116

Alright, so simple problem here. I'm working on a simple back up code. It works fine except if the files have spaces in them. This is how I'm finding files and adding them to a tar archive:

find . -type f | xargs tar -czvf backup.tar.gz

The problem is when the file has a space in the name because tar thinks that it's a folder. Basically is there a way I can add quotes around the results from find? Or a different way to fix this?

shgnInc
  • 2,054
  • 1
  • 23
  • 34
Caleb Kester
  • 1,650
  • 2
  • 14
  • 16
  • 12
    The best way to use `find ... | xargs ...` is to use the -print0/-0 parameter on each: `find -print0 ... | xargs -0 ...`. This will cause the filenames to be separated by a null character, which means you can have spaces or newlines or other weird stuff in your filenames and it will still work. – porges May 05 '11 at 02:12
  • 8
    There is a problem with using xargs and tar this way when you have a large number of files, xargs will repeatedly invoke tar -c, and that will keep overwriting your archive, and the result is you won't have all the files you expect. See [this more detailed explanation](http://www.devdaily.com/blog/post/linux-unix/using-find-xargs-tar-create-huge-archive-cygwin-linux-unix) and [my answer](http://stackoverflow.com/a/12305562/296829) below. – Steve Kehlet Sep 06 '12 at 17:43

10 Answers10

228

Use this:

find . -type f -print0 | tar -czvf backup.tar.gz --null -T -

It will:

  • deal with files with spaces, newlines, leading dashes, and other funniness
  • handle an unlimited number of files
  • won't repeatedly overwrite your backup.tar.gz like using tar -c with xargs will do when you have a large number of files

Also see:

Community
  • 1
  • 1
Steve Kehlet
  • 6,156
  • 5
  • 39
  • 40
  • 1
    how would you do this if you wanted to pipe your find through sed a few times first? e.g. find . -print0 | sed /backups/d | tar.... – Brad Parks Nov 15 '13 at 09:56
  • 10
    Note that if have multiple conditions you need to add parenthesis. Otherwise the `-print0` applies to the last expression only. E.g. `find . \( -type f -o -name '*.c' \) -print0 | ...` – nimrodm Apr 28 '15 at 11:53
  • 1
    For fun, here's a Windows version of this using cygwin: `c:\cygwin\bin\find . -regextype posix-egrep -regex '.*(sln^|vcxproj^|filters)$' -print0 | c:\cygwin\bin\tar -cvf MS_Projects.tar --null -T -` – Jon Jan 05 '17 at 20:34
  • 1
    @Steve can you please explain what is '-' option at the end of tar command. I cannot find it in GNU tar's man page. – shaffooo Mar 01 '17 at 15:46
  • Sure thing, it's a parameter to `-T`, and it means read the file names from standard input: If you give a single dash as a file name for `--files-from', (i.e., you specify either --files-from=- or -T -), then the file names are read from standard input – Steve Kehlet Mar 02 '17 at 18:35
  • What does the `--null`? – tommy.carstensen Dec 10 '17 at 14:44
  • @tommy.carstensen it handles the null-separated file names generated by the `-print0` arg to `find`, which makes all this able to handle filenames with spaces, newlines, etc. – Steve Kehlet Dec 11 '17 at 16:32
15

There could be another way to achieve what you want. Basically,

  1. Use the find command to output path to whatever files you're looking for. Redirect stdout to a filename of your choosing.

  2. Then tar with the -T option which allows it to take a list of file locations (the one you just created with find!)

    find . -name "*.whatever" > yourListOfFiles
tar -cvf yourfile.tar -T yourListOfFiles
errorprone
  • 421
  • 7
  • 10
8

Try running:

    find . -type f | xargs -d "\n" tar -czvf backup.tar.gz
gsteff
  • 4,764
  • 1
  • 20
  • 17
7

Why not:

tar czvf backup.tar.gz *

Sure it's clever to use find and then xargs, but you're doing it the hard way.

Update: Porges has commented with a find-option that I think is a better answer than my answer, or the other one: find -print0 ... | xargs -0 ....

Warren P
  • 65,725
  • 40
  • 181
  • 316
  • My full code will back up only items that are modified in the past day. Since its a daily back up I don't want to have repeated information to save on file size (I also have a full back up every 15 days). – Caleb Kester May 06 '11 at 04:18
  • To make this a better SO question, I would ask the question about "reliably using find, xargs, and tar together". Your title and question don't really specify that you need find and xargs, and yet you do. – Warren P May 07 '11 at 20:03
  • `xargs ... tar c ...` will overwrite the first archive created if the list of files is too long and `xargs` will execute `tar` for second time! To avoid overwriting you can use `xargs -x` but then the archive could be incomplete. Alternative could be to first `tar c ...` and then possibly repeatedly `tar r ...`. (my contribution to the reliability :) – pabouk - Ukraine stay strong Oct 12 '13 at 20:41
4

If you have multiple files or directories and you want to zip them into independent *.gz file you can do this. Optional -type f -atime 

find -name "httpd-log*.txt" -type f -mtime +1 -exec tar -vzcf {}.gz {} \;

This will compress 

httpd-log01.txt
httpd-log02.txt

to

httpd-log01.txt.gz
httpd-log02.txt.gz
Kalibur x
  • 41
  • 1
3

Would add a comment to @Steve Kehlet post but need 50 rep (RIP).

For anyone that has found this post through numerous googling, I found a way to not only find specific files given a time range, but also NOT include the relative paths OR whitespaces that would cause tarring errors. (THANK YOU SO MUCH STEVE.)

find . -name "*.pdf" -type f -mtime 0 -printf "%f\0" | tar -czvf /dir/zip.tar.gz --null -T -

  1. . relative directory

  2. -name "*.pdf" look for pdfs (or any file type)

  3. -type f type to look for is a file

  4. -mtime 0 look for files created in last 24 hours

  5. -printf "%f\0" Regular -print0 OR -printf "%f" did NOT work for me. From man pages:

This quoting is performed in the same way as for GNU ls. This is not the same quoting mechanism as the one used for -ls and -fls. If you are able to decide what format to use for the output of find then it is normally better to use '\0' as a terminator than to use newline, as file names can contain white space and newline characters.

  1. -czvf create archive, filter the archive through gzip , verbosely list files processed, archive name

Edit 2019-08-14: I would like to add, that I was also able to use essentially use the same command in my comment, just using tar itself:

tar -czvf /archiveDir/test.tar.gz --newer-mtime=0 --ignore-failed-read *.pdf

Needed --ignore-failed-read in-case there were no new PDFs for today.

user3472383
  • 33
  • 4
2

Why not give something like this a try: tar cvf scala.tar `find src -name *.scala`

Frank Eggink
  • 29
  • 4
2

Another solution as seen here:

find var/log/ -iname "anaconda.*" -exec tar -cvzf file.tar.gz {} +
tommy.carstensen
  • 8,962
  • 15
  • 65
  • 108
1

The best solution seem to be to create a file list and then archive files because you can use other sources and do something else with the list.

For example this allows using the list to calculate size of the files being archived:

#!/bin/sh

backupFileName="backup-big-$(date +"%Y%m%d-%H%M")"
backupRoot="/var/www"
backupOutPath=""

archivePath=$backupOutPath$backupFileName.tar.gz
listOfFilesPath=$backupOutPath$backupFileName.filelist

#
# Make a list of files/directories to archive
#
echo "" > $listOfFilesPath
echo "${backupRoot}/uploads" >> $listOfFilesPath
echo "${backupRoot}/extra/user/data" >> $listOfFilesPath
find "${backupRoot}/drupal_root/sites/" -name "files" -type d >> $listOfFilesPath

#
# Size calculation
#
sizeForProgress=`
cat $listOfFilesPath | while read nextFile;do
    if [ ! -z "$nextFile" ]; then
        du -sb "$nextFile"
    fi
done | awk '{size+=$1} END {print size}'
`

#
# Archive with progress
#
## simple with dump of all files currently archived
#tar -czvf $archivePath -T $listOfFilesPath
## progress bar
sizeForShow=$(($sizeForProgress/1024/1024))
echo -e "\nRunning backup [source files are $sizeForShow MiB]\n"
tar -cPp -T $listOfFilesPath | pv -s $sizeForProgress | gzip > $archivePath
Nux
  • 9,276
  • 5
  • 59
  • 72
0

Big warning on several of the solutions (and your own test) :

When you do : anything | xargs something

xargs will try to fit "as many arguments as possible" after "something", but then you may end up with multiple invocations of "something".

So your attempt: find ... | xargs tar czvf file.tgz may end up overwriting "file.tgz" at each invocation of "tar" by xargs, and you end up with only the last invocation! (the chosen solution uses a GNU -T special parameter to avoid the problem, but not everyone has that GNU tar available)

You could do instead:

find . -type f -print0 | xargs -0 tar -rvf backup.tar
gzip backup.tar

Proof of the problem on cygwin:

$ mkdir test
$ cd test
$ seq 1 10000 | sed -e "s/^/long_filename_/" | xargs touch 
    # create the files
$ seq 1 10000 | sed -e "s/^/long_filename_/" | xargs tar czvf archive.tgz
    # will invoke tar several time as it can'f fit 10000 long filenames into 1
$ tar tzvf archive.tgz | wc -l
60
    # in my own machine, I end up with only the 60 last filenames, 
    # as the last invocation of tar by xargs overwrote the previous one(s)

# proper way to invoke tar: with -r  (which append to an existing tar file, whereas c would overwrite it)
# caveat: you can't have it compressed (you can't add to a compressed archive)
$ seq 1 10000 | sed -e "s/^/long_filename_/" | xargs tar rvf archive.tar #-r, and without z
$ gzip archive.tar
$ tar tzvf archive.tar.gz | wc -l
10000 
  # we have all our files, despite xargs making several invocations of the tar command

Note: that behavior of xargs is a well know diccifulty, and it is also why, when someone wants to do :

find .... | xargs grep "regex"

they intead have to write it:

find ..... | xargs grep "regex" /dev/null

That way, even if the last invocation of grep by xargs appends only 1 filename, grep sees at least 2 filenames (as each time it has: /dev/null, where it won't find anything, and the filename(s) appended by xargs after it) and thus will always display the file names when something maches "regex". Otherwise you may end up with the last results showing matches without a filename in front.

Olivier Dulac
  • 3,695
  • 16
  • 31