[asterisk-dev] [Code Review] 4594: Asterisk manager output escape message control characters
warren smith
reviewboard at asterisk.org
Wed Apr 8 14:11:19 CDT 2015
> On April 8, 2015, 5:05 p.m., Mark Michelson wrote:
> > I have a few problems with this review:
> >
> > * I don't see in this review or in the patch you uploaded to the linked issue where ast_escape_c() or astman_escape_output() are defined.
> > * This patch performs a lot of heap allocations and does not check for potential memory allocation failure.
> > * This patch places the responsibility of escaping strings on all callers of ast_manager_event(). If escaping strings is an essential part of sending manager events, then it should be built into the function that sends manager events. Otherwise, it's easy for a programmer to forget to perform the escaping before outputting strings. If escaping of manager events is in the same location where manager events are sent, you may also be able to get away with fewer heap allocations, too, which is another bonus.
The patch was missing a bunch of the code from line ending problems, it should all be there now.
I very much agree with your third point. My initial attempt was to add a format string parameter for %M to escape manager output, such as:
manager_event(SOME_EVENT_FLAG, "EventName",
"KeyOne: %M\r\nKeyTwo: %s\r\n", val1, val2);
Which takes care of the allocation and everything. It works well, however you can't tell gcc about added format params at compile time (that I know of at least), and since -Wformat is used anytime the %M shows up it throws a warning, which makes it unusable.
What do you think about changing the manager event handing so that it always escapes %s (using the formatter like %M), and then having something like %S for an unescaped string? This would leave all the changes inside of manager and the only applications that would have to be changed are the very few that add dynamic event keys like:
manager_event(SOME_EVENT_FLAG, "EventName",
"KeyOne: %M\r\nKeyTwo: %s\r\n%s", val1, val2, eventstring);
I would update the documentation for the manager_event functions to let people know %s is escaped, and %S is unescaped - but it would make it so that it would be done correctly by default. It shouldn't throw the compile time warnings either since it would be done in the va_start() / va_end() section of main/manager.c's __ast_manager_event_multichan(), however if someone were to use an unescaped string of %S it would still throw a warning...
- warren
-----------------------------------------------------------
This is an automatically generated e-mail. To reply, visit:
https://reviewboard.asterisk.org/r/4594/#review15129
-----------------------------------------------------------
On April 8, 2015, 5:51 p.m., warren smith wrote:
>
> -----------------------------------------------------------
> This is an automatically generated e-mail. To reply, visit:
> https://reviewboard.asterisk.org/r/4594/
> -----------------------------------------------------------
>
> (Updated April 8, 2015, 5:51 p.m.)
>
>
> Review request for Asterisk Developers.
>
>
> Bugs: asterisk-24934
> https://issues.asterisk.org/jira/browse/asterisk-24934
>
>
> Repository: Asterisk
>
>
> Description
> -------
>
> Asterisk manager output is created using printf formatting, like:
>
> manager_event(SOME_EVENT_FLAG, "EventName",
> "KeyOne: %s\r\nKeyTwo: %s\r\n", val1, val2);
>
> This causes problems when the values themselves contain control characters like carriage return and newline, so that applications parsing the output will interpret this as a new key, or the end of an event. An example of this is having a callerid contain "\r\n\r\n". This ends the event, and the keys for the same event are interpreted as a new message, and any keys below are missed for the real event.
>
> I've included a patch that provides a ast_escape_c() function which takes a string, then returns a pointer to a new string that has the c characters escaped (i.e., newline into \n). I've modified the calls to the manager_event functions (manager_event, ast_manager_event, ast_manager_event_multichan) so that values that could be set by a user are escaped. The string values that as far as I know aren't user-created were left as-is, like channel names and uniqueid.
>
> There are quite a few calls to the manager event functions and I've double checked to make sure all memory allocations are freed after creating the escaped string. I also had added an ast_replace_string function which i didn't end up using, and added an ast_escape_output function which just calls ast_escape_c. An alternative would be to replace the sequence "\r\n" with the escaped version, rather than the individual characters.
>
> I'm testing on our asterisk 11 install and this fixes the parsing bugs we run into from messed up callerids and things like agent names containing return + newline.
>
>
> Diffs
> -----
>
> http://svn.asterisk.org/svn/asterisk/branches/11/res/res_xmpp.c 433419
> http://svn.asterisk.org/svn/asterisk/branches/11/res/res_musiconhold.c 433419
> http://svn.asterisk.org/svn/asterisk/branches/11/res/res_fax.c 433419
> http://svn.asterisk.org/svn/asterisk/branches/11/res/res_agi.c 433419
> http://svn.asterisk.org/svn/asterisk/branches/11/pbx/pbx_realtime.c 433419
> http://svn.asterisk.org/svn/asterisk/branches/11/main/utils.c 433419
> http://svn.asterisk.org/svn/asterisk/branches/11/main/test.c 433419
> http://svn.asterisk.org/svn/asterisk/branches/11/main/pbx.c 433419
> http://svn.asterisk.org/svn/asterisk/branches/11/main/manager.c 433419
> http://svn.asterisk.org/svn/asterisk/branches/11/main/features.c 433419
> http://svn.asterisk.org/svn/asterisk/branches/11/main/channel.c 433419
> http://svn.asterisk.org/svn/asterisk/branches/11/main/cdr.c 433419
> http://svn.asterisk.org/svn/asterisk/branches/11/include/asterisk/utils.h 433419
> http://svn.asterisk.org/svn/asterisk/branches/11/include/asterisk/strings.h 433419
> http://svn.asterisk.org/svn/asterisk/branches/11/include/asterisk/presencestate.h 433419
> http://svn.asterisk.org/svn/asterisk/branches/11/include/asterisk/manager.h 433419
> http://svn.asterisk.org/svn/asterisk/branches/11/funcs/func_presencestate.c 433419
> http://svn.asterisk.org/svn/asterisk/branches/11/funcs/func_global.c 433419
> http://svn.asterisk.org/svn/asterisk/branches/11/channels/chan_sip.c 433419
> http://svn.asterisk.org/svn/asterisk/branches/11/channels/chan_local.c 433419
> http://svn.asterisk.org/svn/asterisk/branches/11/channels/chan_iax2.c 433419
> http://svn.asterisk.org/svn/asterisk/branches/11/channels/chan_agent.c 433419
> http://svn.asterisk.org/svn/asterisk/branches/11/cel/cel_manager.c 433419
> http://svn.asterisk.org/svn/asterisk/branches/11/cdr/cdr_manager.c 433419
> http://svn.asterisk.org/svn/asterisk/branches/11/apps/app_voicemail.c 433419
> http://svn.asterisk.org/svn/asterisk/branches/11/apps/app_userevent.c 433419
> http://svn.asterisk.org/svn/asterisk/branches/11/apps/app_stack.c 433419
> http://svn.asterisk.org/svn/asterisk/branches/11/apps/app_queue.c 433419
> http://svn.asterisk.org/svn/asterisk/branches/11/apps/app_meetme.c 433419
> http://svn.asterisk.org/svn/asterisk/branches/11/apps/app_dial.c 433419
> http://svn.asterisk.org/svn/asterisk/branches/11/apps/app_confbridge.c 433419
>
> Diff: https://reviewboard.asterisk.org/r/4594/diff/
>
>
> Testing
> -------
>
> In our production environment I have been running the patch for about 5 days. The parsing issues we have had in the past are now resolved.
>
>
> Thanks,
>
> warren smith
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.digium.com/pipermail/asterisk-dev/attachments/20150408/7bd4c4c9/attachment-0001.html>
More information about the asterisk-dev
mailing list