chore: restructure README with required policy sections

This commit is contained in:
2026-02-25 19:47:34 +07:00
parent d0fe5e7334
commit 73f1073d61

376
README.md
View File

@@ -1,324 +1,138 @@
# pixa caching image reverse proxy server
# pixa
This is a web service written in go that is designed to proxy images from
source URLs, optionally resizing or transforming them, and serving the
results. Both the source images as well as the transformed images are
cached. The images served to the client are cached a configurable interval
so that subsequent requests to the same path on the pixa server are served
from disk without origin server requests or additional processing.
pixa is a GPL-3.0-licensed Go web server by
[@sneak](https://sneak.berlin) that proxies images from upstream
sources, optionally resizing or transforming them, and serves the
results. Both source and transformed images are cached to disk so that
subsequent requests are served without origin fetches or additional
processing.
# storage
## Getting Started
* unaltered source file straight from upstream:
* `<statedir>/cache/src-content/<ab>/<cd>/<abcdef0123... sha256 of source content>`
* source path metadata
* `<statedir>/cache/src-metadata/<hostname>/<sha256 of path component>.json`
* fetch time
* all original resp headers
* original request
* sha256 hash
```bash
# clone and build
git clone https://git.eeqj.de/sneak/pixa.git
cd pixa
make build
Note that multiple source paths may reference the same content blob. We
won't do refcounting here, we'll use the state database for that.
# run with a config file
./bin/pixad --config config.example.yml
* database:
* `<statedir>/state.sqlite3`
# or build and run via Docker
make docker
docker run -p 8080:8080 pixad:latest
```
* output documents:
* `<statedir>/cache/dst-content/<ab>/<cd>/<abcd... sha256 of output content>`
## Rationale
While the database is the long-term authority on what we have in the output
cache, we must aggressively cache in-process the mapping between requests
and output content hashes so as to serve as a maximally efficient caching
proxy for extremely popular/hot request paths. The goal is the ability to
easily support 1-5k r/s.
Image-heavy web applications need a fast, caching reverse proxy that
can resize and transcode images on the fly. pixa fills that role as a
single, self-contained binary with no external runtime dependencies
beyond libvips. It supports HMAC-SHA256 signed URLs with expiration to
prevent abuse, and whitelisted source hosts for open access.
# Routes
## Design
/img/<size>/<orig host>/<orig path>?signature=<sig>&format=<format>
### Storage
Images are only fetched from origins using TLS. Origin certificates must be
valid at time of fetch.
- **Source content**:
`<statedir>/cache/src-content/<ab>/<cd>/<sha256 of source content>`
- **Source metadata**:
`<statedir>/cache/src-metadata/<hostname>/<sha256 of path>.json`
(fetch time, original headers, request, content hash)
- **Database**: `<statedir>/state.sqlite3` (SQLite)
- **Output documents**:
`<statedir>/cache/dst-content/<ab>/<cd>/<sha256 of output content>`
<format> is one of 'orig', 'png', 'jpeg', 'webp'
Multiple source paths may reference the same content blob; the
database tracks references rather than using filesystem refcounting.
In-process caching of request-to-output mappings targets 1-5k r/s.
<size> is one of 'orig' or '<x resolution>x<y resolution>'
### Routes
# Source Hosts
```
/v1/image/<host>/<path>/<size>.<format>?sig=<signature>&exp=<expiration>
```
Source hosts may be whitelisted in the pixa configuration. If not in the
explicit whitelist, a signature using a shared secret must be appended.
Images are only fetched from origins using TLS with valid certificates.
## Signature Specification
- `<format>`: one of `orig`, `png`, `jpeg`, `webp`
- `<size>`: `orig` or `<width>x<height>` (e.g. `800x600`)
Signatures use HMAC-SHA256 and include an expiration timestamp to prevent replay attacks.
### Source Hosts
### Signed Data Format
Source hosts may be whitelisted in the configuration. Non-whitelisted
hosts require an HMAC-SHA256 signature.
The signature is computed over a colon-separated string:
#### Signature Specification
Signatures use HMAC-SHA256 and include an expiration timestamp to
prevent replay attacks.
**Signed data format** (colon-separated):
```
HMAC-SHA256(secret, "host:path:query:width:height:format:expiration")
```
Where:
- `host` - Source origin hostname (e.g., `cdn.example.com`)
- `path` - Source path (e.g., `/photos/cat.jpg`)
- `query` - Source query string, empty string if none
- `width` - Requested width in pixels, `0` for original
- `height` - Requested height in pixels, `0` for original
- `format` - Output format (jpeg, png, webp, avif, gif, orig)
- `expiration` - Unix timestamp when signature expires
### URL Format with Signature
- `host` — source origin hostname (e.g. `cdn.example.com`)
- `path` — source path (e.g. `/photos/cat.jpg`)
- `query` — source query string, empty string if none
- `width` — requested width in pixels, `0` for original
- `height` — requested height in pixels, `0` for original
- `format` — output format (jpeg, png, webp, avif, gif, orig)
- `expiration` — Unix timestamp when signature expires
```
/v1/image/<host>/<path>/<size>.<format>?sig=<signature>&exp=<expiration>
```
### Example
For a request to resize `https://cdn.example.com/photos/cat.jpg` to 800x600 WebP
with expiration at Unix timestamp 1704067200:
1. Build the signature input:
```
cdn.example.com:/photos/cat.jpg::800:600:webp:1704067200
```
**Example:** resize
`https://cdn.example.com/photos/cat.jpg` to 800x600 WebP with
expiration 1704067200:
1. Build input:
`cdn.example.com:/photos/cat.jpg::800:600:webp:1704067200`
2. Compute HMAC-SHA256 with your secret key
3. Base64URL-encode the result
4. URL:
`/v1/image/cdn.example.com/photos/cat.jpg/800x600.webp?sig=<base64url>&exp=1704067200`
4. Final URL:
```
/v1/image/cdn.example.com/photos/cat.jpg/800x600.webp?sig=<base64url>&exp=1704067200
```
**Whitelist patterns:**
### Whitelist Patterns
- **Exact match**: `cdn.example.com` — matches only that host
- **Suffix match**: `.example.com` — matches `cdn.example.com`,
`images.example.com`, and `example.com`
The whitelist supports two pattern types:
- **Exact match**: `cdn.example.com` - matches only that host
- **Suffix match**: `.example.com` - matches `cdn.example.com`, `images.example.com`, and `example.com`
### Configuration
# configuration
Configured via YAML file (`--config`). Key settings:
* access-control-allow-origin config
* source host whitelist
* upstream fetch timeout
* upstream max response size
* downstream timeout
* downstream max request size
* downstream max response size
* internal processing timeout
* referer blacklist
- `access_control_allow_origin` — CORS origin
- `source_host_whitelist` — list of allowed upstream hosts
- `upstream_fetch_timeout` — timeout for origin requests
- `upstream_max_response_size` — max origin response size
- `downstream_timeout` — client response timeout
- `signing_key` — HMAC secret for URL signatures
# Design Review & Recommendations
See `config.example.yml` for all options with defaults.
## Security Concerns
### Architecture
### Critical
- **HMAC signature scheme is undefined** - The "FIXME" for signature
construction is a blocker. Recommend HMAC-SHA256 over the full path:
`HMAC-SHA256(secret, "/<size>/<host>/<path>?format=<format>")`
- **No signature expiration** - Signatures should include a timestamp to
prevent indefinite replay. Add `&expires=<unix_ts>` and include it in the
HMAC input
- **Path traversal risk** - Ensure `<orig path>` cannot contain `..`
sequences or be used to access unintended resources on origin
- **SSRF potential** - Even with TLS requirement, internal/private IPs
(10.x, 172.16.x, 192.168.x, 127.x, ::1, link-local) must be blocked to
prevent server-side request forgery
- **Open redirect via Host header** - Validate that requests cannot be
manipulated to cache content under incorrect keys
- **Dependency injection**: Uber fx
- **HTTP router**: go-chi
- **Image processing**: govips (CGO wrapper for libvips)
- **Database**: SQLite via modernc.org/sqlite
- **Static assets**: embedded via `//go:embed`
- **Metrics**: Prometheus
- **Logging**: stdlib slog
### Important
- **No authentication for cache purge** - If cache invalidation is needed, it requires auth
- **Response header sanitization** - Strip sensitive headers from upstream before forwarding (X-Powered-By, Server, etc.)
- **Content-Type validation** - Verify upstream Content-Type matches expected image types before processing
- **Maximum image dimensions** - Limit output dimensions to prevent resource exhaustion (e.g., max 4096x4096)
## TODO
## URL Route Improvements
See [TODO.md](TODO.md) for the full prioritized task list.
Current: `/img/<size>/<orig host>/<orig path>?signature=<sig>&format=<format>`
## License
### Recommended Scheme
```
/v1/image/<host>/<path>/<width>x<height>.<format>?sig=<sig>&exp=<expires>
```
GPL-3.0. See [LICENSE](LICENSE).
The size+format segment (e.g., `800x600.webp`) is appended to the source path and stripped when constructing the upstream request. This pattern is unambiguous (regex: `(\d+x\d+|orig)\.(webp|jpg|jpeg|png|avif)$`) and won't collide with real paths.
## Author
**Size options:**
- `800x600.<format>` - resize to 800x600
- `0x0.<format>` - original size, format conversion only
- `orig.<format>` - original size, format conversion only (human-friendly alias)
**Benefits:**
- API versioning (`/v1/`) allows breaking changes later
- Human-readable URLs that can be manually constructed for whitelisted domains
- Format as extension is intuitive and CDN-friendly
### Examples
**Basic resize and convert:**
```
/v1/image/cdn.example.com/photos/cat.jpg/800x600.webp?sig=abc123&exp=1704067200
```
Fetches `https://cdn.example.com/photos/cat.jpg`, resizes to 800x600, converts to webp.
**Source URL with query parameters:**
```
/v1/image/cdn.example.com/photos/cat.jpg%3Farg1=val1%26arg2=val2/800x600.webp?sig=abc123&exp=1704067200
```
Fetches `https://cdn.example.com/photos/cat.jpg?arg1=val1&arg2=val2`, resizes to 800x600, converts to webp.
Note: The source query string must be URL-encoded (`?` → `%3F`, `&` → `%26`) to avoid ambiguity with pixa's own query parameters.
**Original size, format conversion only:**
```
/v1/image/cdn.example.com/photos/cat.jpg/orig.webp?sig=abc123&exp=1704067200
/v1/image/cdn.example.com/photos/cat.jpg/0x0.webp?sig=abc123&exp=1704067200
```
Both fetch the original image and convert to webp without resizing.
## Additional Formats
### Output Formats to Support
- `avif` - Superior compression, growing browser support
- `gif` - For animated image passthrough (with frame limit)
- `svg` - Passthrough only, no resizing (vector)
### Input Format Whitelist (MIME types to accept)
- `image/jpeg`
- `image/png`
- `image/webp`
- `image/gif`
- `image/avif`
- `image/svg+xml` (passthrough or rasterize)
- **Reject all others** - Especially `image/x-*`, `application/*`
### Input Validation
- Verify magic bytes match declared Content-Type
- Maximum input file size (e.g., 50MB)
- Maximum input dimensions (e.g., 16384x16384)
- Reject files with embedded scripts (SVG sanitization)
## Rate Limiting
### Per-IP Limits
- Requests per second (e.g., 10 req/s burst, 100 req/min sustained)
- Concurrent connections (e.g., 50 per IP)
### Global Limits
- Total concurrent upstream fetches (prevent origin overwhelm)
- Per-origin fetch rate limiting (be a good citizen)
- Cache miss rate limiting (prevent cache-busting attacks)
### Response
- Return `429 Too Many Requests` with `Retry-After` header
- Consider `X-RateLimit-*` headers for transparency
## Additional Features for 1.0
### Must Have
- **Health check endpoint** - `/health` or `/healthz` for load balancers
- **Metrics endpoint** - `/metrics` (Prometheus format) for observability
- **Graceful shutdown** - Drain connections on SIGTERM
- **Request ID/tracing** - `X-Request-ID` header propagation
- **Cache-Control headers** - Proper `Cache-Control`, `ETag`, `Last-Modified` on responses
- **Vary header** - `Vary: Accept` if doing content negotiation
### Should Have
- **Auto-format selection** - If `format=auto`, pick best format based on `Accept` header
- **Quality parameter** - `&q=85` for lossy format quality control
- **Fit modes** - `fit=cover|contain|fill|inside|outside` for resize behavior
- **Background color** - For transparent-to-JPEG conversion
- **Blur/sharpen** - Common post-resize operations
- **Watermarking** - Optional overlay support
### Nice to Have
- **Cache warming API** - Pre-populate cache for known images
- **Cache stats API** - Hit/miss rates, storage usage
- **Admin UI** - Simple dashboard for monitoring
## Configuration Additions
```yaml
server:
listen: ":8080"
read_timeout: 30s
write_timeout: 60s
max_header_bytes: 8192
cache:
directory: "/var/cache/pixa"
max_size_gb: 100
ttl: 168h # 7 days
negative_ttl: 5m # Cache 404s briefly
upstream:
timeout: 30s
max_response_size: 52428800 # 50MB
max_concurrent: 100
user_agent: "Pixa/1.0"
processing:
max_input_pixels: 268435456 # 16384x16384
max_output_dimension: 4096
default_quality: 85
strip_metadata: true # Remove EXIF etc.
security:
hmac_secret: "${PIXA_HMAC_SECRET}" # From env
signature_ttl: 3600 # 1 hour
blocked_networks:
- "10.0.0.0/8"
- "172.16.0.0/12"
- "192.168.0.0/16"
- "127.0.0.0/8"
- "::1/128"
- "fc00::/7"
rate_limit:
per_ip_rps: 10
per_ip_burst: 50
per_origin_rps: 100
cors:
allowed_origins: ["*"] # Or specific list
allowed_methods: ["GET", "HEAD", "OPTIONS"]
max_age: 86400
```
## Error Handling
### HTTP Status Codes
- `400` - Bad request (invalid parameters, malformed URL)
- `403` - Forbidden (invalid/expired signature, blocked origin)
- `404` - Origin returned 404 (cache negative response briefly)
- `413` - Payload too large (origin image exceeds limits)
- `415` - Unsupported media type (origin returned non-image)
- `422` - Unprocessable (valid image but cannot transform as requested)
- `429` - Rate limited
- `500` - Internal error
- `502` - Bad gateway (origin connection failed)
- `503` - Service unavailable (overloaded)
- `504` - Gateway timeout (origin timeout)
### Error Response Format
```json
{
"error": "invalid_signature",
"message": "Signature has expired",
"request_id": "abc123"
}
```
## Quick Wins
1. **Conditional requests** - Support `If-None-Match` / `If-Modified-Since` to return `304 Not Modified`
2. **HEAD support** - Allow clients to check image metadata without downloading
3. **Canonical URLs** - Redirect non-canonical requests to prevent cache fragmentation
4. **Debug header** - `X-Pixa-Cache: HIT|MISS|STALE` for debugging
5. **Robots.txt** - Serve a robots.txt to prevent search engine crawling of proxy URLs
[@sneak](https://sneak.berlin)