5. Attacks That Can Be Mitigated with End-to-End Security
This section describes the RDMAP/DDP attacks where the only solution
is to implement some form of end-to-end security. The analysis
includes a detailed description of each attack, what is being
attacked, and a description of the countermeasures that can be taken
to thwart the attack.
Some forms of attack involve modifying the RDMAP or DDP payload by a
network-based attacker or involve monitoring the traffic to discover
private information. An effective tool to ensure confidentiality is
to encrypt the data stream through mechanisms, such as IPsec
encryption. Additionally, authentication protocols, such as IPsec
authentication, are an effective tool to ensure the remote entity is
who they claim to be, as well as ensuring that the payload is
unmodified as it traverses the network.
Note that connection setup and tear down is presumed to be done in
stream mode (i.e., no RDMA encapsulation of the payload), so there
are no new attacks related to connection setup/tear down beyond what
is already present in the LLP (e.g., TCP or SCTP). Note, however,
that RDMAP/DDP parameters may be exchanged in stream mode, and if
they are corrupted by an attacker unintended consequences will
result. Therefore, any existing mitigations for LLP Spoofing,
Tampering, Repudiation, Information Disclosure, Denial of Service, or
Elevation of Privilege continue to apply (and are out of scope of
this document). Thus, the analysis in this section focuses on
attacks that are present, regardless of the LLP Stream type.
Tampering is any modification of the legitimate traffic (machine
internal or network). Spoofing attack is a special case of tampering
where the attacker falsifies an identity of the Remote Peer (identity
can be an IP address, machine name, ULP level identity, etc.).
Spoofing attacks can be launched by the Remote Peer, or by a
network-based attacker. A network-based spoofing attack applies to
all Remote Peers. This section analyzes the various types of
spoofing attacks applicable to RDMAP and DDP.
A network-based attacker can impersonate a legal RDMAP/DDP Peer (by
spoofing a legal IP address). This can either be done as a blind
attack (see [RFC3552]) or by establishing an RDMAP/DDP Stream with
the victim. Because an RDMAP/DDP Stream requires an LLP Stream to be
fully initialized (e.g., for [RFC793], it is in the ESTABLISHED
state), existing transport layer protection mechanisms against blind
attacks remain in place.
For a blind attack to succeed, it requires the attacker to inject a
valid transport layer segment (e.g., for TCP, it must match at least
the 4-tuple as well as guess a sequence number within the window)
while also guessing valid RDMAP or DDP parameters. There are many
ways to attack the RDMAP/DDP protocol if the transport protocol is
assumed to be vulnerable. For example, for Tagged Messages, this
entails guessing the STag and TO values. If the attacker wishes to
simply terminate the connection, it can do so by correctly guessing
the transport and network layer values, and providing an invalid
STag. Per the DDP specification, if an invalid STag is received, the
Stream is torn down and the Remote Peer is notified with an error.
If an attacker wishes to overwrite an Advertised Buffer, it must
successfully guess the correct STag and TO. Given that the TO will
often start at zero, this is straightforward. The value of the STag
should be chosen at random, as discussed in Section 6.1.1, Using an
STag on a Different Stream. For Untagged Messages, if the MSN is
invalid then the connection may be torn down. If it is valid, then
the receive buffers can be corrupted.
End-to-end authentication (e.g., IPsec or ULP authentication)
provides protection against either the blind attack or the connected
5.1.2. Stream Hijacking
Stream hijacking happens when a network-based attacker eavesdrops on
the LLP connection through the Stream establishment phase, and waits
until the authentication phase (if such a phase exists) is completed
successfully. The attacker then spoofs the IP address and re-directs
the Stream from the victim to its own machine. For example, an
attacker can wait until an iSCSI authentication is completed
successfully, and then hijack the iSCSI Stream.
The best protection against this form of attack is end-to-end
integrity protection and authentication, such as IPsec, to prevent
spoofing. Another option is to provide a physically segregated
network for security. Discussion of physical security is out of
scope for this document.
Because the connection and/or Stream itself is established by the
LLP, some LLPs are more difficult to hijack than others. Please see
the relevant LLP documentation on security issues around connection
and/or Stream hijacking.
5.1.3. Man-in-the-Middle Attack
If a network-based attacker has the ability to delete or modify
packets that will still be accepted by the LLP (e.g., TCP sequence
number is correct), then the Stream can be exposed to a man-in-the-
middle attack. One style of attack is for the man-in-the-middle to
send Tagged Messages (either RDMAP or DDP). If it can discover a
buffer that has been exposed for STag enabled access, then the man-
in-the-middle can use an RDMA Read operation to read the contents of
the associated Data Buffer, perform an RDMA Write Operation to modify
the contents of the associated Data Buffer, or invalidate the STag to
disable further access to the buffer.
The best protection against this form of attack is end-to-end
integrity protection and authentication, such as IPsec, to prevent
spoofing or tampering. If authentication and integrity protections
are not used, then physical protection must be employed to prevent
Because the connection/Stream itself is established by the LLP, some
LLPs are more exposed to man-in-the-middle attack than others.
Please see the relevant LLP documentation on security issues around
connection and/or Stream hijacking.
Another approach is to restrict access to only the local subnet/link,
and provide some mechanism to limit access, such as physical security
or 802.1.x. This model is an extremely limited deployment scenario,
and will not be further examined here.
5.2. Tampering - Network-Based Modification of Buffer Content
This is actually a man-in-the-middle attack, but only on the content
of the buffer, as opposed to the man-in-the-middle attack presented
above, where both the signaling and content can be modified. See
Section 5.1.3, Man-in-the-Middle Attack.
5.3. Information Disclosure - Network-Based Eavesdropping
An attacker that is able to eavesdrop on the network can read the
content of all read and write accesses to a Peer's buffers. To
prevent information disclosure, the read/written data must be
encrypted. See also Section 5.1.3, Man-in-the-Middle Attack. The
encryption can be done either by the ULP, or by a protocol that can
provide security services to RDMAP and DDP (e.g., IPsec).
5.4. Specific Requirements for Security Services
Generally speaking, Stream confidentiality protects against
eavesdropping. Stream and/or session authentication and integrity
protection is a counter measurement against various spoofing and
tampering attacks. The effectiveness of authentication and integrity
against a specific attack depends on whether the authentication is
machine level authentication (such as IPsec), or ULP authentication.
5.4.1. Introduction to Security Options
The following security services can be applied to an RDMAP/DDP
1. Session confidentiality - Protects against eavesdropping (Section
2. Per-packet data source authentication - Protects against the
following spoofing attacks: network-based impersonation (Section
5.1.1) and Stream hijacking (Section 5.1.2).
3. Per-packet integrity - Protects against tampering done by
network-based modification of buffer content (Section 5.2) and
when combined with authentication, also protects against man-in-
the-middle attacks (Section 5.1.3).
4. Packet sequencing - protects against replay attacks, which is a
special case of the above tampering attack.
If an RDMAP/DDP Stream may be subject to impersonation attacks, or
Stream hijacking attacks, it is recommended that the Stream be
authenticated, integrity protected, and protected from replay
attacks; it may use confidentiality protection to protect from
eavesdropping (in case the RDMAP/DDP Stream traverses a public
IPsec is a protocol suite that is used to secure communication at the
network layer between two peers. The IPsec protocol suite is
specified within the IP Security Architecture [RFC2401], IKE
[RFC2409], IPsec Authentication Header (AH) [RFC2402], and IPsec
Encapsulating Security Payload (ESP) [RFC2406] documents. IKE is the
key management protocol, while AH and ESP are used to protect IP
traffic. Please see those RFCs for a complete description of the
IPsec is capable of providing the above security services for IP and
TCP traffic, respectively. ULP protocols are able to provide only
part of the above security services.
5.4.2. TLS Is Inappropriate for DDP/RDMAP Security
TLS [RFC4346] provides Stream authentication, integrity and
confidentiality for TCP based ULPs. TLS supports one-way (server
only) or mutual certificates based authentication.
If TLS is layered underneath RDMAP, TLS's connection orientation
makes TLS inappropriate for DDP/RDMA security. If a stream cipher or
block cipher in CBC mode is used for bulk encryption, then a packet
can be decrypted only after all the packets preceding it have already
arrived. If TLS is used to protect DDP/RDMAP traffic, then TCP must
gather all out-of-order packets before TLS can decrypt them. Only
after this is done can RDMAP/DDP place them into the ULP buffer.
Thus, one of the primary features of DDP/RDMAP - enabling
implementations to have a flow-through architecture with little to no
buffering - cannot be achieved if TLS is used to protect the data
If TLS is layered on top of RDMAP or DDP, TLS does not protect the
RDMAP and/or DDP headers. Thus, a man-in-the-middle attack can still
occur by modifying the RDMAP/DDP header to place the data into the
wrong buffer, thus effectively corrupting the data stream.
For these reasons, it is not RECOMMENDED that TLS be layered on top
of RDMAP or DDP.
5.4.3. DTLS and RDDP
DTLS [DTLS] provides security services for datagram protocols,
including unreliable datagram protocols. These services include
anti-replay based on a mechanism adapted from IPsec that is intended
to operate on packets as they are received from the network. For
these and other reasons, DTLS is best applied to RDDP by employing
DTLS beneath TCP, yielding a layering of RDDP over TCP over DTLS over
UDP/IP. Such a layering inserts DTLS at roughly the same level in
the protocol stack as IPsec, making DTLS's security services an
alternative to IPsec's services from an RDDP standpoint.
For RDDP, IPsec is the better choice for a security framework, and
hence is mandatory-to-implement (as specified elsewhere in this
document). An important contributing factor to the specification of
IPsec rather than DTLS is that the non-RDDP versions of two initial
adopters of RDDP (iSCSI [iSCSI][iSER] and NFSv4 [NFSv4][NFSv4.1]) are
compatible with IPsec but neither of these protocols currently uses
either TLS or DTLS. For the specific case of iSCSI, IPsec is the
basis for mandatory-to-implement security services [RFC3723].
Therefore, this document and the RDDP protocol specifications contain
mandatory implementation requirements for IPsec rather than for DTLS.
5.4.4. ULPs That Provide Security
ULPs that provide integrated security but wish to leverage lower-
layer protocol security, should be aware of security concerns around
correlating a specific channel's security mechanisms to the
authentication performed by the ULP. See [NFSv4CHANNEL] for
additional information on a promising approach called "channel
binding". From [NFSv4CHANNEL]:
"The concept of channel bindings allows applications to prove that
the end-points of two secure channels at different network layers
are the same by binding authentication at one channel to the
session protection at the other channel. The use of channel
bindings allows applications to delegate session protection to
lower layers, which may significantly improve performance for some
5.4.5. Requirements for IPsec Encapsulation of DDP
The IP Storage working group has spent significant time and effort to
define the normative IPsec requirements for IP Storage [RFC3723].
Portions of that specification are applicable to a wide variety of
protocols, including the RDDP protocol suite. In order not to
replicate this effort, an RNIC implementation MUST follow the
requirements defined in RFC 3723, Section 2.3 and Section 5,
including the associated normative references for those sections.
Note that this means that support for IPSEC ESP mode is normative.
Additionally, since IPsec acceleration hardware may only be able to
handle a limited number of active IKE Phase 2 SAs, Phase 2 delete
messages may be sent for idle SAs as a means of keeping the number of
active Phase 2 SAs to a minimum. The receipt of an IKE Phase 2
delete message MUST NOT be interpreted as a reason for tearing down a
DDP/RDMA Stream. Rather, it is preferable to leave the Stream up,
and if additional traffic is sent on it, to bring up another IKE
Phase 2 SA to protect it. This avoids the potential for continually
bringing Streams up and down.
Note that there are serious security issues if IPsec is not
implemented end-to-end. For example, if IPsec is implemented as a
tunnel in the middle of the network, any hosts between the Peer and
the IPsec tunneling device can freely attack the unprotected Stream.
The IPsec requirements for RDDP are based on the version of IPsec
specified in RFC 2401 [RFC2401] and related RFCs, as profiled by RFC
3723 [RFC3723], despite the existence of a newer version of IPsec
specified in RFC 4301 [RFC4301] and related RFCs. One of the
important early applications of the RDDP protocols is their use with
iSCSI [iSER]; RDDP's IPsec requirements follow those of IPsec in
order to facilitate that usage by allowing a common profile of IPsec
to be used with iSCSI and the RDDP protocols. In the future, RFC
3723 may be updated to the newer version of IPsec; the IPsec security
requirements of any such update should apply uniformly to iSCSI and
the RDDP protocols.
6. Attacks from Remote Peers
This section describes remote attacks that are possible against the
RDMA system defined in Figure 1 - RDMA Security Model and the RNIC
Engine resources defined in Section 2.2. The analysis includes a
detailed description of each attack, what is being attacked, and a
description of the countermeasures that can be taken to thwart the
The attacks are classified into five categories: Spoofing, Tampering,
Information Disclosure, Denial of Service (DoS) attacks, and
Elevation of Privileges. As mentioned previously, tampering is any
modification of the legitimate traffic (machine internal or network).
A spoofing attack is a special case of tampering where the attacker
falsifies an identity of the Remote Peer (identity can be an IP
address, machine name, ULP level identity, etc.).
This section analyzes the various types of spoofing attacks
applicable to RDMAP and DDP. Spoofing attacks can be launched by the
Remote Peer or by a network-based attacker. For countermeasures
against a network-based attacker, see Section 5, Attacks That Can Be
Mitigated with End-to-End Security.
6.1.1. Using an STag on a Different Stream
One style of attack from the Remote Peer is for it to attempt to use
STag values that it is not authorized to use. Note that if the
Remote Peer sends an invalid STag to the Local Peer, per the DDP and
RDMAP specifications, the Stream must be torn down. Thus, the threat
exists if an STag has been enabled for Remote Access on one Stream
and a Remote Peer is able to use it on an unrelated Stream. If the
attack is successful, the attacker could potentially be able to
either perform RDMA Read operations to read the contents of the
associated Data Buffer, perform RDMA Write operations to modify the
contents of the associated data buffer, or invalidate the STag to
disable further access to the buffer.
An attempt by a Remote Peer to access a buffer with an STag on a
different Stream in the same Protection Domain may or may not be an
attack, depending on whether resource sharing is intended (i.e.,
whether the Streams shared Partial Mutual Trust). For some ULPs,
using an STag on multiple Streams within the same Protection Domain
could be desired behavior. For other ULPs, attempting to use an STag
on a different Stream could be considered an attack. Since this
varies by ULP, a ULP typically would need to be able to control the
scope of the STag.
In the case where an implementation does not share resources between
Streams (including STags), this attack can be defeated by assigning
each Stream to a different Protection Domain. Before allowing remote
access to the buffer, the Protection Domain of the Stream where the
access attempt was made is matched against the Protection Domain of
the STag. If the Protection Domains do not match, access to the
buffer is denied, an error is generated, and the RDMAP Stream
associated with the attacking Stream is terminated.
For implementations that share resources between multiple Streams, it
may not be practical to separate each Stream into its own Protection
Domain. In this case, the ULP can still limit the scope of any of
the STags to a single Stream (if it is enabling it for remote
access). If the STag scope has been limited to a single Stream, any
attempt to use that STag on a different Stream will result in an
error, and the RDMAP Stream is terminated.
Thus, for implementations that do not share STags between Streams,
each Stream MUST either be in a separate Protection Domain or the
scope of an STag MUST be limited to a single Stream.
An RNIC MUST ensure that a specific Stream in a specific Protection
Domain cannot access an STag in a different Protection Domain.
An RNIC MUST ensure that, if an STag is limited in scope to a single
Stream, no other Stream can use the STag.
An additional issue may be unintended sharing of STags (i.e., a bug
in the ULP) or a bug in the Remote Peer that causes an off-by-one
STag to be used. For additional protection, an implementation should
allocate STags in such a fashion that it is difficult to predict the
next allocated STag number, and also ensure that STags are reused at
as slow a rate as possible. Any allocation method that would lead to
intentional or unintentional reuse of an STag by the peer should be
avoided (e.g., a method that always starts with a given STag and
monotonically increases it for each new allocation, or a method that
always uses the same STag for each operation).
A Remote Peer or a network-based attacker can attempt to tamper with
the contents of Data Buffers on a Local Peer that have been enabled
for remote write access. The types of tampering attacks from a
Remote Peer are outlined in the sections that follow. For
countermeasures against a network-based attacker, see Section 5,
Attacks That Can Be Mitigated with End-to-End Security.
6.2.1. Buffer Overrun - RDMA Write or Read Response
This attack is an attempt by the Remote Peer to perform an RDMA Write
or RDMA Read Response to memory outside of the valid length range of
the Data Buffer enabled for remote write access. This attack can
occur even when no resources are shared across Streams. This issue
can also arise if the ULP has a bug.
The countermeasure for this type of attack must be in the RNIC
implementation, leveraging the STag. When the local ULP specifies to
the RNIC the base address and the umber of bytes in the buffer that
it wishes to make accessible, the RNIC must ensure that the base and
bounds check are applied to any access to the buffer referenced by
the STag before the STag is enabled for access. When an RDMA data
transfer operation (which includes an STag) arrives on a Stream, a
base and bounds byte granularity access check must be performed to
ensure that the operation accesses only memory locations within the
buffer described by that STag.
Thus an RNIC implementation MUST ensure that a Remote Peer is not
able to access memory outside of the buffer specified when the STag
was enabled for remote access.
6.2.2. Modifying a Buffer after Indication
This attack can occur if a Remote Peer attempts to modify the
contents of an STag referenced buffer by performing an RDMA Write or
an RDMA Read Response after the Remote Peer has indicated to the
Local Peer or local ULP (by a variety of means) that the STag Data
Buffer contents are ready for use. This attack can occur even when
no resources are shared across Streams. Note that a bug in a Remote
Peer, or network-based tampering, could also result in this problem.
For example, assume that the STag referenced buffer contains ULP
control information as well as ULP payload, and the ULP sequence of
operation is to first validate the control information and then
perform operations on the control information. If the Remote Peer
can perform an additional RDMA Write or RDMA Read Response (thus,
changing the buffer) after the validity checks have been completed
but before the control data is operated on, the Remote Peer could
force the ULP down operational paths that were never intended.
The local ULP can protect itself from this type of attack by revoking
remote access when the original data transfer has completed and
before it validates the contents of the buffer. The local ULP can do
this either by explicitly revoking remote access rights for the STag
when the Remote Peer indicates the operation has completed, or by
checking to make sure the Remote Peer invalidated the STag through
the RDMAP Remote Invalidate capability. If the Remote Peer did not
invalidate the STag, the local ULP then explicitly revokes the STag
remote access rights. (See Section 6.4.5, Remote Invalidate an STag
Shared on Multiple Streams for a definition of Remote Invalidate.)
The local ULP SHOULD follow the above procedure to protect the buffer
before it validates the contents of the buffer (or uses the buffer in
An RNIC MUST ensure that network packets using the STag for a
previously Advertised Buffer can no longer modify the buffer after
the ULP revokes remote access rights for the specific STag.
6.2.3. Multiple STags to Access the Same Buffer
See Section 6.3.6 Using Multiple STags That Alias to the Same Buffer,
for this analysis.
6.3. Information Disclosure
The main potential source for information disclosure is through a
local buffer that has been enabled for remote access. If the buffer
can be probed by a Remote Peer on another Stream, then there is
potential for information disclosure.
The potential attacks that could result in unintended information
disclosure and countermeasures are detailed in the following
6.3.1. Probing Memory Outside of the Buffer Bounds
This is essentially the same attack as described in Section 6.2.1,
Buffer Overrun - RDMA Write or Read Response, except that an RDMA
Read Request is used to mount the attack. The same countermeasure
6.3.2. Using RDMA Read to Access Stale Data
If a buffer is being used for some combination of reads and writes
(either remote or local), and is exposed to a Remote Peer with at
least remote read access rights before it is initialized with the
correct data, there is a potential race condition where the Remote
Peer can view the prior contents of the buffer. This becomes a
security issue if the prior contents of the buffer were not intended
to be shared with the Remote Peer.
To eliminate this race condition, the local ULP SHOULD ensure that no
stale data is contained in the buffer before remote read access
rights are granted (this can be done by zeroing the contents of the
memory, for example). This ensures that the Remote Peer cannot
access the buffer until the stale data has been removed.
6.3.3. Accessing a Buffer after the Transfer
If the Remote Peer has remote read access to a buffer and, by some
mechanism, tells the local ULP that the transfer has been completed,
but the local ULP does not disable remote access to the buffer before
modifying the data, it is possible for the Remote Peer to retrieve
the new data.
This is similar to the attack defined in Section 6.2.2, Modifying a
Buffer after Indication. The same countermeasures apply. In
addition, the local ULP SHOULD grant remote read access rights only
for the amount of time needed to retrieve the data.
6.3.4. Accessing Unintended Data with a Valid STag
If the ULP enables remote access to a buffer using an STag that
references the entire buffer, but intends only a portion of the
buffer to be accessed, it is possible for the Remote Peer to access
the other parts of the buffer anyway.
To prevent this attack, the ULP SHOULD set the base and bounds of the
buffer when the STag is initialized to expose only the data to be
6.3.5. RDMA Read into an RDMA Write Buffer
One form of disclosure can occur if the access rights on the buffer
enabled remote read, when only remote write access was intended. If
the buffer contained ULP data, or data from a transfer on an
unrelated Stream, the Remote Peer could retrieve the data through an
RDMA Read operation. Note that an RNIC implementation is not
required to support STags that have both read and write access.
The most obvious countermeasure for this attack is to not grant
remote read access if the buffer is intended to be write-only. Then
the Remote Peer would not be able to retrieve data associated with
the buffer. An attempt to do so would result in an error and the
RDMAP Stream associated with the Stream would be terminated.
Thus, if a ULP only intends a buffer to be exposed for remote write
access, it MUST set the access rights to the buffer to only enable
remote write access. Note that this requirement is not meant to
restrict the use of zero-length RDMA Reads. Zero-length RDMA Reads
do not expose ULP data. Because they are intended to be used as a
mechanism to ensure that all RDMA Writes have been received, and do
not even require a valid STag, their use is permitted even if a
buffer has only been enabled for write access.
6.3.6. Using Multiple STags That Alias to the Same Buffer
Multiple STags that alias to the same buffer at the same time can
result in unintentional information disclosure if the STags are used
by different, mutually untrusted Remote Peers. This model applies
specifically to client/server communication, where the server is
communicating with multiple clients, each of which do not mutually
trust each other.
If only read access is enabled, then the local ULP has complete
control over information disclosure. Thus, a server that intended to
expose the same data (i.e., buffer) to multiple clients by using
multiple STags to the same buffer creates no new security issues
beyond what has already been described in this document. Note that
if the server did not intend to expose the same data to the clients,
it should use separate buffers for each client (and separate STags).
When one STag has remote read access enabled and a different STag has
remote write access enabled to the same buffer, it is possible for
one Remote Peer to view the contents that have been written by
another Remote Peer.
If both STags have remote write access enabled and the two Remote
Peers do not mutually trust each other, it is possible for one Remote
Peer to overwrite the contents that have been written by the other
Thus, a ULP with multiple Remote Peers that do not share Partial
Mutual Trust MUST NOT grant write access to the same buffer through
different STags. A buffer should be exposed to only one untrusted
Remote Peer at a time to ensure that no information disclosure or
information tampering occurs between peers.
6.4. Denial of Service (DOS)
A DOS attack is one of the primary security risks of RDMAP. This is
because RNIC resources are valuable and scarce, and many ULP
environments require communication with untrusted Remote Peers. If
the Remote Peer can be authenticated or the ULP payload encrypted,
clearly, the DOS profile can be reduced. For the purposes of this
analysis, it is assumed that the RNIC must be able to operate in
untrusted environments, which are open to DOS-style attacks.
Denial of service attacks against RNIC resources are not the typical
unknown party spraying packets at a random host (such as a TCP SYN
attack). Because the connection/Stream must be fully established
(e.g., a 3-message transport layer handshake has occurred), the
attacker must be able to both send and receive messages over that
connection/Stream, or be able to guess a valid packet on an existing
This section outlines the potential attacks and the countermeasures
available for dealing with each attack.
6.4.1. RNIC Resource Consumption
This section covers attacks that fall into the general category of a
local ULP attempting to unfairly allocate scarce (i.e., bounded) RNIC
resources. The local ULP may be attempting to allocate resources on
its own behalf, or on behalf of a Remote Peer. Resources that fall
into this category include Protection Domains, Stream Context Memory,
Translation and Protection Tables, and STag namespace. These can be
due to attacks by currently active local ULPs or ones that allocated
resources earlier but are now idle.
This type of attack can occur regardless of whether resources are
shared across Streams.
The allocation of all scarce resources MUST be placed under the
control of a Privileged Resource Manager. This allows the Privileged
Resource Manager to:
* prevent a local ULP from allocating more than its fair share of
* detect if a Remote Peer is attempting to launch a DOS attack by
attempting to create an excessive number of Streams (with
associated resources) and take corrective action (such as
refusing the request or applying network layer filters against
the Remote Peer).
This analysis assumes that the Resource Manager is responsible for
handing out Protection Domains, and that RNIC implementations will
provide enough Protection Domains to allow the Resource Manager to be
able to assign a unique Protection Domain for each unrelated,
untrusted local ULP (for a bounded, reasonable number of local ULPs).
This analysis further assumes that the Resource Manager implements
policies to ensure that untrusted local ULPs are not able to consume
all the Protection Domains through a DOS attack. Note that
Protection Domain consumption cannot result from a DOS attack
launched by a Remote Peer, unless a local ULP is acting on the Remote
6.4.2. Resource Consumption by Idle ULPs
The simplest form of a DOS attack, given a fixed amount of resources,
is for the Remote Peer to create an RDMAP Stream to a Local Peer,
request dedicated resources, and then do no actual work. This allows
the Remote Peer to be very light weight (i.e., only negotiate
resources, but do no data transfer) and consumes a disproportionate
amount of resources at the Local Peer.
A general countermeasure for this style of attack is to monitor
active RDMAP Streams and, if resources are getting low, to reap the
resources from RDMAP Streams that are not transferring data and
possibly terminate the Stream. This would presumably be under
Refer to Section 6.4.1 for the analysis and countermeasures for this
style of attack on the following RNIC resources: Stream Context
Memory, Page Translation Tables, and STag namespace.
Note that some RNIC resources are not at risk of this type of attack
from a Remote Peer because an attack requires the Remote Peer to send
messages in order to consume the resource. Receive Data Buffers,
Completion Queue, and RDMA Read Request Queue resources are examples.
These resources are, however, at risk from a local ULP that attempts
to allocate resources, then goes idle. This could also be created if
the ULP negotiates the resource levels with the Remote Peer, which
causes the Local Peer to consume resources; however, the Remote Peer
never sends data to consume them. The general countermeasure
described in this section can be used to free resources allocated by
an idle Local Peer.
6.4.3. Resource Consumption by Active ULPs
This section describes DOS attacks from Local and Remote Peers that
are actively exchanging messages. Attacks on each RDMA NIC resource
are examined and specific countermeasures are identified. Note that
attacks on Stream Context Memory, Page Translation Tables, and STag
namespace are covered in Section 6.4.1, RNIC Resource Consumption, so
they are not included here.
22.214.171.124. Multiple Streams Sharing Receive Buffers
The Remote Peer can attempt to consume more than its fair share of
receive Data Buffers (i.e., Untagged Buffers for DDP or Send Type
Messages for RDMAP) if receive buffers are shared across multiple
If resources are not shared across multiple Streams, then this attack
is not possible because the Remote Peer will not be able to consume
more buffers than were allocated to the Stream. The worst case
scenario is that the Remote Peer can consume more receive buffers
than the local ULP allowed, resulting in no buffers being available,
which could cause the Remote Peer's Stream to the Local Peer to be
torn down, and all allocated resources to be released.
If local receive Data Buffers are shared among multiple Streams, then
the Remote Peer can attempt to consume more than its fair share of
the receive buffers, causing a different Stream to be short of
receive buffers, and thus, possibly causing the other Stream to be
torn down. For example, if the Remote Peer sent enough one-byte
Untagged Messages, they might be able to consume all locally shared,
receive queue resources with little effort on their part.
One method the Local Peer could use is to recognize that a Remote
Peer is attempting to use more than its fair share of resources and
terminate the Stream (causing the allocated resources to be
released). However, if the Local Peer is sufficiently slow, it may
be possible for the Remote Peer to still mount a denial of service
attack. One countermeasure that can protect against this attack is
implementing a low-water notification. The low-water notification
alerts the ULP if the number of buffers in the receive queue is less
than a threshold.
If all the following conditions are true, then the Local Peer or
local ULP can size the amount of local receive buffers posted on the
receive queue to ensure a DOS attack can be stopped.
* A low-water notification is enabled, and
* The Local Peer is able to bound the amount of time that it takes
to replenish receive buffers, and
* The Local Peer maintains statistics to determine which Remote
Peer is consuming buffers.
The above conditions enable the low-water notification to arrive
before resources are depleted, and thus, the Local Peer or local ULP
can take corrective action (e.g., terminate the Stream of the
attacking Remote Peer).
A different, but similar, attack is if the Remote Peer sends a
significant number of out-of-order packets and the RNIC has the
ability to use the ULP buffer (i.e., the Untagged Buffer for DDP or
the buffer consumed by a Send Type Message for RDMAP) as a reassembly
buffer. In this case, the Remote Peer can consume a significant
number of ULP buffers, but never send enough data to enable the ULP
buffer to be completed to the ULP.
An effective countermeasure is to create a high-water notification
that alerts the ULP if there is more than a specified number of
receive buffers "in process" (partially consumed, but not completed).
The notification is generated when more than the specified number of
buffers are in process simultaneously on a specific Stream (i.e.,
packets have started to arrive for the buffer, but the buffer has not
yet been delivered to the ULP).
A different countermeasure is for the RNIC Engine to provide the
capability to limit the Remote Peer's ability to consume receive
buffers on a per Stream basis. Unfortunately, this requires a large
amount of state to be tracked in each RNIC on a per Stream basis.
Thus, if an RNIC Engine provides the ability to share receive buffers
across multiple Streams, the combination of the RNIC Engine and the
Privileged Resource Manager MUST be able to detect if the Remote Peer
is attempting to consume more than its fair share of resources so
that the Local Peer or local ULP can apply countermeasures to detect
and prevent the attack.
126.96.36.199. Remote or Local Peer Attacking a Shared CQ
For an overview of the shared CQ attack model, see Section 7.1.
The Remote Peer can attack a shared CQ by consuming more than its
fair share of CQ entries by using one of the following methods:
* The ULP protocol allows the Remote Peer to cause the local ULP to
reserve a specified number of CQ entries, possibly leaving
insufficient entries for other Streams that are sharing the CQ.
* If the Remote Peer, Local Peer, or local ULP (or any combination)
can attack the CQ by overwhelming the CQ with completions, then
completion processing on other Streams sharing that Completion
Queue can be affected (e.g., the Completion Queue overflows and
The first method of attack can be avoided if the ULP does not allow a
Remote Peer to reserve CQ entries, or if there is a trusted
intermediary, such as a Privileged Resource Manager. Unfortunately,
it is often unrealistic not to allow a Remote Peer to reserve CQ
entries, particularly if the number of completion entries is
dependent on other ULP negotiated parameters, such as the amount of
buffering required by the ULP. Thus, an implementation MUST
implement a Privileged Resource Manager to control the allocation of
CQ entries. See Section 2.1, Components, for a definition of a
Privileged Resource Manager.
One way that a Local or Remote Peer can attempt to overwhelm a CQ
with completions is by sending minimum length RDMAP/DDP Messages to
cause as many completions (receive completions for the Remote Peer,
send completions for the Local Peer) per second as possible. If it
is the Remote Peer attacking, and we assume that the Local Peer's
receive queue(s) do not run out of receive buffers (if they do, then
this is a different attack, documented in Section 188.8.131.52 Multiple
Streams Sharing Receive Buffers), then it might be possible for the
Remote Peer to consume more than its fair share of Completion Queue
entries. Depending upon the CQ implementation, this could either
cause the CQ to overflow (if it is not large enough to handle all the
completions generated) or for another Stream not to be able to
generate CQ entries (if the RNIC had flow control on generation of CQ
entries into the CQ). In either case, the CQ will stop functioning
correctly, and any Streams expecting completions on the CQ will stop
This attack can occur regardless of whether all the Streams
associated with the CQ are in the same or different Protection
Domains - the key issue is that the number of Completion Queue
entries is less than the number of all outstanding operations that
can cause a completion.
The Local Peer can protect itself from this type of attack using
either of the following methods:
* Size the CQ to the appropriate level, as specified below (note
that if the CQ currently exists and needs to be resized, resizing
the CQ is not required to succeed in all cases, so the CQ resize
should be done before sizing the Send Queue and Receive Queue on
the Stream), OR
* Grant fewer resources than the Remote Peer requested (not
supplying the number of Receive Data Buffers requested).
The proper sizing of the CQ is dependent on whether the local ULP(s)
will post as many resources to the various queues as the size of the
queue enables. If the local ULP(s) can be trusted to post a number
of resources that is smaller than the size of the specific resource's
queue, then a correctly sized CQ means that the CQ is large enough to
hold completion status for all the outstanding Data Buffers (both
send and receive buffers), or:
CQ_MIN_SIZE = SUM(MaxPostedOnEachRQ)
MaxPostedOnEachRQ = the maximum number of requests that
can cause a completion that will be posted on a
specific Receive Queue.
MaxPostedOnEachSRQ = the maximum number of requests that
can cause a completion that will be posted on a
specific Shared Receive Queue.
MaxPostedOnEachSQ = the maximum number of requests that
can cause a completion that will be posted on a
specific Send Queue.
If the local ULP must be able to completely fill the queues, or
cannot be trusted to observe a limit smaller than the queues, then
the CQ must be sized to accommodate the maximum number of operations
that it is possible to post at any one time. Thus, the equation
CQ_MIN_SIZE = SUM(SizeOfEachRQ)
SizeOfEachRQ = the maximum number of requests that
can cause a completion that can ever be posted
on a specific Receive Queue.
SizeOfEachSRQ = the maximum number of requests that
can cause a completion that can ever be posted
on a specific Shared Receive Queue.
SizeOfEachSQ = the maximum number of requests that
can cause a completion that can ever be posted
on a specific Send Queue.
MaxPosted*OnEach*Q and SizeOfEach*Q vary on a per Stream or per
Shared Receive Queue basis.
If the ULP is sharing a CQ across multiple Streams that do not share
Partial Mutual Trust, then the ULP MUST implement a mechanism to
ensure that the Completion Queue does not overflow. Note that it is
possible to share CQs even if the Remote Peers accessing the CQs are
untrusted if either of the above two formulas are implemented. If
the ULP can be trusted not to post more than MaxPostedOnEachRQ,
MaxPostedOnEachSRQ, and MaxPostedOnEachSQ, then the first formula
applies. If the ULP cannot be trusted to obey the limit, then the
second formula applies.
184.108.40.206. Attacking the RDMA Read Request Queue
The RDMA Read Request Queue can be attacked if the Remote Peer sends
more RDMA Read Requests than the depth of the RDMA Read Request Queue
at the Local Peer. If the RDMA Read Request Queue is a shared
resource, this could corrupt the queue. If the queue is not shared,
then the worst case is that the current Stream is no longer
functional (e.g., torn down). One approach to solving the shared
RDMA Read Request Queue would be to create thresholds, similar to
those described in Section 220.127.116.11, Multiple Streams Sharing Receive
Buffers. A simpler approach is to not share RDMA Read Request Queue
resources among Streams or to enforce hard limits of consumption per
Stream. Thus, RDMA Read Request Queue resource consumption MUST be
controlled by the Privileged Resource Manager such that RDMAP/DDP
Streams that do not share Partial Mutual Trust do not share RDMA Read
Request Queue resources.
If the issue is a bug in the Remote Peer's implementation, but not a
malicious attack, the issue can be solved by requiring the Remote
Peer's RNIC to throttle RDMA Read Requests. By properly configuring
the Stream at the Remote Peer through a trusted agent, the RNIC can
be made not to transmit RDMA Read Requests that exceed the depth of
the RDMA Read Request Queue at the Local Peer. If the Stream is
correctly configured, and if the Remote Peer submits more requests
than the Local Peer's RDMA Read Request Queue can handle, the
requests would be queued at the Remote Peer's RNIC until previous
requests complete. If the Remote Peer's Stream is not configured
correctly, the RDMAP Stream is terminated when more RDMA Read
Requests arrive at the Local Peer than the Local Peer can handle
(assuming that the prior paragraph's recommendation is implemented).
Thus, an RNIC implementation SHOULD provide a mechanism to cap the
number of outstanding RDMA Read Requests. The configuration of this
limit is outside the scope of this document.
6.4.4. Exercise of Non-Optimal Code Paths
Another form of a DOS attack is to attempt to exercise data paths
that can consume a disproportionate amount of resources. An example
might be if error cases are handled on a "slow path" (consuming
either host or RNIC computational resources), and an attacker
generates excessive numbers of errors in an attempt to consume these
resources. Note that for most RDMAP or DDP errors, the attacking
Stream will simply be torn down. Thus, for this form of attack to be
effective, the Remote Peer needs to exercise data paths that do not
cause the Stream to be torn down.
If an RNIC implementation contains "slow paths" that do not result in
the tear down of the Stream, it is recommended that an implementation
provide the ability to detect the above condition and allow an
administrator to act, including potentially administratively tearing
down the RDMAP Stream associated with the Stream that is exercising
data paths, which consume a disproportionate amount of resources.
6.4.5. Remote Invalidate an STag Shared on Multiple Streams
If a Local Peer has enabled an STag for remote access, the Remote
Peer could attempt to remotely invalidate the STag by using the RDMAP
Send with Invalidate or Send with SE and Invalidate Message. If the
STag is only valid on the current Stream, then the only side effect
is that the Remote Peer can no longer use the STag; thus, there are
no security issues.
If the STag is valid across multiple Streams, then the Remote Peer
can prevent other Streams from using that STag by using the Remote
Thus, if RDDP Streams do not share Partial Mutual Trust (i.e., the
Remote Peer may attempt to remotely invalidate the STag prematurely),
the ULP MUST NOT enable an STag that would be valid across multiple
6.4.6. Remote Peer Attacking an Unshared CQ
The Remote Peer can attack an unshared CQ if the Local Peer does not
size the CQ correctly. For example, if the Local Peer enables the CQ
to handle completions of received buffers, and the receive buffer
queue is longer than the Completion Queue, then an overflow can
potentially occur. The effect on the attacker's Stream is
catastrophic. However, if an RNIC does not have the proper
protections in place, then an attack to overflow the CQ can also
cause corruption and/or termination of an unrelated Stream. Thus, an
RNIC MUST ensure that if a CQ overflows, any Streams that do not use
the CQ MUST remain unaffected.
6.5. Elevation of Privilege
The RDMAP/DDP Security Architecture explicitly differentiates between
three levels of privilege: Non-Privileged, Privileged, and the
Privileged Resource Manager. If a Non-Privileged ULP is able to
elevate its privilege level to a Privileged ULP, then mapping a
physical address list to an STag can provide local and remote access
to any physical address location on the node. If a Privileged Mode
ULP is able to promote itself to be a Resource Manager, then it is
possible for it to perform denial of service type attacks where
substantial amounts of local resources could be consumed.
In general, elevation of privilege is a local implementation specific
issue and is thus outside the scope of this document.