| commit | 6f7051920a63678746eceff01c868368cbe3e705 | [log] [tgz] |
|---|---|---|
| author | Andy Green <andy@warmcat.com> | Sun Oct 20 07:12:51 2019 +0100 |
| committer | Andy Green <andy@warmcat.com> | Sun Oct 20 07:13:09 2019 +0100 |
| tree | 871c1cef68c7100334a1939fd950c1986ba198c4 | |
| parent | da2d585583bec1f4314efc9989460ccdf8c5b672 [diff] |
cgi: call init_destroy even though not on available roles list cgi is a role, but it's not an externally-selectable role... it's only used when the mount indicates it should be. That means it's not in the list of EVERY_AVAILABLE_ROLE... make sure its zombie child reaping timer is started at context creation and stopped at destruction.
Libwebsockets is a simple-to-use, pure C library providing client and server for http/1, http/2, websockets and other protocols in a security-minded, lightweight, configurable, scalable and flexible way. It's easy to build and cross-build via cmake and is suitable for tasks from embedded RTOS through mass cloud serving.
50 minimal examples for various scenarios, CC0-licensed (public domain) for cut-and-paste, allow you to get started quickly.
lws_system: DHCP clientDHCP client is now another network service that can be integrated into lws, with LWS_WITH_SYS_DHCP_CLIENT at CMake. When enabled, the lws_system state is held at DHCP until at least one registered network interface acquires a usable set of DHCP information including ip, subnet mask, router / gateway address and at least one DNS server.
See the api-test-dhcp Minimal Example for how to use.
lws_retryUDP support in lws has new helper that allow lws_retry to be applied for retry, and the ability to synthesize rx and tx udp packetloss systemwide to confirm retry strategies. Since multiple transactions may be in flight on one UDP socket, the support relies on an lws_sul in the transaction object to manage the transaction retries individually.
See READMEs/README.udp.md for details.
lws_system: system state and notification handlersLws now has the concept of systemwide state held in the context... this is to manage that there may be multiple steps that need the network before it's possible for the user code to operate normally. The steps defined are
CONTEXT_CREATED, INITIALIZED, IFACE_COLDPLUG, DHCP, TIME_VALID, POLICY_VALID, REGISTERED, AUTH1, AUTH2, OPERATIONAL and POLICY_INVALID. OPERATIONAL is the state where user code can run normally.
User and other parts of lws can hook notifier callbacks to receive and be able to veto system state changes, either definitively or because they have been triggered to perform a step asynchronously and will move the state on themselves when it completes.
By default just after context creation, lws attempts to move straight to OPERATIONAL. If no notifier interecepts it, it will succeed to do that and operate in a backwards-compatible way. Enabling various features like lws ntpclient also enable notifiers that hold progress at the related state until their operation completes successfully, eg, not able to enter TIME_VALID until ntpclient has the time.
See READMEs/README.lws_system.md for details.
lws_system: HAL ops structLws allows you to define a standardized ops struct at context creation time so your user code can get various information like device serial number without embedding system-specific code throughout the user code. It can also perform some generic functions like requesting a device reboot.
See READMEs/README.lws_system.md for details.
lws_system: ntpclientOptional lws system service enabled by cmake -DLWS_WITH_SYS_NTPCLIENT intercepts the lws_system TIME_VALID state and performs ntpclient to get the date and time before entering TIME_VALID. This allows user code to validate tls certificates correctly knowing the current date and time by the time it reached OPERATIONAL.
Lws now allows you to apply a policy for how long a network connection may go without seeing something on it that confirms it‘s still valid in the sense of passing traffic cohernetly both ways. There’s a global policy in the context which defaults to 5m before it produces a PING if possible, and 5m10 before the connection will be hung up, user code can override this in the context, vhost (for server) and client connection info (for client).
An api lws_validity_confirmed(wsi) is provided so user code can indicate that it observed traffic that must mean the connection is passing traffic in both directions to and from the peer. In the absence of these confirmations lws will generate PINGs and take PONGs as the indication of validity.
lws_system: Async DNS supportMaster now provides optional Asynchronous (ie, nonblocking) recursive DNS resolving. Enable with -DLWS_WITH_SYS_ASYNC_DNS=1 at cmake. This provides a quite sophisticated ipv4 + ipv6 capable resolver that autodetects the dns server on several platforms and operates a UDP socket to its port 53 to produce and parse DNS packets from the event loop. And of course, it's extremely compact.
It broadly follows the getaddrinfo style api, but instead of creating the results on the heap for each caller, it caches a single result according to the TTL and then provides refcounted const pointers to the cached result to callers. While there are references on the cached result it can't be reaped.
See READMEs/README.async-dns.md for detailed information on how it works, along with api-tests/api-test-async-dns minimal example.
You can now opt to measure and store us-resolution statistics on effective latencies for client operations, and easily spool them to a file in a format suitable for gnuplot, or handle in your own callback. Enable -DLWS_WITH_DETAILED_LATENCY=1 in cmake to build it into lws.
If you are concerned about operation latency or potential blocking from user code, or behaviour under load, or latency variability on specific platforms, you can get real numbers on your platform using this.
Timings for all aspects of events on connections are recorded, including the time needed for name resolution, setting up the connection, tls negotiation on both client and server sides, and each read and write.
See READMEs/README.detailed-latency.md for how to use it.
Lws master now makes much better use of the DNS results for ipv4 and ipv6... it will iterate through them automatically making the best use it can of what's provided and attempting new connections for each potentially usable one in turn before giving up on the whole client connection attempt.
If ipv6 is disabled at cmake it can only use A / ipv4 records, but if ipv6 is enabled, it tries both; if only ipv6 is enabled it promotes ipv4 to ::ffff:1.2.3.4 IPv4-in-IPv6 addresses.
An internal union lws_sockaddr46 that combines struct sockaddr_in and struct sockaddr_in6 is now public, and there are helpers that can parse (using lws_tokenize) any valid numeric representation for ipv4 and ipv6 either into byte arrays and lengths, or directly to and from lws_sockaddr46.
Lws now supports the convention that half-closing an h2 http stream may make the stream ‘immortal’, in terms of not being bound by normal timeouts. For the client side, there's an api that can be applied to the client stream to make it transition to this “read-only” long poll mode.
See READMEs/README.h2-long-poll.md for full details, including how to test it with the minimal examples.
H1 is not so simple to parse because the header length is not known until it has been fully parsed. The next header, or http body may be directly coalesced with the header as well. Lws has supported bulk h1 parsing from a buffer for a long time, but on clientside due to interactions with http proxying it had been stuck parsing the header bytewise out of the tls buffer. In master, everything now bulk parses from a buffer and uses a buflist to pass leftovers through the event loop cleanly.
lws_sul time refactorJust before v3.2 there was a big refactor about how lws handles time. It now explicitly schedules anything that may happen in the future on a single, sorted linked-list, at us resolution. When entering a poll wait (or returning to an event lib loop) it checks the interval between now and the earliest event on the list to figure out how long to wait if there are no network events. For the event loop case, it sets a native event lib timer to enforce it.
See READMEs/README.lws_sul.md for more details and a handy api where you can schedule your own arbitrary callbacks using this system.
Libwebsockets master is now under the MIT license. See ./LICENSE.
This is the libwebsockets C library for lightweight websocket clients and servers. For support, visit
and consider joining the project mailing list at
https://libwebsockets.org/mailman/listinfo/libwebsockets
You can get the latest version of the library from git:
Doxygen API docs for master: https://libwebsockets.org/lws-api-doc-master/html/index.html