A DNSSEC validation error was introduced in the fix for CVE-2017-15107
Backport the upstream fix to the fix (a simple typo)
Signed-off-by: Kevin Darbyshire-Bryant <ldir@darbyshire-bryant.me.uk>
CVE-2017-15107
An interesting problem has turned up in DNSSEC validation. It turns out
that NSEC records expanded from wildcards are allowed, so a domain can
include an NSEC record for *.example.org and an actual query reply could
expand that to anything in example.org and still have it signed by the
signature for the wildcard. So, for example
!.example.org NSEC zz.example.org
is fine.
The problem is that most implementers (your author included, but also
the Google public DNS people, powerdns and Unbound) then took that
record to prove the nothing exists between !.example.org and
zz.example.org, whereas in fact it only provides that proof between
*.example.org and zz.example.org.
This gives an attacker a way to prove that anything between
!.example.org and *.example.org doesn't exists, when it may well do so.
Signed-off-by: Kevin Darbyshire-Bryant <ldir@darbyshire-bryant.me.uk>
Dnsmasq used SIGHUP to do too many things: 1) set dnssec time validation
enabled, 2) bump SOA zone serial, 3) clear dns cache, 4) reload hosts
files, 5) reload resolvers/servers files.
Many subsystems within LEDE can send SIGHUP to dnsmasq: 1) ntpd hotplug
(to indicate time is valid for dnssec) 2) odhcpd (to indicate a
new/removed host - typically DHCPv6 leases) 3) procd on interface state
changes 4) procd on system config state changes, 5) service reload.
If dnssec time validation is enabled before the system clock has been
set to a sensible time, name resolution will fail. Because name
resolution fails, ntpd is unable to resolve time server names to
addresses, so is unable to set time. Classic chicken/egg.
Since commits 23bba9cb33 (service reload) &
4f02285d8b (system config) make it more
likely a SIGHUP will be sent for events other than 'ntpd has set time'
it is more likely that an errant 'name resolution is failing for
everything' situation will be encountered.
Fortunately the upstream dnsmasq people agree and have moved 'check
dnssec timestamp enable' from SIGHUP handler to SIGINT.
Backport the upstream patch to use SIGINT.
ntpd hotplug script updated to use SIGINT.
Signed-off-by: Kevin Darbyshire-Bryant <ldir@darbyshire-bryant.me.uk>
Send a SIGHUP signal via procd to the dnsmasq service so the instance(s)
re-read(s) the /tmp/hosts/dhcp config.
Signed-off-by: Florian Eckert <fe@dev.tdt.de>
Add config option which allows to enable/disable DHCP support at compile
time. Make DHCPv6 support dependant on DHCP support as DHCPv6 support
implies having DHCP support.
Signed-off-by: Hans Dedecker <dedeckeh@gmail.com>
Different invocations of the dnsmasq init script (e.g. at startup by procd)
will rewrite the dhcp host file which might result into dnsmasq reading an
empty dhcp host file as it is being rewritten by the dnsmasq init script.
Let the dnsmasq init script first write to a temp dhcp host file so it does
not overwrite the contents of the existing dhcp host file.
Reported-by: Hartmut Birr <e9hack@gmail.com>
Signed-off-by: Hans Dedecker <dedeckeh@gmail.com>
If all configured dns servers return refused in response to a query in
strict mode; dnsmasq will end up in an infinite loop retransmitting the
dns query resulting into high CPU load.
Problem is fixed by checking for the end of a dns server list iteration
in strict mode.
Signed-off-by: Hans Dedecker <dedeckeh@gmail.com>
Without this change, the instance-specific conf-file is being added to procd_add_jail_mount,
but not used by dnsmasq.
Signed-off-by: Emerson Pinter <dev@pinter.com.br>
CPE ids helps to tracks CVE in packages.
https://cpe.mitre.org/specification/
Thanks to swalker for CPE to package mapping and
keep tracking CVEs.
Acked-by: Jo-Philipp Wich <jo@mein.io>
Signed-off-by: Alexander Couzens <lynxis@fe80.eu>
This patch adds a parser for the uci representation of
dnsmasq's "-a | --listen-address" option.
In summary, this option forces dnsmasq to listen on the
given IP address(es). Both interface and listen-address
options may be given, in which case the set of both
interfaces and addresses is used.
Note that if no interface option is given, but listen_address is,
dnsmasq will not automatically listen on the loopback interface.
To achieve this, the loopback IP addresses, 127.0.0.1 and/or ::1
must be explicitly added.
This option is useful for ujailed dnsmasq instances, that would
otherwise fail to work properly, because listening to the
"This host on this network" address (aka 0.0.0.0 see rfc1700 page 4)
may not be allowed.
Signed-off-by: Christian Lamparter <chunkeey@gmail.com>
Signed-off-by: Hans Dedecker <dedeckeh@gmail.com> (PKG_RELEASE increase)
Init script won't append --no-dhcp-interface option if interface
protocol is one of: ncm, directip, qmi, mbim.
This is caused by IP address assigned to dynamically created netifd
interfaces. As a result there's no netmask assigned to the main
interface and dhcp_add() function returns prematurely.
By moving network subnet check we can ensure that --no-dhcp-interface is
properly generated for wwan interfaces.
Signed-off-by: Marcin Jurkowski <marcin1j@gmail.com>
Signed-off-by: Hans Dedecker <dedeckeh@gmail.com> [PKG_RELEASE increase; move network checks]
With the introduction of the ubus notifications, we would now fail building
dnsmasq with external toolchains that don't automatically search for headers.
Pass TARGET_CPPFLAGS to the Makefile to resolve that.
Fixes: 34a206bc11 ("dnsmasq: add ubus notifications for new leases")
Signed-off-by: Florian Fainelli <f.fainelli@gmail.com>
Don't return arcount=1 if EDNS0 RR won't fit in the packet.
Omitting the EDNS0 RR but setting arcount gives a malformed packet.
Also, don't accept UDP packet size less than 512 in received EDNS0.
Signed-off-by: Kevin Darbyshire-Bryant <ldir@darbyshire-bryant.me.uk>
Remove LEDE partial fix for CVE-2017-13704.
Backport official fix from upstream.
Signed-off-by: Kevin Darbyshire-Bryant <ldir@darbyshire-bryant.me.uk>
Signed-off-by: Hans Dedecker <dedeckeh@gmail.com> (PKG_RELEASE increase)
Fix SIGSEGV in rfc1035.c answer_request() line 1228 where memset()
is called with header & limit pointing at the same address and thus
tries to clear memory from before the buffer begins.
answer_request() is called with an invalid edns packet size provided by
the client. Ensure the udp_size provided by the client is bounded by
512 and configured maximum as per RFC 6891 6.2.3 "Values lower than 512
MUST be treated as equal to 512"
The client that exposed the problem provided a payload udp size of 0.
Signed-off-by: Kevin Darbyshire-Bryant <kevin@darbyshire-bryant.me.uk>
Acked-by: Hans Dedecker <dedeckeh@gmail.com>
Currently, dnsmasq support assigning multiple tags to a host record
(--dhcp-host), but we only support only 1 tag for a host. The commit
makes the following config to be valid:
config host
option name 'computer'
option mac '00:11:22:33:44:55'
option ip '192.168.1.100'
list tag 'vendor_class'
list tag 'vendor_id'
config tag 'vendor_class'
list dhcp_option 'option:vendor-class,00:...<omitted>'
config tag 'vendor_id'
option force '1'
list dhcp_option 'option:vendor-id-encap,00:...<omitted>'
Signed-off-by: Kuang Rufan <kuangrufan@pset.suntec.net>
Signed-off-by: Hans Dedecker <dedeckeh@gmail.com>
Remove ping check in DHCPDISCOVER case as too many buggy clients leave
an interface in configured state causing the ping check to fail.
Signed-off-by: Hans Dedecker <dedeckeh@gmail.com>
Commit 5cd88f4 "dnsmasq: remove use of uci state for getting network ifname"
broke the ability to specify unmanaged network device names for inclusion
and exclusion in the uci configuration.
Restore support for raw device names by falling back to the input value
when "network_get_device" yields no result.
Fixes FS#876.
Signed-off-by: Jo-Philipp Wich <jo@mein.io>
This is functionally the same as --server, but provides some syntactic sugar to
make specifying address-to-name queries easier.
For example --rev-server=1.2.3.0/24,192.168.0.1 is exactly equivalent to
--server=/3.2.1.in-addr.arpa/192.168.0.1
Signed-off-by: DUPONCHEEL Sébastien <sebastien.duponcheel@corp.ovh.com>
Don't start ping-check of address in DHCP discover if there already
exists a lease for the address. It has been reported under some
circumstances android and netbooted windows devices can reply to
ICMP pings if they have a lease and thus block the allocation of
the IP address the device already has during boot.
Signed-off-by: Hans Dedecker <dedeckeh@gmail.com>
dnsmasq can match tags in its dhcp-range configuration, this commit adds
the option to configure it in the dhcp section
uci configuration:
config dhcp 'lan'
option interface 'lan'
list tag 'blue'
list tag '!red'
option start '10'
option limit '150'
option leasetime '12h'
generated dnsmasq configuration:
dhcp-range=tag:blue,tag:!red,set:lan,192.168.1.10,192.168.1.159,255.255.255.0,12h
Signed-off-by: Grégoire Delattre <gregoire.delattre@gmail.com>
With this patch the dnsmasq init script manages resolv.conf if and only if
when dnsmasq will listen on 127.0.0.1#53 (is main resolver instance).
Also, resolvfile is now set irrespective of the value of noresolv.
Fixes (partially) FS#785
Signed-off-by: Paul Oranje <por@xs4all.nl>
'non-wildcard' interfaces enables dnsmasq's '--bind-dynamic' mode. This
binds to interfaces rather than wildcard addresses *and* keeps track of
interface comings/goings via a unique Linux api.
Quoting dnsmasq's author "bind-dynamic (bind individual addresses, keep
up with changes in interface config) ... On linux, there's actually no
sane reason not to use --bind-dynamic, and it's only not the default for
historical reasons."
Let's change history, well on LEDE at least, and change the default!
Signed-off-by: Kevin Darbyshire-Bryant <kevin@darbyshire-bryant.me.uk>
Commit b32689afd6 added support for dhcp-script hook.
Adding dhcp-script config option results into two instances of dnsmasq being run
which triggered oom issues on platforms having low memory.
The dnsmasq dhcp-script config option will now only be added if at least one of the
dhcp, tftp, neigh hotplug dirs has a regular hotplug file or if the dhcpscript uci
config option is specified.
Signed-off-by: Hans Dedecker <dedeckeh@gmail.com>
Adds a script which acts as a hook for when dnsmasq creates/destroys a
lease, or completes a TFTP file transfer. The hook loops through scripts
in appropriate directories inside '/etc/hotplug.d', executing each one with
the same arguments supplied by dnsmasq.
In case dnsmasq is jailed by ujail the dhcp-script hook will not work as
expected as ujail does not yet support executing a script within a jail.
Signed-off-by: Nick Brassel <nick@tzarc.org>
Signed-off-by: Hans Dedecker <dedeckeh@gmail.com>
Some small tweaks and improvements :
9828ab1 Fix compiler warning.
f77700a Fix compiler warning.
0fbd980 Fix compiler warning.
43cdf1c Remove automatic IDN support when building i18n.
ff19b1a Fix &/&& confusion.
2aaea18 Add .gitattributes to substitute VERSION on export.
Signed-off-by: Hans Dedecker <dedeckeh@gmail.com>
When in ra server mode, configure nameservers passed in router
announcements from the dns value (which is already used by odhcpd).
This also fixes FS#677 by using the global IPv6 address of the router
instead of the link local address (if no nameservers are configured).
Signed-off-by: Arjen de Korte <build+lede@de-korte.org>
Signed-off-by: Hans Dedecker <dedeckeh@gmail.com> [PKG_RELEASE increase]
This patch adds the interface-name option for each dhcp config
in /etc/config/dhcp.
With the interface_name option users can define a DNS name for each dhcp section
that will be resolved by dnsmasq with the underlaying interface address.
For example:
config dhcp 'lan'
option interface 'lan'
...
list interface_name 'home.lan'
...
Signed-off-by: Daniel Danzberger <daniel@dd-wrt.com>
Signed-off-by: Hans Dedecker <dedeckeh@gmail.com> [PKG_RELEASE increase]
A number of small tweaks & improvements on the way to a final release.
Most notable:
Improve DHCPv4 address-in-use check.
Remove the recently introduced RFC-6842 (Client-ids in DHCP replies)
support as it turns out some clients are getting upset.
Signed-off-by: Kevin Darbyshire-Bryant <kevin@darbyshire-bryant.me.uk>
RFC 6761 defines a number of top level domains should not be forwarded
to the Internet's domain servers since they are not responsible for
those domains.
This change adds a list of domains that will be blocked when 'boguspriv'
is used and augments that which is already blocked by dnsmasq's notion
of 'local service' using '--bogus-priv' i.e. RFC 1918 private addresses
and IPv6 prefixes as defined in RFC 6303.
To make this configurable rather than hard coded in dnsmasq's init
script, a new file /usr/share/dnsmasq/rfc6761.conf is conditionally
included.
The default file matches the RFC 6761 recommendation along with a few
other top level domains that should not be forwarded to the Internet.
Compile & run tested Archer C7 v2
Signed-off-by: Kevin Darbyshire-Bryant <kevin@darbyshire-bryant.me.uk>
--bogus-priv now applies to IPv6 prefixes as specified in RFC6303 - this
is significantly friendlier to upstream servers.
CNAME fix in auth mode - A domain can only have a CNAME if it has no
other records
Drop 2 patches now included upstream.
Compile & run tested Archer C7 v2
Signed-off-by: Kevin Darbyshire-Bryant <kevin@darbyshire-bryant.me.uk>
New test release (since test1) includes 2 LEDE patches that are
upstream and may be dropped, along with many spelling fixes.
Add forthcoming 2017 root zone trust anchor to trust-anchors.conf.
Backport 2 patches that just missed test3:
Reduce logspam of those domains handled locally 'local addresses only'
Implement RFC-6842 (Client-ids in DHCP replies)
Compile & run tested Archer C7 v2
Signed-off-by: Kevin Darbyshire-Bryant <kevin@darbyshire-bryant.me.uk>
Bump to dnsmasq 2.77test1 - this includes a number of fixes since 2.76
and allows dropping of 2 LEDE carried patches.
Notable fix in rrfilter code when talking to Nominum's DNS servers
especially with DNSSEC.
A patch to switch dnsmasq back to 'soft fail' for SERVFAIL responses
from dns servers is also included. This mean dnsmasq tries all
configured servers before giving up.
A 'localise queries' enhancement has also been backported (it will
appear in test2/rc'n') this is especially important if using the
recently imported to LEDE 'use dnsmasq standalone' feature 9525743c
I have been following dnsmasq HEAD ever since 2.76 release.
Compile & Run tested: ar71xx, Archer C7 v2
Tested-by: Kevin Darbyshire-Bryant <kevin@darbyshire-bryant.me.uk>
Signed-off-by: Kevin Darbyshire-Bryant <kevin@darbyshire-bryant.me.uk>
dnsmasq has sufficient services to meet the needs of DHCP
and RA with IP6 for single router router users. This is
the most common use for consumer routers. Its reenforced
as most ISP tend to only DHCP-PD /64. dnsmasq has year
over year demonstrated great flexibility in its option
set, and support for off-standard DHCP clients.
odhcpd has enhanced capabilities focused on IP6 such
as DHCP/RA relay and NDP proxy. However, it is not as
flexible in its option set. odhcpd is not as forgiving
with off-standard DHCP clients. Some points may represent
a long term TODO list, but it is the state currently.
These changes make any such combination possible. Already
odhcpd can be set as the main dhcp server. Now odhcpd
can be removed or disabled and dnsmasq will take over
if DHCPv6 compiled in. The existing DHCPv6 and RA UCI
are translated into dnsmasq.conf. The changes focus on
'--dhcp-range', '--dhcp-host', and '--dhcp-options'.
DHCP host ID is least 16 bits [::1000-::FFFF], but
leaves low range for typical infrastructure assignments.
dnsmasq accepts DHCPv6 options in the tranditional
'--dhcp-option' put they must be prefixed 'option6:'.
dnsmasq will also discover SLAAC DNS entries from DHCPv4
clients MAC, and confirm with a ping at least renew.
Long term TODO include improving use of dnsmasq relay
options for DHCPv4 and DHCPv6 in parallel. It would also
be possible to preconfigure DHCP-PD in host-with-options
records for fixed infrastructure.
Signed-off-by: Eric Luehrsen <ericluehrsen@hotmail.com>
[Jo-Philipp Wich: emit proper IPv6 hostid format in dhcp-host directive]
Signed-off-by: Jo-Philipp Wich <jo@mein.io>
ref commit 612e2276b4
ref commit ec63e3bf13
'option add_local_hostname' scripted implementation statically assigns
this host in auto generated host file at init. If IFUP or other signals
do not occur, then address changes are not tracked. The script doesn't
apply all the addresses at an interface. This may make logs obscure.
The script only puts the bare host name (maybe not FQDN) in host file,
but if '--exapandhosts' is enabled, then /etc/hosts entries will be
suffixed, and "127.0.0.1 localhost" becomes "localhost.lan".
dnsmasq provides an option to perform this function, but it is rather
greedy. '--interface-name=<name>,<iface>' will assign the name to all
IP on the specified interface (except link local). This is a useful
feature, but some setups depend on the original restrictive behavior.
'option add_local_fqdn' is added to enhance the feature set, but
if not entered or empty string, then it will default to original
option and behavior. This new option has a few settings. At each
increased setting the most detailed name becomes the PTR record:
0 - same as add_local_hostname 0 or disabled
1 - same as add_local_hostname 1
2 - assigns the bare host name to all IP w/ --dnsmasq-interface
3 - assigns the FQDN and host to all IP w/ --dnsmasq-interface
4 - assigns <iface>.<host>.<domain> and above w/ --dnsmasq-nterface
'option add_wan_fqdn' is added to run the same procedure on
inferred WAN intefaces. If an interface has 'config dhcp' and
'option ignore 1' set, then it is considered WAN. The original
option would only run on DHCP serving interfaces.
Signed-off-by: Eric Luehrsen <ericluehrsen@hotmail.com>
This causes problem when a FQDN is configured in /etc/config/system. The
domain name will appear twice in reverse DNS.
Next to that, there seems to be a bug in dnsmasq. From the manual page:
--interface-name=<name>,<interface>[/4|/6]
Return a DNS record associating the name with the primary address
on the given interface. This flag specifies an A or AAAA record for the
given name in the same way as an /etc/hosts line, except that the address
is not constant, but taken from the given interface. The interface may be
followed by "/4" or "/6" to specify that only IPv4 or IPv6 addresses
of the interface should be used. If the interface is down, not configured
or non-existent, an empty record is returned. The matching PTR record is
also created, mapping the interface address to the name. More than one name
may be associated with an interface address by repeating the flag; in that
case the first instance is used for the reverse address-to-name mapping.
It does not just create an A/AAAA record for the primary address, it creates
one for all addresses. And what is worse, it seems to actually resolve to the
non-primary address first. This is quite annoying when you use floating IP
addresses (e.g. VRRP), because when the floating IP is on the other device,
SSH failes due to incorrect entry in the known hosts file.
I know that this is not a common setup, but it would be nice if there was an
option to restore the previous behaviour, rather than just forcing this new
feature on everybody.
Reported-by: Stijn Tintel <stijn@linux-ipv6.be>
Signed-off-by: Hans Dedecker <dedeckeh@gmail.com>
'add_local_hostname' previous implementation may drop some addresses.
Soft addition of IP6 addresses may not cause a reload or restart event.
dnsmasq '--interface-name' robustly applies DNS to all addresses per
interface (except fe80::/10).
Change UCI 'add_local_hostname' to expand during each interface assignement
during add_dhcp().
Assign '<iface>.<host>.<domain>' as true name (reflexive A, AAAA, and PTR).
Assign '<host>.<domain>' and '<host>' as convinience aliases (no PTR, not
technically CNAME).
This is accomplished with the '--interface-name' order, first is PTR.
We could also assign each <ip4/6>.<iface>.<host>.<domain> to the respective
dual stack on the interface.
That seemed excessive so it was skipped (/4 or /6 suffix to the interface).
Add UCI 'add_wan_hostname' similar to 'add_local_hostname' function for
external WAN.
WAN IP4 are less often named by the ISP and rarely WAN IP6 due to complexity.
For logs, LuCI connection graph, and other uses assigning a WAN name is desired.
'add_local_hostname' only applies with DHCP and 'add_wam_hostname' only applies
without DHCP. Common residential users will want to set both options TRUE.
Businesses will probably have global DNS, static IP, and 'add_wan_hostname' FALSE.
Signed-off-by: Eric Luehrsen <ericluehrsen@hotmail.com>
Let dnsmasq read all hosts files in /tmp/hosts directory by specifying
/tmp/hosts as argument of --addn-host
Signed-off-by: Hans Dedecker <dedeckeh@gmail.com>
Adds the mac address of the DNS requestor to DNS queries which
are forwarded upstream and can be used to do filtering by the
upstream servers. This only works if the requestor is on the
same subnet as the dnsmasq server
The addmac parameter can hold the following values:
0 : mac address is not added
1 : mac address is added in binary format
base64 : mac address is added base64 encoded
text: : mac address is added in human readable format
as hex and colons
Signed-off-by: Hans Dedecker <dedeckeh@gmail.com>
Permit users of the full variant to disable the NO_ID *.bind pseudo
domain masking.
Defaulted 'on' in all variants.
Signed-off-by: Kevin Darbyshire-Bryant <kevin@darbyshire-bryant.me.uk>
Don't expose dnsmasq version & other data to clients via the *.bind
pseudo domain. This uses a new 'NO_ID' compile time option which has been
discussed and submitted upstream.
This is an alternate to replacing version with 'unknown' which affects
the version reported to syslog and 'dnsmasq --version'
Run time tested with & without NO_ID on Archer C7 v2
Signed-off-by: Kevin Darbyshire-Bryant <kevin@darbyshire-bryant.me.uk>
