In a nutshell:
- FTP connections consist of two streams: command (control) and data
- When a socket is opened and code does not explicitly bind to a port, the OS will consume a high numbered port in its ephemeral port range (
sysctl -a | grep ip_local_port_range)
- Non-root processes must use ports >1024
Active Mode FTP
Client connects to server over port 21 and establishes a command stream:
client:32198 -> server:21
Client must send or receive some data, so it informs the server to connect back to it on some port. To do this, it sends a PORT command similar to this:
This is the client informing the server that the server may connect back to the client (at address 18.104.22.168 on port (5 * 256) + 6 = 1286.
server:20 -> client:1286
Typically, this is where you see active mode FTP sessions die; with respect to firewalls and load balancers, the traffic normally flowing from client -> server is expected, but connections initiated by the server to the client are often denied (load balancers are often smart enough to tie associate this data stream with the existing command stream).
Your understanding about needing to open a port range on your firewall to facilitate this behavior is absolutely correct.
Passive Mode FTP
In this scenario, the client establishes a command session as before:
client:56221 -> server:21
But when data is communicated, the client sends a
PASV command primitive. The server responds with an IP:Port combination that the client should connect back to (in a similar format to the
PORT request earlier. So the client then connects to the server as follows:
client:12347 -> server:4566
This bypasses the firewall issues described above for active mode because the connections are established in a traditional and expected manner.
The downside to passive mode is that it consumes more sockets on the server. Issuing frequent
PASV primitives in a heavily loaded environment could eventually lead to port exhaustion. (Sockets will linger in a
TIME_WAIT state for some OS-defined amount of time (~2 minutes I think on Redhat systems).
About your problem
Unless you really are suffering from port exhaustion issues, it is highly unusual that passive mode would fail and active mode would more often succeed. Usually is is the other way around. If you are able to post more about the errors you occasionally receive, we can debug it further.
I would recommend using passive mode wherever possible, and so I would recommend looking into the passive-mode specific failures to find the root cause of your issue before conceding to using active mode FTP.