lfs_cleanerd
—
garbage collect a log-structured file system
lfs_cleanerd |
[-bcDdfmqs ] [-i
segment-number] [-J
raw-device] [-l
load-threshhold] [-n
number-of-segments] [-r
report-frequency] [-S
semaphore-address] [-t
timeout] node |
The lfs_cleanerd
command starts a daemon process which
garbage-collects the log-structured file system residing at the point named by
node in the global file system namespace. This command
is normally executed by
mount_lfs(8) when the
log-structured file system is mounted. The daemon will exit within a few
minutes of when the file system it was cleaning is unmounted.
Garbage collection on a log-structured file system is done by
scanning the file system's segments for active, i.e. referenced, data and
copying it to new segments. When all of the active data in a given segment
has been copied to a new segment that segment can be marked as empty, thus
reclaiming the space taken by the inactive data which was in it.
The following options are available:
-b
- Use bytes written, rather than segments read, when determining how many
segments to clean at once.
-c
- Coalescing mode. For each live inode, check to see if it has too many
blocks that are not contiguous, and if it does, rewrite it. After a single
pass through the filesystem the cleaner will exit. This option has been
reported to corrupt file data; do not use it.
-D
- Stay in the foreground, do not become a daemon process. Does not print
additional debugging information (in contrast to
-d
).
-d
- Run in debug mode. Do not become a daemon process, and print debugging
information. More
-d
s give more detailed
debugging information.
-f
- Use filesystem idle time as the criterion for aggressive cleaning, instead
of system load.
-i
segment-number
- Invalidate the segment with segment number
segment-number. This option is used by
resize_lfs(8), and
should not be specified on the command line.
-J
raw device
- Specify the raw device that the cleaner is to work against rather than
trying to figure it out from the mount point. This is mostly useful when
the cleaner is compiled into
rump_lfs(8), and the ATF
test framework.
-l
load-threshhold
- Clean more aggressively when the system load is below the given
threshhold. The default threshhold is 0.2.
-m
- Does nothing. This option is present for historical compatibility.
-n
number-of-segments
- Clean this number of segments at a time: that is, pass this many segments'
blocks through a single call to lfs_markv, or, if
-b
was also given, pass this many segments' worth
of blocks through a single call to lfs_markv.
-q
- Quit after cleaning once.
-r
report-frequency
- Give an efficiency report after every
report-frequency times through the main loop.
-S
semaphore address
- When the cleaner code is compiled into
rump_lfs(8), and the ATF
frame work, this option allows for a synchronization semaphore to be
specified. This option is not available in the stand alone
lfs_cleanerd(8).
-s
- When cleaning the file system, send only a few blocks through lfs_markv at
a time. Don't use this option.
-t
timeout
- Poll the filesystem every timeout seconds, looking
for opportunities to clean. The default is 300, that is, five minutes.
Note that
lfs_cleanerd
will be automatically
awakened when the filesystem is active, so it is not usually necessary to
set timeout to a low value.
The lfs_cleanerd
utility first appeared in
4.4BSD.