We are focusing on fixing issues and security vulnerabilities to the latest and greatest version of Iris. Iris users SHOULD always update their backend code to the latest version of Iris API, there is a complete HISTORY file which acts as a code migration assistant.
Please report (suspected) security vulnerabilities to iris-go@outlook.com. You will receive a response from us within 3-4 working days. If the issue is confirmed, we will release a new minor version as soon as possible depending on complexity but historically within a few hours or days.
Some Iris defaults favour "it works out of the box" over "it is locked down". They are kept for backwards compatibility, and each has an opt-in hardening. Review this list when you put an application on the public internet.
| Default | What it means | Harden with |
|---|---|---|
cors.New() reflects any Origin and advertises Access-Control-Allow-Credentials: true |
A page on any domain can read your authenticated responses with fetch(..., {credentials: "include"}) |
cors.Strict(), and AllowOrigin/AllowOrigins |
ContextOptions.BodyMaxMemory is 0 (unlimited) |
A client picks the size of the allocation your handler makes | Set BodyMaxMemory, or use the bodylimit middleware |
compress.Handler decompresses without a ceiling |
A ~1MB gzip body expands to ~1GB; a handful of requests exhausts memory | compress.Safe, or compress.Limited(n) |
cache.DefaultOptions marks responses public, max-age=3600 |
A CDN or proxy may store a per-user response and serve it to someone else | cache.PrivateOptions (credentialed requests are already skipped) |
httpcache keys do not include Authorization or cookies |
Two users' identical requests collapse onto one cache entry | Name the header in Options.Headers and every cookie the request sends in Options.Cookies: a request with credentials is skipped until the key includes each one of them. Register the cache handler before sessions.Handler, which otherwise writes its cookie after the page was stored. A response whose Cache-Control carries private, no-store or no-cache is never stored, so send private from a per-user handler |
accesslog logs the request body |
Passwords, tokens and card numbers land in your logs | Set AccessLog.Redact, or RequestBody = false |
methodoverride accepts _method from the query string and form body |
A cross-site form POST can reach a DELETE-only route |
methodoverride.New(methodoverride.Options{HeadersOnly: true}) |
fileserver serves dotfiles |
.git/config and .env are published if they sit under the served root |
Options.AssetValidator = fileserver.BlockHidden |
pprof and monitor have no built-in auth |
Goroutine dumps and host metrics are readable by anyone who finds the route | Wrap them in basicauth or ipaccess, or bind them to an internal listener |
Context.Addr() returns the client-supplied proxy header value |
Not usable for access control | Use Context.RemoteAddr(), as ipaccess does, and only enable RemoteAddrHeaders behind a proxy that overwrites them |
| Sessions do not rotate the id on login | Session fixation | Call Session.RegenerateID after any privilege change |
Two more things worth knowing:
- Proxy headers are opt-in and all-or-nothing.
Options.RemoteAddrHeadersis empty by default, andContext.RemoteAddr()is then the address of the connection. Once you set it, every client can send those headers, so only enable it behind a proxy you control. A proxy appends the address it got the request from, soRemoteAddr()reads each header from right to left: the first public address is the result, without its port, and a private address is skipped as one of your own proxies. An entry that is not an address (unknown,unix:) ends the reading of that header with no result. The next header is read then, and at last the address of the connection is used. Theipaccessmiddleware decides onRemoteAddr(), so what follows holds for its allowlist and its blocklist too. What the client wrote is on the left, and it can still be the result in three cases. The first is a header in which every entry to its right is a private address, as for a request that reached your proxy from inside your network. The second isRemoteAddrHeadersForce, which returns the first address from the left of a header whose entries are all addresses and none of them public. The third is a header your proxy does not write: the proxy passes it on as the client sent it, and it is read when it is named first inRemoteAddrHeaders, or when the headers named before it give no public address. A proxy that overwrites the header closes the first two, and naming only headers your proxy writes closes the third.Context.Addr()needs the overwriting proxy in any case, because it reads from the left. The result is not always the client either. A proxy with a public address in front of your own, a CDN for example, is the first public address from the right, soRemoteAddr()returns the CDN and not the client it got the request from, until you add its IPv4 range toRemoteAddrPrivateSubnets. - MVC controllers are singletons. A controller is resolved once and its methods are bound into handlers at registration. Fields are shared across all concurrent requests; keep per-request state on the Context.