Common Implementation Anti-Patterns Related to Domain Name System (DNS) Resource Record (RR) ProcessingForescout TechnologiesJohn F. Kennedylaan, 2Eindhoven5612ABNetherlandsstanislav.dashevskyi@forescout.comForescout TechnologiesJohn F. Kennedylaan, 2Eindhoven5612ABNetherlandsdaniel.dossantos@forescout.comForescout TechnologiesJohn F. Kennedylaan, 2Eindhoven5612ABNetherlandsjos.wetzels@forescout.comForescout TechnologiesJohn F. Kennedylaan, 2Eindhoven5612ABNetherlandsamine.amri@forescout.comvulnerabilitiesvulnerability
This memo describes common vulnerabilities related to Domain Name
System (DNS) resource record (RR) processing as seen in several DNS
client implementations. These vulnerabilities may lead to successful
Denial-of-Service and Remote Code Execution attacks against the
affected software. Where applicable, violations of RFC 1035 are
mentioned.Status of This Memo
This document is not an Internet Standards Track specification; it is
published for informational purposes.
This is a contribution to the RFC Series, independently of any
other RFC stream. The RFC Editor has chosen to publish this
document at its discretion and makes no statement about its value
for implementation or deployment. Documents approved for
publication by the RFC Editor are not candidates for any level of
Internet Standard; see Section 2 of RFC 7841.
Information about the current status of this document, any
errata, and how to provide feedback on it may be obtained at
.
Copyright Notice
Copyright (c) 2022 IETF Trust and the persons identified as the
document authors. All rights reserved.
This document is subject to BCP 78 and the IETF Trust's Legal
Provisions Relating to IETF Documents
() in effect on the date of
publication of this document. Please review these documents
carefully, as they describe your rights and restrictions with
respect to this document.
Table of Contents
. Introduction
. Compression Pointer and Offset Validation
. Label and Name Length Validation
. Null-Terminator Placement Validation
. Response Data Length Validation
. Record Count Validation
. Security Considerations
. IANA Considerations
. References
. Normative References
. Informative References
Acknowledgements
Authors' Addresses
Introduction
Major vulnerabilities in DNS implementations recently became evident and raised attention to this protocol as an important attack vector, as discussed in , , and
, the latter being a set of 7 critical issues affecting the DNS
forwarder "dnsmasq".
The authors of this memo have analyzed the DNS client implementations
of several major TCP/IP protocol stacks and found a set of
vulnerabilities that share common implementation flaws
(anti-patterns). These flaws are related to processing DNS resource records (RRs)
(discussed in ) and may lead to critical security
vulnerabilities.
While implementation flaws may differ from one software project to
another, these anti-patterns are highly likely to span
multiple implementations. In fact, one of the first "Common Vulnerabilities and Exposures" (CVE) documents related to
one of the anti-patterns dates back to the year 2000.
The observations are not limited to DNS client implementations.
Any software that processes DNS RRs may be affected, such as
firewalls, intrusion detection systems, or general-purpose DNS packet
dissectors (e.g., the DNS dissector in Wireshark; see ). Similar issues may
also occur in DNS-over-HTTPS and DNS-over-TLS
implementations. However, any implementation that deals with the DNS
wire format is subject to the considerations discussed in this document. and briefly mention some of these
anti-patterns, but the main purpose of this memo is to provide
technical details behind these anti-patterns, so that the common
mistakes can be eradicated.
We provide general recommendations on mitigating the anti-patterns.
We also suggest that all implementations should drop
malicious/malformed DNS replies and (optionally) log them.Compression Pointer and Offset Validation defines the DNS message compression scheme that can be used
to reduce the size of messages. When it is used, an entire domain
name or several name labels are replaced with a (compression) pointer
to a prior occurrence of the same name.
The compression pointer is a combination of two octets: the two most
significant bits are set to 1, and the remaining 14 bits are the
OFFSET field. This field specifies the offset from the beginning of
the DNS header, at which another domain name or label is located:
+--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+
| 1 1| OFFSET |
+--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+--+
The message compression scheme explicitly allows a domain name to be
represented as one of the following: (1) a sequence of unpacked labels ending with a zero
octet, (2) a pointer, or (3) a sequence of labels ending with a pointer.
However, does not explicitly state that blindly following
compression pointers of any kind can be harmful , as we
could not have had any assumptions about various implementations
that would follow.
Yet, any DNS packet parser that attempts to decompress domain names
without validating the value of OFFSET is likely susceptible to
memory corruption bugs and buffer overruns. These bugs make it easier to perform
Denial-of-Service attacks and may result in successful Remote Code
Execution attacks.
Pseudocode that illustrates a typical example of a broken domain name
parsing implementation is shown below ():
Such implementations typically have a dedicated function for
decompressing domain names (for example, see and
). Among other parameters, these functions may
accept a pointer to the beginning of the first name label within an
RR ("name") and a pointer to the beginning of the DNS payload to be
used as a starting point for the compression pointer
("dns_payload"). The destination buffer for the domain name
("name_buffer") is typically limited to 255 bytes as per
and can be allocated either in the stack or in the heap
memory region.
The code of the function in reads the domain name
label by label from an RR until it reaches the NUL octet ("0x00") that
signifies the end of a domain name. If the current label length octet
("label_len_octet") is a compression pointer, the code extracts the
value of the compression offset and uses it to "jump" to another
label length octet. If the current label length octet is not a
compression pointer, the label bytes will be copied into the name
buffer, and the number of bytes copied will correspond to the value
of the current label length octet. After the copy operation, the code
will move on to the next label length octet.
The first issue with this implementation is due to unchecked
compression offset values. The second issue is due to the absence of
checks that ensure that a pointer will eventually arrive at a
decompressed domain label. We describe these issues in more detail
below. states that a compression pointer is "a pointer to a prior occurance [sic] of the same name." Also, according to ,
the maximum size of DNS packets that can be sent over UDP
is limited to 512 octets.
The pseudocode in violates these constraints, as it will
accept a compression pointer that forces the code to read outside the
bounds of a DNS packet. For instance, a compression pointer set to
"0xffff" will produce an offset of 16383 octets, which is most
definitely pointing to a label length octet somewhere past the bounds of the
original DNS packet. Supplying such offset values will most likely
cause memory corruption issues and may lead to Denial-of-Service
conditions (e.g., a Null pointer dereference after "label_len_octet"
is set to an invalid address in memory). For additional examples,
see , , and .
The pseudocode in allows jumping from a compression
pointer to another compression pointer and does not restrict the
number of such jumps. That is, if a label length octet that is
currently being parsed is a compression pointer, the code will
perform a jump to another label, and if that other label is a
compression pointer as well, the code will perform another jump, and
so forth until it reaches a decompressed label. This may lead to
unforeseen side effects that result in security issues.Consider the DNS packet excerpt illustrated below:
+----+----+----+----+----+----+----+----+----+----+----+----+
+0x00 | ID | FLAGS | QDCOUNT | ANCOUNT | NSCOUNT | ARCOUNT |
+----+----+----+----+----+----+----+----+----+----+----+----+
->+0x0c |0xc0|0x0c| TYPE | CLASS |0x04| t | e | s | t |0x03|
| +----+--|-+----+----+----+----+----+----+----+----+----+----+
| +0x18 | c | o| | m |0x00| TYPE | CLASS | ................ |
| +----+--|-+----+----+----+----+----+----+----+----+----+----+
| |
-----------------
The packet begins with a DNS header at offset +0x00, and its DNS
payload contains several RRs. The first RR begins at an offset of
12 octets (+0x0c); its first label length octet is set to the
value "0xc0", which indicates that it is a compression pointer. The
compression pointer offset is computed from the two octets "0xc00c"
and is equal to 12. Since the broken implementation in
follows this offset value blindly, the pointer will jump back to
the first octet of the first RR (+0x0c) over and over again. The
code in will enter an infinite-loop state, since it will
never leave the "TRUE" branch of the "while" loop.
Apart from achieving infinite loops, the implementation flaws in
make it possible to achieve various pointer loops that have
other undesirable effects. For instance, consider the DNS packet excerpt shown
below:
+----+----+----+----+----+----+----+----+----+----+----+----+
+0x00 | ID | FLAGS | QDCOUNT | ANCOUNT | NSCOUNT | ARCOUNT |
+----+----+----+----+----+----+----+----+----+----+----+----+
->+0x0c |0x04| t | e | s | t |0xc0|0x0c| ...................... |
| +----+----+----+----+----+----+--|-+----+----+----+----+----+
| |
------------------------------------------
With such a domain name, the implementation in will first
copy the domain label at offset "0xc0" ("test"); it will then
fetch the next label length octet, which happens to be a compression pointer
("0xc0"). The compression pointer offset is computed from the two
octets "0xc00c" and is equal to 12 octets. The code will jump back
to offset "0xc0" where the first label "test" is located. The
code will again copy the "test" label and then jump back to it,
following the compression pointer, over and over again. does not contain any logic that restricts multiple jumps
from the same compression pointer and does not ensure that no more
than 255 octets are copied into the name buffer ("name_buffer"). In
fact,
the code will continue to write the label "test" into it,
overwriting the name buffer and the stack of the heap metadata.
attackers would have a significant degree of freedom in
constructing shell code, since they can create arbitrary copy chains
with various combinations of labels and compression pointers.
Therefore, blindly following compression pointers may lead not only to
Denial-of-Service conditions, as pointed out by , but also to successful Remote
Code Execution attacks, as there may be other implementation issues present
within the corresponding code.
Some implementations may not follow , which states:
The first two bits are ones. This allows a pointer to be distinguished
from a label, since the label must begin with two zero bits because
labels are restricted to 63 octets or less. (The 10 and 01 combinations
are reserved for future use.)
Figures and show pseudocode that implements two functions that check whether a given octet is a compression pointer; shows a correct implementation, and shows an incorrect (broken) implementation.
The correct implementation () ensures that the two most
significant bits of an octet are both set, while the broken
implementation () would consider an octet with only one of
the two bits set to be a compression pointer. This is likely an
implementation mistake rather than an intended violation of
, because there are no benefits in supporting such
compression pointer values. The implementations related to
and had a broken
compression pointer check, similar to the code shown in .
While incorrect implementations alone do not lead to vulnerabilities,
they may have unforeseen side effects when combined with other
vulnerabilities. For instance, the first octet of the value "0x4130"
may be incorrectly interpreted as a label length by a broken
implementation. Such a label length (65) is invalid and is larger
than 63 (as per ); a packet that has this value should
be discarded. However, the function shown in will
consider "0x41" to be a valid compression pointer, and the packet
may pass the validation steps.
This might give attackers additional leverage for constructing
payloads and circumventing the existing DNS packet validation
mechanisms.
The first occurrence of a compression pointer in an RR (an octet with
the two highest bits set to 1) must resolve to an octet within a DNS
record with a value that is greater than 0 (i.e., it must not be a
Null-terminator) and less than 64. The offset at which this octet is
located must be smaller than the offset at which the compression
pointer is located; once an implementation makes sure of that,
compression pointer loops can never occur.
In small DNS implementations (e.g., embedded TCP/IP stacks),
support for nested compression pointers (pointers that point to a
compressed name) should be discouraged: there is very little to be
gained in terms of performance versus the high probability of
introducing errors such as those discussed above.
The code that implements domain name parsing should check the offset
with respect to not only the bounds of a packet but also its
position with respect to the compression pointer in question. A
compression pointer must not be "followed" more than once. We have
seen several implementations using a check that ensures that
a compression pointer is not followed more than several times. A
better alternative may be to ensure that the target of a compression
pointer is always located before the location of the pointer in the
packet.Label and Name Length Validation restricts the length of name labels to 63 octets and
lengths of domain names to 255 octets (i.e., label octets and label
length octets). Some implementations do not explicitly enforce these
restrictions.
Consider the function "copy_domain_name()" shown in below.
The function is a variant of the "decompress_domain_name()" function
(), with the difference that it does not support compressed
labels and only copies decompressed labels into the name buffer.
This implementation does not explicitly check for the value of the
label length octet: this value can be up to 255 octets, and a single
label can fill the name buffer. Depending on the memory layout of the
target, how the name buffer is allocated, and the size of the
malformed packet, it is possible to trigger various memory corruption
issues.
Both Figures and restrict the size of the name buffer to 255
octets; however, there are no restrictions on the actual number of
octets that will be copied into this buffer. In this particular case,
a subsequent copy operation (if another label is present in the
packet) will write past the name buffer, allowing heap
or stack metadata to be overwritten in a controlled manner.
Similar examples of vulnerable implementations can be found in the
code relevant to , , and
.
As a general recommendation, a domain label length octet must have a
value of more than 0 and less than 64 . If this is not the case,
an invalid value has been provided within the packet, or a value at an
invalid position might be interpreted as a domain name length due to other
errors in the packet (e.g., misplaced Null-terminator or invalid
compression pointer).
The number of domain label characters must correspond to the value of
the domain label octet. To avoid possible errors when interpreting
the characters of a domain label, developers may consider
recommendations for the preferred domain name syntax outlined in
.
The domain name length must not be more than 255 octets, including
the size of decompressed domain names. The NUL octet ("0x00") must
be present at the end of the domain name and must be within the maximum name
length (255 octets).Null-Terminator Placement Validation
A domain name must end with a NUL ("0x00") octet, as per .
The implementations shown in Figures and assume that this is the
case for the RRs that they process; however, names that do not have a
NUL octet placed at the proper position within an RR are not
discarded.
This issue is closely related to the absence of label and name length
checks. For example, the logic behind Figures and will continue
to copy octets into the name buffer until a NUL octet is
encountered. This octet can be placed at an arbitrary position
within an RR or not placed at all.
Consider the pseudocode function shown in . The function
returns the length of a domain name ("name") in octets to be used
elsewhere (e.g., to allocate a name buffer of a certain size): for
compressed domain names, the function returns 2; for decompressed
names, it returns their true length using the "strlen(3)" function.
"strlen(3)" is a standard C library function that returns the length
of a given sequence of characters terminated by the NUL ("0x00")
octet. Since this function also expects names to be explicitly
Null-terminated, the return value "strlen(3)" may also be controlled
by attackers. Through the value of "name_len", attackers may control
the allocation of internal buffers or specify the number by octets
copied into these buffers, or they may perform other operations, depending on the
implementation specifics.
The absence of explicit checks for placement of the NUL octet may also
facilitate controlled memory reads and writes. An example of
vulnerable implementations can be found in the code relevant to
, , , and
.
As a general recommendation for mitigating such issues, developers
should never trust user data to be Null-terminated. For example, to
fix/mitigate the issue shown in the code in , developers should use
the function "strnlen(3)", which reads at most X characters (the second
argument of the function), and ensure that X is not larger than the
buffer allocated for the name.Response Data Length Validation
As stated in , every RR contains a variable-length string of
octets that contains the retrieved resource data (RDATA) (e.g., an IP
address that corresponds to a domain name in question). The length of
the RDATA field is regulated by the resource data length field
(RDLENGTH), which is also present in an RR.
Implementations that process RRs may not check for the validity of
the RDLENGTH field value when retrieving RDATA. Failing to do so may
lead to out-of-bound read issues, whose impact may
vary significantly, depending on the implementation specifics. We have
observed instances of Denial-of-Service conditions and information
leaks.
Therefore, the value of the data length byte in response DNS records
(RDLENGTH) must reflect the number of bytes available in the field
that describes the resource (RDATA). The format of RDATA must
conform to the TYPE and CLASS fields of the RR.
Examples of vulnerable implementations can be found in the code
relevant to , , and .Record Count Validation
According to , the DNS header contains four two-octet
fields that specify the amount of question records (QDCOUNT), answer
records (ANCOUNT), authority records (NSCOUNT), and additional
records (ARCOUNT). illustrates a recurring implementation anti-pattern for a
function that processes DNS RRs. The function "process_dns_records()"
extracts the value of ANCOUNT ("num_answers") and the pointer to the
DNS data payload ("data_ptr"). The function processes answer records
in a loop, decrementing the "num_answers" value after processing each
record until the value of "num_answers" becomes zero. For
simplicity, we assume that there is only one domain name per answer.
Inside the loop, the code calculates the domain name length
("name_length") and adjusts the data payload pointer ("data_ptr") by the
offset that corresponds to "name_length + 1", so that the pointer
lands on the first answer record. Next, the answer record is
retrieved and processed, and the "num_answers" value is decremented.
If the ANCOUNT number retrieved from the header
("dns_header->ancount") is not checked against the amount of data
available in the packet and it is, for example, larger than the number of
answer records available, the data pointer ("data_ptr") will read outside
the bounds of the packet. This may result in Denial-of-Service
conditions.
In this section, we used an example of processing answer records.
However, the same logic is often reused for implementing the
processing of other types of records, e.g., the number of question
(QDCOUNT), authority (NSCOUNT), and additional (ARCOUNT) records. The
specified numbers of these records must correspond to the actual data
present within the packet. Therefore, all record count fields must
be checked before fully parsing the contents of a packet.
Specifically, recommends that such malformed
DNS packets should be dropped and (optionally) logged.
Examples of vulnerable implementations can be found in the code
relevant to , , , and
.Security Considerations
Security issues are discussed throughout this memo; it
discusses implementation flaws (anti-patterns) that affect the
functionality of processing DNS RRs. The presence of such
anti-patterns leads to bugs that cause buffer overflows,
read-out-of-bounds, and infinite-loop issues. These issues have the
following security impacts: information leaks, Denial-of-Service attacks, and
Remote Code Execution attacks.
This document lists general recommendations for the developers of DNS
record parsing functionality that allow those developers to prevent such
implementation flaws, e.g., by rigorously checking the data received
over the wire before processing it.IANA Considerations
This document has no IANA actions. Please see
for a complete review of the IANA considerations
introduced by DNS.ReferencesNormative ReferencesDomain names - implementation and specificationThis RFC is the revised specification of the protocol and format used in the implementation of the Domain Name System. It obsoletes RFC-883. This memo documents the details of the domain name client - server communication.DNS Proxy Implementation GuidelinesThis document provides guidelines for the implementation of DNS proxies, as found in broadband gateways and other similar network devices. This document specifies an Internet Best Current Practices for the Internet Community, and requests discussion and suggestions for improvements.Informative ReferencesCVE-2000-0333: A denial-of-service vulnerability in tcpdump, Ethereal, and other sniffer packages via malformed DNS packetsCommon Vulnerabilities and ExposuresCVE-2017-9345: An infinite loop in the DNS dissector of WiresharkCommon Vulnerabilities and ExposuresCVE-2020-15795: A denial-of-service and remote code execution vulnerability DNS domain name label parsing functionality of Nucleus NETCommon Vulnerabilities and ExposuresCVE-2020-17440 A denial-of-service vulnerability in the DNS name parsing implementation of uIPCommon Vulnerabilities and ExposuresCVE-2020-24334: An out-of-bounds read and denial-of-service vulnerability in the DNS response parsing functionality of uIPCommon Vulnerabilities and ExposuresCVE-2020-24335: A memory corruption vulnerability in domain name parsing routines of uIPCommon Vulnerabilities and ExposuresCVE-2020-24336: A buffer overflow vulnerability in the DNS implementation of Contiki and Contiki-NGCommon Vulnerabilities and ExposuresCVE-2020-24338: A denial-of-service and remote code execution vulnerability in the DNS domain name record decompression functionality of picoTCPCommon Vulnerabilities and ExposuresCVE-2020-24339: An out-of-bounds read and denial-of-service vulnerability in the DNS domain name record decompression functionality of picoTCPCommon Vulnerabilities and ExposuresCVE-2020-24340: An out-of-bounds read and denial-of-service vulnerability in the DNS response parsing functionality of picoTCPCommon Vulnerabilities and ExposuresCVE-2020-24383: An information leak and denial-of-service vulnerability while parsing mDNS resource records in FNETCommon Vulnerabilities and ExposuresCVE-2020-25107: A denial-of-service and remote code execution vulnerability in the DNS implementation of Ethernut Nut/OSCommon Vulnerabilities and ExposuresCVE-2020-25108: A denial-of-service and remote code execution vulnerability in the DNS implementation of Ethernut Nut/OSCommon Vulnerabilities and ExposuresCVE-2020-25109: A denial-of-service and remote code execution vulnerability in the DNS implementation of Ethernut Nut/OSCommon Vulnerabilities and ExposuresCVE-2020-25110: A denial-of-service and remote code execution vulnerability in the DNS implementation of Ethernut Nut/OSCommon Vulnerabilities and ExposuresCVE-2020-25767: An out-of-bounds read and denial-of-service vulnerability in the DNS name parsing routine of HCC Embedded NicheStackCommon Vulnerabilities and ExposuresCVE-2020-27009: A denial-of-service and remote code execution vulnerability DNS domain name record decompression functionality of Nucleus NETCommon Vulnerabilities and ExposuresCVE-2020-27736: An information leak and denial-of-service vulnerability in the DNS name parsing functionality of Nucleus NETCommon Vulnerabilities and ExposuresCVE-2020-27737: An information leak and denial-of-service vulnerability in the DNS response parsing functionality of Nucleus NETCommon Vulnerabilities and ExposuresCVE-2020-27738: A denial-of-service and remote code execution vulnerability DNS domain name record decompression functionality of Nucleus NETCommon Vulnerabilities and ExposuresA New Scheme for the Compression of Domain NamesUniversitaet BielefeldThe compression of domain names in DNS messages was introduced in
[RFC1035]. Although some remarks were made about applicability to
future defined resource record types, no method has been deployed yet
to support interoperable DNS compression for RR types specified since
then.
Work in ProgressDNSpooq: Cache Poisoning and RCE in Popular DNS Forwarder dnsmasqJSOF Technical ReportDomain Name System (DNS) IANA ConsiderationsThis document specifies Internet Assigned Numbers Authority (IANA) parameter assignment considerations for the allocation of Domain Name System (DNS) resource record types, CLASSes, operation codes, error codes, DNS protocol message header bits, and AFSDB resource record subtypes. It obsoletes RFC 6195 and updates RFCs 1183, 2845, 2930, and 3597.Specification for DNS over Transport Layer Security (TLS)This document describes the use of Transport Layer Security (TLS) to provide privacy for DNS. Encryption provided by TLS eliminates opportunities for eavesdropping and on-path tampering with DNS queries in the network, such as discussed in RFC 7626. In addition, this document specifies two usage profiles for DNS over TLS and provides advice on performance considerations to minimize overhead from using TCP and TLS with DNS.This document focuses on securing stub-to-recursive traffic, as per the charter of the DPRIVE Working Group. It does not prevent future applications of the protocol to recursive-to-authoritative traffic.DNS Queries over HTTPS (DoH)This document defines a protocol for sending DNS queries and getting DNS responses over HTTPS. Each DNS query-response pair is mapped into an HTTP exchange.DNS Cache Poisoning Attack Reloaded: Revolutions with Side ChannelsProc. 2020 ACM SIGSAC Conference on Computer and Communications Security, CCS '20CVE-2020-1350: A remote code execution vulnerability in Windows Domain Name System serversCommon Vulnerabilities and ExposuresAcknowledgements
We would like to thank , who has greatly contributed to
the research that led to the creation of this document.Authors' AddressesForescout TechnologiesJohn F. Kennedylaan, 2Eindhoven5612ABNetherlandsstanislav.dashevskyi@forescout.comForescout TechnologiesJohn F. Kennedylaan, 2Eindhoven5612ABNetherlandsdaniel.dossantos@forescout.comForescout TechnologiesJohn F. Kennedylaan, 2Eindhoven5612ABNetherlandsjos.wetzels@forescout.comForescout TechnologiesJohn F. Kennedylaan, 2Eindhoven5612ABNetherlandsamine.amri@forescout.com