On Mon, Dec 28, 2009 at 6:36 PM, Martin Bligh <[email protected]> wrote:
> > From the web server you mean. > > Well, sort of, but I think the web server is running as another user, > so changing > the owner bits won't fix anything. If we copy a directory as read-only, > then > try to write in it, we'll be up the creek too? > > > I thought about these issues too but AFAIK what > > you're describing should happen if we disable the chmod after a > get_file() > > (used when receiving files from the client to the server) not after the > > send_file() (used to send files from the server to the client, and this > is what > > John's patch changes). > > Aaaah, good point. missed that. > > > Or maybe I missed a similar problem for send_file(). > > I think there's an implicit (maybe bad) assumption, that what we're sending > is better set up and known, what we get from the client could be any old > gunk. > > That is the general assumption; plus, in practice most of the things we're sending don't need to be writable anyway, just readable, so even if lots of stuff was being sent as read-only it probably wouldn't cause much of a problem. But in any case, I don't see how these send_file changes could be causing the read-only problems you're seeing, since as a general rule all copies to test machines use send_file while all copies from test machines use get_file, and it looks like you're running into problems with the latter case. None of these changes should've impacted the permissions on files copied back; even if there was some sort of ripple effect (where the new initial permissions of files sent to the remote machine were leading to restricted permissions when copying them back) it shouldn't matter since the get_file code still does an explicit chmod on the destinations. I don't think this is a new problem; looking at my results directory, I do see a mix of 444 and 644 files. But I see that exact same mix in older results directories from a couple months ago. -- John
_______________________________________________ Autotest mailing list [email protected] http://test.kernel.org/cgi-bin/mailman/listinfo/autotest
