[asterisk-bugs] [JIRA] (ASTERISK-30258) Dialing API: Cancel a running async thread, does not always cancel all calls

[Frederic LE FOLL (JIRA)]

    [ https://issues.asterisk.org/jira/browse/ASTERISK-30258?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=260406#comment-260406 ] 

Frederic LE FOLL commented on ASTERISK-30258:
---------------------------------------------

I propose to submit a patch that adds soft hangups in ast_dial_join() when called parties are not yet running the application. The solution seems to work but needs confirmation.

> Dialing API: Cancel a running async thread, does not always cancel all calls
> ----------------------------------------------------------------------------
>
>                 Key: ASTERISK-30258
>                 URL: https://issues.asterisk.org/jira/browse/ASTERISK-30258
>             Project: Asterisk
>          Issue Type: Bug
>      Security Level: None
>          Components: Core/Dial
>    Affects Versions: 18.8.0
>            Reporter: Frederic LE FOLL
>
> A race condition may occur in ast_dial_join() function (Dialing API: Cancel async thread), if caller cancels its call while a called party has just answered but has not yet started running its application.
> This was detected with Page() application, calling approx 20 called parties with auto-answer, and caller almost immediately canceling its call (e.g. within 10 to 50 milliseconds).
> Called party that has answered just before caller hangup:
> - has not yet the is_running_app flag up, thus ast_dial_join() does not do Soft Hangup,
> - but has left the waitfor loop, thus pthread_kill() does not have the expected effect.
> Observation: when this race condition occurs for Page application, then :
> - called parties (with one ast_dial_run() per called party) do not get cancelled, they start executing the application (ConfBridge for Page) and never get kicked out, because caller never got connected to ConfBridge,
> - caller Channel gets stuck in ast_dial_join() because pthread_join(thread, NULL) never ends, except if all called called devices hangup.
> This was seen with Asterisk 18.6 but it should be reproductible with other versions, since ast_dial_join() has not changed in recent versions.



--
This message was sent by Atlassian JIRA
(v6.2#6252)