Parse The Request (HTTP)

The stage of the From Scratch HTTP build right after opening a connection.

Before you start. This series builds a small copy of an HTTP server, the program that answers when a browser or a tool like curl asks a website for a page. The last stage opened a socket - the two-way connection a browser and a server talk over - and read whatever raw bytes (the raw chunks of data a network hands over, with no meaning attached yet) arrived off it without understanding a single one of them; this stage is where those bytes finally become a request the program can act on.

The idea in 30 seconds

HTTP’s header section doesn’t say how big it is; the only way to know it’s finished is to search for a blank line, not count to a number.
A labelled field like Host: example.com is called a header, and the blank line under the last one is the only mark that the header section is done.
The program doing the searching is called a parser, and it can be handed the blank line’s four characters split across two separate deliveries.
A parser must remember what it has already searched and start the next search a few characters early, or a split blank line can leave a request hanging forever.
Because the ending is found by searching rather than counting, two parsers reading the same request can disagree about where it stops - and that disagreement is called request smuggling.
A parser that gets two different answers to “how long is this” has to refuse the message and close the connection, not guess.

Read the story · Skip to the numbers