| Age | Commit message (Collapse) | Author |
|---|
|
configuring the same IP.
Found the hard way by afresh1
|
|
defaults before validating the times to prevent excessive logging.
Found the hard way & OK brynet
|
|
probably stuck in some way and the user wants a mostly clean slate.
If we already have an IP address transition to state REBOOTING so that
we no longer unicast dhcp requests. We will then try to reacquire our
lease twice before giving up and transition to INIT and send dhcp
discover messages accepting any IP address.
|
|
whether the address we received in our lease is already configured.
In the case I observed, no default route was added to the routing table
even though the server provided both an address and a route option.
As it happened the leased address was already configured on the interface.
This should not prevent routing table updates, but it did.
ok florian
|
|
resolvd(8), slaacd(8) and dhcpleased(8) are different from other daemons
in that there must only be a single instance.
resolvd already does this, adjust slaacd and dhcpleased accordingly while
moving the lockfile paths under /dev/ such that they work early on boot and
don't run into races should /var be (un)mounted between daemon starts.
Locking is especially required in the installer where all three daemons are
started every time the "(I)nstall, (U)pgrade, (A)utoinstall or (S)hell? "
prompt is entered, i.e. restarting installation or dropping into a shell
and back into the prompt again would start multiple instances.
To avoid expected lockfile error messages in between installer prompts,
discard standard error when starting the autoconf daemons; none of them
has other potential failure cases in installer mode before daemon(3)izing.
Input sthen deraadt
OK deraadt
|
|
as mandated by RFC3442.
Pointed out by, initial diff, testing & OK bket@
|
|
fails to report the path that the failure occured on. Suggested by
deraadt@ after some tech discussion.
Work done and verified by Ashton Fagg <ashton@fagg.id.au>
ok deraadt@ semarie@ claudio@
|
|
file for the installer.
|
|
|
|
in previous.
|
|
For this we need to be able to handle multiple routes being sent from
the engine to the main process as well as to the control tool.
The configuration of the various cases (default route, directly
connected routes, non-default route via a gateway) was inspired by
dhclient's set_routes() and should behave the same way.
Tested by Uwe Werler
|
|
the control socket instead of fatal().
OK deraadt
|
|
switching from chroot("/var/empty") to unveil("/", "").
This is just an extra pair of suspenders since these processes
pledge(2) to not access the filesystem.
OK deraadt
|
|
|
|
padding because the ethernet header in front is only 14 bytes.
Found the hard way by me while testing on sparc64.
Solution suggested by & OK deraadt
|
|
values as specified in RFC2131 section 4.4.5. Allows my Comtrend VI-3223u
to work.
OK florian@
|
|
Doing so implies support for it, but dhcpleased(8) currently ingores it
entirely and does not configure any route from it.
As per RFC 3442 servers SHOULD NOT respond with a "routers" option when
"classless-static-routes" is set.
dhcpd(8)/dhcpd.conf(5) follows that, hence requesting but not using static
routes results in not installing any routes at all.
Stop signaling support for this option and only request "routers" such that
dhcpleased continues to install a default route and properly ignores the
unsupported option if used by the server.
Report from Uwe Werler <uwe @ werler dot is> about a default route not
being set when requesting the "classless-static-routes" dhcp-options(5)
from dhcpd(8), thanks!
OK florian
|
|
need to provide the address of the interface behind which the default
router is in case they are on the same subnet otherwise the kernel
can't figure out which route we are talking about
This happens for example when your wifi and wired networks are bridged.
Pointed out by claudio some time ago.
|
|
interoperable with BOOTP we should also send packets that have a
minimum size of 300.
I haven't seen a DHCP server that actually enforces this except the
one in vmd(8), but it doesn't cost us much and prevents hair pulling
later on when we find one in the wild.
OK deraadt
|
|
|
|
getifaddrs on every route message.
This also allows us to drop the route pledge since we only need to
fetch the interface state with getifaddrs on startup.
|
|
state of the machine on startup using ioctl(2) and getifaddrs(3).
We can then update this state with information provided by route
messages. We still need getifaddrs(3) to check if the layer 2 address
has changed.
This simplifies error handling (what should we do if ioctl(2) fails?),
reduces kernel round trips (no need to ask the kernel again for
information RTM_IFINFO provided already) and prevents a theoretical
race between RTM_IFINFO and getaddrinfo(3).
In a fast link state UP -> DOWN -> UP transition RTM_IFINFO informs us
that the link went down but we were not using this information but
rather looked at getifaddrs(3) information which might see the link as
already up again. We would then do nothing while we should try to get
a new lease.
By storing all interface information in the frontend process we can
skip imsgs to the engine process if we get an RTM_IFINFO without
relevant changes for us.
|
|
|
|
it.
|
|
AF_LINK and skip one ioctl.
OK benno
|
|
behind -vv or by deleting unneeded output.
While here reword some debug output to make it more useful.
(There is more to be done here.)
|
|
handles this for us by doing a state transition if we have been stuck
in "rebooting" or "requesting" for too long.
Makes the code a bit simpler and we only have one place were we need
to special case the timeout cap.
|
|
they don't like them instead of sending a DHCPNAK. Found the hard way
by benno who didn't want to wait 127 seconds.
Due to another bug dhcpleased would have exit through a fatal() in the
frontend process if he had waited long enough for a Rebooting -> Init
transition because we didn't deconfigure our IP address and thus
didn't close our UDP socket. Upon configuring a new IP address we would
open a new UDP socket send it to the frontend which would then fatal()
due to an unexpected fd passed in.
Aproporiate timings are rather underspecified in RFC 2131. Instead of
doing an exponential backoff up to 64 in the "Rebooting" and
"Requesting" state only go up to 2 for a total of 3 packets and total
timeout of 3 seconds before going into "Init" state and sending a
DHCPDISCOVER.
To prevent the fatal() in the frontend process we reshuffle the state
transition into the "Init" state and deconfigure the IP when
appropriate.
|
|
I'm worried we could see packets we shouldn't during a small time window.
|
|
|
|
when the lease directory does not exist.
This means that dhcpleased(8) will no longer request a previously
configured IP address from the dhcp server and will fall back to
DHCPDISCOVER which requests any IP address from the dhcp server.
This likely makes diskless(8) work with dhcpleased(8).
A normal diskless(8) setup has only / mounted via nfs when
dhcpleased(8) starts. /var exists but nothing is mounted there yet,
meaning /var/db/dhcpleased does not exist so lease files are disabled.
dhcpleased(8) sends a DHCPDISCOVER to request any IP address but since
the dhcp server has (very likely) a 'fixed-address' configured we get
the same IP back that is already configured.
If /var/db/dhcpleased/ exists on / (and /var is *NOT* mounted later)
in a diskless(8) setup, care must be taken that the root file system is
not shared between machines.
If /var/db/dhcpleased/ exists on / and /var on NFS is mounted over
this later bad things probably happen. This is a configuration error
and must befixed.
discussed with deraadt@
Actuall tests on existing diskless(8) setups would be appreciated.
|
|
|
|
ok florian@
|
|
deraadt@ pointed out that dhcpleased is too quiet.
|
|
state_transition() and iface_timeout(). Let the compiler help us by
emitting a warning when we missed one (-Wswitch).
Reminded by jsg who pointed out that gcc is quite confused and thinks
there is an out of bounds access in if_state_name[] in the default
case. There is not, if_state_name[] and enum if_state have to be kept
in sync.
(Note that -Wswitch is not a silver bullet, it just happens to work
here.)
|
|
themselves. This way the iface object is in a consistent state.
For consistency we should also withdraw rdns first, then deconfigure
the interface.
Lastly make sure we parse the lease file on a down -> up transition if
we had a lease before and it had expired while the interface was in
down state. Otherwise we'd send a dhcpdiscover requesting any IP
address while we really should send a dhcprequest asking for our
previous IP back.
|
|
from and explain why we are a bit more agressive during startup.
While here make the math a bit easier on the eyes.
|
|
got lost shortly before initial import.
While here explicitly delete the default route when deconfiguring an
interface. There might be corner cases where the stack will not tear
it down for us when we remove the IP address.
|
|
pointed out by deraadt
|
|
the engine process for parsing instead of passing an fd.
Let's us tighten the engine's pledge back down to "stdio".
|
|
|
|
from servers.
dhcpleased(8) follows the well known three process design of all our
privsep daemons. It uses pledge(2) and unveil(2) to restrict access
further. In particular the "engine" process, responsible for parsing
of untrusted data, is pledge'd "stdio". It cannot access the outside
world nor the filesystem at all.
Like slaacd(8) for IPv6 it will be always running and acquire addresses
for all interface with the autoconf4 flag set.
The flag can be set by "ifconfig $if inet autoconf" or by adding
"inet autoconf" to /etc/hostname.if. An existing "dhcp" line should
be removed.
Various iterations tested by deraadt@
The hardest part, finding a name, was handled by jmatthew@ & otto@
"get to it :)" deraadt@
|
