Segfault using TAO SSL connection with dev poll reactor

[pierrebizz]

Version

The version of ACE and/or TAO you are using
ace[core,ssl,tao]:x64-linux -> 7.0.11#2

compiled using vcpkg

Host machine and operating system

NAME="Red Hat Enterprise Linux"
VERSION="8.7 (Ootpa)"
ID="rhel"
ID_LIKE="fedora"
VERSION_ID="8.7"
PLATFORM_ID="platform:el8"
PRETTY_NAME="Red Hat Enterprise Linux 8.7 (Ootpa)"

Compiler name and version (including patch level)

gcc (GCC) 8.5.0 20210514 (Red Hat 8.5.0-16)

Description

When connecting using SSL, the second connection crashes with the following stack trace:

Linking with the debug version of ACE/TAO does not crash; I'm posting here not necessarily to obtain a solution, but just to see if someone had the problem before and if someone has found a solution for it.

Thank you,

Pierre Bisaillon

(gdb) bt
#0  operator>= (tv2=..., tv1=...)
    at /home/ec2-user/vcpkg/buildtrees/ace/src/ACE-TAO-7-453b9ce9a9.clean/ace/Time_Value.inl:123
#1  ACE_Countdown_Time_T<ACE_FPointer_Time_Policy>::stop (this=0x7fa53c24ca40)
    at /home/ec2-user/vcpkg/buildtrees/ace/src/ACE-TAO-7-453b9ce9a9.clean/ace/Countdown_Time_T.cpp:48
#2  ACE_Countdown_Time_T<ACE_FPointer_Time_Policy>::~ACE_Countdown_Time_T (this=0x7fa53c24ca40,
    __in_chrg=<optimized out>)
    at /home/ec2-user/vcpkg/buildtrees/ace/src/ACE-TAO-7-453b9ce9a9.clean/ace/Countdown_Time_T.cpp:26
#3  ACE_SSL_SOCK_Acceptor::accept (this=this@entry=0x5c29fe8, new_stream=..., remote_addr=remote_addr@entry=0x0,
    timeout=timeout@entry=0x7fa53c24cad0, restart=restart@entry=true, reset_new_handle=<optimized out>)
    at /home/ec2-user/vcpkg/buildtrees/ace/src/ACE-TAO-7-453b9ce9a9.clean/ace/SSL/SSL_SOCK_Acceptor.cpp:177
#4  0x00000000013cc54a in TAO::SSLIOP::Accept_Strategy::accept_svc_handler (this=0x5c29fe0, svc_handler=0x7fa4e401d0e0)
    at /home/ec2-user/vcpkg/buildtrees/ace/src/ACE-TAO-7-453b9ce9a9.clean/ace/Svc_Handler.cpp:245
#5  0x00000000013aff59 in ACE_Acceptor<TAO::SSLIOP::Connection_Handler, ACE_SSL_SOCK_Acceptor>::handle_input (
    this=0x5c46308, listener=8)
    at /home/ec2-user/vcpkg/buildtrees/ace/src/ACE-TAO-7-453b9ce9a9.clean/ace/OS_Errno.inl:19
#6  0x000000000133f508 in ACE_Dev_Poll_Reactor::upcall (this=<optimized out>, handle=<optimized out>,
    callback=<optimized out>, event_handler=<optimized out>)
    at /home/ec2-user/vcpkg/buildtrees/ace/src/ACE-TAO-7-453b9ce9a9.clean/ace/Dev_Poll_Reactor.inl:84
#7  ACE_Dev_Poll_Reactor::dispatch_io_event (this=0x5c19df0, guard=...)
    at /home/ec2-user/vcpkg/buildtrees/ace/src/ACE-TAO-7-453b9ce9a9.clean/ace/Dev_Poll_Reactor.cpp:1298
#8  0x00000000013408f9 in ACE_Dev_Poll_Reactor::handle_events (this=0x5c19df0, max_wait_time=0x0)
    at /home/ec2-user/vcpkg/buildtrees/ace/src/ACE-TAO-7-453b9ce9a9.clean/ace/Dev_Poll_Reactor.cpp:1007
#9  0x000000000128d3bd in ACE_Reactor::handle_events (max_wait_time=0x0, this=0x5c2bda0)
    at /home/ec2-user/vcpkg/buildtrees/ace/src/ACE-TAO-7-453b9ce9a9.clean/ace/Reactor.inl:120
#10 TAO_ORB_Core::run (this=0x5c18490, tv=0x0, perform_work=0)
    at /home/ec2-user/vcpkg/buildtrees/ace/src/ACE-TAO-7-453b9ce9a9.clean/TAO/tao/ORB_Core.cpp:2314
#11 0x00000000005579d7 in TextmlORBTask::svc() ()
    at /home/ec2-user/vcpkg/buildtrees/ace/src/ACE-TAO-7-453b9ce9a9.clean/ace/Time_Value.inl:61
#12 0x0000000000b88aad in ACE_Task_Base::svc_run (args=0x7ffed8702200)

[jwillemsen]

I see that you are using the ACE_Dev_Poll_Reactor, that reactor is still experimental for TAO, see http://bugzilla.dre.vanderbilt.edu for issues that still exist. Have you tried to use the default reactor?

[pierrebizz]

We tried; our server opens many files, so we had to change the FD_SETSIZE so that the select of the default reactor would work; but using vckpg to get ACE, we ran into the problem of having the same FD_SETSIZE all around.

[jwillemsen]

That is the big disadvantage of ACE/TAO through vcpkg or any other pre packaging, you can't tweak ACE/TAO to your needs, something I always do for each project. Can you build ACE/TAO yourself and retest with the default reactor?