1

Secure IoT Routing: Selective Forwarding Attacks

and Trust-based Defenses in RPL Network
Jun Jiang, Yuhong Liu†

Department of Computer Science and Engineering, Santa Clara University, Santa Clara, CA, USA

{jjiang2, yhliu}@[Link]
arXiv:2201.06937v1 [[Link]] 18 Jan 2022

Abstract—IPv6 Routing Protocol for Low Power and Lossy and can remain undetected for a longer time, causing long-
Networks (RPL) is an essential routing protocol to enable com- term damage to the network. Despite their impact, existing
munications for IoT networks with low power devices. RPL uses selective forwarding attacks are still lack of flexibility in terms
an objective function and routing constraints to find an optimized
routing path for each node in the network. However, recent of dynamically identifying victim nodes and adjusting packet
research has shown that topological attacks, such as selective forward rates according to the state of the network.
forwarding attacks, pose great challenges to the secure routing On the other hand, there are three major types of defense
of IoT networks. Many conventional secure routing solutions, on mechanisms against selective forwarding attacks on RPL.
the other hand, are computationally heavy to be directly applied The first type is to build a multi-path routing network to
in resource-constrained IoT networks. There is an urgent need
to develop lightweight secure routing solutions for IoT networks. ensure the integrity of information transmission [11]. These
In this paper, we first design and implement a series of advanced mechanisms often require excessive resources to maintain the
selective forwarding attacks from the attack perspective, which backup paths for nodes. The second type is distributed defense
can flexibly select the type and percentage of forwarding packets mechanisms, which deploy a monitor module on individual
in an energy efficient way, and even bad-mouth other innocent nodes [12]. However, these defense mechanisms often lead to
nodes in the network. Experiment results show that the proposed
attacks can maximize the attack consequences (i.e. number of significant extra energy consumption at the monitoring nodes.
dropped packets) while maintaining undetected. Moreover, we In addition, it is challenging to ensure that all the monitoring
propose a lightweight trust-based defense solution to detect and nodes are long-term reliable and honest in reporting their
eliminate malicious selective forwarding nodes from the network. neighbors’ behaviors. The third type is the centralized defense
The results show that the proposed defense solution can achieve mechanisms [13], where a central node is employed to monitor
high detection accuracy with very limited extra energy usage (i.e.
3.4%). and analyze malicious behaviors in the network. Nevertheless,
the central node has to be deployed at a core location in the
Index Terms—Selective forwarding attack, RPL, Routing se- network to ensure coverage across the entire network, and may
curity, IoT, Trust be easily misled by complex bad-mouthing attacks.
In this work, we aim to advance current studies from both
I. I NTRODUCTION the attack and defense perspectives. Specifically, from the
attack aspect, we propose an advanced selective forwarding
With the rapid adoption of Internet of Things (IoT) devices
attack model, which can dynamically launch three types of
around the world [1], many of these devices are resource-
malicious behaviors: (1) flexibly dropping packets from se-
constrained [2]. Since the existing Internet Protocols (IP)
lected types of protocols, (2) adjusting the packet forward
are too complex to be directly implemented on resource-
rate based on the average network packet forward rate to stay
constrained IoT devices [3], the Internet Engineering Task
stealthy, and (3) dynamically selecting specific children nodes
Force (IETF) designs a lightweight IPv6 protocol with a series
for bad-mouthing attacks. Furthermore, these attack behaviors
of core protocols to ensure efficient and secure communica-
can be combined to significantly increase the damages to the
tions, such as IPv6 over Low-power Wireless Personal Area
network and reduce the risk of being detected by state-of-the-
Networks (6LoWPAN) [4] and Routing Protocol for Low
art defense mechanisms.
Power and Lossy Networks (RPL) [5]. In particular, RPL,
Furthermore, from the defense aspect, we propose a novel
as the core routing protocol for resource constrained IoT
centralized trust-based defense mechanism to combat selective
networks, has been adopted by a variety of applications, such
forwarding attacks in RPL networks. Compared to distributed
as healthcare [6], smart grid [7], and smart city [8], etc.
defense mechanisms, the proposed scheme can significantly
Due to its popularity, RPL becomes an attractive attack tar-
save the energy consumption for anomaly detection by de-
get [9]. One attack that can cause massive damage to the RPL
ploying the defense computation only at the root node. Unlike
network is selective forwarding attack [10], where attackers
other centralized defense schemes, the proposed scheme takes
interrupt network data flows by selectively dropping network
advantage of RPL’s tree-based network topology to evaluate
packets. Compared to blackhole attacks which simply drop
the trustworthiness of nodes and avoids the introduction of a
all packets, selective forwarding attacks are more deceptive
trusted third-party node. Furthermore, by carefully designing
† : Corresponding author the detection/notification module, the proposed scheme can
 2

effectively balance the trade-off between detection delay and and stability of the network by changing the topology of the
energy consumption. network, such as sinkhole attacks [10], blackhole attacks [13]
The main contributions of this paper are as follows. and selective forwarding attack [18].
• This work proposes an advanced selective forwarding Selective forwarding attack is one of most devastating type
attack with three types of behaviors against RPL network. of attacks [19]–[21], which can cause severe damage to the
Malicious nodes in the proposed attack model cannot network. However, in many of these attacks, the attackers
only flexibly choose the type of packets to drop, but also set a fixed number of dropped packets or blindly attack all
control the packet forward rates dynamically. As a result, data packets, which increases the risk of being detected by
these attackers are able to launch more stealthy attacks to defense mechanisms. In some advanced attacks [10], [12], the
avoid being detected. Furthermore, malicious nodes can proposed attacks interrupt the victim node’s communications
also bad-mouth other normal nodes to cause false alarms by only forwarding RPL control messages while dropping all
in the system. The experiment results show that it can [10] or partial data packets [12]. Different from these existing
effectively evade RPL self-defense mechanism and state- attacks, in this study, the proposed selective forwarding attacks
of-the-art defense mechanisms. can perform more flexible malicious behaviors and even bad-
• This work proposes a novel centralized trust-based de- mouth other innocent nodes to mislead state-of-the-art defense
fense mechanism. In particular, the proposed trust model schemes.
integrates a self-trust value, which reflects a node’s
trustworthiness in its packet forwarding behavior, and B. RPL Network Defenses
a tree-based descendant trust value, which takes ad- The original design of RPL protocol has some basic security
vantage of the RPL network topology to prevent bad- schemes, such as the local and global repair mechanisms [5],
mouthing attacks. Furthermore, the proposed defense which can be triggered by changes of network topology, e.g. a
mechanism is deployed on the root node, which can link failure. However, these basic security repair mechanisms
effectively reduce the total energy consumption caused are far from adequate to resist the rapid evolving security
by distributed anomaly monitoring scheme, and eliminate attacks [22].
the security risks caused by the introduction of third- Beyond the basic repair mechanisms, there are mainly three
party devices. Experiment results show that the proposed categories of defenses. The first type is to establish multiple
scheme achieves high detection accuracy and low energy routing paths for each node to avoid selective forwarding
consumption. attacks. In [11], the authors propose a secure routing protocol
• This work proposes a novel anomaly report mechanism. M-RPL, which establishes a hierarchical cluster network and
Instead of using UDP packets, ICMPv6 control message backup paths for different clusters in the route discovery phase.
is chosen to send information about malicious nodes. In [23], the authors establish a temporary backup path for a
The reporting mechanism only starts when root node node based on its packet delivery ratio. In [24], the authors pro-
detects the malicious behaviors in the network. This pose to use the principle of Packet Replication and Elimination
reporting mechanism cannot only ensure that nodes in the (PRE), through IEEE 802.15.4 Time-Slotted Channel Hopping
network can be notified promptly, but also avoid causing (TSCH) as media access, to create parallel paths from nodes
information broadcast storms. to the root node. Although the multi-path mechanism can
The rest of this paper is organized as follows. Section II effectively resist attacks, due to the introduction of redundant
discusses existing selective forwarding attacks and defense network routes and extra resources to maintain the backup
mechanisms in RPL networks. Section III introduces pre- paths, these defense schemes often cause significant increases
liminaries of the RPL protocol. Section IV and V discuss in nodes’ energy consumption.
the proposed selective forwarding attack and the defense The second type of defenses is distributed defense mech-
mechanism in details. The results of the experiment are given anisms that build monitor module on each network node.
in Section VI, followed by a conclusion in Section VII. Due to its easy implementation, this type of defenses is the
most popular one. In [12], [25], [26], the authors propose
various distributed trust-based mechanisms, where each indi-
II. RELATED WORK
vidual node monitors its neighbors’ incoming and forward-
A. Selective Forwarding Attacks in RPL Networks ing traffic and calculates their trust values. In [27], [28], a
RPL network faces a variety of security threats, which distributed, collaborative and layered trust-based IDS (T-IDS)
are mainly divided into three categories [14], [15]. The first and a Metric-based RPL Trustworthiness Scheme (MRTS) are
type of attack is resource attack, such as flooding attacks proposed respectively, where each node monitors and cooper-
[16] and increased rank attacks [17]. In these attacks, the ates with its neighbors to detect and report intrusions. These
attacker aims to exhaust the victim node’s energy and reduce distributed mechanisms, however, have to be deployed on each
its lifetime by misleading it to execute a large number of IoT nodes, leading to significant extra energy expenditure.
unnecessary instructions. The second type of attack is traffic On the other hand, it also presents a new challenge, as how
attack, such as sniffing attacks [14] and identity attacks [10], resource-constrained nodes can provide long-term reliable and
where the attacker’s main goal is to eavesdrop or manipulate honest reports on their neighbors’ behaviors.
the network’s traffic. The third type of attack is network The third type is centralized defense mechanisms. In [13],
topology attack where the attacker undermines the security the authors use a SVELTE intrusion detection system (IDS),
 3

where a 6LoWPAN Mapper is placed on the IPv6 Border

Router (6BR) to monitor and analyze the malicious behaviors
in the network. In [10], the authors propose a lightweight
heartbeat protocol, in which the root node detects the mali-
cious node by periodically exchanging an echo signal with all
Fig. 1: Format of DIO message
its children nodes. In [29], authors introduce a control layer
to achieve the hierarchical trust-based mechanism “CTrust- parent node. Based on the objective function, an optimal path
RPL” to monitor the nodes’ behaviors. These centralized from any leaf node to the root node can be constructed.
detection mechanisms are often energy-efficient since only a
limited number of nodes are involved for behavior monitoring. A. Control Messages in RPL Network
However, since the root node cannot directly monitor each There are four main types of control messages [5] to
node in the network, it can be easily misled by some malicious establish the DODAG, which are DODAG Information Object
attacks, such as bad-mouthing attacks [30], where malicious (DIO), DODAG Information Solicitation (DIS), Destination
nodes blame their parent/children nodes for packet dropping. Advertisement Object (DAO) and Destination Advertisement
As a result, it is challenging for such defense schemes to be Object Acknowledgement (DAO-ACK). Particularly, DIO is
robust against complex bad-mouthing strategies. the most frequently used message, sent by the root node to
In our prior conference paper [31], we have studied black- all other nodes in the network, to advertise network structures
hole attacks and proposed a centralized defense mechanism for DODAG discovery, assembly and maintenance. Therefore,
at the root node. On this basis, we further study the selective to rapidly report anomaly detection results while avoiding
forwarding attacks and defense scheme in this work. Unlike extra overhead introduced by anomaly reporting messages, we
the blackhole attack model, which blindly drops all packets, propose to insert detection results into the DIO message to
the attack model in this study launches attacks with three distribute to all nodes in the network. The format of a DIO
malicious behaviors, (1) flexibly select packet protocol, (2) message is shown in Figure 1, which includes RPL InstanceID,
dynamically adjust packet forward rate, (3) bad-mouth the Rank value, DODAG ID, Destination Advertisement Trigger
victim children nodes. As a result, it can effectively hide Sequence Number (DTSN), etc. Non-root nodes must adver-
the attack behaviors and cause long-term network damage. tise and remain the values in DIO message, except for the
In addition, we propose a new lightweight centralized trust- update of the fields Rank and DTSN.
based defense scheme to defend against selective forwarding
attacks. Compared to our prior work, where root node simply
B. Default Security Mechanisms in RPL Networks
uses the average packet forward rate as the trust value for
each node, the defense scheme proposed in this work is more RPL networks can adopt standard mechanisms to ensure
comprehensive by (1) integrating self-trust value and tree- message integrity and confidentiality at different layers of the
based descendant trust value, (2) introducing a beta-based trust protocol stack. For example, the standardized IEEE 802.15.4
framework with a discounting factor to gradually reduce the security, lightweight 6LoWPAN compression IPsec [32], and
impact of previous behaviors, and (3) assigning asymmetric Datagram TLS (DTLS) are adopted to ensure security at the
discounting factors for good/bad behaviors, so that a node’s data link layer, IP layer and transport layer, respectively.
bad behavior is remembered for a longer time. Furthermore, In addition, RPL adopts some simple repair mechanisms to
more comprehensive experiments are performed to evaluate recover from three network failures, including routing topol-
the performance of the proposed attacks and defenses from ogy failure, link failure and node failure. When a small number
different aspects. of failures happen, the local repair mechanism starts. The
local repair mechanism allows the nodes, which are impacted
III. P RELIMINARIES : RPL P ROTOCOL by failures, to detach from the original DODAG and change
As the attacks and defenses proposed in this work are their rank values to infinite, then re-join the DODAG again.
based on RPL protocols, in this section, we briefly introduce After multiple local repair mechanisms are initiated, the RPL
some basis of RPL protocol. The Routing Protocol for LLN protocol performs a global repair to rebuild the entire DODAG
(RPL) is defined by the IETF’s Routing Over Low power and network by increasing the DODAG version number. Please
Lossy Networks (ROLL) Working Group. In particular, each note that the RPL uses a trickle timer to handle inconsistencies
RPL network may contain multiple RPL instances. Each RPL in the RPL DODAG. When the RPL network is stable, the
instance may contain multiple Destination Oriented Directed interval of the trickle timer will increase exponentially. When
Acyclic Graph (DODAG). In a DODAG, the root node, which a network inconsistency is detected, such as a loop generation,
is usually the most powerful node, is responsible for storing the trickle timer is reset.
and managing the routing paths. Non-root nodes can be added As briefly discussed in Section II-B, these basic defense
to one or more DODAGs. mechanisms are far from adequate when advanced attacks are
RPL is a hierarchical-based routing protocol that relies on launched.
a DAG structure to exchange data among network nodes.
Consequently, the parent and children relationship is essential IV. P ROPOSED S ELECTIVE F ORWARDING ATTACK
for routing in a RPL network. Each DODAG has a specific Among diverse attacks against RPL networks, some attacks
objective function, which defines how each node selects its aim to cause as severe damage as possible to the network
 4

Algorithm 1: Protocol-based Attack Algorithm 2: Packet Forward Rate-based Attack

Input: P acket (the Struct of IPv6 packet that need to Input: P acket (the Struct of IPv6 packet that need to
∆t
be forwarded) be forwarded), Fm (PFR of the malicious
1 if P [Link] 6= RP L then node within duration ∆t), F̄ ∆t (Estimated
2 Drop packet; network average PFR within duration ∆t)
∆t ∆t
3 end 1 if Fm = (F̄ + ε) then
2 Drop this packet;
3 end

within a short time period, such as blackhole attacks [31].

These attacks are often easily detected and isolated by the
defense mechanisms due to the aggressiveness of the malicious where Ri∆t and Si∆t represent the number of packets received
nodes’ behaviors. Therefore, selective forwarding attacks, and forwarded by node Ni within time duration ∆t, respec-
which can interrupt network communications in a flexible and tively. In the proposed attack, a malicious node estimates the
stealthy way, are often launched to cause long term network network average PFR (i.e. F̄ ∆t ) based on its neighbors’ PFR,
damages. In this study, we propose an advanced selective as shown below
forwarding attack model with three different types of selective Pp
∆t F ∆t
behaviors. F̄ ≈ i=1 i (2)
p
1) Protocol-based Attack: We propose to selectively drop
network packets according to their protocol types. Specifi- where p represents the number of neighbors of the malicious
cally, we propose to drop only data packets (i.e. non-ICMP node.
packets) to achieve attack stealthiness. This is because in With a larger value of p and a longer time duration ∆t, the
RPL networks, the loss of IPv6-based control messages (i.e. malicious node can achieve a more accurate estimation of F̄ ∆t ,
ICMP packets) will cause inconsistencies in network routing which, however, will also cause extra energy consumption
topology and trigger the RPL repair mechanisms. However, and time delay. The proposed attack can flexibly adjust the
since data massages are transmitted based on UDP, the loss trade-off according to specific attack scenarios. As shown in
of such messages is difficult to be detected by RPL’s self- Algorithm 2, before forwarding a packet, the malicious node
defense mechanisms. In such cases, the RPL’s self-recovery determines whether to drop the packet by comparing the its
mechanisms will not be triggered [10]. current PFR with the observed network average PFR plus ε.
As shown in Algorithm 1, before the malicious node for- The value of ε can be adjusted according to the aggressiveness
wards a packet, it first determines whether the type of the of the attack.
packet is a data packet. If so, this packet can be dropped.
3) Bad-mouthing Attack: The proposed attack can arbitrar-
Please note that, to enable selective behaviors based on
ily choose one or multiple children nodes to achieve bad-
protocol types, the malicious node only needs to check the
mouthing attack. In bad-mouthing attack, the malicious node
header part of the messages, which will not incur significant
can frame the victim node (i.e. one of its children nodes) up
processing power.
by discarding data packets from the victim node.
2) Packet Forward Rate-based Attack: We assume that the
For this attack, the most challenging part is to selectively
data messages transmitted between nodes are encrypted, so
choose the victim nodes and attack strategy. Since blindly se-
that the attacker can control whether the malicious nodes
lecting children nodes to attack or attacking all children nodes
discard the data packet but cannot change the content of
increase the risk of the attacker being exposed to the detection
the data packet. The malicious nodes can dynamically adjust
mechanism, we propose to only select specific children nodes
their packet forward rates (PFR) according to the network
as the victim nodes. More importantly, the attacker can flexibly
conditions. It is not easy to determine an appropriate PFR,
choose victim nodes that are either located at critical network
which can cause non-trivial damages to the network while
positions, or requiring minimum attack effort. For example,
avoiding being detected. In this study, we propose to achieve
as shown in Algorithm 3, the malicious node identifies its
this goal by controlling the PFR to be slightly above the
children nodes with lower PFR as the victims as badmouthing
average PFR of the network. Specifically, the malicious node
these victims requires dropping less number of packets (i.e.
estimates the average PFR by monitoring all its neighbors’
less attack effort).
incoming and outgoing packets, and ensures that its PFR is
slightly above the average PFR of the network (i.e. by a After identifying the ideal victim node, the malicious node
small value ε). Please note this PFR will be dynamically can dynamically discard the victim’s packets, misleading the
updated based on the changes of the average network PFR. root node to identify the victim as a malicious node that drops
Consequently, the malicious node can hide itself while still packets. If multiple malicious nodes coordinately launch attack
causing long term damage to the network. The equation to at the same time, the false alarm rate will significantly increase
calculate a node Ni ’s PFR is shown below for most trust-based defense solutions.
Please note that although we discuss these three attack
Si∆t behaviors independently for the sake of clarification, these
Fi∆t = (1) attacks can be flexibly integrated to cause more damage.
Ri∆t
 5

Algorithm 3: Bad-mouthing
Input: P acket (the Struct of IPv6 packet that need to
be forwarded), Nm (the Struct of malicious
node), F̄ ∆t (Estimated neighbor’s average
PFR), ChList (list of children nodes),
N umV ict (number of victim children nodes)
1 sort ChList in ascending order of PFR;
2 for n = 0; n ≤ N umV ict; n + + do
3 if P [Link] acAddr = ChList[n].M acAddr
and P [Link] acAddr = Nm .M acAddr
and FChList[n] >= F̄ ∆t then
4 Drop packet;
5 break;
6 end Fig. 2: Packet Propagation in the Proposed System
7 end
model binary events. In our case, we consider whether a node
can successfully forward a packet or not as a random binary
event. Then based on the prior observations on the number of
V. L IGHTWEIGHT T RUST- BASED D EFENSE S CHEME successful and failed packet forwarding events, the probability
In this section, we propose a lightweight trust-based defense of this node to successfully forward the next packet can be
scheme, which is deployed on the root-node, against selective estimated as the expected value of the beta distribution, as
forwarding attacks, as shown in Figure 2. The input for the shown in equation (3), where S and L represent the number
defense scheme is the data packets received from non-root of packets successfully sent or lost by node ni , respectively.
nodes. The defense scheme includes three major modules. The S+1
detection module analyzes the trust value of each node based Tis = (3)
(S + L + 2)
on the received data packets, whose propagation path is shown
by the solid black line in Figure 2. After malicious nodes From equation (3), we can observe that a node’s self-trust
(e.g. node 4 in Figure 2) are identified, the notification module value increases when more packets are successfully forwarded,
encapsulates such information in DIO packets and notifies all or drops when more packets are lost.
the nodes in the network, as indicated by the brown dashed However, the basic beta-based trust model cannot capture
line in Figure 2. In the isolation module, children nodes of alternative behavior attacks [34], where malicious nodes alter-
the identified malicious nodes (e.g. node 7 in Figure 2) can natively perform good behaviors to accumulate high trust val-
isolate the malicious nodes and re-select their parent nodes ues and bad behaviors to interrupt network traffic. To prevent
based on received DIO messages. For example, the changed such attacks, we propose to introduce temporal information to
propagation path of data packets from node 7 is shown as the discount a node’s packet forwarding behaviors performed long
blue dashed line in Figure 2. time ago. Specifically, we introduce a discounting factor f to
In the rest of this section, we first introduce the design of the gradually forget a node’s behavior over time, so that a behavior
trust model, which is the core of the proposed defense solution. with a smaller discounting factor indicates a lower influence
Then, we discuss each module of the proposed scheme in on the node’s trust value. The calculation of the discounting
details. factor for the k th behavior of a node is shown below.
fk = e−λ(q−k) (4)
A. Trust Evaluation Model where q represents the total number of behaviors performed
An advanced trust model is designed to evaluate the by the node so far, including both successful and failed
anomaly of each node’s behavior. In a RPL network, the root forwarding behaviors; and (q − k) represents the total number
node tracks the behaviors of each individual node ni and of behaviors performed after behavior k. In other words, each
dynamically calculates the trust value, which is denoted as time when a node performs a new behavior, all previous
Ti . The trust value falls in the range from zero to one. When behaviors will have their (q − k) value increased by 1,
a node’s trust value is below a trust threshold δ, it will be resulting in a smaller discounting factor. Please note that the
identified as a malicious node. latest behavior will always have its ∆k = 0, leading to its
In the proposed trust model, the overall trust value Ti of a discounting factor value as 1. In addition, the forgetting speed
node ni is composed of two parts: a self-trust value Tis , which λ is a constant value. A larger λ value will result in a smaller
is to capture failures of packet forwarding, and a tree-based discounting factor and thus a higher forgetting speed.
descendant trust value Tid , which is to capture bad-mouthing More importantly, to punish bad behaviors further, we
attacks. propose to design an asymmetric trust model so that past good
1) Self-trust Value: The proposed scheme adopts the Beta behaviors can be quickly forgotten while past bad behaviors
trust model [33] as its basis. Beta distribution is a family will be remembered for a longer time. To achieve this goal,
of continuous probability distributions that are often used to two different λ values (i.e. λg and λb ) are adopted to separately
 6

discount past good and bad behaviors respectively, where between nodes are encrypted, meaning that malicious nodes on
λg > λb . Therefore, the self-trust value of node i with q the routing path cannot change the content of the transmitted
s
behaviors (i.e. Ti,q ) is calculated as follows. data information. Second, all data messages generated by non-
Pq root nodes are transmitted through the root node to the external
k=1 ak fk,g +1
s
Ti,q = Pq Pq (5) network (e.g. the Internet). This is a reasonable assumption for
k=1 (1 − ak )fk,b + k=1 ak fk,g + 2 most RPL networks [10], [35].
In equation (5), ak value is 1 if behavior k is good, or 0 if Current RPL protocol does not support the root node to
behavior k is bad. record and track the packets sent by non-root nodes. To address
2) Tree-based Descendant Trust Value: Due to the re- this challenge, we introduce a sequence number, which is
stricted hierarchical structure of RPL networks, only parent stored in the first byte of the data payload sent by each node.
nodes will forward data packets for their children nodes. It is In particular, the sequence number is increased by one each
very easy for a malicious parent node to control the packet time when the source node sends out a packet. The root node
transmissions of one of its children nodes to launch bad- estimates the packet forward rate for each source node based
mouthing attacks against this child. By only considering self- on the number of received data packets and the corresponding
trust value, the victim child node’s trust will drop while the sequence numbers.
malicious parent’s trust value remains the same. To further In addition, since frequently calculating the trust value for
defeat such attacks, we propose to also introduce a tree-based each node greatly increases the workload of the root node, we
descendant trust value for each node, which considers the trust introduce a sliding time window. The root node only calculates
value of its direct descendants. The descendant trust value Tid the trust value of each node once in each sliding time window.
is defined as follows. The length of the window can be determined according to the
P network status, such as the battery capacity of the root node
d
(wj Tjs )
Ti = (6) and the sensitivity of the trust value.
Ci By calculating the trust value of each node, the root node
where Tjs is the self-trust value of node ni ’s child node nj . can identify the possible malicious nodes according to Al-
Parameter wj represents the weight of the child node nj , which gorithm 4. Confirmed malicious nodes will be added to the
is determined by the number of data packet received by nj per “blacklist”. However, because the nodes may suffer from bad-
time period. The greater number of data packets received by mouthing attacks from their parent nodes, it may lead to high
nj means larger weight assigned to node nj . In addition, Ci false alarm rate if we directly add all nodes with low trust
denotes the total number of packets received by node ni . values to the “blacklist”.
Please note that in the proposed scheme, a node’s descen- To reduce the false alarm rate caused by bad-mouthing
dent trust value only depends on its children’s self-trust value. attacks, we propose to add a “watchlist” and a trust recovery
Since RPL is a tree-like network, if a node’s descendent trust time period. When the trust value of a node is lower than the
value also considers its children’s descendent trust values, threshold for the first time, it will be added to the “watchlist”
it will lead to a recursive counting, where leaf nodes’ trust as a suspicious node. If a suspicious node is required to change
values are over-emphasized. Since the descendant trust value its parent for further investigation, the root node will reset a
is mainly designed to prevent bad-mouthing attacks, which recovery timer and track if the suspicious node’s trust value
can only effectively attack children nodes, we propose to not can recover after the parent change action. Please note that
recursively count it. the length of the recovery timer can be determined according
When the malicious node bad mouth any of its children to the specific network status. A longer timer leads to longer
nodes, its own descendant trust value will be decreased. More- detection delay but lower false alarm rate. Within the recovery
over, the descendant trust value will significantly decrease if time period, if the trust value of the suspicious node recovers
the number of the victim children nodes increases. back to the threshold, indicating that changing its parent node
3) Aggregated Trust Value: Finally, the aggregated trust stops the anomaly, this node is considered as a normal node.
value Ti of the node ni is the combination of the self-trust Then its parent will be identified as the malicious node and
value Tis and the descendant trust value Tid . The calculation added to the “blacklist”. Otherwise, the suspicious node is
of the aggregated trust value is shown below. identified as a malicious node and moved to the “blacklist”.
Ti = ws Tis + wd Tid (7)
C. Notification Module
where ws and wd are the weights for self-trust value and
descendant trust value. This aggregated trust value will serve After a malicious node is identified, the root node needs a
as the major criteria to identify suspicious nodes in the reliable way to notify all the children nodes of the malicious
network. node while avoiding information storms. This is challenging
since the RPL network follows a strict tree-like topology for
data packet forwarding, and a node can only receive data
B. Detection Module packets from its parent. It means that if data packets are used
In this section, we present the detection module, which in- to disseminate the notifications, these packets will be simply
volves the above proposed trust model as its core. Specifically, dropped by malicious nodes and never reach their children
we make two assumptions. First, the data messages transmitted nodes.
 7

Algorithm 4: Malicious nodes detection Algorithm 5: Malicious nodes notification

Input: trust threshold Tth , P arentChangedList (The Input: blacklist, watchlist
nodes in the list have changed parent.) 1 INITIAL:P arentChangingList[] = empty,
Output: blacklist (containing malicious node ID), P arentChangedList[] = False;
watchlist (containing suspicious node ID), 2 while (watchlist not empty or blacklist not empty)
1 INITIAL: watchlist[] = empty, blacklist[] = empty, and SendT ime do
timeperiod = constant value, T rustRecoveryT ime[] 3 if watchlist not empty then
= constant value; 4 P [Link]
2 while timeperiod expires do 5 (FindMaxRankNode(watchlist));
3 foreach j ∈ nodes do 6 broadcast(P arentChangingList);
4 if Tj < Tth then 7 foreach i ∈ P arentChangingList do
5 if node nj not in watchlist then 8 P arentChangedList[i] = True;
6 [Link](nj ); 9 end
7 else 10 end
8 if P arentChangedList[j] = True and 11 if blacklist not empty then
T rustRecoveryT ime[j] expires then 12 broadcast(blacklist);
9 [Link].(nj ); 13 end
10 [Link](nj ); 14 end
P arentChangedList[j] = False;
11 end
12 end Algorithm 6: Isolation malicious nodes
13 else Input: P acket (received ICMP6 control message
14 if P arentChangedList[j] = True and which contains malicious node ID Mid or
node nj in watchlist then suspicious node SU Sid ), Parent ID Pid , Node
15 [Link](nj .old parent); ID Nid
16 [Link](nj ); 1 if P [Link] = Pid or P [Link] Sid = Nid then
17 P arentChangedList[j] = False; 2 RplRemoveParent(Pid);
18 end 3 PreferredParent = RplSelectParent(dag);
19 end 4 broadcast(P acket);
20 end 5 else
21 end
6 broadcast(P acket);
7 end

tion process for non-relevant nodes in the network.

In addition, the root node finds sub-trees, which can cover
Fig. 3: The proposed notification message minimum number of nodes in the “watchlist”, based on RPL
tree-like topology. Then, the root node adds the nodes with the
In this work, we propose to use the control messages (i.e. highest rank value in each sub-tree to “ParentChangingList”,
ICMPv6 messages) to disseminate these notifications, which which is the function “FindMaxRankNode” in Algorithm 5.
can reach the children of a malicious node through other
neighbor nodes. In particular, we recommend using the first
byte in the payload of the ICMPv6 control messages to store
the node ID, as shown in Figure 3. The first bit can be used D. Isolation Module
to distinguish the suspicious node, which is set to 0, and
identified malicious node, which is set to 1. As shown in Algorithm 6, when a non-root node in the
In order to reduce the energy consumption of the network, network receives the notification message, it checks whether
the root node executes the notification module right after the the node information in notification is its parent or itself. If
detection module. As shown in Algorithm 5, there are two any of two cases is true, this node removes the current parent
lists: a “blacklist” and a “ParentChangingList”. The root node from its parent list. Then, it re-selects its preferred parent and
marks the node information in the two lists separately. The broadcasts this ICMPv6 control message to its neighbors. If
“blacklist” is used to notify all the children nodes, whose not, the node broadcasts the ICMPv6 control message directly
parents are identified malicious nodes in the “blacklist”, to to all its neighbors.
re-select their parent node. The suspicious nodes, which are After these three modules are completed, the malicious
required to change parents, are added into the “ParentChang- nodes launching selective forwarding attacks will be aban-
ingList”. This distinction can avoid unnecessary parent selec- doned and isolated from the network.
 8

TABLE I: Simulation Parameters

Parameter Value
Simulation platform Contiki/Cooja 3.0
MAC CSMA/CA
Transport UDP/IPv6
Emulated nodes Z1 mote
Simulation coverage area 130 m * 130 m
Total number of nodes 15
Malicious nodes 3
TX range 50 m
Interference range 100 m
Packet size 46-byte
Data packet period 60 seconds
Routing protocol RPL
Fig. 4: Example of RPL Network Topology Network protocol IP based
Simulation time 150 minutes
VI. E XPERIMENT AND R ESULT A NALYSIS Link failure model UDGM with distance

A. Experiment Set Up
This work adopts Cooja, which is a network simulator of most recent PFR. In the third scheme (i.e. def scheme), since
Contiki OS [36], as our experimental platform. Specifically, the default RPL security scheme allows a node to re-select its
fifteen nodes are randomly deployed in the experiments as parent node when network failures (e.g. link failure and node
shown in Figure 4, including a root node (i.e. node 1), failure) are detected, the root node identifies malicious nodes
eleven legitimate non-root nodes (i.e. node 2 to node 11) by checking whether a node is discarded by its children.
and three malicious nodes (i.e. node 12 to node 14). Min- The effectiveness of proposed attack model is illustrated
imum Rank with Hysteresis Objective Function (MRHOF) in Figure 5. In all sub-figures of Figure 5, the def scheme
is selected as the objective function, where children nodes shows the lowest performance with 0.44 average AUC. This
select their preferred parents according to ranks and Expected is because the proposed attacks only drop data packets, which
Transmission Count (ETX) values. The simulation parameters rarely cause failures in network routing. Although avg scheme
are summarized in TABLE I. and rec scheme show slightly better performances (with an
Based on the experiments, the performance of the proposed average AUC as 0.70 and 0.65 respectively in Figure 5 (a)-
attack and defense scheme are tested and then compared with (b)), their performances significantly drop (with an average
other state-of-the-art works. Specifically, this work adopts AUC as 0.36) when the proportion of bad-mouthing attacks
receiver operating characteristic (ROC) curve as the major increases to above 75%, as shown in Figure 5 (c)-(d). This
performance metric because it can effectively reflect the trade- is because these two schemes consider a node with low PFR
off between the detection rate and the false alarm rate when (i.e. either average PFR or the most recent PFR) as a malicious
different thresholds are adopted. In each ROC curve, the x-axis node, which can be taken advantage by bad-mouthing attacks
and y-axis represent the detection rate and false alarm rate, to frame up the victim nodes.
respectively. The area under ROC curve (AUC) represents the
accuracy. Larger area under the curve (i.e. the higher the AUC
C. Performance of Proposed Defense Modules
value) indicates better performance.
In this sub-section, we evaluate the effectiveness of each
critical strategy proposed for the defense scheme. In particular,
B. Performance of Proposed Attacks these strategies include (1) discounting factor, (2) asymmetric
In this experiment, malicious nodes may either selectively forgetting speed, and (3) integration of self-trust and de-
drop the victim node’s data packets based on their perceived scendant trust. Furthermore, the proposed attack models are
average network PFR, or launch bad-mouthing attacks against launched with 50% selective forwarding behaviors and 50%
a specific child node. To illustrate the impact of different bad-mouthing behaviors.
attacks, the ratio of bad-mouthing attacks to the total number 1) Effectiveness of Discounting Factor: In this subsection,
of attacks in the network is divided into four cases, 25%, 50%, λg and λb are set as the same value λ. By changing the values
75% and 100%. of λ, Figure 6 illustrates its impact on the performance of
The performance of the proposed attacks is evaluated the proposed defense scheme. Specifically, four different λ
against three defense schemes. In the first two schemes, the values (i.e. 0, 0.3, 0.8, 100) are applied so that the discounting
root node identifies the malicious nodes by comparing the PFR factor fk ranges in the interval [0, 1]. From Figure 6, it can be
of each non-root node with a threshold value. The nodes with observed that the defense scheme shows the best performance
lower PFR are identified as malicious. The difference is that (e.g. AUC = 0.630) when λ = 0.3.
the first scheme (i.e. avg scheme) uses a node’s average PRF, Specifically, when λ = 100, fk = e−λ(q−k) = e−100(q−k) ,
while the second scheme (i.e. rec scheme) uses only a node’s which is approximately 0 for any q − k > 0. It indicates
 9

1.0 Savg (AUC 0.431) 1.0 Savg (AUC 0.603) 1.0 1.0
Sdef (AUC 0.450) Sdef (AUC 0.430)
0.8 Srec (AUC 0.423)
0.8 Srec (AUC 0.567)
0.8 0.8
Detection Rate

Detection Rate

Detection Rate

Detection Rate
0.6 0.6 0.6 0.6

0.4 0.4 0.4 λb = 0, λg = 0.1, AUC = 0.671 0.4 λb = 0, λg = 0.2, AUC = 0.758
λb = 0.1, λg = 0.2, AUC = 0.664 λb = 0.1, λg = 0.3, AUC = 0.709
0.2 0.2 0.2 λb = 0.2, λg = 0.3, AUC = 0.671 0.2 λb = 0.2, λg = 0.4, AUC = 0.679
λb = 0.3, λg = 0.4, AUC = 0.632 λb = 0.3, λg = 0.5, AUC = 0.659
λb = 0.4, λg = 0.5, AUC = 0.626 λb = 0.4, λg = 0.6, AUC = 0.628
0.0 0.0 0.0 0.0
0.0 0.2 0.4 0.6 0.8 1.0 0.0 0.2 0.4 0.6 0.8 1.0 0.0 0.2 0.4 0.6 0.8 1.0 0.0 0.2 0.4 0.6 0.8 1.0
False Alarm Rate False Alarm Rate False Alarm Rate False Alarm Rate

(a) 25% bad mouthing attacks (b) 50% bad mouthing attacks (a) offset = 0.1 (b) offset = 0.2

1.0 1.0 Savg (AUC 0.285) 1.0 1.0

Sdef (AUC 0.471)
0.8 0.8 0.8 0.8
Srec (AUC 0.306)

Detection Rate

Detection Rate
Detection Rate

Detection Rate

0.6 0.6 0.6 0.6

0.4 0.4 0.4 λb = 0, λg = 0.3, AUC = 0.754 0.4 λb = 0, λg = 0.4, AUC = 0.734
λb = 0.1, λg = 0.4, AUC = 0.705 λb = 0.1, λg = 0.5, AUC = 0.706
Savg (AUC 0.799)
0.2 0.2 0.2 λb = 0.2, λg = 0.5, AUC = 0.680 0.2 λb = 0.2, λg = 0.6, AUC = 0.674
Sdef (AUC 0.418) λb = 0.3, λg = 0.6, AUC = 0.643 λb = 0.3, λg = 0.7, AUC = 0.666
Srec (AUC 0.728) λb = 0.4, λg = 0.7, AUC = 0.631 λb = 0.4, λg = 0.8, AUC = 0.639
0.0 0.0 0.0 0.0
0.0 0.2 0.4 0.6 0.8 1.0 0.0 0.2 0.4 0.6 0.8 1.0 0.0 0.2 0.4 0.6 0.8 1.0 0.0 0.2 0.4 0.6 0.8 1.0
False Alarm Rate False Alarm Rate False Alarm Rate False Alarm Rate

(c) 75% bad mouthing attacks (d) 100% bad mouthing attacks (c) offset = 0.3 (d) offset = 0.4

Fig. 5: Performance of the proposed attack model against three Fig. 7: Effectiveness of asymmetric forgetting speeds. Offset
defense schemes. Savg , Sdef and Srec represents avg scheme, represents the difference between forgetting speeds λb and λg .
def scheme and rec scheme, respectively.

validating the effectiveness of the proposed defense scheme.

1.0 2) Effectiveness of Asymmetric Forgetting Speeds: Next,
we evaluate the effectiveness of asymmetric forgetting speeds
0.8 in Figure 7. Observed from Figure 6, the defense scheme
with λ = 0.3 shows the best performance. Therefore, in this
Detection Rate

subsection, the value range of λb is set from 0 to 0.4. As we

0.6 propose to forget bad behavior slower, the λg values are set to
be 0.1, 0.2, 0.3 and 0.4 higher than λb . The results are shown
0.4 in Figure 7.
λb = λg = 0, AUC = 0.564 Comparing the schemes in Figure 6 and Figure 7, it can be
0.2 λb = λg = 0.3, AUC = 0.630 observed that 79% of the schemes with asymmetric forgetting
λb = λg = 0.8, AUC = 0.611 speeds in Figure 7 perform better than the best scheme (i.e.
λb = λg = 100, AUC = 0.558 λb = λg = 0.3) in Figure 6. In particular, the AUC of
0.0 the best scheme (i.e. λb = 0 and λg = 0.2) in Figure 7
0.0 0.2 0.4 0.6 0.8 1.0 (b) is 20% higher than that of the best scheme in Figure 6.
False Alarm Rate This observation validates the effectiveness of the asymmetric
forgetting speed.
Fig. 6: Effectiveness of various forgetting speeds Furthermore, an appropriate offset (e.g. 0.2) between the
two forgetting speeds makes the scheme perform better.
that the defense scheme only remembers the most recent Specifically, the scheme with λb = 0 and λg = 0.2 in Figure
behavior for trust evaluation. In such settings, the defense 7 (b) performs the best with the AUC as 0.758, which is 13%,
scheme can mistakenly identify normal nodes with accidental 0.5%, and 3% higher than the highest AUCs in Figure 7 (a),
packet losses as malicious nodes and therefore results in high (c), and (d). When the offset is too large, the scheme will be
false alarm rates, as shown by the green dotted curve in too sensitive to bad behaviors, which may increase the false
Figure 6. On the other hand, when the value of λ is 0, the alarm rate. When the offset is too small, the performance will
defense scheme, which remembers all previous behaviors, also be very similar to that of the scheme with identical λb and λg ,
performs worse, as shown by the blue dash dot line in Figure which is less effective.
6. This is because malicious nodes can easily mislead the 3) Effectiveness of Integrating Self-Trust and Descendant
defense scheme by accumulating high trust values through Trust: To detect bad-mouthing attacks, where the parent node
good behaviors performed long time ago. frames its children nodes by discarding their data packets, we
With an appropriate λ value, the proposed defense scheme propose to integrate self-trust (T s ) and descendant trust (T d ).
can achieve high performance. Particularly, as shown in Figure This subsection aims to evaluate the trade-off by adjusting
5 (b), when λ = 0.3, the AUC of the scheme with is 4% and the weights of these two aspects. In particular, ws and wd
11% higher than that of the avg and rec schemes respectively, represent the weights assigned to T s and T d , respectively,
 10

1.0 1.0

0.8 0.8
Detection Rate

Detection Rate
0.6 0.6

0.4 ws = 0.9, wd = 0.1, AUC = 0.647 0.4

ws = 0.7, wd = 0.3, AUC = 0.680 HB (AUC 0.511)
0.2 ws = 0.5, wd = 0.5, AUC = 0.713 0.2 TPPR (AUC 0.691)
ws = 0.3, wd = 0.7, AUC = 0.758 Proposed scheme (AUC 0.758)
ws = 0.1, wd = 0.9, AUC = 0.733 MRHOF (AUC 0.430)
0.0 0.0
0.0 0.2 0.4 0.6 0.8 1.0 0.0 0.2 0.4 0.6 0.8 1.0
False Alarm Rate False Alarm Rate

Fig. 8: Comparison of effectiveness of combination of two Fig. 9: Comparison of performance of different scheme based
trust factors on proposed attack model
9 4.0
where the sum of ws and wd equals one. The values of ws
and wd are in the range of [0.1, 0.9] with the step interval as 8
Power Consumption
3.5
Detection Delay

Detection Delay (log10t, s)

0.1. The values of λb and λg are fixed as 0 and 0.2 (i.e. the
optimal values from Figure 7), respectively.
Power Consumption (mW)
7 3.0
From Figure 8, the proposed defense scheme with ws = 0.3 6
2.5
and wd = 0.7 achieves the best performance. Furthermore,
when ws ∈ [0.2, 0.4] and wd ∈ [0.6, 0.8], the AUC of the 5
2.0
schemes are higher than 0.746, which validates the robustness 4
of the proposed defense scheme. 1.5
More importantly, the AUC of the best scheme in Figure 3
1.0
8 is 26% and 34% higher than the avg scheme and rec 2
scheme in Figure 5 respectively. This observation validates
1 0.5
that the introduction of descendant trust value (T d) enables the
proposed defense scheme to effectively detect bad-mouthing 0 0.0
attacks in the network. MRHOF Proposed Scheme HP TPRP
Some of the schemes in Figure 8 have an inflection point
in the detection rate range of [0.55, 0.65], indicating that the Fig. 10: Power consumption and detection delay comparison
increment of detection rate slows down. This is because this based on proposed attack mode
experiment places a limited number of nodes, including three
malicious nodes, to prevent exceeding the capacity of the messages and thus cannot effectively capture the loss of data
simulation platform (e.g. memory overflow). packets. Furthermore, when the false alarm rate is lower than
0.4, the detection rate of TPRP scheme stays high. This is
because the TPRP scheme can effectively defend against bad-
D. Overall Performance Comparison mouthing attacks by requiring each node in the network to
This section compares the proposed defense scheme with monitor the sending and receiving packets of its neighbor
the state-of-the-art defense schemes, including: (1) the RPL nodes. However, the detection rate of TPRP scheme cannot
default recovering scheme: MRHOF [5], (2) a centralized be significantly improved. This is because it adopts average
scheme: Heartbeat protocol (HP) [10], and (3) a distributed PFR as the trust value, which cannot effectively capture attacks
scheme: Trust-Aware RPL Routing Protocol (TPRP) [12]. alternatively perform good and bad behaviors.
All the schemes are applied on the same network topology 2) Energy Consumption and Detection Delay: Figure 10
and settings. In addition, the proposed defense scheme with compares the detection delay and energy consumption of le-
λb = 0, λg = 0.2, ws = 0.3 and wd = 0.7 is launched. gitimate non-root nodes for different defense schemes. Specif-
1) Detection Accuracy: As shown in Figure 9, the pro- ically, the proposed attacks are launched against these defense
posed scheme yields the best overall performance among four schemes. Each defense scheme has two bars, representing its
defense schemes. Its AUC is 76%, 48% and 10%, higher than energy consumption and detection delay respectively.
that of MRHOF, HP and TPRP, respectively. From Figure 10, we can observe that the proposed defense
In particular, the HP and MRHOF schemes fail to differ- scheme consumes very limited power, only 3.4% more than the
entiate malicious nodes from normal ones, because these two default RPL recovery scheme (i.e. MRHOF). This is because
schemes detect malicious nodes based on replies of the control the proposed scheme adopts the centralized design, where non-
 11

root nodes do not need to monitor neighbor nodes’ activities, descendant trust value based on the RPL tree-like topology.
but only forward notification packets on detected malicious Experimental results show that compared to the state-of-the-art
nodes. defense solutions, the proposed defense scheme can effectively
On the other hand, the HP and TPRP schemes are causing detect advanced proposed attacks with very limited energy
extra 50% and 957% power respectively when compared to consumption.
the default MRHOF scheme. Although the HP scheme also
adopts a centralized defense mechanism, it frequently launches R EFERENCES
a “request and reply” process between the root node and non- [1] K. Rose, S. Eldridge, and L. Chapin, “The internet of things: An
root nodes, which increases power consumption. Furthermore, overview,” The internet society (ISOC), vol. 80, pp. 1–50, 2015.
the power consumption of TPRP scheme is the highest due [2] F. Pereira, R. Correia, P. Pinho, S. I. Lopes, and N. B. Carvalho, “Chal-
lenges in resource-constrained iot devices: Energy and communication
to its distributed design, which requires each non-root node as critical success factors for future iot deployment,” Sensors, vol. 20,
in the network to monitor, analyze, and share the activities of no. 22, p. 6420, 2020.
its neighbors. These requirements significantly increase extra [3] B. Marques and M. Ricardo, “Energy-efficient node selection in
work time and computational costs for each node, leading to application-driven wsn,” Wireless Networks, vol. 23, no. 3, pp. 889–918,
2017.
much higher power consumption. [4] A. Le, J. Loo, A. Lasebae, M. Aiash, and Y. Luo, “6lowpan: a study
The detection delay is calculated based on only the suc- on qos security threats and countermeasures using intrusion detection
system approach,” International Journal of Communication Systems,
cessful detection of each scheme. In other words, if a de- vol. 25, no. 9, pp. 1189–1212, 2012.
tection scheme can only detect two malicious nodes out of [5] T. Winter, P. Thubert, A. Brandt, J. W. Hui, R. Kelsey, P. Levis, K. Pister,
three, the detection delay is the average delay of the two R. Struik, J.-P. Vasseur, R. K. Alexander et al., “Rpl: Ipv6 routing
protocol for low-power and lossy networks.” rfc, vol. 6550, pp. 1–157,
successful detection. As shown in Figure 10, the MRHOF and 2012.
HP schemes show small detection delay, which, however, is [6] F. Gara, L. B. Saad, R. B. Ayed, and B. Tourancheau, “Rpl protocol
calculated based on the very limited malicious nodes that can adapted for healthcare and medical applications,” in 2015 International
wireless communications and mobile computing conference (IWCMC).
be detected. Furthermore, there is a relatively large variation in IEEE, 2015, pp. 690–695.
the detection delay of the HP scheme. This is because the HP [7] J. Nassar, M. Berthomé, J. Dubrulle, N. Gouvy, N. Mitton, and
scheme relies on the exchange of “request-reply” messages to B. Quoitin, “Multiple instances qos routing in rpl: Application to smart
grids,” Sensors, vol. 18, no. 8, p. 2472, 2018.
detect anomaly. The detection delay may vary based on the [8] S. Junior, A. Riker, B. Silvestre, W. Moreira, A. Oliveira-Jr, and
frequency of the request messages. A higher frequency may V. Borges, “Dynasti—dynamic multiple rpl instances for multiple iot
lead to smaller detection delay, but higher power consumption. applications in smart city,” Sensors, vol. 20, no. 11, p. 3130, 2020.
[9] P. Pongle and G. Chavan, “A survey: Attacks on rpl and 6lowpan in
In addition, the proposed scheme yields similar detection iot,” in 2015 International conference on pervasive computing (ICPC).
delay as the TPRP scheme, but a larger variation. This is IEEE, 2015, pp. 1–6.
because in the TPRP scheme, rather than relying on notifi- [10] L. Wallgren, S. Raza, and T. Voigt, “Routing attacks and counter-
measures in the rpl-based internet of things,” International Journal of
cations from the root node, each node directly monitors its Distributed Sensor Networks, vol. 9, no. 8, p. 794326, 2013.
neighbors’ behaviors, resulting in a relatively stable delay. The [11] G. Ma, X. Li, Q. Pei, and Z. Li, “A security routing protocol for internet
proposed scheme, however, can only detect anomaly when the of things based on rpl,” in 2017 International Conference on Networking
and Network Applications (NaNA). IEEE, 2017, pp. 209–213.
data packets arriving at the root node show abnormal patterns, [12] D. Airehrour, J. Gutierrez, and S. K. Ray, “A trust-aware rpl routing
which may vary according to the network’s data rate and the protocol to detect blackhole and selective forwarding attacks,” Journal
source node’s network location. of Telecommunications and the Digital Economy, vol. 5, no. 1, pp. 50–
69, 2017.
In summary, compared to the state-of-the-art defense [13] S. Raza, L. Wallgren, and T. Voigt, “Svelte: Real-time intrusion detection
schemes, the proposed scheme yields much higher detection in the internet of things,” Ad hoc networks, vol. 11, no. 8, pp. 2661–
accuracy (i.e. 10% higher than the second highest one) and 2674, 2013.
[14] A. Mayzaud, R. Badonnel, and I. Chrisment, “A taxonomy of attacks in
lower energy consumption (i.e. 31% lower than that of the rpl-based internet of things,” International Journal of Network Security,
second lowest one). Although its detection delay is higher vol. 18, no. 3, pp. 459–473, 2016.
than other schemes (i.e. the MRHOF and HP schemes), it [15] Z. A. Almusaylim, A. Alhumam, and N. Jhanjhi, “Proposing a secure rpl
based internet of things routing protocol: a review,” Ad Hoc Networks,
is practical to be applied in a low power low data rate RPL vol. 101, p. 102096, 2020.
network. [16] A. Le, J. Loo, Y. Luo, and A. Lasebae, “The impacts of internal threats
towards routing protocol for low power and lossy network performance,”
in 2013 IEEE Symposium on computers and communications (ISCC).
VII. C ONCLUSION IEEE, 2013, pp. 000 789–000 794.
[17] W. Xie, M. Goyal, H. Hosseini, J. Martocci, Y. Bashir, E. Baccelli, and
In RPL networks, malicious nodes can damage routing A. Durresi, “Routing loops in dag-based low power and lossy networks,”
paths by selectively dropping packets. In this paper, we in 2010 24th IEEE International Conference on Advanced Information
Networking and Applications. IEEE, 2010, pp. 888–895.
propose an advanced selective forwarding attack with three [18] F. Gara, L. B. Saad, and R. B. Ayed, “An intrusion detection system
flexible attack behaviors, including protocol selection, packet for selective forwarding attack in ipv6-based mobile wsns,” in 2017
forward rate selection and bad-mouthing with children nodes 13th International Wireless Communications and Mobile Computing
Conference (IWCMC). IEEE, 2017, pp. 276–281.
selection. The flexibility enables the malicious nodes to hide [19] Y. Hu, Y. Wu, H. Wang et al., “Detection of insider selective forwarding
their attack behaviors and maximize the long term attack attack based on monitor node and trust mechanism in wsn,” Wireless
impact. Furthermore, we propose a new centralized trust-based Sensor Network, vol. 6, no. 11, p. 237, 2014.
[20] A. Mathur, T. Newe, and M. Rao, “Defence against black hole and
defense scheme, which consists of a self-trust based on the selective forwarding attacks for medical wsns in the iot,” Sensors,
beta trust model with asymmetric forgetting speeds, and a vol. 16, no. 1, p. 118, 2016.
 12

[21] J. Ren, Y. Zhang, K. Zhang, and X. Shen, “Adaptive and channel-aware

detection of selective forwarding attacks in wireless sensor networks,”
IEEE Transactions on Wireless Communications, vol. 15, no. 5, pp.
3718–3731, 2016.
[22] D. Airehrour, J. Gutierrez, and S. K. Ray, “Securing rpl routing protocol
from blackhole attacks using a trust-based mechanism,” in 2016 26th
International Telecommunication Networks and Applications Conference
(ITNAC). IEEE, 2016, pp. 115–120.
[23] M. A. Lodhi, A. Rehman, M. M. Khan, and F. B. Hussain, “Multiple path
rpl for low power lossy networks,” in 2015 IEEE Asia Pacific Conference
on Wireless and Mobile (APWiMob). IEEE, 2015, pp. 279–284.
[24] T. L. Jenschke, R.-A. Koutsiamanis, G. Z. Papadopoulos, and N. Mon-
tavont, “Multi-path selection in rpl based on replication and elimina-
tion,” in International Conference on Ad-Hoc Networks and Wireless.
Springer, 2018, pp. 15–26.
[25] C. Pu and S. Hajjar, “Mitigating forwarding misbehaviors in rpl-based
low power and lossy networks,” in 2018 15th IEEE Annual Consumer
Communications & Networking Conference (CCNC). IEEE, 2018, pp.
1–6.
[26] Z. A. Khan, J. Ullrich, A. G. Voyiatzis, and P. Herrmann, “A trust-based
resilient routing mechanism for the internet of things,” in Proceedings
of the 12th International Conference on Availability, Reliability and
Security, 2017, pp. 1–6.
[27] F. Medjek, D. Tandjaoui, I. Romdhani, and N. Djedjig, “A trust-based
intrusion detection system for mobile rpl based networks,” in 2017
IEEE International Conference on Internet of Things (iThings) and
IEEE Green Computing and Communications (GreenCom) and IEEE
Cyber, Physical and Social Computing (CPSCom) and IEEE Smart Data
(SmartData). IEEE, 2017, pp. 735–742.
[28] N. Djedjig, D. Tandjaoui, F. Medjek, and I. Romdhani, “Trust-aware
and cooperative routing protocol for iot security,” Journal of Information
Security and Applications, vol. 52, p. 102467, 2020.
[29] T. ul Hassan, M. Asim, T. Baker, J. Hassan, and N. Tariq, “Ctrust-rpl:
A control layer-based trust mechanism for supporting secure routing
in routing protocol for low power and lossy networks-based internet
of things applications,” Transactions on Emerging Telecommunications
Technologies, vol. 32, no. 3, p. e4224, 2021.
[30] A. K. Gautam and R. Kumar, “An efficient packet reachability-based
trust management scheme in wireless sensor networks,” in Safety,
Security, and Reliability of Robotic Systems. CRC Press, 2020, pp.
141–154.
[31] J. Jiang, Y. Liu, and B. Dezfouli, “A root-based defense mechanism
against rpl blackhole attacks in internet of things networks,” in 2018
Asia-Pacific Signal and Information Processing Association Annual
Summit and Conference (APSIPA ASC). IEEE, 2018, pp. 1194–1199.
[32] S. Raza, S. Duquennoy, T. Chung, D. Yazar, T. Voigt, and U. Roedig,
“Securing communication in 6lowpan with compressed ipsec,” in 2011
International Conference on Distributed Computing in Sensor Systems
and Workshops (DCOSS). IEEE, 2011, pp. 1–8.
[33] A. Josang and R. Ismail, “The beta reputation system,” in Proceedings
of the 15th bled electronic commerce conference, vol. 5, 2002, pp. 2502–
2511.
[34] N. Labraoui, M. Gueroui, and L. Sekhri, “On-off attacks mitigation
against trust systems in wireless sensor networks,” in IFIP International
Conference on Computer Science and its Applications. Springer, 2015,
pp. 406–415.
[35] D. Shreenivas, S. Raza, and T. Voigt, “Intrusion detection in the rpl-
connected 6lowpan networks,” in Proceedings of the 3rd ACM interna-
tional workshop on IoT privacy, trust, and security, 2017, pp. 31–38.
[36] Y. B. Zikria, M. K. Afzal, F. Ishmanov, S. W. Kim, and H. Yu, “A
survey on routing protocols supported by the contiki internet of things
operating system,” Future Generation Computer Systems, vol. 82, pp.
200–219, 2018.