Question about SSL_ERROR_WANT_WRITE

Previous Topic Next Topic
 
classic Classic list List threaded Threaded
3 messages Options
Reply | Threaded
Open this post in threaded view
|

Question about SSL_ERROR_WANT_WRITE

Cosmin Apreutesei
Hello,

I have a question regarding SSL_write() and returning SSL_ERROR_WANT_WRITE from the write callback.

_After_ SSL_write() returns with SSL_ERROR_WANT_WRITE (because my write callback returned  SSL_ERROR_WANT_WRITE), can I _then_ send the data given to the calback and then call SSL_write() again (with the same arguments) and then in the second call to the callback return the number of bytes written? Is that a supported use of the API? (I'm asking because that's the only way I can use the API, I can't send the data inside the callback, I need to send it outside the callback, see below for why).

In other words, is it guaranteed that on that second call to  SSL_write(), SSL will want to send the exact same data that it tried before when it failed, and not change its mind about it wants to send? Because technically, since SSL_ERROR_WANT_WRITE implies that "no data was sent", the state machine might as well advance and send something different at a later time (because it received data or something inside expired or whatever).

---- 

Why I need this: I'm using IOCP and LuaJIT which means I have two limitations: 

1) Because I'm using a completion API as opposed to a readiness API, I can't just tell OpenSSL when the socket is writable and let it write to it, I need to write the data myself.

2) because LuaJIT doesn't allow me to yield from inside a C callback, I can't do async I/O inside the callback, I can only do it in between calls to SSL_read()/SSL_write().

Any suggestions appreciated, thanks!

Reply | Threaded
Open this post in threaded view
|

Re: Question about SSL_ERROR_WANT_WRITE

Kurt Roeckx
On Thu, Dec 10, 2020 at 05:14:00PM +0200, Cosmin Apreutesei wrote:

> Hello,
>
> I have a question regarding SSL_write() and returning SSL_ERROR_WANT_WRITE
> from the write callback.
>
> _After_ SSL_write() returns with SSL_ERROR_WANT_WRITE (because my write
> callback returned  SSL_ERROR_WANT_WRITE), can I _then_ send the data given
> to the calback and then call SSL_write() again (with the same arguments)
> and then in the second call to the callback return the number of bytes
> written? Is that a supported use of the API? (I'm asking because that's the
> only way I can use the API, I can't send the data inside the callback, I
> need to send it outside the callback, see below for why).
>
> In other words, is it guaranteed that on that second call to  SSL_write(),
> SSL will want to send the exact same data that it tried before when it
> failed, and not change its mind about it wants to send? Because
> technically, since SSL_ERROR_WANT_WRITE implies that "no data was sent",
> the state machine might as well advance and send something different at a
> later time (because it received data or something inside expired or
> whatever).

I don't fully understand your question, but the manpage says:

WARNINGS
       When a write function call has to be repeated because
SSL_get_error(3) returned SSL_ERROR_WANT_READ or
SSL_ERROR_WANT_WRITE, it must be repeated with the same arguments.
The data that was passed might have been partially processed.
When SSL_MODE_ACCEPT_MOVING_WRITE_BUFFER was set using
SSL_CTX_set_mode(3) the pointer can be different, but the data and
length should still be the same.

Does that answer your question?


Kurt

Reply | Threaded
Open this post in threaded view
|

Re: Question about SSL_ERROR_WANT_WRITE

Kyle Hamilton
In reply to this post by Cosmin Apreutesei
If you get SSL_ERROR_WANT_WRITE, call the same function with the same parameters and same buffer content immediately. (Same with SSL_ERROR_WANT_READ.)

If you need to, stash those parameters in variables for ease of reference.  But don't do anything else on the SSL layer until you get a different return value. (If you implement your own BIO layer, do what you need to do in support of what you're asked to do for the raw I/O. But don't do anything with the SSL layer until its internal state has moved past the need to read or write to or from the BIO.)

-Kyle H

On Thu, Dec 10, 2020, 09:14 Cosmin Apreutesei <[hidden email]> wrote:
Hello,

I have a question regarding SSL_write() and returning SSL_ERROR_WANT_WRITE from the write callback.

_After_ SSL_write() returns with SSL_ERROR_WANT_WRITE (because my write callback returned  SSL_ERROR_WANT_WRITE), can I _then_ send the data given to the calback and then call SSL_write() again (with the same arguments) and then in the second call to the callback return the number of bytes written? Is that a supported use of the API? (I'm asking because that's the only way I can use the API, I can't send the data inside the callback, I need to send it outside the callback, see below for why).

In other words, is it guaranteed that on that second call to  SSL_write(), SSL will want to send the exact same data that it tried before when it failed, and not change its mind about it wants to send? Because technically, since SSL_ERROR_WANT_WRITE implies that "no data was sent", the state machine might as well advance and send something different at a later time (because it received data or something inside expired or whatever).

---- 

Why I need this: I'm using IOCP and LuaJIT which means I have two limitations: 

1) Because I'm using a completion API as opposed to a readiness API, I can't just tell OpenSSL when the socket is writable and let it write to it, I need to write the data myself.

2) because LuaJIT doesn't allow me to yield from inside a C callback, I can't do async I/O inside the callback, I can only do it in between calls to SSL_read()/SSL_write().

Any suggestions appreciated, thanks!