RFC 2616: Hypertext Transfer Protocol -- HTTP/1.1

Versions

The protocol has been revised over time. A version is identified as HTTP/# where # is the version number. This article covers aspects of all versions but provides primary coverage for HTTP/0.9, HTTP/1.0, and HTTP/1.1. Separate articles cover HTTP/2 and HTTP/3 in detail.

In HTTP/1.0, a separate TCP connection to the same server is made for every resource request.

In HTTP/1.1, instead a TCP connection can be reused to make multiple resource requests (i.e. of HTML pages, frames, images, scripts, stylesheets, etc.). HTTP/1.1 communications therefore experience less latency as the establishment of TCP connections presents considerable overhead, especially under high traffic conditions.

Enhancements added with HTTP/2 allow for less latency and, in most cases, higher speeds than HTTP/1.1 communications. HTTP/2 adds support for:

a compressed binary representation of metadata (HTTP headers) instead of a textual one, so that headers require much less space;

a single TCP/IP (usually encrypted) connection per accessed server domain instead of 2 to 8 TCP/IP connections;

one or more bidirectional streams per TCP/IP connection in which HTTP requests and responses are broken down and transmitted in small packets to almost solve the problem of the HOLB (head-of-line blocking);

a push capability to allow server application to send data to clients whenever new data is available (without forcing clients to request periodically new data to server by using polling methods).

HTTP/3 uses QUIC + UDP transport protocols instead of TCP. Only the IP layer is used (which UDP, like TCP, builds on). This slightly improves the average speed of communications and avoids the occasional problem of TCP connection congestion that can temporarily block or slow down the data flow of all its streams (another form of "head of line blocking").

Use

HTTP/2 is supported by 71% of websites (34.1% HTTP/2 + 36.9% HTTP/3 with backwards compatibility) and supported by almost all web browsers (over 98% of users). It is also supported by major web servers over Transport Layer Security (TLS) using an Application-Layer Protocol Negotiation (ALPN) extension where TLS 1.2 or newer is required.

HTTP/3 is used on 40% of websites and is supported by most web browsers, i.e. (at least partially) supported by 97% of users. HTTP/3 uses QUIC instead of TCP for the underlying transport protocol. Like HTTP/2, it does not obsolete previous major versions of the protocol. In 2019, support for HTTP/3 was first added to Cloudflare and Chrome and also enabled in Firefox. HTTP/3 has lower latency for real-world web pages and loads faster than HTTP/2, in some cases over three times faster than HTTP/1.1, which is still commonly the only protocol enabled.

HTTPS, the secure variant of HTTP, is used by 90.1% of websites.

Technology

Transport layer

HTTP presumes an underlying and reliable transport layer protocol. The standard choice of the underlying protocol prior to HTTP/3 is Transmission Control Protocol (TCP). HTTP/3 uses a different transport layer called QUIC, which provides reliability on top of the unreliable User Datagram Protocol (UDP). HTTP/1.1 and earlier have been adapted to be used over plain unreliable UDP in multicast and unicast situations, forming HTTPMU and HTTPU. They are used in UPnP and Simple Service Discovery Protocol (SSDP), two protocols usually run on a local area network.

Data exchange

HTTP is a stateless application-level protocol and it requires a reliable network transport connection to exchange data between client and server. In HTTP implementations, TCP/IP connections are used using well-known ports (typically port 80 if the connection is unencrypted or port 443 if the connection is encrypted, see also List of TCP and UDP port numbers). In HTTP/2, a TCP/IP connection plus multiple protocol channels are used. In HTTP/3, the application transport protocol QUIC over UDP is used.

Data is exchanged through a sequence of request–response messages which are exchanged by a session layer transport connection. An HTTP client initially tries to establish a connection, real or virtual, with a server. An HTTP server listening on the port accepts the connection and then waits for a client's request message. The client sends its HTTP request message. Upon receiving the request the server sends back an HTTP response message, which includes header(s) plus a body if it is required. The body of this response message is typically the requested resource, although an error message or other information may also be returned. At any time and for many reasons, either the client or server can close the connection. Closing a connection is usually advertised by one or more HTTP headers in the last request or response.

In HTTP/0.9, the TCP/IP connection is always closed after server response has been sent, so it is never persistent.

In HTTP/1.0, the TCP/IP connection should always be closed by server after a response has been sent.

In HTTP/1.1, a keep-alive-mechanism was officially introduced so that a connection could be reused for more than one request/response. Such persistent connections reduce request latency perceptibly because the client does not need to re-negotiate the TCP 3-Way-Handshake connection after the first request has been sent. Another positive side effect is that, in general, the connection becomes faster with time due to TCP's slow-start-mechanism.

HTTP/1.1 added also HTTP pipelining in order to further reduce lag time when using persistent connections by allowing clients to send multiple requests before waiting for each response. This optimization was never considered really safe because a few web servers and many proxy servers, specially transparent proxy servers placed in Internet / Intranets between clients and servers, did not handle pipelined requests properly (they served only the first request discarding the others, they closed the connection because they saw more data after the first request or some proxies even returned responses out of order etc.). Because of this, only HEAD and some GET requests (i.e. limited to real file requests and so with URLs without query string used as a command, etc.) could be pipelined in a safe and idempotent mode. After many years of struggling with the problems introduced by enabling pipelining, this feature was first disabled and then removed from most browsers. Pipelining was also removed because of the announced adoption of HTTP/2.

Application session

As a stateless protocol, HTTP does not require the web server to retain information or status about each user for the duration of multiple requests. If a web application needs an application session, it implements it via HTTP cookies, hidden variables in a web form or another mechanism.

Typically, to start a session, an interactive login is performed, and to end a session, a logout is requested by the user. These kind of operations use a custom authentication mechanism, not HTTP authentication.

Authentication

HTTP provides multiple authentication schemes such as basic access authentication and digest access authentication which operate via a challenge–response mechanism whereby the server identifies and issues a challenge before serving the requested content.

HTTP provides a general framework for access control and authentication, via an extensible set of challenge–response authentication schemes, which can be used by a server to challenge a client request and by a client to provide authentication information.

The authentication mechanisms described above belong to the HTTP protocol and are managed by client and server HTTP software (if configured to require authentication before allowing client access to one or more web resources), and not by the web applications using an application session.

The HTTP authentication specification includes realms that provide an arbitrary, implementation-specific construct for further dividing resources common to a given root URI. The realm value string, if present, is combined with the canonical root URI to form the protection space component of the challenge. This in effect allows the server to define separate authentication scopes under one root URI.

Encrypted connection

The most popular way of establishing an encrypted HTTP connection is HTTPS. Two other methods for establishing an encrypted HTTP connection also exist: Secure Hypertext Transfer Protocol, and using the HTTP/1.1 Upgrade header to specify an upgrade to TLS. Browser support for these two is, however, nearly non-existent.

Message format

This section describes messages for HTTP/1.1. Later versions, HTTP/2 and HTTP/3, use a binary protocol, where headers are encoded in a single HEADERS and zero or more CONTINUATION frames using HPACK (HTTP/2) or QPACK (HTTP/3), which both provide efficient header compression. The request or response line from HTTP/1 has also been replaced by several pseudo-header fields, each beginning with a colon (:).

At the highest level, a message consists of a header followed by a body.

Header

A header consists of lines of ASCII text; each terminated with a carriage return and line feed sequence. The layout for both a request and a response header is:

Start line

Structured data that differs for request vs. response.

Header fields

Zero or more header field lines (at least 1 for HTTP/1.1); see below.

Empty line

Marks the end of the header.

Body

A body consists of data in any format; not limited to ASCII. The format must match that specified by the Content-Type header field if the message contains one. A body is optional or, in other words, can be blank.

Entity

Before HTTP/2, the term entity was used to mean the body plus header fields that describe the body. In particular, not all headers were considered part of the entity. The term entity header referred to a header that was considered part of the entity, and sometimes the body was called the entity body. Modern documentation uses body and header without using entity.

Header field

A header field represents metadata about the containing message. Examples include how the body is encoded (via Content-Encoding), session verification and identification of the client (as in browser cookies, IP address, user-agent) or their anonymity thereof (VPN or proxy masking, user-agent spoofing), how the server should handle data (as in Do-Not-Track or Global Privacy Control), and the age (the time it has resided in a shared cache) of the document being downloaded. Generally, the information of a header field is used by software and not shown to the user.

A header field line is formatted as a name-value pair with a colon separator. Whitespace is not allowed around the name, but leading and trailing whitespace is ignored for the value part. Unlike a method name that must match exactly (case-sensitive), a header field name is matched ignoring case although often shown with each word capitalized. For example, the following are header fields for Host and Accept-Language.

Host: www.example.com

Accept-Language: en

The standards do not limit the size of a header field or the number of fields in a message. However, most servers, clients, and proxy software impose limits for practical and security reasons. For example, the Apache 2.3 server by default limits the size of each field to 8190 bytes, and there can be at most 100 header fields in a single request.

Although deprecated by RFC 7230, in the past, long lines could be split into multiple lines with a continuation line starting with a space or tab character.

Request

A request is sent by a client to a server. The start line includes a method name, a request URI and the protocol version with a single space between each field. The following request start line specifies method GET, URI /customer/123 and protocol version HTTP/1.1:

GET /customer/123 HTTP/1.1

Request header fields allow the client to pass additional information beyond the request line, acting as request modifiers (similarly to the parameters of a procedure). They give information about the client, about the target resource, or about the expected handling of the request. In the HTTP/1.1 protocol, all header fields except Host are optional.

A request line containing only the path name is accepted by servers to maintain compatibility with HTTP clients before the HTTP/1.0 specification in RFC 1945.

The protocol structures transaction as operating on resources. What a resource represents, whether pre-existing data or data that is generated dynamically, depends on the implementation of the server. Often, the resource corresponds to a file or the output of an executable running on the server.

A request identifies a method (sometimes informally called verb) to classify the desired action to be performed on a resource. The HTTP/1.0 specification defined the GET, HEAD, and POST methods as well as listing the PUT, DELETE, LINK and UNLINK methods under additional methods. However, the HTTP/1.1 specification added five new methods: PUT, DELETE, CONNECT, OPTIONS, and TRACE. Any client can use any method and the server can be configured to support any combination of methods. If a method is unknown to an intermediate, it will be treated as an unsafe and non-idempotent method. There is no limit to the number of methods that can be defined, which allows for future methods to be specified without breaking existing infrastructure. For example, WebDAV defined seven new methods and RFC 5789 specified the PATCH method. A general-purpose web server is required to implement at least GET and HEAD, and all other methods are considered optional by the specification.

Method names are case sensitive. This is in contrast to HTTP header field names which are case-insensitive.

Response

A response is sent to the client by the server. The start line of a response consists of the protocol version, a status code and optionally a reason phrase with fields separated by a single space character. The following response start line specifies protocol version HTTP/1.1, status code 400 and reason phrase Bad Request.

HTTP/1.1 400 Bad Request

Response header fields allow the server to pass additional information beyond the status line, acting as response modifiers. They give information about the server or about further access to the target resource or related resources. Each response header field has a defined meaning which can be further refined by the semantics of the request method or response status code.

The status code is a three-digit, decimal, integer value that represents the disposition of the server's attempt to satisfy the client's request. Generally, a client handles a response primarily based on the status code and secondarily on response header fields. A client may not understand each status code that a server reports but it must understand the class as indicated by the first digit and treat an unrecognized code as equivalent to the x00 code of that class. The classes are as follows:

1XX informational

The request was received, continuing process.

2XX successful

The request was successfully received, understood, and accepted.

3XX redirection

Further action needs to be taken in order to complete the request.

4XX client error

The request cannot be fulfilled due to an issue that the client might be able to control.

5XX server error

Example

The following demonstrates an HTTP/1.1 request-response transaction for a server at www.example.com, port 80. HTTP/1.0 would use the same messages except for a few missing headers. HTTP/2 and HTTP/3 would use the same request-response mechanism but with different representations for HTTP headers.

The following is a request with no body. It consists of a start line, 6 header fields and a blank line – each terminated with a carriage return and line feed sequence. The Host header field distinguishes between various DNS names sharing a single IP address, allowing name-based virtual hosting. While optional in HTTP/1.0, it is mandatory in HTTP/1.1.

Although not clear in the representation above (due to limitations of this wiki), the blank line at the end results in ending in two line terminator sequences. Represented as a stream of characters, a shorted version of above shows this more clearly with <CRLF> representing a line terminator sequence: GET / HTTP/1.1<CRLF>Host: www.example.com<CRLF><CRLF>.

In the following response, the ETag (entity tag) header field is used to determine if a cached version of the requested resource is identical to the current version of the resource on the server. The Content-Type header field specifies the Internet media type of the data conveyed by the HTTP message, and Content-Length indicates its length in bytes. The HTTP/1.1 webserver publishes its ability to respond to requests for a byte range of the resource by including Accept-Ranges: bytes. This is useful, if the client needs to have only certain portions of a resource sent by the server, which is called byte serving. When Connection: close is sent, it means that the web server will close the TCP connection immediately after the end of the transfer of this response.

Most of the header fields are optional but some are mandatory. When header Content-Length is missing from a response with a body, then this should be considered an error in HTTP/1.0 but it may not be an error in HTTP/1.1 if header Transfer-Encoding: chunked is present. Chunked transfer encoding uses a chunk size of 0 to mark the end of the content. Some old implementations of HTTP/1.0 omitted the header Content-Length when the length of the body was not known at the beginning of the response and so the transfer of data to client continued until server closed the socket.

Similar protocols

Gopher protocol

A content delivery protocol that was displaced by HTTP in the early 1990s.

SPDY protocol

An alternative to HTTP developed at Google, superseded by HTTP/2.

Gemini protocol

A Gopher-inspired protocol which mandates privacy-related features.

History

Tim Berners-Lee and his team at CERN are credited with inventing HTTP, along with HTML and the associated technology for a web server and a client user interface called web browser. Berners-Lee designed HTTP in order to help with the adoption of his other idea: the "WorldWideWeb" project, which was first proposed in 1989, now known as the World Wide Web. Development of HTTP was initiated in 1989 and summarized in a simple document describing the behavior of a client and a server using the first HTTP version, named 0.9. That version was subsequently developed, eventually becoming the public 1.0. Development of early HTTP Request for Comments (RFC) documents started a few years later in a coordinated effort by the Internet Engineering Task Force (IETF) and the World Wide Web Consortium (W3C), with work later moving to the IETF.

The first web server went live in 1990. The protocol used had only one method, namely GET, which would request a page from a server. The response from the server was always an HTML page.

HTTP/0.9

In 1991, the first documented official version of HTTP was written as a plain document, less than 700 words long, and this version was named HTTP/0.9, which supported only GET method, allowing clients to only retrieve HTML documents from the server, but not supporting any other file formats or information upload.

HTTP/1.0-draft

Since 1992, a new document was written to specify the evolution of the basic protocol towards its next full version. It supported both the simple request method of the 0.9 version and the full GET request that included the client HTTP version. This was the first of the many unofficial HTTP/1.0 drafts that preceded the final work on HTTP/1.0.

W3C HTTP Working Group

After having decided that new features of HTTP protocol were required and that they had to be fully documented as official RFC documents, in early 1995, the HTTP Working Group (HTTP WG, led by Dave Raggett) was constituted with the aim to standardize and expand the protocol with extended operations, extended negotiation, richer meta-information, tied with a security protocol which became more efficient by adding additional methods and header fields.

The HTTP WG planned to revise and publish new versions of the protocol as HTTP/1.0 and HTTP/1.1 within 1995, but, because of the many revisions, that timeline lasted much more than one year.

The HTTP WG planned also to specify a far future version of HTTP called HTTP-NG (HTTP Next Generation) that would have solved all remaining problems, of previous versions, related to performances, low latency responses, etc. but this work started only a few years later and it was never completed.

HTTP/1.0

In May 1996, RFC 1945 was published as a final HTTP/1.0 revision of what had been used in previous 4 years as a pre-standard HTTP/1.0-draft which was already used by many web browsers and web servers.

In early 1996 developers started to even include unofficial extensions of the HTTP/1.0 protocol (i.e. keep-alive connections, etc.) into their products by using drafts of the upcoming HTTP/1.1 specifications.

HTTP/1.1

Since early 1996, major web browsers and web server developers also started to implement new features specified by pre-standard HTTP/1.1 drafts specifications. End-user adoption of the new versions of browsers and servers was rapid. In March 1996, one web hosting company reported that over 40% of browsers in use on the Internet used the new HTTP/1.1 header "Host" to enable virtual hosting, and that by June 1996, 65% of all browsers accessing their servers were pre-standard HTTP/1.1 compliant.

In January 1997, RFC 2068 was officially released as HTTP/1.1 specifications.

In June 1999, RFC 2616 was released to include all improvements and updates based on previous (obsolete) HTTP/1.1 specifications.

W3C HTTP-NG Working Group

Resuming the old 1995 plan of previous HTTP Working Group, in 1997 an HTTP-NG Working Group was formed to develop a new HTTP protocol named HTTP-NG (HTTP New Generation). A few proposals / drafts were produced for the new protocol to use multiplexing of HTTP transactions inside a single TCP/IP connection, but in 1999, the group stopped its activity passing the technical problems to IETF.

IETF HTTP Working Group restarted

In 2007, the IETF HTTP Working Group (HTTP WG bis or HTTPbis) was restarted firstly to revise and clarify previous HTTP/1.1 specifications and secondly to write and refine future HTTP/2 specifications (named httpbis).

SPDY

In 2009, Google announced SPDY – a binary protocol they developed to speed up web traffic between browsers and servers. In many tests, using SPDY was indeed faster than using HTTP/1.1. SPDY was integrated into Google's Chromium and then into other major web browsers. Some of the ideas about multiplexing HTTP streams over a single TCP connection were taken from various sources, including the work of W3C HTTP-NG Working Group.

HTTP/2

In 2012, HTTP Working Group (HTTPbis) announced the need for a new protocol; initially considering aspects of SPDY and eventually deciding to derive the new protocol from SPDY. In May 2015, HTTP/2 was published as RFC 7540. The protocol was quickly adopted by web browsers already supporting SPDY and more slowly by web servers.

2014 updates to HTTP/1.1

In June 2014, the HTTP Working Group released an updated six-part HTTP/1.1 specification obsoleting RFC 2616:

RFC 7230 – "Hypertext Transfer Protocol (HTTP/1.1): Message Syntax and Routing," Obsolete.

RFC 7231 – "Hypertext Transfer Protocol (HTTP/1.1): Semantics and Content," Obsolete.

RFC 7232 – "Hypertext Transfer Protocol (HTTP/1.1): Conditional Requests," Obsolete.

RFC 7233 – "Hypertext Transfer Protocol (HTTP/1.1): Range Requests," Obsolete.

RFC 7234 – "Hypertext Transfer Protocol (HTTP/1.1): Caching," Obsolete.

RFC 7235 – "Hypertext Transfer Protocol (HTTP/1.1): Authentication," Obsolete.

HTTP/0.9 Deprecation

In 2014, HTTP/0.9 was deprecated for servers supporting version HTTP/1.1 (and higher):Since HTTP/0.9 did not support header fields in a request, there is no mechanism for it to support name-based virtual hosts (selection of resource by inspection of the Host header field). Any server that implements name-based virtual hosts ought to disable support for HTTP/0.9. Most requests that appear to be HTTP/0.9 are, in fact, badly constructed HTTP/1.x requests caused by a client failing to properly encode the request-target.

Since 2016 many product managers and developers of user agents (browsers, etc.) and web servers have begun planning to gradually deprecate and dismiss support for HTTP/0.9 protocol, mainly for the following reasons:

it is so simple that an RFC document was never written (there is only the original document);

it has no HTTP headers and lacks many other features that nowadays are required for minimal security reasons;

it has not been widespread since 1999..2000 (because of HTTP/1.0 and HTTP/1.1) and is commonly used only by some very old network hardware, i.e. routers, etc.

As of 2022, HTTP/0.9 support has not been officially, completely deprecated and is still present in many web servers and browsers (for server responses only), even if usually disabled. It is unclear how long it will take to decommission HTTP/0.9.

HTTP/3

In 2020, the first drafts of HTTP/3 were published and major web browsers and web servers started to adopt it. On 6 June 2022, IETF standardized HTTP/3 as RFC 9114.

Updates and refactoring in 2022

In June 2022, RFC documents were published that deprecated many of the previous documents and introducing a few minor changes and a refactoring of HTTP semantics description into a separate document.

RFC 9110 – "HTTP Semantics," Internet Standard 97.

RFC 9111 – "HTTP Caching," Internet Standard 98.

RFC 9112 – "HTTP/1.1," Internet Standard 99.

RFC 9113 – "HTTP/2," Proposed Standard.

RFC 9114 – "HTTP/3," Proposed Standard. (See also the section above.)

RFC 9204 – "QPACK: Field Compression for HTTP/3," Proposed Standard.

RFC 9218 – "Extensible Prioritization Scheme for HTTP," Proposed Standard.