[edk2-devel] Http redirection handling using HttpDxe driver

[Vladimir Olovyannikov via groups.io]

> -----Original Message-----
> From: Sean Brogan <spbrogan at outlook.com>
> Sent: Friday, August 28, 2020 8:40 PM
> To: devel at edk2.groups.io; vladimir.olovyannikov at broadcom.com; Rabeda,
> Maciej <maciej.rabeda at linux.intel.com>
> Cc: Laszlo Ersek <lersek at redhat.com>
> Subject: Re: [edk2-devel] Http redirection handling using HttpDxe driver
>
> I think this is the similar to issue 1674 which was reported in early
> 2019.
> https://bugzilla.tianocore.org/show_bug.cgi?id=1674
Thanks Sean,
Indeed, it looks very similar.
For now, I ended up destroying the child explicitly and recreating it if
reconfigure is needed (redirection), inside my app.
This does not affect the common code and works perfectly for the app.

Thank you,
Vladimir
>
> I'll check internally to find out what we ended up doing to work around
> it.  I
> think we closed the socket and opened a new one.  This was less than ideal
> and is not intuitive given the API.
>
> Thanks
> Sean
>
>
>
>
>
>
> On 8/28/2020 10:04 AM, Vladimir Olovyannikov via groups.io wrote:
> >> -----Original Message-----
> >> From: Rabeda, Maciej <maciej.rabeda at linux.intel.com>
> >> Sent: Friday, August 28, 2020 8:27 AM
> >> To: Vladimir Olovyannikov <vladimir.olovyannikov at broadcom.com>
> >> Cc: devel at edk2.groups.io
> >> Subject: Re: Http redirection handling using HttpDxe driver
> >>
> >> Hi Vladimir,
> >>
> >> I got your message, I will look into it next week.
> >> Have a great weekend!
> >>
> > Thank you Maciej,
> > I looked through the code and found that my redirection assumption was
> > incorrect.
> > Apparently, one must destroy the socket before resubmitting a request
> > for a different server.
> > I am going to submit patchset v8. Please review it when you have a
> > chance.
> > @Laszlo
> > This does not affect a direct download when there is no redirection,
> > which you tested.
> >
> > Have a great weekend!
> >
> > Thank you,
> > Vladimir
> >> Maciej
> >>
> >> On 27-Aug-20 21:23, Vladimir Olovyannikov wrote:
> >>> Hi Maciej,
> >>>
> >>> I would like to get your input (and the HttpDxe driver guys) on the
> >>> issue I see with Http redirection (using HttpDynamicCommand)
> >>> Basically, let's say that we connect to a server "server.com"
> >>>
> >>> Request:
> >>> GET / HTTP/1.1
> >>> Host: server.com
> >>>
> >>> Reply:
> >>> HTTP/1.1 301 Moved permanently
> >>> Location: www.server.com
> >>>
> >>> By RFC, the client reissues the request like following and connects
> >>> to
> >>> www.server.com:
> >>> Request:
> >>> GET / HTTP/1.1
> >>> Host: www.server.com
> >>>
> >>> Reply:
> >>> HTTP/1.1 200 OK
> >>>
> >>> I have an issue of getting EFI_ACCESS_DENIED in SockInterface.c on
> >>> the redirection.
> >>> The logic is:
> >>> 1. Send a request
> >>> 2. Parse the response. If there is a redirection code returned, then
> >>>    a) cancel Http connection
> >>>    b) resubmit to the proper host with proper host header.
> >>>
> >>> Obviously, there are no issues with a). However, there is an issue
> >>> with b) When the second request is being sent, EfiHttpRequest sets
> >>> Configure and Reconfigure flags,  closes and cancels Http connection
> >>> with HttpCloseConnection/EfiHttpCancel.
> >>> Everything is OK at this point. Note that the socket is closed, but
> >>> it's ConfiguredState is 1, and is not changed.
> >>> Now, HttpInitSession() is called with Configure flag.
> >>>     It calls into HttpConfigureTcp4(), which in turn calls
> >>>       Tcp4Configure().
> >>>       Here everything is fine until
> >>>         SockConfigure is called.
> >>>         The SockConfigure checks if ConfigureState is "not configured"
> >>> (see above, the ConfigureState is not changed on Http session
> >> cancel/close).
> >>>          So, the SockConfigure returns EFI_ACCESS_DENIED error as
> >>> the socket is configured, and this is returned to HttpConfigureTcp4().
> >>>     Which, in turn, returns an error up, and the Http in
> >>> HttpDynamicCommand gets this error, reports it and aborts the
> operation.
> >>>
> >>> If I change the line in HttpConfigureTcp4, line 1108 in HttpProto.c
> >>> from if (EFI_ERROR (Status)) {
> >>>     DEBUG ((.....));
> >>>     return Status;
> >>> }
> >>> to
> >>> if (EFI_ERROR (Status) && (Status != EFI_ACCESS_DENIED)) {
> >>>     DEBUG ((...));
> >>>     return Status;
> >>> }
> >>> then EFI_SUCCESS is returned, socket placeholder is reused, and
> >>> multiple redirections works properly (I tried this scheme:
> >>> server.com->www.server.com->www1.server.com->got content).
> >>> I suspect, it is wrong to make an assumption that the ACCESS_DEINED
> >>> returned from Tcp4Configure() would be because of the socket.
> >>> Maybe, I am missing something? Can you help?
> >>>
> >>> Thank you,
> >>> Vladimir
> >
> > 
> >

-=-=-=-=-=-=-=-=-=-=-=-
Groups.io Links: You receive all messages sent to this group.

View/Reply Online (#64795): https://edk2.groups.io/g/devel/message/64795
Mute This Topic: https://groups.io/mt/76459258/1813853
Group Owner: devel+owner at edk2.groups.io
Unsubscribe: https://edk2.groups.io/g/devel/unsub  [edk2-devel-archive at redhat.com]
-=-=-=-=-=-=-=-=-=-=-=-