Hi, I have found a part of the gid problem. In quite some cases, the gid in the packages was not random but the id of my user marcus. This is because the packages assume the Linux behaviour of mkdir() and open(..., O_CREAT):
ulysses:/tmp# mkdir foo ulysses:/tmp# chgrp marcus foo ulysses:/tmp# ls -l insgesamt 4 drwxr-xr-x 2 root marcus 4096 26. Apr 01:44 foo/ ulysses:/tmp# cd foo ulysses:/tmp/foo# mkdir bar ulysses:/tmp/foo# touch baz ulysses:/tmp/foo# ls -l insgesamt 4 drwxr-xr-x 2 root root 4096 26. Apr 01:44 bar/ -rw-r--r-- 1 root root 0 26. Apr 01:44 baz ulysses:/tmp/foo# ls -l .. insgesamt 4 drwxr-xr-x 3 root marcus 4096 26. Apr 01:44 foo/ While they will fail on the Hurd behaviour: ulysses:/tmp# mkdir foo ulysses:/tmp# chgrp marcus foo ulysses:/tmp# ls -l insgesamt 4 drwxr-xr-x 2 root marcus 4096 26. Apr 01:44 foo/ ulysses:/tmp# cd foo ulysses:/tmp/foo# mkdir bar ulysses:/tmp/foo# touch baz ulysses:/tmp/foo# ls -l insgesamt 4 drwxr-xr-x 2 root marcus 4096 26. Apr 01:44 bar/ -rw-r--r-- 1 root marcus 0 26. Apr 01:44 baz ulysses:/tmp/foo# ls -l .. insgesamt 4 drwxr-xr-x 3 root marcus 4096 26. Apr 01:44 foo/ The POSIX standard says (in mkdir and O_CREAT): The directory's user ID shall be set to the process' effective user ID. The directory's group ID shall be set to the group ID of the parent directory or to the effective group ID of the process. Implementations shall provide a way to initialize the directory's group ID to the group ID of the parent directory. Implementations may, but need not, provide an implementation-defined way to initialize the directory's group ID to the effective group ID of the calling process. So, both behaviours are correct according to the standard, and this is clearly a bug in the software that it assumes that the group ID of a node created by a root owned process is root. The packages which make this assumption need to be fixed to at least create the top level directory in which they install the final files with a proper group ID. The requirement in the second paragraph seems just to be the sgid flag. The optional feature is not provided by us, or is it? Although our behaviour is standard conform, I want to verify with you that it is really what we want. One remark is that defaulting to giving it the group id of the parent directory makes the sgid flag a bit pointless. So it seems to be more featureful to default to giving the gid the gid of the process. This also will be in line with the popular Linux kernel, so is les surprising for many users coming from there, and relieves us from fixing the broken package build scripts (I am not sure how many there are. Potentially many, but as long as you build as root in root owned directories we won't easily notice most of them). I guess we are imitating some BSD behaviour here? What do modern BSDs do? Thanks, Marcus -- `Rhubarb is no Egyptian god.' Debian http://www.debian.org [EMAIL PROTECTED] Marcus Brinkmann GNU http://www.gnu.org [EMAIL PROTECTED] [EMAIL PROTECTED] http://www.marcus-brinkmann.de _______________________________________________ Bug-hurd mailing list [EMAIL PROTECTED] http://mail.gnu.org/mailman/listinfo/bug-hurd
