Send A in response to service discovery

This allows other mdns software to discover our hostname when they
start.

Signed-off-by: Rafał Miłecki <rafal@milecki.pl>

Use A(AAA) when querying for domain on cache expire

When cache entry is (halfly) expired we send query to get an update. So
far we were sending all queries using PTR records but for cached domains
it should be A(AAA) instead.

Signed-off-by: Rafał Miłecki <rafal@milecki.pl>
Acked-by: John Crispin <john@phrozen.org>

Don't try sending unicast responses with multicast address

So far we were doing that when destination IP address wasn't specified.
Now we have all places in code cleaned up, stop accepting this and print
an error if needed.

Signed-off-by: Rafał Miłecki <rafal@milecki.pl>

Use unicast IP address when sending unicast service reply

For each protocol (IPv4 and IPv6) we have two interfaces (sockets): one
for unicast and one for multicast. If we noticed CLASS_UNICAST in the
multicast query we were switching to unicast interface for sending
reply.

The problem was not passing destination IP address. It was resulting in
sending packet to multicast IP using unicast interface. As we don't
set IP_MULTICAST_TTL / IPV6_MULTICAST_HOPS for unicast ones TTL was 1
and packets were ignored.

Signed-off-by: Rafał Miłecki <rafal@milecki.pl>

Prepare support for sending unicast DNS answers

This adds extra argument to dns_send_answer & interface_send_packet
functions. For now we pass NULL-s only.

Signed-off-by: Rafał Miłecki <rafal@milecki.pl>

Print error if parsing service JSON fails

Signed-off-by: Rafał Miłecki <rafal@milecki.pl>

Set TTL to 255 for outgoing IPv4 unicast packets

1) We SHOULD do this according to the RFC 6762 (Section 11, "Source
   Address Check").
2) We already do it for IPv6 unicast and both multicast interfaces
3) When receiving IPv4 packet we ignore it if TTL doesn't equal 255

In other words this is a requirement for working unicast communication
between 2 umdns instances (which doesn't seem to work yet though).

Signed-off-by: Rafał Miłecki <rafal@milecki.pl>
Acked-by: John Crispin <john@phrozen.org>

Fix sending replies to PTR questions

When we receive PTR question it includes hostname (instance), e.g.:
mdnsd: parse_question (391): Q -> PTR lede._http._tcp.local

First of all we should check if it matches hostname we use before trying
to reply. Secondly service_reply expects service with domain appended
(without hostname/instance) so we need to strip received string out of
hostname before passing it.

Signed-off-by: Rafał Miłecki <rafal@milecki.pl>
Acked-by: John Crispin <john@phrozen.org>

Add helper for checking if cached entry is for a host

Signed-off-by: Rafał Miłecki <rafal@milecki.pl>

Drop declaration of non-existing cache_lookup_name function

Signed-off-by: Rafał Miłecki <rafal@milecki.pl>

Make sdudp string const

It musn't be changed.

Signed-off-by: Rafał Miłecki <rafal@milecki.pl>

Rename cache scanning to updating

What we were do is querying over all interfaces for all cached entries.
This isn't real scanning but rather updating (the cache).

Signed-off-by: Rafał Miłecki <rafal@milecki.pl>
Acked-by: John Crispin <john@phrozen.org>

Set TTL to the new value when updating cached service entry

In the simplest case host may change TTL to some different value. It's
the most important for the goodbye packets though.

Consider situation when host (for which we have no no cached entries)
restarts. First it sends goodbye with TTL 0 (we cache it), then seconds
later it announces with standard TTL values (but we keep TTL 0).
It results in wrong cache state and could cause dropping valid entries.

Signed-off-by: Rafał Miłecki <rafal@milecki.pl>
Acked-by: John Crispin <john@phrozen.org>

Load service JSONs from /etc/umdns/

It matches recent project rename & uses directory that sholud be
persistent. Using /tmp/ could result in user having to setup JSONs after
every reboot.

Signed-off-by: Rafał Miłecki <rafal@milecki.pl>
Acked-by: John Crispin <john@phrozen.org>

Rename project to umdns

Using mdns name is a bit confusing as it's really close to Apple's mdnsd
binary name from their mDNSResponder project.
To make things worse OpenWrt's project was hosted in the mdnsd.git repo.

Rename this project to umdns which follows other projects naming schema
and is unique enough.

Signed-off-by: Rafał Miłecki <rafal@milecki.pl>
Acked-by: John Crispin <john@phrozen.org>

Drop one-liner service_announce function

Now we have service_announce simplified so much there is no reason to
keep it as a helper.

Signed-off-by: Rafał Miłecki <rafal@milecki.pl>
Acked-by: John Crispin <john@phrozen.org>

Don't use _services._dns-sd._tcp.local

It seems mdns(d) was trying to support queries for two records with
following names:
1) _services._dns-sd._udp.local
2) _services._dns-sd._tcp.local

According to the RFC 6763 Section 9 only the first one should be used
and response PTR records should include services of both protocols: UDP
and TCP.

This fixes discovering TCP services as in practice no other software was
sending queries with _services._dns-sd._tcp.local.

Signed-off-by: Rafał Miłecki <rafal@milecki.pl>
Acked-by: John Crispin <john@phrozen.org>

Simplify UDP vs. TCP handling in service_announce_services

Out of 3 calls of this function only one doesn't have TCP vs. UDP
hardcoded. It's easier to move string check to that place and make this
function take "int tcp" argument instead.

Signed-off-by: Rafał Miłecki <rafal@milecki.pl>

Check correct attributes when loading service from blob

We access SERVICE_TXT so we should check for this entry. Checking
SERVICE_SERVICE doesn't make sense anyway as it's verified few lines
above (in the same funcion).

Signed-off-by: Rafał Miłecki <rafal@milecki.pl>
Acked-by: John Crispin <john@phrozen.org>

Avoid duplicated iteration over services when announcing

In function service_announce_services we iterate over services and
handle every single one so calling service_reply which also iterates
doesn't make sense.

In simple cases it was just wasting CPU cycles. We got service, we were
passing its name and we were looking for it again comparing names.
In cases with one service available on few different ports it was worse.
For every iterated service we were announcing that one and all other
sharing the same name. It resulted in sending n^2 records.

Signed-off-by: Rafał Miłecki <rafal@milecki.pl>
Acked-by: John Crispin <john@phrozen.org>

Move matching logic out of service_reply_single function

There is only 1 place in project calling this function with the match
argument. It's easier to add prooper condition in this place and have
service_reply_single simplified by taking 1 less argument.

Signed-off-by: Rafał Miłecki <rafal@milecki.pl>

Send A(AAA) records when announcing

This allows device to be discovered by its local hostname.

Signed-off-by: Rafał Miłecki <rafal@milecki.pl>
Acked-by: John Crispin <john@phrozen.org>

Fix/change comment format about port 5353

It wasn't ANSI C comment and there were 2 unneeded spaces.

Signed-off-by: Rafał Miłecki <rafal@milecki.pl>

Don't call dns_reply_a from service_reply

This simplifies code without changing any behavior. Having this call in
service_reply required checking two conditions and was making code flow
harder to follow.

There are 2 more service_reply calls in the project:
1) In service_announce_services where we iterate over &services and
   every element of this list has "service" field filled. It means match
   argument was never NULL and dns_reply_a was never called from there.
2) In parse_question which also receives some name (there is a proper
   check in the dns_handle_packet). No call there neither.

So after all there was only 1 place that was indeed calling dns_reply_a.

Signed-off-by: Rafał Miłecki <rafal@milecki.pl>
Acked-by: John Crispin <john@phrozen.org>

Rename service_reply_a to dns_reply_a and move it to proper file

This function doesn't really do anything service specify, it just sends
an A(AAA) records. It could probably be used even without any services
registered.

Signed-off-by: Rafał Miłecki <rafal@milecki.pl>
Acked-by: John Crispin <john@phrozen.org>

Add simple "Fall through" comment to the announce_timer function switch

It's a common practice to add such comments to make it clear break
instruction was skipped on purpose.

Signed-off-by: Rafał Miłecki <rafal@milecki.pl>
Acked-by: John Crispin <john@phrozen.org>

Change dns_send_question function arg from unicast to multicast

This trivial patch just reverses argument logic to make it a bit more
consistent with struct interface which contains "multicast" field. This
hopefully will make typos less likely and code easier to follow.

Signed-off-by: Rafał Miłecki <rafal@milecki.pl>
Acked-by: John Crispin <john@phrozen.org>

do not assume that dns structs are aligned

They do not have a fixed alignment in the packet. ARMv5 has separate
instructions for unaligned and aligned 16-bit load/store. The aligned
ones have undefined behavior if the addresses are unaligned, and that
completely breaks packet parsing.

Signed-off-by: Johannes Wegener <mail@johanneswegener.de>
Signed-off-by: Felix Fietkau <nbd@nbd.name>

mdnsd: the service is ignoring questions

When a question is incoming the service was incorrectly resettig the
query timeout regardless of it being the right service. This causes
the code to never answer the questions it is being asked.

Reported-by: Cristian Morales Vega <cristian@samknows.com>
Signed-off-by: John Crispin <john@phrozen.org>

cmake: Search for libjson-c

Since we use libblobmsg_json which uses libjson-c internally, make sure
we find this library.

Signed-off-by: Florian Fainelli <f.fainelli@gmail.com>

Fix IPv6 read

This is a quick fix. It may still not work when the interface has
multiple IPs, and this also applies to IPv4.

Signed-off-by: Cristian Morales Vega <cristian@samknows.com>

Revert "mdnsd: interface: enable looped back messages"

This reverts commit 8a70b2b47dd328f8180e5ecaa7bdc817f574a81b.

mdnsd: interface: enable looped back messages

When the IP_MULTICAST_LOOP/IPV6_MULTICAST_LOOP socket options are not enabled,
locally generated queries are ignored by mdnsd; This prevents local
applications from being able to discover locally published services.

Signed-off-by: Eyal Birger <eyal.birger@gmail.com>

interface: include libubox/utils.h

Signed-off-by: Felix Fietkau <nbd@openwrt.org>

service: remove obsolete uci include statements

Signed-off-by: Felix Fietkau <nbd@openwrt.org>

ubus: add keep functionality to set_config-call

Signed-off-by: Steven Barth <steven@midlink.org>

cache: fix cache record refreshing

Signed-off-by: Steven Barth <steven@midlink.org>

ubus: add ubus fetch call

Signed-off-by: Steven Barth <steven@midlink.org>

dns: use different buffers for encoding and decoding names

dns_send_question was ocassionally called with a name from name_buf

Signed-off-by: Steven Barth <steven@midlink.org>

cache: also cache PTR records

Signed-off-by: Steven Barth <steven@midlink.org>

cache: cache rdata of SRV-records as well

Signed-off-by: Steven Barth <steven@midlink.org>

Use monotonic clock for timeouts

Signed-off-by: Steven Barth <steven@midlink.org>

Add query ubus call

Signed-off-by: Steven Barth <steven@midlink.org>

add a basic cache refresh logic

Signed-off-by: John Crispin <blogic@openwrt.org>

add a iface pointer to services and records

Signed-off-by: John Crispin <blogic@openwrt.org>

entry->service

Signed-off-by: John Crispin <blogic@openwrt.org>

print ttl when receiving a PTR record

Signed-off-by: John Crispin <blogic@openwrt.org>

basic support for the cache flush bit

Signed-off-by: John Crispin <blogic@openwrt.org>

add src ip validation

Signed-off-by: John Crispin <blogic@openwrt.org>

silently drop unicast questions that dont originate from port 5353

Signed-off-by: John Crispin <blogic@openwrt.org>

propagate sockaddr and port form the receive function to the parser

Signed-off-by: John Crispin <blogic@openwrt.org>

fix free after use on shutdown

Signed-off-by: John Crispin <blogic@openwrt.org>

add good bye messages

Signed-off-by: John Crispin <blogic@openwrt.org>

add SO_BINDTODEVICE support

Signed-off-by: John Crispin <blogic@openwrt.org>

saner debug output

Signed-off-by: John Crispin <blogic@openwrt.org>

ipv6 cleanup

Signed-off-by: John Crispin <blogic@openwrt.org>

propagate unicast bit

Signed-off-by: John Crispin <blogic@openwrt.org>

nicer debug output on rx

Signed-off-by: John Crispin <blogic@openwrt.org>

add unicast listener sockets

Signed-off-by: John Crispin <blogic@openwrt.org>

interface.c cleanup

Signed-off-by: John Crispin <blogic@openwrt.org>

remove debug code

Signed-off-by: John Crispin <blogic@openwrt.org>

validate incoming interface

Signed-off-by: John Crispin <blogic@openwrt.org>

sane error message when read fails

Signed-off-by: John Crispin <blogic@openwrt.org>

add pktinfo to rx path

Signed-off-by: John Crispin <blogic@openwrt.org>

add -4 and -6 options

Signed-off-by: John Crispin <blogic@openwrt.org>

load service info from procd

Signed-off-by: John Crispin <blogic@openwrt.org>

split the service_load() function into 2. this allows us to reuse the parsing
code when reading services via ubus

Signed-off-by: John Crispin <blogic@openwrt.org>

change the service json format this allows one instance to register multiple services of the same kind but with different ports

Signed-off-by: John Crispin <blogic@openwrt.org>

send unsolicitated answers when a new service appears

Signed-off-by: John Crispin <blogic@openwrt.org>

remove some debug code

Signed-off-by: John Crispin <blogic@openwrt.org>

dont delete records but set ttl to 1

-> 10.2.  Announcements to Flush Outdated Cache Entries

Signed-off-by: John Crispin <blogic@openwrt.org>

add the rfc

Signed-off-by: John Crispin <blogic@openwrt.org>

add goodbye messages when service disappear

Signed-off-by: John Crispin <blogic@openwrt.org>

more ttl related changes. prereq for goodbye messages

Signed-off-by: John Crispin <blogic@openwrt.org>

dns_add_answer() now has a ttl parameter

Signed-off-by: John Crispin <blogic@openwrt.org>

properly announce all ips of an interface when asked

Signed-off-by: John Crispin <blogic@openwrt.org>

add ipv6 support

Signed-off-by: John Crispin <blogic@openwrt.org>

fix potential null pointer deref when packets are forged

Signed-off-by: John Crispin <blogic@openwrt.org>

fix potential buffer overflow when txt records are forged

Signed-off-by: John Crispin <blogic@openwrt.org>

abort the reply parsing if an error is encountered

Signed-off-by: John Crispin <blogic@openwrt.org>

gracefully exit if the cmdline is bogus

Signed-off-by: John Crispin <blogic@openwrt.org>

remove now unused service-types index

Signed-off-by: John Crispin <blogic@openwrt.org>

fix entry duplication bug

Signed-off-by: John Crispin <blogic@openwrt.org>

fix off by one in record find code

Signed-off-by: John Crispin <blogic@openwrt.org>

fix use after free

Signed-off-by: John Crispin <blogic@openwrt.org>

remove linux specific includes

Signed-off-by: Felix Fietkau <nbd@openwrt.org>

mdns: remove dependencies on asm/byteorder.h, fix constant endian swap initialization

Signed-off-by: Felix Fietkau <nbd@openwrt.org>

service.c: remove unused service_attr_list

Signed-off-by: Felix Fietkau <nbd@openwrt.org>

rename set_interfaces to set_config, more stuff can be added here

Signed-off-by: Felix Fietkau <nbd@openwrt.org>

allow starting without an active interface

Signed-off-by: Felix Fietkau <nbd@openwrt.org>

add support for dynamic reconfiguration of interfaces

Signed-off-by: Felix Fietkau <nbd@openwrt.org>

fix segfault while dumping records

Signed-off-by: Felix Fietkau <nbd@openwrt.org>

simplify memory management for dns answers: use a blob_buf to cache entries

Signed-off-by: Felix Fietkau <nbd@openwrt.org>

use lookup array for dns_type_string()

Signed-off-by: Felix Fietkau <nbd@openwrt.org>

reuse mdns_buf for cache_answer()

Signed-off-by: Felix Fietkau <nbd@openwrt.org>

move signal_setup() to main.c

Signed-off-by: Felix Fietkau <nbd@openwrt.org>

make iface_name static

Signed-off-by: Felix Fietkau <nbd@openwrt.org>

clean up hostname handling, make service_name() static

Signed-off-by: Felix Fietkau <nbd@openwrt.org>

fix excessive stack usage

Signed-off-by: Felix Fietkau <nbd@openwrt.org>

fix a few format string bugs

Signed-off-by: Felix Fietkau <nbd@openwrt.org>