[pve-devel] [PATCH widget-toolkit] fix external linking to products by setting cookie SameSite attribute to lax

[Max Carrara]

On Mon Oct 7, 2024 at 5:02 PM CEST, Dominik Csapak wrote:
> We introduced the 'strict' setting when browsers warned about our
> cookies not having any SameSite setting [0]
>
> While this works in general, it had an unforeseen side effect:
>
> When linking to a URL of our products, the cookie does not get sent on
> the initial page load, leading to the username/CSRFPreventionToken not
> being set in the index response.
>
> Our UI code interprets this as being logged out (e.g. because the ticket
> is not valid) and clears the cookie, displaying the login window.
>
> The MDN reference[1] says that setting it to 'lax' is similar to
> 'strict', but sends the cookie when navigating *to* our origin even from
> other sites, which is what we want when linking from elsewhere. (This
> would have also been the default if we wouldn't have set any attribute).
>
> 0: https://lore.proxmox.com/pve-devel/20230315162630.289768-1-m.carrara@proxmox.com/
> 1: https://developer.mozilla.org/en-US/docs/Web/HTTP/Cookies#SameSite_attribute
>
> Signed-off-by: Dominik Csapak <d.csapak at proxmox.com>

As mentioned off list, this seems fine to me, as our API doesn't have
any side effects during GET requests (otherwise, we'd have different
problems anyway ...) and we only modify, delete etc. via other methods.

Though, as you probably saw in the series I had sent in back then, I
also set SameSite=Strict in a couple other places:

- pve-apiclient/trees/master/src/PVE/APIClient/LWP.pm
- pve-http-server/trees/master/src/PVE/APIServer/Formatter.pm
- pve-http-server/trees/master/src/PVE/APIServer/Formatter/Bootstrap.pm

I guess it would make sense to set them to `lax` there too?

> ---
> Alternatively, we could try to modify the client code to retry with the
> content of the cookie (which the client still has), to generate a
> ticket. Though I'm not sure that will work correctly, since I observed
> some even stricter behaviour in Firefox ESR, where not only the
> navigation part did not send the cookie, but also subsequent requests
> did not do that (I did not test for long, so this may have been an
> artifact of a different issue).
>
> I'd generally prefer this solution, since it's less intrusive and
> restores the behaviour we had before the change, altough I'm not against
> implementing a retry in the client code if that's preferred.
>
>  src/Utils.js | 4 ++--
>  1 file changed, 2 insertions(+), 2 deletions(-)
>
> diff --git a/src/Utils.js b/src/Utils.js
> index 7dd034a..b68c0f4 100644
> --- a/src/Utils.js
> +++ b/src/Utils.js
> @@ -317,7 +317,7 @@ utilities: {
>  	// that way the cookie gets deleted after the browser window is closed
>  	if (data.ticket) {
>  	    Proxmox.CSRFPreventionToken = data.CSRFPreventionToken;
> -	    Ext.util.Cookies.set(Proxmox.Setup.auth_cookie_name, data.ticket, null, '/', null, true, "strict");
> +	    Ext.util.Cookies.set(Proxmox.Setup.auth_cookie_name, data.ticket, null, '/', null, true, "lax");
>  	}
>  
>  	if (data.token) {
> @@ -343,7 +343,7 @@ utilities: {
>  	    return;
>  	}
>  	// ExtJS clear is basically the same, but browser may complain if any cookie isn't "secure"
> -	Ext.util.Cookies.set(Proxmox.Setup.auth_cookie_name, "", new Date(0), null, null, true, "strict");
> +	Ext.util.Cookies.set(Proxmox.Setup.auth_cookie_name, "", new Date(0), null, null, true, "lax");
>  	window.localStorage.removeItem("ProxmoxUser");
>      },
>