HTTP/2 with curl

HTTP/2 Spec http2 explained

Build prerequisites

  • nghttp2
  • OpenSSL, libressl, BoringSSL, NSS, GnutTLS, mbedTLS, wolfSSL or Schannel with a new enough version.


libcurl uses this 3rd party library for the low level protocol handling parts. The reason for this is that HTTP/2 is much more complex at that layer than HTTP/1.1 (which we implement on our own) and that nghttp2 is an already existing and well functional library.

We require at least version 1.12.0.

Over an http:// URL

If CURLOPT_HTTP_VERSION is set to CURL_HTTP_VERSION_2_0, libcurl will include an upgrade header in the initial request to the host to allow upgrading to HTTP/2.

Possibly we can later introduce an option that will cause libcurl to fail if not possible to upgrade. Possibly we introduce an option that makes libcurl use HTTP/2 at once over http://

Over an https:// URL

If CURLOPT_HTTP_VERSION is set to CURL_HTTP_VERSION_2_0, libcurl will use ALPN (or NPN) to negotiate which protocol to continue with. Possibly introduce an option that will cause libcurl to fail if not possible to use HTTP/2.

CURL_HTTP_VERSION_2TLS was added in 7.47.0 as a way to ask libcurl to prefer HTTP/2 for HTTPS but stick to 1.1 by default for plain old HTTP connections.

ALPN is the TLS extension that HTTP/2 is expected to use. The NPN extension is for a similar purpose, was made prior to ALPN and is used for SPDY so early HTTP/2 servers are implemented using NPN before ALPN support is widespread.

CURLOPT_SSL_ENABLE_ALPN and CURLOPT_SSL_ENABLE_NPN are offered to allow applications to explicitly disable ALPN or NPN.

SSL libs

The challenge is the ALPN and NPN support and all our different SSL backends. You may need a fairly updated SSL library version for it to provide the necessary TLS features. Right now we support:

  • OpenSSL: ALPN and NPN
  • libressl: ALPN and NPN
  • BoringSSL: ALPN and NPN
  • NSS: ALPN and NPN
  • GnuTLS: ALPN
  • mbedTLS: ALPN
  • Schannel: ALPN
  • wolfSSL: ALPN
  • Secure Transport: ALPN


Starting in 7.43.0, libcurl fully supports HTTP/2 multiplexing, which is the term for doing multiple independent transfers over the same physical TCP connection.

To take advantage of multiplexing, you need to use the multi interface and set CURLMOPT_PIPELINING to CURLPIPE_MULTIPLEX. With that bit set, libcurl will attempt to re-use existing HTTP/2 connections and just add a new stream over that when doing subsequent parallel requests.

While libcurl sets up a connection to a HTTP server there is a period during which it doesn‘t know if it can pipeline or do multiplexing and if you add new transfers in that period, libcurl will default to start new connections for those transfers. With the new option CURLOPT_PIPEWAIT (added in 7.43.0), you can ask that a transfer should rather wait and see in case there’s a connection for the same host in progress that might end up being possible to multiplex on. It favours keeping the number of connections low to the cost of slightly longer time to first byte transferred.


We hide HTTP/2's binary nature and convert received HTTP/2 traffic to headers in HTTP 1.1 style. This allows applications to work unmodified.

curl tool

curl offers the --http2 command line option to enable use of HTTP/2.

curl offers the --http2-prior-knowledge command line option to enable use of HTTP/2 without HTTP/1.1 Upgrade.

Since 7.47.0, the curl tool enables HTTP/2 by default for HTTPS connections.

curl tool limitations

The command line tool won‘t do any HTTP/2 multiplexing even though libcurl supports it, simply because the curl tool is not written to take advantage of the libcurl API that’s necessary for this (the multi interface). We have an outstanding TODO item for this and you can help us make it happen.

The command line tool also doesn‘t support HTTP/2 server push for the same reason it doesn’t do multiplexing: it needs to use the multi interface for that so that multiplexing is supported.

HTTP Alternative Services

Alt-Svc is an extension with a corresponding frame (ALTSVC) in HTTP/2 that tells the client about an alternative “route” to the same content for the same origin server that you get the response from. A browser or long-living client can use that hint to create a new connection asynchronously. For libcurl, we may introduce a way to bring such clues to the application and/or let a subsequent request use the alternate route automatically.

Detailed in RFC 7838