Open Radar

 
Area:
Something not on this list

Summary:
Hello,

please find attached a minimal test case that shows a deadlock when closing a kqueue file descritpor with close() from another thread.

The original implementation was a periodic timer that did some periodic work and is stoppable while blocked in kevent() syscall.

Once the process is deadlocked it is interruptible and a debugger cannot attach. Running spindump reveals the following two callstacks:

  Thread 0x7617c            DispatchQueue 1           1000 samples (1-1000)     priority 31 (base 31)
  1000  start + 1 (libdyld.dylib + 21077) [0x7fffa67a9255]
    1000  main + 27 (main.c:61,14 in kqueue_test_case + 3819) [0x100000eeb]
      1000  run_test + 88 (main.c:54,5 in kqueue_test_case + 3768) [0x100000eb8]
        1000  close + 10 (libsystem_kernel.dylib + 108370) [0x7fffa68d8752]
         *1000  hndl_unix_scall64 + 22 (kernel + 671558) [0xffffff80002a3f46]
           *1000  unix_syscall64 + 586 (kernel + 6452202) [0xffffff80008273ea]
             *1000  close_nocancel + 266 (kernel + 5595690) [0xffffff800075622a]
               *1000  close_internal_locked + 563 (kernel + 5577891) [0xffffff8000751ca3]
                 *1000  fileproc_drain + 261 (kernel + 5578613) [0xffffff8000751f75]
                   *1000  ??? (kernel + 5813123) [0xffffff800078b383]
                     *1000  lck_mtx_sleep + 132 (kernel + 1049444) [0xffffff8000300364]
                       *1000  thread_block_reason + 222 (kernel + 1091230) [0xffffff800030a69e]
                         *1000  ??? (kernel + 1095803) [0xffffff800030b87b]
                           *1000  machine_switch_context + 206 (kernel + 2102494) [0xffffff80004014de]

  Thread 0x761d4            1000 samples (1-1000)     priority 31 (base 31)
  1000  thread_start + 13 (libsystem_pthread.dylib + 13189) [0x100117385]
    1000  _pthread_start + 286 (libsystem_pthread.dylib + 15260) [0x100117b9c]
      1000  _pthread_body + 180 (libsystem_pthread.dylib + 15440) [0x100117c50]
        1000  kevent + 10 (libsystem_kernel.dylib + 110122) [0x7fffa68d8e2a]
         *1000  ??? (kernel + 5638944) [0xffffff8000760b20]

Steps to Reproduce:
Compile and run the attached project on Sierra with Xcode 8. Start the test case.

Expected Results:
A never ending stream of TEST invocations without deadlocks.

Actual Results:
TEST * fd=3 . x
TEST * fd=3 . x
TEST * fd=3 <deadlocks here>

Version:
Darwin gjasny-mbp.local 16.4.0 Darwin Kernel Version 16.4.0: Thu Dec 22 22:53:21 PST 2016; root:xnu-3789.41.3~3/RELEASE_X86_64 x86_64

Xcode 8.2.1
Build version 8C1002

Notes:


Configuration:
This did not happen on El Capitan with Xcode 7

Attachments:
'kqueue_test_case.zip' was successfully uploaded.