dhcpv4: don't copy reqopts around

Ok, this is a bit weird, dhcpv4_handle_msg() passes reqopts to dhcpv4_lease(),
which stashes a copy of the reqopts in the struct dhcp_assignment, and then
dhcpv4_handle_msg() uses the stashed copy instead.

The only reason a copy is made seems to be so that the requested options can
later be reported via ubus (in "ubus call dhcp ipv4leases"), but it is hard to
see a use-case for that (the DHCP server has already replied to the client by
the time the lease is available over ubus) and we don't do anything similar for
DHCPv6, so remove the extra copying.

Signed-off-by: David Härdeman <[email protected]>
Link: https://github.com/openwrt/odhcpd/pull/278
Signed-off-by: Álvaro Fernández Rojas <[email protected]>

dhcpv4: more refactoring of dhcpv4_handle_msg()

Move some gnarly conditionals into a switch in preparation of later patches.

Signed-off-by: David Härdeman <[email protected]>
Link: https://github.com/openwrt/odhcpd/pull/278
Signed-off-by: Álvaro Fernández Rojas <[email protected]>

dhcpv4: clarify variable names in dhcpv4_handle_msg()

Make it clearer which variables relate to the request, and which ones relate to
the reply that we're constructing.

Also fix the missing endianness conversion for req_leasetime (not so severe
though, a bogus leasetime value will be overridden with limits that are set
per-lease or per-interface).

Signed-off-by: David Härdeman <[email protected]>
Link: https://github.com/openwrt/odhcpd/pull/278
Signed-off-by: Álvaro Fernández Rojas <[email protected]>

dhcpv4: some more cleanups to dhcpv4_handle_msg()

Most notably, neither hostname, nor reqopts actually need to be copied from
req, they are used in a read-only fashion throughout the rest of the function
(and any functions it calls). Also switch some complicated if-else constructs
over to switch-case.

Signed-off-by: David Härdeman <[email protected]>
Link: https://github.com/openwrt/odhcpd/pull/278
Signed-off-by: Álvaro Fernández Rojas <[email protected]>

dhcpv4: preparations for iovec usage

dhcpv4.c currently builds DHCPv4 messages by lots of repeated calls to the
dhcpv4_put() method, which basically does a whole lot of copies to the stack.
The following set of patches convert dhcpv4.c over to use iovecs, which avoids
almost all of the stack allocations and copies.

The first step is to lay the groundwork by changing dhcpv4_send_reply() and the
corresponding dhcpv6_4o6_send_reply() in dhcpv6.c so that they both take an
iovec instead of a plain buffer. Also introduce the first (very very simple)
iovec for dhcpv4_handle_msg() which just contains the whole packet, plus the
end marker as a separate vector element. It will be filled out further in
subsequent patches.

Signed-off-by: David Härdeman <[email protected]>
Link: https://github.com/openwrt/odhcpd/pull/278
Signed-off-by: Álvaro Fernández Rojas <[email protected]>

odhcdp: use a more suitable clock

In my testing, CLOCK_MONOTONIC_COARSE uses about 1/3 of the CPU cycles.

And the accuracy is completely irrelevant.

Signed-off-by: David Härdeman <[email protected]>
Link: https://github.com/openwrt/odhcpd/pull/282
Signed-off-by: Álvaro Fernández Rojas <[email protected]>

dhcpv6: change dhcpv6 message type check in relay

When compiled, the switch instruction jump table O(1) is more efficient
than multiple logical comparisons in the if block (4-5 checks).

Signed-off-by: Paul Donald <[email protected]>
Link: https://github.com/openwrt/odhcpd/pull/279
Signed-off-by: Álvaro Fernández Rojas <[email protected]>

dhcpv6: move dhcpv6 message type check for early exit

Avoid doing a bunch of work if the message type we received is not a
client type and does not warrant a reply. Exit early.

Clarify and comment message type handling.

Signed-off-by: Paul Donald <[email protected]>
Link: https://github.com/openwrt/odhcpd/pull/279
Signed-off-by: Álvaro Fernández Rojas <[email protected]>

odhcpd: add a simple build script

Inspired from the ci script and the scripts I've been using. Should make it a
little bit easier for people who want to contribute to odhcpd.

Signed-off-by: David Härdeman <[email protected]>
Link: https://github.com/openwrt/odhcpd/pull/280
Signed-off-by: Álvaro Fernández Rojas <[email protected]>

github: improve CI

- Remove unneeded CMAKE_SYSTEM_PROCESSOR.
- Add summary table with sizes (in bytes) for each arch/variant.
- Upload generated binaries as artifacts.
- Add OpenWrt formalities.

Signed-off-by: Álvaro Fernández Rojas <[email protected]>

odhcpd: shrink binary size by creating a logging function

Moving the logging to a real function helps shrink the binary size back to what
it was before the logging patches.

Signed-off-by: David Härdeman <[email protected]>
Link: https://github.com/openwrt/odhcpd/pull/273
Signed-off-by: Álvaro Fernández Rojas <[email protected]>

odhcpd: support stderr logging

This is just for debugging purposes. Also, make sure that a cmdline loglevel
doesn't get overwritten by what's in the cfg file (it's customary that
command-line options take precedence over configuration file settings).

Signed-off-by: David Härdeman <[email protected]>
Link: https://github.com/openwrt/odhcpd/pull/273
Signed-off-by: Álvaro Fernández Rojas <[email protected]>

odhcpd: add log helpers

These a just a bit more succinct than their syslog equivalents. Also, this lays
the ground for the next patch.

Signed-off-by: David Härdeman <[email protected]>
Link: https://github.com/openwrt/odhcpd/pull/273
Signed-off-by: Álvaro Fernández Rojas <[email protected]>

config: cap dhcpv6_pd_min_len to max instead of only logging error

Signed-off-by: Paul Donald <[email protected]>
Link: https://github.com/openwrt/odhcpd/pull/225
Signed-off-by: Álvaro Fernández Rojas <[email protected]>

config: clamp dhcpv6_hostid_len instead of only logging an error

Signed-off-by: Paul Donald <[email protected]>
Link: https://github.com/openwrt/odhcpd/pull/225
Signed-off-by: Álvaro Fernández Rojas <[email protected]>

config: clamp ra_mtu into 1280-65535 range

(ipv6 packets can be up to 4GB in size...)

The old logic had a logic error of || instead of &&

Signed-off-by: Paul Donald <[email protected]>
Link: https://github.com/openwrt/odhcpd/pull/225
Signed-off-by: Álvaro Fernández Rojas <[email protected]>

config: cap ra_retranstime and warn instead of only logging an error

Set to the currently defined maximum of 60,000msec via a define.

Signed-off-by: Paul Donald <[email protected]>
Link: https://github.com/openwrt/odhcpd/pull/225
Signed-off-by: Álvaro Fernández Rojas <[email protected]>

config: cap ra_hoplimit to maximum and warn instead of logging an error

Signed-off-by: Paul Donald <[email protected]>
Link: https://github.com/openwrt/odhcpd/pull/225
Signed-off-by: Álvaro Fernández Rojas <[email protected]>

config: cap ra_reachabletime to RFC maximum instead of logging error

Signed-off-by: Paul Donald <[email protected]>
Link: https://github.com/openwrt/odhcpd/pull/225
Signed-off-by: Álvaro Fernández Rojas <[email protected]>

config: drop double size lease times; they are all UINT32_MAX;

This now prevents implicit 64 bit->32 bit truncation which may flag
compiler errors later on down the road.

All of the variables receiving from parse_leasetime() are uint32_t
anyway, so max 136 years of valid lease time will have to suffice :)

Signed-off-by: Paul Donald <[email protected]>
Link: https://github.com/openwrt/odhcpd/pull/225
Signed-off-by: Álvaro Fernández Rojas <[email protected]>

router: redefine ra_mininterval and ra_maxinterval as uint32_t

They never store negative values.

Signed-off-by: Paul Donald <[email protected]>
Link: https://github.com/openwrt/odhcpd/pull/225
Signed-off-by: Álvaro Fernández Rojas <[email protected]>

config: clamp ra_mininterval, ra_maxinterval, ra_lifetime at load time

clamp values to RFC defined limits.

First set ra_maxinterval, then ra_mininterval adjusts based on max.

Signed-off-by: Paul Donald <[email protected]>
Link: https://github.com/openwrt/odhcpd/pull/225
Signed-off-by: Álvaro Fernández Rojas <[email protected]>

router: refactor calc_ra_lifetime; redefine ra_lifetime as uint32_t

ra_lifetime no longer holds negative values, because we no longer do
'init' when we do calc_ra_lifetime, instead we do init at init time.

Now ra_lifetime holds only >0 values.

Signed-off-by: Paul Donald <[email protected]>
Link: https://github.com/openwrt/odhcpd/pull/225
Signed-off-by: Álvaro Fernández Rojas <[email protected]>

config: do MaxRtrAdvInterval init at (ra_maxinterval) init time

Signed-off-by: Paul Donald <[email protected]>
Link: https://github.com/openwrt/odhcpd/pull/225
Signed-off-by: Álvaro Fernández Rojas <[email protected]>

router: Apply updated values from RFC9096 (updates RFC4861) to RA/ND

https://datatracker.ietf.org/doc/html/rfc9096#section-3.4

Signed-off-by: Paul Donald <[email protected]>
Link: https://github.com/openwrt/odhcpd/pull/225
Signed-off-by: Álvaro Fernández Rojas <[email protected]>

router: Apply updated values from RFC8319 (updates RFC4861) to RA/ND

https://www.rfc-editor.org/rfc/rfc8319#section-4

Signed-off-by: Paul Donald <[email protected]>
Link: https://github.com/openwrt/odhcpd/pull/225
Signed-off-by: Álvaro Fernández Rojas <[email protected]>

config: refactor parse_leasetime() - branch amount remains same

Also make 's' value a noop.

Signed-off-by: Paul Donald <[email protected]>
Link: https://github.com/openwrt/odhcpd/pull/225
Signed-off-by: Álvaro Fernández Rojas <[email protected]>

github: fix CMAKE_SYSTEM_PROCESSOR copy&paste

CMAKE_SYSTEM_PROCESSOR was incorrectly defined as matrix.packages instead
of matrix.arch.

Fixes: 288206c9a2ed ("github: add CI build")
Signed-off-by: Álvaro Fernández Rojas <[email protected]>

github: add CI build

Add Github CI supporting different architectures and odhcpd build options.

Signed-off-by: Álvaro Fernández Rojas <[email protected]>

odhcpd: fix a compilation error

odhcpd_get_interface_dns_addr() calls odhcpd_get_interface_linklocal_addr(),
the former takes a const struct interface ptr, the latter takes a non-const
struct interface ptr.

The end result is an unhappy compiler.

Signed-off-by: David Härdeman <[email protected]>
Link: https://github.com/openwrt/odhcpd/pull/272
Signed-off-by: Álvaro Fernández Rojas <[email protected]>

odhcpd: allow assignments to be reassigned

In a nutshell, this allows static leases which have only been configured
with a DUID (so no IAID specified) to be reassigned to any client with
the same DUID (but potentially different IAID). This matches how
multiple MAC addresses are handled in the DHCPv4 case, and supports the
use case for e.g. moving from WiFi to ethernet.

On the other hand, it is probably not what one would want for e.g. a
server that has several NICs connected to the same network. For those
cases, defining static leases with DUID and IAID anyway seems
preferrable (and it will also ensure a static interface<->IPv6 addr
mapping).

Signed-off-by: David Härdeman <[email protected]>
Link: https://github.com/openwrt/odhcpd/pull/255
Signed-off-by: Álvaro Fernández Rojas <[email protected]>

odhcpd: support multiple per-client DUIDs

This patch allows for multiple per-client DHCPv6 DUIDs. One use-case for
multiple DUIDs is dual-boot hosts, where e.g. Windows creates it's own DUID and
Linux (say, systemd-networkd) creates it's own DUID, and the two cannot be
synced in a simple manner. This also makes DUIDs similar to multiple per-host
MAC addresses (which is already supported by dnsmasq for DHCPv4 leases).

Note that the "old" behaviour is still kept, in other words, where a
simple DUID (as opposed to a "DUID%IAID") static association is defined,
the address will be given out on a first-come-first-serve basis.

Signed-off-by: David Härdeman <[email protected]>
Link: https://github.com/openwrt/odhcpd/pull/255
Signed-off-by: Álvaro Fernández Rojas <[email protected]>

odhcpd: support assignments on the basis of IAID

With this patch, IAIDs are actually taken into account when creating
assignments, which allows per-IAID assignments.

The old odhcpd behaviour is still kept for assignments which only
specify a DUID. That behaviour is a first-come-first-serve basis, where
the first request (no matter what the IAID is) will get the static
address, and later requests with different IAIDs will get a NoAddrsAvail
error message (in other words, assignments are not re-assigned and
random adresses are not provided when the "main" address is already
assigned, if the DUID is known in a static config).

The old odhcpd behaviour can be described as follows:

Imagine a "dhcp" configuration file with a single static assignment:

config host
option name 'test'
option ip '192.168.44.2'
option mac '00:00:00:00:ca:fe'
option duid '0003000100000000cafe'
option hostid '02'

And odhcpd running on an interface with IPv6 addr fd00:aaaa::1/48, using
a simple test client (command line parameters should be self-evident
here, foo-client is the client part of a veth interface pair, odhcpd is
listening to the peer interface foo-server):

$ sudo dhcpdig --iaid=0x01 --duid=f000 foo-client
    IPv6: fd00:aaaa::9c5
$ sudo dhcpdig --iaid=0x01 --duid=f000 foo-client
    IPv6: fd00:aaaa::9c5
$ sudo dhcpdig --iaid=0x02 --duid=f000 foo-client
    IPv6: fd00:aaaa::817
$ sudo dhcpdig --iaid=0x02 --duid=f000 foo-client
    IPv6: fd00:aaaa::817
$ sudo dhcpdig --iaid=0x01 --duid=f000 foo-client
    IPv6: fd00:aaaa::9c5
$ sudo dhcpdig --iaid=0x01 --duid=0003000100000000cafe foo-client
    IPv6: fd00:aaaa::2
$ sudo dhcpdig --iaid=0x01 --duid=0003000100000000cafe foo-client
    IPv6: fd00:aaaa::2
$ sudo dhcpdig --iaid=0x02 --duid=0003000100000000cafe foo-client
  STATUS: NoAddrsAvail (2)

IOW, unknown DUIDs get randomly assigned per-IAID IPv6 addresses. Known
statically determined DUIDs will be awarded to the first client
attempting a lease, no matter what the IAID is, but the IAID will be
remembered and subsequent requests with other IAIDs will fail.

With this patch applied, assume a "dhcp" configuration file with the
following assignments:

config host
option name 'testcafe'
option ip '192.168.44.2'
option mac '00:00:00:00:ca:fe'
option hostid '02'
option duid '0003000100000000cafe%123'

config host
option name 'testbeef'
option ip '192.168.44.3'
option mac '00:00:00:00:be:ef'
option hostid '03'
option duid '0003000100000000beef'

config host
option name 'testfood'
option ip '192.168.44.4'
option mac '00:00:00:00:f0:0d'
option hostid '04'
option duid '0003000100000000f00d'

config host
option name 'testfood2'
option ip '192.168.44.5'
option mac '00:00:00:00:f0:0d'
option hostid '05'
option duid '0003000100000000f00d%123'

Now, using the same test client:

$ sudo ./build/dhcpdig --iaid=0x000 --duid=0003000100000000cafe foo-client
    IPv6: fd00:aaaa::9c4
$ sudo ./build/dhcpdig --iaid=0x123 --duid=0003000100000000cafe foo-client
    IPv6: fd00:aaaa::2
$ sudo ./build/dhcpdig --iaid=0x000 --duid=0003000100000000cafe foo-client
    IPv6: fd00:aaaa::9c4
$ sudo ./build/dhcpdig --iaid=0xabc --duid=0003000100000000beef foo-client
    IPv6: fd00:aaaa::3
$ sudo ./build/dhcpdig --iaid=0x123 --duid=0003000100000000beef foo-client
  STATUS: NoAddrsAvail (2)
$ sudo ./build/dhcpdig --iaid=0xabc --duid=0003000100000000f00d foo-client
    IPv6: fd00:aaaa::4
$ sudo ./build/dhcpdig --iaid=0xabc --duid=0003000100000000f00d foo-client
    IPv6: fd00:aaaa::4
$ sudo ./build/dhcpdig --iaid=0x123 --duid=0003000100000000f00d foo-client
    IPv6: fd00:aaaa::5

See also: https://github.com/openwrt/odhcpd/issues/175

Signed-off-by: David Härdeman <[email protected]>
Link: https://github.com/openwrt/odhcpd/pull/255
Signed-off-by: Álvaro Fernández Rojas <[email protected]>

odhcpd: support IAIDs for static DHCPv6 leases

Extend the string format for duids to alternatively support
"<duid>%<iaid>", which makes it possible to define separate static
leases, e.g. for a client which is connected to the same network with
more than one interface (which will both have the same DUID, as it is a
per-host identifier, but different IAIDs).

Note that this only wires up the new format, actually using it to make
DHCPv6-IA decisions will be implemented in subsequent patches.

Signed-off-by: David Härdeman <[email protected]>
Link: https://github.com/openwrt/odhcpd/pull/255
Signed-off-by: Álvaro Fernández Rojas <[email protected]>

odhcpd: break up complex matching logic

Like Winnie the Pooh, “I am a Bear of Very Little Brain, and long words
Bother me”.

Simplify the assignment checking logic by breaking it up a bit. This is
in preparation for subsequent patches.

Signed-off-by: David Härdeman <[email protected]>
Link: https://github.com/openwrt/odhcpd/pull/255
Signed-off-by: Álvaro Fernández Rojas <[email protected]>

odhcpd: document the ubus interface

I guess this might be subject to change (for example, we probably want to
also broadcast events for DHCPv6 leases in the future, and I guess the
ack/release/expire events that are currently broadcast might benefit from
being renamed to something that makes it clearer that they are DHCPv4
specific events), but this is the current ubus interface.

Signed-off-by: David Härdeman <[email protected]>
Link: https://github.com/openwrt/odhcpd/pull/270
Signed-off-by: Álvaro Fernández Rojas <[email protected]>

dhcpv4: generate dbus events on lease expiry

We already generate events when a lease is acquired and released, so it seems
consistent to also generate events when a lease expires.

Signed-off-by: David Härdeman <[email protected]>
Link: https://github.com/openwrt/odhcpd/pull/270
Signed-off-by: Álvaro Fernández Rojas <[email protected]>

dhcpv4: fix ubus events

Note that req->ciaddr which was used to generate the broadcast message is
completely under client control and isn't checked, meaning that a
buggy/malicious client could cause broadcast messages containing an arbitrary
IP address.

While addressing this, move the broadcast message generation into
dhcpv4_lease(), so that the function can release the assignment straight away
and return NULL (it seems to return the released assignment just so that it can
be broadcast and later reaped by the expiry timer).

Signed-off-by: David Härdeman <[email protected]>
Link: https://github.com/openwrt/odhcpd/pull/270
Signed-off-by: Álvaro Fernández Rojas <[email protected]>

odhcpd: remove mac_len argument to ubus_bcast_dhcp_event()

All the places that call ubus_bcast_dhcp_event() have already checked that the
hardware adress is a MAC address, and it's not like we support anything else,
so remove the length argument.

The function is actually DHCPv4 specific, but that's a topic for another time.

Signed-off-by: David Härdeman <[email protected]>
Link: https://github.com/openwrt/odhcpd/pull/270
Signed-off-by: Álvaro Fernández Rojas <[email protected]>

odhcpd: fix ubus support flag in help msg

The cmake definition is a bit sneaky, the cmake variable is "UBUS", but it
leads to "WITH_UBUS" being defined for the compilation.

Signed-off-by: David Härdeman <[email protected]>
Link: https://github.com/openwrt/odhcpd/pull/270
Signed-off-by: Álvaro Fernández Rojas <[email protected]>

odhcpd: reduce use of WITH_UBUS defines in code

By defining some dummy functions in the header file, we can avoid sprinkling
WITH_UBUS defines through the code, which helps readability and also ensures
that compilation errors (e.g. when a variable is renamed) is caught in builds
with and without ubus.

Signed-off-by: David Härdeman <[email protected]>
Link: https://github.com/openwrt/odhcpd/pull/270/
Signed-off-by: Álvaro Fernández Rojas <[email protected]>

ndp: fix macOS IPv6 compatibility by using link-local source addresses

macOS ignores NDP packets that don't originate from link-local addresses,
causing IPv6 connectivity issues with odhcpd. This change ensures NDP
packets (Neighbor Advertisements and ICMP Echo Requests) are sent using
link-local source addresses for RFC 4861 compliance.

Changes:
* Add ndp_from_link_local configuration flag (defaults to true)
* Add odhcpd_send_with_src() to allow explicit source address control
* Add odhcpd_try_send_with_src() helper to eliminate code duplication
* Add odhcpd_get_interface_linklocal_addr() with caching for performance
* Update send_na() and ping6() to use link-local source addresses when
  enabled
* Add RFC 4861, §4.2 comments explaining the mandated behavior
* Maintain backward compatibility with fallback behavior

Fixes: openwrt/openwrt#7561 #202
Signed-off-by: Stephen Groat <[email protected]>
Link: https://github.com/openwrt/odhcpd/pull/242
Signed-off-by: Álvaro Fernández Rojas <[email protected]>

router: fix SLAAC on subnets > 64

Signed-off-by: Álvaro Fernández Rojas <[email protected]>

dhcpv4: simplify dhcpv4_setup_interface()

In the disable case, the function would close the interface uloop fd, free
assignments, check if the interface uloop fd should be closed, and return 0.

Simplify the function by bailing early in the disable case, which also removes
one level of indentation for the gist of the function body.

Signed-off-by: David Härdeman <[email protected]>
Link: https://github.com/openwrt/odhcpd/pull/266
Signed-off-by: Álvaro Fernández Rojas <[email protected]>

dhcpv4: make the cookie explicit in struct dhcpv4_message

This makes the struct more explicit while allowing a number of magic offsets to
be removed.

Note that adding the packed attribute to struct dhcpv4_message is necessary
because the incoming data is cast directly to the struct. Adding the attribute
will give the struct an alignment of 1 instead of 4 (the normal alignment of
struct in_addr and uint32_t).

In the beginning of dhcpv4_handle_msg(), we cast "void *data" to struct
dhcpv4_message, but we can't know the alignment of "*data" (meaning it has
alignment 1). And by casting "*data" to a struct, we're actually promising the
compiler that "*data" has the same alignment as the struct.

Later, the address of members of struct dhcpv4_message are passed to other
functions, but since struct dhcpv4_message can be at any memory location, it's
not a given that the pointers are aligned. This issue only becomes visible once
the packed attribute is added, which is why the temporary (aligned) struct
in_addr variables are added in the WITH_UBUS blocks.

Signed-off-by: David Härdeman <[email protected]>
Link: https://github.com/openwrt/odhcpd/pull/266
Signed-off-by: Álvaro Fernández Rojas <[email protected]>

dhcpv4: simplify dhcpv4_fr_send() a bit

Some minor cleanups to dhcpv4_fr_send()

Signed-off-by: David Härdeman <[email protected]>
Link: https://github.com/openwrt/odhcpd/pull/266
Signed-off-by: Álvaro Fernández Rojas <[email protected]>

dhcpv4: rename cookie variable

The name is potentially confusing given that DHCPv4 includes the concept of a
magic cookie.  The variable is actually used as a cursor/write ptr, so rename
it to something more descriptive.

Signed-off-by: David Härdeman <[email protected]>
Link: https://github.com/openwrt/odhcpd/pull/266
Signed-off-by: Álvaro Fernández Rojas <[email protected]>

dhcpv4: remove some magic numbers

Add a bunch of #define's to dhcpv4.h and use the defined constants to reduce
the number of magic constants.

Signed-off-by: David Härdeman <[email protected]>
Link: https://github.com/openwrt/odhcpd/pull/266
Signed-off-by: Álvaro Fernández Rojas <[email protected]>

dhcpv4: update dhcpv4_msg_to_string()

Simplify and make it cover all currently known message types.

Signed-off-by: David Härdeman <[email protected]>
Link: https://github.com/openwrt/odhcpd/pull/266
Signed-off-by: Álvaro Fernández Rojas <[email protected]>

dhcpv4: dhcpv4_fr_rand_delay() fixups

Move constants to dhcpv4.h, and make their names more descriptive. Also, use
abs() instead of labs(), since "msecs" is an int.

Signed-off-by: David Härdeman <[email protected]>
Link: https://github.com/openwrt/odhcpd/pull/266
Signed-off-by: Álvaro Fernández Rojas <[email protected]>

ubus: improve correspondence between DHCPv[46]

The same parameter is called "accept-reconf-nonce" in the "ipv4leases" case and
"accept-reconf" in the "ipv6leases". I couldn't find anything in the OpenWrt
trees which depended on either naming, but renaming the IPv4 case seems to be
the safer bet since it is not part of the standard installation.

Signed-off-by: David Härdeman <[email protected]>
Link: https://github.com/openwrt/odhcpd/pull/267
Signed-off-by: Álvaro Fernández Rojas <[email protected]>

ubus: minor correctness fix

Now, blobmsg_close_table() and blobmsg_close_array() happen to be identical, so
the actual code doesn't change at all, but this still caused me some
headscratching when I first saw it.

Signed-off-by: David Härdeman <[email protected]>
Link: https://github.com/openwrt/odhcpd/pull/267
Signed-off-by: Álvaro Fernández Rojas <[email protected]>

ubus: don't expose ipv4leases if not supported

This risks confusing other clients, since the "ipv4leases" method will return
an empty list. Better to not have the method available on dbus at all, so
clients will get an error when odhcpd is built without IPv4 support.

Signed-off-by: David Härdeman <[email protected]>
Link: https://github.com/openwrt/odhcpd/pull/267
Signed-off-by: Álvaro Fernández Rojas <[email protected]>

odhcpd: print compiled-in features in help message

This is mostly useful for hacking, as a way to quickly determine from the command-line
how odhcpd has been built. But I also plan to use it in LuCI [1].

[1] https://github.com/openwrt/luci/blob/master/modules/luci-base/root/usr/share/rpcd/ucode/luci

Signed-off-by: David Härdeman <[email protected]>
Link: https://github.com/openwrt/odhcpd/pull/268
Signed-off-by: Álvaro Fernández Rojas <[email protected]>

dhcpv4: rename a variable in dhcpv4_assign()

Rename "buf" to something more descriptive.

Signed-off-by: David Härdeman <[email protected]>
Link: https://github.com/openwrt/odhcpd/pull/264
Signed-off-by: Álvaro Fernández Rojas <[email protected]>

dhcpv4: move the remaining forward declaration

Down to where it is clearer why it is needed (there's a circular dependency
between dhcpv4_fr_delay_timer() and dhcpv4_fr_rand_delay()).

Signed-off-by: David Härdeman <[email protected]>
Link: https://github.com/openwrt/odhcpd/pull/264
Signed-off-by: Álvaro Fernández Rojas <[email protected]>

dhcpv4: move find_assignment_by_hwaddr()

Move find_assignment_by_hwaddr() down to just before the sole user
(dhcpv4_lease()).

Signed-off-by: David Härdeman <[email protected]>
Link: https://github.com/openwrt/odhcpd/pull/264
Signed-off-by: Álvaro Fernández Rojas <[email protected]>

dhcpv4: reorder some more functions

Move dhcpv4_free_assignment(), dhcpv4_insert_assignment(), dhcpv4_assign() and
dhcpv4_lease() further up, allowing more forward declarations to be removed.

Signed-off-by: David Härdeman <[email protected]>
Link: https://github.com/openwrt/odhcpd/pull/264
Signed-off-by: Álvaro Fernández Rojas <[email protected]>

dhcpv4: move struct dhcpv4_dnr to header

This helps declutter dhcpv4.c a little bit.

Signed-off-by: David Härdeman <[email protected]>
Link: https://github.com/openwrt/odhcpd/pull/264
Signed-off-by: Álvaro Fernández Rojas <[email protected]>

dhcpv4: move dhcpv4_fr_stop() up

Thus avoiding another forward declaration.

Signed-off-by: David Härdeman <[email protected]>
Link: https://github.com/openwrt/odhcpd/pull/264
Signed-off-by: Álvaro Fernández Rojas <[email protected]>

dhcpv4: move dhcpv4_free_assignment()

Move dhcpv4_free_assignment() further down so it is grouped with the functions
creating the assignment.

Signed-off-by: David Härdeman <[email protected]>
Link: https://github.com/openwrt/odhcpd/pull/264
Signed-off-by: Álvaro Fernández Rojas <[email protected]>

dhcpv4: replace ip4toa() with inet_ntop()

Replace custom ip4toa() function with inet_ntop().

Signed-off-by: David Härdeman <[email protected]>
Link: https://github.com/openwrt/odhcpd/pull/264
Signed-off-by: Álvaro Fernández Rojas <[email protected]>

dhcpv4: move and rename handle_dhcpv4()

Move handle_dhcpv4() below the function it calls (dhcpv4_handle_msg()) and
rename it to dhcpv4_handle_dgram() for consistency with other function names.

Signed-off-by: David Härdeman <[email protected]>
Link: https://github.com/openwrt/odhcpd/pull/264
Signed-off-by: Álvaro Fernández Rojas <[email protected]>

dhcpv4: rename handle_addrlist_change()

Rename handle_addrlist_change() to dhcpv4_addrlist_change() for more
consistency with the other function names.

Signed-off-by: David Härdeman <[email protected]>
Link: https://github.com/openwrt/odhcpd/pull/264
Signed-off-by: Álvaro Fernández Rojas <[email protected]>

dhcpv4: rename setup_dhcpv4_addresses()

Rename setup_dhcpv4_addresses() to dhcpv4_setup_addresses() for consistency
with the rest of the function names.

Signed-off-by: David Härdeman <[email protected]>
Link: https://github.com/openwrt/odhcpd/pull/264
Signed-off-by: Álvaro Fernández Rojas <[email protected]>

dhcpv4: reorder more functions

Move dhcpv4_setup_interface(), setup_dhcpv4_addresses() and
handle_addrlist_change() towards the end of dhcpv4.c, closer to where they are
actually used, and remove two forward declarations.

Signed-off-by: David Härdeman <[email protected]>
Link: https://github.com/openwrt/odhcpd/pull/264
Signed-off-by: Álvaro Fernández Rojas <[email protected]>

dhcpv4: rename valid_until_cb()

Renaming valid_until_cb() to dhcpv4_valid_until_cb() is more consistent with
the other function names and is helpful if the function ever shows up in a
stack trace.

Signed-off-by: David Härdeman <[email protected]>
Link: https://github.com/openwrt/odhcpd/pull/264
Signed-off-by: Álvaro Fernández Rojas <[email protected]>

dhcpv4: move dhcpv4_init() to end of dhcpv4.c

This removes the need for some forward declarations. Also, inline the two
structs that are only used inside dhcpv4_init().

Signed-off-by: David Härdeman <[email protected]>
Link: https://github.com/openwrt/odhcpd/pull/264
Signed-off-by: Álvaro Fernández Rojas <[email protected]>

ndp: Allow NS loopback for master iface

This commit modifies handle_solicit() in ndp.c to correct IPv6 relay
handling of Neighbor Solicitation (NS) requests when no upstream
solicitation is received.

Background: In IPv6 relay mode, odhcpd discovers local devices only upon
receiving upstream NS packets. If no upstream NS arrives (e.g. because the
upstream router’s neighbor cache is still valid or no solicitation was
ever sent), OpenWrt may attempt neighbor resolution via the master (WAN)
interface instead of the LAN, leaving local devices undiscoverable and
breaking connectivity.

- When an NS packet is sent by the host's master interface, do not
  immediately return; instead continue searching slave interfaces for the
  target neighbor.
- When odhcpd responds to NS packets, add a check to prevent replying to
  NS packets that were sent by the host itself.

Signed-off-by: Haoyi Ci [email protected]
Link: https://github.com/openwrt/odhcpd/pull/240
Signed-off-by: Álvaro Fernández Rojas <[email protected]>

router: log “Sending a RA on lan” at LOG_DEBUG

This reduces log spam when dhcp.odhcpd.loglevel ≥ 5, as neighbor
discovery router advertisements are routinely generated by timer and in
response to router solicitations.

Signed-off-by: Mark Mentovai <[email protected]>
Link: https://github.com/openwrt/odhcpd/pull/191
Signed-off-by: Álvaro Fernández Rojas <[email protected]>

odhcpd: update cmake file

Now that the minimum cmake version has been bumped to 3.13, the cmake file can
be modernized a bit. Although it might look like a lot of changes, most of them
are quite straightforward.

Every library is located in a consistent manner (allowing command-line
overrides of library locations, useful for local development, and also makes it
trivial to do static linking, if desired).

The compiler flags have been broken up to have one per line (making it easy in
the future to add/remove flags, which would create a simple one-line diff).

Although it might look like this bumps the C standard level, cmake is actually
smart enough to pick an earlier version if C11 isn't supported by the compiler
(quite unlikely on any versions of gcc currently in use in OpenWrt and even on
old distros).

Signed-off-by: David Härdeman <[email protected]>
Link: https://github.com/openwrt/odhcpd/pull/258

odhcpd: convert README to markdown

Convert the README file to GitHub style markdown. I don't think it affects the
readability of the plain text file much (or not at all, rather). But it makes
it much easier to read for people looking at the GitHub repo.

I've intentionally kept the fancy markup at a minimum and also haven't changed
any help text messages in order to not get bogged down in how each help message
should read. That is something which can be addressed later.

Signed-off-by: David Härdeman <[email protected]>
Link: https://github.com/openwrt/odhcpd/pull/263
Signed-off-by: Álvaro Fernández Rojas <[email protected]>

odhcpd: allow the use of an alternative cfg file

This is mainly useful for development/testing, since it obviates the
need for a cfg file below the /etc hierarchy.

Also, fixup the help message (long options like "--help" are not
supported).

Signed-off-by: David Härdeman <[email protected]>
Link: https://github.com/openwrt/odhcpd/pull/257
Signed-off-by: Álvaro Fernández Rojas <[email protected]>

odhcpd: remove confusing #defines

Right now, odhcpd.h contains lines like "#define hwaddr mac", which is a great
footgun when hacking on odhcpd (like when creating a function with a local
variable named "hwaddr" and later getting compiler errors talking about how
"mac" has the wrong type...in a function which has no variable named "mac").

So, remove these definitions. I'm not sure the chosen variable names are the
most descriptive, but I erred on the side of caution and picked the alternative
that required the least amount of changes. We can always change the names of
the variables later.

Signed-off-by: David Härdeman <[email protected]>
Link: https://github.com/openwrt/odhcpd/pull/261
Signed-off-by: Álvaro Fernández Rojas <[email protected]>

odhcpd: improve RFC9096 § 3.5 SLAAC compliance

When a CE router provides LAN-side address-configuration information
via SLAAC:

*  A CE router sending RAs that advertise prefixes belonging to a
   dynamically learned prefix (e.g., via DHCPv6-PD) SHOULD record, on
   stable storage, the list of prefixes being advertised via PIOs on
   each network segment and the state of the "A" and "L" flags of the
   corresponding PIOs.

*  Upon changes to the advertised prefixes, and after bootstrapping,
   the CE router advertising prefix information via SLAAC proceeds as
   follows:

   -  Any prefixes that were previously advertised by the CE router
      via PIOs in RA messages, but that have now become stale, MUST
      be advertised with PIOs that have the "Valid Lifetime" and the
      "Preferred Lifetime" set to 0 and the "A" and "L" bits
      unchanged.

   -  The aforementioned advertisements MUST be performed for at
      least the "Valid Lifetime" previously employed for such
      prefixes.

This should be enabled by default with a folder in "/tmp" to avoid forcing
flash writes for all the users and at least handle the stale PIOs on PPPoe
reconnections. For example:
  uci set dhcp.odhcpd.piofolder="/tmp/odhcpd-piofolder"
  uci commit dhcp

There's a new ubus call which allows getting all the IPv6 RA PIOs, with
information about their lifetime and stale status:
  ubus call dhcp ipv6ra
  {
    "interfaces": {
      "br-lan": [
        {
          "lifetime": 5390,
          "prefix": "2001:db8:0:100::/64",
          "stale": true
        },
        {
          "prefix": "2001:db8:0:200::/64",
          "stale": false
        },
        {
          "prefix": "fd00::/64",
          "stale": false
        }
      ],
      "eth1.10": [
        {
          "lifetime": 5390,
          "prefix": "2001:db8:0:110::/64",
          "stale": true
        },
        {
          "prefix": "2001:db8:0:210::/64",
          "stale": false
        },
        {
          "prefix": "fd00:0:0:10::/64",
          "stale": false
        }
      ]
    }
  }

The current implementation performs a flash write per interface whenever
a new prefix is added or an existing prefix becomes stale.
Since some ISPs assign static or semi-static prefixes to their customers,
it should be up to the user to enable flash writes when needed.

Link: https://github.com/openwrt/odhcpd/pull/256
Signed-off-by: Álvaro Fernández Rojas <[email protected]>

dhcpv4: allow lease takeover (bugfix)

Multiple MAC addresses are supported per static lease (congruent to dnsmasq)
Request on different MAC frees any old assignments for a given lease.

Signed-off-by: David Härdeman <[email protected]>

odhcpd: bump minimum cmake version

This removes warnings with more recent versions of cmake.

3.13 is already required by e.g. ubus, so it seems like a reasonable
minimum.

Signed-off-by: David Härdeman <[email protected]>

odhcpd: apply RFC9096 recommended lifetimes

To address issues with hosts retaining stale IPv6 configuration, this change
implements the Recommended Option Lifetimes Configuration Values from
RFC 9096-Section 4.

The `ra_lifetime` is set to 2700s (45min) by default, and the DHCPv6-PD is now
capped at 2700s for preferred lifetime and 5400s (90min) for valid lifetime.

The following changes are introduced:

- The `ra_useleasetime` option is removed.
- New options `max_preferred_lifetime` and `max_valid_lifetime` are added to
  configure the upper limits for prefix lifetimes.
- The Router Advertisement, DNS, and DNR option lifetimes are now dynamically
  adjusted based on the advertised prefix lifetimes to ensure consistency.

These changes help to mitigate issues with stale configurations on client
devices, particularly after network renumbering events.

Signed-off-by: Aviana Cruz <[email protected]>

odhcpd: support multiple per-client MAC addresses

Allow the configuration of multiple per-client DHCPv4 MAC addresses. This is
for feature parity with dnsmasq.

Closes: #221
Closes: openwrt/openwrt#9598
Signed-off-by: David Härdeman <[email protected]>

router: replace ssize_t with size_t

Both valid_addr_cnt and invalid_addr_cnt should be defined with size_t
instead of ssize_t because addr6_len and invalid_addr6_len are defined
with size_t.

Signed-off-by: Álvaro Fernández Rojas <[email protected]>

dhcpv6: add ipv6 pxe support

1. Implement PxE in separate files where possible
2. Update README
3. User can add IPv6 PxE entries in `/etc/config/dhcp`.
4. The new section type is "boot6".
5. The compulsory "url" string specifies the URL to the bootable image.
6. The optional "arch" integer specifies the expected client machine type.
7. The last "boot6" section without "arch" defines the default bootable image.

Signed-off-by: Hang Zhou <[email protected]>

router: move pref64 calculations to the config stage

If pref64 is misconfigured, it'll print a LOG_WARNING message every time a RA
is generated, which seems a bit excessive.

When fixing that, I noticed that all work is done for every RA, so I ended up
moving the gist over to the config stage where it is done once.

Signed-off-by: David Härdeman <[email protected]>

odhcpd: add a helper function for addr6/prefix parsing

Which allows a couple of variations of the same code to be removed...
the helper function also won't modify the input string.

Signed-off-by: David Härdeman <[email protected]>

netlink: fix a memory leak

valgrind noted that addrs is allocated in netlink_get_interface_linklocal(),
but never freed in one case. While I was looking at that, I found a mysterious
double-realloc in netlink.c.

Signed-off-by: David Härdeman <[email protected]>

odhcpd: make the IPv6 RA DNR lifetime configurable

Add a fake SVC parameter "_lifetime=" which allows explicit configuration of
the maximum time in seconds over which a given ADN announcement is valid. In
particular, this allows announcing ADNs which should no longer be used (zero
lifetime, see rfc9463, §6.1).

In the absence of an explicit lifetime, the previous logic is kept (i.e. the
lifetime is set to 3 * MaxRtrAdvInterval, as per §6.1 of RFC9463).

Signed-off-by: David Härdeman <[email protected]>

odhcpd: add DNR (RFC 9463) support

Discovery of Network-designated Resolvers (DNR) allows devices on the network
to discover encrypted DNS resolvers, which has so far required either manual
configuration or other approaches (like systemd-resolved's "opportunistic"
mode).

To enable DNR, a new uci parameter has been added, which needs to contain at
the very least, the priority (1-65535, lower = higher priority) and the server
hostname (Authentication Domain Name, ADN, to use the wording of RFC9463):

config dhcp 'lan'
        …
        list dnr '100 foobar.example.com'

Optionally (and preferably), a comma-separated list of IP addresses and
SvcParams can also be specified, like this (line wrapping added):

config dhcp 'lan'
        …
        list dnr '100 resolver1.example.com
                    fda7:ab54:69fb::1,fda7:ab54:69fb::2,10.0.0.1
                    alpn=dot port=853'
        list dnr '200 resolver2.example.com
                    fda7:ab54:69fb::2,10.0.1.1,10.0.1.2
                    alpn=dot port=853'

Client support is on it's way (e.g. in systemd PR #30952 or in the Windows
Insiders program).

Signed-off-by: David Härdeman <[email protected]>

dhcpv4: store reqopts as uint8_t

Storing request options as chars means that it's difficult to handle option
codes > 127 due to the signedness of char (and options aren't chars, they're
binary data).

Signed-off-by: David Härdeman <[email protected]>

config: set RFC defaults for preferred lifetime

Update preferred lifetime default value in accordance with RFC4861.

Signed-off-by: Paul Donald <[email protected]>
[ fix comment style ]
Signed-off-by: Christian Marangi <[email protected]>

router: disambiguate and clarify 'no route' messages

Better describe "no route" messages.

Signed-off-by: Paul Donald <[email protected]>
[ improve commit title ]
Signed-off-by: Christian Marangi <[email protected]>

router: rename minvalid to lowest_found_lifetime

The variable is used across all prefix lifetimes, serving as a low-tide
mark. It stores the lowest lifetime among all prefixes.

Signed-off-by: Paul Donald <[email protected]>
[ fix comment format ]
Signed-off-by: Christian Marangi <[email protected]>

treewide: normalize spaces to tabs

Normalize space to tabs.

Signed-off-by: Paul Donald <[email protected]>
[ improve commit title and description ]
Signed-off-by: Christian Marangi <[email protected]>

treewide: spell-fixes and new comments for extra clarification

Fix some spell mistake and add new comments for extra clatification.

Signed-off-by: Paul Donald <[email protected]>
Reviewed-by: Daniel Golle <[email protected]>
[ improve commit title ]
Signed-off-by: Christian Marangi <[email protected]>

router: limit prefix preferred_lt to valid_lt in accordance with RFC4861

Follow-up fix for bc9d317f2921 ("dhcpv6-ia: fix invalid preferred
lifetime").

https://www.rfc-editor.org/rfc/rfc4861#page-44

Fixes: bc9d317f2921 ("dhcpv6-ia: fix invalid preferred lifetime")
Signed-off-by: Paul Donald <[email protected]>
Reviewed-by: Daniel Golle <[email protected]>
[ fix comment format and improve commit title ]
Signed-off-by: Christian Marangi <[email protected]>

router: inherit user-assigned preferred_lifetime

Inherit preferred_lifetime value irrespective of whether ra_useleasetime
is set or not.

User-provided values for preferred_lifetime are now assigned, instead of
ignored.

Before:
==
ICMPv6 Option (Prefix information : fd51:1c2a:8909::/64)
    Type: Prefix information (3)
    Length: 4 (32 bytes)
    Prefix Length: 64
    Flag: 0xc0, On-link flag(L), Autonomous address-configuration flag(A)
    Valid Lifetime: Infinity (4294967295)
    Preferred Lifetime: Infinity (4294967295)
    Reserved
    Prefix: fd51:1c2a:8909::
==

After (preferred_lifetime set to 7 minutes):
==
ICMPv6 Option (Prefix information : fd51:1c2a:8909::/64)
    Type: Prefix information (3)
    Length: 4 (32 bytes)
    Prefix Length: 64
    Flag: 0xc0, On-link flag(L), Autonomous address-configuration flag(A)
    Valid Lifetime: Infinity (4294967295)
    Preferred Lifetime: 420
    Reserved
    Prefix: fd51:1c2a:8909::
==

Signed-off-by: Paul Donald <[email protected]>
Reviewed-by: Daniel Golle <[email protected]>
[ fix comment format and improve commit description ]
Signed-off-by: Christian Marangi <[email protected]>

treewide: refactor valid to valid_lt (lifetime)

Refactor "valid" (valid what?) to "valid_lt".

Signed-off-by: Paul Donald <[email protected]>
Reviewed-by: Daniel Golle <[email protected]>
[ improve commit title ]
Signed-off-by: Christian Marangi <[email protected]>

treewide: refactor pref(erred) to preferred_lt (lifetime)

Refactor "preferred" (preferred what?) and "pref" (obscure) to
"preferred_lt".

It is now more difficult to conflate prefix, preference and other "pref"
related terminology with preferred_lifetime.

Signed-off-by: Paul Donald <[email protected]>
Reviewed-by: Daniel Golle <[email protected]>
[ improve commit title ]
Signed-off-by: Christian Marangi <[email protected]>

config: make sure timer is not on the timeouts list before freeing

Signed-off-by: Colin Whittaker <[email protected]>
Signed-off-by: Chad Monroe <[email protected]>

add hostsfile output in addition to statefile

a92c0a7 made the temporary state/leasefile hidden so that an atomic
change was made and dnsmasq only saw the new file on rename.  A
misguided optimisation was made to only rename the temporary file if
something had changed.  Unfortunately only address and hostnames were
considered in the change, lease durations were not.

As a result it was possible for LUCI which consumes the state/leasefile
to report DHCPv6 leases had expired when they had not.

Revert the optimisation so that the file rename occurs irrespective of
content change, this keeps LUCI reporting of state/lease expiry correct.

This leaves us back with hosts file/dnsmasq update problem. Solve this
by writing out a separate hosts file.  Update this file using the
original IP/Hostname change logic that prompts calling the 'lease'
script.

odhcpd config now supports a string 'hostsfile' which defines the path
and name of the hosts file in an identical manner to 'leasefile'.  A
state 'leasefile' must be defined IF a 'hostsfile' is also required.

eg.

leasefile '/tmp/odhcpdstate'
hostsfile '/tmp/hosts/odhcpdhosts'

Signed-off-by: Kevin Darbyshire-Bryant <[email protected]>

dhcpv4: improve error when a prefix is too long

If a user tries to enable dhcpv4 on an interface with a /29, odhcp won't work.
The logs will only contain a message that doesn't help identify the problem.
It'd be idea to support any prefix with a valid pool, but at least this would
point a confused user in the right direction.

Signed-off-by: Ross Vandegrift <[email protected]>

odhcpd: add support for dhcpv6_pd_min_len parameter

The dhcpv6_pd_min_len configuration clamps the requested prefix
delegation to be at least as big as the option.  This allows a
router to manage the size of each downstream router's prefix
delegation length independently from the delegating interface's
prefix length.

This behavior is an implementation choice permitted by the RFCs.
The delegating router (us) is not required to honor the hint
(RFC3633, section 11.2, we MAY choose to use the information in the
option; RFC8168, section 3.2 has several SHOULDs about desired
choices for selecting a prefix to delegate).

This configuration allows us to conserve prefix space so that any
single router can't grab too much of it.  Consider if we have an
interface with a /56 prefix.  A requesting router could ask for a
/58 and take 1/4 of our total address space.  But if we set a
minimum of /60, we can limit each requesting router to get only 1/16
of our total address space.

sample config:

config dhcp 'pd'
    ...
    option dhcpv6_pd_min_len '60'

Signed-off-by: John Kohl <[email protected]>
[ use different comment style and fix commit description ]
Signed-off-by: Christian Marangi <[email protected]>