Whether you’re really passionate about RPC, MQTT, Matrix or wayland, tell us more about the protocols or open standards you have strong opinions on!

  • Mango@lemmy.world
    link
    fedilink
    arrow-up
    9
    ·
    8 months ago

    FTP

    Seriously guys, let’s share files the old fashioned way. Without bullshit.

    • Badabinski@kbin.social
      link
      fedilink
      arrow-up
      8
      ·
      edit-2
      8 months ago

      I’d like to interject for a moment. What you’re referring to as FTP is, in fact, smelly hot garbage.

      For context, I wrote this while waiting for a migraine to pass. I was angry at my brain for ruining my morning, and I like to shit on FTP. It’s fun to be hyperbolic. I don’t intend for this to be an attack on you, I was just bored and decided to write this ridiculous rant to pass the time.

      I must once again rant about FTP. I’ve no idea if you’re serious about liking it or you’re just taking the piss, but seeing those three letters surrounded by whitespace reminds me of all the bad things in the world.

      FTP is, as I’ve said, smelly hot garbage, and the infrastructure built to support FTP is even worse. Why? Well, one reason is that FTP has the most idiotic networking model conceivable. To see how crazy it is, let’s compare to a more sane protocol, like HTTP (for simplicity’s sake, I’ll do HTTP/1.1). First, you get the underlying transport protocol stuff and probably SSL. The HTTP client opens a connection from some local ephemeral port to the destination server on port 80/443/whatever and does all the normal protocol things (so syn->synack->ack and Client Hello -> Server Hello+server cert -> client kex+change cipher -> change cipher -> encrypted data). FTP does TCP too! Same same so far (minus SSL, unless you’re using FTPS). Next, the HTTP client goes like this:

      GET /index.html HTTP/1.1
      Host: www.whatever.the.fuck
      # a bunch of other headers
      
      

      and you know what fucking happens here? The fucking server responds with the data and a response code on the same goddamn TCP connection. You get a big, glorious response over the nice connection you established:

      200 OK
      # a bunch of headers and shit
      
      HERE'S YOUR DAMN DATA NERD
      
      

      So that’s nice, and the client you’re using to read this used that flow (or an evolution of that flow if you’re using HTTP/2 or HTTP/3). So what does FTP do? It does one of two really stupid things depending on whether you’re using active or passive mode. Active mode is the default for the protocol (although not the default for most clients), so let’s analyze that! First, your FTP client initiates a TCP connection to your server on port 21 (by default), and then the server just sends this:

      <--- 220 Rebex FTP Server ready.
      
      

      ok, that kinda came out of nowhere. You’re probably using a modern client that saves you from all of the godawful footguns, so it then asks the server what it supports:

      ---> FEAT
      <--- 211-Supported extensions:
      <---  AUTH TLS;SSL;
      <---  CDUP
      <---  CLNT
      # A whole bunch of other 4 letter acronyms. If I was writing an FTP server, I'd make it swear at the user since there are a lot of fun 4 letter words
      
      

      There’s some other bullshit we don’t care about right now, although highlights include sending the username and password in plain text. There’s also ASCII vs binary mode. WE’LL GET BACK TO THAT. :|

      So then we want to do a LIST. You know what happens in active mode? Your computer opens up some random fucking TCP port. It then instructs the FTP server to CONNECT TO YOUR GODDAMN COMPUTER. Your computer is the server, and the other side is now the client. I would post a more detailed overview of the FTP commands, but most servers on the internet disable active mode because it’s a goddamn liability. All of the sudden, your computer has to be internet facing with open firewall ports, and that’s just a whole heap of shit.

      I’m probably not blowing many minds right now because people know about this shit. I just want to mention that this is how FTP was built. The data plane and control plane are separate, and back in 19XX when this shit was invented, you could trust your fellows on ARPANET and NAT didn’t exist and sure HAM radio operators here’s the entire goddamn 44.0.0.0/8 block for you to do packet switched radio. A simple protocol for simple times, back before we knew what was good and what was bad.

      So, active mode sucks! PASV is the future, and is the default on basically all modern clients and servers! Passive mode works exactly the same as the above, except when the client goes to LIST, the server opens some random TCP port (I’ve often seen something like 44000-44010) and tells the client, “hey you, connect to 1.2.3.4:44000 to get you your tasty data.” Sounds great, right? Well, there’s a problem that I actually touched on in my last paragraph. Back when this dogshit was first squeezed out in the 70s, everyone had a public address. There were SO MANY addresses! 4 billion addresses? We’ll never use all of those! That is clearly not the case anymore. We don’t have enough addresses, and now we have this wonderful thing called NAT.

      Continued in part 2.

      • Badabinski@kbin.social
        link
        fedilink
        arrow-up
        7
        ·
        edit-2
        8 months ago

        PART 2.

        NAT, much like the city of Phoenix, is a monument to man’s arrogance. Fuck NAT and fuck FTP. If your FTP server is listening directly on a public IP address hooked up directly to a proper router, then none of this applies. If you’re anything like me, the last company I worked for (a small startup), or my current company (many many thousands of employees making software you know and may or may not hate, making many billions of dollars a year), then the majority of your servers are living in RFC1918 space. Traffic from the internet is making it to them via NAT (or NAT with extra steps, i.e. L4 load balancers).

        A request comes in for $PUBLIC_IP TCP port 21 and is forwarded to your failure of a boxen at 10.0.54.187. Your FTP server is a big stupid idiot and doesn’t know this. It thinks that it’s king shit and has its own public IP address. Therefore, when it’s deciding what ADDR:PORT it’s going to tell the stupid FTP client to connect to, it just looks at one of the adapters on the box and says “oh, I’ll tell this client on the internet to connect to 10.0.54.187:44007” and then I fucking cry. The FTP client is an idiot, but the IP stack on the client’s home/business router is not and says “oh, that’s an address living in RFC1918 space, I shouldn’t send that out over the internet” and they don’t get the results of their LIST.

        So, how do you fix this? Well, you fix it by not using FTP. Use SFTP USE SFTP USE SFTP FOR GOD’S SAKE. But since this world is a shit fucking place, you have two options. The best option is to configure your FTP server to lie about its IP address. Rather than being honest about what a fool it is, you can tell it to send your public IP address to the client rather than the network adapter IP address. Does your public IP address change? Fuck you, you get to write a daemon that checks for that shit, rewrites your FTP server config, and HUPs the bastard (or SIGTERMs it if your server sucks and can’t do a live config reload).

        Let’s say that you don’t want to do that. Let’s say you work at a small company with a small business internet plan that gives you static IPs but a shitty modem. Let’s say that you don’t know what FTP is or how it works and your boss told you to get it set up ASAP and it’s not working (because the client over in Bendoverville Arkansas is being told to connect to a 10.x.x.x address) and it surely must be your ISP’s fault. So you call up Comcast Business/AT&T/Verizon/Whoeverthefuck and you complain at their technicians for hours and hours, and eventually you get connected to a human that knows what the problem is and tells you how to configure your stupid FTP server to lie like a little sinner. The big telco megacorps don’t like that. They don’t want to waste all those hours, and they don’t want to hire too many people who can figure that shit out because it’s expensive. You wanna know what those fucking asshole companies did?

        Continued in part 3.

        • Badabinski@kbin.social
          link
          fedilink
          arrow-up
          6
          ·
          edit-2
          8 months ago

          PART 3.
          They made their STUPID MODEMS FUCK WITH THE FTP PACKETS. I have personally experienced this with Comcast Business. The stupid piece of shit DOCSIS modem they provide intercepts the FTP packet from your server saying “oh, connect to this address: x.x.x.x:44010” and they rewrite the fucking address to the public IP. There is no way to turn just this horse piss off. Now, for average business customers, this probably saved Comcast a bunch of money in support calls. However, if you’re using the so-called bridge mode on that degenerate piece of shit-wrapped-silicon (where rather than allowing the modem to give you a DHCP address, you just configure your system to have one of the addresses in the /29 space and the modem detects that and says oh okay don’t NAT traffic when it’s going to this address, just rewrite the MAC and shunt it over the right interface), then something funny happens. The modem still rewrites the contents of the packet, but it uses the wrong fucking IP address! Because the public IP that your server is running on is no longer available to the modem, the modem just chooses another fucking address. Then, the client tries to connect to 1.2.3.5 instead of 1.2.3.4 where your server is listening, the modem says “hey I’m 1.2.3.5 and you can fuck off, I’m dropping your SYN for port 44010”, and I get an angry call from the client asking why they can’t download their files using this worthless protocol. I remember having a conversation like this:

          Me: “Just use SFTP on port 22!”
          Client: “No! FTP is faster/more secure/good enough for my grandfather good enough for me/corporate won’t allow port 22.”
          Me: “Comcast is fucking me right now. What if we lied and served SFTP over port 21?”
          # we try it
          Client: “It’s not working! I can’t even connect!”

          I couldn’t connect either. I couldn’t connect to anything. Trying to do SFTP over port 21 caused the stupid fucking modem to CRASH.

          Are you starting to see what the problem is? It’s like Microsoft preserving bugs in Windows APIs so that shitty software doesn’t break, and then they end up doing crazy gymnastics to accomodate old shit like the Windows 8 -> Windows 10 thing where they couldn’t use “Windows 9” because that would confuse software into thinking it was running “Windows 95” or “Windows 98”. FTP has some bugfuck crazy design decisions that we’ve collectively decided to just “work around,” and it leads to fucking gymnastics.

          Speaking of bugfuck crazy design decisions, FTP’s default file transfer mode intentionally mangles data!

          Continued in part 4.

          • Badabinski@kbin.social
            link
            fedilink
            arrow-up
            7
            ·
            edit-2
            8 months ago

            PART 4.

            You expect a file transfer program to reliably and faithfully transfer your files, byte-for-byte, from one system to another. FTP spits in your face and shits on your chest. You know how Linux uses LF (i.e. \n) for newlines and Windows uses CRLF (i.e. \r\n) for newlines? Pretty annoying, right? Well, FTP’s ASCII mode will automatically rip off those \r characters for you! Sounds pretty sweet, right? Fuck no it’s not. All of the sudden, your file checksums have changed. If you pass the same file back to a Windows user with a different and more sane file transfer system, then they get a broken file because FTP didn’t mind its own fucking business. If you have a CRLF file and need an LF file, just explicitly use dos2unix. Wanna go the other way? unix2dos. The tool has been around since 1989 and it’s great.

            Now, what if you’re not transferring text, but instead are transferring a picture of a cute cat? What if your binary data happens to have 0x0D0x0A somewhere in it? Well, ASCII mode will happily translate that to 0x0A and fucking ruin your adorable cat picture that you were going to share with your depressed significant other in an attempt to cheer them up. Now the ruined JPEG will remind them of the futility of their situation and they’ll slide even deeper into cold emptiness. Thanks, FTP.

            You can tell your client to use binary mode and this problem goes away! In fact, modern clients do this automatically so your SO gets to see the adorable fuzzy cat picture. But let’s just stop and think about this. Why use a protocol that is dangerous by default? Why use a protocol that supports no form of security (unless you’re using fucking godawful FTPS or FTP over SSH)? Why use a protocol that is so broken by design that small business hardware has been designed to try to unfuck it? Is it faster? I mean, not really. SFTP has encryption/decryption overhead, but your CPU is so fast that you’d need to transfer at 25+ Gb/s to notice it. Is it easier? Fuck no it’s not easier, look at all of the stupid footguns I’ve just mentioned. Is it simpler? The line protocol is simple, but so is HTTP, and HTTP has a much simpler control flow path (merging the data and control planes is objectively the right thing to do in this context). And shit, you want a simple protocol for cases where you don’t have a lot of CPU power? Use fucking TFTP. It’s dogshit, but it was intentionally designed to be dogshit so that a fucking potato could receive data with it.

            There is no task that is currently being done with FTP that couldn’t be done more easily, more securely, and more quickly with some other protocol (like fucking SSH and SFTP, which is now built into fucking Windows for god’s sake). Fuck FTP.

            • aksdb@lemmy.world
              link
              fedilink
              arrow-up
              2
              ·
              8 months ago

              Have you considered publishing that as a book? (/s)

              You are insane… in a good way. I love it. Fantastic read and I had to chuckle a few times.

            • Mango@lemmy.world
              link
              fedilink
              arrow-up
              1
              ·
              8 months ago

              I read the first two and kinda gave up my dude. Here’s my deal. I get that it’s bad under the hood. What else can I use that lets me and a friend pretend we just have folders in each other’s computers with just a port forward, IP, and a password?

              • Natanael@slrpnk.net
                link
                fedilink
                arrow-up
                1
                ·
                8 months ago

                That’s not even the type of setup you should use. Use a VPN of the type designed for games and IoT stuff, like ZeroTier, n2n, and more. Then you set up a local file share using something like Samba, only accessible by the people who can connect to your local network via the VPN.

                The public facing VPN code will be MUCH more hardened against attack than your typical sharing tool with port forwarding.

                  • Natanael@slrpnk.net
                    link
                    fedilink
                    arrow-up
                    1
                    ·
                    8 months ago

                    If you set up port forwarding for file shares you must keep setting it up again for every new service.

                    If you set up a VPN once then you’re simply done. Every new service you set up is available directly.

                  • BaldProphet@kbin.social
                    link
                    fedilink
                    arrow-up
                    1
                    arrow-down
                    1
                    ·
                    8 months ago

                    ZeroTier is pretty easy to set up, but at the point where you’re worrying about “barriers to sharing” you should probably using a cloud service anyway.

    • lemmyreader@lemmy.ml
      link
      fedilink
      English
      arrow-up
      4
      ·
      8 months ago

      In that case, I’d like to chime in and add NFS to this list. The often overlooked jewel of the glorious past days. /j

      • Mango@lemmy.world
        link
        fedilink
        arrow-up
        2
        ·
        8 months ago

        So like… If I had a game installed on your computer, my computer could treat that game as if it’s local and load files over the Internet like it’s just reading my disk?

        That is cool as fuck.