[asterisk-bugs] [JIRA] (ASTERISK-26974) res_pjsip: Deadlock in T.38 framehook

[Ross Beer (JIRA)]

    [ https://issues.asterisk.org/jira/browse/ASTERISK-26974?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=236824#comment-236824 ] 

Ross Beer edited comment on ASTERISK-26974 at 5/2/17 9:00 AM:
--------------------------------------------------------------

I have just had an out-of-memory issue while running Asterisk Asterisk GIT-13-13.15.0-rc1-121-gbf7cf10M Patched with 5561 + 5560.

Asterisk crashed but didn't generate a core dump, however, the following is shown the log:

{noformat}
May  2 14:14:50 host01 kernel: Out of memory: Kill process 24767 (asterisk) score 904 or sacrifice child
May  2 14:14:50 host1 kernel: Killed process 24767 (asterisk) total-vm:27741128kB, anon-rss:15694096kB, file-rss:0kB
{noformat}

Memory Debug shoes the memory growing:
{noformat}
memory show summary res_pjsip_t38.c
       160 bytes in          4 allocations by t38_attach_framehook() line   529 of res_pjsip_t38.c
       448 bytes in          4 allocations by t38_state_get_or_alloc() line   237 of res_pjsip_t38.c
       608 bytes allocated in 8 selected allocations

  22326454 bytes in all allocations
    216784 bytes in deferred free large allocations
     16719 bytes in deferred free small allocations
    233503 bytes in deferred free allocations
  22559957 bytes in all allocations and deferred free allocations
{noformat}


was (Author: rossbeer):
I have just had an out-of-memory issue while running Asterisk Asterisk GIT-13-13.15.0-rc1-121-gbf7cf10M Patched with 5561 + 5560.

I am currently running a MALLOC_DEBUG to confirm where the issue is. Asterisk crashed but didn't generate a core dump, however, the following is shown the log:

May  2 14:14:50 host01 kernel: Out of memory: Kill process 24767 (asterisk) score 904 or sacrifice child
May  2 14:14:50 host1 kernel: Killed process 24767 (asterisk) total-vm:27741128kB, anon-rss:15694096kB, file-rss:0kB


> res_pjsip: Deadlock in T.38 framehook
> -------------------------------------
>
>                 Key: ASTERISK-26974
>                 URL: https://issues.asterisk.org/jira/browse/ASTERISK-26974
>             Project: Asterisk
>          Issue Type: Bug
>      Security Level: None
>          Components: Resources/res_pjsip_t38
>    Affects Versions: 13.15.0
>            Reporter: Richard Mudgett
>            Assignee: Richard Mudgett
>         Attachments: backtrace-threads-PJSIP.txt
>
>
> A deadlock can happen while T.38 is active and a re-INVITE happens.  The thread processing the re-INVITE is trying to update the channel's format capabilities while the channel thread is trying to send a frame.  The thread processing the re-INVITE has the session's media container lock and wants the channel lock and the channel thread has the channel lock and wants the session's media container lock.



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