4.2 Objective 2: Maintain the
Integrity of Filesystems
Over the course of time, active filesystems
can develop problems, such as:
-
A filesystem fills to capacity, causing
programs or perhaps the entire system to fail.
-
A filesystem is corrupted, perhaps due to a
power failure or system crash.
-
A filesystem runs out of inodes, meaning
that new filesystem objects cannot be created.
Carefully monitoring and checking Linux
filesystems on a regular basis can help prevent and correct
these types of problems.
4.2.1 Monitoring Free Disk Space
and Inodes
A read/write filesystem isn't much good if it
grows to the point that it won't accept any more files. This
could happen if the filesystem fills to capacity or runs out
of inodes.
Inodes are the data structures within
filesystems that describe files on disk. Every filesystem
contains a finite number of inodes, set when the filesystem is
created. This number is also the maximum number of files that
the filesystem can accommodate. Because filesystems are
created with a huge number of inodes, you'll probably never
create as many files as it would take to run out of inodes.
However, it is possible to run out of inodes if a partition
contains many small files.
It is important to prevent space and inode
shortages from occurring on system partitions. The df command gives you the information
you need on the status of both disk space utilization and
inode utilization.
Syntaxdf [options] [directories]
Description
Display overall disk utilization
information for mounted filesystems on directories.
Usually, directories are device files for partitions,
such as /dev/hda1, but using another file or directory
name yields information on the partition that holds the file
or directory. If directories are omitted, information
for mounted filesystems on all devices in /etc/fstab
are displayed.
Frequently used options
- -h
-
Displays results in a human-readable
format, including suffixes such as M (megabytes)
and G (gigabytes).
- -i
-
Displays information on remaining inodes
rather than the default disk space information.
Example 1
Check disk space utilization on all
filesystems: # df -h
Filesystem Size Used Avail Use% Mounted on
/dev/sda1 387M 56M 311M 15% /
/dev/sda5 296M 5.2M 276M 2% /boot
/dev/sda9 1.9G 406M 1.4G 22% /home
/dev/sda6 53M 12M 39M 23% /root
/dev/sda10 99M 104k 93M 0% /tmp
/dev/sda8 972M 507M 414M 55% /usr
/dev/sda7 296M 9.3M 272M 3% /var
This example shows that of the seven
filesystems mounted by default, none exceeds 55 percent
capacity.
Example 2
Check the same filesystems for inode
utilization: # df -i
Filesystem Inodes IUsed IFree IUse% Mounted on
/dev/sda1 102800 7062 95738 7% /
/dev/sda5 78312 29 78283 0% /boot
/dev/sda9 514000 934 513066 0% /home
/dev/sda6 14056 641 13415 5% /root
/dev/sda10 26104 60 26044 0% /tmp
/dev/sda8 257040 36700 220340 14% /usr
/dev/sda7 78312 269 78043 0% /var
Among these partitions, the largest
consumption of inodes is a mere 14 percent. It is clear that
none of the filesystems is anywhere near consuming the maximum
number of inodes available. Note that the /usr
partition (with 14 percent of inodes used) has used 55 percent
of the disk space. With utilization like this, the /usr
volume will most likely fill to capacity long before the
inodes are exhausted.
Example 3
Quickly determine which partition the current
working directory (represented simply by a single dot) is
located: # df .
/dev/sda1 102800 7062 95738 7% /
When a filesystem is nearing capacity, files
may simply be deleted to make additional space available.
However, in the rare case in which an inode shortage occurs,
the filesystem must be recreated with a larger number of
inodes unless a significant number of files can be deleted.
4.2.2 Monitoring Disk Usage
Have you ever found yourself wondering,
"Where did all the disk space go?" Some operating systems make
answering this question surprisingly difficult using only
native tools. On Linux, the du
command can help display disk utilization information on a
per-directory basis and perhaps answer that question. du recursively examines directories
and reports detailed or summarized information on the amount
of space consumed.
Syntaxdu [options] [directories]
Description
Display disk utilization information for
directories. If directories are omitted, the
current working directory is searched.
Frequently used options
- -a
-
Shows all files, not just
directories.
- -c
-
Produces a grand total for all listed
items.
- -h
-
Displays results in a human-readable
format, including suffixes such as M (megabytes)
and G (gigabytes).
- -s
-
Prints a summary for each of the
directories specified, instead of totals for each
subdirectory found recursively.
- -S
-
Excludes subdirectories from counts and
totals, limiting totals to directories.
Example 1
Examine disk utilization in
/etc/rc.d: # du /etc/rc.d
882 /etc/rc.d/init.d
1 /etc/rc.d/rc0.d
1 /etc/rc.d/rc1.d
1 /etc/rc.d/rc2.d
1 /etc/rc.d/rc3.d
1 /etc/rc.d/rc4.d
1 /etc/rc.d/rc5.d
1 /etc/rc.d/rc6.d
904 /etc/rc.d
Example 2
Display utilization by files in /etc,
including subdirectories beneath it: # du -s /etc
13002 /etc
Example 3
Display utilization by files in /etc,
but not in subdirectories beneath it: # du -Ss /etc
1732 /etc
Example 4
Show a summary of all subdirectories under
/home, with human-readable output: # du -csh /home/*
42k /home/bsmith
1.5M /home/httpd
9.5M /home/jdean
42k /home/jdoe
12k /home/lost+found
1.0k /home/samba
11M total
This result shows that 11 MB of total disk
space is used.
Example 5
Show the same summary, but sort the results
to display in order of largest to smallest disk utilization:
# du -cs /home/* | sort -nr
11386 total
9772 jdean
1517 httpd
42 jdoe
42 bsmith
12 lost+found
1 samba
This result shows that user jdean is
consuming the largest amount of space. Note that the
human-readable format does not sort in this way, since sort is unaware of the human-readable
size specifications.
4.2.3 Checking Filesystem
Integrity
No matter how stable, computers do fail, even
due to something as simple as a power cable being accidentally
unplugged. Unfortunately, such an interruption can make a mess
of a filesystem. If a disk write operation is aborted before
it completes, the data in transit could be lost, and the
portions of the disk that were allocated for it are left
marked as used. In addition, filesystem writes are cached in
memory, and a power loss or other crash prevents the kernel
from synchronizing the cache with the disk. Both of these
scenarios lead to inconsistencies in the filesystem and must
be corrected to ensure reliable operation.
Filesystems are checked with fsck. Like mkfs, fsck is a front-end to filesystem-
specific utilities -- including fsck.ext2, which is a
link to the e2fsck program (see
its manpage for detailed information).
Part of the information written on disk to
describe a filesystem is known as the superblock,
written in block 1 of the partition. If this area of the disk
is corrupted, the filesystem is inaccessible. Because the
superblock is so important, copies of it are made in the
filesystem at regular intervals, by default every 8192 blocks.
The first superblock copy is located at block 8193, the second
copy is at block 16385, and so on. As you'll see, fsck can use the information in the
superblock copies to restore the main superblock.
Syntaxfsck [options] [-t type] [fs-options] filesystems
Description
Check
filesystems for errors and optionally correct them. By
default, fsck assumes the
ext2 filesystem type and runs interactively, pausing to
ask for permission before applying fixes.
Frequently used options for fsck
- -A
-
Run checks on all filesystems specified in
/etc/fstab. This option is intended for use at boot
time, before filesystems are mounted.
- -N
-
Don't execute, but show what would be
done.
- -t type
-
Specify the type of filesystem to check;
the default is ext2. The value of type
determines which filesystem-specific checker is called.
Frequently used options for
e2fsck
- -b superblock
-
Use an alternative copy of the
superblock. In interactive mode, e2fsck automatically uses
alternative superblocks. Typically, you'll try -b 8193 in non-interactive mode to
restore a bad superblock.
- -c
-
Check for bad blocks.
- -f
-
Force a check, even if the filesystem looks
clean.
- -p
-
Automatically repair the filesystem without
prompting.
- -y
-
Answers "yes" to all interactive prompts,
allowing e2fsck to be used
noninteractively.
Example 1
Check the ext2 filesystem on
/dev/hda5, which is not mounted: # fsck /dev/hda5
[/sbin/fsck.ext2 -- ] fsck.ext2 /dev/hda5
Parallelizing fsck version 1.14 (9-Jan-1999)
e2fsck 1.14, 9-Jan-1999 for EXT2 FS 0.5b, 95/08/09
/dev/hda5: clean, 1011/34136 files, 4360/136521 blocks
The partition was clean, so fsck didn't really check it.
Example 2
Force a check: # fsck -f /dev/hda5
Parallelizing fsck version 1.14 (9-Jan-1999)
e2fsck 1.14, 9-Jan-1999 for EXT2 FS 0.5b, 95/08/09
Pass 1: Checking inodes, blocks, and sizes
Pass 2: Checking directory structure
Pass 3: Checking directory connectivity
Pass 4: Checking reference counts
Pass 5: Checking group summary information
/dev/hda5: 1011/34136 files (0.1% non-contiguous),
4360/136521 blocks
Example 3
Force another check, this time with verbose
output: # fsck -fv /dev/hda5
Parallelizing fsck version 1.14 (9-Jan-1999)
e2fsck 1.14, 9-Jan-1999 for EXT2 FS 0.5b, 95/08/09
Pass 1: Checking inodes, blocks, and sizes
Pass 2: Checking directory structure
Pass 3: Checking directory connectivity
Pass 4: Checking reference counts
Pass 5: Checking group summary information
1011 inodes used (2%)
1 non-contiguous inodes (0.1%)
# of inodes with ind/dind/tind blocks: 0/0/0
4360 blocks used (3%)
0 bad blocks
1000 regular files
2 directories
0 character device files
0 block device files
0 fifos
0 links
0 symbolic links (0 fast symbolic links)
0 sockets
--------
1002 files
Example 4
Allow fsck to
automatically perform all repairs on a damaged filesystem by
specifying the -y option to run
the command automatically: [root@smp /mnt]# fsck -y /dev/hda5
Parallelizing fsck version 1.14 (9-Jan-1999)
e2fsck 1.14, 9-Jan-1999 for EXT2 FS 0.5b, 95/08/09
Couldn't find ext2 superblock, trying backup blocks...
/dev/hda5 was not cleanly unmounted, check forced.
Pass 1: Checking inodes, blocks, and sizes
Pass 2: Checking directory structure
Pass 3: Checking directory connectivity
Pass 4: Checking reference counts
Pass 5: Checking group summary information
Block bitmap differences: +1 +2 +3 +4
Fix? yes
Inode bitmap differences: +1 +2 +3 +4 +5 +6
Fix? yes
/dev/hda5: ***** FILE SYSTEM WAS MODIFIED *****
/dev/hda5: 1011/34136 files (0.1% non-contiguous),
4360/136521 blocks
When Linux boots, the kernel performs a check
of all filesystems in /etc/fstab using the -A option to fsck. Any
filesystems that were not cleanly unmounted are checked. If
that check finds any significant errors, the system drops into
single-user mode so you can run fsck manually. Unfortunately, unless
you have detailed knowledge of the inner workings of the
filesystem, there's little you can do other than to have fsck do all of the repairs. As a
result, it is common to use the -y option and hope for the best.
|
Familiarity with du, df, and fsck is important. Be sure you
understand the differences between the commands and when
each is used. | 
|