rsync - Oct 2009 - --copy-unsafe-links does not work for "double" symlinks

Hi,

as documented, if you use rsync with --copy-unsafe-links, and copy a
directory with a symlink pointing outside of the copied tree, the
referent of the symlink is copied.

Now, assume that the directory contains a symlink, which points to a
(2nd) directory outside of the copied tree.
If this (2nd) directory contains another symlink, which also points outside
of the copied tree, rsync unfortunately only copies this symlink and *not*
the referent.

I don't know if this is a rsync-problem or a symlink-problem.

(rsync  version 3.0.3  protocol version 30, on Debian Lenny)


Example:

- host 1 directory-structure ("empty"):
    drwx------ ./1

- host 2 directory-structure:
    drwx------ ./2
    -rw------- ./2/file
    lrwxrwxrwx ./2/dir -> ../foo
    drwx------ ./foo
    -rw------- ./foo/foofile
    lrwxrwxrwx ./foo/bar -> ../bar
    drwx------ ./bar
    -rw------- ./bar/barfile

- mirror host 2 to host 1:
    $ rsync -av --copy-unsafe-links 2/ 1/
    sending incremental file list
    ./
    file
    dir/
    dir/bar -> ../bar/
    dir/foofile

    sent 206 bytes  received 60 bytes  532.00 bytes/sec
    total size is 7  speedup is 0.03

Now, host 1 contains:
    drwx------ ./1
    -rw------- ./1/file
    drwx------ ./1/dir
    -rw------- ./1/dir/foofile
    lrwxrwxrws ./1/dir/bar -> ../bar
"file", "dir" and "foofile" are ok; but
"bar" is a dangling symlink, and
"barfile" was not copied.

I would have expected that host 1 would contain:
    drwx------ ./1
    -rw------- ./1/file
    drwx------ ./1/dir
    -rw------- ./1/dir/foofile
    drwx------ ./1/dir/bar/
    -rw------- ./1/dir/bar/barfile
But this is not the case.

Is this a bug, or did I miss something?


thanks,
Roland

On Tue, 2009-10-06 at 01:47 +0200, Roland Koebler wrote:
> as documented, if you use rsync with --copy-unsafe-links, and copy a
> directory with a symlink pointing outside of the copied tree, the
> referent of the symlink is copied.
> 
> Now, assume that the directory contains a symlink, which points to a
> (2nd) directory outside of the copied tree.
> If this (2nd) directory contains another symlink, which also points outside
> of the copied tree, rsync unfortunately only copies this symlink and *not*
> the referent.
> Example:
> 
> - host 1 directory-structure ("empty"):
>     drwx------ ./1
> 
> - host 2 directory-structure:
>     drwx------ ./2
>     -rw------- ./2/file
>     lrwxrwxrwx ./2/dir -> ../foo
>     drwx------ ./foo
>     -rw------- ./foo/foofile
>     lrwxrwxrwx ./foo/bar -> ../bar
>     drwx------ ./bar
>     -rw------- ./bar/barfile
> 
> - mirror host 2 to host 1:
>     $ rsync -av --copy-unsafe-links 2/ 1/
>     sending incremental file list
>     ./
>     file
>     dir/
>     dir/bar -> ../bar/
>     dir/foofile
> 
>     sent 206 bytes  received 60 bytes  532.00 bytes/sec
>     total size is 7  speedup is 0.03
> 
> Now, host 1 contains:
>     drwx------ ./1
>     -rw------- ./1/file
>     drwx------ ./1/dir
>     -rw------- ./1/dir/foofile
>     lrwxrwxrws ./1/dir/bar -> ../bar
> "file", "dir" and "foofile" are ok; but
"bar" is a dangling symlink, and
> "barfile" was not copied.
Ah, yes.  The problem is that rsync is evaluating the safety of the
symlink it sees at "dir/bar" as if "dir" were a real
directory, ini
which case the symlink would indeed be safe.  To fix this, rsync would
need to remember the last component of the file-list path that was an
unsafe symlink and count only the path components after that point in
symlink safety checks.

I've made a first attempt at implementing this in the
"wip/copy-unsafe-links-double" branch of my rsync repository at:

https://mattmccutchen.net/rsync/rsync.git/

-- 
Matt