Bug #19922: ObjectSpace.dump_all in multiple threads in parallel causes SEGV

• Status: Open
• Assignee: -
• Target version: -
• Backport: 3.0: UNKNOWN, 3.1: UNKNOWN, 3.2: UNKNOWN

[wanabe (_ wanabe)]

ObjectSpace.dump_all in multiple threads in parallel causes SEGV.
The reproduce code is here:

$ ruby -robjspace -e 'def dump; 10.times { ObjectSpace.dump_all } end; Thread.new { dump }; dump'
/home/wanabe/.rbenv/versions/master/lib/ruby/3.3.0+0/objspace.rb:100: [BUG] Segmentation fault at 0x0000000000000000
(snip)

I attached the full log.

I assumed that:

• Thread1
  • calls rb_objspace_reachable_objects_from_root()
  • changes GET_RACTOR()->mfd: mfd1
  • calls dump_flush()
    • changes current thread due to IO
• Thread2
  • calls rb_objspace_reachable_objects_from_root()
  • changes GET_RACTOR()->mfd: mfd2
  • calls dump_flush()
    • changes current thread due to IO
• Thread1
  • gets unexpected GET_RACTOR()->mfd: mfd2

[kjtsanaktsidis (KJ Tsanaktsidis)]

I think your diagnosis is correct, and that https://github.com/ruby/ruby/pull/8858 should fix this - does it look reasonable to you?

[wanabe (_ wanabe)]

Thank you. https://github.com/ruby/ruby/pull/8858 resolves my issue.

[byroot (Jean Boussier)]

• Status changed from Open to Closed

Fixed in 76dc327eeffefe02577999fe5f8215f762a581b6 and 9a62fd3cbae2ebb60e2f9cad782af1ad18db4319

[byroot (Jean Boussier)]

• Status changed from Closed to Open

Reopening because the fix caused a failure on ruby-ci, so we'll revert for now.