DOKK / manpages / debian 10 / jdupes / jdupes.1.en
JDUPES(1) General Commands Manual JDUPES(1)

jdupes - finds and performs actions upon duplicate files

jdupes [ options ] FILES and/or DIRECTORIES ...

Searches the given path(s) for duplicate files. Such files are found by comparing file sizes, then partial and full file hashes, followed by a byte-by-byte comparison. The default behavior with no other "action options" specified (delete, summarize, link, dedupe, etc.) is to print sets of matching files.

-@ --loud
output annoying low-level debug info while running
-0 --printnull
when printing matches, use null bytes instead of CR/LF bytes, just like 'find -print0' does. This has no effect with any action mode other than the default "print matches" (delete, link, etc. will still print normal line endings in the output.)
-1 --one-file-system
do not match files that are on different filesystems or devices
exclude hidden files from consideration
issue the btrfs same-extents ioctl to trigger a deduplication on disk. The program must be built with btrfs support for this option to be available
set the I/O chunk size manually; larger values may improve performance on rotating media by reducing the number of head seeks required, but also increases memory usage and can reduce performance in some cases
if this feature is compiled in, show debugging statistics and info at the end of program execution
prompt user for files to preserve, deleting all others (see CAVEATS below)
omit the first file in each set of matches
normally, when two or more files point to the same disk area they are treated as non-duplicates; this option will change this behavior
displays help
reverse (invert) the sort order of matches
isolate each command-line parameter from one another; only match if the files are under different parameter specifications
replace all duplicate files with hardlinks to the first file in each set of duplicates
summarize duplicate file information
print matches and summarize the duplicate file information at the end
when used together with --delete, preserve the first file in each set of duplicates and delete the others without prompting the user
exclude zero-length files from consideration; this option is the default behavior and does nothing (also see -z/--zeromatch)
parameter order preservation is more important than the chosen sort; this is particularly useful with the -N option to ensure that automatic deletion behaves in a controllable way
order files according to WORD: time - sort by modification time name - sort by filename (default)
don't consider files with different owner/group or permission bits as duplicates
print extra information to stdout; valid options are: early - matches that pass early size/permission/link/etc. checks partial - files whose partial hashes match fullhash - files whose full hashes match
[WARNING: RISK OF DATA LOSS, SEE CAVEATS] skip byte-for-byte verification of duplicate pairs (use hashes only)
hide progress indicator
for each directory given after this option follow subdirectories encountered within (note the ':' at the end of option; see the Examples section below for further explanation)
for every directory given follow subdirectories encountered within
replace all duplicate files with symlinks to the first file in each set of duplicates
show size of duplicate files
follow symlinked directories
[WARNING: EXTREME RISK OF DATA LOSS, SEE CAVEATS] match based on hash of first block of file data, ignoring the rest
display jdupes version and compilation feature flags
exclude files of size less than SIZE from consideration, or if SIZE is prefixed with a '+' i.e. jdupes -x +226 [files] then exclude files larger than SIZE. Suffixes K/M/G can be used.
exclude files based on specified criteria; supported specs are:
Match only if size is greater (+), less than (-), or equal to (=) the specified number, with an optional multiplier suffix. The +/- and = specifiers can be combined; ex :"size+=:4K" will match if size is greater than or equal to four kilobytes (4096 bytes). Suffixes supported are K/M/G/T/P/E with a B or iB extension (all case-insensitive); no extension or an IB extension specify binary multipliers while a B extension specifies decimal multipliers (ex: 4K or 4KiB = 4096, 4KB = 4000.)
consider zero-length files to be duplicates; this replaces the old default behavior when -n was not specified
if the user aborts the program (as with CTRL-C) act on the matches that were found before the abort was received. For example, if -L and -Z are specified, all matches found prior to the abort will be hard linked. The default behavior without -Z is to abort without taking any actions.

A set of arrows are used in hard linking to show what action was taken on each link candidate. These arrows are as follows:

This file was successfully hard linked to the first file in the duplicate chain
This file was successfully symlinked to the first file in the chain
This file was already a hard link to the first file in the chain
Linking this file failed due to an error during the linking process

Duplicate files are listed together in groups with each file displayed on a separate line. The groups are then separated from each other by blank lines.

will follow subdirectories under b, but not those under a.
will follow subdirectories under both a and b.
will always place 'dir1' results first in any match set (where relevant)

Using -1 or --one-file-system prevents matches that cross filesystems, but a more relaxed form of this option may be added that allows cross-matching for all filesystems that each parameter is present on.

When using -d or --delete, care should be taken to insure against accidental data loss.

-Z or --softabort used to be --hardabort in jdupes prior to v1.5 and had the opposite behavior. Defaulting to taking action on abort is probably not what most users would expect. The decision to invert rather than reassign to a different option was made because this feature was still fairly new at the time of the change.

The -O or --paramorder option allows the user greater control over what appears in the first position of a match set, specifically for keeping the -N option from deleting all but one file in a set in a seemingly random way. All directories specified on the command line will be used as the sorting order of result sets first, followed by the sorting algorithm set by the -o or --order option. This means that the order of all match pairs for a single directory specification will retain the old sorting behavior even if this option is specified.

When used together with options -s or --symlink, a user could accidentally preserve a symlink while deleting the file it points to.

The -Q or --quick option only reads each file once, hashes it, and performs comparisons based solely on the hashes. There is a small but significant risk of a hash collision which is the purpose of the failsafe byte-for-byte comparison that this option explicitly bypasses. Do not use it on ANY data set for which any amount of data loss is unacceptable. This option is not included in the help text for the program due to its risky nature. You have been warned!

The -T or --partial-only option produces results based on a hash of the first block of file data in each file, ignoring everything else in the file. Partial hash checks have always been an important exclusion step in the jdupes algorithm, usually hashing the first 4096 bytes of data and allowing files that are different at the start to be rejected early. In certain scenarios it may be a useful heuristic for a user to see that a set of files has the same size and the same starting data, even if the remaining data does not match; one example of this would be comparing files with data blocks that are damaged or missing such as an incomplete file transfer or checking a data recovery against known-good copies to see what damaged data can be deleted in favor of restoring the known-good copy. This option is meant to be used with informational actions and can result in EXTREME DATA LOSS if used with options that delete files, create hard links, or perform other destructive actions on data based on the matching output. Because of the potential for massive data destruction, this option MUST BE SPECIFIED TWICE to take effect and will error out if it is only specified once.

Using the -C or --chunksize option to override I/O chunk size can increase performance on rotating storage media by reducing "head thrashing," reading larger amounts of data sequentially from each file. This tunable size can have bad side effects; the default size maximizes algorithmic performance without regard to the I/O characteristics of any given device and uses a modest amount of memory, but other values may greatly increase memory usage or incur a lot more system call overhead. Try several different values to see how they affect performance for your hardware and data set. This option does not affect match results in any way, so even if it slows down the file matching process it will not hurt anything.

Send all bug reports to or use the Issue tracker at

jdupes is a fork of 'fdupes' which is maintained by and contains extra code copyrighted by Jody Bruchon <>

Based on 'fdupes' created by Adrian Lopez <>