clj-ebpf.rate-limit — org.clojars.pgdad/clj-ebpf 0.6.7

Token bucket rate limiting helpers for BPF programs.

Provides comprehensive token bucket rate limiting implementation for XDP/TC programs. Common use cases include:

• Source IP rate limiting (DDoS protection)
• Backend rate limiting (overload protection)
• API rate limiting
• Connection rate limiting

Token Bucket Algorithm:

1. Each bucket has: tokens (current count), last_update (timestamp)
2. Config has: rate (tokens per second), burst (max tokens)
3. On each packet: a. Calculate elapsed time since last_update b. Add new tokens: elapsed_ns * rate / 1e9 c. Cap tokens at burst d. If tokens >= 1, consume and allow e. Else, rate limit (drop)

Data Structures:

• Config map (array): 16 bytes per entry (rate: 8, burst: 8)
• Bucket map (LRU hash): 16 bytes per entry (tokens: 8, last_update: 8)
• Token scale: 1000 (for sub-second precision)

Usage: (require '[clj-ebpf.rate-limit :as rl])

;; Build rate limit check (rl/build-rate-limit-check config-map-fd 0 bucket-map-fd -16 -48 :pass :drop)

BUCKET-OFF-LAST-UPDATE clj

BUCKET-OFF-TOKENS clj

build-rate-limit-check clj

(build-rate-limit-check config-map-fd
                        config-index
                        bucket-map-fd
                        key-stack-off
                        scratch-stack-off
                        skip-label
                        drop-label)

Generate BPF instructions for token bucket rate limit check.

Implements token bucket algorithm:

1. Load config from config-map at config-index
2. If rate == 0, skip (rate limiting disabled)
3. Load/create bucket from bucket-map using key at key-stack-off
4. Calculate elapsed time since last update
5. Add tokens: new_tokens = old_tokens + elapsed_us * rate / 1e6
6. Cap at burst
7. If tokens >= TOKEN-SCALE (1 token), consume and continue
8. Else jump to drop-label

Args: config-map-fd: FD for rate_limit_config array map config-index: Index in config map (e.g., 0 for source, 1 for backend) bucket-map-fd: FD for bucket LRU hash map key-stack-off: Stack offset where lookup key is stored scratch-stack-off: Stack offset for scratch space (needs 32 bytes) skip-label: Label to jump to if rate limiting disabled/passed drop-label: Label to jump to if rate limited

Register usage: r0: Scratch, return values r1-r5: Helper call arguments r6: Saved bucket pointer r7: Saved tokens value r8: Rate/config values r9: Burst/temp values

Stack usage (relative to scratch-stack-off): offset 0-7: config index for lookup offset 8-15: new bucket initialization (tokens) offset 16-23: new bucket initialization (last_update)

Returns: Vector of instructions/pseudo-instructions

build-simple-rate-limit clj

(build-simple-rate-limit bucket-map-fd
                         key-stack-off
                         scratch-stack-off
                         rate
                         burst
                         skip-label
                         drop-label)

Simplified rate limiting that doesn't use a config map.

Rate and burst are hardcoded, suitable for single-purpose rate limiters.

Args: bucket-map-fd: FD for bucket LRU hash map key-stack-off: Stack offset where lookup key is stored scratch-stack-off: Stack offset for scratch space (needs 24 bytes) rate: Tokens per second (unscaled, e.g., 100 for 100 req/s) burst: Maximum tokens (unscaled, e.g., 200) skip-label: Label to jump to on pass drop-label: Label to jump to on rate limit

Returns: Vector of instructions/pseudo-instructions

CONFIG-OFF-BURST clj

CONFIG-OFF-RATE clj

encode-rate-limit-config clj

(encode-rate-limit-config rate burst)

Encode rate limit config for userspace map update.

Args: rate: Tokens per second (unscaled) burst: Maximum tokens (unscaled)

Returns: Byte array suitable for map update (16 bytes)

MAX-ELAPSED-US clj

NS-PER-SEC clj

NS-PER-US clj

rate-disabled-config clj

(rate-disabled-config)

Create config that disables rate limiting.

Returns: Byte array with rate=0, burst=0

TOKEN-SCALE clj

US-PER-SEC clj