/* SPDX-License-Identifier: LGPL-2.1-or-later */ #include #include #include "capability-util.h" #include "errno-util.h" #include "fd-util.h" #include "missing_network.h" #include "missing_socket.h" #include "resolved-dns-stub.h" #include "socket-netlink.h" #include "socket-util.h" #include "stdio-util.h" #include "string-table.h" /* The MTU of the loopback device is 64K on Linux, advertise that as maximum datagram size, but subtract the Ethernet, * IP and UDP header sizes */ #define ADVERTISE_DATAGRAM_SIZE_MAX (65536U-14U-20U-8U) /* On the extra stubs, use a more conservative choice */ #define ADVERTISE_EXTRA_DATAGRAM_SIZE_MAX DNS_PACKET_UNICAST_SIZE_LARGE_MAX static int manager_dns_stub_fd_extra(Manager *m, DnsStubListenerExtra *l, int type); static int manager_dns_stub_fd(Manager *m, int family, const union in_addr_union *listen_address, int type); static void dns_stub_listener_extra_hash_func(const DnsStubListenerExtra *a, struct siphash *state) { assert(a); siphash24_compress(&a->mode, sizeof(a->mode), state); siphash24_compress(&a->family, sizeof(a->family), state); siphash24_compress(&a->address, FAMILY_ADDRESS_SIZE(a->family), state); siphash24_compress(&a->port, sizeof(a->port), state); } static int dns_stub_listener_extra_compare_func(const DnsStubListenerExtra *a, const DnsStubListenerExtra *b) { int r; assert(a); assert(b); r = CMP(a->mode, b->mode); if (r != 0) return r; r = CMP(a->family, b->family); if (r != 0) return r; r = memcmp(&a->address, &b->address, FAMILY_ADDRESS_SIZE(a->family)); if (r != 0) return r; return CMP(a->port, b->port); } DEFINE_HASH_OPS_WITH_KEY_DESTRUCTOR( dns_stub_listener_extra_hash_ops, DnsStubListenerExtra, dns_stub_listener_extra_hash_func, dns_stub_listener_extra_compare_func, dns_stub_listener_extra_free); int dns_stub_listener_extra_new( Manager *m, DnsStubListenerExtra **ret) { DnsStubListenerExtra *l; l = new(DnsStubListenerExtra, 1); if (!l) return -ENOMEM; *l = (DnsStubListenerExtra) { .manager = m, }; *ret = TAKE_PTR(l); return 0; } DnsStubListenerExtra *dns_stub_listener_extra_free(DnsStubListenerExtra *p) { if (!p) return NULL; p->udp_event_source = sd_event_source_disable_unref(p->udp_event_source); p->tcp_event_source = sd_event_source_disable_unref(p->tcp_event_source); hashmap_free(p->queries_by_packet); return mfree(p); } static void stub_packet_hash_func(const DnsPacket *p, struct siphash *state) { assert(p); siphash24_compress(&p->protocol, sizeof(p->protocol), state); siphash24_compress(&p->family, sizeof(p->family), state); siphash24_compress(&p->sender, sizeof(p->sender), state); siphash24_compress(&p->ipproto, sizeof(p->ipproto), state); siphash24_compress(&p->sender_port, sizeof(p->sender_port), state); siphash24_compress(DNS_PACKET_HEADER(p), sizeof(DnsPacketHeader), state); /* We don't bother hashing the full packet here, just the header */ } static int stub_packet_compare_func(const DnsPacket *x, const DnsPacket *y) { int r; r = CMP(x->protocol, y->protocol); if (r != 0) return r; r = CMP(x->family, y->family); if (r != 0) return r; r = memcmp(&x->sender, &y->sender, sizeof(x->sender)); if (r != 0) return r; r = CMP(x->ipproto, y->ipproto); if (r != 0) return r; r = CMP(x->sender_port, y->sender_port); if (r != 0) return r; return memcmp(DNS_PACKET_HEADER(x), DNS_PACKET_HEADER(y), sizeof(DnsPacketHeader)); } DEFINE_HASH_OPS(stub_packet_hash_ops, DnsPacket, stub_packet_hash_func, stub_packet_compare_func); static int reply_add_with_rrsig( DnsAnswer **reply, DnsResourceRecord *rr, int ifindex, DnsAnswerFlags flags, DnsResourceRecord *rrsig, bool with_rrsig) { int r; assert(reply); assert(rr); r = dns_answer_add_extend(reply, rr, ifindex, flags, rrsig); if (r < 0) return r; if (with_rrsig && rrsig) { r = dns_answer_add_extend(reply, rrsig, ifindex, flags, NULL); if (r < 0) return r; } return 0; } static int dns_stub_collect_answer_by_question( DnsAnswer **reply, DnsAnswer *answer, DnsQuestion *question, bool with_rrsig) { /* Add RRSIG RR matching each RR */ DnsAnswerItem *item; int r; assert(reply); /* Copies all RRs from 'answer' into 'reply', if they match 'question'. */ DNS_ANSWER_FOREACH_ITEM(item, answer) { /* We have a question, let's see if this RR matches it */ r = dns_question_matches_rr(question, item->rr, NULL); if (r < 0) return r; if (!r) { /* Maybe there's a CNAME/DNAME in here? If so, that's an answer too */ r = dns_question_matches_cname_or_dname(question, item->rr, NULL); if (r < 0) return r; if (!r) continue; } /* Mask the section info, we want the primary answers to always go without section * info, so that it is added to the answer section when we synthesize a reply. */ r = reply_add_with_rrsig( reply, item->rr, item->ifindex, item->flags & ~DNS_ANSWER_MASK_SECTIONS, item->rrsig, with_rrsig); if (r < 0) return r; } return 0; } static int dns_stub_collect_answer_by_section( DnsAnswer **reply, DnsAnswer *answer, DnsAnswerFlags section, DnsAnswer *exclude1, DnsAnswer *exclude2, bool with_dnssec) { /* Include DNSSEC RRs. RRSIG, NSEC, … */ DnsAnswerItem *item; int r; assert(reply); /* Copies all RRs from 'answer' into 'reply', if they originate from the specified section. Also, * avoid any RRs listed in 'exclude'. */ DNS_ANSWER_FOREACH_ITEM(item, answer) { if (dns_answer_contains(exclude1, item->rr) || dns_answer_contains(exclude2, item->rr)) continue; if (!with_dnssec && dns_type_is_dnssec(item->rr->key->type)) continue; if (((item->flags ^ section) & DNS_ANSWER_MASK_SECTIONS) != 0) continue; r = reply_add_with_rrsig( reply, item->rr, item->ifindex, item->flags, item->rrsig, with_dnssec); if (r < 0) return r; } return 0; } static int dns_stub_assign_sections( DnsQuery *q, DnsQuestion *question, bool edns0_do) { int r; assert(q); assert(question); /* Let's assign the 'answer' RRs we collected to their respective sections in the reply datagram. We * try to reproduce a section assignment similar to what the upstream DNS server responded to us. We * use the DNS_ANSWER_SECTION_xyz flags to match things up, which is where the original upstream's * packet section assignment is stored in the DnsAnswer object. Not all RRs in the 'answer' objects * come with section information though (for example, because they were synthesized locally, and not * from a DNS packet). To deal with that we extend the assignment logic a bit: anything from the * 'answer' object that directly matches the original question is always put in the ANSWER section, * regardless if it carries section info, or what that section info says. Then, anything from the * 'answer' objects that is from the ANSWER or AUTHORITY sections, and wasn't already added to the * ANSWER section is placed in the AUTHORITY section. Everything else from either object is added to * the ADDITIONAL section. */ /* Include all RRs that directly answer the question in the answer section */ r = dns_stub_collect_answer_by_question( &q->reply_answer, q->answer, question, edns0_do); if (r < 0) return r; /* Include all RRs that originate from the authority sections, and aren't already listed in the * answer section, in the authority section */ r = dns_stub_collect_answer_by_section( &q->reply_authoritative, q->answer, DNS_ANSWER_SECTION_AUTHORITY, q->reply_answer, NULL, edns0_do); if (r < 0) return r; /* Include all RRs that originate from the answer or additional sections in the additional section * (except if already listed in the other two sections). Also add all RRs with no section marking. */ r = dns_stub_collect_answer_by_section( &q->reply_additional, q->answer, DNS_ANSWER_SECTION_ANSWER, q->reply_answer, q->reply_authoritative, edns0_do); if (r < 0) return r; r = dns_stub_collect_answer_by_section( &q->reply_additional, q->answer, DNS_ANSWER_SECTION_ADDITIONAL, q->reply_answer, q->reply_authoritative, edns0_do); if (r < 0) return r; r = dns_stub_collect_answer_by_section( &q->reply_additional, q->answer, 0, q->reply_answer, q->reply_authoritative, edns0_do); if (r < 0) return r; return 0; } static int dns_stub_make_reply_packet( DnsPacket **ret, size_t max_size, DnsQuestion *q, bool *ret_truncated) { _cleanup_(dns_packet_unrefp) DnsPacket *p = NULL; bool tc = false; int r; assert(ret); r = dns_packet_new(&p, DNS_PROTOCOL_DNS, 0, max_size); if (r < 0) return r; r = dns_packet_append_question(p, q); if (r == -EMSGSIZE) tc = true; else if (r < 0) return r; if (ret_truncated) *ret_truncated = tc; else if (tc) return -EMSGSIZE; DNS_PACKET_HEADER(p)->qdcount = htobe16(dns_question_size(q)); *ret = TAKE_PTR(p); return 0; } static int dns_stub_add_reply_packet_body( DnsPacket *p, DnsAnswer *answer, DnsAnswer *authoritative, DnsAnswer *additional, bool edns0_do, /* Client expects DNSSEC RRs? */ bool *truncated) { unsigned n_answer = 0, n_authoritative = 0, n_additional = 0; bool tc = false; int r; assert(p); /* Add the three sections to the packet. If the answer section doesn't fit we'll signal that as * truncation. If the authoritative section doesn't fit and we are in DNSSEC mode, also signal * truncation. In all other cases where things don't fit don't signal truncation, as for those cases * the dropped RRs should not be essential. */ r = dns_packet_append_answer(p, answer, &n_answer); if (r == -EMSGSIZE) tc = true; else if (r < 0) return r; else { r = dns_packet_append_answer(p, authoritative, &n_authoritative); if (r == -EMSGSIZE) { if (edns0_do) tc = true; } else if (r < 0) return r; else { r = dns_packet_append_answer(p, additional, &n_additional); if (r < 0 && r != -EMSGSIZE) return r; } } if (tc) { if (!truncated) return -EMSGSIZE; *truncated = true; } DNS_PACKET_HEADER(p)->ancount = htobe16(n_answer); DNS_PACKET_HEADER(p)->nscount = htobe16(n_authoritative); DNS_PACKET_HEADER(p)->arcount = htobe16(n_additional); return 0; } static const char *nsid_string(void) { static char buffer[SD_ID128_STRING_MAX + STRLEN(".resolved.systemd.io")] = ""; sd_id128_t id; int r; /* Let's generate a string that we can use as RFC5001 NSID identifier. The string shall identify us * as systemd-resolved, and return a different string for each resolved instance without leaking host * identity. Hence let's use a fixed suffix that identifies resolved, and a prefix generated from the * machine ID but from which the machine ID cannot be determined. * * Clients can use this to determine whether an answer is originating locally or is proxied from * upstream. */ if (!isempty(buffer)) return buffer; r = sd_id128_get_machine_app_specific( SD_ID128_MAKE(ed,d3,12,5d,16,b9,41,f9,a1,49,5f,ab,15,62,ab,27), &id); if (r < 0) { log_debug_errno(r, "Failed to determine machine ID, ignoring: %m"); return NULL; } xsprintf(buffer, SD_ID128_FORMAT_STR ".resolved.systemd.io", SD_ID128_FORMAT_VAL(id)); return buffer; } static int dns_stub_finish_reply_packet( DnsPacket *p, uint16_t id, int rcode, bool tc, /* set the Truncated bit? */ bool aa, /* set the Authoritative Answer bit? */ bool rd, /* set the Recursion Desired bit? */ bool add_opt, /* add an OPT RR to this packet? */ bool edns0_do, /* set the EDNS0 DNSSEC OK bit? */ bool ad, /* set the DNSSEC authenticated data bit? */ bool cd, /* set the DNSSEC checking disabled bit? */ uint16_t max_udp_size, /* The maximum UDP datagram size to advertise to clients */ bool nsid) { /* whether to add NSID */ int r; assert(p); if (add_opt) { r = dns_packet_append_opt(p, max_udp_size, edns0_do, /* include_rfc6975 = */ false, nsid ? nsid_string() : NULL, rcode, NULL); if (r == -EMSGSIZE) /* Hit the size limit? then indicate truncation */ tc = true; else if (r < 0) return r; } else { /* If the client can't to EDNS0, don't do DO either */ edns0_do = false; /* If we don't do EDNS, clamp the rcode to 4 bit */ if (rcode > 0xF) rcode = DNS_RCODE_SERVFAIL; } /* Don't set the CD bit unless DO is on, too */ if (!edns0_do) cd = false; /* Note that we allow the AD bit to be set even if client didn't signal DO, as per RFC 6840, section * 5.7 */ DNS_PACKET_HEADER(p)->id = id; DNS_PACKET_HEADER(p)->flags = htobe16(DNS_PACKET_MAKE_FLAGS( 1 /* qr */, 0 /* opcode */, aa /* aa */, tc /* tc */, rd /* rd */, 1 /* ra */, ad /* ad */, cd /* cd */, rcode)); return 0; } static bool address_is_proxy(int family, const union in_addr_union *a) { assert(a); /* Returns true if the specified address is the DNS "proxy" stub, i.e. where we unconditionally enable bypass mode */ if (family != AF_INET) return false; return be32toh(a->in.s_addr) == INADDR_DNS_PROXY_STUB; } static int find_socket_fd( Manager *m, DnsStubListenerExtra *l, int family, const union in_addr_union *listen_address, int type) { assert(m); /* Finds the right socket to use for sending. If we know the extra listener, otherwise go via the * address to send from */ if (l) return manager_dns_stub_fd_extra(m, l, type); return manager_dns_stub_fd(m, family, listen_address, type); } static int dns_stub_send( Manager *m, DnsStubListenerExtra *l, DnsStream *s, DnsPacket *p, DnsPacket *reply) { int r; assert(m); assert(p); assert(reply); if (s) r = dns_stream_write_packet(s, reply); else { int fd, ifindex; fd = find_socket_fd(m, l, p->family, &p->destination, SOCK_DGRAM); if (fd < 0) return fd; if (address_is_proxy(p->family, &p->destination)) /* Force loopback iface if this is the loopback proxy stub * and ifindex was normalized to 0 by manager_recv(). */ ifindex = p->ifindex ?: LOOPBACK_IFINDEX; else /* Force loopback iface if this is the main listener stub. */ ifindex = l ? p->ifindex : LOOPBACK_IFINDEX; /* Note that it is essential here that we explicitly choose the source IP address for this * packet. This is because otherwise the kernel will choose it automatically based on the * routing table and will thus pick 127.0.0.1 rather than 127.0.0.53/54. */ r = manager_send(m, fd, ifindex, p->family, &p->sender, p->sender_port, &p->destination, reply); } if (r < 0) return log_debug_errno(r, "Failed to send reply packet: %m"); return 0; } static int dns_stub_reply_with_edns0_do(DnsQuery *q) { assert(q); /* Reply with DNSSEC DO set? Only if client supports it; and we did any DNSSEC verification * ourselves, or consider the data fully authenticated because we generated it locally, or the client * set cd */ return DNS_PACKET_DO(q->request_packet) && (q->answer_dnssec_result >= 0 || /* we did proper DNSSEC validation … */ dns_query_fully_authenticated(q) || /* … or we considered it authentic otherwise … */ DNS_PACKET_CD(q->request_packet)); /* … or client set CD */ } static void dns_stub_suppress_duplicate_section_rrs(DnsQuery *q) { /* If we follow a CNAME/DNAME chain we might end up populating our sections with redundant RRs * because we built up the sections from multiple reply packets (one from each CNAME/DNAME chain * element). E.g. it could be that an RR that was included in the first reply's additional section * ends up being relevant as main answer in a subsequent reply in the chain. Let's clean this up, and * remove everything in the "higher priority" sections from the "lower priority" sections. * * Note that this removal matches by RR keys instead of the full RRs. This is because RRsets should * always end up in one section fully or not at all, but never be split among sections. * * Specifically: we remove ANSWER section RRs from the AUTHORITATIVE and ADDITIONAL sections, as well * as AUTHORITATIVE section RRs from the ADDITIONAL section. */ dns_answer_remove_by_answer_keys(&q->reply_authoritative, q->reply_answer); dns_answer_remove_by_answer_keys(&q->reply_additional, q->reply_answer); dns_answer_remove_by_answer_keys(&q->reply_additional, q->reply_authoritative); } static int dns_stub_send_reply( DnsQuery *q, int rcode) { _cleanup_(dns_packet_unrefp) DnsPacket *reply = NULL; bool truncated, edns0_do; int r; assert(q); edns0_do = dns_stub_reply_with_edns0_do(q); /* let's check if we shall reply with EDNS0 DO? */ r = dns_stub_make_reply_packet( &reply, DNS_PACKET_PAYLOAD_SIZE_MAX(q->request_packet), q->request_packet->question, &truncated); if (r < 0) return log_debug_errno(r, "Failed to build reply packet: %m"); dns_stub_suppress_duplicate_section_rrs(q); r = dns_stub_add_reply_packet_body( reply, q->reply_answer, q->reply_authoritative, q->reply_additional, edns0_do, &truncated); if (r < 0) return log_debug_errno(r, "Failed to append reply packet body: %m"); r = dns_stub_finish_reply_packet( reply, DNS_PACKET_ID(q->request_packet), rcode, truncated, dns_query_fully_authoritative(q), DNS_PACKET_RD(q->request_packet), !!q->request_packet->opt, edns0_do, (DNS_PACKET_AD(q->request_packet) || DNS_PACKET_DO(q->request_packet)) && dns_query_fully_authenticated(q), DNS_PACKET_CD(q->request_packet), q->stub_listener_extra ? ADVERTISE_EXTRA_DATAGRAM_SIZE_MAX : ADVERTISE_DATAGRAM_SIZE_MAX, dns_packet_has_nsid_request(q->request_packet) > 0 && !q->stub_listener_extra); if (r < 0) return log_debug_errno(r, "Failed to build failure packet: %m"); return dns_stub_send(q->manager, q->stub_listener_extra, q->request_stream, q->request_packet, reply); } static int dns_stub_send_failure( Manager *m, DnsStubListenerExtra *l, DnsStream *s, DnsPacket *p, int rcode, bool authenticated) { _cleanup_(dns_packet_unrefp) DnsPacket *reply = NULL; bool truncated; int r; assert(m); assert(p); r = dns_stub_make_reply_packet( &reply, DNS_PACKET_PAYLOAD_SIZE_MAX(p), p->question, &truncated); if (r < 0) return log_debug_errno(r, "Failed to make failure packet: %m"); r = dns_stub_finish_reply_packet( reply, DNS_PACKET_ID(p), rcode, truncated, false, DNS_PACKET_RD(p), !!p->opt, DNS_PACKET_DO(p), (DNS_PACKET_AD(p) || DNS_PACKET_DO(p)) && authenticated, DNS_PACKET_CD(p), l ? ADVERTISE_EXTRA_DATAGRAM_SIZE_MAX : ADVERTISE_DATAGRAM_SIZE_MAX, dns_packet_has_nsid_request(p) > 0 && !l); if (r < 0) return log_debug_errno(r, "Failed to build failure packet: %m"); return dns_stub_send(m, l, s, p, reply); } static int dns_stub_patch_bypass_reply_packet( DnsPacket **ret, /* Where to place the patched packet */ DnsPacket *original, /* The packet to patch */ DnsPacket *request) { /* The packet the patched packet shall look like a reply to */ _cleanup_(dns_packet_unrefp) DnsPacket *c = NULL; int r; assert(ret); assert(original); assert(request); r = dns_packet_dup(&c, original); if (r < 0) return r; /* Extract the packet, so that we know where the OPT field is */ r = dns_packet_extract(c); if (r < 0) return r; /* Copy over the original client request ID, so that we can make the upstream query look like our own reply. */ DNS_PACKET_HEADER(c)->id = DNS_PACKET_HEADER(request)->id; /* Patch in our own maximum datagram size, if EDNS0 was on */ r = dns_packet_patch_max_udp_size(c, ADVERTISE_DATAGRAM_SIZE_MAX); if (r < 0) return r; /* Lower all TTLs by the time passed since we received the datagram. */ if (timestamp_is_set(original->timestamp)) { r = dns_packet_patch_ttls(c, original->timestamp); if (r < 0) return r; } /* Our upstream connection might have supported larger DNS requests than our downstream one, hence * set the TC bit if our reply is larger than what the client supports, and truncate. */ if (c->size > DNS_PACKET_PAYLOAD_SIZE_MAX(request)) { log_debug("Artificially truncating stub response, as advertised size of client is smaller than upstream one."); dns_packet_truncate(c, DNS_PACKET_PAYLOAD_SIZE_MAX(request)); DNS_PACKET_HEADER(c)->flags = htobe16(be16toh(DNS_PACKET_HEADER(c)->flags) | DNS_PACKET_FLAG_TC); } *ret = TAKE_PTR(c); return 0; } static void dns_stub_query_complete(DnsQuery *query) { _cleanup_(dns_query_freep) DnsQuery *q = query; int r; assert(q); assert(q->request_packet); if (q->question_bypass) { /* This is a bypass reply. If so, let's propagate the upstream packet, if we have it and it * is regular DNS. (We can't do this if the upstream packet is LLMNR or mDNS, since the * packets are not 100% compatible.) */ if (q->answer_full_packet && q->answer_full_packet->protocol == DNS_PROTOCOL_DNS) { _cleanup_(dns_packet_unrefp) DnsPacket *reply = NULL; r = dns_stub_patch_bypass_reply_packet(&reply, q->answer_full_packet, q->request_packet); if (r < 0) log_debug_errno(r, "Failed to patch bypass reply packet: %m"); else (void) dns_stub_send(q->manager, q->stub_listener_extra, q->request_stream, q->request_packet, reply); return; } } /* Take all data from the current reply, and merge it into the three reply sections we are building * up. We do this before processing CNAME redirects, so that we gradually build up our sections, and * and keep adding all RRs in the CNAME chain. */ r = dns_stub_assign_sections( q, dns_query_question_for_protocol(q, DNS_PROTOCOL_DNS), dns_stub_reply_with_edns0_do(q)); if (r < 0) return (void) log_debug_errno(r, "Failed to assign sections: %m"); switch (q->state) { case DNS_TRANSACTION_SUCCESS: { bool first = true; for (;;) { int cname_result; cname_result = dns_query_process_cname_one(q); if (cname_result == -ELOOP) { /* CNAME loop, let's send what we already have */ log_debug("Detected CNAME loop, returning what we already have."); (void) dns_stub_send_reply(q, q->answer_rcode); break; } if (cname_result < 0) { log_debug_errno(cname_result, "Failed to process CNAME: %m"); break; } if (cname_result == DNS_QUERY_NOMATCH) { /* This answer doesn't contain any RR that would answer our question * positively, i.e. neither directly nor via CNAME. */ if (first) /* We never followed a CNAME and the answer doesn't match our * question at all? Then this is final, the empty answer is the * answer. */ break; /* Otherwise, we already followed a CNAME once within this packet, and the * packet doesn't answer our question. In that case let's restart the query, * now with the redirected question. We'll */ r = dns_query_go(q); if (r < 0) return (void) log_debug_errno(r, "Failed to restart query: %m"); TAKE_PTR(q); return; } r = dns_stub_assign_sections( q, dns_query_question_for_protocol(q, DNS_PROTOCOL_DNS), dns_stub_reply_with_edns0_do(q)); if (r < 0) return (void) log_debug_errno(r, "Failed to assign sections: %m"); if (cname_result == DNS_QUERY_MATCH) /* A match? Then we are done, let's return what we got */ break; /* We followed a CNAME. and collected the RRs that answer the redirected question * successfully. Let's not try to do this again. */ assert(cname_result == DNS_QUERY_CNAME); first = false; } _fallthrough_; } case DNS_TRANSACTION_RCODE_FAILURE: (void) dns_stub_send_reply(q, q->answer_rcode); break; case DNS_TRANSACTION_NOT_FOUND: (void) dns_stub_send_reply(q, DNS_RCODE_NXDOMAIN); break; case DNS_TRANSACTION_TIMEOUT: case DNS_TRANSACTION_ATTEMPTS_MAX_REACHED: /* Propagate a timeout as a no packet, i.e. that the client also gets a timeout */ break; case DNS_TRANSACTION_NO_SERVERS: /* We're not configured to give answers for this question. Refuse it. */ (void) dns_stub_send_reply(q, DNS_RCODE_REFUSED); break; case DNS_TRANSACTION_RR_TYPE_UNSUPPORTED: /* This RR Type is not implemented */ (void) dns_stub_send_reply(q, DNS_RCODE_NOTIMP); break; case DNS_TRANSACTION_INVALID_REPLY: case DNS_TRANSACTION_ERRNO: case DNS_TRANSACTION_ABORTED: case DNS_TRANSACTION_DNSSEC_FAILED: case DNS_TRANSACTION_NO_TRUST_ANCHOR: case DNS_TRANSACTION_NETWORK_DOWN: case DNS_TRANSACTION_NO_SOURCE: case DNS_TRANSACTION_STUB_LOOP: (void) dns_stub_send_reply(q, DNS_RCODE_SERVFAIL); break; case DNS_TRANSACTION_NULL: case DNS_TRANSACTION_PENDING: case DNS_TRANSACTION_VALIDATING: default: assert_not_reached(); } } static int dns_stub_stream_complete(DnsStream *s, int error) { assert(s); log_debug_errno(error, "DNS TCP connection terminated, destroying queries: %m"); for (;;) { DnsQuery *q; q = set_first(s->queries); if (!q) break; dns_query_free(q); } /* This drops the implicit ref we keep around since it was allocated, as incoming stub connections * should be kept as long as the client wants to. */ dns_stream_unref(s); return 0; } static void dns_stub_process_query(Manager *m, DnsStubListenerExtra *l, DnsStream *s, DnsPacket *p) { uint64_t protocol_flags = SD_RESOLVED_PROTOCOLS_ALL; _cleanup_(dns_query_freep) DnsQuery *q = NULL; Hashmap **queries_by_packet; DnsQuery *existing; bool bypass = false; int r; assert(m); assert(p); assert(p->protocol == DNS_PROTOCOL_DNS); if (!l && /* l == NULL if this is the main stub */ !address_is_proxy(p->family, &p->destination) && /* don't restrict needlessly for 127.0.0.54 */ (in_addr_is_localhost(p->family, &p->sender) <= 0 || in_addr_is_localhost(p->family, &p->destination) <= 0)) { log_warning("Got packet on unexpected (i.e. non-localhost) IP range, ignoring."); return; } if (manager_packet_from_our_transaction(m, p)) { log_debug("Got our own packet looped back, ignoring."); return; } queries_by_packet = l ? &l->queries_by_packet : &m->stub_queries_by_packet; existing = hashmap_get(*queries_by_packet, p); if (existing && dns_packet_equal(existing->request_packet, p)) { log_debug("Got repeat packet from client, ignoring."); return; } r = dns_packet_extract(p); if (r < 0) { log_debug_errno(r, "Failed to extract resources from incoming packet, ignoring packet: %m"); dns_stub_send_failure(m, l, s, p, DNS_RCODE_FORMERR, false); return; } if (!DNS_PACKET_VERSION_SUPPORTED(p)) { log_debug("Got EDNS OPT field with unsupported version number."); dns_stub_send_failure(m, l, s, p, DNS_RCODE_BADVERS, false); return; } if (dns_type_is_obsolete(dns_question_first_key(p->question)->type)) { log_debug("Got message with obsolete key type, refusing."); dns_stub_send_failure(m, l, s, p, DNS_RCODE_REFUSED, false); return; } if (dns_type_is_zone_transer(dns_question_first_key(p->question)->type)) { log_debug("Got request for zone transfer, refusing."); dns_stub_send_failure(m, l, s, p, DNS_RCODE_REFUSED, false); return; } if (!DNS_PACKET_RD(p)) { /* If the "rd" bit is off (i.e. recursion was not requested), then refuse operation */ log_debug("Got request with recursion disabled, refusing."); dns_stub_send_failure(m, l, s, p, DNS_RCODE_REFUSED, false); return; } r = hashmap_ensure_allocated(queries_by_packet, &stub_packet_hash_ops); if (r < 0) { log_oom(); return; } if (address_is_proxy(p->family, &p->destination)) { _cleanup_free_ char *dipa = NULL; r = in_addr_to_string(p->family, &p->destination, &dipa); if (r < 0) return (void) log_error_errno(r, "Failed to format destination address: %m"); log_debug("Got request to DNS proxy address 127.0.0.54, enabling bypass logic."); bypass = true; protocol_flags = SD_RESOLVED_DNS|SD_RESOLVED_NO_ZONE; /* Turn off mDNS/LLMNR for proxy stub. */ } else if ((DNS_PACKET_DO(p) && DNS_PACKET_CD(p))) { log_debug("Got request with DNSSEC checking disabled, enabling bypass logic."); bypass = true; } if (bypass) r = dns_query_new(m, &q, NULL, NULL, p, 0, protocol_flags| SD_RESOLVED_NO_CNAME| SD_RESOLVED_NO_SEARCH| SD_RESOLVED_NO_VALIDATE| SD_RESOLVED_REQUIRE_PRIMARY| SD_RESOLVED_CLAMP_TTL); else r = dns_query_new(m, &q, p->question, p->question, NULL, 0, protocol_flags| SD_RESOLVED_NO_SEARCH| (DNS_PACKET_DO(p) ? SD_RESOLVED_REQUIRE_PRIMARY : 0)| SD_RESOLVED_CLAMP_TTL); if (r < 0) { log_error_errno(r, "Failed to generate query object: %m"); dns_stub_send_failure(m, l, s, p, DNS_RCODE_SERVFAIL, false); return; } q->request_packet = dns_packet_ref(p); q->request_stream = dns_stream_ref(s); /* make sure the stream stays around until we can send a reply through it */ q->stub_listener_extra = l; q->complete = dns_stub_query_complete; if (s) { /* Remember which queries belong to this stream, so that we can cancel them when the stream * is disconnected early */ r = set_ensure_put(&s->queries, NULL, q); if (r < 0) { log_oom(); return; } assert(r > 0); } /* Add the query to the hash table we use to determine repeat packets now. We don't care about * failures here, since in the worst case we'll not recognize duplicate incoming requests, which * isn't particularly bad. */ (void) hashmap_put(*queries_by_packet, q->request_packet, q); r = dns_query_go(q); if (r < 0) { log_error_errno(r, "Failed to start query: %m"); dns_stub_send_failure(m, l, s, p, DNS_RCODE_SERVFAIL, false); return; } log_debug("Processing query..."); TAKE_PTR(q); } static int on_dns_stub_packet_internal(sd_event_source *s, int fd, uint32_t revents, Manager *m, DnsStubListenerExtra *l) { _cleanup_(dns_packet_unrefp) DnsPacket *p = NULL; int r; r = manager_recv(m, fd, DNS_PROTOCOL_DNS, &p); if (r <= 0) return r; if (dns_packet_validate_query(p) > 0) { log_debug("Got DNS stub UDP query packet for id %u", DNS_PACKET_ID(p)); dns_stub_process_query(m, l, NULL, p); } else log_debug("Invalid DNS stub UDP packet, ignoring."); return 0; } static int on_dns_stub_packet(sd_event_source *s, int fd, uint32_t revents, void *userdata) { return on_dns_stub_packet_internal(s, fd, revents, userdata, NULL); } static int on_dns_stub_packet_extra(sd_event_source *s, int fd, uint32_t revents, void *userdata) { DnsStubListenerExtra *l = ASSERT_PTR(userdata); return on_dns_stub_packet_internal(s, fd, revents, l->manager, l); } static int on_dns_stub_stream_packet(DnsStream *s, DnsPacket *p) { assert(s); assert(s->manager); assert(p); if (dns_packet_validate_query(p) > 0) { log_debug("Got DNS stub TCP query packet for id %u", DNS_PACKET_ID(p)); dns_stub_process_query(s->manager, s->stub_listener_extra, s, p); } else log_debug("Invalid DNS stub TCP packet, ignoring."); return 0; } static int on_dns_stub_stream_internal(sd_event_source *s, int fd, uint32_t revents, Manager *m, DnsStubListenerExtra *l) { DnsStream *stream; int cfd, r; cfd = accept4(fd, NULL, NULL, SOCK_NONBLOCK|SOCK_CLOEXEC); if (cfd < 0) { if (ERRNO_IS_ACCEPT_AGAIN(errno)) return 0; return -errno; } r = dns_stream_new(m, &stream, DNS_STREAM_STUB, DNS_PROTOCOL_DNS, cfd, NULL, on_dns_stub_stream_packet, dns_stub_stream_complete, DNS_STREAM_STUB_TIMEOUT_USEC); if (r < 0) { safe_close(cfd); return r; } stream->stub_listener_extra = l; /* We let the reference to the stream dangle here, it will be dropped later by the complete callback. */ return 0; } static int on_dns_stub_stream(sd_event_source *s, int fd, uint32_t revents, void *userdata) { return on_dns_stub_stream_internal(s, fd, revents, userdata, NULL); } static int on_dns_stub_stream_extra(sd_event_source *s, int fd, uint32_t revents, void *userdata) { DnsStubListenerExtra *l = ASSERT_PTR(userdata); return on_dns_stub_stream_internal(s, fd, revents, l->manager, l); } static int set_dns_stub_common_socket_options(int fd, int family) { int r; assert(fd >= 0); assert(IN_SET(family, AF_INET, AF_INET6)); r = setsockopt_int(fd, SOL_SOCKET, SO_REUSEADDR, true); if (r < 0) return r; r = socket_set_recvpktinfo(fd, family, true); if (r < 0) return r; r = socket_set_recvttl(fd, family, true); if (r < 0) return r; return 0; } static int set_dns_stub_common_tcp_socket_options(int fd) { int r; assert(fd >= 0); r = setsockopt_int(fd, IPPROTO_TCP, TCP_FASTOPEN, 5); /* Everybody appears to pick qlen=5, let's do the same here. */ if (r < 0) log_debug_errno(r, "Failed to enable TCP_FASTOPEN on TCP listening socket, ignoring: %m"); r = setsockopt_int(fd, IPPROTO_TCP, TCP_NODELAY, true); if (r < 0) log_debug_errno(r, "Failed to enable TCP_NODELAY mode, ignoring: %m"); return 0; } static int manager_dns_stub_fd( Manager *m, int family, const union in_addr_union *listen_addr, int type) { sd_event_source **event_source; _cleanup_close_ int fd = -EBADF; union sockaddr_union sa; int r; assert(m); assert(listen_addr); if (type == SOCK_DGRAM) event_source = address_is_proxy(family, listen_addr) ? &m->dns_proxy_stub_udp_event_source : &m->dns_stub_udp_event_source; else if (type == SOCK_STREAM) event_source = address_is_proxy(family, listen_addr) ? &m->dns_proxy_stub_tcp_event_source : &m->dns_stub_tcp_event_source; else return -EPROTONOSUPPORT; if (*event_source) return sd_event_source_get_io_fd(*event_source); fd = socket(family, type | SOCK_CLOEXEC | SOCK_NONBLOCK, 0); if (fd < 0) return -errno; r = set_dns_stub_common_socket_options(fd, family); if (r < 0) return r; if (type == SOCK_STREAM) { r = set_dns_stub_common_tcp_socket_options(fd); if (r < 0) return r; } /* Set slightly different socket options for the non-proxy and the proxy binding. The former we want * to be accessible only from the local host, for the latter it's OK if people use NAT redirects or * so to redirect external traffic to it. */ if (!address_is_proxy(family, listen_addr)) { /* Make sure no traffic from outside the local host can leak to onto this socket */ r = socket_bind_to_ifindex(fd, LOOPBACK_IFINDEX); if (r < 0) return r; r = socket_set_ttl(fd, family, 1); if (r < 0) return r; } else if (type == SOCK_DGRAM) { /* Turn off Path MTU Discovery for UDP, for security reasons. See socket_disable_pmtud() for * a longer discussion. (We only do this for sockets that are potentially externally * accessible, i.e. the proxy stub one. For the non-proxy one we instead set the TTL to 1, * see above, so that packets don't get routed at all.) */ r = socket_disable_pmtud(fd, family); if (r < 0) log_debug_errno(r, "Failed to disable UDP PMTUD, ignoring: %m"); r = socket_set_recvfragsize(fd, family, true); if (r < 0) log_debug_errno(r, "Failed to enable fragment size reception, ignoring: %m"); } r = sockaddr_set_in_addr(&sa, family, listen_addr, 53); if (r < 0) return r; if (bind(fd, &sa.sa, sizeof(sa.in)) < 0) return -errno; if (type == SOCK_STREAM && listen(fd, SOMAXCONN_DELUXE) < 0) return -errno; r = sd_event_add_io(m->event, event_source, fd, EPOLLIN, type == SOCK_DGRAM ? on_dns_stub_packet : on_dns_stub_stream, m); if (r < 0) return r; r = sd_event_source_set_io_fd_own(*event_source, true); if (r < 0) return r; (void) sd_event_source_set_description(*event_source, type == SOCK_DGRAM ? "dns-stub-udp" : "dns-stub-tcp"); return TAKE_FD(fd); } static int manager_dns_stub_fd_extra(Manager *m, DnsStubListenerExtra *l, int type) { _cleanup_free_ char *pretty = NULL; _cleanup_close_ int fd = -EBADF; union sockaddr_union sa; int r; assert(m); assert(l); assert(IN_SET(type, SOCK_DGRAM, SOCK_STREAM)); sd_event_source **event_source = type == SOCK_DGRAM ? &l->udp_event_source : &l->tcp_event_source; if (*event_source) return sd_event_source_get_io_fd(*event_source); if (!have_effective_cap(CAP_NET_BIND_SERVICE) && dns_stub_listener_extra_port(l) < 1024) { log_warning("Missing CAP_NET_BIND_SERVICE capability, not creating extra stub listener on port %hu.", dns_stub_listener_extra_port(l)); return 0; } if (l->family == AF_INET) sa = (union sockaddr_union) { .in.sin_family = l->family, .in.sin_port = htobe16(dns_stub_listener_extra_port(l)), .in.sin_addr = l->address.in, }; else sa = (union sockaddr_union) { .in6.sin6_family = l->family, .in6.sin6_port = htobe16(dns_stub_listener_extra_port(l)), .in6.sin6_addr = l->address.in6, }; fd = socket(l->family, type | SOCK_CLOEXEC | SOCK_NONBLOCK, 0); if (fd < 0) { r = -errno; goto fail; } r = set_dns_stub_common_socket_options(fd, l->family); if (r < 0) goto fail; if (type == SOCK_STREAM) { r = set_dns_stub_common_tcp_socket_options(fd); if (r < 0) goto fail; } /* Do not set IP_TTL for extra DNS stub listeners, as the address may not be local and in that case * people may want ttl > 1. */ r = socket_set_freebind(fd, l->family, true); if (r < 0) goto fail; if (type == SOCK_DGRAM) { r = socket_disable_pmtud(fd, l->family); if (r < 0) log_debug_errno(r, "Failed to disable UDP PMTUD, ignoring: %m"); r = socket_set_recvfragsize(fd, l->family, true); if (r < 0) log_debug_errno(r, "Failed to enable fragment size reception, ignoring: %m"); } r = RET_NERRNO(bind(fd, &sa.sa, SOCKADDR_LEN(sa))); if (r < 0) goto fail; if (type == SOCK_STREAM && listen(fd, SOMAXCONN_DELUXE) < 0) { r = -errno; goto fail; } r = sd_event_add_io(m->event, event_source, fd, EPOLLIN, type == SOCK_DGRAM ? on_dns_stub_packet_extra : on_dns_stub_stream_extra, l); if (r < 0) goto fail; r = sd_event_source_set_io_fd_own(*event_source, true); if (r < 0) goto fail; (void) sd_event_source_set_description(*event_source, type == SOCK_DGRAM ? "dns-stub-udp-extra" : "dns-stub-tcp-extra"); if (DEBUG_LOGGING) { (void) in_addr_port_to_string(l->family, &l->address, l->port, &pretty); log_debug("Listening on %s socket %s.", type == SOCK_DGRAM ? "UDP" : "TCP", strnull(pretty)); } return TAKE_FD(fd); fail: assert(r < 0); (void) in_addr_port_to_string(l->family, &l->address, l->port, &pretty); return log_warning_errno(r, r == -EADDRINUSE ? "Another process is already listening on %s socket %s: %m" : "Failed to listen on %s socket %s: %m", type == SOCK_DGRAM ? "UDP" : "TCP", strnull(pretty)); } int manager_dns_stub_start(Manager *m) { int r; assert(m); if (m->dns_stub_listener_mode == DNS_STUB_LISTENER_NO) log_debug("Not creating stub listener."); else if (!have_effective_cap(CAP_NET_BIND_SERVICE)) log_warning("Missing CAP_NET_BIND_SERVICE capability, not creating stub listener on port 53."); else { static const struct { uint32_t addr; int socket_type; } stub_sockets[] = { { INADDR_DNS_STUB, SOCK_DGRAM }, { INADDR_DNS_STUB, SOCK_STREAM }, { INADDR_DNS_PROXY_STUB, SOCK_DGRAM }, { INADDR_DNS_PROXY_STUB, SOCK_STREAM }, }; log_debug("Creating stub listener using %s.", m->dns_stub_listener_mode == DNS_STUB_LISTENER_UDP ? "UDP" : m->dns_stub_listener_mode == DNS_STUB_LISTENER_TCP ? "TCP" : "UDP/TCP"); for (size_t i = 0; i < ELEMENTSOF(stub_sockets); i++) { union in_addr_union a = { .in.s_addr = htobe32(stub_sockets[i].addr), }; if (m->dns_stub_listener_mode == DNS_STUB_LISTENER_UDP && stub_sockets[i].socket_type == SOCK_STREAM) continue; if (m->dns_stub_listener_mode == DNS_STUB_LISTENER_TCP && stub_sockets[i].socket_type == SOCK_DGRAM) continue; r = manager_dns_stub_fd(m, AF_INET, &a, stub_sockets[i].socket_type); if (r < 0) { _cleanup_free_ char *busy_socket = NULL; if (asprintf(&busy_socket, "%s socket " IPV4_ADDRESS_FMT_STR ":53", stub_sockets[i].socket_type == SOCK_DGRAM ? "UDP" : "TCP", IPV4_ADDRESS_FMT_VAL(a.in)) < 0) return log_oom(); if (IN_SET(r, -EADDRINUSE, -EPERM)) { log_warning_errno(r, r == -EADDRINUSE ? "Another process is already listening on %s.\n" "Turning off local DNS stub support." : "Failed to listen on %s: %m.\n" "Turning off local DNS stub support.", busy_socket); manager_dns_stub_stop(m); break; } return log_error_errno(r, "Failed to listen on %s: %m", busy_socket); } } } if (!ordered_set_isempty(m->dns_extra_stub_listeners)) { DnsStubListenerExtra *l; log_debug("Creating extra stub listeners."); ORDERED_SET_FOREACH(l, m->dns_extra_stub_listeners) { if (FLAGS_SET(l->mode, DNS_STUB_LISTENER_UDP)) (void) manager_dns_stub_fd_extra(m, l, SOCK_DGRAM); if (FLAGS_SET(l->mode, DNS_STUB_LISTENER_TCP)) (void) manager_dns_stub_fd_extra(m, l, SOCK_STREAM); } } return 0; } void manager_dns_stub_stop(Manager *m) { assert(m); m->dns_stub_udp_event_source = sd_event_source_disable_unref(m->dns_stub_udp_event_source); m->dns_stub_tcp_event_source = sd_event_source_disable_unref(m->dns_stub_tcp_event_source); m->dns_proxy_stub_udp_event_source = sd_event_source_disable_unref(m->dns_proxy_stub_udp_event_source); m->dns_proxy_stub_tcp_event_source = sd_event_source_disable_unref(m->dns_proxy_stub_tcp_event_source); } static const char* const dns_stub_listener_mode_table[_DNS_STUB_LISTENER_MODE_MAX] = { [DNS_STUB_LISTENER_NO] = "no", [DNS_STUB_LISTENER_UDP] = "udp", [DNS_STUB_LISTENER_TCP] = "tcp", [DNS_STUB_LISTENER_YES] = "yes", }; DEFINE_STRING_TABLE_LOOKUP_WITH_BOOLEAN(dns_stub_listener_mode, DnsStubListenerMode, DNS_STUB_LISTENER_YES);