From 3d0386f27ca66379acf50199e1d1298386eeeeb8 Mon Sep 17 00:00:00 2001 From: Daniel Baumann Date: Mon, 6 May 2024 02:55:53 +0200 Subject: Adding upstream version 3.2.1. Signed-off-by: Daniel Baumann --- lib/README.rst | 314 ++++++++ lib/cache/api.c | 889 +++++++++++++++++++++++ lib/cache/api.h | 201 ++++++ lib/cache/cdb_api.h | 68 ++ lib/cache/cdb_lmdb.c | 710 ++++++++++++++++++ lib/cache/cdb_lmdb.h | 23 + lib/cache/entry_list.c | 293 ++++++++ lib/cache/entry_pkt.c | 224 ++++++ lib/cache/entry_rr.c | 146 ++++ lib/cache/impl.h | 412 +++++++++++ lib/cache/knot_pkt.c | 106 +++ lib/cache/nsec1.c | 511 +++++++++++++ lib/cache/nsec3.c | 501 +++++++++++++ lib/cache/peek.c | 724 +++++++++++++++++++ lib/cookies/alg_containers.c | 72 ++ lib/cookies/alg_containers.h | 49 ++ lib/cookies/alg_sha.c | 122 ++++ lib/cookies/alg_sha.h | 30 + lib/cookies/control.h | 49 ++ lib/cookies/helper.c | 296 ++++++++ lib/cookies/helper.h | 86 +++ lib/cookies/lru_cache.c | 70 ++ lib/cookies/lru_cache.h | 69 ++ lib/cookies/nonce.c | 34 + lib/cookies/nonce.h | 43 ++ lib/defines.h | 102 +++ lib/dnssec.c | 473 ++++++++++++ lib/dnssec.h | 152 ++++ lib/dnssec/nsec.c | 540 ++++++++++++++ lib/dnssec/nsec.h | 105 +++ lib/dnssec/nsec3.c | 776 ++++++++++++++++++++ lib/dnssec/nsec3.h | 82 +++ lib/dnssec/signature.c | 307 ++++++++ lib/dnssec/signature.h | 42 ++ lib/dnssec/ta.c | 185 +++++ lib/dnssec/ta.h | 87 +++ lib/generic/README.rst | 60 ++ lib/generic/array.h | 166 +++++ lib/generic/lru.c | 236 ++++++ lib/generic/lru.h | 249 +++++++ lib/generic/map.c | 354 +++++++++ lib/generic/map.h | 115 +++ lib/generic/pack.h | 249 +++++++ lib/generic/queue.c | 124 ++++ lib/generic/queue.h | 260 +++++++ lib/generic/set.h | 105 +++ lib/generic/trie.c | 912 +++++++++++++++++++++++ lib/generic/trie.h | 150 ++++ lib/layer.h | 107 +++ lib/layer/cache.c | 31 + lib/layer/iterate.c | 1135 +++++++++++++++++++++++++++++ lib/layer/iterate.h | 36 + lib/layer/validate.c | 1133 +++++++++++++++++++++++++++++ lib/lib.mk | 110 +++ lib/module.c | 170 +++++ lib/module.h | 111 +++ lib/nsrep.c | 563 +++++++++++++++ lib/nsrep.h | 188 +++++ lib/resolve.c | 1648 ++++++++++++++++++++++++++++++++++++++++++ lib/resolve.h | 332 +++++++++ lib/rplan.c | 308 ++++++++ lib/rplan.h | 221 ++++++ lib/utils.c | 1084 +++++++++++++++++++++++++++ lib/utils.h | 495 +++++++++++++ lib/zonecut.c | 597 +++++++++++++++ lib/zonecut.h | 173 +++++ 66 files changed, 20315 insertions(+) create mode 100644 lib/README.rst create mode 100644 lib/cache/api.c create mode 100644 lib/cache/api.h create mode 100644 lib/cache/cdb_api.h create mode 100644 lib/cache/cdb_lmdb.c create mode 100644 lib/cache/cdb_lmdb.h create mode 100644 lib/cache/entry_list.c create mode 100644 lib/cache/entry_pkt.c create mode 100644 lib/cache/entry_rr.c create mode 100644 lib/cache/impl.h create mode 100644 lib/cache/knot_pkt.c create mode 100644 lib/cache/nsec1.c create mode 100644 lib/cache/nsec3.c create mode 100644 lib/cache/peek.c create mode 100644 lib/cookies/alg_containers.c create mode 100644 lib/cookies/alg_containers.h create mode 100644 lib/cookies/alg_sha.c create mode 100644 lib/cookies/alg_sha.h create mode 100644 lib/cookies/control.h create mode 100644 lib/cookies/helper.c create mode 100644 lib/cookies/helper.h create mode 100644 lib/cookies/lru_cache.c create mode 100644 lib/cookies/lru_cache.h create mode 100644 lib/cookies/nonce.c create mode 100644 lib/cookies/nonce.h create mode 100644 lib/defines.h create mode 100644 lib/dnssec.c create mode 100644 lib/dnssec.h create mode 100644 lib/dnssec/nsec.c create mode 100644 lib/dnssec/nsec.h create mode 100644 lib/dnssec/nsec3.c create mode 100644 lib/dnssec/nsec3.h create mode 100644 lib/dnssec/signature.c create mode 100644 lib/dnssec/signature.h create mode 100644 lib/dnssec/ta.c create mode 100644 lib/dnssec/ta.h create mode 100644 lib/generic/README.rst create mode 100644 lib/generic/array.h create mode 100644 lib/generic/lru.c create mode 100644 lib/generic/lru.h create mode 100644 lib/generic/map.c create mode 100644 lib/generic/map.h create mode 100644 lib/generic/pack.h create mode 100644 lib/generic/queue.c create mode 100644 lib/generic/queue.h create mode 100644 lib/generic/set.h create mode 100644 lib/generic/trie.c create mode 100644 lib/generic/trie.h create mode 100644 lib/layer.h create mode 100644 lib/layer/cache.c create mode 100644 lib/layer/iterate.c create mode 100644 lib/layer/iterate.h create mode 100644 lib/layer/validate.c create mode 100644 lib/lib.mk create mode 100644 lib/module.c create mode 100644 lib/module.h create mode 100644 lib/nsrep.c create mode 100644 lib/nsrep.h create mode 100644 lib/resolve.c create mode 100644 lib/resolve.h create mode 100644 lib/rplan.c create mode 100644 lib/rplan.h create mode 100644 lib/utils.c create mode 100644 lib/utils.h create mode 100644 lib/zonecut.c create mode 100644 lib/zonecut.h (limited to 'lib') diff --git a/lib/README.rst b/lib/README.rst new file mode 100644 index 0000000..2905676 --- /dev/null +++ b/lib/README.rst @@ -0,0 +1,314 @@ +********************* +Knot Resolver library +********************* + +Requirements +============ + +* libknot_ 2.0 (Knot DNS high-performance DNS library.) + +For users +========= + +The library as described provides basic services for name resolution, which should cover the usage, +examples are in the :ref:`resolve API ` documentation. + +.. tip:: If you're migrating from ``getaddrinfo()``, see *"synchronous"* API, but the library offers iterative API as well to plug it into your event loop for example. + +.. _lib-layers: + +For developers +============== + +The resolution process starts with the functions in :ref:`resolve.c `, they are responsible for: + +* reacting to state machine state (i.e. calling consume layers if we have an answer ready) +* interacting with the library user (i.e. asking caller for I/O, accepting queries) +* fetching assets needed by layers (i.e. zone cut) + +This is the *driver*. The driver is not meant to know *"how"* the query resolves, but rather *"when"* to execute *"what"*. + +.. image:: ../doc/resolution.png + :align: center + +On the other side are *layers*. They are responsible for dissecting the packets and informing the driver about the results. For example, a *produce* layer generates query, a *consume* layer validates answer. + +.. tip:: Layers are executed asynchronously by the driver. If you need some asset beforehand, you can signalize the driver using returning state or current query flags. For example, setting a flag ``AWAIT_CUT`` forces driver to fetch zone cut information before the packet is consumed; setting a ``RESOLVED`` flag makes it pop a query after the current set of layers is finished; returning ``FAIL`` state makes it fail current query. + +Layers can also change course of resolution, for example by appending additional queries. + +.. code-block:: lua + + consume = function (state, req, answer) + answer = kres.pkt_t(answer) + if answer:qtype() == kres.type.NS then + req = kres.request_t(req) + local qry = req:push(answer:qname(), kres.type.SOA, kres.class.IN) + qry.flags.AWAIT_CUT = true + end + return state + end + +This **doesn't** block currently processed query, and the newly created sub-request will start as soon as driver finishes processing current. In some cases you might need to issue sub-request and process it **before** continuing with the current, i.e. validator may need a DNSKEY before it can validate signatures. In this case, layers can yield and resume afterwards. + +.. code-block:: lua + + consume = function (state, req, answer) + answer = kres.pkt_t(answer) + if state == kres.YIELD then + print('continuing yielded layer') + return kres.DONE + else + if answer:qtype() == kres.type.NS then + req = kres.request_t(req) + local qry = req:push(answer:qname(), kres.type.SOA, kres.class.IN) + qry.flags.AWAIT_CUT = true + print('planned SOA query, yielding') + return kres.YIELD + end + return state + end + end + +The ``YIELD`` state is a bit special. When a layer returns it, it interrupts current walk through the layers. When the layer receives it, +it means that it yielded before and now it is resumed. This is useful in a situation where you need a sub-request to determine whether current answer is valid or not. + +Writing layers +============== + +.. warning:: FIXME: this dev-docs section is outdated! Better see comments in files instead, for now. + +The resolver :ref:`library ` leverages the processing API from the libknot to separate packet processing code into layers. + +.. note:: This is only crash-course in the library internals, see the resolver :ref:`library ` documentation for the complete overview of the services. + +The library offers following services: + +- :ref:`Cache ` - MVCC cache interface for retrieving/storing resource records. +- :ref:`Resolution plan ` - Query resolution plan, a list of partial queries (with hierarchy) sent in order to satisfy original query. This contains information about the queries, nameserver choice, timing information, answer and its class. +- :ref:`Nameservers ` - Reputation database of nameservers, this serves as an aid for nameserver choice. + +A processing layer is going to be called by the query resolution driver for each query, +so you're going to work with :ref:`struct kr_request ` as your per-query context. +This structure contains pointers to resolution context, resolution plan and also the final answer. + +.. code-block:: c + + int consume(kr_layer_t *ctx, knot_pkt_t *pkt) + { + struct kr_request *req = ctx->req; + struct kr_query *qry = req->current_query; + } + +This is only passive processing of the incoming answer. If you want to change the course of resolution, say satisfy a query from a local cache before the library issues a query to the nameserver, you can use states (see the :ref:`Static hints ` for example). + +.. code-block:: c + + int produce(kr_layer_t *ctx, knot_pkt_t *pkt) + { + struct kr_request *req = ctx->req; + struct kr_query *qry = req->current_query; + + /* Query can be satisfied locally. */ + if (can_satisfy(qry)) { + /* This flag makes the resolver move the query + * to the "resolved" list. */ + qry->flags.RESOLVED = true; + return KR_STATE_DONE; + } + + /* Pass-through. */ + return ctx->state; + } + +It is possible to not only act during the query resolution, but also to view the complete resolution plan afterwards. This is useful for analysis-type tasks, or *"per answer"* hooks. + +.. code-block:: c + + int finish(kr_layer_t *ctx) + { + struct kr_request *req = ctx->req; + struct kr_rplan *rplan = req->rplan; + + /* Print the query sequence with start time. */ + char qname_str[KNOT_DNAME_MAXLEN]; + struct kr_query *qry = NULL + WALK_LIST(qry, rplan->resolved) { + knot_dname_to_str(qname_str, qry->sname, sizeof(qname_str)); + printf("%s at %u\n", qname_str, qry->timestamp); + } + + return ctx->state; + } + +APIs in Lua +=========== + +The APIs in Lua world try to mirror the C APIs using LuaJIT FFI, with several differences and enhancements. +There is not comprehensive guide on the API yet, but you can have a look at the bindings_ file. + +Elementary types and constants +------------------------------ + +* States are directly in ``kres`` table, e.g. ``kres.YIELD, kres.CONSUME, kres.PRODUCE, kres.DONE, kres.FAIL``. +* DNS classes are in ``kres.class`` table, e.g. ``kres.class.IN`` for Internet class. +* DNS types are in ``kres.type`` table, e.g. ``kres.type.AAAA`` for AAAA type. +* DNS rcodes types are in ``kres.rcode`` table, e.g. ``kres.rcode.NOERROR``. +* Packet sections (QUESTION, ANSWER, AUTHORITY, ADDITIONAL) are in the ``kres.section`` table. + +Working with domain names +------------------------- + +The internal API usually works with domain names in label format, you can convert between text and wire freely. + +.. code-block:: lua + + local dname = kres.str2dname('business.se') + local strname = kres.dname2str(dname) + +Working with resource records +----------------------------- + +Resource records are stored as tables. + +.. code-block:: lua + + local rr = { owner = kres.str2dname('owner'), + ttl = 0, + class = kres.class.IN, + type = kres.type.CNAME, + rdata = kres.str2dname('someplace') } + print(kres.rr2str(rr)) + +RRSets in packet can be accessed using FFI, you can easily fetch single records. + +.. code-block:: lua + + local rrset = { ... } + local rr = rrset:get(0) -- Return first RR + print(kres.dname2str(rr:owner())) + print(rr:ttl()) + print(kres.rr2str(rr)) + +Working with packets +-------------------- + +Packet is the data structure that you're going to see in layers very often. They consists of a header, and four sections: QUESTION, ANSWER, AUTHORITY, ADDITIONAL. The first section is special, as it contains the query name, type, and class; the rest of the sections contain RRSets. + +First you need to convert it to a type known to FFI and check basic properties. Let's start with a snippet of a *consume* layer. + +.. code-block:: lua + + consume = function (state, req, pkt) + pkt = kres.pkt_t(answer) + print('rcode:', pkt:rcode()) + print('query:', kres.dname2str(pkt:qname()), pkt:qclass(), pkt:qtype()) + if pkt:rcode() ~= kres.rcode.NOERROR then + print('error response') + end + end + +You can enumerate records in the sections. + +.. code-block:: lua + + local records = pkt:section(kres.section.ANSWER) + for i = 1, #records do + local rr = records[i] + if rr.type == kres.type.AAAA then + print(kres.rr2str(rr)) + end + end + +During *produce* or *begin*, you might want to want to write to packet. Keep in mind that you have to write packet sections in sequence, +e.g. you can't write to ANSWER after writing AUTHORITY, it's like stages where you can't go back. + +.. code-block:: lua + + pkt:rcode(kres.rcode.NXDOMAIN) + -- Clear answer and write QUESTION + pkt:recycle() + pkt:question('\7blocked', kres.class.IN, kres.type.SOA) + -- Start writing data + pkt:begin(kres.section.ANSWER) + -- Nothing in answer + pkt:begin(kres.section.AUTHORITY) + local soa = { owner = '\7blocked', ttl = 900, class = kres.class.IN, type = kres.type.SOA, rdata = '...' } + pkt:put(soa.owner, soa.ttl, soa.class, soa.type, soa.rdata) + +Working with requests +--------------------- + +The request holds information about currently processed query, enabled options, cache, and other extra data. +You primarily need to retrieve currently processed query. + +.. code-block:: lua + + consume = function (state, req, pkt) + req = kres.request_t(req) + print(req.options) + print(req.state) + + -- Print information about current query + local current = req:current() + print(kres.dname2str(current.owner)) + print(current.stype, current.sclass, current.id, current.flags) + end + +In layers that either begin or finalize, you can walk the list of resolved queries. + +.. code-block:: lua + + local last = req:resolved() + print(last.stype) + +As described in the layers, you can not only retrieve information about current query, but also push new ones or pop old ones. + +.. code-block:: lua + + -- Push new query + local qry = req:push(pkt:qname(), kres.type.SOA, kres.class.IN) + qry.flags.AWAIT_CUT = true + + -- Pop the query, this will erase it from resolution plan + req:pop(qry) + + +.. _libknot: https://gitlab.labs.nic.cz/knot/knot-dns/tree/master/src/libknot +.. _bindings: https://gitlab.labs.nic.cz/knot/knot-resolver/blob/master/daemon/lua/kres.lua + + +Significant Lua API changes +--------------------------- + +Incompatible changes since 3.0.0 +~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ + +In the main ``kres.*`` lua binding, there was only change in struct knot_rrset_t: + +- constructor now accepts TTL as additional parameter (defaulting to zero) +- add_rdata() doesn't accept TTL anymore (and will throw an error if passed) + +In case you used knot_* functions and structures bound to lua: + +- knot_dname_is_sub(a, b): knot_dname_in_bailiwick(a, b) > 0 +- knot_rdata_rdlen(): knot_rdataset_at().len +- knot_rdata_data(): knot_rdataset_at().data +- knot_rdata_array_size(): offsetof(struct knot_data_t, data) + knot_rdataset_at().len +- struct knot_rdataset: field names were renamed to .count and .rdata +- some functions got inlined from headers, but you can use their kr_* clones: + kr_rrsig_sig_inception(), kr_rrsig_sig_expiration(), kr_rrsig_type_covered(). + Note that these functions now accept knot_rdata_t* instead of a pair + knot_rdataset_t* and size_t - you can use knot_rdataset_at() for that. + +- knot_rrset_add_rdata() doesn't take TTL parameter anymore +- knot_rrset_init_empty() was inlined, but in lua you can use the constructor +- knot_rrset_ttl() was inlined, but in lua you can use :ttl() method instead + +- knot_pkt_qname(), _qtype(), _qclass(), _rr(), _section() were inlined, + but in lua you can use methods instead, e.g. myPacket:qname() +- knot_pkt_free() takes knot_pkt_t* instead of knot_pkt_t**, but from lua + you probably didn't want to use that; constructor ensures garbage collection. + + +.. |---| unicode:: U+02014 .. em dash diff --git a/lib/cache/api.c b/lib/cache/api.c new file mode 100644 index 0000000..c0591d6 --- /dev/null +++ b/lib/cache/api.c @@ -0,0 +1,889 @@ +/* Copyright (C) 2014-2017 CZ.NIC, z.s.p.o. + + This program is free software: you can redistribute it and/or modify + it under the terms of the GNU General Public License as published by + the Free Software Foundation, either version 3 of the License, or + (at your option) any later version. + + This program is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the + GNU General Public License for more details. + + You should have received a copy of the GNU General Public License + along with this program. If not, see . + */ + +#include +#include +#include +#include +#include +#include +#include + +#include +#include +#include +#include + +#include "contrib/cleanup.h" +#include "contrib/ucw/lib.h" +#include "lib/cache/api.h" +#include "lib/cache/cdb_lmdb.h" +#include "lib/defines.h" +#include "lib/generic/trie.h" +#include "lib/resolve.h" +#include "lib/rplan.h" +#include "lib/utils.h" + +#include "lib/cache/impl.h" + +/* TODO: + * - Reconsider when RRSIGs are put in and retrieved from the cache. + * Currently it's always done, which _might_ be spurious, depending + * on how kresd will use the returned result. + * There's also the "problem" that kresd ATM does _not_ ask upstream + * with DO bit in some cases. + */ + + +/** Cache version */ +static const uint16_t CACHE_VERSION = 5; +/** Key size */ +#define KEY_HSIZE (sizeof(uint8_t) + sizeof(uint16_t)) +#define KEY_SIZE (KEY_HSIZE + KNOT_DNAME_MAXLEN) + + +/** @internal Forward declarations of the implementation details + * \param optout[out] Set *optout = true; when encountering an opt-out NSEC3 (optional). */ +static ssize_t stash_rrset(struct kr_cache *cache, const struct kr_query *qry, + const knot_rrset_t *rr, const knot_rrset_t *rr_sigs, uint32_t timestamp, + uint8_t rank, trie_t *nsec_pmap, bool *has_optout); +/** Preliminary checks before stash_rrset(). Don't call if returns <= 0. */ +static int stash_rrset_precond(const knot_rrset_t *rr, const struct kr_query *qry/*logs*/); + +/** @internal Removes all records from cache. */ +static inline int cache_clear(struct kr_cache *cache) +{ + cache->stats.delete += 1; + return cache_op(cache, clear); +} + +/** @internal Open cache db transaction and check internal data version. */ +static int assert_right_version(struct kr_cache *cache) +{ + /* Check cache ABI version. */ + /* CACHE_KEY_DEF: to avoid collisions with kr_cache_match(). */ + uint8_t key_str[4] = "VERS"; + knot_db_val_t key = { .data = key_str, .len = sizeof(key_str) }; + knot_db_val_t val = { NULL, 0 }; + int ret = cache_op(cache, read, &key, &val, 1); + if (ret == 0 && val.len == sizeof(CACHE_VERSION) + && memcmp(val.data, &CACHE_VERSION, sizeof(CACHE_VERSION)) == 0) { + ret = kr_error(EEXIST); + } else { + int oldret = ret; + /* Version doesn't match. Recreate cache and write version key. */ + ret = cache_op(cache, count); + if (ret != 0) { /* Non-empty cache, purge it. */ + kr_log_info("[ ][cach] incompatible cache database detected, purging\n"); + if (oldret) { + kr_log_verbose("bad ret: %d\n", oldret); + } else if (val.len != sizeof(CACHE_VERSION)) { + kr_log_verbose("bad length: %d\n", (int)val.len); + } else { + uint16_t ver; + memcpy(&ver, val.data, sizeof(ver)); + kr_log_verbose("bad version: %d\n", (int)ver); + } + ret = cache_clear(cache); + } + /* Either purged or empty. */ + if (ret == 0) { + /* Key/Val is invalidated by cache purge, recreate it */ + val.data = /*const-cast*/(void *)&CACHE_VERSION; + val.len = sizeof(CACHE_VERSION); + ret = cache_op(cache, write, &key, &val, 1); + } + } + kr_cache_sync(cache); + return ret; +} + +int kr_cache_open(struct kr_cache *cache, const struct kr_cdb_api *api, struct kr_cdb_opts *opts, knot_mm_t *mm) +{ + if (!cache) { + return kr_error(EINVAL); + } + /* Open cache */ + if (!api) { + api = kr_cdb_lmdb(); + } + cache->api = api; + int ret = cache->api->open(&cache->db, opts, mm); + if (ret != 0) { + return ret; + } + memset(&cache->stats, 0, sizeof(cache->stats)); + cache->ttl_min = KR_CACHE_DEFAULT_TTL_MIN; + cache->ttl_max = KR_CACHE_DEFAULT_TTL_MAX; + /* Check cache ABI version */ + kr_cache_make_checkpoint(cache); + (void)assert_right_version(cache); + + char *fpath; + ret = asprintf(&fpath, "%s/data.mdb", opts->path); + if (ret > 0) { + kr_cache_emergency_file_to_remove = fpath; + } else { + assert(false); /* non-critical, but still */ + } + return 0; +} + +const char *kr_cache_emergency_file_to_remove = NULL; + + +#define cache_isvalid(cache) ((cache) && (cache)->api && (cache)->db) + +void kr_cache_close(struct kr_cache *cache) +{ + if (cache_isvalid(cache)) { + cache_op(cache, close); + cache->db = NULL; + } + free(/*const-cast*/(char*)kr_cache_emergency_file_to_remove); + kr_cache_emergency_file_to_remove = NULL; +} + +int kr_cache_sync(struct kr_cache *cache) +{ + if (!cache_isvalid(cache)) { + return kr_error(EINVAL); + } + if (cache->api->sync) { + return cache_op(cache, sync); + } + return kr_ok(); +} + +int kr_cache_insert_rr(struct kr_cache *cache, const knot_rrset_t *rr, const knot_rrset_t *rrsig, uint8_t rank, uint32_t timestamp) +{ + int err = stash_rrset_precond(rr, NULL); + if (err <= 0) { + return kr_ok(); + } + ssize_t written = stash_rrset(cache, NULL, rr, rrsig, timestamp, rank, NULL, NULL); + /* Zone's NSEC* parames aren't updated, but that's probably OK + * for kr_cache_insert_rr() */ + if (written >= 0) { + return kr_ok(); + } + + return (int) written; +} + +int kr_cache_clear(struct kr_cache *cache) +{ + if (!cache_isvalid(cache)) { + return kr_error(EINVAL); + } + int ret = cache_clear(cache); + if (ret == 0) { + kr_cache_make_checkpoint(cache); + ret = assert_right_version(cache); + } + return ret; +} + +/* When going stricter, BEWARE of breaking entry_h_consistent_NSEC() */ +struct entry_h * entry_h_consistent(knot_db_val_t data, uint16_t type) +{ + (void) type; /* unused, for now */ + if (!data.data) return NULL; + /* Length checks. */ + if (data.len < offsetof(struct entry_h, data)) + return NULL; + const struct entry_h *eh = data.data; + if (eh->is_packet) { + uint16_t pkt_len; + if (data.len < offsetof(struct entry_h, data) + sizeof(pkt_len)) { + return NULL; + } + memcpy(&pkt_len, eh->data, sizeof(pkt_len)); + if (data.len < offsetof(struct entry_h, data) + sizeof(pkt_len) + + pkt_len) { + return NULL; + } + } + + bool ok = true; + ok = ok && kr_rank_check(eh->rank); + ok = ok && (!kr_rank_test(eh->rank, KR_RANK_BOGUS) + || eh->is_packet); + ok = ok && (eh->is_packet || !eh->has_optout); + + return ok ? /*const-cast*/(struct entry_h *)eh : NULL; +} + +int32_t get_new_ttl(const struct entry_h *entry, const struct kr_query *qry, + const knot_dname_t *owner, uint16_t type, uint32_t now) +{ + int32_t diff = now - entry->time; + if (diff < 0) { + /* We may have obtained the record *after* the request started. */ + diff = 0; + } + int32_t res = entry->ttl - diff; + if (res < 0 && owner && qry && qry->stale_cb) { + /* Stale-serving decision, delegated to a callback. */ + int res_stale = qry->stale_cb(res, owner, type, qry); + if (res_stale >= 0) + return res_stale; + } + return res; +} + +int32_t kr_cache_ttl(const struct kr_cache_p *peek, const struct kr_query *qry, + const knot_dname_t *name, uint16_t type) +{ + const struct entry_h *eh = peek->raw_data; + return get_new_ttl(eh, qry, name, type, qry->timestamp.tv_sec); +} + +/** Check that no label contains a zero character, incl. a log trace. + * + * We refuse to work with those, as LF and our cache keys might become ambiguous. + * Assuming uncompressed name, as usual. + * CACHE_KEY_DEF + */ +static bool check_dname_for_lf(const knot_dname_t *n, const struct kr_query *qry/*logging*/) +{ + const bool ret = knot_dname_size(n) == strlen((const char *)n) + 1; + if (!ret) { WITH_VERBOSE(qry) { + auto_free char *n_str = kr_dname_text(n); + VERBOSE_MSG(qry, "=> skipping zero-containing name %s\n", n_str); + } } + return ret; +} + +/** Return false on types to be ignored. Meant both for sname and direct cache requests. */ +static bool check_rrtype(uint16_t type, const struct kr_query *qry/*logging*/) +{ + const bool ret = !knot_rrtype_is_metatype(type) + && type != KNOT_RRTYPE_RRSIG; + if (!ret) { WITH_VERBOSE(qry) { + auto_free char *type_str = kr_rrtype_text(type); + VERBOSE_MSG(qry, "=> skipping RR type %s\n", type_str); + } } + return ret; +} + +/** Like key_exact_type() but omits a couple checks not holding for pkt cache. */ +knot_db_val_t key_exact_type_maypkt(struct key *k, uint16_t type) +{ + assert(check_rrtype(type, NULL)); + switch (type) { + case KNOT_RRTYPE_RRSIG: /* no RRSIG query caching, at least for now */ + assert(false); + return (knot_db_val_t){ NULL, 0 }; + /* xNAME lumped into NS. */ + case KNOT_RRTYPE_CNAME: + case KNOT_RRTYPE_DNAME: + type = KNOT_RRTYPE_NS; + default: + break; + } + + int name_len = k->buf[0]; + k->buf[name_len + 1] = 0; /* make sure different names can never match */ + k->buf[name_len + 2] = 'E'; /* tag for exact name+type matches */ + memcpy(k->buf + name_len + 3, &type, 2); + k->type = type; + /* CACHE_KEY_DEF: key == dname_lf + '\0' + 'E' + RRTYPE */ + return (knot_db_val_t){ k->buf + 1, name_len + 4 }; +} + + +/** The inside for cache_peek(); implementation separated to ./peek.c */ +int peek_nosync(kr_layer_t *ctx, knot_pkt_t *pkt); +/** function for .produce phase */ +int cache_peek(kr_layer_t *ctx, knot_pkt_t *pkt) +{ + struct kr_request *req = ctx->req; + struct kr_query *qry = req->current_query; + /* We first check various exit-conditions and then call the _real function. */ + + if (!kr_cache_is_open(&req->ctx->cache) + || ctx->state & (KR_STATE_FAIL|KR_STATE_DONE) || qry->flags.NO_CACHE + || (qry->flags.CACHE_TRIED && !qry->stale_cb) + || !check_rrtype(qry->stype, qry) /* LATER: some other behavior for some of these? */ + || qry->sclass != KNOT_CLASS_IN) { + return ctx->state; /* Already resolved/failed or already tried, etc. */ + } + /* ATM cache only peeks for qry->sname and that would be useless + * to repeat on every iteration, so disable it from now on. + * LATER(optim.): assist with more precise QNAME minimization. */ + qry->flags.CACHE_TRIED = true; + + if (qry->stype == KNOT_RRTYPE_NSEC) { + VERBOSE_MSG(qry, "=> skipping stype NSEC\n"); + return ctx->state; + } + if (!check_dname_for_lf(qry->sname, qry)) { + return ctx->state; + } + + int ret = peek_nosync(ctx, pkt); + kr_cache_sync(&req->ctx->cache); + return ret; +} + + + +/** It's simply inside of cycle taken out to decrease indentation. \return error code. */ +static int stash_rrarray_entry(ranked_rr_array_t *arr, int arr_i, + const struct kr_query *qry, struct kr_cache *cache, + int *unauth_cnt, trie_t *nsec_pmap, bool *has_optout); +/** Stash a single nsec_p. \return 0 (errors are ignored). */ +static int stash_nsec_p(const knot_dname_t *dname, const char *nsec_p_v, + struct kr_request *req); + +/** The whole .consume phase for the cache module. */ +int cache_stash(kr_layer_t *ctx, knot_pkt_t *pkt) +{ + struct kr_request *req = ctx->req; + struct kr_query *qry = req->current_query; + struct kr_cache *cache = &req->ctx->cache; + + /* Note: we cache even in KR_STATE_FAIL. For example, + * BOGUS answer can go to +cd cache even without +cd request. */ + if (!kr_cache_is_open(cache) || !qry + || qry->flags.CACHED || !check_rrtype(knot_pkt_qtype(pkt), qry) + || qry->sclass != KNOT_CLASS_IN) { + return ctx->state; + } + /* Do not cache truncated answers, at least for now. LATER */ + if (knot_wire_get_tc(pkt->wire)) { + return ctx->state; + } + /* Stash individual records. */ + ranked_rr_array_t *selected[] = kr_request_selected(req); + int unauth_cnt = 0; + trie_t *nsec_pmap = trie_create(&req->pool); + if (!nsec_pmap) { + assert(!ENOMEM); + goto finally; + } + bool has_optout = false; + /* ^^ DNSSEC_OPTOUT is not fired in cases like `com. A`, + * but currently we don't stash separate NSEC3 proving that. */ + for (int psec = KNOT_ANSWER; psec <= KNOT_ADDITIONAL; ++psec) { + ranked_rr_array_t *arr = selected[psec]; + /* uncached entries are located at the end */ + for (ssize_t i = arr->len - 1; i >= 0; --i) { + ranked_rr_array_entry_t *entry = arr->at[i]; + if (entry->qry_uid != qry->uid) { + continue; + /* TODO: probably safe to break but maybe not worth it */ + } + int ret = stash_rrarray_entry(arr, i, qry, cache, &unauth_cnt, + nsec_pmap, &has_optout); + if (ret) { + VERBOSE_MSG(qry, "=> stashing RRs errored out\n"); + goto finally; + } + /* LATER(optim.): maybe filter out some type-rank combinations + * that won't be useful as separate RRsets. */ + } + } + + trie_it_t *it; + for (it = trie_it_begin(nsec_pmap); !trie_it_finished(it); trie_it_next(it)) { + stash_nsec_p((const knot_dname_t *)trie_it_key(it, NULL), + (const char *)*trie_it_val(it), req); + } + trie_it_free(it); + /* LATER(optim.): typically we also have corresponding NS record in the list, + * so we might save a cache operation. */ + + stash_pkt(pkt, qry, req, has_optout); + +finally: + if (unauth_cnt) { + VERBOSE_MSG(qry, "=> stashed also %d nonauth RRsets\n", unauth_cnt); + }; + kr_cache_sync(cache); + return ctx->state; /* we ignore cache-stashing errors */ +} + +/** Preliminary checks before stash_rrset(). Don't call if returns <= 0. */ +static int stash_rrset_precond(const knot_rrset_t *rr, const struct kr_query *qry/*logs*/) +{ + if (!rr || rr->rclass != KNOT_CLASS_IN) { + assert(!EINVAL); + return kr_error(EINVAL); + } + if (!check_rrtype(rr->type, qry)) { + return kr_ok(); + } + if (!check_dname_for_lf(rr->owner, qry)) { + return kr_ok(); + } + return 1/*proceed*/; +} + +static ssize_t stash_rrset(struct kr_cache *cache, const struct kr_query *qry, + const knot_rrset_t *rr, const knot_rrset_t *rr_sigs, uint32_t timestamp, + uint8_t rank, trie_t *nsec_pmap, bool *has_optout) +{ + assert(stash_rrset_precond(rr, qry) > 0); + if (!cache) { + assert(!EINVAL); + return kr_error(EINVAL); + } + + const int wild_labels = rr_sigs == NULL ? 0 : + knot_dname_labels(rr->owner, NULL) - knot_rrsig_labels(rr_sigs->rrs.rdata); + if (wild_labels < 0) { + return kr_ok(); + } + const knot_dname_t *encloser = rr->owner; /**< the closest encloser name */ + for (int i = 0; i < wild_labels; ++i) { + encloser = knot_wire_next_label(encloser, NULL); + } + int ret = 0; + + /* Construct the key under which RRs will be stored, + * and add corresponding nsec_pmap item (if necessary). */ + struct key k_storage, *k = &k_storage; + knot_db_val_t key; + switch (rr->type) { + case KNOT_RRTYPE_NSEC3: + /* Skip "suspicious" or opt-out NSEC3 sets. */ + if (rr->rrs.count != 1) return kr_ok(); + if (KNOT_NSEC3_FLAG_OPT_OUT & knot_nsec3_flags(rr->rrs.rdata)) { + if (has_optout) *has_optout = true; + return kr_ok(); + } + /* fall through */ + case KNOT_RRTYPE_NSEC: + if (!kr_rank_test(rank, KR_RANK_SECURE)) { + /* Skip any NSEC*s that aren't validated. */ + return kr_ok(); + } + if (!rr_sigs || !rr_sigs->rrs.count || !rr_sigs->rrs.rdata) { + assert(!EINVAL); + return kr_error(EINVAL); + } + const knot_dname_t *signer = knot_rrsig_signer_name(rr_sigs->rrs.rdata); + const int signer_size = knot_dname_size(signer); + k->zlf_len = signer_size - 1; + + void **npp = nsec_pmap == NULL ? NULL + : trie_get_ins(nsec_pmap, (const char *)signer, signer_size); + assert(!nsec_pmap || (npp && ENOMEM)); + if (rr->type == KNOT_RRTYPE_NSEC) { + key = key_NSEC1(k, encloser, wild_labels); + break; + } + + assert(rr->type == KNOT_RRTYPE_NSEC3); + const knot_rdata_t * const rdata = rr->rrs.rdata; + if (rdata->len <= 4) return kr_error(EILSEQ); /*< data from outside; less trust */ + const int np_dlen = nsec_p_rdlen(rdata->data); + if (np_dlen > rdata->len) return kr_error(EILSEQ); + key = key_NSEC3(k, encloser, nsec_p_mkHash(rdata->data)); + if (npp && !*npp) { + *npp = mm_alloc(&qry->request->pool, np_dlen); + if (!*npp) { + assert(!ENOMEM); + break; + } + memcpy(*npp, rdata->data, np_dlen); + } + break; + default: + ret = kr_dname_lf(k->buf, encloser, wild_labels); + if (ret) { + assert(!ret); + return kr_error(ret); + } + key = key_exact_type(k, rr->type); + } + + /* Compute materialized sizes of the new data. */ + const knot_rdataset_t *rds_sigs = rr_sigs ? &rr_sigs->rrs : NULL; + const int rr_ssize = rdataset_dematerialize_size(&rr->rrs); + assert(rr_ssize == to_even(rr_ssize)); + knot_db_val_t val_new_entry = { + .data = NULL, + .len = offsetof(struct entry_h, data) + rr_ssize + + rdataset_dematerialize_size(rds_sigs), + }; + + /* Prepare raw memory for the new entry. */ + ret = entry_h_splice(&val_new_entry, rank, key, k->type, rr->type, + rr->owner, qry, cache, timestamp); + if (ret) return kr_ok(); /* some aren't really errors */ + assert(val_new_entry.data); + + const uint32_t ttl = rr->ttl; + /* FIXME: consider TTLs and expirations of RRSIGs as well, just in case. */ + + /* Write the entry itself. */ + struct entry_h *eh = val_new_entry.data; + memset(eh, 0, offsetof(struct entry_h, data)); + eh->time = timestamp; + eh->ttl = MAX(MIN(ttl, cache->ttl_max), cache->ttl_min); + eh->rank = rank; + if (rdataset_dematerialize(&rr->rrs, eh->data) + || rdataset_dematerialize(rds_sigs, eh->data + rr_ssize)) { + /* minimize the damage from incomplete write; TODO: better */ + eh->time = 0; + eh->ttl = 0; + eh->rank = 0; + assert(false); + } + assert(entry_h_consistent(val_new_entry, rr->type)); + + #if 0 /* Occasionally useful when debugging some kinds of changes. */ + { + kr_cache_sync(cache); + knot_db_val_t val = { NULL, 0 }; + ret = cache_op(cache, read, &key, &val, 1); + if (ret != kr_error(ENOENT)) { // ENOENT might happen in some edge case, I guess + assert(!ret); + entry_list_t el; + entry_list_parse(val, el); + } + } + #endif + + /* Update metrics */ + cache->stats.insert += 1; + + /* Verbose-log some not-too-common cases. */ + WITH_VERBOSE(qry) { if (kr_rank_test(rank, KR_RANK_AUTH) + || rr->type == KNOT_RRTYPE_NS) { + auto_free char *type_str = kr_rrtype_text(rr->type), + *encl_str = kr_dname_text(encloser); + VERBOSE_MSG(qry, "=> stashed %s%s %s, rank 0%.2o, " + "%d B total, incl. %d RRSIGs\n", + (wild_labels ? "*." : ""), encl_str, type_str, rank, + (int)val_new_entry.len, (rr_sigs ? rr_sigs->rrs.count : 0) + ); + } } + + return (ssize_t) val_new_entry.len; +} + +static int stash_rrarray_entry(ranked_rr_array_t *arr, int arr_i, + const struct kr_query *qry, struct kr_cache *cache, + int *unauth_cnt, trie_t *nsec_pmap, bool *has_optout) +{ + ranked_rr_array_entry_t *entry = arr->at[arr_i]; + if (entry->cached) { + return kr_ok(); + } + const knot_rrset_t *rr = entry->rr; + if (rr->type == KNOT_RRTYPE_RRSIG) { + return kr_ok(); /* reduce verbose logging from the following call */ + } + int ret = stash_rrset_precond(rr, qry); + if (ret <= 0) { + return ret; + } + + /* Try to find corresponding signatures, always. LATER(optim.): speed. */ + ranked_rr_array_entry_t *entry_rrsigs = NULL; + const knot_rrset_t *rr_sigs = NULL; + for (ssize_t j = arr->len - 1; j >= 0; --j) { + /* TODO: ATM we assume that some properties are the same + * for all RRSIGs in the set (esp. label count). */ + ranked_rr_array_entry_t *e = arr->at[j]; + bool ok = e->qry_uid == qry->uid && !e->cached + && e->rr->type == KNOT_RRTYPE_RRSIG + && knot_rrsig_type_covered(e->rr->rrs.rdata) == rr->type + && knot_dname_is_equal(rr->owner, e->rr->owner); + if (!ok) continue; + entry_rrsigs = e; + rr_sigs = e->rr; + break; + } + + ssize_t written = stash_rrset(cache, qry, rr, rr_sigs, qry->timestamp.tv_sec, + entry->rank, nsec_pmap, has_optout); + if (written < 0) { + kr_log_error("[%05u.%02u][cach] stash failed, ret = %d\n", qry->request->uid, + qry->uid, ret); + return (int) written; + } + + if (written > 0) { + /* Mark entry as cached for the rest of the query processing */ + entry->cached = true; + if (entry_rrsigs) { + entry_rrsigs->cached = true; + } + if (!kr_rank_test(entry->rank, KR_RANK_AUTH) && rr->type != KNOT_RRTYPE_NS) { + *unauth_cnt += 1; + } + } + + return kr_ok(); +} + +static int stash_nsec_p(const knot_dname_t *dname, const char *nsec_p_v, + struct kr_request *req) +{ + const struct kr_query *qry = req->current_query; + struct kr_cache *cache = &req->ctx->cache; + uint32_t valid_until = qry->timestamp.tv_sec + cache->ttl_max; + /* LATER(optim.): be more precise here ^^ and reduce calls. */ + static const int32_t ttl_margin = 3600; + const uint8_t *nsec_p = (const uint8_t *)nsec_p_v; + int data_stride = sizeof(valid_until) + nsec_p_rdlen(nsec_p); + + unsigned int log_hash = 0xFeeeFeee; /* this type is simpler for printf args */ + auto_free char *log_dname = NULL; + WITH_VERBOSE(qry) { + log_hash = nsec_p_v ? nsec_p_mkHash((const uint8_t *)nsec_p_v) : 0; + log_dname = kr_dname_text(dname); + } + /* Find what's in the cache. */ + struct key k_storage, *k = &k_storage; + int ret = kr_dname_lf(k->buf, dname, false); + if (ret) return kr_error(ret); + knot_db_val_t key = key_exact_type(k, KNOT_RRTYPE_NS); + knot_db_val_t val_orig = { NULL, 0 }; + ret = cache_op(cache, read, &key, &val_orig, 1); + if (ret && ret != -ABS(ENOENT)) { + VERBOSE_MSG(qry, "=> EL read failed (ret: %d)\n", ret); + return kr_ok(); + } + /* Prepare new entry_list_t so we can just write at el[0]. */ + entry_list_t el; + int log_refresh_by = 0; + if (ret == -ABS(ENOENT)) { + memset(el, 0, sizeof(el)); + } else { + ret = entry_list_parse(val_orig, el); + if (ret) { + VERBOSE_MSG(qry, "=> EL parse failed (ret: %d)\n", ret); + return kr_error(0); + } + /* Find the index to replace. */ + int i_replace = ENTRY_APEX_NSECS_CNT - 1; + for (int i = 0; i < ENTRY_APEX_NSECS_CNT; ++i) { + if (el[i].len != data_stride) continue; + if (nsec_p && memcmp(nsec_p, (uint8_t *)el[i].data + sizeof(uint32_t), + data_stride - sizeof(uint32_t)) != 0) { + continue; + } + /* Save a cache operation if TTL extended only a little. */ + uint32_t valid_orig; + memcpy(&valid_orig, el[i].data, sizeof(valid_orig)); + const int32_t ttl_extended_by = valid_until - valid_orig; + if (ttl_extended_by < ttl_margin) { + VERBOSE_MSG(qry, + "=> nsec_p stash for %s skipped (extra TTL: %d, hash: %x)\n", + log_dname, ttl_extended_by, log_hash); + return kr_ok(); + } + i_replace = i; + log_refresh_by = ttl_extended_by; + break; + } + /* Shift the other indices: move the first `i_replace` blocks + * by one position. */ + if (i_replace) { + memmove(&el[1], &el[0], sizeof(el[0]) * i_replace); + } + } + /* Prepare old data into a buffer. See entry_h_splice() for why. LATER(optim.) */ + el[0].len = data_stride; + el[0].data = NULL; + knot_db_val_t val; + val.len = entry_list_serial_size(el), + val.data = mm_alloc(&req->pool, val.len), + entry_list_memcpy(val.data, el); + /* Prepare the new data chunk */ + memcpy(el[0].data, &valid_until, sizeof(valid_until)); + if (nsec_p) { + memcpy((uint8_t *)el[0].data + sizeof(valid_until), nsec_p, + data_stride - sizeof(valid_until)); + } + /* Write it all to the cache */ + ret = cache_op(cache, write, &key, &val, 1); + if (ret || !val.data) { + VERBOSE_MSG(qry, "=> EL write failed (ret: %d)\n", ret); + return kr_ok(); + } + if (log_refresh_by) { + VERBOSE_MSG(qry, "=> nsec_p stashed for %s (refresh by %d, hash: %x)\n", + log_dname, log_refresh_by, log_hash); + } else { + VERBOSE_MSG(qry, "=> nsec_p stashed for %s (new, hash: %x)\n", + log_dname, log_hash); + } + return kr_ok(); +} + + +static int peek_exact_real(struct kr_cache *cache, const knot_dname_t *name, uint16_t type, + struct kr_cache_p *peek) +{ + if (!check_rrtype(type, NULL) || !check_dname_for_lf(name, NULL)) { + return kr_error(ENOTSUP); + } + struct key k_storage, *k = &k_storage; + + int ret = kr_dname_lf(k->buf, name, false); + if (ret) return kr_error(ret); + + knot_db_val_t key = key_exact_type(k, type); + knot_db_val_t val = { NULL, 0 }; + ret = cache_op(cache, read, &key, &val, 1); + if (!ret) ret = entry_h_seek(&val, type); + if (ret) return kr_error(ret); + + const struct entry_h *eh = entry_h_consistent(val, type); + if (!eh || eh->is_packet) { + // TODO: no packets, but better get rid of whole kr_cache_peek_exact(). + return kr_error(ENOENT); + } + *peek = (struct kr_cache_p){ + .time = eh->time, + .ttl = eh->ttl, + .rank = eh->rank, + .raw_data = val.data, + .raw_bound = knot_db_val_bound(val), + }; + return kr_ok(); +} +int kr_cache_peek_exact(struct kr_cache *cache, const knot_dname_t *name, uint16_t type, + struct kr_cache_p *peek) +{ /* Just wrap with extra verbose logging. */ + const int ret = peek_exact_real(cache, name, type, peek); + if (false && VERBOSE_STATUS) { /* too noisy for usual --verbose */ + auto_free char *type_str = kr_rrtype_text(type), + *name_str = kr_dname_text(name); + const char *result_str = (ret == kr_ok() ? "hit" : + (ret == kr_error(ENOENT) ? "miss" : "error")); + VERBOSE_MSG(NULL, "_peek_exact: %s %s %s (ret: %d)", + type_str, name_str, result_str, ret); + } + return ret; +} + +int kr_cache_remove(struct kr_cache *cache, const knot_dname_t *name, uint16_t type) +{ + if (!cache_isvalid(cache)) { + return kr_error(EINVAL); + } + if (!cache->api->remove) { + return kr_error(ENOSYS); + } + struct key k_storage, *k = &k_storage; + int ret = kr_dname_lf(k->buf, name, false); + if (ret) return kr_error(ret); + + knot_db_val_t key = key_exact_type(k, type); + return cache_op(cache, remove, &key, 1); +} + +int kr_cache_match(struct kr_cache *cache, const knot_dname_t *name, + bool exact_name, knot_db_val_t keyval[][2], int maxcount) +{ + if (!cache_isvalid(cache)) { + return kr_error(EINVAL); + } + if (!cache->api->match) { + return kr_error(ENOSYS); + } + + struct key k_storage, *k = &k_storage; + + int ret = kr_dname_lf(k->buf, name, false); + if (ret) return kr_error(ret); + + // use a mock type + knot_db_val_t key = key_exact_type(k, KNOT_RRTYPE_A); + /* CACHE_KEY_DEF */ + key.len -= sizeof(uint16_t); /* the type */ + if (!exact_name) { + key.len -= 2; /* '\0' 'E' */ + if (name[0] == '\0') ++key.len; /* the root name is special ATM */ + } + return cache_op(cache, match, &key, keyval, maxcount); +} + +int kr_unpack_cache_key(knot_db_val_t key, knot_dname_t *buf, uint16_t *type) +{ + if (key.data == NULL || buf == NULL || type == NULL) { + return kr_error(EINVAL); + } + + int len = -1; + const char *tag, *key_data = key.data; + for (tag = key_data + 1; tag < key_data + key.len; ++tag) { + /* CACHE_KEY_DEF */ + if (tag[-1] == '\0' && (tag == key_data + 1 || tag[-2] == '\0')) { + if (tag[0] != 'E') return kr_error(EINVAL); + len = tag - 1 - key_data; + break; + } + } + + if (len == -1 || len > KNOT_DNAME_MAXLEN) { + return kr_error(EINVAL); + } + + int ret = knot_dname_lf2wire(buf, len, key.data); + if (ret < 0) { + return kr_error(ret); + } + + /* CACHE_KEY_DEF: jump over "\0 E/1" */ + memcpy(type, tag + 1, sizeof(uint16_t)); + + return kr_ok(); +} + + +int kr_cache_remove_subtree(struct kr_cache *cache, const knot_dname_t *name, + bool exact_name, int maxcount) +{ + if (!cache_isvalid(cache)) { + return kr_error(EINVAL); + } + + knot_db_val_t keyval[maxcount][2], keys[maxcount]; + int ret = kr_cache_match(cache, name, exact_name, keyval, maxcount); + if (ret <= 0) { /* ENOENT -> nothing to remove */ + return (ret == KNOT_ENOENT) ? 0 : ret; + } + const int count = ret; + /* Duplicate the key strings, as deletion may invalidate the pointers. */ + int i; + for (i = 0; i < count; ++i) { + keys[i].len = keyval[i][0].len; + keys[i].data = malloc(keys[i].len); + if (!keys[i].data) { + ret = kr_error(ENOMEM); + goto cleanup; + } + memcpy(keys[i].data, keyval[i][0].data, keys[i].len); + } + ret = cache->api->remove(cache->db, keys, count); +cleanup: + kr_cache_sync(cache); /* Sync even after just kr_cache_match(). */ + /* Free keys */ + while (--i >= 0) { + free(keys[i].data); + } + return ret; +} + diff --git a/lib/cache/api.h b/lib/cache/api.h new file mode 100644 index 0000000..61bf796 --- /dev/null +++ b/lib/cache/api.h @@ -0,0 +1,201 @@ +/* Copyright (C) 2014-2017 CZ.NIC, z.s.p.o. + + This program is free software: you can redistribute it and/or modify + it under the terms of the GNU General Public License as published by + the Free Software Foundation, either version 3 of the License, or + (at your option) any later version. + + This program is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the + GNU General Public License for more details. + + You should have received a copy of the GNU General Public License + along with this program. If not, see . + */ + +#pragma once + +#include +#include +#include +#include "lib/cache/cdb_api.h" +#include "lib/defines.h" +#include "contrib/ucw/config.h" /*uint*/ + +/** When knot_pkt is passed from cache without ->wire, this is the ->size. */ +static const size_t PKT_SIZE_NOWIRE = -1; + + +#include "lib/module.h" +/* Prototypes for the 'cache' module implementation. */ +int cache_peek(kr_layer_t *ctx, knot_pkt_t *pkt); +int cache_stash(kr_layer_t *ctx, knot_pkt_t *pkt); + + +/** + * Cache structure, keeps API, instance and metadata. + */ +struct kr_cache +{ + knot_db_t *db; /**< Storage instance */ + const struct kr_cdb_api *api; /**< Storage engine */ + struct { + uint32_t hit; /**< Number of cache hits */ + uint32_t miss; /**< Number of cache misses */ + uint32_t insert; /**< Number of insertions */ + uint32_t delete; /**< Number of deletions */ + } stats; + + uint32_t ttl_min, ttl_max; /**< TTL limits */ + + /* A pair of stamps for detection of real-time shifts during runtime. */ + struct timeval checkpoint_walltime; /**< Wall time on the last check-point. */ + uint64_t checkpoint_monotime; /**< Monotonic milliseconds on the last check-point. */ +}; + +/** + * Open/create cache with provided storage options. + * @param cache cache structure to be initialized + * @param api storage engine API + * @param opts storage-specific options (may be NULL for default) + * @param mm memory context. + * @return 0 or an error code + */ +KR_EXPORT +int kr_cache_open(struct kr_cache *cache, const struct kr_cdb_api *api, struct kr_cdb_opts *opts, knot_mm_t *mm); + +/** + * Path to cache file to remove on critical out-of-space error. (do NOT modify it) + */ +KR_EXPORT extern +const char *kr_cache_emergency_file_to_remove; + +/** + * Close persistent cache. + * @note This doesn't clear the data, just closes the connection to the database. + * @param cache structure + */ +KR_EXPORT +void kr_cache_close(struct kr_cache *cache); + +/** Run after a row of operations to release transaction/lock if needed. */ +KR_EXPORT +int kr_cache_sync(struct kr_cache *cache); + +/** + * Return true if cache is open and enabled. + */ +static inline bool kr_cache_is_open(struct kr_cache *cache) +{ + return cache->db != NULL; +} + +/** (Re)set the time pair to the current values. */ +static inline void kr_cache_make_checkpoint(struct kr_cache *cache) +{ + cache->checkpoint_monotime = kr_now(); + gettimeofday(&cache->checkpoint_walltime, NULL); +} + +/** + * Insert RRSet into cache, replacing any existing data. + * @param cache cache structure + * @param rr inserted RRSet + * @param rrsig RRSIG for inserted RRSet (optional) + * @param rank rank of the data + * @param timestamp current time + * @return 0 or an errcode + */ +KR_EXPORT +int kr_cache_insert_rr(struct kr_cache *cache, const knot_rrset_t *rr, const knot_rrset_t *rrsig, uint8_t rank, uint32_t timestamp); + +/** + * Clear all items from the cache. + * @param cache cache structure + * @return 0 or an errcode + */ +KR_EXPORT +int kr_cache_clear(struct kr_cache *cache); + + +/* ** This interface is temporary. ** */ + +struct kr_cache_p { + uint32_t time; /**< The time of inception. */ + uint32_t ttl; /**< TTL at inception moment. Assuming it fits into int32_t ATM. */ + uint8_t rank; /**< See enum kr_rank */ + struct { + /* internal: pointer to eh struct */ + void *raw_data, *raw_bound; + }; +}; +KR_EXPORT +int kr_cache_peek_exact(struct kr_cache *cache, const knot_dname_t *name, uint16_t type, + struct kr_cache_p *peek); +/* Parameters (qry, name, type) are used for timestamp and stale-serving decisions. */ +KR_EXPORT +int32_t kr_cache_ttl(const struct kr_cache_p *peek, const struct kr_query *qry, + const knot_dname_t *name, uint16_t type); + +KR_EXPORT +int kr_cache_materialize(knot_rdataset_t *dst, const struct kr_cache_p *ref, + knot_mm_t *pool); + + +/** + * Remove an entry from cache. + * @param cache cache structure + * @param name dname + * @param type rr type + * @return number of deleted records, or negative error code + * @note only "exact hits" are considered ATM, and + * some other information may be removed alongside. + */ +KR_EXPORT +int kr_cache_remove(struct kr_cache *cache, const knot_dname_t *name, uint16_t type); + +/** + * Get keys matching a dname lf prefix + * @param cache cache structure + * @param name dname + * @param exact_name whether to only consider exact name matches + * @param keyval matched key-value pairs + * @param maxcount limit on the number of returned key-value pairs + * @return result count or an errcode + * @note the cache keys are matched by prefix, i.e. it very much depends + * on their structure; CACHE_KEY_DEF. + */ +KR_EXPORT +int kr_cache_match(struct kr_cache *cache, const knot_dname_t *name, + bool exact_name, knot_db_val_t keyval[][2], int maxcount); + +/** + * Remove a subtree in cache. It's like _match but removing them instead of returning. + * @return number of deleted entries or an errcode + */ +KR_EXPORT +int kr_cache_remove_subtree(struct kr_cache *cache, const knot_dname_t *name, + bool exact_name, int maxcount); + +/** + * Find the closest cached zone apex for a name (in cache). + * @param is_DS start searching one name higher + * @return the number of labels to remove from the name, or negative error code + * @note timestamp is found by a syscall, and stale-serving is not considered + */ +KR_EXPORT +int kr_cache_closest_apex(struct kr_cache *cache, const knot_dname_t *name, bool is_DS, + knot_dname_t **apex); + +/** + * Unpack dname and type from db key + * @param key db key representation + * @param buf output buffer of domain name in dname format + * @param type output for type + * @return length of dname or an errcode + * @note only "exact hits" are considered ATM, moreover xNAME records + * are "hidden" as NS. (see comments in struct entry_h) + */ +KR_EXPORT +int kr_unpack_cache_key(knot_db_val_t key, knot_dname_t *buf, uint16_t *type); diff --git a/lib/cache/cdb_api.h b/lib/cache/cdb_api.h new file mode 100644 index 0000000..814a5f5 --- /dev/null +++ b/lib/cache/cdb_api.h @@ -0,0 +1,68 @@ +/* Copyright (C) 2016-2017 CZ.NIC, z.s.p.o. + + This program is free software: you can redistribute it and/or modify + it under the terms of the GNU General Public License as published by + the Free Software Foundation, either version 3 of the License, or + (at your option) any later version. + + This program is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the + GNU General Public License for more details. + + You should have received a copy of the GNU General Public License + along with this program. If not, see . +*/ + +#pragma once + +#include + +/* Cache options. */ +struct kr_cdb_opts { + const char *path; /*!< Cache URI path. */ + size_t maxsize; /*!< Suggested cache size in bytes. */ +}; + +/*! Cache database API. + * This is a simplified version of generic DB API from libknot, + * that is tailored to caching purposes. + */ +struct kr_cdb_api { + const char *name; + + /* Context operations */ + + int (*open)(knot_db_t **db, struct kr_cdb_opts *opts, knot_mm_t *mm); + void (*close)(knot_db_t *db); + int (*count)(knot_db_t *db); + int (*clear)(knot_db_t *db); + + /** Run after a row of operations to release transaction/lock if needed. */ + int (*sync)(knot_db_t *db); + + /* Data access */ + + int (*read)(knot_db_t *db, const knot_db_val_t *key, knot_db_val_t *val, + int maxcount); + int (*write)(knot_db_t *db, const knot_db_val_t *key, knot_db_val_t *val, + int maxcount); + + /** Remove maxcount keys. + * \returns the number of succesfully removed keys or the first error code + * It returns on first error, but ENOENT is not considered an error. */ + int (*remove)(knot_db_t *db, knot_db_val_t keys[], int maxcount); + + /* Specialised operations */ + + /** Find key-value pairs that are prefixed by the given key, limited by maxcount. + * \return the number of pairs or negative error. */ + int (*match)(knot_db_t *db, knot_db_val_t *key, knot_db_val_t keyval[][2], int maxcount); + /** Not implemented ATM. */ + int (*prune)(knot_db_t *db, int maxcount); + + /** Less-or-equal search (lexicographic ordering). + * On successful return, key->data and val->data point to DB-owned data. + * return: 0 for equality, > 0 for less, < 0 kr_error */ + int (*read_leq)(knot_db_t *db, knot_db_val_t *key, knot_db_val_t *val); +}; diff --git a/lib/cache/cdb_lmdb.c b/lib/cache/cdb_lmdb.c new file mode 100644 index 0000000..621d2e8 --- /dev/null +++ b/lib/cache/cdb_lmdb.c @@ -0,0 +1,710 @@ +/* Copyright (C) 2016-2017 CZ.NIC, z.s.p.o. + + This program is free software: you can redistribute it and/or modify + it under the terms of the GNU General Public License as published by + the Free Software Foundation, either version 3 of the License, or + (at your option) any later version. + + This program is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the + GNU General Public License for more details. + + You should have received a copy of the GNU General Public License + along with this program. If not, see . +*/ + +#include +#include +#include +#include +#include +#include +#include +#include +#include + +#include "contrib/cleanup.h" +#include "lib/cache/cdb_lmdb.h" +#include "lib/cache/api.h" +#include "lib/utils.h" + + +/* Defines */ +#define LMDB_DIR_MODE 0770 +#define LMDB_FILE_MODE 0660 + +struct lmdb_env +{ + size_t mapsize; + MDB_dbi dbi; + MDB_env *env; + + /** Cached transactions + * + * - only one of (ro,rw) may be active at once + * - non-NULL .ro may be active or reset + * - non-NULL .rw is always active + */ + struct { + bool ro_active, ro_curs_active; + MDB_txn *ro, *rw; + MDB_cursor *ro_curs; + } txn; +}; + +/** @brief Convert LMDB error code. */ +static int lmdb_error(int error) +{ + /* _BAD_TXN may happen with overfull DB, + * even during mdb_get with a single fork :-/ */ + if (error == MDB_BAD_TXN) { + kr_log_info("[cache] MDB_BAD_TXN, probably overfull\n"); + error = ENOSPC; + } + switch (error) { + case MDB_SUCCESS: + return kr_ok(); + case MDB_NOTFOUND: + return kr_error(ENOENT); + case ENOSPC: + case MDB_MAP_FULL: + case MDB_TXN_FULL: + return kr_error(ENOSPC); + default: + kr_log_error("[cache] LMDB error: %s\n", mdb_strerror(error)); + return kr_error(error); + } +} + +/** Conversion between knot and lmdb structs for values. */ +static inline knot_db_val_t val_mdb2knot(MDB_val v) +{ + return (knot_db_val_t){ .len = v.mv_size, .data = v.mv_data }; +} +static inline MDB_val val_knot2mdb(knot_db_val_t v) +{ + return (MDB_val){ .mv_size = v.len, .mv_data = v.data }; +} + + +/*! \brief Set the environment map size. + * \note This also sets the maximum database size, see \fn mdb_env_set_mapsize + */ +static int set_mapsize(MDB_env *env, size_t map_size) +{ + long page_size = sysconf(_SC_PAGESIZE); + if (page_size <= 0) { + return KNOT_ERROR; + } + + /* Round to page size. */ + map_size = (map_size / page_size) * page_size; + int ret = mdb_env_set_mapsize(env, map_size); + if (ret != MDB_SUCCESS) { + return lmdb_error(ret); + } + + return 0; +} + +#define FLAG_RENEW (2*MDB_RDONLY) +/** mdb_txn_begin or _renew + handle MDB_MAP_RESIZED. + * + * The retrying logic for MDB_MAP_RESIZED is so ugly that it has its own function. + * \note this assumes no transactions are active + * \return MDB_ errcode, not usual kr_error(...) + */ +static int txn_get_noresize(struct lmdb_env *env, unsigned int flag, MDB_txn **txn) +{ + assert(!env->txn.rw && (!env->txn.ro || !env->txn.ro_active)); + int ret; + if (flag == FLAG_RENEW) { + ret = mdb_txn_renew(*txn); + } else { + ret = mdb_txn_begin(env->env, NULL, flag, txn); + } + if (ret != MDB_MAP_RESIZED) { + return ret; + } + //:unlikely + /* Another process increased the size; let's try to recover. */ + kr_log_info("[cache] detected size increased by another process\n"); + ret = mdb_env_set_mapsize(env->env, 0); + if (ret != MDB_SUCCESS) { + return ret; + } + if (flag == FLAG_RENEW) { + ret = mdb_txn_renew(*txn); + } else { + ret = mdb_txn_begin(env->env, NULL, flag, txn); + } + return ret; +} + +/** Obtain a transaction. (they're cached in env->txn) */ +static int txn_get(struct lmdb_env *env, MDB_txn **txn, bool rdonly) +{ + assert(env && txn); + if (env->txn.rw) { + /* Reuse the *open* RW txn even if only reading is requested. + * We leave the management of this to the cdb_sync command. + * The user may e.g. want to do some reads between the writes. */ + *txn = env->txn.rw; + return kr_ok(); + } + + if (!rdonly) { + /* avoid two active transactions */ + if (env->txn.ro && env->txn.ro_active) { + mdb_txn_reset(env->txn.ro); + env->txn.ro_active = false; + env->txn.ro_curs_active = false; + } + int ret = txn_get_noresize(env, 0/*RW*/, &env->txn.rw); + if (ret == MDB_SUCCESS) { + *txn = env->txn.rw; + assert(*txn); + } + return lmdb_error(ret); + } + + /* Get an active RO txn and return it. */ + int ret = MDB_SUCCESS; + if (!env->txn.ro) { //:unlikely + ret = txn_get_noresize(env, MDB_RDONLY, &env->txn.ro); + } else if (!env->txn.ro_active) { + ret = txn_get_noresize(env, FLAG_RENEW, &env->txn.ro); + } + if (ret != MDB_SUCCESS) { + return lmdb_error(ret); + } + env->txn.ro_active = true; + *txn = env->txn.ro; + assert(*txn); + return kr_ok(); +} + +static int cdb_sync(knot_db_t *db) +{ + struct lmdb_env *env = db; + int ret = kr_ok(); + if (env->txn.rw) { + ret = lmdb_error(mdb_txn_commit(env->txn.rw)); + env->txn.rw = NULL; /* the transaction got freed even in case of errors */ + } else if (env->txn.ro && env->txn.ro_active) { + mdb_txn_reset(env->txn.ro); + env->txn.ro_active = false; + env->txn.ro_curs_active = false; + } + return ret; +} + +/** Obtain a read-only cursor (and a read-only transaction). */ +static int txn_curs_get(struct lmdb_env *env, MDB_cursor **curs) +{ + assert(env && curs); + if (env->txn.ro_curs_active) { + goto success; + } + /* Only in a read-only txn; TODO: it's a bit messy/coupled */ + if (env->txn.rw) { + int ret = cdb_sync(env); + if (ret) return ret; + } + MDB_txn *txn = NULL; + int ret = txn_get(env, &txn, true); + if (ret) return ret; + + if (env->txn.ro_curs) { + ret = mdb_cursor_renew(txn, env->txn.ro_curs); + } else { + ret = mdb_cursor_open(txn, env->dbi, &env->txn.ro_curs); + } + if (ret) return ret; + env->txn.ro_curs_active = true; +success: + assert(env->txn.ro_curs_active && env->txn.ro && env->txn.ro_active + && !env->txn.rw); + *curs = env->txn.ro_curs; + assert(*curs); + return kr_ok(); +} + +static void free_txn_ro(struct lmdb_env *env) +{ + if (env->txn.ro) { + mdb_txn_abort(env->txn.ro); + env->txn.ro = NULL; + } + if (env->txn.ro_curs) { + mdb_cursor_close(env->txn.ro_curs); + env->txn.ro_curs = NULL; + } +} + +/*! \brief Close the database. */ +static void cdb_close_env(struct lmdb_env *env) +{ + assert(env && env->env); + + /* Get rid of any transactions. */ + cdb_sync(env); + free_txn_ro(env); + + mdb_env_sync(env->env, 1); + mdb_dbi_close(env->env, env->dbi); + mdb_env_close(env->env); + memset(env, 0, sizeof(*env)); +} + +/*! \brief Open database environment. */ +static int cdb_open_env(struct lmdb_env *env, unsigned flags, const char *path, size_t mapsize) +{ + int ret = mkdir(path, LMDB_DIR_MODE); + if (ret == -1 && errno != EEXIST) { + return kr_error(errno); + } + + MDB_env *mdb_env = NULL; + ret = mdb_env_create(&mdb_env); + if (ret != MDB_SUCCESS) { + return lmdb_error(ret); + } + + ret = set_mapsize(mdb_env, mapsize); + if (ret != 0) { + mdb_env_close(mdb_env); + return ret; + } + + ret = mdb_env_open(mdb_env, path, flags, LMDB_FILE_MODE); + if (ret != MDB_SUCCESS) { + mdb_env_close(mdb_env); + return lmdb_error(ret); + } + + /* Keep the environment pointer. */ + env->env = mdb_env; + env->mapsize = mapsize; + return 0; +} + +static int cdb_open(struct lmdb_env *env, const char *path, size_t mapsize) +{ + /* Cache doesn't require durability, we can be + * loose with the requirements as a tradeoff for speed. */ + const unsigned flags = MDB_WRITEMAP | MDB_MAPASYNC | MDB_NOTLS; + int ret = cdb_open_env(env, flags, path, mapsize); + if (ret != 0) { + return ret; + } + + /* Open the database. */ + MDB_txn *txn = NULL; + ret = mdb_txn_begin(env->env, NULL, 0, &txn); + if (ret != MDB_SUCCESS) { + mdb_env_close(env->env); + return lmdb_error(ret); + } + + ret = mdb_dbi_open(txn, NULL, 0, &env->dbi); + if (ret != MDB_SUCCESS) { + mdb_txn_abort(txn); + mdb_env_close(env->env); + return lmdb_error(ret); + } + + ret = mdb_txn_commit(txn); + if (ret != MDB_SUCCESS) { + mdb_env_close(env->env); + return lmdb_error(ret); + } + + return 0; +} + +static int cdb_init(knot_db_t **db, struct kr_cdb_opts *opts, knot_mm_t *pool) +{ + if (!db || !opts) { + return kr_error(EINVAL); + } + + struct lmdb_env *env = malloc(sizeof(*env)); + if (!env) { + return kr_error(ENOMEM); + } + memset(env, 0, sizeof(struct lmdb_env)); + + /* Clear stale lockfiles. */ + auto_free char *lockfile = kr_strcatdup(2, opts->path, "/.cachelock"); + if (lockfile) { + if (unlink(lockfile) == 0) { + kr_log_info("[cache] cleared stale lockfile '%s'\n", lockfile); + } else if (errno != ENOENT) { + kr_log_info("[cache] failed to clear stale lockfile '%s': %s\n", lockfile, + strerror(errno)); + } + } + + /* Open the database. */ + int ret = cdb_open(env, opts->path, opts->maxsize); + if (ret != 0) { + free(env); + return ret; + } + + *db = env; + return 0; +} + +static void cdb_deinit(knot_db_t *db) +{ + struct lmdb_env *env = db; + cdb_close_env(env); + free(env); +} + +static int cdb_count(knot_db_t *db) +{ + struct lmdb_env *env = db; + MDB_txn *txn = NULL; + int ret = txn_get(env, &txn, true); + if (ret != 0) { + return ret; + } + + MDB_stat stat; + ret = mdb_stat(txn, env->dbi, &stat); + + return (ret == MDB_SUCCESS) ? stat.ms_entries : lmdb_error(ret); +} + +static int cdb_clear(knot_db_t *db) +{ + struct lmdb_env *env = db; + /* First try mdb_drop() to clear the DB; this may fail with ENOSPC. */ + /* If we didn't do this, explicit cache.clear() ran on an instance + * would lead to the instance detaching from the cache of others, + * until they reopened cache explicitly or cleared it for some reason. + */ + { + MDB_txn *txn = NULL; + int ret = txn_get(env, &txn, false); + if (ret == kr_ok()) { + ret = lmdb_error(mdb_drop(txn, env->dbi, 0)); + if (ret == kr_ok()) { + ret = cdb_sync(db); + } + if (ret == kr_ok()) { + return ret; + } + } + kr_log_info("[cache] clearing error, falling back\n"); + } + + /* We are about to switch to a different file, so end all txns, to be sure. */ + (void) cdb_sync(db); + free_txn_ro(db); + + /* Since there is no guarantee that there will be free + * pages to hold whole dirtied db for transaction-safe clear, + * we simply remove the database files and reopen. + * We can afford this since other readers will continue to read + * from removed file, but will reopen when encountering next + * error. */ + mdb_filehandle_t fd = -1; + int ret = mdb_env_get_fd(env->env, &fd); + if (ret != MDB_SUCCESS) { + return lmdb_error(ret); + } + const char *path = NULL; + ret = mdb_env_get_path(env->env, &path); + if (ret != MDB_SUCCESS) { + return lmdb_error(ret); + } + + auto_free char *mdb_datafile = kr_strcatdup(2, path, "/data.mdb"); + auto_free char *mdb_lockfile = kr_strcatdup(2, path, "/lock.mdb"); + auto_free char *lockfile = kr_strcatdup(2, path, "/.cachelock"); + if (!mdb_datafile || !mdb_lockfile || !lockfile) { + return kr_error(ENOMEM); + } + /* Find if we get a lock on lockfile. */ + ret = open(lockfile, O_CREAT|O_EXCL|O_RDONLY, S_IRUSR); + if (ret == -1) { + kr_log_error("[cache] clearing failed to get ./.cachelock; retry later\n"); + /* As we're out of space (almost certainly - mdb_drop didn't work), + * we will retry on the next failing write operation. */ + return kr_error(errno); + } + close(ret); + /* We acquired lockfile. Now find whether *.mdb are what we have open now. */ + struct stat old_stat, new_stat; + if (fstat(fd, &new_stat) || stat(mdb_datafile, &old_stat)) { + ret = errno; + unlink(lockfile); + return kr_error(ret); + } + /* Remove underlying files only if current open environment + * points to file on the disk. Otherwise just reopen as someone + * else has already removed the files. + */ + if (old_stat.st_dev == new_stat.st_dev && old_stat.st_ino == new_stat.st_ino) { + kr_log_verbose("[cache] clear: identical files, unlinking\n"); + // coverity[toctou] + unlink(mdb_datafile); + unlink(mdb_lockfile); + } else + kr_log_verbose("[cache] clear: not identical files, reopening\n"); + /* Keep copy as it points to current handle internals. */ + auto_free char *path_copy = strdup(path); + size_t mapsize = env->mapsize; + cdb_close_env(env); + ret = cdb_open(env, path_copy, mapsize); + /* Environment updated, release lockfile. */ + unlink(lockfile); + return ret; +} + +static int cdb_readv(knot_db_t *db, const knot_db_val_t *key, knot_db_val_t *val, + int maxcount) +{ + struct lmdb_env *env = db; + MDB_txn *txn = NULL; + int ret = txn_get(env, &txn, true); + if (ret) { + return ret; + } + + for (int i = 0; i < maxcount; ++i) { + /* Convert key structs */ + MDB_val _key = val_knot2mdb(key[i]); + MDB_val _val = val_knot2mdb(val[i]); + ret = mdb_get(txn, env->dbi, &_key, &_val); + if (ret != MDB_SUCCESS) { + ret = lmdb_error(ret); + if (ret == kr_error(ENOSPC)) { + /* we're likely to be forced to cache clear anyway */ + ret = kr_error(ENOENT); + } + return ret; + } + /* Update the result. */ + val[i] = val_mdb2knot(_val); + } + return kr_ok(); +} + +static int cdb_write(struct lmdb_env *env, MDB_txn **txn, const knot_db_val_t *key, + knot_db_val_t *val, unsigned flags) +{ + /* Convert key structs and write */ + MDB_val _key = val_knot2mdb(*key); + MDB_val _val = val_knot2mdb(*val); + int ret = mdb_put(*txn, env->dbi, &_key, &_val, flags); + + /* Try to recover from doing too much writing in a single transaction. */ + if (ret == MDB_TXN_FULL) { + ret = cdb_sync(env); + if (ret) { + ret = txn_get(env, txn, false); + } + if (ret) { + ret = mdb_put(*txn, env->dbi, &_key, &_val, flags); + } + } + if (ret != MDB_SUCCESS) { + return lmdb_error(ret); + } + + /* Update the result. */ + val->data = _val.mv_data; + val->len = _val.mv_size; + return kr_ok(); +} + +static int cdb_writev(knot_db_t *db, const knot_db_val_t *key, knot_db_val_t *val, + int maxcount) +{ + struct lmdb_env *env = db; + MDB_txn *txn = NULL; + int ret = txn_get(env, &txn, false); + + for (int i = 0; ret == kr_ok() && i < maxcount; ++i) { + /* This is LMDB specific optimisation, + * if caller specifies value with NULL data and non-zero length, + * LMDB will preallocate the entry for caller and leave write + * transaction open, caller is responsible for syncing thus committing transaction. + */ + unsigned mdb_flags = 0; + if (val[i].len > 0 && val[i].data == NULL) { + mdb_flags |= MDB_RESERVE; + } + ret = cdb_write(env, &txn, &key[i], &val[i], mdb_flags); + } + + return ret; +} + +static int cdb_remove(knot_db_t *db, knot_db_val_t keys[], int maxcount) +{ + struct lmdb_env *env = db; + MDB_txn *txn = NULL; + int ret = txn_get(env, &txn, false); + int deleted = 0; + + for (int i = 0; ret == kr_ok() && i < maxcount; ++i) { + MDB_val _key = val_knot2mdb(keys[i]); + MDB_val val = { 0, NULL }; + ret = lmdb_error(mdb_del(txn, env->dbi, &_key, &val)); + if (ret == kr_ok()) + deleted++; + else if (ret == KNOT_ENOENT) + ret = kr_ok(); /* skip over non-existing entries */ + } + + return ret < 0 ? ret : deleted; +} + +static int cdb_match(knot_db_t *db, knot_db_val_t *key, knot_db_val_t keyval[][2], int maxcount) +{ + struct lmdb_env *env = db; + MDB_txn *txn = NULL; + int ret = txn_get(env, &txn, true); + if (ret != 0) { + return ret; + } + + /* LATER(optim.): use txn_curs_get() instead, to save resources. */ + MDB_cursor *cur = NULL; + ret = mdb_cursor_open(txn, env->dbi, &cur); + if (ret != 0) { + return lmdb_error(ret); + } + + MDB_val cur_key = val_knot2mdb(*key); + MDB_val cur_val = { 0, NULL }; + ret = mdb_cursor_get(cur, &cur_key, &cur_val, MDB_SET_RANGE); + if (ret != MDB_SUCCESS) { + mdb_cursor_close(cur); + return lmdb_error(ret); + } + + int results = 0; + while (ret == MDB_SUCCESS) { + /* Retrieve current key and compare with prefix */ + if (cur_key.mv_size < key->len || memcmp(cur_key.mv_data, key->data, key->len) != 0) { + break; + } + /* Add to result set */ + if (results < maxcount) { + keyval[results][0] = val_mdb2knot(cur_key); + keyval[results][1] = val_mdb2knot(cur_val); + ++results; + } else { + break; + } + ret = mdb_cursor_get(cur, &cur_key, &cur_val, MDB_NEXT); + } + + mdb_cursor_close(cur); + return results; +} + + +static int cdb_prune(knot_db_t *db, int limit) +{ + return -1; +#if 0 + /* Sync in-flight transactions */ + cdb_sync(db); + + /* Prune old records */ + struct lmdb_env *env = db; + MDB_txn *txn = NULL; + int ret = txn_get(env, &txn, false); + if (ret != 0) { + return ret; + } + + MDB_cursor *cur = NULL; + ret = mdb_cursor_open(txn, env->dbi, &cur); + if (ret != 0) { + return lmdb_error(ret); + } + + MDB_val cur_key, cur_val; + ret = mdb_cursor_get(cur, &cur_key, &cur_val, MDB_FIRST); + if (ret != 0) { + mdb_cursor_close(cur); + return lmdb_error(ret); + } + + int results = 0; + struct timeval now; + gettimeofday(&now, NULL); + while (ret == 0 && results < limit) { + /* Ignore special namespaces. */ + if (cur_key.mv_size < 2 || ((const char *)cur_key.mv_data)[0] == 'V') { + ret = mdb_cursor_get(cur, &cur_key, &cur_val, MDB_NEXT); + continue; + } + /* Check entry age. */ + struct kr_cache_entry *entry = cur_val.mv_data; + if (entry->timestamp > now.tv_sec || + (now.tv_sec - entry->timestamp) < entry->ttl) { + ret = mdb_cursor_get(cur, &cur_key, &cur_val, MDB_NEXT); + continue; + } + /* Remove entry */ + mdb_cursor_del(cur, 0); + ++results; + ret = mdb_cursor_get(cur, &cur_key, &cur_val, MDB_NEXT); + } + mdb_cursor_close(cur); + return ret < 0 ? ret : results; +#endif +} + +static int cdb_read_leq(knot_db_t *env, knot_db_val_t *key, knot_db_val_t *val) +{ + assert(env && key && key->data && val); + MDB_cursor *curs = NULL; + int ret = txn_curs_get(env, &curs); + if (ret) return ret; + + MDB_val key2_m = val_knot2mdb(*key); + MDB_val val2_m = { 0, NULL }; + ret = mdb_cursor_get(curs, &key2_m, &val2_m, MDB_SET_RANGE); + if (ret) return lmdb_error(ret); + /* test for equality //:unlikely */ + if (key2_m.mv_size == key->len + && memcmp(key2_m.mv_data, key->data, key->len) == 0) { + ret = 0; /* equality */ + goto success; + } + /* we must be greater than key; do one step to smaller */ + ret = mdb_cursor_get(curs, &key2_m, &val2_m, MDB_PREV); + if (ret) return lmdb_error(ret); + ret = 1; +success: + /* finalize the output */ + *key = val_mdb2knot(key2_m); + *val = val_mdb2knot(val2_m); + return ret; +} + + +const struct kr_cdb_api *kr_cdb_lmdb(void) +{ + static const struct kr_cdb_api api = { + "lmdb", + cdb_init, cdb_deinit, cdb_count, cdb_clear, cdb_sync, + cdb_readv, cdb_writev, cdb_remove, + cdb_match, cdb_prune, + cdb_read_leq + }; + + return &api; +} diff --git a/lib/cache/cdb_lmdb.h b/lib/cache/cdb_lmdb.h new file mode 100644 index 0000000..bc2c7d6 --- /dev/null +++ b/lib/cache/cdb_lmdb.h @@ -0,0 +1,23 @@ +/* Copyright (C) 2016-2017 CZ.NIC, z.s.p.o. + + This program is free software: you can redistribute it and/or modify + it under the terms of the GNU General Public License as published by + the Free Software Foundation, either version 3 of the License, or + (at your option) any later version. + + This program is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the + GNU General Public License for more details. + + You should have received a copy of the GNU General Public License + along with this program. If not, see . +*/ + +#pragma once + +#include "lib/cache/cdb_api.h" +#include "lib/defines.h" + +KR_EXPORT KR_CONST +const struct kr_cdb_api *kr_cdb_lmdb(void); diff --git a/lib/cache/entry_list.c b/lib/cache/entry_list.c new file mode 100644 index 0000000..6a5001c --- /dev/null +++ b/lib/cache/entry_list.c @@ -0,0 +1,293 @@ +/* Copyright (C) 2017 CZ.NIC, z.s.p.o. + + This program is free software: you can redistribute it and/or modify + it under the terms of the GNU General Public License as published by + the Free Software Foundation, either version 3 of the License, or + (at your option) any later version. + + This program is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the + GNU General Public License for more details. + + You should have received a copy of the GNU General Public License + along with this program. If not, see . + */ + +/** @file + * Implementation of chaining in struct entry_h. Prototypes in ./impl.h + */ + +#include "lib/cache/impl.h" +#include "lib/utils.h" + + +static int entry_h_len(knot_db_val_t val); + + +void entry_list_memcpy(struct entry_apex *ea, entry_list_t list) +{ + assert(ea); + memset(ea, 0, offsetof(struct entry_apex, data)); + ea->has_ns = list[EL_NS ].len; + ea->has_cname = list[EL_CNAME ].len; + ea->has_dname = list[EL_DNAME ].len; + for (int i = 0; i < ENTRY_APEX_NSECS_CNT; ++i) { + ea->nsecs[i] = list[i].len == 0 ? 0 : + (list[i].len == 4 ? 1 : 3); + } + uint8_t *it = ea->data; + for (int i = 0; i < EL_LENGTH; ++i) { + if (list[i].data) { + memcpy(it, list[i].data, list[i].len); + /* LATER(optim.): coalesce consecutive writes? */ + } else { + list[i].data = it; + } + it += to_even(list[i].len); + } +} + +int entry_list_parse(const knot_db_val_t val, entry_list_t list) +{ + const bool ok = val.data && val.len && list; + if (!ok) { + assert(!EINVAL); + return kr_error(EINVAL); + } + /* Parse the apex itself (nsec parameters). */ + const struct entry_apex *ea = entry_apex_consistent(val); + if (!ea) { + return kr_error(EILSEQ); + } + const uint8_t *it = ea->data, + *it_bound = knot_db_val_bound(val); + for (int i = 0; i < ENTRY_APEX_NSECS_CNT; ++i) { + if (it > it_bound) { + return kr_error(EILSEQ); + } + list[i].data = (void *)it; + switch (ea->nsecs[i]) { + case 0: + list[i].len = 0; + break; + case 1: + list[i].len = sizeof(uint32_t); /* just timestamp */ + break; + case 3: { /* timestamp + NSEC3PARAM wire */ + if (it + sizeof(uint32_t) + 4 > it_bound) { + return kr_error(EILSEQ); + } + list[i].len = sizeof(uint32_t) + + nsec_p_rdlen(it + sizeof(uint32_t)); + break; + } + default: + return kr_error(EILSEQ); + }; + it += to_even(list[i].len); + } + /* Parse every entry_h. */ + for (int i = ENTRY_APEX_NSECS_CNT; i < EL_LENGTH; ++i) { + list[i].data = (void *)it; + bool has_type; + switch (i) { + case EL_NS: has_type = ea->has_ns; break; + case EL_CNAME: has_type = ea->has_cname; break; + case EL_DNAME: has_type = ea->has_dname; break; + default: assert(false); return kr_error(EINVAL); /* something very bad */ + } + if (!has_type) { + list[i].len = 0; + continue; + } + if (it >= it_bound) { + assert(!EILSEQ); + return kr_error(EILSEQ); + } + const int len = entry_h_len( + (knot_db_val_t){ .data = (void *)it, .len = it_bound - it }); + if (len < 0) { + assert(false); + return kr_error(len); + } + list[i].len = len; + it += to_even(len); + } + assert(it == it_bound); + return kr_ok(); +} + +/** Given a valid entry header, find its length (i.e. offset of the next entry). + * \param val The beginning of the data and the bound (read only). + */ +static int entry_h_len(const knot_db_val_t val) +{ + const bool ok = val.data && ((ssize_t)val.len) > 0; + if (!ok) return kr_error(EINVAL); + const struct entry_h *eh = val.data; + const uint8_t *d = eh->data; /* iterates over the data in entry */ + const uint8_t *data_bound = knot_db_val_bound(val); + if (d >= data_bound) return kr_error(EILSEQ); + if (!eh->is_packet) { /* Positive RRset + its RRsig set (may be empty). */ + int sets = 2; + while (sets-- > 0) { + d += rdataset_dematerialized_size(d); + if (d > data_bound) { + assert(!EILSEQ); + return kr_error(EILSEQ); + } + } + } else { /* A "packet" (opaque ATM). */ + uint16_t len; + if (d + sizeof(len) > data_bound) return kr_error(EILSEQ); + memcpy(&len, d, sizeof(len)); + d += 2 + to_even(len); + } + if (d > data_bound) { + assert(!EILSEQ); + return kr_error(EILSEQ); + } + return d - (uint8_t *)val.data; +} + +struct entry_apex * entry_apex_consistent(knot_db_val_t val) +{ + //XXX: check lengths, etc. + return val.data; +} + +/* See the header file. */ +int entry_h_seek(knot_db_val_t *val, uint16_t type) +{ + int i = -1; + switch (type) { + case KNOT_RRTYPE_NS: i = EL_NS; break; + case KNOT_RRTYPE_CNAME: i = EL_CNAME; break; + case KNOT_RRTYPE_DNAME: i = EL_DNAME; break; + default: return kr_ok(); + } + + entry_list_t el; + int ret = entry_list_parse(*val, el); + if (ret) return ret; + *val = el[i]; + return val->len ? kr_ok() : kr_error(ENOENT); +} + +static int cache_write_or_clear(struct kr_cache *cache, const knot_db_val_t *key, + knot_db_val_t *val, const struct kr_query *qry) +{ + int ret = cache_op(cache, write, key, val, 1); + if (!ret) return kr_ok(); + /* Clear cache if overfull. It's nontrivial to do better with LMDB. + * LATER: some garbage-collection mechanism. */ + if (ret == kr_error(ENOSPC)) { + ret = kr_cache_clear(cache); + const char *msg = "[cache] clearing because overfull, ret = %d\n"; + if (ret) { + kr_log_error(msg, ret); + } else { + kr_log_info(msg, ret); + ret = kr_error(ENOSPC); + } + return ret; + } + VERBOSE_MSG(qry, "=> failed backend write, ret = %d\n", ret); + return kr_error(ret ? ret : ENOSPC); +} + + +/* See the header file. */ +int entry_h_splice( + knot_db_val_t *val_new_entry, uint8_t rank, + const knot_db_val_t key, const uint16_t ktype, const uint16_t type, + const knot_dname_t *owner/*log only*/, + const struct kr_query *qry, struct kr_cache *cache, uint32_t timestamp) +{ + //TODO: another review, perhaps incuding the API + const bool ok = val_new_entry && val_new_entry->len > 0; + if (!ok) { + assert(!EINVAL); + return kr_error(EINVAL); + } + + int i_type; + switch (type) { + case KNOT_RRTYPE_NS: i_type = EL_NS; break; + case KNOT_RRTYPE_CNAME: i_type = EL_CNAME; break; + case KNOT_RRTYPE_DNAME: i_type = EL_DNAME; break; + default: i_type = 0; + } + + /* Get eh_orig (original entry), and also el list if multi-entry case. */ + const struct entry_h *eh_orig = NULL; + entry_list_t el; + int ret = -1; + if (!kr_rank_test(rank, KR_RANK_SECURE) || ktype == KNOT_RRTYPE_NS) { + knot_db_val_t val; + ret = cache_op(cache, read, &key, &val, 1); + if (i_type) { + if (!ret) ret = entry_list_parse(val, el); + if (ret) memset(el, 0, sizeof(el)); + val = el[i_type]; + } + /* val is on the entry, in either case (or error) */ + if (!ret) { + eh_orig = entry_h_consistent(val, type); + } + } else { + /* We want to fully overwrite the entry, so don't even read it. */ + memset(el, 0, sizeof(el)); + } + + if (!kr_rank_test(rank, KR_RANK_SECURE) && eh_orig) { + /* If equal rank was accepted, spoofing a *single* answer would be + * enough to e.g. override NS record in AUTHORITY section. + * This way they would have to hit the first answer + * (whenever TTL nears expiration). + * Stale-serving is NOT considered, but TTL 1 would be considered + * as expiring anyway, ... */ + int32_t old_ttl = get_new_ttl(eh_orig, qry, NULL, 0, timestamp); + if (old_ttl > 0 && !is_expiring(eh_orig->ttl, old_ttl) + && rank <= eh_orig->rank) { + WITH_VERBOSE(qry) { + auto_free char *type_str = kr_rrtype_text(type), + *owner_str = kr_dname_text(owner); + VERBOSE_MSG(qry, "=> not overwriting %s %s\n", + type_str, owner_str); + } + return kr_error(EEXIST); + } + } + + if (!i_type) { + /* The non-list types are trivial now. */ + return cache_write_or_clear(cache, &key, val_new_entry, qry); + } + /* Now we're in trouble. In some cases, parts of data to be written + * is an lmdb entry that may be invalidated by our write request. + * (lmdb does even in-place updates!) Therefore we copy all into a buffer. + * (We don't bother deallocating from the mempool.) + * LATER(optim.): do this only when neccessary, or perhaps another approach. + * This is also complicated by the fact that the val_new_entry part + * is to be written *afterwards* by the caller. + */ + el[i_type] = (knot_db_val_t){ + .len = val_new_entry->len, + .data = NULL, /* perhaps unclear in the entry_h_splice() API */ + }; + knot_db_val_t val = { + .len = entry_list_serial_size(el), + .data = NULL, + }; + void *buf = mm_alloc(&qry->request->pool, val.len); + entry_list_memcpy(buf, el); + ret = cache_write_or_clear(cache, &key, &val, qry); + if (ret) return kr_error(ret); + memcpy(val.data, buf, val.len); /* we also copy the "empty" space, but well... */ + val_new_entry->data = (uint8_t *)val.data + + ((uint8_t *)el[i_type].data - (uint8_t *)buf); + return kr_ok(); +} + diff --git a/lib/cache/entry_pkt.c b/lib/cache/entry_pkt.c new file mode 100644 index 0000000..00e73d4 --- /dev/null +++ b/lib/cache/entry_pkt.c @@ -0,0 +1,224 @@ +/* Copyright (C) 2017 CZ.NIC, z.s.p.o. + + This program is free software: you can redistribute it and/or modify + it under the terms of the GNU General Public License as published by + the Free Software Foundation, either version 3 of the License, or + (at your option) any later version. + + This program is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the + GNU General Public License for more details. + + You should have received a copy of the GNU General Public License + along with this program. If not, see . + */ + +/** @file + * Implementation of packet-caching. Prototypes in ./impl.h + * + * The packet is stashed in entry_h::data as uint16_t length + full packet wire format. + */ + +#include "lib/utils.h" +#include "lib/layer/iterate.h" /* kr_response_classify */ +#include "lib/cache/impl.h" + + +/** Compute TTL for a packet. Generally it's minimum TTL, with extra conditions. */ +static uint32_t packet_ttl(const knot_pkt_t *pkt, bool is_negative) +{ + bool has_ttl = false; + uint32_t ttl = UINT32_MAX; + /* Find minimum entry TTL in the packet or SOA minimum TTL. */ + for (knot_section_t i = KNOT_ANSWER; i <= KNOT_ADDITIONAL; ++i) { + const knot_pktsection_t *sec = knot_pkt_section(pkt, i); + for (unsigned k = 0; k < sec->count; ++k) { + const knot_rrset_t *rr = knot_pkt_rr(sec, k); + if (is_negative) { + /* Use SOA minimum TTL for negative answers. */ + if (rr->type == KNOT_RRTYPE_SOA) { + return MIN(rr->ttl, knot_soa_minimum(rr->rrs.rdata)); + } else { + continue; /* Use SOA only for negative answers. */ + } + } + if (knot_rrtype_is_metatype(rr->type)) { + continue; /* Skip metatypes. */ + } + ttl = MIN(ttl, rr->ttl); + } + } + /* If no valid TTL present, go with zero (will get clamped to minimum). */ + return has_ttl ? ttl : 0; +} + + +void stash_pkt(const knot_pkt_t *pkt, const struct kr_query *qry, + const struct kr_request *req, const bool has_optout) +{ + /* In some cases, stash also the packet. */ + const bool is_negative = kr_response_classify(pkt) + & (PKT_NODATA|PKT_NXDOMAIN); + const struct kr_qflags * const qf = &qry->flags; + const bool want_negative = qf->DNSSEC_INSECURE || !qf->DNSSEC_WANT || has_optout; + const bool want_pkt = qf->DNSSEC_BOGUS /*< useful for +cd answers */ + || (is_negative && want_negative); + + if (!want_pkt || !knot_wire_get_aa(pkt->wire) + || pkt->parsed != pkt->size /*< malformed packet; still can't detect KNOT_EFEWDATA */ + ) { + return; + } + + /* Compute rank. If cd bit is set or we got answer via non-validated + * forwarding, make the rank bad; otherwise it depends on flags. + * TODO: probably make validator attempt validation even with +cd. */ + uint8_t rank = KR_RANK_AUTH; + const bool risky_vldr = is_negative && qf->FORWARD && qf->CNAME; + /* ^^ CNAME'ed NXDOMAIN answer in forwarding mode can contain + * unvalidated records; original commit: d6e22f476. */ + if (knot_wire_get_cd(req->qsource.packet->wire) || qf->STUB || risky_vldr) { + kr_rank_set(&rank, KR_RANK_OMIT); + } else { + if (qf->DNSSEC_BOGUS) { + kr_rank_set(&rank, KR_RANK_BOGUS); + } else if (qf->DNSSEC_INSECURE) { + kr_rank_set(&rank, KR_RANK_INSECURE); + } else if (!qf->DNSSEC_WANT) { + /* no TAs at all, leave _RANK_AUTH */ + } else if (has_optout) { + /* All bad cases should be filtered above, + * at least the same way as pktcache in kresd 1.5.x. */ + kr_rank_set(&rank, KR_RANK_SECURE); + } else assert(false); + } + + const uint16_t pkt_type = knot_pkt_qtype(pkt); + const knot_dname_t *owner = knot_pkt_qname(pkt); /* qname can't be compressed */ + + // LATER: nothing exists under NXDOMAIN. Implement that (optionally)? +#if 0 + if (knot_wire_get_rcode(pkt->wire) == KNOT_RCODE_NXDOMAIN + /* && !qf->DNSSEC_INSECURE */ ) { + pkt_type = KNOT_RRTYPE_NS; + } +#endif + + /* Construct the key under which the pkt will be stored. */ + struct key k_storage, *k = &k_storage; + knot_db_val_t key; + int ret = kr_dname_lf(k->buf, owner, false); + if (ret) { + /* A server might (incorrectly) reply with QDCOUNT=0. */ + assert(owner == NULL); + return; + } + key = key_exact_type_maypkt(k, pkt_type); + + /* For now we stash the full packet byte-exactly as it came from upstream. */ + const uint16_t pkt_size = pkt->size; + knot_db_val_t val_new_entry = { + .data = NULL, + .len = offsetof(struct entry_h, data) + sizeof(pkt_size) + pkt->size, + }; + /* Prepare raw memory for the new entry and fill it. */ + struct kr_cache *cache = &req->ctx->cache; + ret = entry_h_splice(&val_new_entry, rank, key, k->type, pkt_type, + owner, qry, cache, qry->timestamp.tv_sec); + if (ret) return; /* some aren't really errors */ + assert(val_new_entry.data); + struct entry_h *eh = val_new_entry.data; + memset(eh, 0, offsetof(struct entry_h, data)); + eh->time = qry->timestamp.tv_sec; + eh->ttl = MAX(MIN(packet_ttl(pkt, is_negative), cache->ttl_max), cache->ttl_min); + eh->rank = rank; + eh->is_packet = true; + eh->has_optout = qf->DNSSEC_OPTOUT; + memcpy(eh->data, &pkt_size, sizeof(pkt_size)); + memcpy(eh->data + sizeof(pkt_size), pkt->wire, pkt_size); + + WITH_VERBOSE(qry) { + auto_free char *type_str = kr_rrtype_text(pkt_type), + *owner_str = kr_dname_text(owner); + VERBOSE_MSG(qry, "=> stashed packet: rank 0%.2o, TTL %d, " + "%s %s (%d B)\n", + eh->rank, eh->ttl, + type_str, owner_str, (int)val_new_entry.len); + } +} + + +int answer_from_pkt(kr_layer_t *ctx, knot_pkt_t *pkt, uint16_t type, + const struct entry_h *eh, const void *eh_bound, uint32_t new_ttl) +{ + struct kr_request *req = ctx->req; + struct kr_query *qry = req->current_query; + + uint16_t pkt_len; + memcpy(&pkt_len, eh->data, sizeof(pkt_len)); + if (pkt_len > pkt->max_size) { + return kr_error(ENOENT); + } + + /* Copy answer and reparse it, but keep the original message id. */ + uint16_t msgid = knot_wire_get_id(pkt->wire); + knot_pkt_clear(pkt); + memcpy(pkt->wire, eh->data + 2, pkt_len); + pkt->size = pkt_len; + int ret = knot_pkt_parse(pkt, 0); + if (ret == KNOT_EFEWDATA || ret == KNOT_EMALF) { + return kr_error(ENOENT); + /* LATER(opt): try harder to avoid stashing such packets */ + } + if (ret != KNOT_EOK) { + assert(!ret); + return kr_error(ret); + } + knot_wire_set_id(pkt->wire, msgid); + + /* Add rank into the additional field. */ + for (size_t i = 0; i < pkt->rrset_count; ++i) { + assert(!pkt->rr[i].additional); + uint8_t *rr_rank = mm_alloc(&pkt->mm, sizeof(*rr_rank)); + if (!rr_rank) { + return kr_error(ENOMEM); + } + *rr_rank = eh->rank; + pkt->rr[i].additional = rr_rank; + } + + /* Adjust TTL in each record. */ + const uint32_t drift = eh->ttl - new_ttl; + for (knot_section_t i = KNOT_ANSWER; i <= KNOT_ADDITIONAL; ++i) { + const knot_pktsection_t *sec = knot_pkt_section(pkt, i); + for (unsigned k = 0; k < sec->count; ++k) { + knot_rrset_t *rrs = // vv FIXME?? + /*const-cast*/(knot_rrset_t *)knot_pkt_rr(sec, k); + /* We need to be careful: due to enforcing minimum TTL + * on packet, some records may be below that value. + * We keep those records at TTL 0. */ + if (rrs->ttl >= drift) { + rrs->ttl -= drift; + } else { + rrs->ttl = 0; + } + } + } + + /* Finishing touches. TODO: perhaps factor out */ + struct kr_qflags * const qf = &qry->flags; + qf->EXPIRING = is_expiring(eh->ttl, new_ttl); + qf->CACHED = true; + qf->NO_MINIMIZE = true; + qf->DNSSEC_INSECURE = kr_rank_test(eh->rank, KR_RANK_INSECURE); + qf->DNSSEC_BOGUS = kr_rank_test(eh->rank, KR_RANK_BOGUS); + if (qf->DNSSEC_INSECURE || qf->DNSSEC_BOGUS) { + qf->DNSSEC_WANT = false; + } + qf->DNSSEC_OPTOUT = eh->has_optout; + VERBOSE_MSG(qry, "=> satisfied by exact packet: rank 0%.2o, new TTL %d\n", + eh->rank, new_ttl); + return kr_ok(); +} + diff --git a/lib/cache/entry_rr.c b/lib/cache/entry_rr.c new file mode 100644 index 0000000..94dd0e8 --- /dev/null +++ b/lib/cache/entry_rr.c @@ -0,0 +1,146 @@ +/* Copyright (C) 2017 CZ.NIC, z.s.p.o. + + This program is free software: you can redistribute it and/or modify + it under the terms of the GNU General Public License as published by + the Free Software Foundation, either version 3 of the License, or + (at your option) any later version. + + This program is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the + GNU General Public License for more details. + + You should have received a copy of the GNU General Public License + along with this program. If not, see . + */ + +/** @file + * Implementation of RRset (de)materialization, i.e. (de)serialization to storage + * format used in cache (some repeated fields are omitted). Prototypes in ./impl.h + */ + +#include "lib/cache/impl.h" + +int rdataset_dematerialize(const knot_rdataset_t *rds, uint8_t * restrict data) +{ + /* FIXME: either give up on even alignment and thus direct usability + * of rdatasets as they are in lmdb, or align inside cdb_* functions + * (request sizes one byte longer and shift iff on an odd address). */ + //if ((size_t)data & 1) VERBOSE_MSG(NULL, "dematerialize: odd address\n"); + //const uint8_t *data0 = data; + if (!data) { + assert(data); + return kr_error(EINVAL); + } + const uint16_t rr_count = rds ? rds->count : 0; + memcpy(data, &rr_count, sizeof(rr_count)); + data += sizeof(rr_count); + if (rr_count) { + size_t size = knot_rdataset_size(rds); + memcpy(data, rds->rdata, size); + data += size; + } + //VERBOSE_MSG(NULL, "dematerialized to %d B\n", (int)(data - data0)); + (void)data; + return kr_ok(); +} + +/** Materialize a knot_rdataset_t from cache with given TTL. + * Return the number of bytes consumed or an error code. + */ +static int rdataset_materialize(knot_rdataset_t * restrict rds, const uint8_t * const data, + const uint8_t *data_bound, knot_mm_t *pool) +{ + assert(rds && data && data_bound && data_bound > data && !rds->rdata + /*&& !((size_t)data & 1)*/); + assert(pool); /* not required, but that's our current usage; guard leaks */ + const uint8_t *d = data; /* iterates over the cache data */ + { + uint16_t rr_count; + memcpy(&rr_count, d, sizeof(rr_count)); + d += sizeof(rr_count); + rds->count = rr_count; + if (!rr_count) { /* avoid mm_alloc(pool, 0); etc. */ + return d - data; + } + } + /* First sum up the sizes for wire format length. */ + const knot_rdataset_t rds_tmp = { + .count = rds->count, + .rdata = (knot_rdata_t *)d, + }; + size_t rds_size = knot_rdataset_size(&rds_tmp); /* TODO: we might overrun here already, + but we need to trust cache anyway...*/ + if (d + rds_size > data_bound) { + VERBOSE_MSG(NULL, "materialize: EILSEQ!\n"); + return kr_error(EILSEQ); + } + rds->rdata = mm_alloc(pool, rds_size); + if (!rds->rdata) { + return kr_error(ENOMEM); + } + memcpy(rds->rdata, d, rds_size); + d += rds_size; + //VERBOSE_MSG(NULL, "materialized from %d B\n", (int)(d - data)); + return d - data; +} + +int kr_cache_materialize(knot_rdataset_t *dst, const struct kr_cache_p *ref, + knot_mm_t *pool) +{ + struct entry_h *eh = ref->raw_data; + return rdataset_materialize(dst, eh->data, ref->raw_bound, pool); +} + + +int entry2answer(struct answer *ans, int id, + const struct entry_h *eh, const uint8_t *eh_bound, + const knot_dname_t *owner, uint16_t type, uint32_t new_ttl) +{ + /* We assume it's zeroed. Do basic sanity check. */ + if (ans->rrsets[id].set.rr || ans->rrsets[id].sig_rds.rdata + || (type == KNOT_RRTYPE_NSEC && ans->nsec_p.raw) + || (type == KNOT_RRTYPE_NSEC3 && !ans->nsec_p.raw) + ) + { + assert(false); + return kr_error(EINVAL); + } + /* Materialize the base RRset. */ + knot_rrset_t *rr = ans->rrsets[id].set.rr + = knot_rrset_new(owner, type, KNOT_CLASS_IN, new_ttl, ans->mm); + if (!rr) { + assert(!ENOMEM); + return kr_error(ENOMEM); + } + int ret = rdataset_materialize(&rr->rrs, eh->data, eh_bound, ans->mm); + if (ret < 0) goto fail; + size_t data_off = ret; + ans->rrsets[id].set.rank = eh->rank; + ans->rrsets[id].set.expiring = is_expiring(eh->ttl, new_ttl); + /* Materialize the RRSIG RRset for the answer in (pseudo-)packet. */ + bool want_rrsigs = true; /* LATER(optim.): might be omitted in some cases. */ + if (want_rrsigs) { + ret = rdataset_materialize(&ans->rrsets[id].sig_rds, eh->data + data_off, + eh_bound, ans->mm); + if (ret < 0) goto fail; + /* Sanity check: we consumed exactly all data. */ + int unused_bytes = eh_bound - (uint8_t *)eh->data - data_off - ret; + if (unused_bytes) { + kr_log_error("[cach] entry2answer ERROR: unused bytes: %d\n", + unused_bytes); + assert(!EILSEQ); + ret = kr_error(EILSEQ); + goto fail; /* to be on the safe side */ + } + } + return kr_ok(); +fail: + assert(/*false*/!ret); + /* Cleanup the item that we might've (partially) written to. */ + knot_rrset_free(ans->rrsets[id].set.rr, ans->mm); + knot_rdataset_clear(&ans->rrsets[id].sig_rds, ans->mm); + memset(&ans->rrsets[id], 0, sizeof(ans->rrsets[id])); + return kr_error(ret); +} + diff --git a/lib/cache/impl.h b/lib/cache/impl.h new file mode 100644 index 0000000..a08f355 --- /dev/null +++ b/lib/cache/impl.h @@ -0,0 +1,412 @@ +/* Copyright (C) 2017 CZ.NIC, z.s.p.o. + + This program is free software: you can redistribute it and/or modify + it under the terms of the GNU General Public License as published by + the Free Software Foundation, either version 3 of the License, or + (at your option) any later version. + + This program is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the + GNU General Public License for more details. + + You should have received a copy of the GNU General Public License + along with this program. If not, see . + */ + +/** @file + * Header internal for cache implementation(s). + * Only LMDB works for now. + */ +#pragma once + +#include +#include + +#include +#include +#include +#include +#include + +#include "contrib/cleanup.h" +#include "contrib/murmurhash3/murmurhash3.h" /* hash() for nsec_p_hash() */ +#include "lib/cache/cdb_api.h" +#include "lib/resolve.h" + +/* Cache entry values - binary layout. + * + * It depends on type which is recognizable by the key. + * Code depending on the contents of the key is marked by CACHE_KEY_DEF. + * + * 'E' entry (exact hit): + * - ktype == NS: struct entry_apex - multiple types inside (NS and xNAME); + * - ktype != NS: struct entry_h + * * is_packet: uint16_t length, the rest is opaque and handled by ./entry_pkt.c + * * otherwise RRset + its RRSIG set (possibly empty). + * '1' or '3' entry (NSEC or NSEC3) + * - struct entry_h, contents is the same as for exact hit + * - flags don't make sense there + */ + +struct entry_h { + uint32_t time; /**< The time of inception. */ + uint32_t ttl; /**< TTL at inception moment. Assuming it fits into int32_t ATM. */ + uint8_t rank : 6; /**< See enum kr_rank */ + bool is_packet : 1; /**< Negative-answer packet for insecure/bogus name. */ + bool has_optout : 1; /**< Only for packets; persisted DNSSEC_OPTOUT. */ + uint8_t _pad; /**< We need even alignment for data now. */ + uint8_t data[]; +}; +struct entry_apex; + +/** Check basic consistency of entry_h for 'E' entries, not looking into ->data. + * (for is_packet the length of data is checked) + */ +struct entry_h * entry_h_consistent(knot_db_val_t data, uint16_t type); + +struct entry_apex * entry_apex_consistent(knot_db_val_t val); + +/** Consistency check, ATM common for NSEC and NSEC3. */ +static inline struct entry_h * entry_h_consistent_NSEC(knot_db_val_t data) +{ + /* ATM it's enough to just extend the checks for exact entries. */ + const struct entry_h *eh = entry_h_consistent(data, KNOT_RRTYPE_NSEC); + bool ok = eh != NULL; + ok = ok && !eh->is_packet && !eh->has_optout; + return ok ? /*const-cast*/(struct entry_h *)eh : NULL; +} + + +/* nsec_p* - NSEC* chain parameters */ + +static inline int nsec_p_rdlen(const uint8_t *rdata) +{ + //TODO: do we really need the zero case? + return rdata ? 5 + rdata[4] : 0; /* rfc5155 4.2 and 3.2. */ +} +static const int NSEC_P_MAXLEN = sizeof(uint32_t) + 5 + 255; // TODO: remove?? + +/** Hash of NSEC3 parameters, used as a tag to separate different chains for same zone. */ +typedef uint32_t nsec_p_hash_t; +static inline nsec_p_hash_t nsec_p_mkHash(const uint8_t *nsec_p) +{ + assert(nsec_p && !(KNOT_NSEC3_FLAG_OPT_OUT & nsec_p[1])); + return hash((const char *)nsec_p, nsec_p_rdlen(nsec_p)); +} + +/** NSEC* parameters for the chain. */ +struct nsec_p { + const uint8_t *raw; /**< Pointer to raw NSEC3 parameters; NULL for NSEC. */ + nsec_p_hash_t hash; /**< Hash of `raw`, used for cache keys. */ + dnssec_nsec3_params_t libknot; /**< Format for libknot; owns malloced memory! */ +}; + + + +/** LATER(optim.): this is overshot, but struct key usage should be cheap ATM. */ +#define KR_CACHE_KEY_MAXLEN (KNOT_DNAME_MAXLEN + 100) /* CACHE_KEY_DEF */ + +struct key { + const knot_dname_t *zname; /**< current zone name (points within qry->sname) */ + uint8_t zlf_len; /**< length of current zone's lookup format */ + + /** Corresponding key type; e.g. NS for CNAME. + * Note: NSEC type is ambiguous (exact and range key). */ + uint16_t type; + /** The key data start at buf+1, and buf[0] contains some length. + * For details see key_exact* and key_NSEC* functions. */ + uint8_t buf[KR_CACHE_KEY_MAXLEN]; + /* LATER(opt.): ^^ probably change the anchoring, so that kr_dname_lf() + * doesn't need to move data after knot_dname_lf(). */ +}; + +static inline size_t key_nwz_off(const struct key *k) +{ + /* CACHE_KEY_DEF: zone name lf + 0 ('1' or '3'). + * NSEC '1' case continues just with the name within zone. */ + return k->zlf_len + 2; +} +static inline size_t key_nsec3_hash_off(const struct key *k) +{ + /* CACHE_KEY_DEF NSEC3: tag (nsec_p_hash_t) + 20 bytes NSEC3 name hash) */ + return key_nwz_off(k) + sizeof(nsec_p_hash_t); +} +/** Hash is always SHA1; I see no plans to standardize anything else. + * https://www.iana.org/assignments/dnssec-nsec3-parameters/dnssec-nsec3-parameters.xhtml#dnssec-nsec3-parameters-3 + */ +static const int NSEC3_HASH_LEN = 20, + NSEC3_HASH_TXT_LEN = 32; + +/** Finish constructing string key for for exact search. + * It's assumed that kr_dname_lf(k->buf, owner, *) had been ran. + */ +knot_db_val_t key_exact_type_maypkt(struct key *k, uint16_t type); + +/** Like key_exact_type_maypkt but with extra checks if used for RRs only. */ +static inline knot_db_val_t key_exact_type(struct key *k, uint16_t type) +{ + switch (type) { + /* Sanity check: forbidden types represented in other way(s). */ + case KNOT_RRTYPE_NSEC: + case KNOT_RRTYPE_NSEC3: + assert(false); + return (knot_db_val_t){ NULL, 0 }; + } + return key_exact_type_maypkt(k, type); +} + + +/* entry_h chaining; implementation in ./entry_list.c */ + +enum { ENTRY_APEX_NSECS_CNT = 2 }; + +/** Header of 'E' entry with ktype == NS. Inside is private to ./entry_list.c + * + * We store xNAME at NS type to lower the number of searches in closest_NS(). + * CNAME is only considered for equal name, of course. + * We also store NSEC* parameters at NS type. + */ +struct entry_apex { + /* ENTRY_H_FLAGS */ + bool has_ns : 1; + bool has_cname : 1; + bool has_dname : 1; + + uint8_t pad_; /**< Weird: 1 byte + 2 bytes + x bytes; let's do 2+2+x. */ + int8_t nsecs[ENTRY_APEX_NSECS_CNT]; /**< values: 0: none, 1: NSEC, 3: NSEC3 */ + uint8_t data[]; + /* XXX: if not first, stamp of last being the first? + * Purpose: save cache operations if rolled the algo/params long ago. */ +}; + +/** Indices for decompressed entry_list_t. */ +enum EL { + EL_NS = ENTRY_APEX_NSECS_CNT, + EL_CNAME, + EL_DNAME, + EL_LENGTH +}; +/** Decompressed entry_apex. It's an array of unparsed entry_h references. + * Note: arrays are passed "by reference" to functions (in C99). */ +typedef knot_db_val_t entry_list_t[EL_LENGTH]; + +static inline uint16_t EL2RRTYPE(enum EL i) +{ + switch (i) { + case EL_NS: return KNOT_RRTYPE_NS; + case EL_CNAME: return KNOT_RRTYPE_CNAME; + case EL_DNAME: return KNOT_RRTYPE_DNAME; + default: assert(false); return 0; + } +} + +/** There may be multiple entries within, so rewind `val` to the one we want. + * + * ATM there are multiple types only for the NS ktype - it also accommodates xNAMEs. + * \note `val->len` represents the bound of the whole list, not of a single entry. + * \note in case of ENOENT, `val` is still rewound to the beginning of the next entry. + * \return error code + * TODO: maybe get rid of this API? + */ +int entry_h_seek(knot_db_val_t *val, uint16_t type); + +/** Prepare space to insert an entry. + * + * Some checks are performed (rank, TTL), the current entry in cache is copied + * with a hole ready for the new entry (old one of the same type is cut out). + * + * \param val_new_entry The only changing parameter; ->len is read, ->data written. + * \return error code + */ +int entry_h_splice( + knot_db_val_t *val_new_entry, uint8_t rank, + const knot_db_val_t key, const uint16_t ktype, const uint16_t type, + const knot_dname_t *owner/*log only*/, + const struct kr_query *qry, struct kr_cache *cache, uint32_t timestamp); + +/** Parse an entry_apex into individual items. @return error code. */ +int entry_list_parse(const knot_db_val_t val, entry_list_t list); + +static inline size_t to_even(size_t n) +{ + return n + (n & 1); +} + +static inline int entry_list_serial_size(const entry_list_t list) +{ + int size = offsetof(struct entry_apex, data); + for (int i = 0; i < EL_LENGTH; ++i) { + size += to_even(list[i].len); + } + return size; +} + +/** Fill contents of an entry_apex. + * + * @note NULL pointers are overwritten - caller may like to fill the space later. + */ +void entry_list_memcpy(struct entry_apex *ea, entry_list_t list); + + + +/* Packet caching; implementation in ./entry_pkt.c */ + +/** Stash the packet into cache (if suitable, etc.) + * \param has_optout whether the packet contains an opt-out NSEC3 */ +void stash_pkt(const knot_pkt_t *pkt, const struct kr_query *qry, + const struct kr_request *req, bool has_optout); + +/** Try answering from packet cache, given an entry_h. + * + * This assumes the TTL is OK and entry_h_consistent, but it may still return error. + * On success it handles all the rest, incl. qry->flags. + */ +int answer_from_pkt(kr_layer_t *ctx, knot_pkt_t *pkt, uint16_t type, + const struct entry_h *eh, const void *eh_bound, uint32_t new_ttl); + + +/** Record is expiring if it has less than 1% TTL (or less than 5s) */ +static inline bool is_expiring(uint32_t orig_ttl, uint32_t new_ttl) +{ + int64_t nttl = new_ttl; /* avoid potential over/under-flow */ + return 100 * (nttl - 5) < orig_ttl; +} + +/** Returns signed result so you can inspect how much stale the RR is. + * + * @param owner name for stale-serving decisions. You may pass NULL to disable stale. + * @note: NSEC* uses zone name ATM; for NSEC3 the owner may not even be knowable. + * @param type for stale-serving. + */ +int32_t get_new_ttl(const struct entry_h *entry, const struct kr_query *qry, + const knot_dname_t *owner, uint16_t type, uint32_t now); + + +/* RRset (de)materialization; implementation in ./entry_rr.c */ + +/** Size of the RR count field */ +#define KR_CACHE_RR_COUNT_SIZE sizeof(uint16_t) + +/** Compute size of serialized rdataset. NULL is accepted as empty set. */ +static inline int rdataset_dematerialize_size(const knot_rdataset_t *rds) +{ + return KR_CACHE_RR_COUNT_SIZE + (rds == NULL ? 0 : knot_rdataset_size(rds)); +} + +static inline int rdataset_dematerialized_size(const uint8_t *data) +{ + knot_rdataset_t rds; + memcpy(&rds.count, data, sizeof(rds.count)); + rds.rdata = (knot_rdata_t *)(data + sizeof(rds.count)); + return sizeof(rds.count) + knot_rdataset_size(&rds); +} + +/** Serialize an rdataset. */ +int rdataset_dematerialize(const knot_rdataset_t *rds, uint8_t * restrict data); + + +/** Partially constructed answer when gathering RRsets from cache. */ +struct answer { + int rcode; /**< PKT_NODATA, etc. */ + struct nsec_p nsec_p; /**< Don't mix different NSEC* parameters in one answer. */ + knot_mm_t *mm; /**< Allocator for rrsets */ + struct answer_rrset { + ranked_rr_array_entry_t set; /**< set+rank for the main data */ + knot_rdataset_t sig_rds; /**< RRSIG data, if any */ + } rrsets[1+1+3]; /**< see AR_ANSWER and friends; only required records are filled */ +}; +enum { + AR_ANSWER = 0, /**< Positive answer record. It might be wildcard-expanded. */ + AR_SOA, /**< SOA record. */ + AR_NSEC, /**< NSEC* covering or matching the SNAME (next closer name in NSEC3 case). */ + AR_WILD, /**< NSEC* covering or matching the source of synthesis. */ + AR_CPE, /**< NSEC3 matching the closest provable encloser. */ +}; + +/** Materialize RRset + RRSIGs into ans->rrsets[id]. + * LATER(optim.): it's slightly wasteful that we allocate knot_rrset_t for the packet + * + * \return error code. They are all bad conditions and "guarded" by assert. + */ +int entry2answer(struct answer *ans, int id, + const struct entry_h *eh, const uint8_t *eh_bound, + const knot_dname_t *owner, uint16_t type, uint32_t new_ttl); + + +/* Preparing knot_pkt_t for cache answer from RRs; implementation in ./knot_pkt.c */ + +/** Prepare answer packet to be filled by RRs (without RR data in wire). */ +int pkt_renew(knot_pkt_t *pkt, const knot_dname_t *name, uint16_t type); + +/** Append RRset + its RRSIGs into the current section (*shallow* copy), with given rank. + * \note it works with empty set as well (skipped) + * \note pkt->wire is not updated in any way + * \note KNOT_CLASS_IN is assumed + */ +int pkt_append(knot_pkt_t *pkt, const struct answer_rrset *rrset, uint8_t rank); + + + +/* NSEC (1) stuff. Implementation in ./nsec1.c */ + +/** Construct a string key for for NSEC (1) predecessor-search. + * \param add_wildcard Act as if the name was extended by "*." + * \note k->zlf_len is assumed to have been correctly set */ +knot_db_val_t key_NSEC1(struct key *k, const knot_dname_t *name, bool add_wildcard); + +/** Closest encloser check for NSEC (1). + * To understand the interface, see the call point. + * \param k space to store key + input: zname and zlf_len + * \return 0: success; >0: try other (NSEC3); <0: exit cache immediately. */ +int nsec1_encloser(struct key *k, struct answer *ans, + const int sname_labels, int *clencl_labels, + knot_db_val_t *cover_low_kwz, knot_db_val_t *cover_hi_kwz, + const struct kr_query *qry, struct kr_cache *cache); + +/** Source of synthesis (SS) check for NSEC (1). + * To understand the interface, see the call point. + * \return 0: continue; <0: exit cache immediately; + * AR_SOA: skip to adding SOA (SS was covered or matched for NODATA). */ +int nsec1_src_synth(struct key *k, struct answer *ans, const knot_dname_t *clencl_name, + knot_db_val_t cover_low_kwz, knot_db_val_t cover_hi_kwz, + const struct kr_query *qry, struct kr_cache *cache); + + +/* NSEC3 stuff. Implementation in ./nsec3.c */ + +/** Construct a string key for for NSEC3 predecessor-search, from an NSEC3 name. + * \note k->zlf_len is assumed to have been correctly set */ +knot_db_val_t key_NSEC3(struct key *k, const knot_dname_t *nsec3_name, + const nsec_p_hash_t nsec_p_hash); + +/** TODO. See nsec1_encloser(...) */ +int nsec3_encloser(struct key *k, struct answer *ans, + const int sname_labels, int *clencl_labels, + const struct kr_query *qry, struct kr_cache *cache); + +/** TODO. See nsec1_src_synth(...) */ +int nsec3_src_synth(struct key *k, struct answer *ans, const knot_dname_t *clencl_name, + const struct kr_query *qry, struct kr_cache *cache); + + + +#define VERBOSE_MSG(qry, ...) QRVERBOSE((qry), "cach", ## __VA_ARGS__) + +/** Shorthand for operations on cache backend */ +#define cache_op(cache, op, ...) (cache)->api->op((cache)->db, ## __VA_ARGS__) + + +static inline uint16_t get_uint16(const void *address) +{ + uint16_t tmp; + memcpy(&tmp, address, sizeof(tmp)); + return tmp; +} + +/** Useful pattern, especially as void-pointer arithmetic isn't standard-compliant. */ +static inline uint8_t * knot_db_val_bound(knot_db_val_t val) +{ + return (uint8_t *)val.data + val.len; +} + diff --git a/lib/cache/knot_pkt.c b/lib/cache/knot_pkt.c new file mode 100644 index 0000000..56836a6 --- /dev/null +++ b/lib/cache/knot_pkt.c @@ -0,0 +1,106 @@ +/* Copyright (C) 2017 CZ.NIC, z.s.p.o. + + This program is free software: you can redistribute it and/or modify + it under the terms of the GNU General Public License as published by + the Free Software Foundation, either version 3 of the License, or + (at your option) any later version. + + This program is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the + GNU General Public License for more details. + + You should have received a copy of the GNU General Public License + along with this program. If not, see . + */ + +/** @file + * Implementation of preparing knot_pkt_t for filling with RRs. + * Prototypes in ./impl.h + */ + +#include "lib/cache/impl.h" + +int pkt_renew(knot_pkt_t *pkt, const knot_dname_t *name, uint16_t type) +{ + /* Update packet question if needed. */ + if (!knot_dname_is_equal(knot_pkt_qname(pkt), name) + || knot_pkt_qtype(pkt) != type || knot_pkt_qclass(pkt) != KNOT_CLASS_IN) { + int ret = kr_pkt_recycle(pkt); + if (ret) return kr_error(ret); + ret = knot_pkt_put_question(pkt, name, KNOT_CLASS_IN, type); + if (ret) return kr_error(ret); + } + + pkt->parsed = pkt->size = PKT_SIZE_NOWIRE; + knot_wire_set_qr(pkt->wire); + knot_wire_set_aa(pkt->wire); + return kr_ok(); +} + +/** Reserve space for additional `count` RRsets. + * \note pkt->rr_info gets correct length but is always zeroed + */ +static int pkt_alloc_space(knot_pkt_t *pkt, int count) +{ + size_t allocd_orig = pkt->rrset_allocd; + if (pkt->rrset_count + count <= allocd_orig) { + return kr_ok(); + } + /* A simple growth strategy, amortized O(count). */ + pkt->rrset_allocd = MAX( + pkt->rrset_count + count, + pkt->rrset_count + allocd_orig); + + pkt->rr = mm_realloc(&pkt->mm, pkt->rr, + sizeof(pkt->rr[0]) * pkt->rrset_allocd, + sizeof(pkt->rr[0]) * allocd_orig); + if (!pkt->rr) { + return kr_error(ENOMEM); + } + /* Allocate pkt->rr_info to be certain, but just leave it zeroed. */ + mm_free(&pkt->mm, pkt->rr_info); + pkt->rr_info = mm_alloc(&pkt->mm, sizeof(pkt->rr_info[0]) * pkt->rrset_allocd); + if (!pkt->rr_info) { + return kr_error(ENOMEM); + } + memset(pkt->rr_info, 0, sizeof(pkt->rr_info[0]) * pkt->rrset_allocd); + return kr_ok(); +} + +int pkt_append(knot_pkt_t *pkt, const struct answer_rrset *rrset, uint8_t rank) +{ + /* allocate space, to be sure */ + int rrset_cnt = (rrset->set.rr->rrs.count > 0) + (rrset->sig_rds.count > 0); + int ret = pkt_alloc_space(pkt, rrset_cnt); + if (ret) return kr_error(ret); + /* write both sets */ + const knot_rdataset_t *rdss[2] = { &rrset->set.rr->rrs, &rrset->sig_rds }; + for (int i = 0; i < rrset_cnt; ++i) { + assert(rdss[i]->count); + /* allocate rank */ + uint8_t *rr_rank = mm_alloc(&pkt->mm, sizeof(*rr_rank)); + if (!rr_rank) return kr_error(ENOMEM); + *rr_rank = (i == 0) ? rank : (KR_RANK_OMIT | KR_RANK_AUTH); + /* rank for RRSIGs isn't really useful: ^^ */ + if (i == 0) { + pkt->rr[pkt->rrset_count] = *rrset->set.rr; + pkt->rr[pkt->rrset_count].additional = rr_rank; + } else { + /* append the RR array */ + pkt->rr[pkt->rrset_count] = (knot_rrset_t){ + .owner = knot_dname_copy(rrset->set.rr->owner, &pkt->mm), + /* ^^ well, another copy isn't really needed */ + .ttl = rrset->set.rr->ttl, + .type = KNOT_RRTYPE_RRSIG, + .rclass = KNOT_CLASS_IN, + .rrs = *rdss[i], + .additional = rr_rank, + }; + } + ++pkt->rrset_count; + ++(pkt->sections[pkt->current].count); + } + return kr_ok(); +} + diff --git a/lib/cache/nsec1.c b/lib/cache/nsec1.c new file mode 100644 index 0000000..74b3d34 --- /dev/null +++ b/lib/cache/nsec1.c @@ -0,0 +1,511 @@ +/* Copyright (C) 2017 CZ.NIC, z.s.p.o. + + This program is free software: you can redistribute it and/or modify + it under the terms of the GNU General Public License as published by + the Free Software Foundation, either version 3 of the License, or + (at your option) any later version. + + This program is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the + GNU General Public License for more details. + + You should have received a copy of the GNU General Public License + along with this program. If not, see . + */ + +/** @file + * Implementation of NSEC (1) handling. Prototypes in ./impl.h + */ + +#include "lib/cache/impl.h" +#include "lib/dnssec/nsec.h" +#include "lib/layer/iterate.h" + + +/** Reconstruct a name into a buffer (assuming length at least KNOT_DNAME_MAXLEN). + * \return kr_ok() or error code (<0). */ +static int dname_wire_reconstruct(knot_dname_t *buf, const struct key *k, + knot_db_val_t kwz) +{ + /* Reconstruct from key: first the ending, then zone name. */ + int ret = knot_dname_lf2wire(buf, kwz.len, kwz.data); + if (ret < 0) { + VERBOSE_MSG(NULL, "=> NSEC: LF2wire ret = %d\n", ret); + assert(false); + return ret; + } + /* The last written byte is the zero label for root -> overwrite. */ + knot_dname_t *zone_start = buf + ret - 1; + assert(*zone_start == '\0'); + ret = knot_dname_to_wire(zone_start, k->zname, KNOT_DNAME_MAXLEN - kwz.len); + if (ret != k->zlf_len + 1) { + assert(false); + return ret < 0 ? ret : kr_error(EILSEQ); + } + return kr_ok(); +} + + +knot_db_val_t key_NSEC1(struct key *k, const knot_dname_t *name, bool add_wildcard) +{ + /* we basically need dname_lf with two bytes added + * on a correct place within the name (the cut) */ + int ret; + const bool ok = k && name + && !(ret = kr_dname_lf(k->buf, name, add_wildcard)); + if (!ok) { + assert(false); + return (knot_db_val_t){ NULL, 0 }; + } + + uint8_t *begin = k->buf + 1 + k->zlf_len; /* one byte after zone's zero */ + uint8_t *end = k->buf + 1 + k->buf[0]; /* we don't use the final zero in key, + * but move it anyway */ + if (end < begin) { + assert(false); + return (knot_db_val_t){ NULL, 0 }; + } + int key_len; + if (end > begin) { + memmove(begin + 2, begin, end - begin); + key_len = k->buf[0] + 1; + } else { + key_len = k->buf[0] + 2; + } + /* CACHE_KEY_DEF: key == zone's dname_lf + '\0' + '1' + dname_lf + * of the name within the zone without the final 0. Iff the latter is empty, + * there's no zero to cut and thus the key_len difference. + */ + begin[0] = 0; + begin[1] = '1'; /* tag for NSEC1 */ + k->type = KNOT_RRTYPE_NSEC; + + /* + VERBOSE_MSG(NULL, "<> key_NSEC1; name: "); + kr_dname_print(name, add_wildcard ? "*." : "" , " "); + kr_log_verbose("(zone name LF length: %d; total key length: %d)\n", + k->zlf_len, key_len); + */ + + return (knot_db_val_t){ k->buf + 1, key_len }; +} + + +/** Assuming that k1 < k4, find where k2 is. (Considers DNS wrap-around.) + * + * \return Intuition: position of k2 among kX. + * 0: k2 < k1; 1: k1 == k2; 2: k1 is a prefix of k2 < k4; + * 3: k1 < k2 < k4 (and not 2); 4: k2 == k4; 5: k2 > k4 + * \note k1.data may be NULL, meaning assumption that k1 < k2 and not a prefix + * (i.e. return code will be > 2) + */ +static int kwz_between(knot_db_val_t k1, knot_db_val_t k2, knot_db_val_t k4) +{ + assert(k2.data && k4.data); + /* CACHE_KEY_DEF; we need to beware of one key being a prefix of another */ + int ret_maybe; /**< result, assuming we confirm k2 < k4 */ + if (k1.data) { + const int cmp12 = memcmp(k1.data, k2.data, MIN(k1.len, k2.len)); + if (cmp12 == 0 && k1.len == k2.len) /* iff k1 == k2 */ + return 1; + if (cmp12 > 0 || (cmp12 == 0 && k1.len > k2.len)) /* iff k1 > k2 */ + return 0; + ret_maybe = cmp12 == 0 ? 2 : 3; + } else { + ret_maybe = 3; + } + if (k4.len == 0) { /* wrap-around */ + return k2.len > 0 ? ret_maybe : 4; + } else { + const int cmp24 = memcmp(k2.data, k4.data, MIN(k2.len, k4.len)); + if (cmp24 == 0 && k2.len == k4.len) /* iff k2 == k4 */ + return 4; + if (cmp24 > 0 || (cmp24 == 0 && k2.len > k4.len)) /* iff k2 > k4 */ + return 5; + return ret_maybe; + } +} + + +/** NSEC1 range search. + * + * \param key Pass output of key_NSEC1(k, ...) + * \param value[out] The raw data of the NSEC cache record (optional; consistency checked). + * \param exact_match[out] Whether the key was matched exactly or just covered (optional). + * \param kwz_low[out] Output the low end of covering NSEC, pointing within DB (optional). + * \param kwz_high[in,out] Storage for the high end of covering NSEC (optional). + * It's only set if !exact_match. + * \param new_ttl[out] New TTL of the NSEC (optional). + * \return Error message or NULL. + * \note The function itself does *no* bitmap checks, e.g. RFC 6840 sec. 4. + */ +static const char * find_leq_NSEC1(struct kr_cache *cache, const struct kr_query *qry, + const knot_db_val_t key, const struct key *k, knot_db_val_t *value, + bool *exact_match, knot_db_val_t *kwz_low, knot_db_val_t *kwz_high, + uint32_t *new_ttl) +{ + /* Do the cache operation. */ + const size_t nwz_off = key_nwz_off(k); + if (!key.data || key.len < nwz_off) { + assert(false); + return "range search ERROR"; + } + knot_db_val_t key_nsec = key; + knot_db_val_t val = { NULL, 0 }; + int ret = cache_op(cache, read_leq, &key_nsec, &val); + if (ret < 0) { + if (ret == kr_error(ENOENT)) { + return "range search miss"; + } else { + assert(false); + return "range search ERROR"; + } + } + if (value) { + *value = val; + } + /* Check consistency, TTL, rank. */ + const bool is_exact = (ret == 0); + if (exact_match) { + *exact_match = is_exact; + } + const struct entry_h *eh = entry_h_consistent_NSEC(val); + if (!eh) { + /* This might be just finding something else than NSEC1 entry, + * in case we searched before the very first one in the zone. */ + return "range search found inconsistent entry"; + } + /* Passing just zone name instead of owner, as we don't + * have it reconstructed at this point. */ + int32_t new_ttl_ = get_new_ttl(eh, qry, k->zname, KNOT_RRTYPE_NSEC, + qry->timestamp.tv_sec); + if (new_ttl_ < 0 || !kr_rank_test(eh->rank, KR_RANK_SECURE)) { + return "range search found stale or insecure entry"; + /* TODO: remove the stale record *and* retry, + * in case we haven't run off. Perhaps start by in_zone check. */ + } + if (new_ttl) { + *new_ttl = new_ttl_; + } + if (kwz_low) { + *kwz_low = (knot_db_val_t){ + .data = (uint8_t *)key_nsec.data + nwz_off, + .len = key_nsec.len - nwz_off, + }; /* CACHE_KEY_DEF */ + } + if (is_exact) { + /* Nothing else to do. */ + return NULL; + } + /* The NSEC starts strictly before our target name; + * now check that it still belongs into that zone. */ + const bool nsec_in_zone = key_nsec.len >= nwz_off + /* CACHE_KEY_DEF */ + && memcmp(key.data, key_nsec.data, nwz_off) == 0; + if (!nsec_in_zone) { + return "range search miss (!nsec_in_zone)"; + } + /* We know it starts before sname, so let's check the other end. + * 1. construct the key for the next name - kwz_hi. */ + /* it's *full* name ATM */ + const knot_rdata_t *next = (const knot_rdata_t *) + (eh->data + KR_CACHE_RR_COUNT_SIZE); + if (KR_CACHE_RR_COUNT_SIZE != 2 || get_uint16(eh->data) == 0) { + assert(false); + return "ERROR"; + /* TODO: more checks? */ + } + /* + WITH_VERBOSE { + VERBOSE_MSG(qry, "=> NSEC: next name: "); + kr_dname_print(next, "", "\n"); + } + */ + knot_dname_t ch_buf[KNOT_DNAME_MAXLEN]; + knot_dname_t *chs = kwz_high ? kwz_high->data : ch_buf; + if (!chs) { + assert(false); + return "EINVAL"; + } + { + /* Lower-case chs; see also RFC 6840 5.1. + * LATER(optim.): we do lots of copying etc. */ + knot_dname_t lower_buf[KNOT_DNAME_MAXLEN]; + ret = knot_dname_to_wire(lower_buf, next->data, + MIN(next->len, KNOT_DNAME_MAXLEN)); + if (ret < 0) { /* _ESPACE */ + return "range search found record with incorrect contents"; + } + knot_dname_to_lower(lower_buf); + ret = kr_dname_lf(chs, lower_buf, false); + } + if (ret) { + assert(false); + return "ERROR"; + } + knot_db_val_t kwz_hi = { /* skip the zone name */ + .data = chs + 1 + k->zlf_len, + .len = chs[0] - k->zlf_len, + }; + assert((ssize_t)(kwz_hi.len) >= 0); + /* 2. do the actual range check. */ + const knot_db_val_t kwz_sname = { + .data = (void *)/*const-cast*/(k->buf + 1 + nwz_off), + .len = k->buf[0] - k->zlf_len, + }; + assert((ssize_t)(kwz_sname.len) >= 0); + bool covers = /* we know for sure that the low end is before kwz_sname */ + 3 == kwz_between((knot_db_val_t){ NULL, 0 }, kwz_sname, kwz_hi); + if (!covers) { + return "range search miss (!covers)"; + } + if (kwz_high) { + *kwz_high = kwz_hi; + } + return NULL; +} + + +int nsec1_encloser(struct key *k, struct answer *ans, + const int sname_labels, int *clencl_labels, + knot_db_val_t *cover_low_kwz, knot_db_val_t *cover_hi_kwz, + const struct kr_query *qry, struct kr_cache *cache) +{ + static const int ESKIP = ABS(ENOENT); + /* Basic sanity check. */ + const bool ok = k && ans && clencl_labels && cover_low_kwz && cover_hi_kwz + && qry && cache; + if (!ok) { + assert(!EINVAL); + return kr_error(EINVAL); + } + + /* Find a previous-or-equal name+NSEC in cache covering the QNAME, + * checking TTL etc. */ + knot_db_val_t key = key_NSEC1(k, qry->sname, false); + knot_db_val_t val = { NULL, 0 }; + bool exact_match; + uint32_t new_ttl; + const char *err = find_leq_NSEC1(cache, qry, key, k, &val, + &exact_match, cover_low_kwz, cover_hi_kwz, &new_ttl); + if (err) { + VERBOSE_MSG(qry, "=> NSEC sname: %s\n", err); + return ESKIP; + } + + /* Get owner name of the record. */ + const knot_dname_t *owner; + knot_dname_t owner_buf[KNOT_DNAME_MAXLEN]; + if (exact_match) { + owner = qry->sname; + } else { + int ret = dname_wire_reconstruct(owner_buf, k, *cover_low_kwz); + if (unlikely(ret)) return ESKIP; + owner = owner_buf; + } + /* Basic checks OK -> materialize data. */ + { + const struct entry_h *nsec_eh = val.data; + int ret = entry2answer(ans, AR_NSEC, nsec_eh, knot_db_val_bound(val), + owner, KNOT_RRTYPE_NSEC, new_ttl); + if (ret) return kr_error(ret); + } + + /* Final checks, split for matching vs. covering our sname. */ + const knot_rrset_t *nsec_rr = ans->rrsets[AR_NSEC].set.rr; + const uint8_t *bm = knot_nsec_bitmap(nsec_rr->rrs.rdata); + uint16_t bm_size = knot_nsec_bitmap_len(nsec_rr->rrs.rdata); + assert(bm); + + if (exact_match) { + if (kr_nsec_bitmap_nodata_check(bm, bm_size, qry->stype, nsec_rr->owner) != 0) { + VERBOSE_MSG(qry, + "=> NSEC sname: match but failed type check\n"); + return ESKIP; + } + /* NODATA proven; just need to add SOA+RRSIG later */ + VERBOSE_MSG(qry, "=> NSEC sname: match proved NODATA, new TTL %d\n", + new_ttl); + ans->rcode = PKT_NODATA; + return kr_ok(); + } /* else */ + + /* Inexact match. First check if sname is delegated by that NSEC. */ + const int nsec_matched = knot_dname_matched_labels(nsec_rr->owner, qry->sname); + const bool is_sub = nsec_matched == knot_dname_labels(nsec_rr->owner, NULL); + if (is_sub && kr_nsec_children_in_zone_check(bm, bm_size) != 0) { + VERBOSE_MSG(qry, "=> NSEC sname: covered but delegated (or error)\n"); + return ESKIP; + } + /* NXDOMAIN proven *except* for wildcards. */ + WITH_VERBOSE(qry) { + auto_free char *owner_str = kr_dname_text(nsec_rr->owner), + *next_str = kr_dname_text(knot_nsec_next(nsec_rr->rrs.rdata)); + VERBOSE_MSG(qry, "=> NSEC sname: covered by: %s -> %s, new TTL %d\n", + owner_str, next_str, new_ttl); + } + + /* Find label count of the closest encloser. + * Both endpoints in an NSEC do exist (though possibly in a child zone) + * and any prefixes of those names as well (empty non-terminals), + * but nothing else exists inside this "triangle". + * + * Note that we have to lower-case the next name for comparison, + * even though we have canonicalized NSEC already; see RFC 6840 5.1. + * LATER(optim.): it might be faster to use the LFs we already have. + */ + knot_dname_t next[KNOT_DNAME_MAXLEN]; + int ret = knot_dname_to_wire(next, knot_nsec_next(nsec_rr->rrs.rdata), sizeof(next)); + if (ret < 0) { + assert(!ret); + return kr_error(ret); + } + knot_dname_to_lower(next); + *clencl_labels = MAX( + nsec_matched, + knot_dname_matched_labels(qry->sname, next) + ); + + /* Empty non-terminals don't need to have + * a matching NSEC record. */ + if (sname_labels == *clencl_labels) { + ans->rcode = PKT_NODATA; + VERBOSE_MSG(qry, + "=> NSEC sname: empty non-terminal by the same RR\n"); + } else { + ans->rcode = PKT_NXDOMAIN; + } + return kr_ok(); +} + +/** Verify non-existence after kwz_between() call. */ +static bool nonexistence_ok(int cmp, const knot_rrset_t *rrs) +{ + if (cmp == 3) { + return true; + } + if (cmp != 2) { + return false; + } + const uint8_t *bm = knot_nsec_bitmap(rrs->rrs.rdata); + uint16_t bm_size = knot_nsec_bitmap_len(rrs->rrs.rdata); + return kr_nsec_children_in_zone_check(bm, bm_size) != 0; +} + +int nsec1_src_synth(struct key *k, struct answer *ans, const knot_dname_t *clencl_name, + knot_db_val_t cover_low_kwz, knot_db_val_t cover_hi_kwz, + const struct kr_query *qry, struct kr_cache *cache) +{ + /* Construct key for the source of synthesis. */ + knot_db_val_t key = key_NSEC1(k, clencl_name, true); + const size_t nwz_off = key_nwz_off(k); + if (!key.data || key.len < nwz_off) { + assert(false); + return kr_error(1); + } + /* Check if our sname-covering NSEC also covers/matches SS. */ + knot_db_val_t kwz = { + .data = (uint8_t *)key.data + nwz_off, + .len = key.len - nwz_off, + }; + assert((ssize_t)(kwz.len) >= 0); + const int cmp = kwz_between(cover_low_kwz, kwz, cover_hi_kwz); + if (nonexistence_ok(cmp, ans->rrsets[AR_NSEC].set.rr)) { + VERBOSE_MSG(qry, "=> NSEC wildcard: covered by the same RR\n"); + return AR_SOA; + } + const knot_rrset_t *nsec_rr = NULL; /**< the wildcard proof NSEC */ + bool exact_match; /**< whether it matches the source of synthesis */ + if (cmp == 1) { + exact_match = true; + nsec_rr = ans->rrsets[AR_NSEC].set.rr; + } else { + /* Try to find the NSEC for SS. */ + knot_db_val_t val = { NULL, 0 }; + knot_db_val_t wild_low_kwz = { NULL, 0 }; + uint32_t new_ttl; + const char *err = find_leq_NSEC1(cache, qry, key, k, &val, + &exact_match, &wild_low_kwz, NULL, &new_ttl); + if (err) { + VERBOSE_MSG(qry, "=> NSEC wildcard: %s\n", err); + return kr_ok(); + } + /* Materialize the record into answer (speculatively). */ + knot_dname_t owner[KNOT_DNAME_MAXLEN]; + int ret = dname_wire_reconstruct(owner, k, wild_low_kwz); + if (ret) return kr_error(ret); + const struct entry_h *nsec_eh = val.data; + ret = entry2answer(ans, AR_WILD, nsec_eh, knot_db_val_bound(val), + owner, KNOT_RRTYPE_NSEC, new_ttl); + if (ret) return kr_error(ret); + nsec_rr = ans->rrsets[AR_WILD].set.rr; + } + + assert(nsec_rr); + const uint32_t new_ttl_log = + kr_verbose_status ? nsec_rr->ttl : -1; + const uint8_t *bm = knot_nsec_bitmap(nsec_rr->rrs.rdata); + uint16_t bm_size = knot_nsec_bitmap_len(nsec_rr->rrs.rdata); + int ret; + struct answer_rrset * const arw = &ans->rrsets[AR_WILD]; + if (!bm) { + assert(false); + ret = kr_error(1); + goto clean_wild; + } + if (!exact_match) { + /* Finish verification that the source of synthesis doesn't exist. */ + const int nsec_matched = + knot_dname_matched_labels(nsec_rr->owner, clencl_name); + /* we don't need to use the full source of synthesis ^ */ + const bool is_sub = + nsec_matched == knot_dname_labels(nsec_rr->owner, NULL); + if (is_sub && kr_nsec_children_in_zone_check(bm, bm_size) != 0) { + VERBOSE_MSG(qry, + "=> NSEC wildcard: covered but delegated (or error)\n"); + ret = kr_ok(); + goto clean_wild; + } + /* We have a record proving wildcard non-existence. */ + WITH_VERBOSE(qry) { + auto_free char *owner_str = kr_dname_text(nsec_rr->owner), + *next_str = kr_dname_text(knot_nsec_next(nsec_rr->rrs.rdata)); + VERBOSE_MSG(qry, "=> NSEC wildcard: covered by: %s -> %s, new TTL %d\n", + owner_str, next_str, new_ttl_log); + } + return AR_SOA; + } + + /* The wildcard exists. Find if it's NODATA - check type bitmap. */ + if (kr_nsec_bitmap_nodata_check(bm, bm_size, qry->stype, nsec_rr->owner) == 0) { + /* NODATA proven; just need to add SOA+RRSIG later */ + WITH_VERBOSE(qry) { + const char *msg_start = "=> NSEC wildcard: match proved NODATA"; + if (arw->set.rr) { + auto_free char *owner_str = kr_dname_text(nsec_rr->owner); + VERBOSE_MSG(qry, "%s: %s, new TTL %d\n", + msg_start, owner_str, new_ttl_log); + } else { + /* don't repeat the RR if it's the same */ + VERBOSE_MSG(qry, "%s, by the same RR\n", msg_start); + } + } + ans->rcode = PKT_NODATA; + return AR_SOA; + + } /* else */ + /* The data probably exists -> don't add this NSEC + * and (later) try to find the real wildcard data */ + VERBOSE_MSG(qry, "=> NSEC wildcard: should exist (or error)\n"); + ans->rcode = PKT_NOERROR; + ret = kr_ok(); +clean_wild: + if (arw->set.rr) { /* we may have matched AR_NSEC */ + knot_rrset_free(arw->set.rr, ans->mm); + arw->set.rr = NULL; + knot_rdataset_clear(&arw->sig_rds, ans->mm); + } + return ret; +} + diff --git a/lib/cache/nsec3.c b/lib/cache/nsec3.c new file mode 100644 index 0000000..d2ae72a --- /dev/null +++ b/lib/cache/nsec3.c @@ -0,0 +1,501 @@ +/* Copyright (C) 2018 CZ.NIC, z.s.p.o. + + This program is free software: you can redistribute it and/or modify + it under the terms of the GNU General Public License as published by + the Free Software Foundation, either version 3 of the License, or + (at your option) any later version. + + This program is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the + GNU General Public License for more details. + + You should have received a copy of the GNU General Public License + along with this program. If not, see . + */ + +/** @file + * Implementation of NSEC3 handling. Prototypes in ./impl.h + */ + +#include "lib/cache/impl.h" + +#include "contrib/base32hex.h" +#include "lib/dnssec/nsec.h" +#include "lib/layer/iterate.h" + +#include + +static const knot_db_val_t VAL_EMPTY = { NULL, 0 }; + +/** Common part: write all but the NSEC3 hash. */ +static knot_db_val_t key_NSEC3_common(struct key *k, const knot_dname_t *zname, + const nsec_p_hash_t nsec_p_hash) +{ + int ret; + const bool ok = k && zname + && !(ret = kr_dname_lf(k->buf, zname, false)); + if (!ok) { + assert(false); + return VAL_EMPTY; + } + + /* CACHE_KEY_DEF: key == zone's dname_lf + '\0' + '3' + nsec_p hash (4B) + * + NSEC3 hash (20B == NSEC3_HASH_LEN binary!) + * LATER(optim.) nsec_p hash: perhaps 2B would give a sufficient probability + * of avoiding collisions. + */ + uint8_t *begin = k->buf + 1 + k->zlf_len; /* one byte after zone's zero */ + begin[0] = 0; + begin[1] = '3'; /* tag for NSEC3 */ + k->type = KNOT_RRTYPE_NSEC3; + memcpy(begin + 2, &nsec_p_hash, sizeof(nsec_p_hash)); + return (knot_db_val_t){ + .data = k->buf + 1, + .len = begin + 2 + sizeof(nsec_p_hash) - (k->buf + 1), + }; +} + +knot_db_val_t key_NSEC3(struct key *k, const knot_dname_t *nsec3_name, + const nsec_p_hash_t nsec_p_hash) +{ + knot_db_val_t val = key_NSEC3_common(k, nsec3_name /*only zname required*/, + nsec_p_hash); + if (!val.data) return val; + int len = base32hex_decode(nsec3_name + 1, nsec3_name[0], + knot_db_val_bound(val), KR_CACHE_KEY_MAXLEN - val.len); + if (len != NSEC3_HASH_LEN) { + return VAL_EMPTY; + } + val.len += len; + return val; +} + +/** Construct a string key for for NSEC3 predecessor-search, from an non-NSEC3 name. + * \note k->zlf_len and k->zname are assumed to have been correctly set */ +static knot_db_val_t key_NSEC3_name(struct key *k, const knot_dname_t *name, + const bool add_wildcard, const struct nsec_p *nsec_p) +{ + bool ok = k && name && nsec_p && nsec_p->raw; + if (!ok) return VAL_EMPTY; + knot_db_val_t val = key_NSEC3_common(k, k->zname, nsec_p->hash); + if (!val.data) return val; + + /* Make `name` point to correctly wildcarded owner name. */ + uint8_t buf[KNOT_DNAME_MAXLEN]; + int name_len; + if (add_wildcard) { + buf[0] = '\1'; + buf[1] = '*'; + name_len = knot_dname_to_wire(buf + 2, name, sizeof(buf) - 2); + if (name_len < 0) return VAL_EMPTY; /* wants wildcard but doesn't fit */ + name = buf; + name_len += 2; + } else { + name_len = knot_dname_size(name); + } + /* Append the NSEC3 hash. */ + const dnssec_binary_t dname = { + .size = name_len, + .data = (uint8_t *)/*const-cast*/name, + }; + + #if 0 // LATER(optim.): this requires a patched libdnssec - tries to realloc() + dnssec_binary_t hash = { + .size = KR_CACHE_KEY_MAXLEN - val.len, + .data = val.data + val.len, + }; + int ret = dnssec_nsec3_hash(&dname, &nsec_p->libknot, &hash); + if (ret != DNSSEC_EOK) return VAL_EMPTY; + assert(hash.size == NSEC3_HASH_LEN); + + #else + dnssec_binary_t hash = { .size = 0, .data = NULL }; + int ret = dnssec_nsec3_hash(&dname, &nsec_p->libknot, &hash); + if (ret != DNSSEC_EOK) return VAL_EMPTY; + if (hash.size != NSEC3_HASH_LEN || !hash.data) { + assert(false); + return VAL_EMPTY; + } + memcpy(knot_db_val_bound(val), hash.data, NSEC3_HASH_LEN); + free(hash.data); + #endif + + val.len += hash.size; + return val; +} + +/** Return h1 < h2, semantically on NSEC3 hashes. */ +static inline bool nsec3_hash_ordered(const uint8_t *h1, const uint8_t *h2) +{ + return memcmp(h1, h2, NSEC3_HASH_LEN) < 0; +} + +/** NSEC3 range search. + * + * \param key Pass output of key_NSEC3(k, ...) + * \param nsec_p Restrict to this NSEC3 parameter-set. + * \param value[out] The raw data of the NSEC3 cache record (optional; consistency checked). + * \param exact_match[out] Whether the key was matched exactly or just covered (optional). + * \param hash_low[out] Output the low end hash of covering NSEC3, pointing within DB (optional). + * \param new_ttl[out] New TTL of the NSEC3 (optional). + * \return Error message or NULL. + * \note The function itself does *no* bitmap checks, e.g. RFC 6840 sec. 4. + */ +static const char * find_leq_NSEC3(struct kr_cache *cache, const struct kr_query *qry, + const knot_db_val_t key, const struct key *k, const struct nsec_p *nsec_p, + knot_db_val_t *value, bool *exact_match, const uint8_t **hash_low, + uint32_t *new_ttl) +{ + /* Do the cache operation. */ + const size_t hash_off = key_nsec3_hash_off(k); + if (!key.data || key.len < hash_off) { + assert(false); + return "range search ERROR"; + } + knot_db_val_t key_found = key; + knot_db_val_t val = { NULL, 0 }; + int ret = cache_op(cache, read_leq, &key_found, &val); + /* ^^ LATER(optim.): incrementing key and doing less-than search + * would probably be slightly more efficient with LMDB, + * but the code complexity would grow considerably. */ + if (ret < 0) { + if (ret == kr_error(ENOENT)) { + return "range search miss"; + } else { + assert(false); + return "range search ERROR"; + } + } + if (value) { + *value = val; + } + /* Check consistency, TTL, rank. */ + const bool is_exact = (ret == 0); + if (exact_match) { + *exact_match = is_exact; + } + const struct entry_h *eh = entry_h_consistent_NSEC(val); + if (!eh) { + /* This might be just finding something else than NSEC3 entry, + * in case we searched before the very first one in the zone. */ + return "range search found inconsistent entry"; + } + /* Passing just zone name instead of owner. */ + int32_t new_ttl_ = get_new_ttl(eh, qry, k->zname, KNOT_RRTYPE_NSEC3, + qry->timestamp.tv_sec); + if (new_ttl_ < 0 || !kr_rank_test(eh->rank, KR_RANK_SECURE)) { + return "range search found stale or insecure entry"; + /* TODO: remove the stale record *and* retry, + * in case we haven't run off. Perhaps start by in_zone check. */ + } + if (new_ttl) { + *new_ttl = new_ttl_; + } + if (hash_low) { + *hash_low = (uint8_t *)key_found.data + hash_off; + } + if (is_exact) { + /* Nothing else to do. */ + return NULL; + } + /* The NSEC3 starts strictly before our target name; + * now check that it still belongs into that zone and chain. */ + const uint8_t *nsec_p_raw = eh->data + KR_CACHE_RR_COUNT_SIZE + + 2 /* RDLENGTH from rfc1034 */; + const int nsec_p_len = nsec_p_rdlen(nsec_p_raw); + const bool same_chain = key_found.len == hash_off + NSEC3_HASH_LEN + /* CACHE_KEY_DEF */ + && memcmp(key.data, key_found.data, hash_off) == 0 + /* exact comparison of NSEC3 parameters */ + && nsec_p_len == nsec_p_rdlen(nsec_p->raw) + && memcmp(nsec_p_raw, nsec_p->raw, nsec_p_len) == 0; + if (!same_chain) { + return "range search miss (!same_chain)"; + } + /* We know it starts before sname, so let's check the other end. + * A. find the next hash and check its length. */ + if (KR_CACHE_RR_COUNT_SIZE != 2 || get_uint16(eh->data) == 0) { + assert(false); + return "ERROR"; + /* TODO: more checks? Also, `next` computation is kinda messy. */ + } + const uint8_t *hash_next = nsec_p_raw + nsec_p_len + + sizeof(uint8_t) /* hash length from rfc5155 */; + if (hash_next[-1] != NSEC3_HASH_LEN) { + return "unexpected next hash length"; + } + /* B. do the actual range check. */ + const uint8_t * const hash_searched = (uint8_t *)key.data + hash_off; + bool covers = /* we know for sure that the low end is before the searched name */ + nsec3_hash_ordered(hash_searched, hash_next) + /* and the wrap-around case */ + || nsec3_hash_ordered(hash_next, (const uint8_t *)key_found.data + hash_off); + if (!covers) { + return "range search miss (!covers)"; + } + return NULL; +} + +/** Extract textual representation of NSEC3 hash from a cache key. + * \param text must have length at least NSEC3_HASH_TXT_LEN+1 (will get 0-terminated). */ +static void key_NSEC3_hash2text(const knot_db_val_t key, char *text) +{ + assert(key.data && key.len > NSEC3_HASH_LEN); + const uint8_t *hash_raw = knot_db_val_bound(key) - NSEC3_HASH_LEN; + /* CACHE_KEY_DEF ^^ */ + int len = base32hex_encode(hash_raw, NSEC3_HASH_LEN, (uint8_t *)text, + NSEC3_HASH_TXT_LEN); + assert(len == NSEC3_HASH_TXT_LEN); (void)len; + text[NSEC3_HASH_TXT_LEN] = '\0'; +} + +/** Reconstruct a name into a buffer (assuming length at least KNOT_DNAME_MAXLEN). + * \return kr_ok() or error code (<0). */ +static int dname_wire_reconstruct(knot_dname_t *buf, const knot_dname_t *zname, + const uint8_t *hash_raw) +{ + int len = base32hex_encode(hash_raw, NSEC3_HASH_LEN, buf + 1, NSEC3_HASH_TXT_LEN); + if (len != NSEC3_HASH_TXT_LEN) { + assert(false); + return kr_error(EINVAL); + } + buf[0] = len; + int ret = knot_dname_to_wire(buf + 1 + len, zname, KNOT_DNAME_MAXLEN - 1 - len); + return ret < 0 ? kr_error(ret) : kr_ok(); +} + +static void nsec3_hash2text(const knot_dname_t *owner, char *text) +{ + assert(owner[0] == NSEC3_HASH_TXT_LEN); + memcpy(text, owner + 1, MIN(owner[0], NSEC3_HASH_TXT_LEN)); + text[NSEC3_HASH_TXT_LEN] = '\0'; +} + +int nsec3_encloser(struct key *k, struct answer *ans, + const int sname_labels, int *clencl_labels, + const struct kr_query *qry, struct kr_cache *cache) +{ + static const int ESKIP = ABS(ENOENT); + /* Basic sanity check. */ + const bool ok = k && k->zname && ans && clencl_labels + && qry && cache; + if (!ok) { + assert(!EINVAL); + return kr_error(EINVAL); + } + + /*** Find the closest encloser - cycle: name starting at sname, + * proceeding while longer than zname, shortening by one label on step. + * We need a pair where a name doesn't exist *and* its parent does. */ + /* LATER(optim.): perhaps iterate in the other order - that + * should help significantly against deep queries where we have + * a shallow proof in the cache. We can also optimize by using + * only exact search unless we had a match in the previous iteration. */ + const int zname_labels = knot_dname_labels(k->zname, NULL); + int last_nxproven_labels = -1; + const knot_dname_t *name = qry->sname; + for (int name_labels = sname_labels; name_labels >= zname_labels; + --name_labels, name += 1 + name[0]) { + /* Find a previous-or-equal NSEC3 in cache covering the name, + * checking TTL etc. */ + const knot_db_val_t key = key_NSEC3_name(k, name, false, &ans->nsec_p); + if (!key.data) continue; + WITH_VERBOSE(qry) { + char hash_txt[NSEC3_HASH_TXT_LEN + 1]; + key_NSEC3_hash2text(key, hash_txt); + VERBOSE_MSG(qry, "=> NSEC3 depth %d: hash %s\n", + name_labels - zname_labels, hash_txt); + } + knot_db_val_t val = { NULL, 0 }; + bool exact_match; + uint32_t new_ttl; + const uint8_t *hash_low; + const char *err = find_leq_NSEC3(cache, qry, key, k, &ans->nsec_p, &val, + &exact_match, &hash_low, &new_ttl); + if (err) { + WITH_VERBOSE(qry) { + auto_free char *name_str = kr_dname_text(name); + VERBOSE_MSG(qry, "=> NSEC3 encloser error for %s: %s\n", + name_str, err); + } + continue; + } + if (exact_match && name_labels != sname_labels + && name_labels + 1 != last_nxproven_labels) { + /* This name exists (checked rank and TTL), and it's + * neither of the two interesting cases, so we do not + * keep searching for non-existence above this name. */ + VERBOSE_MSG(qry, + "=> NSEC3 encloser: only found existence of an ancestor\n"); + return ESKIP; + } + /* Optimization: avoid the rest of the last iteration if pointless. */ + if (!exact_match && name_labels == zname_labels + && last_nxproven_labels != name_labels + 1) { + break; + } + + /* Basic checks OK -> materialize data, cleaning any previous + * records on that answer index (unsuccessful attempts). */ + knot_dname_t owner[KNOT_DNAME_MAXLEN]; + { + int ret = dname_wire_reconstruct(owner, k->zname, hash_low); + if (unlikely(ret)) continue; + } + const int ans_id = (exact_match && name_labels + 1 == last_nxproven_labels) + ? AR_CPE : AR_NSEC; + { + const struct entry_h *nsec_eh = val.data; + memset(&ans->rrsets[ans_id], 0, sizeof(ans->rrsets[ans_id])); + int ret = entry2answer(ans, ans_id, nsec_eh, knot_db_val_bound(val), + owner, KNOT_RRTYPE_NSEC3, new_ttl); + if (ret) return kr_error(ret); + } + + if (!exact_match) { + /* Non-existence proven, but we don't know if `name` + * is the next closer name. + * Note: we don't need to check for the sname being + * delegated away by this record, as with NSEC3 only + * *exact* match on an ancestor could do that. */ + last_nxproven_labels = name_labels; + WITH_VERBOSE(qry) { + char hash_low_txt[NSEC3_HASH_TXT_LEN + 1]; + nsec3_hash2text(owner, hash_low_txt); + VERBOSE_MSG(qry, + "=> NSEC3 depth %d: covered by %s -> TODO, new TTL %d\n", + name_labels - zname_labels, hash_low_txt, new_ttl); + } + continue; + } + + /* Exactly matched NSEC3: two cases, one after another. */ + const knot_rrset_t *nsec_rr = ans->rrsets[ans_id].set.rr; + const uint8_t *bm = knot_nsec3_bitmap(nsec_rr->rrs.rdata); + uint16_t bm_size = knot_nsec3_bitmap_len(nsec_rr->rrs.rdata); + assert(bm); + if (name_labels == sname_labels) { + if (kr_nsec_bitmap_nodata_check(bm, bm_size, qry->stype, + nsec_rr->owner) != 0) { + VERBOSE_MSG(qry, + "=> NSEC3 sname: match but failed type check\n"); + return ESKIP; + } + /* NODATA proven; just need to add SOA+RRSIG later */ + VERBOSE_MSG(qry, + "=> NSEC3 sname: match proved NODATA, new TTL %d\n", + new_ttl); + ans->rcode = PKT_NODATA; + return kr_ok(); + + } /* else */ + + assert(name_labels + 1 == last_nxproven_labels); + if (kr_nsec_children_in_zone_check(bm, bm_size) != 0) { + VERBOSE_MSG(qry, + "=> NSEC3 encloser: found but delegated (or error)\n"); + return ESKIP; + } + /* NXDOMAIN proven *except* for wildcards. */ + WITH_VERBOSE(qry) { + auto_free char *name_str = kr_dname_text(name); + VERBOSE_MSG(qry, + "=> NSEC3 encloser: confirmed as %s, new TTL %d\n", + name_str, new_ttl); + } + *clencl_labels = name_labels; + ans->rcode = PKT_NXDOMAIN; + /* Avoid repeated NSEC3 - remove either if the hashes match. + * This is very unlikely in larger zones: 1/size (per attempt). + * Well, deduplication would happen anyway when the answer + * from cache is read by kresd (internally). */ + if (unlikely(0 == memcmp(ans->rrsets[AR_NSEC].set.rr->owner + 1, + ans->rrsets[AR_CPE ].set.rr->owner + 1, + NSEC3_HASH_LEN))) { + memset(&ans->rrsets[AR_CPE], 0, sizeof(ans->rrsets[AR_CPE])); + /* LATER(optim.): perhaps check this earlier and avoid some work? */ + } + return kr_ok(); + } + + /* We've ran out of options. */ + if (last_nxproven_labels > 0) { + /* We didn't manage to prove existence of the closest encloser, + * meaning the only chance left is a *positive* wildcard record. */ + *clencl_labels = last_nxproven_labels - 1; + ans->rcode = PKT_NXDOMAIN; + /* FIXME: review */ + } + return ESKIP; +} + +int nsec3_src_synth(struct key *k, struct answer *ans, const knot_dname_t *clencl_name, + const struct kr_query *qry, struct kr_cache *cache) +{ + /* Find a previous-or-equal NSEC3 in cache covering or matching + * the source of synthesis, checking TTL etc. */ + const knot_db_val_t key = key_NSEC3_name(k, clencl_name, true, &ans->nsec_p); + if (!key.data) return kr_error(1); + WITH_VERBOSE(qry) { + char hash_txt[NSEC3_HASH_TXT_LEN + 1]; + key_NSEC3_hash2text(key, hash_txt); + VERBOSE_MSG(qry, "=> NSEC3 wildcard: hash %s\n", hash_txt); + } + knot_db_val_t val = { NULL, 0 }; + bool exact_match; + uint32_t new_ttl; + const uint8_t *hash_low; + const char *err = find_leq_NSEC3(cache, qry, key, k, &ans->nsec_p, &val, + &exact_match, &hash_low, &new_ttl); + if (err) { + VERBOSE_MSG(qry, "=> NSEC3 wildcard: %s\n", err); + return kr_ok(); + } + + /* LATER(optim.): avoid duplicities in answer. */ + + /* Basic checks OK -> materialize the data (speculatively). */ + knot_dname_t owner[KNOT_DNAME_MAXLEN]; + { + int ret = dname_wire_reconstruct(owner, k->zname, hash_low); + if (unlikely(ret)) return kr_ok(); + const struct entry_h *nsec_eh = val.data; + ret = entry2answer(ans, AR_WILD, nsec_eh, knot_db_val_bound(val), + owner, KNOT_RRTYPE_NSEC3, new_ttl); + if (ret) return kr_error(ret); + } + const knot_rrset_t *nsec_rr = ans->rrsets[AR_WILD].set.rr; + + if (!exact_match) { + /* The record proves wildcard non-existence. */ + WITH_VERBOSE(qry) { + char hash_low_txt[NSEC3_HASH_TXT_LEN + 1]; + nsec3_hash2text(owner, hash_low_txt); + VERBOSE_MSG(qry, + "=> NSEC3 wildcard: covered by %s -> TODO, new TTL %d\n", + hash_low_txt, new_ttl); + } + return AR_SOA; + } + + /* The wildcard exists. Find if it's NODATA - check type bitmap. */ + const uint8_t *bm = knot_nsec3_bitmap(nsec_rr->rrs.rdata); + uint16_t bm_size = knot_nsec3_bitmap_len(nsec_rr->rrs.rdata); + assert(bm); + if (kr_nsec_bitmap_nodata_check(bm, bm_size, qry->stype, nsec_rr->owner) == 0) { + /* NODATA proven; just need to add SOA+RRSIG later */ + VERBOSE_MSG(qry, "=> NSEC3 wildcard: match proved NODATA, new TTL %d\n", + new_ttl); + ans->rcode = PKT_NODATA; + return AR_SOA; + + } /* else */ + /* The data probably exists -> don't add this NSEC3 + * and (later) try to find the real wildcard data */ + VERBOSE_MSG(qry, "=> NSEC3 wildcard: should exist (or error)\n"); + ans->rcode = PKT_NOERROR; + memset(&ans->rrsets[AR_WILD], 0, sizeof(ans->rrsets[AR_WILD])); + return kr_ok(); +} + diff --git a/lib/cache/peek.c b/lib/cache/peek.c new file mode 100644 index 0000000..1af1627 --- /dev/null +++ b/lib/cache/peek.c @@ -0,0 +1,724 @@ +/* Copyright (C) 2018 CZ.NIC, z.s.p.o. + + This program is free software: you can redistribute it and/or modify + it under the terms of the GNU General Public License as published by + the Free Software Foundation, either version 3 of the License, or + (at your option) any later version. + + This program is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the + GNU General Public License for more details. + + You should have received a copy of the GNU General Public License + along with this program. If not, see . + */ + +#include "lib/cache/impl.h" + +#include "lib/dnssec/ta.h" +#include "lib/layer/iterate.h" + +/* The whole file only exports peek_nosync(). + * Forwards for larger chunks of code: */ + +static int found_exact_hit(kr_layer_t *ctx, knot_pkt_t *pkt, knot_db_val_t val, + uint8_t lowest_rank); +static int closest_NS(struct kr_cache *cache, struct key *k, entry_list_t el, + struct kr_query *qry, bool only_NS, bool is_DS); +static int answer_simple_hit(kr_layer_t *ctx, knot_pkt_t *pkt, uint16_t type, + const struct entry_h *eh, const void *eh_bound, uint32_t new_ttl); +static int try_wild(struct key *k, struct answer *ans, const knot_dname_t *clencl_name, + uint16_t type, uint8_t lowest_rank, + const struct kr_query *qry, struct kr_cache *cache); + +static int peek_encloser( + struct key *k, struct answer *ans, int sname_labels, + uint8_t lowest_rank, const struct kr_query *qry, struct kr_cache *cache); + + +static int nsec_p_init(struct nsec_p *nsec_p, knot_db_val_t nsec_p_entry, bool with_knot) +{ + const size_t stamp_len = sizeof(uint32_t); + if (nsec_p_entry.len <= stamp_len) { /* plain NSEC if equal */ + nsec_p->raw = NULL; + nsec_p->hash = 0; + return kr_ok(); + } + nsec_p->raw = (uint8_t *)nsec_p_entry.data + stamp_len; + nsec_p->hash = nsec_p_mkHash(nsec_p->raw); + if (!with_knot) return kr_ok(); + /* Convert NSEC3 params to another format. */ + const dnssec_binary_t rdata = { + .size = nsec_p_rdlen(nsec_p->raw), + .data = (uint8_t *)/*const-cast*/nsec_p->raw, + }; + int ret = dnssec_nsec3_params_from_rdata(&nsec_p->libknot, &rdata); + return ret == DNSSEC_EOK ? kr_ok() : kr_error(ret); +} + +static void nsec_p_cleanup(struct nsec_p *nsec_p) +{ + dnssec_binary_free(&nsec_p->libknot.salt); + /* We don't really need to clear it, but it's not large. (`salt` zeroed above) */ + memset(nsec_p, 0, sizeof(*nsec_p)); +} + +/** Compute new TTL for nsec_p entry, using SOA serial arith. + * \param new_ttl (optionally) write the new TTL (even if negative) + * \return error code, e.g. kr_error(ESTALE) */ +static int nsec_p_ttl(knot_db_val_t entry, const uint32_t timestamp, int32_t *new_ttl) +{ + if (!entry.data) { + assert(!EINVAL); + return kr_error(EINVAL); + } + uint32_t stamp; + if (!entry.len) { + return kr_error(ENOENT); + } + if (entry.len < sizeof(stamp)) { + assert(!EILSEQ); + return kr_error(EILSEQ); + } + memcpy(&stamp, entry.data, sizeof(stamp)); + int32_t newttl = stamp - timestamp; + if (new_ttl) *new_ttl = newttl; + return newttl < 0 ? kr_error(ESTALE) : kr_ok(); +} + +static uint8_t get_lowest_rank(const struct kr_request *req, const struct kr_query *qry) +{ + /* TODO: move rank handling into the iterator (DNSSEC_* flags)? */ + const bool allow_unverified = + knot_wire_get_cd(req->qsource.packet->wire) || qry->flags.STUB; + /* in stub mode we don't trust RRs anyway ^^ */ + if (qry->flags.NONAUTH) { + return KR_RANK_INITIAL; + /* Note: there's little sense in validation status for non-auth records. + * In case of using NONAUTH to get NS IPs, knowing that you ask correct + * IP doesn't matter much for security; it matters whether you can + * validate the answers from the NS. + */ + } else if (!allow_unverified) { + /* Records not present under any TA don't have their security + * verified at all, so we also accept low ranks in that case. */ + const bool ta_covers = kr_ta_covers_qry(req->ctx, qry->sname, qry->stype); + /* ^ TODO: performance? TODO: stype - call sites */ + if (ta_covers) { + return KR_RANK_INSECURE | KR_RANK_AUTH; + } /* else falltrhough */ + } + return KR_RANK_INITIAL | KR_RANK_AUTH; +} + + +/** Almost whole .produce phase for the cache module. + * \note we don't transition to KR_STATE_FAIL even in case of "unexpected errors". + */ +int peek_nosync(kr_layer_t *ctx, knot_pkt_t *pkt) +{ + struct kr_request *req = ctx->req; + struct kr_query *qry = req->current_query; + struct kr_cache *cache = &req->ctx->cache; + + struct key k_storage, *k = &k_storage; + int ret = kr_dname_lf(k->buf, qry->sname, false); + if (unlikely(ret)) { + assert(false); + return ctx->state; + } + + const uint8_t lowest_rank = get_lowest_rank(req, qry); + + /**** 1. find the name or the closest (available) zone, not considering wildcards + **** 1a. exact name+type match (can be negative answer in insecure zones) */ + { + knot_db_val_t key = key_exact_type_maypkt(k, qry->stype); + knot_db_val_t val = { NULL, 0 }; + ret = cache_op(cache, read, &key, &val, 1); + if (!ret) { + /* found an entry: test conditions, materialize into pkt, etc. */ + ret = found_exact_hit(ctx, pkt, val, lowest_rank); + } + } + if (ret && ret != -abs(ENOENT)) { + VERBOSE_MSG(qry, "=> exact hit error: %d %s\n", ret, kr_strerror(ret)); + assert(false); + return ctx->state; + } else if (!ret) { + return KR_STATE_DONE; + } + + /**** 1b. otherwise, find the longest prefix zone/xNAME (with OK time+rank). [...] */ + k->zname = qry->sname; + ret = kr_dname_lf(k->buf, k->zname, false); /* LATER(optim.): probably remove */ + if (unlikely(ret)) { + assert(false); + return ctx->state; + } + entry_list_t el; + ret = closest_NS(cache, k, el, qry, false, qry->stype == KNOT_RRTYPE_DS); + if (ret) { + assert(ret == kr_error(ENOENT)); + if (ret != kr_error(ENOENT) || !el[0].len) { + return ctx->state; + } + } + switch (k->type) { + case KNOT_RRTYPE_CNAME: { + const knot_db_val_t v = el[EL_CNAME]; + assert(v.data && v.len); + const int32_t new_ttl = get_new_ttl(v.data, qry, qry->sname, + KNOT_RRTYPE_CNAME, qry->timestamp.tv_sec); + ret = answer_simple_hit(ctx, pkt, KNOT_RRTYPE_CNAME, v.data, + knot_db_val_bound(v), new_ttl); + /* TODO: ^^ cumbersome code; we also recompute the TTL */ + return ret == kr_ok() ? KR_STATE_DONE : ctx->state; + } + case KNOT_RRTYPE_DNAME: + VERBOSE_MSG(qry, "=> DNAME not supported yet\n"); // LATER + return ctx->state; + } + + /* We have to try proving from NSEC*. */ + auto_free char *log_zname = NULL; + WITH_VERBOSE(qry) { + log_zname = kr_dname_text(k->zname); + if (!el[0].len) { + VERBOSE_MSG(qry, "=> no NSEC* cached for zone: %s\n", log_zname); + } + } + +#if 0 + if (!eh) { /* fall back to root hints? */ + ret = kr_zonecut_set_sbelt(req->ctx, &qry->zone_cut); + if (ret) return ctx->state; + assert(!qry->zone_cut.parent); + + //VERBOSE_MSG(qry, "=> using root hints\n"); + //qry->flags.AWAIT_CUT = false; + return ctx->state; + } + + /* Now `eh` points to the closest NS record that we've found, + * and that's the only place to start - we may either find + * a negative proof or we may query upstream from that point. */ + kr_zonecut_set(&qry->zone_cut, k->zname); + ret = kr_make_query(qry, pkt); // TODO: probably not yet - qname minimization + if (ret) return ctx->state; +#endif + + /** Structure for collecting multiple NSEC* + RRSIG records, + * in preparation for the answer, and for tracking the progress. */ + struct answer ans; + memset(&ans, 0, sizeof(ans)); + ans.mm = &pkt->mm; + const int sname_labels = knot_dname_labels(qry->sname, NULL); + + /* Try the NSEC* parameters in order, until success. + * Let's not mix different parameters for NSEC* RRs in a single proof. */ + for (int i = 0; ;) { + int32_t log_new_ttl = -123456789; /* visually recognizable value */ + ret = nsec_p_ttl(el[i], qry->timestamp.tv_sec, &log_new_ttl); + if (!ret || VERBOSE_STATUS) { + nsec_p_init(&ans.nsec_p, el[i], !ret); + } + if (ret) { + VERBOSE_MSG(qry, "=> skipping zone: %s, %s, hash %x;" + "new TTL %d, ret %d\n", + log_zname, (ans.nsec_p.raw ? "NSEC3" : "NSEC"), + (unsigned)ans.nsec_p.hash, (int)log_new_ttl, ret); + /* no need for nsec_p_cleanup() in this case */ + goto cont; + } + VERBOSE_MSG(qry, "=> trying zone: %s, %s, hash %x\n", + log_zname, (ans.nsec_p.raw ? "NSEC3" : "NSEC"), + (unsigned)ans.nsec_p.hash); + /**** 2. and 3. inside */ + ret = peek_encloser(k, &ans, sname_labels, + lowest_rank, qry, cache); + nsec_p_cleanup(&ans.nsec_p); + if (!ret) break; + if (ret < 0) return ctx->state; + cont: + /* Otherwise we try another nsec_p, if available. */ + if (++i == ENTRY_APEX_NSECS_CNT) return ctx->state; + /* clear possible partial answers in `ans` (no need to deallocate) */ + ans.rcode = 0; + memset(&ans.rrsets, 0, sizeof(ans.rrsets)); + } + + /**** 4. add SOA iff needed */ + if (ans.rcode != PKT_NOERROR) { + /* Assuming k->buf still starts with zone's prefix, + * look up the SOA in cache. */ + k->buf[0] = k->zlf_len; + knot_db_val_t key = key_exact_type(k, KNOT_RRTYPE_SOA); + knot_db_val_t val = { NULL, 0 }; + ret = cache_op(cache, read, &key, &val, 1); + const struct entry_h *eh; + if (ret || !(eh = entry_h_consistent(val, KNOT_RRTYPE_SOA))) { + assert(ret); /* only want to catch `eh` failures */ + VERBOSE_MSG(qry, "=> SOA missed\n"); + return ctx->state; + } + /* Check if the record is OK. */ + int32_t new_ttl = get_new_ttl(eh, qry, k->zname, KNOT_RRTYPE_SOA, + qry->timestamp.tv_sec); + if (new_ttl < 0 || eh->rank < lowest_rank || eh->is_packet) { + VERBOSE_MSG(qry, "=> SOA unfit %s: rank 0%.2o, new TTL %d\n", + (eh->is_packet ? "packet" : "RR"), + eh->rank, new_ttl); + return ctx->state; + } + /* Add the SOA into the answer. */ + ret = entry2answer(&ans, AR_SOA, eh, knot_db_val_bound(val), + k->zname, KNOT_RRTYPE_SOA, new_ttl); + if (ret) return ctx->state; + } + + /* Find our target RCODE. */ + int real_rcode; + switch (ans.rcode) { + case PKT_NODATA: + case PKT_NOERROR: /* positive wildcarded response */ + real_rcode = KNOT_RCODE_NOERROR; + break; + case PKT_NXDOMAIN: + real_rcode = KNOT_RCODE_NXDOMAIN; + break; + default: + assert(false); + case 0: /* i.e. nothing was found */ + /* LATER(optim.): zone cut? */ + VERBOSE_MSG(qry, "=> cache miss\n"); + return ctx->state; + } + + if (pkt_renew(pkt, qry->sname, qry->stype) + || knot_pkt_begin(pkt, KNOT_ANSWER) + ) { + assert(false); + return ctx->state; + } + knot_wire_set_rcode(pkt->wire, real_rcode); + + bool expiring = false; // TODO + VERBOSE_MSG(qry, "=> writing RRsets: "); + for (int i = 0; i < sizeof(ans.rrsets) / sizeof(ans.rrsets[0]); ++i) { + if (i == 1) knot_pkt_begin(pkt, KNOT_AUTHORITY); + if (!ans.rrsets[i].set.rr) continue; + expiring = expiring || ans.rrsets[i].set.expiring; + ret = pkt_append(pkt, &ans.rrsets[i], ans.rrsets[i].set.rank); + if (ret) { + assert(false); + return ctx->state; + } + kr_log_verbose(kr_rank_test(ans.rrsets[i].set.rank, KR_RANK_SECURE) + ? "+" : "-"); + } + kr_log_verbose("\n"); + + /* Finishing touches. */ + struct kr_qflags * const qf = &qry->flags; + qf->EXPIRING = expiring; + qf->CACHED = true; + qf->NO_MINIMIZE = true; + + return KR_STATE_DONE; +} + +/** + * This is where the high-level "business logic" of aggressive cache is. + * \return 0: success (may need SOA); >0: try other nsec_p; <0: exit cache immediately. + */ +static int peek_encloser( + struct key *k, struct answer *ans, const int sname_labels, + uint8_t lowest_rank, const struct kr_query *qry, struct kr_cache *cache) +{ + /** Start of NSEC* covering the sname; + * it's part of key - the one within zone (read only) */ + knot_db_val_t cover_low_kwz = { NULL, 0 }; + knot_dname_t cover_hi_storage[KNOT_DNAME_MAXLEN]; + /** End of NSEC* covering the sname. */ + knot_db_val_t cover_hi_kwz = { + .data = cover_hi_storage, + .len = sizeof(cover_hi_storage), + }; + + /**** 2. Find a closest (provable) encloser (of sname). */ + int clencl_labels = -1; + bool clencl_is_tentative = false; + if (!ans->nsec_p.raw) { /* NSEC */ + int ret = nsec1_encloser(k, ans, sname_labels, &clencl_labels, + &cover_low_kwz, &cover_hi_kwz, qry, cache); + if (ret) return ret; + } else { + int ret = nsec3_encloser(k, ans, sname_labels, &clencl_labels, + qry, cache); + clencl_is_tentative = ret == ABS(ENOENT) && clencl_labels >= 0; + /* ^^ Last chance: *positive* wildcard record under this clencl. */ + if (ret && !clencl_is_tentative) return ret; + } + + /* We should have either a match or a cover at this point. */ + if (ans->rcode != PKT_NODATA && ans->rcode != PKT_NXDOMAIN) { + assert(false); + return kr_error(EINVAL); + } + const bool ncloser_covered = ans->rcode == PKT_NXDOMAIN; + + /** Name of the closest (provable) encloser. */ + const knot_dname_t *clencl_name = qry->sname; + for (int l = sname_labels; l > clencl_labels; --l) + clencl_name = knot_wire_next_label(clencl_name, NULL); + + /**** 3. source of synthesis checks, in case the next closer name was covered. + **** 3a. We want to query for NSEC* of source of synthesis (SS) or its + * predecessor, providing us with a proof of its existence or non-existence. */ + if (ncloser_covered && !ans->nsec_p.raw) { + int ret = nsec1_src_synth(k, ans, clencl_name, + cover_low_kwz, cover_hi_kwz, qry, cache); + if (ret == AR_SOA) return 0; + assert(ret <= 0); + if (ret) return ret; + + } else if (ncloser_covered && ans->nsec_p.raw && !clencl_is_tentative) { + int ret = nsec3_src_synth(k, ans, clencl_name, qry, cache); + if (ret == AR_SOA) return 0; + assert(ret <= 0); + if (ret) return ret; + + } /* else (!ncloser_covered) so no wildcard checks needed, + * as we proved that sname exists. */ + + /**** 3b. find wildcarded answer, if next closer name was covered + * and we don't have a full proof yet. (common for NSEC*) */ + if (!ncloser_covered) + return kr_ok(); /* decrease indentation */ + /* Construct key for exact qry->stype + source of synthesis. */ + int ret = kr_dname_lf(k->buf, clencl_name, true); + if (ret) { + assert(!ret); + return kr_error(ret); + } + const uint16_t types[] = { qry->stype, KNOT_RRTYPE_CNAME }; + for (int i = 0; i < (2 - (qry->stype == KNOT_RRTYPE_CNAME)); ++i) { + ret = try_wild(k, ans, clencl_name, types[i], + lowest_rank, qry, cache); + if (ret == kr_ok()) { + return kr_ok(); + } else if (ret != -ABS(ENOENT) && ret != -ABS(ESTALE)) { + assert(false); + return kr_error(ret); + } + /* else continue */ + } + /* Neither attempt succeeded, but the NSEC* proofs were found, + * so skip trying other parameters, as it seems very unlikely + * to turn out differently than by the same wildcard search. */ + return -ABS(ENOENT); +} + + +static int answer_simple_hit(kr_layer_t *ctx, knot_pkt_t *pkt, uint16_t type, + const struct entry_h *eh, const void *eh_bound, uint32_t new_ttl) +#define CHECK_RET(ret) do { \ + if ((ret) < 0) { assert(false); return kr_error((ret)); } \ +} while (false) +{ + struct kr_request *req = ctx->req; + struct kr_query *qry = req->current_query; + + /* All OK, so start constructing the (pseudo-)packet. */ + int ret = pkt_renew(pkt, qry->sname, qry->stype); + CHECK_RET(ret); + + /* Materialize the sets for the answer in (pseudo-)packet. */ + struct answer ans; + memset(&ans, 0, sizeof(ans)); + ans.mm = &pkt->mm; + ret = entry2answer(&ans, AR_ANSWER, eh, eh_bound, + qry->sname, type, new_ttl); + CHECK_RET(ret); + /* Put links to the materialized data into the pkt. */ + ret = pkt_append(pkt, &ans.rrsets[AR_ANSWER], eh->rank); + CHECK_RET(ret); + + /* Finishing touches. */ + struct kr_qflags * const qf = &qry->flags; + qf->EXPIRING = is_expiring(eh->ttl, new_ttl); + qf->CACHED = true; + qf->NO_MINIMIZE = true; + qf->DNSSEC_INSECURE = kr_rank_test(eh->rank, KR_RANK_INSECURE); + if (qf->DNSSEC_INSECURE) { + qf->DNSSEC_WANT = false; + } + VERBOSE_MSG(qry, "=> satisfied by exact %s: rank 0%.2o, new TTL %d\n", + (type == KNOT_RRTYPE_CNAME ? "CNAME" : "RRset"), + eh->rank, new_ttl); + return kr_ok(); +} +#undef CHECK_RET + + +/** TODO: description; see the single call site for now. */ +static int found_exact_hit(kr_layer_t *ctx, knot_pkt_t *pkt, knot_db_val_t val, + uint8_t lowest_rank) +{ + struct kr_request *req = ctx->req; + struct kr_query *qry = req->current_query; + + int ret = entry_h_seek(&val, qry->stype); + if (ret) return ret; + const struct entry_h *eh = entry_h_consistent(val, qry->stype); + if (!eh) { + assert(false); + return kr_error(ENOENT); + // LATER: recovery in case of error, perhaps via removing the entry? + // LATER(optim): pehaps optimize the zone cut search + } + + int32_t new_ttl = get_new_ttl(eh, qry, qry->sname, qry->stype, + qry->timestamp.tv_sec); + if (new_ttl < 0 || eh->rank < lowest_rank) { + /* Positive record with stale TTL or bad rank. + * LATER(optim.): It's unlikely that we find a negative one, + * so we might theoretically skip all the cache code. */ + + VERBOSE_MSG(qry, "=> skipping exact %s: rank 0%.2o (min. 0%.2o), new TTL %d\n", + eh->is_packet ? "packet" : "RR", eh->rank, lowest_rank, new_ttl); + return kr_error(ENOENT); + } + + const uint8_t *eh_bound = knot_db_val_bound(val); + if (eh->is_packet) { + /* Note: we answer here immediately, even if it's (theoretically) + * possible that we could generate a higher-security negative proof. + * Rank is high-enough so we take it to save time searching. */ + return answer_from_pkt (ctx, pkt, qry->stype, eh, eh_bound, new_ttl); + } else { + return answer_simple_hit(ctx, pkt, qry->stype, eh, eh_bound, new_ttl); + } +} + + +/** Try to satisfy via wildcard (positively). See the single call site. */ +static int try_wild(struct key *k, struct answer *ans, const knot_dname_t *clencl_name, + const uint16_t type, const uint8_t lowest_rank, + const struct kr_query *qry, struct kr_cache *cache) +{ + knot_db_val_t key = key_exact_type(k, type); + /* Find the record. */ + knot_db_val_t val = { NULL, 0 }; + int ret = cache_op(cache, read, &key, &val, 1); + if (!ret) { + ret = entry_h_seek(&val, type); + } + if (ret) { + if (ret != -ABS(ENOENT)) { + VERBOSE_MSG(qry, "=> wildcard: hit error %d %s\n", + ret, strerror(abs(ret))); + assert(false); + } + WITH_VERBOSE(qry) { + auto_free char *clencl_str = kr_dname_text(clencl_name), + *type_str = kr_rrtype_text(type); + VERBOSE_MSG(qry, "=> wildcard: not found: *.%s %s\n", + clencl_str, type_str); + } + return ret; + } + /* Check if the record is OK. */ + const struct entry_h *eh = entry_h_consistent(val, type); + if (!eh) { + assert(false); + return kr_error(ret); + // LATER: recovery in case of error, perhaps via removing the entry? + } + int32_t new_ttl = get_new_ttl(eh, qry, qry->sname, type, qry->timestamp.tv_sec); + /* ^^ here we use the *expanded* wildcard name */ + if (new_ttl < 0 || eh->rank < lowest_rank || eh->is_packet) { + /* Wildcard record with stale TTL, bad rank or packet. */ + VERBOSE_MSG(qry, "=> wildcard: skipping %s, rank 0%.2o, new TTL %d\n", + eh->is_packet ? "packet" : "RR", eh->rank, new_ttl); + return -ABS(ESTALE); + } + /* Add the RR into the answer. */ + ret = entry2answer(ans, AR_ANSWER, eh, knot_db_val_bound(val), + qry->sname, type, new_ttl); + VERBOSE_MSG(qry, "=> wildcard: answer expanded, ret = %d, new TTL %d\n", + ret, (int)new_ttl); + if (ret) return kr_error(ret); + ans->rcode = PKT_NOERROR; + return kr_ok(); +} + +int kr_cache_closest_apex(struct kr_cache *cache, const knot_dname_t *name, bool is_DS, + knot_dname_t ** apex) +{ + if (!cache || !cache->db || !name || !apex || *apex) { + assert(!EINVAL); + return kr_error(EINVAL); + } + struct key k_storage, *k = &k_storage; + int ret = kr_dname_lf(k->buf, name, false); + if (ret) + return kr_error(ret); + entry_list_t el_; + k->zname = name; + ret = closest_NS(cache, k, el_, NULL, true, is_DS); + if (ret && ret != -abs(ENOENT)) + return ret; + *apex = knot_dname_copy(k->zname, NULL); + if (!*apex) + return kr_error(ENOMEM); + return kr_ok(); +} + +/** \internal for closest_NS. Check suitability of a single entry, setting k->type if OK. + * \return error code, negative iff whole list should be skipped. + */ +static int check_NS_entry(struct key *k, knot_db_val_t entry, int i, + bool exact_match, bool is_DS, + const struct kr_query *qry, uint32_t timestamp); + +/** + * Find the longest prefix zone/xNAME (with OK time+rank), starting at k->*. + * + * The found type is returned via k->type; the values are returned in el. + * \note we use k->type = KNOT_RRTYPE_NS also for the nsec_p result. + * \param qry can be NULL (-> gettimeofday(), but you lose the stale-serve hook) + * \param only_NS don't consider xNAMEs + * \return error code + */ +static int closest_NS(struct kr_cache *cache, struct key *k, entry_list_t el, + struct kr_query *qry, const bool only_NS, const bool is_DS) +{ + /* get the current timestamp */ + uint32_t timestamp; + if (qry) { + timestamp = qry->timestamp.tv_sec; + } else { + struct timeval tv; + if (gettimeofday(&tv, NULL)) return kr_error(errno); + timestamp = tv.tv_sec; + } + + int zlf_len = k->buf[0]; + + // LATER(optim): if stype is NS, we check the same value again + bool exact_match = true; + bool need_zero = true; + /* Inspect the NS/xNAME entries, shortening by a label on each iteration. */ + do { + k->buf[0] = zlf_len; + knot_db_val_t key = key_exact_type(k, KNOT_RRTYPE_NS); + knot_db_val_t val; + int ret = cache_op(cache, read, &key, &val, 1); + if (ret == -abs(ENOENT)) goto next_label; + if (ret) { + assert(!ret); + if (need_zero) memset(el, 0, sizeof(entry_list_t)); + return kr_error(ret); + } + + /* Check consistency, find any type; + * using `goto` for shortening by another label. */ + ret = entry_list_parse(val, el); + if (ret) { + assert(!ret); // do something about it? + goto next_label; + } + need_zero = false; + /* More types are possible; try in order. + * For non-fatal failures just "continue;" to try the next type. */ + const int el_count = only_NS ? EL_NS + 1 : EL_LENGTH; + for (int i = 0; i < el_count; ++i) { + ret = check_NS_entry(k, el[i], i, exact_match, is_DS, + qry, timestamp); + if (ret < 0) goto next_label; else + if (!ret) { + /* We found our match. */ + k->zlf_len = zlf_len; + return kr_ok(); + } + } + + next_label: + /* remove one more label */ + exact_match = false; + if (k->zname[0] == 0) { + /* We miss root NS in cache, but let's at least assume it exists. */ + k->type = KNOT_RRTYPE_NS; + k->zlf_len = zlf_len; + assert(zlf_len == 0); + if (need_zero) memset(el, 0, sizeof(entry_list_t)); + return kr_error(ENOENT); + } + zlf_len -= (k->zname[0] + 1); + k->zname += (k->zname[0] + 1); + k->buf[zlf_len + 1] = 0; + } while (true); +} + +static int check_NS_entry(struct key *k, const knot_db_val_t entry, const int i, + const bool exact_match, const bool is_DS, + const struct kr_query *qry, uint32_t timestamp) +{ + const int ESKIP = ABS(ENOENT); + if (!entry.len + /* On a zone cut we want DS from the parent zone. */ + || (i <= EL_NS && exact_match && is_DS) + /* CNAME is interesting only if we + * directly hit the name that was asked. + * Note that we want it even in the DS case. */ + || (i == EL_CNAME && !exact_match) + /* DNAME is interesting only if we did NOT + * directly hit the name that was asked. */ + || (i == EL_DNAME && exact_match) + ) { + return ESKIP; + } + + uint16_t type; + if (i < ENTRY_APEX_NSECS_CNT) { + type = KNOT_RRTYPE_NS; + int32_t log_new_ttl = -123456789; /* visually recognizable value */ + const int err = nsec_p_ttl(entry, timestamp, &log_new_ttl); + if (err) { + VERBOSE_MSG(qry, + "=> skipping unfit nsec_p: new TTL %d, error %d\n", + (int)log_new_ttl, err); + return ESKIP; + } + } else { + type = EL2RRTYPE(i); + /* Find the entry for the type, check positivity, TTL */ + const struct entry_h *eh = entry_h_consistent(entry, type); + if (!eh) { + VERBOSE_MSG(qry, "=> EH not consistent\n"); + assert(false); + return kr_error(EILSEQ); + } + const int32_t log_new_ttl = get_new_ttl(eh, qry, k->zname, type, timestamp); + const uint8_t rank_min = KR_RANK_INSECURE | KR_RANK_AUTH; + const bool ok = /* For NS any kr_rank is accepted, + * as insecure or even nonauth is OK */ + (type == KNOT_RRTYPE_NS || eh->rank >= rank_min) + /* Not interested in negative bogus or outdated RRs. */ + && !eh->is_packet && log_new_ttl >= 0; + WITH_VERBOSE(qry) { if (!ok) { + auto_free char *type_str = kr_rrtype_text(type); + const char *packet_str = eh->is_packet ? "packet" : "RR"; + VERBOSE_MSG(qry, + "=> skipping unfit %s %s: rank 0%.2o, new TTL %d\n", + type_str, packet_str, eh->rank, (int)log_new_ttl); + } } + if (!ok) return ESKIP; + } + k->type = type; + return kr_ok(); +} + diff --git a/lib/cookies/alg_containers.c b/lib/cookies/alg_containers.c new file mode 100644 index 0000000..a80487d --- /dev/null +++ b/lib/cookies/alg_containers.c @@ -0,0 +1,72 @@ +/* Copyright (C) 2016-2017 CZ.NIC, z.s.p.o. + + This program is free software: you can redistribute it and/or modify + it under the terms of the GNU General Public License as published by + the Free Software Foundation, either version 3 of the License, or + (at your option) any later version. + + This program is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the + GNU General Public License for more details. + + You should have received a copy of the GNU General Public License + along with this program. If not, see . + */ + +#include +#include +#include + +#include + +#include "lib/cookies/alg_containers.h" +#include "lib/cookies/alg_sha.h" + +const struct knot_cc_alg *kr_cc_alg_get(int id) +{ + /* + * Client algorithm identifiers are used to index this array of + * pointers. + */ + static const struct knot_cc_alg *const cc_algs[] = { + /* 0 */ &knot_cc_alg_fnv64, + /* 1 */ &knot_cc_alg_hmac_sha256_64 + }; + + if (id >= 0 && id < 2) { + return cc_algs[id]; + } + + return NULL; +} + +const knot_lookup_t kr_cc_alg_names[] = { + { 0, "FNV-64" }, + { 1, "HMAC-SHA256-64" }, + { -1, NULL } +}; + +const struct knot_sc_alg *kr_sc_alg_get(int id) +{ + /* + * Server algorithm identifiers are used to index this array of + * pointers. + */ + static const struct knot_sc_alg *const sc_algs[] = { + /* 0 */ &knot_sc_alg_fnv64, + /* 1 */ &knot_sc_alg_hmac_sha256_64 + }; + + if (id >= 0 && id < 2) { + return sc_algs[id]; + } + + return NULL; +} + +const knot_lookup_t kr_sc_alg_names[] = { + { 0, "FNV-64" }, + { 1, "HMAC-SHA256-64" }, + { -1, NULL } +}; diff --git a/lib/cookies/alg_containers.h b/lib/cookies/alg_containers.h new file mode 100644 index 0000000..1e2aef6 --- /dev/null +++ b/lib/cookies/alg_containers.h @@ -0,0 +1,49 @@ +/* Copyright (C) 2016-2017 CZ.NIC, z.s.p.o. + + This program is free software: you can redistribute it and/or modify + it under the terms of the GNU General Public License as published by + the Free Software Foundation, either version 3 of the License, or + (at your option) any later version. + + This program is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the + GNU General Public License for more details. + + You should have received a copy of the GNU General Public License + along with this program. If not, see . + */ + +#pragma once + +#include +#include +#include + +#include "lib/defines.h" + +/** + * @brief Returns pointer to client cookie algorithm. + * + * @param id algorithm identifier as defined by lookup table + * @return pointer to algorithm structure with given id or NULL on error + */ +KR_EXPORT +const struct knot_cc_alg *kr_cc_alg_get(int id); + +/** Binds client algorithm identifiers onto names. */ +KR_EXPORT +extern const knot_lookup_t kr_cc_alg_names[]; + +/** + * @brief Returns pointer to server cookie algorithm. + * + * @param id algorithm identifier as defined by lookup table + * @return pointer to algorithm structure with given id or NULL on error + */ +KR_EXPORT +const struct knot_sc_alg *kr_sc_alg_get(int id); + +/** Binds server algorithm identifiers onto names. */ +KR_EXPORT +extern const knot_lookup_t kr_sc_alg_names[]; diff --git a/lib/cookies/alg_sha.c b/lib/cookies/alg_sha.c new file mode 100644 index 0000000..e23c983 --- /dev/null +++ b/lib/cookies/alg_sha.c @@ -0,0 +1,122 @@ +/* Copyright (C) 2016-2017 CZ.NIC, z.s.p.o. + + This program is free software: you can redistribute it and/or modify + it under the terms of the GNU General Public License as published by + the Free Software Foundation, either version 3 of the License, or + (at your option) any later version. + + This program is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the + GNU General Public License for more details. + + You should have received a copy of the GNU General Public License + along with this program. If not, see . + */ + +#include +#include +#include +#include + +#include +#include + +#include "lib/cookies/alg_sha.h" +#include "lib/utils.h" + +/** + * @brief Update hash value. + * + * @param ctx HMAC-SHA256 context to be updated. + * @param sa Socket address. + */ +static inline void update_hash(struct hmac_sha256_ctx *ctx, + const struct sockaddr *sa) +{ + assert(ctx && sa); + + int addr_len = kr_inaddr_len(sa); + const uint8_t *addr = (uint8_t *)kr_inaddr(sa); + + if (addr && addr_len > 0) { + hmac_sha256_update(ctx, addr_len, addr); + } +} + +/** + * @brief Compute client cookie using HMAC-SHA256-64. + * @note At least one of the arguments must be non-null. + * @param input input parameters + * @param cc_out buffer for computed client cookie + * @param cc_len buffer size + * @return Non-zero size of written data on success, 0 in case of a failure. + */ +static uint16_t cc_gen_hmac_sha256_64(const struct knot_cc_input *input, + uint8_t *cc_out, uint16_t cc_len) +{ + if (!knot_cc_input_is_valid(input) || + !cc_out || cc_len < KNOT_OPT_COOKIE_CLNT) { + return 0; + } + + struct hmac_sha256_ctx ctx; + hmac_sha256_set_key(&ctx, input->secret_len, input->secret_data); + + if (input->clnt_sockaddr) { + update_hash(&ctx, input->clnt_sockaddr); + } + + if (input->srvr_sockaddr) { + update_hash(&ctx, input->srvr_sockaddr); + } + + /* KNOT_OPT_COOKIE_CLNT <= SHA256_DIGEST_SIZE */ + + hmac_sha256_digest(&ctx, KNOT_OPT_COOKIE_CLNT, cc_out); + + return KNOT_OPT_COOKIE_CLNT; +} + +#define SRVR_HMAC_SHA256_64_HASH_SIZE 8 + +/** + * @brief Compute server cookie hash using HMAC-SHA256-64). + * @note Server cookie = nonce | time | HMAC-SHA256-64( server secret, client cookie | nonce| time | client IP ) + * @param input data to compute cookie from + * @param hash_out hash output buffer + * @param hash_len buffer size + * @return Non-zero size of written data on success, 0 in case of a failure. + */ +static uint16_t sc_gen_hmac_sha256_64(const struct knot_sc_input *input, + uint8_t *hash_out, uint16_t hash_len) +{ + if (!knot_sc_input_is_valid(input) || + !hash_out || hash_len < SRVR_HMAC_SHA256_64_HASH_SIZE) { + return 0; + } + + struct hmac_sha256_ctx ctx; + hmac_sha256_set_key(&ctx, input->srvr_data->secret_len, + input->srvr_data->secret_data); + + hmac_sha256_update(&ctx, input->cc_len, input->cc); + + if (input->nonce && input->nonce_len) { + hmac_sha256_update(&ctx, input->nonce_len, input->nonce); + } + + if (input->srvr_data->clnt_sockaddr) { + update_hash(&ctx, input->srvr_data->clnt_sockaddr); + } + + /* SRVR_HMAC_SHA256_64_HASH_SIZE < SHA256_DIGEST_SIZE */ + + hmac_sha256_digest(&ctx, SRVR_HMAC_SHA256_64_HASH_SIZE, hash_out); + + return SRVR_HMAC_SHA256_64_HASH_SIZE; +} + +const struct knot_cc_alg knot_cc_alg_hmac_sha256_64 = { KNOT_OPT_COOKIE_CLNT, cc_gen_hmac_sha256_64 }; + +const struct knot_sc_alg knot_sc_alg_hmac_sha256_64 = { SRVR_HMAC_SHA256_64_HASH_SIZE, sc_gen_hmac_sha256_64 }; diff --git a/lib/cookies/alg_sha.h b/lib/cookies/alg_sha.h new file mode 100644 index 0000000..5b0d0c9 --- /dev/null +++ b/lib/cookies/alg_sha.h @@ -0,0 +1,30 @@ +/* Copyright (C) 2016-2017 CZ.NIC, z.s.p.o. + + This program is free software: you can redistribute it and/or modify + it under the terms of the GNU General Public License as published by + the Free Software Foundation, either version 3 of the License, or + (at your option) any later version. + + This program is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the + GNU General Public License for more details. + + You should have received a copy of the GNU General Public License + along with this program. If not, see . + */ + +#pragma once + +#include +#include + +#include "lib/defines.h" + +/* These structures are not meant to be part of public interface. */ + +/** HMAC-SHA256-64 client cookie algorithm. */ +extern const struct knot_cc_alg knot_cc_alg_hmac_sha256_64; + +/** HMAC-SHA256-64 server cookie algorithm. */ +extern const struct knot_sc_alg knot_sc_alg_hmac_sha256_64; diff --git a/lib/cookies/control.h b/lib/cookies/control.h new file mode 100644 index 0000000..1bf503f --- /dev/null +++ b/lib/cookies/control.h @@ -0,0 +1,49 @@ +/* Copyright (C) 2016-2017 CZ.NIC, z.s.p.o. + + This program is free software: you can redistribute it and/or modify + it under the terms of the GNU General Public License as published by + the Free Software Foundation, either version 3 of the License, or + (at your option) any later version. + + This program is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the + GNU General Public License for more details. + + You should have received a copy of the GNU General Public License + along with this program. If not, see . + */ + +#pragma once + +#include +#include +#include + +#include "lib/defines.h" + +/** Holds secret quantity. */ +struct kr_cookie_secret { + size_t size; /*!< Secret quantity size. */ + uint8_t data[]; /*!< Secret quantity data. */ +}; + +/** Holds settings that have direct influence on cookie values computation. */ +struct kr_cookie_comp { + struct kr_cookie_secret *secr; /*!< Secret data. */ + int alg_id; /*!< Cookie algorithm identifier. */ +}; + +/** Holds settings that control client/server cookie behaviour. */ +struct kr_cookie_settings { + bool enabled; /**< Enable/disables DNS cookies functionality. */ + + struct kr_cookie_comp current; /**< Current cookie settings. */ + struct kr_cookie_comp recent; /**< Recent cookie settings. */ +}; + +/** DNS cookies controlling structure. */ +struct kr_cookie_ctx { + struct kr_cookie_settings clnt; /**< Client settings. */ + struct kr_cookie_settings srvr; /**< Server settings. */ +}; diff --git a/lib/cookies/helper.c b/lib/cookies/helper.c new file mode 100644 index 0000000..e07b8cb --- /dev/null +++ b/lib/cookies/helper.c @@ -0,0 +1,296 @@ +/* Copyright (C) 2016-2017 CZ.NIC, z.s.p.o. + + This program is free software: you can redistribute it and/or modify + it under the terms of the GNU General Public License as published by + the Free Software Foundation, either version 3 of the License, or + (at your option) any later version. + + This program is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the + GNU General Public License for more details. + + You should have received a copy of the GNU General Public License + along with this program. If not, see . + */ + +#include +#include +#include + +#include "lib/cookies/helper.h" +#include "lib/defines.h" + +/** + * @brief Check whether there is a cached cookie that matches the current + * client cookie. + */ +static const uint8_t *peek_and_check_cc(kr_cookie_lru_t *cache, const void *sa, + const uint8_t *cc, uint16_t cc_len) +{ + assert(cache && sa && cc && cc_len); + + const uint8_t *cached_opt = kr_cookie_lru_get(cache, sa); + if (!cached_opt) { + return NULL; + } + + const uint8_t *cached_cc = knot_edns_opt_get_data((uint8_t *) cached_opt); + + if (cc_len == KNOT_OPT_COOKIE_CLNT && + 0 == memcmp(cc, cached_cc, cc_len)) { + return cached_opt; + } + + return NULL; +} + +/** + * @brief Put a client cookie into the RR Set. + */ +static int opt_rr_put_cookie(knot_rrset_t *opt_rr, uint8_t *data, + uint16_t data_len, knot_mm_t *mm) +{ + assert(opt_rr && data && data_len > 0); + + const uint8_t *cc = NULL, *sc = NULL; + uint16_t cc_len = 0, sc_len = 0; + + int ret = knot_edns_opt_cookie_parse(data, data_len, &cc, &cc_len, + &sc, &sc_len); + if (ret != KNOT_EOK) { + return kr_error(EINVAL); + } + assert(data_len == cc_len + sc_len); + + uint16_t cookies_size = data_len; + uint8_t *cookies_data = NULL; + + ret = knot_edns_reserve_unique_option(opt_rr, KNOT_EDNS_OPTION_COOKIE, + cookies_size, &cookies_data, mm); + if (ret != KNOT_EOK) { + return kr_error(EINVAL); + } + assert(cookies_data != NULL); + + cookies_size = knot_edns_opt_cookie_write(cc, cc_len, sc, sc_len, + cookies_data, cookies_size); + if (cookies_size == 0) { + return kr_error(EINVAL); + } + assert(cookies_size == data_len); + + return kr_ok(); +} + +/** + * @brief Puts entire EDNS option into the RR Set. + */ +static int opt_rr_put_cookie_opt(knot_rrset_t *opt_rr, uint8_t *option, knot_mm_t *mm) +{ + assert(opt_rr && option); + + uint16_t opt_code = knot_edns_opt_get_code(option); + if (opt_code != KNOT_EDNS_OPTION_COOKIE) { + return kr_error(EINVAL); + } + + uint16_t opt_len = knot_edns_opt_get_length(option); + uint8_t *opt_data = knot_edns_opt_get_data(option); + if (!opt_data || opt_len == 0) { + return kr_error(EINVAL); + } + + return opt_rr_put_cookie(opt_rr, opt_data, opt_len, mm); +} + +int kr_request_put_cookie(const struct kr_cookie_comp *clnt_comp, + kr_cookie_lru_t *cookie_cache, + const struct sockaddr *clnt_sa, + const struct sockaddr *srvr_sa, + struct kr_request *req) +{ + if (!clnt_comp || !req) { + return kr_error(EINVAL); + } + + if (!req->ctx->opt_rr) { + return kr_ok(); + } + + if (!clnt_comp->secr || (clnt_comp->alg_id < 0) || !cookie_cache) { + return kr_error(EINVAL); + } + + /* + * Generate client cookie from client address, server address and + * secret quantity. + */ + struct knot_cc_input input = { + .clnt_sockaddr = clnt_sa, + .srvr_sockaddr = srvr_sa, + .secret_data = clnt_comp->secr->data, + .secret_len = clnt_comp->secr->size + }; + uint8_t cc[KNOT_OPT_COOKIE_CLNT]; + uint16_t cc_len = KNOT_OPT_COOKIE_CLNT; + const struct knot_cc_alg *cc_alg = kr_cc_alg_get(clnt_comp->alg_id); + if (!cc_alg) { + return kr_error(EINVAL); + } + assert(cc_alg->gen_func); + cc_len = cc_alg->gen_func(&input, cc, cc_len); + if (cc_len != KNOT_OPT_COOKIE_CLNT) { + return kr_error(EINVAL); + } + + const uint8_t *cached_cookie = peek_and_check_cc(cookie_cache, + srvr_sa, cc, cc_len); + + /* Add cookie option. */ + int ret; + if (cached_cookie) { + ret = opt_rr_put_cookie_opt(req->ctx->opt_rr, + (uint8_t *)cached_cookie, + req->ctx->pool); + } else { + ret = opt_rr_put_cookie(req->ctx->opt_rr, cc, cc_len, + req->ctx->pool); + } + + return ret; +} + +int kr_answer_write_cookie(struct knot_sc_input *sc_input, + const struct kr_nonce_input *nonce, + const struct knot_sc_alg *alg, knot_pkt_t *pkt) +{ + if (!sc_input || !sc_input->cc || sc_input->cc_len == 0) { + return kr_error(EINVAL); + } + + if (!sc_input->srvr_data || !sc_input->srvr_data->clnt_sockaddr || + !sc_input->srvr_data->secret_data || + !sc_input->srvr_data->secret_len) { + return kr_error(EINVAL); + } + + if (!nonce) { + return kr_error(EINVAL); + } + + if (!alg || !alg->hash_size || !alg->hash_func) { + return kr_error(EINVAL); + } + + if (!pkt || !pkt->opt_rr) { + return kr_error(EINVAL); + } + + uint16_t nonce_len = KR_NONCE_LEN; + uint16_t hash_len = alg->hash_size; + + /* + * Space for cookie is reserved inside the EDNS OPT RR of + * the answer packet. + */ + uint8_t *cookie = NULL; + uint16_t cookie_len = knot_edns_opt_cookie_data_len(sc_input->cc_len, + nonce_len + hash_len); + if (cookie_len == 0) { + return kr_error(EINVAL); + } + + int ret = knot_edns_reserve_unique_option(pkt->opt_rr, + KNOT_EDNS_OPTION_COOKIE, + cookie_len, &cookie, + &pkt->mm); + if (ret != KNOT_EOK) { + return kr_error(ENOMEM); + } + assert(cookie != NULL); + + /* + * Function knot_edns_opt_cookie_data_len() returns the sum of its + * parameters or zero. Anyway, let's check again. + */ + if (cookie_len < (sc_input->cc_len + nonce_len + hash_len)) { + return kr_error(EINVAL); + } + + /* Copy client cookie data portion. */ + memcpy(cookie, sc_input->cc, sc_input->cc_len); + + if (nonce_len) { + /* Write nonce data portion. */ + kr_nonce_write_wire(cookie + sc_input->cc_len, nonce_len, + nonce); + /* Adjust input for written nonce value. */ + sc_input->nonce = cookie + sc_input->cc_len; + sc_input->nonce_len = nonce_len; + } + + hash_len = alg->hash_func(sc_input, + cookie + sc_input->cc_len + nonce_len, + hash_len); + /* Zero nonce values. */ + sc_input->nonce = NULL; + sc_input->nonce_len = 0; + + return (hash_len != 0) ? kr_ok() : kr_error(EINVAL); +} + +int kr_pkt_set_ext_rcode(knot_pkt_t *pkt, uint16_t whole_rcode) +{ + /* + * RFC6891 6.1.3 -- extended RCODE forms the upper 8 bits of whole + * 12-bit RCODE (together with the 4 bits of 'normal' RCODE). + * + * | 11 10 09 08 07 06 05 04 | 03 02 01 00 | + * | 12-bit whole RCODE | + * | 8-bit extended RCODE | 4-bit RCODE | + */ + + if (!pkt || !knot_pkt_has_edns(pkt)) { + return kr_error(EINVAL); + } + + uint8_t rcode = whole_rcode & 0x0f; + uint8_t ext_rcode = whole_rcode >> 4; + knot_wire_set_rcode(pkt->wire, rcode); + knot_edns_set_ext_rcode(pkt->opt_rr, ext_rcode); + + return kr_ok(); +} + +uint8_t *kr_no_question_cookie_query(const knot_pkt_t *pkt) +{ + if (!pkt || knot_wire_get_qdcount(pkt->wire) > 0) { + return false; + } + + if (knot_wire_get_qr(pkt->wire) != 0 || !pkt->opt_rr) { + return false; + } + + return knot_edns_get_option(pkt->opt_rr, KNOT_EDNS_OPTION_COOKIE); +} + +int kr_parse_cookie_opt(uint8_t *cookie_opt, struct knot_dns_cookies *cookies) +{ + if (!cookie_opt || !cookies) { + return kr_error(EINVAL); + } + + const uint8_t *cookie_data = knot_edns_opt_get_data(cookie_opt); + uint16_t cookie_len = knot_edns_opt_get_length(cookie_opt); + if (!cookie_data || cookie_len == 0) { + return kr_error(EINVAL); + } + + int ret = knot_edns_opt_cookie_parse(cookie_data, cookie_len, + &cookies->cc, &cookies->cc_len, + &cookies->sc, &cookies->sc_len); + + return (ret == KNOT_EOK) ? kr_ok() : kr_error(EINVAL); +} diff --git a/lib/cookies/helper.h b/lib/cookies/helper.h new file mode 100644 index 0000000..e5659ce --- /dev/null +++ b/lib/cookies/helper.h @@ -0,0 +1,86 @@ +/* Copyright (C) 2016-2017 CZ.NIC, z.s.p.o. + + This program is free software: you can redistribute it and/or modify + it under the terms of the GNU General Public License as published by + the Free Software Foundation, either version 3 of the License, or + (at your option) any later version. + + This program is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the + GNU General Public License for more details. + + You should have received a copy of the GNU General Public License + along with this program. If not, see . + */ + +#pragma once + +#include + +#include "lib/cookies/alg_containers.h" +#include "lib/cookies/control.h" +#include "lib/cookies/lru_cache.h" +#include "lib/cookies/nonce.h" +#include "lib/defines.h" +#include "lib/resolve.h" + +/** + * @brief Updates DNS cookie in the request EDNS options. + * @note This function must be called before the request packet is finalised. + * @param clnt_comp client cookie control structure + * @param cookie_cache cookie cache + * @param clnt_sa client socket address + * @param srvr_sa server socket address + * @param req name resolution request + * @return kr_ok() or error code + */ +KR_EXPORT +int kr_request_put_cookie(const struct kr_cookie_comp *clnt_comp, + kr_cookie_lru_t *cookie_cache, + const struct sockaddr *clnt_sa, + const struct sockaddr *srvr_sa, + struct kr_request *req); + +/** + * @brief Inserts a cookie option into the OPT RR. It does not write any + * wire data. + * @note The content of @a sc_input is modified. Any pre-set nonce value is + * ignored. After retuning its nonce value will be null. + * @param sc_input data needed to compute server cookie, nonce is ignored + * @param nonce nonce value that is actually used + * @param alg hash algorithm + * @param pkt DNS response packet + */ +KR_EXPORT +int kr_answer_write_cookie(struct knot_sc_input *sc_input, + const struct kr_nonce_input *nonce, + const struct knot_sc_alg *alg, knot_pkt_t *pkt); + +/** + * @brief Set RCODE and extended RCODE. + * @param pkt DNS packet + * @param whole_rcode RCODE value + * @return kr_ok() or error code + */ +KR_EXPORT +int kr_pkt_set_ext_rcode(knot_pkt_t *pkt, uint16_t whole_rcode); + +/** + * @brief Check whether packet is a server cookie request according to + * RFC7873 5.4. + * @param pkt Packet to be examined. + * @return Pointer to entire cookie option if is a server cookie query, NULL on + * errors or if packet doesn't contain cookies or if QDCOUNT > 0. + */ +KR_EXPORT +uint8_t *kr_no_question_cookie_query(const knot_pkt_t *pkt); + +/** + * @brief Parse cookies from cookie option. + * @param cookie_opt Cookie option. + * @param cookies Cookie structure to be set. + * @return kr_ok() on success, error if cookies are malformed. + */ +KR_EXPORT +int kr_parse_cookie_opt(uint8_t *cookie_opt, struct knot_dns_cookies *cookies); diff --git a/lib/cookies/lru_cache.c b/lib/cookies/lru_cache.c new file mode 100644 index 0000000..8ea97ab --- /dev/null +++ b/lib/cookies/lru_cache.c @@ -0,0 +1,70 @@ +/* Copyright (C) 2016-2017 CZ.NIC, z.s.p.o. + + This program is free software: you can redistribute it and/or modify + it under the terms of the GNU General Public License as published by + the Free Software Foundation, either version 3 of the License, or + (at your option) any later version. + + This program is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the + GNU General Public License for more details. + + You should have received a copy of the GNU General Public License + along with this program. If not, see . + */ + +#include +#include + +#include "lib/cookies/lru_cache.h" +#include "lib/utils.h" + +const uint8_t *kr_cookie_lru_get(kr_cookie_lru_t *cache, + const struct sockaddr *sa) +{ + if (!cache || !sa) { + return NULL; + } + + int addr_len = kr_inaddr_len(sa); + const char *addr = kr_inaddr(sa); + if (!addr || addr_len <= 0) { + return NULL; + } + + struct cookie_opt_data *cached = lru_get_try(cache, addr, addr_len); + return cached ? cached->opt_data : NULL; +} + +int kr_cookie_lru_set(kr_cookie_lru_t *cache, const struct sockaddr *sa, + uint8_t *opt) +{ + if (!cache || !sa) { + return kr_error(EINVAL); + } + + if (!opt) { + return kr_ok(); + } + + int addr_len = kr_inaddr_len(sa); + const char *addr = kr_inaddr(sa); + if (!addr || addr_len <= 0) { + return kr_error(EINVAL); + } + + uint16_t opt_size = KNOT_EDNS_OPTION_HDRLEN + + knot_edns_opt_get_length(opt); + + if (opt_size > KR_COOKIE_OPT_MAX_LEN) { + return kr_error(EINVAL); + } + + struct cookie_opt_data *cached = lru_get_new(cache, addr, addr_len, NULL); + if (cached) { + memcpy(cached->opt_data, opt, opt_size); + } + + return kr_ok(); +} diff --git a/lib/cookies/lru_cache.h b/lib/cookies/lru_cache.h new file mode 100644 index 0000000..8ce0a01 --- /dev/null +++ b/lib/cookies/lru_cache.h @@ -0,0 +1,69 @@ +/* Copyright (C) 2016-2017 CZ.NIC, z.s.p.o. + + This program is free software: you can redistribute it and/or modify + it under the terms of the GNU General Public License as published by + the Free Software Foundation, either version 3 of the License, or + (at your option) any later version. + + This program is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the + GNU General Public License for more details. + + You should have received a copy of the GNU General Public License + along with this program. If not, see . + */ + +#pragma once + +#include +#include + +#if defined(ENABLE_COOKIES) +#include +#include +#else +#define KNOT_OPT_COOKIE_CLNT 8 +#define KNOT_OPT_COOKIE_SRVR_MAX 32 +#endif /* defined(ENABLE_COOKIES) */ + +#include "lib/defines.h" +#include "lib/generic/lru.h" + +/** Maximal size of a cookie option. */ +#define KR_COOKIE_OPT_MAX_LEN (KNOT_EDNS_OPTION_HDRLEN + KNOT_OPT_COOKIE_CLNT + KNOT_OPT_COOKIE_SRVR_MAX) + +/** + * Cookie option entry. + */ +struct cookie_opt_data { + uint8_t opt_data[KR_COOKIE_OPT_MAX_LEN]; +}; + +/** + * DNS cookies tracking. + */ +typedef lru_t(struct cookie_opt_data) kr_cookie_lru_t; + +/** + * @brief Obtain LRU cache entry. + * + * @param cache cookie LRU cache + * @param sa socket address serving as key + * @return pointer to cached option or NULL if not found or error occurred + */ +KR_EXPORT +const uint8_t *kr_cookie_lru_get(kr_cookie_lru_t *cache, + const struct sockaddr *sa); + +/** + * @brief Stores cookie option into LRU cache. + * + * @param cache cookie LRU cache + * @param sa socket address serving as key + * @param opt cookie option to be stored + * @return kr_ok() or error code + */ +KR_EXPORT +int kr_cookie_lru_set(kr_cookie_lru_t *cache, const struct sockaddr *sa, + uint8_t *opt); diff --git a/lib/cookies/nonce.c b/lib/cookies/nonce.c new file mode 100644 index 0000000..4ad72c8 --- /dev/null +++ b/lib/cookies/nonce.c @@ -0,0 +1,34 @@ +/* Copyright (C) 2016-2017 CZ.NIC, z.s.p.o. + + This program is free software: you can redistribute it and/or modify + it under the terms of the GNU General Public License as published by + the Free Software Foundation, either version 3 of the License, or + (at your option) any later version. + + This program is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the + GNU General Public License for more details. + + You should have received a copy of the GNU General Public License + along with this program. If not, see . + */ + +#include + +#include "contrib/wire.h" +#include "lib/cookies/nonce.h" + +uint16_t kr_nonce_write_wire(uint8_t *buf, uint16_t buf_len, + const struct kr_nonce_input *input) +{ + if (!buf || buf_len < KR_NONCE_LEN || !input) { + return 0; + } + + wire_write_u32(buf, input->rand); + wire_write_u32(buf + sizeof(uint32_t), input->time); + buf_len = 2 * sizeof(uint32_t); + + return buf_len; +} diff --git a/lib/cookies/nonce.h b/lib/cookies/nonce.h new file mode 100644 index 0000000..99c54d3 --- /dev/null +++ b/lib/cookies/nonce.h @@ -0,0 +1,43 @@ +/* Copyright (C) 2016-2017 CZ.NIC, z.s.p.o. + + This program is free software: you can redistribute it and/or modify + it under the terms of the GNU General Public License as published by + the Free Software Foundation, either version 3 of the License, or + (at your option) any later version. + + This program is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the + GNU General Public License for more details. + + You should have received a copy of the GNU General Public License + along with this program. If not, see . + */ + +#pragma once + +#include "lib/defines.h" + +/* RFC7873 Appendix B.2 mentions an algorithm using two values before the + * actual server cookie hash. */ + +/** Nonce value length. */ +#define KR_NONCE_LEN 8 + +/** Input data to generate nonce from. */ +struct kr_nonce_input { + uint32_t rand; /**< some random value */ + uint32_t time; /**< time stamp */ +}; + +/** + * @brief Writes server cookie nonce value into given buffer. + * + * @param buf buffer to write nonce data in wire format into + * @param buf_len buffer size + * @param input data to generate wire data from + * @return non-zero size of written data on success, 0 on failure + */ +KR_EXPORT +uint16_t kr_nonce_write_wire(uint8_t *buf, uint16_t buf_len, + const struct kr_nonce_input *input); diff --git a/lib/defines.h b/lib/defines.h new file mode 100644 index 0000000..84da059 --- /dev/null +++ b/lib/defines.h @@ -0,0 +1,102 @@ +/* Copyright (C) 2014-2017 CZ.NIC, z.s.p.o. + + This program is free software: you can redistribute it and/or modify + it under the terms of the GNU General Public License as published by + the Free Software Foundation, either version 3 of the License, or + (at your option) any later version. + + This program is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the + GNU General Public License for more details. + + You should have received a copy of the GNU General Public License + along with this program. If not, see . + */ + +#pragma once + +#include +#include +#include +#include + +/* Function attributes */ +#if __GNUC__ >= 4 +#define KR_EXPORT __attribute__ ((visibility ("default"))) +#define KR_CONST __attribute__((__const__)) +#define KR_PURE __attribute__((__pure__)) +#define KR_NORETURN __attribute__((__noreturn__)) +#define KR_COLD __attribute__((__cold__)) +#define KR_PRINTF(n) __attribute__((format (printf, n, (n+1)))) +#else +#define KR_EXPORT +#define KR_CONST +#define KR_PURE +#define KR_NORETURN +#define KR_COLD +#define KR_PRINTF(n) +#endif + +#ifndef uint /* Redefining typedef is a C11 feature. */ +typedef unsigned int uint; +#define uint uint +#endif + +/* + * Error codes. + */ +#define kr_ok() 0 +/* Mark as cold to mark all branches as unlikely. */ +static inline int KR_COLD kr_error(int x) { + return x <= 0 ? x : -x; +} +#define kr_strerror(x) strerror(abs(x)) + +/* + * Connection limits. + * @cond internal + */ +#define KR_CONN_RTT_MAX 2000 /* Timeout for network activity */ +#define KR_CONN_RETRY 200 /* Retry interval for network activity */ +#define KR_ITER_LIMIT 100 /* Built-in iterator limit */ +#define KR_RESOLVE_TIME_LIMIT 10000 /* Upper limit for resolution time of single query, ms */ +#define KR_CNAME_CHAIN_LIMIT 40 /* Built-in maximum CNAME chain length */ +#define KR_TIMEOUT_LIMIT 4 /* Maximum number of retries after timeout. */ +#define KR_QUERY_NSRETRY_LIMIT 4 /* Maximum number of retries per query. */ + +/* + * Defines. + */ +#define KR_DNS_PORT 53 +#define KR_DNS_TLS_PORT 853 +#define KR_EDNS_VERSION 0 +#define KR_EDNS_PAYLOAD 4096 /* Default UDP payload (max unfragmented UDP is 1452B) */ +#define KR_CACHE_DEFAULT_TTL_MIN (5) /* avoid bursts of queries */ +#define KR_CACHE_DEFAULT_TTL_MAX (6 * 24 * 3600) /* 6 days, like the root NS TTL */ + +#define KR_DNAME_STR_MAXLEN (KNOT_DNAME_TXT_MAXLEN + 1) +#define KR_RRTYPE_STR_MAXLEN (16 + 1) + +/* + * Address sanitizer hints. + */ +#if !defined(__SANITIZE_ADDRESS__) && defined(__has_feature) +# if __has_feature(address_sanitizer) +# define __SANITIZE_ADDRESS__ 1 +# endif +#endif +#if defined(__SANITIZE_ADDRESS__) +void __asan_poison_memory_region(void const volatile *addr, size_t size); +void __asan_unpoison_memory_region(void const volatile *addr, size_t size); +#define kr_asan_poison(addr, size) __asan_poison_memory_region((addr), (size)) +#define kr_asan_unpoison(addr, size) __asan_unpoison_memory_region((addr), (size)) +#define kr_asan_custom_poison(fn, addr) fn ##_poison((addr)) +#define kr_asan_custom_unpoison(fn, addr) fn ##_unpoison((addr)) +#else +#define kr_asan_poison(addr, size) +#define kr_asan_unpoison(addr, size) +#define kr_asan_custom_poison(fn, addr) +#define kr_asan_custom_unpoison(fn, addr) +#endif +/* @endcond */ diff --git a/lib/dnssec.c b/lib/dnssec.c new file mode 100644 index 0000000..4f8ad8a --- /dev/null +++ b/lib/dnssec.c @@ -0,0 +1,473 @@ +/* Copyright (C) 2015-2017 CZ.NIC, z.s.p.o. + + This program is free software: you can redistribute it and/or modify + it under the terms of the GNU General Public License as published by + the Free Software Foundation, either version 3 of the License, or + (at your option) any later version. + + This program is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the + GNU General Public License for more details. + + You should have received a copy of the GNU General Public License + along with this program. If not, see . + */ + +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include + +#include "lib/defines.h" +#include "lib/dnssec/nsec.h" +#include "lib/dnssec/nsec3.h" +#include "lib/dnssec/signature.h" +#include "lib/dnssec.h" +#include "lib/resolve.h" + +/* forward */ +static int kr_rrset_validate_with_key(kr_rrset_validation_ctx_t *vctx, + const knot_rrset_t *covered, size_t key_pos, const struct dseckey *key); + +void kr_crypto_init(void) +{ + dnssec_crypto_init(); +} + +void kr_crypto_cleanup(void) +{ + dnssec_crypto_cleanup(); +} + +void kr_crypto_reinit(void) +{ + dnssec_crypto_reinit(); +} + +#define FLG_WILDCARD_EXPANSION 0x01 /**< Possibly generated by using wildcard expansion. */ + +/** + * Check the RRSIG RR validity according to RFC4035 5.3.1 . + * @param flags The flags are going to be set according to validation result. + * @param cov_labels Covered RRSet owner label count. + * @param rrsigs rdata containing the signatures. + * @param key_owner Associated DNSKEY's owner. + * @param key_rdata Associated DNSKEY's rdata. + * @param keytag Used key tag. + * @param zone_name The name of the zone cut. + * @param timestamp Validation time. + */ +static int validate_rrsig_rr(int *flags, int cov_labels, + const knot_rdata_t *rrsigs, + const knot_dname_t *key_owner, const knot_rdata_t *key_rdata, + uint16_t keytag, + const knot_dname_t *zone_name, uint32_t timestamp, + kr_rrset_validation_ctx_t *vctx) +{ + if (!flags || !rrsigs || !key_owner || !key_rdata || !zone_name) { + return kr_error(EINVAL); + } + /* bullet 5 */ + if (knot_rrsig_sig_expiration(rrsigs) < timestamp) { + vctx->rrs_counters.expired++; + return kr_error(EINVAL); + } + /* bullet 6 */ + if (knot_rrsig_sig_inception(rrsigs) > timestamp) { + vctx->rrs_counters.notyet++; + return kr_error(EINVAL); + } + /* bullet 2 */ + const knot_dname_t *signer_name = knot_rrsig_signer_name(rrsigs); + if (!signer_name || !knot_dname_is_equal(signer_name, zone_name)) { + vctx->rrs_counters.signer_invalid++; + return kr_error(EAGAIN); + } + /* bullet 4 */ + { + int rrsig_labels = knot_rrsig_labels(rrsigs); + if (rrsig_labels > cov_labels) { + vctx->rrs_counters.labels_invalid++; + return kr_error(EINVAL); + } + if (rrsig_labels < cov_labels) { + *flags |= FLG_WILDCARD_EXPANSION; + } + } + + /* bullet 7 */ + if ((!knot_dname_is_equal(key_owner, signer_name)) || + (knot_dnskey_alg(key_rdata) != knot_rrsig_alg(rrsigs)) || + (keytag != knot_rrsig_key_tag(rrsigs))) { + vctx->rrs_counters.key_invalid++; + return kr_error(EINVAL); + } + /* bullet 8 */ + /* Checked somewhere else. */ + /* bullet 9 and 10 */ + /* One of the requirements should be always fulfilled. */ + + return kr_ok(); +} + +/** + * Returns the number of labels that have been added by wildcard expansion. + * @param expanded Expanded wildcard. + * @param rrsigs RRSet containing the signatures. + * @param sig_pos Specifies the signature within the RRSIG RRSet. + * @return Number of added labels, -1 on error. + */ +static inline int wildcard_radix_len_diff(const knot_dname_t *expanded, + const knot_rdata_t *rrsig) +{ + if (!expanded || !rrsig) { + return -1; + } + + return knot_dname_labels(expanded, NULL) - knot_rrsig_labels(rrsig); +} + +int kr_rrset_validate(kr_rrset_validation_ctx_t *vctx, const knot_rrset_t *covered) +{ + if (!vctx) { + return kr_error(EINVAL); + } + if (!vctx->pkt || !covered || !vctx->keys || !vctx->zone_name) { + return kr_error(EINVAL); + } + + memset(&vctx->rrs_counters, 0, sizeof(vctx->rrs_counters)); + for (unsigned i = 0; i < vctx->keys->rrs.count; ++i) { + int ret = kr_rrset_validate_with_key(vctx, covered, i, NULL); + if (ret == 0) { + return ret; + } + } + + return kr_error(ENOENT); +} + +/** + * Validate RRSet using a specific key. + * @param vctx Pointer to validation context. + * @param covered RRSet covered by a signature. It must be in canonical format. + * @param key_pos Position of the key to be validated with. + * @param key Key to be used to validate. + * If NULL, then key from DNSKEY RRSet is used. + * @return 0 or error code, same as vctx->result. + */ +static int kr_rrset_validate_with_key(kr_rrset_validation_ctx_t *vctx, + const knot_rrset_t *covered, + size_t key_pos, const struct dseckey *key) +{ + const knot_pkt_t *pkt = vctx->pkt; + const knot_rrset_t *keys = vctx->keys; + const knot_dname_t *zone_name = vctx->zone_name; + uint32_t timestamp = vctx->timestamp; + bool has_nsec3 = vctx->has_nsec3; + struct dseckey *created_key = NULL; + + /* It's just caller's approximation that the RR is in that particular zone. + * We MUST guard against attempts of zones signing out-of-bailiwick records. */ + if (knot_dname_in_bailiwick(covered->owner, zone_name) < 0) { + vctx->result = kr_error(ENOENT); + return vctx->result; + } + + const knot_rdata_t *key_rdata = knot_rdataset_at(&keys->rrs, key_pos); + if (key == NULL) { + int ret = kr_dnssec_key_from_rdata(&created_key, keys->owner, + key_rdata->data, key_rdata->len); + if (ret != 0) { + vctx->result = ret; + return vctx->result; + } + key = created_key; + } + uint16_t keytag = dnssec_key_get_keytag((dnssec_key_t *)key); + int covered_labels = knot_dname_labels(covered->owner, NULL); + if (knot_dname_is_wildcard(covered->owner)) { + /* The asterisk does not count, RFC4034 3.1.3, paragraph 3. */ + --covered_labels; + } + + for (uint16_t i = 0; i < vctx->rrs->len; ++i) { + /* Consider every RRSIG that matches owner and covers the class/type. */ + const knot_rrset_t *rrsig = vctx->rrs->at[i]->rr; + if (rrsig->type != KNOT_RRTYPE_RRSIG) { + continue; + } + if ((covered->rclass != rrsig->rclass) || !knot_dname_is_equal(covered->owner, rrsig->owner)) { + continue; + } + knot_rdata_t *rdata_j = rrsig->rrs.rdata; + for (uint16_t j = 0; j < rrsig->rrs.count; ++j, rdata_j = knot_rdataset_next(rdata_j)) { + int val_flgs = 0; + int trim_labels = 0; + if (knot_rrsig_type_covered(rdata_j) != covered->type) { + continue; + } + vctx->rrs_counters.matching_name_type++; + int retv = validate_rrsig_rr(&val_flgs, covered_labels, rdata_j, + keys->owner, key_rdata, keytag, + zone_name, timestamp, vctx); + if (retv == kr_error(EAGAIN)) { + kr_dnssec_key_free(&created_key); + vctx->result = retv; + return retv; + } else if (retv != 0) { + continue; + } + if (val_flgs & FLG_WILDCARD_EXPANSION) { + trim_labels = wildcard_radix_len_diff(covered->owner, rdata_j); + if (trim_labels < 0) { + break; + } + } + if (kr_check_signature(rdata_j, (dnssec_key_t *) key, covered, trim_labels) != 0) { + vctx->rrs_counters.crypto_invalid++; + continue; + } + if (val_flgs & FLG_WILDCARD_EXPANSION) { + int ret = 0; + if (!has_nsec3) { + ret = kr_nsec_wildcard_answer_response_check(pkt, KNOT_AUTHORITY, covered->owner); + } else { + ret = kr_nsec3_wildcard_answer_response_check(pkt, KNOT_AUTHORITY, covered->owner, trim_labels - 1); + if (ret == kr_error(KNOT_ERANGE)) { + ret = 0; + vctx->flags |= KR_DNSSEC_VFLG_OPTOUT; + } + } + if (ret != 0) { + vctx->rrs_counters.nsec_invalid++; + continue; + } + vctx->flags |= KR_DNSSEC_VFLG_WEXPAND; + } + /* Validated with current key, OK */ + kr_dnssec_key_free(&created_key); + vctx->result = kr_ok(); + return vctx->result; + } + } + /* No applicable key found, cannot be validated. */ + kr_dnssec_key_free(&created_key); + vctx->result = kr_error(ENOENT); + return vctx->result; +} + +static bool kr_ds_algo_support(const knot_rrset_t *ta) +{ + knot_rdata_t *rdata_i = ta->rrs.rdata; + for (uint16_t i = 0; i < ta->rrs.count; + ++i, rdata_i = knot_rdataset_next(rdata_i)) { + if (dnssec_algorithm_digest_support(knot_ds_digest_type(rdata_i)) + && dnssec_algorithm_key_support(knot_ds_alg(rdata_i))) { + return true; + } + } + return false; +} + +int kr_dnskeys_trusted(kr_rrset_validation_ctx_t *vctx, const knot_rrset_t *ta) +{ + const knot_pkt_t *pkt = vctx->pkt; + const knot_rrset_t *keys = vctx->keys; + + const bool ok = pkt && keys && ta && ta->rrs.count && ta->rrs.rdata + && ta->type == KNOT_RRTYPE_DS; + if (!ok) { + assert(false); + return kr_error(EINVAL); + } + + /* Check if at least one DS has a usable algorithm pair. */ + if (!kr_ds_algo_support(ta)) { + /* See RFC6840 5.2. */ + return vctx->result = kr_error(DNSSEC_INVALID_DS_ALGORITHM); + } + + /* RFC4035 5.2, bullet 1 + * The supplied DS record has been authenticated. + * It has been validated or is part of a configured trust anchor. + */ + memset(&vctx->rrs_counters, 0, sizeof(vctx->rrs_counters)); + for (uint16_t i = 0; i < keys->rrs.count; ++i) { + /* RFC4035 5.3.1, bullet 8 */ /* ZSK */ + /* LATER(optim.): more efficient way to iterate than _at() */ + const knot_rdata_t *krr = knot_rdataset_at(&keys->rrs, i); + if (!kr_dnssec_key_zsk(krr->data) || kr_dnssec_key_revoked(krr->data)) { + continue; + } + + struct dseckey *key = NULL; + if (kr_dnssec_key_from_rdata(&key, keys->owner, krr->data, krr->len) != 0) { + continue; + } + if (kr_authenticate_referral(ta, (dnssec_key_t *) key) != 0) { + kr_dnssec_key_free(&key); + continue; + } + if (kr_rrset_validate_with_key(vctx, keys, i, key) != 0) { + kr_dnssec_key_free(&key); + continue; + } + kr_dnssec_key_free(&key); + assert (vctx->result == 0); + return vctx->result; + } + + /* No useable key found */ + vctx->result = kr_error(ENOENT); + return vctx->result; +} + +bool kr_dnssec_key_zsk(const uint8_t *dnskey_rdata) +{ + return wire_read_u16(dnskey_rdata) & 0x0100; +} + +bool kr_dnssec_key_ksk(const uint8_t *dnskey_rdata) +{ + return wire_read_u16(dnskey_rdata) & 0x0001; +} + +/** Return true if the DNSKEY is revoked. */ +bool kr_dnssec_key_revoked(const uint8_t *dnskey_rdata) +{ + return wire_read_u16(dnskey_rdata) & 0x0080; +} + +int kr_dnssec_key_tag(uint16_t rrtype, const uint8_t *rdata, size_t rdlen) +{ + if (!rdata || rdlen == 0 || (rrtype != KNOT_RRTYPE_DS && rrtype != KNOT_RRTYPE_DNSKEY)) { + return kr_error(EINVAL); + } + if (rrtype == KNOT_RRTYPE_DS) { + return wire_read_u16(rdata); + } else if (rrtype == KNOT_RRTYPE_DNSKEY) { + struct dseckey *key = NULL; + int ret = kr_dnssec_key_from_rdata(&key, NULL, rdata, rdlen); + if (ret != 0) { + return ret; + } + uint16_t keytag = dnssec_key_get_keytag((dnssec_key_t *)key); + kr_dnssec_key_free(&key); + return keytag; + } else { + return kr_error(EINVAL); + } +} + +int kr_dnssec_key_match(const uint8_t *key_a_rdata, size_t key_a_rdlen, + const uint8_t *key_b_rdata, size_t key_b_rdlen) +{ + dnssec_key_t *key_a = NULL, *key_b = NULL; + int ret = kr_dnssec_key_from_rdata((struct dseckey **)&key_a, NULL, key_a_rdata, key_a_rdlen); + if (ret != 0) { + return ret; + } + ret = kr_dnssec_key_from_rdata((struct dseckey **)&key_b, NULL, key_b_rdata, key_b_rdlen); + if (ret != 0) { + dnssec_key_free(key_a); + return ret; + } + /* If the algorithm and the public key match, we can be sure + * that they are the same key. */ + ret = kr_error(ENOENT); + dnssec_binary_t pk_a, pk_b; + if (dnssec_key_get_algorithm(key_a) == dnssec_key_get_algorithm(key_b) && + dnssec_key_get_pubkey(key_a, &pk_a) == DNSSEC_EOK && + dnssec_key_get_pubkey(key_b, &pk_b) == DNSSEC_EOK) { + if (pk_a.size == pk_b.size && memcmp(pk_a.data, pk_b.data, pk_a.size) == 0) { + ret = 0; + } + } + dnssec_key_free(key_a); + dnssec_key_free(key_b); + return ret; +} + +int kr_dnssec_key_from_rdata(struct dseckey **key, const knot_dname_t *kown, const uint8_t *rdata, size_t rdlen) +{ + if (!key || !rdata || rdlen == 0) { + return kr_error(EINVAL); + } + + dnssec_key_t *new_key = NULL; + const dnssec_binary_t binary_key = { + .size = rdlen, + .data = (uint8_t *)rdata + }; + + int ret = dnssec_key_new(&new_key); + if (ret != DNSSEC_EOK) { + return kr_error(ENOMEM); + } + ret = dnssec_key_set_rdata(new_key, &binary_key); + if (ret != DNSSEC_EOK) { + dnssec_key_free(new_key); + return kr_error(ret); + } + if (kown) { + ret = dnssec_key_set_dname(new_key, kown); + if (ret != DNSSEC_EOK) { + dnssec_key_free(new_key); + return kr_error(ENOMEM); + } + } + + *key = (struct dseckey *) new_key; + return kr_ok(); +} + +void kr_dnssec_key_free(struct dseckey **key) +{ + assert(key); + + dnssec_key_free((dnssec_key_t *) *key); + *key = NULL; +} + +int kr_dnssec_matches_name_and_type(const ranked_rr_array_t *rrs, uint32_t qry_uid, + const knot_dname_t *name, uint16_t type) +{ + int ret = kr_error(ENOENT); + for (size_t i = 0; i < rrs->len; ++i) { + const ranked_rr_array_entry_t *entry = rrs->at[i]; + const knot_rrset_t *nsec = entry->rr; + if (entry->qry_uid != qry_uid || entry->yielded) { + continue; + } + if (nsec->type != KNOT_RRTYPE_NSEC && + nsec->type != KNOT_RRTYPE_NSEC3) { + continue; + } + if (!kr_rank_test(entry->rank, KR_RANK_SECURE)) { + continue; + } + if (nsec->type == KNOT_RRTYPE_NSEC) { + ret = kr_nsec_matches_name_and_type(nsec, name, type); + } else { + ret = kr_nsec3_matches_name_and_type(nsec, name, type); + } + if (ret == kr_ok()) { + return kr_ok(); + } else if (ret != kr_error(ENOENT)) { + return ret; + } + } + return ret; +} diff --git a/lib/dnssec.h b/lib/dnssec.h new file mode 100644 index 0000000..b54908c --- /dev/null +++ b/lib/dnssec.h @@ -0,0 +1,152 @@ +/* Copyright (C) 2015-2017 CZ.NIC, z.s.p.o. + + This program is free software: you can redistribute it and/or modify + it under the terms of the GNU General Public License as published by + the Free Software Foundation, either version 3 of the License, or + (at your option) any later version. + + This program is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the + GNU General Public License for more details. + + You should have received a copy of the GNU General Public License + along with this program. If not, see . + */ + +#pragma once + +#include "lib/defines.h" +#include "lib/utils.h" +#include + +/** + * Initialise cryptographic back-end. + */ +KR_EXPORT +void kr_crypto_init(void); + +/** + * De-initialise cryptographic back-end. + */ +KR_EXPORT +void kr_crypto_cleanup(void); + +/** + * Re-initialise cryptographic back-end. + * @note Must be called after fork() in the child. + */ +KR_EXPORT +void kr_crypto_reinit(void); + +/** Opaque DNSSEC key pointer. */ +struct dseckey; + +#define KR_DNSSEC_VFLG_WEXPAND 0x01 +#define KR_DNSSEC_VFLG_OPTOUT 0x02 + +/** DNSSEC validation context. */ +struct kr_rrset_validation_ctx { + const knot_pkt_t *pkt; /*!< Packet to be validated. */ + ranked_rr_array_t *rrs; /*!< List of preselected RRs to be validated. */ + knot_section_t section_id; /*!< Section to work with. */ + const knot_rrset_t *keys; /*!< DNSKEY RRSet. */ + const knot_dname_t *zone_name; /*!< Name of the zone containing the RRSIG RRSet. */ + uint32_t timestamp; /*!< Validation time. */ + bool has_nsec3; /*!< Whether to use NSEC3 validation. */ + uint32_t qry_uid; /*!< Current query uid. */ + uint32_t flags; /*!< Output - Flags. */ + uint32_t err_cnt; /*!< Output - Number of validation failures. */ + int result; /*!< Output - 0 or error code. */ + struct { + unsigned int matching_name_type; /*!< Name + type matches */ + unsigned int expired; + unsigned int notyet; + unsigned int signer_invalid; /*!< Signer is not zone apex */ + unsigned int labels_invalid; /*!< Number of labels in RRSIG */ + unsigned int key_invalid; /*!< Algorithm/keytag/key owner */ + unsigned int crypto_invalid; + unsigned int nsec_invalid; + } rrs_counters; /*!< Error counters for single RRset validation. */ +}; + +typedef struct kr_rrset_validation_ctx kr_rrset_validation_ctx_t; + +/** + * Validate RRSet. + * @param vctx Pointer to validation context. + * @param covered RRSet covered by a signature. It must be in canonical format. + * @return 0 or error code, same as vctx->result. + */ +int kr_rrset_validate(kr_rrset_validation_ctx_t *vctx, + const knot_rrset_t *covered); + +/** + * Check whether the DNSKEY rrset matches the supplied trust anchor RRSet. + * @param vctx Pointer to validation context. + * @param ta Trust anchor RRSet against which to validate the DNSKEY RRSet. + * @return 0 or error code, same as vctx->result. In particular, + * DNSSEC_INVALID_DS_ALGORITHM if *each* DS records is unusable + * due to unimplemented DNSKEY or DS algorithm. + */ +int kr_dnskeys_trusted(kr_rrset_validation_ctx_t *vctx, const knot_rrset_t *ta); + +/** Return true if the DNSKEY can be used as a ZSK. */ +KR_EXPORT KR_PURE +bool kr_dnssec_key_zsk(const uint8_t *dnskey_rdata); + +/** Return true if the DNSKEY indicates being KSK (=> has SEP). */ +KR_EXPORT KR_PURE +bool kr_dnssec_key_ksk(const uint8_t *dnskey_rdata); + +/** Return true if the DNSKEY is revoked. */ +KR_EXPORT KR_PURE +bool kr_dnssec_key_revoked(const uint8_t *dnskey_rdata); + +/** Return DNSKEY tag. + * @param rrtype RR type (either DS or DNSKEY are supported) + * @param rdata Key/digest RDATA. + * @param rdlen RDATA length. + * @return Key tag (positive number), or an error code + */ +KR_EXPORT KR_PURE +int kr_dnssec_key_tag(uint16_t rrtype, const uint8_t *rdata, size_t rdlen); + +/** Return 0 if the two keys are identical. + * @note This compares RDATA only, algorithm and public key must match. + * @param key_a_rdata First key RDATA + * @param key_a_rdlen First key RDATA length + * @param key_b_rdata Second key RDATA + * @param key_b_rdlen Second key RDATA length + * @return 0 if they match or an error code + */ +KR_EXPORT KR_PURE +int kr_dnssec_key_match(const uint8_t *key_a_rdata, size_t key_a_rdlen, + const uint8_t *key_b_rdata, size_t key_b_rdlen); + +/** + * Construct a DNSSEC key. + * @param key Pointer to be set to newly created DNSSEC key. + * @param kown DNSKEY owner name. + * @param rdata DNSKEY RDATA + * @param rdlen DNSKEY RDATA length + * @return 0 or error code; in particular: DNSSEC_INVALID_KEY_ALGORITHM + */ +int kr_dnssec_key_from_rdata(struct dseckey **key, const knot_dname_t *kown, const uint8_t *rdata, size_t rdlen); + +/** + * Frees the DNSSEC key. + * @param key Pointer to freed key. + */ +void kr_dnssec_key_free(struct dseckey **key); + +/** + * Checks whether NSEC/NSEC3 RR selected by iterator matches the supplied name and type. + * @param rrs Records selected by iterator. + * @param qry_uid Query unique identifier where NSEC/NSEC3 belongs to. + * @param name Name to be checked. + * @param type Type to be checked. + * @return 0 or error code. + */ +int kr_dnssec_matches_name_and_type(const ranked_rr_array_t *rrs, uint32_t qry_uid, + const knot_dname_t *name, uint16_t type); diff --git a/lib/dnssec/nsec.c b/lib/dnssec/nsec.c new file mode 100644 index 0000000..4ae13b2 --- /dev/null +++ b/lib/dnssec/nsec.c @@ -0,0 +1,540 @@ +/* Copyright (C) 2015-2017 CZ.NIC, z.s.p.o. + + This program is free software: you can redistribute it and/or modify + it under the terms of the GNU General Public License as published by + the Free Software Foundation, either version 3 of the License, or + (at your option) any later version. + + This program is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the + GNU General Public License for more details. + + You should have received a copy of the GNU General Public License + along with this program. If not, see . + */ + +#include +#include + +#include +#include +#include +#include +#include +#include +#include +#include + +#include "lib/defines.h" +#include "lib/dnssec/nsec.h" + + +int kr_nsec_children_in_zone_check(const uint8_t *bm, uint16_t bm_size) +{ + if (!bm) { + return kr_error(EINVAL); + } + const bool parent_side = + dnssec_nsec_bitmap_contains(bm, bm_size, KNOT_RRTYPE_DNAME) + || (dnssec_nsec_bitmap_contains(bm, bm_size, KNOT_RRTYPE_NS) + && !dnssec_nsec_bitmap_contains(bm, bm_size, KNOT_RRTYPE_SOA) + ); + return parent_side ? abs(ENOENT) : kr_ok(); + /* LATER: after refactoring, probably also check if signer name equals owner, + * but even without that it's not possible to attack *correctly* signed zones. + */ +} + +/** + * Check whether the NSEC RR proves that there is no closer match for . + * @param nsec NSEC RRSet. + * @param sname Searched name. + * @return 0 if proves, >0 if not (abs(ENOENT)), or error code (<0). + */ +static int nsec_covers(const knot_rrset_t *nsec, const knot_dname_t *sname) +{ + assert(nsec && sname); + if (knot_dname_cmp(sname, nsec->owner) <= 0) { + return abs(ENOENT); /* 'sname' before 'owner', so can't be covered */ + } + + /* If NSEC 'owner' >= 'next', it means that there is nothing after 'owner' */ + /* We have to lower-case it with libknot >= 2.7; see also RFC 6840 5.1. */ + knot_dname_t next[KNOT_DNAME_MAXLEN]; + int ret = knot_dname_to_wire(next, knot_nsec_next(nsec->rrs.rdata), sizeof(next)); + if (ret < 0) { + assert(!ret); + return kr_error(ret); + } + knot_dname_to_lower(next); + + const bool is_last_nsec = knot_dname_cmp(nsec->owner, next) >= 0; + const bool in_range = is_last_nsec || knot_dname_cmp(sname, next) < 0; + if (!in_range) { + return abs(ENOENT); + } + /* Before returning kr_ok(), we have to check a special case: + * sname might be under delegation from owner and thus + * not in the zone of this NSEC at all. + */ + if (knot_dname_in_bailiwick(sname, nsec->owner) <= 0) { + return kr_ok(); + } + const uint8_t *bm = knot_nsec_bitmap(nsec->rrs.rdata); + uint16_t bm_size = knot_nsec_bitmap_len(nsec->rrs.rdata); + + return kr_nsec_children_in_zone_check(bm, bm_size); +} + +#define FLG_NOEXIST_RRTYPE (1 << 0) /**< exists, does not exist. */ +#define FLG_NOEXIST_RRSET (1 << 1) /**< does not exist. */ +#define FLG_NOEXIST_WILDCARD (1 << 2) /**< No wildcard covering exists. */ +#define FLG_NOEXIST_CLOSER (1 << 3) /**< Wildcard covering exists, but doesn't match STYPE. */ + + +/** + * According to set flags determine whether NSEC proving + * RRset or RRType non-existense has been found. + * @param f Flags to inspect. + * @return True if required NSEC exists. + */ +#define kr_nsec_rrset_noexist(f) \ + ((f) & (FLG_NOEXIST_RRTYPE | FLG_NOEXIST_RRSET)) +/** + * According to set flags determine whether wildcard non-existense + * has been proven. + * @param f Flags to inspect. + * @return True if wildcard not exists. + */ +#define kr_nsec_wcard_noexist(f) ((f) & FLG_NOEXIST_WILDCARD) + +/** + * According to set flags determine whether authenticated denial of existence has been proven. + * @param f Flags to inspect. + * @return True if denial of existence proven. + */ +#define kr_nsec_existence_denied(f) \ + ((kr_nsec_rrset_noexist(f)) && (kr_nsec_wcard_noexist(f))) + +/** + * Name error response check (RFC4035 3.1.3.2; RFC4035 5.4, bullet 2). + * @note Returned flags must be checked in order to prove denial. + * @param flags Flags to be set according to check outcome. + * @param nsec NSEC RR. + * @param name Name to be checked. + * @param pool + * @return 0 or error code. + */ +static int name_error_response_check_rr(int *flags, const knot_rrset_t *nsec, + const knot_dname_t *name) +{ + assert(flags && nsec && name); + + if (nsec_covers(nsec, name) == 0) { + *flags |= FLG_NOEXIST_RRSET; + } + + /* Try to find parent wildcard that is proved by this NSEC. */ + uint8_t namebuf[KNOT_DNAME_MAXLEN]; + int ret = knot_dname_to_wire(namebuf, name, sizeof(namebuf)); + if (ret < 0) + return ret; + knot_dname_t *ptr = namebuf; + while (ptr[0]) { + /* Remove leftmost label and replace it with '\1*'. */ + ptr = (uint8_t *) knot_wire_next_label(ptr, NULL); + if (!ptr) { + return kr_error(EINVAL); + } + *(--ptr) = '*'; + *(--ptr) = 1; + /* True if this wildcard provably doesn't exist. */ + if (nsec_covers(nsec, ptr) == 0) { + *flags |= FLG_NOEXIST_WILDCARD; + break; + } + /* Remove added leftmost asterisk. */ + ptr += 2; + } + + return kr_ok(); +} + +int kr_nsec_name_error_response_check(const knot_pkt_t *pkt, knot_section_t section_id, + const knot_dname_t *sname) +{ + const knot_pktsection_t *sec = knot_pkt_section(pkt, section_id); + if (!sec || !sname) { + return kr_error(EINVAL); + } + + int flags = 0; + for (unsigned i = 0; i < sec->count; ++i) { + const knot_rrset_t *rrset = knot_pkt_rr(sec, i); + if (rrset->type != KNOT_RRTYPE_NSEC) { + continue; + } + int ret = name_error_response_check_rr(&flags, rrset, sname); + if (ret != 0) { + return ret; + } + } + + return kr_nsec_existence_denied(flags) ? kr_ok() : kr_error(ENOENT); +} + +/** + * Returns the labels from the covering RRSIG RRs. + * @note The number must be the same in all covering RRSIGs. + * @param nsec NSEC RR. + * @param sec Packet section. + * @param Number of labels or (negative) error code. + */ +static int coverign_rrsig_labels(const knot_rrset_t *nsec, const knot_pktsection_t *sec) +{ + assert(nsec && sec); + + int ret = kr_error(ENOENT); + + for (unsigned i = 0; i < sec->count; ++i) { + const knot_rrset_t *rrset = knot_pkt_rr(sec, i); + if ((rrset->type != KNOT_RRTYPE_RRSIG) || + (!knot_dname_is_equal(rrset->owner, nsec->owner))) { + continue; + } + + knot_rdata_t *rdata_j = rrset->rrs.rdata; + for (uint16_t j = 0; j < rrset->rrs.count; + ++j, rdata_j = knot_rdataset_next(rdata_j)) { + if (knot_rrsig_type_covered(rdata_j) != KNOT_RRTYPE_NSEC) { + continue; + } + + if (ret < 0) { + ret = knot_rrsig_labels(rdata_j); + } else { + if (ret != knot_rrsig_labels(rdata_j)) { + return kr_error(EINVAL); + } + } + } + } + + return ret; +} + + +int kr_nsec_bitmap_nodata_check(const uint8_t *bm, uint16_t bm_size, uint16_t type, const knot_dname_t *owner) +{ + const int NO_PROOF = abs(ENOENT); + if (!bm || !owner) { + return kr_error(EINVAL); + } + if (dnssec_nsec_bitmap_contains(bm, bm_size, type)) { + return NO_PROOF; + } + + if (type != KNOT_RRTYPE_CNAME + && dnssec_nsec_bitmap_contains(bm, bm_size, KNOT_RRTYPE_CNAME)) { + return NO_PROOF; + } + /* Special behavior around zone cuts. */ + switch (type) { + case KNOT_RRTYPE_DS: + /* Security feature: in case of DS also check for SOA + * non-existence to be more certain that we don't hold + * a child-side NSEC by some mistake (e.g. when forwarding). + * See RFC4035 5.2, next-to-last paragraph. + * This doesn't apply for root DS as it doesn't exist in DNS hierarchy. + */ + if (owner[0] != '\0' && dnssec_nsec_bitmap_contains(bm, bm_size, KNOT_RRTYPE_SOA)) { + return NO_PROOF; + } + break; + case KNOT_RRTYPE_CNAME: + /* Exception from the `default` rule. It's perhaps disputable, + * but existence of CNAME at zone apex is not allowed, so we + * consider a parent-side record to be enough to prove non-existence. */ + break; + default: + /* Parent-side delegation record isn't authoritative for non-DS; + * see RFC6840 4.1. */ + if (dnssec_nsec_bitmap_contains(bm, bm_size, KNOT_RRTYPE_NS) + && !dnssec_nsec_bitmap_contains(bm, bm_size, KNOT_RRTYPE_SOA)) { + return NO_PROOF; + } + /* LATER(opt): perhaps short-circuit test if we repeat it here. */ + } + + return kr_ok(); +} + +/** + * Attempt to prove NODATA given a matching NSEC. + * @param flags Flags to be set according to check outcome. + * @param nsec NSEC RR. + * @param type Type to be checked. + * @return 0 on success, abs(ENOENT) for no proof, or error code (<0). + * @note It's not a *full* proof, of course (wildcards, etc.) + * @TODO returning result via `flags` is just ugly. + */ +static int no_data_response_check_rrtype(int *flags, const knot_rrset_t *nsec, + uint16_t type) +{ + assert(flags && nsec); + + const uint8_t *bm = knot_nsec_bitmap(nsec->rrs.rdata); + uint16_t bm_size = knot_nsec_bitmap_len(nsec->rrs.rdata); + int ret = kr_nsec_bitmap_nodata_check(bm, bm_size, type, nsec->owner); + if (ret == kr_ok()) { + *flags |= FLG_NOEXIST_RRTYPE; + } + return ret <= 0 ? ret : kr_ok(); +} + +/** + * Perform check for RR type wildcard existence denial according to RFC4035 5.4, bullet 1. + * @param flags Flags to be set according to check outcome. + * @param nsec NSEC RR. + * @param sec Packet section to work with. + * @return 0 or error code. + */ +static int no_data_wildcard_existence_check(int *flags, const knot_rrset_t *nsec, + const knot_pktsection_t *sec) +{ + assert(flags && nsec && sec); + + int rrsig_labels = coverign_rrsig_labels(nsec, sec); + if (rrsig_labels < 0) { + return rrsig_labels; + } + int nsec_labels = knot_dname_labels(nsec->owner, NULL); + if (nsec_labels < 0) { + return nsec_labels; + } + + if (rrsig_labels == nsec_labels) { + *flags |= FLG_NOEXIST_WILDCARD; + } + + return kr_ok(); +} + +/** + * Perform check for NSEC wildcard existence that covers sname and + * have no stype bit set. + * @param pkt Packet structure to be processed. + * @param sec Packet section to work with. + * @param sname Queried domain name. + * @param stype Queried type. + * @return 0 or error code. + */ +static int wildcard_match_check(const knot_pkt_t *pkt, const knot_pktsection_t *sec, + const knot_dname_t *sname, uint16_t stype) +{ + if (!sec || !sname) { + return kr_error(EINVAL); + } + + int flags = 0; + for (unsigned i = 0; i < sec->count; ++i) { + const knot_rrset_t *rrset = knot_pkt_rr(sec, i); + if (rrset->type != KNOT_RRTYPE_NSEC) { + continue; + } + if (!knot_dname_is_wildcard(rrset->owner)) { + continue; + } + if (!knot_dname_is_equal(rrset->owner, sname)) { + int wcard_labels = knot_dname_labels(rrset->owner, NULL); + int common_labels = knot_dname_matched_labels(rrset->owner, sname); + int rrsig_labels = coverign_rrsig_labels(rrset, sec); + if (wcard_labels < 1 || + common_labels != wcard_labels - 1 || + common_labels != rrsig_labels) { + continue; + } + } + int ret = no_data_response_check_rrtype(&flags, rrset, stype); + if (ret != 0) { + return ret; + } + } + return (flags & FLG_NOEXIST_RRTYPE) ? kr_ok() : kr_error(ENOENT); +} + +int kr_nsec_no_data_response_check(const knot_pkt_t *pkt, knot_section_t section_id, + const knot_dname_t *sname, uint16_t stype) +{ + const knot_pktsection_t *sec = knot_pkt_section(pkt, section_id); + if (!sec || !sname) { + return kr_error(EINVAL); + } + + int flags = 0; + for (unsigned i = 0; i < sec->count; ++i) { + const knot_rrset_t *rrset = knot_pkt_rr(sec, i); + if (rrset->type != KNOT_RRTYPE_NSEC) { + continue; + } + if (knot_dname_is_equal(rrset->owner, sname)) { + int ret = no_data_response_check_rrtype(&flags, rrset, stype); + if (ret != 0) { + return ret; + } + } + } + + return (flags & FLG_NOEXIST_RRTYPE) ? kr_ok() : kr_error(ENOENT); +} + +int kr_nsec_wildcard_answer_response_check(const knot_pkt_t *pkt, knot_section_t section_id, + const knot_dname_t *sname) +{ + const knot_pktsection_t *sec = knot_pkt_section(pkt, section_id); + if (!sec || !sname) { + return kr_error(EINVAL); + } + + for (unsigned i = 0; i < sec->count; ++i) { + const knot_rrset_t *rrset = knot_pkt_rr(sec, i); + if (rrset->type != KNOT_RRTYPE_NSEC) { + continue; + } + if (nsec_covers(rrset, sname) == 0) { + return kr_ok(); + } + } + + return kr_error(ENOENT); +} + +int kr_nsec_existence_denial(const knot_pkt_t *pkt, knot_section_t section_id, + const knot_dname_t *sname, uint16_t stype) +{ + const knot_pktsection_t *sec = knot_pkt_section(pkt, section_id); + if (!sec || !sname) { + return kr_error(EINVAL); + } + + int flags = 0; + for (unsigned i = 0; i < sec->count; ++i) { + const knot_rrset_t *rrset = knot_pkt_rr(sec, i); + if (rrset->type != KNOT_RRTYPE_NSEC) { + continue; + } + /* NSEC proves that name exists, but has no data (RFC4035 4.9, 1) */ + if (knot_dname_is_equal(rrset->owner, sname)) { + no_data_response_check_rrtype(&flags, rrset, stype); + } else { + /* NSEC proves that name doesn't exist (RFC4035, 4.9, 2) */ + name_error_response_check_rr(&flags, rrset, sname); + } + no_data_wildcard_existence_check(&flags, rrset, sec); + } + if (kr_nsec_existence_denied(flags)) { + /* denial of existence proved accordignly to 4035 5.4 - + * NSEC proving either rrset non-existance or + * qtype non-existance has been found, + * and no wildcard expansion occurred. + */ + return kr_ok(); + } else if (kr_nsec_rrset_noexist(flags)) { + /* NSEC proving either rrset non-existance or + * qtype non-existance has been found, + * but wildcard expansion occurs. + * Try to find matching wildcard and check + * corresponding types. + */ + return wildcard_match_check(pkt, sec, sname, stype); + } + return kr_error(ENOENT); +} + +int kr_nsec_ref_to_unsigned(const knot_pkt_t *pkt) +{ + int nsec_found = 0; + const knot_pktsection_t *sec = knot_pkt_section(pkt, KNOT_AUTHORITY); + if (!sec) { + return kr_error(EINVAL); + } + for (unsigned i = 0; i < sec->count; ++i) { + const knot_rrset_t *ns = knot_pkt_rr(sec, i); + if (ns->type == KNOT_RRTYPE_DS) { + return kr_error(EEXIST); + } + if (ns->type != KNOT_RRTYPE_NS) { + continue; + } + nsec_found = 0; + for (unsigned j = 0; j < sec->count; ++j) { + const knot_rrset_t *nsec = knot_pkt_rr(sec, j); + if (nsec->type == KNOT_RRTYPE_DS) { + return kr_error(EEXIST); + } + if (nsec->type != KNOT_RRTYPE_NSEC) { + continue; + } + /* nsec found + * check if owner name matches the delegation name + */ + if (!knot_dname_is_equal(nsec->owner, ns->owner)) { + /* nsec does not match the delegation */ + continue; + } + nsec_found = 1; + const uint8_t *bm = knot_nsec_bitmap(nsec->rrs.rdata); + uint16_t bm_size = knot_nsec_bitmap_len(nsec->rrs.rdata); + if (!bm) { + return kr_error(EINVAL); + } + if (dnssec_nsec_bitmap_contains(bm, bm_size, + KNOT_RRTYPE_NS) && + !dnssec_nsec_bitmap_contains(bm, bm_size, + KNOT_RRTYPE_DS) && + !dnssec_nsec_bitmap_contains(bm, bm_size, + KNOT_RRTYPE_SOA)) { + /* rfc4035, 5.2 */ + return kr_ok(); + } + } + if (nsec_found) { + /* nsec which owner matches + * the delegation name was found, + * but nsec type bitmap contains wrong types + */ + return kr_error(EINVAL); + } else { + /* nsec that matches delegation was not found */ + return kr_error(DNSSEC_NOT_FOUND); + } + } + + return kr_error(EINVAL); +} + +int kr_nsec_matches_name_and_type(const knot_rrset_t *nsec, + const knot_dname_t *name, uint16_t type) +{ + /* It's not secure enough to just check a single bit for (some) other types, + * but we don't (currently) only use this API for NS. See RFC 6840 sec. 4. + */ + if (type != KNOT_RRTYPE_NS || !nsec || !name) { + assert(!EINVAL); + return kr_error(EINVAL); + } + if (!knot_dname_is_equal(nsec->owner, name)) { + return kr_error(ENOENT); + } + const uint8_t *bm = knot_nsec_bitmap(nsec->rrs.rdata); + uint16_t bm_size = knot_nsec_bitmap_len(nsec->rrs.rdata); + if (!bm) { + return kr_error(EINVAL); + } + if (dnssec_nsec_bitmap_contains(bm, bm_size, type)) { + return kr_ok(); + } else { + return kr_error(ENOENT); + } +} diff --git a/lib/dnssec/nsec.h b/lib/dnssec/nsec.h new file mode 100644 index 0000000..9439a88 --- /dev/null +++ b/lib/dnssec/nsec.h @@ -0,0 +1,105 @@ +/* Copyright (C) 2015-2017 CZ.NIC, z.s.p.o. + + This program is free software: you can redistribute it and/or modify + it under the terms of the GNU General Public License as published by + the Free Software Foundation, either version 3 of the License, or + (at your option) any later version. + + This program is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the + GNU General Public License for more details. + + You should have received a copy of the GNU General Public License + along with this program. If not, see . + */ + +#pragma once + +#include + +/** + * Check bitmap that child names are contained in the same zone. + * @note see RFC6840 4.1. + * @param bm Bitmap from NSEC or NSEC3. + * @param bm_size Bitmap size. + * @return 0 if they are, >0 if not (abs(ENOENT)), <0 on error. + */ +int kr_nsec_children_in_zone_check(const uint8_t *bm, uint16_t bm_size); + +/** + * Check an NSEC or NSEC3 bitmap for NODATA for a type. + * @param bm Bitmap. + * @param bm_size Bitmap size. + * @param type RR type to check. + * @param owner NSEC record owner. + * @note This includes special checks for zone cuts, e.g. from RFC 6840 sec. 4. + * @return 0, abs(ENOENT) (no proof), kr_error(EINVAL) + */ +int kr_nsec_bitmap_nodata_check(const uint8_t *bm, uint16_t bm_size, uint16_t type, const knot_dname_t *owner); + +/** + * Name error response check (RFC4035 3.1.3.2; RFC4035 5.4, bullet 2). + * @note No RRSIGs are validated. + * @param pkt Packet structure to be processed. + * @param section_id Packet section to be processed. + * @param sname Name to be checked. + * @return 0 or error code. + */ +int kr_nsec_name_error_response_check(const knot_pkt_t *pkt, knot_section_t section_id, + const knot_dname_t *sname); + +/** + * No data response check (RFC4035 3.1.3.1; RFC4035 5.4, bullet 1). + * @param pkt Packet structure to be processed. + * @param section_id Packet section to be processed. + * @param sname Name to be checked. + * @param stype Type to be checked. + * @return 0 or error code. + */ +int kr_nsec_no_data_response_check(const knot_pkt_t *pkt, knot_section_t section_id, + const knot_dname_t *sname, uint16_t stype); + +/** + * Wildcard answer response check (RFC4035 3.1.3.3). + * @param pkt Packet structure to be processed. + * @param section_id Packet section to be processed. + * @param sname Name to be checked. + * @return 0 or error code. + */ +int kr_nsec_wildcard_answer_response_check(const knot_pkt_t *pkt, knot_section_t section_id, + const knot_dname_t *sname); + +/** + * Authenticated denial of existence according to RFC4035 5.4. + * @note No RRSIGs are validated. + * @param pkt Packet structure to be processed. + * @param section_id Packet section to be processed. + * @param sname Queried domain name. + * @param stype Queried type. + * @return 0 or error code. + */ +int kr_nsec_existence_denial(const knot_pkt_t *pkt, knot_section_t section_id, + const knot_dname_t *sname, uint16_t stype); + +/** + * Referral to unsigned subzone check (RFC4035 5.2). + * @note No RRSIGs are validated. + * @param pkt Packet structure to be processed. + * @return 0 or error code: + * DNSSEC_NOT_FOUND - neither ds nor nsec records + * were not found. + * EEXIST - ds record was found. + * EINVAL - bogus. + */ +int kr_nsec_ref_to_unsigned(const knot_pkt_t *pkt); + +/** + * Checks whether supplied NSEC RR matches the supplied name and type. + * @param nsec NSEC RR. + * @param name Name to be checked. + * @param type Type to be checked. Only use with NS! TODO (+copy&paste NSEC3) + * @return 0 or error code. + */ +int kr_nsec_matches_name_and_type(const knot_rrset_t *nsec, + const knot_dname_t *name, uint16_t type); diff --git a/lib/dnssec/nsec3.c b/lib/dnssec/nsec3.c new file mode 100644 index 0000000..4c2ef95 --- /dev/null +++ b/lib/dnssec/nsec3.c @@ -0,0 +1,776 @@ +/* Copyright (C) 2015-2017 CZ.NIC, z.s.p.o. + + This program is free software: you can redistribute it and/or modify + it under the terms of the GNU General Public License as published by + the Free Software Foundation, either version 3 of the License, or + (at your option) any later version. + + This program is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the + GNU General Public License for more details. + + You should have received a copy of the GNU General Public License + along with this program. If not, see . + */ + +#include +#include + +#include +#include +#include +#include +#include +#include +#include + +#include "lib/defines.h" +#include "lib/dnssec/nsec.h" +#include "lib/dnssec/nsec3.h" + +#define OPT_OUT_BIT 0x01 + +//#define FLG_CLOSEST_ENCLOSER (1 << 0) +#define FLG_CLOSEST_PROVABLE_ENCLOSER (1 << 1) +#define FLG_NAME_COVERED (1 << 2) +#define FLG_NAME_MATCHED (1 << 3) +#define FLG_TYPE_BIT_MISSING (1 << 4) +#define FLG_CNAME_BIT_MISSING (1 << 5) + +/** + * Obtains NSEC3 parameters from RR. + * @param params NSEC3 parameters structure to be set. + * @param nsec3 NSEC3 RR containing the parameters. + * @return 0 or error code. + */ +static int nsec3_parameters(dnssec_nsec3_params_t *params, const knot_rrset_t *nsec3) +{ + assert(params && nsec3); + + const knot_rdata_t *rr = knot_rdataset_at(&nsec3->rrs, 0); + assert(rr); + + /* Every NSEC3 RR contains data from NSEC3PARAMS. */ + const size_t SALT_OFFSET = 5; /* First 5 octets contain { Alg, Flags, Iterations, Salt length } */ + dnssec_binary_t rdata = { + .size = SALT_OFFSET + (size_t)knot_nsec3_salt_len(nsec3->rrs.rdata), + .data = /*const-cast*/(uint8_t *)rr->data, + }; + if (rdata.size > rr->len) + return kr_error(EMSGSIZE); + + int ret = dnssec_nsec3_params_from_rdata(params, &rdata); + if (ret != DNSSEC_EOK) { + return kr_error(EINVAL); + } + + return kr_ok(); +} + +/** + * Computes a hash of a given domain name. + * @param hash Resulting hash, must be freed. + * @param params NSEC3 parameters. + * @param name Domain name to be hashed. + * @return 0 or error code. + */ +static int hash_name(dnssec_binary_t *hash, const dnssec_nsec3_params_t *params, + const knot_dname_t *name) +{ + assert(hash && params); + if (!name) + return kr_error(EINVAL); + + dnssec_binary_t dname = { + .size = knot_dname_size(name), + .data = (uint8_t *) name, + }; + + int ret = dnssec_nsec3_hash(&dname, params, hash); + if (ret != DNSSEC_EOK) { + return kr_error(EINVAL); + } + + return kr_ok(); +} + +/** + * Read hash from NSEC3 owner name and store its binary form. + * @param hash Buffer to be written. + * @param max_hash_size Maximal has size. + * @param nsec3 NSEC3 RR. + * @return 0 or error code. + */ +static int read_owner_hash(dnssec_binary_t *hash, size_t max_hash_size, const knot_rrset_t *nsec3) +{ + assert(hash && nsec3); + assert(hash->data); + + int32_t ret = base32hex_decode(nsec3->owner + 1, nsec3->owner[0], hash->data, max_hash_size); + if (ret < 0) { + return kr_error(EILSEQ); + } + hash->size = ret; + + return kr_ok(); +} + +#define MAX_HASH_BYTES 64 +/** + * Closest (provable) encloser match (RFC5155 7.2.1, bullet 1). + * @param flags Flags to be set according to check outcome. + * @param nsec3 NSEC3 RR. + * @param name Name to be checked. + * @param skipped Number of skipped labels to find closest (provable) match. + * @return 0 or error code. + */ +static int closest_encloser_match(int *flags, const knot_rrset_t *nsec3, + const knot_dname_t *name, unsigned *skipped) +{ + assert(flags && nsec3 && name && skipped); + + uint8_t hash_data[MAX_HASH_BYTES] = {0, }; + dnssec_binary_t owner_hash = { 0, hash_data }; + dnssec_nsec3_params_t params = { 0, }; + dnssec_binary_t name_hash = { 0, }; + + int ret = read_owner_hash(&owner_hash, MAX_HASH_BYTES, nsec3); + if (ret != 0) { + goto fail; + } + + ret = nsec3_parameters(¶ms, nsec3); + if (ret != 0) { + goto fail; + } + + /* Root label has no encloser */ + if (!name[0]) { + ret = kr_error(ENOENT); + goto fail; + } + + const knot_dname_t *encloser = knot_wire_next_label(name, NULL); + *skipped = 1; + + while(encloser) { + ret = hash_name(&name_hash, ¶ms, encloser); + if (ret != 0) { + goto fail; + } + + if ((owner_hash.size == name_hash.size) && + (memcmp(owner_hash.data, name_hash.data, owner_hash.size) == 0)) { + dnssec_binary_free(&name_hash); + *flags |= FLG_CLOSEST_PROVABLE_ENCLOSER; + break; + } + + dnssec_binary_free(&name_hash); + + if (!encloser[0]) + break; + encloser = knot_wire_next_label(encloser, NULL); + ++(*skipped); + } + + ret = kr_ok(); + +fail: + if (params.salt.data) { + dnssec_nsec3_params_free(¶ms); + } + if (name_hash.data) { + dnssec_binary_free(&name_hash); + } + return ret; +} + +/** + * Checks whether NSEC3 RR covers the supplied name (RFC5155 7.2.1, bullet 2). + * @param flags Flags to be set according to check outcome. + * @param nsec3 NSEC3 RR. + * @param name Name to be checked. + * @return 0 or error code. + */ +static int covers_name(int *flags, const knot_rrset_t *nsec3, const knot_dname_t *name) +{ + assert(flags && nsec3 && name); + + uint8_t hash_data[MAX_HASH_BYTES] = { 0, }; + dnssec_binary_t owner_hash = { 0, hash_data }; + dnssec_nsec3_params_t params = { 0, }; + dnssec_binary_t name_hash = { 0, }; + + int ret = read_owner_hash(&owner_hash, MAX_HASH_BYTES, nsec3); + if (ret != 0) { + goto fail; + } + + ret = nsec3_parameters(¶ms, nsec3); + if (ret != 0) { + goto fail; + } + + ret = hash_name(&name_hash, ¶ms, name); + if (ret != 0) { + goto fail; + } + + uint8_t next_size = knot_nsec3_next_len(nsec3->rrs.rdata); + const uint8_t *next_hash = knot_nsec3_next(nsec3->rrs.rdata); + + if ((next_size > 0) && (owner_hash.size == next_size) && (name_hash.size == next_size)) { + /* All hash lengths must be same. */ + const uint8_t *ownrd = owner_hash.data; + const uint8_t *nextd = next_hash; + int covered = 0; + int greater_then_owner = (memcmp(ownrd, name_hash.data, next_size) < 0); + int less_then_next = (memcmp(name_hash.data, nextd, next_size) < 0); + if (memcmp(ownrd, nextd, next_size) < 0) { + /* + * 0 (...) owner ... next (...) MAX + * ^ + * name + * ==> + * (owner < name) && (name < next) + */ + covered = ((greater_then_owner) && (less_then_next)); + } else { + /* + * owner ... MAX, 0 ... next + * ^ ^ ^ + * name name name + * => + * (owner < name) || (name < next) + */ + covered = ((greater_then_owner) || (less_then_next)); + } + + if (covered) { + *flags |= FLG_NAME_COVERED; + + uint8_t nsec3_flags = knot_nsec3_flags(nsec3->rrs.rdata); + if (nsec3_flags & ~OPT_OUT_BIT) { + /* RFC5155 3.1.2 */ + ret = kr_error(EINVAL); + } else { + ret = kr_ok(); + } + } + } + +fail: + if (params.salt.data) { + dnssec_nsec3_params_free(¶ms); + } + if (name_hash.data) { + dnssec_binary_free(&name_hash); + } + return ret; +} + +/** + * Checks whether NSEC3 RR has the opt-out bit set. + * @param flags Flags to be set according to check outcome. + * @param nsec3 NSEC3 RR. + * @param name Name to be checked. + * @return 0 or error code. + */ +static bool has_optout(const knot_rrset_t *nsec3) +{ + if (!nsec3) { + return false; + } + + uint8_t nsec3_flags = knot_nsec3_flags(nsec3->rrs.rdata); + if (nsec3_flags & ~OPT_OUT_BIT) { + /* RFC5155 3.1.2 */ + return false; + } + + return nsec3_flags & OPT_OUT_BIT; +} + +/** + * Checks whether NSEC3 RR matches the supplied name. + * @param flags Flags to be set according to check outcome. + * @param nsec3 NSEC3 RR. + * @param name Name to be checked. + * @return 0 if matching, >0 if not (abs(ENOENT)), or error code (<0). + */ +static int matches_name(const knot_rrset_t *nsec3, const knot_dname_t *name) +{ + assert(nsec3 && name); + + uint8_t hash_data[MAX_HASH_BYTES] = { 0, }; + dnssec_binary_t owner_hash = { 0, hash_data }; + dnssec_nsec3_params_t params = { 0, }; + dnssec_binary_t name_hash = { 0, }; + + int ret = read_owner_hash(&owner_hash, MAX_HASH_BYTES, nsec3); + if (ret != 0) { + goto fail; + } + + ret = nsec3_parameters(¶ms, nsec3); + if (ret != 0) { + goto fail; + } + + ret = hash_name(&name_hash, ¶ms, name); + if (ret != 0) { + goto fail; + } + + if ((owner_hash.size == name_hash.size) && + (memcmp(owner_hash.data, name_hash.data, owner_hash.size) == 0)) { + ret = kr_ok(); + } else { + ret = abs(ENOENT); + } + +fail: + if (params.salt.data) { + dnssec_nsec3_params_free(¶ms); + } + if (name_hash.data) { + dnssec_binary_free(&name_hash); + } + return ret; +} +#undef MAX_HASH_BYTES + +/** + * Prepends an asterisk label to given name. + * + * @param tgt Target buffer to write domain name into. + * @param name Name to be added to the asterisk. + * @return Size of the resulting name or error code. + */ +static int prepend_asterisk(uint8_t *tgt, size_t maxlen, const knot_dname_t *name) +{ + assert(maxlen >= 3); + memcpy(tgt, "\1*", 3); + return knot_dname_to_wire(tgt + 2, name, maxlen - 2); +} + +/** + * Closest encloser proof (RFC5155 7.2.1). + * @note No RRSIGs are validated. + * @param pkt Packet structure to be processed. + * @param section_id Packet section to be processed. + * @param sname Name to be checked. + * @param encloser_name Returned matching encloser name, if found. + * @param matching_encloser_nsec3 Pointer to matching encloser NSEC RRSet. + * @param covering_next_nsec3 Pointer to covering next closer NSEC3 RRSet. + * @return 0 or error code. + */ +static int closest_encloser_proof(const knot_pkt_t *pkt, + knot_section_t section_id, + const knot_dname_t *sname, + const knot_dname_t **encloser_name, + const knot_rrset_t **matching_encloser_nsec3, + const knot_rrset_t **covering_next_nsec3) +{ + const knot_pktsection_t *sec = knot_pkt_section(pkt, section_id); + if (!sec || !sname) { + return kr_error(EINVAL); + } + + const knot_rrset_t *matching = NULL; + const knot_rrset_t *covering = NULL; + + int flags = 0; + const knot_dname_t *next_closer = NULL; + for (unsigned i = 0; i < sec->count; ++i) { + const knot_rrset_t *rrset = knot_pkt_rr(sec, i); + if (rrset->type != KNOT_RRTYPE_NSEC3) { + continue; + } + /* Also skip the NSEC3-to-match an ancestor of sname if it's + * a parent-side delegation, as that would mean the owner + * does not really exist (authoritatively in this zone, + * even in case of opt-out). + */ + const uint8_t *bm = knot_nsec3_bitmap(rrset->rrs.rdata); + uint16_t bm_size = knot_nsec3_bitmap_len(rrset->rrs.rdata); + if (kr_nsec_children_in_zone_check(bm, bm_size) != 0) { + continue; /* no fatal errors from bad RRs */ + } + /* Match the NSEC3 to sname or one of its ancestors. */ + unsigned skipped = 0; + flags = 0; + int ret = closest_encloser_match(&flags, rrset, sname, &skipped); + if (ret != 0) { + return ret; + } + if (!(flags & FLG_CLOSEST_PROVABLE_ENCLOSER)) { + continue; + } + matching = rrset; + /* Construct the next closer name and try to cover it. */ + --skipped; + next_closer = sname; + for (unsigned j = 0; j < skipped; ++j) { + assert(next_closer[0]); + next_closer = knot_wire_next_label(next_closer, NULL); + } + for (unsigned j = 0; j < sec->count; ++j) { + const knot_rrset_t *rrset_j = knot_pkt_rr(sec, j); + if (rrset_j->type != KNOT_RRTYPE_NSEC3) { + continue; + } + ret = covers_name(&flags, rrset_j, next_closer); + if (ret != 0) { + return ret; + } + if (flags & FLG_NAME_COVERED) { + covering = rrset_j; + break; + } + } + if (flags & FLG_NAME_COVERED) { + break; + } + flags = 0; // + } + + if ((flags & FLG_CLOSEST_PROVABLE_ENCLOSER) && (flags & FLG_NAME_COVERED) && next_closer) { + if (encloser_name && next_closer[0]) { + *encloser_name = knot_wire_next_label(next_closer, NULL); + } + if (matching_encloser_nsec3) { + *matching_encloser_nsec3 = matching; + } + if (covering_next_nsec3) { + *covering_next_nsec3 = covering; + } + return kr_ok(); + } + + return kr_error(ENOENT); +} + +/** + * Check whether any NSEC3 RR covers a wildcard RR at the closer encloser. + * @param pkt Packet structure to be processed. + * @param section_id Packet section to be processed. + * @param encloser Closest (provable) encloser domain name. + * @return 0 or error code: + * KNOT_ERANGE - NSEC3 RR (that covers a wildcard) + * has been found, but has opt-out flag set; + * otherwise - error. + */ +static int covers_closest_encloser_wildcard(const knot_pkt_t *pkt, knot_section_t section_id, + const knot_dname_t *encloser) +{ + const knot_pktsection_t *sec = knot_pkt_section(pkt, section_id); + if (!sec || !encloser) { + return kr_error(EINVAL); + } + + uint8_t wildcard[KNOT_DNAME_MAXLEN]; + wildcard[0] = 1; + wildcard[1] = '*'; + int encloser_len = knot_dname_size(encloser); + if (encloser_len < 0) { + return encloser_len; + } + memcpy(wildcard + 2, encloser, encloser_len); + + int flags = 0; + for (unsigned i = 0; i < sec->count; ++i) { + const knot_rrset_t *rrset = knot_pkt_rr(sec, i); + if (rrset->type != KNOT_RRTYPE_NSEC3) { + continue; + } + int ret = covers_name(&flags, rrset, wildcard); + if (ret != 0) { + return ret; + } + if (flags & FLG_NAME_COVERED) { + return has_optout(rrset) ? + kr_error(KNOT_ERANGE) : kr_ok(); + } + } + + return kr_error(ENOENT); +} + +int kr_nsec3_name_error_response_check(const knot_pkt_t *pkt, knot_section_t section_id, + const knot_dname_t *sname) +{ + const knot_dname_t *encloser = NULL; + const knot_rrset_t *covering_next_nsec3 = NULL; + int ret = closest_encloser_proof(pkt, section_id, sname, + &encloser, NULL, &covering_next_nsec3); + if (ret != 0) { + return ret; + } + ret = covers_closest_encloser_wildcard(pkt, section_id, encloser); + if (ret != 0) { + /* OK, but NSEC3 for wildcard at encloser has opt-out; + * or error */ + return ret; + } + /* Closest encloser proof is OK and + * NSEC3 for wildcard has been found and optout flag is not set. + * Now check if NSEC3 that covers next closer name has opt-out. */ + return has_optout(covering_next_nsec3) ? + kr_error(KNOT_ERANGE) : kr_ok(); +} + +/** + * Search the packet section for a matching NSEC3 with nodata-proving bitmap. + * @param pkt Packet structure to be processed. + * @param section_id Packet section to be processed. + * @param sname Name to be checked. + * @param stype Type to be checked. + * @return 0 or error code. + * @note This does NOT check the opt-out case if type is DS; + * see RFC 5155 8.6. + */ +static int nodata_find(const knot_pkt_t *pkt, knot_section_t section_id, + const knot_dname_t *name, const uint16_t type) +{ + const knot_pktsection_t *sec = knot_pkt_section(pkt, section_id); + if (!sec || !name) { + return kr_error(EINVAL); + } + + for (unsigned i = 0; i < sec->count; ++i) { + const knot_rrset_t *nsec3 = knot_pkt_rr(sec, i); + /* Records causing any errors are simply skipped. */ + if (nsec3->type != KNOT_RRTYPE_NSEC3 + || matches_name(nsec3, name) != kr_ok()) { + continue; + /* LATER(optim.): we repeatedly recompute the hash of `name` */ + } + + const uint8_t *bm = knot_nsec3_bitmap(nsec3->rrs.rdata); + uint16_t bm_size = knot_nsec3_bitmap_len(nsec3->rrs.rdata); + if (kr_nsec_bitmap_nodata_check(bm, bm_size, type, nsec3->owner) == kr_ok()) { + return kr_ok(); + } + } + + return kr_error(ENOENT); +} + +/** + * Check whether NSEC3 RR matches a wildcard at the closest encloser and has given type bit missing. + * @param pkt Packet structure to be processed. + * @param section_id Packet section to be processed. + * @param encloser Closest (provable) encloser domain name. + * @param stype Type to be checked. + * @return 0 or error code. + */ +static int matches_closest_encloser_wildcard(const knot_pkt_t *pkt, knot_section_t section_id, + const knot_dname_t *encloser, uint16_t stype) +{ + const knot_pktsection_t *sec = knot_pkt_section(pkt, section_id); + if (!sec || !encloser) { + return kr_error(EINVAL); + } + + uint8_t wildcard[KNOT_DNAME_MAXLEN]; /**< the source of synthesis */ + int ret = prepend_asterisk(wildcard, sizeof(wildcard), encloser); + if (ret < 0) { + return ret; + } + assert(ret >= 3); + return nodata_find(pkt, section_id, wildcard, stype); +} + +int kr_nsec3_wildcard_answer_response_check(const knot_pkt_t *pkt, knot_section_t section_id, + const knot_dname_t *sname, int trim_to_next) +{ + const knot_pktsection_t *sec = knot_pkt_section(pkt, section_id); + if (!sec || !sname) { + return kr_error(EINVAL); + } + + /* Compute the next closer name. */ + for (int i = 0; i < trim_to_next; ++i) { + assert(sname[0]); + sname = knot_wire_next_label(sname, NULL); + } + + int flags = 0; + for (unsigned i = 0; i < sec->count; ++i) { + const knot_rrset_t *rrset = knot_pkt_rr(sec, i); + if (rrset->type != KNOT_RRTYPE_NSEC3) { + continue; + } + int ret = covers_name(&flags, rrset, sname); + if (ret != 0) { + return ret; + } + if (flags & FLG_NAME_COVERED) { + return has_optout(rrset) ? + kr_error(KNOT_ERANGE) : kr_ok(); + } + } + + return kr_error(ENOENT); +} + + +int kr_nsec3_no_data(const knot_pkt_t *pkt, knot_section_t section_id, + const knot_dname_t *sname, uint16_t stype) +{ + /* DS record may be also matched by an existing NSEC3 RR. */ + int ret = nodata_find(pkt, section_id, sname, stype); + if (ret == 0) { + /* Satisfies RFC5155 8.5 and 8.6, both first paragraph. */ + return ret; + } + + /* Find closest provable encloser. */ + const knot_dname_t *encloser_name = NULL; + const knot_rrset_t *covering_next_nsec3 = NULL; + ret = closest_encloser_proof(pkt, section_id, sname, &encloser_name, + NULL, &covering_next_nsec3); + if (ret != 0) { + return ret; + } + + assert(encloser_name && covering_next_nsec3); + ret = matches_closest_encloser_wildcard(pkt, section_id, + encloser_name, stype); + if (ret == 0) { + /* Satisfies RFC5155 8.7 */ + if (has_optout(covering_next_nsec3)) { + /* Opt-out is detected. + * Despite the fact that all records + * in the packet can be properly signed, + * AD bit must not be set due to rfc5155 9.2. + * Return appropriate code to the caller */ + ret = kr_error(KNOT_ERANGE); + } + return ret; + } + + if (!has_optout(covering_next_nsec3)) { + /* Bogus */ + ret = kr_error(ENOENT); + } else { + /* + * Satisfies RFC5155 8.6 (QTYPE == DS), 2nd paragraph. + * Also satisfies ERRATA 3441 8.5 (QTYPE != DS), 3rd paragraph. + * - (wildcard) empty nonterminal + * derived from unsecure delegation. + * Denial of existence can not be proven. + * Set error code to proceed unsecure. + */ + ret = kr_error(KNOT_ERANGE); + } + + return ret; +} + +int kr_nsec3_ref_to_unsigned(const knot_pkt_t *pkt) +{ + const knot_pktsection_t *sec = knot_pkt_section(pkt, KNOT_AUTHORITY); + if (!sec) { + return kr_error(EINVAL); + } + for (unsigned i = 0; i < sec->count; ++i) { + const knot_rrset_t *ns = knot_pkt_rr(sec, i); + if (ns->type == KNOT_RRTYPE_DS) { + return kr_error(EEXIST); + } + if (ns->type != KNOT_RRTYPE_NS) { + continue; + } + + int flags = 0; + bool nsec3_found = false; + for (unsigned j = 0; j < sec->count; ++j) { + const knot_rrset_t *nsec3 = knot_pkt_rr(sec, j); + if (nsec3->type == KNOT_RRTYPE_DS) { + return kr_error(EEXIST); + } + if (nsec3->type != KNOT_RRTYPE_NSEC3) { + continue; + } + nsec3_found = true; + /* nsec3 found, check if owner name matches the delegation name. + * Just skip in case of *any* errors. */ + if (matches_name(nsec3, ns->owner) != kr_ok()) { + continue; + } + + const uint8_t *bm = knot_nsec3_bitmap(nsec3->rrs.rdata); + uint16_t bm_size = knot_nsec3_bitmap_len(nsec3->rrs.rdata); + if (!bm) { + return kr_error(EINVAL); + } + if (dnssec_nsec_bitmap_contains(bm, bm_size, + KNOT_RRTYPE_NS) && + !dnssec_nsec_bitmap_contains(bm, bm_size, + KNOT_RRTYPE_DS) && + !dnssec_nsec_bitmap_contains(bm, bm_size, + KNOT_RRTYPE_SOA)) { + /* Satisfies rfc5155, 8.9. paragraph 2 */ + return kr_ok(); + } + } + if (!nsec3_found) { + return kr_error(DNSSEC_NOT_FOUND); + } + if (flags & FLG_NAME_MATCHED) { + /* nsec3 which owner matches + * the delegation name was found, + * but nsec3 type bitmap contains wrong types + */ + return kr_error(EINVAL); + } + /* nsec3 that matches the delegation was not found. + * Check rfc5155, 8.9. paragraph 4. + * Find closest provable encloser. + */ + const knot_dname_t *encloser_name = NULL; + const knot_rrset_t *covering_next_nsec3 = NULL; + int ret = closest_encloser_proof(pkt, KNOT_AUTHORITY, ns->owner, + &encloser_name, NULL, &covering_next_nsec3); + if (ret != 0) { + return kr_error(EINVAL); + } + + if (has_optout(covering_next_nsec3)) { + return kr_error(KNOT_ERANGE); + } else { + return kr_error(EINVAL); + } + } + return kr_error(EINVAL); +} + +int kr_nsec3_matches_name_and_type(const knot_rrset_t *nsec3, + const knot_dname_t *name, uint16_t type) +{ + /* It's not secure enough to just check a single bit for (some) other types, + * but we don't (currently) only use this API for NS. See RFC 6840 sec. 4. + */ + if (type != KNOT_RRTYPE_NS) { + assert(!EINVAL); + return kr_error(EINVAL); + } + int ret = matches_name(nsec3, name); + if (ret) { + return kr_error(ret); + } + const uint8_t *bm = knot_nsec3_bitmap(nsec3->rrs.rdata); + uint16_t bm_size = knot_nsec3_bitmap_len(nsec3->rrs.rdata); + if (!bm) { + return kr_error(EINVAL); + } + if (dnssec_nsec_bitmap_contains(bm, bm_size, type)) { + return kr_ok(); + } else { + return kr_error(ENOENT); + } +} diff --git a/lib/dnssec/nsec3.h b/lib/dnssec/nsec3.h new file mode 100644 index 0000000..527ccce --- /dev/null +++ b/lib/dnssec/nsec3.h @@ -0,0 +1,82 @@ +/* Copyright (C) 2015-2017 CZ.NIC, z.s.p.o. + + This program is free software: you can redistribute it and/or modify + it under the terms of the GNU General Public License as published by + the Free Software Foundation, either version 3 of the License, or + (at your option) any later version. + + This program is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the + GNU General Public License for more details. + + You should have received a copy of the GNU General Public License + along with this program. If not, see . + */ + +#pragma once + +#include + +/** + * Name error response check (RFC5155 7.2.2). + * @note No RRSIGs are validated. + * @param pkt Packet structure to be processed. + * @param section_id Packet section to be processed. + * @param sname Name to be checked. + * @return 0 or error code. + */ +int kr_nsec3_name_error_response_check(const knot_pkt_t *pkt, knot_section_t section_id, + const knot_dname_t *sname); + +/** + * Wildcard answer response check (RFC5155 7.2.6). + * @param pkt Packet structure to be processed. + * @param section_id Packet section to be processed. + * @param sname Name to be checked. + * @param trim_to_next Number of labels to remove to obtain next closer name. + * @return 0 or error code: + * KNOT_ERANGE - NSEC3 RR that covers a wildcard + * has been found, but has opt-out flag set; + * otherwise - error. + */ +int kr_nsec3_wildcard_answer_response_check(const knot_pkt_t *pkt, knot_section_t section_id, + const knot_dname_t *sname, int trim_to_next); + +/** + * Authenticated denial of existence according to RFC5155 8.5 and 8.7. + * @note No RRSIGs are validated. + * @param pkt Packet structure to be processed. + * @param section_id Packet section to be processed. + * @param sname Queried domain name. + * @param stype Queried type. + * @return 0 or error code: + * DNSSEC_NOT_FOUND - neither ds nor nsec records + * were not found. + * KNOT_ERANGE - denial of existence can't be proven + * due to opt-out, otherwise - bogus. + */ +int kr_nsec3_no_data(const knot_pkt_t *pkt, knot_section_t section_id, + const knot_dname_t *sname, uint16_t stype); + +/** + * Referral to unsigned subzone check (RFC5155 8.9). + * @note No RRSIGs are validated. + * @param pkt Packet structure to be processed. + * @return 0 or error code: + * KNOT_ERANGE - denial of existence can't be proven + * due to opt-out. + * EEXIST - ds record was found. + * EINVAL - bogus. + */ +int kr_nsec3_ref_to_unsigned(const knot_pkt_t *pkt); + +/** + * Checks whether supplied NSEC3 RR matches the supplied name and NS type. + * @param nsec3 NSEC3 RR. + * @param name Name to be checked. + * @param type Type to be checked. Only use with NS! TODO + * @return 0 or error code. + */ +int kr_nsec3_matches_name_and_type(const knot_rrset_t *nsec3, + const knot_dname_t *name, uint16_t type); diff --git a/lib/dnssec/signature.c b/lib/dnssec/signature.c new file mode 100644 index 0000000..356bf14 --- /dev/null +++ b/lib/dnssec/signature.c @@ -0,0 +1,307 @@ +/* Copyright (C) 2015-2017 CZ.NIC, z.s.p.o. + + This program is free software: you can redistribute it and/or modify + it under the terms of the GNU General Public License as published by + the Free Software Foundation, either version 3 of the License, or + (at your option) any later version. + + This program is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the + GNU General Public License for more details. + + You should have received a copy of the GNU General Public License + along with this program. If not, see . + */ + +#include +#include +#include + +#include +#include +#include +#include +#include +#include +#include +#include +#include + +#include "lib/defines.h" +#include "lib/utils.h" +#include "lib/dnssec/signature.h" + +#include "contrib/wire.h" + +static int authenticate_ds(const dnssec_key_t *key, dnssec_binary_t *ds_rdata, uint8_t digest_type) +{ + /* Compute DS RDATA from the DNSKEY. */ + dnssec_binary_t computed_ds = { 0, }; + int ret = dnssec_key_create_ds(key, digest_type, &computed_ds); + if (ret != DNSSEC_EOK) { + goto fail; + } + + /* DS records contain algorithm, key tag and the digest. + * Therefore the comparison of the two DS is sufficient. + */ + ret = (ds_rdata->size == computed_ds.size) && + (memcmp(ds_rdata->data, computed_ds.data, ds_rdata->size) == 0); + ret = ret ? kr_ok() : kr_error(ENOENT); + +fail: + dnssec_binary_free(&computed_ds); + return kr_error(ret); +} + +int kr_authenticate_referral(const knot_rrset_t *ref, const dnssec_key_t *key) +{ + assert(ref && key); + if (ref->type != KNOT_RRTYPE_DS) { + return kr_error(EINVAL); + } + + /* Try all possible DS records */ + int ret = 0; + knot_rdata_t *rd = ref->rrs.rdata; + for (uint16_t i = 0; i < ref->rrs.count; ++i) { + dnssec_binary_t ds_rdata = { + .size = rd->len, + .data = rd->data + }; + ret = authenticate_ds(key, &ds_rdata, knot_ds_digest_type(rd)); + if (ret == 0) { /* Found a good DS */ + return kr_ok(); + } + rd = knot_rdataset_next(rd); + } + + return kr_error(ret); +} + +/** + * Adjust TTL in wire format. + * @param wire RR Set in wire format. + * @param wire_size Size of the wire data portion. + * @param new_ttl TTL value to be set for all RRs. + * @return 0 or error code. + */ +static int adjust_wire_ttl(uint8_t *wire, size_t wire_size, uint32_t new_ttl) +{ + assert(wire); + assert(sizeof(uint16_t) == 2); + assert(sizeof(uint32_t) == 4); + uint16_t rdlen; + + int ret; + + new_ttl = htonl(new_ttl); + + size_t i = 0; + /* RR wire format in RFC1035 3.2.1 */ + while(i < wire_size) { + ret = knot_dname_size(wire + i); + if (ret < 0) { + return ret; + } + i += ret + 4; + memcpy(wire + i, &new_ttl, sizeof(uint32_t)); + i += sizeof(uint32_t); + + memcpy(&rdlen, wire + i, sizeof(uint16_t)); + rdlen = ntohs(rdlen); + i += sizeof(uint16_t) + rdlen; + + assert(i <= wire_size); + } + + return kr_ok(); +} + +/*! + * \brief Add RRSIG RDATA without signature to signing context. + * + * Requires signer name in RDATA in canonical form. + * + * \param ctx Signing context. + * \param rdata Pointer to RRSIG RDATA. + * + * \return Error code, KNOT_EOK if successful. + */ +#define RRSIG_RDATA_SIGNER_OFFSET 18 +static int sign_ctx_add_self(dnssec_sign_ctx_t *ctx, const uint8_t *rdata) +{ + assert(ctx); + assert(rdata); + + int result; + + // static header + + dnssec_binary_t header = { + .data = (uint8_t *)rdata, + .size = RRSIG_RDATA_SIGNER_OFFSET, + }; + + result = dnssec_sign_add(ctx, &header); + if (result != DNSSEC_EOK) { + return result; + } + + // signer name + + const uint8_t *rdata_signer = rdata + RRSIG_RDATA_SIGNER_OFFSET; + dnssec_binary_t signer = { 0 }; + signer.data = knot_dname_copy(rdata_signer, NULL); + signer.size = knot_dname_size(signer.data); + + result = dnssec_sign_add(ctx, &signer); + free(signer.data); + + return result; +} +#undef RRSIG_RDATA_SIGNER_OFFSET + +/*! + * \brief Add covered RRs to signing context. + * + * Requires all DNAMEs in canonical form and all RRs ordered canonically. + * + * \param ctx Signing context. + * \param covered Covered RRs. + * + * \return Error code, KNOT_EOK if successful. + */ +static int sign_ctx_add_records(dnssec_sign_ctx_t *ctx, const knot_rrset_t *covered, + uint32_t orig_ttl, int trim_labels) +{ + if (!ctx || !covered || trim_labels < 0) { + return kr_error(EINVAL); + } + + // huge block of rrsets can be optionally created + static uint8_t wire_buffer[KNOT_WIRE_MAX_PKTSIZE]; + int written = knot_rrset_to_wire(covered, wire_buffer, sizeof(wire_buffer), NULL); + if (written < 0) { + return written; + } + + /* Set original ttl. */ + int ret = adjust_wire_ttl(wire_buffer, written, orig_ttl); + if (ret != 0) { + return ret; + } + + if (!trim_labels) { + const dnssec_binary_t wire_binary = { + .size = written, + .data = wire_buffer + }; + return dnssec_sign_add(ctx, &wire_binary); + } + + /* RFC4035 5.3.2 + * Remove leftmost labels and replace them with '*.' + * for each RR in covered. + */ + uint8_t *beginp = wire_buffer; + for (uint16_t i = 0; i < covered->rrs.count; ++i) { + /* RR(i) = name | type | class | OrigTTL | RDATA length | RDATA */ + for (int j = 0; j < trim_labels; ++j) { + assert(beginp[0]); + beginp = (uint8_t *) knot_wire_next_label(beginp, NULL); + assert(beginp != NULL); + } + *(--beginp) = '*'; + *(--beginp) = 1; + const size_t rdatalen_offset = knot_dname_size(beginp) + /* name */ + sizeof(uint16_t) + /* type */ + sizeof(uint16_t) + /* class */ + sizeof(uint32_t); /* OrigTTL */ + const uint8_t *rdatalen_ptr = beginp + rdatalen_offset; + const uint16_t rdata_size = wire_read_u16(rdatalen_ptr); + const size_t rr_size = rdatalen_offset + + sizeof(uint16_t) + /* RDATA length */ + rdata_size; /* RDATA */ + const dnssec_binary_t wire_binary = { + .size = rr_size, + .data = beginp + }; + ret = dnssec_sign_add(ctx, &wire_binary); + if (ret != 0) { + break; + } + beginp += rr_size; + } + return ret; +} + +/*! + * \brief Add all data covered by signature into signing context. + * + * RFC 4034: The signature covers RRSIG RDATA field (excluding the signature) + * and all matching RR records, which are ordered canonically. + * + * Requires all DNAMEs in canonical form and all RRs ordered canonically. + * + * \param ctx Signing context. + * \param rrsig_rdata RRSIG RDATA with populated fields except signature. + * \param covered Covered RRs. + * + * \return Error code, KNOT_EOK if successful. + */ +/* TODO -- Taken from knot/src/knot/dnssec/rrset-sign.c. Re-write for better fit needed. */ +static int sign_ctx_add_data(dnssec_sign_ctx_t *ctx, const uint8_t *rrsig_rdata, + const knot_rrset_t *covered, uint32_t orig_ttl, int trim_labels) +{ + int result = sign_ctx_add_self(ctx, rrsig_rdata); + if (result != KNOT_EOK) { + return result; + } + + return sign_ctx_add_records(ctx, covered, orig_ttl, trim_labels); +} + +int kr_check_signature(const knot_rdata_t *rrsig, + const dnssec_key_t *key, const knot_rrset_t *covered, + int trim_labels) +{ + if (!rrsig || !key || !dnssec_key_can_verify(key)) { + return kr_error(EINVAL); + } + + int ret = 0; + dnssec_sign_ctx_t *sign_ctx = NULL; + dnssec_binary_t signature = { + .data = /*const-cast*/(uint8_t*)knot_rrsig_signature(rrsig), + .size = knot_rrsig_signature_len(rrsig), + }; + if (!signature.data || !signature.size) { + ret = kr_error(EINVAL); + goto fail; + } + + if (dnssec_sign_new(&sign_ctx, key) != 0) { + ret = kr_error(ENOMEM); + goto fail; + } + + uint32_t orig_ttl = knot_rrsig_original_ttl(rrsig); + + if (sign_ctx_add_data(sign_ctx, rrsig->data, covered, orig_ttl, trim_labels) != 0) { + ret = kr_error(ENOMEM); + goto fail; + } + + if (dnssec_sign_verify(sign_ctx, &signature) != 0) { + ret = kr_error(EBADMSG); + goto fail; + } + + ret = kr_ok(); + +fail: + dnssec_sign_free(sign_ctx); + return ret; +} diff --git a/lib/dnssec/signature.h b/lib/dnssec/signature.h new file mode 100644 index 0000000..b756483 --- /dev/null +++ b/lib/dnssec/signature.h @@ -0,0 +1,42 @@ +/* Copyright (C) 2015-2017 CZ.NIC, z.s.p.o. + + This program is free software: you can redistribute it and/or modify + it under the terms of the GNU General Public License as published by + the Free Software Foundation, either version 3 of the License, or + (at your option) any later version. + + This program is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the + GNU General Public License for more details. + + You should have received a copy of the GNU General Public License + along with this program. If not, see . + */ + +#pragma once + +#include +#include + +/** + * Performs referral authentication according to RFC4035 5.2, bullet 2 + * @param ref Referral RRSet. Currently only DS can be used. + * @param key Already parsed key. + * @return 0 or error code. In particular: DNSSEC_INVALID_DS_ALGORITHM + * in case *all* DSs in ref use an unimplemented algorithm. + */ +int kr_authenticate_referral(const knot_rrset_t *ref, const dnssec_key_t *key); + +/** + * Check the signature of the supplied RRSet. + * @param rrsigs RRSet containing signatures. + * @param pos Index of the signature record in the signature RRSet. + * @param key Key to be used to validate the signature. + * @param covered The covered RRSet. + * @param trim_labels Number of the leftmost labels to be removed and replaced with '*.'. + * @return 0 if signature valid, error code else. + */ +int kr_check_signature(const knot_rdata_t *rrsig, + const dnssec_key_t *key, const knot_rrset_t *covered, + int trim_labels); diff --git a/lib/dnssec/ta.c b/lib/dnssec/ta.c new file mode 100644 index 0000000..12629c9 --- /dev/null +++ b/lib/dnssec/ta.c @@ -0,0 +1,185 @@ +/* Copyright (C) 2014-2017 CZ.NIC, z.s.p.o. + + This program is free software: you can redistribute it and/or modify + it under the terms of the GNU General Public License as published by + the Free Software Foundation, either version 3 of the License, or + (at your option) any later version. + + This program is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the + GNU General Public License for more details. + + You should have received a copy of the GNU General Public License + along with this program. If not, see . + */ + +#include +#include +#include +#include +#include +#include +#include + +#include "lib/defines.h" +#include "lib/dnssec.h" +#include "lib/dnssec/ta.h" +#include "lib/resolve.h" +#include "lib/utils.h" + +knot_rrset_t *kr_ta_get(map_t *trust_anchors, const knot_dname_t *name) +{ + return map_get(trust_anchors, (const char *)name); +} + +const knot_dname_t *kr_ta_get_longest_name(map_t *trust_anchors, const knot_dname_t *name) +{ + while(name) { + if (kr_ta_get(trust_anchors, name)) { + return name; + } + if (name[0] == '\0') { + break; + } + name = knot_wire_next_label(name, NULL); + } + return NULL; +} + +/* @internal Create DS from DNSKEY, caller MUST free dst if successful. */ +static int dnskey2ds(dnssec_binary_t *dst, const knot_dname_t *owner, const uint8_t *rdata, uint16_t rdlen) +{ + dnssec_key_t *key = NULL; + int ret = dnssec_key_new(&key); + if (ret) goto cleanup; + /* Create DS from DNSKEY and reinsert */ + const dnssec_binary_t key_data = { .size = rdlen, .data = (uint8_t *)rdata }; + ret = dnssec_key_set_rdata(key, &key_data); + if (ret) goto cleanup; + /* Accept only keys with Zone and SEP flags that aren't revoked, + * as a precaution. RFC 5011 also utilizes these flags. + * TODO: kr_dnssec_key_* names are confusing. */ + const bool flags_ok = kr_dnssec_key_zsk(rdata) && !kr_dnssec_key_revoked(rdata); + if (!flags_ok) { + auto_free char *owner_str = kr_dname_text(owner); + kr_log_error("[ ta ] refusing to trust %s DNSKEY because of flags %d\n", + owner_str, dnssec_key_get_flags(key)); + ret = kr_error(EILSEQ); + goto cleanup; + } else if (!kr_dnssec_key_ksk(rdata)) { + auto_free char *owner_str = kr_dname_text(owner); + int flags = dnssec_key_get_flags(key); + kr_log_info("[ ta ] warning: %s DNSKEY is missing the SEP bit; " + "flags %d instead of %d\n", + owner_str, flags, flags + 1/*a little ugly*/); + } + ret = dnssec_key_set_dname(key, owner); + if (ret) goto cleanup; + ret = dnssec_key_create_ds(key, DNSSEC_KEY_DIGEST_SHA256, dst); +cleanup: + dnssec_key_free(key); + return kr_error(ret); +} + +/* @internal Insert new TA to trust anchor set, rdata MUST be of DS type. */ +static int insert_ta(map_t *trust_anchors, const knot_dname_t *name, + uint32_t ttl, const uint8_t *rdata, uint16_t rdlen) +{ + bool is_new_key = false; + knot_rrset_t *ta_rr = kr_ta_get(trust_anchors, name); + if (!ta_rr) { + ta_rr = knot_rrset_new(name, KNOT_RRTYPE_DS, KNOT_CLASS_IN, ttl, NULL); + is_new_key = true; + } + /* Merge-in new key data */ + if (!ta_rr || (rdlen > 0 && knot_rrset_add_rdata(ta_rr, rdata, rdlen, NULL) != 0)) { + knot_rrset_free(ta_rr, NULL); + return kr_error(ENOMEM); + } + if (is_new_key) { + return map_set(trust_anchors, (const char *)name, ta_rr); + } + return kr_ok(); +} + +int kr_ta_add(map_t *trust_anchors, const knot_dname_t *name, uint16_t type, + uint32_t ttl, const uint8_t *rdata, uint16_t rdlen) +{ + if (!trust_anchors || !name) { + return kr_error(EINVAL); + } + + /* DS/DNSEY types are accepted, for DNSKEY we + * need to compute a DS digest. */ + if (type == KNOT_RRTYPE_DS) { + return insert_ta(trust_anchors, name, ttl, rdata, rdlen); + } else if (type == KNOT_RRTYPE_DNSKEY) { + dnssec_binary_t ds_rdata = { 0, }; + int ret = dnskey2ds(&ds_rdata, name, rdata, rdlen); + if (ret != 0) { + return ret; + } + ret = insert_ta(trust_anchors, name, ttl, ds_rdata.data, ds_rdata.size); + dnssec_binary_free(&ds_rdata); + return ret; + } else { /* Invalid type for TA */ + return kr_error(EINVAL); + } +} + +int kr_ta_covers(map_t *trust_anchors, const knot_dname_t *name) +{ + while(name) { + if (kr_ta_get(trust_anchors, name)) { + return true; + } + if (name[0] == '\0') { + return false; + } + name = knot_wire_next_label(name, NULL); + } + return false; +} + +bool kr_ta_covers_qry(struct kr_context *ctx, const knot_dname_t *name, + const uint16_t type) +{ + assert(ctx && name); + if (type == KNOT_RRTYPE_DS && name[0] != '\0') { + /* DS is parent-side record, so the parent name needs to be covered. */ + name = knot_wire_next_label(name, NULL); + if (!name) { + assert(false); + return false; + } + } + return kr_ta_covers(&ctx->trust_anchors, name) + && !kr_ta_covers(&ctx->negative_anchors, name); +} + +/* Delete record data */ +static int del_record(const char *k, void *v, void *ext) +{ + knot_rrset_t *ta_rr = v; + if (ta_rr) { + knot_rrset_free(ta_rr, NULL); + } + return 0; +} + +int kr_ta_del(map_t *trust_anchors, const knot_dname_t *name) +{ + knot_rrset_t *ta_rr = kr_ta_get(trust_anchors, name); + if (ta_rr) { + del_record(NULL, ta_rr, NULL); + map_del(trust_anchors, (const char *)name); + } + return kr_ok(); +} + +void kr_ta_clear(map_t *trust_anchors) +{ + map_walk(trust_anchors, del_record, NULL); + map_clear(trust_anchors); +} diff --git a/lib/dnssec/ta.h b/lib/dnssec/ta.h new file mode 100644 index 0000000..dcc55e7 --- /dev/null +++ b/lib/dnssec/ta.h @@ -0,0 +1,87 @@ +/* Copyright (C) 2015-2017 CZ.NIC, z.s.p.o. + + This program is free software: you can redistribute it and/or modify + it under the terms of the GNU General Public License as published by + the Free Software Foundation, either version 3 of the License, or + (at your option) any later version. + + This program is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the + GNU General Public License for more details. + + You should have received a copy of the GNU General Public License + along with this program. If not, see . + */ + +#pragma once + +#include "lib/generic/map.h" +#include + +/** + * Find TA RRSet by name. + * @param trust_anchors trust store + * @param name name of the TA + * @return non-empty RRSet or NULL + */ +KR_EXPORT +knot_rrset_t *kr_ta_get(map_t *trust_anchors, const knot_dname_t *name); + +/** + * Add TA to trust store. DS or DNSKEY types are supported. + * @param trust_anchors trust store + * @param name name of the TA + * @param type RR type of the TA (DS or DNSKEY) + * @param ttl + * @param rdata + * @param rdlen + * @return 0 or an error + */ +KR_EXPORT +int kr_ta_add(map_t *trust_anchors, const knot_dname_t *name, uint16_t type, + uint32_t ttl, const uint8_t *rdata, uint16_t rdlen); + +/** + * Return true if the name is below/at any TA in the store. + * This can be useful to check if it's possible to validate a name beforehand. + * @param trust_anchors trust store + * @param name name of the TA + * @return boolean + */ +KR_EXPORT KR_PURE +int kr_ta_covers(map_t *trust_anchors, const knot_dname_t *name); + +struct kr_context; +/** + * A wrapper around kr_ta_covers that is aware of negative TA and types. + */ +KR_EXPORT KR_PURE +bool kr_ta_covers_qry(struct kr_context *ctx, const knot_dname_t *name, + const uint16_t type); + +/** + * Remove TA from trust store. + * @param trust_anchors trust store + * @param name name of the TA + * @return 0 or an error + */ +KR_EXPORT +int kr_ta_del(map_t *trust_anchors, const knot_dname_t *name); + +/** + * Clear trust store. + * @param trust_anchors trust store + */ +KR_EXPORT +void kr_ta_clear(map_t *trust_anchors); + +/** + * Return TA with the longest name that covers given name. + * @param trust_anchors trust store + * @param name name of the TA + * @return pointer to name or NULL. + if not NULL, points inside the name parameter. + */ +KR_EXPORT +const knot_dname_t *kr_ta_get_longest_name(map_t *trust_anchors, const knot_dname_t *name); diff --git a/lib/generic/README.rst b/lib/generic/README.rst new file mode 100644 index 0000000..7adff86 --- /dev/null +++ b/lib/generic/README.rst @@ -0,0 +1,60 @@ +Generics library +---------------- + +This small collection of "generics" was born out of frustration that I couldn't find no +such thing for C. It's either bloated, has poor interface, null-checking is absent or +doesn't allow custom allocation scheme. BSD-licensed (or compatible) code is allowed here, +as long as it comes with a test case in `tests/test_generics.c`. + +* array_ - a set of simple macros to make working with dynamic arrays easier. +* queue_ - a FIFO + LIFO queue. +* map_ - a `Crit-bit tree`_ key-value map implementation (public domain) that comes with tests. +* set_ - set abstraction implemented on top of ``map`` (unused now). +* pack_ - length-prefixed list of objects (i.e. array-list). +* lru_ - LRU-like hash table +* trie_ - a trie-based key-value map, taken from knot-dns + +array +~~~~~ + +.. doxygenfile:: array.h + :project: libkres + +queue +~~~~~ + +.. doxygenfile:: queue.h + :project: libkres + +map +~~~ + +.. doxygenfile:: map.h + :project: libkres + +set +~~~ + +.. doxygenfile:: set.h + :project: libkres + +pack +~~~~ + +.. doxygenfile:: pack.h + :project: libkres + +lru +~~~ + +.. doxygenfile:: lru.h + :project: libkres + +trie +~~~~ + +.. doxygenfile:: trie.h + :project: libkres + + +.. _`Crit-bit tree`: https://cr.yp.to/critbit.html diff --git a/lib/generic/array.h b/lib/generic/array.h new file mode 100644 index 0000000..ece4dd1 --- /dev/null +++ b/lib/generic/array.h @@ -0,0 +1,166 @@ +/* Copyright (C) 2015-2017 CZ.NIC, z.s.p.o. + + This program is free software: you can redistribute it and/or modify + it under the terms of the GNU General Public License as published by + the Free Software Foundation, either version 3 of the License, or + (at your option) any later version. + + This program is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the + GNU General Public License for more details. + + You should have received a copy of the GNU General Public License + along with this program. If not, see . + */ + +/** + * + * @file array.h + * @brief A set of simple macros to make working with dynamic arrays easier. + * + * @note The C has no generics, so it is implemented mostly using macros. + * Be aware of that, as direct usage of the macros in the evaluating macros + * may lead to different expectations: + * + * @code{.c} + * MIN(array_push(arr, val), other) + * @endcode + * + * May evaluate the code twice, leading to unexpected behaviour. + * This is a price to pay for the absence of proper generics. + * + * # Example usage: + * + * @code{.c} + * array_t(const char*) arr; + * array_init(arr); + * + * // Reserve memory in advance + * if (array_reserve(arr, 2) < 0) { + * return ENOMEM; + * } + * + * // Already reserved, cannot fail + * array_push(arr, "princess"); + * array_push(arr, "leia"); + * + * // Not reserved, may fail + * if (array_push(arr, "han") < 0) { + * return ENOMEM; + * } + * + * // It does not hide what it really is + * for (size_t i = 0; i < arr.len; ++i) { + * printf("%s\n", arr.at[i]); + * } + * + * // Random delete + * array_del(arr, 0); + * @endcode + * \addtogroup generics + * @{ + */ + +#pragma once +#include + +/** Simplified Qt containers growth strategy. */ +static inline size_t array_next_count(size_t want) +{ + if (want < 2048) { + return (want < 20) ? want + 4 : want * 2; + } else { + return want + 2048; + } +} + +/** @internal Incremental memory reservation */ +static inline int array_std_reserve(void *baton, char **mem, size_t elm_size, size_t want, size_t *have) +{ + if (*have >= want) { + return 0; + } + /* Simplified Qt containers growth strategy */ + size_t next_size = array_next_count(want); + void *mem_new = realloc(*mem, next_size * elm_size); + if (mem_new != NULL) { + *mem = mem_new; + *have = next_size; + return 0; + } + return -1; +} + +/** @internal Wrapper for stdlib free. */ +static inline void array_std_free(void *baton, void *p) +{ + free(p); +} + +/** Declare an array structure. */ +#define array_t(type) struct {type * at; size_t len; size_t cap; } + +/** Zero-initialize the array. */ +#define array_init(array) ((array).at = NULL, (array).len = (array).cap = 0) + +/** Free and zero-initialize the array (plain malloc/free). */ +#define array_clear(array) \ + array_clear_mm(array, array_std_free, NULL) + +/** Make the array empty and free pointed-to memory. + * Mempool usage: pass mm_free and a knot_mm_t* . */ +#define array_clear_mm(array, free, baton) \ + (free)((baton), (array).at), array_init(array) + +/** Reserve capacity for at least n elements. + * @return 0 if success, <0 on failure */ +#define array_reserve(array, n) \ + array_reserve_mm(array, n, array_std_reserve, NULL) + +/** Reserve capacity for at least n elements. + * Mempool usage: pass kr_memreserve and a knot_mm_t* . + * @return 0 if success, <0 on failure */ +#define array_reserve_mm(array, n, reserve, baton) \ + (reserve)((baton), (char **) &(array).at, sizeof((array).at[0]), (n), &(array).cap) + +/** + * Push value at the end of the array, resize it if necessary. + * Mempool usage: pass kr_memreserve and a knot_mm_t* . + * @note May fail if the capacity is not reserved. + * @return element index on success, <0 on failure + */ +#define array_push_mm(array, val, reserve, baton) \ + (int)((array).len < (array).cap ? ((array).at[(array).len] = val, (array).len++) \ + : (array_reserve_mm(array, ((array).cap + 1), reserve, baton) < 0 ? -1 \ + : ((array).at[(array).len] = val, (array).len++))) + +/** + * Push value at the end of the array, resize it if necessary (plain malloc/free). + * @note May fail if the capacity is not reserved. + * @return element index on success, <0 on failure + */ +#define array_push(array, val) \ + array_push_mm(array, val, array_std_reserve, NULL) + +/** + * Pop value from the end of the array. + */ +#define array_pop(array) \ + (array).len -= 1 + +/** + * Remove value at given index. + * @return 0 on success, <0 on failure + */ +#define array_del(array, i) \ + (int)((i) < (array).len ? ((array).len -= 1,(array).at[i] = (array).at[(array).len], 0) : -1) + +/** + * Return last element of the array. + * @warning Undefined if the array is empty. + */ +#define array_tail(array) \ + (array).at[(array).len - 1] + +/** @} */ diff --git a/lib/generic/lru.c b/lib/generic/lru.c new file mode 100644 index 0000000..12bba4e --- /dev/null +++ b/lib/generic/lru.c @@ -0,0 +1,236 @@ +/* Copyright (C) 2016-2018 CZ.NIC, z.s.p.o. + + This program is free software: you can redistribute it and/or modify + it under the terms of the GNU General Public License as published by + the Free Software Foundation, either version 3 of the License, or + (at your option) any later version. + + This program is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the + GNU General Public License for more details. + + You should have received a copy of the GNU General Public License + along with this program. If not, see . + */ + +#include "lib/generic/lru.h" +#include "contrib/murmurhash3/murmurhash3.h" + +typedef struct lru_group lru_group_t; + +struct lru_item { + uint16_t key_len, val_len; /**< Two bytes should be enough for our purposes. */ + char data[]; + /**< Place for both key and value. + * + * We use "char" to satisfy the C99+ aliasing rules. + * See C99 section 6.5 Expressions, paragraph 7. + * Any type can be accessed through char-pointer, + * so we can use a common struct definition + * for all types being held. + */ +}; + +/** @internal Compute offset of value in struct lru_item. */ +static uint val_offset(uint key_len) +{ + uint key_end = offsetof(struct lru_item, data) + key_len; + // align it to the closest multiple of four + return round_power(key_end, 2); +} + +/** @internal Return pointer to value in an item. */ +static void * item_val(struct lru_item *it) +{ + return it->data + val_offset(it->key_len) - offsetof(struct lru_item, data); +} + +/** @internal Compute the size of an item. ATM we don't align/pad the end of it. */ +static uint item_size(uint key_len, uint val_len) +{ + return val_offset(key_len) + val_len; +} + +/** @internal Free each item. */ +KR_EXPORT void lru_free_items_impl(struct lru *lru) +{ + assert(lru); + for (size_t i = 0; i < (1 << (size_t)lru->log_groups); ++i) { + lru_group_t *g = &lru->groups[i]; + for (int j = 0; j < LRU_ASSOC; ++j) + mm_free(lru->mm, g->items[j]); + } +} + +/** @internal See lru_apply. */ +KR_EXPORT void lru_apply_impl(struct lru *lru, lru_apply_fun f, void *baton) +{ + bool ok = lru && f; + if (!ok) { + assert(false); + return; + } + for (size_t i = 0; i < (1 << (size_t)lru->log_groups); ++i) { + lru_group_t *g = &lru->groups[i]; + for (uint j = 0; j < LRU_ASSOC; ++j) { + struct lru_item *it = g->items[j]; + if (!it) + continue; + enum lru_apply_do ret = + f(it->data, it->key_len, item_val(it), baton); + switch(ret) { + case LRU_APPLY_DO_EVICT: // evict + mm_free(lru->mm, it); + g->items[j] = NULL; + g->counts[j] = 0; + g->hashes[j] = 0; + break; + default: + assert(ret == LRU_APPLY_DO_NOTHING); + } + } + } +} + +/** @internal See lru_create. */ +KR_EXPORT struct lru * lru_create_impl(uint max_slots, knot_mm_t *mm_array, knot_mm_t *mm) +{ + assert(max_slots); + if (!max_slots) + return NULL; + // let lru->log_groups = ceil(log2(max_slots / (float) assoc)) + // without trying for efficiency + uint group_count = (max_slots - 1) / LRU_ASSOC + 1; + uint log_groups = 0; + for (uint s = group_count - 1; s; s /= 2) + ++log_groups; + group_count = 1 << log_groups; + assert(max_slots <= group_count * LRU_ASSOC && group_count * LRU_ASSOC < 2 * max_slots); + + size_t size = offsetof(struct lru, groups[group_count]); + struct lru *lru = mm_alloc(mm_array, size); + if (unlikely(lru == NULL)) + return NULL; + *lru = (struct lru){ + .mm = mm, + .mm_array = mm_array, + .log_groups = log_groups, + }; + // zeros are a good init + memset(lru->groups, 0, size - offsetof(struct lru, groups)); + return lru; +} + +/** @internal Decrement all counters within a group. */ +static void group_dec_counts(lru_group_t *g) { + g->counts[LRU_TRACKED] = LRU_TRACKED; + for (uint i = 0; i < LRU_TRACKED + 1; ++i) + if (likely(g->counts[i])) + --g->counts[i]; +} + +/** @internal Increment a counter within a group. */ +static void group_inc_count(lru_group_t *g, int i) { + if (likely(++(g->counts[i]))) + return; + g->counts[i] = -1; + // We could've decreased or halved all of them, but let's keep the max. +} + +/** @internal Implementation of both getting and insertion. + * Note: val_len is only meaningful if do_insert. + * *is_new is only meaningful when return value isn't NULL, contains + * true when returned lru entry has been allocated right now + * if return value is NULL, *is_new remains untouched. + */ +KR_EXPORT void * lru_get_impl(struct lru *lru, const char *key, uint key_len, + uint val_len, bool do_insert, bool *is_new) +{ + bool ok = lru && (key || !key_len) && key_len <= UINT16_MAX + && (!do_insert || val_len <= UINT16_MAX); + if (!ok) { + assert(false); + return NULL; // reasonable fallback when not debugging + } + bool is_new_entry = false; + // find the right group + uint32_t khash = hash(key, key_len); + uint16_t khash_top = khash >> 16; + lru_group_t *g = &lru->groups[khash & ((1 << lru->log_groups) - 1)]; + struct lru_item *it = NULL; + uint i; + // scan the *stored* elements in the group + for (i = 0; i < LRU_ASSOC; ++i) { + if (g->hashes[i] == khash_top) { + it = g->items[i]; + if (likely(it && it->key_len == key_len + && (key_len == 0 || memcmp(it->data, key, key_len) == 0))) { + /* Found a key, but trying to insert a value larger than available + * space in the allocated slot, so the entry must be resized to fit. */ + if (unlikely(do_insert && val_len > it->val_len)) { + goto insert; + } else { + goto found; // to reduce huge nesting depth + } + } + } + } + // key not found; first try an empty/counted-out place to insert + if (do_insert) + for (i = 0; i < LRU_ASSOC; ++i) + if (g->items[i] == NULL || g->counts[i] == 0) + goto insert; + // check if we track key's count at least + for (i = LRU_ASSOC; i < LRU_TRACKED; ++i) { + if (g->hashes[i] == khash_top) { + group_inc_count(g, i); + if (!do_insert) + return NULL; + // check if we trumped some stored key + for (uint j = 0; j < LRU_ASSOC; ++j) + if (unlikely(g->counts[i] > g->counts[j])) { + // evict key j, i.e. swap with i + --g->counts[i]; // we increment it below + SWAP(g->counts[i], g->counts[j]); + SWAP(g->hashes[i], g->hashes[j]); + i = j; + goto insert; + } + return NULL; + } + } + // not found at all: decrement all counts but only on every LRU_TRACKED occasion + if (g->counts[LRU_TRACKED]) + --g->counts[LRU_TRACKED]; + else + group_dec_counts(g); + return NULL; +insert: // insert into position i (incl. key) + assert(i < LRU_ASSOC); + g->hashes[i] = khash_top; + it = g->items[i]; + uint new_size = item_size(key_len, val_len); + if (it == NULL || new_size != item_size(it->key_len, it->val_len)) { + // (re)allocate + mm_free(lru->mm, it); + it = g->items[i] = mm_alloc(lru->mm, new_size); + if (it == NULL) + return NULL; + } + it->key_len = key_len; + it->val_len = val_len; + if (key_len > 0) { + memcpy(it->data, key, key_len); + } + memset(item_val(it), 0, val_len); // clear the value + is_new_entry = true; +found: // key and hash OK on g->items[i]; now update stamps + assert(i < LRU_ASSOC); + group_inc_count(g, i); + if (is_new) { + *is_new = is_new_entry; + } + return item_val(g->items[i]); +} + diff --git a/lib/generic/lru.h b/lib/generic/lru.h new file mode 100644 index 0000000..b5c9bcd --- /dev/null +++ b/lib/generic/lru.h @@ -0,0 +1,249 @@ +/* Copyright (C) 2016-2018 CZ.NIC, z.s.p.o. + + This program is free software: you can redistribute it and/or modify + it under the terms of the GNU General Public License as published by + the Free Software Foundation, either version 3 of the License, or + (at your option) any later version. + + This program is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the + GNU General Public License for more details. + + You should have received a copy of the GNU General Public License + along with this program. If not, see . + */ +/** + * @file lru.h + * @brief A lossy cache. + * + * @note The implementation tries to keep frequent keys and avoid others, + * even if "used recently", so it may refuse to store it on lru_get_new(). + * It uses hashing to split the problem pseudo-randomly into smaller groups, + * and within each it tries to approximate relative usage counts of several + * most frequent keys/hashes. This tracking is done for *more* keys than + * those that are actually stored. + * + * Example usage: + * @code{.c} + * // Define new LRU type + * typedef lru_t(int) lru_int_t; + * + * // Create LRU + * lru_int_t *lru; + * lru_create(&lru, 5, NULL, NULL); + * + * // Insert some values + * int *pi = lru_get_new(lru, "luke", strlen("luke"), NULL); + * if (pi) + * *pi = 42; + * pi = lru_get_new(lru, "leia", strlen("leia"), NULL); + * if (pi) + * *pi = 24; + * + * // Retrieve values + * int *ret = lru_get_try(lru, "luke", strlen("luke"), NULL); + * if (!ret) printf("luke dropped out!\n"); + * else printf("luke's number is %d\n", *ret); + * + * char *enemies[] = {"goro", "raiden", "subzero", "scorpion"}; + * for (int i = 0; i < 4; ++i) { + * int *val = lru_get_new(lru, enemies[i], strlen(enemies[i]), NULL); + * if (val) + * *val = i; + * } + * + * // We're done + * lru_free(lru); + * @endcode + * + * \addtogroup generics + * @{ + */ + +#pragma once + +#include +#include +#include + +#include "contrib/ucw/lib.h" +#include "lib/utils.h" +#include "libknot/mm_ctx.h" + +/* ================================ Interface ================================ */ + +/** @brief The type for LRU, parametrized by value type. */ +#define lru_t(type) \ + union { \ + type *pdata_t; /* only the *type* information is used */ \ + struct lru lru; \ + } + +/** + * @brief Allocate and initialize an LRU with default associativity. + * + * The real limit on the number of slots can be a bit larger but less than double. + * + * @param ptable pointer to a pointer to the LRU + * @param max_slots number of slots + * @param mm_ctx_array memory context to use for the huge array, NULL for default + * @param mm_ctx memory context to use for individual key-value pairs, NULL for default + * + * @note The pointers to memory contexts need to remain valid + * during the whole life of the structure (or be NULL). + */ +#define lru_create(ptable, max_slots, mm_ctx_array, mm_ctx) do { \ + (void)(((__typeof__((*(ptable))->pdata_t))0) == (void *)0); /* typecheck lru_t */ \ + *(ptable) = (__typeof__(*(ptable))) \ + lru_create_impl((max_slots), (mm_ctx_array), (mm_ctx)); \ + } while (false) + +/** @brief Free an LRU created by lru_create (it can be NULL). */ +#define lru_free(table) \ + lru_free_impl(&(table)->lru) + +/** @brief Reset an LRU to the empty state (but preserve any settings). */ +#define lru_reset(table) \ + lru_reset_impl(&(table)->lru) + +/** + * @brief Find key in the LRU and return pointer to the corresponding value. + * + * @param table pointer to LRU + * @param key_ lookup key + * @param len_ key length + * @return pointer to data or NULL if not found + */ +#define lru_get_try(table, key_, len_) \ + (__typeof__((table)->pdata_t)) \ + lru_get_impl(&(table)->lru, (key_), (len_), -1, false, NULL) + +/** + * @brief Return pointer to value, inserting if needed (zeroed). + * + * @param table pointer to LRU + * @param key_ lookup key + * @param len_ key lengthkeys + * @param is_new pointer to bool to store result of operation + * (true if entry is newly added, false otherwise; can be NULL). + * @return pointer to data or NULL (can be even if memory could be allocated!) + */ +#define lru_get_new(table, key_, len_, is_new) \ + (__typeof__((table)->pdata_t)) \ + lru_get_impl(&(table)->lru, (key_), (len_), \ + sizeof(*(table)->pdata_t), true, is_new) + +/** + * @brief Apply a function to every item in LRU. + * + * @param table pointer to LRU + * @param function enum lru_apply_do (*function)(const char *key, uint len, val_type *val, void *baton) + * See enum lru_apply_do for the return type meanings. + * @param baton extra pointer passed to each function invocation + */ +#define lru_apply(table, function, baton) do { \ + lru_apply_fun_g(fun_dummy, __typeof__(*(table)->pdata_t)) = 0; \ + (void)(fun_dummy == (function)); /* produce a warning with incompatible function type */ \ + lru_apply_impl(&(table)->lru, (lru_apply_fun)(function), (baton)); \ + } while (false) + +/** @brief Possible actions to do with an element. */ +enum lru_apply_do { + LRU_APPLY_DO_NOTHING, + LRU_APPLY_DO_EVICT, + /* maybe more in future*/ +}; + +/** + * @brief Return the real capacity - maximum number of keys holdable within. + * + * @param table pointer to LRU + */ +#define lru_capacity(table) lru_capacity_impl(&(table)->lru) + + +/** @brief Round the value up to a multiple of (1 << power). */ +static inline uint round_power(uint size, uint power) +{ + uint res = ((size - 1) & ~((1 << power) - 1)) + (1 << power); + assert(__builtin_ctz(res) >= power); + assert(size <= res && res < size + (1 << power)); + return res; +} + + +/* ======================== Inlined part of implementation ======================== */ +/** @cond internal */ + +#define lru_apply_fun_g(name, val_type) \ + enum lru_apply_do (*(name))(const char *key, uint len, val_type *val, void *baton) +typedef lru_apply_fun_g(lru_apply_fun, void); + +#if __GNUC__ >= 4 + #define CACHE_ALIGNED __attribute__((aligned(64))) +#else + #define CACHE_ALIGNED +#endif + +struct lru; +void lru_free_items_impl(struct lru *lru); +struct lru * lru_create_impl(uint max_slots, knot_mm_t *mm_array, knot_mm_t *mm); +void * lru_get_impl(struct lru *lru, const char *key, uint key_len, + uint val_len, bool do_insert, bool *is_new); +void lru_apply_impl(struct lru *lru, lru_apply_fun f, void *baton); + +struct lru_item; + +#if SIZE_MAX > (1 << 32) + /** @internal The number of keys stored within each group. */ + #define LRU_ASSOC 3 +#else + #define LRU_ASSOC 4 +#endif +/** @internal The number of hashes tracked within each group: 10-1 or 12-1. */ +#define LRU_TRACKED ((64 - sizeof(size_t) * LRU_ASSOC) / 4 - 1) + +struct lru_group { + uint16_t counts[LRU_TRACKED+1]; /*!< Occurrence counters; the last one is special. */ + uint16_t hashes[LRU_TRACKED+1]; /*!< Top halves of hashes; the last one is unused. */ + struct lru_item *items[LRU_ASSOC]; /*!< The full items. */ +} CACHE_ALIGNED; + +/* The sizes are chosen so lru_group just fits into a single x86 cache line. */ +static_assert(64 == sizeof(struct lru_group) + && 64 == LRU_ASSOC * sizeof(void*) + (LRU_TRACKED+1) * 4, + "bad sizing for your sizeof(void*)"); + +struct lru { + struct knot_mm *mm, /**< Memory context to use for keys. */ + *mm_array; /**< Memory context to use for this structure itself. */ + uint log_groups; /**< Logarithm of the number of LRU groups. */ + struct lru_group groups[] CACHE_ALIGNED; /**< The groups of items. */ +}; + +/** @internal See lru_free. */ +static inline void lru_free_impl(struct lru *lru) +{ + if (!lru) + return; + lru_free_items_impl(lru); + mm_free(lru->mm_array, lru); +} + +/** @internal See lru_reset. */ +static inline void lru_reset_impl(struct lru *lru) +{ + lru_free_items_impl(lru); + memset(lru->groups, 0, sizeof(lru->groups[0]) * (1 << lru->log_groups)); +} + +/** @internal See lru_capacity. */ +static inline uint lru_capacity_impl(struct lru *lru) +{ + assert(lru); + return (1 << lru->log_groups) * LRU_ASSOC; +} + +/** @endcond */ +/** @} (addtogroup generics) */ diff --git a/lib/generic/map.c b/lib/generic/map.c new file mode 100644 index 0000000..3e06520 --- /dev/null +++ b/lib/generic/map.c @@ -0,0 +1,354 @@ +/* + * critbit89 - A crit-bit tree implementation for strings in C89 + * Written by Jonas Gehring + * Implemented key-value storing by Marek Vavrusa + * + * The code makes the assumption that malloc returns pointers aligned at at + * least a two-byte boundary. Since the C standard requires that malloc return + * pointers that can store any type, there are no commonly-used toolchains for + * which this assumption is false. + * + * See https://github.com/agl/critbit/blob/master/critbit.pdf for reference. + */ + +#include +#include +#include + +#include "map.h" +#include "lib/utils.h" + + /* Exports */ +#if defined _WIN32 || defined __CYGWIN__ + #define EXPORT __attribute__ ((dllexport)) +#else + #define EXPORT __attribute__ ((visibility ("default"))) +#endif + +#ifdef _MSC_VER /* MSVC */ + typedef unsigned __int8 uint8_t; + typedef unsigned __int32 uint32_t; + #ifdef _WIN64 + typedef signed __int64 intptr_t; + #else + typedef _W64 signed int intptr_t; + #endif +#else /* Not MSVC */ + #include +#endif + +typedef struct { + void* value; + uint8_t key[]; +} cb_data_t; + +typedef struct { + void *child[2]; + uint32_t byte; + uint8_t otherbits; +} cb_node_t; + +/* Return true if ptr is internal node. */ +static inline int ref_is_internal(const uint8_t *p) +{ + return 1 & (intptr_t)p; +} + +/* Get internal node. */ +static inline cb_node_t *ref_get_internal(uint8_t *p) +{ + return (cb_node_t *)(p - 1); +} + +/* Static helper functions */ +static void cbt_traverse_delete(map_t *map, void *top) +{ + uint8_t *p = top; + if (ref_is_internal(p)) { + cb_node_t *q = ref_get_internal(p); + cbt_traverse_delete(map, q->child[0]); + cbt_traverse_delete(map, q->child[1]); + mm_free(map->pool, q); + } else { + mm_free(map->pool, p); + } +} + +static int cbt_traverse_prefixed(void *top, + int (*callback)(const char *, void *, void *), void *baton) +{ + uint8_t *p = top; + cb_data_t *x = (cb_data_t *)top; + + if (ref_is_internal(p)) { + cb_node_t *q = ref_get_internal(p); + int ret = 0; + + ret = cbt_traverse_prefixed(q->child[0], callback, baton); + if (ret != 0) { + return ret; + } + ret = cbt_traverse_prefixed(q->child[1], callback, baton); + if (ret != 0) { + return ret; + } + return 0; + } + + return (callback)((const char *)x->key, x->value, baton); +} + +static cb_data_t *cbt_make_data(map_t *map, const uint8_t *str, size_t len, void *value) +{ + cb_data_t *x = mm_alloc(map->pool, sizeof(cb_data_t) + len); + if (x != NULL) { + x->value = value; + memcpy(x->key, str, len); + } + return x; +} + +/*! Like map_contains, but also set the value, if passed and found. */ +static int cbt_get(map_t *map, const char *str, void **value) +{ + const uint8_t *ubytes = (void *)str; + const size_t ulen = strlen(str); + uint8_t *p = map->root; + cb_data_t *x = NULL; + + if (p == NULL) { + return 0; + } + + while (ref_is_internal(p)) { + cb_node_t *q = ref_get_internal(p); + uint8_t c = 0; + int direction; + + if (q->byte < ulen) { + c = ubytes[q->byte]; + } + direction = (1 + (q->otherbits | c)) >> 8; + + p = q->child[direction]; + } + + x = (cb_data_t *)p; + if (strcmp(str, (const char *)x->key) == 0) { + if (value != NULL) { + *value = x->value; + } + return 1; + } + + return 0; +} + +/*! Returns non-zero if map contains str */ +EXPORT int map_contains(map_t *map, const char *str) +{ + return cbt_get(map, str, NULL); +} + +EXPORT void *map_get(map_t *map, const char *str) +{ + void *v = NULL; + cbt_get(map, str, &v); + return v; +} + +EXPORT int map_set(map_t *map, const char *str, void *val) +{ + const uint8_t *const ubytes = (void *)str; + const size_t ulen = strlen(str); + uint8_t *p = map->root; + uint8_t c = 0, *x = NULL; + uint32_t newbyte = 0; + uint32_t newotherbits = 0; + int direction = 0, newdirection = 0; + cb_node_t *newnode = NULL; + cb_data_t *data = NULL; + void **wherep = NULL; + + if (p == NULL) { + map->root = cbt_make_data(map, (const uint8_t *)str, ulen + 1, val); + if (map->root == NULL) { + return ENOMEM; + } + return 0; + } + + while (ref_is_internal(p)) { + cb_node_t *q = ref_get_internal(p); + c = 0; + if (q->byte < ulen) { + c = ubytes[q->byte]; + } + direction = (1 + (q->otherbits | c)) >> 8; + + p = q->child[direction]; + } + + data = (cb_data_t *)p; + for (newbyte = 0; newbyte < ulen; ++newbyte) { + if (data->key[newbyte] != ubytes[newbyte]) { + newotherbits = data->key[newbyte] ^ ubytes[newbyte]; + goto different_byte_found; + } + } + + if (data->key[newbyte] != 0) { + newotherbits = data->key[newbyte]; + goto different_byte_found; + } + data->value = val; + return 1; + +different_byte_found: + newotherbits |= newotherbits >> 1; + newotherbits |= newotherbits >> 2; + newotherbits |= newotherbits >> 4; + newotherbits = (newotherbits & ~(newotherbits >> 1)) ^ 255; + c = data->key[newbyte]; + newdirection = (1 + (newotherbits | c)) >> 8; + + newnode = mm_alloc(map->pool, sizeof(cb_node_t)); + if (newnode == NULL) { + return ENOMEM; + } + + x = (uint8_t *)cbt_make_data(map, ubytes, ulen + 1, val); + if (x == NULL) { + mm_free(map->pool, newnode); + return ENOMEM; + } + + newnode->byte = newbyte; + newnode->otherbits = newotherbits; + newnode->child[1 - newdirection] = x; + + /* Insert into map */ + wherep = &map->root; + for (;;) { + cb_node_t *q; + p = *wherep; + if (!ref_is_internal(p)) { + break; + } + + q = ref_get_internal(p); + if (q->byte > newbyte) { + break; + } + if (q->byte == newbyte && q->otherbits > newotherbits) { + break; + } + + c = 0; + if (q->byte < ulen) { + c = ubytes[q->byte]; + } + direction = (1 + (q->otherbits | c)) >> 8; + wherep = q->child + direction; + } + + newnode->child[newdirection] = *wherep; + *wherep = (void *)(1 + (char *)newnode); + return 0; +} + +/*! Deletes str from the map, returns 0 on success */ +EXPORT int map_del(map_t *map, const char *str) +{ + const uint8_t *ubytes = (void *)str; + const size_t ulen = strlen(str); + uint8_t *p = map->root; + void **wherep = NULL, **whereq = NULL; + cb_node_t *q = NULL; + cb_data_t *data = NULL; + int direction = 0; + + if (map->root == NULL) { + return 1; + } + wherep = &map->root; + + while (ref_is_internal(p)) { + uint8_t c = 0; + whereq = wherep; + q = ref_get_internal(p); + + if (q->byte < ulen) { + c = ubytes[q->byte]; + } + direction = (1 + (q->otherbits | c)) >> 8; + wherep = q->child + direction; + p = *wherep; + } + + data = (cb_data_t *)p; + if (strcmp(str, (const char *)data->key) != 0) { + return 1; + } + mm_free(map->pool, p); + + if (!whereq) { + map->root = NULL; + return 0; + } + + *whereq = q->child[1 - direction]; + mm_free(map->pool, q); + return 0; +} + +/*! Clears the given map */ +EXPORT void map_clear(map_t *map) +{ + if (map->root) { + cbt_traverse_delete(map, map->root); + } + map->root = NULL; +} + +/*! Calls callback for all strings in map with the given prefix */ +EXPORT int map_walk_prefixed(map_t *map, const char *prefix, + int (*callback)(const char *, void *, void *), void *baton) +{ + if (!map) { + return 0; + } + + const uint8_t *ubytes = (void *)prefix; + const size_t ulen = strlen(prefix); + uint8_t *p = map->root; + uint8_t *top = p; + cb_data_t *data = NULL; + + if (p == NULL) { + return 0; + } + + while (ref_is_internal(p)) { + cb_node_t *q = ref_get_internal(p); + uint8_t c = 0; + int direction; + + if (q->byte < ulen) { + c = ubytes[q->byte]; + } + direction = (1 + (q->otherbits | c)) >> 8; + + p = q->child[direction]; + if (q->byte < ulen) { + top = p; + } + } + + data = (cb_data_t *)p; + if (strlen((const char *)data->key) < ulen || memcmp(data->key, prefix, ulen) != 0) { + return 0; /* No strings match */ + } + + return cbt_traverse_prefixed(top, callback, baton); +} diff --git a/lib/generic/map.h b/lib/generic/map.h new file mode 100644 index 0000000..73ce4c0 --- /dev/null +++ b/lib/generic/map.h @@ -0,0 +1,115 @@ +/* + * critbit89 - A crit-bit map implementation for strings in C89 + * Written by Jonas Gehring + */ + +/** + * @file map.h + * @brief A Crit-bit tree key-value map implementation. + * + * @warning If the user provides a custom allocator, it must return addresses aligned to 2B boundary. + * + * # Example usage: + * + * @code{.c} + * map_t map = map_make(NULL); + * + * // Custom allocator (optional) + * map.malloc = &mymalloc; + * map.baton = &mymalloc_context; + * + * // Insert k-v pairs + * int values = { 42, 53, 64 }; + * if (map_set(&map, "princess", &values[0]) != 0 || + * map_set(&map, "prince", &values[1]) != 0 || + * map_set(&map, "leia", &values[2]) != 0) { + * fail(); + * } + * + * // Test membership + * if (map_contains(&map, "leia")) { + * success(); + * } + * + * // Prefix search + * int i = 0; + * int count(const char *k, void *v, void *ext) { (*(int *)ext)++; return 0; } + * if (map_walk_prefixed(map, "princ", count, &i) == 0) { + * printf("%d matches\n", i); + * } + * + * // Delete + * if (map_del(&map, "badkey") != 0) { + * fail(); // No such key + * } + * + * // Clear the map + * map_clear(&map); + * @endcode + * + * \addtogroup generics + * @{ + */ + +#pragma once + +#include + +#ifdef __cplusplus +extern "C" { +#endif + +struct knot_mm; /* avoid the unnecessary include */ + +/** Main data structure */ +typedef struct { + void *root; + struct knot_mm *pool; +} map_t; + +/** Creates an new empty critbit map. Pass NULL for malloc+free. */ +static inline map_t map_make(struct knot_mm *pool) +{ + return (map_t){ .root = NULL, .pool = pool }; +} + +/** Returns non-zero if map contains str */ +int map_contains(map_t *map, const char *str); + +/** Returns value if map contains str. Note: NULL may mean two different things. */ +void *map_get(map_t *map, const char *str); + +/** Inserts str into map. Returns 0 if new, 1 if replaced, or ENOMEM. */ +int map_set(map_t *map, const char *str, void *val); + +/** Deletes str from the map, returns 0 on suceess */ +int map_del(map_t *map, const char *str); + +/** Clears the given map */ +void map_clear(map_t *map); + +/** + * Calls callback for all strings in map + * See @fn map_walk_prefixed() for documentation on parameters. + */ +#define map_walk(map, callback, baton) \ + map_walk_prefixed((map), "", (callback), (baton)) + +/** + * Calls callback for all strings in map with the given prefix. + * Returns value immediately if a callback returns nonzero. + * + * @param map + * @param prefix required string prefix (empty => all strings) + * @param callback callback parameters are (key, value, baton) + * @param baton passed uservalue + */ +int map_walk_prefixed(map_t *map, const char *prefix, + int (*callback)(const char *, void *, void *), void *baton); + + +#ifdef __cplusplus +} +#endif + +/** @} */ diff --git a/lib/generic/pack.h b/lib/generic/pack.h new file mode 100644 index 0000000..dc7a975 --- /dev/null +++ b/lib/generic/pack.h @@ -0,0 +1,249 @@ +/* Copyright (C) 2015-2017 CZ.NIC, z.s.p.o. + + This program is free software: you can redistribute it and/or modify + it under the terms of the GNU General Public License as published by + the Free Software Foundation, either version 3 of the License, or + (at your option) any later version. + + This program is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the + GNU General Public License for more details. + + You should have received a copy of the GNU General Public License + along with this program. If not, see . + */ + +/** + * @file pack.h + * @brief A length-prefixed list of objects, also an array list. + * + * Each object is prefixed by item length, unlike array this structure + * permits variable-length data. It is also equivallent to forward-only list + * backed by an array. + * + * @note Maximum object size is 2^16 bytes, see ::pack_objlen_t + * @TODO If some mistake happens somewhere, the access may end up in an infinite loop. + * (equality comparison on pointers) + * + * # Example usage: + * + * @code{.c} + * pack_t pack; + * pack_init(pack); + * + * // Reserve 2 objects, 6 bytes total + * pack_reserve(pack, 2, 4 + 2); + * + * // Push 2 objects + * pack_obj_push(pack, U8("jedi"), 4) + * pack_obj_push(pack, U8("\xbe\xef"), 2); + * + * // Iterate length-value pairs + * uint8_t *it = pack_head(pack); + * while (it != pack_tail(pack)) { + * uint8_t *val = pack_obj_val(it); + * it = pack_obj_next(it); + * } + * + * // Remove object + * pack_obj_del(pack, U8("jedi"), 4); + * + * pack_clear(pack); + * @endcode + * + * \addtogroup generics + * @{ + */ + +#pragma once + +#include +#include +#include "array.h" + +#ifdef __cplusplus +extern "C" { +#endif + +/** Packed object length type. */ +typedef uint16_t pack_objlen_t; + +/** Pack is defined as an array of bytes */ +typedef array_t(uint8_t) pack_t; + +/** Zero-initialize the pack. */ +#define pack_init(pack) \ + array_init(pack) + +/** Make the pack empty and free pointed-to memory (plain malloc/free). */ +#define pack_clear(pack) \ + array_clear(pack) + +/** Make the pack empty and free pointed-to memory. + * Mempool usage: pass mm_free and a knot_mm_t* . */ +#define pack_clear_mm(pack, free, baton) \ + array_clear_mm((pack), (free), (baton)) + +/** Reserve space for *additional* objects in the pack (plain malloc/free). + * @return 0 if success, <0 on failure */ +#define pack_reserve(pack, objs_count, objs_len) \ + pack_reserve_mm((pack), (objs_count), (objs_len), array_std_reserve, NULL) + +/** Reserve space for *additional* objects in the pack. + * Mempool usage: pass kr_memreserve and a knot_mm_t* . + * @return 0 if success, <0 on failure */ +#define pack_reserve_mm(pack, objs_count, objs_len, reserve, baton) \ + array_reserve_mm((pack), (pack).len + (sizeof(pack_objlen_t)*(objs_count) + (objs_len)), (reserve), (baton)) + +/** Return pointer to first packed object. + * + * Recommended way to iterate: + * for (uint8_t *it = pack_head(pack); it != pack_tail(pack); it = pack_obj_next(it)) + */ +#define pack_head(pack) \ + ((pack).len > 0 ? &((pack).at[0]) : NULL) + +/** Return pack end pointer. */ +#define pack_tail(pack) \ + ((pack).len > 0 ? &((pack).at[(pack).len]) : NULL) + +/** Return packed object length. */ +static inline pack_objlen_t pack_obj_len(uint8_t *it) +{ + pack_objlen_t len = 0; + if (it != NULL) + memcpy(&len, it, sizeof(len)); + return len; +} + +/** Return packed object value. */ +static inline uint8_t *pack_obj_val(uint8_t *it) +{ + if (it == NULL) { + assert(it); + return NULL; + } + return it + sizeof(pack_objlen_t); +} + +/** Return pointer to next packed object. */ +static inline uint8_t *pack_obj_next(uint8_t *it) +{ + if (it == NULL) { + assert(it); + return NULL; + } + return pack_obj_val(it) + pack_obj_len(it); +} + +/** Return pointer to the last packed object. */ +static inline uint8_t *pack_last(pack_t pack) +{ + if (pack.len == 0) { + return NULL; + } + uint8_t *it = pack_head(pack); + uint8_t *tail = pack_tail(pack); + while (true) { + uint8_t *next = pack_obj_next(it); + if (next == tail) { + return it; + } + it = next; + } +} + +/** Push object to the end of the pack + * @return 0 on success, negative number on failure + */ +static inline int pack_obj_push(pack_t *pack, const uint8_t *obj, pack_objlen_t len) +{ + if (pack == NULL || obj == NULL) { + assert(false); + return kr_error(EINVAL); + } + size_t packed_len = len + sizeof(len); + if (pack->len + packed_len > pack->cap) { + return kr_error(ENOSPC); + } + + uint8_t *endp = pack->at + pack->len; + memcpy(endp, (char *)&len, sizeof(len)); + memcpy(endp + sizeof(len), obj, len); + pack->len += packed_len; + return 0; +} + +/** Returns a pointer to packed object. + * @return pointer to packed object or NULL + */ +static inline uint8_t *pack_obj_find(pack_t *pack, const uint8_t *obj, pack_objlen_t len) +{ + if (pack == NULL || obj == NULL) { + assert(obj != NULL); + return NULL; + } + uint8_t *endp = pack_tail(*pack); + uint8_t *it = pack_head(*pack); + while (it != endp) { + uint8_t *val = pack_obj_val(it); + if (pack_obj_len(it) == len && memcmp(obj, val, len) == 0) { + return it; + } + it = pack_obj_next(it); + } + return NULL; +} + +/** Delete object from the pack + * @return 0 on success, negative number on failure + */ +static inline int pack_obj_del(pack_t *pack, const uint8_t *obj, pack_objlen_t len) +{ + if (pack == NULL || obj == NULL) { + assert(obj != NULL); + return kr_error(EINVAL); + } + uint8_t *endp = pack_tail(*pack); + uint8_t *it = pack_obj_find(pack, obj, len); + if (it) { + size_t packed_len = len + sizeof(len); + memmove(it, it + packed_len, endp - it - packed_len); + pack->len -= packed_len; + return 0; + } + return -1; +} + +/** Clone a pack, replacing destination pack; (*dst == NULL) is valid input. + * @return kr_error(ENOMEM) on allocation failure. */ +static inline int pack_clone(pack_t **dst, const pack_t *src, knot_mm_t *pool) +{ + if (!dst || !src) { + assert(false); + return kr_error(EINVAL); + } + /* Get a valid pack_t. */ + if (!*dst) { + *dst = mm_alloc(pool, sizeof(pack_t)); + if (!*dst) return kr_error(ENOMEM); + pack_init(**dst); + /* Clone data only if needed */ + if (src->len == 0) return kr_ok(); + } + /* Replace the contents of the pack_t. */ + int ret = array_reserve_mm(**dst, src->len, kr_memreserve, pool); + if (ret < 0) { + return kr_error(ENOMEM); + } + memcpy((*dst)->at, src->at, src->len); + (*dst)->len = src->len; + return kr_ok(); +} + +#ifdef __cplusplus +} +#endif + +/** @} */ diff --git a/lib/generic/queue.c b/lib/generic/queue.c new file mode 100644 index 0000000..45657c7 --- /dev/null +++ b/lib/generic/queue.c @@ -0,0 +1,124 @@ +/* Copyright (C) 2018 CZ.NIC, z.s.p.o. + + This program is free software: you can redistribute it and/or modify + it under the terms of the GNU General Public License as published by + the Free Software Foundation, either version 3 of the License, or + (at your option) any later version. + + This program is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the + GNU General Public License for more details. + + You should have received a copy of the GNU General Public License + along with this program. If not, see . + */ + +#include "lib/generic/queue.h" +#include + +KR_EXPORT void queue_init_impl(struct queue *q, size_t item_size) +{ + q->len = 0; + q->item_size = item_size; + q->head = q->tail = NULL; + /* Take 128 B (two x86 cache lines), except a small margin + * that the allocator can use for its overhead. + * Normally (64-bit pointers) this means 16 B header + 13*8 B data. */ + q->chunk_cap = (128 - offsetof(struct queue_chunk, data) + - sizeof(size_t) + ) / item_size; + if (!q->chunk_cap) q->chunk_cap = 1; /* item_size big enough by itself */ +} + +KR_EXPORT void queue_deinit_impl(struct queue *q) +{ + assert(q); + struct queue_chunk *p = q->head; + while (p != NULL) { + struct queue_chunk *pf = p; + p = p->next; + free(pf); + } +#ifndef NDEBUG + memset(q, 0, sizeof(*q)); +#endif +} + +static struct queue_chunk * queue_chunk_new(const struct queue *q) +{ + struct queue_chunk *c = malloc(offsetof(struct queue_chunk, data) + + q->chunk_cap * q->item_size); + if (unlikely(!c)) abort(); // simplify stuff + memset(c, 0, offsetof(struct queue_chunk, data)); + c->cap = q->chunk_cap; + /* ->begin and ->end are zero, i.e. we optimize for _push + * and not _push_head, by default. */ + return c; +} + +/* Return pointer to the space for the new element. */ +KR_EXPORT void * queue_push_impl(struct queue *q) +{ + assert(q); + struct queue_chunk *t = q->tail; // shorthand + if (unlikely(!t)) { + assert(!q->head && !q->len); + q->head = q->tail = t = queue_chunk_new(q); + } else + if (t->end == t->cap) { + if (t->begin * 2 >= t->cap) { + /* Utilization is below 50%, so let's shift (no overlap). */ + memcpy(t->data, t->data + t->begin * q->item_size, + (t->end - t->begin) * q->item_size); + t->end -= t->begin; + t->begin = 0; + } else { + /* Let's grow the tail by another chunk. */ + assert(!t->next); + t->next = queue_chunk_new(q); + t = q->tail = t->next; + } + } + assert(t->end < t->cap); + ++(q->len); + ++(t->end); + return t->data + q->item_size * (t->end - 1); +} + +/* Return pointer to the space for the new element. */ +KR_EXPORT void * queue_push_head_impl(struct queue *q) +{ + /* When we have choice, we optimize for further _push_head, + * i.e. when shifting or allocating a chunk, + * we store items on the tail-end of the chunk. */ + assert(q); + struct queue_chunk *h = q->head; // shorthand + if (unlikely(!h)) { + assert(!q->tail && !q->len); + h = q->head = q->tail = queue_chunk_new(q); + h->begin = h->end = h->cap; + } else + if (h->begin == 0) { + if (h->end * 2 <= h->cap) { + /* Utilization is below 50%, so let's shift (no overlap). + * Computations here are simplified due to h->begin == 0. */ + const int cnt = h->end; + memcpy(h->data + (h->cap - cnt) * q->item_size, h->data, + cnt * q->item_size); + h->begin = h->cap - cnt; + h->end = h->cap; + } else { + /* Let's grow the head by another chunk. */ + h = queue_chunk_new(q); + h->next = q->head; + q->head = h; + h->begin = h->end = h->cap; + } + } + assert(h->begin > 0); + --(h->begin); + ++(q->len); + return h->data + q->item_size * h->begin; +} + diff --git a/lib/generic/queue.h b/lib/generic/queue.h new file mode 100644 index 0000000..755e759 --- /dev/null +++ b/lib/generic/queue.h @@ -0,0 +1,260 @@ +/* Copyright (C) 2018 CZ.NIC, z.s.p.o. + + This program is free software: you can redistribute it and/or modify + it under the terms of the GNU General Public License as published by + the Free Software Foundation, either version 3 of the License, or + (at your option) any later version. + + This program is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the + GNU General Public License for more details. + + You should have received a copy of the GNU General Public License + along with this program. If not, see . + */ +/** + * @file queue.h + * @brief A queue, usable for FIFO and LIFO simultaneously. + * + * Both the head and tail of the queue can be accessed and pushed to, + * but only the head can be popped from. + * + * @note The implementation uses a singly linked list of blocks + * where each block stores an array of values (for better efficiency). + * + * Example usage: + * @code{.c} + // define new queue type, and init a new queue instance + typedef queue_t(int) queue_int_t; + queue_int_t q; + queue_init(q); + // do some operations + queue_push(q, 1); + queue_push(q, 2); + queue_push(q, 3); + queue_push(q, 4); + queue_pop(q); + assert(queue_head(q) == 2); + assert(queue_tail(q) == 4); + + // you may iterate + typedef queue_it_t(int) queue_it_int_t; + for (queue_it_int_t it = queue_it_begin(q); !queue_it_finished(it); + queue_it_next(it)) { + ++queue_it_val(it); + } + assert(queue_tail(q) == 5); + + queue_push_head(q, 0); + ++queue_tail(q); + assert(queue_tail(q) == 6); + // free it up + queue_deinit(q); + + // you may use dynamic allocation for the type itself + queue_int_t *qm = malloc(sizeof(queue_int_t)); + queue_init(*qm); + queue_deinit(*qm); + free(qm); + * @endcode + * + * \addtogroup generics + * @{ + */ + +#pragma once + +#include "lib/defines.h" +#include "contrib/ucw/lib.h" +#include +#include +#include +#include +#include + +/** @brief The type for queue, parametrized by value type. */ +#define queue_t(type) \ + union { \ + type *pdata_t; /* only the *type* information is used */ \ + struct queue queue; \ + } + +/** @brief Initialize a queue. You can malloc() it the usual way. */ +#define queue_init(q) do { \ + (void)(((__typeof__(((q).pdata_t)))0) == (void *)0); /* typecheck queue_t */ \ + queue_init_impl(&(q).queue, sizeof(*(q).pdata_t)); \ + } while (false) + +/** @brief De-initialize a queue: make it invalid and free any inner allocations. */ +#define queue_deinit(q) \ + queue_deinit_impl(&(q).queue) + +/** @brief Push data to queue's tail. (Type-safe version; use _impl() otherwise.) */ +#define queue_push(q, data) \ + *((__typeof__((q).pdata_t)) queue_push_impl(&(q).queue)) = data + +/** @brief Push data to queue's head. (Type-safe version; use _impl() otherwise.) */ +#define queue_push_head(q, data) \ + *((__typeof__((q).pdata_t)) queue_push_head_impl(&(q).queue)) = data + +/** @brief Remove the element at the head. + * The queue must not be empty. */ +#define queue_pop(q) \ + queue_pop_impl(&(q).queue) + +/** @brief Return a "reference" to the element at the head (it's an L-value). + * The queue must not be empty. */ +#define queue_head(q) \ + ( *(__typeof__((q).pdata_t)) queue_head_impl(&(q).queue) ) + +/** @brief Return a "reference" to the element at the tail (it's an L-value). + * The queue must not be empty. */ +#define queue_tail(q) \ + ( *(__typeof__((q).pdata_t)) queue_tail_impl(&(q).queue) ) + +/** @brief Return the number of elements in the queue (very efficient). */ +#define queue_len(q) \ + ((const size_t)(q).queue.len) + + +/** @brief Type for queue iterator, parametrized by value type. + * It's a simple structure that owns no other resources. + * You may NOT use it after doing any push or pop (without _begin again). */ +#define queue_it_t(type) \ + union { \ + type *pdata_t; /* only the *type* information is used */ \ + struct queue_it iter; \ + } + +/** @brief Initialize a queue iterator at the head of the queue. + * If you use this in assignment (instead of initialization), + * you will unfortunately need to add corresponding type-cast in front. + * Beware: there's no type-check between queue and iterator! */ +#define queue_it_begin(q) \ + { .iter = queue_it_begin_impl(&(q).queue) } + +/** @brief Return a "reference" to the current element (it's an L-value) . */ +#define queue_it_val(it) \ + ( *(__typeof__((it).pdata_t)) queue_it_val_impl(&(it).iter) ) + +/** @brief Test if the iterator has gone past the last element. + * If it has, you may not use _val or _next. */ +#define queue_it_finished(it) \ + queue_it_finished_impl(&(it).iter) + +/** @brief Advance the iterator to the next element. */ +#define queue_it_next(it) \ + queue_it_next_impl(&(it).iter) + + + +/* ====================== Internal for the implementation ================== */ +/** @cond internal */ + +struct queue; +/* Non-inline functions are exported to be usable from daemon. */ +void queue_init_impl(struct queue *q, size_t item_size); +void queue_deinit_impl(struct queue *q); +void * queue_push_impl(struct queue *q); +void * queue_push_head_impl(struct queue *q); + +struct queue_chunk; +struct queue { + size_t len; + uint16_t chunk_cap, item_size; + struct queue_chunk *head, *tail; +}; + +struct queue_chunk { + struct queue_chunk *next; /*< head -> ... -> tail */ + int16_t begin, end, cap, pad_; /*< indices: zero is closest to head */ + /*< We could fit into uint8_t for example, but the choice of (3+1)*2 bytes + * is a compromise between wasting space and getting a good alignment. + * In particular, queue_t(type*) will store the pointers on addresses + * aligned to the pointer size, in both 64-bit and 32-bit platforms. + */ + char data[]; + /**< The item data. We use "char" to satisfy the C99+ aliasing rules. + * See C99 section 6.5 Expressions, paragraph 7. + * Any type can be accessed through char-pointer, + * so we can use a common struct definition + * for all types being held. + */ +}; + +static inline void * queue_head_impl(const struct queue *q) +{ + assert(q); + struct queue_chunk *h = q->head; + if (unlikely(!h)) + return NULL; + assert(h->end > h->begin); + return h->data + h->begin * q->item_size; +} + +static inline void * queue_tail_impl(const struct queue *q) +{ + assert(q); + struct queue_chunk *t = q->tail; + if (unlikely(!t)) + return NULL; + assert(t->end > t->begin); + return t->data + (t->end - 1) * q->item_size; +} + +static inline void queue_pop_impl(struct queue *q) +{ + assert(q); + struct queue_chunk *h = q->head; + assert(h && h->end > h->begin); + if (h->end - h->begin == 1) { + /* removing the last element in the chunk */ + q->head = h->next; + free(h); + } else { + ++(h->begin); + } + --(q->len); +} + + +struct queue_it { + struct queue_chunk *chunk; + int16_t pos, item_size; +}; + +static inline struct queue_it queue_it_begin_impl(struct queue *q) +{ + assert(q); + return (struct queue_it){ + .chunk = q->head, + .pos = q->head ? q->head->begin : -1, + .item_size = q->item_size, + }; +} + +static inline bool queue_it_finished_impl(struct queue_it *it) +{ + return it->chunk == NULL || it->pos >= it->chunk->end; +} + +static inline void * queue_it_val_impl(struct queue_it *it) +{ + assert(!queue_it_finished_impl(it)); + return it->chunk->data + it->pos * it->item_size; +} + +static inline void queue_it_next_impl(struct queue_it *it) +{ + assert(!queue_it_finished_impl(it)); + ++(it->pos); + if (it->pos < it->chunk->end) + return; + it->chunk = it->chunk->next; + it->pos = it->chunk ? it->chunk->begin : -1; +} + +/** @endcond (internal) */ +/** @} (addtogroup generics) */ + diff --git a/lib/generic/set.h b/lib/generic/set.h new file mode 100644 index 0000000..332c1aa --- /dev/null +++ b/lib/generic/set.h @@ -0,0 +1,105 @@ +/* Copyright (C) 2015-2017 CZ.NIC, z.s.p.o. + + This program is free software: you can redistribute it and/or modify + it under the terms of the GNU General Public License as published by + the Free Software Foundation, either version 3 of the License, or + (at your option) any later version. + + This program is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the + GNU General Public License for more details. + + You should have received a copy of the GNU General Public License + along with this program. If not, see . + */ + +/** + * @file set.h + * @brief A set abstraction implemented on top of map. + * + * @note The API is based on map.h, see it for more examples. + * + * # Example usage: + * + * @code{.c} + * set_t set = set_make(NULL); + * + * // Insert keys + * if (set_add(&set, "princess") != 0 || + * set_add(&set, "prince") != 0 || + * set_add(&set, "leia") != 0) { + * fail(); + * } + * + * // Test membership + * if (set_contains(&set, "leia")) { + * success(); + * } + * + * // Prefix search + * int i = 0; + * int count(const char *s, void *n) { (*(int *)n)++; return 0; } + * if (set_walk_prefixed(set, "princ", count, &i) == 0) { + * printf("%d matches\n", i); + * } + * + * // Delete + * if (set_del(&set, "badkey") != 0) { + * fail(); // No such key + * } + * + * // Clear the set + * set_clear(&set); + * @endcode + * + * \addtogroup generics + * @{ + */ + +#pragma once + +#include +#include "map.h" + +#ifdef __cplusplus +extern "C" { +#endif + +typedef map_t set_t; +typedef int (set_walk_cb)(const char *, void *); + +/*! Creates an new, empty critbit set */ +#define set_make \ + map_make + +/*! Returns non-zero if set contains str */ +#define set_contains(set, str) \ + map_contains((set), (str)) + +/*! Inserts str into set. Returns 0 if new, 1 if already present, or ENOMEM. */ +#define set_add(set, str) \ + map_set((set), (str), (void *)1) + +/*! Deletes str from the set, returns 0 on suceess */ +#define set_del(set, str) \ + map_del((set), (str)) + +/*! Clears the given set */ +#define set_clear(set) \ + map_clear(set) + +/*! Calls callback for all strings in map */ +#define set_walk(set, callback, baton) \ + map_walk_prefixed((set), "", (callback), (baton)) + +/*! Calls callback for all strings in set with the given prefix */ +#define set_walk_prefixed(set, prefix, callback, baton) \ + map_walk_prefixed((set), (prefix), (callback), (baton)) + + +#ifdef __cplusplus +} +#endif + +/** @} */ diff --git a/lib/generic/trie.c b/lib/generic/trie.c new file mode 100644 index 0000000..0009eef --- /dev/null +++ b/lib/generic/trie.c @@ -0,0 +1,912 @@ +/* Copyright (C) 2016-2018 CZ.NIC, z.s.p.o. + + This program is free software: you can redistribute it and/or modify + it under the terms of the GNU General Public License as published by + the Free Software Foundation, either version 3 of the License, or + (at your option) any later version. + + This program is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the + GNU General Public License for more details. + + You should have received a copy of the GNU General Public License + along with this program. If not, see . + + The code originated from https://github.com/fanf2/qp/blob/master/qp.c + at revision 5f6d93753. + */ + +#include +#include +#include + +#include "lib/generic/trie.h" +#include "lib/utils.h" +#include "contrib/ucw/lib.h" + +#if defined(__i386) || defined(__x86_64) || defined(_M_IX86) \ + || (defined(__BYTE_ORDER__) && defined(__ORDER_LITTLE_ENDIAN) \ + && __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__) + + /*! + * \brief Use a pointer alignment hack to save memory. + * + * When on, isbranch() relies on the fact that in leaf_t the first pointer + * is aligned on multiple of 4 bytes and that the flags bitfield is + * overlaid over the lowest two bits of that pointer. + * Neither is really guaranteed by the C standards; the second part should + * be OK with x86_64 ABI and most likely any other little-endian platform. + * It would be possible to manipulate the right bits portably, but it would + * complicate the code nontrivially. C++ doesn't even guarantee type-punning. + * In debug mode we check this works OK when creating a new trie instance. + */ + #define FLAGS_HACK 1 +#else + #define FLAGS_HACK 0 +#endif + +typedef unsigned char byte; +#ifndef uint +typedef unsigned int uint; +#define uint uint +#endif +typedef uint bitmap_t; /*! Bit-maps, using the range of 1<<0 to 1<<16 (inclusive). */ + +typedef struct { + uint32_t len; // 32 bits are enough for key lengths; probably even 16 bits would be. + char chars[]; +} tkey_t; + +/*! \brief Leaf of trie. */ +typedef struct { + #if !FLAGS_HACK + byte flags; + #endif + tkey_t *key; /*!< The pointer must be aligned to 4-byte multiples! */ + trie_val_t val; +} leaf_t; + +/*! \brief A trie node is either leaf_t or branch_t. */ +typedef union node node_t; + +/*! + * \brief Branch node of trie. + * + * - The flags distinguish whether the node is a leaf_t (0), or a branch + * testing the more-important nibble (1) or the less-important one (2). + * - It stores the index of the byte that the node tests. The combined + * value (index*4 + flags) increases in branch nodes as you go deeper + * into the trie. All the keys below a branch are identical up to the + * nibble identified by the branch. Indices have to be stored because + * we skip any branch nodes that would have a single child. + * (Consequently, the skipped parts of key have to be validated in a leaf.) + * - The bitmap indicates which subtries are present. The present child nodes + * are stored in the twigs array (with no holes between them). + * - To simplify storing keys that are prefixes of each other, the end-of-string + * position is treated as another nibble value, ordered before all others. + * That affects the bitmap and twigs fields. + * + * \note The branch nodes are never allocated individually, but they are + * always part of either the root node or the twigs array of the parent. + */ +typedef struct { + #if FLAGS_HACK + uint32_t flags : 2, + bitmap : 17; /*!< The first bitmap bit is for end-of-string child. */ + #else + byte flags; + uint32_t bitmap; + #endif + uint32_t index; + node_t *twigs; +} branch_t; + +union node { + leaf_t leaf; + branch_t branch; +}; + +struct trie { + node_t root; // undefined when weight == 0, see empty_root() + size_t weight; + knot_mm_t mm; +}; + +/*! \brief Make the root node empty (debug-only). */ +static inline void empty_root(node_t *root) { +#ifndef NDEBUG + *root = (node_t){ .branch = { + .flags = 3, // invalid value that fits + .bitmap = 0, + .index = -1, + .twigs = NULL + } }; +#endif +} + +/*! \brief Check that unportable code works OK (debug-only). */ +static void assert_portability(void) { +#if FLAGS_HACK + assert(((union node){ .leaf = { + .key = (tkey_t *)(((uint8_t *)NULL) + 1), + .val = NULL + } }).branch.flags == 1); +#endif +} + +/*! \brief Propagate error codes. */ +#define ERR_RETURN(x) \ + do { \ + int err_code_ = x; \ + if (unlikely(err_code_ != KNOT_EOK)) \ + return err_code_; \ + } while (false) + +/*! + * \brief Count the number of set bits. + * + * \TODO This implementation may be relatively slow on some HW. + */ +static uint bitmap_weight(bitmap_t w) +{ + assert((w & ~((1 << 17) - 1)) == 0); // using the least-important 17 bits + return __builtin_popcount(w); +} + +/*! \brief Only keep the lowest bit in the bitmap (least significant -> twigs[0]). */ +static bitmap_t bitmap_lowest_bit(bitmap_t w) +{ + assert((w & ~((1 << 17) - 1)) == 0); // using the least-important 17 bits + return 1 << __builtin_ctz(w); +} + +/*! \brief Test flags to determine type of this node. */ +static bool isbranch(const node_t *t) +{ + uint f = t->branch.flags; + assert(f <= 2); + return f != 0; +} + +/*! \brief Make a bitmask for testing a branch bitmap. */ +static bitmap_t nibbit(byte k, uint flags) +{ + uint shift = (2 - flags) << 2; + uint nibble = (k >> shift) & 0xf; + return 1 << (nibble + 1/*because of prefix keys*/); +} + +/*! \brief Extract a nibble from a key and turn it into a bitmask. */ +static bitmap_t twigbit(const node_t *t, const char *key, uint32_t len) +{ + assert(isbranch(t)); + uint i = t->branch.index; + + if (i >= len) + return 1 << 0; // leaf position + + return nibbit((byte)key[i], t->branch.flags); +} + +/*! \brief Test if a branch node has a child indicated by a bitmask. */ +static bool hastwig(const node_t *t, bitmap_t bit) +{ + assert(isbranch(t)); + return t->branch.bitmap & bit; +} + +/*! \brief Compute offset of an existing child in a branch node. */ +static uint twigoff(const node_t *t, bitmap_t b) +{ + assert(isbranch(t)); + return bitmap_weight(t->branch.bitmap & (b - 1)); +} + +/*! \brief Get pointer to a particular child of a branch node. */ +static node_t* twig(node_t *t, uint i) +{ + assert(isbranch(t)); + return &t->branch.twigs[i]; +} + +/*! + * \brief For a branch nod, compute offset of a child and child count. + * + * Having this separate might be meaningful for performance optimization. + */ +#define TWIGOFFMAX(off, max, t, b) do { \ + off = twigoff(t, b); \ + max = bitmap_weight(t->branch.bitmap); \ + } while(0) + +/*! \brief Simple string comparator. */ +static int key_cmp(const char *k1, uint32_t k1_len, const char *k2, uint32_t k2_len) +{ + int ret = memcmp(k1, k2, MIN(k1_len, k2_len)); + if (ret != 0) { + return ret; + } + + /* Key string is equal, compare lengths. */ + if (k1_len == k2_len) { + return 0; + } else if (k1_len < k2_len) { + return -1; + } else { + return 1; + } +} + +trie_t* trie_create(knot_mm_t *mm) +{ + assert_portability(); + trie_t *trie = mm_alloc(mm, sizeof(trie_t)); + if (trie != NULL) { + empty_root(&trie->root); + trie->weight = 0; + if (mm != NULL) + trie->mm = *mm; + else + mm_ctx_init(&trie->mm); + } + return trie; +} + +/*! \brief Free anything under the trie node, except for the passed pointer itself. */ +static void clear_trie(node_t *trie, knot_mm_t *mm) +{ + if (!isbranch(trie)) { + mm_free(mm, trie->leaf.key); + } else { + branch_t *b = &trie->branch; + int len = bitmap_weight(b->bitmap); + for (int i = 0; i < len; ++i) + clear_trie(b->twigs + i, mm); + mm_free(mm, b->twigs); + } +} + +void trie_free(trie_t *tbl) +{ + if (tbl == NULL) + return; + if (tbl->weight) + clear_trie(&tbl->root, &tbl->mm); + mm_free(&tbl->mm, tbl); +} + +void trie_clear(trie_t *tbl) +{ + assert(tbl); + if (!tbl->weight) + return; + clear_trie(&tbl->root, &tbl->mm); + empty_root(&tbl->root); + tbl->weight = 0; +} + +size_t trie_weight(const trie_t *tbl) +{ + assert(tbl); + return tbl->weight; +} + +struct found { + leaf_t *l; /**< the found leaf (NULL if not found) */ + branch_t *p; /**< the leaf's parent (if exists) */ + bitmap_t b; /**< bit-mask with a single bit marking l under p */ +}; +/** Search trie for an item with the given key (equality only). */ +static struct found find_equal(trie_t *tbl, const char *key, uint32_t len) +{ + assert(tbl); + struct found ret0; + memset(&ret0, 0, sizeof(ret0)); + if (!tbl->weight) + return ret0; + /* Current node and parent while descending (returned values basically). */ + node_t *t = &tbl->root; + branch_t *p = NULL; + bitmap_t b = 0; + while (isbranch(t)) { + __builtin_prefetch(t->branch.twigs); + b = twigbit(t, key, len); + if (!hastwig(t, b)) + return ret0; + p = &t->branch; + t = twig(t, twigoff(t, b)); + } + if (key_cmp(key, len, t->leaf.key->chars, t->leaf.key->len) != 0) + return ret0; + return (struct found) { + .l = &t->leaf, + .p = p, + .b = b, + }; +} +/** Find item with the first key (lexicographical order). */ +static struct found find_first(trie_t *tbl) +{ + assert(tbl); + if (!tbl->weight) { + struct found ret0; + memset(&ret0, 0, sizeof(ret0)); + return ret0; + } + /* Current node and parent while descending (returned values basically). */ + node_t *t = &tbl->root; + branch_t *p = NULL; + while (isbranch(t)) { + p = &t->branch; + t = &p->twigs[0]; + } + return (struct found) { + .l = &t->leaf, + .p = p, + .b = p ? bitmap_lowest_bit(p->bitmap) : 0, + }; +} + +trie_val_t* trie_get_try(trie_t *tbl, const char *key, uint32_t len) +{ + struct found found = find_equal(tbl, key, len); + return found.l ? &found.l->val : NULL; +} + +trie_val_t* trie_get_first(trie_t *tbl, char **key, uint32_t *len) +{ + struct found found = find_first(tbl); + if (!found.l) + return NULL; + if (key) + *key = found.l->key->chars; + if (len) + *len = found.l->key->len; + return &found.l->val; +} + +/** Delete the found element (if any) and return value (unless NULL is passed) */ +static int del_found(trie_t *tbl, struct found found, trie_val_t *val) +{ + if (!found.l) + return KNOT_ENOENT; + mm_free(&tbl->mm, found.l->key); + if (val != NULL) + *val = found.l->val; // we return trie_val_t directly when deleting + --tbl->weight; + branch_t * const p = found.p; // short-hand + if (unlikely(!p)) { // whole trie was a single leaf + assert(tbl->weight == 0); + empty_root(&tbl->root); + return KNOT_EOK; + } + // remove leaf t as child of p; get child index via pointer arithmetic + int ci = ((union node *)found.l) - p->twigs, + cc = bitmap_weight(p->bitmap); // child count + assert(ci >= 0 && ci < cc); + + if (cc == 2) { // collapse binary node p: move the other child to this node + node_t *twigs = p->twigs; + (*(union node *)p) = twigs[1 - ci]; // it might be a leaf or branch + mm_free(&tbl->mm, twigs); + return KNOT_EOK; + } + memmove(p->twigs + ci, p->twigs + ci + 1, sizeof(node_t) * (cc - ci - 1)); + p->bitmap &= ~found.b; + node_t *twigs = mm_realloc(&tbl->mm, p->twigs, sizeof(node_t) * (cc - 1), + sizeof(node_t) * cc); + if (likely(twigs != NULL)) + p->twigs = twigs; + /* We can ignore mm_realloc failure, only beware that next time + * the prev_size passed to it wouldn't be correct; TODO? */ + return KNOT_EOK; +} + +int trie_del(trie_t *tbl, const char *key, uint32_t len, trie_val_t *val) +{ + struct found found = find_equal(tbl, key, len); + return del_found(tbl, found, val); +} + +int trie_del_first(trie_t *tbl, char *key, uint32_t *len, trie_val_t *val) +{ + struct found found = find_first(tbl); + if (!found.l) + return KNOT_ENOENT; + if (key) { + if (!len) + return KNOT_EINVAL; + if (*len < found.l->key->len) + return kr_error(ENOSPC); + memcpy(key, found.l->key->chars, found.l->key->len); + } + if (len) { // makes sense even with key == NULL + *len = found.l->key->len; + } + return del_found(tbl, found, val); +} + +/*! + * \brief Stack of nodes, storing a path down a trie. + * + * The structure also serves directly as the public trie_it_t type, + * in which case it always points to the current leaf, unless we've finished + * (i.e. it->len == 0). + */ +typedef struct trie_it { + node_t* *stack; /*!< The stack; malloc is used directly instead of mm. */ + uint32_t len; /*!< Current length of the stack. */ + uint32_t alen; /*!< Allocated/available length of the stack. */ + /*! \brief Initial storage for \a stack; it should fit in many use cases. */ + node_t* stack_init[60]; +} nstack_t; + +/*! \brief Create a node stack containing just the root (or empty). */ +static void ns_init(nstack_t *ns, trie_t *tbl) +{ + assert(tbl); + ns->stack = ns->stack_init; + ns->alen = sizeof(ns->stack_init) / sizeof(ns->stack_init[0]); + if (tbl->weight) { + ns->len = 1; + ns->stack[0] = &tbl->root; + } else { + ns->len = 0; + } +} + +/*! \brief Free inside of the stack, i.e. not the passed pointer itself. */ +static void ns_cleanup(nstack_t *ns) +{ + assert(ns && ns->stack); + if (likely(ns->stack == ns->stack_init)) + return; + free(ns->stack); + #ifndef NDEBUG + ns->stack = NULL; + ns->alen = 0; + #endif +} + +/*! \brief Allocate more space for the stack. */ +static int ns_longer_alloc(nstack_t *ns) +{ + ns->alen *= 2; + size_t new_size = sizeof(nstack_t) + ns->alen * sizeof(node_t *); + node_t **st; + if (ns->stack == ns->stack_init) { + st = malloc(new_size); + if (st != NULL) + memcpy(st, ns->stack, ns->len * sizeof(node_t *)); + } else { + st = realloc(ns->stack, new_size); + } + if (st == NULL) + return KNOT_ENOMEM; + ns->stack = st; + return KNOT_EOK; +} + +/*! \brief Ensure the node stack can be extended by one. */ +static inline int ns_longer(nstack_t *ns) +{ + // get a longer stack if needed + if (likely(ns->len < ns->alen)) + return KNOT_EOK; + return ns_longer_alloc(ns); // hand-split the part suitable for inlining +} + +/*! + * \brief Find the "branching point" as if searching for a key. + * + * The whole path to the point is kept on the passed stack; + * always at least the root will remain on the top of it. + * Beware: the precise semantics of this function is rather tricky. + * The top of the stack will contain: the corresponding leaf if exact match is found; + * or the immediate node below a branching-point-on-edge or the branching-point itself. + * + * \param info Set position of the point of first mismatch (in index and flags). + * \param first Set the value of the first non-matching character (from trie), + * optionally; end-of-string character has value -256 (that's why it's int). + * Note: the character is converted to *unsigned* char (i.e. 0..255), + * as that's the ordering used in the trie. + * + * \return KNOT_EOK or KNOT_ENOMEM. + */ +static int ns_find_branch(nstack_t *ns, const char *key, uint32_t len, + branch_t *info, int *first) +{ + assert(ns && ns->len && info); + // First find some leaf with longest matching prefix. + while (isbranch(ns->stack[ns->len - 1])) { + ERR_RETURN(ns_longer(ns)); + node_t *t = ns->stack[ns->len - 1]; + __builtin_prefetch(t->branch.twigs); + bitmap_t b = twigbit(t, key, len); + // Even if our key is missing from this branch we need to + // keep iterating down to a leaf. It doesn't matter which + // twig we choose since the keys are all the same up to this + // index. Note that blindly using twigoff(t, b) can cause + // an out-of-bounds index if it equals twigmax(t). + uint i = hastwig(t, b) ? twigoff(t, b) : 0; + ns->stack[ns->len++] = twig(t, i); + } + tkey_t *lkey = ns->stack[ns->len-1]->leaf.key; + // Find index of the first char that differs. + uint32_t index = 0; + while (index < MIN(len,lkey->len)) { + if (key[index] != lkey->chars[index]) + break; + else + ++index; + } + info->index = index; + if (first) + *first = lkey->len > index ? (unsigned char)lkey->chars[index] : -256; + // Find flags: which half-byte has matched. + uint flags; + if (index == len && len == lkey->len) { // found equivalent key + info->flags = flags = 0; + goto success; + } + if (likely(index < MIN(len,lkey->len))) { + byte k2 = (byte)lkey->chars[index]; + byte k1 = (byte)key[index]; + flags = ((k1 ^ k2) & 0xf0) ? 1 : 2; + } else { // one is prefix of another + flags = 1; + } + info->flags = flags; + // now go up the trie from the current leaf + branch_t *t; + do { + if (unlikely(ns->len == 1)) + goto success; // only the root stays on the stack + t = (branch_t*)ns->stack[ns->len - 2]; + if (t->index < index || (t->index == index && t->flags < flags)) + goto success; + --ns->len; + } while (true); +success: + #ifndef NDEBUG // invariants on successful return + assert(ns->len); + if (isbranch(ns->stack[ns->len - 1])) { + t = &ns->stack[ns->len - 1]->branch; + assert(t->index > index || (t->index == index && t->flags >= flags)); + } + if (ns->len > 1) { + t = &ns->stack[ns->len - 2]->branch; + assert(t->index < index || (t->index == index + && (t->flags < flags || (t->flags == 1 && flags == 0)))); + } + #endif + return KNOT_EOK; +} + +/*! + * \brief Advance the node stack to the last leaf in the subtree. + * + * \return KNOT_EOK or KNOT_ENOMEM. + */ +static int ns_last_leaf(nstack_t *ns) +{ + assert(ns); + do { + ERR_RETURN(ns_longer(ns)); + node_t *t = ns->stack[ns->len - 1]; + if (!isbranch(t)) + return KNOT_EOK; + int lasti = bitmap_weight(t->branch.bitmap) - 1; + assert(lasti >= 0); + ns->stack[ns->len++] = twig(t, lasti); + } while (true); +} + +/*! + * \brief Advance the node stack to the first leaf in the subtree. + * + * \return KNOT_EOK or KNOT_ENOMEM. + */ +static int ns_first_leaf(nstack_t *ns) +{ + assert(ns && ns->len); + do { + ERR_RETURN(ns_longer(ns)); + node_t *t = ns->stack[ns->len - 1]; + if (!isbranch(t)) + return KNOT_EOK; + ns->stack[ns->len++] = twig(t, 0); + } while (true); +} + +/*! + * \brief Advance the node stack to the leaf that is previous to the current node. + * + * \note Prefix leaf under the current node DOES count (if present; perhaps questionable). + * \return KNOT_EOK on success, KNOT_ENOENT on not-found, or possibly KNOT_ENOMEM. + */ +static int ns_prev_leaf(nstack_t *ns) +{ + assert(ns && ns->len > 0); + + node_t *t = ns->stack[ns->len - 1]; + if (hastwig(t, 1 << 0)) { // the prefix leaf + t = twig(t, 0); + ERR_RETURN(ns_longer(ns)); + ns->stack[ns->len++] = t; + return KNOT_EOK; + } + + do { + if (ns->len < 2) + return KNOT_ENOENT; // root without empty key has no previous leaf + t = ns->stack[ns->len - 1]; + node_t *p = ns->stack[ns->len - 2]; + int pindex = t - p->branch.twigs; // index in parent via pointer arithmetic + assert(pindex >= 0 && pindex <= 16); + if (pindex > 0) { // t isn't the first child -> go down the previous one + ns->stack[ns->len - 1] = twig(p, pindex - 1); + return ns_last_leaf(ns); + } + // we've got to go up again + --ns->len; + } while (true); +} + +/*! + * \brief Advance the node stack to the leaf that is successor to the current node. + * + * \note Prefix leaf or anything else under the current node DOES count. + * \return KNOT_EOK on success, KNOT_ENOENT on not-found, or possibly KNOT_ENOMEM. + */ +static int ns_next_leaf(nstack_t *ns) +{ + assert(ns && ns->len > 0); + + node_t *t = ns->stack[ns->len - 1]; + if (isbranch(t)) + return ns_first_leaf(ns); + do { + if (ns->len < 2) + return KNOT_ENOENT; // not found, as no more parent is available + t = ns->stack[ns->len - 1]; + node_t *p = ns->stack[ns->len - 2]; + int pindex = t - p->branch.twigs; // index in parent via pointer arithmetic + assert(pindex >= 0 && pindex <= 16); + int pcount = bitmap_weight(p->branch.bitmap); + if (pindex + 1 < pcount) { // t isn't the last child -> go down the next one + ns->stack[ns->len - 1] = twig(p, pindex + 1); + return ns_first_leaf(ns); + } + // we've got to go up again + --ns->len; + } while (true); +} + +int trie_get_leq(trie_t *tbl, const char *key, uint32_t len, trie_val_t **val) +{ + assert(tbl && val); + *val = NULL; // so on failure we can just return; + if (tbl->weight == 0) + return KNOT_ENOENT; + { // Intentionally un-indented; until end of function, to bound cleanup attr. + // First find a key with longest-matching prefix + __attribute__((cleanup(ns_cleanup))) + nstack_t ns_local; + ns_init(&ns_local, tbl); + nstack_t *ns = &ns_local; + branch_t bp; + int un_leaf; // first unmatched character in the leaf + ERR_RETURN(ns_find_branch(ns, key, len, &bp, &un_leaf)); + int un_key = bp.index < len ? (unsigned char)key[bp.index] : -256; + node_t *t = ns->stack[ns->len - 1]; + if (bp.flags == 0) { // found exact match + *val = &t->leaf.val; + return KNOT_EOK; + } + // Get t: the last node on matching path + if (isbranch(t) && t->branch.index == bp.index && t->branch.flags == bp.flags) { + // t is OK + } else { + // the top of the stack was the first unmatched node -> step up + if (ns->len == 1) { + // root was unmatched already + if (un_key < un_leaf) + return KNOT_ENOENT; + ERR_RETURN(ns_last_leaf(ns)); + goto success; + } + --ns->len; + t = ns->stack[ns->len - 1]; + } + // Now we re-do the first "non-matching" step in the trie + // but try the previous child if key was less (it may not exist) + bitmap_t b = twigbit(t, key, len); + int i = hastwig(t, b) + ? twigoff(t, b) - (un_key < un_leaf) + : twigoff(t, b) - 1 /*twigoff returns successor when !hastwig*/; + if (i >= 0) { + ERR_RETURN(ns_longer(ns)); + ns->stack[ns->len++] = twig(t, i); + ERR_RETURN(ns_last_leaf(ns)); + } else { + ERR_RETURN(ns_prev_leaf(ns)); + } +success: + assert(!isbranch(ns->stack[ns->len - 1])); + *val = &ns->stack[ns->len - 1]->leaf.val; + return 1; + } +} + +/*! \brief Initialize a new leaf, copying the key, and returning failure code. */ +static int mk_leaf(node_t *leaf, const char *key, uint32_t len, knot_mm_t *mm) +{ + tkey_t *k = mm_alloc(mm, sizeof(tkey_t) + len); + #if FLAGS_HACK + assert(((uintptr_t)k) % 4 == 0); // we need an aligned pointer + #endif + if (unlikely(!k)) + return KNOT_ENOMEM; + k->len = len; + memcpy(k->chars, key, len); + leaf->leaf = (leaf_t){ + #if !FLAGS_HACK + .flags = 0, + #endif + .val = NULL, + .key = k + }; + return KNOT_EOK; +} + +trie_val_t* trie_get_ins(trie_t *tbl, const char *key, uint32_t len) +{ + assert(tbl); + // First leaf in an empty tbl? + if (unlikely(!tbl->weight)) { + if (unlikely(mk_leaf(&tbl->root, key, len, &tbl->mm))) + return NULL; + ++tbl->weight; + return &tbl->root.leaf.val; + } + { // Intentionally un-indented; until end of function, to bound cleanup attr. + // Find the branching-point + __attribute__((cleanup(ns_cleanup))) + nstack_t ns_local; + ns_init(&ns_local, tbl); + nstack_t *ns = &ns_local; + branch_t bp; // branch-point: index and flags signifying the longest common prefix + int k2; // the first unmatched character in the leaf + if (unlikely(ns_find_branch(ns, key, len, &bp, &k2))) + return NULL; + node_t *t = ns->stack[ns->len - 1]; + if (bp.flags == 0) // the same key was already present + return &t->leaf.val; + node_t leaf; + if (unlikely(mk_leaf(&leaf, key, len, &tbl->mm))) + return NULL; + + if (isbranch(t) && bp.index == t->branch.index && bp.flags == t->branch.flags) { + // The node t needs a new leaf child. + bitmap_t b1 = twigbit(t, key, len); + assert(!hastwig(t, b1)); + uint s, m; TWIGOFFMAX(s, m, t, b1); // new child position and original child count + node_t *twigs = mm_realloc(&tbl->mm, t->branch.twigs, + sizeof(node_t) * (m + 1), sizeof(node_t) * m); + if (unlikely(!twigs)) + goto err_leaf; + memmove(twigs + s + 1, twigs + s, sizeof(node_t) * (m - s)); + twigs[s] = leaf; + t->branch.twigs = twigs; + t->branch.bitmap |= b1; + ++tbl->weight; + return &twigs[s].leaf.val; + } else { + // We need to insert a new binary branch with leaf at *t. + // Note: it works the same for the case where we insert above root t. + #ifndef NDEBUG + if (ns->len > 1) { + node_t *pt = ns->stack[ns->len - 2]; + assert(hastwig(pt, twigbit(pt, key, len))); + } + #endif + node_t *twigs = mm_alloc(&tbl->mm, sizeof(node_t) * 2); + if (unlikely(!twigs)) + goto err_leaf; + node_t t2 = *t; // Save before overwriting t. + t->branch.flags = bp.flags; + t->branch.index = bp.index; + t->branch.twigs = twigs; + bitmap_t b1 = twigbit(t, key, len); + bitmap_t b2 = unlikely(k2 == -256) ? (1 << 0) : nibbit(k2, bp.flags); + t->branch.bitmap = b1 | b2; + *twig(t, twigoff(t, b1)) = leaf; + *twig(t, twigoff(t, b2)) = t2; + ++tbl->weight; + return &twig(t, twigoff(t, b1))->leaf.val; + }; +err_leaf: + mm_free(&tbl->mm, leaf.leaf.key); + return NULL; + } +} + +/*! \brief Apply a function to every trie_val_t*, in order; a recursive solution. */ +static int apply_trie(node_t *t, int (*f)(trie_val_t *, void *), void *d) +{ + assert(t); + if (!isbranch(t)) + return f(&t->leaf.val, d); + int child_count = bitmap_weight(t->branch.bitmap); + for (int i = 0; i < child_count; ++i) + ERR_RETURN(apply_trie(twig(t, i), f, d)); + return KNOT_EOK; +} + +int trie_apply(trie_t *tbl, int (*f)(trie_val_t *, void *), void *d) +{ + assert(tbl && f); + if (!tbl->weight) + return KNOT_EOK; + return apply_trie(&tbl->root, f, d); +} + +/* These are all thin wrappers around static Tns* functions. */ +trie_it_t* trie_it_begin(trie_t *tbl) +{ + assert(tbl); + trie_it_t *it = malloc(sizeof(nstack_t)); + if (!it) + return NULL; + ns_init(it, tbl); + if (it->len == 0) // empty tbl + return it; + if (ns_first_leaf(it)) { + ns_cleanup(it); + free(it); + return NULL; + } + return it; +} + +void trie_it_next(trie_it_t *it) +{ + assert(it && it->len); + if (ns_next_leaf(it) != KNOT_EOK) + it->len = 0; +} + +bool trie_it_finished(trie_it_t *it) +{ + assert(it); + return it->len == 0; +} + +void trie_it_free(trie_it_t *it) +{ + if (!it) + return; + ns_cleanup(it); + free(it); +} + +const char* trie_it_key(trie_it_t *it, size_t *len) +{ + assert(it && it->len); + node_t *t = it->stack[it->len - 1]; + assert(!isbranch(t)); + tkey_t *key = t->leaf.key; + if (len) + *len = key->len; + return key->chars; +} + +trie_val_t* trie_it_val(trie_it_t *it) +{ + assert(it && it->len); + node_t *t = it->stack[it->len - 1]; + assert(!isbranch(t)); + return &t->leaf.val; +} diff --git a/lib/generic/trie.h b/lib/generic/trie.h new file mode 100644 index 0000000..0550e95 --- /dev/null +++ b/lib/generic/trie.h @@ -0,0 +1,150 @@ +/* Copyright (C) 2017-2018 CZ.NIC, z.s.p.o. + + This program is free software: you can redistribute it and/or modify + it under the terms of the GNU General Public License as published by + the Free Software Foundation, either version 3 of the License, or + (at your option) any later version. + + This program is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the + GNU General Public License for more details. + + You should have received a copy of the GNU General Public License + along with this program. If not, see . + */ + +#pragma once + +#include +#include + +#include +#include "lib/defines.h" + +/*! + * \brief Native API of QP-tries: + * + * - keys are char strings, not necessarily zero-terminated, + * the structure copies the contents of the passed keys + * - values are void* pointers, typically you get an ephemeral pointer to it + * - key lengths are limited by 2^32-1 ATM + * + * XXX EDITORS: trie.{h,c} are synced from + * https://gitlab.labs.nic.cz/knot/knot-dns/tree/68352fc969/src/contrib/qp-trie + * only with tiny adjustments, mostly #includes and KR_EXPORT. + */ + +/*! \brief Element value. */ +typedef void* trie_val_t; + +/*! \brief Opaque structure holding a QP-trie. */ +typedef struct trie trie_t; + +/*! \brief Opaque type for holding a QP-trie iterator. */ +typedef struct trie_it trie_it_t; + +/*! \brief Create a trie instance. Pass NULL to use malloc+free. */ +KR_EXPORT +trie_t* trie_create(knot_mm_t *mm); + +/*! \brief Free a trie instance. */ +KR_EXPORT +void trie_free(trie_t *tbl); + +/*! \brief Clear a trie instance (make it empty). */ +KR_EXPORT +void trie_clear(trie_t *tbl); + +/*! \brief Return the number of keys in the trie. */ +KR_EXPORT +size_t trie_weight(const trie_t *tbl); + +/*! \brief Search the trie, returning NULL on failure. */ +KR_EXPORT +trie_val_t* trie_get_try(trie_t *tbl, const char *key, uint32_t len); + +/*! + * \brief Return pointer to the minimum. Optionally with key and its length. */ +KR_EXPORT +trie_val_t* trie_get_first(trie_t *tbl, char **key, uint32_t *len); + +/*! \brief Search the trie, inserting NULL trie_val_t on failure. */ +KR_EXPORT +trie_val_t* trie_get_ins(trie_t *tbl, const char *key, uint32_t len); + +/*! + * \brief Search for less-or-equal element. + * + * \param tbl Trie. + * \param key Searched key. + * \param len Key length. + * \param val Must be valid; it will be set to NULL if not found or errored. + * \return KNOT_EOK for exact match, 1 for previous, KNOT_ENOENT for not-found, + * or KNOT_E*. + */ +KR_EXPORT +int trie_get_leq(trie_t *tbl, const char *key, uint32_t len, trie_val_t **val); + +/*! + * \brief Apply a function to every trie_val_t, in order. + * + * \param d Parameter passed as the second argument to f(). + * \return First nonzero from f() or zero (i.e. KNOT_EOK). + */ +int trie_apply(trie_t *tbl, int (*f)(trie_val_t *, void *), void *d); + +/*! + * \brief Remove an item, returning KNOT_EOK if succeeded or KNOT_ENOENT if not found. + * + * If val!=NULL and deletion succeeded, the deleted value is set. + */ +KR_EXPORT +int trie_del(trie_t *tbl, const char *key, uint32_t len, trie_val_t *val); + +/*! + * \brief Remove the first item, returning KNOT_EOK on success. + * + * You may optionally get the key and/or value. + * The key is copied, so you need to pass sufficient len, + * otherwise kr_error(ENOSPC) is returned. + */ +KR_EXPORT +int trie_del_first(trie_t *tbl, char *key, uint32_t *len, trie_val_t *val); + +/*! \brief Create a new iterator pointing to the first element (if any). */ +KR_EXPORT +trie_it_t* trie_it_begin(trie_t *tbl); + +/*! + * \brief Advance the iterator to the next element. + * + * Iteration is in ascending lexicographical order. + * In particular, the empty string would be considered as the very first. + * + * \note You may not use this function if the trie's key-set has been modified + * during the lifetime of the iterator (modifying values only is OK). + */ +KR_EXPORT +void trie_it_next(trie_it_t *it); + +/*! \brief Test if the iterator has gone past the last element. */ +KR_EXPORT +bool trie_it_finished(trie_it_t *it); + +/*! \brief Free any resources of the iterator. It's OK to call it on NULL. */ +KR_EXPORT +void trie_it_free(trie_it_t *it); + +/*! + * \brief Return pointer to the key of the current element. + * + * \note The optional len is uint32_t internally but size_t is better for our usage, + * as it is without an additional type conversion. + */ +KR_EXPORT +const char* trie_it_key(trie_it_t *it, size_t *len); + +/*! \brief Return pointer to the value of the current element (writable). */ +KR_EXPORT +trie_val_t* trie_it_val(trie_it_t *it); diff --git a/lib/layer.h b/lib/layer.h new file mode 100644 index 0000000..0909cb7 --- /dev/null +++ b/lib/layer.h @@ -0,0 +1,107 @@ +/* Copyright (C) 2014-2017 CZ.NIC, z.s.p.o. + + This program is free software: you can redistribute it and/or modify + it under the terms of the GNU General Public License as published by + the Free Software Foundation, either version 3 of the License, or + (at your option) any later version. + + This program is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the + GNU General Public License for more details. + + You should have received a copy of the GNU General Public License + along with this program. If not, see . + */ + +#pragma once + +#include "lib/defines.h" +#include "lib/utils.h" + +#ifdef NOVERBOSELOG + #define QRVERBOSE(query, cls, ...) +#else + /** Print a debug message related to resolution. + * \param _query associated kr_query, may be NULL + * \param _cls identifying string, typically of length exactly four (padded) + * \param ... printf-compatible list of parameters + */ + #define QRVERBOSE(_query, _cls, ...) do { \ + const struct kr_query *_qry = (_query); \ + if (kr_log_trace_enabled(_qry)) { \ + kr_log_trace(_qry, (_cls), __VA_ARGS__); \ + } else if (VERBOSE_STATUS) { \ + kr_log_qverbose_impl(_qry, (_cls), __VA_ARGS__); \ + } \ + } while (false) +#endif + +/** Layer processing states. Only one value at a time (but see TODO). + * + * Each state represents the state machine transition, + * and determines readiness for the next action. + * See struct kr_layer_api for the actions. + * + * TODO: the cookie module sometimes sets (_FAIL | _DONE) on purpose (!) + */ +enum kr_layer_state { + KR_STATE_CONSUME = 1 << 0, /*!< Consume data. */ + KR_STATE_PRODUCE = 1 << 1, /*!< Produce data. */ + KR_STATE_DONE = 1 << 2, /*!< Finished successfully. */ + KR_STATE_FAIL = 1 << 3, /*!< Error. */ + KR_STATE_YIELD = 1 << 4, /*!< Paused, waiting for a sub-query. */ +}; + +/* Forward declarations. */ +struct kr_layer_api; + +/** Packet processing context. */ +typedef struct kr_layer { + int state; /*!< The current state; bitmap of enum kr_layer_state. */ + struct kr_request *req; /*!< The corresponding request. */ + const struct kr_layer_api *api; +} kr_layer_t; + +/** Packet processing module API. All functions return the new kr_layer_state. */ +struct kr_layer_api { + /** Start of processing the DNS request. */ + int (*begin)(kr_layer_t *ctx); + + int (*reset)(kr_layer_t *ctx); + + /** Paired to begin, called both on successes and failures. */ + int (*finish)(kr_layer_t *ctx); + + /** Processing an answer from upstream or the answer to the request. */ + int (*consume)(kr_layer_t *ctx, knot_pkt_t *pkt); + + /** Produce either an answer to the request or a query for upstream (or fail). */ + int (*produce)(kr_layer_t *ctx, knot_pkt_t *pkt); + + /** Finalises the outbound query packet with the knowledge of the IP addresses. + * The checkout layer doesn't persist the state, so canceled subrequests + * don't affect the resolution or rest of the processing. */ + int (*checkout)(kr_layer_t *ctx, knot_pkt_t *packet, struct sockaddr *dst, int type); + + /** Finalises the answer. + * Last chance to affect what will get into the answer, including EDNS.*/ + int (*answer_finalize)(kr_layer_t *ctx); + + /** The module can store anything in here. */ + void *data; + + /** Internal to ./daemon/ffimodule.c. */ + int cb_slots[]; +}; + +typedef struct kr_layer_api kr_layer_api_t; + +/** Pickled layer state (api, input, state). */ +struct kr_layer_pickle { + struct kr_layer_pickle *next; + const struct kr_layer_api *api; + knot_pkt_t *pkt; + unsigned state; +}; + diff --git a/lib/layer/cache.c b/lib/layer/cache.c new file mode 100644 index 0000000..c7bbc1a --- /dev/null +++ b/lib/layer/cache.c @@ -0,0 +1,31 @@ +/* Copyright (C) 2017 CZ.NIC, z.s.p.o. + + This program is free software: you can redistribute it and/or modify + it under the terms of the GNU General Public License as published by + the Free Software Foundation, either version 3 of the License, or + (at your option) any later version. + + This program is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the + GNU General Public License for more details. + + You should have received a copy of the GNU General Public License + along with this program. If not, see . + */ + +#include "lib/module.h" +#include "lib/cache/api.h" + +/** Module implementation. */ +const kr_layer_api_t *cache_layer(struct kr_module *module) +{ + static const kr_layer_api_t _layer = { + .produce = &cache_peek, + .consume = &cache_stash, + }; + + return &_layer; +} + +KR_MODULE_EXPORT(cache) diff --git a/lib/layer/iterate.c b/lib/layer/iterate.c new file mode 100644 index 0000000..cf57cc5 --- /dev/null +++ b/lib/layer/iterate.c @@ -0,0 +1,1135 @@ +/* Copyright (C) 2014-2017 CZ.NIC, z.s.p.o. + + This program is free software: you can redistribute it and/or modify + it under the terms of the GNU General Public License as published by + the Free Software Foundation, either version 3 of the License, or + (at your option) any later version. + + This program is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the + GNU General Public License for more details. + + You should have received a copy of the GNU General Public License + along with this program. If not, see . + */ + +/** @file iterate.c + * + * This builtin module is mainly active in the consume phase. + * Primary responsibilities: + * - Classify the packet as auth/nonauth and change its AA flag accordingly. + * - Pick interesting RRs to kr_request::answ_selected and ::auth_selected, + * NEW: and classify their rank, except for validation status. + * - Update kr_query::zone_cut (in case of referral). + * - Interpret CNAMEs. + * - Prepare the followup query - either inline or as another kr_query + * (CNAME jumps create a new "sibling" query). + */ + +#include +#include +#include + +#include +#include +#include +#include + +#include "lib/layer/iterate.h" +#include "lib/resolve.h" +#include "lib/rplan.h" +#include "lib/defines.h" +#include "lib/nsrep.h" +#include "lib/module.h" +#include "lib/dnssec/ta.h" + +#define VERBOSE_MSG(...) QRVERBOSE(req->current_query, "iter", __VA_ARGS__) +#define QVERBOSE_MSG(qry, ...) QRVERBOSE(qry, "iter", __VA_ARGS__) + +/* Iterator often walks through packet section, this is an abstraction. */ +typedef int (*rr_callback_t)(const knot_rrset_t *, unsigned, struct kr_request *); + +/** Return minimized QNAME/QTYPE for current zone cut. */ +static const knot_dname_t *minimized_qname(struct kr_query *query, uint16_t *qtype) +{ + /* Minimization disabled. */ + const knot_dname_t *qname = query->sname; + if (qname[0] == '\0' || query->flags.NO_MINIMIZE || query->flags.STUB) { + return qname; + } + + /* Minimize name to contain current zone cut + 1 label. */ + int cut_labels = knot_dname_labels(query->zone_cut.name, NULL); + int qname_labels = knot_dname_labels(qname, NULL); + while(qname[0] && qname_labels > cut_labels + 1) { + qname = knot_wire_next_label(qname, NULL); + qname_labels -= 1; + } + + /* Hide QTYPE if minimized. */ + if (qname != query->sname) { + *qtype = KNOT_RRTYPE_NS; + } + + return qname; +} + +/** Answer is paired to query. */ +static bool is_paired_to_query(const knot_pkt_t *answer, struct kr_query *query) +{ + uint16_t qtype = query->stype; + const knot_dname_t *qname = minimized_qname(query, &qtype); + + return query->id == knot_wire_get_id(answer->wire) && + knot_wire_get_qdcount(answer->wire) > 0 && + query->sclass == knot_pkt_qclass(answer) && + qtype == knot_pkt_qtype(answer) && + knot_dname_is_equal(qname, knot_pkt_qname(answer)); +} + +/** Relaxed rule for AA, either AA=1 or SOA matching zone cut is required. */ +static bool is_authoritative(const knot_pkt_t *answer, struct kr_query *query) +{ + if (knot_wire_get_aa(answer->wire)) { + return true; + } + + const knot_pktsection_t *ns = knot_pkt_section(answer, KNOT_AUTHORITY); + for (unsigned i = 0; i < ns->count; ++i) { + const knot_rrset_t *rr = knot_pkt_rr(ns, i); + if (rr->type == KNOT_RRTYPE_SOA + && knot_dname_in_bailiwick(rr->owner, query->zone_cut.name) >= 0) { + return true; + } + } + +#ifndef STRICT_MODE + /* Last resort to work around broken auths, if the zone cut is at/parent of the QNAME. */ + if (knot_dname_is_equal(query->zone_cut.name, knot_pkt_qname(answer))) { + return true; + } +#endif + + /* Some authoritative servers are hopelessly broken, allow lame answers in permissive mode. */ + if (query->flags.PERMISSIVE) { + return true; + } + + return false; +} + +int kr_response_classify(const knot_pkt_t *pkt) +{ + const knot_pktsection_t *an = knot_pkt_section(pkt, KNOT_ANSWER); + switch (knot_wire_get_rcode(pkt->wire)) { + case KNOT_RCODE_NOERROR: + return (an->count == 0) ? PKT_NODATA : PKT_NOERROR; + case KNOT_RCODE_NXDOMAIN: + return PKT_NXDOMAIN; + case KNOT_RCODE_REFUSED: + return PKT_REFUSED; + default: + return PKT_ERROR; + } +} + +/** @internal Filter ANY or loopback addresses. */ +static bool is_valid_addr(const uint8_t *addr, size_t len) +{ + if (len == sizeof(struct in_addr)) { + /* Filter ANY and 127.0.0.0/8 */ + uint32_t ip_host = ntohl(*(const uint32_t *)(addr)); + if (ip_host == 0 || (ip_host & 0xff000000) == 0x7f000000) { + return false; + } + } else if (len == sizeof(struct in6_addr)) { + struct in6_addr ip6_mask; + memset(&ip6_mask, 0, sizeof(ip6_mask)); + /* All except last byte are zeroed, last byte defines ANY/::1 */ + if (memcmp(addr, ip6_mask.s6_addr, sizeof(ip6_mask.s6_addr) - 1) == 0) { + return (addr[len - 1] > 1); + } + } + return true; +} + +/** @internal Update NS address from record \a rr. Return _FAIL on error. */ +static int update_nsaddr(const knot_rrset_t *rr, struct kr_query *query, int *glue_cnt) +{ + if (rr->type == KNOT_RRTYPE_A || rr->type == KNOT_RRTYPE_AAAA) { + const knot_rdata_t *rdata = rr->rrs.rdata; + const int a_len = rr->type == KNOT_RRTYPE_A + ? sizeof(struct in_addr) : sizeof(struct in6_addr); + if (a_len != rdata->len) { + QVERBOSE_MSG(query, "<= ignoring invalid glue, length %d != %d\n", + (int)rdata->len, a_len); + return KR_STATE_FAIL; + } + char name_str[KR_DNAME_STR_MAXLEN]; + char addr_str[INET6_ADDRSTRLEN]; + WITH_VERBOSE(query) { + const int af = (rr->type == KNOT_RRTYPE_A) ? AF_INET : AF_INET6; + knot_dname_to_str(name_str, rr->owner, sizeof(name_str)); + name_str[sizeof(name_str) - 1] = 0; + inet_ntop(af, rdata->data, addr_str, sizeof(addr_str)); + } + if (!(query->flags.ALLOW_LOCAL) && + !is_valid_addr(rdata->data, rdata->len)) { + QVERBOSE_MSG(query, "<= ignoring invalid glue for " + "'%s': '%s'\n", name_str, addr_str); + return KR_STATE_CONSUME; /* Ignore invalid addresses */ + } + int ret = kr_zonecut_add(&query->zone_cut, rr->owner, rdata->data, rdata->len); + if (ret != 0) { + return KR_STATE_FAIL; + } + + ++*glue_cnt; /* reduced verbosity */ + /* QVERBOSE_MSG(query, "<= using glue for " + "'%s': '%s'\n", name_str, addr_str); + */ + } + return KR_STATE_CONSUME; +} + +/** @internal From \a pkt, fetch glue records for name \a ns, and update the cut etc. + * + * \param glue_cnt the number of accepted addresses (to be incremented) + */ +static void fetch_glue(knot_pkt_t *pkt, const knot_dname_t *ns, bool in_bailiwick, + struct kr_request *req, const struct kr_query *qry, int *glue_cnt) +{ + ranked_rr_array_t *selected[] = kr_request_selected(req); + for (knot_section_t i = KNOT_ANSWER; i <= KNOT_ADDITIONAL; ++i) { + const knot_pktsection_t *sec = knot_pkt_section(pkt, i); + for (unsigned k = 0; k < sec->count; ++k) { + const knot_rrset_t *rr = knot_pkt_rr(sec, k); + if (!knot_dname_is_equal(ns, rr->owner)) { + continue; + } + if ((rr->type != KNOT_RRTYPE_A) && + (rr->type != KNOT_RRTYPE_AAAA)) { + continue; + } + + uint8_t rank = (in_bailiwick && i == KNOT_ANSWER) + ? (KR_RANK_INITIAL | KR_RANK_AUTH) : KR_RANK_OMIT; + (void) kr_ranked_rrarray_add(selected[i], rr, rank, + false, qry->uid, &req->pool); + + if ((rr->type == KNOT_RRTYPE_A) && + (req->ctx->options.NO_IPV4)) { + continue; + } + if ((rr->type == KNOT_RRTYPE_AAAA) && + (req->ctx->options.NO_IPV6)) { + continue; + } + (void) update_nsaddr(rr, req->current_query, glue_cnt); + } + } +} + +/** Attempt to find glue for given nameserver name (best effort). */ +static bool has_glue(knot_pkt_t *pkt, const knot_dname_t *ns) +{ + for (knot_section_t i = KNOT_ANSWER; i <= KNOT_ADDITIONAL; ++i) { + const knot_pktsection_t *sec = knot_pkt_section(pkt, i); + for (unsigned k = 0; k < sec->count; ++k) { + const knot_rrset_t *rr = knot_pkt_rr(sec, k); + if (knot_dname_is_equal(ns, rr->owner) && + (rr->type == KNOT_RRTYPE_A || rr->type == KNOT_RRTYPE_AAAA)) { + return true; + } + } + } + return false; +} + +/** @internal Update the cut with another NS(+glue) record. + * @param current_cut is cut name before this packet. + * @return _DONE if cut->name changes, _FAIL on error, and _CONSUME otherwise. */ +static int update_cut(knot_pkt_t *pkt, const knot_rrset_t *rr, + struct kr_request *req, const knot_dname_t *current_cut, + int *glue_cnt) +{ + struct kr_query *qry = req->current_query; + struct kr_zonecut *cut = &qry->zone_cut; + int state = KR_STATE_CONSUME; + + /* New authority MUST be at/below the authority of the current cut; + * also qname must be below new authority; + * otherwise it's a possible cache injection attempt. */ + const bool ok = knot_dname_in_bailiwick(rr->owner, current_cut) >= 0 + && knot_dname_in_bailiwick(qry->sname, rr->owner) >= 0; + if (!ok) { + VERBOSE_MSG("<= authority: ns outside bailiwick\n"); +#ifdef STRICT_MODE + return KR_STATE_FAIL; +#else + /* Workaround: ignore out-of-bailiwick NSs for authoritative answers, + * but fail for referrals. This is important to detect lame answers. */ + if (knot_pkt_section(pkt, KNOT_ANSWER)->count == 0) { + state = KR_STATE_FAIL; + } + return state; +#endif + } + + /* Update zone cut name */ + if (!knot_dname_is_equal(rr->owner, cut->name)) { + /* Remember parent cut and descend to new (keep keys and TA). */ + struct kr_zonecut *parent = mm_alloc(&req->pool, sizeof(*parent)); + if (parent) { + memcpy(parent, cut, sizeof(*parent)); + kr_zonecut_init(cut, rr->owner, &req->pool); + cut->key = parent->key; + cut->trust_anchor = parent->trust_anchor; + cut->parent = parent; + } else { + kr_zonecut_set(cut, rr->owner); + } + state = KR_STATE_DONE; + } + + /* Fetch glue for each NS */ + knot_rdata_t *rdata_i = rr->rrs.rdata; + for (unsigned i = 0; i < rr->rrs.count; + ++i, rdata_i = knot_rdataset_next(rdata_i)) { + const knot_dname_t *ns_name = knot_ns_name(rdata_i); + /* Glue is mandatory for NS below zone */ + if (knot_dname_in_bailiwick(ns_name, rr->owner) >= 0 + && !has_glue(pkt, ns_name)) { + const char *msg = + "<= authority: missing mandatory glue, skipping NS"; + WITH_VERBOSE(qry) { + auto_free char *ns_str = kr_dname_text(ns_name); + VERBOSE_MSG("%s %s\n", msg, ns_str); + } + continue; + } + int ret = kr_zonecut_add(cut, ns_name, NULL, 0); + assert(!ret); (void)ret; + + /* Choose when to use glue records. */ + const bool in_bailiwick = + knot_dname_in_bailiwick(ns_name, current_cut) >= 0; + bool do_fetch; + if (qry->flags.PERMISSIVE) { + do_fetch = true; + } else if (qry->flags.STRICT) { + /* Strict mode uses only mandatory glue. */ + do_fetch = knot_dname_in_bailiwick(ns_name, cut->name) >= 0; + } else { + /* Normal mode uses in-bailiwick glue. */ + do_fetch = in_bailiwick; + } + if (do_fetch) { + fetch_glue(pkt, ns_name, in_bailiwick, req, qry, glue_cnt); + } + } + + return state; +} + +/** Compute rank appropriate for RRs present in the packet. + * @param answer whether the RR is from answer or authority section + * @param is_nonauth: from referral or forwarding (etc.) */ +static uint8_t get_initial_rank(const knot_rrset_t *rr, const struct kr_query *qry, + const bool answer, const bool is_nonauth) +{ + /* For RRSIGs, ensure the KR_RANK_AUTH flag corresponds to the signed RR. */ + uint16_t type = kr_rrset_type_maysig(rr); + + if (qry->flags.CACHED) { + return rr->additional ? *(uint8_t *)rr->additional : KR_RANK_OMIT; + /* ^^ Current use case for "cached" RRs without rank: hints module. */ + } + if (answer || type == KNOT_RRTYPE_DS + || type == KNOT_RRTYPE_SOA /* needed for aggressive negative caching */ + || type == KNOT_RRTYPE_NSEC || type == KNOT_RRTYPE_NSEC3) { + /* We almost always want these validated, and it should be possible. */ + return KR_RANK_INITIAL | KR_RANK_AUTH; + } + /* Be aggressive: try to validate anything else (almost never extra latency). */ + return KR_RANK_TRY; + /* TODO: this classifier of authoritativity may not be perfect yet. */ +} + +static int pick_authority(knot_pkt_t *pkt, struct kr_request *req, bool to_wire) +{ + struct kr_query *qry = req->current_query; + const knot_pktsection_t *ns = knot_pkt_section(pkt, KNOT_AUTHORITY); + + const knot_dname_t *zonecut_name = qry->zone_cut.name; + bool referral = !knot_wire_get_aa(pkt->wire); + if (referral) { + /* zone cut already updated by process_authority() + * use parent zonecut name */ + zonecut_name = qry->zone_cut.parent ? qry->zone_cut.parent->name : qry->zone_cut.name; + to_wire = false; + } + + for (unsigned i = 0; i < ns->count; ++i) { + const knot_rrset_t *rr = knot_pkt_rr(ns, i); + if (rr->rclass != KNOT_CLASS_IN + || knot_dname_in_bailiwick(rr->owner, zonecut_name) < 0) { + continue; + } + uint8_t rank = get_initial_rank(rr, qry, false, + qry->flags.FORWARD || referral); + int ret = kr_ranked_rrarray_add(&req->auth_selected, rr, + rank, to_wire, qry->uid, &req->pool); + if (ret != kr_ok()) { + return ret; + } + } + + return kr_ok(); +} + +static int process_authority(knot_pkt_t *pkt, struct kr_request *req) +{ + struct kr_query *qry = req->current_query; + assert(!(qry->flags.STUB)); + + int result = KR_STATE_CONSUME; + if (qry->flags.FORWARD) { + return result; + } + + const knot_pktsection_t *ns = knot_pkt_section(pkt, KNOT_AUTHORITY); + const knot_pktsection_t *an = knot_pkt_section(pkt, KNOT_ANSWER); + +#ifdef STRICT_MODE + /* AA, terminate resolution chain. */ + if (knot_wire_get_aa(pkt->wire)) { + return KR_STATE_CONSUME; + } +#else + /* Work around servers sending back CNAME with different delegation and no AA. */ + if (an->count > 0 && ns->count > 0) { + const knot_rrset_t *rr = knot_pkt_rr(an, 0); + if (rr->type == KNOT_RRTYPE_CNAME) { + return KR_STATE_CONSUME; + } + /* Work around for these NSs which are authoritative both for + * parent and child and mixes data from both zones in single answer */ + if (knot_wire_get_aa(pkt->wire) && + (rr->type == qry->stype) && + (knot_dname_is_equal(rr->owner, qry->sname))) { + return KR_STATE_CONSUME; + } + } +#endif + /* Remember current bailiwick for NS processing. */ + const knot_dname_t *current_zone_cut = qry->zone_cut.name; + bool ns_record_exists = false; + int glue_cnt = 0; + /* Update zone cut information. */ + for (unsigned i = 0; i < ns->count; ++i) { + const knot_rrset_t *rr = knot_pkt_rr(ns, i); + if (rr->type == KNOT_RRTYPE_NS) { + ns_record_exists = true; + int state = update_cut(pkt, rr, req, current_zone_cut, &glue_cnt); + switch(state) { + case KR_STATE_DONE: result = state; break; + case KR_STATE_FAIL: return state; break; + default: /* continue */ break; + } + } else if (rr->type == KNOT_RRTYPE_SOA + && knot_dname_in_bailiwick(rr->owner, qry->zone_cut.name) > 0) { + /* SOA below cut in authority indicates different authority, + * but same NS set. */ + qry->zone_cut.name = knot_dname_copy(rr->owner, &req->pool); + } + } + + /* Nameserver is authoritative for both parent side and the child side of the + * delegation may respond with an NS record in the answer section, and still update + * the zone cut (e.g. what a.gtld-servers.net would respond for `com NS`) */ + if (!ns_record_exists && knot_wire_get_aa(pkt->wire)) { + for (unsigned i = 0; i < an->count; ++i) { + const knot_rrset_t *rr = knot_pkt_rr(an, i); + if (rr->type == KNOT_RRTYPE_NS + && knot_dname_in_bailiwick(rr->owner, qry->zone_cut.name) > 0) { + /* NS below cut in authority indicates different authority, + * but same NS set. */ + qry->zone_cut.name = knot_dname_copy(rr->owner, &req->pool); + } + } + } + + if (glue_cnt) { + VERBOSE_MSG("<= loaded %d glue addresses\n", glue_cnt); + } + + + if ((qry->flags.DNSSEC_WANT) && (result == KR_STATE_CONSUME)) { + if (knot_wire_get_aa(pkt->wire) == 0 && + knot_wire_get_ancount(pkt->wire) == 0 && + ns_record_exists) { + /* Unhelpful referral + Prevent from validating as an authoritative answer */ + result = KR_STATE_DONE; + } + } + + /* CONSUME => Unhelpful referral. + * DONE => Zone cut updated. */ + return result; +} + +static void finalize_answer(knot_pkt_t *pkt, struct kr_query *qry, struct kr_request *req) +{ + /* Finalize header */ + knot_pkt_t *answer = req->answer; + knot_wire_set_rcode(answer->wire, knot_wire_get_rcode(pkt->wire)); +} + +static int unroll_cname(knot_pkt_t *pkt, struct kr_request *req, bool referral, const knot_dname_t **cname_ret) +{ + struct kr_query *query = req->current_query; + assert(!(query->flags.STUB)); + /* Process answer type */ + const knot_pktsection_t *an = knot_pkt_section(pkt, KNOT_ANSWER); + const knot_dname_t *cname = NULL; + const knot_dname_t *pending_cname = query->sname; + unsigned cname_chain_len = 0; + bool is_final = (query->parent == NULL); + uint32_t iter_count = 0; + bool strict_mode = (query->flags.STRICT); + do { + /* CNAME was found at previous iteration, but records may not follow the correct order. + * Try to find records for pending_cname owner from section start. */ + cname = pending_cname; + pending_cname = NULL; + const int cname_labels = knot_dname_labels(cname, NULL); + for (unsigned i = 0; i < an->count; ++i) { + const knot_rrset_t *rr = knot_pkt_rr(an, i); + + /* Skip the RR if its owner+type doesn't interest us. */ + const uint16_t type = kr_rrset_type_maysig(rr); + const bool type_OK = rr->type == query->stype || type == query->stype + || type == KNOT_RRTYPE_CNAME || type == KNOT_RRTYPE_DNAME; + /* TODO: actually handle DNAMEs */ + if (rr->rclass != KNOT_CLASS_IN || !type_OK + || !knot_dname_is_equal(rr->owner, cname) + || knot_dname_in_bailiwick(rr->owner, query->zone_cut.name) < 0) { + continue; + } + + if (rr->type == KNOT_RRTYPE_RRSIG) { + int rrsig_labels = knot_rrsig_labels(rr->rrs.rdata); + if (rrsig_labels > cname_labels) { + /* clearly wrong RRSIG, don't pick it. + * don't fail immediately, + * let validator work. */ + continue; + } + if (rrsig_labels < cname_labels) { + query->flags.DNSSEC_WEXPAND = true; + } + } + + /* Process records matching current SNAME */ + int state = KR_STATE_FAIL; + bool to_wire = false; + if (is_final) { + /* if not referral, mark record to be written to final answer */ + to_wire = !referral; + } else { + int cnt_ = 0; + state = update_nsaddr(rr, query->parent, &cnt_); + if (state == KR_STATE_FAIL) { + return state; + } + } + uint8_t rank = get_initial_rank(rr, query, true, + query->flags.FORWARD || referral); + state = kr_ranked_rrarray_add(&req->answ_selected, rr, + rank, to_wire, query->uid, &req->pool); + if (state != kr_ok()) { + return KR_STATE_FAIL; + } + /* Jump to next CNAME target */ + if ((query->stype == KNOT_RRTYPE_CNAME) || (rr->type != KNOT_RRTYPE_CNAME)) { + continue; + } + cname_chain_len += 1; + pending_cname = knot_cname_name(rr->rrs.rdata); + if (!pending_cname) { + break; + } + if (cname_chain_len > an->count || cname_chain_len > KR_CNAME_CHAIN_LIMIT) { + VERBOSE_MSG("<= too long cname chain\n"); + return KR_STATE_FAIL; + } + /* Don't use pending_cname immediately. + * There are can be records for "old" cname. */ + } + if (!pending_cname) { + break; + } + if (knot_dname_is_equal(cname, pending_cname)) { + VERBOSE_MSG("<= cname chain loop\n"); + return KR_STATE_FAIL; + } + /* In strict mode, explicitly fetch each CNAME target. */ + if (strict_mode) { + cname = pending_cname; + break; + } + /* Information outside bailiwick is not trusted. */ + if (knot_dname_in_bailiwick(pending_cname, query->zone_cut.name) < 0) { + cname = pending_cname; + break; + } + /* The validator still can't handle multiple zones in one answer, + * so we only follow if a single label is replaced. + * TODO: this still isn't 100%, as the target might have a NS+DS, + * possibly leading to a SERVFAIL for the in-bailiwick name. */ + const int pending_labels = knot_dname_labels(pending_cname, NULL); + if (pending_labels != cname_labels) { + cname = pending_cname; + break; + } + if (knot_dname_matched_labels(pending_cname, cname) != + (cname_labels - 1)) { + cname = pending_cname; + break; + } + } while (++iter_count < KR_CNAME_CHAIN_LIMIT); + if (iter_count >= KR_CNAME_CHAIN_LIMIT) { + VERBOSE_MSG("<= too long cname chain\n"); + return KR_STATE_FAIL; + } + *cname_ret = cname; + return kr_ok(); +} + +static int process_referral_answer(knot_pkt_t *pkt, struct kr_request *req) +{ + const knot_dname_t *cname = NULL; + int state = unroll_cname(pkt, req, true, &cname); + if (state != kr_ok()) { + return KR_STATE_FAIL; + } + struct kr_query *query = req->current_query; + if (!(query->flags.CACHED)) { + /* If not cached (i.e. got from upstream) + * make sure that this is not an authoritative answer + * (even with AA=1) for other layers. + * There can be answers with AA=1, + * empty answer section and NS in authority. + * Clearing of AA prevents them from + * caching in the packet cache. + * If packet already cached, don't touch him. */ + knot_wire_clear_aa(pkt->wire); + } + state = pick_authority(pkt, req, false); + return state == kr_ok() ? KR_STATE_DONE : KR_STATE_FAIL; +} + +static int process_final(knot_pkt_t *pkt, struct kr_request *req, + const knot_dname_t *cname) +{ + const int pkt_class = kr_response_classify(pkt); + struct kr_query *query = req->current_query; + ranked_rr_array_t *array = &req->answ_selected; + for (size_t i = 0; i < array->len; ++i) { + const knot_rrset_t *rr = array->at[i]->rr; + if (!knot_dname_is_equal(rr->owner, cname)) { + continue; + } + if ((rr->rclass != query->sclass) || + (rr->type != query->stype)) { + continue; + } + const bool to_wire = ((pkt_class & (PKT_NXDOMAIN|PKT_NODATA)) != 0); + const int state = pick_authority(pkt, req, to_wire); + if (state != kr_ok()) { + return KR_STATE_FAIL; + } + if (!array->at[i]->to_wire) { + const size_t last_idx = array->len - 1; + size_t j = i; + ranked_rr_array_entry_t *entry = array->at[i]; + /* Relocate record to the end, after current cname */ + while (j < last_idx) { + array->at[j] = array->at[j + 1]; + ++j; + } + array->at[last_idx] = entry; + entry->to_wire = true; + } + finalize_answer(pkt, query, req); + return KR_STATE_DONE; + } + return kr_ok(); +} + +static int process_answer(knot_pkt_t *pkt, struct kr_request *req) +{ + struct kr_query *query = req->current_query; + + /* Response for minimized QNAME. Note that current iterator's minimization + * is only able ask one label below a zone cut. + * NODATA => may be empty non-terminal, retry (found zone cut) + * NOERROR => found zone cut, retry, except the case described below + * NXDOMAIN => parent is zone cut, retry as a workaround for bad authoritatives + */ + const bool is_final = (query->parent == NULL); + const int pkt_class = kr_response_classify(pkt); + const knot_dname_t * pkt_qname = knot_pkt_qname(pkt); + if (!knot_dname_is_equal(pkt_qname, query->sname) && + (pkt_class & (PKT_NOERROR|PKT_NXDOMAIN|PKT_REFUSED|PKT_NODATA))) { + /* Check for parent server that is authoritative for child zone, + * several CCTLDs where the SLD and TLD have the same name servers */ + const knot_pktsection_t *ans = knot_pkt_section(pkt, KNOT_ANSWER); + if ((pkt_class & (PKT_NOERROR)) && ans->count > 0 && + knot_dname_is_equal(pkt_qname, query->zone_cut.name)) { + VERBOSE_MSG("<= continuing with qname minimization\n"); + } else { + /* fall back to disabling minimization */ + VERBOSE_MSG("<= retrying with non-minimized name\n"); + query->flags.NO_MINIMIZE = true; + } + return KR_STATE_CONSUME; + } + + /* This answer didn't improve resolution chain, therefore must be authoritative (relaxed to negative). */ + if (!is_authoritative(pkt, query)) { + if (!(query->flags.FORWARD) && + pkt_class & (PKT_NXDOMAIN|PKT_NODATA)) { + VERBOSE_MSG("<= lame response: non-auth sent negative response\n"); + return KR_STATE_FAIL; + } + } + + const knot_dname_t *cname = NULL; + /* Process answer type */ + int state = unroll_cname(pkt, req, false, &cname); + if (state != kr_ok()) { + return state; + } + /* Make sure that this is an authoritative answer (even with AA=0) for other layers */ + knot_wire_set_aa(pkt->wire); + /* Either way it resolves current query. */ + query->flags.RESOLVED = true; + /* Follow canonical name as next SNAME. */ + if (!knot_dname_is_equal(cname, query->sname)) { + /* Check if target record has been already copied */ + query->flags.CNAME = true; + if (is_final) { + state = process_final(pkt, req, cname); + if (state != kr_ok()) { + return state; + } + } else if ((query->flags.FORWARD) && + ((query->stype == KNOT_RRTYPE_DS) || + (query->stype == KNOT_RRTYPE_NS))) { + /* CNAME'ed answer for DS or NS subquery. + * Treat it as proof of zonecut nonexistance. */ + return KR_STATE_DONE; + } + VERBOSE_MSG("<= cname chain, following\n"); + /* Check if the same query was followed in the same CNAME chain. */ + for (const struct kr_query *q = query->cname_parent; q != NULL; + q = q->cname_parent) { + if (q->sclass == query->sclass && + q->stype == query->stype && + knot_dname_is_equal(q->sname, cname)) { + VERBOSE_MSG("<= cname chain loop\n"); + return KR_STATE_FAIL; + } + } + struct kr_query *next = kr_rplan_push(&req->rplan, query->parent, cname, query->sclass, query->stype); + if (!next) { + return KR_STATE_FAIL; + } + next->flags.AWAIT_CUT = true; + + /* Copy transitive flags from original query to CNAME followup. */ + next->flags.TRACE = query->flags.TRACE; + next->flags.ALWAYS_CUT = query->flags.ALWAYS_CUT; + next->flags.NO_MINIMIZE = query->flags.NO_MINIMIZE; + next->flags.NO_THROTTLE = query->flags.NO_THROTTLE; + + if (query->flags.FORWARD) { + next->forward_flags.CNAME = true; + if (query->parent == NULL) { + state = kr_nsrep_copy_set(&next->ns, &query->ns); + if (state != kr_ok()) { + return KR_STATE_FAIL; + } + } + } + next->cname_parent = query; + /* Want DNSSEC if and only if it's posible to secure + * this name (i.e. iff it is covered by a TA) */ + if (kr_ta_covers_qry(req->ctx, cname, query->stype)) { + next->flags.DNSSEC_WANT = true; + } else { + next->flags.DNSSEC_WANT = false; + } + if (!(query->flags.FORWARD) || + (query->flags.DNSSEC_WEXPAND)) { + state = pick_authority(pkt, req, false); + if (state != kr_ok()) { + return KR_STATE_FAIL; + } + } + } else if (!query->parent) { + /* Answer for initial query */ + const bool to_wire = ((pkt_class & (PKT_NXDOMAIN|PKT_NODATA)) != 0); + state = pick_authority(pkt, req, to_wire); + if (state != kr_ok()) { + return KR_STATE_FAIL; + } + finalize_answer(pkt, query, req); + } else { + /* Answer for sub-query; DS, IP for NS etc. + * It may contains NSEC \ NSEC3 records for + * data non-existence or wc expansion proving. + * If yes, they must be validated by validator. + * If no, authority section is unuseful. + * dnssec\nsec.c & dnssec\nsec3.c use + * rrsets from incoming packet. + * validator uses answer_selected & auth_selected. + * So, if nsec\nsec3 records are present in authority, + * pick_authority() must be called. + * TODO refactor nsec\nsec3 modules to work with + * answer_selected & auth_selected instead of incoming pkt. */ + bool auth_is_unuseful = true; + const knot_pktsection_t *ns = knot_pkt_section(pkt, KNOT_AUTHORITY); + for (unsigned i = 0; i < ns->count; ++i) { + const knot_rrset_t *rr = knot_pkt_rr(ns, i); + if (rr->type == KNOT_RRTYPE_NSEC || + rr->type == KNOT_RRTYPE_NSEC3) { + auth_is_unuseful = false; + break; + } + } + if (!auth_is_unuseful) { + state = pick_authority(pkt, req, false); + if (state != kr_ok()) { + return KR_STATE_FAIL; + } + } + } + return KR_STATE_DONE; +} + +/** @internal like process_answer() but for the STUB mode. */ +static int process_stub(knot_pkt_t *pkt, struct kr_request *req) +{ + struct kr_query *query = req->current_query; + assert(query->flags.STUB); + /* Pick all answer RRs. */ + const knot_pktsection_t *an = knot_pkt_section(pkt, KNOT_ANSWER); + for (unsigned i = 0; i < an->count; ++i) { + const knot_rrset_t *rr = knot_pkt_rr(an, i); + int err = kr_ranked_rrarray_add(&req->answ_selected, rr, + KR_RANK_OMIT | KR_RANK_AUTH, true, query->uid, &req->pool); + /* KR_RANK_AUTH: we don't have the records directly from + * an authoritative source, but we do trust the server and it's + * supposed to only send us authoritative records. */ + if (err != kr_ok()) { + return KR_STATE_FAIL; + } + } + + knot_wire_set_aa(pkt->wire); + query->flags.RESOLVED = true; + /* Pick authority RRs. */ + int pkt_class = kr_response_classify(pkt); + const bool to_wire = ((pkt_class & (PKT_NXDOMAIN|PKT_NODATA)) != 0); + int err = pick_authority(pkt, req, to_wire); + if (err != kr_ok()) { + return KR_STATE_FAIL; + } + + finalize_answer(pkt, query, req); + return KR_STATE_DONE; +} + + +/** Error handling, RFC1034 5.3.3, 4d. */ +static int resolve_error(knot_pkt_t *pkt, struct kr_request *req) +{ + return KR_STATE_FAIL; +} + +/* State-less single resolution iteration step, not needed. */ +static int reset(kr_layer_t *ctx) { return KR_STATE_PRODUCE; } + +/* Set resolution context and parameters. */ +static int begin(kr_layer_t *ctx) +{ + if (ctx->state & (KR_STATE_DONE|KR_STATE_FAIL)) { + return ctx->state; + } + /* + * RFC7873 5.4 extends the QUERY operation code behaviour in order to + * be able to generate requests for server cookies. Such requests have + * QDCOUNT equal to zero and must contain a cookie option. + * Server cookie queries must be handled by the cookie module/layer + * before this layer. + */ + const knot_pkt_t *pkt = ctx->req->qsource.packet; + if (!pkt || knot_wire_get_qdcount(pkt->wire) == 0) { + return KR_STATE_FAIL; + } + + struct kr_query *qry = ctx->req->current_query; + /* Avoid any other classes, and avoid any meta-types ~~except for ANY~~. */ + if (qry->sclass != KNOT_CLASS_IN + || (knot_rrtype_is_metatype(qry->stype) + /* && qry->stype != KNOT_RRTYPE_ANY hmm ANY seems broken ATM */)) { + knot_wire_set_rcode(ctx->req->answer->wire, KNOT_RCODE_NOTIMPL); + return KR_STATE_FAIL; + } + + return reset(ctx); +} + +int kr_make_query(struct kr_query *query, knot_pkt_t *pkt) +{ + /* Minimize QNAME (if possible). */ + uint16_t qtype = query->stype; + const knot_dname_t *qname = minimized_qname(query, &qtype); + + /* Form a query for the authoritative. */ + knot_pkt_clear(pkt); + int ret = knot_pkt_put_question(pkt, qname, query->sclass, qtype); + if (ret != KNOT_EOK) { + return ret; + } + + /* Query built, expect answer. */ + query->id = kr_rand_bytes(2); + /* We must respect https://tools.ietf.org/html/rfc7766#section-6.2.1 + * - When sending multiple queries over a TCP connection, clients MUST NOT + * reuse the DNS Message ID of an in-flight query on that connection. + * + * So, if query is going to be sent over TCP connection + * this id can be changed to avoid duplication with query that already was sent + * but didn't receive answer yet. + */ + knot_wire_set_id(pkt->wire, query->id); + pkt->parsed = pkt->size; + + return kr_ok(); +} + +static int prepare_query(kr_layer_t *ctx, knot_pkt_t *pkt) +{ + assert(pkt && ctx); + struct kr_request *req = ctx->req; + struct kr_query *query = req->current_query; + if (!query || ctx->state & (KR_STATE_DONE|KR_STATE_FAIL)) { + return ctx->state; + } + + /* Make query */ + int ret = kr_make_query(query, pkt); + if (ret != 0) { + return KR_STATE_FAIL; + } + + WITH_VERBOSE(query) { + KR_DNAME_GET_STR(name_str, query->sname); + KR_RRTYPE_GET_STR(type_str, query->stype); + QVERBOSE_MSG(query, "'%s' type '%s' new uid was assigned .%02u, parent uid .%02u\n", + name_str, type_str, req->rplan.next_uid, + query->parent ? query->parent->uid : 0); + } + + query->uid = req->rplan.next_uid; + req->rplan.next_uid += 1; + + return KR_STATE_CONSUME; +} + +static int resolve_badmsg(knot_pkt_t *pkt, struct kr_request *req, struct kr_query *query) +{ + +#ifndef STRICT_MODE + /* Work around broken auths/load balancers */ + if (query->flags.SAFEMODE) { + return resolve_error(pkt, req); + } else if (query->flags.NO_MINIMIZE) { + query->flags.SAFEMODE = true; + return KR_STATE_DONE; + } else { + query->flags.NO_MINIMIZE = true; + return KR_STATE_DONE; + } +#else + return resolve_error(pkt, req); +#endif +} + +static int resolve_notimpl(knot_pkt_t *pkt, struct kr_request *req, struct kr_query *qry) +{ + if (qry->stype == KNOT_RRTYPE_RRSIG && qry->parent != NULL) { + /* RRSIG subquery have got NOTIMPL. + * Possible scenario - same NS is autoritative for child and parent, + * but child isn't signed. + * We got delegation to parent, + * then NS responded as NS for child zone. + * Answer contained record been requested, but no RRSIGs, + * Validator issued RRSIG query then. If qname is zone name, + * we can get NOTIMPL. Ask for DS to find out security status. + * TODO - maybe it would be better to do this in validator, when + * RRSIG revalidation occurs. + */ + struct kr_rplan *rplan = &req->rplan; + struct kr_query *next = kr_rplan_push(rplan, qry->parent, qry->sname, + qry->sclass, KNOT_RRTYPE_DS); + if (!next) { + return KR_STATE_FAIL; + } + kr_zonecut_set(&next->zone_cut, qry->parent->zone_cut.name); + kr_zonecut_copy(&next->zone_cut, &qry->parent->zone_cut); + kr_zonecut_copy_trust(&next->zone_cut, &qry->parent->zone_cut); + next->flags.DNSSEC_WANT = true; + qry->flags.RESOLVED = true; + return KR_STATE_DONE; + } + return resolve_badmsg(pkt, req, qry); +} + +/** Resolve input query or continue resolution with followups. + * + * This roughly corresponds to RFC1034, 5.3.3 4a-d. + */ +static int resolve(kr_layer_t *ctx, knot_pkt_t *pkt) +{ + assert(pkt && ctx); + struct kr_request *req = ctx->req; + struct kr_query *query = req->current_query; + if (!query) { + return ctx->state; + } + + WITH_VERBOSE(query) { + if (query->flags.TRACE) { + auto_free char *pkt_text = kr_pkt_text(pkt); + VERBOSE_MSG("<= answer received: \n%s\n", pkt_text); + } + } + + if (query->flags.RESOLVED || query->flags.BADCOOKIE_AGAIN) { + return ctx->state; + } + + /* Check for packet processing errors first. + * Note - we *MUST* check if it has at least a QUESTION, + * otherwise it would crash on accessing QNAME. */ +#ifdef STRICT_MODE + if (pkt->parsed < pkt->size) { + VERBOSE_MSG("<= pkt contains excessive data\n"); + return resolve_badmsg(pkt, req, query); + } else +#endif + if (pkt->parsed <= KNOT_WIRE_HEADER_SIZE) { + VERBOSE_MSG("<= malformed response\n"); + return resolve_badmsg(pkt, req, query); + } else if (!is_paired_to_query(pkt, query)) { + VERBOSE_MSG("<= ignoring mismatching response\n"); + /* Force TCP, to work around authoritatives messing up question + * without yielding to spoofed responses. */ + query->flags.TCP = true; + return resolve_badmsg(pkt, req, query); + } else if (knot_wire_get_tc(pkt->wire)) { + VERBOSE_MSG("<= truncated response, failover to TCP\n"); + if (query) { + /* Fail if already on TCP. */ + if (query->flags.TCP) { + VERBOSE_MSG("<= TC=1 with TCP, bailing out\n"); + return resolve_error(pkt, req); + } + query->flags.TCP = true; + } + return KR_STATE_CONSUME; + } + +#ifndef NOVERBOSELOG + const knot_lookup_t *rcode = knot_lookup_by_id(knot_rcode_names, knot_wire_get_rcode(pkt->wire)); +#endif + + /* Check response code. */ + switch(knot_wire_get_rcode(pkt->wire)) { + case KNOT_RCODE_NOERROR: + case KNOT_RCODE_NXDOMAIN: + break; /* OK */ + case KNOT_RCODE_REFUSED: + case KNOT_RCODE_SERVFAIL: { + if (query->flags.STUB) { + /* Pass through in stub mode */ + break; + } + VERBOSE_MSG("<= rcode: %s\n", rcode ? rcode->name : "??"); + query->fails += 1; + if (query->fails >= KR_QUERY_NSRETRY_LIMIT) { + query->fails = 0; /* Reset per-query counter. */ + return resolve_error(pkt, req); + } else { + if (!query->flags.FORWARD) { + query->flags.NO_MINIMIZE = true; /* Drop minimisation as a safe-guard. */ + } + return KR_STATE_CONSUME; + } + } + case KNOT_RCODE_FORMERR: + VERBOSE_MSG("<= rcode: %s\n", rcode ? rcode->name : "??"); + return resolve_badmsg(pkt, req, query); + case KNOT_RCODE_NOTIMPL: + VERBOSE_MSG("<= rcode: %s\n", rcode ? rcode->name : "??"); + return resolve_notimpl(pkt, req, query); + default: + VERBOSE_MSG("<= rcode: %s\n", rcode ? rcode->name : "??"); + return resolve_error(pkt, req); + } + + /* Forwarding/stub mode is special. */ + if (query->flags.STUB) { + return process_stub(pkt, req); + } + + /* Resolve authority to see if it's referral or authoritative. */ + int state = process_authority(pkt, req); + switch(state) { + case KR_STATE_CONSUME: /* Not referral, process answer. */ + VERBOSE_MSG("<= rcode: %s\n", rcode ? rcode->name : "??"); + state = process_answer(pkt, req); + break; + case KR_STATE_DONE: /* Referral */ + state = process_referral_answer(pkt,req); + VERBOSE_MSG("<= referral response, follow\n"); + break; + default: + break; + } + + return state; +} + +/** Module implementation. */ +const kr_layer_api_t *iterate_layer(struct kr_module *module) +{ + static const kr_layer_api_t _layer = { + .begin = &begin, + .reset = &reset, + .consume = &resolve, + .produce = &prepare_query + }; + return &_layer; +} + +KR_MODULE_EXPORT(iterate) + +#undef VERBOSE_MSG diff --git a/lib/layer/iterate.h b/lib/layer/iterate.h new file mode 100644 index 0000000..a9395bf --- /dev/null +++ b/lib/layer/iterate.h @@ -0,0 +1,36 @@ +/* Copyright (C) 2014-2017 CZ.NIC, z.s.p.o. + + This program is free software: you can redistribute it and/or modify + it under the terms of the GNU General Public License as published by + the Free Software Foundation, either version 3 of the License, or + (at your option) any later version. + + This program is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the + GNU General Public License for more details. + + You should have received a copy of the GNU General Public License + along with this program. If not, see . + */ + +#pragma once + +#include "lib/layer.h" +#include "lib/rplan.h" + +/* Packet classification. */ +enum { + PKT_NOERROR = 1 << 0, /* Positive response */ + PKT_NODATA = 1 << 1, /* No data response */ + PKT_NXDOMAIN = 1 << 2, /* Negative response */ + PKT_REFUSED = 1 << 3, /* Refused response */ + PKT_ERROR = 1 << 4 /* Bad message */ +}; + +/** Classify response by type. */ +int kr_response_classify(const knot_pkt_t *pkt); + +/** Make next iterative query. */ +KR_EXPORT +int kr_make_query(struct kr_query *query, knot_pkt_t *pkt); diff --git a/lib/layer/validate.c b/lib/layer/validate.c new file mode 100644 index 0000000..55c3ad7 --- /dev/null +++ b/lib/layer/validate.c @@ -0,0 +1,1133 @@ +/* Copyright (C) 2014-2017 CZ.NIC, z.s.p.o. + + This program is free software: you can redistribute it and/or modify + it under the terms of the GNU General Public License as published by + the Free Software Foundation, either version 3 of the License, or + (at your option) any later version. + + This program is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the + GNU General Public License for more details. + + You should have received a copy of the GNU General Public License + along with this program. If not, see . + */ + +#include +#include +#include +#include +#include + +#include +#include +#include +#include +#include + +#include "lib/dnssec/nsec.h" +#include "lib/dnssec/nsec3.h" +#include "lib/dnssec.h" +#include "lib/layer.h" +#include "lib/resolve.h" +#include "lib/rplan.h" +#include "lib/utils.h" +#include "lib/defines.h" +#include "lib/module.h" + +#define VERBOSE_MSG(qry, ...) QRVERBOSE(qry, "vldr", __VA_ARGS__) + +#define MAX_REVALIDATION_CNT 2 + +/** + * Search in section for given type. + * @param sec Packet section. + * @param type Type to search for. + * @return True if found. + */ +static bool section_has_type(const knot_pktsection_t *sec, uint16_t type) +{ + if (!sec) { + return false; + } + + for (unsigned i = 0; i < sec->count; ++i) { + const knot_rrset_t *rr = knot_pkt_rr(sec, i); + if (rr->type == type) { + return true; + } + } + + return false; +} + +static bool pkt_has_type(const knot_pkt_t *pkt, uint16_t type) +{ + if (!pkt) { + return false; + } + + if (section_has_type(knot_pkt_section(pkt, KNOT_ANSWER), type)) { + return true; + } + if (section_has_type(knot_pkt_section(pkt, KNOT_AUTHORITY), type)) { + return true; + } + return section_has_type(knot_pkt_section(pkt, KNOT_ADDITIONAL), type); +} + +static void log_bogus_rrsig(kr_rrset_validation_ctx_t *vctx, const struct kr_query *qry, + const knot_rrset_t *rr, const char *msg) { + WITH_VERBOSE(qry) { + auto_free char *name_text = kr_dname_text(rr->owner); + auto_free char *type_text = kr_rrtype_text(rr->type); + VERBOSE_MSG(qry, ">< %s: %s %s " + "(%u matching RRSIGs, %u expired, %u not yet valid, " + "%u invalid signer, %u invalid label count, %u invalid key, " + "%u invalid crypto, %u invalid NSEC)\n", + msg, name_text, type_text, vctx->rrs_counters.matching_name_type, + vctx->rrs_counters.expired, vctx->rrs_counters.notyet, + vctx->rrs_counters.signer_invalid, vctx->rrs_counters.labels_invalid, + vctx->rrs_counters.key_invalid, vctx->rrs_counters.crypto_invalid, + vctx->rrs_counters.nsec_invalid); + } +} + +static int validate_section(kr_rrset_validation_ctx_t *vctx, const struct kr_query *qry, + knot_mm_t *pool) +{ + if (!vctx) { + return kr_error(EINVAL); + } + + /* Can't use qry->zone_cut.name directly, as this name can + * change when updating cut information before validation. + */ + vctx->zone_name = vctx->keys ? vctx->keys->owner : NULL; + + int validation_result = 0; + for (ssize_t i = 0; i < vctx->rrs->len; ++i) { + ranked_rr_array_entry_t *entry = vctx->rrs->at[i]; + const knot_rrset_t *rr = entry->rr; + + if (entry->yielded || vctx->qry_uid != entry->qry_uid) { + continue; + } + + if (kr_rank_test(entry->rank, KR_RANK_OMIT) + || kr_rank_test(entry->rank, KR_RANK_SECURE)) { + continue; /* these are already OK */ + } + + if (rr->type == KNOT_RRTYPE_RRSIG) { + const knot_dname_t *signer_name = knot_rrsig_signer_name(rr->rrs.rdata); + if (!knot_dname_is_equal(vctx->zone_name, signer_name)) { + kr_rank_set(&entry->rank, KR_RANK_MISMATCH); + vctx->err_cnt += 1; + break; + } + kr_rank_set(&entry->rank, KR_RANK_OMIT); + continue; + } + + uint8_t rank_orig = entry->rank; + validation_result = kr_rrset_validate(vctx, rr); + if (validation_result == kr_ok()) { + kr_rank_set(&entry->rank, KR_RANK_SECURE); + + } else if (kr_rank_test(rank_orig, KR_RANK_TRY)) { + log_bogus_rrsig(vctx, qry, rr, + "failed to validate non-authoritative data but continuing"); + vctx->result = kr_ok(); + kr_rank_set(&entry->rank, KR_RANK_TRY); + /* ^^ BOGUS would be more accurate, but it might change + * to MISMATCH on revalidation, e.g. in test val_referral_nods :-/ + */ + + } else if (validation_result == kr_error(ENOENT)) { + /* no RRSIGs found */ + kr_rank_set(&entry->rank, KR_RANK_MISSING); + vctx->err_cnt += 1; + log_bogus_rrsig(vctx, qry, rr, "no valid RRSIGs found"); + } else { + kr_rank_set(&entry->rank, KR_RANK_BOGUS); + vctx->err_cnt += 1; + log_bogus_rrsig(vctx, qry, rr, "bogus signatures"); + } + } + return kr_ok(); +} + +static int validate_records(struct kr_request *req, knot_pkt_t *answer, knot_mm_t *pool, bool has_nsec3) +{ + struct kr_query *qry = req->current_query; + if (!qry->zone_cut.key) { + VERBOSE_MSG(qry, "<= no DNSKEY, can't validate\n"); + return kr_error(EBADMSG); + } + + kr_rrset_validation_ctx_t vctx = { + .pkt = answer, + .rrs = &req->answ_selected, + .section_id = KNOT_ANSWER, + .keys = qry->zone_cut.key, + .zone_name = qry->zone_cut.name, + .timestamp = qry->timestamp.tv_sec, + .qry_uid = qry->uid, + .has_nsec3 = has_nsec3, + .flags = 0, + .err_cnt = 0, + .result = 0 + }; + + int ret = validate_section(&vctx, qry, pool); + req->answ_validated = (vctx.err_cnt == 0); + if (ret != kr_ok()) { + return ret; + } + + uint32_t an_flags = vctx.flags; + vctx.rrs = &req->auth_selected; + vctx.section_id = KNOT_AUTHORITY; + vctx.flags = 0; + vctx.err_cnt = 0; + vctx.result = 0; + + ret = validate_section(&vctx, qry, pool); + req->auth_validated = (vctx.err_cnt == 0); + if (ret != kr_ok()) { + return ret; + } + + /* Records were validated. + * If there is wildcard expansion in answer, + * or optout - flag the query. + */ + if (an_flags & KR_DNSSEC_VFLG_WEXPAND) { + qry->flags.DNSSEC_WEXPAND = true; + } + if (an_flags & KR_DNSSEC_VFLG_OPTOUT) { + qry->flags.DNSSEC_OPTOUT = true; + } + + return ret; +} + +static int validate_keyset(struct kr_request *req, knot_pkt_t *answer, bool has_nsec3) +{ + /* Merge DNSKEY records from answer that are below/at current cut. */ + struct kr_query *qry = req->current_query; + bool updated_key = false; + const knot_pktsection_t *an = knot_pkt_section(answer, KNOT_ANSWER); + for (unsigned i = 0; i < an->count; ++i) { + const knot_rrset_t *rr = knot_pkt_rr(an, i); + if (rr->type != KNOT_RRTYPE_DNSKEY + || knot_dname_in_bailiwick(rr->owner, qry->zone_cut.name) < 0) { + continue; + } + /* Merge with zone cut (or replace ancestor key). */ + if (!qry->zone_cut.key || !knot_dname_is_equal(qry->zone_cut.key->owner, rr->owner)) { + qry->zone_cut.key = knot_rrset_copy(rr, qry->zone_cut.pool); + if (!qry->zone_cut.key) { + return kr_error(ENOMEM); + } + updated_key = true; + } else { + int ret = knot_rdataset_merge(&qry->zone_cut.key->rrs, + &rr->rrs, qry->zone_cut.pool); + if (ret != 0) { + knot_rrset_free(qry->zone_cut.key, qry->zone_cut.pool); + qry->zone_cut.key = NULL; + return ret; + } + updated_key = true; + } + } + + /* Check if there's a key for current TA. */ + if (updated_key && !(qry->flags.CACHED)) { + + kr_rrset_validation_ctx_t vctx = { + .pkt = answer, + .rrs = &req->answ_selected, + .section_id = KNOT_ANSWER, + .keys = qry->zone_cut.key, + .zone_name = qry->zone_cut.name, + .timestamp = qry->timestamp.tv_sec, + .qry_uid = qry->uid, + .has_nsec3 = has_nsec3, + .flags = 0, + .result = 0 + }; + int ret = kr_dnskeys_trusted(&vctx, qry->zone_cut.trust_anchor); + if (ret != 0) { + if (ret != kr_error(DNSSEC_INVALID_DS_ALGORITHM) && + ret != kr_error(EAGAIN)) { + log_bogus_rrsig(&vctx, qry, qry->zone_cut.key, "bogus key"); + } + knot_rrset_free(qry->zone_cut.key, qry->zone_cut.pool); + qry->zone_cut.key = NULL; + return ret; + } + + if (vctx.flags & KR_DNSSEC_VFLG_WEXPAND) { + qry->flags.DNSSEC_WEXPAND = true; + } + if (vctx.flags & KR_DNSSEC_VFLG_OPTOUT) { + qry->flags.DNSSEC_OPTOUT = true; + } + + } + return kr_ok(); +} + +static knot_rrset_t *update_ds(struct kr_zonecut *cut, const knot_pktsection_t *sec) +{ + /* Aggregate DS records (if using multiple keys) */ + knot_rrset_t *new_ds = NULL; + for (unsigned i = 0; i < sec->count; ++i) { + const knot_rrset_t *rr = knot_pkt_rr(sec, i); + if (rr->type != KNOT_RRTYPE_DS) { + continue; + } + int ret = 0; + if (new_ds) { + ret = knot_rdataset_merge(&new_ds->rrs, &rr->rrs, cut->pool); + } else { + new_ds = knot_rrset_copy(rr, cut->pool); + if (!new_ds) { + return NULL; + } + } + if (ret != 0) { + knot_rrset_free(new_ds, cut->pool); + return NULL; + } + } + return new_ds; +} + +static void mark_insecure_parents(const struct kr_query *qry) +{ + /* If there is a chain of DS queries mark all of them, + * then mark first non-DS parent. + * Stop if parent is waiting for ns address. + * NS can be located at unsigned zone, but still will return + * valid DNSSEC records for initial query. */ + struct kr_query *parent = qry->parent; + while (parent && !parent->flags.AWAIT_IPV4 && !parent->flags.AWAIT_IPV6) { + parent->flags.DNSSEC_WANT = false; + parent->flags.DNSSEC_INSECURE = true; + if (parent->stype != KNOT_RRTYPE_DS && + parent->stype != KNOT_RRTYPE_RRSIG) { + break; + } + parent = parent->parent; + } +} + +static int update_parent_keys(struct kr_request *req, uint16_t answer_type) +{ + struct kr_query *qry = req->current_query; + struct kr_query *parent = qry->parent; + assert(parent); + switch(answer_type) { + case KNOT_RRTYPE_DNSKEY: + VERBOSE_MSG(qry, "<= parent: updating DNSKEY\n"); + parent->zone_cut.key = knot_rrset_copy(qry->zone_cut.key, parent->zone_cut.pool); + if (!parent->zone_cut.key) { + return KR_STATE_FAIL; + } + break; + case KNOT_RRTYPE_DS: + VERBOSE_MSG(qry, "<= parent: updating DS\n"); + if (qry->flags.DNSSEC_INSECURE) { /* DS non-existence proven. */ + mark_insecure_parents(qry); + } else if (qry->flags.DNSSEC_NODS && !qry->flags.FORWARD) { + if (qry->flags.DNSSEC_OPTOUT) { + mark_insecure_parents(qry); + } else { + int ret = kr_dnssec_matches_name_and_type(&req->auth_selected, qry->uid, + qry->sname, KNOT_RRTYPE_NS); + if (ret == kr_ok()) { + mark_insecure_parents(qry); + } + } + } else if (qry->flags.DNSSEC_NODS && qry->flags.FORWARD) { + int ret = kr_dnssec_matches_name_and_type(&req->auth_selected, qry->uid, + qry->sname, KNOT_RRTYPE_NS); + if (ret == kr_ok()) { + mark_insecure_parents(qry); + } + } else { /* DS existence proven. */ + parent->zone_cut.trust_anchor = knot_rrset_copy(qry->zone_cut.trust_anchor, parent->zone_cut.pool); + if (!parent->zone_cut.trust_anchor) { + return KR_STATE_FAIL; + } + } + break; + default: break; + } + return kr_ok(); +} + +static int update_delegation(struct kr_request *req, struct kr_query *qry, knot_pkt_t *answer, bool has_nsec3) +{ + struct kr_zonecut *cut = &qry->zone_cut; + + /* RFC4035 3.1.4. authoritative must send either DS or proof of non-existence. + * If it contains neither, resolver must query the parent for the DS (RFC4035 5.2.). + * If DS exists, the referral is OK, + * otherwise referral is bogus (or an attempted downgrade attack). + */ + + + unsigned section = KNOT_ANSWER; + const bool referral = !knot_wire_get_aa(answer->wire); + if (referral) { + section = KNOT_AUTHORITY; + } else if (knot_pkt_qtype(answer) == KNOT_RRTYPE_DS && + !(qry->flags.CNAME) && + (knot_wire_get_rcode(answer->wire) != KNOT_RCODE_NXDOMAIN)) { + section = KNOT_ANSWER; + } else { /* N/A */ + return kr_ok(); + } + + int ret = 0; + const knot_dname_t *proved_name = knot_pkt_qname(answer); + /* Aggregate DS records (if using multiple keys) */ + knot_rrset_t *new_ds = update_ds(cut, knot_pkt_section(answer, section)); + if (!new_ds) { + /* No DS provided, check for proof of non-existence. */ + if (!has_nsec3) { + if (referral) { + /* Check if it is referral to unsigned, rfc4035 5.2 */ + ret = kr_nsec_ref_to_unsigned(answer); + } else { + /* No-data answer */ + ret = kr_nsec_existence_denial(answer, KNOT_AUTHORITY, proved_name, KNOT_RRTYPE_DS); + } + } else { + if (referral) { + /* Check if it is referral to unsigned, rfc5155 8.9 */ + ret = kr_nsec3_ref_to_unsigned(answer); + } else { + /* No-data answer, QTYPE is DS, rfc5155 8.6 */ + ret = kr_nsec3_no_data(answer, KNOT_AUTHORITY, proved_name, KNOT_RRTYPE_DS); + } + if (ret == kr_error(KNOT_ERANGE)) { + /* Not bogus, going insecure due to optout */ + ret = 0; + } + } + + if (referral && qry->stype != KNOT_RRTYPE_DS && + ret == DNSSEC_NOT_FOUND) { + /* referral, + * qtype is not KNOT_RRTYPE_DS, NSEC\NSEC3 were not found. + * Check if DS already was fetched. */ + knot_rrset_t *ta = cut->trust_anchor; + if (knot_dname_is_equal(cut->name, ta->owner)) { + /* DS is OK */ + ret = 0; + } + } else if (ret != 0) { + VERBOSE_MSG(qry, "<= bogus proof of DS non-existence\n"); + qry->flags.DNSSEC_BOGUS = true; + } else if (proved_name[0] != '\0') { /* don't go to insecure for . DS */ + VERBOSE_MSG(qry, "<= DS doesn't exist, going insecure\n"); + qry->flags.DNSSEC_NODS = true; + /* Rank the corresponding nonauth NS as insecure. */ + for (int i = 0; i < req->auth_selected.len; ++i) { + ranked_rr_array_entry_t *ns = req->auth_selected.at[i]; + if (ns->qry_uid != qry->uid + || !ns->rr + || ns->rr->type != KNOT_RRTYPE_NS) { + continue; + } + if (!referral && !knot_dname_is_equal(qry->sname, ns->rr->owner)) { + continue; + } + /* Found the record. Note: this is slightly fragile + * in case there were more NS records in the packet. + * As it is now for referrals, kr_nsec*_ref_to_unsigned consider + * (only) the first NS record in the packet. */ + if (!kr_rank_test(ns->rank, KR_RANK_AUTH)) { /* sanity */ + ns->rank = KR_RANK_INSECURE; + } + break; + } + } + return ret; + } else if (qry->flags.FORWARD && qry->parent) { + struct kr_query *parent = qry->parent; + parent->zone_cut.name = knot_dname_copy(qry->sname, parent->zone_cut.pool); + } + + /* Extend trust anchor */ + VERBOSE_MSG(qry, "<= DS: OK\n"); + cut->trust_anchor = new_ds; + return ret; +} + +static const knot_dname_t *find_first_signer(ranked_rr_array_t *arr) +{ + for (size_t i = 0; i < arr->len; ++i) { + ranked_rr_array_entry_t *entry = arr->at[i]; + const knot_rrset_t *rr = entry->rr; + if (entry->yielded || + (!kr_rank_test(entry->rank, KR_RANK_INITIAL) && + !kr_rank_test(entry->rank, KR_RANK_TRY) && + !kr_rank_test(entry->rank, KR_RANK_MISMATCH))) { + continue; + } + if (rr->type == KNOT_RRTYPE_RRSIG) { + return knot_rrsig_signer_name(rr->rrs.rdata); + } + } + return NULL; +} + +static const knot_dname_t *signature_authority(struct kr_request *req) +{ + const knot_dname_t *signer_name = find_first_signer(&req->answ_selected); + if (!signer_name) { + signer_name = find_first_signer(&req->auth_selected); + } + return signer_name; +} + +static int rrsig_not_found(kr_layer_t *ctx, const knot_rrset_t *rr) +{ + struct kr_request *req = ctx->req; + struct kr_query *qry = req->current_query; + + /* Parent-side record, so don't ask for RRSIG. + * We won't receive it anyway. */ + if (qry->stype == KNOT_RRTYPE_DS) { + return KR_STATE_FAIL; + } + + struct kr_zonecut *cut = &qry->zone_cut; + const knot_dname_t *cut_name_start = qry->zone_cut.name; + bool use_cut = true; + if (knot_dname_in_bailiwick(rr->owner, cut_name_start) < 0) { + int zone_labels = knot_dname_labels(qry->zone_cut.name, NULL); + int matched_labels = knot_dname_matched_labels(qry->zone_cut.name, rr->owner); + int skip_labels = zone_labels - matched_labels; + while (skip_labels--) { + cut_name_start = knot_wire_next_label(cut_name_start, NULL); + } + /* try to find the name wanted among ancestors */ + use_cut = false; + while (cut->parent) { + cut = cut->parent; + if (knot_dname_is_equal(cut_name_start, cut->name)) { + use_cut = true; + break; + } + }; + } + struct kr_rplan *rplan = &req->rplan; + struct kr_query *next = kr_rplan_push(rplan, qry, rr->owner, rr->rclass, KNOT_RRTYPE_RRSIG); + if (!next) { + return KR_STATE_FAIL; + } + kr_zonecut_init(&next->zone_cut, cut_name_start, &req->pool); + if (use_cut) { + kr_zonecut_copy(&next->zone_cut, cut); + kr_zonecut_copy_trust(&next->zone_cut, cut); + } else { + next->flags.AWAIT_CUT = true; + } + if (qry->flags.FORWARD) { + next->flags.AWAIT_CUT = false; + } + next->flags.DNSSEC_WANT = true; + return KR_STATE_YIELD; +} + +static int check_validation_result(kr_layer_t *ctx, ranked_rr_array_t *arr) +{ + int ret = KR_STATE_DONE; + struct kr_request *req = ctx->req; + struct kr_query *qry = req->current_query; + ranked_rr_array_entry_t *invalid_entry = NULL; + for (size_t i = 0; i < arr->len; ++i) { + ranked_rr_array_entry_t *entry = arr->at[i]; + if (entry->yielded || entry->qry_uid != qry->uid) { + continue; + } + if (kr_rank_test(entry->rank, KR_RANK_MISMATCH)) { + invalid_entry = entry; + break; + } else if (kr_rank_test(entry->rank, KR_RANK_MISSING) && + !invalid_entry) { + invalid_entry = entry; + } else if (kr_rank_test(entry->rank, KR_RANK_OMIT)) { + continue; + } else if (!kr_rank_test(entry->rank, KR_RANK_SECURE) && + !invalid_entry) { + invalid_entry = entry; + } + } + + if (!invalid_entry) { + return ret; + } + + if (!kr_rank_test(invalid_entry->rank, KR_RANK_SECURE) && + (++(invalid_entry->revalidation_cnt) > MAX_REVALIDATION_CNT)) { + VERBOSE_MSG(qry, "<= continuous revalidation, fails\n"); + qry->flags.DNSSEC_BOGUS = true; + return KR_STATE_FAIL; + } + + const knot_rrset_t *rr = invalid_entry->rr; + if (kr_rank_test(invalid_entry->rank, KR_RANK_MISMATCH)) { + const knot_dname_t *signer_name = knot_rrsig_signer_name(rr->rrs.rdata); + if (knot_dname_in_bailiwick(signer_name, qry->zone_cut.name) > 0) { + qry->zone_cut.name = knot_dname_copy(signer_name, &req->pool); + qry->flags.AWAIT_CUT = true; + } else if (!knot_dname_is_equal(signer_name, qry->zone_cut.name)) { + if (qry->zone_cut.parent) { + memcpy(&qry->zone_cut, qry->zone_cut.parent, sizeof(qry->zone_cut)); + } else { + qry->flags.AWAIT_CUT = true; + } + qry->zone_cut.name = knot_dname_copy(signer_name, &req->pool); + } + VERBOSE_MSG(qry, ">< cut changed (new signer), needs revalidation\n"); + ret = KR_STATE_YIELD; + } else if (kr_rank_test(invalid_entry->rank, KR_RANK_MISSING)) { + ret = rrsig_not_found(ctx, rr); + } else if (!kr_rank_test(invalid_entry->rank, KR_RANK_SECURE)) { + qry->flags.DNSSEC_BOGUS = true; + ret = KR_STATE_FAIL; + } + + return ret; +} + +static bool check_empty_answer(kr_layer_t *ctx, knot_pkt_t *pkt) +{ + struct kr_request *req = ctx->req; + struct kr_query *qry = req->current_query; + ranked_rr_array_t *arr = &req->answ_selected; + size_t num_entries = 0; + for (size_t i = 0; i < arr->len; ++i) { + ranked_rr_array_entry_t *entry = arr->at[i]; + const knot_rrset_t *rr = entry->rr; + if (rr->type == KNOT_RRTYPE_RRSIG && qry->stype != KNOT_RRTYPE_RRSIG) { + continue; + } + if (entry->qry_uid == qry->uid) { + ++num_entries; + } + } + const knot_pktsection_t *an = knot_pkt_section(pkt, KNOT_ANSWER); + return ((an->count != 0) && (num_entries == 0)) ? false : true; +} + +static int unsigned_forward(kr_layer_t *ctx, knot_pkt_t *pkt) +{ + struct kr_request *req = ctx->req; + struct kr_query *qry = req->current_query; + const uint16_t qtype = knot_pkt_qtype(pkt); + const uint8_t pkt_rcode = knot_wire_get_rcode(pkt->wire); + bool nods = false; + bool ns_exist = true; + for (int i = 0; i < req->rplan.resolved.len; ++i) { + struct kr_query *q = req->rplan.resolved.at[i]; + if (q->sclass == qry->sclass && + q->stype == KNOT_RRTYPE_DS && + knot_dname_is_equal(q->sname, qry->sname)) { + nods = true; + if (!(q->flags.DNSSEC_OPTOUT)) { + int ret = kr_dnssec_matches_name_and_type(&req->auth_selected, q->uid, + qry->sname, KNOT_RRTYPE_NS); + ns_exist = (ret == kr_ok()); + } + } + } + + if (nods && ns_exist && qtype == KNOT_RRTYPE_NS) { + qry->flags.DNSSEC_WANT = false; + qry->flags.DNSSEC_INSECURE = true; + if (qry->forward_flags.CNAME) { + assert(qry->cname_parent); + qry->cname_parent->flags.DNSSEC_WANT = false; + qry->cname_parent->flags.DNSSEC_INSECURE = true; + } else if (pkt_rcode == KNOT_RCODE_NOERROR && qry->parent != NULL) { + const knot_pktsection_t *sec = knot_pkt_section(pkt, KNOT_ANSWER); + const knot_rrset_t *rr = knot_pkt_rr(sec, 0); + if (rr->type == KNOT_RRTYPE_NS) { + qry->parent->zone_cut.name = knot_dname_copy(rr->owner, &req->pool); + qry->parent->flags.DNSSEC_WANT = false; + qry->parent->flags.DNSSEC_INSECURE = true; + } + } + while (qry->parent) { + qry = qry->parent; + qry->flags.DNSSEC_WANT = false; + qry->flags.DNSSEC_INSECURE = true; + if (qry->forward_flags.CNAME) { + assert(qry->cname_parent); + qry->cname_parent->flags.DNSSEC_WANT = false; + qry->cname_parent->flags.DNSSEC_INSECURE = true; + } + } + return KR_STATE_DONE; + } + + if (ctx->state == KR_STATE_YIELD) { + return KR_STATE_DONE; + } + + if (!nods && qtype != KNOT_RRTYPE_DS) { + struct kr_rplan *rplan = &req->rplan; + struct kr_query *next = kr_rplan_push(rplan, qry, qry->sname, qry->sclass, KNOT_RRTYPE_DS); + if (!next) { + return KR_STATE_FAIL; + } + kr_zonecut_set(&next->zone_cut, qry->zone_cut.name); + kr_zonecut_copy_trust(&next->zone_cut, &qry->zone_cut); + next->flags.DNSSEC_WANT = true; + } + + return KR_STATE_YIELD; +} + +static int check_signer(kr_layer_t *ctx, knot_pkt_t *pkt) +{ + struct kr_request *req = ctx->req; + struct kr_query *qry = req->current_query; + const knot_dname_t *ta_name = qry->zone_cut.trust_anchor ? qry->zone_cut.trust_anchor->owner : NULL; + const knot_dname_t *signer = signature_authority(req); + if (ta_name && (!signer || !knot_dname_is_equal(ta_name, signer))) { + /* check all newly added RRSIGs */ + if (!signer) { + if (qry->flags.FORWARD) { + return unsigned_forward(ctx, pkt); + } + /* Not a DNSSEC-signed response. */ + if (ctx->state == KR_STATE_YIELD) { + /* Already yielded for revalidation. + * It means that trust chain is OK and + * transition to INSECURE hasn't occurred. + * Let the validation logic ask about RRSIG. */ + return KR_STATE_DONE; + } + /* Ask parent for DS + * to prove transition to INSECURE. */ + const uint16_t qtype = knot_pkt_qtype(pkt); + const knot_dname_t *qname = knot_pkt_qname(pkt); + if (qtype == KNOT_RRTYPE_NS && + knot_dname_in_bailiwick(qname, qry->zone_cut.name) > 0) { + /* Server is authoritative + * for both parent and child, + * and child zone is not signed. */ + qry->zone_cut.name = knot_dname_copy(qname, &req->pool); + } + } else if (knot_dname_in_bailiwick(signer, qry->zone_cut.name) > 0) { + if (!(qry->flags.FORWARD)) { + /* Key signer is below current cut, advance and refetch keys. */ + qry->zone_cut.name = knot_dname_copy(signer, &req->pool); + } else { + /* Check if DS does not exist. */ + struct kr_query *q = kr_rplan_find_resolved(&req->rplan, NULL, + signer, qry->sclass, KNOT_RRTYPE_DS); + if (q && q->flags.DNSSEC_NODS) { + qry->flags.DNSSEC_WANT = false; + qry->flags.DNSSEC_INSECURE = true; + if (qry->parent) { + qry->parent->flags.DNSSEC_WANT = false; + qry->parent->flags.DNSSEC_INSECURE = true; + } + } else if (qry->stype != KNOT_RRTYPE_DS) { + struct kr_rplan *rplan = &req->rplan; + struct kr_query *next = kr_rplan_push(rplan, qry, qry->sname, + qry->sclass, KNOT_RRTYPE_DS); + if (!next) { + return KR_STATE_FAIL; + } + kr_zonecut_set(&next->zone_cut, qry->zone_cut.name); + kr_zonecut_copy_trust(&next->zone_cut, &qry->zone_cut); + next->flags.DNSSEC_WANT = true; + } + } + } else if (!knot_dname_is_equal(signer, qry->zone_cut.name)) { + /* Key signer is above the current cut, so we can't validate it. This happens when + a server is authoritative for both grandparent, parent and child zone. + Ascend to parent cut, and refetch authority for signer. */ + if (qry->zone_cut.parent) { + memcpy(&qry->zone_cut, qry->zone_cut.parent, sizeof(qry->zone_cut)); + } else { + qry->flags.AWAIT_CUT = true; + } + qry->zone_cut.name = knot_dname_copy(signer, &req->pool); + } + + /* zone cut matches, but DS/DNSKEY doesn't => refetch. */ + VERBOSE_MSG(qry, ">< cut changed, needs revalidation\n"); + if ((qry->flags.FORWARD) && qry->stype != KNOT_RRTYPE_DS) { + struct kr_rplan *rplan = &req->rplan; + struct kr_query *next = kr_rplan_push(rplan, qry, signer, + qry->sclass, KNOT_RRTYPE_DS); + if (!next) { + return KR_STATE_FAIL; + } + kr_zonecut_set(&next->zone_cut, qry->zone_cut.name); + kr_zonecut_copy_trust(&next->zone_cut, &qry->zone_cut); + next->flags.DNSSEC_WANT = true; + return KR_STATE_YIELD; + } + if (!(qry->flags.FORWARD)) { + return KR_STATE_YIELD; + } + } + return KR_STATE_DONE; +} + +/** Change ranks of RRs from this single iteration: + * _INITIAL or _TRY or _MISSING -> rank_to_set. + * + * Optionally do this only in a `bailiwick` (if not NULL). + * Iterator shouldn't have selected such records, but we check to be sure. */ +static void rank_records(kr_layer_t *ctx, enum kr_rank rank_to_set, + const knot_dname_t *bailiwick) +{ + struct kr_request *req = ctx->req; + struct kr_query *qry = req->current_query; + ranked_rr_array_t *ptrs[2] = { &req->answ_selected, &req->auth_selected }; + for (size_t i = 0; i < 2; ++i) { + ranked_rr_array_t *arr = ptrs[i]; + for (size_t j = 0; j < arr->len; ++j) { + ranked_rr_array_entry_t *entry = arr->at[j]; + if (entry->qry_uid != qry->uid) { + continue; + } + if (bailiwick && knot_dname_in_bailiwick(entry->rr->owner, + bailiwick) < 0) { + continue; + } + if (kr_rank_test(entry->rank, KR_RANK_INITIAL) + || kr_rank_test(entry->rank, KR_RANK_TRY) + || kr_rank_test(entry->rank, KR_RANK_MISSING)) { + kr_rank_set(&entry->rank, rank_to_set); + } + } + } +} + +static void check_wildcard(kr_layer_t *ctx) +{ + struct kr_request *req = ctx->req; + struct kr_query *qry = req->current_query; + ranked_rr_array_t *ptrs[2] = { &req->answ_selected, &req->auth_selected }; + + for (int i = 0; i < 2; ++i) { + ranked_rr_array_t *arr = ptrs[i]; + for (ssize_t j = 0; j < arr->len; ++j) { + ranked_rr_array_entry_t *entry = arr->at[j]; + const knot_rrset_t *rrsigs = entry->rr; + + if (qry->uid != entry->qry_uid) { + continue; + } + + if (rrsigs->type != KNOT_RRTYPE_RRSIG) { + continue; + } + + int owner_labels = knot_dname_labels(rrsigs->owner, NULL); + + knot_rdata_t *rdata_k = rrsigs->rrs.rdata; + for (int k = 0; k < rrsigs->rrs.count; + ++k, rdata_k = knot_rdataset_next(rdata_k)) { + if (knot_rrsig_labels(rdata_k) != owner_labels) { + qry->flags.DNSSEC_WEXPAND = true; + } + } + } + } +} + +/** Just for wildcard_adjust_to_wire() */ +static bool rr_is_for_wildcard(const ranked_rr_array_entry_t *entry) +{ + switch (kr_rrset_type_maysig(entry->rr)) { + case KNOT_RRTYPE_NSEC: + case KNOT_RRTYPE_NSEC3: + return true; + default: + return false; + } +} +/** In case of wildcard expansion, mark required authority RRs by to_wire. */ +static int wildcard_adjust_to_wire(struct kr_request *req, const struct kr_query *qry) +{ + if (!qry->parent && qry->flags.DNSSEC_WEXPAND) { + return kr_ranked_rrarray_set_wire(&req->auth_selected, true, + qry->uid, true, &rr_is_for_wildcard); + } + return kr_ok(); +} + +static int validate(kr_layer_t *ctx, knot_pkt_t *pkt) +{ + int ret = 0; + struct kr_request *req = ctx->req; + struct kr_query *qry = req->current_query; + + /* Ignore faulty or unprocessed responses. */ + if (ctx->state & (KR_STATE_FAIL|KR_STATE_CONSUME)) { + return ctx->state; + } + + /* Pass-through if user doesn't want secure answer or stub. */ + /* @todo: Validating stub resolver mode. */ + if (qry->flags.STUB) { + rank_records(ctx, KR_RANK_OMIT, NULL); + return ctx->state; + } + uint8_t pkt_rcode = knot_wire_get_rcode(pkt->wire); + if ((qry->flags.FORWARD) && + pkt_rcode != KNOT_RCODE_NOERROR && + pkt_rcode != KNOT_RCODE_NXDOMAIN) { + do { + qry->flags.DNSSEC_BOGUS = true; + if (qry->cname_parent) { + qry->cname_parent->flags.DNSSEC_BOGUS = true; + } + qry = qry->parent; + } while (qry); + ctx->state = KR_STATE_DONE; + return ctx->state; + } + + if (!(qry->flags.DNSSEC_WANT)) { + const bool is_insec = qry->flags.CACHED && qry->flags.DNSSEC_INSECURE; + if ((qry->flags.DNSSEC_INSECURE)) { + rank_records(ctx, KR_RANK_INSECURE, qry->zone_cut.name); + } + if (is_insec && qry->parent != NULL) { + /* We have got insecure answer from cache. + * Mark parent(s) as insecure. */ + mark_insecure_parents(qry); + VERBOSE_MSG(qry, "<= cached insecure response, going insecure\n"); + ctx->state = KR_STATE_DONE; + } else if (ctx->state == KR_STATE_YIELD) { + /* Transition to unsecure state + occurred during revalidation. + if state remains YIELD, answer will not be cached. + Let cache layers to work. */ + ctx->state = KR_STATE_DONE; + } + return ctx->state; + } + + /* Pass-through if CD bit is set. */ + if (knot_wire_get_cd(req->qsource.packet->wire)) { + check_wildcard(ctx); + wildcard_adjust_to_wire(req, qry); + rank_records(ctx, KR_RANK_OMIT, NULL); + return ctx->state; + } + /* Answer for RRSIG may not set DO=1, but all records MUST still validate. */ + bool use_signatures = (knot_pkt_qtype(pkt) != KNOT_RRTYPE_RRSIG); + if (!(qry->flags.CACHED) && !knot_pkt_has_dnssec(pkt) && !use_signatures) { + VERBOSE_MSG(qry, "<= got insecure response\n"); + qry->flags.DNSSEC_BOGUS = true; + return KR_STATE_FAIL; + } + + /* Check if this is a DNSKEY answer, check trust chain and store. */ + uint16_t qtype = knot_pkt_qtype(pkt); + bool has_nsec3 = pkt_has_type(pkt, KNOT_RRTYPE_NSEC3); + const knot_pktsection_t *an = knot_pkt_section(pkt, KNOT_ANSWER); + const bool referral = (an->count == 0 && !knot_wire_get_aa(pkt->wire)); + + if (!(qry->flags.CACHED) && knot_wire_get_aa(pkt->wire)) { + /* Check if answer if not empty, + * but iterator has not selected any records. */ + if (!check_empty_answer(ctx, pkt)) { + VERBOSE_MSG(qry, "<= no useful RR in authoritative answer\n"); + qry->flags.DNSSEC_BOGUS = true; + return KR_STATE_FAIL; + } + /* Track difference between current TA and signer name. + * This indicates that the NS is auth for both parent-child, + * and we must update DS/DNSKEY to validate it. + */ + ret = check_signer(ctx, pkt); + if (ret != KR_STATE_DONE) { + return ret; + } + if (qry->flags.FORWARD && qry->flags.DNSSEC_INSECURE) { + return KR_STATE_DONE; + } + } + + if (knot_wire_get_aa(pkt->wire) && qtype == KNOT_RRTYPE_DNSKEY) { + ret = validate_keyset(req, pkt, has_nsec3); + if (ret == kr_error(EAGAIN)) { + VERBOSE_MSG(qry, ">< cut changed, needs revalidation\n"); + return KR_STATE_YIELD; + } else if (ret == kr_error(DNSSEC_INVALID_DS_ALGORITHM)) { + VERBOSE_MSG(qry, ">< all DS entries use unsupported algorithm pairs, going insecure\n"); + /* ^ the message is a bit imprecise to avoid being too verbose */ + qry->flags.DNSSEC_WANT = false; + qry->flags.DNSSEC_INSECURE = true; + rank_records(ctx, KR_RANK_INSECURE, qry->zone_cut.name); + mark_insecure_parents(qry); + return KR_STATE_DONE; + } else if (ret != 0) { + VERBOSE_MSG(qry, "<= bad keys, broken trust chain\n"); + qry->flags.DNSSEC_BOGUS = true; + return KR_STATE_FAIL; + } + } + + /* Validate non-existence proof if not positive answer. + * In case of CNAME, iterator scheduled a sibling query for the target, + * so we just drop the negative piece of information and don't try to prove it. + * TODO: not ideal; with aggressive cache we'll at least avoid the extra packet. */ + if (!qry->flags.CACHED && pkt_rcode == KNOT_RCODE_NXDOMAIN && !qry->flags.CNAME) { + /* @todo If knot_pkt_qname(pkt) is used instead of qry->sname then the tests crash. */ + if (!has_nsec3) { + ret = kr_nsec_name_error_response_check(pkt, KNOT_AUTHORITY, qry->sname); + } else { + ret = kr_nsec3_name_error_response_check(pkt, KNOT_AUTHORITY, qry->sname); + } + if (has_nsec3 && (ret == kr_error(KNOT_ERANGE))) { + /* NXDOMAIN proof is OK, + * but NSEC3 that covers next closer name + * (or wildcard at next closer name) has opt-out flag. + * RFC5155 9.2; AD flag can not be set */ + qry->flags.DNSSEC_OPTOUT = true; + VERBOSE_MSG(qry, "<= can't prove NXDOMAIN due to optout, going insecure\n"); + } else if (ret != 0) { + VERBOSE_MSG(qry, "<= bad NXDOMAIN proof\n"); + qry->flags.DNSSEC_BOGUS = true; + return KR_STATE_FAIL; + } + } + + /* @todo WTH, this needs API that just tries to find a proof and the caller + * doesn't have to worry about NSEC/NSEC3 + * @todo rework this + * CNAME: same as the NXDOMAIN case above */ + if (!qry->flags.CACHED && pkt_rcode == KNOT_RCODE_NOERROR && !qry->flags.CNAME) { + bool no_data = (an->count == 0 && knot_wire_get_aa(pkt->wire)); + if (no_data) { + /* @todo + * ? quick mechanism to determine which check to preform first + * ? merge the functionality together to share code/resources + */ + if (!has_nsec3) { + ret = kr_nsec_existence_denial(pkt, KNOT_AUTHORITY, knot_pkt_qname(pkt), knot_pkt_qtype(pkt)); + } else { + ret = kr_nsec3_no_data(pkt, KNOT_AUTHORITY, knot_pkt_qname(pkt), knot_pkt_qtype(pkt)); + } + if (ret != 0) { + if (has_nsec3 && (ret == kr_error(KNOT_ERANGE))) { + VERBOSE_MSG(qry, "<= can't prove NODATA due to optout, going insecure\n"); + qry->flags.DNSSEC_OPTOUT = true; + /* Could not return from here, + * we must continue, validate NSEC\NSEC3 and + * call update_parent_keys() to mark + * parent queries as insecured */ + } else { + VERBOSE_MSG(qry, "<= bad NODATA proof\n"); + qry->flags.DNSSEC_BOGUS = true; + return KR_STATE_FAIL; + } + } + } + } + + /* Validate all records, fail as bogus if it doesn't match. + * Do not revalidate data from cache, as it's already trusted. */ + if (!(qry->flags.CACHED)) { + ret = validate_records(req, pkt, req->rplan.pool, has_nsec3); + if (ret != 0) { + /* something exceptional - no DNS key, empty pointers etc + * normally it shoudn't happen */ + VERBOSE_MSG(qry, "<= couldn't validate RRSIGs\n"); + qry->flags.DNSSEC_BOGUS = true; + return KR_STATE_FAIL; + } + /* check validation state and spawn subrequests */ + if (!req->answ_validated) { + ret = check_validation_result(ctx, &req->answ_selected); + if (ret != KR_STATE_DONE) { + return ret; + } + } + if (!req->auth_validated) { + ret = check_validation_result(ctx, &req->auth_selected); + if (ret != KR_STATE_DONE) { + return ret; + } + } + } + + wildcard_adjust_to_wire(req, qry); + + /* Check and update current delegation point security status. */ + ret = update_delegation(req, qry, pkt, has_nsec3); + if (ret == DNSSEC_NOT_FOUND && qry->stype != KNOT_RRTYPE_DS) { + if (ctx->state == KR_STATE_YIELD) { + VERBOSE_MSG(qry, "<= can't validate referral\n"); + qry->flags.DNSSEC_BOGUS = true; + return KR_STATE_FAIL; + } else { + /* Check the trust chain and query DS\DNSKEY if needed. */ + VERBOSE_MSG(qry, "<= DS\\NSEC was not found, querying for DS\n"); + return KR_STATE_YIELD; + } + } else if (ret != 0) { + return KR_STATE_FAIL; + } else if (pkt_rcode == KNOT_RCODE_NOERROR && + referral && + ((!qry->flags.DNSSEC_WANT && qry->flags.DNSSEC_INSECURE) || + (qry->flags.DNSSEC_NODS))) { + /* referral with proven DS non-existance */ + qtype = KNOT_RRTYPE_DS; + } + /* Update parent query zone cut */ + if (qry->parent) { + if (update_parent_keys(req, qtype) != 0) { + return KR_STATE_FAIL; + } + } + + if (qry->flags.FORWARD && qry->parent) { + if (pkt_rcode == KNOT_RCODE_NXDOMAIN) { + qry->parent->forward_flags.NO_MINIMIZE = true; + } + } + VERBOSE_MSG(qry, "<= answer valid, OK\n"); + return KR_STATE_DONE; +} +/** Module implementation. */ +const kr_layer_api_t *validate_layer(struct kr_module *module) +{ + static const kr_layer_api_t _layer = { + .consume = &validate, + }; + /* Store module reference */ + return &_layer; +} + +int validate_init(struct kr_module *module) +{ + return kr_ok(); +} + +KR_MODULE_EXPORT(validate) + +#undef VERBOSE_MSG diff --git a/lib/lib.mk b/lib/lib.mk new file mode 100644 index 0000000..8ac4e91 --- /dev/null +++ b/lib/lib.mk @@ -0,0 +1,110 @@ +libkres_SOURCES := \ + lib/cache/api.c \ + lib/cache/cdb_lmdb.c \ + lib/cache/entry_list.c \ + lib/cache/entry_pkt.c \ + lib/cache/entry_rr.c \ + lib/cache/knot_pkt.c \ + lib/cache/nsec1.c \ + lib/cache/nsec3.c \ + lib/cache/peek.c \ + lib/dnssec.c \ + lib/dnssec/nsec.c \ + lib/dnssec/nsec3.c \ + lib/dnssec/signature.c \ + lib/dnssec/ta.c \ + lib/generic/lru.c \ + lib/generic/map.c \ + lib/generic/queue.c \ + lib/generic/trie.c \ + lib/layer/cache.c \ + lib/layer/iterate.c \ + lib/layer/validate.c \ + lib/module.c \ + lib/nsrep.c \ + lib/resolve.c \ + lib/rplan.c \ + lib/utils.c \ + lib/zonecut.c + +libkres_HEADERS := \ + lib/cache/api.h \ + lib/cache/cdb_api.h \ + lib/cache/cdb_lmdb.h \ + lib/cache/impl.h \ + lib/defines.h \ + lib/dnssec.h \ + lib/dnssec/nsec.h \ + lib/dnssec/nsec3.h \ + lib/dnssec/signature.h \ + lib/dnssec/ta.h \ + lib/generic/array.h \ + lib/generic/lru.h \ + lib/generic/map.h \ + lib/generic/pack.h \ + lib/generic/queue.h \ + lib/generic/trie.h \ + lib/layer.h \ + lib/layer/iterate.h \ + lib/module.h \ + lib/nsrep.h \ + lib/resolve.h \ + lib/rplan.h \ + lib/utils.h \ + lib/zonecut.h + +# Dependencies +libkres_DEPEND := $(contrib) +libkres_CFLAGS := -fPIC $(lmdb_CFLAGS) +libkres_LIBS := $(contrib_TARGET) $(libknot_LIBS) $(libdnssec_LIBS) $(lmdb_LIBS) \ + $(libuv_LIBS) $(gnutls_LIBS) +libkres_TARGET := -L$(abspath lib) -lkres + +ifeq ($(ENABLE_COOKIES),yes) +libkres_SOURCES += \ + lib/cookies/alg_containers.c \ + lib/cookies/alg_sha.c \ + lib/cookies/helper.c \ + lib/cookies/lru_cache.c \ + lib/cookies/nonce.c + +libkres_HEADERS += \ + lib/cookies/alg_containers.h \ + lib/cookies/alg_sha.h \ + lib/cookies/control.h \ + lib/cookies/helper.h \ + lib/cookies/lru_cache.h \ + lib/cookies/nonce.h + +libkres_LIBS += $(nettle_LIBS) +endif + +# Make library +ifeq ($(BUILDMODE), static) +$(eval $(call make_static,libkres,lib,yes)) +else +$(eval $(call make_lib,libkres,lib,yes,$(ABIVER))) +endif + +# Generate pkg-config file +libkres.pc: + @echo 'prefix='$(PREFIX) > $@ + @echo 'exec_prefix=$${prefix}' >> $@ + @echo 'libdir='$(LIBDIR) >> $@ + @echo 'includedir='$(INCLUDEDIR) >> $@ + @echo 'Name: libkres' >> $@ + @echo 'Description: Knot Resolver library' >> $@ + @echo 'URL: https://www.knot-resolver.cz' >> $@ + @echo 'Version: $(VERSION)' >> $@ + @echo 'Libs: -L$${libdir} -lkres' >> $@ + @echo 'Cflags: -I$${includedir}' >> $@ +libkres-pcinstall: libkres.pc libkres-install + $(INSTALL) -d -m 755 $(DESTDIR)$(PKGCONFIGDIR) + $(INSTALL) -m 644 $< $(DESTDIR)$(PKGCONFIGDIR) + +# Targets +lib: $(libkres) +lib-install: libkres-install libkres-pcinstall +lib-clean: libkres-clean + +.PHONY: lib lib-install lib-clean libkres.pc diff --git a/lib/module.c b/lib/module.c new file mode 100644 index 0000000..68abc84 --- /dev/null +++ b/lib/module.c @@ -0,0 +1,170 @@ +/* Copyright (C) 2015-2017 CZ.NIC, z.s.p.o. + + This program is free software: you can redistribute it and/or modify + it under the terms of the GNU General Public License as published by + the Free Software Foundation, either version 3 of the License, or + (at your option) any later version. + + This program is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the + GNU General Public License for more details. + + You should have received a copy of the GNU General Public License + along with this program. If not, see . + */ + +#include +#include +#include +#include + +#include "lib/defines.h" +#include "lib/utils.h" +#include "lib/module.h" + +/* List of embedded modules */ +const kr_layer_api_t *iterate_layer(struct kr_module *module); +const kr_layer_api_t *validate_layer(struct kr_module *module); +const kr_layer_api_t *cache_layer(struct kr_module *module); +static const struct kr_module embedded_modules[] = { + { "iterate", NULL, NULL, NULL, iterate_layer, NULL, NULL, NULL }, + { "validate", NULL, NULL, NULL, validate_layer, NULL, NULL, NULL }, + { "cache", NULL, NULL, NULL, cache_layer, NULL, NULL, NULL }, +}; + +/** Library extension. */ +#if defined(__APPLE__) + #define LIBEXT ".dylib" +#elif _WIN32 + #define LIBEXT ".dll" +#else + #define LIBEXT ".so" +#endif + + +/** Load prefixed symbol. */ +static void *load_symbol(void *lib, const char *prefix, const char *name) +{ + auto_free char *symbol = kr_strcatdup(2, prefix, name); + return dlsym(lib, symbol); +} + +static int load_library(struct kr_module *module, const char *name, const char *path) +{ + assert(module && name && path); + /* Absolute or relative path (then only library search path is used). */ + auto_free char *lib_path = kr_strcatdup(4, path, "/", name, LIBEXT); + if (lib_path == NULL) { + return kr_error(ENOMEM); + } + + /* Workaround for buggy _fini/__attribute__((destructor)) and dlclose(), + * this keeps the library mapped until the program finishes though. */ + module->lib = dlopen(lib_path, RTLD_NOW | RTLD_NODELETE); + if (module->lib) { + return kr_ok(); + } + + return kr_error(ENOENT); +} + +const struct kr_module * kr_module_embedded(const char *name) +{ + for (unsigned i = 0; i < sizeof(embedded_modules)/sizeof(embedded_modules[0]); ++i) { + if (strcmp(name, embedded_modules[i].name) == 0) + return embedded_modules + i; + } + return NULL; +} + +/** Load C module symbols. */ +static int load_sym_c(struct kr_module *module, uint32_t api_required) +{ + #pragma GCC diagnostic push + #pragma GCC diagnostic ignored "-Wpedantic" /* casts after load_symbol() */ + /* Check if it's embedded first */ + const struct kr_module *embedded = kr_module_embedded(module->name); + if (embedded) { + module->init = embedded->init; + module->deinit = embedded->deinit; + module->config = embedded->config; + module->layer = embedded->layer; + module->props = embedded->props; + return kr_ok(); + } + /* Load dynamic library module */ + auto_free char *m_prefix = kr_strcatdup(2, module->name, "_"); + + /* Check ABI version, return error on mismatch. */ + module_api_cb *api = load_symbol(module->lib, m_prefix, "api"); + if (api == NULL) { + return kr_error(ENOENT); + } + if (api() != api_required) { + return kr_error(ENOTSUP); + } + + /* Load ABI by symbol names. */ + #define ML(symname) module->symname = \ + load_symbol(module->lib, m_prefix, #symname) + ML(init); + ML(deinit); + ML(config); + ML(layer); + ML(props); + #undef ML + + return kr_ok(); + #pragma GCC diagnostic pop +} + +int kr_module_load(struct kr_module *module, const char *name, const char *path) +{ + if (module == NULL || name == NULL) { + return kr_error(EINVAL); + } + + /* Initialize, keep userdata */ + void *data = module->data; + memset(module, 0, sizeof(struct kr_module)); + module->data = data; + module->name = strdup(name); + if (module->name == NULL) { + return kr_error(ENOMEM); + } + + /* Search for module library. */ + if (!path || load_library(module, name, path) != 0) { + module->lib = RTLD_DEFAULT; + } + + /* Try to load module ABI. */ + int ret = load_sym_c(module, KR_MODULE_API); + if (ret == 0 && module->init) { + ret = module->init(module); + } + if (ret != 0) { + kr_module_unload(module); + } + + return ret; +} + +void kr_module_unload(struct kr_module *module) +{ + if (module == NULL) { + return; + } + + if (module->deinit) { + module->deinit(module); + } + + if (module->lib && module->lib != RTLD_DEFAULT) { + dlclose(module->lib); + } + + free(module->name); + memset(module, 0, sizeof(struct kr_module)); +} diff --git a/lib/module.h b/lib/module.h new file mode 100644 index 0000000..295c1f9 --- /dev/null +++ b/lib/module.h @@ -0,0 +1,111 @@ +/* Copyright (C) 2015-2017 CZ.NIC, z.s.p.o. + + This program is free software: you can redistribute it and/or modify + it under the terms of the GNU General Public License as published by + the Free Software Foundation, either version 3 of the License, or + (at your option) any later version. + + This program is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the + GNU General Public License for more details. + + You should have received a copy of the GNU General Public License + along with this program. If not, see . + */ + +/** @file + * Module API definition and functions for (un)loading modules. + */ + +#pragma once + +#include "lib/defines.h" +#include "lib/utils.h" +#include "lib/layer.h" + +struct kr_module; +struct kr_prop; + + +/** + * Export module API version (place this at the end of your module). + * + * @param module module name (f.e. hints) + */ +#define KR_MODULE_EXPORT(module) \ + KR_EXPORT uint32_t module ## _api() { return KR_MODULE_API; } +#define KR_MODULE_API ((uint32_t) 0x20180401) + +typedef uint32_t (module_api_cb)(void); + + +/** + * Module representation. + * + * The five symbols (init, ...) may be defined by the module as name_init(), etc; + * all are optional and missing symbols are represented as NULLs; + */ +struct kr_module { + char *name; + + /** Constructor. Called after loading the module. @return error code. */ + int (*init)(struct kr_module *self); + /** Destructor. Called before unloading the module. @return error code. */ + int (*deinit)(struct kr_module *self); + /** Configure with encoded JSON (NULL if missing). @return error code. */ + int (*config)(struct kr_module *self, const char *input); + /** Get a pointer to packet processing API specs. See docs on that type. */ + const kr_layer_api_t * (*layer)(struct kr_module *self); + /** Get a pointer to list of properties, terminated by { NULL, NULL, NULL }. */ + const struct kr_prop * (*props)(void); + + void *lib; /**< Shared library handle or RTLD_DEFAULT */ + void *data; /**< Custom data context. */ +}; + +/** + * Module property callback. Input and output is passed via a JSON encoded in a string. + * + * @param env pointer to the lua engine, i.e. struct engine *env (TODO: explicit type) + * @param input parameter (NULL if missing/nil on lua level) + * @return a free-form JSON output (malloc-ated) + * @note see l_trampoline() implementation for details about the input/output conversion. + */ +typedef char *(kr_prop_cb)(void *env, struct kr_module *self, const char *input); + +/** + * Module property (named callable). + */ +struct kr_prop { + kr_prop_cb *cb; + const char *name; + const char *info; +}; + + +/** + * Load a C module instance into memory. + * + * @param module module structure + * @param name module name + * @param path module search path + * @return 0 or an error + */ +KR_EXPORT +int kr_module_load(struct kr_module *module, const char *name, const char *path); + +/** + * Unload module instance. + * + * @param module module structure + */ +KR_EXPORT +void kr_module_unload(struct kr_module *module); + +/** + * Get embedded module prototype by name (or NULL). + */ +KR_EXPORT +const struct kr_module * kr_module_embedded(const char *name); + diff --git a/lib/nsrep.c b/lib/nsrep.c new file mode 100644 index 0000000..2af12f2 --- /dev/null +++ b/lib/nsrep.c @@ -0,0 +1,563 @@ +/* Copyright (C) 2014-2017 CZ.NIC, z.s.p.o. + + This program is free software: you can redistribute it and/or modify + it under the terms of the GNU General Public License as published by + the Free Software Foundation, either version 3 of the License, or + (at your option) any later version. + + This program is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the + GNU General Public License for more details. + + You should have received a copy of the GNU General Public License + along with this program. If not, see . + */ + +#include +#include +#include +#include + +#include + +#include "lib/nsrep.h" +#include "lib/rplan.h" +#include "lib/resolve.h" +#include "lib/defines.h" +#include "lib/generic/pack.h" +#include "contrib/ucw/lib.h" + +/** Some built-in unfairness ... */ +#ifndef FAVOUR_IPV6 +#define FAVOUR_IPV6 20 /* 20ms bonus for v6 */ +#endif + +/** @internal Macro to set address structure. */ +#define ADDR_SET(sa, family, addr, len, port) do {\ + memcpy(&sa ## _addr, (addr), (len)); \ + sa ## _family = (family); \ + sa ## _port = htons(port); \ +} while (0) + +/** Update nameserver representation with current name/address pair. */ +static void update_nsrep(struct kr_nsrep *ns, size_t pos, uint8_t *addr, size_t addr_len, int port) +{ + if (addr == NULL) { + ns->addr[pos].ip.sa_family = AF_UNSPEC; + return; + } + + /* Rotate previous addresses to the right. */ + memmove(ns->addr + pos + 1, ns->addr + pos, (KR_NSREP_MAXADDR - pos - 1) * sizeof(ns->addr[0])); + + switch(addr_len) { + case sizeof(struct in_addr): + ADDR_SET(ns->addr[pos].ip4.sin, AF_INET, addr, addr_len, port); break; + case sizeof(struct in6_addr): + ADDR_SET(ns->addr[pos].ip6.sin6, AF_INET6, addr, addr_len, port); break; + default: assert(0); break; + } +} + +static void update_nsrep_set(struct kr_nsrep *ns, const knot_dname_t *name, uint8_t *addr[], unsigned score) +{ + /* NSLIST is not empty, empty NS cannot be a leader. */ + if (!addr[0] && ns->addr[0].ip.sa_family != AF_UNSPEC) { + return; + } + /* Set new NS leader */ + ns->name = name; + ns->score = score; + for (size_t i = 0; i < KR_NSREP_MAXADDR; ++i) { + if (addr[i]) { + void *addr_val = pack_obj_val(addr[i]); + size_t len = pack_obj_len(addr[i]); + update_nsrep(ns, i, addr_val, len, KR_DNS_PORT); + } else { + break; + } + } +} + +#undef ADDR_SET + +/** + * \param addr_set pack with one IP address per element */ +static unsigned eval_addr_set(const pack_t *addr_set, struct kr_context *ctx, + struct kr_qflags opts, unsigned score, uint8_t *addr[]) +{ + kr_nsrep_rtt_lru_t *rtt_cache = ctx->cache_rtt; + kr_nsrep_rtt_lru_entry_t *rtt_cache_entry_ptr[KR_NSREP_MAXADDR] = { NULL, }; + assert (KR_NSREP_MAXADDR >= 2); + unsigned rtt_cache_entry_score[KR_NSREP_MAXADDR] = { score, KR_NS_MAX_SCORE + 1, }; + uint64_t now = kr_now(); + + /* Name server is better candidate if it has address record. */ + for (uint8_t *it = pack_head(*addr_set); it != pack_tail(*addr_set); + it = pack_obj_next(it)) { + void *val = pack_obj_val(it); + size_t len = pack_obj_len(it); + unsigned favour = 0; + bool is_valid = false; + /* Check if the address isn't disabled. */ + if (len == sizeof(struct in6_addr)) { + is_valid = !(opts.NO_IPV6); + favour = FAVOUR_IPV6; + } else if (len == sizeof(struct in_addr)) { + is_valid = !(opts.NO_IPV4); + } else { + assert(!EINVAL); + is_valid = false; + } + + if (!is_valid) { + continue; + } + + /* Get score for the current address. */ + kr_nsrep_rtt_lru_entry_t *cached = rtt_cache ? + lru_get_try(rtt_cache, val, len) : + NULL; + unsigned cur_addr_score = KR_NS_GLUED; + if (cached) { + cur_addr_score = cached->score; + if (cached->score >= KR_NS_TIMEOUT) { + /* If NS once was marked as "timeouted", + * it won't participate in NS elections + * at least ctx->cache_rtt_tout_retry_interval milliseconds. */ + uint64_t elapsed = now - cached->tout_timestamp; + elapsed = elapsed > UINT_MAX ? UINT_MAX : elapsed; + if (elapsed > ctx->cache_rtt_tout_retry_interval) { + /* Select this NS for probing in this particular query, + * but don't change the cached score. + * For other queries this NS will remain "timeouted". */ + cur_addr_score = KR_NS_LONG - 1; + } + } + } + + for (size_t i = 0; i < KR_NSREP_MAXADDR; ++i) { + if (cur_addr_score >= KR_NS_TIMEOUT) { + /* We can't use favour here. + * If all of the conditions below are true + * + * rtt_cache_entry_score[i] < KR_NS_TIMEOUT + * rtt_cache_entry_score[i] + favour > KR_NS_TIMEOUT + * cur_addr_score < rtt_cache_entry_score[i] + favour + * + * we will prefer "certainly dead" cur_addr_score + * instead of "almost dead, but alive" rtt_cache_entry_score[i] + */ + if (cur_addr_score >= rtt_cache_entry_score[i]) { + continue; + } + } + if (cur_addr_score >= KR_NS_TIMEOUT + || cur_addr_score < rtt_cache_entry_score[i] + favour) { + /* Shake down previous contenders */ + for (size_t j = KR_NSREP_MAXADDR - 1; j > i; --j) { + addr[j] = addr[j - 1]; + rtt_cache_entry_ptr[j] = rtt_cache_entry_ptr[j - 1]; + rtt_cache_entry_score[j] = rtt_cache_entry_score[j - 1]; + } + addr[i] = it; + rtt_cache_entry_score[i] = cur_addr_score; + rtt_cache_entry_ptr[i] = cached; + break; + } + } + } + + /* At this point, rtt_cache_entry_ptr contains up to KR_NSREP_MAXADDR + * pointers to the rtt cache entries with the best scores for the given addr_set. + * Check if there are timeouted NS. */ + + for (size_t i = 0; i < KR_NSREP_MAXADDR; ++i) { + if (rtt_cache_entry_ptr[i] == NULL) + continue; + if (rtt_cache_entry_ptr[i]->score < KR_NS_TIMEOUT) + continue; + + uint64_t elapsed = now - rtt_cache_entry_ptr[i]->tout_timestamp; + elapsed = elapsed > UINT_MAX ? UINT_MAX : elapsed; + if (elapsed <= ctx->cache_rtt_tout_retry_interval) + continue; + + /* rtt_cache_entry_ptr[i] points to "timeouted" rtt cache entry. + * The period of the ban on participation in elections has expired. */ + + if (VERBOSE_STATUS) { + void *val = pack_obj_val(addr[i]); + size_t len = pack_obj_len(addr[i]); + char sa_str[INET6_ADDRSTRLEN]; + int af = (len == sizeof(struct in6_addr)) ? AF_INET6 : AF_INET; + inet_ntop(af, val, sa_str, sizeof(sa_str)); + kr_log_verbose("[ ][nsre] probing timeouted NS: %s, score %i\n", + sa_str, rtt_cache_entry_ptr[i]->score); + } + + rtt_cache_entry_ptr[i]->tout_timestamp = now; + } + + return rtt_cache_entry_score[0]; +} + +static int eval_nsrep(const knot_dname_t *owner, const pack_t *addr_set, struct kr_query *qry) +{ + struct kr_nsrep *ns = &qry->ns; + struct kr_context *ctx = ns->ctx; + unsigned score = KR_NS_MAX_SCORE; + unsigned reputation = 0; + uint8_t *addr_choice[KR_NSREP_MAXADDR] = { NULL, }; + + /* Fetch NS reputation */ + if (ctx->cache_rep) { + unsigned *cached = lru_get_try(ctx->cache_rep, (const char *)owner, + knot_dname_size(owner)); + if (cached) { + reputation = *cached; + } + } + + /* Favour nameservers with unknown addresses to probe them, + * otherwise discover the current best address for the NS. */ + if (addr_set->len == 0) { + score = KR_NS_UNKNOWN; + /* If the server doesn't have IPv6, give it disadvantage. */ + if (reputation & KR_NS_NOIP6) { + score += FAVOUR_IPV6; + /* If the server is unknown but has rep record, treat it as timeouted */ + if (reputation & KR_NS_NOIP4) { + score = KR_NS_UNKNOWN; + /* Try to start with clean slate */ + if (!(qry->flags.NO_IPV6)) { + reputation &= ~KR_NS_NOIP6; + } + if (!(qry->flags.NO_IPV4)) { + reputation &= ~KR_NS_NOIP4; + } + } + } + } else { + score = eval_addr_set(addr_set, ctx, qry->flags, score, addr_choice); + } + + /* Probabilistic bee foraging strategy (naive). + * The fastest NS is preferred by workers until it is depleted (timeouts or degrades), + * at the same time long distance scouts probe other sources (low probability). + * Servers on TIMEOUT will not have probed at all. + * Servers with score above KR_NS_LONG will have periodically removed from + * reputation cache, so that kresd can reprobe them. */ + if (score >= KR_NS_TIMEOUT) { + return kr_ok(); + } else if (score <= ns->score && + (score < KR_NS_LONG || qry->flags.NO_THROTTLE)) { + update_nsrep_set(ns, owner, addr_choice, score); + ns->reputation = reputation; + } else if (kr_rand_coin(1, 10) && + !kr_rand_coin(score, KR_NS_MAX_SCORE)) { + /* With 10% chance probe server with a probability + * given by its RTT / MAX_RTT. */ + update_nsrep_set(ns, owner, addr_choice, score); + ns->reputation = reputation; + return 1; /* Stop evaluation */ + } else if (ns->score > KR_NS_MAX_SCORE) { + /* Check if any server was already selected. + * If no, pick current server and continue evaluation. */ + update_nsrep_set(ns, owner, addr_choice, score); + ns->reputation = reputation; + } + + return kr_ok(); +} + +int kr_nsrep_set(struct kr_query *qry, size_t index, const struct sockaddr *sock) +{ + if (!qry) { + return kr_error(EINVAL); + } + if (index >= KR_NSREP_MAXADDR) { + return kr_error(ENOSPC); + } + + if (!sock) { + qry->ns.name = (const uint8_t *)""; + qry->ns.addr[index].ip.sa_family = AF_UNSPEC; + return kr_ok(); + } + + switch (sock->sa_family) { + case AF_INET: + if (qry->flags.NO_IPV4) { + return kr_error(ENOENT); + } + qry->ns.addr[index].ip4 = *(const struct sockaddr_in *)sock; + break; + case AF_INET6: + if (qry->flags.NO_IPV6) { + return kr_error(ENOENT); + } + qry->ns.addr[index].ip6 = *(const struct sockaddr_in6 *)sock; + break; + default: + qry->ns.addr[index].ip.sa_family = AF_UNSPEC; + return kr_error(EINVAL); + } + + qry->ns.name = (const uint8_t *)""; + /* Reset score on first entry */ + if (index == 0) { + qry->ns.score = KR_NS_UNKNOWN; + qry->ns.reputation = 0; + } + + /* Retrieve RTT from cache */ + struct kr_context *ctx = qry->ns.ctx; + kr_nsrep_rtt_lru_entry_t *rtt_cache_entry = ctx + ? lru_get_try(ctx->cache_rtt, kr_inaddr(sock), kr_family_len(sock->sa_family)) + : NULL; + if (rtt_cache_entry) { + qry->ns.score = MIN(qry->ns.score, rtt_cache_entry->score); + } + + return kr_ok(); +} + +#define ELECT_INIT(ns, ctx_) do { \ + (ns)->ctx = (ctx_); \ + (ns)->addr[0].ip.sa_family = AF_UNSPEC; \ + (ns)->reputation = 0; \ + (ns)->score = KR_NS_MAX_SCORE + 1; \ +} while (0) + +int kr_nsrep_elect(struct kr_query *qry, struct kr_context *ctx) +{ + if (!qry || !ctx) { + //assert(!EINVAL); + return kr_error(EINVAL); + } + + struct kr_nsrep *ns = &qry->ns; + ELECT_INIT(ns, ctx); + + int ret = kr_ok(); + trie_it_t *it; + for (it = trie_it_begin(qry->zone_cut.nsset); !trie_it_finished(it); + trie_it_next(it)) { + ret = eval_nsrep(/* we trust it's a correct dname */ + (const knot_dname_t *)trie_it_key(it, NULL), + (const pack_t *)*trie_it_val(it), qry); + if (ret) break; + } + trie_it_free(it); + + if (qry->ns.score <= KR_NS_MAX_SCORE && qry->ns.score >= KR_NS_LONG) { + /* This is a low-reliability probe, + * go with TCP to get ICMP reachability check. */ + qry->flags.TCP = true; + } + return ret; +} + +int kr_nsrep_elect_addr(struct kr_query *qry, struct kr_context *ctx) +{ + if (!qry || !ctx) { + //assert(!EINVAL); + return kr_error(EINVAL); + } + + /* Get address list for this NS */ + struct kr_nsrep *ns = &qry->ns; + ELECT_INIT(ns, ctx); + pack_t *addr_set = kr_zonecut_find(&qry->zone_cut, ns->name); + if (!addr_set) { + return kr_error(ENOENT); + } + /* Evaluate addr list */ + uint8_t *addr_choice[KR_NSREP_MAXADDR] = { NULL, }; + unsigned score = eval_addr_set(addr_set, ctx, qry->flags, ns->score, addr_choice); + update_nsrep_set(ns, ns->name, addr_choice, score); + return kr_ok(); +} + +#undef ELECT_INIT + +int kr_nsrep_update_rtt(struct kr_nsrep *ns, const struct sockaddr *addr, + unsigned score, kr_nsrep_rtt_lru_t *cache, int umode) +{ + if (!cache || umode > KR_NS_MAX || umode < 0) { + return kr_error(EINVAL); + } + + /* Get `addr`, and later its raw string. */ + if (addr) { + /* Caller provided specific address, OK. */ + } else if (ns != NULL) { + addr = &ns->addr[0].ip; + } else { + assert(false && "kr_nsrep_update_rtt: don't know what address to update"); + return kr_error(EINVAL); + } + const char *addr_in = kr_inaddr(addr); + size_t addr_len = kr_inaddr_len(addr); + if (!addr_in || addr_len <= 0) { + assert(false && "kr_nsrep_update_rtt: incorrect address"); + return kr_error(EINVAL); + } + + bool is_new_entry = false; + kr_nsrep_rtt_lru_entry_t *cur = lru_get_new(cache, addr_in, addr_len, + (&is_new_entry)); + if (!cur) { + return kr_ok(); + } + if (score <= KR_NS_GLUED) { + score = KR_NS_GLUED + 1; + } + /* If there's nothing to update, we reset it unless KR_NS_UPDATE_NORESET + * mode was requested. New items are zeroed by LRU automatically. */ + if (is_new_entry && umode != KR_NS_UPDATE_NORESET) { + umode = KR_NS_RESET; + } + unsigned new_score = 0; + /* Update score, by default smooth over last two measurements. */ + switch (umode) { + case KR_NS_UPDATE: + case KR_NS_UPDATE_NORESET: + new_score = (cur->score + score) / 2; break; + case KR_NS_RESET: new_score = score; break; + case KR_NS_ADD: new_score = MIN(KR_NS_MAX_SCORE - 1, cur->score + score); break; + case KR_NS_MAX: new_score = MAX(cur->score, score); break; + default: return kr_error(EINVAL); + } + /* Score limits */ + if (new_score > KR_NS_MAX_SCORE) { + new_score = KR_NS_MAX_SCORE; + } + if (new_score >= KR_NS_TIMEOUT && cur->score < KR_NS_TIMEOUT) { + /* Set the timestamp only when NS became "timeouted" */ + cur->tout_timestamp = kr_now(); + } + cur->score = new_score; + return kr_ok(); +} + +int kr_nsrep_update_rep(struct kr_nsrep *ns, unsigned reputation, kr_nsrep_lru_t *cache) +{ + if (!ns || !cache ) { + return kr_error(EINVAL); + } + + /* Store in the struct */ + ns->reputation = reputation; + /* Store reputation in the LRU cache */ + unsigned *cur = lru_get_new(cache, (const char *)ns->name, + knot_dname_size(ns->name), NULL); + if (cur) { + *cur = reputation; + } + return kr_ok(); +} + +int kr_nsrep_copy_set(struct kr_nsrep *dst, const struct kr_nsrep *src) +{ + if (!dst || !src ) { + return kr_error(EINVAL); + } + + memcpy(dst, src, sizeof(struct kr_nsrep)); + dst->name = (const uint8_t *)""; + dst->score = KR_NS_UNKNOWN; + dst->reputation = 0; + + return kr_ok(); +} + +int kr_nsrep_sort(struct kr_nsrep *ns, struct kr_context *ctx) +{ + if (!ns || !ctx) { + assert(false); + return kr_error(EINVAL); + } + + kr_nsrep_rtt_lru_t *rtt_cache = ctx->cache_rtt; + + ns->reputation = 0; + ns->score = KR_NS_MAX_SCORE + 1; + + if (ns->addr[0].ip.sa_family == AF_UNSPEC) { + return kr_error(EINVAL); + } + + /* Compute the scores. Unfortunately there's no space for scores + * along the addresses. */ + unsigned scores[KR_NSREP_MAXADDR]; + int i; + bool timeouted_address_is_already_selected = false; + for (i = 0; i < KR_NSREP_MAXADDR; ++i) { + const struct sockaddr *sa = &ns->addr[i].ip; + if (sa->sa_family == AF_UNSPEC) { + break; + } + kr_nsrep_rtt_lru_entry_t *rtt_cache_entry = lru_get_try(rtt_cache, + kr_inaddr(sa), + kr_family_len(sa->sa_family)); + if (!rtt_cache_entry) { + scores[i] = 1; /* prefer unknown to probe RTT */ + } else if (rtt_cache_entry->score < KR_NS_FWD_TIMEOUT) { + /* some probability to bump bad ones up for re-probe */ + scores[i] = rtt_cache_entry->score; + /* The lower the rtt, the more likely it will be selected. */ + if (!kr_rand_coin(rtt_cache_entry->score, KR_NS_FWD_TIMEOUT)) { + scores[i] = 1; + } + } else { + uint64_t now = kr_now(); + uint64_t elapsed = now - rtt_cache_entry->tout_timestamp; + scores[i] = KR_NS_MAX_SCORE + 1; + elapsed = elapsed > UINT_MAX ? UINT_MAX : elapsed; + if (elapsed > ctx->cache_rtt_tout_retry_interval && + !timeouted_address_is_already_selected) { + scores[i] = 1; + rtt_cache_entry->tout_timestamp = now; + timeouted_address_is_already_selected = true; + } + } + + /* Give advantage to IPv6. */ + if (scores[i] <= KR_NS_MAX_SCORE && sa->sa_family == AF_INET) { + scores[i] += FAVOUR_IPV6; + } + + if (VERBOSE_STATUS) { + kr_log_verbose("[ ][nsre] score %d for %s;\t cached RTT: %d\n", + scores[i], kr_straddr(sa), + rtt_cache_entry ? rtt_cache_entry->score : -1); + } + } + + /* Select-sort the addresses. */ + const int count = i; + for (i = 0; i < count - 1; ++i) { + /* find min from i onwards */ + int min_i = i; + for (int j = i + 1; j < count; ++j) { + if (scores[j] < scores[min_i]) { + min_i = j; + } + } + /* swap the indices */ + if (min_i != i) { + SWAP(scores[min_i], scores[i]); + SWAP(ns->addr[min_i], ns->addr[i]); + } + } + + if (count > 0) { + ns->score = scores[0]; + ns->reputation = 0; + } + + return kr_ok(); +} diff --git a/lib/nsrep.h b/lib/nsrep.h new file mode 100644 index 0000000..cc35ca4 --- /dev/null +++ b/lib/nsrep.h @@ -0,0 +1,188 @@ +/* Copyright (C) 2014-2017 CZ.NIC, z.s.p.o. + + This program is free software: you can redistribute it and/or modify + it under the terms of the GNU General Public License as published by + the Free Software Foundation, either version 3 of the License, or + (at your option) any later version. + + This program is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the + GNU General Public License for more details. + + You should have received a copy of the GNU General Public License + along with this program. If not, see . + */ + +#pragma once + +#include +#include +#include +#include + +#include "lib/defines.h" +#include "lib/generic/lru.h" + +struct kr_query; + +/** + * NS RTT score (special values). + * @note RTT is measured in milliseconds. + */ +enum kr_ns_score { + KR_NS_MAX_SCORE = 20 * KR_CONN_RTT_MAX, /* max possible value */ + KR_NS_FWD_TIMEOUT = (95 * 10000) / 100, /* timeout for upstream recursor, + * 95 percents from max resolution time */ + KR_NS_TIMEOUT = (95 * KR_CONN_RTT_MAX) / 100, /* timeout for upstream auth */ + KR_NS_LONG = (3 * KR_NS_TIMEOUT) / 4, + KR_NS_UNKNOWN = KR_NS_TIMEOUT / 2, + KR_NS_PENALTY = 100, + KR_NS_GLUED = 10 +}; + +/** + * See kr_nsrep_update_rtt() + */ +#define KR_NS_DEAD (((KR_NS_TIMEOUT * 4) + 3) / 3) +#define KR_NS_FWD_DEAD (((KR_NS_FWD_TIMEOUT * 4) + 3) / 3) + +/** If once NS was marked as "timeouted", it won't participate in NS elections + * at least KR_NS_TIMEOUT_RETRY_INTERVAL milliseconds (now: one second). */ +#define KR_NS_TIMEOUT_RETRY_INTERVAL 1000 + +/** + * NS QoS flags. + */ +enum kr_ns_rep { + KR_NS_NOIP4 = 1 << 0, /**< NS has no IPv4 */ + KR_NS_NOIP6 = 1 << 1, /**< NS has no IPv6 */ + KR_NS_NOEDNS = 1 << 2 /**< NS has no EDNS support */ +}; + +/** + * NS RTT update modes. + * First update is always KR_NS_RESET unless + * KR_NS_UPDATE_NORESET mode had choosen. + */ +enum kr_ns_update_mode { + KR_NS_UPDATE = 0, /**< Update as smooth over last two measurements */ + KR_NS_UPDATE_NORESET, /**< Same as KR_NS_UPDATE, but disable fallback to + * KR_NS_RESET on newly added entries. + * Zero is used as initial value. */ + KR_NS_RESET, /**< Set to given value */ + KR_NS_ADD, /**< Increment current value */ + KR_NS_MAX /**< Set to maximum of current/proposed value. */ +}; + +struct kr_nsrep_rtt_lru_entry { + unsigned score; /* combined rtt */ + uint64_t tout_timestamp; /* The time when score became + * greater or equal then KR_NS_TIMEOUT. + * Is meaningful only when score >= KR_NS_TIMEOUT */ +}; + +typedef struct kr_nsrep_rtt_lru_entry kr_nsrep_rtt_lru_entry_t; + +/** + * NS QoS tracking. + */ +typedef lru_t(kr_nsrep_rtt_lru_entry_t) kr_nsrep_rtt_lru_t; + +/** + * NS reputation tracking. + */ +typedef lru_t(unsigned) kr_nsrep_lru_t; + +/* Maximum count of addresses probed in one go (last is left empty) */ +#define KR_NSREP_MAXADDR 4 + +/** + * Name server representation. + * Contains extra information about the name server, e.g. score + * or other metadata. + */ +struct kr_nsrep +{ + unsigned score; /**< NS score */ + unsigned reputation; /**< NS reputation */ + const knot_dname_t *name; /**< NS name */ + struct kr_context *ctx; /**< Resolution context */ + union inaddr addr[KR_NSREP_MAXADDR]; /**< NS address(es) */ +}; + +/** + * Set given NS address. (Very low-level access to the list.) + * @param qry updated query + * @param index index of the updated target + * @param sock socket address to use (sockaddr_in or sockaddr_in6 or NULL) + * @return 0 or an error code, in particular kr_error(ENOENT) for net.ipvX + */ +KR_EXPORT +int kr_nsrep_set(struct kr_query *qry, size_t index, const struct sockaddr *sock); + +/** + * Elect best nameserver/address pair from the nsset. + * @param qry updated query + * @param ctx resolution context + * @return 0 or an error code + */ +KR_EXPORT +int kr_nsrep_elect(struct kr_query *qry, struct kr_context *ctx); + +/** + * Elect best nameserver/address pair from the nsset. + * @param qry updated query + * @param ctx resolution context + * @return 0 or an error code + */ +KR_EXPORT +int kr_nsrep_elect_addr(struct kr_query *qry, struct kr_context *ctx); + +/** + * Update NS address RTT information. + * + * @brief In KR_NS_UPDATE mode reputation is smoothed over last N measurements. + * + * @param ns updated NS representation + * @param addr chosen address (NULL for first) + * @param score new score (i.e. RTT), see enum kr_ns_score + * @param cache RTT LRU cache + * @param umode update mode (KR_NS_UPDATE or KR_NS_RESET or KR_NS_ADD) + * @return 0 on success, error code on failure + */ +KR_EXPORT +int kr_nsrep_update_rtt(struct kr_nsrep *ns, const struct sockaddr *addr, + unsigned score, kr_nsrep_rtt_lru_t *cache, int umode); + +/** + * Update NSSET reputation information. + * + * @param ns updated NS representation + * @param reputation combined reputation flags, see enum kr_ns_rep + * @param cache LRU cache + * @return 0 on success, error code on failure + */ +KR_EXPORT +int kr_nsrep_update_rep(struct kr_nsrep *ns, unsigned reputation, kr_nsrep_lru_t *cache); +/** + * Copy NSSET reputation information and resets score. + * + * @param dst updated NS representation + * @param src source NS representation + * @return 0 on success, error code on failure + */ +int kr_nsrep_copy_set(struct kr_nsrep *dst, const struct kr_nsrep *src); + +/** + * Sort addresses in the query nsrep list by cached RTT. + * if RTT is greater then KR_NS_TIMEOUT, address will placed at the beginning of the + * nsrep list once in cache.ns_tout() milliseconds. Otherwise it will be sorted + * as if it has cached RTT equal to KR_NS_MAX_SCORE + 1. + * @param ns updated kr_nsrep + * @param ctx name resolution context. + * @return 0 or an error code + * @note ns reputation is zeroed and score is set to KR_NS_MAX_SCORE + 1. + */ +KR_EXPORT +int kr_nsrep_sort(struct kr_nsrep *ns, struct kr_context *ctx); diff --git a/lib/resolve.c b/lib/resolve.c new file mode 100644 index 0000000..8718f25 --- /dev/null +++ b/lib/resolve.c @@ -0,0 +1,1648 @@ +/* Copyright (C) 2014-2017 CZ.NIC, z.s.p.o. + + This program is free software: you can redistribute it and/or modify + it under the terms of the GNU General Public License as published by + the Free Software Foundation, either version 3 of the License, or + (at your option) any later version. + + This program is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the + GNU General Public License for more details. + + You should have received a copy of the GNU General Public License + along with this program. If not, see . + */ + +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include "lib/resolve.h" +#include "lib/layer.h" +#include "lib/rplan.h" +#include "lib/layer/iterate.h" +#include "lib/dnssec/ta.h" +#include "lib/dnssec.h" +#if defined(ENABLE_COOKIES) +#include "lib/cookies/control.h" +#include "lib/cookies/helper.h" +#include "lib/cookies/nonce.h" +#else /* Define compatibility macros */ +#define KNOT_EDNS_OPTION_COOKIE 10 +#endif /* defined(ENABLE_COOKIES) */ + +#define VERBOSE_MSG(qry, ...) QRVERBOSE((qry), "resl", __VA_ARGS__) + +bool kr_rank_check(uint8_t rank) +{ + switch (rank & ~KR_RANK_AUTH) { + case KR_RANK_INITIAL: + case KR_RANK_OMIT: + case KR_RANK_TRY: + case KR_RANK_INDET: + case KR_RANK_BOGUS: + case KR_RANK_MISMATCH: + case KR_RANK_MISSING: + case KR_RANK_INSECURE: + case KR_RANK_SECURE: + return true; + default: + return false; + } +} + +/** @internal Set @a yielded to all RRs with matching @a qry_uid. */ +static void set_yield(ranked_rr_array_t *array, const uint32_t qry_uid, const bool yielded) +{ + for (unsigned i = 0; i < array->len; ++i) { + ranked_rr_array_entry_t *entry = array->at[i]; + if (entry->qry_uid == qry_uid) { + entry->yielded = yielded; + } + } +} + +/** + * @internal Defer execution of current query. + * The current layer state and input will be pushed to a stack and resumed on next iteration. + */ +static int consume_yield(kr_layer_t *ctx, knot_pkt_t *pkt) +{ + struct kr_request *req = ctx->req; + size_t pkt_size = pkt->size; + if (knot_pkt_has_tsig(pkt)) { + pkt_size += pkt->tsig_wire.len; + } + knot_pkt_t *pkt_copy = knot_pkt_new(NULL, pkt_size, &req->pool); + struct kr_layer_pickle *pickle = mm_alloc(&req->pool, sizeof(*pickle)); + if (pickle && pkt_copy && knot_pkt_copy(pkt_copy, pkt) == 0) { + struct kr_query *qry = req->current_query; + pickle->api = ctx->api; + pickle->state = ctx->state; + pickle->pkt = pkt_copy; + pickle->next = qry->deferred; + qry->deferred = pickle; + set_yield(&req->answ_selected, qry->uid, true); + set_yield(&req->auth_selected, qry->uid, true); + return kr_ok(); + } + return kr_error(ENOMEM); +} +static int begin_yield(kr_layer_t *ctx) { return kr_ok(); } +static int reset_yield(kr_layer_t *ctx) { return kr_ok(); } +static int finish_yield(kr_layer_t *ctx) { return kr_ok(); } +static int produce_yield(kr_layer_t *ctx, knot_pkt_t *pkt) { return kr_ok(); } +static int checkout_yield(kr_layer_t *ctx, knot_pkt_t *packet, struct sockaddr *dst, int type) { return kr_ok(); } +static int answer_finalize_yield(kr_layer_t *ctx) { return kr_ok(); } + +/** @internal Macro for iterating module layers. */ +#define RESUME_LAYERS(from, r, qry, func, ...) \ + (r)->current_query = (qry); \ + for (size_t i = (from); i < (r)->ctx->modules->len; ++i) { \ + struct kr_module *mod = (r)->ctx->modules->at[i]; \ + if (mod->layer) { \ + struct kr_layer layer = {.state = (r)->state, .api = mod->layer(mod), .req = (r)}; \ + if (layer.api && layer.api->func) { \ + (r)->state = layer.api->func(&layer, ##__VA_ARGS__); \ + if ((r)->state == KR_STATE_YIELD) { \ + func ## _yield(&layer, ##__VA_ARGS__); \ + break; \ + } \ + } \ + } \ + } /* Invalidate current query. */ \ + (r)->current_query = NULL + +/** @internal Macro for starting module iteration. */ +#define ITERATE_LAYERS(req, qry, func, ...) RESUME_LAYERS(0, req, qry, func, ##__VA_ARGS__) + +/** @internal Find layer id matching API. */ +static inline size_t layer_id(struct kr_request *req, const struct kr_layer_api *api) { + module_array_t *modules = req->ctx->modules; + for (size_t i = 0; i < modules->len; ++i) { + struct kr_module *mod = modules->at[i]; + if (mod->layer && mod->layer(mod) == api) { + return i; + } + } + return 0; /* Not found, try all. */ +} + +/* @internal We don't need to deal with locale here */ +KR_CONST static inline bool isletter(unsigned chr) +{ return (chr | 0x20 /* tolower */) - 'a' <= 'z' - 'a'; } + +/* Randomize QNAME letter case. + * This adds 32 bits of randomness at maximum, but that's more than an average domain name length. + * https://tools.ietf.org/html/draft-vixie-dnsext-dns0x20-00 + */ +static void randomized_qname_case(knot_dname_t * restrict qname, uint32_t secret) +{ + if (secret == 0) { + return; + } + assert(qname); + const int len = knot_dname_size(qname) - 2; /* Skip first, last label. */ + for (int i = 0; i < len; ++i) { + if (isletter(*++qname)) { + *qname ^= ((secret >> (i & 31)) & 1) * 0x20; + } + } +} + +/** Invalidate current NS/addr pair. */ +static int invalidate_ns(struct kr_rplan *rplan, struct kr_query *qry) +{ + if (qry->ns.addr[0].ip.sa_family != AF_UNSPEC) { + const char *addr = kr_inaddr(&qry->ns.addr[0].ip); + int addr_len = kr_inaddr_len(&qry->ns.addr[0].ip); + return kr_zonecut_del(&qry->zone_cut, qry->ns.name, addr, addr_len); + } else { + return kr_zonecut_del_all(&qry->zone_cut, qry->ns.name); + } +} + +/** This turns of QNAME minimisation if there is a non-terminal between current zone cut, and name target. + * It save several minimization steps, as the zone cut is likely final one. + */ +static void check_empty_nonterms(struct kr_query *qry, knot_pkt_t *pkt, struct kr_cache *cache, uint32_t timestamp) +{ + // FIXME cleanup, etc. +#if 0 + if (qry->flags.NO_MINIMIZE) { + return; + } + + const knot_dname_t *target = qry->sname; + const knot_dname_t *cut_name = qry->zone_cut.name; + if (!target || !cut_name) + return; + + struct kr_cache_entry *entry = NULL; + /* @note: The non-terminal must be direct child of zone cut (e.g. label distance <= 2), + * otherwise this would risk leaking information to parent if the NODATA TTD > zone cut TTD. */ + int labels = knot_dname_labels(target, NULL) - knot_dname_labels(cut_name, NULL); + while (target[0] && labels > 2) { + target = knot_wire_next_label(target, NULL); + --labels; + } + for (int i = 0; i < labels; ++i) { + int ret = kr_cache_peek(cache, KR_CACHE_PKT, target, KNOT_RRTYPE_NS, &entry, ×tamp); + if (ret == 0) { /* Either NXDOMAIN or NODATA, start here. */ + /* @todo We could stop resolution here for NXDOMAIN, but we can't because of broken CDNs */ + qry->flags.NO_MINIMIZE = true; + kr_make_query(qry, pkt); + break; + } + assert(target[0]); + target = knot_wire_next_label(target, NULL); + } + kr_cache_sync(cache); +#endif +} + +static int ns_fetch_cut(struct kr_query *qry, const knot_dname_t *requested_name, + struct kr_request *req, knot_pkt_t *pkt) +{ + /* It can occur that here parent query already have + * provably insecured zonecut which not in the cache yet. */ + struct kr_qflags pflags; + if (qry->parent) { + pflags = qry->parent->flags; + } + const bool is_insecured = qry->parent != NULL + && !(pflags.AWAIT_IPV4 || pflags.AWAIT_IPV6) + && (pflags.DNSSEC_INSECURE || pflags.DNSSEC_NODS); + + /* Want DNSSEC if it's possible to secure this name + * (e.g. is covered by any TA) */ + if (is_insecured) { + /* If parent is unsecured we don't want DNSSEC + * even if cut name is covered by TA. */ + qry->flags.DNSSEC_WANT = false; + qry->flags.DNSSEC_INSECURE = true; + VERBOSE_MSG(qry, "=> going insecure because parent query is insecure\n"); + } else if (kr_ta_covers_qry(req->ctx, qry->zone_cut.name, KNOT_RRTYPE_NS)) { + qry->flags.DNSSEC_WANT = true; + } else { + qry->flags.DNSSEC_WANT = false; + VERBOSE_MSG(qry, "=> going insecure because there's no covering TA\n"); + } + + struct kr_zonecut cut_found; + kr_zonecut_init(&cut_found, requested_name, req->rplan.pool); + /* Cut that has been found can differs from cut that has been requested. + * So if not already insecured, + * try to fetch ta & keys even if initial cut name not covered by TA */ + bool secured = !is_insecured; + int ret = kr_zonecut_find_cached(req->ctx, &cut_found, requested_name, + qry, &secured); + if (ret == kr_error(ENOENT)) { + /* No cached cut found, start from SBELT + * and issue priming query. */ + kr_zonecut_deinit(&cut_found); + ret = kr_zonecut_set_sbelt(req->ctx, &qry->zone_cut); + if (ret != 0) { + return KR_STATE_FAIL; + } + VERBOSE_MSG(qry, "=> using root hints\n"); + qry->flags.AWAIT_CUT = false; + return KR_STATE_DONE; + } else if (ret != kr_ok()) { + kr_zonecut_deinit(&cut_found); + return KR_STATE_FAIL; + } + + /* Find out security status. + * Go insecure if the zone cut is provably insecure */ + if ((qry->flags.DNSSEC_WANT) && !secured) { + VERBOSE_MSG(qry, "=> NS is provably without DS, going insecure\n"); + qry->flags.DNSSEC_WANT = false; + qry->flags.DNSSEC_INSECURE = true; + } + /* Zonecut name can change, check it again + * to prevent unnecessary DS & DNSKEY queries */ + if (!(qry->flags.DNSSEC_INSECURE) && + kr_ta_covers_qry(req->ctx, cut_found.name, KNOT_RRTYPE_NS)) { + qry->flags.DNSSEC_WANT = true; + } else { + qry->flags.DNSSEC_WANT = false; + } + /* Check if any DNSKEY found for cached cut */ + if (qry->flags.DNSSEC_WANT && cut_found.key == NULL && + kr_zonecut_is_empty(&cut_found)) { + /* Cut found and there are no proofs of zone insecurity. + * But no DNSKEY found and no glue fetched. + * We have got circular dependency - must fetch A\AAAA + * from authoritative, but we have no key to verify it. */ + kr_zonecut_deinit(&cut_found); + if (requested_name[0] != '\0' ) { + /* If not root - try next label */ + return KR_STATE_CONSUME; + } + /* No cached cut & keys found, start from SBELT */ + ret = kr_zonecut_set_sbelt(req->ctx, &qry->zone_cut); + if (ret != 0) { + return KR_STATE_FAIL; + } + VERBOSE_MSG(qry, "=> using root hints\n"); + qry->flags.AWAIT_CUT = false; + return KR_STATE_DONE; + } + /* Use the found zone cut. */ + kr_zonecut_move(&qry->zone_cut, &cut_found); + /* Check if there's a non-terminal between target and current cut. */ + struct kr_cache *cache = &req->ctx->cache; + check_empty_nonterms(qry, pkt, cache, qry->timestamp.tv_sec); + /* Cut found */ + return KR_STATE_PRODUCE; +} + +static int ns_resolve_addr(struct kr_query *qry, struct kr_request *param) +{ + struct kr_rplan *rplan = ¶m->rplan; + struct kr_context *ctx = param->ctx; + + + /* Start NS queries from root, to avoid certain cases + * where a NS drops out of cache and the rest is unavailable, + * this would lead to dependency loop in current zone cut. + * Prefer IPv6 and continue with IPv4 if not available. + */ + uint16_t next_type = 0; + if (!(qry->flags.AWAIT_IPV6) && + !(ctx->options.NO_IPV6)) { + next_type = KNOT_RRTYPE_AAAA; + qry->flags.AWAIT_IPV6 = true; + } else if (!(qry->flags.AWAIT_IPV4) && + !(ctx->options.NO_IPV4)) { + next_type = KNOT_RRTYPE_A; + qry->flags.AWAIT_IPV4 = true; + /* Hmm, no useable IPv6 then. */ + qry->ns.reputation |= KR_NS_NOIP6; + kr_nsrep_update_rep(&qry->ns, qry->ns.reputation, ctx->cache_rep); + } + /* Bail out if the query is already pending or dependency loop. */ + if (!next_type || kr_rplan_satisfies(qry->parent, qry->ns.name, KNOT_CLASS_IN, next_type)) { + /* Fall back to SBELT if root server query fails. */ + if (!next_type && qry->zone_cut.name[0] == '\0') { + VERBOSE_MSG(qry, "=> fallback to root hints\n"); + kr_zonecut_set_sbelt(ctx, &qry->zone_cut); + qry->flags.NO_THROTTLE = true; /* Pick even bad SBELT servers */ + return kr_error(EAGAIN); + } + /* No IPv4 nor IPv6, flag server as unusable. */ + VERBOSE_MSG(qry, "=> unresolvable NS address, bailing out\n"); + qry->ns.reputation |= KR_NS_NOIP4 | KR_NS_NOIP6; + kr_nsrep_update_rep(&qry->ns, qry->ns.reputation, ctx->cache_rep); + invalidate_ns(rplan, qry); + return kr_error(EHOSTUNREACH); + } + /* Push new query to the resolution plan */ + struct kr_query *next = + kr_rplan_push(rplan, qry, qry->ns.name, KNOT_CLASS_IN, next_type); + if (!next) { + return kr_error(ENOMEM); + } + next->flags.NONAUTH = true; + + /* At the root level with no NS addresses, add SBELT subrequest. */ + int ret = 0; + if (qry->zone_cut.name[0] == '\0') { + ret = kr_zonecut_set_sbelt(ctx, &next->zone_cut); + if (ret == 0) { /* Copy TA and key since it's the same cut to avoid lookup. */ + kr_zonecut_copy_trust(&next->zone_cut, &qry->zone_cut); + kr_zonecut_set_sbelt(ctx, &qry->zone_cut); /* Add SBELT to parent in case query fails. */ + qry->flags.NO_THROTTLE = true; /* Pick even bad SBELT servers */ + } + } else { + next->flags.AWAIT_CUT = true; + } + return ret; +} + +static int edns_put(knot_pkt_t *pkt, bool reclaim) +{ + if (!pkt->opt_rr) { + return kr_ok(); + } + if (reclaim) { + /* Reclaim reserved size. */ + int ret = knot_pkt_reclaim(pkt, knot_edns_wire_size(pkt->opt_rr)); + if (ret != 0) { + return ret; + } + } + /* Write to packet. */ + assert(pkt->current == KNOT_ADDITIONAL); + return knot_pkt_put(pkt, KNOT_COMPR_HINT_NONE, pkt->opt_rr, KNOT_PF_FREE); +} + +/** Removes last EDNS OPT RR written to the packet. */ +static int edns_erase_and_reserve(knot_pkt_t *pkt) +{ + /* Nothing to be done. */ + if (!pkt || !pkt->opt_rr) { + return 0; + } + + /* Fail if the data are located elsewhere than at the end of packet. */ + if (pkt->current != KNOT_ADDITIONAL || + pkt->opt_rr != &pkt->rr[pkt->rrset_count - 1]) { + return -1; + } + + size_t len = knot_rrset_size(pkt->opt_rr); + int16_t rr_removed = pkt->opt_rr->rrs.count; + /* Decrease rrset counters. */ + pkt->rrset_count -= 1; + pkt->sections[pkt->current].count -= 1; + pkt->size -= len; + knot_wire_add_arcount(pkt->wire, -rr_removed); /* ADDITIONAL */ + + pkt->opt_rr = NULL; + + /* Reserve the freed space. */ + return knot_pkt_reserve(pkt, len); +} + +static int edns_create(knot_pkt_t *pkt, knot_pkt_t *template, struct kr_request *req) +{ + pkt->opt_rr = knot_rrset_copy(req->ctx->opt_rr, &pkt->mm); + size_t wire_size = knot_edns_wire_size(pkt->opt_rr); +#if defined(ENABLE_COOKIES) + if (req->ctx->cookie_ctx.clnt.enabled || + req->ctx->cookie_ctx.srvr.enabled) { + wire_size += KR_COOKIE_OPT_MAX_LEN; + } +#endif /* defined(ENABLE_COOKIES) */ + if (req->qsource.flags.tls) { + if (req->ctx->tls_padding == -1) + /* FIXME: we do not know how to reserve space for the + * default padding policy, since we can't predict what + * it will select. So i'm just guessing :/ */ + wire_size += KNOT_EDNS_OPTION_HDRLEN + 512; + if (req->ctx->tls_padding >= 2) + wire_size += KNOT_EDNS_OPTION_HDRLEN + req->ctx->tls_padding; + } + return knot_pkt_reserve(pkt, wire_size); +} + +static int answer_prepare(struct kr_request *req, knot_pkt_t *query) +{ + knot_pkt_t *answer = req->answer; + if (knot_pkt_init_response(answer, query) != 0) { + return kr_error(ENOMEM); /* Failed to initialize answer */ + } + /* Handle EDNS in the query */ + if (knot_pkt_has_edns(query)) { + answer->opt_rr = knot_rrset_copy(req->ctx->opt_rr, &answer->mm); + if (answer->opt_rr == NULL){ + return kr_error(ENOMEM); + } + /* Set DO bit if set (DNSSEC requested). */ + if (knot_pkt_has_dnssec(query)) { + knot_edns_set_do(answer->opt_rr); + } + } + return kr_ok(); +} + +/** @return error code, ignoring if forced to truncate the packet. */ +static int write_extra_records(const rr_array_t *arr, uint16_t reorder, knot_pkt_t *answer) +{ + for (size_t i = 0; i < arr->len; ++i) { + int err = knot_pkt_put_rotate(answer, 0, arr->at[i], reorder, 0); + if (err != KNOT_EOK) { + return err == KNOT_ESPACE ? kr_ok() : kr_error(err); + } + } + return kr_ok(); +} + +/** + * @param all_secure optionally &&-combine security of written RRs into its value. + * (i.e. if you pass a pointer to false, it will always remain) + * @param all_cname optionally output if all written RRs are CNAMEs and RRSIGs of CNAMEs + * @return error code, ignoring if forced to truncate the packet. + */ +static int write_extra_ranked_records(const ranked_rr_array_t *arr, uint16_t reorder, + knot_pkt_t *answer, bool *all_secure, bool *all_cname) +{ + const bool has_dnssec = knot_pkt_has_dnssec(answer); + bool all_sec = true; + bool all_cn = (all_cname != NULL); /* optim.: init as false if not needed */ + int err = kr_ok(); + + for (size_t i = 0; i < arr->len; ++i) { + ranked_rr_array_entry_t * entry = arr->at[i]; + if (!entry->to_wire) { + continue; + } + knot_rrset_t *rr = entry->rr; + if (!has_dnssec) { + if (rr->type != knot_pkt_qtype(answer) && knot_rrtype_is_dnssec(rr->type)) { + continue; + } + } + err = knot_pkt_put_rotate(answer, 0, rr, reorder, 0); + if (err != KNOT_EOK) { + if (err == KNOT_ESPACE) { + err = kr_ok(); + } + break; + } + + if (rr->type != KNOT_RRTYPE_RRSIG) { + all_sec = all_sec && kr_rank_test(entry->rank, KR_RANK_SECURE); + } + all_cn = all_cn && kr_rrset_type_maysig(entry->rr) == KNOT_RRTYPE_CNAME; + } + + if (all_secure) { + *all_secure = *all_secure && all_sec; + } + if (all_cname) { + *all_cname = all_cn; + } + return err; +} + +/** @internal Add an EDNS padding RR into the answer if requested and required. */ +static int answer_padding(struct kr_request *request) +{ + if (!request || !request->answer || !request->ctx) { + assert(false); + return kr_error(EINVAL); + } + int32_t padding = request->ctx->tls_padding; + knot_pkt_t *answer = request->answer; + knot_rrset_t *opt_rr = answer->opt_rr; + int32_t pad_bytes = -1; + + if (padding == -1) { /* use the default padding policy from libknot */ + pad_bytes = knot_pkt_default_padding_size(answer, opt_rr); + } + if (padding >= 2) { + int32_t max_pad_bytes = knot_edns_get_payload(opt_rr) - (answer->size + knot_rrset_size(opt_rr)); + pad_bytes = MIN(knot_edns_alignment_size(answer->size, knot_rrset_size(opt_rr), padding), + max_pad_bytes); + } + + if (pad_bytes >= 0) { + uint8_t zeros[MAX(1, pad_bytes)]; + memset(zeros, 0, sizeof(zeros)); + int r = knot_edns_add_option(opt_rr, KNOT_EDNS_OPTION_PADDING, + pad_bytes, zeros, &answer->mm); + if (r != KNOT_EOK) { + knot_rrset_clear(opt_rr, &answer->mm); + return kr_error(r); + } + } + return kr_ok(); +} + +static int answer_fail(struct kr_request *request) +{ + knot_pkt_t *answer = request->answer; + int ret = kr_pkt_clear_payload(answer); + knot_wire_clear_ad(answer->wire); + knot_wire_clear_aa(answer->wire); + knot_wire_set_rcode(answer->wire, KNOT_RCODE_SERVFAIL); + if (ret == 0 && answer->opt_rr) { + /* OPT in SERVFAIL response is still useful for cookies/additional info. */ + knot_pkt_begin(answer, KNOT_ADDITIONAL); + answer_padding(request); /* Ignore failed padding in SERVFAIL answer. */ + ret = edns_put(answer, false); + } + return ret; +} + +static int answer_finalize(struct kr_request *request, int state) +{ + struct kr_rplan *rplan = &request->rplan; + knot_pkt_t *answer = request->answer; + + /* Always set SERVFAIL for bogus answers. */ + if (state == KR_STATE_FAIL && rplan->pending.len > 0) { + struct kr_query *last = array_tail(rplan->pending); + if ((last->flags.DNSSEC_WANT) && (last->flags.DNSSEC_BOGUS)) { + return answer_fail(request); + } + } + + struct kr_query *last = rplan->resolved.len > 0 ? array_tail(rplan->resolved) : NULL; + /* TODO ^^^^ this is slightly fragile */ + + /* AD flag. We can only change `secure` from true to false. + * Be conservative. Primary approach: check ranks of all RRs in wire. + * Only "negative answers" need special handling. */ + bool secure = last != NULL && state == KR_STATE_DONE /*< suspicious otherwise */ + && knot_pkt_qtype(answer) != KNOT_RRTYPE_RRSIG; + if (last && (last->flags.STUB)) { + secure = false; /* don't trust forwarding for now */ + } + if (last && (last->flags.DNSSEC_OPTOUT)) { + VERBOSE_MSG(NULL, "AD: opt-out\n"); + secure = false; /* the last answer is insecure due to opt-out */ + } + + const uint16_t reorder = last ? last->reorder : 0; + bool answ_all_cnames = false/*arbitrary*/; + if (request->answ_selected.len > 0) { + assert(answer->current <= KNOT_ANSWER); + /* Write answer records. */ + if (answer->current < KNOT_ANSWER) { + knot_pkt_begin(answer, KNOT_ANSWER); + } + if (write_extra_ranked_records(&request->answ_selected, reorder, + answer, &secure, &answ_all_cnames)) + { + return answer_fail(request); + } + } + + /* Write authority records. */ + if (answer->current < KNOT_AUTHORITY) { + knot_pkt_begin(answer, KNOT_AUTHORITY); + } + if (write_extra_ranked_records(&request->auth_selected, reorder, + answer, &secure, NULL)) { + return answer_fail(request); + } + /* Write additional records. */ + knot_pkt_begin(answer, KNOT_ADDITIONAL); + if (write_extra_records(&request->additional, reorder, answer)) { + return answer_fail(request); + } + /* Write EDNS information */ + if (answer->opt_rr) { + if (request->qsource.flags.tls) { + if (answer_padding(request) != kr_ok()) { + return answer_fail(request); + } + } + knot_pkt_begin(answer, KNOT_ADDITIONAL); + int ret = knot_pkt_put(answer, KNOT_COMPR_HINT_NONE, + answer->opt_rr, KNOT_PF_FREE); + if (ret != KNOT_EOK) { + return answer_fail(request); + } + } + + if (!last) secure = false; /*< should be no-op, mostly documentation */ + /* AD: "negative answers" need more handling. */ + if (kr_response_classify(answer) != PKT_NOERROR + /* Additionally check for CNAME chains that "end in NODATA", + * as those would also be PKT_NOERROR. */ + || (answ_all_cnames && knot_pkt_qtype(answer) != KNOT_RRTYPE_CNAME)) { + + secure = secure && last->flags.DNSSEC_WANT + && !last->flags.DNSSEC_BOGUS && !last->flags.DNSSEC_INSECURE; + } + + if (secure) { + struct kr_query *cname_parent = last->cname_parent; + while (cname_parent != NULL) { + if (cname_parent->flags.DNSSEC_OPTOUT) { + secure = false; + break; + } + cname_parent = cname_parent->cname_parent; + } + } + + /* No detailed analysis ATM, just _SECURE or not. + * LATER: request->rank might better be computed in validator's finish phase. */ + VERBOSE_MSG(last, "AD: request%s classified as SECURE\n", secure ? "" : " NOT"); + request->rank = secure ? KR_RANK_SECURE : KR_RANK_INITIAL; + + /* Clear AD if not secure. ATM answer has AD=1 if requested secured answer. */ + if (!secure) { + knot_wire_clear_ad(answer->wire); + } + + return kr_ok(); +} + +static int query_finalize(struct kr_request *request, struct kr_query *qry, knot_pkt_t *pkt) +{ + int ret = 0; + knot_pkt_begin(pkt, KNOT_ADDITIONAL); + if (!(qry->flags.SAFEMODE)) { + /* Remove any EDNS records from any previous iteration. */ + ret = edns_erase_and_reserve(pkt); + if (ret == 0) { + ret = edns_create(pkt, request->answer, request); + } + if (ret == 0) { + /* Stub resolution (ask for +rd and +do) */ + if (qry->flags.STUB) { + knot_wire_set_rd(pkt->wire); + if (knot_pkt_has_dnssec(request->qsource.packet)) { + knot_edns_set_do(pkt->opt_rr); + } + if (knot_wire_get_cd(request->qsource.packet->wire)) { + knot_wire_set_cd(pkt->wire); + } + /* Full resolution (ask for +cd and +do) */ + } else if (qry->flags.FORWARD) { + knot_wire_set_rd(pkt->wire); + knot_edns_set_do(pkt->opt_rr); + knot_wire_set_cd(pkt->wire); + } else if (qry->flags.DNSSEC_WANT) { + knot_edns_set_do(pkt->opt_rr); + knot_wire_set_cd(pkt->wire); + } + } + } + return ret; +} + +int kr_resolve_begin(struct kr_request *request, struct kr_context *ctx, knot_pkt_t *answer) +{ + /* Initialize request */ + request->ctx = ctx; + request->answer = answer; + request->options = ctx->options; + request->state = KR_STATE_CONSUME; + request->current_query = NULL; + array_init(request->additional); + array_init(request->answ_selected); + array_init(request->auth_selected); + array_init(request->add_selected); + request->answ_validated = false; + request->auth_validated = false; + request->rank = KR_RANK_INITIAL; + request->trace_log = NULL; + request->trace_finish = NULL; + + /* Expect first query */ + kr_rplan_init(&request->rplan, request, &request->pool); + return KR_STATE_CONSUME; +} + +static int resolve_query(struct kr_request *request, const knot_pkt_t *packet) +{ + struct kr_rplan *rplan = &request->rplan; + const knot_dname_t *qname = knot_pkt_qname(packet); + uint16_t qclass = knot_pkt_qclass(packet); + uint16_t qtype = knot_pkt_qtype(packet); + struct kr_query *qry = NULL; + struct kr_context *ctx = request->ctx; + struct kr_cookie_ctx *cookie_ctx = ctx ? &ctx->cookie_ctx : NULL; + + if (qname != NULL) { + qry = kr_rplan_push(rplan, NULL, qname, qclass, qtype); + } else if (cookie_ctx && cookie_ctx->srvr.enabled && + knot_wire_get_qdcount(packet->wire) == 0 && + knot_pkt_has_edns(packet) && + knot_pkt_edns_option(packet, KNOT_EDNS_OPTION_COOKIE)) { + /* Plan empty query only for cookies. */ + qry = kr_rplan_push_empty(rplan, NULL); + } + if (!qry) { + return KR_STATE_FAIL; + } + + if (qname != NULL) { + /* Deferred zone cut lookup for this query. */ + qry->flags.AWAIT_CUT = true; + /* Want DNSSEC if it's posible to secure this name (e.g. is covered by any TA) */ + if ((knot_wire_get_ad(packet->wire) || knot_pkt_has_dnssec(packet)) && + kr_ta_covers_qry(request->ctx, qname, qtype)) { + qry->flags.DNSSEC_WANT = true; + } + } + + /* Initialize answer packet */ + knot_pkt_t *answer = request->answer; + knot_wire_set_qr(answer->wire); + knot_wire_clear_aa(answer->wire); + knot_wire_set_ra(answer->wire); + knot_wire_set_rcode(answer->wire, KNOT_RCODE_NOERROR); + + assert(request->qsource.packet); + if (knot_wire_get_cd(request->qsource.packet->wire)) { + knot_wire_set_cd(answer->wire); + } else if (qry->flags.DNSSEC_WANT) { + knot_wire_set_ad(answer->wire); + } + + /* Expect answer, pop if satisfied immediately */ + ITERATE_LAYERS(request, qry, begin); + if ((request->state & KR_STATE_DONE) != 0) { + kr_rplan_pop(rplan, qry); + } else if (qname == NULL) { + /* it is an empty query which must be resolved by + `begin` layer of cookie module. + If query isn't resolved, fail. */ + request->state = KR_STATE_FAIL; + } + return request->state; +} + +KR_PURE static bool kr_inaddr_equal(const struct sockaddr *a, const struct sockaddr *b) +{ + const int a_len = kr_inaddr_len(a); + const int b_len = kr_inaddr_len(b); + return a_len == b_len && memcmp(kr_inaddr(a), kr_inaddr(b), a_len) == 0; +} + +static void update_nslist_rtt(struct kr_context *ctx, struct kr_query *qry, const struct sockaddr *src) +{ + /* Do not track in safe mode. */ + if (qry->flags.SAFEMODE) { + return; + } + + /* Calculate total resolution time from the time the query was generated. */ + uint64_t elapsed = kr_now() - qry->timestamp_mono; + elapsed = elapsed > UINT_MAX ? UINT_MAX : elapsed; + + /* NSs in the preference list prior to the one who responded will be penalised + * with the RETRY timer interval. This is because we know they didn't respond + * for N retries, so their RTT must be at least N * RETRY. + * The NS in the preference list that responded will have RTT relative to the + * time when the query was sent out, not when it was originated. + */ + for (size_t i = 0; i < KR_NSREP_MAXADDR; ++i) { + const struct sockaddr *addr = &qry->ns.addr[i].ip; + if (addr->sa_family == AF_UNSPEC) { + break; + } + /* If this address is the source of the answer, update its RTT */ + if (kr_inaddr_equal(src, addr)) { + kr_nsrep_update_rtt(&qry->ns, addr, elapsed, ctx->cache_rtt, KR_NS_UPDATE); + WITH_VERBOSE(qry) { + char addr_str[INET6_ADDRSTRLEN]; + inet_ntop(addr->sa_family, kr_inaddr(addr), addr_str, sizeof(addr_str)); + VERBOSE_MSG(qry, "<= server: '%s' rtt: %"PRIu64" ms\n", + addr_str, elapsed); + } + } else { + /* Response didn't come from this IP, but we know the RTT must be at least + * several RETRY timer tries, e.g. if we have addresses [a, b, c] and we have + * tried [a, b] when the answer from 'a' came after 350ms, then we know + * that 'b' didn't respond for at least 350 - (1 * 300) ms. We can't say that + * its RTT is 50ms, but we can say that its score shouldn't be less than 50. */ + kr_nsrep_update_rtt(&qry->ns, addr, elapsed, ctx->cache_rtt, KR_NS_MAX); + WITH_VERBOSE(qry) { + char addr_str[INET6_ADDRSTRLEN]; + inet_ntop(addr->sa_family, kr_inaddr(addr), addr_str, sizeof(addr_str)); + VERBOSE_MSG(qry, "<= server: '%s' rtt: >= %"PRIu64" ms\n", + addr_str, elapsed); + } + } + /* Subtract query start time from elapsed time */ + if (elapsed < KR_CONN_RETRY) { + break; + } + elapsed = elapsed - KR_CONN_RETRY; + } +} + +static void update_nslist_score(struct kr_request *request, struct kr_query *qry, const struct sockaddr *src, knot_pkt_t *packet) +{ + struct kr_context *ctx = request->ctx; + /* On successful answer, update preference list RTT and penalise timer */ + if (request->state != KR_STATE_FAIL) { + /* Update RTT information for preference list */ + update_nslist_rtt(ctx, qry, src); + /* Do not complete NS address resolution on soft-fail. */ + const int rcode = packet ? knot_wire_get_rcode(packet->wire) : 0; + if (rcode != KNOT_RCODE_SERVFAIL && rcode != KNOT_RCODE_REFUSED) { + qry->flags.AWAIT_IPV6 = false; + qry->flags.AWAIT_IPV4 = false; + } else { /* Penalize SERVFAILs. */ + kr_nsrep_update_rtt(&qry->ns, src, KR_NS_PENALTY, ctx->cache_rtt, KR_NS_ADD); + } + } +} + +static bool resolution_time_exceeded(struct kr_query *qry, uint64_t now) +{ + uint64_t resolving_time = now - qry->creation_time_mono; + if (resolving_time > KR_RESOLVE_TIME_LIMIT) { + WITH_VERBOSE(qry) { + VERBOSE_MSG(qry, "query resolution time limit exceeded\n"); + } + return true; + } + return false; +} + +int kr_resolve_consume(struct kr_request *request, const struct sockaddr *src, knot_pkt_t *packet) +{ + struct kr_rplan *rplan = &request->rplan; + + /* Empty resolution plan, push packet as the new query */ + if (packet && kr_rplan_empty(rplan)) { + if (answer_prepare(request, packet) != 0) { + return KR_STATE_FAIL; + } + return resolve_query(request, packet); + } + + /* Different processing for network error */ + struct kr_query *qry = array_tail(rplan->pending); + /* Check overall resolution time */ + if (resolution_time_exceeded(qry, kr_now())) { + return KR_STATE_FAIL; + } + bool tried_tcp = (qry->flags.TCP); + if (!packet || packet->size == 0) { + if (tried_tcp) { + request->state = KR_STATE_FAIL; + } else { + qry->flags.TCP = true; + } + } else { + /* Packet cleared, derandomize QNAME. */ + knot_dname_t *qname_raw = knot_pkt_qname(packet); + if (qname_raw && qry->secret != 0) { + randomized_qname_case(qname_raw, qry->secret); + } + request->state = KR_STATE_CONSUME; + if (qry->flags.CACHED) { + ITERATE_LAYERS(request, qry, consume, packet); + } else { + /* Fill in source and latency information. */ + request->upstream.rtt = kr_now() - qry->timestamp_mono; + request->upstream.addr = src; + ITERATE_LAYERS(request, qry, consume, packet); + /* Clear temporary information */ + request->upstream.addr = NULL; + request->upstream.rtt = 0; + } + } + + /* Track RTT for iterative answers */ + if (src && !(qry->flags.CACHED)) { + update_nslist_score(request, qry, src, packet); + } + /* Resolution failed, invalidate current NS. */ + if (request->state == KR_STATE_FAIL) { + invalidate_ns(rplan, qry); + qry->flags.RESOLVED = false; + } + + /* Pop query if resolved. */ + if (request->state == KR_STATE_YIELD) { + return KR_STATE_PRODUCE; /* Requery */ + } else if (qry->flags.RESOLVED) { + kr_rplan_pop(rplan, qry); + } else if (!tried_tcp && (qry->flags.TCP)) { + return KR_STATE_PRODUCE; /* Requery over TCP */ + } else { /* Clear query flags for next attempt */ + qry->flags.CACHED = false; + if (!request->options.TCP) { + qry->flags.TCP = false; + } + } + + ITERATE_LAYERS(request, qry, reset); + + /* Do not finish with bogus answer. */ + if (qry->flags.DNSSEC_BOGUS) { + return KR_STATE_FAIL; + } + + return kr_rplan_empty(&request->rplan) ? KR_STATE_DONE : KR_STATE_PRODUCE; +} + +/** @internal Spawn subrequest in current zone cut (no minimization or lookup). */ +static struct kr_query *zone_cut_subreq(struct kr_rplan *rplan, struct kr_query *parent, + const knot_dname_t *qname, uint16_t qtype) +{ + struct kr_query *next = kr_rplan_push(rplan, parent, qname, parent->sclass, qtype); + if (!next) { + return NULL; + } + kr_zonecut_set(&next->zone_cut, parent->zone_cut.name); + if (kr_zonecut_copy(&next->zone_cut, &parent->zone_cut) != 0 || + kr_zonecut_copy_trust(&next->zone_cut, &parent->zone_cut) != 0) { + return NULL; + } + next->flags.NO_MINIMIZE = true; + if (parent->flags.DNSSEC_WANT) { + next->flags.DNSSEC_WANT = true; + } + return next; +} + +static int forward_trust_chain_check(struct kr_request *request, struct kr_query *qry, bool resume) +{ + struct kr_rplan *rplan = &request->rplan; + map_t *trust_anchors = &request->ctx->trust_anchors; + map_t *negative_anchors = &request->ctx->negative_anchors; + + if (qry->parent != NULL && + !(qry->forward_flags.CNAME) && + !(qry->flags.DNS64_MARK) && + knot_dname_in_bailiwick(qry->zone_cut.name, qry->parent->zone_cut.name) >= 0) { + return KR_STATE_PRODUCE; + } + + assert(qry->flags.FORWARD); + + if (!trust_anchors) { + qry->flags.AWAIT_CUT = false; + return KR_STATE_PRODUCE; + } + + if (qry->flags.DNSSEC_INSECURE) { + qry->flags.AWAIT_CUT = false; + return KR_STATE_PRODUCE; + } + + if (qry->forward_flags.NO_MINIMIZE) { + qry->flags.AWAIT_CUT = false; + return KR_STATE_PRODUCE; + } + + const knot_dname_t *start_name = qry->sname; + if ((qry->flags.AWAIT_CUT) && !resume) { + qry->flags.AWAIT_CUT = false; + const knot_dname_t *longest_ta = kr_ta_get_longest_name(trust_anchors, qry->sname); + if (longest_ta) { + start_name = longest_ta; + qry->zone_cut.name = knot_dname_copy(start_name, qry->zone_cut.pool); + qry->flags.DNSSEC_WANT = true; + } else { + qry->flags.DNSSEC_WANT = false; + return KR_STATE_PRODUCE; + } + } + + bool has_ta = (qry->zone_cut.trust_anchor != NULL); + knot_dname_t *ta_name = (has_ta ? qry->zone_cut.trust_anchor->owner : NULL); + bool refetch_ta = (!has_ta || !knot_dname_is_equal(qry->zone_cut.name, ta_name)); + bool is_dnskey_subreq = kr_rplan_satisfies(qry, ta_name, KNOT_CLASS_IN, KNOT_RRTYPE_DNSKEY); + bool refetch_key = has_ta && (!qry->zone_cut.key || !knot_dname_is_equal(ta_name, qry->zone_cut.key->owner)); + if (refetch_key && !is_dnskey_subreq) { + struct kr_query *next = zone_cut_subreq(rplan, qry, ta_name, KNOT_RRTYPE_DNSKEY); + if (!next) { + return KR_STATE_FAIL; + } + return KR_STATE_DONE; + } + + int name_offset = 1; + const knot_dname_t *wanted_name; + bool nods, ds_req, ns_req, minimized, ns_exist; + do { + wanted_name = start_name; + ds_req = false; + ns_req = false; + ns_exist = true; + + int cut_labels = knot_dname_labels(qry->zone_cut.name, NULL); + int wanted_name_labels = knot_dname_labels(wanted_name, NULL); + while (wanted_name[0] && wanted_name_labels > cut_labels + name_offset) { + wanted_name = knot_wire_next_label(wanted_name, NULL); + wanted_name_labels -= 1; + } + minimized = (wanted_name != qry->sname); + + for (int i = 0; i < request->rplan.resolved.len; ++i) { + struct kr_query *q = request->rplan.resolved.at[i]; + if (q->parent == qry && + q->sclass == qry->sclass && + (q->stype == KNOT_RRTYPE_DS || q->stype == KNOT_RRTYPE_NS) && + knot_dname_is_equal(q->sname, wanted_name)) { + if (q->stype == KNOT_RRTYPE_DS) { + ds_req = true; + if (q->flags.CNAME) { + ns_exist = false; + } else if (!(q->flags.DNSSEC_OPTOUT)) { + int ret = kr_dnssec_matches_name_and_type(&request->auth_selected, q->uid, + wanted_name, KNOT_RRTYPE_NS); + ns_exist = (ret == kr_ok()); + } + } else { + if (q->flags.CNAME) { + ns_exist = false; + } + ns_req = true; + } + } + } + + if (ds_req && ns_exist && !ns_req && (minimized || resume)) { + struct kr_query *next = zone_cut_subreq(rplan, qry, wanted_name, + KNOT_RRTYPE_NS); + if (!next) { + return KR_STATE_FAIL; + } + return KR_STATE_DONE; + } + + if (qry->parent == NULL && (qry->flags.CNAME) && + ds_req && ns_req) { + return KR_STATE_PRODUCE; + } + + /* set `nods` */ + if ((qry->stype == KNOT_RRTYPE_DS) && + knot_dname_is_equal(wanted_name, qry->sname)) { + nods = true; + } else if (resume && !ds_req) { + nods = false; + } else if (!minimized && qry->stype != KNOT_RRTYPE_DNSKEY) { + nods = true; + } else { + nods = ds_req; + } + name_offset += 1; + } while (ds_req && (ns_req || !ns_exist) && minimized); + + /* Disable DNSSEC if it enters NTA. */ + if (kr_ta_get(negative_anchors, wanted_name)){ + VERBOSE_MSG(qry, ">< negative TA, going insecure\n"); + qry->flags.DNSSEC_WANT = false; + } + + /* Enable DNSSEC if enters a new island of trust. */ + bool want_secured = (qry->flags.DNSSEC_WANT) && + !knot_wire_get_cd(request->qsource.packet->wire); + if (!(qry->flags.DNSSEC_WANT) && + !knot_wire_get_cd(request->qsource.packet->wire) && + kr_ta_get(trust_anchors, wanted_name)) { + qry->flags.DNSSEC_WANT = true; + want_secured = true; + WITH_VERBOSE(qry) { + KR_DNAME_GET_STR(qname_str, wanted_name); + VERBOSE_MSG(qry, ">< TA: '%s'\n", qname_str); + } + } + + if (want_secured && !qry->zone_cut.trust_anchor) { + knot_rrset_t *ta_rr = kr_ta_get(trust_anchors, wanted_name); + if (!ta_rr) { + char name[] = "\0"; + ta_rr = kr_ta_get(trust_anchors, (knot_dname_t*)name); + } + if (ta_rr) { + qry->zone_cut.trust_anchor = knot_rrset_copy(ta_rr, qry->zone_cut.pool); + } + } + + has_ta = (qry->zone_cut.trust_anchor != NULL); + ta_name = (has_ta ? qry->zone_cut.trust_anchor->owner : NULL); + refetch_ta = (!has_ta || !knot_dname_is_equal(wanted_name, ta_name)); + if (!nods && want_secured && refetch_ta) { + struct kr_query *next = zone_cut_subreq(rplan, qry, wanted_name, + KNOT_RRTYPE_DS); + if (!next) { + return KR_STATE_FAIL; + } + return KR_STATE_DONE; + } + + /* Try to fetch missing DNSKEY. + * Do not fetch if this is a DNSKEY subrequest to avoid circular dependency. */ + is_dnskey_subreq = kr_rplan_satisfies(qry, ta_name, KNOT_CLASS_IN, KNOT_RRTYPE_DNSKEY); + refetch_key = has_ta && (!qry->zone_cut.key || !knot_dname_is_equal(ta_name, qry->zone_cut.key->owner)); + if (want_secured && refetch_key && !is_dnskey_subreq) { + struct kr_query *next = zone_cut_subreq(rplan, qry, ta_name, KNOT_RRTYPE_DNSKEY); + if (!next) { + return KR_STATE_FAIL; + } + return KR_STATE_DONE; + } + + return KR_STATE_PRODUCE; +} + +/* @todo: Validator refactoring, keep this in driver for now. */ +static int trust_chain_check(struct kr_request *request, struct kr_query *qry) +{ + struct kr_rplan *rplan = &request->rplan; + map_t *trust_anchors = &request->ctx->trust_anchors; + map_t *negative_anchors = &request->ctx->negative_anchors; + + /* Disable DNSSEC if it enters NTA. */ + if (kr_ta_get(negative_anchors, qry->zone_cut.name)){ + VERBOSE_MSG(qry, ">< negative TA, going insecure\n"); + qry->flags.DNSSEC_WANT = false; + qry->flags.DNSSEC_INSECURE = true; + } + if (qry->flags.DNSSEC_NODS) { + /* This is the next query iteration with minimized qname. + * At previous iteration DS non-existance has been proven */ + qry->flags.DNSSEC_NODS = false; + qry->flags.DNSSEC_WANT = false; + qry->flags.DNSSEC_INSECURE = true; + } + /* Enable DNSSEC if entering a new (or different) island of trust, + * and update the TA RRset if required. */ + bool want_secured = (qry->flags.DNSSEC_WANT) && + !knot_wire_get_cd(request->qsource.packet->wire); + knot_rrset_t *ta_rr = kr_ta_get(trust_anchors, qry->zone_cut.name); + if (!knot_wire_get_cd(request->qsource.packet->wire) && ta_rr) { + qry->flags.DNSSEC_WANT = true; + want_secured = true; + + if (qry->zone_cut.trust_anchor == NULL + || !knot_dname_is_equal(qry->zone_cut.trust_anchor->owner, qry->zone_cut.name)) { + mm_free(qry->zone_cut.pool, qry->zone_cut.trust_anchor); + qry->zone_cut.trust_anchor = knot_rrset_copy(ta_rr, qry->zone_cut.pool); + + WITH_VERBOSE(qry) { + KR_DNAME_GET_STR(qname_str, ta_rr->owner); + VERBOSE_MSG(qry, ">< TA: '%s'\n", qname_str); + } + } + } + + /* Try to fetch missing DS (from above the cut). */ + const bool has_ta = (qry->zone_cut.trust_anchor != NULL); + const knot_dname_t *ta_name = (has_ta ? qry->zone_cut.trust_anchor->owner : NULL); + const bool refetch_ta = !has_ta || !knot_dname_is_equal(qry->zone_cut.name, ta_name); + if (want_secured && refetch_ta) { + /* @todo we could fetch the information from the parent cut, but we don't remember that now */ + struct kr_query *next = kr_rplan_push(rplan, qry, qry->zone_cut.name, qry->sclass, KNOT_RRTYPE_DS); + if (!next) { + return KR_STATE_FAIL; + } + next->flags.AWAIT_CUT = true; + next->flags.DNSSEC_WANT = true; + return KR_STATE_DONE; + } + /* Try to fetch missing DNSKEY (either missing or above current cut). + * Do not fetch if this is a DNSKEY subrequest to avoid circular dependency. */ + const bool is_dnskey_subreq = kr_rplan_satisfies(qry, ta_name, KNOT_CLASS_IN, KNOT_RRTYPE_DNSKEY); + const bool refetch_key = has_ta && (!qry->zone_cut.key || !knot_dname_is_equal(ta_name, qry->zone_cut.key->owner)); + if (want_secured && refetch_key && !is_dnskey_subreq) { + struct kr_query *next = zone_cut_subreq(rplan, qry, ta_name, KNOT_RRTYPE_DNSKEY); + if (!next) { + return KR_STATE_FAIL; + } + return KR_STATE_DONE; + } + + return KR_STATE_PRODUCE; +} + +/** @internal Check current zone cut status and credibility, spawn subrequests if needed. */ +static int zone_cut_check(struct kr_request *request, struct kr_query *qry, knot_pkt_t *packet) +/* TODO: using cache on this point in this way just isn't nice; remove in time */ +{ + /* Stub mode, just forward and do not solve cut. */ + if (qry->flags.STUB) { + return KR_STATE_PRODUCE; + } + + /* Forwarding to upstream resolver mode. + * Since forwarding targets already are in qry->ns - + * cut fetching is not needed. */ + if (qry->flags.FORWARD) { + return forward_trust_chain_check(request, qry, false); + } + if (!(qry->flags.AWAIT_CUT)) { + /* The query was resolved from cache. + * Spawn DS \ DNSKEY requests if needed and exit */ + return trust_chain_check(request, qry); + } + + /* The query wasn't resolved from cache, + * now it's the time to look up closest zone cut from cache. */ + struct kr_cache *cache = &request->ctx->cache; + if (!kr_cache_is_open(cache)) { + int ret = kr_zonecut_set_sbelt(request->ctx, &qry->zone_cut); + if (ret != 0) { + return KR_STATE_FAIL; + } + VERBOSE_MSG(qry, "=> no cache open, using root hints\n"); + qry->flags.AWAIT_CUT = false; + return KR_STATE_DONE; + } + + const knot_dname_t *requested_name = qry->sname; + /* If at/subdomain of parent zone cut, start from its encloser. + * This is for case when we get to a dead end + * (and need glue from parent), or DS refetch. */ + if (qry->parent) { + const knot_dname_t *parent = qry->parent->zone_cut.name; + if (parent[0] != '\0' + && knot_dname_in_bailiwick(qry->sname, parent) >= 0) { + requested_name = knot_wire_next_label(parent, NULL); + } + } else if ((qry->stype == KNOT_RRTYPE_DS) && (qry->sname[0] != '\0')) { + /* If this is explicit DS query, start from encloser too. */ + requested_name = knot_wire_next_label(requested_name, NULL); + } + + int state = KR_STATE_FAIL; + do { + state = ns_fetch_cut(qry, requested_name, request, packet); + if (state == KR_STATE_DONE || state == KR_STATE_FAIL) { + return state; + } else if (state == KR_STATE_CONSUME) { + requested_name = knot_wire_next_label(requested_name, NULL); + } + } while (state == KR_STATE_CONSUME); + + /* Update minimized QNAME if zone cut changed */ + if (qry->zone_cut.name && qry->zone_cut.name[0] != '\0' && !(qry->flags.NO_MINIMIZE)) { + if (kr_make_query(qry, packet) != 0) { + return KR_STATE_FAIL; + } + } + qry->flags.AWAIT_CUT = false; + + /* Check trust chain */ + return trust_chain_check(request, qry); +} + +int kr_resolve_produce(struct kr_request *request, struct sockaddr **dst, int *type, knot_pkt_t *packet) +{ + struct kr_rplan *rplan = &request->rplan; + unsigned ns_election_iter = 0; + + /* No query left for resolution */ + if (kr_rplan_empty(rplan)) { + return KR_STATE_FAIL; + } + /* If we have deferred answers, resume them. */ + struct kr_query *qry = array_tail(rplan->pending); + if (qry->deferred != NULL) { + /* @todo: Refactoring validator, check trust chain before resuming. */ + int state = 0; + if (((qry->flags.FORWARD) == 0) || + ((qry->stype == KNOT_RRTYPE_DS) && (qry->flags.CNAME))) { + state = trust_chain_check(request, qry); + } else { + state = forward_trust_chain_check(request, qry, true); + } + + switch(state) { + case KR_STATE_FAIL: return KR_STATE_FAIL; + case KR_STATE_DONE: return KR_STATE_PRODUCE; + default: break; + } + VERBOSE_MSG(qry, "=> resuming yielded answer\n"); + struct kr_layer_pickle *pickle = qry->deferred; + request->state = KR_STATE_YIELD; + set_yield(&request->answ_selected, qry->uid, false); + set_yield(&request->auth_selected, qry->uid, false); + RESUME_LAYERS(layer_id(request, pickle->api), request, qry, consume, pickle->pkt); + if (request->state != KR_STATE_YIELD) { + /* No new deferred answers, take the next */ + qry->deferred = pickle->next; + } + } else { + /* Caller is interested in always tracking a zone cut, even if the answer is cached + * this is normally not required, and incurrs another cache lookups for cached answer. */ + if (qry->flags.ALWAYS_CUT) { + if (!(qry->flags.STUB)) { + switch(zone_cut_check(request, qry, packet)) { + case KR_STATE_FAIL: return KR_STATE_FAIL; + case KR_STATE_DONE: return KR_STATE_PRODUCE; + default: break; + } + } + } + /* Resolve current query and produce dependent or finish */ + request->state = KR_STATE_PRODUCE; + ITERATE_LAYERS(request, qry, produce, packet); + if (request->state != KR_STATE_FAIL && knot_wire_get_qr(packet->wire)) { + /* Produced an answer from cache, consume it. */ + qry->secret = 0; + request->state = KR_STATE_CONSUME; + ITERATE_LAYERS(request, qry, consume, packet); + } + } + switch(request->state) { + case KR_STATE_FAIL: return request->state; + case KR_STATE_CONSUME: break; + case KR_STATE_DONE: + default: /* Current query is done */ + if (qry->flags.RESOLVED && request->state != KR_STATE_YIELD) { + kr_rplan_pop(rplan, qry); + } + ITERATE_LAYERS(request, qry, reset); + return kr_rplan_empty(rplan) ? KR_STATE_DONE : KR_STATE_PRODUCE; + } + + + /* This query has RD=0 or is ANY, stop here. */ + if (qry->stype == KNOT_RRTYPE_ANY || + !knot_wire_get_rd(request->qsource.packet->wire)) { + VERBOSE_MSG(qry, "=> qtype is ANY or RD=0, bail out\n"); + return KR_STATE_FAIL; + } + + /* Update zone cut, spawn new subrequests. */ + if (!(qry->flags.STUB)) { + int state = zone_cut_check(request, qry, packet); + switch(state) { + case KR_STATE_FAIL: return KR_STATE_FAIL; + case KR_STATE_DONE: return KR_STATE_PRODUCE; + default: break; + } + } + +ns_election: + + /* If the query has already selected a NS and is waiting for IPv4/IPv6 record, + * elect best address only, otherwise elect a completely new NS. + */ + if(++ns_election_iter >= KR_ITER_LIMIT) { + VERBOSE_MSG(qry, "=> couldn't converge NS selection, bail out\n"); + return KR_STATE_FAIL; + } + + const struct kr_qflags qflg = qry->flags; + const bool retry = qflg.TCP || qflg.BADCOOKIE_AGAIN; + if (qflg.AWAIT_IPV4 || qflg.AWAIT_IPV6) { + kr_nsrep_elect_addr(qry, request->ctx); + } else if (qflg.FORWARD || qflg.STUB) { + kr_nsrep_sort(&qry->ns, request->ctx); + if (qry->ns.score > KR_NS_MAX_SCORE) { + /* At the moment all NS have bad reputation. + * But there can be existing connections*/ + VERBOSE_MSG(qry, "=> no valid NS left\n"); + return KR_STATE_FAIL; + } + } else if (!qry->ns.name || !retry) { /* Keep NS when requerying/stub/badcookie. */ + /* Root DNSKEY must be fetched from the hints to avoid chicken and egg problem. */ + if (qry->sname[0] == '\0' && qry->stype == KNOT_RRTYPE_DNSKEY) { + kr_zonecut_set_sbelt(request->ctx, &qry->zone_cut); + qry->flags.NO_THROTTLE = true; /* Pick even bad SBELT servers */ + } + kr_nsrep_elect(qry, request->ctx); + if (qry->ns.score > KR_NS_MAX_SCORE) { + if (kr_zonecut_is_empty(&qry->zone_cut)) { + VERBOSE_MSG(qry, "=> no NS with an address\n"); + } else { + VERBOSE_MSG(qry, "=> no valid NS left\n"); + } + if (!qry->flags.NO_NS_FOUND) { + qry->flags.NO_NS_FOUND = true; + } else { + ITERATE_LAYERS(request, qry, reset); + kr_rplan_pop(rplan, qry); + } + return KR_STATE_PRODUCE; + } + } + + /* Resolve address records */ + if (qry->ns.addr[0].ip.sa_family == AF_UNSPEC) { + int ret = ns_resolve_addr(qry, request); + if (ret != 0) { + qry->flags.AWAIT_IPV6 = false; + qry->flags.AWAIT_IPV4 = false; + qry->flags.TCP = false; + qry->ns.name = NULL; + goto ns_election; /* Must try different NS */ + } + ITERATE_LAYERS(request, qry, reset); + return KR_STATE_PRODUCE; + } + + /* Randomize query case (if not in safe mode or turned off) */ + qry->secret = (qry->flags.SAFEMODE || qry->flags.NO_0X20) + ? 0 : kr_rand_bytes(sizeof(qry->secret)); + knot_dname_t *qname_raw = knot_pkt_qname(packet); + randomized_qname_case(qname_raw, qry->secret); + + /* + * Additional query is going to be finalized when calling + * kr_resolve_checkout(). + */ + qry->timestamp_mono = kr_now(); + *dst = &qry->ns.addr[0].ip; + *type = (qry->flags.TCP) ? SOCK_STREAM : SOCK_DGRAM; + return request->state; +} + +#if defined(ENABLE_COOKIES) +/** Update DNS cookie data in packet. */ +static bool outbound_request_update_cookies(struct kr_request *req, + const struct sockaddr *src, + const struct sockaddr *dst) +{ + assert(req); + + /* RFC7873 4.1 strongly requires server address. */ + if (!dst) { + return false; + } + + struct kr_cookie_settings *clnt_sett = &req->ctx->cookie_ctx.clnt; + + /* Cookies disabled or packet has no EDNS section. */ + if (!clnt_sett->enabled) { + return true; + } + + /* + * RFC7873 4.1 recommends using also the client address. The matter is + * also discussed in section 6. + */ + + kr_request_put_cookie(&clnt_sett->current, req->ctx->cache_cookie, + src, dst, req); + + return true; +} +#endif /* defined(ENABLE_COOKIES) */ + +int kr_resolve_checkout(struct kr_request *request, const struct sockaddr *src, + struct sockaddr *dst, int type, knot_pkt_t *packet) +{ + /* @todo: Update documentation if this function becomes approved. */ + + struct kr_rplan *rplan = &request->rplan; + + if (knot_wire_get_qr(packet->wire) != 0) { + return kr_ok(); + } + + /* No query left for resolution */ + if (kr_rplan_empty(rplan)) { + return kr_error(EINVAL); + } + struct kr_query *qry = array_tail(rplan->pending); + +#if defined(ENABLE_COOKIES) + /* Update DNS cookies in request. */ + if (type == SOCK_DGRAM) { /* @todo: Add cookies also over TCP? */ + /* + * The actual server IP address is needed before generating the + * actual cookie. If we don't know the server address then we + * also don't know the actual cookie size. + */ + if (!outbound_request_update_cookies(request, src, dst)) { + return kr_error(EINVAL); + } + } +#endif /* defined(ENABLE_COOKIES) */ + + int ret = query_finalize(request, qry, packet); + if (ret != 0) { + return kr_error(EINVAL); + } + + /* Track changes in minimization secret to enable/disable minimization */ + uint32_t old_minimization_secret = qry->secret; + + /* Run the checkout layers and cancel on failure. + * The checkout layer doesn't persist the state, so canceled subrequests + * don't affect the resolution or rest of the processing. */ + int state = request->state; + ITERATE_LAYERS(request, qry, checkout, packet, dst, type); + if (request->state == KR_STATE_FAIL) { + request->state = state; /* Restore */ + return kr_error(ECANCELED); + } + + /* Randomize query case (if secret changed) */ + knot_dname_t *qname = (knot_dname_t *)knot_pkt_qname(packet); + if (qry->secret != old_minimization_secret) { + randomized_qname_case(qname, qry->secret); + } + + /* Write down OPT unless in safemode */ + if (!(qry->flags.SAFEMODE)) { + ret = edns_put(packet, true); + if (ret != 0) { + return kr_error(EINVAL); + } + } + + WITH_VERBOSE(qry) { + + KR_DNAME_GET_STR(qname_str, knot_pkt_qname(packet)); + KR_DNAME_GET_STR(zonecut_str, qry->zone_cut.name); + KR_RRTYPE_GET_STR(type_str, knot_pkt_qtype(packet)); + + for (size_t i = 0; i < KR_NSREP_MAXADDR; ++i) { + struct sockaddr *addr = &qry->ns.addr[i].ip; + if (addr->sa_family == AF_UNSPEC) { + break; + } + if (!kr_inaddr_equal(dst, addr)) { + continue; + } + const char *ns_str = kr_straddr(addr); + VERBOSE_MSG(qry, + "=> id: '%05u' querying: '%s' score: %u zone cut: '%s' " + "qname: '%s' qtype: '%s' proto: '%s'\n", + qry->id, ns_str ? ns_str : "", qry->ns.score, zonecut_str, + qname_str, type_str, (qry->flags.TCP) ? "tcp" : "udp"); + + break; + }} + + return kr_ok(); +} + +int kr_resolve_finish(struct kr_request *request, int state) +{ + /* Finalize answer and construct wire-buffer. */ + ITERATE_LAYERS(request, NULL, answer_finalize); + if (request->state == KR_STATE_FAIL) { + state = KR_STATE_FAIL; + } else if (answer_finalize(request, state) != 0) { + state = KR_STATE_FAIL; + } + + /* Error during processing, internal failure */ + if (state != KR_STATE_DONE) { + knot_pkt_t *answer = request->answer; + if (knot_wire_get_rcode(answer->wire) == KNOT_RCODE_NOERROR) { + knot_wire_clear_ad(answer->wire); + knot_wire_clear_aa(answer->wire); + knot_wire_set_rcode(answer->wire, KNOT_RCODE_SERVFAIL); + } + } + + request->state = state; + ITERATE_LAYERS(request, NULL, finish); + +#ifndef NOVERBOSELOG + struct kr_rplan *rplan = &request->rplan; + struct kr_query *last = kr_rplan_last(rplan); + VERBOSE_MSG(last, "finished: %d, queries: %zu, mempool: %zu B\n", + request->state, rplan->resolved.len, (size_t) mp_total_size(request->pool.ctx)); +#endif + + /* Trace request finish */ + if (request->trace_finish) { + request->trace_finish(request); + } + + /* Uninstall all tracepoints */ + request->trace_finish = NULL; + request->trace_log = NULL; + + return KR_STATE_DONE; +} + +struct kr_rplan *kr_resolve_plan(struct kr_request *request) +{ + if (request) { + return &request->rplan; + } + return NULL; +} + +knot_mm_t *kr_resolve_pool(struct kr_request *request) +{ + if (request) { + return &request->pool; + } + return NULL; +} + +#undef VERBOSE_MSG diff --git a/lib/resolve.h b/lib/resolve.h new file mode 100644 index 0000000..60d80bb --- /dev/null +++ b/lib/resolve.h @@ -0,0 +1,332 @@ +/* Copyright (C) 2014-2017 CZ.NIC, z.s.p.o. + + This program is free software: you can redistribute it and/or modify + it under the terms of the GNU General Public License as published by + the Free Software Foundation, either version 3 of the License, or + (at your option) any later version. + + This program is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the + GNU General Public License for more details. + + You should have received a copy of the GNU General Public License + along with this program. If not, see . + */ + +#pragma once + +#include +#include +#include + +#include "lib/cookies/control.h" +#include "lib/cookies/lru_cache.h" +#include "lib/layer.h" +#include "lib/generic/map.h" +#include "lib/generic/array.h" +#include "lib/nsrep.h" +#include "lib/rplan.h" +#include "lib/module.h" +#include "lib/cache/api.h" + +/** + * @file resolve.h + * @brief The API provides an API providing a "consumer-producer"-like interface to enable + * user to plug it into existing event loop or I/O code. + * + * # Example usage of the iterative API: + * + * @code{.c} + * + * // Create request and its memory pool + * struct kr_request req = { + * .pool = { + * .ctx = mp_new (4096), + * .alloc = (mm_alloc_t) mp_alloc + * } + * }; + * + * // Setup and provide input query + * int state = kr_resolve_begin(&req, ctx, final_answer); + * state = kr_resolve_consume(&req, query); + * + * // Generate answer + * while (state == KR_STATE_PRODUCE) { + * + * // Additional query generate, do the I/O and pass back answer + * state = kr_resolve_produce(&req, &addr, &type, query); + * while (state == KR_STATE_CONSUME) { + * int ret = sendrecv(addr, proto, query, resp); + * + * // If I/O fails, make "resp" empty + * state = kr_resolve_consume(&request, addr, resp); + * knot_pkt_clear(resp); + * } + * knot_pkt_clear(query); + * } + * + * // "state" is either DONE or FAIL + * kr_resolve_finish(&request, state); + * + * @endcode + */ + + +/** + * RRset rank - for cache and ranked_rr_*. + * + * The rank meaning consists of one independent flag - KR_RANK_AUTH, + * and the rest have meaning of values where only one can hold at any time. + * You can use one of the enums as a safe initial value, optionally | KR_RANK_AUTH; + * otherwise it's best to manipulate ranks via the kr_rank_* functions. + * + * @note The representation is complicated by restrictions on integer comparison: + * - AUTH must be > than !AUTH + * - AUTH INSECURE must be > than AUTH (because it attempted validation) + * - !AUTH SECURE must be > than AUTH (because it's valid) + * + * See also: + * https://tools.ietf.org/html/rfc2181#section-5.4.1 + * https://tools.ietf.org/html/rfc4035#section-4.3 + */ +enum kr_rank { + /* Initial-like states. No validation has been attempted (yet). */ + KR_RANK_INITIAL = 0, /**< Did not attempt to validate. It's assumed + compulsory to validate (or prove insecure). */ + KR_RANK_OMIT, /**< Do not attempt to validate. + (And don't consider it a validation failure.) */ + KR_RANK_TRY, /**< Attempt to validate, but failures are non-fatal. */ + + /* Failure states. These have higher value because they have more information. */ + KR_RANK_INDET = 4, /**< Unable to determine whether it should be secure. */ + KR_RANK_BOGUS, /**< Ought to be secure but isn't. */ + KR_RANK_MISMATCH, + KR_RANK_MISSING, /**< Unable to obtain a good signature. */ + + /** Proven to be insecure, i.e. we have a chain of trust from TAs + * that cryptographically denies the possibility of existence + * of a positive chain of trust from the TAs to the record. */ + KR_RANK_INSECURE = 8, + + /** Authoritative data flag; the chain of authority was "verified". + * Even if not set, only in-bailiwick stuff is acceptable, + * i.e. almost authoritative (example: mandatory glue and its NS RR). */ + KR_RANK_AUTH = 16, + + KR_RANK_SECURE = 32, /**< Verified whole chain of trust from the closest TA. */ + /* @note Rank must not exceed 6 bits */ +}; + +/** Check that a rank value is valid. Meant for assertions. */ +bool kr_rank_check(uint8_t rank) KR_PURE; + +/** Test the presence of any flag/state in a rank, i.e. including KR_RANK_AUTH. */ +static inline bool kr_rank_test(uint8_t rank, uint8_t kr_flag) +{ + assert(kr_rank_check(rank) && kr_rank_check(kr_flag)); + if (kr_flag == KR_RANK_AUTH) { + return rank & KR_RANK_AUTH; + } + assert(!(kr_flag & KR_RANK_AUTH)); + /* The rest are exclusive values - exactly one has to be set. */ + return (rank & ~KR_RANK_AUTH) == kr_flag; +} + +/** Set the rank state. The _AUTH flag is kept as it was. */ +static inline void kr_rank_set(uint8_t *rank, uint8_t kr_flag) +{ + assert(rank && kr_rank_check(*rank)); + assert(kr_rank_check(kr_flag) && !(kr_flag & KR_RANK_AUTH)); + *rank = kr_flag | (*rank & KR_RANK_AUTH); +} + + +/** @cond internal Array of modules. */ +typedef array_t(struct kr_module *) module_array_t; +/* @endcond */ + +/** + * Name resolution context. + * + * Resolution context provides basic services like cache, configuration and options. + * + * @note This structure is persistent between name resolutions and may + * be shared between threads. + */ +struct kr_context +{ + struct kr_qflags options; + knot_rrset_t *opt_rr; + map_t trust_anchors; + map_t negative_anchors; + struct kr_zonecut root_hints; + struct kr_cache cache; + kr_nsrep_rtt_lru_t *cache_rtt; + unsigned cache_rtt_tout_retry_interval; + kr_nsrep_lru_t *cache_rep; + module_array_t *modules; + /* The cookie context structure should not be held within the cookies + * module because of better access. */ + struct kr_cookie_ctx cookie_ctx; + kr_cookie_lru_t *cache_cookie; + int32_t tls_padding; /**< See net.tls_padding in ../daemon/README.rst -- -1 is "true" (default policy), 0 is "false" (no padding) */ + knot_mm_t *pool; +}; + +/* Kept outside, because kres-gen.lua can't handle this depth + * (and lines here were too long anyway). */ +struct kr_request_qsource_flags { + bool tcp:1; /**< true if the request is on TCP (or TLS); only meaningful if (dst_addr). */ + bool tls:1; /**< true if the request is on TLS; only meaningful if (dst_addr). */ +}; + +/** + * Name resolution request. + * + * Keeps information about current query processing between calls to + * processing APIs, i.e. current resolved query, resolution plan, ... + * Use this instead of the simple interface if you want to implement + * multiplexing or custom I/O. + * + * @note All data for this request must be allocated from the given pool. + */ +struct kr_request { + struct kr_context *ctx; + knot_pkt_t *answer; + struct kr_query *current_query; /**< Current evaluated query. */ + struct { + /** Address that originated the request. NULL for internal origin. */ + const struct sockaddr *addr; + /** Address that accepted the request. NULL for internal origin. */ + const struct sockaddr *dst_addr; + const knot_pkt_t *packet; + struct kr_request_qsource_flags flags; /**< See definition above. */ + size_t size; /**< query packet size */ + } qsource; + struct { + unsigned rtt; /**< Current upstream RTT */ + const struct sockaddr *addr; /**< Current upstream address */ + } upstream; /**< Upstream information, valid only in consume() phase */ + struct kr_qflags options; + int state; + ranked_rr_array_t answ_selected; + ranked_rr_array_t auth_selected; + ranked_rr_array_t add_selected; + rr_array_t additional; + bool answ_validated; /**< internal to validator; beware of caching, etc. */ + bool auth_validated; /**< see answ_validated ^^ ; TODO */ + + /** Overall rank for the request. + * + * Values from kr_rank, currently just KR_RANK_SECURE and _INITIAL. + * Only read this in finish phase and after validator, please. + * Meaning of _SECURE: all RRs in answer+authority are _SECURE, + * including any negative results implied (NXDOMAIN, NODATA). + */ + uint8_t rank; + + struct kr_rplan rplan; + trace_log_f trace_log; /**< Logging tracepoint */ + trace_callback_f trace_finish; /**< Request finish tracepoint */ + int vars_ref; /**< Reference to per-request variable table. LUA_NOREF if not set. */ + knot_mm_t pool; + unsigned int uid; /** for logging purposes only */ + void *daemon_context; /** pointer to worker from daemon. Can be used in modules. */ +}; + +/** Initializer for an array of *_selected. */ +#define kr_request_selected(req) { \ + [KNOT_ANSWER] = &(req)->answ_selected, \ + [KNOT_AUTHORITY] = &(req)->auth_selected, \ + [KNOT_ADDITIONAL] = &(req)->add_selected, \ + } + +/** + * Begin name resolution. + * + * @note Expects a request to have an initialized mempool, the "answer" packet will + * be kept during the resolution and will contain the final answer at the end. + * + * @param request request state with initialized mempool + * @param ctx resolution context + * @param answer allocated packet for final answer + * @return CONSUME (expecting query) + */ +KR_EXPORT +int kr_resolve_begin(struct kr_request *request, struct kr_context *ctx, knot_pkt_t *answer); + +/** + * Consume input packet (may be either first query or answer to query originated from kr_resolve_produce()) + * + * @note If the I/O fails, provide an empty or NULL packet, this will make iterator recognize nameserver failure. + * + * @param request request state (awaiting input) + * @param src [in] packet source address + * @param packet [in] input packet + * @return any state + */ +KR_EXPORT +int kr_resolve_consume(struct kr_request *request, const struct sockaddr *src, knot_pkt_t *packet); + +/** + * Produce either next additional query or finish. + * + * If the CONSUME is returned then dst, type and packet will be filled with + * appropriate values and caller is responsible to send them and receive answer. + * If it returns any other state, then content of the variables is undefined. + * + * @param request request state (in PRODUCE state) + * @param dst [out] possible address of the next nameserver + * @param type [out] possible used socket type (SOCK_STREAM, SOCK_DGRAM) + * @param packet [out] packet to be filled with additional query + * @return any state + */ +KR_EXPORT +int kr_resolve_produce(struct kr_request *request, struct sockaddr **dst, int *type, knot_pkt_t *packet); + +/** + * Finalises the outbound query packet with the knowledge of the IP addresses. + * + * @note The function must be called before actual sending of the request packet. + * + * @param request request state (in PRODUCE state) + * @param src address from which the query is going to be sent + * @param dst address of the name server + * @param type used socket type (SOCK_STREAM, SOCK_DGRAM) + * @param packet [in,out] query packet to be finalised + * @return kr_ok() or error code + */ +KR_EXPORT +int kr_resolve_checkout(struct kr_request *request, const struct sockaddr *src, + struct sockaddr *dst, int type, knot_pkt_t *packet); + +/** + * Finish resolution and commit results if the state is DONE. + * + * @note The structures will be deinitialized, but the assigned memory pool is not going to + * be destroyed, as it's owned by caller. + * + * @param request request state + * @param state either DONE or FAIL state + * @return DONE + */ +KR_EXPORT +int kr_resolve_finish(struct kr_request *request, int state); + +/** + * Return resolution plan. + * @param request request state + * @return pointer to rplan + */ +KR_EXPORT KR_PURE +struct kr_rplan *kr_resolve_plan(struct kr_request *request); + +/** + * Return memory pool associated with request. + * @param request request state + * @return mempool + */ +KR_EXPORT KR_PURE +knot_mm_t *kr_resolve_pool(struct kr_request *request); + diff --git a/lib/rplan.c b/lib/rplan.c new file mode 100644 index 0000000..6ed8e4e --- /dev/null +++ b/lib/rplan.c @@ -0,0 +1,308 @@ +/* Copyright (C) 2014-2017 CZ.NIC, z.s.p.o. + + This program is free software: you can redistribute it and/or modify + it under the terms of the GNU General Public License as published by + the Free Software Foundation, either version 3 of the License, or + (at your option) any later version. + + This program is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the + GNU General Public License for more details. + + You should have received a copy of the GNU General Public License + along with this program. If not, see . + */ + +#include +#include + +#include "lib/rplan.h" +#include "lib/resolve.h" +#include "lib/cache/api.h" +#include "lib/defines.h" +#include "lib/layer.h" + +#define VERBOSE_MSG(qry, ...) QRVERBOSE(qry, "plan", __VA_ARGS__) +#define QUERY_PROVIDES(q, name, cls, type) \ + ((q)->sclass == (cls) && (q)->stype == type && knot_dname_is_equal((q)->sname, name)) + +inline static unsigned char chars_or(const unsigned char a, const unsigned char b) +{ + return a | b; +} + +/** Bits set to 1 in variable b will be set to zero in variable a. */ +inline static unsigned char chars_mask(const unsigned char a, const unsigned char b) +{ + return a & ~b; +} + +/** Apply mod(a, b) to every byte a, b from fl1, fl2 and return result in fl1. */ +inline static void kr_qflags_mod(struct kr_qflags *fl1, struct kr_qflags fl2, + unsigned char mod(const unsigned char a, const unsigned char b)) +{ + if (!fl1) abort(); + union { + struct kr_qflags flags; + /* C99 section 6.5.3.4: sizeof(char) == 1 */ + unsigned char chars[sizeof(struct kr_qflags)]; + } tmp1, tmp2; + /* The compiler should be able to optimize all this into simple ORs. */ + tmp1.flags = *fl1; + tmp2.flags = fl2; + for (size_t i = 0; i < sizeof(struct kr_qflags); ++i) { + tmp1.chars[i] = mod(tmp1.chars[i], tmp2.chars[i]); + } + *fl1 = tmp1.flags; +} + +/** + * Set bits from variable fl2 in variable fl1. + * Bits which are not set in fl2 are not modified in fl1. + * + * @param[in,out] fl1 + * @param[in] fl2 + */ +void kr_qflags_set(struct kr_qflags *fl1, struct kr_qflags fl2) +{ + kr_qflags_mod(fl1, fl2, chars_or); +} + +/** + * Clear bits from variable fl2 in variable fl1. + * Bits which are not set in fl2 are not modified in fl1. + * + * @param[in,out] fl1 + * @param[in] fl2 + */ +void kr_qflags_clear(struct kr_qflags *fl1, struct kr_qflags fl2) +{ + kr_qflags_mod(fl1, fl2, chars_mask); +} + +static struct kr_query *query_create(knot_mm_t *pool, const knot_dname_t *name, uint32_t uid) +{ + struct kr_query *qry = mm_alloc(pool, sizeof(struct kr_query)); + if (qry == NULL) { + return NULL; + } + + memset(qry, 0, sizeof(struct kr_query)); + if (name != NULL) { + qry->sname = knot_dname_copy(name, pool); + if (qry->sname == NULL) { + mm_free(pool, qry); + return NULL; + } + } + + knot_dname_to_lower(qry->sname); + qry->uid = uid; + return qry; +} + +static void query_free(knot_mm_t *pool, struct kr_query *qry) +{ + kr_zonecut_deinit(&qry->zone_cut); + mm_free(pool, qry->sname); + mm_free(pool, qry); +} + +int kr_rplan_init(struct kr_rplan *rplan, struct kr_request *request, knot_mm_t *pool) +{ + if (rplan == NULL) { + return KNOT_EINVAL; + } + + memset(rplan, 0, sizeof(struct kr_rplan)); + + rplan->pool = pool; + rplan->request = request; + array_init(rplan->pending); + array_init(rplan->resolved); + rplan->next_uid = 0; + return KNOT_EOK; +} + +void kr_rplan_deinit(struct kr_rplan *rplan) +{ + if (rplan == NULL) { + return; + } + + for (size_t i = 0; i < rplan->pending.len; ++i) { + query_free(rplan->pool, rplan->pending.at[i]); + } + for (size_t i = 0; i < rplan->resolved.len; ++i) { + query_free(rplan->pool, rplan->resolved.at[i]); + } + array_clear_mm(rplan->pending, mm_free, rplan->pool); + array_clear_mm(rplan->resolved, mm_free, rplan->pool); +} + +bool kr_rplan_empty(struct kr_rplan *rplan) +{ + if (rplan == NULL) { + return true; + } + + return rplan->pending.len == 0; +} + +static struct kr_query *kr_rplan_push_query(struct kr_rplan *rplan, + struct kr_query *parent, + const knot_dname_t *name) +{ + if (rplan == NULL) { + return NULL; + } + + /* Make sure there's enough space */ + int ret = array_reserve_mm(rplan->pending, rplan->pending.len + 1, kr_memreserve, rplan->pool); + if (ret != 0) { + return NULL; + } + + struct kr_query *qry = query_create(rplan->pool, name, rplan->next_uid); + if (qry == NULL) { + return NULL; + } + rplan->next_uid += 1; + /* Class and type must be set outside this function. */ + qry->flags = rplan->request->options; + qry->parent = parent; + qry->request = rplan->request; + qry->ns.ctx = rplan->request->ctx; + qry->ns.addr[0].ip.sa_family = AF_UNSPEC; + gettimeofday(&qry->timestamp, NULL); + qry->timestamp_mono = kr_now(); + qry->creation_time_mono = parent ? parent->creation_time_mono : qry->timestamp_mono; + kr_zonecut_init(&qry->zone_cut, (const uint8_t *)"", rplan->pool); + qry->reorder = qry->flags.REORDER_RR ? kr_rand_bytes(sizeof(qry->reorder)) : 0; + + /* When forwarding, keep the nameserver addresses. */ + if (parent && parent->flags.FORWARD && qry->flags.FORWARD) { + ret = kr_nsrep_copy_set(&qry->ns, &parent->ns); + if (ret) { + query_free(rplan->pool, qry); + return NULL; + } + } + + array_push(rplan->pending, qry); + + return qry; +} + +struct kr_query *kr_rplan_push_empty(struct kr_rplan *rplan, struct kr_query *parent) +{ + if (rplan == NULL) { + return NULL; + } + + struct kr_query *qry = kr_rplan_push_query(rplan, parent, NULL); + if (qry == NULL) { + return NULL; + } + + WITH_VERBOSE(qry) { + VERBOSE_MSG(qry, "plan '%s' type '%s' uid [%05u.%02u]\n", "", "", + qry->request ? qry->request->uid : 0, qry->uid); + } + return qry; +} + +struct kr_query *kr_rplan_push(struct kr_rplan *rplan, struct kr_query *parent, + const knot_dname_t *name, uint16_t cls, uint16_t type) +{ + if (rplan == NULL || name == NULL) { + return NULL; + } + + struct kr_query *qry = kr_rplan_push_query(rplan, parent, name); + if (qry == NULL) { + return NULL; + } + + qry->sclass = cls; + qry->stype = type; + + WITH_VERBOSE(qry) { + KR_DNAME_GET_STR(name_str, name); + KR_RRTYPE_GET_STR(type_str, type); + VERBOSE_MSG(parent, "plan '%s' type '%s' uid [%05u.%02u]\n", + name_str, type_str, + qry->request ? qry->request->uid : 0, qry->uid); + } + return qry; +} + +int kr_rplan_pop(struct kr_rplan *rplan, struct kr_query *qry) +{ + if (rplan == NULL || qry == NULL) { + return KNOT_EINVAL; + } + + /* Make sure there's enough space */ + int ret = array_reserve_mm(rplan->resolved, rplan->resolved.len + 1, kr_memreserve, rplan->pool); + if (ret != 0) { + return ret; + } + + /* Find the query, it will likely be on top */ + for (size_t i = rplan->pending.len; i > 0; i--) { + if (rplan->pending.at[i - 1] == qry) { + array_del(rplan->pending, i - 1); + array_push(rplan->resolved, qry); + break; + } + } + return KNOT_EOK; +} + +bool kr_rplan_satisfies(struct kr_query *closure, const knot_dname_t *name, uint16_t cls, uint16_t type) +{ + while (name && closure) { + if (QUERY_PROVIDES(closure, name, cls, type)) { + return true; + } + closure = closure->parent; + } + return false; +} + +struct kr_query *kr_rplan_resolved(struct kr_rplan *rplan) +{ + if (rplan->resolved.len == 0) { + return NULL; + } + return array_tail(rplan->resolved); +} + +struct kr_query *kr_rplan_last(struct kr_rplan *rplan) +{ + if (!kr_rplan_empty(rplan)) { + return array_tail(rplan->pending); + } + + return kr_rplan_resolved(rplan); +} + +struct kr_query *kr_rplan_find_resolved(struct kr_rplan *rplan, struct kr_query *parent, + const knot_dname_t *name, uint16_t cls, uint16_t type) +{ + struct kr_query *ret = NULL; + for (int i = 0; i < rplan->resolved.len; ++i) { + struct kr_query *q = rplan->resolved.at[i]; + if (q->stype == type && q->sclass == cls && + (parent == NULL || q->parent == parent) && + knot_dname_is_equal(q->sname, name)) { + ret = q; + break; + } + } + return ret; +} + +#undef VERBOSE_MSG diff --git a/lib/rplan.h b/lib/rplan.h new file mode 100644 index 0000000..21b0b0b --- /dev/null +++ b/lib/rplan.h @@ -0,0 +1,221 @@ +/* Copyright (C) 2014-2017 CZ.NIC, z.s.p.o. + + This program is free software: you can redistribute it and/or modify + it under the terms of the GNU General Public License as published by + the Free Software Foundation, either version 3 of the License, or + (at your option) any later version. + + This program is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the + GNU General Public License for more details. + + You should have received a copy of the GNU General Public License + along with this program. If not, see . + */ + +#pragma once + +#include +#include +#include + +#include "lib/cache/api.h" +#include "lib/zonecut.h" +#include "lib/nsrep.h" + +/** Query flags */ +struct kr_qflags { + bool NO_MINIMIZE : 1; /**< Don't minimize QNAME. */ + bool NO_THROTTLE : 1; /**< No query/slow NS throttling. */ + bool NO_IPV6 : 1; /**< Disable IPv6 */ + bool NO_IPV4 : 1; /**< Disable IPv4 */ + bool TCP : 1; /**< Use TCP for this query. */ + bool RESOLVED : 1; /**< Query is resolved. Note that kr_query gets + * RESOLVED before following a CNAME chain; see .CNAME. */ + bool AWAIT_IPV4 : 1; /**< Query is waiting for A address. */ + bool AWAIT_IPV6 : 1; /**< Query is waiting for AAAA address. */ + bool AWAIT_CUT : 1; /**< Query is waiting for zone cut lookup */ + bool SAFEMODE : 1; /**< Don't use fancy stuff (EDNS, 0x20, ...) */ + bool CACHED : 1; /**< Query response is cached. */ + bool NO_CACHE : 1; /**< No cache for lookup; exception: finding NSs and subqueries. */ + bool EXPIRING : 1; /**< Query response is cached, but expiring. */ + bool ALLOW_LOCAL : 1; /**< Allow queries to local or private address ranges. */ + bool DNSSEC_WANT : 1; /**< Want DNSSEC secured answer; exception: +cd, + * i.e. knot_wire_set_cd(request->answer->wire). */ + bool DNSSEC_BOGUS : 1; /**< Query response is DNSSEC bogus. */ + bool DNSSEC_INSECURE : 1;/**< Query response is DNSSEC insecure. */ + bool DNSSEC_CD : 1; /**< Instruction to set CD bit in request. */ + bool STUB : 1; /**< Stub resolution, accept received answer as solved. */ + bool ALWAYS_CUT : 1; /**< Always recover zone cut (even if cached). */ + bool DNSSEC_WEXPAND : 1; /**< Query response has wildcard expansion. */ + bool PERMISSIVE : 1; /**< Permissive resolver mode. */ + bool STRICT : 1; /**< Strict resolver mode. */ + bool BADCOOKIE_AGAIN : 1;/**< Query again because bad cookie returned. */ + bool CNAME : 1; /**< Query response contains CNAME in answer section. */ + bool REORDER_RR : 1; /**< Reorder cached RRs. */ + bool TRACE : 1; /**< Also log answers if --verbose. */ + bool NO_0X20 : 1; /**< Disable query case randomization . */ + bool DNSSEC_NODS : 1; /**< DS non-existance is proven */ + bool DNSSEC_OPTOUT : 1; /**< Closest encloser proof has optout */ + bool NONAUTH : 1; /**< Non-authoritative in-bailiwick records are enough. + * TODO: utilize this also outside cache. */ + bool FORWARD : 1; /**< Forward all queries to upstream; validate answers. */ + bool DNS64_MARK : 1; /**< Internal mark for dns64 module. */ + bool CACHE_TRIED : 1; /**< Internal to cache module. */ + bool NO_NS_FOUND : 1; /**< No valid NS found during last PRODUCE stage. */ +}; + +/** Combine flags together. This means set union for simple flags. */ +KR_EXPORT +void kr_qflags_set(struct kr_qflags *fl1, struct kr_qflags fl2); + +/** Remove flags. This means set-theoretic difference. */ +KR_EXPORT +void kr_qflags_clear(struct kr_qflags *fl1, struct kr_qflags fl2); + +/** Callback for serve-stale decisions. + * @param ttl the expired TTL (i.e. it's < 0) + * @return the adjusted TTL (typically 1) or < 0. + */ +typedef int32_t (*kr_stale_cb)(int32_t ttl, const knot_dname_t *owner, uint16_t type, + const struct kr_query *qry); + +/** + * Single query representation. + */ +struct kr_query { + struct kr_query *parent; + knot_dname_t *sname; /**< The name to resolve - lower-cased, uncompressed. */ + uint16_t stype; + uint16_t sclass; + uint16_t id; + struct kr_qflags flags, forward_flags; + uint32_t secret; + uint16_t fails; + uint16_t reorder; /**< Seed to reorder (cached) RRs in answer or zero. */ + uint64_t creation_time_mono; /* The time of query's creation (milliseconds). + * Or time of creation of an oldest + * ancestor if it is a subquery. */ + uint64_t timestamp_mono; /**< Time of query created or time of + * query to upstream resolver (milliseconds). */ + struct timeval timestamp; /**< Real time for TTL+DNSSEC checks (.tv_sec only). */ + struct kr_zonecut zone_cut; + struct kr_layer_pickle *deferred; + uint32_t uid; /**< Query iteration number, unique within the kr_rplan. */ + /** Pointer to the query that originated this one because of following a CNAME (or NULL). */ + struct kr_query *cname_parent; + struct kr_request *request; /**< Parent resolution request. */ + kr_stale_cb stale_cb; /**< See the type */ + /* Beware: this must remain the last, because of lua bindings. */ + struct kr_nsrep ns; +}; + +/** @cond internal Array of queries. */ +typedef array_t(struct kr_query *) kr_qarray_t; +/* @endcond */ + +/** + * Query resolution plan structure. + * + * The structure most importantly holds the original query, answer and the + * list of pending queries required to resolve the original query. + * It also keeps a notion of current zone cut. + */ +struct kr_rplan { + kr_qarray_t pending; /**< List of pending queries. */ + kr_qarray_t resolved; /**< List of resolved queries. */ + struct kr_request *request; /**< Parent resolution request. */ + knot_mm_t *pool; /**< Temporary memory pool. */ + uint32_t next_uid; /**< Next value for kr_query::uid (incremental). */ +}; + +/** + * Initialize resolution plan (empty). + * @param rplan plan instance + * @param request resolution request + * @param pool ephemeral memory pool for whole resolution + */ +KR_EXPORT +int kr_rplan_init(struct kr_rplan *rplan, struct kr_request *request, knot_mm_t *pool); + +/** + * Deinitialize resolution plan, aborting any uncommited transactions. + * @param rplan plan instance + */ +KR_EXPORT +void kr_rplan_deinit(struct kr_rplan *rplan); + +/** + * Return true if the resolution plan is empty (i.e. finished or initialized) + * @param rplan plan instance + * @return true or false + */ +KR_EXPORT KR_PURE +bool kr_rplan_empty(struct kr_rplan *rplan); + +/** + * Push empty query to the top of the resolution plan. + * @note This query serves as a cookie query only. + * @param rplan plan instance + * @param parent query parent (or NULL) + * @return query instance or NULL + */ +KR_EXPORT +struct kr_query *kr_rplan_push_empty(struct kr_rplan *rplan, + struct kr_query *parent); + +/** + * Push a query to the top of the resolution plan. + * @note This means that this query takes precedence before all pending queries. + * @param rplan plan instance + * @param parent query parent (or NULL) + * @param name resolved name + * @param cls resolved class + * @param type resolved type + * @return query instance or NULL + */ +KR_EXPORT +struct kr_query *kr_rplan_push(struct kr_rplan *rplan, struct kr_query *parent, + const knot_dname_t *name, uint16_t cls, uint16_t type); + +/** + * Pop existing query from the resolution plan. + * @note Popped queries are not discarded, but moved to the resolved list. + * @param rplan plan instance + * @param qry resolved query + * @return 0 or an error + */ +KR_EXPORT +int kr_rplan_pop(struct kr_rplan *rplan, struct kr_query *qry); + +/** + * Return true if resolution chain satisfies given query. + */ +KR_EXPORT KR_PURE +bool kr_rplan_satisfies(struct kr_query *closure, const knot_dname_t *name, uint16_t cls, uint16_t type); + +/** Return last resolved query. */ +KR_EXPORT KR_PURE +struct kr_query *kr_rplan_resolved(struct kr_rplan *rplan); + +/** + * Return last query (either currently being solved or last resolved). + * This is necessary to retrieve the last query in case of resolution failures (e.g. time limit reached). + */ +KR_EXPORT KR_PURE +struct kr_query *kr_rplan_last(struct kr_rplan *rplan); + + +/** + * Check if a given query already resolved. + * @param rplan plan instance + * @param parent query parent (or NULL) + * @param name resolved name + * @param cls resolved class + * @param type resolved type + * @return query instance or NULL + */ +KR_EXPORT KR_PURE +struct kr_query *kr_rplan_find_resolved(struct kr_rplan *rplan, struct kr_query *parent, + const knot_dname_t *name, uint16_t cls, uint16_t type); diff --git a/lib/utils.c b/lib/utils.c new file mode 100644 index 0000000..fe9ab03 --- /dev/null +++ b/lib/utils.c @@ -0,0 +1,1084 @@ +/* Copyright (C) 2014-2017 CZ.NIC, z.s.p.o. + + This program is free software: you can redistribute it and/or modify + it under the terms of the GNU General Public License as published by + the Free Software Foundation, either version 3 of the License, or + (at your option) any later version. + + This program is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the + GNU General Public License for more details. + + You should have received a copy of the GNU General Public License + along with this program. If not, see . + */ + +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include + +#include "lib/defines.h" +#include "lib/utils.h" +#include "lib/generic/array.h" +#include "lib/nsrep.h" +#include "lib/module.h" +#include "lib/resolve.h" + + +/* Always compile-in log symbols, even if disabled. */ +#undef kr_verbose_status +#undef kr_verbose_set +#undef kr_log_verbose + +/* Logging & debugging */ +bool kr_verbose_status = false; + +void *mm_realloc(knot_mm_t *mm, void *what, size_t size, size_t prev_size) +{ + if (mm) { + void *p = mm->alloc(mm->ctx, size); + if (p == NULL) { + return NULL; + } else { + if (what) { + memcpy(p, what, + prev_size < size ? prev_size : size); + } + mm_free(mm, what); + return p; + } + } else { + return realloc(what, size); + } +} + +void *mm_malloc(void *ctx, size_t n) +{ + (void)ctx; + return malloc(n); +} + +/* + * Macros. + */ +#define strlen_safe(x) ((x) ? strlen(x) : 0) + +/** + * @internal Convert 16bit unsigned to string, keeps leading spaces. + * @note Always fills dst length = 5 + * Credit: http://computer-programming-forum.com/46-asm/7aa4b50bce8dd985.htm + */ +static inline int u16tostr(uint8_t *dst, uint16_t num) +{ + uint32_t tmp = num * (((1 << 28) / 10000) + 1) - (num / 4); + for(size_t i = 0; i < 5; i++) { + dst[i] = '0' + (char) (tmp >> 28); + tmp = (tmp & 0x0fffffff) * 10; + } + return 5; +} + +/* + * Cleanup callbacks. + */ + +static void kres_gnutls_log(int level, const char *message) +{ + kr_log_verbose("[gnutls] (%d) %s", level, message); +} + +bool kr_verbose_set(bool status) +{ +#ifndef NOVERBOSELOG + kr_verbose_status = status; + + /* gnutls logs messages related to our TLS and also libdnssec, + * and the logging is set up in a global way only */ + if (status) { + gnutls_global_set_log_function(kres_gnutls_log); + } + gnutls_global_set_log_level(status ? 5 : 0); +#endif + return kr_verbose_status; +} + +void kr_log_verbose(const char *fmt, ...) +{ + if (kr_verbose_status) { + va_list args; + va_start(args, fmt); + vprintf(fmt, args); + va_end(args); + fflush(stdout); + } +} + +void kr_log_qverbose_impl(const struct kr_query *qry, const char *cls, const char *fmt, ...) +{ + unsigned ind = 0; + for (const struct kr_query *q = qry; q; q = q->parent) + ind += 2; + uint32_t qry_uid = qry ? qry->uid : 0; + uint32_t req_uid = qry && qry->request ? qry->request->uid : 0; + /* Simplified kr_log_verbose() calls, first prefix then passed fmt... + * Calling it would take about the same amount of code. */ + printf("[%05u.%02u][%s] %*s", req_uid, qry_uid, cls, ind, ""); + va_list args; + va_start(args, fmt); + vprintf(fmt, args); + va_end(args); + fflush(stdout); +} + +bool kr_log_trace(const struct kr_query *query, const char *source, const char *fmt, ...) +{ + if (!kr_log_trace_enabled(query)) { + return false; + } + + auto_free char *msg = NULL; + + va_list args; + va_start(args, fmt); + int len = vasprintf(&msg, fmt, args); + va_end(args); + + /* Check formatting result before logging */ + if (len < 0) { + return false; + } + + query->request->trace_log(query, source, msg); + return true; +} + +char* kr_strcatdup(unsigned n, ...) +{ + if (n < 1) { + return NULL; + } + + /* Calculate total length */ + size_t total_len = 0; + va_list vl; + va_start(vl, n); + for (unsigned i = 0; i < n; ++i) { + char *item = va_arg(vl, char *); + const size_t new_len = total_len + strlen_safe(item); + if (unlikely(new_len < total_len)) { + va_end(vl); + return NULL; + } + total_len = new_len; + } + va_end(vl); + + /* Allocate result and fill */ + char *result = NULL; + if (total_len > 0) { + if (unlikely(total_len + 1 == 0)) return NULL; + result = malloc(total_len + 1); + } + if (result) { + char *stream = result; + va_start(vl, n); + for (unsigned i = 0; i < n; ++i) { + char *item = va_arg(vl, char *); + if (item) { + size_t len = strlen(item); + memcpy(stream, item, len + 1); + stream += len; + } + } + va_end(vl); + } + + return result; +} + +int kr_memreserve(void *baton, char **mem, size_t elm_size, size_t want, size_t *have) +{ + if (*have >= want) { + return 0; + } else { + knot_mm_t *pool = baton; + size_t next_size = array_next_count(want); + void *mem_new = mm_alloc(pool, next_size * elm_size); + if (mem_new != NULL) { + memcpy(mem_new, *mem, (*have)*(elm_size)); + mm_free(pool, *mem); + *mem = mem_new; + *have = next_size; + return 0; + } + } + return -1; +} + +static int pkt_recycle(knot_pkt_t *pkt, bool keep_question) +{ + /* The maximum size of a header + query name + (class, type) */ + uint8_t buf[KNOT_WIRE_HEADER_SIZE + KNOT_DNAME_MAXLEN + 2 * sizeof(uint16_t)]; + + /* Save header and the question section */ + size_t base_size = KNOT_WIRE_HEADER_SIZE; + if (keep_question) { + base_size += knot_pkt_question_size(pkt); + } + assert(base_size <= sizeof(buf)); + memcpy(buf, pkt->wire, base_size); + + /* Clear the packet and its auxiliary structures */ + knot_pkt_clear(pkt); + + /* Restore header and question section and clear counters */ + pkt->size = base_size; + memcpy(pkt->wire, buf, base_size); + knot_wire_set_qdcount(pkt->wire, keep_question); + knot_wire_set_ancount(pkt->wire, 0); + knot_wire_set_nscount(pkt->wire, 0); + knot_wire_set_arcount(pkt->wire, 0); + + /* Reparse question */ + knot_pkt_begin(pkt, KNOT_ANSWER); + return knot_pkt_parse_question(pkt); +} + +int kr_pkt_recycle(knot_pkt_t *pkt) +{ + return pkt_recycle(pkt, false); +} + +int kr_pkt_clear_payload(knot_pkt_t *pkt) +{ + return pkt_recycle(pkt, knot_wire_get_qdcount(pkt->wire)); +} + +int kr_pkt_put(knot_pkt_t *pkt, const knot_dname_t *name, uint32_t ttl, + uint16_t rclass, uint16_t rtype, const uint8_t *rdata, uint16_t rdlen) +{ + /* LATER(opt.): there's relatively lots of copying, but ATM kr_pkt_put() + * isn't considered to be used in any performance-critical parts (just lua). */ + if (!pkt || !name) { + return kr_error(EINVAL); + } + /* Create empty RR */ + knot_rrset_t rr; + knot_rrset_init(&rr, knot_dname_copy(name, &pkt->mm), rtype, rclass, ttl); + /* Create RDATA */ + knot_rdata_t *rdata_tmp = mm_alloc(&pkt->mm, offsetof(knot_rdata_t, data) + rdlen); + knot_rdata_init(rdata_tmp, rdlen, rdata); + knot_rdataset_add(&rr.rrs, rdata_tmp, &pkt->mm); + mm_free(&pkt->mm, rdata_tmp); /* we're always on mempool for now, but whatever */ + /* Append RR */ + return knot_pkt_put(pkt, 0, &rr, KNOT_PF_FREE); +} + +void kr_pkt_make_auth_header(knot_pkt_t *pkt) +{ + assert(pkt && pkt->wire); + knot_wire_clear_ad(pkt->wire); + knot_wire_set_aa(pkt->wire); +} + +const char *kr_inaddr(const struct sockaddr *addr) +{ + if (!addr) { + return NULL; + } + switch (addr->sa_family) { + case AF_INET: return (const char *)&(((const struct sockaddr_in *)addr)->sin_addr); + case AF_INET6: return (const char *)&(((const struct sockaddr_in6 *)addr)->sin6_addr); + default: return NULL; + } +} + +int kr_inaddr_family(const struct sockaddr *addr) +{ + if (!addr) + return AF_UNSPEC; + return addr->sa_family; +} + +int kr_inaddr_len(const struct sockaddr *addr) +{ + if (!addr) { + return kr_error(EINVAL); + } + return kr_family_len(addr->sa_family); +} + +int kr_sockaddr_len(const struct sockaddr *addr) +{ + if (!addr) { + return kr_error(EINVAL); + } + switch (addr->sa_family) { + case AF_INET: return sizeof(struct sockaddr_in); + case AF_INET6: return sizeof(struct sockaddr_in6); + default: return kr_error(EINVAL); + } +} + +int kr_sockaddr_cmp(const struct sockaddr *left, const struct sockaddr *right) +{ + if (!left || !right) { + return kr_error(EINVAL); + } + if (left->sa_family != right->sa_family) { + return kr_error(EFAULT); + } + if (left->sa_family == AF_INET) { + struct sockaddr_in *left_in = (struct sockaddr_in *)left; + struct sockaddr_in *right_in = (struct sockaddr_in *)right; + if (left_in->sin_addr.s_addr != right_in->sin_addr.s_addr) { + return kr_error(EFAULT); + } + if (left_in->sin_port != right_in->sin_port) { + return kr_error(EFAULT); + } + } else if (left->sa_family == AF_INET6) { + struct sockaddr_in6 *left_in6 = (struct sockaddr_in6 *)left; + struct sockaddr_in6 *right_in6 = (struct sockaddr_in6 *)right; + if (memcmp(&left_in6->sin6_addr, &right_in6->sin6_addr, + sizeof(struct in6_addr)) != 0) { + return kr_error(EFAULT); + } + if (left_in6->sin6_port != right_in6->sin6_port) { + return kr_error(EFAULT); + } + } else { + return kr_error(ENOENT); + } + return kr_ok(); +} + +uint16_t kr_inaddr_port(const struct sockaddr *addr) +{ + if (!addr) { + return 0; + } + switch (addr->sa_family) { + case AF_INET: return ntohs(((const struct sockaddr_in *)addr)->sin_port); + case AF_INET6: return ntohs(((const struct sockaddr_in6 *)addr)->sin6_port); + default: return 0; + } +} + +void kr_inaddr_set_port(struct sockaddr *addr, uint16_t port) +{ + if (!addr) { + return; + } + switch (addr->sa_family) { + case AF_INET: ((struct sockaddr_in *)addr)->sin_port = htons(port); + case AF_INET6: ((struct sockaddr_in6 *)addr)->sin6_port = htons(port); + default: break; + } +} + +int kr_inaddr_str(const struct sockaddr *addr, char *buf, size_t *buflen) +{ + if (!addr || !buf || !buflen) { + return kr_error(EINVAL); + } + + if (!inet_ntop(addr->sa_family, kr_inaddr(addr), buf, *buflen)) { + return kr_error(errno); + } + const int len = strlen(buf); + const int len_need = len + 1 + 5 + 1; + if (len_need > *buflen) { + *buflen = len_need; + return kr_error(ENOSPC); + } + *buflen = len_need; + buf[len] = '#'; + u16tostr((uint8_t *)&buf[len + 1], kr_inaddr_port(addr)); + buf[len_need - 1] = 0; + return kr_ok(); +} + +int kr_straddr_family(const char *addr) +{ + if (!addr) { + return kr_error(EINVAL); + } + if (strchr(addr, ':')) { + return AF_INET6; + } + return AF_INET; +} + +int kr_family_len(int family) +{ + switch (family) { + case AF_INET: return sizeof(struct in_addr); + case AF_INET6: return sizeof(struct in6_addr); + default: return kr_error(EINVAL); + } +} + +struct sockaddr * kr_straddr_socket(const char *addr, int port) +{ + switch (kr_straddr_family(addr)) { + case AF_INET: { + struct sockaddr_in *res = malloc(sizeof(*res)); + if (uv_ip4_addr(addr, port, res) >= 0) { + return (struct sockaddr *)res; + } else { + free(res); + return NULL; + } + } + case AF_INET6: { + struct sockaddr_in6 *res = malloc(sizeof(*res)); + if (uv_ip6_addr(addr, port, res) >= 0) { + return (struct sockaddr *)res; + } else { + free(res); + return NULL; + } + } + default: + return NULL; + } +} + +int kr_straddr_subnet(void *dst, const char *addr) +{ + if (!dst || !addr) { + return kr_error(EINVAL); + } + /* Parse subnet */ + int bit_len = 0; + int family = kr_straddr_family(addr); + auto_free char *addr_str = strdup(addr); + char *subnet = strchr(addr_str, '/'); + if (subnet) { + *subnet = '\0'; + subnet += 1; + bit_len = strtol(subnet, NULL, 10); + /* Check client subnet length */ + const int max_len = (family == AF_INET6) ? 128 : 32; + if (bit_len < 0 || bit_len > max_len) { + return kr_error(ERANGE); + } + } else { + /* No subnet, use maximal subnet length. */ + bit_len = (family == AF_INET6) ? 128 : 32; + } + /* Parse address */ + int ret = inet_pton(family, addr_str, dst); + if (ret < 0) { + return kr_error(EILSEQ); + } + + return bit_len; +} + +int kr_straddr_split(const char *addr, char *buf, size_t buflen, uint16_t *port) +{ + const int base = 10; + long p = 0; + size_t addrlen = strlen(addr); + char *p_start = strchr(addr, '@'); + char *p_end; + + if (!p_start) { + p_start = strchr(addr, '#'); + } + + if (p_start) { + if (p_start[1] != '\0'){ + p = strtol(p_start + 1, &p_end, base); + if (*p_end != '\0' || p <= 0 || p > UINT16_MAX) { + return kr_error(EINVAL); + } + } + addrlen = p_start - addr; + } + + /* Check if address is valid. */ + if (addrlen >= INET6_ADDRSTRLEN) { + return kr_error(EINVAL); + } + + char str[INET6_ADDRSTRLEN]; + struct sockaddr_storage ss; + + memcpy(str, addr, addrlen); str[addrlen] = '\0'; + + int family = kr_straddr_family(str); + if (family == kr_error(EINVAL) || !inet_pton(family, str, &ss)) { + return kr_error(EINVAL); + } + + /* Address and port contains valid values, return it to caller */ + if (buf) { + if (addrlen >= buflen) { + return kr_error(ENOSPC); + } + memcpy(buf, addr, addrlen); buf[addrlen] = '\0'; + } + if (port) { + *port = (uint16_t)p; + } + + return kr_ok(); +} + +int kr_straddr_join(const char *addr, uint16_t port, char *buf, size_t *buflen) +{ + if (!addr || !buf || !buflen) { + return kr_error(EINVAL); + } + + struct sockaddr_storage ss; + int family = kr_straddr_family(addr); + if (family == kr_error(EINVAL) || !inet_pton(family, addr, &ss)) { + return kr_error(EINVAL); + } + + int len = strlen(addr); + if (len + 6 >= *buflen) { + return kr_error(ENOSPC); + } + + memcpy(buf, addr, len + 1); + buf[len] = '#'; + u16tostr((uint8_t *)&buf[len + 1], port); + len += 6; + buf[len] = 0; + *buflen = len; + + return kr_ok(); +} + +int kr_bitcmp(const char *a, const char *b, int bits) +{ + /* We're using the function from lua directly, so at least for now + * we avoid crashing on bogus inputs. Meaning: NULL is ordered before + * anything else, and negative length is the same as zero. + * TODO: review the call sites and probably remove the checks. */ + if (bits <= 0 || (!a && !b)) { + return 0; + } else if (!a) { + return -1; + } else if (!b) { + return 1; + } + + assert((a && b && bits >= 0) || bits == 0); + /* Compare part byte-divisible part. */ + const size_t chunk = bits / 8; + int ret = memcmp(a, b, chunk); + if (ret != 0) { + return ret; + } + a += chunk; + b += chunk; + bits -= chunk * 8; + /* Compare last partial byte address block. */ + if (bits > 0) { + const size_t shift = (8 - bits); + ret = ((uint8_t)(*a >> shift) - (uint8_t)(*b >> shift)); + } + return ret; +} + +int kr_rrkey(char *key, uint16_t class, const knot_dname_t *owner, + uint16_t type, uint16_t additional) +{ + if (!key || !owner) { + return kr_error(EINVAL); + } + uint8_t *key_buf = (uint8_t *)key; + int ret = u16tostr(key_buf, class); + if (ret <= 0) { + return ret; + } + key_buf += ret; + ret = knot_dname_to_wire(key_buf, owner, KNOT_DNAME_MAXLEN); + if (ret <= 0) { + return ret; + } + knot_dname_to_lower(key_buf); + key_buf += ret - 1; + ret = u16tostr(key_buf, type); + if (ret <= 0) { + return ret; + } + key_buf += ret; + ret = u16tostr(key_buf, additional); + if (ret <= 0) { + return ret; + } + key_buf[ret] = '\0'; + return (char *)&key_buf[ret] - key; +} + +/** Return whether two RRsets match, i.e. would form the same set; see ranked_rr_array_t */ +static inline bool rrsets_match(const knot_rrset_t *rr1, const knot_rrset_t *rr2) +{ + bool match = rr1->type == rr2->type && rr1->rclass == rr2->rclass; + if (match && rr2->type == KNOT_RRTYPE_RRSIG) { + match = match && knot_rrsig_type_covered(rr1->rrs.rdata) + == knot_rrsig_type_covered(rr2->rrs.rdata); + } + match = match && knot_dname_is_equal(rr1->owner, rr2->owner); + return match; +} + +/** Ensure that an index in a ranked array won't cause "duplicate" RRsets on wire. + * + * Other entries that would form the same RRset get to_wire = false. + * See also rrsets_match. + */ +static int to_wire_ensure_unique(ranked_rr_array_t *array, size_t index) +{ + bool ok = array && index < array->len; + if (!ok) { + assert(false); + return kr_error(EINVAL); + } + + const struct ranked_rr_array_entry *e0 = array->at[index]; + if (!e0->to_wire) { + return kr_ok(); + } + + for (ssize_t i = array->len - 1; i >= 0; --i) { + /* ^ iterate backwards, as the end is more likely in CPU caches */ + struct ranked_rr_array_entry *ei = array->at[i]; + if (ei->qry_uid == e0->qry_uid /* assumption: no duplicates within qry */ + || !ei->to_wire /* no use for complex comparison if @to_wire */ + ) { + continue; + } + if (rrsets_match(ei->rr, e0->rr)) { + ei->to_wire = false; + } + } + return kr_ok(); +} + +int kr_ranked_rrarray_add(ranked_rr_array_t *array, const knot_rrset_t *rr, + uint8_t rank, bool to_wire, uint32_t qry_uid, knot_mm_t *pool) +{ + /* rr always has one record per rrset + * check if another rrset with the same + * rclass/type/owner combination exists within current query + * and merge if needed */ + for (ssize_t i = array->len - 1; i >= 0; --i) { + ranked_rr_array_entry_t *stashed = array->at[i]; + if (stashed->yielded) { + break; + } + if (stashed->qry_uid != qry_uid) { + break; + } + if (!rrsets_match(stashed->rr, rr)) { + continue; + } + /* Found the entry to merge with. Check consistency and merge. */ + bool ok = stashed->rank == rank && !stashed->cached; + if (!ok) { + assert(false); + return kr_error(EEXIST); + } + /* It may happen that an RRset is first considered useful + * (to_wire = false, e.g. due to being part of glue), + * and later we may find we also want it in the answer. */ + stashed->to_wire = stashed->to_wire || to_wire; + + return knot_rdataset_merge(&stashed->rr->rrs, &rr->rrs, pool); + } + + /* No stashed rrset found, add */ + int ret = array_reserve_mm(*array, array->len + 1, kr_memreserve, pool); + if (ret != 0) { + return kr_error(ENOMEM); + } + + ranked_rr_array_entry_t *entry = mm_alloc(pool, sizeof(ranked_rr_array_entry_t)); + if (!entry) { + return kr_error(ENOMEM); + } + knot_rrset_t *copy = knot_rrset_copy(rr, pool); + if (!copy) { + mm_free(pool, entry); + return kr_error(ENOMEM); + } + + entry->qry_uid = qry_uid; + entry->rr = copy; + entry->rank = rank; + entry->revalidation_cnt = 0; + entry->cached = false; + entry->yielded = false; + entry->to_wire = to_wire; + if (array_push(*array, entry) < 0) { + /* Silence coverity. It shouldn't be possible to happen, + * due to the array_reserve_mm call above. */ + mm_free(pool, entry); + return kr_error(ENOMEM); + } + + return to_wire_ensure_unique(array, array->len - 1); +} + +int kr_ranked_rrarray_set_wire(ranked_rr_array_t *array, bool to_wire, + uint32_t qry_uid, bool check_dups, + bool (*extraCheck)(const ranked_rr_array_entry_t *)) +{ + for (size_t i = 0; i < array->len; ++i) { + ranked_rr_array_entry_t *entry = array->at[i]; + if (entry->qry_uid != qry_uid) { + continue; + } + if (extraCheck != NULL && !extraCheck(entry)) { + continue; + } + entry->to_wire = to_wire; + if (check_dups) { + int ret = to_wire_ensure_unique(array, i); + if (ret) return ret; + } + } + return kr_ok(); +} + + +static char *callprop(struct kr_module *module, const char *prop, const char *input, void *env) +{ + if (!module || !module->props || !prop) { + return NULL; + } + for (const struct kr_prop *p = module->props(); p && p->name; ++p) { + if (p->cb != NULL && strcmp(p->name, prop) == 0) { + return p->cb(env, module, input); + } + } + return NULL; +} + +char *kr_module_call(struct kr_context *ctx, const char *module, const char *prop, const char *input) +{ + if (!ctx || !ctx->modules || !module || !prop) { + return NULL; + } + module_array_t *mod_list = ctx->modules; + for (size_t i = 0; i < mod_list->len; ++i) { + struct kr_module *mod = mod_list->at[i]; + if (strcmp(mod->name, module) == 0) { + return callprop(mod, prop, input, ctx); + } + } + return NULL; +} + +static void flags_to_str(char *dst, const knot_pkt_t *pkt, size_t maxlen) +{ + int offset = 0; + int ret = 0; + struct { + uint8_t (*get) (const uint8_t *packet); + char name[3]; + } flag[7] = { + {knot_wire_get_qr, "qr"}, + {knot_wire_get_aa, "aa"}, + {knot_wire_get_rd, "rd"}, + {knot_wire_get_ra, "ra"}, + {knot_wire_get_tc, "tc"}, + {knot_wire_get_ad, "ad"}, + {knot_wire_get_cd, "cd"} + }; + for (int i = 0; i < 7; ++i) { + if (!flag[i].get(pkt->wire)) { + continue; + } + ret = snprintf(dst + offset, maxlen, "%s ", flag[i].name); + if (ret <= 0 || ret >= maxlen) { + dst[0] = 0; + return; + } + offset += ret; + maxlen -= ret; + } + dst[offset] = 0; +} + +static char *print_section_opt(struct mempool *mp, char *endp, const knot_rrset_t *rr, const uint8_t rcode) +{ + uint8_t ercode = knot_edns_get_ext_rcode(rr); + uint16_t ext_rcode_id = knot_edns_whole_rcode(ercode, rcode); + const char *ext_rcode_str = "Unused"; + const knot_lookup_t *ext_rcode; + + if (ercode > 0) { + ext_rcode = knot_lookup_by_id(knot_rcode_names, ext_rcode_id); + if (ext_rcode != NULL) { + ext_rcode_str = ext_rcode->name; + } else { + ext_rcode_str = "Unknown"; + } + } + + return mp_printf_append(mp, endp, + ";; EDNS PSEUDOSECTION:\n;; " + "Version: %u; flags: %s; UDP size: %u B; ext-rcode: %s\n\n", + knot_edns_get_version(rr), + (knot_edns_do(rr) != 0) ? "do" : "", + knot_edns_get_payload(rr), + ext_rcode_str); + +} + +char *kr_pkt_text(const knot_pkt_t *pkt) +{ + if (!pkt) { + return NULL; + } + + struct mempool *mp = mp_new(512); + + static const char * snames[] = { + ";; ANSWER SECTION", ";; AUTHORITY SECTION", ";; ADDITIONAL SECTION" + }; + char flags[32]; + uint8_t pkt_rcode = knot_wire_get_rcode(pkt->wire); + uint8_t pkt_opcode = knot_wire_get_opcode(pkt->wire); + const char *rcode_str = "Unknown"; + const char *opcode_str = "Unknown"; + const knot_lookup_t *rcode = knot_lookup_by_id(knot_rcode_names, pkt_rcode); + const knot_lookup_t *opcode = knot_lookup_by_id(knot_opcode_names, pkt_opcode); + uint16_t qry_id = knot_wire_get_id(pkt->wire); + uint16_t qdcount = knot_wire_get_qdcount(pkt->wire); + + if (rcode != NULL) { + rcode_str = rcode->name; + } + if (opcode != NULL) { + opcode_str = opcode->name; + } + flags_to_str(flags, pkt, sizeof(flags)); + + char *ptr = mp_printf(mp, + ";; ->>HEADER<<- opcode: %s; status: %s; id: %hu\n" + ";; Flags: %s QUERY: %hu; ANSWER: %hu; " + "AUTHORITY: %hu; ADDITIONAL: %hu\n\n", + opcode_str, rcode_str, qry_id, + flags, + qdcount, + knot_wire_get_ancount(pkt->wire), + knot_wire_get_nscount(pkt->wire), + knot_wire_get_arcount(pkt->wire)); + + if (knot_pkt_has_edns(pkt)) { + ptr = print_section_opt(mp, ptr, pkt->opt_rr, knot_wire_get_rcode(pkt->wire)); + } + + if (qdcount == 1) { + KR_DNAME_GET_STR(qname, knot_pkt_qname(pkt)); + KR_RRTYPE_GET_STR(rrtype, knot_pkt_qtype(pkt)); + ptr = mp_printf_append(mp, ptr, ";; QUESTION SECTION\n%s\t\t%s\n", qname, rrtype); + } else if (qdcount > 1) { + ptr = mp_printf_append(mp, ptr, ";; Warning: unsupported QDCOUNT %hu\n", qdcount); + } + + for (knot_section_t i = KNOT_ANSWER; i <= KNOT_ADDITIONAL; ++i) { + const knot_pktsection_t *sec = knot_pkt_section(pkt, i); + if (sec->count == 0 || knot_pkt_rr(sec, 0)->type == KNOT_RRTYPE_OPT) { + /* OPT RRs are _supposed_ to be the last ^^, if they appear */ + continue; + } + + ptr = mp_printf_append(mp, ptr, "\n%s\n", snames[i - KNOT_ANSWER]); + for (unsigned k = 0; k < sec->count; ++k) { + const knot_rrset_t *rr = knot_pkt_rr(sec, k); + if (rr->type == KNOT_RRTYPE_OPT) { + continue; + } + auto_free char *rr_text = kr_rrset_text(rr); + ptr = mp_printf_append(mp, ptr, "%s", rr_text); + } + } + + /* Close growing buffer and duplicate result before deleting */ + char *result = strdup(ptr); + mp_delete(mp); + return result; +} + +char *kr_rrset_text(const knot_rrset_t *rr) +{ + if (!rr) { + return NULL; + } + + /* Note: knot_rrset_txt_dump will double the size until the rrset fits */ + size_t bufsize = 128; + char *buf = malloc(bufsize); + int ret = knot_rrset_txt_dump(rr, &buf, &bufsize, &KNOT_DUMP_STYLE_DEFAULT); + if (ret < 0) { + free(buf); + return NULL; + } + + return buf; +} + +uint64_t kr_now() +{ + return uv_now(uv_default_loop()); +} + +const char *kr_strptime_diff(const char *format, const char *time1_str, + const char *time0_str, double *diff) { + assert(format != NULL); + assert(time1_str != NULL); + assert(time0_str != NULL); + assert(diff != NULL); + + struct tm time1_tm; + time_t time1_u; + struct tm time0_tm; + time_t time0_u; + + char *err = strptime(time1_str, format, &time1_tm); + if (err == NULL || err != time1_str + strlen(time1_str)) + return "strptime failed for time1"; + time1_tm.tm_isdst = -1; /* determine if DST is active or not */ + time1_u = mktime(&time1_tm); + if (time1_u == (time_t)-1) + return "mktime failed for time1"; + + err = strptime(time0_str, format, &time0_tm); + if (err == NULL || err != time0_str + strlen(time0_str)) + return "strptime failed for time0"; + time0_tm.tm_isdst = -1; /* determine if DST is active or not */ + time0_u = mktime(&time0_tm); + if (time0_u == (time_t)-1) + return "mktime failed for time0"; + *diff = difftime(time1_u, time0_u); + + return NULL; +} + +int knot_dname_lf2wire(knot_dname_t * const dst, uint8_t len, const uint8_t *lf) +{ + knot_dname_t *d = dst; /* moving "cursor" as we write it out */ + bool ok = d && (len == 0 || lf); + if (!ok) { + assert(false); + return kr_error(EINVAL); + } + /* we allow the final zero byte to be omitted */ + if (!len) { + goto finish; + } + if (lf[len - 1]) { + ++len; + } + /* convert the name, one label at a time */ + int label_end = len - 1; /* index of the zero byte after the current label */ + while (label_end >= 0) { + /* find label_start */ + int i = label_end - 1; + while (i >= 0 && lf[i]) + --i; + const int label_start = i + 1; /* index of the first byte of the current label */ + const int label_len = label_end - label_start; + assert(label_len >= 0); + if (label_len > 63 || label_len <= 0) + return kr_error(EILSEQ); + /* write the label */ + *d = label_len; + ++d; + memcpy(d, lf + label_start, label_len); + d += label_len; + /* next label */ + label_end = label_start - 1; + } +finish: + *d = 0; /* the final zero */ + ++d; + return d - dst; +} + +static void rnd_noerror(void *data, uint size) +{ + int ret = gnutls_rnd(GNUTLS_RND_NONCE, data, size); + if (ret) { + kr_log_error("gnutls_rnd(): %s\n", gnutls_strerror(ret)); + abort(); + } +} +void kr_rnd_buffered(void *data, uint size) +{ + /* static circular buffer, from index _begin (inclusive) to _end (exclusive) */ + static uint8_t buf[512/8]; /* gnutls_rnd() works on blocks of 512 bits (chacha) */ + static uint buf_begin = sizeof(buf); + + if (unlikely(size > sizeof(buf))) { + rnd_noerror(data, size); + return; + } + /* Start with contiguous chunk, possibly until the end of buffer. */ + const uint size1 = MIN(size, sizeof(buf) - buf_begin); + uint8_t *d = data; + memcpy(d, buf + buf_begin, size1); + if (size1 == size) { + buf_begin += size1; + return; + } + d += size1; + size -= size1; + /* Refill the whole buffer, and finish by another contiguous chunk. */ + rnd_noerror(buf, sizeof(buf)); + memcpy(d, buf, size); + buf_begin = size; +} + +void kr_rrset_init(knot_rrset_t *rrset, knot_dname_t *owner, + uint16_t type, uint16_t rclass, uint32_t ttl) +{ + assert(rrset); + knot_rrset_init(rrset, owner, type, rclass, ttl); +} +uint16_t kr_pkt_qclass(const knot_pkt_t *pkt) +{ + return knot_pkt_qclass(pkt); +} +uint16_t kr_pkt_qtype(const knot_pkt_t *pkt) +{ + return knot_pkt_qtype(pkt); +} +uint32_t kr_rrsig_sig_inception(const knot_rdata_t *rdata) +{ + return knot_rrsig_sig_inception(rdata); +} +uint32_t kr_rrsig_sig_expiration(const knot_rdata_t *rdata) +{ + return knot_rrsig_sig_expiration(rdata); +} +uint16_t kr_rrsig_type_covered(const knot_rdata_t *rdata) +{ + return knot_rrsig_type_covered(rdata); +} + diff --git a/lib/utils.h b/lib/utils.h new file mode 100644 index 0000000..21eabac --- /dev/null +++ b/lib/utils.h @@ -0,0 +1,495 @@ +/* Copyright (C) 2014-2017 CZ.NIC, z.s.p.o. + + This program is free software: you can redistribute it and/or modify + it under the terms of the GNU General Public License as published by + the Free Software Foundation, either version 3 of the License, or + (at your option) any later version. + + This program is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the + GNU General Public License for more details. + + You should have received a copy of the GNU General Public License + along with this program. If not, see . + */ + +#pragma once + +#include +#include +#include +#include +#include +#include + +#include +#include +#include + +#include +#include +#include +#include + +#include "lib/generic/array.h" +#include "lib/defines.h" + +struct kr_query; +struct kr_request; + +/* + * Logging and debugging. + */ + +/** @brief Callback for request events. */ +typedef void (*trace_callback_f)(struct kr_request *request); +/** @brief Callback for request logging handler. */ +typedef void (*trace_log_f)(const struct kr_query *query, const char *source, const char *msg); + +#define kr_log_info(...) do { printf(__VA_ARGS__); fflush(stdout); } while(0) +#define kr_log_error(...) fprintf(stderr, ## __VA_ARGS__) + +/* Always export these, but override direct calls by macros conditionally. */ +/** Whether in --verbose mode. Only use this for reading. */ +KR_EXPORT extern bool kr_verbose_status; + +/** Set --verbose mode. Not available if compiled with -DNOVERBOSELOG. */ +KR_EXPORT bool kr_verbose_set(bool status); + +/** Log a message if in --verbose mode. */ +KR_EXPORT KR_PRINTF(1) +void kr_log_verbose(const char *fmt, ...); + +/** Utility for QRVERBOSE - use that instead. */ +KR_EXPORT KR_PRINTF(3) +void kr_log_qverbose_impl(const struct kr_query *qry, const char *cls, const char *fmt, ...); + +/** + * @brief Return true if the query has request log handler installed. + */ +#define kr_log_trace_enabled(query) (__builtin_expect( \ + (query) && (query)->request && (query)->request->trace_log, \ + false)) + +/** + * Log a message through the request log handler. + * @param query current query + * @param source message source + * @param fmt message format + * @return true if the message was logged + */ +KR_EXPORT KR_PRINTF(3) +bool kr_log_trace(const struct kr_query *query, const char *source, const char *fmt, ...); + +#ifdef NOVERBOSELOG +/* Efficient compile-time disabling of verbose messages. */ +#define kr_verbose_status false +#define kr_verbose_set(x) +#endif + +/** Block run in --verbose mode; optimized when not run. */ +#define VERBOSE_STATUS __builtin_expect(kr_verbose_status, false) +#define WITH_VERBOSE(query) if(__builtin_expect(kr_verbose_status || kr_log_trace_enabled(query), false)) +#define kr_log_verbose if(VERBOSE_STATUS) kr_log_verbose + +#define KR_DNAME_GET_STR(dname_str, dname) \ + char dname_str[KR_DNAME_STR_MAXLEN]; \ + knot_dname_to_str(dname_str, (dname), sizeof(dname_str)); \ + dname_str[sizeof(dname_str) - 1] = 0; + +#define KR_RRTYPE_GET_STR(rrtype_str, rrtype) \ + char rrtype_str[KR_RRTYPE_STR_MAXLEN]; \ + knot_rrtype_to_string((rrtype), rrtype_str, sizeof(rrtype_str)); \ + rrtype_str[sizeof(rrtype_str) - 1] = 0; + +/* C11 compatibility, but without any implementation so far. */ +#ifndef static_assert +#define static_assert(cond, msg) +#endif + +/** @cond Memory alloc routines */ +static inline void *mm_alloc(knot_mm_t *mm, size_t size) +{ + if (mm) return mm->alloc(mm->ctx, size); + else return malloc(size); +} +static inline void mm_free(knot_mm_t *mm, const void *what) +{ + if (mm) { + if (mm->free) + mm->free((void *)what); + } + else free((void *)what); +} + +/** Realloc implementation using memory context. */ +KR_EXPORT +void *mm_realloc(knot_mm_t *mm, void *what, size_t size, size_t prev_size); + +/** Trivial malloc() wrapper. */ +void *mm_malloc(void *ctx, size_t n); + +/** Initialize mm with standard malloc+free. */ +static inline void mm_ctx_init(knot_mm_t *mm) +{ + mm->ctx = NULL; + mm->alloc = mm_malloc; + mm->free = free; +} +/* @endcond */ + +/** Return time difference in miliseconds. + * @note based on the _BSD_SOURCE timersub() macro */ +static inline long time_diff(struct timeval *begin, struct timeval *end) { + struct timeval res = { + .tv_sec = end->tv_sec - begin->tv_sec, + .tv_usec = end->tv_usec - begin->tv_usec + }; + if (res.tv_usec < 0) { + --res.tv_sec; + res.tv_usec += 1000000; + } + return res.tv_sec * 1000 + res.tv_usec / 1000; +} + +/** @cond internal Array types */ +struct kr_context; + +typedef array_t(knot_rrset_t *) rr_array_t; +struct ranked_rr_array_entry { + uint32_t qry_uid; + uint8_t rank; /**< enum kr_rank */ + uint8_t revalidation_cnt; + bool cached : 1; /**< Set to true if the entry was written into cache */ + bool yielded : 1; + bool to_wire : 1; /**< whether to be put into the answer */ + bool expiring : 1; /**< low remaining TTL; see is_expiring; only used in cache ATM */ + knot_rrset_t *rr; +}; +typedef struct ranked_rr_array_entry ranked_rr_array_entry_t; + +/** Array of RRsets coming from multiple queries; for struct kr_request. + * + * Notes: + * - RRSIGs are only considered to form an RRset when the types covered match; + * cache-related code relies on that! + * - RRsets from the same packet (qry_uid) get merged. + */ +typedef array_t(ranked_rr_array_entry_t *) ranked_rr_array_t; +/* @endcond */ + +/** Concatenate N strings. */ +KR_EXPORT +char* kr_strcatdup(unsigned n, ...); + +/** You probably want kr_rand_* convenience functions instead. + * This is a buffered version of gnutls_rnd(GNUTLS_RND_NONCE, ..) */ +KR_EXPORT +void kr_rnd_buffered(void *data, unsigned int size); + +/** Return a few random bytes. */ +static inline uint64_t kr_rand_bytes(unsigned int size) +{ + uint64_t result; + if (size <= 0 || size > sizeof(result)) { + kr_log_error("kr_rand_bytes(): EINVAL\n"); + abort(); + } + uint8_t data[sizeof(result)]; + kr_rnd_buffered(data, size); + /* I would have liked to dump the random data into a size_t directly, + * but that would work well only on little-endian machines, + * so intsead I hope that the compiler will optimize this out. + * (Tested via reading assembly from usual gcc -O2 setup.) + * Alternatively we could waste more rnd bytes, but that seemed worse. */ + result = 0; + for (unsigned int i = 0; i < size; ++ i) { + result |= ((size_t)data[i]) << (i * 8); + } + return result; +} + +/** Throw a pseudo-random coin, succeeding approximately with probability nomin/denomin. + * - low precision, only one byte of randomness (or none with extreme parameters) + * - tip: use !kr_rand_coin() to get the complementary probability + */ +static inline bool kr_rand_coin(unsigned int nomin, unsigned int denomin) +{ + /* This function might be called with non-constant values + * so we try to handle odd corner cases instead of crash. */ + if (nomin >= denomin) + return true; + else if (nomin <= 0) + return false; + + /* threshold = how many parts from 256 are a success */ + unsigned int threshold = (nomin * 256 + /*rounding*/ denomin / 2) / denomin; + if (threshold == 0) threshold = 1; + if (threshold == 256) threshold = 255; + return (kr_rand_bytes(1) < threshold); +} + +/** Memory reservation routine for knot_mm_t */ +KR_EXPORT +int kr_memreserve(void *baton, char **mem, size_t elm_size, size_t want, size_t *have); + +/** @internal Fast packet reset. */ +KR_EXPORT +int kr_pkt_recycle(knot_pkt_t *pkt); + +/** @internal Clear packet payload. */ +KR_EXPORT +int kr_pkt_clear_payload(knot_pkt_t *pkt); + +/** Construct and put record to packet. */ +KR_EXPORT +int kr_pkt_put(knot_pkt_t *pkt, const knot_dname_t *name, uint32_t ttl, + uint16_t rclass, uint16_t rtype, const uint8_t *rdata, uint16_t rdlen); + +/** Set packet header suitable for authoritative answer. (for policy module) */ +KR_EXPORT +void kr_pkt_make_auth_header(knot_pkt_t *pkt); + +/** Simple storage for IPx address or AF_UNSPEC. */ +union inaddr { + struct sockaddr ip; + struct sockaddr_in ip4; + struct sockaddr_in6 ip6; +}; + +/** Address bytes for given family. */ +KR_EXPORT KR_PURE +const char *kr_inaddr(const struct sockaddr *addr); +/** Address family. */ +KR_EXPORT KR_PURE +int kr_inaddr_family(const struct sockaddr *addr); +/** Address length for given family, i.e. sizeof(struct in*_addr). */ +KR_EXPORT KR_PURE +int kr_inaddr_len(const struct sockaddr *addr); +/** Sockaddr length for given family, i.e. sizeof(struct sockaddr_in*). */ +KR_EXPORT KR_PURE +int kr_sockaddr_len(const struct sockaddr *addr); +/** Compare two given sockaddr. + * return 0 - addresses are equal, error code otherwise. + */ +KR_EXPORT KR_PURE +int kr_sockaddr_cmp(const struct sockaddr *left, const struct sockaddr *right); +/** Port. */ +KR_EXPORT KR_PURE +uint16_t kr_inaddr_port(const struct sockaddr *addr); +/** Set port. */ +KR_EXPORT +void kr_inaddr_set_port(struct sockaddr *addr, uint16_t port); + +/** Write string representation for given address as "#". + * \param[in] addr the raw address + * \param[out] buf the buffer for output string + * \param[in,out] buflen the available(in) and utilized(out) length, including \0 */ +KR_EXPORT +int kr_inaddr_str(const struct sockaddr *addr, char *buf, size_t *buflen); + +/** Return address type for string. */ +KR_EXPORT KR_PURE +int kr_straddr_family(const char *addr); +/** Return address length in given family (struct in*_addr). */ +KR_EXPORT KR_CONST +int kr_family_len(int family); +/** Create a sockaddr* from string+port representation (also accepts IPv6 link-local). */ +KR_EXPORT +struct sockaddr * kr_straddr_socket(const char *addr, int port); +/** Parse address and return subnet length (bits). + * @warning 'dst' must be at least `sizeof(struct in6_addr)` long. */ +KR_EXPORT +int kr_straddr_subnet(void *dst, const char *addr); + +/** Splits ip address specified as "addr@port" or "addr#port" into addr and port + * and performs validation. + * @note if #port part isn't present, then port will be set to 0. + * buf and\or port can be set to NULL. + * @return kr_error(EINVAL) - addr part doesn't contains valid ip address or + * #port part is out-of-range (either < 0 either > UINT16_MAX) + * kr_error(ENOSP) - buflen is too small + */ +KR_EXPORT +int kr_straddr_split(const char *addr, char *buf, size_t buflen, uint16_t *port); +/** Formats ip address and port in "addr#port" format. + * and performs validation. + * @note Port always formatted as five-character string with leading zeros. + * @return kr_error(EINVAL) - addr or buf is NULL or buflen is 0 or + * addr doesn't contain a valid ip address + * kr_error(ENOSP) - buflen is too small + */ +KR_EXPORT +int kr_straddr_join(const char *addr, uint16_t port, char *buf, size_t *buflen); + +/** Compare memory bitwise. The semantics is "the same" as for memcmp(). + * The partial byte is considered with more-significant bits first, + * so this is e.g. suitable for comparing IP prefixes. */ +KR_EXPORT KR_PURE +int kr_bitcmp(const char *a, const char *b, int bits); + +/** @internal RR map flags. */ +static const uint8_t KEY_FLAG_RRSIG = 0x02; +static inline uint8_t KEY_FLAG_RANK(const char *key) + { return ((uint8_t)(key[0])) >> 2; } +static inline bool KEY_COVERING_RRSIG(const char *key) + { return ((uint8_t)(key[0])) & KEY_FLAG_RRSIG; } + +/* Stash key = {[5] class, [1-255] owner, [5] type, [5] additional, [1] \x00 } */ +#define KR_RRKEY_LEN (16 + KNOT_DNAME_MAXLEN) +/** Create unique null-terminated string key for RR. + * @param key Destination buffer for key size, MUST be KR_RRKEY_LEN or larger. + * @param class RR class. + * @param owner RR owner name. + * @param type RR type. + * @param additional flags (for instance can be used for storing covered type + * when RR type is RRSIG). + * @return key length if successful or an error + * */ +KR_EXPORT +int kr_rrkey(char *key, uint16_t class, const knot_dname_t *owner, + uint16_t type, uint16_t additional); + +/** @internal Add RRSet copy to ranked RR array. */ +KR_EXPORT +int kr_ranked_rrarray_add(ranked_rr_array_t *array, const knot_rrset_t *rr, + uint8_t rank, bool to_wire, uint32_t qry_uid, knot_mm_t *pool); + +/** @internal Mark the RRSets from particular query as + * "have (not) to be recorded in the final answer". + * @param array RRSet array. + * @param to_wire Records must be\must not be recorded in final answer. + * @param qry_uid Query uid. + * @param check_dups When to_wire is true, try to avoid duplicate RRSets. + * @param extraCheck optional function checking whether to consider the record + * @return 0 or an error + */ +int kr_ranked_rrarray_set_wire(ranked_rr_array_t *array, bool to_wire, + uint32_t qry_uid, bool check_dups, + bool (*extraCheck)(const ranked_rr_array_entry_t *)); + +KR_PURE +char *kr_pkt_text(const knot_pkt_t *pkt); + +KR_PURE +char *kr_rrset_text(const knot_rrset_t *rr); + +KR_PURE +static inline char *kr_dname_text(const knot_dname_t *name) { + return knot_dname_to_str_alloc(name); +} + +KR_CONST +static inline char *kr_rrtype_text(const uint16_t rrtype) { + char type_str[32] = {0}; + knot_rrtype_to_string(rrtype, type_str, sizeof(type_str)); + return strdup(type_str); +} + +/** + * Call module property. + */ +KR_EXPORT +char *kr_module_call(struct kr_context *ctx, const char *module, const char *prop, const char *input); + +/** Swap two places. Note: the parameters need to be without side effects. */ +#define SWAP(x, y) do { /* http://stackoverflow.com/a/3982430/587396 */ \ + unsigned char swap_temp[sizeof(x) == sizeof(y) ? (ssize_t)sizeof(x) : -1]; \ + memcpy(swap_temp, &y, sizeof(x)); \ + memcpy(&y, &x, sizeof(x)); \ + memcpy(&x, swap_temp, sizeof(x)); \ + } while(0) + +/** Return the (covered) type of an nonempty RRset. */ +static inline uint16_t kr_rrset_type_maysig(const knot_rrset_t *rr) +{ + assert(rr && rr->rrs.count && rr->rrs.rdata); + uint16_t type = rr->type; + if (type == KNOT_RRTYPE_RRSIG) + type = knot_rrsig_type_covered(rr->rrs.rdata); + return type; +} + +/** Printf onto the lua stack, avoiding additional copy (thin wrapper). */ +KR_PRINTF(2) +static inline const char *lua_push_printf(lua_State *L, const char *fmt, ...) +{ + va_list args; + va_start(args, fmt); + const char *ret = lua_pushvfstring(L, fmt, args); + va_end(args); + return ret; +} + +/** @internal Return string representation of addr. + * @note return pointer to static string + */ +static inline char *kr_straddr(const struct sockaddr *addr) +{ + assert(addr != NULL); + /* We are the sinle-threaded application */ + static char str[INET6_ADDRSTRLEN + 1 + 5 + 1]; + size_t len = sizeof(str); + int ret = kr_inaddr_str(addr, str, &len); + return ret != kr_ok() || len == 0 ? NULL : str; +} + +/** The current time in monotonic milliseconds. + * + * \note it may be outdated in case of long callbacks; see uv_now(). + */ +KR_EXPORT +uint64_t kr_now(); + +/** Convert name from lookup format to wire. See knot_dname_lf + * + * \note len bytes are read and len+1 are written with *normal* LF, + * but it's also allowed that the final zero byte is omitted in LF. + * \return the number of bytes written (>0) or error code (<0) + */ +int knot_dname_lf2wire(knot_dname_t *dst, uint8_t len, const uint8_t *lf); + +/** Patched knot_dname_lf. LF for "." has length zero instead of one, for consistency. + * (TODO: consistency?) + * \param add_wildcard append the wildcard label + * \note packet is always NULL + */ +static inline int kr_dname_lf(uint8_t *dst, const knot_dname_t *src, bool add_wildcard) +{ + knot_dname_storage_t right_aligned_dst; + uint8_t *right_aligned_dname_start = knot_dname_lf(src, right_aligned_dst); + if (!right_aligned_dname_start) { + return kr_error(EINVAL); + } + int len = right_aligned_dname_start[0]; + assert(right_aligned_dname_start + 1 + len - KNOT_DNAME_MAXLEN == right_aligned_dst); + memcpy(dst + 1, right_aligned_dname_start + 1, len); + if (add_wildcard) { + if (len + 2 > KNOT_DNAME_MAXLEN) + return kr_error(ENOSPC); + dst[len + 1] = '*'; + dst[len + 2] = '\0'; + len += 2; + } + dst[0] = len; + return KNOT_EOK; +} + +/** + * Difference between two calendar times specified as strings. + * \param format[in] format for strptime + * \param diff[out] result from C difftime(time1, time0) + */ +KR_EXPORT +const char *kr_strptime_diff(const char *format, const char *time1_str, + const char *time0_str, double *diff); + +/* Trivial non-inline wrappers, to be used in lua. */ +KR_EXPORT void kr_rrset_init(knot_rrset_t *rrset, knot_dname_t *owner, + uint16_t type, uint16_t rclass, uint32_t ttl); +KR_EXPORT uint16_t kr_pkt_qclass(const knot_pkt_t *pkt); +KR_EXPORT uint16_t kr_pkt_qtype(const knot_pkt_t *pkt); +KR_EXPORT uint32_t kr_rrsig_sig_inception(const knot_rdata_t *rdata); +KR_EXPORT uint32_t kr_rrsig_sig_expiration(const knot_rdata_t *rdata); +KR_EXPORT uint16_t kr_rrsig_type_covered(const knot_rdata_t *rdata); diff --git a/lib/zonecut.c b/lib/zonecut.c new file mode 100644 index 0000000..5e54d97 --- /dev/null +++ b/lib/zonecut.c @@ -0,0 +1,597 @@ +/* Copyright (C) 2014-2017 CZ.NIC, z.s.p.o. + + This program is free software: you can redistribute it and/or modify + it under the terms of the GNU General Public License as published by + the Free Software Foundation, either version 3 of the License, or + (at your option) any later version. + + This program is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the + GNU General Public License for more details. + + You should have received a copy of the GNU General Public License + along with this program. If not, see . + */ + +#include "lib/zonecut.h" + +#include "contrib/cleanup.h" +#include "lib/defines.h" +#include "lib/generic/pack.h" +#include "lib/layer.h" +#include "lib/resolve.h" +#include "lib/rplan.h" + +#include +#include +#include + +#define VERBOSE_MSG(qry, ...) QRVERBOSE(qry, "zcut", __VA_ARGS__) + +/** Information for one NS name + address type. */ +typedef enum { + AI_UNINITED = 0, + AI_REPUT, /**< Don't use this addrset, due to: cache_rep, NO_IPV6, ... + * cache_rep approximates various problems when fetching the RRset. */ + AI_CYCLED, /**< Skipped due to cycle detection; see implementation for details. */ + AI_LAST_BAD = AI_CYCLED, /** bad states: <= AI_LAST_BAD */ + AI_UNKNOWN, /**< Don't know status of this RRset; various reasons. */ + AI_EMPTY, /**< No usable address (may mean e.g. just NODATA). */ + AI_OK, /**< At least one usable address. + * LATER: we might be interested whether it's only glue. */ +} addrset_info_t; + + +static void update_cut_name(struct kr_zonecut *cut, const knot_dname_t *name) +{ + if (knot_dname_is_equal(name, cut->name)) { + return; + } + knot_dname_t *next_name = knot_dname_copy(name, cut->pool); + mm_free(cut->pool, cut->name); + cut->name = next_name; +} + +int kr_zonecut_init(struct kr_zonecut *cut, const knot_dname_t *name, knot_mm_t *pool) +{ + if (!cut || !name) { + return kr_error(EINVAL); + } + + memset(cut, 0, sizeof(*cut)); + cut->name = knot_dname_copy(name, pool); + cut->pool = pool; + cut->nsset = trie_create(pool); + return cut->name && cut->nsset ? kr_ok() : kr_error(ENOMEM); +} + +/** Completely free a pack_t. */ +static inline void free_addr_set(pack_t *pack, knot_mm_t *pool) +{ + if (unlikely(!pack)) { + /* promised we don't store NULL packs */ + assert(false); + return; + } + pack_clear_mm(*pack, mm_free, pool); + mm_free(pool, pack); +} +/** Trivial wrapper for use in trie_apply, due to ugly casting. */ +static int free_addr_set_cb(trie_val_t *v, void *pool) +{ + free_addr_set(*v, pool); + return kr_ok(); +} + +void kr_zonecut_deinit(struct kr_zonecut *cut) +{ + if (!cut) { + return; + } + mm_free(cut->pool, cut->name); + if (cut->nsset) { + trie_apply(cut->nsset, free_addr_set_cb, cut->pool); + trie_free(cut->nsset); + } + knot_rrset_free(cut->key, cut->pool); + knot_rrset_free(cut->trust_anchor, cut->pool); +} + +void kr_zonecut_move(struct kr_zonecut *to, const struct kr_zonecut *from) +{ + if (!to || !from) abort(); + kr_zonecut_deinit(to); + memcpy(to, from, sizeof(*to)); +} + +void kr_zonecut_set(struct kr_zonecut *cut, const knot_dname_t *name) +{ + if (!cut || !name) { + return; + } + knot_rrset_t *key, *ta; + key = cut->key; cut->key = NULL; + ta = cut->trust_anchor; cut->trust_anchor = NULL; + kr_zonecut_deinit(cut); + kr_zonecut_init(cut, name, cut->pool); + cut->key = key; + cut->trust_anchor = ta; +} + +int kr_zonecut_copy(struct kr_zonecut *dst, const struct kr_zonecut *src) +{ + if (!dst || !src) { + return kr_error(EINVAL); + } + if (!dst->nsset) { + dst->nsset = trie_create(dst->pool); + } + /* Copy the contents, one by one. */ + int ret = kr_ok(); + trie_it_t *it; + for (it = trie_it_begin(src->nsset); !trie_it_finished(it); trie_it_next(it)) { + size_t klen; + const char * const k = trie_it_key(it, &klen); + pack_t **new_pack = (pack_t **)trie_get_ins(dst->nsset, k, klen); + if (!new_pack) { + ret = kr_error(ENOMEM); + break; + } + const pack_t *old_pack = *trie_it_val(it); + ret = pack_clone(new_pack, old_pack, dst->pool); + if (ret) break; + } + trie_it_free(it); + return ret; +} + +int kr_zonecut_copy_trust(struct kr_zonecut *dst, const struct kr_zonecut *src) +{ + knot_rrset_t *key_copy = NULL; + knot_rrset_t *ta_copy = NULL; + + if (src->key) { + key_copy = knot_rrset_copy(src->key, dst->pool); + if (!key_copy) { + return kr_error(ENOMEM); + } + } + + if (src->trust_anchor) { + ta_copy = knot_rrset_copy(src->trust_anchor, dst->pool); + if (!ta_copy) { + knot_rrset_free(key_copy, dst->pool); + return kr_error(ENOMEM); + } + } + + knot_rrset_free(dst->key, dst->pool); + dst->key = key_copy; + knot_rrset_free(dst->trust_anchor, dst->pool); + dst->trust_anchor = ta_copy; + + return kr_ok(); +} + +int kr_zonecut_add(struct kr_zonecut *cut, const knot_dname_t *ns, const void *data, int len) +{ + if (!cut || !ns || !cut->nsset || (data && len <= 0)) { + assert(!EINVAL); + return kr_error(EINVAL); + } + /* Disabled; add_reverse_pair() misuses this for domain name in rdata. */ + if (false && data && len != sizeof(struct in_addr) + && len != sizeof(struct in6_addr)) { + assert(!EINVAL); + return kr_error(EINVAL); + } + + /* Get a pack_t for the ns. */ + pack_t **pack = (pack_t **)trie_get_ins(cut->nsset, (const char *)ns, knot_dname_size(ns)); + if (!pack) return kr_error(ENOMEM); + if (*pack == NULL) { + *pack = mm_alloc(cut->pool, sizeof(pack_t)); + if (*pack == NULL) return kr_error(ENOMEM); + pack_init(**pack); + } + /* Insert data (if has any) */ + if (data == NULL) { + return kr_ok(); + } + /* Check for duplicates */ + if (pack_obj_find(*pack, data, len)) { + return kr_ok(); + } + /* Push new address */ + int ret = pack_reserve_mm(**pack, 1, len, kr_memreserve, cut->pool); + if (ret != 0) { + return kr_error(ENOMEM); + } + return pack_obj_push(*pack, data, len); +} + +int kr_zonecut_del(struct kr_zonecut *cut, const knot_dname_t *ns, const void *data, int len) +{ + if (!cut || !ns || (data && len <= 0)) { + return kr_error(EINVAL); + } + + /* Find the address list. */ + int ret = kr_ok(); + pack_t *pack = kr_zonecut_find(cut, ns); + if (pack == NULL) { + return kr_error(ENOENT); + } + /* Remove address from the pack. */ + if (data) { + ret = pack_obj_del(pack, data, len); + } + /* No servers left, remove NS from the set. */ + if (pack->len == 0) { + free_addr_set(pack, cut->pool); + ret = trie_del(cut->nsset, (const char *)ns, knot_dname_size(ns), NULL); + assert(ret == 0); /* only KNOT_ENOENT and that *can't* happen */ + return (ret == 0) ? kr_ok() : kr_error(ret); + } + + return ret; +} + +int kr_zonecut_del_all(struct kr_zonecut *cut, const knot_dname_t *ns) +{ + if (!cut || !ns) { + return kr_error(EINVAL); + } + + /* Find the address list; then free and remove it. */ + pack_t *pack; + int ret = trie_del(cut->nsset, (const char *)ns, knot_dname_size(ns), + (trie_val_t *)&pack); + if (ret) { /* deletion failed */ + assert(ret == KNOT_ENOENT); + return kr_error(ENOENT); + } + free_addr_set(pack, cut->pool); + return kr_ok(); +} + +pack_t *kr_zonecut_find(struct kr_zonecut *cut, const knot_dname_t *ns) +{ + if (!cut || !ns) { + return NULL; + } + trie_val_t *val = trie_get_try(cut->nsset, (const char *)ns, knot_dname_size(ns)); + /* we get pointer to the pack_t pointer */ + return val ? (pack_t *)*val : NULL; +} + +static int has_address(trie_val_t *v, void *baton_) +{ + const pack_t *pack = *v; + const bool found = pack != NULL && pack->len != 0; + return found; +} + +bool kr_zonecut_is_empty(struct kr_zonecut *cut) +{ + if (!cut || !cut->nsset) { + assert(false); + return true; + } + return !trie_apply(cut->nsset, has_address, NULL); +} + +int kr_zonecut_set_sbelt(struct kr_context *ctx, struct kr_zonecut *cut) +{ + if (!ctx || !cut || !ctx->root_hints.nsset) { + return kr_error(EINVAL); + } + + trie_apply(cut->nsset, free_addr_set_cb, cut->pool); + trie_clear(cut->nsset); + + const uint8_t *const dname_root = (const uint8_t *)/*sign-cast*/(""); + update_cut_name(cut, dname_root); + /* Copy root hints from resolution context. */ + return kr_zonecut_copy(cut, &ctx->root_hints); +} + +/** Fetch address for zone cut. Any rank is accepted (i.e. glue as well). */ +static addrset_info_t fetch_addr(pack_t *addrs, const knot_dname_t *ns, uint16_t rrtype, + knot_mm_t *mm_pool, const struct kr_query *qry) +// LATER(optim.): excessive data copying +{ + int rdlen; + switch (rrtype) { + case KNOT_RRTYPE_A: + rdlen = 4; + break; + case KNOT_RRTYPE_AAAA: + rdlen = 16; + break; + default: + assert(!EINVAL); + return AI_UNKNOWN; + } + + struct kr_context *ctx = qry->request->ctx; + struct kr_cache_p peek; + if (kr_cache_peek_exact(&ctx->cache, ns, rrtype, &peek) != 0) { + return AI_UNKNOWN; + } + int32_t new_ttl = kr_cache_ttl(&peek, qry, ns, rrtype); + if (new_ttl < 0) { + return AI_UNKNOWN; + } + + knot_rrset_t cached_rr; + knot_rrset_init(&cached_rr, /*const-cast*/(knot_dname_t *)ns, rrtype, + KNOT_CLASS_IN, new_ttl); + if (kr_cache_materialize(&cached_rr.rrs, &peek, mm_pool) < 0) { + return AI_UNKNOWN; + } + + /* Reserve memory in *addrs. Implementation detail: + * pack_t cares for lengths, so we don't store those in the data. */ + const size_t pack_extra_size = knot_rdataset_size(&cached_rr.rrs) + - cached_rr.rrs.count * offsetof(knot_rdata_t, len); + int ret = pack_reserve_mm(*addrs, cached_rr.rrs.count, pack_extra_size, + kr_memreserve, mm_pool); + if (ret) abort(); /* ENOMEM "probably" */ + + int usable_cnt = 0; + addrset_info_t result = AI_EMPTY; + knot_rdata_t *rd = cached_rr.rrs.rdata; + for (uint16_t i = 0; i < cached_rr.rrs.count; ++i, rd = knot_rdataset_next(rd)) { + if (unlikely(rd->len != rdlen)) { + VERBOSE_MSG(qry, "bad NS address length %d for rrtype %d, skipping\n", + (int)rd->len, (int)rrtype); + continue; + } + /* Check RTT cache - whether the IP is usable or not. */ + kr_nsrep_rtt_lru_entry_t *rtt_e = ctx->cache_rtt + ? lru_get_try(ctx->cache_rtt, (const char *)rd->data, rd->len) + : NULL; + const bool unusable = rtt_e && rtt_e->score >= KR_NS_TIMEOUT + && qry->creation_time_mono + < rtt_e->tout_timestamp + ctx->cache_rtt_tout_retry_interval; + if (!unusable) { + result = AI_OK; + ++usable_cnt; + } + + ret = pack_obj_push(addrs, rd->data, rd->len); + assert(!ret); /* didn't fit because of incorrectly reserved memory */ + /* LATER: for now we lose quite some information here, + * as keeping it would need substantial changes on other places, + * and it turned out to be premature optimization (most likely). + * We might e.g. skip adding unusable addresses, + * and either keep some rtt information associated + * or even finish up choosing the set to send packets to. + * Overall there's some overlap with nsrep.c functionality. + */ + } + if (usable_cnt != cached_rr.rrs.count) { + VERBOSE_MSG(qry, "usable NS addresses: %d/%d\n", + usable_cnt, cached_rr.rrs.count); + } + return result; +} + +/** Fetch best NS for zone cut. */ +static int fetch_ns(struct kr_context *ctx, struct kr_zonecut *cut, + const knot_dname_t *name, const struct kr_query *qry, + uint8_t * restrict rank) +{ + struct kr_cache_p peek; + int ret = kr_cache_peek_exact(&ctx->cache, name, KNOT_RRTYPE_NS, &peek); + if (ret != 0) { + return ret; + } + /* Note: we accept *any* rank from the cache. We assume that nothing + * completely untrustworthy could get into the cache, e.g out-of-bailiwick + * records that weren't validated. + */ + *rank = peek.rank; + + int32_t new_ttl = kr_cache_ttl(&peek, qry, name, KNOT_RRTYPE_NS); + if (new_ttl < 0) { + return kr_error(ESTALE); + } + /* Materialize the rdataset temporarily, for simplicity. */ + knot_rdataset_t ns_rds = { 0, NULL }; + ret = kr_cache_materialize(&ns_rds, &peek, cut->pool); + if (ret < 0) { + return ret; + } + + /* Insert name servers for this zone cut, addresses will be looked up + * on-demand (either from cache or iteratively) */ + bool all_bad = true; /**< All NSs (seen so far) are in a bad state. */ + knot_rdata_t *rdata_i = ns_rds.rdata; + for (unsigned i = 0; i < ns_rds.count; + ++i, rdata_i = knot_rdataset_next(rdata_i)) { + const knot_dname_t *ns_name = knot_ns_name(rdata_i); + const size_t ns_size = knot_dname_size(ns_name); + + /* Get a new pack within the nsset. */ + pack_t **pack = (pack_t **)trie_get_ins(cut->nsset, + (const char *)ns_name, ns_size); + if (!pack) return kr_error(ENOMEM); + assert(!*pack); /* not critical, really */ + *pack = mm_alloc(cut->pool, sizeof(pack_t)); + if (!*pack) return kr_error(ENOMEM); + pack_init(**pack); + + addrset_info_t infos[2]; + /* Fetch NS reputation and decide whether to prefetch A/AAAA records. */ + unsigned *cached = lru_get_try(ctx->cache_rep, + (const char *)ns_name, ns_size); + unsigned reputation = (cached) ? *cached : 0; + infos[0] = (reputation & KR_NS_NOIP4) || qry->flags.NO_IPV4 + ? AI_REPUT + : fetch_addr(*pack, ns_name, KNOT_RRTYPE_A, cut->pool, qry); + infos[1] = (reputation & KR_NS_NOIP6) || qry->flags.NO_IPV6 + ? AI_REPUT + : fetch_addr(*pack, ns_name, KNOT_RRTYPE_AAAA, cut->pool, qry); + + #if 0 /* rather unlikely to be useful unless changing some zcut code */ + WITH_VERBOSE(qry) { + auto_free char *ns_name_txt = kr_dname_text(ns_name); + VERBOSE_MSG(qry, "NS %s infos: %d, %d\n", + ns_name_txt, (int)infos[0], (int)infos[1]); + } + #endif + + /* AI_CYCLED checks. + * If an ancestor query has its zone cut in the state that + * it's looking for name or address(es) of some NS(s), + * we want to avoid doing so with a NS that lies under its cut. + * Instead we need to consider such names unusable in the cut (for now). */ + if (infos[0] != AI_UNKNOWN && infos[1] != AI_UNKNOWN) { + /* Optimization: the following loop would be pointless. */ + all_bad = false; + continue; + } + for (const struct kr_query *aq = qry; aq->parent; aq = aq->parent) { + const struct kr_qflags *aqpf = &aq->parent->flags; + if ( (aqpf->AWAIT_CUT && aq->stype == KNOT_RRTYPE_NS) + || (aqpf->AWAIT_IPV4 && aq->stype == KNOT_RRTYPE_A) + || (aqpf->AWAIT_IPV6 && aq->stype == KNOT_RRTYPE_AAAA)) { + if (knot_dname_in_bailiwick(ns_name, + aq->parent->zone_cut.name)) { + for (int j = 0; j < 2; ++j) + if (infos[j] == AI_UNKNOWN) + infos[j] = AI_CYCLED; + break; + } + } else { + /* This ancestor waits for other reason that + * NS name or address, so we're out of a direct cycle. */ + break; + } + } + all_bad = all_bad && infos[0] <= AI_LAST_BAD && infos[1] <= AI_LAST_BAD; + } + + if (all_bad) { WITH_VERBOSE(qry) { + auto_free char *name_txt = kr_dname_text(name); + VERBOSE_MSG(qry, "cut %s: all NSs bad, count = %d\n", + name_txt, (int)ns_rds.count); + } } + knot_rdataset_clear(&ns_rds, cut->pool); + return all_bad ? ELOOP : kr_ok(); +} + +/** + * Fetch secure RRSet of given type. + */ +static int fetch_secure_rrset(knot_rrset_t **rr, struct kr_cache *cache, + const knot_dname_t *owner, uint16_t type, knot_mm_t *pool, + const struct kr_query *qry) +{ + if (!rr) { + assert(!EINVAL); + return kr_error(EINVAL); + } + /* peek, check rank and TTL */ + struct kr_cache_p peek; + int ret = kr_cache_peek_exact(cache, owner, type, &peek); + if (ret != 0) { + return ret; + } + if (!kr_rank_test(peek.rank, KR_RANK_SECURE)) { + return kr_error(ENOENT); + } + int32_t new_ttl = kr_cache_ttl(&peek, qry, owner, type); + if (new_ttl < 0) { + return kr_error(ESTALE); + } + /* materialize a new RRset */ + knot_rrset_free(*rr, pool); + *rr = mm_alloc(pool, sizeof(knot_rrset_t)); + if (*rr == NULL) { + return kr_error(ENOMEM); + } + owner = knot_dname_copy(/*const-cast*/(knot_dname_t *)owner, pool); + if (!owner) { + mm_free(pool, *rr); + *rr = NULL; + return kr_error(ENOMEM); + } + knot_rrset_init(*rr, /*const-cast*/(knot_dname_t *)owner, type, + KNOT_CLASS_IN, new_ttl); + ret = kr_cache_materialize(&(*rr)->rrs, &peek, pool); + if (ret < 0) { + knot_rrset_free(*rr, pool); + *rr = NULL; + return ret; + } + + return kr_ok(); +} + +int kr_zonecut_find_cached(struct kr_context *ctx, struct kr_zonecut *cut, + const knot_dname_t *name, const struct kr_query *qry, + bool * restrict secured) +{ + if (!ctx || !cut || !name) { + //assert(false); + return kr_error(EINVAL); + } + /* I'm not sure whether the caller always passes a clean state; + * mixing doesn't seem to make sense in any case, so let's clear it. + * We don't bother freeing the packs, as they're on mempool. */ + trie_clear(cut->nsset); + /* Copy name as it may overlap with cut name that is to be replaced. */ + knot_dname_t *qname = knot_dname_copy(name, cut->pool); + if (!qname) { + return kr_error(ENOMEM); + } + /* Start at QNAME. */ + int ret; + const knot_dname_t *label = qname; + while (true) { + /* Fetch NS first and see if it's insecure. */ + uint8_t rank = 0; + const bool is_root = (label[0] == '\0'); + ret = fetch_ns(ctx, cut, label, qry, &rank); + if (ret == 0) { + /* Flag as insecure if cached as this */ + if (kr_rank_test(rank, KR_RANK_INSECURE)) { + *secured = false; + } + /* Fetch DS and DNSKEY if caller wants secure zone cut */ + int ret_ds = 1, ret_dnskey = 1; + if (*secured || is_root) { + ret_ds = fetch_secure_rrset(&cut->trust_anchor, &ctx->cache, + label, KNOT_RRTYPE_DS, cut->pool, qry); + ret_dnskey = fetch_secure_rrset(&cut->key, &ctx->cache, + label, KNOT_RRTYPE_DNSKEY, cut->pool, qry); + } + update_cut_name(cut, label); + WITH_VERBOSE(qry) { + auto_free char *label_str = kr_dname_text(label); + VERBOSE_MSG(qry, + "found cut: %s (rank 0%.2o return codes: DS %d, DNSKEY %d)\n", + label_str, rank, ret_ds, ret_dnskey); + } + ret = kr_ok(); + break; + } /* else */ + + trie_clear(cut->nsset); + /* Subtract label from QNAME. */ + if (!is_root) { + label = knot_wire_next_label(label, NULL); + } else { + ret = kr_error(ENOENT); + break; + } + } + + kr_cache_sync(&ctx->cache); + mm_free(cut->pool, qname); + return ret; +} diff --git a/lib/zonecut.h b/lib/zonecut.h new file mode 100644 index 0000000..2808c66 --- /dev/null +++ b/lib/zonecut.h @@ -0,0 +1,173 @@ +/* Copyright (C) 2014-2017 CZ.NIC, z.s.p.o. + + This program is free software: you can redistribute it and/or modify + it under the terms of the GNU General Public License as published by + the Free Software Foundation, either version 3 of the License, or + (at your option) any later version. + + This program is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the + GNU General Public License for more details. + + You should have received a copy of the GNU General Public License + along with this program. If not, see . + */ + +#pragma once + +#include "lib/cache/api.h" +#include "lib/defines.h" +#include "lib/generic/pack.h" +#include "lib/generic/trie.h" + +struct kr_rplan; +struct kr_context; + +/** + * Current zone cut representation. +*/ +struct kr_zonecut { + knot_dname_t *name; /**< Zone cut name. */ + knot_rrset_t* key; /**< Zone cut DNSKEY. */ + knot_rrset_t* trust_anchor; /**< Current trust anchor. */ + struct kr_zonecut *parent; /**< Parent zone cut. */ + trie_t *nsset; /**< Map of nameserver => address_set (pack_t). */ + knot_mm_t *pool; /**< Memory pool. */ +}; + +/** + * Populate root zone cut with SBELT. + * @param cut zone cut + * @param name + * @param pool + * @return 0 or error code + */ +KR_EXPORT +int kr_zonecut_init(struct kr_zonecut *cut, const knot_dname_t *name, knot_mm_t *pool); + +/** + * Clear the structure and free the address set. + * @param cut zone cut + */ +KR_EXPORT +void kr_zonecut_deinit(struct kr_zonecut *cut); + +/** + * Move a zonecut, transferring ownership of any pointed-to memory. + * @param to the target - it gets deinit-ed + * @param from the source - not modified, but shouldn't be used afterward + */ +KR_EXPORT +void kr_zonecut_move(struct kr_zonecut *to, const struct kr_zonecut *from); + +/** + * Reset zone cut to given name and clear address list. + * @note This clears the address list even if the name doesn't change. TA and DNSKEY don't change. + * @param cut zone cut to be set + * @param name new zone cut name + */ +KR_EXPORT +void kr_zonecut_set(struct kr_zonecut *cut, const knot_dname_t *name); + +/** + * Copy zone cut, including all data. Does not copy keys and trust anchor. + * @param dst destination zone cut + * @param src source zone cut + * @return 0 or an error code; If it fails with kr_error(ENOMEM), + * it may be in a half-filled state, but it's safe to deinit... + * @note addresses for names in `src` get replaced and others are left as they were. + */ +KR_EXPORT +int kr_zonecut_copy(struct kr_zonecut *dst, const struct kr_zonecut *src); + +/** + * Copy zone trust anchor and keys. + * @param dst destination zone cut + * @param src source zone cut + * @return 0 or an error code + */ +KR_EXPORT +int kr_zonecut_copy_trust(struct kr_zonecut *dst, const struct kr_zonecut *src); + +/** + * Add address record to the zone cut. + * + * The record will be merged with existing data, + * it may be either A/AAAA type. + * + * @param cut zone cut to be populated + * @param ns nameserver name + * @param data typically knot_rdata_t::data + * @param len typically knot_rdata_t::len + * @return 0 or error code + */ +KR_EXPORT +int kr_zonecut_add(struct kr_zonecut *cut, const knot_dname_t *ns, const void *data, int len); + +/** + * Delete nameserver/address pair from the zone cut. + * @param cut + * @param ns name server name + * @param data typically knot_rdata_t::data + * @param len typically knot_rdata_t::len + * @return 0 or error code + */ +KR_EXPORT +int kr_zonecut_del(struct kr_zonecut *cut, const knot_dname_t *ns, const void *data, int len); + +/** + * Delete all addresses associated with the given name. + * @param cut + * @param ns name server name + * @return 0 or error code + */ +KR_EXPORT +int kr_zonecut_del_all(struct kr_zonecut *cut, const knot_dname_t *ns); + +/** + * Find nameserver address list in the zone cut. + * + * @note This can be used for membership test, a non-null pack is returned + * if the nameserver name exists. + * + * @param cut + * @param ns name server name + * @return pack of addresses or NULL + */ +KR_EXPORT KR_PURE +pack_t *kr_zonecut_find(struct kr_zonecut *cut, const knot_dname_t *ns); + +/** + * Populate zone cut with a root zone using SBELT :rfc:`1034` + * + * @param ctx resolution context (to fetch root hints) + * @param cut zone cut to be populated + * @return 0 or error code + */ +KR_EXPORT +int kr_zonecut_set_sbelt(struct kr_context *ctx, struct kr_zonecut *cut); + +/** + * Populate zone cut address set from cache. + * + * @param ctx resolution context (to fetch data from LRU caches) + * @param cut zone cut to be populated + * @param name QNAME to start finding zone cut for + * @param qry query for timestamp and stale-serving decisions + * @param secured set to true if want secured zone cut, will return false if it is provably insecure + * @return 0 or error code (ENOENT if it doesn't find anything) + */ +KR_EXPORT +int kr_zonecut_find_cached(struct kr_context *ctx, struct kr_zonecut *cut, + const knot_dname_t *name, const struct kr_query *qry, + bool * restrict secured); +/** + * Check if any address is present in the zone cut. + * + * @param cut zone cut to check + * @return true/false + */ +KR_EXPORT +bool kr_zonecut_is_empty(struct kr_zonecut *cut); + -- cgit v1.2.3