HTTP multipart/form-data. What happends when binary data has no string representation?

Question

I want to write an HTTP implementation.

I've been looking around for a few days about sending files over HTTP with Content-Type: multipart/form-data, and I'm really interested about how browsers (or any HTTP client) creates that kind of request.

I already took a look at a lots of questions about it here at stackoverflow like:
How does HTTP file upload work?
What does enctype='multipart/form-data' mean?

I dig into RFCs 2616 (and newer versions), 2046, etc. But I didn't find a clear answer (obviously I did not get the idea behind it).

At most articles and answers I found this piece of request string, that's is simple to me to interpret, all these things are documented at RFCs...

POST /upload?upload_progress_id=12344 HTTP/1.1
Host: localhost:3000
Content-Length: 1325
Origin: http://localhost:3000
... other headers ...
Content-Type: multipart/form-data; boundary=----WebKitFormBoundaryePkpFF7tjBAqx29L

------WebKitFormBoundaryePkpFF7tjBAqx29L
Content-Disposition: form-data; name="MAX_FILE_SIZE"

100000
------WebKitFormBoundaryePkpFF7tjBAqx29L
Content-Disposition: form-data; name="uploadedfile"; filename="hello.o"
Content-Type: application/x-object

... contents of file goes here ...
------WebKitFormBoundaryePkpFF7tjBAqx29L--

...and it would be simple to implement an HTTP client to construct a piece of string that way in any language.

The problem becomes at ... contents of file goes here ..., there's little information about what "contents of file" is. I know it's binary data with a certain type and encoding, but It's difficult to think out of string data, how I would add a piece of binary data that has no string representation inside a string.

I would like to see examples of low level implementations of HTTP protocol with any language. And maybe in depth explanations about binary data transfer over HTTP, how client creates requests and how server read/parse it.

PD. I know this question my look a duplicate but most of the answers are not focused on explaining binary data transfer (like media).

Answer 1

You should not try to handle strings on this part of the body, you should send binary data, see it as reading bytes from the resource and sending theses bytes unaltered.

So especially no encoding applied, no utf-8, no base64, HTTP is not a protocol with an ascii7 restriction like smtp, where base64 encoding is applied to ensure only ascii7 characters are used.

There is, by definition, no string version of this data, and looking at raw HTTP transfer (with wireshark for example) you should see binary data, bytes, stuff.

This is why most HTTP servers uses C to manage HTTP, they parse the HTTP communication byte per byte (as the protocol headers are ascii 7 only, certainly not multibytes characters) and they can also read/write arbitrary binary data for the body quite easily (or even using system calls like readfile to let the kernel manage the binary part).

Now, about examples.

When you use Content-Length and no multipart stuff the body is exactly (content-length) bytes long, so the client parsing your sent data will just read this number of bytes and will treat this whole raw data as the body content (which may have a mime type and and encoding information, but that's just informations for layers set on top of the HTTP protocol).

When you use Transfer-Encoding: chunked, the raw binary body is separated into pieces, each part is then prefixed by an hexadecimal number (the size of the chunk) and the end of line marker. With a final null marker at the end.

If we take the wikipedia example:

4\r\n
Wiki\r\n
5\r\n
pedia\r\n
E\r\n
 in\r\n
\r\n
chunks.\r\n
0\r\n
\r\n

We could replace each ascii7 letter by any byte, even a byte that would have no ascii7 representation, Ill use a * character for each real body byte:

4\r\n
****\r\n
5\r\n
*****\r\n
E\r\n
**************\r\n
0\r\n
\r\n

All the other characters are part of the HTTP protocol (here a chunked body transmission). I could also use a \n representation of binary data, and send only the null byte for each byte of the body, that would be:

4\r\n
\0\0\0\0\0\r\n
5\r\n
\0\0\0\0\0\0\r\n
E\r\n
\0\0\0\0\0\0\0\0\0\0\0\0\0\0\r\n
0\r\n
\r\n

That's just a representation, we could also use \xNN or \NN representations, in reality these are bytes, 8 bits (too lazy to write the 0/1 representation of this body :-) ).

If the text of the example, instead of being:

Wikipedia in\r\n
\r\n
chunks.

It could have been a more complex one, with multibytes characters (here a é in utf-8):

Wikipédia in\r\n
\r\n
chunks.

This é is in fact 11000011:10101001 in utf-8, two bytes: \xc3\xa9 in \xNN representation), instead of the simple 01100101 / \x65 / echaracter. The HTTP body is now (see that second chunk size is 6 and not 5):

4\r\n
Wiki\r\n
6\r\n
p\xc3\xa9dia\r\n
E\r\n
 in\r\n
\r\n
chunks.\r\n
0\r\n
\r\n

But this is only valid if the source data was effectively in utf-8, could have been another encoding. By default, unless you have some specific configuration settings available in your web server where you enforce a conversion of the source document in a specific encoding, that's not really the job of the web server to convert the source document, you take what you have, and you maybe add an header to tell the client what encoding was defined on the source document.

Finally we have the multipart way of transmitting the body, like in your question, it's a lot like the chunked version, except here boundaries and intermediary headers are used, but for the binary data between these boundaries, headers, and line endings control characters it is the same rule, everything inside are just bytes...