Bug 72296

Summary:	docs: don't suggest "*" as a hostname unless it is portable
Product:	dbus	Reporter:	Chengwei Yang <chengwei.yang.cn>
Component:	core	Assignee:	Simon McVittie <smcv>
Status:	RESOLVED FIXED	QA Contact:	D-Bus Maintainers <dbus>
Severity:	enhancement
Priority:	low	CC:	alban.crequy, chengwei.yang.cn, msniko14, thiago, walters
Version:	1.5	Keywords:	patch
Hardware:	All
OS:	All
Whiteboard:	review?
i915 platform:		i915 features:
Attachments:	Do not claim that '*' resolves to INADDR_ANY on all OSs

Description Chengwei Yang 2013-12-04 07:45:15 UTC

If a dbus-daemon configured to listen on tcp, like <listen>tcp:host=*,port=0</listen>, it will fail to start like

$ ./tmp_install/bin/dbus-daemon --session --print-address= --print-pid=
Failed to start message bus: Failed to lookup host/port: "*:0": hostname nor servname provided, or not known (8)

For more test cases, see #bug28979

Comment 1 Chengwei Yang 2013-12-04 07:54:20 UTC

The FreeBSD 9.1 man page getaddrinfo(3) says something about hostname below.

     The hostname and servname arguments are either pointers to NUL-terminated
     strings or the null pointer.  An acceptable value for hostname is either
     a valid host name or a numeric host address string consisting of a dotted
     decimal IPv4 address or an IPv6 address.

So "*" isn't an acceptable value for hostname.

Comment 2 Chengwei Yang 2013-12-04 09:40:42 UTC

Just tried to find how the users learn that "*" is a valid value for host, in tcp/nonce-tcp transport. However, I didn't find that in both DBus Spec or dbus-daemon(1), so I'm curious, the only said in dbus-daemon is that

    The bind option can also take a special name '*' to cause the bus to listen 
    on all local address (INADDR_ANY)

Comment 3 Simon McVittie 2013-12-04 12:50:58 UTC

Which OS(s) do consider "*" to be a valid hostname?

As far as I'm concerned, the preferred syntax to express "listen on every IPv4 address" is 0.0.0.0 (which I believe is portable); omitting the "host" part entirely might also work.

If "host=*,port=3000" happens to work on Windows, or on platforms with GNU libc, or some other subset of platforms (because their libc resolver resolves * to 0.0.0.0), we shouldn't forbid that; but I also think we shouldn't recommend it, or go out of our way to make it more portable.

Analogously: localhost resolving to 127.0.0.1 is common, but not universal, and we should use host=127.0.0.1 in our defaults and documentation.

Comment 4 Simon McVittie 2014-09-11 14:05:09 UTC

The only mention of '*' that I can find is this, in dbus-daemon(1):

    <para>tcp/nonce-tcp addresses also allow a bind=hostname option,
    [...]
    or '*'
    to listen on all interfaces simultaneously.

    <para>Example: &lt;listen&gt;tcp:host=localhost,bind=*,port=0&lt;/listen&gt;</para>

Comment 5 Simon McVittie 2014-09-11 14:05:48 UTC

Created attachment 106134 [details] [review]
Do not claim that '*' resolves to INADDR_ANY on all OSs

Comment 6 Simon McVittie 2014-10-13 13:03:05 UTC

This documentation fix is a month old and has not had any comments, so I assume nobody has any objection to it. Fixed in git for 1.9.2.

Use of freedesktop.org services, including Bugzilla, is subject to our Code of Conduct. How we collect and use information is described in our Privacy Policy.