Re: sharedrepository=group not working
On Dec 3, 2018, at 8:50 PM, Jeff King wrote: > > I don't suppose this is leaving those incoming-* directories sitting > around so we can inspect their permissions (it's suppose to clean them > up, so I doubt it). If you're up for it, it might be interesting to > patch Git to inspect the umask and "ls -l" the objects/ directory at the > problematic moment. The interesting point is when we call into > tmp-objdir.c:setup_tmp_objdir(). The problem was that Apache was setting my umask to 113! With that: + ls -ldF ./objects/incoming-w7agmb/pack objects/incoming-w7agmb ls: cannot access ./objects/incoming-w7agmb/pack: Permission denied drw---S--- 2 apache cvs 4096 Dec 3 21:14 objects/incoming-w7agmb/ error: remote unpack failed: unable to create temporary object directory With 002 it succeeds: + ls -ldF ./objects/incoming-IbGS6h/pack objects/incoming-IbGS6h drwx--S--- 3 apache cvs 4096 Dec 3 21:19 objects/incoming-IbGS6h/ drwxrwsr-x 2 apache cvs 4096 Dec 3 21:19 ./objects/incoming-IbGS6h/pack/ So I fixed my umask and got it working, but maybe a test for "your umask is dumb" is worthwhile. Thanks for your help! -- Jamie Zawinski https://www.jwz.org/ https://www.dnalounge.com/
Re: sharedrepository=group not working
On Mon, Dec 03, 2018 at 08:19:12PM -0800, Jamie Zawinski wrote: > On Dec 3, 2018, at 8:09 PM, Jeff King wrote: > > > > but it works fine. Might there be some effective-uid trickiness with the > > way the server side of git is invoked? Or is this a network mount where > > the filesystem uid might not match the process uid? > > Huh. They're on the same ext4 fs (it's an AWS EBS sc1 volume, but I > think that still counts as "not a network mount" as far as Linux is > concerned.) Yeah, I think we can discount any oddness there. > The way I was seeing this fail was a CGI invoking "git push", as user > "httpd" (and I verified that when the cgi was invoked, "groups" > reported that "httpd" was a member of group "cvs") but when I tried to > reproduce the error with "sudo -u apache git push" it didn't fail. So > possibly something hinky is going on with group permissions when httpd > invokes git, but I did verify that whoami, groups and pwd were as > expected, so I couldn't tell what that might be... (Oh, I didn't check > what umask was, but it should have been 022...) Hrm. I don't think group permissions would even matter. We asked to mkdir() with 0700 anyway, so we know they'd be zero. But a funny umask does seem like a likely candidate for causing the problem. We asked for 0700, but if there were bits set in umask (say, 0200 or something), that would restrict that further. And it would explain what you're seeing (inability to write into a directory we just created), and it might have worked with previous versions (which was less strict on the group permissions). I don't suppose this is leaving those incoming-* directories sitting around so we can inspect their permissions (it's suppose to clean them up, so I doubt it). If you're up for it, it might be interesting to patch Git to inspect the umask and "ls -l" the objects/ directory at the problematic moment. The interesting point is when we call into tmp-objdir.c:setup_tmp_objdir(). -Peff
Re: sharedrepository=group not working
On Dec 3, 2018, at 8:19 PM, Jamie Zawinski wrote: > > (Oh, I didn't check what umask was, but it should have been 022...) Typo, I mean to say 002. -- Jamie Zawinski https://www.jwz.org/ https://www.dnalounge.com/
Re: sharedrepository=group not working
On Dec 3, 2018, at 8:09 PM, Jeff King wrote: > > but it works fine. Might there be some effective-uid trickiness with the > way the server side of git is invoked? Or is this a network mount where > the filesystem uid might not match the process uid? Huh. They're on the same ext4 fs (it's an AWS EBS sc1 volume, but I think that still counts as "not a network mount" as far as Linux is concerned.) The way I was seeing this fail was a CGI invoking "git push", as user "httpd" (and I verified that when the cgi was invoked, "groups" reported that "httpd" was a member of group "cvs") but when I tried to reproduce the error with "sudo -u apache git push" it didn't fail. So possibly something hinky is going on with group permissions when httpd invokes git, but I did verify that whoami, groups and pwd were as expected, so I couldn't tell what that might be... (Oh, I didn't check what umask was, but it should have been 022...) -- Jamie Zawinski https://www.jwz.org/ https://www.dnalounge.com/
Re: sharedrepository=group not working
On Mon, Dec 03, 2018 at 07:27:13PM -0800, Jamie Zawinski wrote: > I think sharedrepository=group stopped working some time between > 2.10.5 (works) and 2.12.4 (does not). 2.19.2 also does not. Hmm. Given the time-frame and the fact that your strace shows problems writing into the objects/incoming-* directory, it's likely caused by 722ff7f876 (receive-pack: quarantine objects until pre-receive accepts, 2016-10-03). The big change there is that instead of writing directly into objects/, we create a temporary objects/incoming-* directory, write there, and then migrate the objects over after we determine they're sane. So in your strace we see the temp directory get created: > mkdir("./objects/incoming-U5EN8D", 0700 > <... mkdir resumed> ) = 0 The permissions are tighter than we ultimately want, but that's OK. This tempdir is just for this process (and its children) to look at, and then we'd eventually migrate the files out. I could definitely imagine there being a bug in which we don't then properly loosen permissions when we move things out of the tempdir, but we don't even get that far. We fail immediately: > mkdir("./objects/incoming-U5EN8D/pack", 0777) = -1 EACCES (Permission denied) That seems strange. The outer directory is only 0700, but the user permissions should be sufficient. Even with the g+s bit set, it should still be owned by the same user, shouldn't it? I tried reproducing your state like this: git init --bare dst.git git -C dst.git config core.sharedrepository group chgrp -R somegroup dst.git find dst.git -type f | xargs chmod g+rw find dst.git -type d | xargs chmod g+srw # push works from original user git clone dst.git client ( cd client && git commit --allow-empty -m foo git push ) # push works from alternate user sudo su anotheruser sh -c ' git clone dst.git /tmp/other && cd /tmp/other && git commit --allow-empty -m foo && git push --receive-pack="strace -e mkdir git-receive-pack" ' but it works fine. Might there be some effective-uid trickiness with the way the server side of git is invoked? Or is this a network mount where the filesystem uid might not match the process uid? -Peff