Luigi Rizzo [Mon, 18 Dec 2006 16:24:44 +0000 (16:24 +0000)]
apply the proposed fix for bug 8602
http://bugs.digium.com/view.php?id=8602
(i am not sure if there is still missing cast in
front of the alloca() call - being a macro this is
probably triggered only when actually used).
Add function ast_str_reset() to reinitialize the
string to an empty string instead of playing with
the internal fields.
Luigi Rizzo [Mon, 18 Dec 2006 11:28:37 +0000 (11:28 +0000)]
Replace ast_build_string with ast_str_*().
On passing remove presumably duplicate code to generate
the message for the manager_hooks:
in the previous version, the message was almost the same as the one sent
to regular sessions, with the exception of the empty line at the end, and
a few (presumably unintentional) differences e.g. timestamps,
debugging, and lowercase headers for "event" and "privilege".
Luigi Rizzo [Sun, 17 Dec 2006 22:57:46 +0000 (22:57 +0000)]
replace ast_build_string() with ast_str_*().
Unless i am very mistaken, function_realtime_read() was
broken in that it would always return an empty string
(because ast_build_string() advanced the pointer to the
end of the string, and there was no reference to the
initial value.
simplify dependency tracking system, using the compiler's built-in method for generating them, and only doing dependency tracking if developer mode is enabled via the configure script
since we really, really have to have autoconfig.h included before all other headers (especially system headers), the Makefile will now force it to happen (this will fix build problems with files like ast_expr2f.c, where we can't control the inclusion order in the file itself)
instead of initializing the curl library every time the CURL() function is invoked, do it only once per thread (this allows multiple calls to CURL() in the dialplan for a channel to run much more quickly, and also to re-use connections to the server) (thanks to JerJer for frequently complaining about this performance problem)
Luigi Rizzo [Fri, 15 Dec 2006 22:08:46 +0000 (22:08 +0000)]
move the dynamic string support in a better place i.e. string.h
While doing this, add a bit of documentation, and slightly
extend the functionality as follows:
+ a max_len of -1 means that we take whatever the current size
is, and never try to extend the buffer;
+ add support for alloca()-ted dynamic strings, which is very
useful for all cases where we do an ast_build_string() now.
Next step is to simplify the interface by using shorter names
(e.g. ast_str as a prefix) and removing the _thread variant
of the functions by saving the threadstorage reference into
the struct ast_str. This can be done by overloading the
'type' field.
Finally, I will do my best to remove the convoluted interface
that results from trying to support platforms without va_copy().
Turn payload_lock into bridge_lock and make it encompass all RTP structure contents that may relate to bridge information, including who we are bridged to.
Payload values on the RTP structure can change AFTER a bridge has started. This comes from the packet handling of the SIP response when indication that it was answered has been sent. Therefore we need to protect this data with a lock when we read/write. (issue #8232 reported by tgrman)
Remove direct RTCP bridging. I've come to the conclusion that we should handle this through the core and not just forward it on. Should solve a few bugs.
Luigi Rizzo [Wed, 13 Dec 2006 22:54:41 +0000 (22:54 +0000)]
clean up function manager_show_dialplan_helper()
reducing indentation and normalizing loops.
While doing this, remove some unused variables,
fix an uninitialized string (idaction), and mark
some places where the behaviour is not what we would expect
(e.g. an empty context is reported as an error same as
a non-existent one). Given that this function is
not in 1.4, the above can be changed without too many
backward compatibility concerns.
Russell Bryant [Tue, 12 Dec 2006 22:27:22 +0000 (22:27 +0000)]
Make chan_zap inform you that your version of zaptel is too old instead of
just failing to compile.
It seems like the proper way to do this would be in the configure script.
However, that wouldn't help existing checkouts unless we forced the configure
script to be executed after any code was changed.
app_externalivr needs a real silence file, and additional
changes to add silence files into core instead of extra
patch provided by bug 8177 with minor additions.
Return non-existant callerid handling to that which it was before. In 1.4 and trunk callerid can be allocated but not have any contents so we have to use ast_strlen_zero before passing it to the relevant functions. (issue #8567 reported by pabelanger)
Luigi Rizzo [Mon, 11 Dec 2006 18:11:58 +0000 (18:11 +0000)]
make sure the argument to ast_malloc() is > 0.
Long explaination:
The behaviour of the underlying malloc(0) differs depending on the
operating system. Some return NULL (SysV behaviour); some still
allocate a small chunk of memory and return a valid pointer (e.g.
traditional BSD); some (e.g. FreeBSD 6.x) return a non-null pointer
that causes a memory fault if used, even just for reading.
Given the above variety, better never call malloc(0).
When doing a fork() and exec(), two problems existed (Issue 8086):
1) Ignored signals stayed ignored after the exec().
2) Signals could possibly fire between the fork() and exec(), causing Asterisk
signal handlers within the child to execute, which caused nasty race conditions.
Russell Bryant [Sat, 9 Dec 2006 16:04:06 +0000 (16:04 +0000)]
Merged revisions 48363 via svnmerge from
https://origsvn.digium.com/svn/asterisk/branches/1.4
........
r48363 | russell | 2006-12-09 10:59:42 -0500 (Sat, 09 Dec 2006) | 8 lines
Use locking when accessing the registrations list. This list is not actually
used very often, so the likelihood of there being a problem is pretty small,
but still possible. For example, if the CLI command to list the registrations
was called at the same time that a reload was occurring and the registrations
list was getting destroyed and rebuilt, a crash could occur.
In passing, go ahead and convert this list to use the linked list macros.
Russell Bryant [Sat, 9 Dec 2006 07:10:55 +0000 (07:10 +0000)]
chan_iax2 has an extremely large function, socket_process(), to handle incoming
frames. The function, before this commit, was roughly 1400 lines long. So, I
am working on breaking this up into functions so that the code is easier to
follow and debug. Also, I will be committing these changes in chunks as I do
them to ease tracking down any potentially introduced problems.
Break out roughly 150 lines from socket_process() and introduce a new function,
socket_process_meta() which handles the parsing of an incoming meta frame.
Also, restructure some of this code a bit to reduce the deep nesting that was
in this code.
Russell Bryant [Sat, 9 Dec 2006 05:04:38 +0000 (05:04 +0000)]
- Fix a few spelling mistakes
- Use sizeof() to pass an array size to a function
- Use a single bit for a variable in the chan_iax2_pvt stuct since that is all
it needs.
- Add some comments about the iaxs, iaxl, and lastused arrays.
Luigi Rizzo [Thu, 7 Dec 2006 16:42:29 +0000 (16:42 +0000)]
- Generalize the function ssl_setup() so that the certificate info
are passed as an argument.
- Update the code in main/http.c to use the new interface
(the diff is large but mostly mechanical, due to the name change of
several variables);
- And since now it is trivial, implement "AMI over TLS", and document
the possible options in manager.conf
- And since the test client (openssl s_client -connect host:port )
does not generate \r\n as a line terminator, make get_input()
also accept just a \n as a line terminator (Mac users: do you
also need the \r-only version ?)
The option parsing in manager.conf is not very efficient, and needs
to be cleaned up and made similar to what we have in http.conf
Doxygen update
- Document cause codes
- Document a bit more on channel variables - global, predefined and local
- Fix some doxygen in channel.h. Adding one comment for two definitions does not
work. They won't be copied to each.
................
r47995 | murf | 2006-11-24 10:40:49 -0700 (Fri, 24 Nov 2006) | 1 line
This fix inspired by a patch supplied in bug 8189, which points out problems with the PLC code
................
r47997 | murf | 2006-11-24 11:17:25 -0700 (Fri, 24 Nov 2006) | 1 line
removed the svnmerge-integrated property from trunk; it's confusing svnmerge in newly created branches
................
r48001 | rizzo | 2006-11-25 02:02:42 -0700 (Sat, 25 Nov 2006) | 5 lines
set pointers to NULL after freeing memory to avoid multiple free()
probably 1.4/1.2 issue as well if someone can look into that.
- Adding comment on suspicious memory allocation. Seems like it's never freed, but I don't
have a clear understanding of the frame allocation/deallocation, so I just mark this
for investigation. (Reported by Ed Guy). We're trying to see if a free() hurts...
- Doxygen comments on p2p rtp bridge stuff. I am a bit worried about shortcutting
rtcp this way, but will need feedback from rtcp gurus. This should work for
video calls too, and possibly UDPTL.
generalize a bit the functions used to create an tcp socket
and then run a service on it.
The code in manager.c does essentially the same things,
so we will be able to reuse the code in here (probably
moving it to netsock.c or another appropriate library file).
................
r48009 | mattf | 2006-11-25 13:30:04 -0700 (Sat, 25 Nov 2006) | 1 line
Updates to show linkset command
................
r48010 | mattf | 2006-11-25 13:54:38 -0700 (Sat, 25 Nov 2006) | 2 lines
Add ss7 show linkset command
................
r48011 | mattf | 2006-11-25 14:32:33 -0700 (Sat, 25 Nov 2006) | 1 line
Make sure we don't send a group reset on a group larger than 32 CICs
................
r48012 | mattf | 2006-11-25 14:35:23 -0700 (Sat, 25 Nov 2006) | 1 line
bug fix
................
r48013 | mattf | 2006-11-25 14:46:58 -0700 (Sat, 25 Nov 2006) | 1 line
Make compiler happier
................
r48014 | mattf | 2006-11-25 14:50:42 -0700 (Sat, 25 Nov 2006) | 1 line
Little fix so we use the right message
................
r48016 | murf | 2006-11-25 17:15:42 -0700 (Sat, 25 Nov 2006) | 9 lines
Merged revisions 48015 via svnmerge from
https://origsvn.digium.com/svn/asterisk/branches/1.4
........
r48015 | murf | 2006-11-25 17:01:34 -0700 (Sat, 25 Nov 2006) | 1 line
A little bit of func_cdr documentation upgrade-- no bug# involved, although 8221 may have inspired it.
........
Merged revisions 48017 via svnmerge from
https://origsvn.digium.com/svn/asterisk/branches/1.4
........
r48017 | murf | 2006-11-25 17:26:16 -0700 (Sat, 25 Nov 2006) | 1 line
might as well also document the raw values of the flag vars
........
................
r48019 | russell | 2006-11-25 23:55:33 -0700 (Sat, 25 Nov 2006) | 6 lines
- Add some comments on thread storage with a brief explanation of what it is
as well as what the motivation is for using it.
- Add a comment by the declaration of ast_inet_ntoa() noting that this function
is not reentrant, and the result of a previous call to the function is no
longer valid after calling it again.
Luigi Rizzo [Wed, 6 Dec 2006 20:46:01 +0000 (20:46 +0000)]
remove duplicated code to start the server threads, use
the infrastructure exposed in http.c earlier today.
As a bonus, now we can restart the session on a different
port just reloading the module.
On passing, fix a bug in the handling of 'enabled' in the configuration
file - previously, a missing "enabled=" line in manager.conf meant
"whatever the state was before" instead of a specific value.
Luigi Rizzo [Wed, 6 Dec 2006 18:45:19 +0000 (18:45 +0000)]
Part of the transformations necessary to add TLS support,
as described in
http://lists.digium.com/pipermail/asterisk-dev/2006-December/025213.html
In detail, this commit does the following:
b) change the function get_input() to use fread() instead of read()
to collect the data. One can still do the ast_wait_for_input() on
the original descriptor returned by accept().
c) change the function send_string() to work on the FILE *.
As a side effect, this change now really guarantees that
we don't spend more than "writetimeout" milliseconds on
each line sent.
d) modify the function action_command() so that it creates a
temporary file descriptor to be passed to ast_cli_command(),
and then read back the data from the temp file and write it
to the output with send_string(). The code is similar to
what is done in generic_http_callback() to support AMI-over-HTTP.
Luigi Rizzo [Wed, 6 Dec 2006 16:17:57 +0000 (16:17 +0000)]
Make externally visible some generic code useful to create
and implement services over tcp and/or tcp-tls.
This commit is nothing more than moving structure definitions
(and documentation) from main/http.c to include/asterisk/http.h
(temporary location until we find a better place), and removing the
'static' qualifier from server_root() and server_start().
The name change (adding the ast_ prefix as a minimum, and then
possibly a more meaningful name) is postponed to future commits.
Jason Parker [Tue, 5 Dec 2006 20:15:37 +0000 (20:15 +0000)]
Expand on r48273 (from issue 8506), to translate more of the fskmodem stuff to English.
r48273 dealt with the comments and such, this deals with the code itself.
(This couldn't have been easily done if it weren't for 48273 - thanks again for that merbanan)
Olle Johansson [Tue, 5 Dec 2006 10:52:53 +0000 (10:52 +0000)]
Reserving flags for coming code (currently in the "videocaps" branch)
implementing T.140 support in RTP.
T.140/RFC 4351 is TDD over IP - text telephony for hearing impaired.
It defines a realtime text chat, much like the old "talk" application
in Unix.
T.140 is character by character in real time. It's not
the same as our current MESSAGE format - that is more like IM, but
can be gatewayed to MESSAGE with a text "codec" if needed.
More patches will follow, as soon as we've separated this code from
the video capabilities functions in the videocaps branch.
If the recording in the database is too large, it will fail to retrieve with
an mmap error. Not too sure why this doesn't happen when we put it in the
database, also, but since that doesn't seem to be broken, I'm not going to fix
it (at least until someone reports it). Solution is to ask for the file in
smaller chunks. (Bug 8385)
If the generic bridge tells us not to retry, and we have a frame to spit out then break the bridge. Props to markit in #asterisk-bugs for bringing this up.
projects / thirdparty / asterisk.git / log
