idnits 2.17.1 draft-schulzrinne-p2prg-rtc-security-00.txt: Checking boilerplate required by RFC 5378 and the IETF Trust (see https://trustee.ietf.org/license-info): ---------------------------------------------------------------------------- No issues found here. Checking nits according to https://www.ietf.org/id-info/1id-guidelines.txt: ---------------------------------------------------------------------------- No issues found here. Checking nits according to https://www.ietf.org/id-info/checklist : ---------------------------------------------------------------------------- ** The document seems to lack an IANA Considerations section. (See Section 2.2 of https://www.ietf.org/id-info/checklist for how to handle the case when there are no actions for IANA.) Miscellaneous warnings: ---------------------------------------------------------------------------- == The copyright year in the IETF Trust and authors Copyright Line does not match the current year -- The document date (February 22, 2009) is 5542 days in the past. Is this intentional? Checking references for intended status: Informational ---------------------------------------------------------------------------- == Unused Reference: 'SHA1' is defined on line 945, but no explicit reference was found in the text -- Obsolete informational reference (is this intentional?): RFC 2818 (Obsoleted by RFC 9110) Summary: 1 error (**), 0 flaws (~~), 2 warnings (==), 2 comments (--). Run idnits with the --verbose option for more detailed information about the items above. -------------------------------------------------------------------------------- 2 Network Working Group H. Schulzrinne 3 Internet-Draft Columbia University 4 Intended status: Informational E. Marocco 5 Expires: August 26, 2009 Telecom Italia 6 E. Ivov 7 SIP Communicator 8 February 22, 2009 10 Security Issues and Solutions in Peer-to-peer Systems for Realtime 11 Communications 12 draft-schulzrinne-p2prg-rtc-security-00 14 Status of this Memo 16 This Internet-Draft is submitted to IETF in full conformance with the 17 provisions of BCP 78 and BCP 79. This document may not be modified, 18 and derivative works of it may not be created, except to format it 19 for publication as an RFC and to translate it into languages other 20 than English. 22 Internet-Drafts are working documents of the Internet Engineering 23 Task Force (IETF), its areas, and its working groups. Note that 24 other groups may also distribute working documents as Internet- 25 Drafts. 27 Internet-Drafts are draft documents valid for a maximum of six months 28 and may be updated, replaced, or obsoleted by other documents at any 29 time. It is inappropriate to use Internet-Drafts as reference 30 material or to cite them other than as "work in progress." 32 The list of current Internet-Drafts can be accessed at 33 http://www.ietf.org/ietf/1id-abstracts.txt. 35 The list of Internet-Draft Shadow Directories can be accessed at 36 http://www.ietf.org/shadow.html. 38 This Internet-Draft will expire on August 26, 2009. 40 Copyright Notice 42 Copyright (c) 2009 IETF Trust and the persons identified as the 43 document authors. All rights reserved. 45 This document is subject to BCP 78 and the IETF Trust's Legal 46 Provisions Relating to IETF Documents 47 (http://trustee.ietf.org/license-info) in effect on the date of 48 publication of this document. Please review these documents 49 carefully, as they describe your rights and restrictions with respect 50 to this document. 52 Abstract 54 Peer-to-peer (P2P) networks offer higher robustness against failure, 55 easier configuration and are generally more economical than their 56 client-server counterparts. It has therefore become reasonable for 57 resource consuming and typically centralized applications like Voice 58 over IP (VoIP) and, in general, realtime communication to adapt and 59 exploit the benefits of P2P. Such a migration needs to address a new 60 set of P2P specific security problems. This document describes some 61 of the known issues found in common P2P networks, analyzing the 62 relevance of such issues and the applicability of existing solutions 63 when using P2P architectures for realtime communication. 65 Table of Contents 67 1. Introduction . . . . . . . . . . . . . . . . . . . . . . . . . 4 68 1.1. Purpose of this document . . . . . . . . . . . . . . . . . 6 69 2. The attackers . . . . . . . . . . . . . . . . . . . . . . . . 6 70 2.1. Incentive of the attacker . . . . . . . . . . . . . . . . 6 71 2.2. Resources available to the attacker . . . . . . . . . . . 7 72 2.3. Victim of the attack . . . . . . . . . . . . . . . . . . . 7 73 2.4. Time of attack . . . . . . . . . . . . . . . . . . . . . . 7 74 3. Admission control . . . . . . . . . . . . . . . . . . . . . . 8 75 4. Determining the position in the overlay . . . . . . . . . . . 9 76 5. Resilience against malicious peers . . . . . . . . . . . . . . 10 77 5.1. Identification of malicious peers . . . . . . . . . . . . 10 78 5.1.1. Proactive identification . . . . . . . . . . . . . . . 10 79 5.1.2. Reactive identification . . . . . . . . . . . . . . . 11 80 5.2. Reputation management systems . . . . . . . . . . . . . . 11 81 5.2.1. Unstructured reputation management . . . . . . . . . . 11 82 5.2.2. Structured reputation management . . . . . . . . . . . 12 83 6. Routing and data integrity . . . . . . . . . . . . . . . . . . 12 84 6.1. Data integrity . . . . . . . . . . . . . . . . . . . . . . 12 85 6.2. Routing integrity . . . . . . . . . . . . . . . . . . . . 13 86 7. Peer-to-peer in realtime communication . . . . . . . . . . . . 13 87 7.1. Admission . . . . . . . . . . . . . . . . . . . . . . . . 14 88 7.1.1. Active vs. passive upgrades . . . . . . . . . . . . . 14 89 7.1.2. When to upgrade . . . . . . . . . . . . . . . . . . . 15 90 7.1.3. Which clients to upgrade . . . . . . . . . . . . . . . 15 91 7.1.4. Incentives for clients . . . . . . . . . . . . . . . . 15 92 7.2. Security . . . . . . . . . . . . . . . . . . . . . . . . . 16 93 7.2.1. Targeted denial of service . . . . . . . . . . . . . . 16 94 7.2.2. Man in the middle attack . . . . . . . . . . . . . . . 16 95 7.2.3. Trust between peers . . . . . . . . . . . . . . . . . 16 96 7.2.4. Routing call signalization . . . . . . . . . . . . . . 17 97 7.2.5. Integrity of location bindings . . . . . . . . . . . . 17 98 7.2.6. Encrypting content . . . . . . . . . . . . . . . . . . 18 99 7.2.7. Other issues . . . . . . . . . . . . . . . . . . . . . 18 100 8. Security Considerations . . . . . . . . . . . . . . . . . . . 19 101 9. Acknowledgments . . . . . . . . . . . . . . . . . . . . . . . 19 102 10. Informative references . . . . . . . . . . . . . . . . . . . . 19 103 Authors' Addresses . . . . . . . . . . . . . . . . . . . . . . . . 22 105 1. Introduction 107 Peer to Peer (P2P) overlays have become quite popular with the advent 108 of file-sharing applications such as Napster [NAPSTER], KaZaa [KAZAA] 109 and BitTorrent [BITTORRENT]. After their success in file-sharing and 110 content distribution [Androutsellis-Theotokis], P2P networks are now 111 also being used for applications such as Voice over IP (VoIP) [SKYPE] 112 [Singh] and television [JOOST] [COOLSTREAM]. However most of these 113 systems are not purely P2P and have centralized components like the 114 login server in Skype [Baset] or moderators and trackers in 115 BitTorrent [Pouwelse]. Securing pure P2P networks is therefore still 116 a field of very active research [Wallach]. P2P overlays can be 117 broadly classified as structured and unstructured. Unstructured 118 overlays are often relatively simple but search operations in them 119 tend to be inefficient. Structured P2P overlays use distributed hash 120 tables (DHT) to perform directed searches which make lookups more 121 efficient in locating data. This document will mostly focus on DHT- 122 based P2P overlays. 124 When analyzing the various attacks that are possible on P2P systems, 125 it is important to first understand the motivation of the attackers 126 as well as the resources (i.e. computation power, access to different 127 IP subnets) that they would have at their disposal. 129 Once the threat has been identified, admission control is the first 130 step towards security [Kim]. Most solutions rely on the assumption 131 that malicious nodes represent a small fraction of all peers. It is 132 therefore important to restrict their number in the overlay. 134 Other P2P specific security problems discussed here include attacks 135 on the routing of queries, targeted denial of service attacks and 136 attacks on data integrity. 138 This document, after discussing some of the main security issues and 139 proposed solutions for P2P systems in general, focuses on one 140 particular application -- realtime communication. The idea behind 141 P2P realtime communication is using the DHTs employed by file-sharing 142 applications, in order to implement services such as registration, 143 user location lookup, and assistance with NAT and firewall traversal. 144 Even if, from a technical point of view, P2P communication services 145 may seem similar to file-sharing, Table 1 shows that some important 146 differences, mostly related to privacy and availability, 147 significantly increase security requirements. 149 +-----------------+-----------------------+-------------------------+ 150 | | File-sharing | Realtime communication | 151 +-----------------+-----------------------+-------------------------+ 152 | Distributed | Shared file locations | User locations are | 153 | database | are indexed in a | indexed in a table | 154 | | table distributed | distributed among | 155 | | among peers; often | peers; rarely more than | 156 | | hundreds or thousands | one per user. | 157 | | per user. | | 158 | Availability | Same files are | Users are unique; | 159 | | usually available at | attacks targeting | 160 | | multiple locations | single users may be | 161 | | and failures | addressed both to the | 162 | | involving single | distributed index and | 163 | | istances are overcame | to the user's device | 164 | | by abundancy of | directly. | 165 | | resources; attacks | | 166 | | targeting single | | 167 | | files need to be | | 168 | | addressed to the | | 169 | | distributed index. | | 170 | Integrity | Attackers may want to | Attackers may want to | 171 | | share corrupted files | impersonate different | 172 | | in place of popular | users in order to | 173 | | content, e.g. to | handle calls directed | 174 | | discourage users from | to them; constitute a | 175 | | acquiring copyrighted | particular threat for | 176 | | material; constitute | the user as, in case of | 177 | | a threat for the | success, the attacker | 178 | | service, but not for | acquires full control | 179 | | the users. | on the victim's | 180 | | | personal | 181 | | | communications. | 182 | Confidentiality | Shared files are, by | Communications are | 183 | | definition, readable | usually meant to be | 184 | | by all users; in some | private and need to be | 185 | | cases encryption is | encrypted; evesdropping | 186 | | used to avoid | may reveal sensitive | 187 | | elements not involved | data and is a serious | 188 | | in the service to | threat for users. | 189 | | detect traffic. | | 190 +-----------------+-----------------------+-------------------------+ 192 Main differences between P2P applications used for file-sharing and 193 for realtime communication. 195 Table 1 197 The rest of the document is organized as follows. In Section 2, we 198 discuss P2P security attackers. We try to elaborate on their 199 motivation, the resources that would generally be available to them, 200 their victims and the timing of their attacks. In Section 3, we 201 discuss admission control problems. In Section 4, we identify the 202 problem of where a node joins in the overlay. In Section 5, we 203 describe problems related to identification of malicious nodes and 204 the dissemination of this information. In Section 6, we describe the 205 issues of routing and data integrity in P2P networks. Finally, in 206 Section 7 we discuss how issues and solutions previously presented 207 apply in P2P overlays for realtime communication. 209 1.1. Purpose of this document 211 This document is partially derived from the article "Peer-to-peer 212 Overlays for Real-Time Communications: Issues and Solutions," 213 published in IEEE Surveys & Tutorials, Vol. 11, No. 1 and originally 214 authored by Dhruv Chopra, Henning Schulzrinne, Enrico Marocco and 215 Emil Ivov. Its goal is to collect feedback from the IRTF community 216 in order to document the advances in the field of security of P2P 217 systems for realtime communications, for the benefit of related 218 standardization activities going on in IETF. 220 2. The attackers 222 2.1. Incentive of the attacker 224 Attacks on networks happen for a variety of reasons such as monetary 225 gain, personal enmity or even for fame in the hacker community. 226 There are quite a few well known cases of denial of service attacks 227 for extortion in the client-server model [McCue]. One of the salient 228 points of the P2P model is that the services it provides have higher 229 robustness against failure. However, such attacks are still possible 230 against individuals within the overlay if the attackers possess 231 sufficient resources. For instance, a network of worm-affected 232 malicious nodes spread across the Internet and controlled by an 233 attacker (often referred as botnet), could simultaneously bombard 234 lookup queries for a particular key in the DHT. The peer responsible 235 for this key would then come under a lot of load and could crash 236 [Sit]. However with replication of key-value pairs at multiple 237 locations, such threats can be mitigated. 239 Attackers may also have other incentives apart from money. With the 240 growth of illegal usage of sharing files with copyrights, record 241 companies have been known to attempt polluting content in the 242 overlays by putting up nodes with corrupt chunks of data but with 243 correct file names to degrade the service [Liang] and in hope that 244 users would get frustrated and stop using the service. Attacks can 245 also be launched by novice attackers who are there attacking the 246 overlay for fun or fame in a community. These are perhaps less 247 likely to be successful or cause damage, since their resources tend 248 to be relatively limited. 250 2.2. Resources available to the attacker 252 Resource constraints play an important role in determining the nature 253 of the attack. An attacker who controls a botnet can use an Internet 254 relay channel and launch distributed denial of service attacks 255 against another node. With respect to attacks where a single node 256 impersonates multiple identities, as in the case of the sybil attack 257 [Douceur] described in Section 4, IP addresses are also an important 258 resource for the attacker since in DHTs such as Chord [Stoica], the 259 position in the overlay is determined by using a base hash function 260 such as SHA-1 [SHA1]on the node's IP address. The cryptographic 261 puzzles [Rowaihy] that are sometimes suggested as a way to deter 262 sybil attacks by making the join process harder are futile against an 263 attacker with a botnet and virtually unlimited computation power. 264 Doucer [Douceur] proves that even with the assumption that attackers 265 only have minimum resources at their disposal, it is not possible to 266 defend against them in a pure P2P system. 268 2.3. Victim of the attack 270 The victim of an attack could be an individual node, a particular 271 content entry or the entire overlay service. If malicious nodes are 272 strategically placed in the overlay, they can block a node from using 273 its services. Attacks could also be launched against specific 274 content [Sit] or even the entire overlay service. For example, if 275 the malicious nodes are randomly placed in the overlay and drop 276 packets or upload malcontent, then the quality of the overlay would 277 deteriorate. 279 2.4. Time of attack 281 A malicious node could start misbehaving as soon as it enters the 282 overlay or it could follow the rules of the overlay for a finite 283 amount of time and then attack. The latter could prove to be more 284 harmful if the overlay design suggests accumulating trust in peers 285 based on the amount of time they have been present and/or not 286 misbehaving. In Kademlia [KADEMLIA], for instance, the routing 287 tables are populated with nodes that have been up for a certain 288 amount of time. While this provides some robustness from attacks in 289 which the malicious nodes start dropping routing requests from the 290 moment they enter, it would take time for the algorithm to adapt to 291 nodes which start misbehaving in a later stage (i.e., after they have 292 been recorded in routing tables). Similarly for reputation 293 management systems, it is important that they adapt to the current 294 behavior of a peer. 296 3. Admission control 298 Admission control depends on who decides whether or not to admit a 299 node and how this permission is granted. Kim et. al [Kim] answer 300 these questions independently of any particular environment or 301 application. They define two basic elements for admission in a peer 302 group, a group charter, which is an electronic document that 303 specifies the procedure of admission into the overlay, and a group 304 authority, which is an entity that can certify group admission. A 305 prospective member first gets a copy of the group charter, satisfies 306 the requirements and approaches the group authority. The group 307 authority then verifies the admission request and grants a group 308 membership certificate. 310 The group charter and authority verification can be provided by a 311 centralized certificate authority or a trusted third party, or it 312 could be provided by the peers themselves (by voting). The former is 313 more practical and tends to make the certification process simpler 314 although it is in violation of the pure P2P model and exposes the 315 system to attacks typical for server-based solutions (e.g., denial of 316 service attacks targeted to the central authority). The latter, the 317 group authority could either be a fixed number of peers or it could 318 be a dynamic number based on the total membership of the group. The 319 authors argue that even if the group charter requires a prospective 320 member to get votes from peers, the group membership certificate must 321 be issued by a distinct entity. The reason for this is that voters 322 need to accompany their votes with a certificate that proves their 323 own membership. Possible signature schemes that could be used in 324 voting such as plain digital signature, threshold signature and 325 accountable subgroup multisignature are also described. Saxena et. 326 al [Saxena] performed experiments with the different signature 327 schemes and suggest the use of plain signatures for groups of 328 moderate size and where bandwidth is not a concern. For larger 329 groups and where bandwidth is a concern, they suggest threshold 330 signature [Kong] and multisignature schemes [Ohta]. 332 Another way of handling admission would be to use mechanisms based on 333 trust and recommendation where each new applicant has to be known and 334 vouched for by at least N existing members. The difficulties that 335 such models represent include identity assertion and preventing bot/ 336 worm attacks. A compromised node could have a valid certificate 337 identifying a trustworthy peer and it would be difficult to detect 338 this. Possible solutions include sending graphic or logic puzzles 339 easily addressed by humans but hard to solve by computers, also known 340 as CAPTCHA [Ahn]. 342 4. Determining the position in the overlay 344 For ring based DHT overlays such as Chord [Stoica], Kademlia 345 [KADEMLIA] and Pastry [PASTRY], when a node joins the overlay, it 346 uses a numeric identifier (ID) to determine its position in the ring. 347 The positioning of a node determines what information it stores and 348 which nodes it serves. To provide a degree of robustness, content 349 and services are often replicated across multiple nodes. However it 350 is possible for an adversary with sufficient resources to undermine 351 the redundancy deployed in the overlay by representing multiple 352 identities. Such an attack is called a sybil attack [Douceur]. This 353 makes the assignment of IDs very important. One possible scheme to 354 tackle such attacks on the ID mapping is to have a temporal mechanism 355 in which nodes need to re-join the network after some time [Condie] 356 [Scheideler]. Such temporal solutions, however have the drawback 357 that they increase the maintenance traffic and possibly deteriorate 358 the efficiency of caching. Danezis et. al [Danezis] suggest 359 mechanisms to mitigate the effect of sybil attacks by reducing the 360 amount of information received from malicious nodes. Their idea is 361 to vary the nodes used for routing with time and thus avoid a trust 362 bottleneck. Other solutions suggest making the joining process 363 harder by introducing cryptographic puzzles as suggested by Rowaihy 364 et. al [Rowaihy]. The assumption is that the adversary has limited 365 computational resources which may not be true if the adversary has 366 control over a botnet. Another drawback of such methods is that non- 367 malicious nodes would also have to perform the extra computations 368 before they can join the overlay. 370 A possible heuristic to hamper sybil attacks is to employ redundancy 371 at nodes with diametrically opposite IDs (in the DHT ID space) 372 instead of successive IDs as in Chord. The idea behind choosing 373 diametrically opposite nodes is based on the fact that a malicious 374 peer can grant admission to others as its successor without them 375 actually possessing the required IP address (whose hash is adjacent 376 to the former's), and then they can cooperate to control access to 377 that part of the ring. If however admission decisions and redundant 378 content (for robustness), also involve nodes which are the furthest 379 away (diametrically opposite) from a given position, then the 380 adversary would require double resources (IP addresses) to attack. 381 This happens because the adversary would need presence in the overlay 382 at two independent positions in the ring. 384 Another approach proposed by Yu et al [Yu]. to limit sybil attacks is 385 based on the usage of the social relations between users. Authors 386 use the fact that as a result of sybil attacks, affected P2P overlays 387 end up containing a large set of sybil nodes connected to the rest of 388 the peers through an irregularly small number of edges. The 389 SybilGuard protocol [Yu] defines a method that allows to discover 390 such kind of discontinuities in the topology by using a special kind 391 of a verifiable random walk and hence without the need of one node 392 having a global vision of the graph. 394 It is also worth mentioning that in DHT overlays using different 395 geometric concepts, (e.g., hypercubes instead of rings), peer 396 positions are usually not related to identifiers. In the content 397 addressable network (CAN) [Ratnasamy], for example, the position of 398 an entering node may be either selected by the node itself, or, with 399 little modification to the original algorithm, assigned by peers 400 already in the overlay. However, even when malicious nodes do not 401 know their position before joining, the overlay is still vulnerable 402 to sybil attacks. 404 5. Resilience against malicious peers 406 Making overlays robust against even a small percentage of malicious 407 nodes is difficult [Castro]. It is therefore important for other 408 peers to identify such nodes and keep track of their number. There 409 are two aspects to this problem. One is the identification itself 410 and the second is the dissemination of this information amongst the 411 peers. Different metrics need to be defined depending on the peer 412 group for the former and reputation management systems are needed for 413 the latter. 415 5.1. Identification of malicious peers 417 For identifying a node as malicious, malicious activity has to be 418 observed first. This could be done in either a proactive way, or a 419 reactive way. 421 5.1.1. Proactive identification 423 When acting proactively, peers perform periodic operations with the 424 purpose of detecting malicious activity. A malicious node could 425 prevent access to content it is responsible for (e.g., by claiming 426 the object doesn't exist), or return references to content that does 427 not match the original queries [Sit]. With this approach, publishers 428 of content can later perform lookups for it at periodic intervals and 429 verify the integrity of whatever is returned. Any inconsistencies 430 could then be interpreted as malicious activity. The problem with 431 proactive identification is the management of the overhead it 432 implies: if checks are performed too often, they may actually hinder 433 scalability, while, if they are performed too rarely, they would 434 probably be useless. 436 5.1.2. Reactive identification 438 In a reactive strategy, the peers perform normal operations and if 439 they happen to detect some malicious activity, then they can label 440 the responsible node as malicious. In a file-sharing application for 441 example, after downloading content from a node, if the peer observes 442 that data does not match its original query it can identify the 443 corresponding node as malicious. Poon et. al [Poon] suggest a 444 strategy based on the forwarding of queries. If routing is done in 445 an iterative way, then dropping of packets, forwarding to an 446 incorrect node and delay in forwarding arouse suspicion and the 447 corresponding peer is identified as malicious. 449 5.2. Reputation management systems 451 Reputation management systems are used to allow peers to share 452 information about other peers based on their own experience and thus 453 help in making better judgments. Most reputation management systems 454 proposed in the literature [Uzun] [Damiani] [Lee] [Kamvar] are for 455 file-sharing applications. In reputation systems, it should not be 456 possible for a misbehaving peer with low reputation to simply rejoin 457 the network with a different ID and therefore start from a clean 458 slate. To counter this, Kwon et. al [Lee] store not only the 459 reputation of a peer but also the reputation of files based on file 460 name and content to avoid spreading of a bad file. Another method is 461 to make the reputation of a new peer the minimum possible [Kamvar]. 462 Kamvar et. al [Kamvar] define five design considerations for 463 reputation management systems; 464 o Self policing. 465 o Anonymity. 466 o No profit to new comers. 467 o Minimal overhead. 468 o Robustness to malicious peers. 470 5.2.1. Unstructured reputation management 472 Unstructured reputation management systems have been proposed by Uzun 473 et. al [Uzun] and Damiani et. al [Damiani]. The basic idea of these 474 is that each peer maintains information about its own experience with 475 other peers and resources, and shares it with others on demand. In 476 the system proposed by Uzun et. al [Uzun], each node maintains trust 477 and distrust vectors for every other node that it has interacted 478 with. When reputation information about a peer is required, a node 479 first checks its local database, and if insufficient information is 480 present, it sends a query to its neighbors just as it would when 481 looking up content. However, such an approach requires peers to get 482 reputation information from as many sources as possible; otherwise, 483 malicious nodes may succesfully place targeted attacks returning 484 false values for their victims. 486 5.2.2. Structured reputation management 488 One of the problems with unstructured reputation management systems 489 is that they either take the feedback from few peers, or if they do 490 from all, then the they incur large traffic overhead. Systems such 491 as those proposed by [Lee] [Kamvar] try to resolve it in a structured 492 manner. The idea of the eigen trust algorithm [Kamvar] for example, 493 is transitivity of trust. If a node trusts peer X then it would also 494 trust the feedback it gives about other peers. A node builds such 495 information in an iterative way. The algorithm has fast convergence 496 properties [Haveliwala]. For maintaining this information in a 497 structured way, the authors use a content addressable network (CAN) 498 DHT [Ratnasamy]. The information of each peer is stored and 499 replicated on different peers to provide robustness against malicious 500 nodes. They also suggest favoring peers probabilistically with high 501 trust values instead of doing it deterministically, to allow new 502 peers to slowly develop a reputation. Eventually, they suggest the 503 use of incentives for peers with high reputation values. 505 6. Routing and data integrity 507 Preserving integrity of routing and data, or, in other words, 508 preventing peers from returning corrupt responses to queries and 509 routing through malicious peers, is an important security issue in 510 P2P networks. The data stored on a P2P overlay depends on the 511 applications that are using it. For file-sharing, this data would be 512 the files themselves, their location, and owner information. For 513 realtime communication, this would include user location bindings and 514 other routing information. We describe such data integrity issues 515 separately in Section 7. 517 6.1. Data integrity 519 For file-sharing applications, insertion of wrong content (e.g. files 520 not matching their names or descriptions) or introduction of corrupt 521 data chunks (often referred to as poisoning and pollution) are a 522 significant problem. Bit-Torrent uses voluntary moderators to weed 523 out bogus files and the SHA-1 algorithm to determine the hash of each 524 piece of a file to allow verification of integrity. If a peer 525 detects a bad chunk, it can download that chunk from another peer. 526 With this strategy, different peers download different pieces of a 527 file before the original peer disappears from the network. However, 528 if a malicious peer modifies the pieces that are only available on it 529 and the original peer disappears, then the object distribution will 530 fail [Zhang]. An analysis of BitTorrent in terms of integrity and 531 performance can be found in the work of Pouwelse et. al [Pouwelse]. 533 6.2. Routing integrity 535 To enhance the integrity of routing, it is important to reduce the 536 number of queries forwarded to malicious nodes. Marti et. al [Marti] 537 developed a system that uses social network information to route 538 queries over trusted nodes. Their algorithm uses trusted nodes to 539 forward queries (if one exists and is closer to the required ID in 540 the ID space). Otherwise they use the regular Chord [Stoica] routing 541 table to forward queries. While their results indicate good average 542 performance, it can not guarantee log$N$ hops for all cases. Danezis 543 et. al [Danezis] suggest a method for routing in the presence of a 544 large number of sybil nodes. Their method is to ensure that a peer 545 queries a diverse set of nodes and does not place too much trust in a 546 node. Both the above works have been described based on Chord. 547 However, unlike Chord, in DHTs like Pastry [PASTRY] and Kademlia 548 [KADEMLIA] there is flexibility in selecting nodes for any row in a 549 peer's routing table. Potentially many nodes have a common ID prefix 550 of a given length and are candidates for routing a given query. To 551 exploit the social network information and still guarantee log(N) 552 hops, a peer should select its friends to route a query, but only 553 when they are present in the appropriate row selected by the DHT 554 algorithm. 556 7. Peer-to-peer in realtime communication 558 The idea of using P2P in realtime communication boils down to 559 distributing centralized entities from conventional architectures 560 over peer-to-peer overlays and thus reducing the costs of deployment 561 and increasing reliability of the different services. Initiatives 562 such as the P2PSIP working group in IETF [P2PSIP] are currently 563 concentrating on achieving this by using a DHT for services such as 564 registration, location lookup, and support for NAT traversal, which 565 are normally handled by dedicated servers. 567 Even if based on the same technology, overlays used for realtime 568 communication differ from those used for file sharing in at least two 569 aspects: 570 o Resource consumption. Contrary to file sharing systems where the 571 DHT is used to store huge amounts of data (even if the distributed 572 database is used only for storing file locations, each user 573 usually indexes hundreds or thousands of files), realtime 574 communication overlays only require a subset of the resources 575 available at any given time as users only register a limited 576 number of locations (rarely more than one). 577 o Confidentiality. While in file sharing applications, where shared 578 files are supposed to be made publicly available, eavesdropping 579 and identity theft do not constitute real threats, in realtime 580 communication, since exchanges of data are usually meant to happen 581 privately, it is essential to have mechanisms to assert identities 582 and to guarantee confidentiality. 584 In this section we go over the admission issues, and security 585 problems discussed in previous sections, and discuss solutions that 586 would be applicable to realtime communication in P2P. 588 7.1. Admission 590 In order to keep as much compatibility with existing user agents as 591 possible, nodes in P2P communication architectures would probably 592 have to participate as either peers or clients. If a node 593 participates as a client, then it would use the overlay network by 594 simply attaching to a peer or a proxy instead of registering with a 595 server. In most cases users would be able to benefit from the 596 overlay by only acting as clients. However, in order to keep the 597 solution scalable, at some point clients would have to be promoted to 598 peers (admission to the DHT). This requires addressing the following 599 issues. 601 7.1.1. Active vs. passive upgrades 603 Most existing P2P networks [KAZAA] [BITTORRENT] [JOOST] would 604 generally make it the responsibility of clients to determine if and 605 when they would apply for becoming peers. A well known exception to 606 this trend is the Skype network [SKYPE], arguably one of the most 607 popular overlay networks used for realtime communications today. 608 Instances of the Skype application are supposed to operate as either 609 super-nodes, directly contributing to the distributed provision of 610 the service, or ordinary-nodes, simply using the service, and the 611 ``promotions'' are decided by the higher levels of the hierarchy 612 [Baset]. Even if there is not much difference for a client whether 613 it has to actively ask for authorization to join an overlay, or 614 passively wait for an invitation, the latter approach has some 615 advantages which fit well in overlays where only a subset of the 616 peers is required to provide the service (as in realtime 617 communication): 618 o An attacker cannot estimate in advance when and if it would be 619 invited to join the overlay as a peer. 620 o Allows peers to perform long-lasting measurements on sets of 621 candidates, in order to accurately select the most appropriate for 622 upgrading and only invite it when they are ``ready'' to do so. 624 The opposite approach, that is when clients initiate the join 625 themselves, adds an extra constraint for the peer that has to act 626 upon the request since it doesn't know if and when the peer would 627 attempt to join again. 628 o Discourages malicious peers from attempting sybil and, more 629 generally, brute force attacks, as only a small ratio of clients 630 has chances to join the overlay (possibly after an accurate 631 examination). 633 7.1.2. When to upgrade 635 In order to answer this question one would have to define some 636 criteria that would allow to determine the load on a peer and a 637 reasonable threshold. When the load exceeds this threshold, a client 638 is invited to become a peer and share the load. The criteria for 639 determining load can be: 640 o Number of clients attached. 641 o Bandwidth usage for DHT maintenance, forwarding requests and 642 responses to and from peers and from the attached clients. 643 o Memory usage for DHT routing table, DHT neighborhood table, 644 application specific data and information about the attached 645 clients. 647 7.1.3. Which clients to upgrade 649 Selecting which clients to upgrade would require defining and keeping 650 track of new metrics. The exact set of metrics and how they 651 influence decisions should be the subject of serious analysis and 652 experimentation. These could be based on the following observations: 653 o Uptime. A peer could easily record the amount of time that it has 654 been maintaining a connection with a client and take it into 655 account when trying to determine whether or not to upgrade it. 656 o Level of activity. It is reasonable to assume that the more a 657 client uses the service (e.g. making phone calls), the less they 658 would be willing to degrade it. 659 o Keeping track of history. Peers could record history of the 660 clients they invite and the way they contribute to the overlay. 661 Other metrics such as public vs. private IP addresses, computation 662 power, and bandwidth should also be taken into account even though 663 they do not necessarily have a direct impact on security. 665 7.1.4. Incentives for clients 667 Clients need to have incentives for accepting upgrades in order to 668 prevent excessive burden on existing peers. One way to handle this 669 would be to maintain separate incentive management through the use of 670 currency or credits. Another option would involve embedding these 671 incentives inside the protocol itself: 673 o Peers share with clients only a fraction of their bandwidth 674 (uplink and downlink). This would result in higher latency when 675 using the services of the overlay as a client and better service 676 quality for peers. 677 o Peers could restrict the number or types of calls that they allow 678 clients to make. 679 Introducing such incentives, however, may turn out to be somewhat 680 risky. Differences in quality would probably be perceptible for end 681 users who would not always be able to understand the difference 682 between the roles that their user agent is playing in the overlay. 683 Such behavior may therefore be interpreted as arbitrary and make the 684 service look unreliable. 686 7.2. Security 688 7.2.1. Targeted denial of service 690 In addition to bombardment with queries as described in Section 2, 691 the denial of service attack against an individual node can be 692 conducted in DHTs used for realtime communications if the peers which 693 surround a particular ID are compromised. These peers which act as 694 proxy servers for the victim, can fake the responses from the victim 695 by sending fictitious error messages back to peers trying to 696 establish a session. Danezis et al.'s solution [Danezis] can also 697 provide protection against such attacks as in their solution peers 698 vary the nodes used in queries. 700 7.2.2. Man in the middle attack 702 The man in the middle attack is well described by Seedorf [Seedorf06] 703 in the particular case of P2PSIP [P2PSIP] and consist of an attack 704 that exploits the lack of integrity when routing information. A 705 malicious node could return IP addresses of other malicious nodes 706 when queried for a particular ID. The requesting peer would then 707 establish a session with a second malicious node which would again 708 return a ``poisoned'' reply. This could go on until the TTL expires 709 and the requester gives up the ``wild goose chase'' [Danezis]. A 710 simple way for entities to verify the correctness of the routing 711 lookup is to employ iterative routing and to check the node-ID of 712 every routing hop that it is returned and it should get closer to the 713 desired ID with every hop. However, this is not a strong check and 714 can be defeated [Seedorf06]. 716 7.2.3. Trust between peers 718 The effect of malicious peers could be mitigated by introducing the 719 concept of trust within an overlay. This can be done in different 720 ways: 722 o Using certificates assigned by an external authority. The 723 drawback with this approach is that it requires a centralized 724 element. 725 o Using certificates reciprocally signed by peers. This mechanism 726 is quite similar to PGP [Zimmermann]; every peer signs 727 certificates of ``friend'' peers and trusts any other peer with a 728 certificate signed by one of its friends. However even though it 729 might be theoretically possible, in reality it is extremely 730 difficult to obtain long enough trust chains. 732 7.2.4. Routing call signalization 734 One way for implementing realtime communication overlays (as we have 735 mentioned in earlier sections) would be to simply replace centralized 736 entities in signalling protocols like SIP [RFC3261] with distributed 737 services. In some cases this might imply reusing existing protocol 738 mechanisms for routing signalling messages. In the case of SIP this 739 would imply regarding peers as SIP proxies. However the design of 740 SIP supposes that such proxies are trusted, and makes it possible for 741 them to fork requests or change their destination, add or remove 742 header fields, act as the remote party, and generally manipulate 743 message content and semantics 745 However, in a P2P environment where messages may be routed through 746 numerous successive peers, some of which might be compromised, it is 747 important not to treat them as trusted proxies. One way to limit 748 what peers can do is by protecting signalling with some kind of end- 749 to-end encryption. 751 Another option would be to extend existing signalling protocols and 752 modify the way they route messages in order to guarantee secure end- 753 to-end transmission. Gurbani et al. define a similar mechanism for 754 SIP called SIPSEC [I-D.gurbani-sip-sipsec]. It allows nodes to 755 establish a secure channel by sending a CONNECT SIP request, and then 756 tunnel all SIP messages through it, adopting a similar mechanism to 757 the one used for upgrading from HTTP to HTTPS [RFC2818]. 759 7.2.5. Integrity of location bindings 761 It is important to ensure that the location that a user registers, 762 usually a (URI, IP) pair, is what is returned to the requesting 763 party. Or the entities that issue the lookup request must be able to 764 verify the integrity of this pair. A pure P2P approach to allow 765 verification of the integrity of location binding information is 766 presented in [Seedorf08]. The idea is for an entity to choose an 767 asymmetric key pair and hash its public key to generate its URI. The 768 entity then signs its present location with its private key and 769 registers with the quadruple (URI, IP, signature, public key). Any 770 entity which looks up for the URI and receives such a quadruple can 771 then verify its integrity by using the public key and the 772 certificate. Another possible merit of such an approach could be 773 that it is possible to identify the malicious nodes and maintain a 774 black list. However, the resulting URIs are not easy to remember and 775 associate with entities. Discovering these URIs and associating them 776 with entities would therefore require some sort of a directory 777 service. The authors suggest using existing authentication 778 infrastructure for this such as a certified web service using SSL 779 which can publish an ``online phone book'' mapping users to URIs. 781 7.2.6. Encrypting content 783 Using P2P overlays for realtime communication implies that content is 784 likely to traverse numerous intermediate peers before reaching its 785 destination. A typical example could be the use of peers as media 786 relays as a way of traversing NATs in VoIP calls. 788 Contrary to publicly shared files, communication sessions are in most 789 cases expected to be private. It is therefore very important to make 790 sure that no media leaves the client application without being 791 encrypted and securely transported through a protocol like SRTP 792 [RFC3711]. However, the extra processing resources required by the 793 encryption algorithms, the management of keying material (e.g., 794 retrieving public keys when interacting with unknown peers) may 795 constitute an expensive task, especially for mobile devices. 797 7.2.7. Other issues 799 Identifying more specific threats related to the P2P realtime 800 communications, would require a clearly defined economic model. 801 Answers to the following questions would be helpful. 802 o To whom do the users pay? 803 o Do the users only pay when accessing the public telephone network? 804 o Is the billing done per call or is it fixed? 805 For instance, the implications of an attack such as taking control 806 over another's user agent or its identity and using it for outbound 807 calls would depend on whether or not this would be economically 808 advantageous for the attacker. Baumann et. al [Baumann] suggests 809 that to prevent unwanted communication costs, gateways for the public 810 telephone network should only be accessible via authenticated servers 811 and dialing authorizations should be enforced. Also it seems that it 812 would be difficult to do billing in a pure P2P manner as it would 813 mean keeping the billing details with untrusted peers. 815 8. Security Considerations 817 This document, informative in nature, discusses some of the security 818 issues of peer-to-peer systems used for realtime communications. 820 9. Acknowledgments 822 The authors are particularly grateful to Dhruv Chopra who contributed 823 to the writing of the article "Peer-to-peer Overlays for Real-Time 824 Communications: Issues and Solutions" (IEEE Surveys & Tutorials, Vol. 825 11, No. 1) this work is partially derived from. 827 10. Informative references 829 [Ahn] Ahn, Luis., Blum, Manuel., and John. Langford, "Telling 830 humans and computers apart automatically". 832 [Androutsellis-Theotokis] 833 Androutsellis-Theotokis, S. and D. Spinellis, "A survey of 834 peer-to-peer content distribution technologies". 836 [BITTORRENT] 837 "BitTorrent", . 839 [Baset] Baset, S. and H. Schulzrinne, "An analysis of the skype 840 peer-to-peer internet telephony protocol". 842 [Baumann] Baumann, R., Cavin, S., and S. Schmid, "Voice Over IP - 843 Security and SPIT". 845 [COOLSTREAM] 846 "COOLSTREAMING", . 848 [Castro] Castro, M., Druschel, P., Ganesh, A., Rowstron, A., and D. 849 Wallach, "Secure routing for structured peer-to-peer 850 overlay networks". 852 [Condie] Condie, T., Kacholia, V., Sankararaman, S., Hellerstein, 853 J., and P. Maniatis, "Maelstorm: Churn as Shelter". 855 [Damiani] Damiani, E., Vimercati, D., Paraboschi, S., Samarati, P., 856 and F. Violante, "A Reputation-Based Approach for Choosing 857 Reliable Resources in Peer-to-Peer Networks". 859 [Danezis] Danezis, G., Lesniewski-Laas, C., Kaashoek, M., and R. 860 Anderson, "Sybil-resistant DHT routing". 862 [Douceur] Douceur, J., "The Sybil Attack". 864 [Haveliwala] 865 Haveliwala, T. and S. Kamvar, "The second value eigenvalue 866 of the google matrix". 868 [I-D.gurbani-sip-sipsec] 869 Gurbani, V., Audet, F., and D. Willis, "The SIPSEC Uniform 870 Resource Identifier (URI)", draft-gurbani-sip-sipsec-01 871 (work in progress), June 2007. 873 [JOOST] "Joost", . 875 [KADEMLIA] 876 Maymounkov, P. and D. Mazi, "Kademlia: A Peer-to-peer 877 Information System Based on the XOR Metric". 879 [KAZAA] "KaZaa", . 881 [Kamvar] Kamvar, S., Garcia-Molina, H., and M. Schlosser, "The 882 EigenTrust Algorithm for Reputation Management in P2P 883 Networks". 885 [Kim] Kim, Y., Mazzocchi, D., and G. Tsudik, "Admission Control 886 in Peer Groups". 888 [Kong] Kong, J., Zerfos, P., Luo, H., Lu, S., and L. Zhang, 889 "Providing robust and ubiquitous security support for 890 MANET". 892 [Lee] Lee, S., Kwon, O., Kim, J., and S. Hong, "A Reputation 893 Management System in Structured Peer-to-Peer Networks". 895 [Liang] Liang, J., Kumar, R., Xi, Y., and K. Ross, "Pollution in 896 p2p file sharing systems". 898 [Marti] Marti, S., Ganesan, P., and H. Garcia-Molina, "SPROUT: P2P 899 Routing with Social Networks". 901 [McCue] McCue, Andy., "Bookie reveals 100,000 cost of denial-of- 902 service extortion attacks", . 905 [NAPSTER] "Napster", . 907 [Ohta] Ohta, K., Micali, S., and L. Reyzin, "Accountable Subgroup 908 Multisignatures". 910 [P2PSIP] "Peer-to-Peer Session Initiation Protocol (P2PSIP) IETF 911 Working Group", 912 . 914 [PASTRY] Rowstron, A. and P. Druschel, "Pastry: Scalable, 915 distributed object location and routing for large-scale 916 peer-to-peer systems". 918 [Poon] Poon, W. and R. Chang, "Robust Forwarding in Structured 919 Peer-to-Peer Overlay Networks". 921 [Pouwelse] 922 Pouwelse, J., Garbacki, P., Epema, D., and H. Sips, "The 923 Bittorent P2P File-Sharing System: Measurements and 924 Analysis". 926 [RFC2818] Rescorla, E., "HTTP Over TLS", RFC 2818, May 2000. 928 [RFC3261] Rosenberg, J., Schulzrinne, H., Camarillo, G., Johnston, 929 A., Peterson, J., Sparks, R., Handley, M., and E. 930 Schooler, "SIP: Session Initiation Protocol", RFC 3261, 931 June 2002. 933 [RFC3711] Baugher, M., McGrew, D., Naslund, M., Carrara, E., and K. 934 Norrman, "The Secure Real-time Transport Protocol (SRTP)", 935 RFC 3711, March 2004. 937 [Ratnasamy] 938 Ratnasamy, S., Francis, P., Handley, M., Karp, R., and S. 939 Shenker, "A Scalable Content-Addressable Network". 941 [Rowaihy] Rowaihy, H., Enck, W., McDaniel, P., and T. Porta, 942 "Limiting Sybil attacks in structured peer-to-peer 943 networks". 945 [SHA1] 180-1, FIPS., "Secure Hash Standard". 947 [SKYPE] "Skype", . 949 [Saxena] Saxena, N., Tsudik, G., and J. Yi, "Admission Control in 950 Peer-to-Peer: Design and Performance Evaluation". 952 [Scheideler] 953 Scheideler, C., "How to Spread Adversarial Nodes?: 954 Rotate!". 956 [Seedorf06] 957 Seedorf, J., "Security Challenges for Peer-to-Peer SIP". 959 [Seedorf08] 960 Seedorf, J., "Using Cryptographically Generated SIP-URIs 961 to Protect the Integrity of Content in P2P-SIP". 963 [Singh] Singh, K. and H. Schulzrinne, "Peer-to-Peer Internet 964 Telephony using SIP". 966 [Sit] Sit, E. and R. Morris, "Security considerations for peer- 967 to-peer distributed hash tables". 969 [Stoica] Stoica, I., Morris, R., Karger, D., Kaashoek, M., and H. 970 Balakrishnan, "Chord: A Scalable Peer-to-peer Lookup 971 Service for Internet Applications". 973 [Uzun] Uzun, E., Pariente, M., and A. Selpk, "A Reputation-Based 974 Trust Management System for P2P Networks". 976 [Wallach] Wallach, D., "A Survey of Peer-to-Peer Security Issues", 977 . 979 [Yu] Yu, H., Kaminsky, M., Gibbons, P., and A. Flaxman, 980 "SybilGuard: Defending Against Sybil Attacks via Social 981 Networks". 983 [Zhang] Zhang, X., Chen, S., and R. Sandhu, "Enhancing Data 984 Authenticity and Integrity in P2P Systems". 986 [Zimmermann] 987 Zimmermann, Philip., "Pretty good privacy: public key 988 encryption for the masses". 990 Authors' Addresses 992 Henning Schulzrinne 993 Columbia University 994 1214 Amsterdam Avenue 995 New York, NY 10027 996 USA 998 Email: hgs@cs.columbia.edu 999 Enrico Marocco 1000 Telecom Italia 1001 Via G. Reiss Romoli, 274 1002 Turin 10148 1003 Italy 1005 Email: enrico.marocco@telecomitalia.it 1007 Emil Ivov 1008 SIP Communicator 1009 4 rue Blaise Pascal 1010 Strasbourg Cedex F-67070 1011 France 1013 Email: emcho@sip-communicator.org