Hooks and interface scripts are now sourced more systematically.
Settings of wireless profiles that depend on hooks are now picked up
correctly by netctl-auto.

Suggested by: Vitor Sakaguti <vis@tutao.de>

This also allows hooks to alter the profile to be loaded.

Hooks can now tell the command to be executed by looking at $Command.
The value will be "start", "stop", or "" (undefined), depending on the
reason a profile is being sourced.

This covers some corner cases where, for instance, IFS does not contain
whitespace characters.

Distinguish between started profiles and online profiles in the output
of the list subcommand of netctl.
To be consistent in symbol usage, the wifi-menu flags have changed too.

The netctl@ service now yields before obtaining an IP address. This
prevents holding up network.target unnecessarily. Waiting for profiles
to obtain an IP address is possible through

1) The netctl-wait-online service
This service waits until all enabled profiles have obtained an address.
It is ordered before network-online.target so that this target is now
correctly implemented in netctl.

2) The wait-online <PROFILE> subcommand to netctl
This command waits for a started profile to obtain an address.

This adds the LinkOptions= variable for use with connection types that
add interfaces (bond, bridge, dummy, macvlan, vlan). This variable is
not portable in the event of moving away from iproute2, so common
options should still get their own variables.

`shift 3` doesn't do anything if there are only 2 parameters.

The main benefit is the possibility to use interface hooks for links
that are created at runtime.

Thomas Bächler authored May 16, 2013 and Jouke Witteveen committed May 17, 2013

When switching networks in auto.action, the addresses are not flushed. This
is especially problematic with stateless ipv6 autoconfigutation, as invalid
IPs may stay around until their (potentially very long) lifetime has expired.

bring_interface_down is always called after ip_unset everywhere else, so
this change does not affect anything else.

V2: Make sure not to flush the link local address.

This allows connection scripts to provide defaults, among other things.

This settles a style for 'case' statements:
The label and closing colons are indented with two spaces,
The body is indented with two more, totalling four spaces.

This commit contains the refactoring and rewriting of code.

This commit contains the moving of files.

This commit removes an undocumented feature where IFACE_{UP,DOWN} would
be evaluated when the interface is brought up or down.
This functionality did not work as expected since the interface could
potentially be brought up/down more than once in one netcfg invocation.
This is apparent in the changes to the bonding code.

Direct invocation as fgrep is deprecated.

quirk() is not used anymore, inarray() doesn't belong in src/network.

The new way (cat) is common for netcfg and more readable.
The old way (echo $(< )) is faster, but that is completely subordinate to readability, here.

$STATE_DIR/last_profile was unused and functionality one could want from it will soon be provided by netcfg-daemon.

This should fix FS#30361 along the way.

After this change, a profile started by a previous version of netcfg can potentially not be brought down anymore. This only concerns the updating process. Running `netcfg clean` should make everything OK again (potentially leaving your network down, but you can now bring up profiles again).

Suggestion by Dave Reisner.

Good drivers indicate when they're up by the IFF_UP flag.
This is the end of the (undocumented) UP_SLEEP variable and introduction of the UP_TIMEOUT (in seconds, default: 5) variable.

This has 'noticeable speed improvement'-potential.

This was the result of commit b6964 which said to fix FS#17406 by reverting to the subshell-less for-loop that is once again removed in this commit.
The real fix to FS#17406 was commit c3276 to the initscripts repository.

Configuration is done through /etc/conf.d/netcfg.
Compatibility with configuration through /etc/rc.conf is dropped.

/etc/rc.d/functions is owned by initscripts on which netcfg does not explicitly depend.
The revised layout is inspired by systemd.

It is now possible to reconnect based on the name of the interface.

Signed-off-by: Rémy Oudompheng <remy@archlinux.org>

Avoid duplicating function names in docstrings, and adopt a
style similar to doxygen.
Signed-off-by: Rémy Oudompheng <remy@archlinux.org>

Signed-off-by: Jesse Young <jesse.young@gmail.com>

Was the intent for is_interface to be passed the INTERFACE as a
parameter (that's how it's called in most or all places)? Or to read
INTERFACE from the current environment? That's how it's implemented.

Also made INTERFACE in bring_interface a local variable. I don't think
this should break anything, but I'm only 95% confident.