embedaddon/dnsmasq/CHANGELOG - view

File: [ELWIX - Embedded LightWeight unIX -] / embedaddon / dnsmasq / CHANGELOG
Revision 1.1.1.5 (vendor branch): download - view: text, annotated - select for diffs - revision graph
Wed Sep 27 11:02:07 2023 UTC (13 months ago) by misho
Branches: elwix, dnsmasq, MAIN
CVS tags: v8_2p1, HEAD

Version 8.2p1

version 2.90 Fix reversion in --rev-server introduced in 2.88 which caused breakage if the prefix length is not exactly divisible by 8 (IPv4) or 4 (IPv6). Fix possible SEGV when there server(s) for a particular domain are configured, but no server which is not qualified for a particular domain. Thanks to Daniel Danzberger for spotting this bug. version 2.89 Fix bug introduced in 2.88 (commit fe91134b) which can result in corruption of the DNS cache internal data structures and logging of "cache internal error". This has only been seen in one place in the wild, and it took considerable effort to even generate a test case to reproduce it, but there's no way to be sure it won't strike, and the effect is to break the cache badly. Installations with DNSSEC enabled are more likely to see the problem, but not running DNSSEC does not guarantee that it won't happen. Thanks to Timo van Roermund for reporting the bug and for his great efforts in chasing it down. version 2.88 Fix bug in --dynamic-host when an interface has /16 IPv4 address. Thanks to Mark Dietzer for spotting this. Add --fast-dns-retry option. This gives dnsmasq the ability to originate retries for upstream DNS queries itself, rather than relying on the downstream client. This is most useful when doing DNSSEC over unreliable upstream networks. It comes with some cost in memory usage and network bandwidth. Add --use-stale-cache option. When set, if a DNS name exists in the cache, but its time-to-live has expired, dnsmasq will return the data anyway. (It attempts to refresh the data with an upstream query after returning the stale data.) This can improve speed and reliability. It comes at the expense of sometimes returning out-of-date data and less efficient cache utilisation, since old data cannot be flushed when its TTL expires, so the cache becomes strictly least-recently-used. Add --port-limit option which allows tuning for robustness in the face of some upstream network errors. Thanks to Prashant Kumar Singh, Ravi Nagayach and Mike Danilov, all of Amazon Web Services, for their efforts in developing this and the stale-cache and fast-retry options. Make --hostsdir (but NOT --dhcp-hostsdir and --dhcp-optsdir) handle removal of whole files or entries within files. Thanks to Dominik Derigs for the initial patches for this. Fix bug, introduced in 2.87, which could result in DNS servers being removed from the configuration when reloading server configuration from DBus, or re-reading /etc/resolv.conf Only servers from the same source should be replaced, but some servers from other sources (i.e., hard coded or another dynamic source) could mysteriously disappear. Thanks to all reporting this, but especially Christopher J. Madsen who reduced the problem to an easily reproducible case which saved much labour in finding it. Add --no-round-robin option. Allow domain names as well as IP addresses when specifying upstream DNS servers. There are some gotchas associated with this (it will mysteriously fail to work if the dnsmasq instance being started is in the path from the system resolver to the DNS), and a seemingly sensible configuration like --server=domain.name@1.2.3.4 is unactionable if domain.name only resolves to an IPv6 address). There are, however, cases where is can be useful. Thanks to Dominik Derigs for the patch. Handle DS records for unsupported crypto algorithms correctly. Such a DS, as long as it is validated, should allow answers in the domain it attests to be returned as unvalidated, and not as a validation error. Optimise reading large numbers of --server options. When re-reading upstream servers from /etc/resolv.conf or other sources that can change dnsmasq tries to avoid memory fragmentation by re-using existing records that are being re-read unchanged. This involves seaching all the server records for each new one installed. During startup this search is pointless, and can cause long start times with thousands of --server options because the work needed is O(n^2). Handle this case more intelligently. Thanks to Ye Zhou for spotting the problem and an initial patch. If we detect that a DNS reply from upstream is malformed don't return it to the requestor; send a SEVFAIL rcode instead. version 2.87 Allow arbitrary prefix lengths in --rev-server and --domain=....,local Replace --address=/#/..... functionality which got missed in the 2.86 domain search rewrite. Add --nftset option, like --ipset but for the newer nftables. Thanks to Chen Zhenge for the patch. Add --filter-A and --filter-AAAA options, to remove IPv4 or IPv6 addresses from DNS answers. Fix crash doing netbooting when --port is set to zero to disable the DNS server. Thanks to Drexl Johannes for the bug report. Generalise --dhcp-relay. Sending via broadcast/multicast is now supported for both IPv4 and IPv6 and the configuration syntax made easier (but backwards compatible). Add snooping of IPv6 prefix-delegations to the DHCP-relay system. Finesse parsing of --dhcp-remoteid and --dhcp-subscrid. To be treated as hex, the pattern must consist of only hex digits AND contain at least one ':'. Thanks to Bengt-Erik Sandstrom who tripped over a pattern consisting of a decimal number which was interpreted surprisingly. Include client address in TFTP file-not-found error reports. Thanks to Stefan Rink for the initial patch, which has been re-worked by me (srk). All bugs mine. Note in manpage the change in behaviour of -address. This behaviour actually changed in v2.86, but was undocumented there. From 2.86 on, (eg) --address=/example.com/1.2.3.4 ONLY applies to A queries. All other types of query will be sent upstream. Pre 2.86, that would catch the whole example.com domain and queries for other types would get a local NODATA answer. The pre-2.86 behaviour is still available, by configuring --address=/example.com/1.2.3.4 --local=/example.com/ Fix problem with binding DHCP sockets to an individual interface. Despite the fact that the system call tales the interface _name_ as a parameter, it actually, binds the socket to interface _index_. Deleting the interface and creating a new one with the same name leaves the socket bound to the old index. (Creating new sockets always allocates a fresh index, they are not reused). We now take this behaviour into account and keep up with changing indexes. Add --conf-script configuration option. Enhance --domain to accept, for instance, --domain=net2.thekelleys.org.uk,eth2 so that hosts get a domain which relects the interface they are attached to in a way which doesn't require hard-coding addresses. Thanks to Sten Spans for the idea. Fix write-after-free error in DHCPv6 server code. CVE-2022-0934 refers. Add the ability to specify destination port in DHCP-relay mode. This change also removes a previous bug where --dhcp-alternate-port would affect the port used to relay _to_ as well as the port being listened on. The new feature allows configuration to provide bug-for-bug compatibility, if required. Thanks to Damian Kaczkowski for the feature suggestion. Bound the value of UDP packet size in the EDNS0 header of forwarded queries to the configured or default value of edns-packet-max. There's no point letting a client set a larger value if we're unable to return the answer. Thanks to Bertie Taylor for pointing out the problem and supplying the patch. Fix problem with the configuration --server=/some.domain/# --address=/#/<ip> --server=<server_ip> This would return <ip> for queries in some.domain, rather than forwarding the query via the default server. Tweak DHCPv6 relay code so that packets relayed towards a server have source address on the server-facing network, not the client facing network. Thanks to Luis Thomas for spotting this and initial patch. version 2.86 Handle DHCPREBIND requests in the DHCPv6 server code. Thanks to Aichun Li for spotting this omission, and the initial patch. Fix bug which caused dnsmasq to lose track of processes forked to handle TCP DNS connections under heavy load. The code checked that at least one free process table slot was available before listening on TCP sockets, but didn't take into account that more than one TCP connection could arrive, so that check was not sufficient to ensure that there would be slots for all new processes. It compounded this error by silently failing to store the process when it did run out of slots. Even when this bug is triggered, all the right things happen, and answers are still returned. Only under very exceptional circumstances, does the bug manifest itself: see https://lists.thekelleys.org.uk/pipermail/dnsmasq-discuss/2021q2/014976.html Thanks to Tijs Van Buggenhout for finding the conditions under which the bug manifests itself, and then working out exactly what was going on. Major rewrite of the DNS server and domain handling code. This should be largely transparent, but it drastically improves performance and reduces memory foot-print when configuring large numbers domains of the form local=/adserver.com/ or local=/adserver.com/# Lookup times now grow as log-to-base-2 of the number of domains, rather than greater than linearly, as before. The change makes multiple addresses associated with a domain work address=/example.com/1.2.3.4 address=/example.com/5.6.7.8 It also handles multiple upstream servers for a domain better; using the same try/retry algorithms as non domain-specific servers. This also applies to DNSSEC-generated queries. Finally, some of the oldest and gnarliest code in dnsmasq has had a significant clean-up. It's far from perfect, but it _is_ better. Revise resource handling for number of concurrent DNS queries. This used to have a global limit, but that has a problem when using different servers for different upstream domains. Queries which are routed by domain to an upstream server which is not responding will build up and trigger the limit, which breaks DNS service for all other domains which could be handled by other servers. The change is to make the limit per server-group, where a server group is the set of servers configured for a particular domain. In the common case, where only default servers are declared, there is no effective change. Improve efficiency of DNSSEC. The sharing point for DNSSEC RR data used to be when it entered the cache, having been validated. After that queries requiring the KEY or DS records would share the cached values. There is a common case in dual-stack hosts that queries for A and AAAA records for the same domain are made simultaneously. If required keys were not in the cache, this would result in two requests being sent upstream for the same key data (and all the subsequent chain-of-trust queries.) Now we combine these requests and elide the duplicates, resulting in fewer queries upstream and better performance. To keep a better handle on what's going on, the "extra" logging mode has been modified to associate queries and answers for DNSSEC queries in the same way as ordinary queries. The requesting address and port have been removed from DNSSEC logging lines, since this is no longer strictly defined. Connection track mark based DNS query filtering. Thanks to Etan Kissling for implementing this It extends query filtering support beyond what is currently possible with the `--ipset` configuration option, by adding support for: 1) Specifying allowlists on a per-client basis, based on their associated Linux connection track mark. 2) Dynamic configuration of allowlists via Ubus. 3) Reporting when a DNS query resolves or is rejected via Ubus. 4) DNS name patterns containing wildcards. Disallowed queries are not forwarded; they are rejected with a REFUSED error code. Allow smaller than 64 prefix lengths in synth-domain, with caveats. --synth-domain=1234:4567::/56,example.com is now valid. Make domains generated by --synth-domain appear in replies when in authoritative mode. Ensure CAP_NET_ADMIN capability is available when conntrack is configured. Thanks to Yick Xie for spotting the lack of this. When --dhcp-hostsfile --dhcp-optsfile and --addn-hosts are given a directory as argument, define the order in which files within that directory are read (alphabetical order of filename). Thanks to Ed Wildgoose for the initial patch and motivation for this. Allow adding IP address to nftables set in addition to ipset. version 2.85 Fix problem with DNS retries in 2.83/2.84. The new logic in 2.83/2.84 which merges distinct requests for the same domain causes problems with clients which do retries as distinct requests (differing IDs and/or source ports.) The retries just get piggy-backed on the first, failed, request. The logic is now changed so that distinct requests for repeated queries still get merged into a single ID/source port, but they now always trigger a re-try upstream. Thanks to Nicholas Mu for his analysis. Tweak sort order of tags in get-version. v2.84 sorts before v2.83, but v2.83 sorts before v2.83rc1 and 2.83rc1 sorts before v2.83test1. This fixes the problem which lead to 2.84 announcing itself as 2.84rc2. Avoid treating a --dhcp-host which has an IPv6 address as eligible for use with DHCPv4 on the grounds that it has no address, and vice-versa. Thanks to Viktor Papp for spotting the problem. (This bug was fixed was back in 2.67, and then regressed in 2.81). Add --dynamic-host option: A and AAAA records which take their network part from the network of a local interface. Useful for routers with dynamically prefixes. Thanks to Fred F for the suggestion. Teach --bogus-nxdomain and --ignore-address to take an IPv4 subnet. Use random source ports where possible if source addresses/interfaces in use. CVE-2021-3448 applies. Thanks to Petr Menšík for spotting this. It's possible to specify the source address or interface to be used when contacting upstream name servers: server=8.8.8.8@1.2.3.4 or server=8.8.8.8@1.2.3.4#66 or server=8.8.8.8@eth0, and all of these have, until now, used a single socket, bound to a fixed port. This was originally done to allow an error (non-existent interface, or non-local address) to be detected at start-up. This means that any upstream servers specified in such a way don't use random source ports, and are more susceptible to cache-poisoning attacks. We now use random ports where possible, even when the source is specified, so server=8.8.8.8@1.2.3.4 or server=8.8.8.8@eth0 will use random source ports. server=8.8.8.8@1.2.3.4#66 or any use of --query-port will use the explicitly configured port, and should only be done with understanding of the security implications. Note that this change changes non-existing interface, or non-local source address errors from fatal to run-time. The error will be logged and communication with the server not possible. Change the method of allocation of random source ports for DNS. Previously, without min-port or max-port configured, dnsmasq would default to the compiled in defaults for those, which are 1024 and 65535. Now, when neither are configured, it defaults instead to the kernel's ephemeral port range, which is typically 32768 to 60999 on Linux systems. This change eliminates the possibility that dnsmasq may be using a registered port > 1024 when a long-running daemon starts up and wishes to claim it. This change does likely slightly reduce the number of random ports and therefore the protection from reply spoofing. The older behaviour can be restored using the min-port and max-port config switches should that be a concern. Scale the size of the DNS random-port pool based on the value of the --dns-forward-max configuration. Tweak TFTP code to check sender of all received packets, as specified in RFC 1350 para 4. Support some wildcard matching of input tags to --tag-if. Thanks to Geoff Back for the idea and the patch. version 2.84 Fix a problem, introduced in 2.83, which could see DNS replies being sent via the wrong socket. On machines running both IPv4 and IPv6 this could result in sporadic messages of the form "failed to send packet: Network is unreachable" and the lost of the query. Since the error is sporadic and of low probability, the client retry would normally succeed. Change HAVE_NETTLEHASH compile-time to HAVE_CRYPTOHASH. version 2.83 Use the values of --min-port and --max-port in outgoing TCP connections to upstream DNS servers. Fix a remote buffer overflow problem in the DNSSEC code. Any dnsmasq with DNSSEC compiled in and enabled is vulnerable to this, referenced by CVE-2020-25681, CVE-2020-25682, CVE-2020-25683 CVE-2020-25687. Be sure to only accept UDP DNS query replies at the address from which the query was originated. This keeps as much entropy in the {query-ID, random-port} tuple as possible, to help defeat cache poisoning attacks. Refer: CVE-2020-25684. Use the SHA-256 hash function to verify that DNS answers received are for the questions originally asked. This replaces the slightly insecure SHA-1 (when compiled with DNSSEC) or the very insecure CRC32 (otherwise). Refer: CVE-2020-25685. Handle multiple identical near simultaneous DNS queries better. Previously, such queries would all be forwarded independently. This is, in theory, inefficient but in practise not a problem, _except_ that is means that an answer for any of the forwarded queries will be accepted and cached. An attacker can send a query multiple times, and for each repeat, another {port, ID} becomes capable of accepting the answer he is sending in the blind, to random IDs and ports. The chance of a successful attack is therefore multiplied by the number of repeats of the query. The new behaviour detects repeated queries and merely stores the clients sending repeats so that when the first query completes, the answer can be sent to all the clients who asked. Refer: CVE-2020-25686. version 2.82 Improve behaviour in the face of network interfaces which come and go and change index. Thanks to Petr Mensik for the patch. Convert hard startup failure on NETLINK_NO_ENOBUFS under qemu-user to a warning. Allow IPv6 addresses ofthe form [::ffff:1.2.3.4] in --dhcp-option. Fix crash under heavy TCP connection load introduced in 2.81. Thanks to Frank for good work chasing this down. Change default lease time for DHCPv6 to one day. Alter calculation of preferred and valid times in router advertisements, so that these do not have a floor applied of the lease time in the dhcp-range if this is not explicitly specified and is merely the default. Thanks to Martin-Éric Racine for suggestions on this. version 2.81 Improve cache behaviour for TCP connections. For ease of implementation, dnsmasq has always forked a new process to handle each incoming TCP connection. A side-effect of this is that any DNS queries answered from TCP connections are not cached: when TCP connections were rare, this was not a problem. With the coming of DNSSEC, it is now the case that some DNSSEC queries have answers which spill to TCP, and if, for instance, this applies to the keys for the root, then those never get cached, and performance is very bad. This fix passes cache entries back from the TCP child process to the main server process, and fixes the problem. Remove the NO_FORK compile-time option, and support for uclinux. In an era where everything has an MMU, this looks like an anachronism, and it adds to (Ok, multiplies!) the combinatorial explosion of compile-time options. Thanks to Kevin Darbyshire-Bryant for the patch. Fix line-counting when reading /etc/hosts and friends; for correct error messages. Thanks to Christian Rosentreter for reporting this. Fix bug in DNS non-terminal code, added in 2.80, which could sometimes cause a NODATA rather than an NXDOMAIN reply. Thanks to Norman Rasmussen, Sven Mueller and Maciej Żenczykowski for spotting and diagnosing the bug and providing patches. Support TCP-fastopen (RFC-7413) on both incoming and outgoing TCP connections, if supported and enabled in the OS. Improve kernel-capability manipulation code under Linux. Dnsmasq now fails early if a required capability is not available, and tries not to request capabilities not required by its configuration. Add --shared-network config. This enables allocation of addresses by the DHCP server in subnets where the server (or relay) does not have an interface on the network in that subnet. Many thanks to kamp.de for sponsoring this feature. Fix broken contrib/lease_tools/dhcp_lease_time.c. A packet validation check got borked in commit 2b38e382 and release 2.80. Thanks to Tomasz Szajner for spotting this. Fix compilation against nettle version 3.5 and later. Fix spurious DNSSEC validation failures when the auth section of a reply contains unsigned RRs from a signed zone, with the exception that NSEC and NSEC3 RRs must always be signed. Thanks to Tore Anderson for spotting and diagnosing the bug. Add --dhcp-ignore-clid. This disables reading of DHCP client identifier option (option 61), so clients are only identified by MAC addresses. Fix a bug which stopped --dhcp-name-match from working when a hostname is supplied in --dhcp-host. Thanks to James Feeney for spotting this. Fix bug which caused very rarely caused zero-length DHCPv6 packets. Thanks to Dereck Higgins for spotting this. Add --tftp-single-port option. Enhance --conf-dir to load files in a deterministic order. Thanks to Evgenii Seliavka for the suggestion and initial patch. In the router advert code, handle case where we have two different interfaces on the same IPv6 net, and we are doing RA/DHCP service on only one of them. Thanks to NIIBE Yutaka for spotting this case and making the initial patch. Support prefixed ranges of ipv6 addresses in dhcp-host. This eases problems chain-netbooting, where each link in the chain requests an address using a different UID. With a single address, only one gets the "static" address, but with this fix, enough addresses can be reserved for all the stages of the boot. Many thanks to Harald Jensås for his work on this idea and earlier patches. Add filtering by tag of --dhcp-host directives. Based on a patch by Harald Jensås. Allow empty server spec in --rev-server, to match --server. Remove DSA signature verification from DNSSEC, as specified in RFC 8624. Thanks to Loganaden Velvindron for the original patch. Add --script-on-renewal option. version 2.80 Add support for RFC 4039 DHCP rapid commit. Thanks to Ashram Method for the initial patch and motivation. Alter the default for dnssec-check-unsigned. Versions of dnsmasq prior to 2.80 defaulted to not checking unsigned replies, and used --dnssec-check-unsigned to switch this on. Such configurations will continue to work as before, but those which used the default of no checking will need to be altered to explicitly select no checking. The new default is because switching off checking for unsigned replies is inherently dangerous. Not only does it open the possiblity of forged replies, but it allows everything to appear to be working even when the upstream namesevers do not support DNSSEC, and in this case no DNSSEC validation at all is occuring. Fix DHCP broken-ness when --no-ping AND --dhcp-sequential-ip are set. Thanks to Daniel Miess for help with this. Add a facilty to store DNS packets sent/recieved in a pcap-format file for later debugging. The file location is given by the --dumpfile option, and a bitmap controlling which packets should be dumped is given by the --dumpmask option. Handle the case of both standard and constructed dhcp-ranges on the same interface better. We don't now contruct a dhcp-range if there's already one specified. This allows the specified interface to have different parameters and avoids advertising the same prefix twice. Thanks to Luis Marsano for spotting this case. Allow zone transfer in authoritative mode if auth-peer is specified, even if auth-sec-servers is not. Thanks to Raphaël Halimi for the suggestion. Fix bug which sometimes caused dnsmasq to wrongly return answers without DNSSEC RRs to queries with the do-bit set, but only when DNSSEC validation was not enabled. Thanks to Petr Menšík for spotting this. Fix missing fatal errors with some malformed options (server, local, address, rebind-domain-ok, ipset, alias). Thanks to Eugene Lozovoy for spotting the problem. Fix crash on startup with a --synth-domain which has no prefix. Introduced in 2.79. Thanks to Andreas Engel for the bug report. Fix missing EDNS0 section in some replies generated by local DNS configuration which confused systemd-resolvd. Thanks to Steve Dodd for characterising the problem. Add --dhcp-name-match config option. Add --caa-record config option. Implement --address=/example.com/# as (more efficient) syntactic sugar for --address=/example.com/0.0.0.0 and --address=/example.com/:: Returning null addresses is a useful technique for ad-blocking. Thanks to Peter Russell for the suggestion. Change anti cache-snooping behaviour with queries with the recursion-desired bit unset. Instead to returning SERVFAIL, we now always forward, and never answer from the cache. This allows "dig +trace" command to work. Include in the example config file a formulation which stops DHCP clients from claiming the DNS name "wpad". This is a fix for the CERT Vulnerability VU#598349. version 2.79 Fix parsing of CNAME arguments, which are confused by extra spaces. Thanks to Diego Aguirre for spotting the bug. Where available, use IP_UNICAST_IF or IPV6_UNICAST_IF to bind upstream servers to an interface, rather than SO_BINDTODEVICE. Thanks to Beniamino Galvani for the patch. Always return a SERVFAIL answer to DNS queries without the recursion desired bit set, UNLESS acting as an authoritative DNS server. This avoids a potential route to cache snooping. Add support for Ed25519 signatures in DNSSEC validation. No longer support RSA/MD5 signatures in DNSSEC validation, since these are not secure. This behaviour is mandated in RFC-6944. Fix incorrect error exit code from dhcp_release6 utility. Thanks Gaudenz Steinlin for the bug report. Use SIGINT (instead of overloading SIGHUP) to turn on DNSSEC time validation when --dnssec-no-timecheck is in use. Note that this is an incompatible change from earlier releases. Allow more than one --bridge-interface option to refer to an interface, so that we can use --bridge-interface=int1,alias1 --bridge-interface=int1,alias2 as an alternative to --bridge-interface=int1,alias1,alias2 Thanks to Neil Jerram for work on this. Fix for DNSSEC with wildcard-derived NSEC records. It's OK for NSEC records to be expanded from wildcards, but in that case, the proof of non-existence is only valid starting at the wildcard name, *.<domain> NOT the name expanded from the wildcard. Without this check it's possible for an attacker to craft an NSEC which wrongly proves non-existence. Thanks to Ralph Dolmans for finding this, and co-ordinating the vulnerability tracking and fix release. CVE-2017-15107 applies. Remove special handling of A-for-A DNS queries. These are no longer a significant problem in the global DNS. http://cs.northwestern.edu/~ychen/Papers/DNS_ToN15.pdf Thanks to Mattias Hellström for the initial patch. Fix failure to delete dynamically created dhcp options from files in -dhcp-optsdir directories. Thanks to Lindgren Fredrik for the bug report. Add to --synth-domain the ability to create names using sequential numbers, as well as encodings of IP addresses. For instance, --synth-domain=thekelleys.org.uk,192.168.0.50,192.168.0.70,internal-* creates 21 domain names of the form internal-4.thekelleys.org.uk over the address range given, with internal-0.thekelleys.org.uk being 192.168.0.50 and internal-20.thekelleys.org.uk being 192.168.0.70 Thanks to Andy Hawkins for the suggestion. Tidy up Crypto code, removing workarounds for ancient versions of libnettle. We now require libnettle 3. version 2.78 Fix logic of appending ".<layer>" to PXE basename. Thanks to Chris Novakovic for the patch. Revert ping-check of address in DHCPDISCOVER if there already exists a lease for the address. Under some circumstances, and netbooted windows installation can reply to pings before if has a DHCP lease and block allocation of the address it already used during netboot. Thanks to Jan Psota for spotting this. Fix DHCP relaying, broken in 2.76 and 2.77 by commit ff325644c7afae2588583f935f4ea9b9694eb52e. Thanks to John Fitzgibbon for the diagnosis and patch. Try other servers if first returns REFUSED when --strict-order active. Thanks to Hans Dedecker for the patch Fix regression in 2.77, ironically added as a security improvement, which resulted in a crash when a DNS query exceeded 512 bytes (or the EDNS0 packet size, if different.) Thanks to Christian Kujau, Arne Woerner Juan Manuel Fernandez and Kevin Darbyshire-Bryant for chasing this one down. CVE-2017-13704 applies. Fix heap overflow in DNS code. This is a potentially serious security hole. It allows an attacker who can make DNS requests to dnsmasq, and who controls the contents of a domain, which is thereby queried, to overflow (by 2 bytes) a heap buffer and either crash, or even take control of, dnsmasq. CVE-2017-14491 applies. Credit to Felix Wilhelm, Fermin J. Serna, Gabriel Campana Kevin Hamacher and Ron Bowes of the Google Security Team for finding this. Fix heap overflow in IPv6 router advertisement code. This is a potentially serious security hole, as a crafted RA request can overflow a buffer and crash or control dnsmasq. Attacker must be on the local network. CVE-2017-14492 applies. Credit to Felix Wilhelm, Fermin J. Serna, Gabriel Campana and Kevin Hamacher of the Google Security Team for finding this. Fix stack overflow in DHCPv6 code. An attacker who can send a DHCPv6 request to dnsmasq can overflow the stack frame and crash or control dnsmasq. CVE-2017-14493 applies. Credit to Felix Wilhelm, Fermin J. Serna, Gabriel Campana Kevin Hamacher and Ron Bowes of the Google Security Team for finding this. Fix information leak in DHCPv6. A crafted DHCPv6 packet can cause dnsmasq to forward memory from outside the packet buffer to a DHCPv6 server when acting as a relay. CVE-2017-14494 applies. Credit to Felix Wilhelm, Fermin J. Serna, Gabriel Campana Kevin Hamacher and Ron Bowes of the Google Security Team for finding this. Fix DoS in DNS. Invalid boundary checks in the add_pseudoheader function allows a memcpy call with negative size An attacker which can send malicious DNS queries to dnsmasq can trigger a DoS remotely. dnsmasq is vulnerable only if one of the following option is specified: --add-mac, --add-cpe-id or --add-subnet. CVE-2017-14496 applies. Credit to Felix Wilhelm, Fermin J. Serna, Gabriel Campana Kevin Hamacher and Ron Bowes of the Google Security Team for finding this. Fix out-of-memory Dos vulnerability. An attacker which can send malicious DNS queries to dnsmasq can trigger memory allocations in the add_pseudoheader function The allocated memory is never freed which leads to a DoS through memory exhaustion. dnsmasq is vulnerable only if one of the following option is specified: --add-mac, --add-cpe-id or --add-subnet. CVE-2017-14495 applies. Credit to Felix Wilhelm, Fermin J. Serna, Gabriel Campana Kevin Hamacher and Ron Bowes of the Google Security Team for finding this. version 2.77 Generate an error when configured with a CNAME loop, rather than a crash. Thanks to George Metz for spotting this problem. Calculate the length of TFTP error reply packet correctly. This fixes a problem when the error message in a TFTP packet exceeds the arbitrary limit of 500 characters. The message was correctly truncated, but not the packet length, so extra data was appended. This is a possible security risk, since the extra data comes from a buffer which is also used for DNS, so that previous DNS queries or replies may be leaked. Thanks to Mozilla for funding the security audit which spotted this bug. Fix logic error in Linux netlink code. This could cause dnsmasq to enter a tight loop on systems with a very large number of network interfaces. Thanks to Ivan Kokshaysky for the diagnosis and patch. Fix problem with --dnssec-timestamp whereby receipt of SIGHUP would erroneously engage timestamp checking. Thanks to Kevin Darbyshire-Bryant for this work. Bump zone serial on reloading /etc/hosts and friends when providing authoritative DNS. Thanks to Harrald Dunkel for spotting this. Handle v4-mapped IPv6 addresses sanely in --synth-domain. These have standard representation like ::ffff:1.2.3.4 and are now converted to names like <prefix>--ffff-1-2-3-4.<domain> Handle binding upstream servers to an interface (--server=1.2.3.4@eth0) when the named interface is destroyed and recreated in the kernel. Thanks to Beniamino Galvani for the patch. Allow wildcard CNAME records in authoritative zones. For example --cname=*.example.com,default.example.com Thanks to Pro Backup for sponsoring this development. Bump the allowed backlog of TCP connections from 5 to 32, and make this a compile-time configurable option. Thanks to Donatas Abraitis for diagnosing this as a potential problem. Add DNSMASQ_REQUESTED_OPTIONS environment variable to the lease-change script. Thanks to ZHAO Yu for the patch. Fix foobar in rrfilter code, that could cause malformed replies, especially when DNSSEC validation on, and the upstream server returns answer with the RRs in a particular order. The only DNS server known to tickle this is Nominum's. Thanks to Dave Täht for spotting the bug and assisting in the fix. Fix the manpage which lied that only the primary address of an interface is used by --interface-name. Make --localise-queries apply to names from --interface-name. Thanks to Kevin Darbyshire-Bryant and Eric Luehrsen for pushing this. Improve connection handling when talking to TCP upstream servers. Specifically, be prepared to open a new TCP connection when we want to make multiple queries but the upstream server accepts fewer queries per connection. Improve logging of upstream servers when there are a lot of "local addresses only" entries. Thanks to Hannu Nyman for the patch. Make --bogus-priv apply to IPv6, for the prefixes specified in RFC6303. Thanks to Kevin Darbyshire-Bryant for work on this. Allow use of MAC addresses with --tftp-unique-root. Thanks to Floris Bos for the patch. Add --dhcp-reply-delay option. Thanks to Floris Bos for the patch. Add mtu setting facility to --ra-param. Thanks to David Flamand for the patch. Capture STDOUT and STDERR output from dhcp-script and log it as part of the dnsmasq log stream. Makes life easier for diagnosing unexpected problems in scripts. Thanks to Petr Mensik for the patch. Generate fatal errors when failing to parse the output of the dhcp-script in "init" mode. Avoids strange errors when the script accidentally emits error messages. Thanks to Petr Mensik for the patch. Make --rev-server for an RFC1918 subnet work even in the presence of the --bogus-priv flag. Thanks to Vladislav Grishenko for the patch. Extend --ra-param mtu: field to allow an interface name. This allows the MTU of a WAN interface to be advertised on the internal interfaces of a router. Thanks to Vladislav Grishenko for the patch. Do ICMP-ping check for address-in-use for DHCPv4 when the client specifies an address in DHCPDISCOVER, and when an address in configured locally. Thanks to Alin Năstac for spotting the problem. Add new DHCP tag "known-othernet" which is set when only a dhcp-host exists for another subnet. Can be used to ensure that privileged hosts are not given "guest" addresses by accident. Thanks to Todd Sanket for the suggestion. Remove historic automatic inclusion of IDN support when building internationalisation support. This doesn't fit now there is a choice of IDN libraries. Be sure to include either -DHAVE_IDN or -DHAVE_LIBIDN2 for IDN support. version 2.76 Include 0.0.0.0/8 in DNS rebind checks. This range translates to hosts on the local network, or, at least, 0.0.0.0 accesses the local host, so could be targets for DNS rebinding. See RFC 5735 section 3 for details. Thanks to Stephen Röttger for the bug report. Enhance --add-subnet to allow arbitrary subnet addresses. Thanks to Ed Barsley for the patch. Respect the --no-resolv flag in inotify code. Fixes bug which caused dnsmasq to fail to start if a resolv-file was a dangling symbolic link, even of --no-resolv set. Thanks to Alexander Kurtz for spotting the problem. Fix crash when an A or AAAA record is defined locally, in a hosts file, and an upstream server sends a reply that the same name is empty. Thanks to Edwin Török for the patch. Fix failure to correctly calculate cache-size when reading a hosts-file fails. Thanks to André Glüpker for the patch. Fix wrong answer to simple name query when --domain-needed set, but no upstream servers configured. Dnsmasq returned REFUSED, in this case, when it should be the same as when upstream servers are configured - NOERROR. Thanks to Allain Legacy for spotting the problem. Return REFUSED when running out of forwarding table slots, not SERVFAIL. Add --max-port configuration. Thanks to Hans Dedecker for the patch. Add --script-arp and two new functions for the dhcp-script. These are "arp" and "arp-old" which announce the arrival and removal of entries in the ARP or neighbour tables. Extend --add-mac to allow a new encoding of the MAC address as base64, by configuring --add-mac=base64 Add --add-cpe-id option. Don't crash with divide-by-zero if an IPv6 dhcp-range is declared as a whole /64. (ie xx::0 to xx::ffff:ffff:ffff:ffff) Thanks to Laurent Bendel for spotting this problem. Add support for a TTL parameter in --host-record and --cname. Add --dhcp-ttl option. Add --tftp-mtu option. Thanks to Patrick McLean for the initial patch. Check return-code of inet_pton() when parsing dhcp-option. Bad addresses could fail to generate errors and result in garbage dhcp-options being sent. Thanks to Marc Branchaud for spotting this. Fix wrong value for EDNS UDP packet size when using --servers-file to define upstream DNS servers. Thanks to Scott Bonar for the bug report. Move the dhcp_release and dhcp_lease_time tools from contrib/wrt to contrib/lease-tools. Add dhcp_release6 to contrib/lease-tools. Many thanks to Sergey Nechaev for this code. To avoid filling logs in configurations which define many upstream nameservers, don't log more that 30 servers. The number to be logged can be changed as SERVERS_LOGGED in src/config.h. Swap the values if BC_EFI and x86-64_EFI in --pxe-service. These were previously wrong due to an error in RFC 4578. If you're using BC_EFI to boot 64-bit EFI machines, you will need to update your config. Add ARM32_EFI and ARM64_EFI as valid architectures in --pxe-service. Fix PXE booting for UEFI architectures. Modify PXE boot sequence in this case to force the client to talk to dnsmasq over port 4011. This makes PXE and especially proxy-DHCP PXE work with these architectures. Workaround problems with UEFI PXE clients. There exist in the wild PXE clients which have problems with PXE boot menus. To work around this, when there's a single --pxe-service which applies to client, then that target will be booted directly, rather then sending a single-item boot menu. Many thanks to Jarek Polok, Michael Kuron and Dreamcat4 for their work on the long-standing UEFI PXE problem. Subtle change in the semantics of "basename" in --pxe-service. The historical behaviour has always been that the actual filename downloaded from the TFTP server is <basename>.<layer> where <layer> is an integer which corresponds to the layer parameter supplied by the client. It's not clear what the function of the "layer" actually is in the PXE protocol, and in practise layer is always zero, so the filename is <basename>.0 The new behaviour is the same as the old, except when <basename> includes a file suffix, in which case the layer suffix is no longer added. This allows sensible suffices to be used, rather then the meaningless ".0". Only in the unlikely event that you have a config with a basename which already has a suffix, is this an incompatible change, since the file downloaded will change from name.suffix.0 to just name.suffix version 2.75 Fix reversion on 2.74 which caused 100% CPU use when a dhcp-script is configured. Thanks to Adrian Davey for reporting the bug and testing the fix. version 2.74 Fix reversion in 2.73 where --conf-file would attempt to read the default file, rather than no file. Fix inotify code to handle dangling symlinks better and not SEGV in some circumstances. DNSSEC fix. In the case of a signed CNAME generated by a wildcard which pointed to an unsigned domain, the wrong status would be logged, and some necessary checks omitted. version 2.73 Fix crash at startup when an empty suffix is supplied to --conf-dir, also trivial memory leak. Thanks to Tomas Hozza for spotting this. Remove floor of 4096 on advertised EDNS0 packet size when DNSSEC in use, the original rationale for this has long gone. Thanks to Anders Kaseorg for spotting this. Use inotify for checking on updates to /etc/resolv.conf and friends under Linux. This fixes race conditions when the files are updated rapidly and saves CPU by noy polling. To build a binary that runs on old Linux kernels without inotify, use make COPTS=-DNO_INOTIFY Fix breakage of --domain=<domain>,<subnet>,local - only reverse queries were intercepted. THis appears to have been broken since 2.69. Thanks to Josh Stone for finding the bug. Eliminate IPv6 privacy addresses and deprecated addresses from the answers given by --interface-name. Note that reverse queries (ie looking for names, given addresses) are not affected. Thanks to Michael Gorbach for the suggestion. Fix crash in DNSSEC code with long RRs. Thanks to Marco Davids for the bug report. Add --ignore-address option. Ignore replies to A-record queries which include the specified address. No error is generated, dnsmasq simply continues to listen for another reply. This is useful to defeat blocking strategies which rely on quickly supplying a forged answer to a DNS request for certain domains, before the correct answer can arrive. Thanks to Glen Huang for the patch. Revisit the part of DNSSEC validation which determines if an unsigned answer is legit, or is in some part of the DNS tree which should be signed. Dnsmasq now works from the DNS root downward looking for the limit of signed delegations, rather than working bottom up. This is both more correct, and less likely to trip over broken nameservers in the unsigned parts of the DNS tree which don't respond well to DNSSEC queries. Add --log-queries=extra option, which makes logs easier to search automatically. Add --min-cache-ttl option. I've resisted this for a long time, on the grounds that disbelieving TTLs is never a good idea, but I've been persuaded that there are sometimes reasons to do it. (Step forward, GFW). To avoid misuse, there's a hard limit on the TTL floor of one hour. Thanks to RinSatsuki for the patch. Cope with multiple interfaces with the same link-local address. (IPv6 addresses are scoped, so this is allowed.) Thanks to Cory Benfield for help with this. Add --dhcp-hostsdir. This allows addition of new host configurations to a running dnsmasq instance much more cheaply than having dnsmasq re-read all its existing configuration each time. Don't reply to DHCPv6 SOLICIT messages if we're not configured to do stateful DHCPv6. Thanks to Win King Wan for the patch. Fix broken DNSSEC validation of ECDSA signatures. Add --dnssec-timestamp option, which provides an automatic way to detect when the system time becomes valid after boot on systems without an RTC, whilst allowing DNS queries before the clock is valid so that NTP can run. Thanks to Kevin Darbyshire-Bryant for developing this idea. Add --tftp-no-fail option. Thanks to Stefan Tomanek for the patch. Fix crash caused by looking up servers.bind, CHAOS text record, when more than about five --servers= lines are in the dnsmasq config. This causes memory corruption which causes a crash later. Thanks to Matt Coddington for sterling work chasing this down. Fix crash on receipt of certain malformed DNS requests. Thanks to Nick Sampanis for spotting the problem. Note that this is could allow the dnsmasq process's memory to be read by an attacker under certain circumstances, so it has a CVE, CVE-2015-3294 Fix crash in authoritative DNS code, if a .arpa zone is declared as authoritative, and then a PTR query which is not to be treated as authoritative arrived. Normally, directly declaring .arpa zone as authoritative is not done, so this crash wouldn't be seen. Instead the relevant .arpa zone should be specified as a subnet in the auth-zone declaration. Thanks to Johnny S. Lee for the bugreport and initial patch. Fix authoritative DNS code to correctly reply to NS and SOA queries for .arpa zones for which we are declared authoritative by means of a subnet in auth-zone. Previously we provided correct answers to PTR queries in such zones (including NS and SOA) but not direct NS and SOA queries. Thanks to Johnny S. Lee for pointing out the problem. Fix logging of DHCPREPLY which should be suppressed by quiet-dhcp6. Thanks to J. Pablo Abonia for spotting the problem. Try and handle net connections with broken fragmentation that lose large UDP packets. If a server times out, reduce the maximum UDP packet size field in the EDNS0 header to 1280 bytes. If it then answers, make that change permanent. Check IPv4-mapped IPv6 addresses when --stop-rebind is active. Thanks to Jordan Milne for spotting this. Allow DHCPv4 options T1 and T2 to be set using --dhcp-option. Thanks to Kevin Benton for patches and work on this. Fix code for DHCPCONFIRM DHCPv6 messages to confirm addresses in the correct subnet, even of not in dynamic address allocation range. Thanks to Steve Hirsch for spotting the problem. Add AddDhcpLease and DeleteDhcpLease DBus methods. Thanks to Nicolas Cavallari for the patch. Allow configuration of router advertisements without the "on-link" bit set. Thanks to Neil Jerram for the patch. Extend --bridge-interface to DHCPv6 and router advertisements. Thanks to Neil Jerram for the patch. version 2.72 Add ra-advrouter mode, for RFC-3775 mobile IPv6 support. Add support for "ipsets" in *BSD, using pf. Thanks to Sven Falempin for the patch. Fix race condition which could lock up dnsmasq when an interface goes down and up rapidly. Thanks to Conrad Kostecki for helping to chase this down. Add DBus methods SetFilterWin2KOption and SetBogusPrivOption Thanks to the Smoothwall project for the patch. Fix failure to build against Nettle-3.0. Thanks to Steven Barth for spotting this and finding the fix. When assigning existing DHCP leases to interfaces by comparing networks, handle the case that two or more interfaces have the same network part, but different prefix lengths (favour the longer prefix length.) Thanks to Lung-Pin Chang for the patch. Add a mode which detects and removes DNS forwarding loops, ie a query sent to an upstream server returns as a new query to dnsmasq, and would therefore be forwarded again, resulting in a query which loops many times before being dropped. Upstream servers which loop back are disabled and this event is logged. Thanks to Smoothwall for their sponsorship of this feature. Extend --conf-dir to allow filtering of files. So --conf-dir=/etc/dnsmasq.d,\*.conf will load all the files in /etc/dnsmasq.d which end in .conf Fix bug when resulted in NXDOMAIN answers instead of NODATA in some circumstances. Fix bug which caused dnsmasq to become unresponsive if it failed to send packets due to a network interface disappearing. Thanks to Niels Peen for spotting this. Fix problem with --local-service option on big-endian platforms Thanks to Richard Genoud for the patch. version 2.71 Subtle change to error handling to help DNSSEC validation when servers fail to provide NODATA answers for non-existent DS records. Tweak code which removes DNSSEC records from answers when not required. Fixes broken answers when additional section has real records in it. Thanks to Marco Davids for the bug report. Fix DNSSEC validation of ANY queries. Thanks to Marco Davids for spotting that too. Fix total DNS failure and 100% CPU use if cachesize set to zero, regression introduced in 2.69. Thanks to James Hunt and the Ubuntu crowd for assistance in fixing this. version 2.70 Fix crash, introduced in 2.69, on TCP request when dnsmasq compiled with DNSSEC support, but running without DNSSEC enabled. Thanks to Manish Sing for spotting that one. Fix regression which broke ipset functionality. Thanks to Wang Jian for the bug report. version 2.69 Implement dynamic interface discovery on *BSD. This allows the constructor: syntax to be used in dhcp-range for DHCPv6 on the BSD platform. Thanks to Matthias Andree for valuable research on how to implement this. Fix infinite loop associated with some --bogus-nxdomain configs. Thanks fogobogo for the bug report. Fix missing RA RDNS option with configuration like --dhcp-option=option6:23,[::] Thanks to Tsachi Kimeldorfer for spotting the problem. Add [fd00::] and [fe80::] as special addresses in DHCPv6 options, analogous to [::]. [fd00::] is replaced with the actual ULA of the interface on the machine running dnsmasq, [fe80::] with the link-local address. Thanks to Tsachi Kimeldorfer for championing this. DNSSEC validation and caching. Dnsmasq needs to be compiled with this enabled, with make dnsmasq COPTS=-DHAVE_DNSSEC this adds dependencies on the nettle crypto library and the gmp maths library. It's possible to have these linked statically with make dnsmasq COPTS='-DHAVE_DNSSEC -DHAVE_DNSSEC_STATIC' which bloats the dnsmasq binary, but saves the size of the shared libraries which are much bigger. To enable, DNSSEC, you will need a set of trust-anchors. Now that the TLDs are signed, this can be the keys for the root zone, and for convenience they are included in trust-anchors.conf in the dnsmasq distribution. You should of course check that these are legitimate and up-to-date. So, adding conf-file=/path/to/trust-anchors.conf dnssec to your config is all that's needed to get things working. The upstream nameservers have to be DNSSEC-capable too, of course. Many ISP nameservers aren't, but the Google public nameservers (8.8.8.8 and 8.8.4.4) are. When DNSSEC is configured, dnsmasq validates any queries for domains which are signed. Query results which are bogus are replaced with SERVFAIL replies, and results which are correctly signed have the AD bit set. In addition, and just as importantly, dnsmasq supplies correct DNSSEC information to clients which are doing their own validation, and caches DNSKEY, DS and RRSIG records, which significantly improve the performance of downstream validators. Setting --log-queries will show DNSSEC in action. If a domain is returned from an upstream nameserver without DNSSEC signature, dnsmasq by default trusts this. This means that for unsigned zone (still the majority) there is effectively no cost for having DNSSEC enabled. Of course this allows an attacker to replace a signed record with a false unsigned record. This is addressed by the --dnssec-check-unsigned flag, which instructs dnsmasq to prove that an unsigned record is legitimate, by finding a secure proof that the zone containing the record is not signed. Doing this has costs (typically one or two extra upstream queries). It also has a nasty failure mode if dnsmasq's upstream nameservers are not DNSSEC capable. Without --dnssec-check-unsigned using such an upstream server will simply result in not queries being validated; with --dnssec-check-unsigned enabled and a DNSSEC-ignorant upstream server, _all_ queries will fail. Note that DNSSEC requires that the local time is valid and accurate, if not then DNSSEC validation will fail. NTP should be running. This presents a problem for routers without a battery-backed clock. To set the time needs NTP to do DNS lookups, but lookups will fail until NTP has run. To address this, there's a flag, --dnssec-no-timecheck which disables the time checks (only) in DNSSEC. When dnsmasq is started and the clock is not synced, this flag should be used. As soon as the clock is synced, SIGHUP dnsmasq. The SIGHUP clears the cache of partially-validated data and resets the no-timecheck flag, so that all DNSSEC checks henceforward will be complete. The development of DNSSEC in dnsmasq was started by Giovanni Bajo, to whom huge thanks are owed. It has been supported by Comcast, whose techfund grant has allowed for an invaluable period of full-time work to get it to a workable state. Add --rev-server. Thanks to Dave Taht for suggesting this. Add --servers-file. Allows dynamic update of upstream servers full access to configuration. Add --local-service. Accept DNS queries only from hosts whose address is on a local subnet, ie a subnet for which an interface exists on the server. This option only has effect if there are no --interface --except-interface, --listen-address or --auth-server options. It is intended to be set as a default on installation, to allow unconfigured installations to be useful but also safe from being used for DNS amplification attacks. Fix crashes in cache_get_cname_target() when dangling CNAMEs encountered. Thanks to Andy and the rt-n56u project for find this and helping to chase it down. Fix wrong RCODE in authoritative DNS replies to PTR queries. The correct answer was included, but the RCODE was set to NXDOMAIN. Thanks to Craig McQueen for spotting this. Make statistics available as DNS queries in the .bind TLD as well as logging them. version 2.68 Use random addresses for DHCPv6 temporary address allocations, instead of algorithmically determined stable addresses. Fix bug which meant that the DHCPv6 DUID was not available in DHCP script runs during the lifetime of the dnsmasq process which created the DUID de-novo. Once the DUID was created and stored in the lease file and dnsmasq restarted, this bug disappeared. Fix bug introduced in 2.67 which could result in erroneous NXDOMAIN returns to CNAME queries. Fix build failures on MacOS X and openBSD. Allow subnet specifications in --auth-zone to be interface names as well as address literals. This makes it possible to configure authoritative DNS when local address ranges are dynamic and works much better than the previous work-around which exempted constructed DHCP ranges from the IP address filtering. As a consequence, that work-around is removed. Under certain circumstances, this change wil break existing configuration: if you're relying on the constructed-range exception, you need to change --auth-zone to specify the same interface as is used to construct your DHCP ranges, probably with a trailing "/6" like this: --auth-zone=example.com,eth0/6 to limit the addresses to IPv6 addresses of eth0. Fix problems when advertising deleted IPv6 prefixes. If the prefix is deleted (rather than replaced), it doesn't get advertised with zero preferred time. Thanks to Tsachi for the bug report. Fix segfault with some locally configured CNAMEs. Thanks to Andrew Childs for spotting the problem. Fix memory leak on re-reading /etc/hosts and friends, introduced in 2.67. Check the arrival interface of incoming DNS and TFTP requests via IPv6, even in --bind-interfaces mode. This isn't possible for IPv4 and can generate scary warnings, but as it's always possible for IPv6 (the API always exists) then we should do it always. Tweak the rules on prefix-lengths in --dhcp-range for IPv6. The new rule is that the specified prefix length must be larger than or equal to the prefix length of the corresponding address on the local interface. version 2.67 Fix crash if upstream server returns SERVFAIL when --conntrack in use. Thanks to Giacomo Tazzari for finding this and supplying the patch. Repair regression in 2.64. That release stopped sending lease-time information in the reply to DHCPINFORM requests, on the correct grounds that it was a standards violation. However, this broke the dnsmasq-specific dhcp_lease_time utility. Now, DHCPINFORM returns lease-time only if it's specifically requested (maintaining standards) and the dhcp_lease_time utility has been taught to ask for it (restoring functionality). Fix --dhcp-match, --dhcp-vendorclass and --dhcp-userclass to work with BOOTP and well as DHCP. Thanks to Peter Korsgaard for spotting the problem. Add --synth-domain. Thanks to Vishvananda Ishaya for suggesting this. Fix failure to compile ipset.c if old kernel headers are in use. Thanks to Eugene Rudoy for pointing this out. Handle IPv4 interface-address labels in Linux. These are often used to emulate the old IP-alias addresses. Before, using --interface=eth0 would service all the addresses of eth0, including ones configured as aliases, which appear in ifconfig as eth0:0. Now, only addresses with the label eth0 are active. This is not backwards compatible: if you want to continue to bind the aliases too, you need to add eg. --interface=eth0:0 to the config. Fix "failed to set SO_BINDTODEVICE on DHCP socket: Socket operation on non-socket" error on startup with configurations which have exactly one --interface option and do RA but _not_ DHCPv6. Thanks to Trever Adams for the bug report. Generalise --interface-name to cope with IPv6 addresses and multiple addresses per interface per address family. Fix option parsing for --dhcp-host, which was generating a spurious error when all seven possible items were included. Thanks to Zhiqiang Wang for the bug report. Remove restriction on prefix-length in --auth-zone. Thanks to Toke Hoiland-Jorgensen for suggesting this. Log when the maximum number of concurrent DNS queries is reached. Thanks to Marcelo Salhab Brogliato for the patch. If wildcards are used in --interface, don't assume that there will only ever be one available interface for DHCP just because there is one at start-up. More may appear, so we can't use SO_BINDTODEVICE. Thanks to Natrio for the bug report. Increase timeout/number of retries in TFTP to accommodate AudioCodes Voice Gateways doing streaming writes to flash. Thanks to Damian Kaczkowski for spotting the problem. Fix crash with empty DHCP string options when adding zero terminator. Thanks to Patrick McLean for the bug report. Allow hostnames to start with a number, as allowed in RFC-1123. Thanks to Kyle Mestery for the patch. Fixes to DHCP FQDN option handling: don't terminate FQDN if domain not known and allow a FQDN option with blank name to request that a FQDN option is returned in the reply. Thanks to Roy Marples for the patch. Make --clear-on-reload apply to setting upstream servers via DBus too. When the address which triggered the construction of an advertised IPv6 prefix disappears, continue to advertise the prefix for up to 2 hours, with the preferred lifetime set to zero. This satisfies RFC 6204 4.3 L-13 and makes things work better if a prefix disappears without being deprecated first. Thanks to Uwe Schindler for persuasively arguing for this. Fix MAC address enumeration on *BSD. Thanks to Brad Smith for the bug report. Support RFC-4242 information-refresh-time options in the reply to DHCPv6 information-request. The lease time of the smallest valid dhcp-range is sent. Thanks to Uwe Schindler for suggesting this. Make --listen-address higher priority than --except-interface in all circumstances. Thanks to Thomas Hood for the bugreport. Provide independent control over which interfaces get TFTP service. If enable-tftp is given a list of interfaces, then TFTP is provided on those. Without the list, the previous behaviour (provide TFTP to the same interfaces we provide DHCP to) is retained. Thanks to Lonnie Abelbeck for the suggestion. Add --dhcp-relay config option. Many thanks to vtsl.net for sponsoring this development. Fix crash with empty tag: in --dhcp-range. Thanks to Kaspar Schleiser for the bug report. Add "baseline" and "bloatcheck" makefile targets, for revealing size changes during development. Thanks to Vladislav Grishenko for the patch. Cope with DHCPv6 clients which send REQUESTs without address options - treat them as SOLICIT with rapid commit. Support identification of clients by MAC address in DHCPv6. When using a relay, the relay must support RFC 6939 for this to work. It always works for directly connected clients. Thanks to Vladislav Grishenko for prompting this feature. Remove the rule for constructed DHCP ranges that the local address must be either the first or last address in the range. This was originally to avoid SLAAC addresses, but we now explicitly autoconfig and privacy addresses instead. Update Polish translation. Thanks to Jan Psota. Fix problem in DHCPv6 vendorclass/userclass matching code. Thanks to Tanguy Bouzeloc for the patch. Update Spanish translation. Thanks to Vicente Soriano. Add --ra-param option. Thanks to Vladislav Grishenko for inspiration on this. Add --add-subnet configuration, to tell upstream DNS servers where the original client is. Thanks to DNSthingy for sponsoring this feature. Add --quiet-dhcp, --quiet-dhcp6 and --quiet-ra. Thanks to Kevin Darbyshire-Bryant for the initial patch. Allow A/AAAA records created by --interface-name to be the target of --cname. Thanks to Hadmut Danisch for the suggestion. Avoid treating a --dhcp-host which has an IPv6 address as eligible for use with DHCPv4 on the grounds that it has no address, and vice-versa. Thanks to Yury Konovalov for spotting the problem. Do a better job caching dangling CNAMEs. Thanks to Yves Dorfsman for spotting the problem. version 2.66 Add the ability to act as an authoritative DNS server. Dnsmasq can now answer queries from the wider 'net with local data, as long as the correct NS records are set up. Only local data is provided, to avoid creating an open DNS relay. Zone transfer is supported, to allow secondary servers to be configured. Add "constructed DHCP ranges" for DHCPv6. This is intended for IPv6 routers which get prefixes dynamically via prefix delegation. With suitable configuration, stateful DHCPv6 and RA can happen automatically as prefixes are delegated and then deprecated, without having to re-write the dnsmasq configuration file or restart the daemon. Thanks to Steven Barth for extensive testing and development work on this idea. Fix crash on startup on Solaris 11. Regression probably introduced in 2.61. Thanks to Geoff Johnstone for the patch. Add code to make behaviour for TCP DNS requests that same as for UDP requests, when a request arrives for an allowed address, but via a banned interface. This change is only active on Linux, since the relevant API is missing (AFAIK) on other platforms. Many thanks to Tomas Hozza for spotting the problem, and doing invaluable discovery of the obscure and undocumented API required for the solution. Don't send the default DHCP option advertising dnsmasq as the local DNS server if dnsmasq is configured to not act as DNS server, or it's configured to a non-standard port. Add DNSMASQ_CIRCUIT_ID, DNSMASQ_SUBSCRIBER_ID, DNSMASQ_REMOTE_ID variables to the environment of the lease-change script (and the corresponding Lua). These hold information inserted into the DHCP request by a DHCP relay agent. Thanks to Lakefield Communications for providing a bounty for this addition. Fixed crash, introduced in 2.64, whilst handling DHCPv6 information-requests with some common configurations. Thanks to Robert M. Albrecht for the bug report and chasing the problem. Add --ipset option. Thanks to Jason A. Donenfeld for the patch. Don't erroneously reject some option names in --dhcp-match options. Thanks to Benedikt Hochstrasser for the bug report. Allow a trailing '*' wildcard in all interface-name configurations. Thanks to Christian Parpart for the patch. Handle the situation where libc headers define SO_REUSEPORT, but the kernel in use doesn't, to cope with the introduction of this option to Linux. Thanks to Rich Felker for the bug report. Update Polish translation. Thanks to Jan Psota. Fix crash if the configured DHCP lease limit is reached. Regression occurred in 2.61. Thanks to Tsachi for the bug report. Update the French translation. Thanks to Gildas le Nadan. version 2.65 Fix regression which broke forwarding of queries sent via TCP which are not for A and AAAA and which were directed to non-default servers. Thanks to Niax for the bug report. Fix failure to build with DHCP support excluded. Thanks to Gustavo Zacarias for the patch. Fix nasty regression in 2.64 which completely broke caching. version 2.64 Handle DHCP FQDN options with all flag bits zero and --dhcp-client-update set. Thanks to Bernd Krumbroeck for spotting the problem. Finesse the check for /etc/hosts names which conflict with DHCP names. Previously a name/address pair in /etc/hosts which didn't match the name/address of a DHCP lease would generate a warning. Now that only happens if there is not also a match. This allows multiple addresses for a name in /etc/hosts with one of them assigned via DHCP. Fix broken vendor-option processing for BOOTP. Thanks to Hans-Joachim Baader for the bug report. Don't report spurious netlink errors, regression in 2.63. Thanks to Vladislav Grishenko for the patch. Flag DHCP or DHCPv6 in startup logging. Thanks to Vladislav Grishenko for the patch. Add SetServersEx method in DBus interface. Thanks to Dan Williams for the patch. Add SetDomainServers method in DBus interface. Thanks to Roy Marples for the patch. Fix build with later Lua libraries. Thanks to Cristian Rodriguez for the patch. Add --max-cache-ttl option. Thanks to Dennis Kaarsemaker for the patch. Fix breakage of --host-record parsing, resulting in infinite loop at startup. Regression in 2.63. Thanks to Haim Gelfenbeyn for spotting this. Set SO_REUSEADDRESS and SO_V6ONLY options on the DHCPv6 socket, this allows multiple instances of dnsmasq on a single machine, in the same way as for DHCPv4. Thanks to Gene Czarcinski and Vladislav Grishenko for work on this. Fix DHCPv6 to do access control correctly when it's configured with --listen-address. Thanks to Gene Czarcinski for sorting this out. Add a "wildcard" dhcp-range which works for any IPv6 subnet, --dhcp-range=::,static Useful for Stateless DHCPv6. Thanks to Vladislav Grishenko for the patch. Don't include lease-time in DHCPACK replies to DHCPINFORM queries, since RFC-2131 says we shouldn't. Thanks to Wouter Ibens for pointing this out. Makefile tweak to do dependency checking on header files. Thanks to Johan Peeters for the patch. Check interface for outgoing unsolicited router advertisements, rather than relying on interface address configuration. Thanks to Gene Czarinski for the patch. Handle better attempts to transmit on interfaces which are still doing DAD, and specifically do not just transmit without setting source address and interface, since this can cause very puzzling effects when a router advertisement goes astray. Thanks again to Gene Czarinski. Get RA timers right when there is more than one dhcp-range on a subnet. version 2.63 Do duplicate dhcp-host address check in --test mode. Check that tftp-root directories are accessible before start-up. Thanks to Daniel Veillard for the initial patch. Allow more than one --tfp-root flag. The per-interface stuff is pointless without that. Add --bind-dynamic. A hybrid mode between the default and --bind-interfaces which copes with dynamically created interfaces. A couple of fixes to the build system for Android. Thanks to Metin Kaya for the patches. Remove the interface:<interface> argument in --dhcp-range, and the interface argument to --enable-tftp. These were a still-born attempt to allow automatic isolated configuration by libvirt, but have never (to my knowledge) been used, had very strange semantics, and have been superseded by other mechanisms. Fixed bug logging filenames when duplicate dhcp-host addresses are found. Thanks to John Hanks for the patch. Fix regression in 2.61 which broke caching of CNAME chains. Thanks to Atul Gupta for the bug report. Allow the target of a --cname flag to be another --cname. Teach DHCPv6 about the RFC 4242 information-refresh-time option, and add parsing if the minutes, hours and days format for options. Thanks to Francois-Xavier Le Bail for the suggestion. Allow "w" (for week) as multiplier in lease times, as well as seconds, minutes, hours and days. Álvaro Gámez Machado spotted the omission. Update French translation. Thanks to Gildas Le Nadan. Allow a DBus service name to be given with --enable-dbus which overrides the default, uk.org.thekelleys.dnsmasq. Thanks to Mathieu Trudel-Lapierre for the patch. Set the "prefix on-link" bit in Router Advertisements. Thanks to Gui Iribarren for the patch. version 2.62 Update German translation. Thanks to Conrad Kostecki. Cope with router-solict packets which don't have a valid source address. Thanks to Vladislav Grishenko for the patch. Fixed bug which caused missing periodic router advertisements with some configurations. Thanks to Vladislav Grishenko for the patch. Fixed bug which broke DHCPv6/RA with prefix lengths which are not divisible by 8. Thanks to Andre Coetzee for spotting this. Fix non-response to router-solicitations when router-advertisement configured, but DHCPv6 not configured. Thanks to Marien Zwart for the patch. Add --dns-rr, to allow arbitrary DNS resource records. Fixed bug which broke RA scheduling when an interface had two addresses in the same network. Thanks to Jim Bos for his help nailing this. version 2.61 Re-write interface discovery code on *BSD to use getifaddrs. This is more portable, more straightforward, and allows us to find the prefix length for IPv6 addresses. Add ra-names, ra-stateless and slaac keywords for DHCPv6. Dnsmasq can now synthesise AAAA records for dual-stack hosts which get IPv6 addresses via SLAAC. It is also now possible to use SLAAC and stateless DHCPv6, and to tell clients to use SLAAC addresses as well as DHCP ones. Thanks to Dave Taht for help with this. Add --dhcp-duid to allow DUID-EN uids to be used. Explicitly send DHCPv6 replies to the correct port, instead of relying on clients to send requests with the correct source address, since at least one client in the wild gets this wrong. Thanks to Conrad Kostecki for help tracking this down. Send a preference value of 255 in DHCPv6 replies when --dhcp-authoritative is in effect. This tells clients not to wait around for other DHCP servers. Better logging of DHCPv6 options. Add --host-record. Thanks to Rob Zwissler for the suggestion. Invoke the DHCP script with action "tftp" when a TFTP file transfer completes. The size of the file, address to which it was sent and complete pathname are supplied. Note that version 2.60 introduced some script incompatibilities associated with DHCPv6, and this is a further change. To be safe, scripts should ignore unknown actions, and if not IPv6-aware, should exit if the environment variable DNSMASQ_IAID is set. The use-case for this is to track netboot/install. Suggestion from Shantanu Gadgil. Update contrib/port-forward/dnsmasq-portforward to reflect the above. Set the environment variable DNSMASQ_LOG_DHCP when running the script id --log-dhcp is in effect, so that script can taylor their logging verbosity. Suggestion from Malte Forkel. Arrange that addresses specified with --listen-address work even if there is no interface carrying the address. This is chiefly useful for IPv4 loopback addresses, where any address in 127.0.0.0/8 is a valid loopback address, but normally only 127.0.0.1 appears on the lo interface. Thanks to Mathieu Trudel-Lapierre for the idea and initial patch. Fix crash, introduced in 2.60, when a DHCPINFORM is received from a network which has no valid dhcp-range. Thanks to Stephane Glondu for the bug report. Add a new DHCP lease time keyword, "deprecated" for --dhcp-range. This is only valid for IPv6, and sets the preferred lease time for both DHCP and RA to zero. The effect is that clients can continue to use the address for existing connections, but new connections will use other addresses, if they exist. This makes hitless renumbering at least possible. Fix bug in address6_available() which caused DHCPv6 lease acquisition to fail if more than one dhcp-range in use. Provide RDNSS and DNSSL data in router advertisements, using the settings provided for DHCP options option6:domain-search and option6:dns-server. Tweak logo/favicon.ico to add some transparency. Thanks to SamLT for work on this. Don't cache data from non-recursive nameservers, since it may erroneously look like a valid CNAME to a non-existent name. Thanks to Ben Winslow for finding this. Call SO_BINDTODEVICE on the DHCP socket(s) when doing DHCP on exactly one interface and --bind-interfaces is set. This makes the OpenStack use-case of one dnsmasq per virtual interface work. This is only available on Linux; it's not supported on other platforms. Thanks to Vishvananda Ishaya and the OpenStack team for the suggestion. Updated French translation. Thanks to Gildas Le Nadan. Give correct from-cache answers to explicit CNAME queries. Thanks to Rob Zwissler for spotting this. Add --tftp-lowercase option. Thanks to Oliver Rath for the patch. Ensure that the DBus DhcpLeaseUpdated events are generated when a lease goes through INIT_REBOOT state, even if the dhcp-script is not in use. Thanks to Antoaneta-Ecaterina Ene for the patch. Fix failure of TFTP over IPv4 on OpenBSD platform. Thanks to Brad Smith for spotting this. version 2.60 Fix compilation problem in Mac OS X Lion. Thanks to Olaf Flebbe for the patch. Fix DHCP when using --listen-address with an IP address which is not the primary address of an interface. Add --dhcp-client-update option. Add Lua integration. Dnsmasq can now execute a DHCP lease-change script written in Lua. This needs to be enabled at compile time by setting HAVE_LUASCRIPT in src/config.h or running "make COPTS=-DHAVE_LUASCRIPT" Thanks to Jan-Piet Mens for the idea and proof-of-concept implementation. Tidied src/config.h to distinguish between platform-dependent compile-time options which are selected automatically, and builder-selectable compile time options. Document the latter better, and describe how to set them from the make command line. Tidied up IPPROTO_IP/SOL_IP (and IPv6 equivalent) confusion. IPPROTO_IP works everywhere now. Set TOS on DHCP sockets, this improves things on busy wireless networks. Thanks to Dave Taht for the patch. Determine VERSION automatically based on git magic: release tags or hash values. Improve start-up speed when reading large hosts files containing many distinct addresses. Fix problem if dnsmasq is started without the stdin, stdout and stderr file descriptors open. This can manifest itself as 100% CPU use. Thanks to Chris Moore for finding this. Fix shell-scripting bug in bld/pkg-wrapper. Thanks to Mark Mitchell for the patch. Allow the TFP server or boot server in --pxe-service, to be a domain name instead of an IP address. This allows for round-robin to multiple servers, in the same way as --dhcp-boot. A good suggestion from Cristiano Cumer. Support BUILDDIR variable in the Makefile. Allows builds for multiple archs from the same source tree with eg. make BUILDDIR=linux (relative to dnsmasq tree) make BUILDDIR=/tmp/openbsd (absolute path) If BUILDDIR is not set, compilation happens in the src directory, as before. Suggestion from Mark Mitchell. Support DHCPv6. Support is there for the sort of things the existing v4 server does, including tags, options, static addresses and relay support. Missing is prefix delegation, which is probably not required in the dnsmasq niche, and an easy way to accept prefix delegations from an upstream DHCPv6 server, which is. Future plans include support for DHCPv6 router option and MAC address option (to make selecting clients by MAC address work like IPv4). These will be added as the standards mature. This code has been tested, but this is the first release, so don't bet the farm on it just yet. Many thanks to all testers who have got it this far. Support IPv6 router advertisements. This is a simple-minded implementation, aimed at providing the vestigial RA needed to go alongside IPv6. Is picks up configuration from the DHCPv6 conf, and should just need enabling with --enable-ra. Fix long-standing wrinkle with --localise-queries that could result in wrong answers when DNS packets arrive via an interface other than the expected one. Thanks to Lorenzo Milesi and John Hanks for spotting this one. Update French translation. Thanks to Gildas Le Nadan. Update Polish translation. Thanks to Jan Psota. version 2.59 Fix regression in 2.58 which caused failure to start up with some combinations of dnsmasq config and IPv6 kernel network config. Thanks to Brielle Bruns for the bug report. Improve dnsmasq's behaviour when network interfaces are still doing duplicate address detection (DAD). Previously, dnsmasq would wait up to 20 seconds at start-up for the DAD state to terminate. This is broken for bridge interfaces on recent Linux kernels, which don't start DAD until the bridge comes up, and so can take arbitrary time. The new behaviour lets dnsmasq poll for an arbitrary time whilst providing service on other interfaces. Thanks to Stephen Hemminger for pointing out the problem. version 2.58 Provide a definition of the SA_SIZE macro where it's missing. Fixes build failure on openBSD. Don't include a zero terminator at the end of messages sent to /dev/log when /dev/log is a datagram socket. Thanks to Didier Rabound for spotting the problem. Add --dhcp-sequential-ip flag, to force allocation of IP addresses in ascending order. Note that the default pseudo-random mode is in general better but some server-deployment applications need this. Fix problem where a server-id of 0.0.0.0 is sent to a client when a dhcp-relay is in use if a client renews a lease after dnsmasq restart and before any clients on the subnet get a new lease. Thanks to Mike Ruiz for assistance in chasing this one down. Don't return NXDOMAIN to an AAAA query if we have CNAME which points to an A record only: NODATA is the correct reply in this case. Thanks to Tom Fernandes for spotting the problem. Relax the need to supply a netmask in --dhcp-range for networks which use a DHCP relay. Whilst this is still desirable, in the absence of a netmask dnsmasq will use a default based on the class (A, B, or C) of the address. This should at least remove a cause of mysterious failure for people using RFC1918 addresses and relays. Add support for Linux conntrack connection marking. If enabled with --conntrack, the connection mark for incoming DNS queries will be copied to the outgoing connections used to answer those queries. This allows clever firewall and accounting stuff. Only available if dnsmasq is compiled with HAVE_CONNTRACK and adds a dependency on libnetfilter-conntrack. Thanks to Ed Wildgoose for the initial idea, testing and sponsorship of this function. Provide a sane error message when someone attempts to match a tag in --dhcp-host. Tweak the behaviour of --domain-needed, to avoid problems with recursive nameservers downstream of dnsmasq. The new behaviour only stops A and AAAA queries, and returns NODATA rather than NXDOMAIN replies. Efficiency fix for very large DHCP configurations, thanks to James Gartrell and Mike Ruiz for help with this. Allow the TFTP-server address in --dhcp-boot to be a domain-name which is looked up in /etc/hosts. This can give multiple IP addresses which are used round-robin, thus doing TFTP server load-balancing. Thanks to Sushil Agrawal for the patch. When two tagged dhcp-options for a particular option number are both valid, use the one which is valid without a tag from the dhcp-range. Allows overriding of the value of a DHCP option for a particular host as well as per-network values. So --dhcp-range=set:interface1,...... --dhcp-host=set:myhost,..... --dhcp-option=tag:interface1,option:nis-domain,"domain1" --dhcp-option=tag:myhost,option:nis-domain,"domain2" will set the NIS-domain to domain1 for hosts in the range, but override that to domain2 for a particular host. Fix bug which resulted in truncated files and timeouts for some TFTP transfers. The bug only occurs with netascii transfers and needs an unfortunate relationship between file size, blocksize and the number of newlines in the last block before it manifests itself. Many thanks to Alkis Georgopoulos for spotting the problem and providing a comprehensive test-case. Fix regression in TFTP server on *BSD platforms introduced in version 2.56, due to confusion with sockaddr length. Many thanks to Loic Pefferkorn for finding this. Support scope-ids in IPv6 addresses of nameservers from /etc/resolv.conf and in --server options. Eg nameserver fe80::202:a412:4512:7bbf%eth0 or server=fe80::202:a412:4512:7bbf%eth0. Thanks to Michael Stapelberg for the suggestion. Update Polish translation, thanks to Jan Psota. Update French translation. Thanks to Gildas Le Nadan. version 2.57 Add patches to allow build under Android. Provide our own header for the DNS protocol, rather than relying on arpa/nameser.h. This has proved more or less defective over the years and the final straw is that it's effectively empty on Android. Fix regression in 2.56 which caused hex constants in configuration to be rejected if they contain the '*' wildcard. Correct wrong casts of arguments to ctype.h functions, isdigit(), isxdigit() etc. Thanks to Matthias Andree for spotting this. Allow build with IDN support independently from i18n. IDN support continues to be included automatically when i18n is included. 'make COPTS=-DHAVE_IDN' is the magic incantation. Modify check on extraneous command line junk (added in 2.56) so that it doesn't complain about extra _empty_ arguments. Otherwise this breaks libvirt. version 2.56 Add a patch to allow dnsmasq to get interface names right in a Solaris zone. Thanks to Dj Padzensky for this. Improve data-type parsing heuristics so that --dhcp-option=option:domain-search,. treats the value as a string and not an IP address. Thanks to Clemens Fischer for spotting that. Add IPv6 support to the TFTP server. Many thanks to Jan 'RedBully' Seiffert for the patches. Log DNS queries at level LOG_INFO, rather then LOG_DEBUG. This makes things consistent with DHCP logging. Thanks to Adam Pribyl for spotting the problem. Ensure that dnsmasq terminates cleanly when using --syslog-async even if it cannot make a connection to the syslogd. Add --add-mac option. This is to support currently experimental DNS filtering facilities. Thanks to Benjamin Petrin for the original patch. Fix bug which meant that tags were ignored in dhcp-range configuration specifying PXE-proxy service. Thanks to Cristiano Cumer for spotting this. Raise an error if there is extra junk, not part of an option, on the command line. Flag a couple of log messages in cache.c as coming from the DHCP subsystem. Thanks to Olaf Westrik for the patch. Omit timestamps from logs when a) logging to stderr and b) --keep-in-foreground is set. The logging facility on the other end of stderr can be assumed to supply them. Thanks to John Hallam for the patch. Don't complain about strings longer than 255 characters in --txt-record, just split the long strings into 255 character chunks instead. Fix crash on double-free. This bug can only happen when dhcp-script is in use and then only in rare circumstances triggered by high DHCP transaction rate and a slow script. Thanks to Ferenc Wagner for finding the problem. Only log that a file has been sent by TFTP after the transfer has completed successfully. A good suggestion from Ferenc Wagner: extend the --domain option to allow this sort of thing: --domain=thekelleys.org.uk,192.168.0.0/24,local which automatically creates --local=/thekelleys.org.uk/ --local=/0.168.192.in-addr.arpa/ Tighten up syntax checking of hex constants in the config file. Thanks to Fred Damen for spotting this. Add dnsmasq logo/icon, contributed by Justin Swift. Many thanks for that. Never cache DNS replies which have the 'cd' bit set, or which result from queries forwarded with the 'cd' bit set. The 'cd' bit instructs a DNSSEC validating server upstream to ignore signature failures and return replies anyway. Without this change it's possible to pollute the dnsmasq cache with bad data by making a query with the 'cd' bit set and subsequent queries would return this data without its being marked as suspect. Thanks to Anders Kaseorg for pointing out this problem. Add --proxy-dnssec flag, for compliance with RFC 4035. Dnsmasq will now clear the 'ad' bit in answers returned from upstream validating nameservers unless this option is set. Allow a filename of "-" for --conf-file to read stdin. Suggestion from Timothy Redaelli. Rotate the order of SRV records in replies, to provide round-robin load balancing when all the priorities are equal. Thanks to Peter McKinney for the suggestion. Edit contrib/MacOSX-launchd/uk.org.thekelleys.dnsmasq.plist so that it doesn't log all queries to a file by default. Thanks again to Peter McKinney. By default, setting an IPv4 address for a domain but not an IPv6 address causes dnsmasq to return a NODATA reply for IPv6 (or vice-versa). So --address=/google.com/1.2.3.4 stops IPv6 queries for *google.com from being forwarded. Make it possible to override this behaviour by defining the semantics if the same domain appears in both --server and --address. In that case, the --address has priority for the address family in which is appears, but the --server has priority of the address family which doesn't appear in --address So: --address=/google.com/1.2.3.4 --server=/google.com/# will return 1.2.3.4 for IPv4 queries for *.google.com but forward IPv6 queries to the normal upstream nameserver. Similarly when setting an IPv6 address only this will allow forwarding of IPv4 queries. Thanks to William for pointing out the need for this. Allow more than one --dhcp-optsfile and --dhcp-hostsfile and make them understand directories as arguments in the same way as --addn-hosts. Suggestion from John Hanks. Ignore rebinding requests for leases we don't know about. Rebind is broadcast, so we might get to overhear a request meant for another DHCP server. NAKing this is wrong. Thanks to Brad D'Hondt for assistance with this. Fix cosmetic bug which produced strange output when dumping cache statistics with some configurations. Thanks to Fedor Kozhevnikov for spotting this. version 2.55 Fix crash when /etc/ethers is in use. Thanks to Gianluigi Tiesi for finding this. Fix crash in netlink_multicast(). Thanks to Arno Wald for finding this one. Allow the empty domain "." in dhcp domain-search (119) options. version 2.54 There is no version 2.54 to avoid confusion with 2.53, which incorrectly identifies itself as 2.54. version 2.53 Fix failure to compile on Debian/kFreeBSD. Thanks to Axel Beckert and Petr Salinger. Fix code to avoid scary strict-aliasing warnings generated by gcc 4.4. Added FAQ entry warning about DHCP failures with Vista when firewalls block 255.255.255.255. Fixed bug which caused bad things to happen if a resolv.conf file which exists is subsequently removed. Thanks to Nikolai Saoukh for the patch. Rationalised the DHCP tag system. Every configuration item which can set a tag does so by adding "set:<tag>" and every configuration item which is conditional on a tag is made so by "tag:<tag>". The NOT operator changes to '!', which is a bit more intuitive too. Dhcp-host directives can set more than one tag now. The old '#' NOT, "net:" prefix and no-prefixes are still honoured, so no existing config file needs to be changed, but the documentation and new-style config files should be much less confusing. Added --tag-if to allow boolean operations on tags. This allows complicated logic to be clearer and more general. A great suggestion from Richard Voigt. Add broadcast/unicast information to DHCP logging. Allow --dhcp-broadcast to be unconditional. Fixed incorrect behaviour with NOT <tag> conditionals in dhcp-options. Thanks to Max Turkewitz for assistance finding this. If we send vendor-class encapsulated options based on the vendor-class supplied by the client, and no explicit vendor-class option is given, echo back the vendor-class from the client. Fix bug which stopped dnsmasq from matching both a circuitid and a remoteid. Thanks to Ignacio Bravo for finding this. Add --dhcp-proxy, which makes it possible to configure dnsmasq to use a DHCP relay agent as a full proxy, with all DHCP messages passing through the proxy. This is useful if the relay adds extra information to the packets it forwards, but cannot be configured with the RFC 5107 server-override option. Added interface:<iface name> part to dhcp-range. The semantics of this are very odd at first sight, but it allows a single line of the form dhcp-range=interface:virt0,192.168.0.4,192.168.0.200 to be added to dnsmasq configuration which then supplies DHCP and DNS services to that interface, without affecting what services are supplied to other interfaces and irrespective of the existence or lack of interface=<interface> lines elsewhere in the dnsmasq configuration. The idea is that such a line can be added automatically by libvirt or equivalent systems, without disturbing any manual configuration. Similarly to the above, allow --enable-tftp=<interface> Allow a TFTP root to be set separately for requests via different interfaces, --tftp-root=<path>,<interface> Correctly handle and log clashes between CNAMES and DNS names being given to DHCP leases. This fixes a bug which caused nonsense IP addresses to be logged. Thanks to Sergei Zhirikov for finding and analysing the problem. Tweak flush_log so as to avoid leaving the log file in non-blocking mode. O_NONBLOCK is a property of the file, not the process/descriptor. Fix contrib/Solaris10/create_package (/usr/man -> /usr/share/man) Thanks to Vita Batrla. Fix a problem where, if a client got a lease, then went to another subnet and got another lease, then moved back, it couldn't resume the old lease, but would instead get a new address. Thanks to Leonardo Rodrigues for spotting this and testing the fix. Fix weird bug which sometimes omitted certain characters from the start of quoted strings in dhcp-options. Thanks to Dayton Turner for spotting the problem. Add facility to redirect some domains to the standard upstream servers: this allows something like --server=/google.com/1.2.3.4 --server=/www.google.com/# which will send queries for *.google.com to 1.2.3.4, except *www.google.com which will be forwarded as usual. Thanks to AJ Weber for prompting this addition. Improve the hash-algorithm used to generate IP addresses from MAC addresses during initial DHCP address allocation. This improves performance when large numbers of hosts with similar MAC addresses all try and get an IP address at the same time. Thanks to Paul Smith for his work on this. Tweak DHCP code so that --bridge-interface can be used to select which IP alias of an interface should be used for DHCP purposes on Linux. If eth0 has an alias eth0:dhcp then adding --bridge-interface=eth0:dhcp,eth0 will use the address of eth0:dhcp to determine the correct subnet for DHCP address allocation. Thanks to Pawel Golaszewski for prompting this and Eric Cooper for further testing. Add --dhcp-generate-names. Suggestion by Ferenc Wagner. Tweak DNS server selection algorithm when there is more than one server available for a domain, eg. --server=/mydomain/1.1.1.1 --server=/mydomain/2.2.2.2 Thanks to Alberto Cuesta-Canada for spotting a weakness here. Add --max-ttl. Thanks to Fredrik Ringertz for the patch. Allow --log-facility=- to force all logging to stderr. Suggestion from Clemens Fischer. Fix regression which caused configuration like --address=/.domain.com/1.2.3.4 to be rejected. The dot to the left of the domain has been implied and not required for a long time, but it should be accepted for backward compatibility. Thanks to Andrew Burcin for spotting this. Add --rebind-domain-ok and --rebind-localhost-ok. Suggestion from Clemens Fischer. Log replies to queries of type TXT, when --log-queries is set. Fix compiler warnings when compiled with -DNO_DHCP. Thanks to Shantanu Gadgil for the patch. Updated French translation. Thanks to Gildas Le Nadan. Updated Polish translation. Thanks to Jan Psota. Updated German translation. Thanks to Matthias Andree. Added contrib/static-arp, thanks to Darren Hoo. Fix corruption of the domain when a name from /etc/hosts overrides one supplied by a DHCP client. Thanks to Fedor Kozhevnikov for spotting the problem. Updated Spanish translation. Thanks to Chris Chatham. version 2.52 Work around a Linux kernel bug which insists that the length of the option passed to setsockopt must be at least sizeof(int) bytes, even if we're calling SO_BINDTODEVICE and the device name is "lo". Note that this is fixed in kernel 2.6.31, but the workaround is harmless and allows earlier kernels to be used. Also fix dnsmasq bug which reported the wrong address when this failed. Thanks to Fedor for finding this. The API for IPv6 PKTINFO changed around Linux kernel 2.6.14. Workaround the case where dnsmasq is compiled against newer headers, but then run on an old kernel: necessary for some *WRT distros. Re-read the set of network interfaces when re-loading /etc/resolv.conf if --bind-interfaces is not set. This handles the case that loopback interfaces do not exist when dnsmasq is first started. Tweak the PXE code to support port 4011. This should reduce broadcasts and make things more reliable when other servers are around. It also improves inter-operability with certain clients. Make a pxe-service configuration with no filename or boot service type legal: this does a local boot. eg. pxe-service=x86PC, "Local boot" Be more conservative in detecting "A for A" queries. Dnsmasq checks if the name in a type=A query looks like a dotted-quad IP address and answers the query itself if so, rather than forwarding it. Previously dnsmasq relied in the library function inet_addr() to convert addresses, and that will accept some things which are confusing in this context, like 1.2.3 or even just 1234. Now we only do A for A processing for four decimal numbers delimited by dots. A couple of tweaks to fix compilation on Solaris. Thanks to Joel Macklow for help with this. Another Solaris compilation tweak, needed for Solaris 2009.06. Thanks to Lee Essen for that. Added extract packaging stuff from Lee Essen to contrib/Solaris10. Increased the default limit on number of leases to 1000 (from 150). This is mainly a defence against DoS attacks, and for the average "one for two class C networks" installation, IP address exhaustion does that just as well. Making the limit greater than the number of IP addresses available in such an installation removes a surprise which otherwise can catch people out. Removed extraneous trailing space in the value of the DNSMASQ_TIME_REMAINING DNSMASQ_LEASE_LENGTH and DNSMASQ_LEASE_EXPIRES environment variables. Thanks to Gildas Le Nadan for spotting this. Provide the network-id tags for a DHCP transaction to the lease-change script in the environment variable DNSMASQ_TAGS. A good suggestion from Gildas Le Nadan. Add support for RFC3925 "Vendor-Identifying Vendor Options". The syntax looks like this: --dhcp-option=vi-encap:<enterprise number>, ......... Add support to --dhcp-match to allow matching against RFC3925 "Vendor-Identifying Vendor Classes". The syntax looks like this: --dhcp-match=tag,vi-encap<enterprise number>, <value> Add some application specific code to assist in implementing the Broadband forum TR069 CPE-WAN specification. The details are in contrib/CPE-WAN/README Increase the default DNS packet size limit to 4096, as recommended by RFC5625 section 4.4.3. This can be reconfigured using --edns-packet-max if needed. Thanks to Francis Dupont for pointing this out. Rewrite query-ids even for TSIG signed packets, since this is allowed by RFC5625 section 4.5. Use getopt_long by default on OS X. It has been supported since version 10.3.0. Thanks to Arek Dreyer for spotting this. Added up-to-date startup configuration for MacOSX/launchd in contrib/MacOSX-launchd. Thanks to Arek Dreyer for providing this. Fix link error when including Dbus but excluding DHCP. Thanks to Oschtan for the bug report. Updated French translation. Thanks to Gildas Le Nadan. Updated Polish translation. Thanks to Jan Psota. Updated Spanish translation. Thanks to Chris Chatham. Fixed confusion about domains, when looking up DHCP hosts in /etc/hosts. This could cause spurious "Ignoring domain..." messages. Thanks to Fedor Kozhevnikov for finding and analysing the problem. version 2.51 Add support for internationalised DNS. Non-ASCII characters in domain names found in /etc/hosts, /etc/ethers and /etc/dnsmasq.conf will be correctly handled by translation to punycode, as specified in RFC3490. This function is only available if dnsmasq is compiled with internationalisation support, and adds a dependency on GNU libidn. Without i18n support, dnsmasq continues to be compilable with just standard tools. Thanks to Yves Dorfsman for the suggestion. Add two more environment variables for lease-change scripts: First, DNSMASQ_SUPPLIED_HOSTNAME; this is set to the hostname supplied by a client, even if the actual hostname used is over-ridden by dhcp-host or dhcp-ignore-names directives. Also DNSMASQ_RELAY_ADDRESS which gives the address of a DHCP relay, if used. Suggestions from Michael Rack. Fix regression which broke echo of relay-agent options. Thanks to Michael Rack for spotting this. Don't treat option 67 as being interchangeable with dhcp-boot parameters if it's specified as dhcp-option-force. Make the code to call scripts on lease-change compile-time optional. It can be switched off by editing src/config.h or building with "make COPTS=-DNO_SCRIPT". Make the TFTP server cope with filenames from Windows/DOS which use '\' as pathname separator. Thanks to Ralf for the patch. Updated Polish translation. Thanks to Jan Psota. Warn if an IP address is duplicated in /etc/ethers. Thanks to Felix Schwarz for pointing this out. Teach --conf-dir to take an option list of file suffices which will be ignored when scanning the directory. Useful for backup files etc. Thanks to Helmut Hullen for the suggestion. Add new DHCP option named tftpserver-address, which corresponds to the third argument of dhcp-boot. This allows the complete functionality of dhcp-boot to be replicated with dhcp-option. Useful when using dhcp-optsfile. Test which upstream nameserver to use every 10 seconds or 50 queries and not just when a query times out and is retried. This should improve performance when there is a slow nameserver in the list. Thanks to Joe for the suggestion. Don't do any PXE processing, even for clients with the correct vendorclass, unless at least one pxe-prompt or pxe-service option is given. This stops dnsmasq interfering with proxy PXE subsystems when it is just the DHCP server. Thanks to Spencer Clark for spotting this. Limit the blocksize used for TFTP transfers to a value which avoids packet fragmentation, based on the MTU of the local interface. Many netboot ROMs can't cope with fragmented packets. Honour dhcp-ignore configuration for PXE and proxy-PXE requests. Thanks to Niels Basjes for the bug report. Updated French translation. Thanks to Gildas Le Nadan. version 2.50 Fix security problem which allowed any host permitted to do TFTP to possibly compromise dnsmasq by remote buffer overflow when TFTP enabled. Thanks to Core Security Technologies and Iván Arce, Pablo Hernán Jorge, Alejandro Pablo Rodriguez, Martín Coco, Alberto Soliño Testa and Pablo Annetta. This problem has Bugtraq id: 36121 and CVE: 2009-2957 Fix a problem which allowed a malicious TFTP client to crash dnsmasq. Thanks to Steve Grubb at Red Hat for spotting this. This problem has Bugtraq id: 36120 and CVE: 2009-2958 version 2.49 Fix regression in 2.48 which disables the lease-change script. Thanks to Jose Luis Duran for spotting this. Log TFTP "file not found" errors. These were not logged, since a normal PXELinux boot generates many of them, but the lack of the messages seems to be more confusing than routinely seeing them when there is no real error. Update Spanish translation. Thanks to Chris Chatham. version 2.48 Archived the extensive, backwards, changelog to CHANGELOG.archive. The current changelog now runs from version 2.43 and runs conventionally. Fixed bug which broke binding of servers to physical interfaces when interface names were longer than four characters. Thanks to MURASE Katsunori for the patch. Fixed netlink code to check that messages come from the correct source, and not another userspace process. Thanks to Steve Grubb for the patch. Maintainability drive: removed bug and missing feature workarounds for some old platforms. Solaris 9, OpenBSD older than 4.1, Glibc older than 2.2, Linux 2.2.x and DBus older than 1.1.x are no longer supported. Don't read included configuration files more than once: allows complex configuration structures without problems. Mark log messages from the various subsystems in dnsmasq: messages from the DHCP subsystem now have the ident string "dnsmasq-dhcp" and messages from TFTP have ident "dnsmasq-tftp". Thanks to Olaf Westrik for the patch. Fix possible infinite DHCP protocol loop when an IP address nailed to a hostname (not a MAC address) and a host sometimes provides the name, sometimes not. Allow --addn-hosts to take a directory: all the files in the directory are read. Thanks to Phil Cornelius for the suggestion. Support --bridge-interface on all platforms, not just BSD. Added support for advanced PXE functions. It's now possible to define a prompt and menu options which will be displayed when a client PXE boots. It's also possible to hand-off booting to other boot servers. Proxy-DHCP, where dnsmasq just supplies the PXE information and another DHCP server does address allocation, is also allowed. See the --pxe-prompt and --pxe-service keywords. Thanks to Alkis Georgopoulos for the suggestion and Guilherme Moro and Michael Brown for assistance. Improvements to DHCP logging. Thanks to Tom Metro for useful suggestions. Add ability to build dnsmasq without DHCP support. To do this, edit src/config.h or build with "make COPTS=-DNO_DHCP". Thanks to Mahavir Jain for the patch. Added --test command-line switch - syntax check configuration files only. Updated French translation. Thanks to Gildas Le Nadan. version 2.47 Updated French translation. Thanks to Gildas Le Nadan. Fixed interface enumeration code to work on NetBSD 5.0. Thanks to Roy Marples for the patch. Updated config.h to use the same location for the lease file on NetBSD as the other *BSD variants. Also allow LEASEFILE and CONFFILE symbols to be overridden in CFLAGS. Handle duplicate address detection on IPv6 more intelligently. In IPv6, an interface can have an address which is not usable, because it is still undergoing DAD (such addresses are marked "tentative"). Attempting to bind to an address in this state returns an error, EADDRNOTAVAIL. Previously, on getting such an error, dnsmasq would silently abandon the address, and never listen on it. Now, it retries once per second for 20 seconds before generating a fatal error. 20 seconds should be long enough for any DAD process to complete, but can be adjusted in src/config.h if necessary. Thanks to Martin Krafft for the bug report. Add DBus introspection. Patch from Jeremy Laine. Update Dbus configuration file. Patch from Colin Walters. Fix for this bug: http://bugs.freedesktop.org/show_bug.cgi?id=18961 Support arbitrarily encapsulated DHCP options, suggestion and initial patch from Samium Gromoff. This is useful for (eg) iPXE, which expect all its private options to be encapsulated inside a single option 175. So, eg, dhcp-option = encap:175, 190, "iscsi-client0" dhcp-option = encap:175, 191, "iscsi-client0-secret" will provide iSCSI parameters to iPXE. Enhance --dhcp-match to allow testing of the contents of a client-sent option, as well as its presence. This application in mind for this is RFC 4578 client-architecture specifiers, but it's generally useful. Joey Korkames suggested the enhancement. Move from using the IP_XMIT_IF ioctl to IP_BOUND_IF on OpenSolaris. Thanks to Bastian Machek for the heads-up. No longer complain about blank lines in /etc/ethers. Thanks to Jon Nelson for the patch. Fix binding of servers to physical devices, eg --server=/domain/1.2.3.4@eth0 which was broken from 2.43 onwards unless --query-port=0 set. Thanks to Peter Naulls for the bug report. Reply to DHCPINFORM requests even when the supplied ciaddr doesn't fall in any dhcp-range. In this case it's not possible to supply a complete configuration, but individually-configured options (eg PAC) may be useful. Allow the source address of an alias to be a range: --alias=192.168.0.0,10.0.0.0,255.255.255.0 maps the whole subnet 192.168.0.0->192.168.0.255 to 10.0.0.0->10.0.0.255, as before. --alias=192.168.0.10-192.168.0.40,10.0.0.0,255.255.255.0 maps only the 192.168.0.10->192.168.0.40 region. Thanks to Ib Uhrskov for the suggestion. Don't dynamically allocate DHCP addresses which may break Windows. Addresses which end in .255 or .0 are broken in Windows even when using supernetting. --dhcp-range=192.168.0.1,192.168.1.254,255,255,254.0 means 192.168.0.255 is a valid IP address, but not for Windows. See Microsoft KB281579. We therefore no longer allocate these addresses to avoid hard-to-diagnose problems. Update Polish translation. Thanks to Jan Psota. Delete the PID-file when dnsmasq shuts down. Note that by this time, dnsmasq is normally not running as root, so this will fail if the PID-file is stored in a root-owned directory; such failure is silently ignored. To take advantage of this feature, the PID-file must be stored in a directory owned and write-able by the user running dnsmasq. version 2.46 Allow --bootp-dynamic to take a netid tag, so that it may be selectively enabled. Thanks to Olaf Westrik for the suggestion. Remove ISC-leasefile reading code. This has been deprecated for a long time, and last time I removed it, it ended up going back by request of one user. This time, it's gone for good; otherwise it would need to be re-worked to support multiple domains (see below). Support DHCP clients in multiple DNS domains. This is a long-standing request. Clients are assigned to a domain based in their IP address. Add --dhcp-fqdn flag, which changes behaviour if DNS names assigned to DHCP clients. When this is set, there must be a domain associated with each client, and only fully-qualified domain names are added to the DNS. The advantage is that the only the FQDN needs to be unique, so that two or more DHCP clients can share a hostname, as long as they are in different domains. Set environment variable DNSMASQ_DOMAIN when invoking lease-change script. This may be useful information to have now that it's variable. Tighten up data-checking code for DNS packet handling. Thanks to Steve Dodd who found certain illegal packets which could crash dnsmasq. No memory overwrite was possible, so this is not a security issue beyond the DoS potential. Update example config dhcp option 47, the previous suggestion generated an illegal, zero-length, option. Thanks to Matthias Andree for finding this. Rewrite hosts-file reading code to remove the limit of 1024 characters per line. John C Meuser found this. Create a net-id tag with the name of the interface on which the DHCP request was received. Fixed minor memory leak in DBus code, thanks to Jeremy Laine for the patch. Emit DBus signals as the DHCP lease database changes. Thanks to Jeremy Laine for the patch. Allow for more that one MAC address in a dhcp-host line. This configuration tells dnsmasq that it's OK to abandon a DHCP lease of the fixed address to one MAC address, if another MAC address in the dhcp-host statement asks for an address. This is useful to give a fixed address to a host which has two network interfaces (say, a laptop with wired and wireless interfaces.) It's very important to ensure that only one interface at a time is up, since dnsmasq abandons the first lease and re-uses the address before the leased time has elapsed. John Gray suggested this. Tweak the response to a DHCP request packet with a wrong server-id when --dhcp-authoritative is set; dnsmasq now returns a DHCPNAK, rather than silently ignoring the packet. Thanks to Chris Marget for spotting this improvement. Add --cname option. This provides a limited alias function, usable for DHCP names. Thanks to AJ Weber for suggestions on this. Updated contrib/webmin with latest version from Neil Fisher. Updated Polish translation. Thanks to Jan Psota. Correct the text names for DHCP options 64 and 65 to be "nis+-domain" and "nis+-servers". Updated Spanish translation. Thanks to Chris Chatham. Force re-reading of /etc/resolv.conf when an "interface up" event occurs. version 2.45 Fix total DNS failure in release 2.44 unless --min-port specified. Thanks to Steven Barth and Grant Coady for bugreport. Also reject out-of-range port spec, which could break things too: suggestion from Gilles Espinasse. version 2.44 Fix crash when unknown client attempts to renew a DHCP lease, problem introduced in version 2.43. Thanks to Carlos Carvalho for help chasing this down. Fix potential crash when a host which doesn't have a lease does DHCPINFORM. Again introduced in 2.43. This bug has never been reported in the wild. Fix crash in netlink code introduced in 2.43. Thanks to Jean Wolter for finding this. Change implementation of min_port to work even if min-port is large. Patch to enable compilation of latest Mac OS X. Thanks to David Gilman. Update Spanish translation. Thanks to Christopher Chatham. version 2.43 Updated Polish translation. Thanks to Jan Psota. Flag errors when configuration options are repeated illegally. Further tweaks for GNU/kFreeBSD Add --no-wrap to msgmerge call - provides nicer .po file format. Honour lease-time spec in dhcp-host lines even for BOOTP. The user is assumed to known what they are doing in this case. (Hosts without the time spec still get infinite leases for BOOTP, over-riding the default in the dhcp-range.) Thanks to Peter Katzmann for uncovering this. Fix problem matching relay-agent ids. Thanks to Michael Rack for the bug report. Add --naptr-record option. Suggestion from Johan Bergquist. Implement RFC 5107 server-id-override DHCP relay agent option. Apply patches from Stefan Kruger for compilation on Solaris 10 under Sun studio. Yet more tweaking of Linux capability code, to suppress pointless wingeing from kernel 2.6.25 and above. Improve error checking during startup. Previously, some errors which occurred during startup would be worked around, with dnsmasq still starting up. Some were logged, some silent. Now, they all cause a fatal error and dnsmasq terminates with a non-zero exit code. The errors are those associated with changing uid and gid, setting process capabilities and writing the pidfile. Thanks to Uwe Gansert and the Suse security team for pointing out this improvement, and Bill Reimers for good implementation suggestions. Provide NO_LARGEFILE compile option to switch off largefile support when compiling against versions of uclibc which don't support it. Thanks to Stephane Billiart for the patch. Implement random source ports for interactions with upstream nameservers. New spoofing attacks have been found against nameservers which do not do this, though it is not clear if dnsmasq is vulnerable, since to doesn't implement recursion. By default dnsmasq will now use a different source port (and socket) for each query it sends upstream. This behaviour can suppressed using the --query-port option, and the old default behaviour restored using --query-port=0. Explicit source-port specifications in --server configs are still honoured. Replace the random number generator, for better security. On most BSD systems, dnsmasq uses the arc4random() RNG, which is secure, but on other platforms, it relied on the C-library RNG, which may be guessable and therefore allow spoofing. This release replaces the libc RNG with the SURF RNG, from Daniel J. Berstein's DJBDNS package. Don't attempt to change user or group or set capabilities if dnsmasq is run as a non-root user. Without this, the change from soft to hard errors when these fail causes problems for non-root daemons listening on high ports. Thanks to Patrick McLean for spotting this. Updated French translation. Thanks to Gildas Le Nadan. version 2.42 The changelog for version 2.42 and earlier is available in CHANGELOG.archive.