Multi-Layer DDoS Mitigation Strategies
2 likes3,259 views
The document outlines multi-layer DDoS mitigation strategies, emphasizing the increasing sophistication of attacks and the importance of diverse protective measures such as load balancers, firewalls, and IDS/IPS systems. It discusses the role of increased connectivity, IoT vulnerabilities, and the necessity for robust application-level and server-layer protections, including the use of third-party scrubbing services and CDNs. The authors provide insights into various approaches and tools available for effectively defending against DDoS threats in a complex attack landscape.
Multi-Layer DDoS Mitigation Strategies
- 1. 1 Multi-Layer DDoS Mitigation Strategies Sagi Brody, CTO @webairsagi sagi@webair.com Logan Best Senior Infrastructure Engineer logan@webair.com
- 2. 2 Multi-Layer DDoS Mitigation Strategies Application Server(s) Load Balancers/Proxies IDS/IPS Network 3rd Party Scrubbing 3rd Party CDN/Proxies Firewall
- 3. 3 Fully Managed Web Stack Application Server Load Balancers IDS/IPS Network Scrubbing CDN/Proxies FW & Cache plugins Memcache, Fail2ban, HAProxy + keepalived, nginx, SOCaaS, ProtectWise, AlertLogic, Snort RTBH, FlowSpec, Outbound filters Network Taps , Analysis, Automated BGP Redirects to CDN in App or via HTTP rewrite MikroTik, PaloAlto, JuniperFirewall
- 4. 4 Why? - New Attack Landscape • DDoS Ransom on the rise with Bitcoins - Attackers becoming more anonymous • Multiple governments issue warnings about DD4BC • Stress Testers • Used to be called “booters’, available on IRC (EFnet anyone?) • DDoS-as-a-Service • Christmas 2014, Lizard Squad “stress tested” PSN and Xbox Live. Sets up lizardstresser.su to monetize. • Inexpensive - as little as $5/attack • Technically legal, RageBooter.net claims legal, provides full phone support. • DDoS Mitigation Companies targeted
- 5. 5 Why? - Unlikely DDoS Originators • IoT • IoCT? Internet of Compromised Things? • Professor in Berlin creates smart home. Lightbulb unintentionally launches DoS against controller, causing home to lock up. • Proliferation of cheap android devices and LTE will result in perfect mobile storm. • 1 billion devices 0.1% vulnerable: • 100 kilobits per second == 100 gigabits per second • 100 pps == 100 million packets per second • Today: • 7.2 Billion cell devices globally • 6.4 Billion IoT devices globally
- 6. 6 More sophisticated attackers • Less total attacks, increase in effectiveness • Attacks are getting more complicated and last longer • More effective L7 Botnets being used
- 7. 7 Our Environment Founded: 1996 Headquarters: Long Island, NY Services Offered: Public, Private & Hybrid Cloud, Dedicated Servers, Colocation, CDN, Security, DRaaS, and more Customers: Enterprise, Healthcare, SMBs, VoIP providers, eCommerce, Information Technology, and more The Webair Value: ➢ Over 18 years providing customers with best-in-class Managed Hosting solutions ➢ High-touch Support ➢ Full ownership of our customer’s infrastructure stack so they can focus on their core business. ➢ Manages most secure and redundant facility east of Manhattan with trans-Atlantic bypass (Tier III, HIPPA/SSAE16)
- 8. 8 Full Stack Management “Outsourced Devops”
- 9. 9 Application Layer • Simple to take sites down with low request volume • Stay under volumetric detection’s radar • Uncached dynamic (db backed) URLs fall over quickly (especially if framework is known) • Poorly written SQL queries • Apps written w/o scale in mind • Cross-site scripting, SQL injection, buffer overflows, file inclusion, oh my… • Search result pages, POST to forms, comments, etc • Botnets storing session cookies, use real ‘hidden’ browsers on infected PCs • Simple HTTP floods lead to state-exhaustion • Part of larger multi-vector attacks
- 10. 10 Application Layer Protection • Framework or App specific plugins - WP (NinjaFirewall, Wordfence) • HTTP server modules - mod_evasive, mod_security • Common caching layers can perform rate limiting based on specific request attributes (nginx, varnish) • Cache for security • In memory - (memcache, redis) • Opcode caching - (APC, xcache, eAccelerator) • SQL - (sphinx search, solr, elasticsearch) • File • Follow best practices for web scale: • Disable all unnecessary logging • Scale-out architecture • In-memory session management • (many more!..)
- 11. 11 Server Layer Linux Kernel /etc/sysctl.conf customization: • Prevent against the common 'syn flood attack' #net.ipv4.tcp_syncookies = 1 • Increase the number of outstanding syn requests allowed #net.ipv4.tcp_max_syn_backlog = 4096 • Reduce # of potential ACK replies (default 5) #net.ipv4.tcp_synack_retries = 3 • Reduce # of connections in TIME_WAIT state #net.ipv4.tcp_fin_timeout = 15 • Enable IP spoofing protection (turn on source route verification) #net.ipv4.conf.all.rp_filter = 1 #net.ipv4.conf.default.rp_filter = 1 • Decrease the time default value for tcp_keepalive_time connection #net.ipv4.tcp_keepalive_time = 1800
- 12. 12 Server Layer IPTables rules: • Rate limit and block syn flooding: #iptables -N syn-flood #iptables -A syn-flood -m limit --limit 100/second --limit-burst 150 -j RETURN #iptables -A syn-flood -j LOG --log-prefix "SYN flood: “ #iptables -A syn-flood -j DROP • Drop excessive RST packets to avoid smurf attacks #iptables -A INPUT -p tcp -m tcp --tcp-flags RST RST -m limit --limit 2/second --limit-burst 2 -j ACCEPT #iptables -A INPUT -p icmp -m icmp --icmp-type address-mask-request -j DROP #iptables -A INPUT -p icmp -m icmp --icmp-type timestamp-request -j DROP #iptables -A INPUT -p icmp -m icmp -j DROP • Drop Invalid Packets #iptables -A INPUT -m state --state INVALID -j DROP #iptables -A FORWARD -m state --state INVALID -j DROP #iptables -A OUTPUT -m state --state INVALID -j DROP
- 13. 13 Server Layer IPTables rules: •Reject spoofed packets (bogon networks, RFC1918) #iptables -A INPUT -s 10.0.0.0/8 -j DROP #iptables -A INPUT -s 169.254.0.0/16 -j DROP #iptables -A INPUT -s 172.16.0.0/12 -j DROP #iptables -A INPUT -d 239.255.255.0/24 -j DROP #iptables -A INPUT -d 255.255.255.255 -j DROP …….etc • Reject ports, protocols, and IPs not in use. Are you really using UDP port 80? • Fail2Ban/SSHGuard • Mindful of NFS overhead (stop logging to NFS!)
- 14. 14 Load Balancers • Load balancers can shield app/servers from attacks • Already operating at L7, ability to perform complex L7 rules and mitigation, https via offload • Provides inherent synflood protection, slowloris, • Traditional appliance solutions too rigid. Next gen ADC (Application delivery controller) to provide LB, WAF, Cache combined • DIY, our kit: Ansible, Debian/RHEL, HAProxy, keepalived, csync, git, automation software for provisioning, configuration, attack defense • Integrate with 3rd party software easily (geoip, automated blacklists) • Tie into caching: (varnish, nginx)
- 15. 15 Load Balancers (HAProxy) • Provides rate limiting on many levels: • Connections per IP • Connection rate per IP • Bandwidth usage too high • Client not respecting RFCs (IE for SMTP) • HTTP request rate (L7) - Block based on • Request string/regex • User-Agent • Any other tracked variables • Vulnerability scanners: • Invalid or truncated requests • Denied or tarpitted requests • Failed authentications • 4xx error pages
- 16. 16 HAProxy examples stick-table type ip size 256k expire 300s peers mypeers store gpc0,http_req_rate(5s),conn_cur,conn_rate(5s),http_err_rate(5s) acl abuse src_http_req_rate gt 50 acl flag_abuser sc1_inc_gpc0 acl ipwhitelist src -f /path/to/ipwhitelist.txt acl ipblacklist src -f /path/to/ipblacklist.txt tcp-request connection track-sc1 src tcp-request connection reject if ipblacklist tcp-request connection accept if ipwhitelist tcp-request connection reject if { sc1_conn_cur ge 30 } tcp-request connection reject if { sc1_conn_rate ge 20 } tcp-request connection reject if { sc1_get_gpc0 gt 0 } tcp-request content reject if { sc1_get_gpc0 gt 0 } http-request deny if ipblacklist http-request allow if ipwhitelist http-request deny if { sc1_get_gpc0 gt 0 } http-request deny if { sc1_http_err_rate ge 20 } flag_abuser http-request deny if abuse flag_abuser
- 17. 17 IDS/IPS • “Threat Monitoring & Mitigation” • Sits behind the FW to catch unprotected malicious traffic • Works in virtual environments as well • Can be tap/span, logs, or libpcap based • Graphs and other cool Visual Dashboards! • No the “be all, end all” solution • API/Developer friendly • SOCaaS • Monitors 24/7/365, SLA backed • Hands on investigations • Detailed intrusion information (kill-chain analysis) • Custom retention periods for all types of compliance • packet replay
- 18. 18 IDS/IPS
- 19. 19 IDS/IPS
- 20. 20 Firewalls • WAF capabilities becoming standard w/ all new FWs • Is WAF even enough? (Palo Alto says no) • L7 SSL inspect via passthrough certs • Many great OSS options still exist • MikroTik: • SSH/GUI w/ built in easy packet sniffing • L7 regex matching • VPN (IPSec, PPTP, SSTP, L2TP), BGP, OSPF, etc.. • Run a whole ISP on it :) • HW appliance, w/ support and built in virtualization • Dynamic address lists based on multitude of rules • ‘Good’ IPv6 support FTW!
- 21. 21 Firewalls
- 22. 22 Network
- 23. 23 Network • Block Martian and Bogon routes (automate if possible) • Unicast Reverse Path Forwarding (w/ full routes) • Filter unknown outbound packets to prevent spoofing + retaliation • Proper flow monitoring at the edge - CLI + GUI • Packet sniffing • Strategy around reflection attacks: • Detect internally • Run scan and patch? • Block outbound if you can or redirect to internal server(s) •IPv6 /64 assignments are bad - NDP exhaustion
- 24. 24 Network • RTBH (remotely triggered blackhole) on your network (but kills dst) • Blackhole srcs upstream via community strings • Automate ^^ via scripts/software • BGP Flowspec: Use BGP to distribute flow specification filters and dynamically filter on routers. • communities based actions (accept, discard, rate-limit, sample, redirect) • Match on combination of source/dest prefix, source/dest port, ICMP type/code, packet size, DSCP, TCP flag, fragment encoding, etc • Leverage ASIC filtering in routers • Available today
- 25. 25 Network
- 26. 26 3rd Party on-demand scrubbing Providers: • Staminus communications • Level3 (legacy BlackLotus) • TATA • Savvis • Verisign • Anyone running Arbor gear?? BGP enabled on the fly mitigation
- 27. 27 3rd Party on-demand scrubbing • On Demand: Enable Mitigation only when needed. “Insurance Policy” • Assume potential attack will always have more BW than you • Many providers now support L7 mitigation via BGP swing • Upgrading capacity as needed via contracts, not hardware • On prem device detects attacks and auto-swings on its own (automation) • “Big Data” play - view details on attack, portal, reports • A ‘required’ value add these days • If BGP cannot be maintained to provider, route withdraws, attack comes back to you. Some providers can instigate their own announcements on request
- 28. 28 3rd Party CDN/Proxies • “Always on” - All connections proxies via 3rd party (for better, or worse) • Usually work well out of the box with popular applications • Simple way to handle many security touch points: • Caching & CDN • WAF capabilities and common injection detection • SSL offload • Custom Rules • Easy to use for Beginners • Enterprise offerings can: • Cache dynamic applications (EdgeCast/VZW + Fast.ly) • Work with Varnish VCLs and other common formats • Mid-tier origin caching • Option of last resort: CAPTCHA for user verification (eww) • Not a complete solution, origins are usually found
- 30. 30 THANK YOU! Sagi Brody, CTO @WebairSagi sagi@webair.com Logan Best Senior Infrastructure/Network Engineer logan@webair.com *Research data provided via Staminus Communications
Editor's Notes
- #2: Speaker: Sagi
- #3: Speaker: Sagi
- #4: Speaker: Sagi
- #5: Speaker: Sagi
- #6: Speaker: Sagi - How exactly do you reboot your home?
- #7: Speaker: Sagi
- #8: Speaker: Sagi
- #9: Speaker: Sagi
- #10: Speaker: Sagi
- #11: Speaker: Sagi
- #12: Speaker: Logan Simple things like per-IP rate limiting, flooding rate limits, and syn cookies should be enabled without a question. The sysctl variables improves the overall network handling efficiency and protects about common SYN/ACK Denial of service attacks. In syslog on server, I saw lots of "nf_conntrack: table full, dropping packet". And netstat showed lots of TIME_WAIT.
- #13: Speaker: Logan
- #14: Speaker: Logan
- #15: Speaker: Logan An application delivery controller (ADC) is a computer network device in a datacenter, often part of an application delivery network (ADN), that helps perform common tasks such as those done by web sites to remove load from the web servers themselves. Many also provide load balancing.
- #16: Speaker: Logan
- #17: Speaker: Logan Our Basic HAP ddos config ACL’s. Many many more options to harden applications of all types
- #18: Speaker: Logan & Sagi
- #19: Speaker: Logan
- #20: Speaker: Sagi
- #21: Speaker: Sagi mention NTP and DNS reflection specifically detail IPv6 Allocation/Assignment flow (Allocate /64 to client, assign /120 etc to vlan)
- #22: Speaker: Sagi
- #23: Speaker: Sagi
- #24: Speaker: Sagi
- #25: Speaker: Sagi
- #26: Speaker: Logan
- #28: Speaker: Sagi