summaryrefslogtreecommitdiff
diff options
context:
space:
mode:
authorLennart Poettering <lennart@poettering.net>2016-01-15 18:18:54 +0100
committerLennart Poettering <lennart@poettering.net>2016-01-17 20:47:46 +0100
commitc3f7000e611b2c08052aca6db47245e77c008ae6 (patch)
tree64094689b3d4b2264d59bc67715e0b943d3560d0
parentafc58cc2fb5841154fe036ee7a6e1c8a06bc5d29 (diff)
resolved: ignore invalid OPT RRs in incoming packets
This validates OPT RRs more rigorously, before honouring them: if we any of the following condition holds, we'll ignore them: a) Multiple OPT RRs in the same message b) OPT RR not owned by the root domain c) OPT RR in the wrong section (Belkin routers do this) d) OPT RR contain rfc6975 algorithm data (Belkin routers do this) e) OPT version is not 0 f) OPT payload doesn't add up with the lengths Note that d) may be an indication that the server just blindly copied OPT data from the response into the reply. RFC6975 data is only supposed to be included in queries, and we do so. It's not supposed to be included in responses (and the RFC is very clear on that). Hence if we get it back in a reply, then the server probably just copied the OPT RR.
-rw-r--r--src/resolve/resolved-dns-packet.c107
1 files changed, 88 insertions, 19 deletions
diff --git a/src/resolve/resolved-dns-packet.c b/src/resolve/resolved-dns-packet.c
index a8a8632491..57cfc1ab9b 100644
--- a/src/resolve/resolved-dns-packet.c
+++ b/src/resolve/resolved-dns-packet.c
@@ -2017,6 +2017,48 @@ fail:
return r;
}
+static bool opt_is_good(DnsResourceRecord *rr, bool *rfc6975) {
+ const uint8_t* p;
+ bool found_dau_dhu_n3u = false;
+ size_t l;
+
+ /* Checks whether the specified OPT RR is well-formed and whether it contains RFC6975 data (which is not OK in
+ * a reply). */
+
+ assert(rr);
+ assert(rr->key->type == DNS_TYPE_OPT);
+
+ /* Check that the version is 0 */
+ if (((rr->ttl >> 16) & UINT32_C(0xFF)) != 0)
+ return false;
+
+ p = rr->opt.data;
+ l = rr->opt.size;
+ while (l > 0) {
+ uint16_t option_code, option_length;
+
+ /* At least four bytes for OPTION-CODE and OPTION-LENGTH are required */
+ if (l < 4U)
+ return false;
+
+ option_code = unaligned_read_be16(p);
+ option_length = unaligned_read_be16(p + 2);
+
+ if (l < option_length + 4U)
+ return false;
+
+ /* RFC 6975 DAU, DHU or N3U fields found. */
+ if (IN_SET(option_code, 5, 6, 7))
+ found_dau_dhu_n3u = true;
+
+ p += option_length + 4U;
+ l -= option_length + 4U;
+ }
+
+ *rfc6975 = found_dau_dhu_n3u;
+ return true;
+}
+
int dns_packet_extract(DnsPacket *p) {
_cleanup_(dns_question_unrefp) DnsQuestion *question = NULL;
_cleanup_(dns_answer_unrefp) DnsAnswer *answer = NULL;
@@ -2064,6 +2106,8 @@ int dns_packet_extract(DnsPacket *p) {
n = DNS_PACKET_RRCOUNT(p);
if (n > 0) {
+ bool bad_opt = false;
+
answer = dns_answer_new(n);
if (!answer) {
r = -ENOMEM;
@@ -2079,35 +2123,57 @@ int dns_packet_extract(DnsPacket *p) {
goto finish;
if (rr->key->type == DNS_TYPE_OPT) {
+ bool has_rfc6975;
+
+ if (p->opt || bad_opt) {
+ /* Multiple OPT RRs? if so, let's ignore all, because there's something wrong
+ * with the server, and if one is valid we wouldn't know which one. */
+ log_debug("Multiple OPT RRs detected, ignoring all.");
+ bad_opt = true;
+ continue;
+ }
if (!dns_name_is_root(DNS_RESOURCE_KEY_NAME(rr->key))) {
- r = -EBADMSG;
- goto finish;
+ /* If the OPT RR qis not owned by the root domain, then it is bad, let's ignore
+ * it. */
+ log_debug("OPT RR is not owned by root domain, ignoring.");
+ bad_opt = true;
+ continue;
+ }
+
+ if (i < DNS_PACKET_ANCOUNT(p) + DNS_PACKET_NSCOUNT(p)) {
+ /* OPT RR is in the wrong section? Some Belkin routers do this. This is a hint
+ * the EDNS implementation is borked, like the Belkin one is, hence ignore
+ * it. */
+ log_debug("OPT RR in wrong section, ignoring.");
+ bad_opt = true;
+ continue;
}
- /* Note that we accept the OPT RR in
- * any section, not just in the
- * additional section, as some routers
- * (Belkin!) blindly copy the OPT RR
- * from the query to the reply packet,
- * and don't get the section right. */
+ if (!opt_is_good(rr, &has_rfc6975)) {
+ log_debug("Malformed OPT RR, ignoring.");
+ bad_opt = true;
+ continue;
+ }
- /* Two OPT RRs? */
- if (p->opt) {
- r = -EBADMSG;
- goto finish;
+ if (has_rfc6975) {
+ /* OPT RR contains RFC6975 algorithm data, then this is indication that the
+ * server just copied the OPT it got from us (which contained that data) back
+ * into the reply. If so, then it doesn't properly support EDNS, as RFC6975
+ * makes it very clear that the algorithm data should only be contained in
+ * questions, never in replies. Crappy Belkin copy the OPT data for example,
+ * hence let's detect this so that we downgrade early. */
+ log_debug("OPT RR contained RFC6975 data, ignoring.");
+ bad_opt = true;
+ continue;
}
p->opt = dns_resource_record_ref(rr);
} else {
- /* According to RFC 4795, section
- * 2.9. only the RRs from the Answer
- * section shall be cached. Hence mark
- * only those RRs as cacheable by
- * default, but not the ones from the
- * Additional or Authority
- * sections. */
+ /* According to RFC 4795, section 2.9. only the RRs from the Answer section shall be
+ * cached. Hence mark only those RRs as cacheable by default, but not the ones from the
+ * Additional or Authority sections. */
r = dns_answer_add(answer, rr, p->ifindex,
(i < DNS_PACKET_ANCOUNT(p) ? DNS_ANSWER_CACHEABLE : 0) |
@@ -2116,6 +2182,9 @@ int dns_packet_extract(DnsPacket *p) {
goto finish;
}
}
+
+ if (bad_opt)
+ p->opt = dns_resource_record_unref(p->opt);
}
p->question = question;