| Age | Commit message (Collapse) | Author | Files | Lines |
|---|
|
* The point of having both ethernet and ethernet-iproute
was presumably to transition slowly from ifconfig/net-tools to
iproute2. But the current codebase mixes the two in handling
ethernet and wireless connections. This commit separates
all the ip calls to the newer connections, and the ifconfig
calls to the older connections. The latter now call set_interface
with up-old, down-old, forcedown-old.
* I'm not urging that ifconfig code sticks around. Just trying
to make the code consistent with what seems to me to be
its implicit design principles.
* the check for NO-CARRIER in ethernet is now also done without
calling iproute2 tools. I saw this technique recommended somewhere
but have no citation. I don't know how many kernel revisions one
can rely on any given part of /sys to remain stable for.
It may be possible to achieve the same goal here in other ways.
If in fact it's worth the effort to keep the ifconfig code
around.
* Began also to integrate the IPROUTE variable introduced in
ethernet, but stopped. If that's integrated everywhere, shouldn't
it make connection=ethernet equivalent to
connection=ethernet-iproute? Is there justification for
using it at some places but not everywhere? I just let it alone
for now.
Signed-off-by: Jim Pryor <profjim@jimpryor.net>
|
|
* try to factor all the ifconfig/ip set dev code to single location
(currently networks/set_interface, though I recommend a different name
for this function because it's too close in name and different
in functionality from set_iface, set_profile).
* added "forcedown" option to set_interface, meaning "just go down
don't worry about 'nodown' quirks"
* even with "forcedown", this factoring would introduce "ip addr
flush dev $INTERFACE" calls to existing code. I assume that's
harmless? or even beneficial?
* one code block (in ethernet_iproute) had stdout,stderr of the
down calls redirected to /dev/null. I've moved that to
set_interface, so now everyone's getting that behavior. I assume
that's appropriate...
* now no more calls to "ip" or "ifconfig" outside of set_interface,
except the checks for NO-CARRIER after ethernet*/set_interface up.
They might also be moved to network/set_interface?
* as of this commit, all the set_interface calls are now using ip
instead of ifconfig. A subsequent commit will isolate the ip-use
to ethernet-iproute and the ifconfig-use to ethernet.
* ALERT: attend to wireless_down, it seems to be calling
set_interface down twice (once in ethernet_down, once in itself).
And neither this commit nor the code it's patching respects
quirk nodown in the ethernet_down block.
Signed-off-by: Jim Pryor <profjim@jimpryor.net>
|
|
Signed-off-by: Jim Pryor <profjim@jimpryor.net>
|
|
Signed-off-by: Jim Pryor <profjim@jimpryor.net>
|
|
* AP= arguments in profiles take precedence over ESSID= arguments
* ESSID= arguments regexps instead of literals. To avoid screwing up
any currently working profiles, I restricted the interpretation as a
regexp to just the new wireless-dbus connection types.
* But a global change involved in implementing this is that the
connection up/down calls will be passed a literal ESSID (which of the
local networks matches the regexp) as an additional argument.
Signed-off-by: Jim Pryor <profjim@jimpryor.net>
|
|
Signed-off-by: Jim Pryor <profjim@jimpryor.net>
|
|
Signed-off-by: Jim Pryor <profjim@jimpryor.net>
|
|
Signed-off-by: Jim Pryor <profjim@jimpryor.net>
|
|
* net-profiles will only attempt to control the profiles it was told to;
if you manually netcfg a different interface, it will be left alone.
(This also helps keep netcfg from stomping on other daemons that might be
hooking into it.)
Signed-off-by: Jim Pryor <profjim@jimpryor.net>
|
|
Signed-off-by: Jim Pryor <profjim@jimpryor.net>
|
|
Signed-off-by: Jim Pryor <profjim@jimpryor.net>
|
|
Signed-off-by: Jim Pryor <profjim@jimpryor.net>
|
|
Signed-off-by: Jim Pryor <profjim@jimpryor.net>
|
|
Signed-off-by: Jim Pryor <profjim@jimpryor.net>
|
|
Signed-off-by: Jim Pryor <profjim@jimpryor.net>
|
|
Signed-off-by: Jim Pryor <profjim@jimpryor.net>
|
|
Arguments to wpa_start shouldn't be expanded until wpa_start receives
them: so should be quoted by all calling functions. I added quotes
throughout. In general, it seems wise to quote except when it's
absolutely certain not to be needed---or you specifically _do_ want
to expand arguments. In this case, it's extra important.
Because this might be missed in later code, I also wrangled wpa_start
so that it doesn't rely on WPA_OPTS being all in $3.
One of these fixes would have sufficed, but what the hell I did both.
Signed-off-by: Jim Pryor <profjim@jimpryor.net>
|
|
* Supply -i switch to wpa_cli
* Have wpa_check call stop_wpa on failure, instead of duplicating
(only some of) the termination code.
* Need brief sleep in termination code.
* Note that functions which call wpa_check don't need to call
stop_wpa themselves, when wpa_check fails.
* Functions which call wpa_check and start_wpa are however
responsible for all error reporting on failure.
* Add some debugging calls to announce calls to wpa_start/wpa_check
Signed-off-by: Jim Pryor <profjim@jimpryor.net>
|
|
Signed-off-by: Jim Pryor <profjim@jimpryor.net>
|
|
These and "netcfg version" don't require root privileges.
Signed-off-by: Jim Pryor <profjim@jimpryor.net>
|
|
* Add contrib/logging.hook
* Update Makefile and manpage
* Propose that hooks go into /etc, not /usr---latter
should be shareable and possibly read-only, not
good spot for user-configurable files like these.
* Since we have hooks/arch as part of the package,
this commit attempts to load hooks from both
/usr/lib/network/hooks and /etc/network.d/hooks.
(Executable hooks in /etc take precedence over
hooks with same name from /usr.)
* Need to load /etc/rc.conf before /etc/rc.d/functions
(e.g. for USECOLORS)
Signed-off-by: Jim Pryor <profjim@jimpryor.net>
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
