We should first shut down TLS so we can signal the other end a clean shutdown
in contrast to a socket failure.
Big thanks to wget(1) to find it. Fantastic tool.
This generates error pages based on IDs. This allows to reuse errors
and add more advanced information to them.
This patch also makes Icecast send in plain text OR HTML based
on the clients Accept:-string.
This allows ICY source clients to log in again with 2.4.x-style config.
ICY does not provide a username. Therefore the username was unset.
This patch sets the username for ICY source connections and allow them
to be handled as if a username was provided.
Configuration tag <shoutcast-user> was added to select the username to set
to. This defaults to "source".
ban and allow file support has been moved into a generic implementation
that can be re-used by later code such as proxy matching or other
blacklisting.
See: #1959
This tries to unbreak the 100-continue thing a bit.
Proper headers are send in case of 100-continue. Also the 200 OK
was moved to the end of the stream.
The client may even be reused. But I'm not sure we only do
'End of stream' in cases we are not on an undefined state of the
client. Do we as of now support anything that give us EOS without
a fatal socket state (error or shutdown)?
At least this unbreaks 100-continue with active TLS that was broken
as hell as we sent the header in clear over the socket. Not checking
if it was transmitted completly or something. Using lowlevel socket
calls (in contrast to the connection level calls that would have
handled the TLS just fine).
While this seem sto be a huge improvement to the situation I kindly
ask you to do a lot testing on this. Both with SOURCE as well as
PUT clients.
We have answered SOURCE/PUT requests with a simple
"HTTP/1.0 200 OK\r\n\r\n" while we should answer with complet
headers. This commit corrects this.
Interoperability:
This will work for two classes of clients:
* Super dumb ones not checking the response and will just send
data and hope the socket is still open.
* Those with a real parser.
* Shoutcast clients (indepnded code path).
* All libshout, curl, wget, ... based clients.
This *could* break:
* Clients that check for the exact returned message. However
as we start with the same string this should work.
Only client that check for the string including the End-of-header
CRLF sequense may break.
Please test careful before release.
This allows the usage of RFC 2817 style upgrade commands.
This is not correctly announced yet.
Can hardly be tested as most clients do not support this.
Will be helpful with TLS support in libshout.
See: #2159, #2152
This allows <resource> (former <alias>) to have only one of the
attribues source and destination. In that case other parameters
will be applied to the client but uri is not mapped.
See #2097
Old-style <authentication> within <mount> didn't work for type="url"
as well as some other parameters due to confusion between "node"
and "child" variable.
Thanks for trilliot for pointing out! Should work now.
closes#2039
Renamed event_config_read() into config_reread_config() and moved
it into cfgfile.c. This allowed to delete event.[ch].
event.[ch] will later be used to implement <event>.
This patch *replaces* the authentication system completly.
What is new:
- <authentication> in mount section is now a container object.
- <authentication> in root and mount section may hold any number of <role>-Tags.
- <role> tags:
Those tags define a 'role' and it's ACL rules.
A role is a instance of an authentication module (see below).
<role> takes the following options. All but type are optional.
- authentication related:
- type: Type of the authentication module (values: anonymous, static, legacy-password, url or htpasswd;
symbolic constants in auth.h)
- name: Name for the role. For later matching. (values: any string; default: (none))
- method: This rule is only active on the given list of HTTP methods.
(list of enum values: methods as recognized by httpp/ (e.g: get,post); default: *)
- ACL related:
- allow-method: Allowed HTTP methods.
(list of enum values: methods as recognized by httpp/ (e.g: get,post); default: get)
- deny-method: Rejected HTTP methods.
(list of enum values: methods as recognized by httpp/ (e.g: get,post); default: *)
- allow-admin: Allowed admin commands. (list of enum values: admin command; default: buildm3u)
- deny-admin: Rejected admin commands. (list of enum values: admin command; default: *)
- allow-web: Allowed web pages. (values: empty or *; default: *)
- deny-web: Rejected web pages. (values: empty or *; default: (empty))
- connections-per-user: maximum number of simultaneous connections per role and username.
This is only active on active sources. (values: unlimited or number of connections; default: unlimited)
- connection-duration: maximum time of a connection. This is only active on active sources.
(values: unlimited or number of secounds; default: unlimited)
<role> takes <option> child tags. <option> tags contain a name and a value option.
Meaning of <option> tags is up to the authentication module.
- <role>s are considered to build a stack. If a role returns with AUTH_NOMATCH the next one will be tried.
- <role>s are tested in this order: mount specific, default mount specific, global, internal fallback.
Internal fallback is set to allow web/ access via GET, POST and HEAD (only GET supported by this time)
and rejects all other requests.
- New authentication module: anonymous
This module matches all requests. No options taken.
- New authentication module: static
This module matches with a static username and password.
It takes two <option>s. One with name="username" and one with name="password" to set username and password.
This replaces old style <*-username> and <*-password> tags.
- New authentication module: legacy-password
This module matches with a statich password.
It takes one <option> with name="password" to set password.
This replaces old ICE and ICY (shoutcast compat mode) authentication.
- Parsing <authentication> in <mount> with a type set in a special way to allow 100% backward compatibility.
- Parsing of <source-password>, <admin-password>, <admin-user>, <relay-password> and <relay-user> in global
<authentication> for 100% backward compatibility.
- <alias> is now proccessed very early. This enables them to be used for all kinds of requests.
To Do List & What does not yet work:
- type="url" auth: mount_add and mount_remove.
This should be replaced by an unique feature I would call '<event>'.
- Admin commands manageauth and manageauth.xsl are disabled as they need more review:
This code needs to be ported to support multiple <role>s per <mount>.
- url authentication module can not yet return AUTH_NOMATCH.
This needs to be reviewed and discussed on how to handle this case best way.
- Default config files needs to be updated to reflect the changes.
As this is quite some political act it should be done in dicussion with the whole team
and permission of the release manager.
- Docs need to be updated to reflect the changes.
How does it work:
Code has been changed so that authentification is done early for all clients.
This allows accessing the ACL data (client->acl) from nearly everywhere in the code.
After accept() and initial client setup the request is parsed. In the next step
all <alias>es are resolved. After this the client is passed for authentication.
After authentication it is passed to the corresponding subsystem depending on kind of request.
All authentication instances have a thread running for doing the authentication.
This thread works on a queue of clients.
Hints for testers:
- Test with default config.
- Test with diffrent authentication modules in <mount>.
- Test shoutcast compatibility mode.
- Test with new style <authentication> and any amount of <role> (zero to quite some).
- Test <alias> lookup on all kinds of objects.
- Test source level credential login into the admin interface.
- Test shoucast style meta data updates.
- Test playlist generation.
Thank you for reading this long commit message. Have fun reading the full patch!
svn path=/icecast/trunk/icecast/; revision=19358
In case of SOURCE we are lenient and thus quite some source clients
don't send a proper content-type, especially if they only support mp3.
This was meant to be introduced in 2.4.0 already, sadly we missed it.
All source clients MUST send proper content-type after migrating to
Icecast HTTP PUT protocol.
closes#2082
svn path=/icecast/trunk/icecast/; revision=19288
Add source IP adress to startup and source exit logging, see #2016
Add mountpoint to some log lines, see #1388
svn path=/icecast/trunk/icecast/; revision=19250
The default mount is a block in the config file that contains settings for
all mount points that do not have a block in configfile themself.
This is implemented by a <mount type="default">-block.
In this case the <mount>-block MUST NOT contain a <mount-name>-subblock.
svn path=/icecast/trunk/icecast/; revision=18902
We are handling it the same as we would handle a SOURCE request.
Due to legacy code, sender MUST send proper content-type header,
if content type is not audio/mpeg!
Can be tested using real-time encoded output and piping it into
| curl -u username:password -H "Content-type: application/ogg" -T - http://localhost:8000/mountname.ogg
Note that this example has ZERO timing, so a simple 'cat *.ogg' will fail.
Whatever feeds the pipe must do it at proper timing for real-time playback!
svn path=/icecast/trunk/icecast/; revision=18888
specifying the TCP send buffer size, it seems that on at least some win32 systems,
the window size stays at 8k (even with registry settings) which could limit available
streaming bandwidth.
svn path=/icecast/trunk/icecast/; revision=15766
issue requests (using ithe stream_auth option) to allow external engines to
determine whether a client can stream or not. Admin requests using source auth
are able to use this mechanism however source clients using the icy protocol
cannot yet.
svn path=/icecast/trunk/icecast/; revision=15621
the overall time is very slow. This lag is mainly down to the sleep that
the connection and fserve threads do to prevent busy looping when idle. The
sleep has to expire even if requests are waiting.
This patch merges the tasks done by the connection and listener thread into
one thread. It allows us to remove some locking, reduce thread usage and
service new connections more quickly as the poll/select acts as the busy
CPU prevention.
svn path=/icecast/trunk/icecast/; revision=15123
provide a mountpoint so you specify a shoutcast-mount in the listen-socket
section. Specifying this will automatically define the port+1 with the same
settings however the original shoutcast-compat approach is maintained.
svn path=/icecast/trunk/icecast/; revision=14045
them. Each listener connection for each request is checked as connected before
performing the request (so that time isn't wasted on slow authentication). Various
name/comment cleanups as well.
svn path=/icecast/trunk/icecast/; revision=13583
need updating. Fixes a few unusual cases, if a mount was removed (with xml reload)
whilst no source was reserved then those stats were left behind.
svn path=/icecast/trunk/icecast/; revision=13538
we'll get queries about the access log. Fix a potential FD leak and add a
small delay into the listener thread if accept fails, the usual cause is an FD
limit being reached triggering logs to fill up and CPU to max out.
svn path=/icecast/trunk/icecast/; revision=13492
