f18a6388fd
fix kore_realloc() to actually behave properly
2013-06-27 09:20:48 +02:00
4b2420097b
- Better spread load between all worker processes.
...
- Introduce own memory management system on top of malloc to keep track
of all our allocations and free's. Later we should introduce a pooling
mechanism for fixed size allocations (http_request comes to mind).
- Introduce ssl_cipher in configuration.
Memory usage is kind of high right now, but it seems its OpenSSL
doing it rather then Kore.
2013-06-27 08:46:18 +02:00
b4a0330a96
- Better spread load between all worker processes.
...
- Introduce own memory management system on top of malloc to keep track
of all our allocations and free's. Later we should introduce a pooling
mechanism for fixed size allocations (http_request comes to mind).
- Introduce ssl_cipher in configuration.
Memory usage is kind of high right now, but it seems its OpenSSL
doing it rather then Kore.
2013-06-27 08:43:07 +02:00
2fc5233358
Rework the way worker processes give each other the accept lock.
...
Instead of waiting until one worker is filled up on connections
the workers find the next lowest loaded worker and will hand
over the lock to them instead. This will cause a nicer spread of load.
Instead of running one accept per event loop, we attempt to accept
as many as worker_max_connections allows.
Refactor net sending/recv code a bit.
2013-06-27 00:22:48 +02:00
124f3ecad9
no need to use kore_log() in a debug context
2013-06-26 15:59:42 +02:00
bf1940225a
everybody loves to tout their own horn.. so introduce a server response header
2013-06-26 16:58:01 +02:00
0dda6f996f
Add a form of synchronization between what worker will be accepting
...
new connections and which ones will not be notified for it.
Fixes the thundering herd problem, and nicely spreads out load between
all the workers equally. A configuration option (workers_max_connections)
is available to tweak how many connections a worker will have before
giving up the accept lock.
Two ways are added to this commit for access locking:
- Locking via semaphores.
- Locking via GCC's builtin atomic methods.
The default is running with semaphores disabled (OpenBSD cannot do
sem_init() with pshared set to 1, which is required).
If you want to use semaphores add KORE_USE_SEMAPHORES to CFLAGS,
and -lpthread to LDFLAGS in the Makefile.
Other fixes:
- BSD: add a timeout to kevent().
- Merge kore_worker_wait together, linux knows waitpid() as well.
- Send the correct SIGQUIT signal to workers instead of SIGINT.
- Fix kore_time_ms().
- Log fatal worker messages in syslog.
- Refactor code even more.
- Do not free our own kore_worker structure.
2013-06-26 16:37:22 +02:00
315f964abd
correct typo
2013-06-26 11:20:25 +02:00
3e5c17b8a3
refactor code quite a bit.
2013-06-26 11:18:32 +02:00
6026a6d4ee
add SNI support, and change domain configuration a bit.
2013-06-24 11:32:45 +02:00
a1b400c400
Add access logging to Kore.
2013-06-24 09:36:40 +02:00
4dff38ebb0
ignore sigpipe in workers
2013-06-19 22:41:00 +02:00
c22eebb6d9
for non spdy clients, attempt to keep the connection open as long as possible (aka, send connection: keep-alive and do not close it till client does).
2013-06-22 20:14:10 +02:00
8b832f0a37
fix off-by-one
2013-06-20 23:30:03 +02:00
c49622e4ae
do not schedule kqueue events if we've reached the limit of changelist.
2013-06-20 23:29:26 +02:00
25f1ab9865
Add BSD kqueue(2) support. Compile with make bsd (or make linux for linux)
2013-06-17 23:39:17 +02:00
e170e916ce
Reload the module in the main process as well when SIGHUP is received.
...
Otherwise new worker processes will not receive the updated module.
2013-06-05 13:50:50 +02:00
b4deea82b4
add kore_buf_appendb() which allows us to append a kore_buf to another one.
...
(releases the kore_buf that is being appended while at it).
2013-06-05 11:27:03 +02:00
a74fffe40c
Introduce certfile and certkey in the configuration to specify where the certificate file and keys are located on a system.
...
Free unused vars in the main process after starting.
2013-06-05 09:47:08 +02:00
32a2035ce9
move kore_server_sslstart() into main process, workers will inherit.
2013-06-05 09:32:53 +02:00
338047a95b
use kore_log() to notify when a module is reloaded
2013-06-05 08:56:37 +02:00
b9f47b0f86
configuration files are now passed using the -c option.
...
allow debug output when the -d flag is specified.
2013-06-05 08:55:07 +02:00
ade34a26c1
move chroot() to workers instead, parent process doesn't require it plus it cannot unlink pidfile otherwise.
...
better logging in syslog
2013-06-05 08:45:51 +02:00
90e1b3a7da
use syslog() for informative messages from all parts of kore.
2013-06-04 23:24:47 +02:00
49b77d3b0e
better proctitles
2013-06-04 17:04:28 +02:00
276d8c5a82
oops bring back chroot
2013-06-04 17:01:06 +02:00
ca437a6cef
set process title
2013-06-04 16:58:13 +02:00
9ef669ff6f
write main process pid to /var/run/kore.pid (changable in configuration)
2013-06-04 16:53:30 +02:00
443b1c8c5f
format
2013-06-04 16:33:35 +02:00
e7db5ee6b1
rename kore_log to kore_debug, and allow one to turn it off.
2013-06-04 16:30:53 +02:00
11fca19923
be less spammy
2013-06-04 16:17:42 +02:00
ab0dc25c61
use sched_setaffinity() to set what CPU each worker process should run on.
2013-06-04 13:54:16 +02:00
8f8ab92521
upon quit time, workers will not accept new connections but will
...
handle all outstanding http requests.
2013-06-04 13:43:11 +02:00
0de28488a6
move from multithreads to single threaded worker processes.
2013-06-04 11:55:38 +02:00
b65cc93426
allow handlers to return KORE_RESULT_RETRY. This will tell the worker to reschedule the page request again at the end of its list. (Allows module creators to write truely nonblocking modules).
2013-05-31 00:40:06 +02:00
7dfa7e6ec0
be carefull when we reload the module to not reload it when workers are inside the module callbacks.
...
do this by implementing a pthread rwlock, and locking it for reading when going into a callback and locking it for writing when we need to reload the mod.
2013-05-31 00:06:54 +02:00
fecbd058cb
rework the worker thread so there's actually time to schedule more then one request at a time on them.
2013-05-31 14:24:00 +02:00
e428886e16
decrement the worker load on HTTP_REQUEST_DELETE
2013-05-31 13:30:51 +02:00
bb4001d119
ok that was wrong, disconnect http clients whne we're done sending or upon error, not immediately after queueing everything.
2013-05-30 21:57:14 +02:00
ca1c884e43
attempt to clear out send buffer before we shutdown ssl connection.
...
remove superfleaous debug
2013-05-30 21:39:01 +02:00
cf6a6550f0
allow onload to be given in the config file.
...
onload specifies what function in your module to call when the module has been loaded or reloaded.
2013-05-30 21:26:39 +02:00
ec5ac40706
wake up the workers once in a while to process anything that is waiting.
2013-05-30 20:55:50 +02:00
f9b3cfcee4
reschedule events if we cannot lock the connection at the time being.
2013-05-30 20:38:25 +02:00
9ad263e287
do not remove disconnected connections until we actually are ready to disconnect them.
2013-05-30 20:07:06 +02:00
9243f409cc
move to a worker based threading approach where we delegate http requests to workers in a round robin basis (later this should be swapped to find the laziest worker and assign the request to that instead).
2013-05-30 19:36:42 +02:00
8478d8df54
add chroot and runas directives so we can chroot and drop privilegs properly
2013-05-04 22:18:27 +02:00
3b30920a60
for now, until we properly handle a persistent HTTP connection for non spdy clients make sure we close the connection after handling the HTTP request.
2013-05-04 21:03:53 +02:00
ce729010f7
- simplify header building using kore_buf_appendf() for normal HTTP requests.
...
- make sure we dont free nb->buf if its NULL. (semantics).
- remove some superfluffy debug.
- make sure we call [inflate|deflate]End when we dc a client that used zlib.
2013-05-04 20:44:16 +02:00
45adae62f7
q->value should be set to NULL if a query parameter is present but no value was set for it.
...
introduce kore_buf_appendv() (much like readv()).
introduce kore_buf_appendf() (printf into buffers).
2013-05-04 19:09:07 +02:00
61b937ac1b
remove unused vars
2013-05-04 16:38:40 +02:00
088d877715
switch dynamic handlers to regex based patterns, much easier.
2013-05-29 14:29:46 +02:00
2290d09d3b
Add dynamic handles, which can be used to partially match a URI and still call a cb handler. This is especially usefull when considering the handlers as a ruleset:
...
static / serve_index
static /foo serve_foo
dynamic / serve_other
/ will be matched to serve_index, while /foo will be matched to serve_foo and /bar will be matched to serve_other for example.
2013-05-29 13:33:32 +02:00
0be748ee0f
use SSL_CTX_use_certificate_chain_file instead when loading certificates so we can properly use our intermediate authority.
2013-05-29 12:27:30 +02:00
1671698f58
increase sbuf length substantially
2013-05-22 12:50:20 +02:00
a88b8de6a2
add proper query argument parsing, modules must call
...
http_populate_arguments() before using http_argument_lookup() to lookup a given argument.
population uses the appropriate query string (from GET or POST) automatically.
2013-05-03 09:09:27 +02:00
b3cdccbc87
add http_post_data_text() which returns the post data NUL-terminated to the caller.
2013-05-03 07:49:40 +02:00
85aad24af5
do not NUL terminate the post data, for later.
2013-05-03 07:43:53 +02:00
2401bd1a9f
support for POST in both spdy and normal connections
2013-05-03 07:35:11 +02:00
879b5716d1
req->method is no longer a string, adapt kore_log()
2013-05-03 01:21:09 +02:00
90588913d4
allow sending SIGHUP to kore, which will then reload its content module
2013-05-03 00:04:06 +02:00
ba2ed8d954
host header might not come in second..
2013-05-02 22:35:29 +02:00
e7feaaba76
properly close a spdy stream once both sides have FLAG_FIN.
2013-05-02 17:30:06 +02:00
4ed4f76e44
allow POST to work with spdy, all POST data is stored in the httpreq->post_data field. (as a kore_buf)
2013-05-02 17:14:07 +02:00
f61bbe8ff4
move buf stuff back into kore.h, wont need it separately
2013-05-02 15:14:00 +02:00
a025f0fa8e
save the http request for spdy later on
2013-05-02 15:06:26 +02:00
84428f7133
move buf stuff into its own header, so we can use it in spdy.h
2013-05-02 14:55:57 +02:00
60ed3e0fd2
the net qeueuing functions don't need to return status codes, simplifies a lot of other code that revolves around those.
...
prepare for POST support
2013-05-02 14:47:02 +02:00
07f2f83def
make sure we do not dereference p if its NULL
2013-05-02 13:47:53 +02:00
807764614b
from now on configuration files must specify a domain for the handlers that follow. This allows for easy subdomain configuration.
...
example:
domain joris.local
static / serve_index
domain .joris.local
static / serve_another_index
2013-05-02 13:30:13 +02:00
53cbc1a21e
make sure we advance the host buffer properly
2013-05-02 12:11:56 +02:00
f1fa108f98
begin processing window_update messages
2013-05-02 10:10:03 +02:00
c2520b2ebf
oops, sending 2 bytes too much.
2013-05-02 09:25:14 +02:00
364dc58219
rework disconnecting clients and fix bug where spdy ocnnections could segfault after disconnecting
2013-05-02 09:10:35 +02:00
fdb9004c6d
do not kill spdy sessions upon receiving unknown frames.
2013-05-02 05:32:56 +02:00
b98c688429
remove debug
2013-05-02 04:44:52 +02:00
2cc4505c8c
correct offset
2013-05-02 03:52:53 +02:00
f46bba50ef
add very basic support for HTTP/1.1.
2013-05-02 03:51:04 +02:00
5f983d575b
add flags to net_recv_queue() and net_send_queue(),.
...
NETBUF_CALL_CB_ALWAYS flag, calls the registered cb everytime data has
been received or sent.
2013-05-02 00:28:49 +02:00
a9ebf37cae
allow modules to add headers to the response before calling http_response()
...
by calling http_response_header_add().
fix wrong overflow check in spdy_stream_get_header().
html_inject now exports last modified timestamp for the files that are
compiled into the module in the format static_mtime_<type>_<name>.
modules can now look into the request headers using http_request_header_get().
2013-05-01 21:16:09 +02:00
a228cdba0e
introduce kore_split_string() to properly split strings apart.
...
introduce kore_date_to_time() to conver http-date formatted strings to time_t.
2013-05-01 20:10:45 +02:00
39afa6c25b
add missing config.c
2013-05-01 17:17:16 +02:00
fa1a3099c1
better error msg
2013-05-01 17:16:09 +02:00
4ecfd81e6e
allow us to pass content-type to http_response()
2013-05-01 16:51:34 +02:00
a39040ba9b
build kore as -rdynamic so we can properly load modules.
...
implement kore_module_handler_find().
2013-05-01 16:11:10 +02:00
d9bbb0b1f1
initial module support
2013-05-01 16:03:48 +02:00
a074258dcc
rework the way events trigger and the way we handle them. follow the manual for once, as edge trigger does not refire until we got an EAGAIN.
...
introduce net_send_flush() and net_recv_flush() for this purpose, we attempt to make as much headway as possible until we reach EAGAIN or until we can simply pickup again later.
should merge all the stuff in http_response() into a single send buffer, **out is in place in net_send_queue() for that purpose.
2013-05-01 13:43:47 +02:00
d59847c448
SPDY works, we can serve content. Now for dynamic modules so we can
...
get stuff on the road.
TODO: add plain HTTP support if SPDY isnt working in the browser.
2013-05-01 12:23:21 +02:00
848704f74b
lots of new stuff, including processing of http requests and an attempt
...
to build an initial spdy response (SYN frame + header block content).
2013-05-01 08:09:04 +02:00
4fc434e909
make sure method, path and host are free'd in case of failure.
2013-05-01 01:23:46 +02:00
05156c9a3a
properly deflate/inflate of name/value blocks. and add ability
...
to get headers from them.
2013-05-01 00:35:33 +02:00
bb3b804222
add beginnings of zlib inflation, not really working. nap time.
2013-04-29 23:35:36 +02:00
071fc4100d
only clear our a netbuf if retain is 0 AND it's been filled.
2013-04-29 07:15:14 +02:00
c8b422d29d
allow us to expand receive buffers automatically so we can keep chaining data into the same netbuf. This gives us the possibility to retain the ctrl_frame by the time we reach the proper cb for the actual frame message.
2013-04-28 23:42:13 +02:00
23c0ec67c6
begin with the ability to read control frames. something feels fishy with epoll() and its triggering of events. I probably got it wrong.
2013-04-28 19:11:44 +02:00
c6c6b1823f
read next protocol properly and remember it for later.
2013-04-21 20:35:47 +02:00
68d0523817
Add initial SSL stuff + bare bones NPN advertisement. So far chrome connects and
...
shows our connection as SPDY/3, good start.
2013-04-21 20:21:46 +02:00
857c3f9121
first commit
2013-04-17 22:34:27 +02:00
Joris Vink