Fix security issues and bugs across Rust and TypeScript

[pcarrier]

Summary

• Timing side-channel in constant_time_eq: The early return on length mismatch leaked whether passphrase/token lengths matched. Fixed in interactive.rs, gateway/lib.rs, and webserver/config.rs by folding the length difference into the comparison byte and always comparing min(len_a, len_b) bytes.
• Passphrase leaked in error messages: WebRTC connection errors in interactive.rs included the plaintext passphrase, which was then forwarded to the browser client. Removed the passphrase from the error format string.
• Integer overflow in PNG size calculation: width * height * 4 was computed as u32 arithmetic in server/lib.rs, which overflows for large surfaces. Cast to usize before multiplication.
• Missing frame size check in WebRTC data channel transport: webrtc.ts parsed frame length headers without an upper bound, allowing a malicious peer to cause unbounded readBuf growth. Added the same 16 MiB cap used in mux.ts and webtransport.ts.
• Stack overflow in btoa(String.fromCharCode(...array)): The spread operator hits the call stack limit for large payloads. Replaced with a loop-based approach in webrtc-share.ts and passphrase-crypto.ts.
• Unchecked JSON.parse on WebSocket messages: Two JSON.parse calls in webrtc-share.ts signaling handlers were not wrapped in try-catch, causing unhandled exceptions on malformed messages. Added error handling.
• Overly broad CORS allowlist: The demo server allowed any *.vercel.app origin (anyone can deploy to Vercel). Replaced with blit.sh / *.blit.sh.
• Non-atomic rate limiter: The INCR + EXPIRE in demo.ts were separate Redis calls. If the process crashed between them, the key would persist forever, permanently blocking that IP. Replaced with a Lua script for atomicity.

Motivation

Security hardening and bug fixes identified via an audit of the Rust and TypeScript codebases.

Testing

These are targeted, low-risk fixes to existing code paths. Each fix addresses a specific, well-understood issue. The constant-time comparison, frame size validation, and JSON parse safety fixes are straightforward defensive changes. The CORS and rate limiter fixes affect the demo server only.

Tag @indent to continue the conversation here.

[github-actions]

Coverage

Crate	Lines	Functions	Regions
alacritty-driver	59.7% (529/886)	62.5% (40/64)	59.8% (754/1261)
browser	0.0% (0/807)	0.0% (0/65)	0.0% (0/1370)
cli	34.2% (1072/3132)	44.3% (152/343)	36.0% (1929/5356)
compositor	1.2% (93/7956)	2.3% (8/347)	1.4% (146/10664)
fonts	76.8% (486/633)	85.5% (47/55)	77.9% (922/1183)
gateway	25.3% (355/1405)	27.6% (35/127)	18.8% (442/2353)
proxy	18.4% (150/815)	21.1% (24/114)	20.4% (260/1273)
remote	71.9% (1940/2699)	81.9% (186/227)	74.5% (3674/4929)
server	18.6% (1904/10232)	33.3% (215/645)	20.0% (3198/16014)
ssh	1.9% (7/374)	3.2% (1/31)	0.7% (4/613)
webrtc-forwarder	0.6% (15/2555)	0.5% (1/185)	0.8% (33/4300)
webserver	54.1% (442/817)	64.7% (75/116)	58.6% (839/1432)
Total	21.6% (6993/32311)	33.8% (784/2319)	24.0% (12201/50748)

[github-actions]

🔗 Preview: https://blit-n46d5v50d-indent.vercel.app