Retaining file permissions with Git

问题:

I want to version control my web server as described in Version control for my web server, by creating a git repo out of my /var/www directory. My hope was that I would then be able to push web content from our dev server to github, pull it to our production server, and spend the rest of the day at the pool.

Apparently a kink in my plan is that Git won't respect file permissions (I haven't tried it, only reading about it now.) I guess this makes sense in that different boxes are liable to have different user/group setups. But if I wanted to force permissions to propagate, knowing my servers are configured the same, do I have any options? Or is there an easier way to approach what I'm trying to do?

回答1:

The git-cache-meta mentioned in SO question "git - how to recover the file permissions git thinks the file should be?" (and the git FAQ) is the more staightforward approach.

The idea is to store in a .git_cache_meta file the permissions of the files and directories.
It is a separate file not versioned directly in the Git repo.

That is why the usage for it is:

$ git bundle create mybundle.bdl master; git-cache-meta --store
$ scp mybundle.bdl .git_cache_meta machine2: 
#then on machine2:
$ git init; git pull mybundle.bdl master; git-cache-meta --apply

So you:

• bundle your repo and save the associated file permissions.
• copy those two files on the remote server
• restore the repo there, and apply the permission

回答2:

Git is Version Control System, created for software development, so from the whole set of modes and permissions it stores only executable bit (for ordinary files) and symlink bit. If you want to store full permissions, you need third party tool, like git-cache-meta (mentioned by VonC), or Metastore (used by etckeeper). Or you can use IsiSetup, which IIRC uses git as backend.

See Interfaces, frontends, and tools page on Git Wiki.

回答3:

This is quite late but might help some others. I do what you want to do by adding two git hooks to my repository.

.git/hooks/pre-commit:

#!/bin/bash
#
# A hook script called by "git commit" with no arguments. The hook should
# exit with non-zero status after issuing an appropriate message if it wants
# to stop the commit.

SELF_DIR=`git rev-parse --show-toplevel`
DATABASE=$SELF_DIR/.permissions

# Clear the permissions database file
> $DATABASE

echo -n "Backing-up file permissions..."

IFS_OLD=$IFS; IFS=$'\n'
for FILE in `git ls-files`
do
   # Save the permissions of all the files in the index
   echo $FILE";"`stat -c "%a;%U;%G" $FILE` >> $DATABASE
done
IFS=$IFS_OLD

# Add the permissions database file to the index
git add $DATABASE

echo "OK"

.git/hooks/post-checkout:

#!/bin/bash

SELF_DIR=`git rev-parse --show-toplevel`
DATABASE=$SELF_DIR/.permissions

echo -n "Restoring file permissions..."

IFS_OLD=$IFS; IFS=$'\n'
while read -r LINE || [[ -n "$LINE" ]];
do
   FILE=`echo $LINE | cut -d ";" -f 1`
   PERMISSIONS=`echo $LINE | cut -d ";" -f 2`
   USER=`echo $LINE | cut -d ";" -f 3`
   GROUP=`echo $LINE | cut -d ";" -f 4`

   # Set the file permissions
   chmod $PERMISSIONS $FILE

   # Set the file owner and groups
   chown $USER:$GROUP $FILE

done < $DATABASE
IFS=$IFS_OLD

echo "OK"

exit 0

The first hook is called when you "commit" and will read the ownership and permissions for all the files in the repository and store them in a file in the root of the repository called .permissions and then add the .permissions file to the commit.

The second hook is called when you "checkout" and will go through the list of files in the .permissions file and restore the ownership and permissions of those files.

• You might need to do the commit and checkout using sudo.
• Make sure the pre-commit and post-checkout scripts have execution permission.

回答4:

In case you are coming into this right now, I've just been through it today and can summarize where this stands. If you did not try this yet, some details here might help.

I think @Omid Ariyan's approach is the best way. Add the pre-commit and post-checkout scripts. DON'T forget to name them exactly the way Omid does and DON'T forget to make them executable. If you forget either of those, they have no effect and you run "git commit" over and over wondering why nothing happens :) Also, if you cut and paste out of the web browser, be careful that the quotation marks and ticks are not altered.

If you run the pre-commit script once (by running a git commit), then the file .permissions will be created. You can add it to the repository and I think it is unnecessary to add it over and over at the end of the pre-commit script. But it does not hurt, I think (hope).

There are a few little issues about the directory name and the existence of spaces in the file names in Omid's scripts. The spaces were a problem here and I had some trouble with the IFS fix. For the record, this pre-commit script did work correctly for me:

#!/bin/bash  

SELF_DIR=`git rev-parse --show-toplevel`
DATABASE=$SELF_DIR/.permissions

# Clear the permissions database file
> $DATABASE

echo -n "Backing-up file permissions..."

IFSold=$IFS
IFS=$'\n'
for FILE  in `git ls-files`
do
   # Save the permissions of all the files in the index
   echo $FILE";"`stat -c "%a;%U;%G" $FILE` >> $DATABASE
done
IFS=${IFSold}
# Add the permissions database file to the index
git add $DATABASE

echo "OK"

Now, what do we get out of this?

The .permissions file is in the top level of the git repo. It has one line per file, here is the top of my example:

$ cat .permissions
.gitignore;660;pauljohn;pauljohn
05.WhatToReport/05.WhatToReport.doc;664;pauljohn;pauljohn
05.WhatToReport/05.WhatToReport.pdf;664;pauljohn;pauljohn

As you can see, we have

filepath;perms;owner;group

In the comments about this approach, one of the posters complains that it only works with same username, and that is technically true, but it is very easy to fix it. Note the post-checkout script has 2 action pieces,

# Set the file permissions
chmod $PERMISSIONS $FILE
# Set the file owner and groups
chown $USER:$GROUP $FILE

So I am only keeping the first one, that's all I need. My user name on the Web server is indeed different, but more importantly you can't run chown unless you are root. Can run "chgrp", however. It is plain enough how to put that to use.

In the first answer in this post, the one that is most widely accepted, the suggestion is so use git-cache-meta, a script that is doing the same work that the pre/post hook scripts here are doing (parsing output from git ls-files). These scripts are easier for me to understand, the git-cache-meta code is rather more elaborate. It is possible to keep git-cache-meta in the path and write pre-commit and post-checkout scripts that would use it.

Spaces in file names are a problem with both of Omid's scripts. In the post-checkout script, you'll know you have the spaces in file names if you see errors like this

$ git checkout -- upload.sh
Restoring file permissions...chmod: cannot access  '04.StartingValuesInLISREL/Open': No such file or directory
chmod: cannot access 'Notebook.onetoc2': No such file or directory
chown: cannot access '04.StartingValuesInLISREL/Open': No such file or directory
chown: cannot access 'Notebook.onetoc2': No such file or directory

I'm checking on solutions for that. Here's something that seems to work, but I've only tested in one case

#!/bin/bash

SELF_DIR=`git rev-parse --show-toplevel`
DATABASE=$SELF_DIR/.permissions

echo -n "Restoring file permissions..."
IFSold=${IFS}
IFS=$
while read -r LINE || [[ -n "$LINE" ]];
do
   FILE=`echo $LINE | cut -d ";" -f 1`
   PERMISSIONS=`echo $LINE | cut -d ";" -f 2`
   USER=`echo $LINE | cut -d ";" -f 3`
   GROUP=`echo $LINE | cut -d ";" -f 4`

   # Set the file permissions
   chmod $PERMISSIONS $FILE
   # Set the file owner and groups
   chown $USER:$GROUP $FILE
done < $DATABASE
IFS=${IFSold}
echo "OK"

exit 0

Since the permissions information is one line at a time, I set IFS to $, so only line breaks are seen as new things.

I read that it is VERY IMPORTANT to set the IFS environment variable back the way it was! You can see why a shell session might go badly if you leave $ as the only separator.

回答5:

In pre-commit/post-checkout an option would be to use "mtree" (FreeBSD), or "fmtree" (Ubuntu) utility which "compares a file hierarchy against a specification, creates a specification for a file hierarchy, or modifies a specification."

The default set are flags, gid, link, mode, nlink, size, time, type, and uid. This can be fitted to the specific purpose with -k switch.

回答6:

I am running on FreeBSD 11.1, the freebsd jail virtualization concept makes the operating system optimal. The current version of Git I am using is 2.15.1, I also prefer to run everything on shell scripts. With that in mind I modified the suggestions above as followed:

git push: .git/hooks/pre-commit

#! /bin/sh -
#
# A hook script called by "git commit" with no arguments. The hook should
# exit with non-zero status after issuing an appropriate message if it wants
# to stop the commit.

SELF_DIR=$(git rev-parse --show-toplevel);
DATABASE=$SELF_DIR/.permissions;

# Clear the permissions database file
> $DATABASE;

printf "Backing-up file permissions...\n";

OLDIFS=$IFS;
IFS=$'\n';
for FILE in $(git ls-files);
do
   # Save the permissions of all the files in the index
    printf "%s;%s\n" $FILE $(stat -f "%Lp;%u;%g" $FILE) >> $DATABASE;
done
IFS=$OLDIFS;

# Add the permissions database file to the index
git add $DATABASE;

printf "OK\n";

git pull: .git/hooks/post-merge

#! /bin/sh -

SELF_DIR=$(git rev-parse --show-toplevel);
DATABASE=$SELF_DIR/.permissions;

printf "Restoring file permissions...\n";

OLDIFS=$IFS;
IFS=$'\n';
while read -r LINE || [ -n "$LINE" ];
do
   FILE=$(printf "%s" $LINE | cut -d ";" -f 1);
   PERMISSIONS=$(printf "%s" $LINE | cut -d ";" -f 2);
   USER=$(printf "%s" $LINE | cut -d ";" -f 3);
   GROUP=$(printf "%s" $LINE | cut -d ";" -f 4);

   # Set the file permissions
   chmod $PERMISSIONS $FILE;

   # Set the file owner and groups
   chown $USER:$GROUP $FILE;

done < $DATABASE
IFS=$OLDIFS

pritnf "OK\n";

exit 0;

If for some reason you need to recreate the script the .permissions file output should have the following format:

.gitignore;644;0;0

For a .gitignore file with 644 permissions given to root:wheel

Notice I had to make a few changes to the stat options.

Enjoy,

回答7:

One addition to @Omid Ariyan's answer is permissions on directories. Add this after the for loop's done in his pre-commit script.

for DIR in $(find ./ -mindepth 1 -type d -not -path "./.git" -not -path "./.git/*" | sed 's@^\./@@')
do
    # Save the permissions of all the files in the index
    echo $DIR";"`stat -c "%a;%U;%G" $DIR` >> $DATABASE
done

This will save directory permissions as well.