Software /
code /
prosody
File
net/cqueues.lua @ 11665:148075532021
net.server_epoll: Prevent stack overflow of opportunistic writes
net.http.files serving a big enough file on a fast enough connection
with opportunistic_writes enabled could trigger a stack overflow through
repeatedly serving more data that immediately gets sent, draining the
buffer and triggering more data to be sent. This also blocked the server
on a single task until completion or an error.
This change prevents nested opportunistic writes, which should prevent
the stack overflow, at the cost of reduced download speed, but this is
unlikely to be noticeable outside of Gbit networks. Speed at the cost of
blocking other processing is not worth it, especially with the risk of
stack overflow.
author | Kim Alvefur <zash@zash.se> |
---|---|
date | Sun, 11 Jul 2021 09:39:21 +0200 |
parent | 10999:37b884d675f7 |
child | 12974:ba409c67353b |
line wrap: on
line source
-- Prosody IM -- Copyright (C) 2014 Daurnimator -- -- This project is MIT/X11 licensed. Please see the -- COPYING file in the source package for more information. -- -- This module allows you to use cqueues with a net.server mainloop -- local server = require "net.server"; local cqueues = require "cqueues"; local timer = require "util.timer"; assert(cqueues.VERSION >= 20150113, "cqueues newer than 20150113 required") -- Create a single top level cqueue local cq; if server.cq then -- server provides cqueues object cq = server.cq; elseif server.watchfd then cq = cqueues.new(); local timeout = timer.add_task(cq:timeout() or 0, function () -- FIXME It should be enough to reschedule this timeout instead of replacing it, but this does not work. See https://issues.prosody.im/1572 assert(cq:loop(0)); return cq:timeout(); end); server.watchfd(cq:pollfd(), function () assert(cq:loop(0)); local t = cq:timeout(); if t then timer.stop(timeout); timeout = timer.add_task(cq:timeout(), function () assert(cq:loop(0)); return cq:timeout(); end); end end); else error "NYI" end return { cq = cq; }