profile/ivi/libwebsockets.git
11 years agoforce client ssl bio nonblocking 72/2972/1
Andy Green [Mon, 28 Jan 2013 09:48:21 +0000 (17:48 +0800)]
force client ssl bio nonblocking

Signed-off-by: Andy Green <andy.green@linaro.org>
11 years agoforce ssl rw bios nonblocking 71/2971/1
Andy Green [Mon, 28 Jan 2013 09:45:34 +0000 (17:45 +0800)]
force ssl rw bios nonblocking

Signed-off-by: Andy Green <andy.green@linaro.org>
11 years agotimeout coverage for ssl accept 70/2970/1
Andy Green [Mon, 28 Jan 2013 09:20:41 +0000 (17:20 +0800)]
timeout coverage for ssl accept

Signed-off-by: Andy Green <andy.green@linaro.org>
11 years agobreak up ssl accept action 69/2969/1
Andy Green [Mon, 28 Jan 2013 04:19:10 +0000 (12:19 +0800)]
break up ssl accept action

Signed-off-by: Andy Green <andy.green@linaro.org>
11 years agoensure accept is nonblocking 68/2968/1
Andy Green [Mon, 28 Jan 2013 03:23:52 +0000 (11:23 +0800)]
ensure accept is nonblocking

Signed-off-by: Andy Green <andy.green@linaro.org>
11 years agoautocreate foreign broadcast sockets on broadcast 67/2967/1
Andy Green [Fri, 25 Jan 2013 09:34:15 +0000 (17:34 +0800)]
autocreate foreign broadcast sockets on broadcast

Also introduce libwebsockets_broadcast_foreign() separate from libwebsockets_broadcast()

Signed-off-by: Andy Green <andy.green@linaro.org>
11 years agotrac 5 sa_restorer deprecated 66/2966/1
Andy Green [Wed, 23 Jan 2013 00:32:03 +0000 (08:32 +0800)]
trac 5 sa_restorer deprecated

Signed-off-by: Andy Green <andy.green@linaro.org>
11 years agotrac 3 document write and context_user 65/2965/1
Andy Green [Tue, 22 Jan 2013 04:40:35 +0000 (12:40 +0800)]
trac 3 document write and context_user

Signed-off-by: Andy Green <andy.green@linaro.org>
11 years agotrac 4 correct libebsocket_service_fd 64/2964/1
Andy Green [Tue, 22 Jan 2013 04:32:11 +0000 (12:32 +0800)]
trac 4 correct libebsocket_service_fd

Signed-off-by: Andy Green <andy.green@linaro.org>
11 years agoupdate numbers for minimal build footprint 63/2963/1
Andy Green [Mon, 21 Jan 2013 05:20:33 +0000 (13:20 +0800)]
update numbers for minimal build footprint

Signed-off-by: Andy Green <andy.green@linaro.org>
11 years agomore LWS_NO_DAEMONIZE 62/2962/1
Andy Green [Mon, 21 Jan 2013 05:06:38 +0000 (13:06 +0800)]
more LWS_NO_DAEMONIZE

Signed-off-by: Andy Green <andy.green@linaro.org>
11 years agouse correct LWS_NO_DAEMONIZE on test server 61/2961/1
Andy Green [Mon, 21 Jan 2013 04:58:04 +0000 (12:58 +0800)]
use correct LWS_NO_DAEMONIZE on test server

Signed-off-by: Andy Green <andy.green@linaro.org>
11 years agobind gcc debug generation to_DEBUG 60/2960/1
Andy Green [Mon, 21 Jan 2013 04:19:30 +0000 (12:19 +0800)]
bind gcc debug generation to_DEBUG

Either generate debug sections in output or optimize aggressively,
controlled by the existing --disable-debug

Signed-off-by: Andy Green <andy.green@linaro.org>
11 years agounionize mutually exclusive wsi members 59/2959/1
Andy Green [Mon, 21 Jan 2013 03:04:23 +0000 (11:04 +0800)]
unionize mutually exclusive wsi members

Large chunks of struct libwebsocket members actually have a mutually
exclusive lifecycle, eg, once the http headers are finished they sit
there unused until the instance is destroyed.

This makes a big improvement in memory efficiency by making four
categories of member: always needed, needed for header processing,
needed for http processing, and needed for ws processing.  The last
three are mutually exclusive and bound into a union inside the wsi.

Care needs taking now at "union transitions", although we zeroed down
the struct at init, the other union siblings have been writing the
same memory by the time later member siblings start to use it.  So
it must be cleared down appropriately when we cross from one
mutually-exclusive use to another.

Signed-off-by: Andy Green <andy.green@linaro.org>
11 years agokey_b64 doesnt need to be in wsi 58/2958/1
Andy Green [Mon, 21 Jan 2013 03:04:49 +0000 (11:04 +0800)]
key_b64 doesnt need to be in wsi

It's only used in some client code and never seen again

Signed-off-by: Andy Green <andy.green@linaro.org>
11 years agoavoid PATH_MAX in bss in daemonize 57/2957/1
Andy Green [Mon, 21 Jan 2013 02:36:12 +0000 (10:36 +0800)]
avoid PATH_MAX in bss in daemonize

PATH_MAX is typically 4KB, let's malloc space for the
actual path instead and just have the pointer in bss

Signed-off-by: Andy Green <andy.green@linaro.org>
11 years agoremove all support for pre v13 protocols 56/2956/1
Andy Green [Mon, 21 Jan 2013 01:53:35 +0000 (09:53 +0800)]
remove all support for pre v13 protocols

Since v13 was defined as the released ietf version the older versions
are deprecated.  This patch strips out everything to do with the older
versions and gets rid of the option to send stuff unmasked.

The in-tree md5 implementation is then also deleted as nothing needs
it any more, 1280 loc are shed in all

Signed-off-by: Andy Green <andy.green@linaro.org>
11 years agodisable private broadcast sockets if enable no fork config option 55/2955/1
Andy Green [Mon, 21 Jan 2013 01:09:52 +0000 (09:09 +0800)]
disable private broadcast sockets if enable no fork config option

The whole thing about count_protocols + 1 broadcast sockets and
associated dummy wsis is a workaround for getting a broadcast from
a different process context, if we are running with --enable-no-fork
then we don't need any of it in.

Signed-off-by: Andy Green <andy.green@linaro.org>
11 years agoalign test server extpoll with library dynamic approach 54/2954/1
Edwin van den Oetelaar [Sun, 20 Jan 2013 12:51:14 +0000 (20:51 +0800)]
align test server extpoll with library dynamic approach

Signed-off-by: Edwin van den Oetelaar <oetelaar.automatisering@gmail.com>
11 years agodocument memory efficiency 53/2953/1
Andy Green [Sun, 20 Jan 2013 12:14:42 +0000 (20:14 +0800)]
document memory efficiency

This puts some numbers of library size with the various --without
and --disable options and about dynamic memory allocation
performance

Signed-off-by: Andy Green <andy.green@linaro.org>
11 years agolog major dynamic allocation info 52/2952/1
Andy Green [Sun, 20 Jan 2013 12:21:54 +0000 (20:21 +0800)]
log major dynamic allocation info

Signed-off-by: Andy Green <andy.green@linaro.org>
11 years agozlib not needed if no extensions 51/2951/1
Andy Green [Sun, 20 Jan 2013 10:26:20 +0000 (18:26 +0800)]
zlib not needed if no extensions

Signed-off-by: Andy Green <andy.green@linaro.org>
11 years agointroduce without extensions 50/2950/1
Andy Green [Sun, 20 Jan 2013 09:08:31 +0000 (17:08 +0800)]
introduce without extensions

The new --without-extensions config flag completely removes all code
and data related to extensions from the build throughout the library
when given.

Signed-off-by: Andy Green <andy.green@linaro.org>
11 years agorefactor README 49/2949/1
Andy Green [Sun, 20 Jan 2013 03:28:06 +0000 (11:28 +0800)]
refactor README

Signed-off-by: Andy Green <andy.green@linaro.org>
11 years agosyslog requires format string 48/2948/1
Edwin van den Oetelaar [Sat, 19 Jan 2013 12:01:01 +0000 (20:01 +0800)]
syslog requires format string

Signed-off-by: Edwin van den Oetelaar <oetelaar.automatisering@gmail.com>
11 years agorevert zlib update 1.2.7 47/2947/1
Andy Green [Sat, 19 Jan 2013 07:11:23 +0000 (15:11 +0800)]
revert zlib update 1.2.7

Signed-off-by: Andy Green <andy.green@linaro.org>
11 years agomake use of lock file 46/2946/1
Andy Green [Sat, 19 Jan 2013 05:56:10 +0000 (13:56 +0800)]
make use of lock file

Signed-off-by: Andy Green <andy.green@linaro.org>
11 years agowindows compatibility changes for private libwebsockets 45/2945/1
FNA [Sat, 19 Jan 2013 05:58:06 +0000 (13:58 +0800)]
windows compatibility changes for private libwebsockets

Signed-off-by: FNA <aqiruse@gmail.com>
11 years agotest server add daemonization flag 44/2944/1
Andy Green [Sat, 19 Jan 2013 03:11:42 +0000 (11:11 +0800)]
test server add daemonization flag

Signed-off-by: Andy Green <andy.green@linaro.org>
11 years agotest server use syslog logging 43/2943/1
Andy Green [Sat, 19 Jan 2013 03:32:18 +0000 (11:32 +0800)]
test server use syslog logging

Signed-off-by: Andy Green <andy.green@linaro.org>
11 years agoallow_use_of_lwsl_logging in user code 42/2942/1
Andy Green [Sat, 19 Jan 2013 03:58:07 +0000 (11:58 +0800)]
allow_use_of_lwsl_logging in user code

Signed-off-by: Andy Green <andy.green@linaro.org>
11 years agohelper api: log through syslog 41/2941/1
Andy Green [Sat, 19 Jan 2013 03:12:16 +0000 (11:12 +0800)]
helper api:  log through syslog

Signed-off-by: Andy Green <andy.green@linaro.org>
11 years agologging select some lwsl_info usage to be lwsl_notice 40/2940/1
Andy Green [Sat, 19 Jan 2013 05:08:17 +0000 (13:08 +0800)]
logging select some lwsl_info usage to be lwsl_notice

Signed-off-by: Andy Green <andy.green@linaro.org>
11 years agoadd lwsl_notice 39/2939/1
Andy Green [Sat, 19 Jan 2013 04:18:07 +0000 (12:18 +0800)]
add lwsl_notice

Signed-off-by: Andy Green <andy.green@linaro.org>
11 years agoexpose log level in emit 38/2938/1
Andy Green [Sat, 19 Jan 2013 03:17:56 +0000 (11:17 +0800)]
expose log level in emit

Signed-off-by: Andy Green <andy.green@linaro.org>
11 years agochange bitfield setting to avoid gcc warnings 37/2937/1
Andy Green [Sat, 19 Jan 2013 03:52:06 +0000 (11:52 +0800)]
change bitfield setting to avoid gcc warnings

Signed-off-by: Andy Green <andy.green@linaro.org>
11 years agoimprove frame_is_binary setting 36/2936/1
Andy Green [Sat, 19 Jan 2013 02:39:35 +0000 (10:39 +0800)]
improve frame_is_binary setting

Gregory Junker <ggjunker@gmail.com> noticed the binary flag is not
getting set right, or at all on client side.  This should improve
matters.

Signed-off-by: Andy Green <andy.green@linaro.org>
11 years agoadd lws_confirm_legit_wsi 35/2935/1
Andy Green [Fri, 18 Jan 2013 14:00:22 +0000 (22:00 +0800)]
add lws_confirm_legit_wsi

Signed-off-by: Andy Green <andy.green@linaro.org>
11 years agorefactor and introduce without server configure option 34/2934/1
Andy Green [Fri, 18 Jan 2013 03:43:21 +0000 (11:43 +0800)]
refactor and introduce without server configure option

Move server-only stuff into their own files and make building
that depend on not having --without-server on the configure

Make fragments in other places conditional as well

Remove client-related members from struct libwebscket when
building LWS_NO_CLIENT

Apps:

normal: build test server, client, fraggle, ping
--without-client: build test server
--without-server: build test client, ping

Signed-off-by: Andy Green <andy.green@linaro.org>
11 years agointrodice tracking if frame is binary 33/2933/1
Andy Green [Fri, 18 Jan 2013 01:49:20 +0000 (09:49 +0800)]
introdice tracking if frame is binary

Signed-off-by: Andy Green <andy.green@linaro.org>
11 years agodifferent compiler warning fixes 32/2932/1
Edwin van den Oetelaar [Fri, 18 Jan 2013 01:20:54 +0000 (09:20 +0800)]
different compiler warning fixes

Solve some unchecked return codes in the new daemonization file

AG adapted a bit

Signed-off-by: Edwin van den Oetelaar <oetelaar.automatisering@gmail.com>
Signed-off-by: Andy Green <andy.green@linaro.org>
11 years agoadd lexical parser for headers 31/2931/1
Andy Green [Thu, 17 Jan 2013 17:55:48 +0000 (01:55 +0800)]
add lexical parser for headers

Profiling what happens during the ab test, one of the hotspots
was strcasecmp in a loop looking for header name matches each time.

This patch introduces a lexical parser that creates a state machine
in 276 bytes that encodes all the known header names.  The fsm is
walked bytewise as chaacters come in... most states do not need any
recursion to match or fail.

The state machine output is cut-and-pasted into parsers.c as an
unsigned char array.

The fsm generator is a bit rough and ready, included in the tree but
not built since normal mortals won't need to touch it.

Signed-off-by: Andy Green <andy.green@linaro.org>
11 years agodeprecate x google mux 30/2930/1
Andy Green [Thu, 17 Jan 2013 11:53:16 +0000 (19:53 +0800)]
deprecate x google mux

Unfortunately this code is beginning to rot due to lack of demand to
provide it and it being disabled by default.

If demand appears we can revert this and resume work on it, otherwise
let's bite the bullet for the moment.

Signed-off-by: Andy Green <andy.green@linaro.org>
11 years agosolve flowcontrol problems 29/2929/1
Andy Green [Thu, 17 Jan 2013 08:50:35 +0000 (16:50 +0800)]
solve flowcontrol problems

Problems with rx flow control implementation were the underlying cause
of the connection stalling issue that was covered up with the udelay()
patch that was removed recently.

This get rx flow control working properly and corrects problems with
fifo management in the test server mirror protocol code too.

The rxfow control api has been changed to just set a flag, so it's very cheap
to call from user code.  After the callbacks that might use the rxflow control
api the flag is checked and any pending actions done.

rx flow control now stops any rx packet coming immediately, with compessed
connections "just what was left in the pipe" might be hundreds of KBytes.  To
implement that the current packet being decoded is copied into a malloc'd buffer
by the rx processing code now.

When rxflow is allows to come again, the buffer is drained and freed before any
new packet content is accepted.

Signed-off-by: Andy Green <andy.green@linaro.org>
11 years agocheck errors on shutdown close 28/2928/1
Andy Green [Thu, 17 Jan 2013 06:46:43 +0000 (14:46 +0800)]
check errors on shutdown close

Also make sure CLOSE doesn't go through extension munging

Reduce wait for close ack to 1s

Signed-off-by: Andy Green <andy.green@linaro.org>
11 years agoreplace hashtable polltable management 27/2927/1
Andy Green [Thu, 17 Jan 2013 04:26:48 +0000 (12:26 +0800)]
replace hashtable polltable management

This rips out the connection hashtable implementation along with
MAX_CLIENTS and replaces it with a dynamically allocated fds array
and lookup table along the same lines as the new extpoll implementation
from Edwin van den Oetelaar.

It detects the max number of file descriptors possible at context init
time and allocates accordingly; this can be externally controlled by
ulimit and the server run as a specific user to facilitate targeting
specific ulimit rules at it.

Many operations that translated between socket descriptors and struct
websocket or pollfd objects have had iteration removed by this patch
and under load will be a lot faster.

Signed-off-by: Andy Green <andy.green@linaro.org>
11 years agouse simple lookup table for extpoll 26/2926/1
Edwin van der Oetelaar [Thu, 17 Jan 2013 03:16:15 +0000 (11:16 +0800)]
use simple lookup table for extpoll

Hash stuff is overkill since Edwin found a max connection limit of 30000 on his
box anyway.  Just use a simple preallocated lookup table and fds array.

AG Modified for style and removed debugging bits

Signed-off-by: Edwin van der Oetelaar <oetelaar.automatisering@gmail.com>
Signed-off-by: Andy Green <andy.green@linaro.org>
11 years agoinclude daemonization file whoops 25/2925/1
Andy Green [Thu, 17 Jan 2013 07:49:27 +0000 (15:49 +0800)]
include daemonization file whoops

Signed-off-by: Andy Green <andy.green@linaro.org>
11 years agoportability dont assume size of tv.t_usec 24/2924/1
Andy Green [Thu, 17 Jan 2013 07:02:02 +0000 (15:02 +0800)]
portability dont assume size of tv.t_usec

Signed-off-by: Andy Green <andy.green@linaro.org>
11 years agoadd disable debug to README configuration options list 23/2923/1
Andy Green [Thu, 17 Jan 2013 02:15:52 +0000 (10:15 +0800)]
add disable debug to README configuration options list

Signed-off-by: Andy Green <andy.green@linaro.org>
11 years agoincluding assert h needed on osx 22/2922/1
Peter Young [Thu, 17 Jan 2013 02:10:10 +0000 (10:10 +0800)]
including assert h needed on osx

Signed-off-by: Peter Young <young40@qq.com>
11 years agojust use limits.h directly 21/2921/1
Peter Young [Thu, 17 Jan 2013 02:08:16 +0000 (10:08 +0800)]
just use limits.h directly

Signed-off-by: Peter Young <young40@qq.com>
11 years agointroduce daemonize 20/2920/1
Andy Green [Thu, 17 Jan 2013 02:05:39 +0000 (10:05 +0800)]
introduce daemonize

Signed-off-by: Andy Green <andy.green@linaro.org>
11 years agoclient allow remote server to accept with no protocol set 19/2919/1
Andy Green [Wed, 16 Jan 2013 16:50:48 +0000 (00:50 +0800)]
client allow remote server to accept with no protocol set

Needed if our clientside can connect to, eg, echo.websocket.org

Signed-off-by: Andy Green <andy.green@linaro.org>
11 years agomove array bounds gcc workaround outside function 18/2918/1
Andy Green [Wed, 16 Jan 2013 09:46:00 +0000 (17:46 +0800)]
move array bounds gcc workaround outside function

Signed-off-by: Andy Green <andy.green@linaro.org>
11 years agoupdate ping test client and stop exposing payload to extensions 17/2917/1
Andy Green [Wed, 16 Jan 2013 06:35:12 +0000 (14:35 +0800)]
update ping test client and stop exposing payload to extensions

Ping and Pong payload in control messages need to be
above the fray of extension payload munging

Signed-off-by: Andy Green <andy.green@linaro.org>
11 years agoexport lswl_hexdump 16/2916/1
Andy Green [Wed, 16 Jan 2013 06:35:27 +0000 (14:35 +0800)]
export lswl_hexdump

Signed-off-by: Andy Green <andy.green@linaro.org>
11 years agoroubustness only return 0 from fd service if handled 15/2915/1
Andy Green [Wed, 16 Jan 2013 05:40:43 +0000 (13:40 +0800)]
roubustness only return 0 from fd service if handled

Otherwise our app outer loop will think the world is ending

Signed-off-by: Andy Green <andy.green@linaro.org>
11 years agoconfigure without client 14/2914/1
Andy Green [Wed, 16 Jan 2013 03:47:40 +0000 (11:47 +0800)]
configure without client

This leverages the refactor patches to introduce the ability to
disable building any client side code in the library or the client
side test apps.

This will be a considerable size saving for embedded server-only
case.

Signed-off-by: Andy Green <andy.green@linaro.org>
11 years agotest client remove usleep 13/2913/1
Andy Green [Wed, 16 Jan 2013 04:32:20 +0000 (12:32 +0800)]
test client remove usleep

Whatever caused the need for this has gone away

Signed-off-by: Andy Green <andy.green@linaro.org>
11 years agorefactor output.c 12/2912/1
Andy Green [Wed, 16 Jan 2013 04:21:29 +0000 (12:21 +0800)]
refactor output.c

Signed-off-by: Andy Green <andy.green@linaro.org>
11 years agorefactor migrate client stuff to client.c 11/2911/1
Andy Green [Wed, 16 Jan 2013 03:53:05 +0000 (11:53 +0800)]
refactor migrate client stuff to client.c

Signed-off-by: Andy Green <andy.green@linaro.org>
11 years agoadd new context arg to libwebsockets_serve_http_file 10/2910/1
Andy Green [Wed, 16 Jan 2013 03:45:38 +0000 (11:45 +0800)]
add new context arg to libwebsockets_serve_http_file

Signed-off-by: Andy Green <andy.green@linaro.org>
11 years agorobustness server dont exit server on accept problems 09/2909/1
Andy Green [Wed, 16 Jan 2013 03:43:53 +0000 (11:43 +0800)]
robustness server dont exit server on accept problems

Error message corrected to %s as well

Signed-off-by: Andy Green <andy.green@linaro.org>
11 years agoworkaround for some gcc array bounds false positive 08/2908/1
Andy Green [Wed, 16 Jan 2013 02:21:34 +0000 (10:21 +0800)]
workaround for some gcc array bounds false positive

Signed-off-by: Andy Green <andy.green@linaro.org>
11 years agoadd logo to test file 07/2907/1
Andy Green [Wed, 16 Jan 2013 02:11:34 +0000 (10:11 +0800)]
add logo to test file

Signed-off-by: Andy Green <andy.green@linaro.org>
11 years agoupdate test server html serving callback to use aepd whitelist approach 06/2906/1
Andy Green [Wed, 16 Jan 2013 02:06:28 +0000 (10:06 +0800)]
update test server html serving callback to use aepd whitelist approach

Signed-off-by: Andy Green <andy.green@linaro.org>
11 years agoadd libwebsockets.org logo to share 05/2905/1
Andy Green [Wed, 16 Jan 2013 02:00:39 +0000 (10:00 +0800)]
add libwebsockets.org logo to share

Signed-off-by: Andy Green <andy.green@linaro.org>
11 years agooptimize http file sending 04/2904/1
Andy Green [Wed, 16 Jan 2013 00:37:48 +0000 (08:37 +0800)]
optimize http file sending

This adapts the approach from the single-packet-per-poll-loop improvement
to sending more packets while the socket can take them.

It still falls back to the multi-state scheme if the socket ever chokes,
which it certainly will on larger files, so it's safe while being highly
efficient at smaller file sizes.

Nor should it significantly add to latency for other sockets, it simply
stuffs the pipe asynchronously as much as the pipe can take.

We also increase the packet payoad size from 512 to 1400 a time.

This reduces the time taken in the 300 connection / 5000 transfers ab test
from >8s to ~3.4s, transferring the same amount of data.

$ ab -t 100 -n 5000 -c 300 'http://127.0.0.1:7681/'
This is ApacheBench, Version 2.3 <$Revision: 1373084 $>
Copyright 1996 Adam Twiss, Zeus Technology Ltd, http://www.zeustech.net/
Licensed to The Apache Software Foundation, http://www.apache.org/

Benchmarking 127.0.0.1 (be patient)
Completed 500 requests
Completed 1000 requests
Completed 1500 requests
Completed 2000 requests
Completed 2500 requests
Completed 3000 requests
Completed 3500 requests
Completed 4000 requests
Completed 4500 requests
Completed 5000 requests
Finished 5000 requests

Server Software:        libwebsockets
Server Hostname:        127.0.0.1
Server Port:            7681

Document Path:          /
Document Length:        8447 bytes

Concurrency Level:      300
Time taken for tests:   3.400 seconds
Complete requests:      5000
Failed requests:        0
Write errors:           0
Total transferred:      42680000 bytes
HTML transferred:       42235000 bytes
Requests per second:    1470.76 [#/sec] (mean)
Time per request:       203.976 [ms] (mean)
Time per request:       0.680 [ms] (mean, across all concurrent requests)
Transfer rate:          12260.17 [Kbytes/sec] received

Connection Times (ms)
              min  mean[+/-sd] median   max
Connect:        7   24  15.6     20     125
Processing:    32  172  50.2    161     407
Waiting:       27  154  49.4    142     386
Total:         81  196  48.3    182     428

Percentage of the requests served within a certain time (ms)
  50%    182
  66%    185
  75%    188
  80%    194
  90%    304
  95%    316
  98%    322
  99%    328
 100%    428 (longest request)

Signed-off-by: Andy Green <andy.green@linaro.org>
11 years agolisten socket more frequent service 03/2903/1
Andy Green [Tue, 15 Jan 2013 23:59:47 +0000 (07:59 +0800)]
listen socket more frequent service

From an idea by Edwin van den Oetelaar <oetelaar.automatisering@gmail.com>

When testing libwebsockets with ab, Edwin found an unexpected bump in
the distribution of latencies, some connections were held back almost
the whole test duration.

http://ml.libwebsockets.org/pipermail/libwebsockets/2013-January/000006.html

Studying the problem revealed that when there are mass pending connections
amongst many active connections, we do not service the listen socket often
enough to clear the backlog, some seem to get stale violating FIFO ordering.

This patch introduces listen socket service "piggybacking", where every n
normal socket service actions we also check the listen socket and deal with
pending connections there.

Normally, it checks the listen socket gratuitously every 10 normal socket
services.  However, if it finds something waiting, it forces a check on the
next normal socket service too by keeping stats on how often something was
waiting.  If the probability of something waiting each time becomes high,
it will allow up to two waiting connections to be serviced for each normal
socket service.

In that way it has low burden in the normal case, but rapidly adapts by
detecting mass connection loads as found in ab.

Signed-off-by: Andy Green <andy.green@linaro.org>
11 years agoadd empty m4 dir as workaround for autoreconf issue 02/2902/1
Andy Green [Tue, 15 Jan 2013 13:56:12 +0000 (21:56 +0800)]
add empty m4 dir as workaround for autoreconf issue

Signed-off-by: Andy Green <andy.green@linaro.org>
11 years agoallow LWS_SOMAXCONN to be defined at configuretime 01/2901/1
Andy Green [Tue, 15 Jan 2013 12:52:29 +0000 (20:52 +0800)]
allow LWS_SOMAXCONN to be defined at configuretime

Default remains at SOMAXCONN, you can force it at configure time
along these lines

./configure CFLAGS="-DLWS_SOMAXCONN=16384"

Signed-off-by: Andy Green <andy.green@linaro.org>
11 years agoextpoll use hashtable for fd tracking 00/2900/1
Andy Green [Tue, 15 Jan 2013 11:44:33 +0000 (19:44 +0800)]
extpoll use hashtable for fd tracking

This implements a much faster, hashtable-based tracking scheme for
external poll fds.

Signed-off-by: Andy Green <andy.green@linaro.org>
11 years agoallow building just the library not the testapps 99/2899/1
Andy Green [Tue, 15 Jan 2013 11:57:06 +0000 (19:57 +0800)]
allow building just the library not the testapps

From an idea by Jack Mitchell <ml@communistcode.co.uk>

Use --without-testapps at configure time to suppress building them

Signed-off-by: Andy Green <andy.green@linaro.org>
11 years agomake sure we have PATH_MAX on some linux toolchains 98/2898/1
Jack Mitchell [Tue, 15 Jan 2013 11:49:05 +0000 (19:49 +0800)]
make sure we have PATH_MAX on some linux toolchains
(AG modified a bit)

Signed-off-by: Jack Mitchell <ml@communistcode.co.uk>
11 years agoextpoll break out of loop when set or clear finds fd 97/2897/1
Edwin van der Oetelaar [Tue, 15 Jan 2013 08:22:34 +0000 (16:22 +0800)]
extpoll break out of loop when set or clear finds fd

Signed-off-by: Edwin van der Oetelaar <oetelaar.automatisering@gmail.com>
11 years agohttp service break into outer loop states 96/2896/1
Andy Green [Tue, 15 Jan 2013 05:40:23 +0000 (13:40 +0800)]
http service break into outer loop states

Previously we sat and looped to dump a file over http protocol.

Actually that's a source of blocking to the other sockets being serviced.

This patch breaks up the file service into a roundtrip around the poll()
loop for each 512-byte packet.  It doesn't make much difference if the
server is idle, but if it's busy it makes sure everyone else is getting
service while the file is sent.

It doesn't try to optimize multiple users of the file or to keep the
descriptor open, the point of this patch is to establish the breaking up
of the file send action into the poll loop.

On the user side, there are two differences:

 - context is now needed in the first argument to libwebsockets_serve_http_file()
that's not too bad since we provide context in the callback.

 - file send is now asynchronous to the user code, you get a new callback coming
in protocol 0 when it's done, LWS_CALLBACK_HTTP_FILE_COMPLETION

libwebsockets-test-server is updated accordingly.

Signed-off-by: Andy Green <andy.green@linaro.org>
11 years agomerge test server extpoll into test server 95/2895/1
Andy Green [Tue, 15 Jan 2013 04:39:48 +0000 (12:39 +0800)]
merge test server extpoll into test server

the -extpoll version of the test server was starting to rot compared to
the test-server.c it was originally based on.

This patch deletes the -extpoll.c version and instead has the test-server.c
source built two different ways in the makefile, once with the define
EXTERNAL_POLL which forces non-fork mode and enables the "by hand"
pollfd array handling.  The resulting binary of that is still called
libwebsockets-test-server-extpoll.

Another problem was that the pollfd array length needs to match MAX_CLIENTS, that
now happens during the build.

Signed-off-by: Andy Green <andy.green@linaro.org>
11 years agooptimize extpoll fd delete 94/2894/1
Edwin van der Oetelaar [Tue, 15 Jan 2013 03:23:05 +0000 (11:23 +0800)]
optimize extpoll fd delete

Previous method of shifting back array by one to cover the deleted
item could be expensive when the list was large and the deleted item
appeared early in it.

This scheme swaps the last guy into the vacant space and reduces the
list size by one.

(AG adapted for style and not to care if n is end guy)

Signed-off-by: Edwin van der Oetelaar <oetelaar.automatisering@gmail.com>
Signed-off-by: Andy Green <andy.green@linaro.org>
11 years agodeal with SSL_ERROR_WANT_ in client connect action 93/2893/1
Andy Green [Sun, 13 Jan 2013 03:05:30 +0000 (11:05 +0800)]
deal with SSL_ERROR_WANT_ in client connect action

"4b0e01f Retry SSL_connect when SSL_get_error requests it. " from David Galeano
noticed the problem that client connect may receive SSL_ERROR_WANT_* from
SSL_connect, which is basically WOULDBLOCK.  That patch tried to deal with it
by blocking in a while(1) until the condition went away.

That's problematic because of it blocks service of anything else (including
the host application sockets in the external socket poll sharing case) for
up to 5s controlled by conditions at one client.

After fiddling with and researching this, the actual problem with the code is
we are not getting the SSL layer error correctly, it is not contained in the
code returned from the Connect api directly.

I was unable to get a renegotiation forced on my modern SSL libs, it complained
about protocol error are reopened the connection instead.  So I think the stuff
found in the docs and the web about the SSL_ERROR_WANT_ is probably not something
we will see in reality (if we check the right error code...)

Signed-off-by: Andy Green <andy.green@linaro.org>
11 years agoadd longlived option to test client 92/2892/1
Andy Green [Sun, 13 Jan 2013 03:58:18 +0000 (11:58 +0800)]
add longlived option to test client

Needed to confirm pending timeouts won't kill the connection, by default
it spams the server with connections that live less than 5s

Signed-off-by: Andy Green <andy.green@linaro.org>
11 years agologging ensure everyone has a newline 91/2891/1
Andy Green [Mon, 14 Jan 2013 05:10:55 +0000 (13:10 +0800)]
logging ensure everyone has a newline

Signed-off-by: Andy Green <andy.green@linaro.org>
11 years agoreplace ifdefs around close socket with compatible_close 90/2890/1
Andy Green [Mon, 14 Jan 2013 07:35:02 +0000 (15:35 +0800)]
replace ifdefs around close socket with compatible_close

Signed-off-by: Andy Green <andy.green@linaro.org>
11 years agossl client certs fix crash 89/2889/1
Larry Hayes [Mon, 14 Jan 2013 17:03:58 +0000 (01:03 +0800)]
ssl client certs fix crash

I run a web socket server that requires clients to present a certificate.

context_ssl_ = libwebsocket_create_context(wssPort_, wssIpAddr_.c_str(), protocols_ssl,

                           libwebsocket_internal_extensions,

                           cert_path.c_str(), key_path.c_str(), -1, -1,

LWS_SERVER_OPTION_REQUIRE_VALID_OPENSSL_CLIENT_CERT);

I am getting a crash in the OpenSSL_verify_callback().

The SSL_get_ex_data() call is returning NULL

I could not find a call to SSL_set_ex_data() for server mode operation.

Has anyone seen this crash in the newer versions?

Signed-off-by: Larry Hayes <larry.hayes@prodeasystems.com>
11 years agoabsorb README.rst into main README and code 88/2888/1
Andy Green [Sun, 13 Jan 2013 01:53:18 +0000 (09:53 +0800)]
absorb README.rst into main README and code

Some of the advice in README.rst became deprecated with recent patches,
the (good) advice about http connection close is better demonstrated
in the code and API docs, and the remainder can go in the main README,
which will have to be refactored itself at some point.

Signed-off-by: Andy Green <andy.green@linaro.org>
11 years agoexpose compiletime constants to setting from configure 87/2887/1
Andy Green [Sat, 12 Jan 2013 15:42:17 +0000 (23:42 +0800)]
expose compiletime constants to setting from configure

This patch allows control of the main compiletime constants in libwebsockets
from the configure commandline.

README is updated with documentation on what's available, how to set them
and the defaults.

The constants are logged with "info" severity (not visible by default) at
context create time.

The zlib constant previously exposed like this is moved to private-libwebsockets.h
so it can be printed along with the rest.

Signed-off-by: Andy Green <andy.green@linaro.org>
11 years agorenable deflate frame buffer expansion fixing DoS 86/2886/1
Andy Green [Sat, 12 Jan 2013 15:09:36 +0000 (23:09 +0800)]
renable deflate frame buffer expansion fixing DoS

This reverts the removal of the deflate_frame code that was crashing after porting
David Galeano's code: he pointed out there's a typo in the merged version causing
the crash which is fixed here.

However the fixed code has a problem, there's no limit (other than int size) to the
amount of memory it will try to malloc, which can allow a DoS of the server by the
client sending malicious compression states that inflate to a large amount.  I have
added checking for OOM already that will avert the segfault that would otherwise follow
but the server will be unusuable if malicious connections were made repeatedly each
forcing it to allocate large buffers and cause small allocations on other connections
to fail.

The patch changes the code to use realloc(), and introduces a configurable limit
on the amount of memory one connection may need for zlib before the server hangs
up the connection.  It defaults to 64KBytes but can be set from ./configure as
described now in the README.

Signed-off-by: Andy Green <andy.green@linaro.org>
11 years agofix config enable name for getifaddrs 85/2885/1
Andy Green [Sat, 12 Jan 2013 15:31:39 +0000 (23:31 +0800)]
fix config enable name for getifaddrs

copy-paste...

Signed-off-by: Andy Green <andy.green@linaro.org>
11 years agointroduce getifaddrs for toolchains without it 84/2884/1
David [Sat, 12 Jan 2013 12:39:47 +0000 (20:39 +0800)]
introduce getifaddrs for toolchains without it

David found that uclibc did not provide this slightly esoteric api
and provided one from BSD that can be built by the library internally.

AG: Made contingent on configure option --enable-builtin-getifaddrs

Signed-off-by: David <cymerio@gmail.com>
Signed-off-by: Andy Green <andy.green@linaro.org>
11 years agoaudit and make all malloc check for OOM 83/2883/1
Andy Green [Sat, 12 Jan 2013 05:21:08 +0000 (13:21 +0800)]
audit and make all malloc check for OOM

Signed-off-by: Andy Green <andy.green@linaro.org>
11 years agologging add timestamp 82/2882/1
Andy Green [Sat, 12 Jan 2013 01:25:07 +0000 (09:25 +0800)]
logging add timestamp

Signed-off-by: Andy Green <andy.green@linaro.org>
11 years agologging extend level set api to allow setting emission function 81/2881/1
Andy Green [Sat, 12 Jan 2013 01:17:42 +0000 (09:17 +0800)]
logging extend level set api to allow setting emission function

Signed-off-by: Andy Green <andy.green@linaro.org>
11 years agoupdate README with info on new logging scheme 80/2880/1
Andy Green [Thu, 10 Jan 2013 14:56:15 +0000 (22:56 +0800)]
update README with info on new logging scheme

Signed-off-by: Andy Green <andy.green@linaro.org>
11 years agoallow enabling debug contexts from test apps 79/2879/1
Andy Green [Thu, 10 Jan 2013 14:28:59 +0000 (22:28 +0800)]
allow enabling debug contexts from test apps

Adds a -d switch to everything so you can set the log level bitfeld.

Signed-off-by: Andy Green <andy.green@linaro.org>
11 years agointroduce logging api and convert all library output to use it 78/2878/1
Andy Green [Thu, 10 Jan 2013 11:50:35 +0000 (19:50 +0800)]
introduce logging api and convert all library output to use it

 - multiple debug context calls lwsl_ err, warn, debug, parser, ext, client

 - api added to set which contexts output to stderr using a bitfield log_level

 - --disable-debug on configure removes all code that is not err or warn severity

 - err and warn contexts always output to stderr unless disabled by log_level

 - err and warn enabled by default in log_level

Signed-off-by: Andy Green <andy@warmcat.com>
11 years agocompile in xcode, privatize debug macro 77/2877/1
Aaron Zinman [Thu, 10 Jan 2013 04:35:18 +0000 (12:35 +0800)]
compile in xcode, privatize debug macro

11 years agoupdate soname and configure to v1.0 76/2876/1
Andy Green [Thu, 10 Jan 2013 04:35:18 +0000 (12:35 +0800)]
update soname and configure to v1.0

Signed-off-by: Andy Green <andy.green@linaro.org>
11 years agocorrect autotools warning 75/2875/1
Andy Green [Thu, 10 Jan 2013 04:36:59 +0000 (12:36 +0800)]
correct autotools warning

Signed-off-by: Andy Green <andy.green@linaro.org>
11 years agozlib code add OOM checks remove buffer expansion on rx path 74/2874/1
Andy Green [Thu, 10 Jan 2013 04:26:13 +0000 (12:26 +0800)]
zlib code add OOM checks remove buffer expansion on rx path

Here testing with the test serer and chrome 25, the buffer expansion
code on Rx was triggered by a valid no data output condition and looped
until it exhausted all memory.

This patch adds OOM check to all malloc()s and removes the buffer expansion
code on the rx path... leaving the code on tx path for now.

Signed-off-by: Andy Green <andy.green@linaro.org>
11 years agoAvoid leaking a socket when SSL_accept fails. 73/2873/1
David Galeano [Thu, 10 Jan 2013 02:45:24 +0000 (10:45 +0800)]
Avoid leaking a socket when SSL_accept fails.