apr-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "William A. Rowe, Jr." <wr...@rowe-clan.net>
Subject Re: SVN CLI and Windows security
Date Sun, 29 Jul 2007 21:54:01 GMT
Erik Huelsmann wrote:
> On 7/25/07, Brad Stiles <bradstiles@bellsouth.net> wrote:
>> I ran into a situation today concerning the command line client and windows security.
 The problem is that the user (a build user performing automated builds on a Windows 2003
VM) attempted to check out a file to a network share and couldn't, apparently because that
user doesn't have access to the share root, even though it does have access to the directory
into which the file should be exported.
>> The command I used was:
>> svn export http://server/trunk/dir1/dir2/file.txt \\server\share\dir1\dir2\dir3\file.txt
--non-interactive --force
>> To which svn.exe responded:
>> svn: Error resolving case of '\\server\share\dir1\dir2\dir3\file.txt'
>> In that tree, the user has no rights to \\server\share\, but has read/write access
to dir1 and below.  If, as that user, I do 'dir \\server\share', I see nothing, but if I do
'dir \\server\share\dir1', I see everything just fine.
>> Now, in our case, that shouldn't be a problem much longer because the network guys
are in the process of changing the access rights on that directory; the build user really
should have access there.  However, I don't think it's unreasonable to assume that there might
be a legitimate reason for such a restriction.
>> Is there something we did wrong here (aside from the rights issue) that would have
allowed the export to happen despite the access rights?  FWIW, 'svn checkout' resulted in
the same type of error.
>> When the network guy saw it, he said, "Maybe they're using the old NT4 method of
stepping through the tree, rather than going straight to the location specified."  Is there
something in whatever layer svn is using for this that accesses the tree step by step rather
than as one entity?  And is it worth "fixing" if it is a problem?
> Well, it's an error that finds its roots in a call to
> apr_filepath_merge with an argument of APR_FILEPATH_TRUENAME. Looking
> at that function, it calls apr_filepath_root, which tries to isolate
> the root in order to see whether it actually exists. That scheme
> breaks down when there are no read/write permissions.

This presumes it's an error at all.  The _TRUENAME flag tells us that we
MUST use the canonical path.  The *only* way to accomplish this is with
read/traverse access through the file path.

The arguements for this approach is that an app (such as a server) often
needs to deal with the single proper name of a resource.  Program Files
may be Progra~1, it may also be Progra~2, or on netware, ProgramF.

> I have no idea whether the APR project feels this is something worth
> to be fixed, or even fixeable. I cc'ed their dev@ list, to make them
> aware if the weren't already.

Of course it won't be fixed, because it is not broken.


View raw message