Storing ownership / device nodes without root
Wesley W.Terpstra
wesley at terpstra.ca
Sat Oct 28 15:15:06 GMT 2006
On Oct 28, 2006, at 12:58 AM, Wayne Davison wrote:
> Yeah, the error checking is not up-to-snuff yet.
Sorry, when you said you'd checked it in, I assumed that meant you
were done with it and the final version was committed.
Now I've seen:
http://archives.free.net.ph/thread/
20061028.062640.ffddea90.en.html#20061028.062640.ffddea90
that nothing could be farther from the truth, and you don't sleep at
all!
>> When I use lsetxattr on a symlink, I get EPERM:
>
> Me too. I currently have symlinks disabled so that we never try to
> set
> the xattr-stat value. It might be better to just silence the error in
> the future.
Since xattr on osx symlinks succeeds, that seems the better choice.
> The latest code
> ensures that the umask can't ask to restrict the file-owner's
> permissions, so this will not happen (and I went ahead and moved the
> set_stat_xattr() call up higher, just in case the ACL settings mess
> things up).
Right. That's a good idea.
> Thanks for the feedback, and please feel free to offer more hints,
> criticisms, and/or patches.
I would send patches, but I find it quite difficult to get patches
against patches.
Is there some sort of script you're using to do this?
For that matter, how are you keeping such a large patchset consistent
when you modify cvs??
Anyway, I just did a number of tests using --link-dest together with
--fake-super (this being the eventual goal after all!)
I created these directories:
kiwiw:~/rsync/tests terpstra$ ls -l base mod
base:
total 20
-rw-r--r-- 1 root terpstra 6 Oct 28 16:40 chgrp
-rw-r--r-- 1 root terpstra 6 Oct 28 16:40 chmod-high-
bits-change
-rw-r--r-- 1 root terpstra 6 Oct 28 16:40 chmod-low-
bits-change
-rw-r--r-- 1 root terpstra 6 Oct 28 16:40 chown
-rw-r--r-- 1 root terpstra 0 Oct 28 16:40 deleted-file
brw-r--r-- 1 root terpstra 14, 6 Oct 28 16:40 device-
becomes-file
brw-r--r-- 1 root terpstra 14, 6 Oct 28 16:40 device-changes
-rw-r--r-- 1 root terpstra 0 Oct 28 16:40 file-becomes-
device
-rw-r--r-- 1 root terpstra 0 Oct 28 16:40 file-becomes-
symlink
lrwxr-xr-x 1 root terpstra 7 Oct 28 16:40 symlink-
becomes-file -> nowhere
-rw-r--r-- 1 root terpstra 5 Oct 28 16:40 unchanged-file
mod:
total 20
-rw-r--r-- 1 root bin 6 Oct 28 16:40 chgrp
-rw-r--r-T 1 root terpstra 6 Oct 28 16:40 chmod-high-
bits-change
-rw-r--rw- 1 root terpstra 6 Oct 28 16:40 chmod-low-
bits-change
-rw-r--r-- 1 www terpstra 6 Oct 28 16:40 chown
-rw-r--r-- 1 root terpstra 0 Oct 28 16:40 created-file
-rw-r--r-- 1 root terpstra 0 Oct 28 16:40 device-
becomes-file
brw-r--r-- 1 root terpstra 14, 7 Oct 28 16:40 device-changes
brw-r--r-- 1 root terpstra 14, 6 Oct 28 16:40 file-becomes-
device
lrwxr-xr-x 1 root terpstra 7 Oct 28 16:40 file-becomes-
symlink -> nowhere
-rw-r--r-- 1 root terpstra 0 Oct 28 16:40 symlink-
becomes-file
-rw-r--r-- 1 root terpstra 5 Oct 28 16:40 unchanged-file
Then sync'd them to the remote end and back with:
kiwiw:~/rsync/tests terpstra$ sudo rsync -a base/ pumpkin::backup/
test/base
kiwiw:~/rsync/tests terpstra$ sudo rsync -a --link-dest=../base mod/
pumpkin::backup/test/mod
kiwiw:~/rsync/tests terpstra$ sudo rsync -a pumpkin::backup/test/mod/
mod2
As hoped, the only file on the remote end with a common inode was
'unchanged-file':
pumpkin:/backup# ls -li test/base test/mod
test/base:
total 60
25231365 -rw-r--r-- 1 backup root 6 Oct 28 16:40 chgrp
25231366 -rw-r--r-- 1 backup root 6 Oct 28 16:40
chmod-high-bits-change
25231367 -rw-r--r-- 1 backup root 6 Oct 28 16:40
chmod-low-bits-change
25231368 -rw-r--r-- 1 backup root 6 Oct 28 16:40 chown
25231369 -rw-r--r-- 1 backup root 0 Oct 28 16:40
deleted-file
25231362 -rw-r--r-- 1 backup root 0 Oct 28 16:40
device-becomes-file
25231363 -rw-r--r-- 1 backup root 0 Oct 28 16:40
device-changes
25231370 -rw-r--r-- 1 backup root 0 Oct 28 16:40 file-
becomes-device
25231371 -rw-r--r-- 1 backup root 0 Oct 28 16:40 file-
becomes-symlink
25231364 lrwxrwxrwx 1 backup root 7 Oct 28 16:59
symlink-becomes-file -> nowhere
25231372 -rw-r--r-- 2 backup root 5 Oct 28 16:40
unchanged-file
test/mod:
total 60
25264130 -rw-r--r-- 1 backup root 6 Oct 28 16:40 chgrp
25264131 -rw-r--r-- 1 backup root 6 Oct 28 16:40
chmod-high-bits-change
25264132 -rw-r--rw- 1 backup root 6 Oct 28 16:40
chmod-low-bits-change
25264133 -rw-r--r-- 1 backup root 6 Oct 28 16:40 chown
25264137 -rw-r--r-- 1 backup root 0 Oct 28 16:40
created-file
25264138 -rw-r--r-- 1 backup root 0 Oct 28 16:40
device-becomes-file
25264134 -rw-r--r-- 1 backup root 0 Oct 28 16:40
device-changes
25264135 -rw-r--r-- 1 backup root 0 Oct 28 16:40 file-
becomes-device
25264136 lrwxrwxrwx 1 backup root 7 Oct 28 16:59 file-
becomes-symlink -> nowhere
25264139 -rw-r--r-- 1 backup root 0 Oct 28 16:40
symlink-becomes-file
25231372 -rw-r--r-- 2 backup root 5 Oct 28 16:40
unchanged-file
Also, it exposed the correct low-bit chmod changes, while masking out
the high bits.
Back on the local machine, the restore worked perfectly (except for
symlinks - grrr!):
kiwiw:~/rsync/tests terpstra$ ls -l mod mod2
mod:
total 20
-rw-r--r-- 1 root bin 6 Oct 28 16:40 chgrp
-rw-r--r-T 1 root terpstra 6 Oct 28 16:40 chmod-high-
bits-change
-rw-r--rw- 1 root terpstra 6 Oct 28 16:40 chmod-low-
bits-change
-rw-r--r-- 1 www terpstra 6 Oct 28 16:40 chown
-rw-r--r-- 1 root terpstra 0 Oct 28 16:40 created-file
-rw-r--r-- 1 root terpstra 0 Oct 28 16:40 device-
becomes-file
brw-r--r-- 1 root terpstra 14, 7 Oct 28 16:40 device-changes
brw-r--r-- 1 root terpstra 14, 6 Oct 28 16:40 file-becomes-
device
lrwxr-xr-x 1 root terpstra 7 Oct 28 16:40 file-becomes-
symlink -> nowhere
-rw-r--r-- 1 root terpstra 0 Oct 28 16:40 symlink-
becomes-file
-rw-r--r-- 1 root terpstra 5 Oct 28 16:40 unchanged-file
mod2:
total 20
-rw-r--r-- 1 root bin 6 Oct 28 16:40 chgrp
-rw-r--r-T 1 root terpstra 6 Oct 28 16:40 chmod-high-
bits-change
-rw-r--rw- 1 root terpstra 6 Oct 28 16:40 chmod-low-
bits-change
-rw-r--r-- 1 www terpstra 6 Oct 28 16:40 chown
-rw-r--r-- 1 root terpstra 0 Oct 28 16:40 created-file
-rw-r--r-- 1 root terpstra 0 Oct 28 16:40 device-
becomes-file
brw-r--r-- 1 root terpstra 14, 7 Oct 28 16:40 device-changes
brw-r--r-- 1 root terpstra 14, 6 Oct 28 16:40 file-becomes-
device
lrwxrwxrwx 1 1001 wheel 7 Oct 28 17:02 file-becomes-
symlink -> nowhere
-rw-r--r-- 1 root terpstra 0 Oct 28 16:40 symlink-
becomes-file
-rw-r--r-- 1 root terpstra 5 Oct 28 16:40 unchanged-file
I also tried the same tests without --fake-super. In that case, the
files chown and chgrp also share an inode. The sticky bit was
exposed, and symlink permissions were still broken.
Conclusion: it works! (mostly)
The only regression I saw over the normal rsync behaviour is the
strange change to the symlink owner. I imagine this is because we
fail to store the owner in the xattr of the symlink (target machine
is linux), and on the way back we therefore use the owner of the file
on the --fake-super machine (1001 in this case). The local end,
having real root, happily sets the symlink owner to 1001, which is
wrong.
Do you see anything else?
More information about the rsync
mailing list