Commit Graph

194 Commits

Author SHA1 Message Date
Yorick de Wid 565bf5bdec Set server version (#157) 2017-01-13 13:03:47 +01:00
Joris Vink 57840a8366 Deal with the Host header in a way IPv6 hosts work
Fixes #164.
2017-01-11 11:01:58 +01:00
Joris Vink 31d14d028a add compile-time configurable mime types.
these types are used for the new builtin asset_serve_* functions.

must be defined in conf/build.conf.
2016-12-26 23:37:05 +01:00
Joris Vink d783a1d22d Add auto generated serving functions for assets.
These functions are created by the cli tool when building
and follow the naming format: asset_serve_<name>_<ext>().

Those serving functions can be used directly in handlers and
callthrough to a http_serveable() function that uses the SHA1
of the asset as its ETag and automatically checks for if-none-match.
2016-12-26 21:15:03 +01:00
Joris Vink facc8b9d6c set req->owner to NULL when the connection removes it. 2016-12-26 20:08:53 +01:00
Joris Vink 0b92afe53d Explicitly set offset to 0 for http_file. 2016-07-27 16:43:05 +02:00
Joris Vink 43fec8678e kore_buf_create -> kore_buf_alloc. 2016-07-14 12:34:29 +02:00
Joris Vink 4ad50caa29 Large changes to the memory subsystem in kore.
- Change pools to use mmap() for allocating regions.
- Change kore_malloc() to use pools for commonly sized objects.
  (split into multiple of 2 buckets, starting at 8 bytes up to 8192).
- Rename kore_mem_free() to kore_free().

The preallocated pools will hold up to 128K of elements per block size.

In case a larger object is to be allocated kore_malloc() will use
malloc() instead.
2016-07-12 13:54:14 +02:00
Joris Vink d30921103b Code cleanup, several API breaking changes in here 2016-07-08 10:03:41 +02:00
Joris Vink 2dfd22a79a Change kore_buf_stringify() a bit.
Takes a size_t pointer as its second argument now, if not
NULL this will be populated with the length of the string
that is being returned.
2016-06-02 07:08:19 +02:00
Joris Vink fb6512259c small style changes. 2016-03-21 15:30:11 +01:00
Ángel González 1bda217d57 Based on #115 pull request, adds http_method_text, similar to http_status_text 2016-03-21 15:12:27 +01:00
Joris Vink f7b0b7c7ac Sometimes content_length wasn't properly updated. 2016-02-22 14:08:21 +01:00
Joris Vink 8fcf4762f4 Improve http_body_recv().
- If we fail to write to our temporary file error instead of fatal.
- Return KORE_RESULT_ERROR on other errors as well.
2016-02-13 15:41:37 +01:00
Joris Vink 12e1820a4f Unbreak normal POST parsing.
It was broken for > 2 fields due to recent large changes,
unfortunately i didn't catch this until now.
2016-02-06 15:32:23 +01:00
Joris Vink f4d00645ed Merge pull request #99 from oneswig/master
Add resource management as part of the kore shutdown process.
2016-02-01 19:51:27 +01:00
Joris Vink 1f5e482b8a Build option changes.
- Build with -O2 unless NOOPT is set to 1.
- Hide -g behind DEBUG instead of always building with it.
- Explicitely set the standard used to c99, use pedantic.
2016-02-01 15:33:40 +01:00
Stig Telfer 55d2451b90 Add cleanup of http_body_path pool 2016-02-01 10:10:04 +00:00
Stig Telfer 2ac6e7d41d Merge branch 'master' into oneswig 2016-01-24 13:46:38 +00:00
Stig Telfer ec73c35952 Add missing C library headers 2016-01-22 11:08:13 +00:00
Joris Vink 192f458f97 remove unused variable. 2016-01-18 11:43:44 +01:00
Joris Vink fcb86ddb8b Massive rework of HTTP layer.
This commit is a flag day, your old modules will almost certainly
need to be updated in order to build properly with these changes.

Summary of changes:

- Offload HTTP bodies to disk if they are large (inspired by #100).
  (disabled by default)
- The http_argument_get* macros now takes an explicit http_request parameter.
- Kore will now throw 404 errors almost immediately after an HTTP request
  has come in instead of waiting until all data has arrived.

API changes:

- http_argument_get* macros now require an explicit http_request parameter.
  (no more magic invokations).
- http_generic_404() is gone
- http_populate_arguments() is gone
- http_body_bytes() is gone
- http_body_text() is gone
- http_body_read() has been added
- http_populate_post() has been added
- http_populate_get() has been added
- http_file_read() has been added
- http_file_rewind() has been added
- http_file_lookup() no longer takes name, fname, data and len parameters.
- http_file_lookup() now returns a struct http_file pointer.
- http_populate_multipart_form() no longer takes an secondary parameter.

New configuration options:

- http_body_disk_offload:
	Number of bytes after which Kore will offload the HTTP body to
	disk instead of retaining it in memory. If 0 this feature is
	disabled. (Default: 0)

- http_body_disk_path:
	The path where Kore will store temporary HTTP body files.
	(this directory does not get created if http_body_disk_offload is 0).

New example:

The upload example has been added, demonstrating how to deal with file
uploads from a multipart form.
2016-01-18 11:30:22 +01:00
Joris Vink e580e6678e Simplify checking filename component. 2016-01-08 21:19:37 +01:00
Joris Vink ce874f09ea Cut memory usage for incoming HTTP bodies in half.
Kore pre-allocates a kore_buf for the full size of the
incoming HTTP body ... but also was passing the full
size to the net_recv_reset() function.

Instead of this, properly read smaller chunks from the
network and append them to the body buffer as they roll in.
2016-01-08 20:56:21 +01:00
Joris Vink 0647901ef5 Improve http_body_max directive a bit.
Allow setting it to 0 which will disable HTTP requests
that have a body (POST/PUT).

Reduce default http_body_max to 1MB by default, 10MB seems large.

Revisit to this code inspired by #100.
2016-01-08 17:54:40 +01:00
Joris Vink 0c47574fe9 If a body is too large send a 413 instead of 411. 2016-01-08 17:47:15 +01:00
Joris Vink f2d8834e8e After receiving a body, prime the netbuf again.
Not doing this results in nothing being read anymore
from the connection after a POST with a content-length of > 0.

Found by Xuning Niu.
2016-01-07 10:40:33 +01:00
Stig Telfer 18d3cc032d rename *_fini to *_cleanup 2016-01-04 21:40:14 +00:00
Joris Vink c4b1206ae3 Bump copyright to 2016. 2016-01-04 12:58:51 +01:00
Stig Telfer 0c51d9da53 Add resource management as part of the kore shutdown process. 2015-12-29 19:39:39 +00:00
Joris Vink 6ac15d37cb Make sure user agent is still set properly. 2015-12-10 09:25:22 +01:00
Joris Vink 4fd6d8a7a4 Correct usage of http_request_header().
Since latest change we no longer need free its result.
2015-11-29 17:22:30 +01:00
Joris Vink 7bdae240cf Change semantics for http_request_header().
The result returned by this function no longer needs to
be freed by the caller.
2015-11-29 14:19:44 +01:00
Joris Vink 769c78a6e8 Introduce NOHTTP=1 build option.
This basically turns off the HTTP layer for Kore. It does not
compile in anything for HTTP.

This allows Kore to be used as a network application platform as well.
Added an example for this called nohttp.

Other changes that sneaked in while hacking on this:
* Use calloc(), kill pendantic malloc option.
* Killed off SPDY/3.1 support completely, will be superseded by http2

Note that comes with massive changes to a lot of the core API
functions provided by Kore, these might break your application.
2015-11-27 16:22:50 +01:00
Joris Vink a20fb00789 req can be NULL if we are coming from an error path. 2015-08-02 16:53:40 +02:00
Joris Vink be3fa17201 Do not send connection:close for websocket upgrades. 2015-06-30 16:43:21 +02:00
Cleve Lendon 18becccd5a Fix indentation. 2015-06-30 18:07:45 +09:00
Cleve Lendon fef3ee8efe Fix websockets. Connection must be Upgrade, not keep-alive, Upgrade 2015-06-26 12:14:01 +09:00
Joris Vink a97e18bbe9 Shuffle some HTTP_REQUEST_NO_CONTENT_LENGTH around 2015-05-19 09:13:29 +02:00
Joris Vink e33710b26a Wrap expression with some additional parenthesis. 2015-05-19 09:04:46 +02:00
Quentin Perez ff4faa98b7 fix-null-dereference
http.c:1214, req->method
http.c:1272, req->flags
2015-05-18 23:42:50 +02:00
Joris Vink 1dd7407a41 Close SPDY streams if len == 0 and SPDY_NO_CLOSE is not set. 2015-05-15 19:16:35 +02:00
Joris Vink cefeaf7992 HTTP layer improvements.
Add HTTP_REQUEST_NO_CONTENT_LENGTH which can be set by
a handler before calling http_response() to avoid Kore
from setting the content-length altogether.

If we are on a SPDY connection do not close the stream
if we do not pass data to http_response().
2015-05-15 19:12:18 +02:00
Joris Vink 09ed61187b Unfuck content-length changes.
We want a content-length unless its 204 or status < 200.
2015-05-13 09:16:06 +02:00
Joris Vink 7ce742f131 Merge pull request #48 from thorduri/content-length
Always write out content-length;
2015-05-13 09:02:58 +02:00
Thordur Bjornsson a62bce3e88 Always write out content-length;
Unless status is 204 or 1xx (informational).
2015-05-12 20:23:47 +02:00
Joris Vink 4f36b534aa Reset fsm_state back to 0 after we're done with it. 2015-05-08 21:19:20 +02:00
Joris Vink e5f6d37cd2 Make parameter validation less noisy. 2015-04-27 15:35:02 +02:00
Joris Vink dc6e6fb1b4 Kill NETBUF_CALL_CB_ALWAYS for incoming HTTP body
Do not let the NETBUF_CALL_CB_ALWAYS flag linger in
the netbuf we reset for the incoming HTTP body.
2015-04-21 11:30:59 +02:00
Joris Vink dfea3cf483 Place path and host in http_request into pools.
Also change the amount of times we end of doing unneccessairy strlen() calls.
2015-04-15 12:40:15 +02:00