages with prefixes in standardized format,
which is useful for log message filtering and dispatching. Each message
line begins with
.Bd -literal -offset indent
.Pa GEOM_VIRSTOR[%d]:
.Li GEOM_VIRSTOR[%d]:
.Ed
.Pp
The number (%d) is message verbosity / importance level, in the range
1 to 15. If a message filtering, dispatching or operator alert system is
used, it is recommended that messages with levels 1 and 2 be taken
seriously (for example, to catch out-of-space conditions as set by
watermark sysctls).
The number
.Pq %d
is message verbosity / importance level, in the range 1 to 15. If a
message filtering, dispatching or operator alert system is used, it is
recommended that messages with levels 1 and 2 be taken seriously
.Pq for example, to catch out-of-space conditions as set by watermark
sysctls .
.Sh SEE ALSO
.Xr geom 4 ,
.Xr geom 8 ,
@ -218,10 +244,40 @@ The
utility appeared in
.Fx 7.0 .
.Sh BUGS
Commands "add" and "remove" contain unavoidable critical sections
which may make the virstor device unusable if a power failure (or
other disruptive event) happens during their execution.
It's recommended to run them when the system is quiescent.
Commands
.Cm add
and
.Cm remove
contain unavoidable critical sections which may make the virstor
device unusable if a power failure
.Pq or other disruptive event
happens during their execution. It's recommended to run them when the
system is quiescent.
.Sh ASSUMPTIONS AND INTERACTION WITH FILE SYSTEMS
There are several assumptions that
.Nm
has in its operation: that the size of the virtual storage device will not
change once it's set, and that the sizes of individual physical storage
components will always remain constant during their existence. For
alternative ways to implement virtual or resizable file systems see
.Xr zfs 1M ,
.Xr gconcat 8 and
.Xr growfs 8 .
.Pp
Note that
.Nm
has nontrivial interaction with file systems which initialize a large
number of on-disk structures during newfs. If such file systems
attempt to spread their structures across the drive media
.Pq like UFS/UFS2 does ,
their efforts will be effectively foiled by sequential allocation of
chunks in
.Nm
and all their structures will be physically allocated at the start
of the first virstor component. This could have a significant impac
t on file system performance
.Pq which can in some rare cases be even positive .
.Sh AUTHOR
.An Ivan Voras Aq ivoras@FreeBSD.org
.An Ivan Voras Aq ivoras@FreeBSD.org
.Pp
Sponsored by Google Summer of Code 2006