L2/L3 Switch Development — Linux-Based L3 Switch Master Program (Fundamentals to Expert)

Write efficient, correct C code for protocol daemons and kernel-adjacent networking programs.

• Pointers & memory layout — stack, heap, BSS; malloc / calloc / free patterns
• Packed structures & unions — mapping C structs directly to protocol headers (__attribute__((packed)))
• Bitwise operations — AND/OR/XOR/shift for flag manipulation, VLAN TCI parsing (CRITICAL)
• Byte-order conversion — htons / ntohs / htonl / ntohl in practice
• Raw socket programming — PF_PACKET SOCK_RAW: send/receive custom Ethernet frames
• File I/O & logging — syslog, file-based event logging for daemons

Implement the core data structures used inside every real switch OS for MAC learning, routing, and packet queuing.

• Hash tables — MAC address FDB lookup, ARP/ND table; collision handling
• Trie / radix tree — Longest Prefix Match (LPM) for IP routing table lookup
• Linked lists — neighbour lists, packet queues, timer wheels
• Ring buffers — packet descriptor rings, producer/consumer queues
• Red-black / AVL trees — ordered key storage for routing databases

Understand how the Linux kernel receives, processes, and forwards packets — the foundation for all switch software development on Linux.

• sk_buff structure — packet buffer lifecycle: NIC driver → netif_receive_skb → netdev → socket
• Kernel vs user space — system call interface, privilege levels, fast-path vs slow-path
• Linux bridge (br_*) — VLAN-aware bridge, br_forward, br_flood, FDB management
• /proc/net & /sys/class/net — inspecting interfaces, FDB, routing, neighbour tables
• Netdevice operations — ndo_start_xmit, ndo_get_stats64, driver model
• Packet processing path — RX interrupt → NAPI poll → GRO → netfilter → routing decision

Write production-quality Linux programs that configure the kernel network stack — the same approach used by FRRouting, Open vSwitch, and switch vendor software.

• System calls — fork, exec, epoll for async event-driven daemons
• Netlink sockets (RTNETLINK) — add/delete routes, neighbours, VLAN interfaces from userspace C
• iproute2 internals — how ip route, ip link, bridge vlan use netlink under the hood
• Multithreading — pthreads, mutex, per-port timer threads for protocol daemons
• POSIX timers & signals — protocol Hello timers, ageing timers, retry logic

Build the packet I/O foundation for all protocol daemon development — capturing control-plane PDUs and injecting responses.

• PF_PACKET SOCK_RAW — binding to interfaces, promiscuous mode, EtherType filtering
• libpcap — BPF filter compilation, pcap_loop, offline PCAP replay for regression testing
• Packet injection — crafting and sending STP BPDUs, LACP PDUs, LLDP frames in C
• Netfilter hooks — NF_INET_PRE_ROUTING, FORWARD, POST_ROUTING for L3 policy enforcement
• Scapy (Python) — rapid protocol test frame generation for lab validation

Understand how a Linux-based switch is architected and how hardware ASIC forwarding integrates with the Linux software stack.

• Switch OS architecture — monolithic vs microservice design (Cisco IOS-XR, SONiC, OpenWrt)
• MAC learning — FDB ageing timer, unknown unicast flood, filtering database isolation per VLAN
• Forwarding & filtering — FDB lookup, unicast/multicast/broadcast decision
• Store-and-forward vs cut-through — latency vs error detection trade-off on ASICs
• Switch fabric & buffering — shared memory, output-queued, VOQ architectures
• Hardware offload — how Linux switchdev and TC-flower offload decisions to ASIC

Master the architectural split that defines every production switch and router — the most asked concept in networking silicon interviews.

• Control plane — routing daemons (FRRouting/Zebra), STP daemon, LACP daemon; slow-path PDU processing
• Data plane — Linux kernel FIB, bridge FDB, iptables chains; hardware ASIC TCAM/FDB
• RIB → FIB synchronisation — Zebra populates kernel FIB via Netlink; FIB pushes to ASIC via SDK
• CPU punt path — which packets the ASIC sends to Linux CPU (BPDUs, ARP, OSPF Hellos)
• Switchdev model — Linux kernel abstraction for offloading L2/L3 decisions to switch ASICs
• SAI (Switch Abstraction Interface) — vendor-neutral ASIC API used in SONiC and Open Compute

Implement VLAN tagging, trunk/access port logic, and per-VLAN FDB isolation to IEEE 802.1Q specification.

• 802.1Q tag structure — TPID (0x8100), PCP, DEI, VID fields in the 4-byte shim header
• Tag insertion & stripping in C — memmove, byte-order, EtherType patching
• Access vs trunk ports — ingress tagging, egress stripping, native VLAN logic
• Per-VLAN FDB isolation — MAC+VID as FDB key; VLAN membership bitmap per port
• QinQ (802.1ad) — double-tagged frames, S-VLAN/C-VLAN, provider bridging use cases
• Linux VLAN interfaces — ip link add link eth0 name eth0.100 type vlan id 100; kernel implementation

Implement the STP/RSTP state machine from the IEEE 802.1D/w specification — the most asked L2 protocol in development interviews.

• Spanning Tree Algorithm — root bridge election (Bridge ID = priority + MAC), path cost, port roles
• BPDU encoding/decoding in C — Configuration BPDU and TCN BPDU wire format (IEEE 802.1D §9.3)
• Port state machine — Blocking → Listening → Learning → Forwarding transitions with timer events
• RSTP (802.1w) — Discarding/Learning/Forwarding states, proposal-agreement handshake, edge ports, <1s convergence
• MSTP (802.1s) — MST regions, IST, per-VLAN tree instances (CIST/MSTI)
• Loop prevention — topology change notification, MAC flush on TCN

Implement IEEE 802.3ad LACP for bonding multiple physical links into a logical aggregate with load balancing.

• LAG concepts — port aggregation, LAG group membership, logical interface abstraction
• LACPDU wire format (IEEE 802.3ad §43) — Actor/Partner TLVs, system priority, key, port priority
• Actor/Partner state machine — Collecting/Distributing transitions, LACP_Activity, LACP_Timeout
• Load balancing algorithms — L2 (src/dst MAC XOR), L3 (IP hash), L4 (port hash)
• Linux bonding driver — mode 4 (802.3ad); teamd LACP daemon source review

Implement the remaining IEEE L2 protocols used in every production enterprise and carrier switch.

• LLDP (IEEE 802.1AB) — TLV encoding/decoding in C, Chassis ID, Port ID, TTL, System Capabilities TLVs
• 802.1X port-based access control — Authenticator state machine, EAP-over-LAN, EAPOL wire format
• IGMP Snooping (RFC 4541) — group membership tracking, querier election, fast-leave, per-VLAN multicast table
• MACsec (IEEE 802.1AE) — L2 encryption concepts, Secure Channels, MKA key negotiation overview

• 4-node VLAN topology: access + trunk ports, 802.1Q tagging verified in Wireshark
• 3-bridge ring: RSTP convergence <1s, BPDU injection to force root re-election
• LACP bonding: Actor/Partner negotiation, load-balance, link failure failover
• LLDP neighbour discovery daemon: parse TLVs, build topology map
• MAC learning table in C: hash + ageing + per-VLAN isolation

Build a software IP forwarding engine from scratch — the core of the L3 part of a Linux switch.

• Static vs dynamic routing — admin distance, route preference hierarchy
• Routing table structure — destination prefix, prefix length, next-hop, egress interface, metric
• LPM (Longest Prefix Match) — trie lookup algorithm; implementing in C for the data plane
• IP forwarding logic in C — TTL decrement, header checksum recalculation, ARP resolution for next-hop
• Linux FIB — ip route internals, fib_result, nexthop objects, ECMP
• L3 switch operation — L3 interface on a VLAN (SVI concept), inter-VLAN routing

Master OSPF implementation — the most important IGP in enterprise and carrier L3 switch deployments.

• Link-state routing — LSDB synchronisation, reliable flooding, SPF algorithm (Dijkstra)
• OSPF packet types — Hello, DBD, LSR, LSU, LSAck; state machine (Down → Full)
• DR/BDR election on broadcast networks (Ethernet segments)
• LSA types — Type 1 Router, Type 2 Network, Type 3 Summary, Type 5 External
• Multi-area OSPF — ABR, ASBR, stub/NSSA areas, route summarisation
• FRRouting ospfd — reading ospfd/ source; how OSPF pushes routes to Zebra RIB → Netlink FIB

Understand BGP path-vector mechanics and FRRouting integration — essential for carrier-grade L3 switch deployments.

• Path vector protocol — AS_PATH loop prevention, session establishment, route advertisement
• BGP FSM — Idle → Connect → Active → OpenSent → OpenConfirm → Established
• Route attributes — AS_PATH, NEXT_HOP, LOCAL_PREF, MED, COMMUNITY
• eBGP vs iBGP — NEXT_HOP handling, route reflectors
• BGP best-path decision process (14-step algorithm)
• FRRouting bgpd → Zebra RIB → Netlink FIB pipeline; route redistribution with OSPF

Round out L3 knowledge with redundancy, label switching, and multicast routing used in real switch deployments.

• VRRP (RFC 5798) — virtual router redundancy, Master/Backup election, <1s failover; implementation in FRR
• MPLS — label shim header, LFIB, LER/LSR roles, LDP session setup, label push/pop/swap
• PIM-SM (Protocol Independent Multicast) — RPT/SPT, join/prune, multicast routing table (MRIB)
• RIP (RFC 2453) — Bellman-Ford, distance vector, split horizon; still used on L3 switch access edges

• Software L3 forwarder in C: trie LPM, TTL decrement, ARP resolution, 3-subnet topology
• 4-router multi-area OSPF on FRRouting; LSA trace; Zebra FIB sync verification
• eBGP peering: LOCAL_PREF / MED policy manipulation; best-path verification
• VRRP failover: 2-router active/standby, <1s gateway switchover
• MPLS LDP: label-switched path setup, Wireshark MPLS shim decode

Understand how switching ASICs work internally and how vendor SDKs expose hardware tables to software.

• ASIC pipeline — ingress parser → TCAM lookup → action → forwarding → egress; NPU pipelines
• Key forwarding tables — L2 FDB (MAC+VLAN), L3 FIB/LPM, ARP/ND table, ACL TCAM
• CPU punt path — exception packets (STP BPDU, ARP, OSPF Hello) sent to Linux kernel; trap rules
• Broadcom switch family — Trident / Tomahawk / Qumran; OpenNSL and OF-DPA SDK APIs
• Marvell switch family — Prestera / Aldrin / AC5; CPSS (Control Plane Support Software) SDK
• SONiC architecture — SAI (Switch Abstraction Interface) as the common ASIC abstraction layer

Program the Broadcom switching ASIC to implement L2 and L3 forwarding tables directly from C using the OpenNSL / BCM SDK.

• OpenNSL initialisation — unit/device init, port configuration, VLAN creation via bcm_vlan_create()
• L2 FDB management — bcm_l2_addr_add(), bcm_l2_addr_delete(), MAC learning callbacks
• L3 interface & egress object — bcm_l3_intf_create(), bcm_l3_egress_create() with next-hop MAC/VLAN
• L3 route table — bcm_l3_route_add() / bcm_l3_route_delete(); LPM route programming
• ACL / Field processing — bcm_field_* APIs for TCAM-based match-action rules
• FRRouting → Zebra → Netlink → OpenNSL pipeline — sync software RIB to ASIC forwarding table

Program Marvell Prestera/Aldrin ASIC using the CPSS SDK — the industry-standard SDK used across enterprise switch platforms.

• CPSS architecture — FAP layer, Generic CPSS, Functional CPSS; device discovery and init
• Port configuration — cpssDxChPortModeSpeedSet(), cpssDxChBrgVlanPortAdd()
• L2 FDB programming — cpssDxChBrgFdbMacEntrySet(); MAC learning mode, ageing enable
• L3 routing — cpssDxChIpLpmIpv4UcPrefixAdd(); ARP entry, next-hop interface config
• Packet capture to CPU — CPU code configuration, trap rules for STP/ARP/OSPF PDUs
• CPSS vs OpenNSL comparison — API style differences; portability considerations

• OpenNSL: create 4 VLANs, 20 static FDB entries, 3 L3 routes on Broadcom simulator
• CPSS: configure VLAN tagging, L3 prefix, ARP trap rule on Marvell simulator
• Full pipeline: FRRouting OSPF → Zebra RIB → Netlink → OpenNSL L3 route sync
• CPU punt path: verify BPDU, ARP, OSPF Hello are correctly trapped to Linux socket

Comprehensive technical and career preparation to land a switch/protocol engineering role at top networking companies.

• Weekly coding tests and IEEE protocol quizzes (throughout the program)
• Lab assignment grading — code quality, correctness, IEEE spec compliance
• Final project presentation and code review with industry mentors
• Resume workshop — highlighting SDK experience, GitHub projects, protocol knowledge
• Mock technical interviews — Broadcom/Marvell/Cisco/Juniper-style protocol & C questions
• Placement drive — direct referrals to hiring managers at networking silicon companies