Passing Configuration to Middleware

[zenazn]

Passing Configuration to Middleware

[dre1080]

Hey @elithrar,
Wouldn't be using nosurf be much easier for CSRF?

[zenazn]

The way I did this last time is something like this:

type Params struct {
    foo int
}

type myMiddleware struct {
    h http.Handler
    c *web.C
    p Params
}

func (m myMiddleware) ServeHTTP(w http.ResponseWriter, r *http.Request) {
    // ...
    m.h.ServeHTTP(w, r)
    // ...
}

func MyMiddleware(p Params) func(*web.C, http.Handler) http.Handler {
    return func(c *web.C, h http.Handler) {
        return myMiddleware{h, c, p}
    }
}

// You can store this in a variable if you want to reference it again
goji.Use(MyMiddleware(Params{
    foo: 4,
}))

The way you're doing it would work as well. I haven't given much thought to which I prefer, but I think the way I wrote above probably will result in one fewer helper struct.

As for wrapping Gorilla, I've found that Gorilla does some... questionable things with global maps, which tends to result in unnecessary lock contention and (more importantly) the tendency to leak memory unless you're careful. I haven't looked very hard at nosurf, but it should be trivial to drop in to Goji and probably doesn't do quite as much spooky action at a distance as Gorilla.

[elithrar]

@zenazn Thanks for the example code. Extending/wrapping http.Handler is actually something I didn't think of at first, and it does look a bit cleaner. I'm definitely aware of the need to call context.ClearHandler(myHandler) to avoid any leaks from gorilla (I know that's a gotcha when not using gorilla/mux alongside), but beyond that I plan to use goji's own request context to pass the token to the wrapped handler and avoid Gorilla's there. nosurf still uses a global context map anyway so I'm not sure it provides any distinct benefits on that front.

@dre1080 @zenazn I like nosurf, but you're also entirely tied to a cookie-based session store. Being able to use server-side sessions (particularly if you are already) is a nice win for security, and you also avoid the 4K limit on session size that browsers enforce on cookies. I'm also not quite sure nosurf's cookie contents are encrypted (as gorilla/sessions allows) from looking through the source, although if you're dealing with any POST data over vanilla HTTP that's your second problem ;)

I'm happy to take advice otherwise, but I think there's some value in providing some Goji-centric anti CSRF middleware.

[zenazn]

RE: nosurf's global map: oh dear.

Anyways, you seem to know what you're doing so I'll stop bothering you with my unfounded opinions :)

[elithrar]

Closing this specific one out - but am still working on the CSRF & sessions middleware when I can find some spare time.