odhcpd: remove fallback DNS search domain

This might require some explanation :)

First, res_init() is marked as deprecated in the Linux man-pages (read: in
glibc), and has been so for a while.

Second, we use musl on OpenWrt, and what does res_init() look like in musl?

 musl/src/network/res_init.c:

int res_init()
{
return 0;
}

 musl/include/resolv.h (commit date 2011-02-12):

/* unused; purely for broken apps */
typedef struct __res_state {

OpenWrt switched to musl sometime in 2015 (from a quick search, don't
quote me on that), and res_init() hasn't worked since then.

Ok, so first I thought I might reimplement res_init(), using the glibc
implementation as inspiration.

glibc's res_init() basically has three sources of domain search data:
 1. The LOCALDOMAIN environment variable
 2. /etc/resolv.conf
 3. gethostname() followed by checking if there's at least one dot in the name

1. The environment variable won't help us, we don't have it on OpenWrt and it
would just be confusing.

2. resolv.conf seems reasonable, but note that it is typically a symlink:
/etc/resolv.conf -> /tmp/resolv.conf -> /tmp/resolv.conf.d/resolv.conf.auto

The latter is created by netifd. Where do we get iface->dns_search from? From
netifd via ubus. In addition, the resolv.conf.auto that netifd generates
includes all dns search domain and we might end up picking a random one (so if
the user has set a domain on the guest network, but not on the main network,
the former might end up being the fallback for the latter, not good).

3. gethostname() will return the hostname, which on OpenWrt is typically set to
exactly that - the hostname and not a FQDN (and this is what the UIs tell the
users to do).

In summary, all these calls to res_init() are pointless right now, and there is
no reasonable fallback once we've failed to get the info we want from netifd.

So, remove the DNS search domain fallback logic if one isn't set. It's not like
DHCPv[46] clients or hosts listening to RAs have any expectation that the
domain search list MUST be defined.

Signed-off-by: David Härdeman <david@hardeman.nu>
Link: https://github.com/openwrt/odhcpd/pull/358
Signed-off-by: Álvaro Fernández Rojas <noltari@gmail.com>

router: remove some log spam in send_router_advert()

Both these messages would otherwise be printed on each RA.

Especially the second one merely means that we're follwing the cfg
explicitly set by the user, so they're not very likely to be interested
to know that we're doing what we've been told to do.

Signed-off-by: David Härdeman <david@hardeman.nu>
Link: https://github.com/openwrt/odhcpd/pull/358
Signed-off-by: Álvaro Fernández Rojas <noltari@gmail.com>

router: improve send_router_advert()

Remove a bunch of memory (re)allocations which are performed
on every single RA. Also remove some pointless casts.

Signed-off-by: David Härdeman <david@hardeman.nu>
Link: https://github.com/openwrt/odhcpd/pull/358
Signed-off-by: Álvaro Fernández Rojas <noltari@gmail.com>

config: add default value for dhcpv6_pd_min_len

Right now, dhcpv6_pd_min_len is zero, meaning any request would be
accepted (as long as it fits what we have available).

That means that with a /56 (a common residential scenario), a downstream
router can request a /57 and gobble up 1/2 of the available address
space. Even a default dhcpv6_pd_min_len value of 60 means that we'd
support 15 PDs in total.

So let's err on the side of caution and set /62 as the default, people
who want larger allocations can easily override the default.

Signed-off-by: David Härdeman <david@hardeman.nu>
Link: https://github.com/openwrt/odhcpd/pull/357
Signed-off-by: Álvaro Fernández Rojas <noltari@gmail.com>

config: allow minimum PD len up to 64

The current limit of [1,62] doesn't really seem to be justified. With
things like RFC9762 (DHCPv6-PD preferred), it isn't unlikely that we'll
see more DHCPv6-PD requests in the future, and not just from actual
routers.

The smallest realistic PD that can be allocated is a /64, so make that
the minimum.

Also, fix a bug in the logic (note that the PD_MIN_LEN_MAX actually
isn't enforced).

Closes: #352
Signed-off-by: David Härdeman <david@hardeman.nu>
Link: https://github.com/openwrt/odhcpd/pull/357
Signed-off-by: Álvaro Fernández Rojas <noltari@gmail.com>

Revert "router: optimize duplicated PIO comparison"

This reverts commit d354ebb66cdcccc3373156468eec1c660fb2922c.

As stated by David Härdeman at [1]:
Note that time_t will have an alignment of 4 (unlikely) or 8, meaning
that struct ra_pio will have padding. A straight memcmp() could
therefore generate the wrong result unless both struct ra_pios that are
compared are guaranteed to have been memset to zero.

Avoid the risk of something going wrong here by doing a straight
comparison of the two members of the struct that we want to compare.

[1]: https://github.com/openwrt/odhcpd/pull/348/changes/d02fce006abd4f0761bf3100c9ddd4ce70847edb

Link: https://github.com/openwrt/odhcpd/pull/354
Signed-off-by: Álvaro Fernández Rojas <noltari@gmail.com>

statefiles: don't consider no hostname as broken

Commit e568bde changed the OAF_BROKEN_HOSTNAME flag to a boolean, but at
the same time changed the semantics slightly so that no hostname was
also considered broken (rather, no hostname means it's neither valid nor
invalid).

So, add a simple fix for this.

Closes: https://github.com/openwrt/odhcpd/issues/350
Signed-off-by: David Härdeman <david@hardeman.nu>
Link: https://github.com/openwrt/odhcpd/pull/351
Signed-off-by: Álvaro Fernández Rojas <noltari@gmail.com>

odhcpd: fix captive_portal_uri reset

odhcpd_reload() calls clean_interface() which does a memset() on the
given struct iface from iface->ra onwards, so captive_portal_uri and
captive_portal_uri_len have to be below iface->ra in order to be zeroed
on reload.

Otherwise, iface->captive_portal_uri will be free():d, but
iface->captive_portal_uri_len could still be > 0.

This is a bit of a footgun (one wouldn't normally expect the order of
struct members to matter), and should probably be dealt with in a more
thorough manner, but I've just done the minimum and added a warning in
src/odhcpd.h for now,

Signed-off-by: David Härdeman <david@hardeman.nu>
Link: https://github.com/openwrt/odhcpd/pull/345
Signed-off-by: Álvaro Fernández Rojas <noltari@gmail.com>

dhcpv4: fix DNS server option

The iface->dhcpv4_own_ip struct is not a struct in_addr, so change the
code so that the iov takes the actual IPv4 address.

Closes: https://github.com/openwrt/odhcpd/issues/344
Signed-off-by: David Härdeman <david@hardeman.nu>
Link: https://github.com/openwrt/odhcpd/pull/345
Signed-off-by: Álvaro Fernández Rojas <noltari@gmail.com>

router: Modify relayed RA PIO P flag according to interface policy

Effectively:
-inherit the P flag from upstream, but retain the possibility to disable
its availability downstream, if any of the following are disabled:

-DHCPv6
-DHCPv6-PD
-DHCPv6-PD Preferred

Signed-off-by: Paul Donald <newtwen+github@gmail.com>
Link: https://github.com/openwrt/odhcpd/pull/317
Signed-off-by: Álvaro Fernández Rojas <noltari@gmail.com>

router: Modify relayed RA PIO A flags according to interface policy

Effectively:
-inherit the A flag from upstream, but retain the possibility to
disable SLAAC downstream.

h/t user @Shine-

Signed-off-by: Paul Donald <newtwen+github@gmail.com>
Link: https://github.com/openwrt/odhcpd/pull/317
Signed-off-by: Álvaro Fernández Rojas <noltari@gmail.com>

README.md: update dhcp ubus events

The names were changed some time ago, but I overlooked the README.md
documentation.

Signed-off-by: David Härdeman <david@hardeman.nu>
Link: https://github.com/openwrt/odhcpd/pull/343
Signed-off-by: Álvaro Fernández Rojas <noltari@gmail.com>

statefiles: don't write empty hosts files

This is mostly cosmetic, skip writing out hosts files for interfaces
where it makes little sense to do so (i.e. where the hosts file will
anyway be empty).

Closes: https://github.com/openwrt/odhcpd/issues/338
Signed-off-by: David Härdeman <david@hardeman.nu>
Link: https://github.com/openwrt/odhcpd/pull/342
Signed-off-by: Álvaro Fernández Rojas <noltari@gmail.com>

Revert "statefiles: fix escape sequence for broken hostname output"

This reverts commit c29aa70914988e09774fc5f49565e8ca88526050.

See https://github.com/openwrt/odhcpd/pull/340#issuecomment-3633182512

Signed-off-by: Álvaro Fernández Rojas <noltari@gmail.com>

statefiles: fix stale pio handling for !ubus

Currently, config.c:odhcpd_reload() will call:
set_config()
// the piodir is set here
set_interface()
config_parse_interface()
statefiles_read_prefix_information()
statefiles_setup_dirfd()
ubus_apply_network()
config_parse_interface()
statefiles_read_prefix_information()

This works when ubus is enabled, because of the extra call to
ubus_apply_network() which will read the pio file after the
dirfd has been set up.

However, it won't work when ubus isn't enabled since it'll
turn ubus_apply_network() into a noop.

Fix this by moving the statefiles_setup_dirfd() calls into
set_config(), i.e. setup the dirfds as soon as the relevant
cfg options have been read.

Signed-off-by: David Härdeman <david@hardeman.nu>
Link: https://github.com/openwrt/odhcpd/pull/333
Signed-off-by: Álvaro Fernández Rojas <noltari@gmail.com>

odhcpd: rename piofolder to piodir

A "folder" is a Windows concept, rename the cfg option from "piofolder"
to "piodir", which will also make it more consistent with other cfg
options.

Signed-off-by: David Härdeman <david@hardeman.nu>
Link: https://github.com/openwrt/odhcpd/pull/333
Signed-off-by: Álvaro Fernández Rojas <noltari@gmail.com>

statefiles: use tmpfile functions for pio files

Reduce code duplication by using the same tmpfile functions
for pio handling as well.

Signed-off-by: David Härdeman <david@hardeman.nu>
Link: https://github.com/openwrt/odhcpd/pull/333
Signed-off-by: Álvaro Fernández Rojas <noltari@gmail.com>

statefiles: rename prefix information functions

This makes it clearer that the functions are in statefiles.c.

Signed-off-by: David Härdeman <david@hardeman.nu>
Link: https://github.com/openwrt/odhcpd/pull/333
Signed-off-by: Álvaro Fernández Rojas <noltari@gmail.com>

config: move pio json handling to statefiles.c

The prefix information files are also examples of statefiles, and there
are some things which can be shared between e.g. the state file logic
and the pio file logic (like the conversion from MONOTIME to time_t and
back, and also the use of tmp files). That's for later patches though,
this just moves the code without any code changes to make subsequent
patches clearer.

Signed-off-by: David Härdeman <david@hardeman.nu>
Link: https://github.com/openwrt/odhcpd/pull/333
Signed-off-by: Álvaro Fernández Rojas <noltari@gmail.com>

statefiles: add a dirfd helper function

The same logic was repeated three times in config.c, move it
to a single function in statefiles.c.

Signed-off-by: David Härdeman <david@hardeman.nu>
Link: https://github.com/openwrt/odhcpd/pull/333
Signed-off-by: Álvaro Fernández Rojas <noltari@gmail.com>

statefiles: add tmp helper functions

Move temporary file creation/finishing to separate functions.

Signed-off-by: David Härdeman <david@hardeman.nu>
Link: https://github.com/openwrt/odhcpd/pull/333
Signed-off-by: Álvaro Fernández Rojas <noltari@gmail.com>

statefiles: fix escape sequence for broken hostname output

Broken hostnames are being reported as "broken\x20-"
instead of "broken -" which I'm guessing is what was
intended.

Signed-off-by: Kevin Darbyshire-Bryant <ldir@darbyshire-bryant.me.uk>
Link: https://github.com/openwrt/odhcpd/pull/340
Signed-off-by: Álvaro Fernández Rojas <noltari@gmail.com>

dhcpv4: don't send zero IPv6-only preferred option

As noted in https://github.com/openwrt/openwrt/issues/21048, if the
ipv6_only_preferred option isn't set (or is set to zero), odhcpd will
still include the IPv6-Only Preferred option (with a zero value) in
DHCPv4 replies, while it should not return the option at all.

Closes: https://github.com/openwrt/openwrt/issues/21048
Signed-off-by: David Härdeman <david@hardeman.nu>
Link: https://github.com/openwrt/odhcpd/pull/341
Signed-off-by: Álvaro Fernández Rojas <noltari@gmail.com>

Revert "dhcpv6-ia: add some noise to the T1 and T2 periods"

Accidental push to wrong branch/repo

This reverts commit 81ea5bfef775b8e2d4aae9826e84a040288864df.

Revert "do not delegate ULA prefixes"

Accidental push to wrong branch - oops

This reverts commit fd4714bb2dfec0fadf5b49f21487f37f4613dc0f.

do not delegate ULA prefixes

dhcpv6-ia: add some noise to the T1 and T2 periods

Without this, all the clients get the same timeouts and try to renew
addresses at the same time.  Reduce the T1 & T2 by a pseudo random 0 to
127 seconds, thus the renew attempts are likely to be spread out a bit
in time.

Signed-off-by: Kevin Darbyshire-Bryant <ldir@darbyshire-bryant.me.uk>

router: skip if branch in PIO clear functions

We can skip the if branch that replaces the expired/duplicated ra_pio
value in the router_clear_*_ra_pio() functions.

Suggested-by: Paul Donald <newtwen+github@gmail.com>
Link: https://github.com/openwrt/odhcpd/pull/336
Signed-off-by: Álvaro Fernández Rojas <noltari@gmail.com>

router: optimize duplicated PIO comparison

Compare prefix and length with a single memcmp instead of performing
separate comparisons.

Suggested-by: Paul Donald <newtwen+github@gmail.com>
Link: https://github.com/openwrt/odhcpd/pull/336
Signed-off-by: Álvaro Fernández Rojas <noltari@gmail.com>

router: remove duplicated PIOs

There can be duplicated PIOs if the user changed the assigned prefix length
in the configuration.

Link: https://github.com/openwrt/odhcpd/pull/336
Signed-off-by: Álvaro Fernández Rojas <noltari@gmail.com>

router: rename router_clear_ra_pio

This function clears the expired RA PIOs, so router_clear_expired_ra_pio
is a better name.

Link: https://github.com/openwrt/odhcpd/pull/336
Signed-off-by: Álvaro Fernández Rojas <noltari@gmail.com>

router: improve RA PIO search logic

We need to properly handle when the user modifies the prefix length
assigned to an interface by:
- Checking if the prefix matches the address taking into account:
    - The biggest length of address and PIO.
    - A minimum PIO of /64.
- Updating addresses and lengths for existing PIOs.

Signed-off-by: Álvaro Fernández Rojas <noltari@gmail.com>

formal: fix workflow permissions

Fix formality check permissions that are needed to post optional
summaries back to the PR.

Link: openwrt/actions-shared-workflows#64
Signed-off-by: George Sapkin <george@sapk.in>
Link: https://github.com/openwrt/odhcpd/pull/325
Signed-off-by: Álvaro Fernández Rojas <noltari@gmail.com>

build: disable static libraries

Allows linking against shared json-c, which reduces the binary size and
matches the OpenWrt package binary.

Signed-off-by: Álvaro Fernández Rojas <noltari@gmail.com>

dhcpv6-ia: fix crash in dhcpv6_free_lease()

Note that dhcpv6_free_lease() can be called with a NULL argument e.g.
from dhcpv6_ia_handle_IAs() when iface->no_dynamic_dhcp is set and a
client without a static lease cfg tries to get a dynamic lease.

Closes: #321
Signed-off-by: David Härdeman <david@hardeman.nu>
Link: https://github.com/openwrt/odhcpd/pull/335
Signed-off-by: Álvaro Fernández Rojas <noltari@gmail.com>

build: fix BUILD_ARGS

Commit ef8de928da7f forced DHCPV4_SUPPORT and UBUS cmake options, but should
have fixed BUILD_ARGS instead (-WITH_UBUS=ON -> -DUBUS=ON).

Fixes: ef8de928da7f ("build: enable DHCPV4_SUPPORT and UBUS")
Signed-off-by: Álvaro Fernández Rojas <noltari@gmail.com>

statefiles: change ipv4 MAC statefile syntax

This changes the MAC representation in the lease/statefile so that it
reads "aa:bb:cc:dd:ee:ff" instead of "aabbccddeeff".

No functional changes, but this makes it easier to parse the statefile
in rpcd-mod-luci (for reporting the MAC addresses in LuCI).

We should probably change lease->hwaddr to be a struct ether_addr,
to make it clearer that we only support MAC addresses as hwaddr,
but that's a topic for a different PR.

Signed-off-by: David Härdeman <david@hardeman.nu>
Link: https://github.com/openwrt/odhcpd/pull/332
Signed-off-by: Álvaro Fernández Rojas <noltari@gmail.com>

odhcpd: remove OAF_BOUND

Replace the OAF_BOUND flag with a simple boolean.

Signed-off-by: David Härdeman <david@hardeman.nu>
Link: https://github.com/openwrt/odhcpd/pull/331
Signed-off-by: Álvaro Fernández Rojas <noltari@gmail.com>

dhcpv6: remove OAF_TENTATIVE

The flag was only set in two places, and never used. Its meaning was
also essentially the inverse of OAF_BOUND.

Signed-off-by: David Härdeman <david@hardeman.nu>
Link: https://github.com/openwrt/odhcpd/pull/331
Signed-off-by: Álvaro Fernández Rojas <noltari@gmail.com>

odhcpd: remove OAF_STATIC

This flag does nothing more than document if lease_cfg is set in the
respective dhcpv[46]_lease structs.

Signed-off-by: David Härdeman <david@hardeman.nu>
Link: https://github.com/openwrt/odhcpd/pull/331
Signed-off-by: Álvaro Fernández Rojas <noltari@gmail.com>

odhcpd: remove OAF_BROKEN_HOSTNAME

A bool is quite a bit more ergonomic (as the diff for this commit
shows).

Signed-off-by: David Härdeman <david@hardeman.nu>
Link: https://github.com/openwrt/odhcpd/pull/331
Signed-off-by: Álvaro Fernández Rojas <noltari@gmail.com>

statefiles: remove hosts entries from statefile

Further simplify statefiles.c by removing the hosts entries from the
statefile (they can be found in the dedicated hosts files instead).

Signed-off-by: David Härdeman <david@hardeman.nu>
Link: https://github.com/openwrt/odhcpd/pull/331
Signed-off-by: Álvaro Fernández Rojas <noltari@gmail.com>

build: enable DHCPV4_SUPPORT and UBUS

Enable DHCPV4_SUPPORT and UBUS in order to correctly build all the source
code.

Signed-off-by: Álvaro Fernández Rojas <noltari@gmail.com>

ubus: fix shadow local variable

Fix leftover shadowed local variable in ubus code from commit a170d63874f2.

src/ubus.c:130:38: error: declaration of 'a' shadows a previous local [-Werror=shadow=local]
  130 |                 struct dhcpv6_lease *a, *border;
      |                                      ^
src/ubus.c:118:15: note: shadowed declaration is here
  118 |         void *a;
      |               ^

Fixes: a170d63874f2 ("src: fix shadowed local variables")
Signed-off-by: Álvaro Fernández Rojas <noltari@gmail.com>

src: fix shadowed local variables

Fix shadowed local variables and enable warning to prevent more shadowed
variables in the future.

Signed-off-by: Álvaro Fernández Rojas <noltari@gmail.com>

dhcpv6-ia: fix a crash when static lease isn't configured

Quoting from https://github.com/openwrt/odhcpd/issues/321:
  [71802.880039] odhcpd[22696]: segfault at 78...

Quoting from https://forum.openwrt.org/t/odhcpd-crash-loop-when-receiving-packet/243015/69:
  [   77.761062] odhcpd[2075]: segfault at 78...

0x78 is the offset of duid_count in struct lease_cfg, so if lease_cfg is
null, we'd end up reading from address 0x78 when trying to read
lease_cfg->duid_count.

This should fix the issue. Thanks to @klipz in the forums for giving me
SSH access to an awesome test sandbox.

Closes: https://github.com/openwrt/odhcpd/issues/321
Signed-off-by: David Härdeman <david@hardeman.nu>
Link: https://github.com/openwrt/odhcpd/pull/328
Signed-off-by: Álvaro Fernández Rojas <noltari@gmail.com>

dhcpv4: support IPv6-only preferred (RFC8925)

This adds support for RFC8925/IPv6-only preferred to the DHCPv4 server.

Closes: https://github.com/openwrt/odhcpd/pull/235
Signed-off-by: David Härdeman <david@hardeman.nu>
Link: https://github.com/openwrt/odhcpd/pull/327
Signed-off-by: Álvaro Fernández Rojas <noltari@gmail.com>

dhcpv6-ia: fix realloc bug

Commit b9db4d7061a08bf82a25222074065cce71973d0c introduced a bug, the
"hostname" variable used for the realloc would shadow the real hostname
defined at the beginning of the function. Fix this by using a different
variable name.

Fixes: b9db4d7061a0 ("dhcpv6: handle realloc failure")
Signed-off-by: David Härdeman <david@hardeman.nu>
Link: https://github.com/openwrt/odhcpd/pull/326
Signed-off-by: Álvaro Fernández Rojas <noltari@gmail.com>

netlink: make it clearer that we're handling realloc correctly

This doesn't really change the code, it just makes it clearer when
grepping for realloc() that failure is handled correctly.

Signed-off-by: David Härdeman <david@hardeman.nu>
Link: https://github.com/openwrt/odhcpd/pull/320
Signed-off-by: Álvaro Fernández Rojas <noltari@gmail.com>

dhcpv6: handle realloc failure

Another realloc failure case.

Signed-off-by: David Härdeman <david@hardeman.nu>
Link: https://github.com/openwrt/odhcpd/pull/320
Signed-off-by: Álvaro Fernández Rojas <noltari@gmail.com>

config: handle realloc failure in piofolder parsing

Another realloc that goes unchecked.

Signed-off-by: David Härdeman <david@hardeman.nu>
Link: https://github.com/openwrt/odhcpd/pull/320
Signed-off-by: Álvaro Fernández Rojas <noltari@gmail.com>

config: fix realloc() error handling for "dhcpv6_raw" option

Check the return value from realloc() to avoid leaking memory.

Signed-off-by: David Härdeman <david@hardeman.nu>
Link: https://github.com/openwrt/odhcpd/pull/320
Signed-off-by: Álvaro Fernández Rojas <noltari@gmail.com>

config: fix realloc() handling for the "upstream" option

Deal properly with realloc() failure.

Signed-off-by: David Härdeman <david@hardeman.nu>
Link: https://github.com/openwrt/odhcpd/pull/320
Signed-off-by: Álvaro Fernández Rojas <noltari@gmail.com>

config: fix (S)NTP realloc handling

Make sure that realloc errors are dealt with properly in the (S)NTP
option parsing.

Signed-off-by: David Härdeman <david@hardeman.nu>
Link: https://github.com/openwrt/odhcpd/pull/320
Signed-off-by: Álvaro Fernández Rojas <noltari@gmail.com>

dhcpv4: iface->dhcpv4_router -> iface->dhcpv4_routers

This makes it clearer that the variable stores multiple addresses, also
fix a potential realloc memleak.

Signed-off-by: David Härdeman <david@hardeman.nu>
Link: https://github.com/openwrt/odhcpd/pull/320
Signed-off-by: Álvaro Fernández Rojas <noltari@gmail.com>

odhcpd: rename iface->search -> iface->dns_search

And fix some bugs/nits in the process, e.g.:
 - realloc() not handled properly in config.c
 - replace a number of magic values
 - in router.c, the DNSSL options has 8 bytes of superfluous padding

Signed-off-by: David Härdeman <david@hardeman.nu>
Link: https://github.com/openwrt/odhcpd/pull/320
Signed-off-by: Álvaro Fernández Rojas <noltari@gmail.com>

odhcpd: consistent dns naming

Rename iface->dns to iface->dns6_addrs
Rename iface->dhcpv4_dns to iface->dns4_addrs

This makes it clearer what is stored in each.

Signed-off-by: David Härdeman <david@hardeman.nu>
Link: https://github.com/openwrt/odhcpd/pull/320
Signed-off-by: Álvaro Fernández Rojas <noltari@gmail.com>

dhcpv4: remove iface->dhcpv4_bcast

This is also covered by iface->dhcpv4_own_ip

Signed-off-by: David Härdeman <david@hardeman.nu>
Link: https://github.com/openwrt/odhcpd/pull/320
Signed-off-by: Álvaro Fernández Rojas <noltari@gmail.com>

dhcpv4: remove iface->dhcpv4_mask

This is already covered by iface->dhcpv4_own_ip

Signed-off-by: David Härdeman <david@hardeman.nu>
Link: https://github.com/openwrt/odhcpd/pull/320
Signed-off-by: Álvaro Fernández Rojas <noltari@gmail.com>

dhcpv4: rename iface->dhcpv4_local -> iface->dhcpv4_own_ip

"local" isn't very informative, the local what?

In addition, use a struct odhpd_ipaddr instead of a struct in_addr, the
former can already store the prefix_len, broadcast and netmask address,
this will allow the separate iface->dhcpv4_bcast and iface->dhcpv4_mask
members to be removed (see subsequent patches).

Signed-off-by: David Härdeman <david@hardeman.nu>
Link: https://github.com/openwrt/odhcpd/pull/320
Signed-off-by: Álvaro Fernández Rojas <noltari@gmail.com>

dhcpv4: simplify pool determination

This simplifies the logic to determine a dynamic dhcp pool range based
on the prefix length (when "start" and "limit" aren't set).

Signed-off-by: David Härdeman <david@hardeman.nu>
Link: https://github.com/openwrt/odhcpd/pull/320
Signed-off-by: Álvaro Fernández Rojas <noltari@gmail.com>

netlink: add netmask for IPv4

Add a netmask member to struct odhcpd_ipaddr, this is mostly for
convenience.

Signed-off-by: David Härdeman <david@hardeman.nu>
Link: https://github.com/openwrt/odhcpd/pull/320
Signed-off-by: Álvaro Fernández Rojas <noltari@gmail.com>

netlink: variable naming cleanup

Correct some comments, use boolean where a boolean value is expected,
rename more "struct odhcpd_ipaddr" variables to indicate that it is a
struct odhcpd_ipaddr and indicate plural/IPv4/IPv6 in the variable
names.

Signed-off-by: David Härdeman <david@hardeman.nu>
Link: https://github.com/openwrt/odhcpd/pull/320
Signed-off-by: Álvaro Fernández Rojas <noltari@gmail.com>

dhcpv4: fix no_dynamic_dhcp

This fixes the support for running with no_dynamic_dhcp in the DHCPv4
server.

Signed-off-by: David Härdeman <david@hardeman.nu>
Link: https://github.com/openwrt/odhcpd/pull/320
Signed-off-by: Álvaro Fernández Rojas <noltari@gmail.com>

odhcpd: rename iface->addr4 -> iface->oaddrs4

Rename iface->addr4 to iface->oaddrs4, to make it clearer from a cursory
glance that this isn't e.g. "struct in_addr" addresses but a number of
"struct odhcpd_ipaddr" addresses.

At the same time, rename iface->addr4_len to iface->oaddrs4_cnt, to make
it clear that this is a count and not e.g. an allocation length.

Signed-off-by: David Härdeman <david@hardeman.nu>
Link: https://github.com/openwrt/odhcpd/pull/320
Signed-off-by: Álvaro Fernández Rojas <noltari@gmail.com>

odhcpd: rename prefix -> prefix_len

Change a bunch of places where a prefix length is stored as "prefix" to
instead call the length "prefix_len", to make it clearer that it is not
a prefix that is stored.

Signed-off-by: David Härdeman <david@hardeman.nu>
Link: https://github.com/openwrt/odhcpd/pull/320
Signed-off-by: Álvaro Fernández Rojas <noltari@gmail.com>

odhcpd: rename union if_addr -> in46_addr

Anyone reading "union if_addr" might be tempted to think it's got
something to do with an interface addr (e.g. a MAC, or something). In
reality, it's an IPv4/IPv6 addr (i.e. struct in_addr or struct
in6_addr), so rename it to make things clearer.

Signed-off-by: David Härdeman <david@hardeman.nu>
Link: https://github.com/openwrt/odhcpd/pull/320
Signed-off-by: Álvaro Fernández Rojas <noltari@gmail.com>

dhcpv4: improve pool var naming

The uci cfg options might have somewhat confusing names, but
that doesn't mean we have to perpetuate those names throughout the
source.

Also, move the checking of whether the pool parameters from when
we are setting up a pool to where the cfg variables are actually set
from config.

Finally, there's no need to store the pool start and end as "struct
in_addr" in struct interface, because the start and end actually aren't
IPv4 addresses.

Signed-off-by: David Härdeman <david@hardeman.nu>
Link: https://github.com/openwrt/odhcpd/pull/320
Signed-off-by: Álvaro Fernández Rojas <noltari@gmail.com>

ndp: correctness fix for BPF filter

Note that AF_PACKET sockets start receiving packets as soon as they are
created.  Thus, a packet can arrive between the creation of the socket
and the time the real filter is installed. Fix this using the same
technique as used in libpcap, i.e. by installing a drop-all filter,
removing any packets from the socket, then installing the real filter
later (an atomic operation which replaces the drop filter).

Signed-off-by: David Härdeman <david@hardeman.nu>
Link: https://github.com/openwrt/odhcpd/pull/319
Signed-off-by: Álvaro Fernández Rojas <noltari@gmail.com>

dhcpv4: add BPF to dhcpv4_setup_interface()

This means the kernel won't even pass packets which we are not
interested in.

Signed-off-by: David Härdeman <david@hardeman.nu>
Link: https://github.com/openwrt/odhcpd/pull/318
Signed-off-by: Álvaro Fernández Rojas <noltari@gmail.com>

dhcpv4: define val where it is used in dhcpv4_setup_interface()

This makes it a tiny bit easier to see immediately what "val" is.

Signed-off-by: David Härdeman <david@hardeman.nu>
Link: https://github.com/openwrt/odhcpd/pull/318
Signed-off-by: Álvaro Fernández Rojas <noltari@gmail.com>

dhcpv4: simplify error handling in dhcpv4_setup_interface()

Remove the "ret" variable which can anyway always just be -1 on error,
and rename the "out" label to "error" to clarify that it's only used on
error.

Signed-off-by: David Härdeman <david@hardeman.nu>
Link: https://github.com/openwrt/odhcpd/pull/318
Signed-off-by: Álvaro Fernández Rojas <noltari@gmail.com>

dhcvp4: use tmp fd in dhcpv4_setup_interface()

This makes the code a little bit easier to read.

Signed-off-by: David Härdeman <david@hardeman.nu>
Link: https://github.com/openwrt/odhcpd/pull/318
Signed-off-by: Álvaro Fernández Rojas <noltari@gmail.com>

dhcpv4: replace ToS precedence

The IPTOS_PREC_* values are deprecated since RFC2474 §4.2.2.1 (Dec 1998)
replaced them with class selector codepoints.

Signed-off-by: David Härdeman <david@hardeman.nu>
Link: https://github.com/openwrt/odhcpd/pull/318
Signed-off-by: Álvaro Fernández Rojas <noltari@gmail.com>

statefiles: fix off-by-one-bug

The string with the final hostsfile name is "%s.%s", so we need two
extra bytes for the separating "." and for the trailing null byte.

Signed-off-by: David Härdeman <david@hardeman.nu>
Link: https://github.com/openwrt/odhcpd/pull/322
Signed-off-by: Álvaro Fernández Rojas <noltari@gmail.com>

all: implement RFC8910 captive portal (CP) option for DHCPv4

https://www.rfc-editor.org/rfc/rfc8910.html

Signed-off-by: Paul Donald <newtwen+github@gmail.com>
Link: https://github.com/openwrt/odhcpd/pull/315
Signed-off-by: Álvaro Fernández Rojas <noltari@gmail.com>

all: implement RFC8910 captive portal (CP) option

https://www.rfc-editor.org/rfc/rfc8910.html

RFC8910 defines a captive portal API URI for a CP client to consume.

With captive_portal_uri set to 'https://test.example.com'

Produces in RA:

ICMPv6 Option (DHCP Captive-Portal)
    Type: DHCP Captive-Portal (37)
    Length: 4 (32 bytes)
    Captive Portal: https://test.example.com

And in DHCPv6 Reply:

Captive Portal
    Option: Captive Portal (103)
    Length: 24
    Captive Portal: https://test.example.com

Signed-off-by: Paul Donald <newtwen+github@gmail.com>
Link: https://github.com/openwrt/odhcpd/pull/315
Signed-off-by: Álvaro Fernández Rojas <noltari@gmail.com>

github: ci: add MIPS64, PowerPC64 and RISCV64

MIPS64, PowerPC64 and RISCV64 are popular OpenWrt archs.
Refactor the sizes build step to generate the table programatically.

Signed-off-by: Álvaro Fernández Rojas <noltari@gmail.com>

router: Modify relayed RA PIO L and RA M/O flags according to interface policy

Effectively:
-clear the L flag
-apply local M and O flags if we are not DHCPv6 relay mode

h/t user @Shine-

Signed-off-by: Paul Donald <newtwen+github@gmail.com>
Link: https://github.com/openwrt/odhcpd/pull/300
Signed-off-by: Álvaro Fernández Rojas <noltari@gmail.com>

dhcpv6: prevent network loop scenario

It's possible for odhcpd to talk with its (odhcp6c) self in a network loop
scenario with strange results. Prevent this by returning early.

Signed-off-by: Paul Donald <newtwen+github@gmail.com>
Link: https://github.com/openwrt/odhcpd/pull/314
Signed-off-by: Álvaro Fernández Rojas <noltari@gmail.com>

dhcpv4: simplify random address selection

Simplify the generation of random addresses, and make the code easier
to follow, by using the same approach as in dhcpv6.

Signed-off-by: David Härdeman <david@hardeman.nu>
Link: https://github.com/openwrt/odhcpd/pull/313
Signed-off-by: Álvaro Fernández Rojas <noltari@gmail.com>

odhcpd: update devel-build.sh

Now that ubus can be disabled at runtime, we can default to building
it in, making it easier to catch compile-time errors in ubus.c.

Signed-off-by: David Härdeman <david@hardeman.nu>
Link: https://github.com/openwrt/odhcpd/pull/312
Signed-off-by: Álvaro Fernández Rojas <noltari@gmail.com>

odhcpd: make ubus optional at runtime

This is mostly for developer convenience, allowing ubus to be built in, but
disabled at will when launching odhcpd.

Signed-off-by: David Härdeman <david@hardeman.nu>
Link: https://github.com/openwrt/odhcpd/pull/312
Signed-off-by: Álvaro Fernández Rojas <noltari@gmail.com>

odhcpd: simplify signal handling

Currently, odhcpd's main() function will:
 - call uloop_init()
 - set up some signal handlers for SIGUSR1/SIGINT/SIGTERM
   (SIGUSR1 is ignored, the other handlers call uloop_end())
 - call odhcpd_run(), which will:
 - set up some signal handlers for SIGTERM/SIGINT/SIGHUP
   (the first two call uloop_end(), SIGHUP calls odhcpd_reload())

Note that uloop_init() will call uloop_setup_signals() which will call
uloop_setup_signals() which will install handlers for SIGINT/SIGTERM
which will set uloop_cancelled to true (e.g. the same thing that
uloop_end() does).

In other words, the signal handlers are modified three times, and since
the default "exit" action is to call uloop_end(), there's nothing that
will actually react to signals until uloop_run() is called, meaning that
the startup will be uninterruptible if e.g. ubus_init() fails since
it'll just retry in a loop with no handling of signals.

So, simplify and fix this by letting uloop handle the signals and
checking if uloop has been told to exit, removing a bunch of
special-purpose code.

Signed-off-by: David Härdeman <david@hardeman.nu>
Link: https://github.com/openwrt/odhcpd/pull/312
Signed-off-by: Álvaro Fernández Rojas <noltari@gmail.com>

src: improve attributes

- Properly guard odhcp6c.h attributes with `#ifndef` to avoid redefining
them when including external headers.
- Also convert the remaining __attribute__ usages to the custom
declarations.
- Consolidate custom declarations with _o_ prefix.

Signed-off-by: Álvaro Fernández Rojas <noltari@gmail.com>
Link: https://github.com/openwrt/odhcpd/pull/307

src: consolidate and improve fallthrough

- Enable fallthrough warnings.
- Use fallthrough attribute instead of comments.
- Drop unneeded fallthroughs (no code between two case statements).

Signed-off-by: Álvaro Fernández Rojas <noltari@gmail.com>
Link: https://github.com/openwrt/odhcpd/pull/307

dhcpv6-ia: add missing limits header

Fix the following build error by including <limits.h>:
src/dhcpv6-ia.c: In function 'handle_addrlist_change':
src/dhcpv6-ia.c:495:53: error: 'INT_MAX' undeclared (first use in this function)
  495 |                                         a->fr_cnt = INT_MAX;
      |                                                     ^~~~~~~
src/dhcpv6-ia.c:33:1: note: 'INT_MAX' is defined in header '<limits.h>'; did you forget to '#include <limits.h>'?
   32 | #include <libubox/md5.h>
  +++ |+#include <limits.h>
   33 |
src/dhcpv6-ia.c:495:53: note: each undeclared identifier is reported only once for each function it appears in
  495 |                                         a->fr_cnt = INT_MAX;
      |                                                     ^~~~~~~

Fixes: 7136fbe390a5 ("dhcpv6-ia: split statefile handling to separate file")
Signed-off-by: Álvaro Fernández Rojas <noltari@gmail.com>
Link: https://github.com/openwrt/odhcpd/pull/311

config: fix memleak during odhcpd reload

- The memset in close_interface reset the pios pointer before it
  could be freed, causing a memory leak. Relocate the free call
  to clean_interface to ensure proper deallocation.

- Use realloc instead of malloc in config_load_ra_pio()
  This function may be called multiple times during odhcpd reload,
  and using malloc without freeing the previous allocation was
  causing memory leaks.

Signed-off-by: Fei Lv <feilv@asrmicro.com>
Link: https://github.com/openwrt/odhcpd/pull/309
Signed-off-by: Álvaro Fernández Rojas <noltari@gmail.com>

dhcpv4: update ubus DHCPv4 events/methods

This is based on the assumption that we don't really have any consumers
of ubus DHCPv4 events (yet).

With that in mind, rename and simplify the event function (yes, we
should add a sibling function for DHCPv6 leases later).

Also, take the chance to improve some naming, and introduce additional
attributes to the ubus event/method (the interface name and the
DUID/IAID, if known).

Signed-off-by: David Härdeman <david@hardeman.nu>
Link: https://github.com/openwrt/odhcpd/pull/306
Signed-off-by: Álvaro Fernández Rojas <noltari@gmail.com>

odhcpd: rename dhcpv6_lease->clid[_data|_len]

The option is called clientid, but the option carries a DUID. So what
we're storing is a DUID, not a CLID.

Signed-off-by: David Härdeman <david@hardeman.nu>
Link: https://github.com/openwrt/odhcpd/pull/306
Signed-off-by: Álvaro Fernández Rojas <noltari@gmail.com>

odhcpd: rename [lease_cfg|dhcpv4_lease]->ipaddr to ipv4

"ipaddr" doesn't say if it's IPv4 or IPv6, and we want to be clear in a
dual-stack codebase.

And at the same time, use a struct in_addr to store the IPv4 address,
which is just less confusing.

Finally, remove all uses of "inet_ntoa()", which is deprecated.

Signed-off-by: David Härdeman <david@hardeman.nu>
Link: https://github.com/openwrt/odhcpd/pull/306
Signed-off-by: Álvaro Fernández Rojas <noltari@gmail.com>

dhcpv4: use an AVL to store leases

An AVL has O(log n) complexity vs O(n) for a linked list in all the
operations we care about (insertion and search).

Also, it makes the code simpler by not having to care about details like
how to sort the leases.

Signed-off-by: David Härdeman <david@hardeman.nu>
Link: https://github.com/openwrt/odhcpd/pull/306
Signed-off-by: Álvaro Fernández Rojas <noltari@gmail.com>

github: ci: add powerpc arch

PowerPC is another popular OpenWrt arch.

Signed-off-by: Álvaro Fernández Rojas <noltari@gmail.com>

github: ci: add cmake build and source directories

Add cmake build and source directories to suppress the following warning:
CMake Warning:
  No source or binary directory provided. Both will be assumed to be the
  same as the current working directory, but note that this warning will
  become a fatal error in future CMake releases.

Signed-off-by: Álvaro Fernández Rojas <noltari@gmail.com>

github: ci: disable json-c tests

Disable BUILD_TESTING to save time when building json-c.

Signed-off-by: Álvaro Fernández Rojas <noltari@gmail.com>

scripts: devel-build: disable json-c tests

Disable BUILD_TESTING to save time when building json-c.

Signed-off-by: Álvaro Fernández Rojas <noltari@gmail.com>

netlink: fix typo in debug msg

As noted by @CasperVector in #115, there's a typo in netlink.c,
checking if true is true isn't very useful :)

Signed-off-by: David Härdeman <david@hardeman.nu>
Link: https://github.com/openwrt/odhcpd/pull/305
Signed-off-by: Álvaro Fernández Rojas <noltari@gmail.com>

src: fix whitespace issues

- Remove double whitespaces.
- Convert whitespace alignments to tabs.
- Fix code comments using whitespaces.

Signed-off-by: Álvaro Fernández Rojas <noltari@gmail.com>

vscode: enable indentation detection

Fixes issues with VSCode forcing whitespaces instead of tabs for indentation.

Fixes: df1824aec66c ("vscode: add tab settings")
Signed-off-by: Álvaro Fernández Rojas <noltari@gmail.com>

odhcpd: add support for "ignore"

This is inspired by @Kasoo's work in PR #234, but slightly different.

The "ip" option now takes "ignore" as a value , but it'll only disable
DHCPv4 for a matching client.

Similarly, the "hostid" option now also understands "ignore", which will
disable DHCPv6 for a matching client.

Of course, this is all based on client-reported MAC
addresses/DUIDs/client IDs/etc, so it's not actually a security feature.

Closes: https://github.com/openwrt/odhcpd/pull/234
Closes: https://github.com/openwrt/odhcpd/issues/198
Signed-off-by: David Härdeman <david@hardeman.nu>
Link: https://github.com/openwrt/odhcpd/pull/303
Signed-off-by: Álvaro Fernández Rojas <noltari@gmail.com>