DragonFly On-Line Manual Pages
DUPEMAP(1) Magic Rescue DUPEMAP(1)
NAME
dupemap - Creates a database of file checksums and uses it to eliminate
duplicates
SYNOPSIS
dupemap [ options ] [ -d database ] operation path...
DESCRIPTION
dupemap recursively scans each path to find checksums of file contents.
Directories are searched through in no particular order. Its actions
depend on whether the -d option is given, and on the operation
parameter, which must be a comma-seperated list of scan, report,
delete:
Without -d
dupemap will take action when it sees the same checksum repeated more
than once, i.e. it simply finds duplicates recursively. The action
depends on operation:
report Report what files are encountered more than once, printing their
names to standard output.
delete[,report]
Delete files that are encountered more than once. Print their
names if report is also given.
WARNING: use the report operation first to see what will be
deleted.
WARNING: You are advised to make a backup of the target first,
e.g. with "cp -al" (for GNU cp) to create hard links
recursively.
With -d
The database argument to -d will denote a database file (see the
"DATABASE" section in this manual for details) to read from or write
to. In this mode, the scan operation should be run on one path,
followed by the report or delete operation on another (not the same!)
path.
scan Add the checksum of each file to database. This operation must
be run initially to create the database. To start over, you
must manually delete the database file(s) (see the "DATABASE"
section).
report Print each file name if its checksum is found in database.
delete[,report]
Delete each file if its checksum is found in database. If
report is also present, print the name of each deleted file.
WARNING: if you run dupemap delete on the same path you just ran
dupemap scan on, it will delete every file! The idea of these
options is to scan one path and delete files in a second path.
WARNING: use the report operation first to see what will be
deleted.
WARNING: You are advised to make a backup of the target first,
e.g. with "cp -al" (for GNU cp) to create hard links
recursively.
OPTIONS
-d database
Use database as an on-disk database to read from or write to.
See the "DESCRIPTION" section above about how this influences
the operation of dupemap.
-I file
Reads input files from file in addition to those listed on the
command line. If file is "-", read from standard input. Each
line will be interpreted as a file name.
The paths given here will NOT be scanned recursively.
Directories will be ignored and symlinks will be followed.
-m minsize
Ignore files below this size.
-M maxsize
Ignore files above this size.
USAGE
General usage
The easiest operations to understand is when the -d option is not
given. To delete all duplicate files in /tmp/recovered-files, do:
$ dupemap delete /tmp/recovered-files
Often, dupemap scan is run to produce a checksum database of all files
in a directory tree. Then dupemap delete is run on another directory,
possibly following dupemap report. For example, to delete all files in
/tmp/recovered-files that already exist in $HOME, do this:
$ dupemap -d homedir.map scan $HOME
$ dupemap -d homedir.map delete,report /tmp/recovered-files
Usage with magicrescue
The main application for dupemap is to take some pain out of performing
undelete operations with magicrescue(1). The reason is that
magicrescue will extract every single file of the specified type on the
block device, so undeleting files requires you to find a few files out
of hundreds, which can take a long time if done manually. What we want
to do is to only extract the documents that don't exist on the file
system already.
In the following scenario, you have accidentally deleted some important
Word documents in Windows. If this were a real-world scenario, then by
all means use The Sleuth Kit. However, magicrescue will work even when
the directory entries were overwritten, i.e. more files were stored in
the same folder later.
You boot into Linux and change to a directory with lots of space.
Mount the Windows partition, preferably read-only (especially with
NTFS), and create the directories we will use.
$ mount -o ro /dev/hda1 /mnt/windows
$ mkdir healthy_docs rescued_docs
Extract all the healthy Word documents with magicrescue and build a
database of their checksums. It may seem a little redundant to send
all the documents through magicrescue first, but the reason is that
this process may modify them (e.g. stripping trailing garbage), and
therefore their checksum will not be the same as the original
documents. Also, it will find documents embedded inside other files,
such as uncompressed zip archives or files with the wrong extension.
$ find /mnt/windows -type f \
|magicrescue -I- -r msoffice -d healthy_docs
$ dupemap -d healthy_docs.map scan healthy_docs
$ rm -rf healthy_docs
Now rescue all "msoffice" documents from the block device and get rid
of everything that's not a *.doc.
$ magicrescue -Mo -r msoffice -d rescued_docs /dev/hda1 \
|grep -v '\.doc$'|xargs rm -f
Remove all the rescued documents that also appear on the file system,
and remove duplicates.
$ dupemap -d healthy_docs.map delete,report rescued_docs
$ dupemap delete,report rescued_docs
The rescued_docs folder should now contain only a few files. This will
be the undeleted files and some documents that were not stored in
contiguous blocks (use that defragger ;-)).
Usage with fsck
In this scenario (based on a true story), you have a hard disk that's
gone bad. You have managed to dd about 80% of the contents into the
file diskimage, and you have an old backup from a few months ago. The
disk is using reiserfs on Linux.
First, use fsck to make the file system usable again. It will find
many nameless files and put them in lost+found. You need to make sure
there is some free space on the disk image, so fsck has something to
work with.
$ cp diskimage diskimage.bak
$ dd if=/dev/zero bs=1M count=2048 >> diskimage
$ reiserfsck --rebuild-tree diskimage
$ mount -o loop diskimage /mnt
$ ls /mnt/lost+found
(tons of files)
Our strategy will be to restore the system with the old backup as a
base and merge the two other sets of files (/mnt/lost+found and /mnt)
into the backup after eliminating duplicates. Therefore we create a
checksum database of the directory we have unpacked the backup in.
$ dupemap -d backup.map scan ~/backup
Next, we eliminate all the files from the rescued image that are also
present in the backup.
$ dupemap -d backup.map delete,report /mnt
We also want to remove duplicates from lost+found, and we want to get
rid of any files that are also present in the other directories in
/mnt.
$ dupemap delete,report /mnt/lost+found
$ ls /mnt|grep -v lost+found|xargs dupemap -d mnt.map scan
$ dupemap -d mnt.map delete,report /mnt/lost+found
This should leave only the files in /mnt that have changed since the
last backup or got corrupted. Particularly, the contents of
/mnt/lost+found should now be reduced enough to manually sort through
them (or perhaps use magicsort(1)).
Primitive intrusion detection
You can use dupemap to see what files change on your system. This is
one of the more exotic uses, and it's only included for inspiration.
First, you map the whole file system.
$ dupemap -d old.map scan /
Then you come back a few days/weeks later and run dupemap report. This
will give you a view of what has not changed. To see what has changed,
you need a list of the whole file system. You can get this list along
with preparing a new map easily. Both lists need to be sorted to be
compared.
$ dupemap -d old.map report /|sort > unchanged_files
$ dupemap -d current.map scan /|sort > current_files
All that's left to do is comparing these files and preparing for next
week. This assumes that the dbm appends the ".db" extension to
database files.
$ diff unchanged_files current_files > changed_files
$ mv current.map.db old.map.db
DATABASE
The actual database file(s) written by dupecheck will have some
relation to the database argument, but most implementations append an
extension. For example, Berkeley DB names the files database.db, while
Solaris and GDBM creates both a database.dir and database.pag file.
dupecheck depends on a database library for storing the checksums. It
currently requires the POSIX-standardized ndbm library, which must be
present on XSI-compliant UNIXes. Implementations are not required to
handle hash key collisions, and a faliure to do that could make
dupecheck delete too many files. I haven't heard of such an
implementation, though.
The current checksum algorithm is the file's CRC32 combined with its
size. Both values are stored in native byte order, and because of
varying type sizes the database is not portable across architectures,
compilers and operating systems.
SEE ALSO
magicrescue(1), weeder(1)
This tool does the same thing weeder does, except that weeder cannot
seem to handle many files without crashing, and it has no largefile
support.
BUGS
There is a tiny chance that two different files can have the same
checksum and size. The probability of this happening is around 1 to
10^14, and since dupemap is part of the Magic Rescue package, which
deals with disaster recovery, that chance becomes an insignificant part
of the game. You should consider this if you apply dupemap to other
applications, especially if they are security-related (see next
paragraph).
It is possible to craft a file to have a known CRC32. You need to keep
this in mind if you use dupemap on untrusted data. A solution to this
could be to implement an option for using MD5 checksums instead.
AUTHOR
Jonas Jensen <jbj@knef.dk>
LATEST VERSION
This tool is part of Magic Rescue. You can find the latest version at
<http://jbj.rapanden.dk/magicrescue/>
1.1.9 2008-06-26 DUPEMAP(1)