Ugly work around to (hopefully) allow encryption to work when receiving multiplexed streams

protect calling applications from our use/abuse of the url character buffer (fixes multiplex sender bug)

Change the input thread to only sleep 5ms at a time when waiting for data. This fixes a bug where the application refuses to die with control+c

Refactor dataout_fifo_queue to be per flow and not per context. Data on it was being shared by multiple flows which resulted in a totally broken buffer.

Nonce should not be generated by generate_flowid function as it sets the LSB to 0
effectively setting the nonce to 31 bits of random data.
Also rand() was not being properly seeded.

Refactor stream-id. We are now tying it to the destination port and not the source port of the internal GRE tunnel.

Group the sender peers together into a single context but keep individual separate contexts per receiver

1) We now use one library context per sender, each with it's own RTCP channel and separate SSRC
2) We now spin a thread for each UDP socket we listen to
3) We now support rist as a form of input in addition to UDP and RTP

Windows Docker image seems to be missing CMake and MacOS fails on downloading mbedtls

Make the required changes in meson.config file for building via subproject
Allow turning off of mbedtls if desired

Allows for accelerated AES on more platforms beside linux, as well as
DTLS support which we can work on for future release.

TODO:
- work out meson logic, test against encryption via linux crypto
and C AES.
- add mbedtls as a subproject in meson and have it build in case library
  is not available (i.e.: Windows builds)

+ don't change our configured ports to track remote when a new client comes in and we are sending

Gijs Peskens authored 4 years ago and Sergio Ammirata committed 4 years ago

Should prevent packets that arrive too late getting NACK'ed (which might also get send if
sender latency buffer is greater than ours)

Add extra periodic error log line for lost packets

Gijs Peskens authored 4 years ago and Sergio Ammirata committed 4 years ago

Be a bit more aggresive in resetting flow: also reset when we dropped too many packets due to buffer being full (fixes rare issue with upipe sender)
Help compiler out by indicating unlikely conditions

Gijs Peskens authored 4 years ago and Sergio Ammirata committed 4 years ago

Also move lots of logging to debug

Still might need to keep more counters and print them from within an interval timer

The peer lookup code does not limit it's search to peers originating from the same socket,
RIST simple profile connections where the RTCP and DATA connections originate from the same
source IP and port combinations thus don't get correctly setup, resulting in a segfault when
NACKS are to be transmitted as the peer->peer_rtcp is not populated.

When a flow times out and get's deleted it's childpeers are not correctly updated
to reflect this, causing a segfault when the (/ a) flow returns on the same peers
(matched via src port + src ip + target port + target ip)

Fixes issue on Windows where we are unable to open a ipv4 wildcard socket.
To open a ipv6 wildcard socket just enter: rist://@[::]:1234, on Linux this
will also receive ipv4 packets.