Call je_dallocx with flags when needed

[duke8253]

Have been seeing some weird ASan outputs like this:

=================================================================
==21402==ERROR: AddressSanitizer: attempting free on address which was not malloc()-ed: 0x7f23e542d000 in thread T52 ([LOG_PREPROC 0]) 
    #0 0x668e70 in __interceptor_free (/opt/oath/trafficserver/9.1/bin/traffic_server+0x668e70)
    #1 0x7f2405d1f76c in ink_freelist_free ../../../../../../_scm/trafficserver9.1_asan/src/tscore/ink_queue.cc:281
    #2 0xb336f7 in LogBuffer::destroy(LogBuffer*&) ../../../../../../_scm/trafficserver9.1_asan/proxy/logging/LogBuffer.h:205
    #3 0xb336f7 in LogFile::preproc_and_try_delete(LogBuffer*) ../../../../../../_scm/trafficserver9.1_asan/proxy/logging/LogFile.cc:497
    #4 0xb5414e in LogBufferManager::preproc_buffers(LogBufferSink*) ../../../../../../_scm/trafficserver9.1_asan/proxy/logging/LogObject.cc:79
    #5 0xb5c27d in LogObject::preproc_buffers(int) ../../../../../../_scm/trafficserver9.1_asan/proxy/logging/LogObject.h:146
    #6 0xb5c27d in LogObjectManager::preproc_buffers(int) ../../../../../../_scm/trafficserver9.1_asan/proxy/logging/LogObject.cc:1084
    #7 0xafc42b in Log::preproc_thread_main(void*) ../../../../../../_scm/trafficserver9.1_asan/proxy/logging/Log.cc:1305
    #8 0xb0894c in LoggingPreprocContinuation::mainEvent(int, void*) ../../../../../../_scm/trafficserver9.1_asan/proxy/logging/Log.cc:266
    #9 0x107d687 in Continuation::handleEvent(int, void*) ../../../../../../_scm/trafficserver9.1_asan/iocore/eventsystem/I_Continuation.h:219
    #10 0x107d687 in Continuation::handleEvent(int, void*) ../../../../../../_scm/trafficserver9.1_asan/iocore/eventsystem/I_Continuation.h:215
    #11 0x107d687 in EThread::execute() ../../../../../../_scm/trafficserver9.1_asan/iocore/eventsystem/UnixEThread.cc:348
    #12 0x1078345 in spawn_thread_internal ../../../../../../_scm/trafficserver9.1_asan/iocore/eventsystem/Thread.cc:92
    #13 0x7f2404303ea4 in start_thread /usr/src/debug/glibc-2.17-c758a686/nptl/pthread_create.c:307
    #14 0x7f24034009fc in clone (/lib64/libc.so.6+0xfe9fc)

Address 0x7f23e542d000 is a wild pointer.
SUMMARY: AddressSanitizer: bad-free (/opt/oath/trafficserver/9.1/bin/traffic_server+0x668e70) in __interceptor_free
Thread T52 ([LOG_PREPROC 0]) created by T0 here:
    #0 0x5ccd0f in pthread_create (/opt/oath/trafficserver/9.1/bin/traffic_server+0x5ccd0f)
    #1 0x10794be in ink_thread_create /sd/workspace/src/git.vzbuilders.com/Edge/build/_build/build_release_posix-x86_64_gcc_8/trafficserver9.1_asan/build/../../../../_scm/trafficserver9.1_asan/include/tscore/ink_thread.h:159
    #2 0x10794be in Thread::start(char const*, void*, unsigned long, std::function<void ()> const&) ../../../../../../_scm/trafficserver9.1_asan/iocore/eventsystem/Thread.cc:109
    #3 0x10879cb in EventProcessor::spawn_thread(Continuation*, char const*, unsigned long) ../../../../../../_scm/trafficserver9.1_asan/iocore/eventsystem/UnixEventProcessor.cc:497
    #4 0xafe0b3 in Log::create_threads() ../../../../../../_scm/trafficserver9.1_asan/proxy/logging/Log.cc:1110
    #5 0xafe61b in Log::init_when_enabled() ../../../../../../_scm/trafficserver9.1_asan/proxy/logging/Log.cc:1082
    #6 0x54eb0e in main ../../../../../_scm/trafficserver9.1_asan/src/traffic_server/traffic_server.cc:2186
    #7 0x7f2403324554 in __libc_start_main ../csu/libc-start.c:266

==21402==ABORTING
=================================================================

This fixes the problem.

[bneradt]

I think this makes sense. To summarize: just as we do some special stuff with jemalloc with allocate, so we have to do some specific de-allocation logic, right?

[duke8253]

I think this makes sense. To summarize: just as we do some special stuff with jemalloc with allocate, so we have to do some specific de-allocation logic, right?

More or less, I must have thought the default free from jemalloc was fine for the flags we were using, since the arena_id part isn't used during the free process, so I removed this part during a cleanup, but apparently I was wrong about that.

[bneradt]

@SolidWallOfCode can give this a review after a quick change to the file we discussed in the PR issue (just pass the flag value to dallocx rather than create a variable).

[bneradt]

Adding this to 9.2.x since this is a bug fix that we should get there.

[zwoop]

Cherry-picked to v9.2.x