rvlan: bugfix: wrs_vlans is not in /home/rubini/bin:/home/rubini/bin/i386:/usr/local/bin:/bin:/usr/bin:/sbin:/usr/sbin:/usr/bin/X11:/usr/TeX/bin at boot
Signed-off-by: Alessandro Rubini <rubini@gnudd.com>

Signed-off-by: Alessandro Rubini <rubini@gnudd.com>

Signed-off-by: Alessandro Rubini <rubini@gnudd.com>

Signed-off-by: Alessandro Rubini <rubini@gnudd.com>

In sh, string comparison for equality is "=", not "==". Bash accepts
either operator but suggests to use "=" for posix conformance.
Our /bin/sh is currently busybox, which accepts "==" like bash, but
if you use ash or dash it fails:

   $ CONFIG_FOO=y
   $ if [ "$CONFIG_FOO" = "y" ]; then echo ok; fi
   ok
   $ if [ "$CONFIG_FOO" == "y" ]; then echo ok; fi
   dash: 3: [: y: unexpected operator
Signed-off-by: Alessandro Rubini <rubini@gnudd.com>

The radius server, when denying authorization, may reply as follows:

   Received response ID 79, code 3, length = 44
       Tunnel-Type:0 = 13
       Tunnel-Medium-Type:0 = IEEE-802
       Framed-Protocol = PPP
       Service-Type = Framed-User

i.e., it misses the vlan number as "Tunnel-Private-Group-Id", while
everything else is in place. In this case we should not obey
dotconfig, but rather keep the noauth_vlan.
Signed-off-by: Alessandro Rubini <rubini@gnudd.com>

Signed-off-by: Alessandro Rubini <rubini@gnudd.com>

also, the mac address is always saved in dev->mac, so we get it
in rvlan-status also for "down" interfaces.

Example:

   wri12 (001bc509084f <-> ): state down, vlan 0, pid 0, fd -1
   wri13 (001bc5090850 <-> ): state down, vlan 0, pid 0, fd -1
   wri14 (001bc5090851 <-> ): state down, vlan 0, pid 0, fd -1
Signed-off-by: Alessandro Rubini <rubini@gnudd.com>

This simply sends SIGUSR1 to radiusvlan and sorts the output file.

Example:

wri2 (001bc5090845 <-> 001bc509007d): state configured, vlan 4094, pid 0, fd -1
wri3 (001bc5090846 <-> 00e04c1bfd8d): state configured, vlan 31, pid 0, fd -1
wri4 (000000000000 <-> ): state down, vlan 0, pid 0, fd -1
wri5 (000000000000 <-> ): state down, vlan 0, pid 0, fd -1
wri6 (000000000000 <-> ): state down, vlan 0, pid 0, fd -1
wri7 (000000000000 <-> ): state down, vlan 0, pid 0, fd -1
wri8 (000000000000 <-> ): state down, vlan 0, pid 0, fd -1
wri9 (000000000000 <-> ): state down, vlan 0, pid 0, fd -1
wri10 (000000000000 <-> ): state down, vlan 0, pid 0, fd -1
wri11 (000000000000 <-> ): state down, vlan 0, pid 0, fd -1
wri12 (000000000000 <-> ): state down, vlan 0, pid 0, fd -1
wri13 (000000000000 <-> ): state down, vlan 0, pid 0, fd -1
wri14 (000000000000 <-> ): state down, vlan 0, pid 0, fd -1
wri15 (000000000000 <-> ): state down, vlan 0, pid 0, fd -1
wri16 (000000000000 <-> ): state down, vlan 0, pid 0, fd -1
wri17 (000000000000 <-> ): state down, vlan 0, pid 0, fd -1
wri18 (000000000000 <-> ): state down, vlan 0, pid 0, fd -1
Signed-off-by: Alessandro Rubini <rubini@gnudd.com>

The user can now send SIGUSR1 to get a snapshot of the current status,
and SIGUSR2 to force reauthentication of all ports (without getting
through stop/start).

Please note that during reauth we do not configure ports to the
auth_vlan network: they would just continue to live on their current
VLAN while the radius server is queried. For those ports where nothing
changed, no service interruption happens.
Signed-off-by: Alessandro Rubini <rubini@gnudd.com>

Signed-off-by: Alessandro Rubini <rubini@gnudd.com>

The problem is exposed when we force all intefaces to "justup", in a
later commit. Some interfaces will have no children to wait for and
would spit this:

   radiusvlan: wait(0): No child processes
Signed-off-by: Alessandro Rubini <rubini@gnudd.com>

When we get one of the very-unlikely errors, we should countinue to
the next interface rather than return to the caller.
Signed-off-by: Alessandro Rubini <rubini@gnudd.com>

The problem with /sys/net/*/carrier, is that the file may be missing.
For example, if the fiber is unplugged.  "operstate" is always there,
and it is either "up" or "down".  If it is ifconfig'd up but without fiber,
the operstate is properly down, so we can use it with no further ado.
Signed-off-by: Alessandro Rubini <rubini@gnudd.com>

Signed-off-by: Alessandro Rubini <rubini@gnudd.com>

We may use the return value later to see if the server is unresponsive,
and pick another one. Anyways this is cleaner, kill() was a hack.
Signed-off-by: Alessandro Rubini <rubini@gnudd.com>

Signed-off-by: Alessandro Rubini <rubini@gnudd.com>

Signed-off-by: Alessandro Rubini <rubini@gnudd.com>

Signed-off-by: Alessandro Rubini <rubini@gnudd.com>

Signed-off-by: Alessandro Rubini <rubini@gnudd.com>

Signed-off-by: Alessandro Rubini <rubini@gnudd.com>

Now that the state machine is known to work, we'd better only report
state changes. So this is the new output for the whole cycle
when a fiber is plugged (including kernel messages on the console):

   wri3: Link up, lpa 0x41a0.
   device wri3 entered promiscuous mode
   FSM: wri3: down -> sniff
   device wri3 left promiscuous mode
   recvfrom(wri3): 0800-90e2ba456c6b  FSM: wri3: sniff -> auth
   dev wri3, got 55 bytes so far
   dev wri3: vlan 31
   FSM: wri3: auth -> config
   FSM: wri3: config -> configured

While this is a port already up twhen the tool starts, in an ssh run:

   FSM: wri17: justup -> sniff
   recvfrom(wri17): 88f7-00267b0003d4
   FSM: wri17: sniff -> auth
   dev wri17, got 185 bytes so far
   dev wri17: vlan 31
   FSM: wri17: auth -> config
   FSM: wri17: config -> configured
Signed-off-by: Alessandro Rubini <rubini@gnudd.com>

Signed-off-by: Alessandro Rubini <rubini@gnudd.com>

Signed-off-by: Alessandro Rubini <rubini@gnudd.com>

Signed-off-by: Alessandro Rubini <rubini@gnudd.com>

Signed-off-by: Alessandro Rubini <rubini@gnudd.com>

Signed-off-by: Alessandro Rubini <rubini@gnudd.com>

This commit fixes errors in pipe closing, and changes the parsing.

NOTE: it also changes the identification of "self" frames. ppsi
sends everything using the eth0 mac address, which is different
from the wriXX port.

This is hacked now by checking the first 5 bytes, but it must be
fixed for real.

Finally, I save both stdin and stdout that I exchange with radclient
into tmp, so we can check what happened during tests.
Signed-off-by: Alessandro Rubini <rubini@gnudd.com>

We'd better not daemonize, so we can use the init.d scripts of hald and
other wr tools, with syslog support.
Signed-off-by: Alessandro Rubini <rubini@gnudd.com>

Signed-off-by: Alessandro Rubini <rubini@gnudd.com>

Signed-off-by: Alessandro Rubini <rubini@gnudd.com>

Signed-off-by: Alessandro Rubini <rubini@gnudd.com>

Signed-off-by: Alessandro Rubini <rubini@gnudd.com>

The full set of /etc/raddb/dictionary* is overkill, we only need
a few symbols from these three files, if we run radclient.
Signed-off-by: Alessandro Rubini <rubini@gnudd.com>

Signed-off-by: Alessandro Rubini <rubini@gnudd.com>

Signed-off-by: Alessandro Rubini <rubini@gnudd.com>

Signed-off-by: Alessandro Rubini <rubini@gnudd.com>

Signed-off-by: Alessandro Rubini <rubini@gnudd.com>

Signed-off-by: Alessandro Rubini <rubini@gnudd.com>

Signed-off-by: Alessandro Rubini <rubini@gnudd.com>