Thread (1 message) 1 message, 1 author, 2021-12-02

Re: [PATCH 1/2] http-backend: CGI error messages need to be output on stdout

From: Junio C Hamano <hidden>
Date: 2021-12-02 07:31:58

Jan Engelhardt [off-list ref] writes:
Accessing a clone-only URL with a browser would always show a blank
page, because the reason string "Request is not supported" was
"is not supported" -> "not supported", I think.
sent to the wrong file descriptor.
I looked at the messages given to not_found() and forbidden().  Most
of them do look like they are meant to be given to the remote user,
but some reveal the server side paths, and I am reluctant to judge
the security implication to start giving them, which have been
hidden so far, to the remote users.

I am not sure how strictly it is enforced these days, but at least
in early days we were fairly paranoid and aimed to make sure that a
remote user cannot tell a repository that does not exist from a
repository that the user does not have access to by throwing random
requests at the server.

http-backend.c:161:		forbidden(hdr, "Unsupported service: getanyfile");

This one should be safe, I would think.

http-backend.c:184:		not_found(hdr, "Cannot open '%s': %s", p, strerror(errno));

The 'p' is a path to a file in the repository like
'objects/alternates', that is ready to be given to open(2), so it
clearly leaks the path on the server.

http-backend.c:260:		forbidden(hdr, "Unsupported service: '%s'", name);
http-backend.c:271:		forbidden(hdr, "Unsupported service: '%s'", name);
http-backend.c:278:		forbidden(hdr, "Service not enabled: '%s'", svc->name);

These I think should be benign, but I have to admit that I didn't
think too deep about them.
Keyboard shortcuts
hback out one level
jnext message in thread
kprevious message in thread
ldrill in
Escclose help / fold thread tree
?toggle this help