Bug #17677: Ractor crashes fork when blocking

• Status: Assigned
• Assignee: ko1 (Koichi Sasada)
• Target version: -
• ruby -v: ruby 3.0.0p0 (2020-12-25 revision 95aff21468) [x86_64-linux]
• Backport: 2.5: UNKNOWN, 2.6: UNKNOWN, 2.7: UNKNOWN, 3.0: UNKNOWN
• Tags: ractor

[delner (David Elner)]

Background

If you create a Ractor which blocks (e.g. receive), then fork the process, the fork will segfault upon completion.

How to reproduce

r2 = Ractor.new do
 Ractor.receive
end
 
sleep(1)
 
puts "Forking..."
 
fork do
 sleep(1)
 puts "End fork."
end
 
loop do
 puts "Main thread."
 sleep(1)
end

Expectation and result

Application prints “Main thread” from main process every second, while fork prints “End fork.” then produces a segfault. Main process continues to run.

Expected fork to not raise a segfault.

<internal:ractor>:267: warning: Ractor is experimental, and the behavior may change in future versions of Ruby! Also there are many implementation issues.
Forking...
Main thread.
Main thread.
End fork.
app/sandbox.rb:80: [BUG]: Device or resource busy (EBUSY)
ruby 3.0.0p0 (2020-12-25 revision 95aff21468) [x86_64-linux]

-- Control frame information -----------------------------------------------
c:0003 p:---- s:0011 e:000010 CFUNC  :fork
c:0002 p:0048 s:0007 E:001b48 EVAL   app/sandbox.rb:80 [FINISH]
c:0001 p:0000 s:0003 E:002590 (none) [FINISH]

-- Ruby level backtrace information ----------------------------------------
app/sandbox.rb:80:in `<main>'
app/sandbox.rb:80:in `fork'

-- C level backtrace information -------------------------------------------
/usr/local/lib/libruby.so.3.0(rb_print_backtrace+0x11) [0x7f9848528cfb] vm_dump.c:758
/usr/local/lib/libruby.so.3.0(rb_vm_bugreport) vm_dump.c:998
/usr/local/lib/libruby.so.3.0(bug_report_end+0x0) [0x7f9848354808] error.c:763
/usr/local/lib/libruby.so.3.0(rb_bug_without_die) error.c:763
/usr/local/lib/libruby.so.3.0(die+0x0) [0x7f98482c6902] error.c:771
/usr/local/lib/libruby.so.3.0(rb_bug) error.c:773
/usr/local/lib/libruby.so.3.0(rb_bug_errno+0x3c) [0x7f9848354a1c] error.c:802
/usr/local/lib/libruby.so.3.0(rb_native_mutex_destroy+0x20) [0x7f98484ce380] thread_pthread.c:444
/usr/local/lib/libruby.so.3.0(rb_native_cond_initialize) (null):0
/usr/local/lib/libruby.so.3.0(ractor_free+0xd) [0x7f984844487d] ractor.c:229
/usr/local/lib/libruby.so.3.0(run_final+0xb) [0x7f9848371c66] gc.c:3670
/usr/local/lib/libruby.so.3.0(finalize_list) gc.c:3689
/usr/local/lib/libruby.so.3.0(rb_objspace_call_finalizer+0x33d) [0x7f984837cc5d] gc.c:3852
/usr/local/lib/libruby.so.3.0(rb_ec_cleanup+0x311) [0x7f984835f0b1] eval.c:184
/usr/local/lib/libruby.so.3.0(ruby_stop+0x9) [0x7f984835f339] eval.c:329
/usr/local/lib/libruby.so.3.0(rb_f_fork+0x1f) [0x7f98484402f8] process.c:4348
/usr/local/lib/libruby.so.3.0(rb_f_fork) process.c:4338
/usr/local/lib/libruby.so.3.0(vm_call_cfunc_with_frame+0x11b) [0x7f984850672b] vm_insnhelper.c:2898
/usr/local/lib/libruby.so.3.0(vm_call_method_each_type+0xf9) [0x7f98485192c9] vm_insnhelper.c:3388
/usr/local/lib/libruby.so.3.0(vm_call_method+0xb4) [0x7f9848519b24] vm_insnhelper.c:3506
/usr/local/lib/libruby.so.3.0(vm_sendish+0xb3) [0x7f984850a3d3] vm_insnhelper.c:4499
/usr/local/lib/libruby.so.3.0(vm_exec_core+0x140) [0x7f98485123e0] insns.def:770
/usr/local/lib/libruby.so.3.0(rb_vm_exec+0x176) [0x7f9848517b26] vm.c:2163
/usr/local/lib/libruby.so.3.0(rb_ec_exec_node+0xd9) [0x7f9848359719] eval.c:317
/usr/local/lib/libruby.so.3.0(ruby_run_node+0x55) [0x7f984835f395] eval.c:375
/usr/local/bin/ruby(main+0x5b) [0x565147e5410b] ./main.c:50

...
Main thread.
Main thread.
Main thread.

Additional notes

This does not happen if a blocking operation does not occur in the Ractor. E.g.

r2 = Ractor.new do
 loop { puts "[#{Process.pid}] Ractor"; sleep(1) }
end

Segfault can also be prevented by invoking close_incoming prior to forking, although this raises another error internally.

It also does not crash on MacOS 10.15.7: ruby 3.0.0p0 (2020-12-25 revision 95aff21468) [x86_64-darwin19].

Suggested solutions

(None)

[hsbt (Hiroshi SHIBATA)]

• Status changed from Open to Assigned
• Assignee set to ko1 (Koichi Sasada)

[hsbt (Hiroshi SHIBATA)]

• Tags set to ractor