[FFmpeg-devel] [PATCH] avformat/tls_schannel: forward AVIO_FLAG_NONBLOCK to tcp stream

[Gyan Doshi]

On 2024-06-18 10:00 pm, Timo Rothenpieler wrote:
> On 11.06.2024 15:10, Timo Rothenpieler wrote:
>> On 03.06.2024 22:28, Timo Rothenpieler wrote:
>>> From: BtbN <btbn at btbn.de>
>>
>> This is fixed locally
>>
>>> Fixes for example rtmps streaming over schannel.
>>> ---
>>>   libavformat/tls_schannel.c | 15 ++++++++++++++-
>>>   1 file changed, 14 insertions(+), 1 deletion(-)
>>>
>>> diff --git a/libavformat/tls_schannel.c b/libavformat/tls_schannel.c
>>> index 214a47a218..7265a9794d 100644
>>> --- a/libavformat/tls_schannel.c
>>> +++ b/libavformat/tls_schannel.c
>>> @@ -113,6 +113,7 @@ static int tls_shutdown_client(URLContext *h)
>>> c->request_flags, 0, 0, NULL, 0, &c->ctxt_handle,
>>> &outbuf_desc, &c->context_flags, &c->ctxt_timestamp);
>>>           if (sspi_ret == SEC_E_OK || sspi_ret == 
>>> SEC_I_CONTEXT_EXPIRED) {
>>> +            s->tcp->flags &= ~AVIO_FLAG_NONBLOCK;
>>>               ret = ffurl_write(s->tcp, outbuf.pvBuffer, 
>>> outbuf.cbBuffer);
>>>               FreeContextBuffer(outbuf.pvBuffer);
>>>               if (ret < 0 || ret != outbuf.cbBuffer)
>>> @@ -316,6 +317,7 @@ static int tls_client_handshake(URLContext *h)
>>>           goto fail;
>>>       }
>>> +    s->tcp->flags &= ~AVIO_FLAG_NONBLOCK;
>>>       ret = ffurl_write(s->tcp, outbuf.pvBuffer, outbuf.cbBuffer);
>>>       FreeContextBuffer(outbuf.pvBuffer);
>>>       if (ret < 0 || ret != outbuf.cbBuffer) {
>>> @@ -416,11 +418,16 @@ static int tls_read(URLContext *h, uint8_t 
>>> *buf, int len)
>>>               }
>>>           }
>>> +        s->tcp->flags &= ~AVIO_FLAG_NONBLOCK;
>>> +        s->tcp->flags |= h->flags & AVIO_FLAG_NONBLOCK;
>>> +
>>>           ret = ffurl_read(s->tcp, c->enc_buf + c->enc_buf_offset,
>>>                            c->enc_buf_size - c->enc_buf_offset);
>>>           if (ret == AVERROR_EOF) {
>>>               c->connection_closed = 1;
>>>               ret = 0;
>>> +        } else if (ret == AVERROR(EAGAIN)) {
>>> +            ret = 0;
>>>           } else if (ret < 0) {
>>>               av_log(h, AV_LOG_ERROR, "Unable to read from socket\n");
>>>               return ret;
>>> @@ -564,8 +571,14 @@ static int tls_write(URLContext *h, const 
>>> uint8_t *buf, int len)
>>>       sspi_ret = EncryptMessage(&c->ctxt_handle, 0, &outbuf_desc, 0);
>>>       if (sspi_ret == SEC_E_OK)  {
>>>           len = outbuf[0].cbBuffer + outbuf[1].cbBuffer + 
>>> outbuf[2].cbBuffer;
>>> +
>>> +        s->tcp->flags &= ~AVIO_FLAG_NONBLOCK;
>>> +        s->tcp->flags |= h->flags & AVIO_FLAG_NONBLOCK;
>>> +
>>>           ret = ffurl_write(s->tcp, data, len);
>>> -        if (ret < 0 || ret != len) {
>>> +        if (ret == AVERROR(EAGAIN)) {
>>> +            goto done;
>>> +        } else if (ret < 0 || ret != len) {
>>>               ret = AVERROR(EIO);
>>>               av_log(h, AV_LOG_ERROR, "Writing encrypted data to 
>>> socket failed\n");
>>>               goto done;
>>
>> ping
>>
>>
>> I'm specifically unsure if implementing the sending-side like this is 
>> valid and would appreciate review from someone familiar with the code 
>> and schannel.

FWIW, I had to do the same for securetransport on a project a couple of 
years back to get rtmps working. Worked fine, and did not get any 
reports of ill-effects.

Regards,
Gyan