• Thomas Haller's avatar
    all: avoid byte ordering issue for IP4Config's Nameservers/WinsServers on D-Bus · 4eeb4b1b
    Thomas Haller authored
    Some properties in NetworkManager's D-Bus API are IPv4 addresses
    in network byte order (big endian). That is problematic.
    It is no problem, when the NetworkManager client runs on the same
    host. That is the case with libnm, which does not support to be used
    remotely for the time being.
    It is a problem for an application that wants to access the D-Bus
    interface of NetworkManager remotely. Possibly, such an application
    would be implemented in two layers:
     - one layer merely remotes D-Bus, without specific knowledge of
       NetworkManager's API.
     - a higher layer which accesses the remote D-Bus interface of NetworkManager.
       Preferably it does so in an agnostic way, regardless of whether it runs
       locally or remotely.
    When using a D-Bus library, all accesses to 32 bit integers are in
    native endianness (regardless of how the integer is actually encoded
    on the lower layers). Likewise, D-Bus does not support annotating integer
    types in non-native endianness. There is no way to annotate an integer
    type "u" to be anything but native order.
    That means, when remoting D-Bus at some point the endianness must be
    But by looking at the D-Bus introspection alone, it is not possible
    to know which property need correction and which don't. One would need
    to understand the meaning of the properties.
    That makes it problematic, because the higher layer of the application,
    which knows that the "Nameservers" property is supposed to be in network
    order, might not easily know, whether it must correct for endianness.
    Deprecate IP4Config properties that are only accessible with a particular
    endianness, and add new properties that expose the same data in an
    agnostic way.
    Note that I added "WinsServerData" to be a plain "as", while
    "NameserverData" is of type "aa{sv}". There is no particularly strong
    reason for these choices, except that I could imagine that it could be
    useful to expose additional information in the future about nameservers
    (e.g. are they received via DHCP or manual configuration?). On the other
    hand, WINS information likely won't get extended in the future.
    Also note, libnm was not modified to use the new D-Bus fields. The
    endianness issue is no problem for libnm, so there is little reason to
    change it (at this point).
nm-ip-config.c 14.4 KB